V0910 09:41:41.414000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/__run_lpar_main__.py", 0]}
V0910 09:41:41.416000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/__par__/meta_only/bootstrap.py", 1]}
V0910 09:41:41.417000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/__par__/bootstrap.py", 2]}
V0910 09:41:41.418000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["<frozen runpy>", 3]}
V0910 09:41:41.419000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/scripts/yimingzhou/voxtral_lowering.py", 4]}
V0910 09:41:41.420000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/export/passes/__init__.py", 5]}
V0910 09:41:41.421000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/utils/_pytree.py", 6]}
V0910 09:41:41.422000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/utils/_stats.py", 7]}
V0910 09:41:41.423000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_subclasses/fake_tensor.py", 8]}
V0910 09:41:41.424000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_ops.py", 9]}
V0910 09:41:41.425000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_prims_common/wrappers.py", 10]}
V0910 09:41:41.426000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_decomp/decompositions.py", 11]}
V0910 09:41:41.427000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_prims/__init__.py", 12]}
V0910 09:41:41.428000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_prims_common/__init__.py", 13]}
V0910 09:41:41.429000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/fx/experimental/symbolic_shapes.py", 14]}
V0910 09:41:41.430000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/fx/experimental/recording.py", 15]}
V0910 09:41:41.430000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_logging/_internal.py", 16]}
V0910 09:41:41.432000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/fx/experimental/symbolic_shapes.py:7190] {"guard_added_fast": {"expr": "Ne(s6, 1)", "user_stack": [], "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 57, "name": "process_model", "filename": 4, "loc": "cuda_ep = move_to_device_pass(ep, \"cuda:0\")"}, {"line": 73, "name": "move_to_device_pass", "filename": 5, "loc": "node.meta[\"val\"] = pytree.tree_map("}, {"line": 1376, "name": "tree_map", "filename": 6, "loc": "return treespec.unflatten(map(func, *flat_args))"}, {"line": 1193, "name": "unflatten", "filename": 6, "loc": "leaves = list(leaves)"}, {"line": 74, "name": "<lambda>", "filename": 5, "loc": "lambda v: v.to(_get_new_device(v.device, location))"}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 872, "name": "__torch_dispatch__", "filename": 8, "loc": "return func(*args, **kwargs)"}, {"line": 841, "name": "__call__", "filename": 9, "loc": "return self._op(*args, **kwargs)"}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1511, "name": "_cached_dispatch_impl", "filename": 8, "loc": "output = self._dispatch_impl(func, types, args, kwargs)"}, {"line": 2635, "name": "_dispatch_impl", "filename": 8, "loc": "decomposition_table[func](*args, **kwargs)"}, {"line": 309, "name": "_fn", "filename": 10, "loc": "result = fn(*args, **kwargs)"}, {"line": 2214, "name": "_to_copy", "filename": 11, "loc": "x_tensor = torch._prims.convert_element_type(x_tensor, dtype)"}, {"line": 841, "name": "__call__", "filename": 9, "loc": "return self._op(*args, **kwargs)"}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1511, "name": "_cached_dispatch_impl", "filename": 8, "loc": "output = self._dispatch_impl(func, types, args, kwargs)"}, {"line": 2657, "name": "_dispatch_impl", "filename": 8, "loc": "func.prim_meta_impl(*args, **kwargs)"}, {"line": 1920, "name": "_convert_element_type_meta", "filename": 12, "loc": "if torch._prims_common.is_non_overlapping_and_dense(a):"}, {"line": 531, "name": "is_non_overlapping_and_dense", "filename": 13, "loc": "return _is_non_overlapping_and_dense_or_false(a.shape, a.stride())"}, {"line": 516, "name": "_is_non_overlapping_and_dense_or_false", "filename": 13, "loc": "return check_contiguous_sizes_strides(sizes, strides, false_if_dde=True)"}, {"line": 282, "name": "check_contiguous_sizes_strides", "filename": 13, "loc": "if maybe_guard_or_false(x == 1):"}, {"line": 1406, "name": "guard_or_false", "filename": 14, "loc": "return _guard_or(a, False)"}, {"line": 1396, "name": "_guard_or", "filename": 14, "loc": "r = sym_node.shape_env.evaluate_sym_node("}, {"line": 7239, "name": "evaluate_sym_node", "filename": 14, "loc": "return self.evaluate_expr("}, {"line": 7339, "name": "evaluate_expr", "filename": 14, "loc": "return self._inner_evaluate_expr("}, {"line": 272, "name": "wrapper", "filename": 15, "loc": "return retlog(fn(*args, **kwargs))"}, {"line": 7362, "name": "_inner_evaluate_expr", "filename": 14, "loc": "return self._evaluate_expr("}, {"line": 7615, "name": "_evaluate_expr", "filename": 14, "loc": "self._log_guard(\"eval\", g, forcing_spec=forcing_spec)"}, {"line": 7190, "name": "_log_guard", "filename": 14, "loc": "trace_structured("}, {"line": 1346, "name": "trace_structured", "filename": 16, "loc": "record[name] = metadata_fn()"}]}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 57, "name": "process_model", "filename": 4, "loc": "cuda_ep = move_to_device_pass(ep, \"cuda:0\")"}, {"line": 73, "name": "move_to_device_pass", "filename": 5, "loc": "node.meta[\"val\"] = pytree.tree_map("}, {"line": 1376, "name": "tree_map", "filename": 6, "loc": "return treespec.unflatten(map(func, *flat_args))"}, {"line": 1193, "name": "unflatten", "filename": 6, "loc": "leaves = list(leaves)"}, {"line": 74, "name": "<lambda>", "filename": 5, "loc": "lambda v: v.to(_get_new_device(v.device, location))"}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 872, "name": "__torch_dispatch__", "filename": 8, "loc": "return func(*args, **kwargs)"}, {"line": 841, "name": "__call__", "filename": 9, "loc": "return self._op(*args, **kwargs)"}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1511, "name": "_cached_dispatch_impl", "filename": 8, "loc": "output = self._dispatch_impl(func, types, args, kwargs)"}, {"line": 2635, "name": "_dispatch_impl", "filename": 8, "loc": "decomposition_table[func](*args, **kwargs)"}, {"line": 309, "name": "_fn", "filename": 10, "loc": "result = fn(*args, **kwargs)"}, {"line": 2214, "name": "_to_copy", "filename": 11, "loc": "x_tensor = torch._prims.convert_element_type(x_tensor, dtype)"}, {"line": 841, "name": "__call__", "filename": 9, "loc": "return self._op(*args, **kwargs)"}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1511, "name": "_cached_dispatch_impl", "filename": 8, "loc": "output = self._dispatch_impl(func, types, args, kwargs)"}, {"line": 2657, "name": "_dispatch_impl", "filename": 8, "loc": "func.prim_meta_impl(*args, **kwargs)"}, {"line": 1920, "name": "_convert_element_type_meta", "filename": 12, "loc": "if torch._prims_common.is_non_overlapping_and_dense(a):"}, {"line": 531, "name": "is_non_overlapping_and_dense", "filename": 13, "loc": "return _is_non_overlapping_and_dense_or_false(a.shape, a.stride())"}, {"line": 516, "name": "_is_non_overlapping_and_dense_or_false", "filename": 13, "loc": "return check_contiguous_sizes_strides(sizes, strides, false_if_dde=True)"}, {"line": 282, "name": "check_contiguous_sizes_strides", "filename": 13, "loc": "if maybe_guard_or_false(x == 1):"}, {"line": 1406, "name": "guard_or_false", "filename": 14, "loc": "return _guard_or(a, False)"}, {"line": 1396, "name": "_guard_or", "filename": 14, "loc": "r = sym_node.shape_env.evaluate_sym_node("}, {"line": 7239, "name": "evaluate_sym_node", "filename": 14, "loc": "return self.evaluate_expr("}, {"line": 7339, "name": "evaluate_expr", "filename": 14, "loc": "return self._inner_evaluate_expr("}, {"line": 272, "name": "wrapper", "filename": 15, "loc": "return retlog(fn(*args, **kwargs))"}, {"line": 7362, "name": "_inner_evaluate_expr", "filename": 14, "loc": "return self._evaluate_expr("}, {"line": 7615, "name": "_evaluate_expr", "filename": 14, "loc": "self._log_guard(\"eval\", g, forcing_spec=forcing_spec)"}, {"line": 7190, "name": "_log_guard", "filename": 14, "loc": "trace_structured("}]}
V0910 09:41:48.238000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1b057b90bbfcf1e35bdb5a148124c0b7"}
	{
	"name": "compile_fx_aot",
	"ts": 1757522508238056.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:41:48.242000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "69b7d388231cae51856d747824d12a89"}
	{
	"name": "inductor_codecache_torch_key",
	"ts": 1757522508242475.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:41:48.244000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "333b25fafccb455f24f288650b18e19e"}
	{
	"name": "inductor_codecache_torch_key",
	"ts": 1757522508244399.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:41:48.305000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_inductor/__init__.py", 17]}
V0910 09:41:48.306000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_inductor/debug.py", 18]}
V0910 09:41:48.307000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_inductor/compile_fx.py", 19]}
V0910 09:41:48.450000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_inductor/compile_fx.py:2322] {"artifact": {"name": "before_pre_grad_graph", "encoding": "string"}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2511, "name": "compile_fx", "filename": 19, "loc": "model_ = run_pre_grad_passes(model_, example_inputs_)"}, {"line": 2322, "name": "run_pre_grad_passes", "filename": 19, "loc": "trace_structured("}], "has_payload": "52d2886661c5d8302c82502bddf2438b"}
	class GraphModule(torch.nn.Module):
	    def forward(self, input_features: "f32[s6, 128, 3000][384000, 3000, 1]cuda:0"):
	        # No stacktrace found for following nodes
	        model_audio_tower_embed_positions_weight: "f32[1500, 1280][1280, 1]cuda:0" = self.model.audio_tower.embed_positions.weight
	        model_audio_tower_conv1_weight: "f32[1280, 128, 3][384, 3, 1]cuda:0" = self.model.audio_tower.conv1.weight
	        model_audio_tower_conv1_bias: "f32[1280][1]cuda:0" = self.model.audio_tower.conv1.bias
	        model_audio_tower_conv2_weight: "f32[1280, 1280, 3][3840, 3, 1]cuda:0" = self.model.audio_tower.conv2.weight
	        model_audio_tower_conv2_bias: "f32[1280][1]cuda:0" = self.model.audio_tower.conv2.bias
	        model_audio_tower_layers_0_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn_layer_norm.weight
	        model_audio_tower_layers_0_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn_layer_norm.bias
	        model_audio_tower_layers_0_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.bias
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.bias
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.bias
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").final_layer_norm.weight
	        model_audio_tower_layers_0_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").final_layer_norm.bias
	        model_audio_tower_layers_0_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.bias
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_0_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.bias
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn_layer_norm.weight
	        model_audio_tower_layers_1_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn_layer_norm.bias
	        model_audio_tower_layers_1_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.bias
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.bias
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.bias
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").final_layer_norm.weight
	        model_audio_tower_layers_1_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").final_layer_norm.bias
	        model_audio_tower_layers_1_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.bias
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_1_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.bias
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn_layer_norm.weight
	        model_audio_tower_layers_2_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn_layer_norm.bias
	        model_audio_tower_layers_2_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.bias
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.bias
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.bias
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").final_layer_norm.weight
	        model_audio_tower_layers_2_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").final_layer_norm.bias
	        model_audio_tower_layers_2_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.bias
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_2_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.bias
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn_layer_norm.weight
	        model_audio_tower_layers_3_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn_layer_norm.bias
	        model_audio_tower_layers_3_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.bias
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.bias
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.bias
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").final_layer_norm.weight
	        model_audio_tower_layers_3_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").final_layer_norm.bias
	        model_audio_tower_layers_3_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.bias
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_3_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.bias
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn_layer_norm.weight
	        model_audio_tower_layers_4_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn_layer_norm.bias
	        model_audio_tower_layers_4_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.bias
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.bias
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.bias
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").final_layer_norm.weight
	        model_audio_tower_layers_4_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").final_layer_norm.bias
	        model_audio_tower_layers_4_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.bias
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_4_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.bias
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn_layer_norm.weight
	        model_audio_tower_layers_5_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn_layer_norm.bias
	        model_audio_tower_layers_5_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.bias
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.bias
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.bias
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").final_layer_norm.weight
	        model_audio_tower_layers_5_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").final_layer_norm.bias
	        model_audio_tower_layers_5_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.bias
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_5_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.bias
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn_layer_norm.weight
	        model_audio_tower_layers_6_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn_layer_norm.bias
	        model_audio_tower_layers_6_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.bias
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.bias
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.bias
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").final_layer_norm.weight
	        model_audio_tower_layers_6_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").final_layer_norm.bias
	        model_audio_tower_layers_6_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.bias
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_6_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.bias
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn_layer_norm.weight
	        model_audio_tower_layers_7_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn_layer_norm.bias
	        model_audio_tower_layers_7_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.bias
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.bias
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.bias
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").final_layer_norm.weight
	        model_audio_tower_layers_7_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").final_layer_norm.bias
	        model_audio_tower_layers_7_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.bias
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_7_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.bias
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn_layer_norm.weight
	        model_audio_tower_layers_8_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn_layer_norm.bias
	        model_audio_tower_layers_8_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.bias
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.bias
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.bias
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").final_layer_norm.weight
	        model_audio_tower_layers_8_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").final_layer_norm.bias
	        model_audio_tower_layers_8_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.bias
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_8_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.bias
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn_layer_norm.weight
	        model_audio_tower_layers_9_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn_layer_norm.bias
	        model_audio_tower_layers_9_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.bias
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.bias
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.bias
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").final_layer_norm.weight
	        model_audio_tower_layers_9_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").final_layer_norm.bias
	        model_audio_tower_layers_9_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.bias
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_9_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.bias
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn_layer_norm.weight
	        model_audio_tower_layers_10_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn_layer_norm.bias
	        model_audio_tower_layers_10_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.bias
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.bias
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.bias
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").final_layer_norm.weight
	        model_audio_tower_layers_10_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").final_layer_norm.bias
	        model_audio_tower_layers_10_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.bias
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_10_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.bias
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn_layer_norm.weight
	        model_audio_tower_layers_11_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn_layer_norm.bias
	        model_audio_tower_layers_11_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.bias
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.bias
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.bias
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").final_layer_norm.weight
	        model_audio_tower_layers_11_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").final_layer_norm.bias
	        model_audio_tower_layers_11_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.bias
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_11_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.bias
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn_layer_norm.weight
	        model_audio_tower_layers_12_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn_layer_norm.bias
	        model_audio_tower_layers_12_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.bias
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.bias
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.bias
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").final_layer_norm.weight
	        model_audio_tower_layers_12_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").final_layer_norm.bias
	        model_audio_tower_layers_12_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.bias
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_12_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.bias
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn_layer_norm.weight
	        model_audio_tower_layers_13_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn_layer_norm.bias
	        model_audio_tower_layers_13_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.bias
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.bias
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.bias
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").final_layer_norm.weight
	        model_audio_tower_layers_13_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").final_layer_norm.bias
	        model_audio_tower_layers_13_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.bias
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_13_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.bias
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn_layer_norm.weight
	        model_audio_tower_layers_14_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn_layer_norm.bias
	        model_audio_tower_layers_14_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.bias
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.bias
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.bias
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").final_layer_norm.weight
	        model_audio_tower_layers_14_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").final_layer_norm.bias
	        model_audio_tower_layers_14_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.bias
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_14_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.bias
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn_layer_norm.weight
	        model_audio_tower_layers_15_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn_layer_norm.bias
	        model_audio_tower_layers_15_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.bias
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.bias
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.bias
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").final_layer_norm.weight
	        model_audio_tower_layers_15_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").final_layer_norm.bias
	        model_audio_tower_layers_15_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.bias
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_15_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.bias
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn_layer_norm.weight
	        model_audio_tower_layers_16_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn_layer_norm.bias
	        model_audio_tower_layers_16_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.bias
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.bias
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.bias
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").final_layer_norm.weight
	        model_audio_tower_layers_16_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").final_layer_norm.bias
	        model_audio_tower_layers_16_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.bias
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_16_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.bias
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn_layer_norm.weight
	        model_audio_tower_layers_17_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn_layer_norm.bias
	        model_audio_tower_layers_17_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.bias
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.bias
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.bias
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").final_layer_norm.weight
	        model_audio_tower_layers_17_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").final_layer_norm.bias
	        model_audio_tower_layers_17_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.bias
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_17_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.bias
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn_layer_norm.weight
	        model_audio_tower_layers_18_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn_layer_norm.bias
	        model_audio_tower_layers_18_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.bias
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.bias
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.bias
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").final_layer_norm.weight
	        model_audio_tower_layers_18_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").final_layer_norm.bias
	        model_audio_tower_layers_18_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.bias
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_18_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.bias
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn_layer_norm.weight
	        model_audio_tower_layers_19_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn_layer_norm.bias
	        model_audio_tower_layers_19_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.bias
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.bias
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.bias
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").final_layer_norm.weight
	        model_audio_tower_layers_19_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").final_layer_norm.bias
	        model_audio_tower_layers_19_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.bias
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_19_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.bias
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn_layer_norm.weight
	        model_audio_tower_layers_20_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn_layer_norm.bias
	        model_audio_tower_layers_20_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.bias
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.bias
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.bias
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").final_layer_norm.weight
	        model_audio_tower_layers_20_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").final_layer_norm.bias
	        model_audio_tower_layers_20_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.bias
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_20_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.bias
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn_layer_norm.weight
	        model_audio_tower_layers_21_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn_layer_norm.bias
	        model_audio_tower_layers_21_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.bias
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.bias
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.bias
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").final_layer_norm.weight
	        model_audio_tower_layers_21_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").final_layer_norm.bias
	        model_audio_tower_layers_21_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.bias
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_21_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.bias
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn_layer_norm.weight
	        model_audio_tower_layers_22_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn_layer_norm.bias
	        model_audio_tower_layers_22_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.bias
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.bias
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.bias
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").final_layer_norm.weight
	        model_audio_tower_layers_22_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").final_layer_norm.bias
	        model_audio_tower_layers_22_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.bias
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_22_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.bias
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn_layer_norm.weight
	        model_audio_tower_layers_23_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn_layer_norm.bias
	        model_audio_tower_layers_23_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.bias
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.bias
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.bias
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").final_layer_norm.weight
	        model_audio_tower_layers_23_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").final_layer_norm.bias
	        model_audio_tower_layers_23_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.bias
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_23_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.bias
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn_layer_norm.weight
	        model_audio_tower_layers_24_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn_layer_norm.bias
	        model_audio_tower_layers_24_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.bias
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.bias
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.bias
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").final_layer_norm.weight
	        model_audio_tower_layers_24_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").final_layer_norm.bias
	        model_audio_tower_layers_24_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.bias
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_24_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.bias
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn_layer_norm.weight
	        model_audio_tower_layers_25_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn_layer_norm.bias
	        model_audio_tower_layers_25_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.bias
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.bias
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.bias
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").final_layer_norm.weight
	        model_audio_tower_layers_25_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").final_layer_norm.bias
	        model_audio_tower_layers_25_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.bias
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_25_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.bias
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn_layer_norm.weight
	        model_audio_tower_layers_26_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn_layer_norm.bias
	        model_audio_tower_layers_26_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.bias
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.bias
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.bias
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").final_layer_norm.weight
	        model_audio_tower_layers_26_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").final_layer_norm.bias
	        model_audio_tower_layers_26_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.bias
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_26_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.bias
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn_layer_norm.weight
	        model_audio_tower_layers_27_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn_layer_norm.bias
	        model_audio_tower_layers_27_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.bias
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.bias
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.bias
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").final_layer_norm.weight
	        model_audio_tower_layers_27_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").final_layer_norm.bias
	        model_audio_tower_layers_27_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.bias
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_27_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.bias
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn_layer_norm.weight
	        model_audio_tower_layers_28_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn_layer_norm.bias
	        model_audio_tower_layers_28_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.bias
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.bias
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.bias
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").final_layer_norm.weight
	        model_audio_tower_layers_28_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").final_layer_norm.bias
	        model_audio_tower_layers_28_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.bias
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_28_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.bias
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn_layer_norm.weight
	        model_audio_tower_layers_29_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn_layer_norm.bias
	        model_audio_tower_layers_29_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.bias
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.bias
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.bias
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").final_layer_norm.weight
	        model_audio_tower_layers_29_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").final_layer_norm.bias
	        model_audio_tower_layers_29_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.bias
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_29_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.bias
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn_layer_norm.weight
	        model_audio_tower_layers_30_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn_layer_norm.bias
	        model_audio_tower_layers_30_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.bias
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.bias
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.bias
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").final_layer_norm.weight
	        model_audio_tower_layers_30_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").final_layer_norm.bias
	        model_audio_tower_layers_30_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.bias
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_30_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.bias
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn_layer_norm.weight
	        model_audio_tower_layers_31_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn_layer_norm.bias
	        model_audio_tower_layers_31_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.bias
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.bias
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.bias
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").final_layer_norm.weight
	        model_audio_tower_layers_31_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").final_layer_norm.bias
	        model_audio_tower_layers_31_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.bias
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_31_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.bias
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original2
	        model_audio_tower_layer_norm_weight: "f32[1280][1]cuda:0" = self.model.audio_tower.layer_norm.weight
	        model_audio_tower_layer_norm_bias: "f32[1280][1]cuda:0" = self.model.audio_tower.layer_norm.bias
	        model_multi_modal_projector_linear_1_parametrizations_weight_original0: "i8[3072, 5120][5120, 1]cuda:0" = self.model.multi_modal_projector.linear_1.parametrizations.weight.original0
	        model_multi_modal_projector_linear_1_parametrizations_weight_original1: "f32[3072, 160][160, 1]cuda:0" = self.model.multi_modal_projector.linear_1.parametrizations.weight.original1
	        model_multi_modal_projector_linear_1_parametrizations_weight_original2: "i8[3072, 160][160, 1]cuda:0" = self.model.multi_modal_projector.linear_1.parametrizations.weight.original2
	        model_multi_modal_projector_linear_2_parametrizations_weight_original0: "i8[3072, 3072][3072, 1]cuda:0" = self.model.multi_modal_projector.linear_2.parametrizations.weight.original0
	        model_multi_modal_projector_linear_2_parametrizations_weight_original1: "f32[3072, 96][96, 1]cuda:0" = self.model.multi_modal_projector.linear_2.parametrizations.weight.original1
	        model_multi_modal_projector_linear_2_parametrizations_weight_original2: "i8[3072, 96][96, 1]cuda:0" = self.model.multi_modal_projector.linear_2.parametrizations.weight.original2
	        
	         # 
	        sym_size_int_2: "Sym(s6)" = torch.ops.aten.sym_size.int(input_features, 0)
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:348 in forward, code: input_features = input_features.to(dtype=self.conv1.weight.dtype, device=self.conv1.weight.device)
	        _assert_tensor_metadata_default = torch.ops.aten._assert_tensor_metadata.default(input_features, dtype = torch.float32, device = 'cuda:0', layout = torch.strided);  _assert_tensor_metadata_default = None
	        to: "f32[s6, 128, 3000][384000, 3000, 1]cuda:0" = torch.ops.aten.to.device(input_features, 'cuda:0', torch.float32);  input_features = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:349 in forward, code: inputs_embeds = nn.functional.gelu(self.conv1(input_features))
	        conv1d: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.conv1d.default(to, model_audio_tower_conv1_weight, model_audio_tower_conv1_bias, [1], [1]);  to = model_audio_tower_conv1_weight = model_audio_tower_conv1_bias = None
	        gelu: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.gelu.default(conv1d);  conv1d = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:350 in forward, code: inputs_embeds = nn.functional.gelu(self.conv2(inputs_embeds))
	        conv1d_1: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.conv1d.default(gelu, model_audio_tower_conv2_weight, model_audio_tower_conv2_bias, [2], [1]);  gelu = model_audio_tower_conv2_weight = model_audio_tower_conv2_bias = None
	        gelu_1: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.gelu.default(conv1d_1);  conv1d_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:351 in forward, code: inputs_embeds = inputs_embeds.permute(0, 2, 1)
	        permute: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.permute.default(gelu_1, [0, 2, 1]);  gelu_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:354 in forward, code: hidden_states = (inputs_embeds + embed_pos).to(inputs_embeds.dtype)
	        add: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(permute, model_audio_tower_embed_positions_weight);  permute = model_audio_tower_embed_positions_weight = None
	        _assert_tensor_metadata_default_1 = torch.ops.aten._assert_tensor_metadata.default(add, dtype = torch.float32, device = 'cuda:0', layout = torch.strided);  _assert_tensor_metadata_default_1 = None
	        to_1: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.to.dtype(add, torch.float32);  add = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:355 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.dropout.default(to_1, 0.0, False);  to_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(dropout, [1280], model_audio_tower_layers_0_self_attn_layer_norm_weight, model_audio_tower_layers_0_self_attn_layer_norm_bias);  model_audio_tower_layers_0_self_attn_layer_norm_weight = model_audio_tower_layers_0_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default = torch.ops.torchao.choose_qparams_affine.default(layer_norm, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default[0]
	        getitem_1: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default[1];  choose_qparams_affine_default = None
	        quantize_affine: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm, [1, 1, 1280], getitem, getitem_1, torch.int8)
	        dequantize_affine: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine, [1, 1, 1280], getitem, getitem_1, torch.int8);  quantize_affine = getitem = getitem_1 = None
	        dequantize_affine_1: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine, dequantize_affine_1, model_audio_tower_layers_0_self_attn_q_proj_bias);  dequantize_affine = dequantize_affine_1 = model_audio_tower_layers_0_self_attn_q_proj_bias = None
	        mul_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear, 0.125);  linear = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_5, [sym_size_int_2, 1500, 20, 64]);  mul_5 = None
	        transpose: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view, 1, 2);  view = None
	        contiguous: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose);  transpose = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_1 = torch.ops.torchao.choose_qparams_affine.default(layer_norm, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_2: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_1[0]
	        getitem_3: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_1[1];  choose_qparams_affine_default_1 = None
	        quantize_affine_1: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm, [1, 1, 1280], getitem_2, getitem_3, torch.int8)
	        dequantize_affine_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_1, [1, 1, 1280], getitem_2, getitem_3, torch.int8);  quantize_affine_1 = getitem_2 = getitem_3 = None
	        dequantize_affine_3: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_2, dequantize_affine_3);  dequantize_affine_2 = dequantize_affine_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_1, [sym_size_int_2, -1, 20, 64]);  linear_1 = None
	        transpose_1: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_1, 1, 2);  view_1 = None
	        contiguous_1: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_1);  transpose_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_2 = torch.ops.torchao.choose_qparams_affine.default(layer_norm, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_4: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_2[0]
	        getitem_5: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_2[1];  choose_qparams_affine_default_2 = None
	        quantize_affine_2: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm, [1, 1, 1280], getitem_4, getitem_5, torch.int8);  layer_norm = None
	        dequantize_affine_4: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_2, [1, 1, 1280], getitem_4, getitem_5, torch.int8);  quantize_affine_2 = getitem_4 = getitem_5 = None
	        dequantize_affine_5: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_4, dequantize_affine_5, model_audio_tower_layers_0_self_attn_v_proj_bias);  dequantize_affine_4 = dequantize_affine_5 = model_audio_tower_layers_0_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_2, [sym_size_int_2, -1, 20, 64]);  linear_2 = None
	        transpose_2: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_2, 1, 2);  view_2 = None
	        contiguous_2: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_2);  transpose_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous, contiguous_1, contiguous_2, scale = 1.0);  contiguous = contiguous_1 = contiguous_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_3: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention, 1, 2);  scaled_dot_product_attention = None
	        contiguous_3: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_3);  transpose_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_3, [sym_size_int_2, 1500, -1]);  contiguous_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_3 = torch.ops.torchao.choose_qparams_affine.default(reshape, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_6: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_3[0]
	        getitem_7: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_3[1];  choose_qparams_affine_default_3 = None
	        quantize_affine_3: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape, [1, 1, 1280], getitem_6, getitem_7, torch.int8);  reshape = None
	        dequantize_affine_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_3, [1, 1, 1280], getitem_6, getitem_7, torch.int8);  quantize_affine_3 = getitem_6 = getitem_7 = None
	        dequantize_affine_7: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_3: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_6, dequantize_affine_7, model_audio_tower_layers_0_self_attn_out_proj_bias);  dequantize_affine_6 = dequantize_affine_7 = model_audio_tower_layers_0_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_3, 0.0, False);  linear_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_9: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(dropout, dropout_1);  dropout = dropout_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_9, [1280], model_audio_tower_layers_0_final_layer_norm_weight, model_audio_tower_layers_0_final_layer_norm_bias);  model_audio_tower_layers_0_final_layer_norm_weight = model_audio_tower_layers_0_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_4 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_1, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_8: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_4[0]
	        getitem_9: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_4[1];  choose_qparams_affine_default_4 = None
	        quantize_affine_4: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_1, [1, 1, 1280], getitem_8, getitem_9, torch.int8);  layer_norm_1 = None
	        dequantize_affine_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_4, [1, 1, 1280], getitem_8, getitem_9, torch.int8);  quantize_affine_4 = getitem_8 = getitem_9 = None
	        dequantize_affine_9: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_0_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_0_fc1_parametrizations_weight_original1, model_audio_tower_layers_0_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_0_fc1_parametrizations_weight_original0 = model_audio_tower_layers_0_fc1_parametrizations_weight_original1 = model_audio_tower_layers_0_fc1_parametrizations_weight_original2 = None
	        linear_4: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_8, dequantize_affine_9, model_audio_tower_layers_0_fc1_bias);  dequantize_affine_8 = dequantize_affine_9 = model_audio_tower_layers_0_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_2: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_4);  linear_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_2: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_2, 0.0, False);  gelu_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_5 = torch.ops.torchao.choose_qparams_affine.default(dropout_2, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_10: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_5[0]
	        getitem_11: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_5[1];  choose_qparams_affine_default_5 = None
	        quantize_affine_5: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_2, [1, 1, 5120], getitem_10, getitem_11, torch.int8);  dropout_2 = None
	        dequantize_affine_10: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_5, [1, 1, 5120], getitem_10, getitem_11, torch.int8);  quantize_affine_5 = getitem_10 = getitem_11 = None
	        dequantize_affine_11: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_0_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_0_fc2_parametrizations_weight_original1, model_audio_tower_layers_0_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_0_fc2_parametrizations_weight_original0 = model_audio_tower_layers_0_fc2_parametrizations_weight_original1 = model_audio_tower_layers_0_fc2_parametrizations_weight_original2 = None
	        linear_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_10, dequantize_affine_11, model_audio_tower_layers_0_fc2_bias);  dequantize_affine_10 = dequantize_affine_11 = model_audio_tower_layers_0_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_3: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_5, 0.0, False);  linear_5 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_14: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_9, dropout_3);  add_9 = dropout_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_14, [1280], model_audio_tower_layers_1_self_attn_layer_norm_weight, model_audio_tower_layers_1_self_attn_layer_norm_bias);  model_audio_tower_layers_1_self_attn_layer_norm_weight = model_audio_tower_layers_1_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_6 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_2, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_12: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_6[0]
	        getitem_13: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_6[1];  choose_qparams_affine_default_6 = None
	        quantize_affine_6: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_2, [1, 1, 1280], getitem_12, getitem_13, torch.int8)
	        dequantize_affine_12: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_6, [1, 1, 1280], getitem_12, getitem_13, torch.int8);  quantize_affine_6 = getitem_12 = getitem_13 = None
	        dequantize_affine_13: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_12, dequantize_affine_13, model_audio_tower_layers_1_self_attn_q_proj_bias);  dequantize_affine_12 = dequantize_affine_13 = model_audio_tower_layers_1_self_attn_q_proj_bias = None
	        mul_42: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_6, 0.125);  linear_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_3: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_42, [sym_size_int_2, 1500, 20, 64]);  mul_42 = None
	        transpose_4: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_3, 1, 2);  view_3 = None
	        contiguous_4: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_4);  transpose_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_7 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_2, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_14: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_7[0]
	        getitem_15: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_7[1];  choose_qparams_affine_default_7 = None
	        quantize_affine_7: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_2, [1, 1, 1280], getitem_14, getitem_15, torch.int8)
	        dequantize_affine_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_7, [1, 1, 1280], getitem_14, getitem_15, torch.int8);  quantize_affine_7 = getitem_14 = getitem_15 = None
	        dequantize_affine_15: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_7: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_14, dequantize_affine_15);  dequantize_affine_14 = dequantize_affine_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_4: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_7, [sym_size_int_2, -1, 20, 64]);  linear_7 = None
	        transpose_5: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_4, 1, 2);  view_4 = None
	        contiguous_5: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_5);  transpose_5 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_8 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_2, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_16: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_8[0]
	        getitem_17: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_8[1];  choose_qparams_affine_default_8 = None
	        quantize_affine_8: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_2, [1, 1, 1280], getitem_16, getitem_17, torch.int8);  layer_norm_2 = None
	        dequantize_affine_16: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_8, [1, 1, 1280], getitem_16, getitem_17, torch.int8);  quantize_affine_8 = getitem_16 = getitem_17 = None
	        dequantize_affine_17: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_16, dequantize_affine_17, model_audio_tower_layers_1_self_attn_v_proj_bias);  dequantize_affine_16 = dequantize_affine_17 = model_audio_tower_layers_1_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_5: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_8, [sym_size_int_2, -1, 20, 64]);  linear_8 = None
	        transpose_6: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_5, 1, 2);  view_5 = None
	        contiguous_6: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_6);  transpose_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_1: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_4, contiguous_5, contiguous_6, scale = 1.0);  contiguous_4 = contiguous_5 = contiguous_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_7: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_1, 1, 2);  scaled_dot_product_attention_1 = None
	        contiguous_7: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_7);  transpose_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_7, [sym_size_int_2, 1500, -1]);  contiguous_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_9 = torch.ops.torchao.choose_qparams_affine.default(reshape_1, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_18: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_9[0]
	        getitem_19: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_9[1];  choose_qparams_affine_default_9 = None
	        quantize_affine_9: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_1, [1, 1, 1280], getitem_18, getitem_19, torch.int8);  reshape_1 = None
	        dequantize_affine_18: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_9, [1, 1, 1280], getitem_18, getitem_19, torch.int8);  quantize_affine_9 = getitem_18 = getitem_19 = None
	        dequantize_affine_19: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_9: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_18, dequantize_affine_19, model_audio_tower_layers_1_self_attn_out_proj_bias);  dequantize_affine_18 = dequantize_affine_19 = model_audio_tower_layers_1_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_4: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_9, 0.0, False);  linear_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_23: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_14, dropout_4);  add_14 = dropout_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_3: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_23, [1280], model_audio_tower_layers_1_final_layer_norm_weight, model_audio_tower_layers_1_final_layer_norm_bias);  model_audio_tower_layers_1_final_layer_norm_weight = model_audio_tower_layers_1_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_10 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_3, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_20: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_10[0]
	        getitem_21: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_10[1];  choose_qparams_affine_default_10 = None
	        quantize_affine_10: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_3, [1, 1, 1280], getitem_20, getitem_21, torch.int8);  layer_norm_3 = None
	        dequantize_affine_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_10, [1, 1, 1280], getitem_20, getitem_21, torch.int8);  quantize_affine_10 = getitem_20 = getitem_21 = None
	        dequantize_affine_21: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_1_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_1_fc1_parametrizations_weight_original1, model_audio_tower_layers_1_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_1_fc1_parametrizations_weight_original0 = model_audio_tower_layers_1_fc1_parametrizations_weight_original1 = model_audio_tower_layers_1_fc1_parametrizations_weight_original2 = None
	        linear_10: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_20, dequantize_affine_21, model_audio_tower_layers_1_fc1_bias);  dequantize_affine_20 = dequantize_affine_21 = model_audio_tower_layers_1_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_3: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_10);  linear_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_5: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_3, 0.0, False);  gelu_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_11 = torch.ops.torchao.choose_qparams_affine.default(dropout_5, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_22: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_11[0]
	        getitem_23: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_11[1];  choose_qparams_affine_default_11 = None
	        quantize_affine_11: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_5, [1, 1, 5120], getitem_22, getitem_23, torch.int8);  dropout_5 = None
	        dequantize_affine_22: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_11, [1, 1, 5120], getitem_22, getitem_23, torch.int8);  quantize_affine_11 = getitem_22 = getitem_23 = None
	        dequantize_affine_23: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_1_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_1_fc2_parametrizations_weight_original1, model_audio_tower_layers_1_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_1_fc2_parametrizations_weight_original0 = model_audio_tower_layers_1_fc2_parametrizations_weight_original1 = model_audio_tower_layers_1_fc2_parametrizations_weight_original2 = None
	        linear_11: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_22, dequantize_affine_23, model_audio_tower_layers_1_fc2_bias);  dequantize_affine_22 = dequantize_affine_23 = model_audio_tower_layers_1_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_11, 0.0, False);  linear_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_28: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_23, dropout_6);  add_23 = dropout_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_4: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_28, [1280], model_audio_tower_layers_2_self_attn_layer_norm_weight, model_audio_tower_layers_2_self_attn_layer_norm_bias);  model_audio_tower_layers_2_self_attn_layer_norm_weight = model_audio_tower_layers_2_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_12 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_4, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_24: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_12[0]
	        getitem_25: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_12[1];  choose_qparams_affine_default_12 = None
	        quantize_affine_12: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_4, [1, 1, 1280], getitem_24, getitem_25, torch.int8)
	        dequantize_affine_24: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_12, [1, 1, 1280], getitem_24, getitem_25, torch.int8);  quantize_affine_12 = getitem_24 = getitem_25 = None
	        dequantize_affine_25: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_12: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_24, dequantize_affine_25, model_audio_tower_layers_2_self_attn_q_proj_bias);  dequantize_affine_24 = dequantize_affine_25 = model_audio_tower_layers_2_self_attn_q_proj_bias = None
	        mul_79: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_12, 0.125);  linear_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_6: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_79, [sym_size_int_2, 1500, 20, 64]);  mul_79 = None
	        transpose_8: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_6, 1, 2);  view_6 = None
	        contiguous_8: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_8);  transpose_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_13 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_4, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_26: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_13[0]
	        getitem_27: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_13[1];  choose_qparams_affine_default_13 = None
	        quantize_affine_13: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_4, [1, 1, 1280], getitem_26, getitem_27, torch.int8)
	        dequantize_affine_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_13, [1, 1, 1280], getitem_26, getitem_27, torch.int8);  quantize_affine_13 = getitem_26 = getitem_27 = None
	        dequantize_affine_27: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_13: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_26, dequantize_affine_27);  dequantize_affine_26 = dequantize_affine_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_7: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_13, [sym_size_int_2, -1, 20, 64]);  linear_13 = None
	        transpose_9: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_7, 1, 2);  view_7 = None
	        contiguous_9: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_9);  transpose_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_14 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_4, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_28: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_14[0]
	        getitem_29: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_14[1];  choose_qparams_affine_default_14 = None
	        quantize_affine_14: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_4, [1, 1, 1280], getitem_28, getitem_29, torch.int8);  layer_norm_4 = None
	        dequantize_affine_28: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_14, [1, 1, 1280], getitem_28, getitem_29, torch.int8);  quantize_affine_14 = getitem_28 = getitem_29 = None
	        dequantize_affine_29: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_28, dequantize_affine_29, model_audio_tower_layers_2_self_attn_v_proj_bias);  dequantize_affine_28 = dequantize_affine_29 = model_audio_tower_layers_2_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_8: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_14, [sym_size_int_2, -1, 20, 64]);  linear_14 = None
	        transpose_10: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_8, 1, 2);  view_8 = None
	        contiguous_10: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_10);  transpose_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_2: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_8, contiguous_9, contiguous_10, scale = 1.0);  contiguous_8 = contiguous_9 = contiguous_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_11: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_2, 1, 2);  scaled_dot_product_attention_2 = None
	        contiguous_11: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_11);  transpose_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_11, [sym_size_int_2, 1500, -1]);  contiguous_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_15 = torch.ops.torchao.choose_qparams_affine.default(reshape_2, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_30: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_15[0]
	        getitem_31: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_15[1];  choose_qparams_affine_default_15 = None
	        quantize_affine_15: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_2, [1, 1, 1280], getitem_30, getitem_31, torch.int8);  reshape_2 = None
	        dequantize_affine_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_15, [1, 1, 1280], getitem_30, getitem_31, torch.int8);  quantize_affine_15 = getitem_30 = getitem_31 = None
	        dequantize_affine_31: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_15: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_30, dequantize_affine_31, model_audio_tower_layers_2_self_attn_out_proj_bias);  dequantize_affine_30 = dequantize_affine_31 = model_audio_tower_layers_2_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_7: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_15, 0.0, False);  linear_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_37: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_28, dropout_7);  add_28 = dropout_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_37, [1280], model_audio_tower_layers_2_final_layer_norm_weight, model_audio_tower_layers_2_final_layer_norm_bias);  model_audio_tower_layers_2_final_layer_norm_weight = model_audio_tower_layers_2_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_16 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_5, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_32: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_16[0]
	        getitem_33: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_16[1];  choose_qparams_affine_default_16 = None
	        quantize_affine_16: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_5, [1, 1, 1280], getitem_32, getitem_33, torch.int8);  layer_norm_5 = None
	        dequantize_affine_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_16, [1, 1, 1280], getitem_32, getitem_33, torch.int8);  quantize_affine_16 = getitem_32 = getitem_33 = None
	        dequantize_affine_33: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_2_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_2_fc1_parametrizations_weight_original1, model_audio_tower_layers_2_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_2_fc1_parametrizations_weight_original0 = model_audio_tower_layers_2_fc1_parametrizations_weight_original1 = model_audio_tower_layers_2_fc1_parametrizations_weight_original2 = None
	        linear_16: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_32, dequantize_affine_33, model_audio_tower_layers_2_fc1_bias);  dequantize_affine_32 = dequantize_affine_33 = model_audio_tower_layers_2_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_4: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_16);  linear_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_8: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_4, 0.0, False);  gelu_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_17 = torch.ops.torchao.choose_qparams_affine.default(dropout_8, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_34: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_17[0]
	        getitem_35: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_17[1];  choose_qparams_affine_default_17 = None
	        quantize_affine_17: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_8, [1, 1, 5120], getitem_34, getitem_35, torch.int8);  dropout_8 = None
	        dequantize_affine_34: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_17, [1, 1, 5120], getitem_34, getitem_35, torch.int8);  quantize_affine_17 = getitem_34 = getitem_35 = None
	        dequantize_affine_35: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_2_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_2_fc2_parametrizations_weight_original1, model_audio_tower_layers_2_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_2_fc2_parametrizations_weight_original0 = model_audio_tower_layers_2_fc2_parametrizations_weight_original1 = model_audio_tower_layers_2_fc2_parametrizations_weight_original2 = None
	        linear_17: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_34, dequantize_affine_35, model_audio_tower_layers_2_fc2_bias);  dequantize_affine_34 = dequantize_affine_35 = model_audio_tower_layers_2_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_9: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_17, 0.0, False);  linear_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_42: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_37, dropout_9);  add_37 = dropout_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_42, [1280], model_audio_tower_layers_3_self_attn_layer_norm_weight, model_audio_tower_layers_3_self_attn_layer_norm_bias);  model_audio_tower_layers_3_self_attn_layer_norm_weight = model_audio_tower_layers_3_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_18 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_6, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_36: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_18[0]
	        getitem_37: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_18[1];  choose_qparams_affine_default_18 = None
	        quantize_affine_18: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_6, [1, 1, 1280], getitem_36, getitem_37, torch.int8)
	        dequantize_affine_36: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_18, [1, 1, 1280], getitem_36, getitem_37, torch.int8);  quantize_affine_18 = getitem_36 = getitem_37 = None
	        dequantize_affine_37: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_18: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_36, dequantize_affine_37, model_audio_tower_layers_3_self_attn_q_proj_bias);  dequantize_affine_36 = dequantize_affine_37 = model_audio_tower_layers_3_self_attn_q_proj_bias = None
	        mul_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_18, 0.125);  linear_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_9: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_116, [sym_size_int_2, 1500, 20, 64]);  mul_116 = None
	        transpose_12: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_9, 1, 2);  view_9 = None
	        contiguous_12: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_12);  transpose_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_19 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_6, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_38: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_19[0]
	        getitem_39: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_19[1];  choose_qparams_affine_default_19 = None
	        quantize_affine_19: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_6, [1, 1, 1280], getitem_38, getitem_39, torch.int8)
	        dequantize_affine_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_19, [1, 1, 1280], getitem_38, getitem_39, torch.int8);  quantize_affine_19 = getitem_38 = getitem_39 = None
	        dequantize_affine_39: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_19: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_38, dequantize_affine_39);  dequantize_affine_38 = dequantize_affine_39 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_10: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_19, [sym_size_int_2, -1, 20, 64]);  linear_19 = None
	        transpose_13: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_10, 1, 2);  view_10 = None
	        contiguous_13: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_13);  transpose_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_20 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_6, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_40: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_20[0]
	        getitem_41: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_20[1];  choose_qparams_affine_default_20 = None
	        quantize_affine_20: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_6, [1, 1, 1280], getitem_40, getitem_41, torch.int8);  layer_norm_6 = None
	        dequantize_affine_40: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_20, [1, 1, 1280], getitem_40, getitem_41, torch.int8);  quantize_affine_20 = getitem_40 = getitem_41 = None
	        dequantize_affine_41: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_40, dequantize_affine_41, model_audio_tower_layers_3_self_attn_v_proj_bias);  dequantize_affine_40 = dequantize_affine_41 = model_audio_tower_layers_3_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_11: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_20, [sym_size_int_2, -1, 20, 64]);  linear_20 = None
	        transpose_14: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_11, 1, 2);  view_11 = None
	        contiguous_14: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_14);  transpose_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_3: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_12, contiguous_13, contiguous_14, scale = 1.0);  contiguous_12 = contiguous_13 = contiguous_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_15: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_3, 1, 2);  scaled_dot_product_attention_3 = None
	        contiguous_15: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_15);  transpose_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_3: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_15, [sym_size_int_2, 1500, -1]);  contiguous_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_21 = torch.ops.torchao.choose_qparams_affine.default(reshape_3, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_42: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_21[0]
	        getitem_43: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_21[1];  choose_qparams_affine_default_21 = None
	        quantize_affine_21: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_3, [1, 1, 1280], getitem_42, getitem_43, torch.int8);  reshape_3 = None
	        dequantize_affine_42: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_21, [1, 1, 1280], getitem_42, getitem_43, torch.int8);  quantize_affine_21 = getitem_42 = getitem_43 = None
	        dequantize_affine_43: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_42, dequantize_affine_43, model_audio_tower_layers_3_self_attn_out_proj_bias);  dequantize_affine_42 = dequantize_affine_43 = model_audio_tower_layers_3_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_10: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_21, 0.0, False);  linear_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_51: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_42, dropout_10);  add_42 = dropout_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_7: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_51, [1280], model_audio_tower_layers_3_final_layer_norm_weight, model_audio_tower_layers_3_final_layer_norm_bias);  model_audio_tower_layers_3_final_layer_norm_weight = model_audio_tower_layers_3_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_22 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_7, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_44: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_22[0]
	        getitem_45: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_22[1];  choose_qparams_affine_default_22 = None
	        quantize_affine_22: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_7, [1, 1, 1280], getitem_44, getitem_45, torch.int8);  layer_norm_7 = None
	        dequantize_affine_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_22, [1, 1, 1280], getitem_44, getitem_45, torch.int8);  quantize_affine_22 = getitem_44 = getitem_45 = None
	        dequantize_affine_45: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_3_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_3_fc1_parametrizations_weight_original1, model_audio_tower_layers_3_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_3_fc1_parametrizations_weight_original0 = model_audio_tower_layers_3_fc1_parametrizations_weight_original1 = model_audio_tower_layers_3_fc1_parametrizations_weight_original2 = None
	        linear_22: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_44, dequantize_affine_45, model_audio_tower_layers_3_fc1_bias);  dequantize_affine_44 = dequantize_affine_45 = model_audio_tower_layers_3_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_5: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_22);  linear_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_11: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_5, 0.0, False);  gelu_5 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_23 = torch.ops.torchao.choose_qparams_affine.default(dropout_11, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_46: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_23[0]
	        getitem_47: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_23[1];  choose_qparams_affine_default_23 = None
	        quantize_affine_23: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_11, [1, 1, 5120], getitem_46, getitem_47, torch.int8);  dropout_11 = None
	        dequantize_affine_46: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_23, [1, 1, 5120], getitem_46, getitem_47, torch.int8);  quantize_affine_23 = getitem_46 = getitem_47 = None
	        dequantize_affine_47: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_3_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_3_fc2_parametrizations_weight_original1, model_audio_tower_layers_3_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_3_fc2_parametrizations_weight_original0 = model_audio_tower_layers_3_fc2_parametrizations_weight_original1 = model_audio_tower_layers_3_fc2_parametrizations_weight_original2 = None
	        linear_23: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_46, dequantize_affine_47, model_audio_tower_layers_3_fc2_bias);  dequantize_affine_46 = dequantize_affine_47 = model_audio_tower_layers_3_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_12: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_23, 0.0, False);  linear_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_56: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_51, dropout_12);  add_51 = dropout_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_56, [1280], model_audio_tower_layers_4_self_attn_layer_norm_weight, model_audio_tower_layers_4_self_attn_layer_norm_bias);  model_audio_tower_layers_4_self_attn_layer_norm_weight = model_audio_tower_layers_4_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_24 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_8, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_48: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_24[0]
	        getitem_49: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_24[1];  choose_qparams_affine_default_24 = None
	        quantize_affine_24: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_8, [1, 1, 1280], getitem_48, getitem_49, torch.int8)
	        dequantize_affine_48: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_24, [1, 1, 1280], getitem_48, getitem_49, torch.int8);  quantize_affine_24 = getitem_48 = getitem_49 = None
	        dequantize_affine_49: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_24: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_48, dequantize_affine_49, model_audio_tower_layers_4_self_attn_q_proj_bias);  dequantize_affine_48 = dequantize_affine_49 = model_audio_tower_layers_4_self_attn_q_proj_bias = None
	        mul_153: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_24, 0.125);  linear_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_12: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_153, [sym_size_int_2, 1500, 20, 64]);  mul_153 = None
	        transpose_16: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_12, 1, 2);  view_12 = None
	        contiguous_16: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_16);  transpose_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_25 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_8, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_50: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_25[0]
	        getitem_51: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_25[1];  choose_qparams_affine_default_25 = None
	        quantize_affine_25: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_8, [1, 1, 1280], getitem_50, getitem_51, torch.int8)
	        dequantize_affine_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_25, [1, 1, 1280], getitem_50, getitem_51, torch.int8);  quantize_affine_25 = getitem_50 = getitem_51 = None
	        dequantize_affine_51: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_25: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_50, dequantize_affine_51);  dequantize_affine_50 = dequantize_affine_51 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_13: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_25, [sym_size_int_2, -1, 20, 64]);  linear_25 = None
	        transpose_17: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_13, 1, 2);  view_13 = None
	        contiguous_17: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_17);  transpose_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_26 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_8, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_52: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_26[0]
	        getitem_53: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_26[1];  choose_qparams_affine_default_26 = None
	        quantize_affine_26: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_8, [1, 1, 1280], getitem_52, getitem_53, torch.int8);  layer_norm_8 = None
	        dequantize_affine_52: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_26, [1, 1, 1280], getitem_52, getitem_53, torch.int8);  quantize_affine_26 = getitem_52 = getitem_53 = None
	        dequantize_affine_53: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_52, dequantize_affine_53, model_audio_tower_layers_4_self_attn_v_proj_bias);  dequantize_affine_52 = dequantize_affine_53 = model_audio_tower_layers_4_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_14: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_26, [sym_size_int_2, -1, 20, 64]);  linear_26 = None
	        transpose_18: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_14, 1, 2);  view_14 = None
	        contiguous_18: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_18);  transpose_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_4: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_16, contiguous_17, contiguous_18, scale = 1.0);  contiguous_16 = contiguous_17 = contiguous_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_19: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_4, 1, 2);  scaled_dot_product_attention_4 = None
	        contiguous_19: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_19);  transpose_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_4: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_19, [sym_size_int_2, 1500, -1]);  contiguous_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_27 = torch.ops.torchao.choose_qparams_affine.default(reshape_4, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_54: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_27[0]
	        getitem_55: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_27[1];  choose_qparams_affine_default_27 = None
	        quantize_affine_27: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_4, [1, 1, 1280], getitem_54, getitem_55, torch.int8);  reshape_4 = None
	        dequantize_affine_54: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_27, [1, 1, 1280], getitem_54, getitem_55, torch.int8);  quantize_affine_27 = getitem_54 = getitem_55 = None
	        dequantize_affine_55: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_27: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_54, dequantize_affine_55, model_audio_tower_layers_4_self_attn_out_proj_bias);  dequantize_affine_54 = dequantize_affine_55 = model_audio_tower_layers_4_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_13: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_27, 0.0, False);  linear_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_65: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_56, dropout_13);  add_56 = dropout_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_9: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_65, [1280], model_audio_tower_layers_4_final_layer_norm_weight, model_audio_tower_layers_4_final_layer_norm_bias);  model_audio_tower_layers_4_final_layer_norm_weight = model_audio_tower_layers_4_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_28 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_9, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_56: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_28[0]
	        getitem_57: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_28[1];  choose_qparams_affine_default_28 = None
	        quantize_affine_28: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_9, [1, 1, 1280], getitem_56, getitem_57, torch.int8);  layer_norm_9 = None
	        dequantize_affine_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_28, [1, 1, 1280], getitem_56, getitem_57, torch.int8);  quantize_affine_28 = getitem_56 = getitem_57 = None
	        dequantize_affine_57: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_4_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_4_fc1_parametrizations_weight_original1, model_audio_tower_layers_4_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_4_fc1_parametrizations_weight_original0 = model_audio_tower_layers_4_fc1_parametrizations_weight_original1 = model_audio_tower_layers_4_fc1_parametrizations_weight_original2 = None
	        linear_28: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_56, dequantize_affine_57, model_audio_tower_layers_4_fc1_bias);  dequantize_affine_56 = dequantize_affine_57 = model_audio_tower_layers_4_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_6: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_28);  linear_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_14: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_6, 0.0, False);  gelu_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_29 = torch.ops.torchao.choose_qparams_affine.default(dropout_14, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_58: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_29[0]
	        getitem_59: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_29[1];  choose_qparams_affine_default_29 = None
	        quantize_affine_29: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_14, [1, 1, 5120], getitem_58, getitem_59, torch.int8);  dropout_14 = None
	        dequantize_affine_58: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_29, [1, 1, 5120], getitem_58, getitem_59, torch.int8);  quantize_affine_29 = getitem_58 = getitem_59 = None
	        dequantize_affine_59: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_4_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_4_fc2_parametrizations_weight_original1, model_audio_tower_layers_4_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_4_fc2_parametrizations_weight_original0 = model_audio_tower_layers_4_fc2_parametrizations_weight_original1 = model_audio_tower_layers_4_fc2_parametrizations_weight_original2 = None
	        linear_29: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_58, dequantize_affine_59, model_audio_tower_layers_4_fc2_bias);  dequantize_affine_58 = dequantize_affine_59 = model_audio_tower_layers_4_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_15: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_29, 0.0, False);  linear_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_70: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_65, dropout_15);  add_65 = dropout_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_10: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_70, [1280], model_audio_tower_layers_5_self_attn_layer_norm_weight, model_audio_tower_layers_5_self_attn_layer_norm_bias);  model_audio_tower_layers_5_self_attn_layer_norm_weight = model_audio_tower_layers_5_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_30 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_10, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_60: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_30[0]
	        getitem_61: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_30[1];  choose_qparams_affine_default_30 = None
	        quantize_affine_30: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_10, [1, 1, 1280], getitem_60, getitem_61, torch.int8)
	        dequantize_affine_60: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_30, [1, 1, 1280], getitem_60, getitem_61, torch.int8);  quantize_affine_30 = getitem_60 = getitem_61 = None
	        dequantize_affine_61: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_60, dequantize_affine_61, model_audio_tower_layers_5_self_attn_q_proj_bias);  dequantize_affine_60 = dequantize_affine_61 = model_audio_tower_layers_5_self_attn_q_proj_bias = None
	        mul_190: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_30, 0.125);  linear_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_15: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_190, [sym_size_int_2, 1500, 20, 64]);  mul_190 = None
	        transpose_20: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_15, 1, 2);  view_15 = None
	        contiguous_20: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_20);  transpose_20 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_31 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_10, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_62: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_31[0]
	        getitem_63: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_31[1];  choose_qparams_affine_default_31 = None
	        quantize_affine_31: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_10, [1, 1, 1280], getitem_62, getitem_63, torch.int8)
	        dequantize_affine_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_31, [1, 1, 1280], getitem_62, getitem_63, torch.int8);  quantize_affine_31 = getitem_62 = getitem_63 = None
	        dequantize_affine_63: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_31: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_62, dequantize_affine_63);  dequantize_affine_62 = dequantize_affine_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_16: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_31, [sym_size_int_2, -1, 20, 64]);  linear_31 = None
	        transpose_21: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_16, 1, 2);  view_16 = None
	        contiguous_21: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_21);  transpose_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_32 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_10, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_64: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_32[0]
	        getitem_65: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_32[1];  choose_qparams_affine_default_32 = None
	        quantize_affine_32: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_10, [1, 1, 1280], getitem_64, getitem_65, torch.int8);  layer_norm_10 = None
	        dequantize_affine_64: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_32, [1, 1, 1280], getitem_64, getitem_65, torch.int8);  quantize_affine_32 = getitem_64 = getitem_65 = None
	        dequantize_affine_65: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_64, dequantize_affine_65, model_audio_tower_layers_5_self_attn_v_proj_bias);  dequantize_affine_64 = dequantize_affine_65 = model_audio_tower_layers_5_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_17: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_32, [sym_size_int_2, -1, 20, 64]);  linear_32 = None
	        transpose_22: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_17, 1, 2);  view_17 = None
	        contiguous_22: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_22);  transpose_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_5: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_20, contiguous_21, contiguous_22, scale = 1.0);  contiguous_20 = contiguous_21 = contiguous_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_23: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_5, 1, 2);  scaled_dot_product_attention_5 = None
	        contiguous_23: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_23);  transpose_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_23, [sym_size_int_2, 1500, -1]);  contiguous_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_33 = torch.ops.torchao.choose_qparams_affine.default(reshape_5, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_66: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_33[0]
	        getitem_67: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_33[1];  choose_qparams_affine_default_33 = None
	        quantize_affine_33: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_5, [1, 1, 1280], getitem_66, getitem_67, torch.int8);  reshape_5 = None
	        dequantize_affine_66: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_33, [1, 1, 1280], getitem_66, getitem_67, torch.int8);  quantize_affine_33 = getitem_66 = getitem_67 = None
	        dequantize_affine_67: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_33: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_66, dequantize_affine_67, model_audio_tower_layers_5_self_attn_out_proj_bias);  dequantize_affine_66 = dequantize_affine_67 = model_audio_tower_layers_5_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_16: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_33, 0.0, False);  linear_33 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_79: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_70, dropout_16);  add_70 = dropout_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_11: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_79, [1280], model_audio_tower_layers_5_final_layer_norm_weight, model_audio_tower_layers_5_final_layer_norm_bias);  model_audio_tower_layers_5_final_layer_norm_weight = model_audio_tower_layers_5_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_34 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_11, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_68: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_34[0]
	        getitem_69: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_34[1];  choose_qparams_affine_default_34 = None
	        quantize_affine_34: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_11, [1, 1, 1280], getitem_68, getitem_69, torch.int8);  layer_norm_11 = None
	        dequantize_affine_68: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_34, [1, 1, 1280], getitem_68, getitem_69, torch.int8);  quantize_affine_34 = getitem_68 = getitem_69 = None
	        dequantize_affine_69: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_5_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_5_fc1_parametrizations_weight_original1, model_audio_tower_layers_5_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_5_fc1_parametrizations_weight_original0 = model_audio_tower_layers_5_fc1_parametrizations_weight_original1 = model_audio_tower_layers_5_fc1_parametrizations_weight_original2 = None
	        linear_34: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_68, dequantize_affine_69, model_audio_tower_layers_5_fc1_bias);  dequantize_affine_68 = dequantize_affine_69 = model_audio_tower_layers_5_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_7: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_34);  linear_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_17: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_7, 0.0, False);  gelu_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_35 = torch.ops.torchao.choose_qparams_affine.default(dropout_17, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_70: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_35[0]
	        getitem_71: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_35[1];  choose_qparams_affine_default_35 = None
	        quantize_affine_35: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_17, [1, 1, 5120], getitem_70, getitem_71, torch.int8);  dropout_17 = None
	        dequantize_affine_70: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_35, [1, 1, 5120], getitem_70, getitem_71, torch.int8);  quantize_affine_35 = getitem_70 = getitem_71 = None
	        dequantize_affine_71: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_5_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_5_fc2_parametrizations_weight_original1, model_audio_tower_layers_5_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_5_fc2_parametrizations_weight_original0 = model_audio_tower_layers_5_fc2_parametrizations_weight_original1 = model_audio_tower_layers_5_fc2_parametrizations_weight_original2 = None
	        linear_35: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_70, dequantize_affine_71, model_audio_tower_layers_5_fc2_bias);  dequantize_affine_70 = dequantize_affine_71 = model_audio_tower_layers_5_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_18: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_35, 0.0, False);  linear_35 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_84: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_79, dropout_18);  add_79 = dropout_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_12: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_84, [1280], model_audio_tower_layers_6_self_attn_layer_norm_weight, model_audio_tower_layers_6_self_attn_layer_norm_bias);  model_audio_tower_layers_6_self_attn_layer_norm_weight = model_audio_tower_layers_6_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_36 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_12, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_72: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_36[0]
	        getitem_73: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_36[1];  choose_qparams_affine_default_36 = None
	        quantize_affine_36: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_12, [1, 1, 1280], getitem_72, getitem_73, torch.int8)
	        dequantize_affine_72: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_36, [1, 1, 1280], getitem_72, getitem_73, torch.int8);  quantize_affine_36 = getitem_72 = getitem_73 = None
	        dequantize_affine_73: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_36: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_72, dequantize_affine_73, model_audio_tower_layers_6_self_attn_q_proj_bias);  dequantize_affine_72 = dequantize_affine_73 = model_audio_tower_layers_6_self_attn_q_proj_bias = None
	        mul_227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_36, 0.125);  linear_36 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_18: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_227, [sym_size_int_2, 1500, 20, 64]);  mul_227 = None
	        transpose_24: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_18, 1, 2);  view_18 = None
	        contiguous_24: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_24);  transpose_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_37 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_12, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_74: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_37[0]
	        getitem_75: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_37[1];  choose_qparams_affine_default_37 = None
	        quantize_affine_37: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_12, [1, 1, 1280], getitem_74, getitem_75, torch.int8)
	        dequantize_affine_74: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_37, [1, 1, 1280], getitem_74, getitem_75, torch.int8);  quantize_affine_37 = getitem_74 = getitem_75 = None
	        dequantize_affine_75: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_37: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_74, dequantize_affine_75);  dequantize_affine_74 = dequantize_affine_75 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_19: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_37, [sym_size_int_2, -1, 20, 64]);  linear_37 = None
	        transpose_25: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_19, 1, 2);  view_19 = None
	        contiguous_25: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_25);  transpose_25 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_38 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_12, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_76: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_38[0]
	        getitem_77: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_38[1];  choose_qparams_affine_default_38 = None
	        quantize_affine_38: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_12, [1, 1, 1280], getitem_76, getitem_77, torch.int8);  layer_norm_12 = None
	        dequantize_affine_76: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_38, [1, 1, 1280], getitem_76, getitem_77, torch.int8);  quantize_affine_38 = getitem_76 = getitem_77 = None
	        dequantize_affine_77: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_76, dequantize_affine_77, model_audio_tower_layers_6_self_attn_v_proj_bias);  dequantize_affine_76 = dequantize_affine_77 = model_audio_tower_layers_6_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_20: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_38, [sym_size_int_2, -1, 20, 64]);  linear_38 = None
	        transpose_26: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_20, 1, 2);  view_20 = None
	        contiguous_26: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_26);  transpose_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_6: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_24, contiguous_25, contiguous_26, scale = 1.0);  contiguous_24 = contiguous_25 = contiguous_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_27: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_6, 1, 2);  scaled_dot_product_attention_6 = None
	        contiguous_27: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_27);  transpose_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_27, [sym_size_int_2, 1500, -1]);  contiguous_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_39 = torch.ops.torchao.choose_qparams_affine.default(reshape_6, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_78: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_39[0]
	        getitem_79: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_39[1];  choose_qparams_affine_default_39 = None
	        quantize_affine_39: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_6, [1, 1, 1280], getitem_78, getitem_79, torch.int8);  reshape_6 = None
	        dequantize_affine_78: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_39, [1, 1, 1280], getitem_78, getitem_79, torch.int8);  quantize_affine_39 = getitem_78 = getitem_79 = None
	        dequantize_affine_79: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_39: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_78, dequantize_affine_79, model_audio_tower_layers_6_self_attn_out_proj_bias);  dequantize_affine_78 = dequantize_affine_79 = model_audio_tower_layers_6_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_19: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_39, 0.0, False);  linear_39 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_93: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_84, dropout_19);  add_84 = dropout_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_13: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_93, [1280], model_audio_tower_layers_6_final_layer_norm_weight, model_audio_tower_layers_6_final_layer_norm_bias);  model_audio_tower_layers_6_final_layer_norm_weight = model_audio_tower_layers_6_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_40 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_13, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_80: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_40[0]
	        getitem_81: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_40[1];  choose_qparams_affine_default_40 = None
	        quantize_affine_40: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_13, [1, 1, 1280], getitem_80, getitem_81, torch.int8);  layer_norm_13 = None
	        dequantize_affine_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_40, [1, 1, 1280], getitem_80, getitem_81, torch.int8);  quantize_affine_40 = getitem_80 = getitem_81 = None
	        dequantize_affine_81: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_6_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_6_fc1_parametrizations_weight_original1, model_audio_tower_layers_6_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_6_fc1_parametrizations_weight_original0 = model_audio_tower_layers_6_fc1_parametrizations_weight_original1 = model_audio_tower_layers_6_fc1_parametrizations_weight_original2 = None
	        linear_40: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_80, dequantize_affine_81, model_audio_tower_layers_6_fc1_bias);  dequantize_affine_80 = dequantize_affine_81 = model_audio_tower_layers_6_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_8: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_40);  linear_40 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_20: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_8, 0.0, False);  gelu_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_41 = torch.ops.torchao.choose_qparams_affine.default(dropout_20, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_82: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_41[0]
	        getitem_83: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_41[1];  choose_qparams_affine_default_41 = None
	        quantize_affine_41: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_20, [1, 1, 5120], getitem_82, getitem_83, torch.int8);  dropout_20 = None
	        dequantize_affine_82: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_41, [1, 1, 5120], getitem_82, getitem_83, torch.int8);  quantize_affine_41 = getitem_82 = getitem_83 = None
	        dequantize_affine_83: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_6_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_6_fc2_parametrizations_weight_original1, model_audio_tower_layers_6_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_6_fc2_parametrizations_weight_original0 = model_audio_tower_layers_6_fc2_parametrizations_weight_original1 = model_audio_tower_layers_6_fc2_parametrizations_weight_original2 = None
	        linear_41: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_82, dequantize_affine_83, model_audio_tower_layers_6_fc2_bias);  dequantize_affine_82 = dequantize_affine_83 = model_audio_tower_layers_6_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_41, 0.0, False);  linear_41 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_98: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_93, dropout_21);  add_93 = dropout_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_98, [1280], model_audio_tower_layers_7_self_attn_layer_norm_weight, model_audio_tower_layers_7_self_attn_layer_norm_bias);  model_audio_tower_layers_7_self_attn_layer_norm_weight = model_audio_tower_layers_7_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_42 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_14, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_84: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_42[0]
	        getitem_85: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_42[1];  choose_qparams_affine_default_42 = None
	        quantize_affine_42: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_14, [1, 1, 1280], getitem_84, getitem_85, torch.int8)
	        dequantize_affine_84: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_42, [1, 1, 1280], getitem_84, getitem_85, torch.int8);  quantize_affine_42 = getitem_84 = getitem_85 = None
	        dequantize_affine_85: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_42: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_84, dequantize_affine_85, model_audio_tower_layers_7_self_attn_q_proj_bias);  dequantize_affine_84 = dequantize_affine_85 = model_audio_tower_layers_7_self_attn_q_proj_bias = None
	        mul_264: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_42, 0.125);  linear_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_21: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_264, [sym_size_int_2, 1500, 20, 64]);  mul_264 = None
	        transpose_28: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_21, 1, 2);  view_21 = None
	        contiguous_28: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_28);  transpose_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_43 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_14, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_86: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_43[0]
	        getitem_87: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_43[1];  choose_qparams_affine_default_43 = None
	        quantize_affine_43: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_14, [1, 1, 1280], getitem_86, getitem_87, torch.int8)
	        dequantize_affine_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_43, [1, 1, 1280], getitem_86, getitem_87, torch.int8);  quantize_affine_43 = getitem_86 = getitem_87 = None
	        dequantize_affine_87: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_43: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_86, dequantize_affine_87);  dequantize_affine_86 = dequantize_affine_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_22: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_43, [sym_size_int_2, -1, 20, 64]);  linear_43 = None
	        transpose_29: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_22, 1, 2);  view_22 = None
	        contiguous_29: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_29);  transpose_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_44 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_14, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_88: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_44[0]
	        getitem_89: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_44[1];  choose_qparams_affine_default_44 = None
	        quantize_affine_44: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_14, [1, 1, 1280], getitem_88, getitem_89, torch.int8);  layer_norm_14 = None
	        dequantize_affine_88: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_44, [1, 1, 1280], getitem_88, getitem_89, torch.int8);  quantize_affine_44 = getitem_88 = getitem_89 = None
	        dequantize_affine_89: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_88, dequantize_affine_89, model_audio_tower_layers_7_self_attn_v_proj_bias);  dequantize_affine_88 = dequantize_affine_89 = model_audio_tower_layers_7_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_23: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_44, [sym_size_int_2, -1, 20, 64]);  linear_44 = None
	        transpose_30: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_23, 1, 2);  view_23 = None
	        contiguous_30: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_30);  transpose_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_7: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_28, contiguous_29, contiguous_30, scale = 1.0);  contiguous_28 = contiguous_29 = contiguous_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_31: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_7, 1, 2);  scaled_dot_product_attention_7 = None
	        contiguous_31: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_31);  transpose_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_7: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_31, [sym_size_int_2, 1500, -1]);  contiguous_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_45 = torch.ops.torchao.choose_qparams_affine.default(reshape_7, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_90: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_45[0]
	        getitem_91: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_45[1];  choose_qparams_affine_default_45 = None
	        quantize_affine_45: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_7, [1, 1, 1280], getitem_90, getitem_91, torch.int8);  reshape_7 = None
	        dequantize_affine_90: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_45, [1, 1, 1280], getitem_90, getitem_91, torch.int8);  quantize_affine_45 = getitem_90 = getitem_91 = None
	        dequantize_affine_91: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_45: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_90, dequantize_affine_91, model_audio_tower_layers_7_self_attn_out_proj_bias);  dequantize_affine_90 = dequantize_affine_91 = model_audio_tower_layers_7_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_22: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_45, 0.0, False);  linear_45 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_107: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_98, dropout_22);  add_98 = dropout_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_15: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_107, [1280], model_audio_tower_layers_7_final_layer_norm_weight, model_audio_tower_layers_7_final_layer_norm_bias);  model_audio_tower_layers_7_final_layer_norm_weight = model_audio_tower_layers_7_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_46 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_15, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_92: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_46[0]
	        getitem_93: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_46[1];  choose_qparams_affine_default_46 = None
	        quantize_affine_46: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_15, [1, 1, 1280], getitem_92, getitem_93, torch.int8);  layer_norm_15 = None
	        dequantize_affine_92: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_46, [1, 1, 1280], getitem_92, getitem_93, torch.int8);  quantize_affine_46 = getitem_92 = getitem_93 = None
	        dequantize_affine_93: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_7_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_7_fc1_parametrizations_weight_original1, model_audio_tower_layers_7_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_7_fc1_parametrizations_weight_original0 = model_audio_tower_layers_7_fc1_parametrizations_weight_original1 = model_audio_tower_layers_7_fc1_parametrizations_weight_original2 = None
	        linear_46: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_92, dequantize_affine_93, model_audio_tower_layers_7_fc1_bias);  dequantize_affine_92 = dequantize_affine_93 = model_audio_tower_layers_7_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_9: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_46);  linear_46 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_23: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_9, 0.0, False);  gelu_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_47 = torch.ops.torchao.choose_qparams_affine.default(dropout_23, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_94: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_47[0]
	        getitem_95: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_47[1];  choose_qparams_affine_default_47 = None
	        quantize_affine_47: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_23, [1, 1, 5120], getitem_94, getitem_95, torch.int8);  dropout_23 = None
	        dequantize_affine_94: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_47, [1, 1, 5120], getitem_94, getitem_95, torch.int8);  quantize_affine_47 = getitem_94 = getitem_95 = None
	        dequantize_affine_95: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_7_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_7_fc2_parametrizations_weight_original1, model_audio_tower_layers_7_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_7_fc2_parametrizations_weight_original0 = model_audio_tower_layers_7_fc2_parametrizations_weight_original1 = model_audio_tower_layers_7_fc2_parametrizations_weight_original2 = None
	        linear_47: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_94, dequantize_affine_95, model_audio_tower_layers_7_fc2_bias);  dequantize_affine_94 = dequantize_affine_95 = model_audio_tower_layers_7_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_24: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_47, 0.0, False);  linear_47 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_112: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_107, dropout_24);  add_107 = dropout_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_16: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_112, [1280], model_audio_tower_layers_8_self_attn_layer_norm_weight, model_audio_tower_layers_8_self_attn_layer_norm_bias);  model_audio_tower_layers_8_self_attn_layer_norm_weight = model_audio_tower_layers_8_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_48 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_16, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_96: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_48[0]
	        getitem_97: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_48[1];  choose_qparams_affine_default_48 = None
	        quantize_affine_48: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_16, [1, 1, 1280], getitem_96, getitem_97, torch.int8)
	        dequantize_affine_96: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_48, [1, 1, 1280], getitem_96, getitem_97, torch.int8);  quantize_affine_48 = getitem_96 = getitem_97 = None
	        dequantize_affine_97: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_48: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_96, dequantize_affine_97, model_audio_tower_layers_8_self_attn_q_proj_bias);  dequantize_affine_96 = dequantize_affine_97 = model_audio_tower_layers_8_self_attn_q_proj_bias = None
	        mul_301: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_48, 0.125);  linear_48 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_24: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_301, [sym_size_int_2, 1500, 20, 64]);  mul_301 = None
	        transpose_32: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_24, 1, 2);  view_24 = None
	        contiguous_32: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_32);  transpose_32 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_49 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_16, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_98: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_49[0]
	        getitem_99: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_49[1];  choose_qparams_affine_default_49 = None
	        quantize_affine_49: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_16, [1, 1, 1280], getitem_98, getitem_99, torch.int8)
	        dequantize_affine_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_49, [1, 1, 1280], getitem_98, getitem_99, torch.int8);  quantize_affine_49 = getitem_98 = getitem_99 = None
	        dequantize_affine_99: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_49: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_98, dequantize_affine_99);  dequantize_affine_98 = dequantize_affine_99 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_25: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_49, [sym_size_int_2, -1, 20, 64]);  linear_49 = None
	        transpose_33: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_25, 1, 2);  view_25 = None
	        contiguous_33: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_33);  transpose_33 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_50 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_16, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_100: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_50[0]
	        getitem_101: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_50[1];  choose_qparams_affine_default_50 = None
	        quantize_affine_50: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_16, [1, 1, 1280], getitem_100, getitem_101, torch.int8);  layer_norm_16 = None
	        dequantize_affine_100: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_50, [1, 1, 1280], getitem_100, getitem_101, torch.int8);  quantize_affine_50 = getitem_100 = getitem_101 = None
	        dequantize_affine_101: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_100, dequantize_affine_101, model_audio_tower_layers_8_self_attn_v_proj_bias);  dequantize_affine_100 = dequantize_affine_101 = model_audio_tower_layers_8_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_26: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_50, [sym_size_int_2, -1, 20, 64]);  linear_50 = None
	        transpose_34: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_26, 1, 2);  view_26 = None
	        contiguous_34: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_34);  transpose_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_8: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_32, contiguous_33, contiguous_34, scale = 1.0);  contiguous_32 = contiguous_33 = contiguous_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_35: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_8, 1, 2);  scaled_dot_product_attention_8 = None
	        contiguous_35: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_35);  transpose_35 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_35, [sym_size_int_2, 1500, -1]);  contiguous_35 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_51 = torch.ops.torchao.choose_qparams_affine.default(reshape_8, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_102: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_51[0]
	        getitem_103: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_51[1];  choose_qparams_affine_default_51 = None
	        quantize_affine_51: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_8, [1, 1, 1280], getitem_102, getitem_103, torch.int8);  reshape_8 = None
	        dequantize_affine_102: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_51, [1, 1, 1280], getitem_102, getitem_103, torch.int8);  quantize_affine_51 = getitem_102 = getitem_103 = None
	        dequantize_affine_103: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_51: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_102, dequantize_affine_103, model_audio_tower_layers_8_self_attn_out_proj_bias);  dequantize_affine_102 = dequantize_affine_103 = model_audio_tower_layers_8_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_25: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_51, 0.0, False);  linear_51 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_121: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_112, dropout_25);  add_112 = dropout_25 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_17: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_121, [1280], model_audio_tower_layers_8_final_layer_norm_weight, model_audio_tower_layers_8_final_layer_norm_bias);  model_audio_tower_layers_8_final_layer_norm_weight = model_audio_tower_layers_8_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_52 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_17, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_104: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_52[0]
	        getitem_105: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_52[1];  choose_qparams_affine_default_52 = None
	        quantize_affine_52: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_17, [1, 1, 1280], getitem_104, getitem_105, torch.int8);  layer_norm_17 = None
	        dequantize_affine_104: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_52, [1, 1, 1280], getitem_104, getitem_105, torch.int8);  quantize_affine_52 = getitem_104 = getitem_105 = None
	        dequantize_affine_105: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_8_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_8_fc1_parametrizations_weight_original1, model_audio_tower_layers_8_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_8_fc1_parametrizations_weight_original0 = model_audio_tower_layers_8_fc1_parametrizations_weight_original1 = model_audio_tower_layers_8_fc1_parametrizations_weight_original2 = None
	        linear_52: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_104, dequantize_affine_105, model_audio_tower_layers_8_fc1_bias);  dequantize_affine_104 = dequantize_affine_105 = model_audio_tower_layers_8_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_10: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_52);  linear_52 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_26: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_10, 0.0, False);  gelu_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_53 = torch.ops.torchao.choose_qparams_affine.default(dropout_26, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_106: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_53[0]
	        getitem_107: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_53[1];  choose_qparams_affine_default_53 = None
	        quantize_affine_53: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_26, [1, 1, 5120], getitem_106, getitem_107, torch.int8);  dropout_26 = None
	        dequantize_affine_106: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_53, [1, 1, 5120], getitem_106, getitem_107, torch.int8);  quantize_affine_53 = getitem_106 = getitem_107 = None
	        dequantize_affine_107: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_8_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_8_fc2_parametrizations_weight_original1, model_audio_tower_layers_8_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_8_fc2_parametrizations_weight_original0 = model_audio_tower_layers_8_fc2_parametrizations_weight_original1 = model_audio_tower_layers_8_fc2_parametrizations_weight_original2 = None
	        linear_53: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_106, dequantize_affine_107, model_audio_tower_layers_8_fc2_bias);  dequantize_affine_106 = dequantize_affine_107 = model_audio_tower_layers_8_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_27: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_53, 0.0, False);  linear_53 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_126: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_121, dropout_27);  add_121 = dropout_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_18: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_126, [1280], model_audio_tower_layers_9_self_attn_layer_norm_weight, model_audio_tower_layers_9_self_attn_layer_norm_bias);  model_audio_tower_layers_9_self_attn_layer_norm_weight = model_audio_tower_layers_9_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_54 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_18, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_108: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_54[0]
	        getitem_109: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_54[1];  choose_qparams_affine_default_54 = None
	        quantize_affine_54: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_18, [1, 1, 1280], getitem_108, getitem_109, torch.int8)
	        dequantize_affine_108: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_54, [1, 1, 1280], getitem_108, getitem_109, torch.int8);  quantize_affine_54 = getitem_108 = getitem_109 = None
	        dequantize_affine_109: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_54: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_108, dequantize_affine_109, model_audio_tower_layers_9_self_attn_q_proj_bias);  dequantize_affine_108 = dequantize_affine_109 = model_audio_tower_layers_9_self_attn_q_proj_bias = None
	        mul_338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_54, 0.125);  linear_54 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_27: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_338, [sym_size_int_2, 1500, 20, 64]);  mul_338 = None
	        transpose_36: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_27, 1, 2);  view_27 = None
	        contiguous_36: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_36);  transpose_36 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_55 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_18, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_110: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_55[0]
	        getitem_111: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_55[1];  choose_qparams_affine_default_55 = None
	        quantize_affine_55: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_18, [1, 1, 1280], getitem_110, getitem_111, torch.int8)
	        dequantize_affine_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_55, [1, 1, 1280], getitem_110, getitem_111, torch.int8);  quantize_affine_55 = getitem_110 = getitem_111 = None
	        dequantize_affine_111: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_55: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_110, dequantize_affine_111);  dequantize_affine_110 = dequantize_affine_111 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_28: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_55, [sym_size_int_2, -1, 20, 64]);  linear_55 = None
	        transpose_37: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_28, 1, 2);  view_28 = None
	        contiguous_37: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_37);  transpose_37 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_56 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_18, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_112: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_56[0]
	        getitem_113: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_56[1];  choose_qparams_affine_default_56 = None
	        quantize_affine_56: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_18, [1, 1, 1280], getitem_112, getitem_113, torch.int8);  layer_norm_18 = None
	        dequantize_affine_112: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_56, [1, 1, 1280], getitem_112, getitem_113, torch.int8);  quantize_affine_56 = getitem_112 = getitem_113 = None
	        dequantize_affine_113: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_112, dequantize_affine_113, model_audio_tower_layers_9_self_attn_v_proj_bias);  dequantize_affine_112 = dequantize_affine_113 = model_audio_tower_layers_9_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_29: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_56, [sym_size_int_2, -1, 20, 64]);  linear_56 = None
	        transpose_38: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_29, 1, 2);  view_29 = None
	        contiguous_38: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_38);  transpose_38 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_9: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_36, contiguous_37, contiguous_38, scale = 1.0);  contiguous_36 = contiguous_37 = contiguous_38 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_39: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_9, 1, 2);  scaled_dot_product_attention_9 = None
	        contiguous_39: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_39);  transpose_39 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_9: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_39, [sym_size_int_2, 1500, -1]);  contiguous_39 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_57 = torch.ops.torchao.choose_qparams_affine.default(reshape_9, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_114: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_57[0]
	        getitem_115: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_57[1];  choose_qparams_affine_default_57 = None
	        quantize_affine_57: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_9, [1, 1, 1280], getitem_114, getitem_115, torch.int8);  reshape_9 = None
	        dequantize_affine_114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_57, [1, 1, 1280], getitem_114, getitem_115, torch.int8);  quantize_affine_57 = getitem_114 = getitem_115 = None
	        dequantize_affine_115: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_57: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_114, dequantize_affine_115, model_audio_tower_layers_9_self_attn_out_proj_bias);  dequantize_affine_114 = dequantize_affine_115 = model_audio_tower_layers_9_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_28: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_57, 0.0, False);  linear_57 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_135: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_126, dropout_28);  add_126 = dropout_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_19: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_135, [1280], model_audio_tower_layers_9_final_layer_norm_weight, model_audio_tower_layers_9_final_layer_norm_bias);  model_audio_tower_layers_9_final_layer_norm_weight = model_audio_tower_layers_9_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_58 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_19, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_116: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_58[0]
	        getitem_117: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_58[1];  choose_qparams_affine_default_58 = None
	        quantize_affine_58: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_19, [1, 1, 1280], getitem_116, getitem_117, torch.int8);  layer_norm_19 = None
	        dequantize_affine_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_58, [1, 1, 1280], getitem_116, getitem_117, torch.int8);  quantize_affine_58 = getitem_116 = getitem_117 = None
	        dequantize_affine_117: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_9_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_9_fc1_parametrizations_weight_original1, model_audio_tower_layers_9_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_9_fc1_parametrizations_weight_original0 = model_audio_tower_layers_9_fc1_parametrizations_weight_original1 = model_audio_tower_layers_9_fc1_parametrizations_weight_original2 = None
	        linear_58: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_116, dequantize_affine_117, model_audio_tower_layers_9_fc1_bias);  dequantize_affine_116 = dequantize_affine_117 = model_audio_tower_layers_9_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_11: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_58);  linear_58 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_29: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_11, 0.0, False);  gelu_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_59 = torch.ops.torchao.choose_qparams_affine.default(dropout_29, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_118: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_59[0]
	        getitem_119: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_59[1];  choose_qparams_affine_default_59 = None
	        quantize_affine_59: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_29, [1, 1, 5120], getitem_118, getitem_119, torch.int8);  dropout_29 = None
	        dequantize_affine_118: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_59, [1, 1, 5120], getitem_118, getitem_119, torch.int8);  quantize_affine_59 = getitem_118 = getitem_119 = None
	        dequantize_affine_119: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_9_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_9_fc2_parametrizations_weight_original1, model_audio_tower_layers_9_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_9_fc2_parametrizations_weight_original0 = model_audio_tower_layers_9_fc2_parametrizations_weight_original1 = model_audio_tower_layers_9_fc2_parametrizations_weight_original2 = None
	        linear_59: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_118, dequantize_affine_119, model_audio_tower_layers_9_fc2_bias);  dequantize_affine_118 = dequantize_affine_119 = model_audio_tower_layers_9_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_59, 0.0, False);  linear_59 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_140: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_135, dropout_30);  add_135 = dropout_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_140, [1280], model_audio_tower_layers_10_self_attn_layer_norm_weight, model_audio_tower_layers_10_self_attn_layer_norm_bias);  model_audio_tower_layers_10_self_attn_layer_norm_weight = model_audio_tower_layers_10_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_60 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_20, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_120: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_60[0]
	        getitem_121: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_60[1];  choose_qparams_affine_default_60 = None
	        quantize_affine_60: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_20, [1, 1, 1280], getitem_120, getitem_121, torch.int8)
	        dequantize_affine_120: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_60, [1, 1, 1280], getitem_120, getitem_121, torch.int8);  quantize_affine_60 = getitem_120 = getitem_121 = None
	        dequantize_affine_121: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_60: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_120, dequantize_affine_121, model_audio_tower_layers_10_self_attn_q_proj_bias);  dequantize_affine_120 = dequantize_affine_121 = model_audio_tower_layers_10_self_attn_q_proj_bias = None
	        mul_375: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_60, 0.125);  linear_60 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_30: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_375, [sym_size_int_2, 1500, 20, 64]);  mul_375 = None
	        transpose_40: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_30, 1, 2);  view_30 = None
	        contiguous_40: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_40);  transpose_40 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_61 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_20, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_122: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_61[0]
	        getitem_123: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_61[1];  choose_qparams_affine_default_61 = None
	        quantize_affine_61: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_20, [1, 1, 1280], getitem_122, getitem_123, torch.int8)
	        dequantize_affine_122: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_61, [1, 1, 1280], getitem_122, getitem_123, torch.int8);  quantize_affine_61 = getitem_122 = getitem_123 = None
	        dequantize_affine_123: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_61: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_122, dequantize_affine_123);  dequantize_affine_122 = dequantize_affine_123 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_31: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_61, [sym_size_int_2, -1, 20, 64]);  linear_61 = None
	        transpose_41: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_31, 1, 2);  view_31 = None
	        contiguous_41: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_41);  transpose_41 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_62 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_20, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_124: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_62[0]
	        getitem_125: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_62[1];  choose_qparams_affine_default_62 = None
	        quantize_affine_62: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_20, [1, 1, 1280], getitem_124, getitem_125, torch.int8);  layer_norm_20 = None
	        dequantize_affine_124: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_62, [1, 1, 1280], getitem_124, getitem_125, torch.int8);  quantize_affine_62 = getitem_124 = getitem_125 = None
	        dequantize_affine_125: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_124, dequantize_affine_125, model_audio_tower_layers_10_self_attn_v_proj_bias);  dequantize_affine_124 = dequantize_affine_125 = model_audio_tower_layers_10_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_32: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_62, [sym_size_int_2, -1, 20, 64]);  linear_62 = None
	        transpose_42: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_32, 1, 2);  view_32 = None
	        contiguous_42: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_42);  transpose_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_10: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_40, contiguous_41, contiguous_42, scale = 1.0);  contiguous_40 = contiguous_41 = contiguous_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_43: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_10, 1, 2);  scaled_dot_product_attention_10 = None
	        contiguous_43: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_43);  transpose_43 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_10: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_43, [sym_size_int_2, 1500, -1]);  contiguous_43 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_63 = torch.ops.torchao.choose_qparams_affine.default(reshape_10, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_126: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_63[0]
	        getitem_127: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_63[1];  choose_qparams_affine_default_63 = None
	        quantize_affine_63: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_10, [1, 1, 1280], getitem_126, getitem_127, torch.int8);  reshape_10 = None
	        dequantize_affine_126: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_63, [1, 1, 1280], getitem_126, getitem_127, torch.int8);  quantize_affine_63 = getitem_126 = getitem_127 = None
	        dequantize_affine_127: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_63: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_126, dequantize_affine_127, model_audio_tower_layers_10_self_attn_out_proj_bias);  dequantize_affine_126 = dequantize_affine_127 = model_audio_tower_layers_10_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_31: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_63, 0.0, False);  linear_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_149: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_140, dropout_31);  add_140 = dropout_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_149, [1280], model_audio_tower_layers_10_final_layer_norm_weight, model_audio_tower_layers_10_final_layer_norm_bias);  model_audio_tower_layers_10_final_layer_norm_weight = model_audio_tower_layers_10_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_64 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_21, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_128: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_64[0]
	        getitem_129: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_64[1];  choose_qparams_affine_default_64 = None
	        quantize_affine_64: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_21, [1, 1, 1280], getitem_128, getitem_129, torch.int8);  layer_norm_21 = None
	        dequantize_affine_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_64, [1, 1, 1280], getitem_128, getitem_129, torch.int8);  quantize_affine_64 = getitem_128 = getitem_129 = None
	        dequantize_affine_129: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_10_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_10_fc1_parametrizations_weight_original1, model_audio_tower_layers_10_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_10_fc1_parametrizations_weight_original0 = model_audio_tower_layers_10_fc1_parametrizations_weight_original1 = model_audio_tower_layers_10_fc1_parametrizations_weight_original2 = None
	        linear_64: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_128, dequantize_affine_129, model_audio_tower_layers_10_fc1_bias);  dequantize_affine_128 = dequantize_affine_129 = model_audio_tower_layers_10_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_12: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_64);  linear_64 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_32: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_12, 0.0, False);  gelu_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_65 = torch.ops.torchao.choose_qparams_affine.default(dropout_32, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_130: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_65[0]
	        getitem_131: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_65[1];  choose_qparams_affine_default_65 = None
	        quantize_affine_65: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_32, [1, 1, 5120], getitem_130, getitem_131, torch.int8);  dropout_32 = None
	        dequantize_affine_130: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_65, [1, 1, 5120], getitem_130, getitem_131, torch.int8);  quantize_affine_65 = getitem_130 = getitem_131 = None
	        dequantize_affine_131: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_10_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_10_fc2_parametrizations_weight_original1, model_audio_tower_layers_10_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_10_fc2_parametrizations_weight_original0 = model_audio_tower_layers_10_fc2_parametrizations_weight_original1 = model_audio_tower_layers_10_fc2_parametrizations_weight_original2 = None
	        linear_65: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_130, dequantize_affine_131, model_audio_tower_layers_10_fc2_bias);  dequantize_affine_130 = dequantize_affine_131 = model_audio_tower_layers_10_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_33: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_65, 0.0, False);  linear_65 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_154: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_149, dropout_33);  add_149 = dropout_33 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_22: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_154, [1280], model_audio_tower_layers_11_self_attn_layer_norm_weight, model_audio_tower_layers_11_self_attn_layer_norm_bias);  model_audio_tower_layers_11_self_attn_layer_norm_weight = model_audio_tower_layers_11_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_66 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_22, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_132: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_66[0]
	        getitem_133: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_66[1];  choose_qparams_affine_default_66 = None
	        quantize_affine_66: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_22, [1, 1, 1280], getitem_132, getitem_133, torch.int8)
	        dequantize_affine_132: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_66, [1, 1, 1280], getitem_132, getitem_133, torch.int8);  quantize_affine_66 = getitem_132 = getitem_133 = None
	        dequantize_affine_133: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_66: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_132, dequantize_affine_133, model_audio_tower_layers_11_self_attn_q_proj_bias);  dequantize_affine_132 = dequantize_affine_133 = model_audio_tower_layers_11_self_attn_q_proj_bias = None
	        mul_412: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_66, 0.125);  linear_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_33: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_412, [sym_size_int_2, 1500, 20, 64]);  mul_412 = None
	        transpose_44: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_33, 1, 2);  view_33 = None
	        contiguous_44: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_44);  transpose_44 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_67 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_22, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_134: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_67[0]
	        getitem_135: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_67[1];  choose_qparams_affine_default_67 = None
	        quantize_affine_67: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_22, [1, 1, 1280], getitem_134, getitem_135, torch.int8)
	        dequantize_affine_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_67, [1, 1, 1280], getitem_134, getitem_135, torch.int8);  quantize_affine_67 = getitem_134 = getitem_135 = None
	        dequantize_affine_135: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_67: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_134, dequantize_affine_135);  dequantize_affine_134 = dequantize_affine_135 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_34: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_67, [sym_size_int_2, -1, 20, 64]);  linear_67 = None
	        transpose_45: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_34, 1, 2);  view_34 = None
	        contiguous_45: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_45);  transpose_45 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_68 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_22, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_136: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_68[0]
	        getitem_137: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_68[1];  choose_qparams_affine_default_68 = None
	        quantize_affine_68: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_22, [1, 1, 1280], getitem_136, getitem_137, torch.int8);  layer_norm_22 = None
	        dequantize_affine_136: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_68, [1, 1, 1280], getitem_136, getitem_137, torch.int8);  quantize_affine_68 = getitem_136 = getitem_137 = None
	        dequantize_affine_137: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_68: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_136, dequantize_affine_137, model_audio_tower_layers_11_self_attn_v_proj_bias);  dequantize_affine_136 = dequantize_affine_137 = model_audio_tower_layers_11_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_35: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_68, [sym_size_int_2, -1, 20, 64]);  linear_68 = None
	        transpose_46: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_35, 1, 2);  view_35 = None
	        contiguous_46: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_46);  transpose_46 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_11: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_44, contiguous_45, contiguous_46, scale = 1.0);  contiguous_44 = contiguous_45 = contiguous_46 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_47: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_11, 1, 2);  scaled_dot_product_attention_11 = None
	        contiguous_47: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_47);  transpose_47 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_11: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_47, [sym_size_int_2, 1500, -1]);  contiguous_47 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_69 = torch.ops.torchao.choose_qparams_affine.default(reshape_11, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_138: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_69[0]
	        getitem_139: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_69[1];  choose_qparams_affine_default_69 = None
	        quantize_affine_69: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_11, [1, 1, 1280], getitem_138, getitem_139, torch.int8);  reshape_11 = None
	        dequantize_affine_138: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_69, [1, 1, 1280], getitem_138, getitem_139, torch.int8);  quantize_affine_69 = getitem_138 = getitem_139 = None
	        dequantize_affine_139: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_69: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_138, dequantize_affine_139, model_audio_tower_layers_11_self_attn_out_proj_bias);  dequantize_affine_138 = dequantize_affine_139 = model_audio_tower_layers_11_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_34: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_69, 0.0, False);  linear_69 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_163: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_154, dropout_34);  add_154 = dropout_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_23: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_163, [1280], model_audio_tower_layers_11_final_layer_norm_weight, model_audio_tower_layers_11_final_layer_norm_bias);  model_audio_tower_layers_11_final_layer_norm_weight = model_audio_tower_layers_11_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_70 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_23, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_140: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_70[0]
	        getitem_141: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_70[1];  choose_qparams_affine_default_70 = None
	        quantize_affine_70: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_23, [1, 1, 1280], getitem_140, getitem_141, torch.int8);  layer_norm_23 = None
	        dequantize_affine_140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_70, [1, 1, 1280], getitem_140, getitem_141, torch.int8);  quantize_affine_70 = getitem_140 = getitem_141 = None
	        dequantize_affine_141: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_11_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_11_fc1_parametrizations_weight_original1, model_audio_tower_layers_11_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_11_fc1_parametrizations_weight_original0 = model_audio_tower_layers_11_fc1_parametrizations_weight_original1 = model_audio_tower_layers_11_fc1_parametrizations_weight_original2 = None
	        linear_70: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_140, dequantize_affine_141, model_audio_tower_layers_11_fc1_bias);  dequantize_affine_140 = dequantize_affine_141 = model_audio_tower_layers_11_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_13: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_70);  linear_70 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_35: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_13, 0.0, False);  gelu_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_71 = torch.ops.torchao.choose_qparams_affine.default(dropout_35, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_142: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_71[0]
	        getitem_143: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_71[1];  choose_qparams_affine_default_71 = None
	        quantize_affine_71: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_35, [1, 1, 5120], getitem_142, getitem_143, torch.int8);  dropout_35 = None
	        dequantize_affine_142: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_71, [1, 1, 5120], getitem_142, getitem_143, torch.int8);  quantize_affine_71 = getitem_142 = getitem_143 = None
	        dequantize_affine_143: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_11_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_11_fc2_parametrizations_weight_original1, model_audio_tower_layers_11_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_11_fc2_parametrizations_weight_original0 = model_audio_tower_layers_11_fc2_parametrizations_weight_original1 = model_audio_tower_layers_11_fc2_parametrizations_weight_original2 = None
	        linear_71: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_142, dequantize_affine_143, model_audio_tower_layers_11_fc2_bias);  dequantize_affine_142 = dequantize_affine_143 = model_audio_tower_layers_11_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_36: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_71, 0.0, False);  linear_71 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_168: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_163, dropout_36);  add_163 = dropout_36 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_24: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_168, [1280], model_audio_tower_layers_12_self_attn_layer_norm_weight, model_audio_tower_layers_12_self_attn_layer_norm_bias);  model_audio_tower_layers_12_self_attn_layer_norm_weight = model_audio_tower_layers_12_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_72 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_24, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_144: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_72[0]
	        getitem_145: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_72[1];  choose_qparams_affine_default_72 = None
	        quantize_affine_72: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_24, [1, 1, 1280], getitem_144, getitem_145, torch.int8)
	        dequantize_affine_144: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_72, [1, 1, 1280], getitem_144, getitem_145, torch.int8);  quantize_affine_72 = getitem_144 = getitem_145 = None
	        dequantize_affine_145: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_72: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_144, dequantize_affine_145, model_audio_tower_layers_12_self_attn_q_proj_bias);  dequantize_affine_144 = dequantize_affine_145 = model_audio_tower_layers_12_self_attn_q_proj_bias = None
	        mul_449: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_72, 0.125);  linear_72 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_36: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_449, [sym_size_int_2, 1500, 20, 64]);  mul_449 = None
	        transpose_48: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_36, 1, 2);  view_36 = None
	        contiguous_48: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_48);  transpose_48 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_73 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_24, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_146: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_73[0]
	        getitem_147: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_73[1];  choose_qparams_affine_default_73 = None
	        quantize_affine_73: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_24, [1, 1, 1280], getitem_146, getitem_147, torch.int8)
	        dequantize_affine_146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_73, [1, 1, 1280], getitem_146, getitem_147, torch.int8);  quantize_affine_73 = getitem_146 = getitem_147 = None
	        dequantize_affine_147: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_73: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_146, dequantize_affine_147);  dequantize_affine_146 = dequantize_affine_147 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_37: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_73, [sym_size_int_2, -1, 20, 64]);  linear_73 = None
	        transpose_49: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_37, 1, 2);  view_37 = None
	        contiguous_49: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_49);  transpose_49 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_74 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_24, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_148: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_74[0]
	        getitem_149: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_74[1];  choose_qparams_affine_default_74 = None
	        quantize_affine_74: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_24, [1, 1, 1280], getitem_148, getitem_149, torch.int8);  layer_norm_24 = None
	        dequantize_affine_148: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_74, [1, 1, 1280], getitem_148, getitem_149, torch.int8);  quantize_affine_74 = getitem_148 = getitem_149 = None
	        dequantize_affine_149: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_74: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_148, dequantize_affine_149, model_audio_tower_layers_12_self_attn_v_proj_bias);  dequantize_affine_148 = dequantize_affine_149 = model_audio_tower_layers_12_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_38: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_74, [sym_size_int_2, -1, 20, 64]);  linear_74 = None
	        transpose_50: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_38, 1, 2);  view_38 = None
	        contiguous_50: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_50);  transpose_50 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_12: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_48, contiguous_49, contiguous_50, scale = 1.0);  contiguous_48 = contiguous_49 = contiguous_50 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_51: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_12, 1, 2);  scaled_dot_product_attention_12 = None
	        contiguous_51: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_51);  transpose_51 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_12: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_51, [sym_size_int_2, 1500, -1]);  contiguous_51 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_75 = torch.ops.torchao.choose_qparams_affine.default(reshape_12, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_150: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_75[0]
	        getitem_151: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_75[1];  choose_qparams_affine_default_75 = None
	        quantize_affine_75: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_12, [1, 1, 1280], getitem_150, getitem_151, torch.int8);  reshape_12 = None
	        dequantize_affine_150: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_75, [1, 1, 1280], getitem_150, getitem_151, torch.int8);  quantize_affine_75 = getitem_150 = getitem_151 = None
	        dequantize_affine_151: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_75: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_150, dequantize_affine_151, model_audio_tower_layers_12_self_attn_out_proj_bias);  dequantize_affine_150 = dequantize_affine_151 = model_audio_tower_layers_12_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_37: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_75, 0.0, False);  linear_75 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_177: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_168, dropout_37);  add_168 = dropout_37 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_25: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_177, [1280], model_audio_tower_layers_12_final_layer_norm_weight, model_audio_tower_layers_12_final_layer_norm_bias);  model_audio_tower_layers_12_final_layer_norm_weight = model_audio_tower_layers_12_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_76 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_25, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_152: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_76[0]
	        getitem_153: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_76[1];  choose_qparams_affine_default_76 = None
	        quantize_affine_76: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_25, [1, 1, 1280], getitem_152, getitem_153, torch.int8);  layer_norm_25 = None
	        dequantize_affine_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_76, [1, 1, 1280], getitem_152, getitem_153, torch.int8);  quantize_affine_76 = getitem_152 = getitem_153 = None
	        dequantize_affine_153: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_12_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_12_fc1_parametrizations_weight_original1, model_audio_tower_layers_12_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_12_fc1_parametrizations_weight_original0 = model_audio_tower_layers_12_fc1_parametrizations_weight_original1 = model_audio_tower_layers_12_fc1_parametrizations_weight_original2 = None
	        linear_76: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_152, dequantize_affine_153, model_audio_tower_layers_12_fc1_bias);  dequantize_affine_152 = dequantize_affine_153 = model_audio_tower_layers_12_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_14: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_76);  linear_76 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_38: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_14, 0.0, False);  gelu_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_77 = torch.ops.torchao.choose_qparams_affine.default(dropout_38, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_154: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_77[0]
	        getitem_155: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_77[1];  choose_qparams_affine_default_77 = None
	        quantize_affine_77: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_38, [1, 1, 5120], getitem_154, getitem_155, torch.int8);  dropout_38 = None
	        dequantize_affine_154: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_77, [1, 1, 5120], getitem_154, getitem_155, torch.int8);  quantize_affine_77 = getitem_154 = getitem_155 = None
	        dequantize_affine_155: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_12_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_12_fc2_parametrizations_weight_original1, model_audio_tower_layers_12_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_12_fc2_parametrizations_weight_original0 = model_audio_tower_layers_12_fc2_parametrizations_weight_original1 = model_audio_tower_layers_12_fc2_parametrizations_weight_original2 = None
	        linear_77: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_154, dequantize_affine_155, model_audio_tower_layers_12_fc2_bias);  dequantize_affine_154 = dequantize_affine_155 = model_audio_tower_layers_12_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_39: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_77, 0.0, False);  linear_77 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_182: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_177, dropout_39);  add_177 = dropout_39 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_182, [1280], model_audio_tower_layers_13_self_attn_layer_norm_weight, model_audio_tower_layers_13_self_attn_layer_norm_bias);  model_audio_tower_layers_13_self_attn_layer_norm_weight = model_audio_tower_layers_13_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_78 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_26, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_156: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_78[0]
	        getitem_157: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_78[1];  choose_qparams_affine_default_78 = None
	        quantize_affine_78: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_26, [1, 1, 1280], getitem_156, getitem_157, torch.int8)
	        dequantize_affine_156: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_78, [1, 1, 1280], getitem_156, getitem_157, torch.int8);  quantize_affine_78 = getitem_156 = getitem_157 = None
	        dequantize_affine_157: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_78: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_156, dequantize_affine_157, model_audio_tower_layers_13_self_attn_q_proj_bias);  dequantize_affine_156 = dequantize_affine_157 = model_audio_tower_layers_13_self_attn_q_proj_bias = None
	        mul_486: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_78, 0.125);  linear_78 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_39: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_486, [sym_size_int_2, 1500, 20, 64]);  mul_486 = None
	        transpose_52: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_39, 1, 2);  view_39 = None
	        contiguous_52: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_52);  transpose_52 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_79 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_26, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_158: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_79[0]
	        getitem_159: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_79[1];  choose_qparams_affine_default_79 = None
	        quantize_affine_79: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_26, [1, 1, 1280], getitem_158, getitem_159, torch.int8)
	        dequantize_affine_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_79, [1, 1, 1280], getitem_158, getitem_159, torch.int8);  quantize_affine_79 = getitem_158 = getitem_159 = None
	        dequantize_affine_159: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_79: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_158, dequantize_affine_159);  dequantize_affine_158 = dequantize_affine_159 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_40: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_79, [sym_size_int_2, -1, 20, 64]);  linear_79 = None
	        transpose_53: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_40, 1, 2);  view_40 = None
	        contiguous_53: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_53);  transpose_53 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_80 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_26, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_160: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_80[0]
	        getitem_161: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_80[1];  choose_qparams_affine_default_80 = None
	        quantize_affine_80: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_26, [1, 1, 1280], getitem_160, getitem_161, torch.int8);  layer_norm_26 = None
	        dequantize_affine_160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_80, [1, 1, 1280], getitem_160, getitem_161, torch.int8);  quantize_affine_80 = getitem_160 = getitem_161 = None
	        dequantize_affine_161: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_160, dequantize_affine_161, model_audio_tower_layers_13_self_attn_v_proj_bias);  dequantize_affine_160 = dequantize_affine_161 = model_audio_tower_layers_13_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_41: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_80, [sym_size_int_2, -1, 20, 64]);  linear_80 = None
	        transpose_54: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_41, 1, 2);  view_41 = None
	        contiguous_54: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_54);  transpose_54 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_13: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_52, contiguous_53, contiguous_54, scale = 1.0);  contiguous_52 = contiguous_53 = contiguous_54 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_55: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_13, 1, 2);  scaled_dot_product_attention_13 = None
	        contiguous_55: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_55);  transpose_55 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_13: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_55, [sym_size_int_2, 1500, -1]);  contiguous_55 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_81 = torch.ops.torchao.choose_qparams_affine.default(reshape_13, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_162: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_81[0]
	        getitem_163: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_81[1];  choose_qparams_affine_default_81 = None
	        quantize_affine_81: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_13, [1, 1, 1280], getitem_162, getitem_163, torch.int8);  reshape_13 = None
	        dequantize_affine_162: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_81, [1, 1, 1280], getitem_162, getitem_163, torch.int8);  quantize_affine_81 = getitem_162 = getitem_163 = None
	        dequantize_affine_163: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_81: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_162, dequantize_affine_163, model_audio_tower_layers_13_self_attn_out_proj_bias);  dequantize_affine_162 = dequantize_affine_163 = model_audio_tower_layers_13_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_40: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_81, 0.0, False);  linear_81 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_191: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_182, dropout_40);  add_182 = dropout_40 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_27: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_191, [1280], model_audio_tower_layers_13_final_layer_norm_weight, model_audio_tower_layers_13_final_layer_norm_bias);  model_audio_tower_layers_13_final_layer_norm_weight = model_audio_tower_layers_13_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_82 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_27, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_164: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_82[0]
	        getitem_165: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_82[1];  choose_qparams_affine_default_82 = None
	        quantize_affine_82: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_27, [1, 1, 1280], getitem_164, getitem_165, torch.int8);  layer_norm_27 = None
	        dequantize_affine_164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_82, [1, 1, 1280], getitem_164, getitem_165, torch.int8);  quantize_affine_82 = getitem_164 = getitem_165 = None
	        dequantize_affine_165: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_13_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_13_fc1_parametrizations_weight_original1, model_audio_tower_layers_13_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_13_fc1_parametrizations_weight_original0 = model_audio_tower_layers_13_fc1_parametrizations_weight_original1 = model_audio_tower_layers_13_fc1_parametrizations_weight_original2 = None
	        linear_82: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_164, dequantize_affine_165, model_audio_tower_layers_13_fc1_bias);  dequantize_affine_164 = dequantize_affine_165 = model_audio_tower_layers_13_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_15: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_82);  linear_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_41: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_15, 0.0, False);  gelu_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_83 = torch.ops.torchao.choose_qparams_affine.default(dropout_41, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_166: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_83[0]
	        getitem_167: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_83[1];  choose_qparams_affine_default_83 = None
	        quantize_affine_83: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_41, [1, 1, 5120], getitem_166, getitem_167, torch.int8);  dropout_41 = None
	        dequantize_affine_166: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_83, [1, 1, 5120], getitem_166, getitem_167, torch.int8);  quantize_affine_83 = getitem_166 = getitem_167 = None
	        dequantize_affine_167: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_13_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_13_fc2_parametrizations_weight_original1, model_audio_tower_layers_13_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_13_fc2_parametrizations_weight_original0 = model_audio_tower_layers_13_fc2_parametrizations_weight_original1 = model_audio_tower_layers_13_fc2_parametrizations_weight_original2 = None
	        linear_83: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_166, dequantize_affine_167, model_audio_tower_layers_13_fc2_bias);  dequantize_affine_166 = dequantize_affine_167 = model_audio_tower_layers_13_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_42: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_83, 0.0, False);  linear_83 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_196: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_191, dropout_42);  add_191 = dropout_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_28: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_196, [1280], model_audio_tower_layers_14_self_attn_layer_norm_weight, model_audio_tower_layers_14_self_attn_layer_norm_bias);  model_audio_tower_layers_14_self_attn_layer_norm_weight = model_audio_tower_layers_14_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_84 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_28, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_168: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_84[0]
	        getitem_169: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_84[1];  choose_qparams_affine_default_84 = None
	        quantize_affine_84: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_28, [1, 1, 1280], getitem_168, getitem_169, torch.int8)
	        dequantize_affine_168: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_84, [1, 1, 1280], getitem_168, getitem_169, torch.int8);  quantize_affine_84 = getitem_168 = getitem_169 = None
	        dequantize_affine_169: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_84: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_168, dequantize_affine_169, model_audio_tower_layers_14_self_attn_q_proj_bias);  dequantize_affine_168 = dequantize_affine_169 = model_audio_tower_layers_14_self_attn_q_proj_bias = None
	        mul_523: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_84, 0.125);  linear_84 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_42: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_523, [sym_size_int_2, 1500, 20, 64]);  mul_523 = None
	        transpose_56: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_42, 1, 2);  view_42 = None
	        contiguous_56: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_56);  transpose_56 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_85 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_28, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_170: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_85[0]
	        getitem_171: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_85[1];  choose_qparams_affine_default_85 = None
	        quantize_affine_85: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_28, [1, 1, 1280], getitem_170, getitem_171, torch.int8)
	        dequantize_affine_170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_85, [1, 1, 1280], getitem_170, getitem_171, torch.int8);  quantize_affine_85 = getitem_170 = getitem_171 = None
	        dequantize_affine_171: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_85: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_170, dequantize_affine_171);  dequantize_affine_170 = dequantize_affine_171 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_43: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_85, [sym_size_int_2, -1, 20, 64]);  linear_85 = None
	        transpose_57: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_43, 1, 2);  view_43 = None
	        contiguous_57: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_57);  transpose_57 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_86 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_28, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_172: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_86[0]
	        getitem_173: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_86[1];  choose_qparams_affine_default_86 = None
	        quantize_affine_86: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_28, [1, 1, 1280], getitem_172, getitem_173, torch.int8);  layer_norm_28 = None
	        dequantize_affine_172: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_86, [1, 1, 1280], getitem_172, getitem_173, torch.int8);  quantize_affine_86 = getitem_172 = getitem_173 = None
	        dequantize_affine_173: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_172, dequantize_affine_173, model_audio_tower_layers_14_self_attn_v_proj_bias);  dequantize_affine_172 = dequantize_affine_173 = model_audio_tower_layers_14_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_44: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_86, [sym_size_int_2, -1, 20, 64]);  linear_86 = None
	        transpose_58: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_44, 1, 2);  view_44 = None
	        contiguous_58: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_58);  transpose_58 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_14: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_56, contiguous_57, contiguous_58, scale = 1.0);  contiguous_56 = contiguous_57 = contiguous_58 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_59: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_14, 1, 2);  scaled_dot_product_attention_14 = None
	        contiguous_59: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_59);  transpose_59 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_59, [sym_size_int_2, 1500, -1]);  contiguous_59 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_87 = torch.ops.torchao.choose_qparams_affine.default(reshape_14, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_174: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_87[0]
	        getitem_175: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_87[1];  choose_qparams_affine_default_87 = None
	        quantize_affine_87: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_14, [1, 1, 1280], getitem_174, getitem_175, torch.int8);  reshape_14 = None
	        dequantize_affine_174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_87, [1, 1, 1280], getitem_174, getitem_175, torch.int8);  quantize_affine_87 = getitem_174 = getitem_175 = None
	        dequantize_affine_175: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_87: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_174, dequantize_affine_175, model_audio_tower_layers_14_self_attn_out_proj_bias);  dequantize_affine_174 = dequantize_affine_175 = model_audio_tower_layers_14_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_43: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_87, 0.0, False);  linear_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_205: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_196, dropout_43);  add_196 = dropout_43 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_29: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_205, [1280], model_audio_tower_layers_14_final_layer_norm_weight, model_audio_tower_layers_14_final_layer_norm_bias);  model_audio_tower_layers_14_final_layer_norm_weight = model_audio_tower_layers_14_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_88 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_29, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_176: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_88[0]
	        getitem_177: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_88[1];  choose_qparams_affine_default_88 = None
	        quantize_affine_88: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_29, [1, 1, 1280], getitem_176, getitem_177, torch.int8);  layer_norm_29 = None
	        dequantize_affine_176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_88, [1, 1, 1280], getitem_176, getitem_177, torch.int8);  quantize_affine_88 = getitem_176 = getitem_177 = None
	        dequantize_affine_177: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_14_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_14_fc1_parametrizations_weight_original1, model_audio_tower_layers_14_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_14_fc1_parametrizations_weight_original0 = model_audio_tower_layers_14_fc1_parametrizations_weight_original1 = model_audio_tower_layers_14_fc1_parametrizations_weight_original2 = None
	        linear_88: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_176, dequantize_affine_177, model_audio_tower_layers_14_fc1_bias);  dequantize_affine_176 = dequantize_affine_177 = model_audio_tower_layers_14_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_16: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_88);  linear_88 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_44: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_16, 0.0, False);  gelu_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_89 = torch.ops.torchao.choose_qparams_affine.default(dropout_44, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_178: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_89[0]
	        getitem_179: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_89[1];  choose_qparams_affine_default_89 = None
	        quantize_affine_89: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_44, [1, 1, 5120], getitem_178, getitem_179, torch.int8);  dropout_44 = None
	        dequantize_affine_178: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_89, [1, 1, 5120], getitem_178, getitem_179, torch.int8);  quantize_affine_89 = getitem_178 = getitem_179 = None
	        dequantize_affine_179: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_14_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_14_fc2_parametrizations_weight_original1, model_audio_tower_layers_14_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_14_fc2_parametrizations_weight_original0 = model_audio_tower_layers_14_fc2_parametrizations_weight_original1 = model_audio_tower_layers_14_fc2_parametrizations_weight_original2 = None
	        linear_89: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_178, dequantize_affine_179, model_audio_tower_layers_14_fc2_bias);  dequantize_affine_178 = dequantize_affine_179 = model_audio_tower_layers_14_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_45: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_89, 0.0, False);  linear_89 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_210: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_205, dropout_45);  add_205 = dropout_45 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_210, [1280], model_audio_tower_layers_15_self_attn_layer_norm_weight, model_audio_tower_layers_15_self_attn_layer_norm_bias);  model_audio_tower_layers_15_self_attn_layer_norm_weight = model_audio_tower_layers_15_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_90 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_30, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_180: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_90[0]
	        getitem_181: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_90[1];  choose_qparams_affine_default_90 = None
	        quantize_affine_90: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_30, [1, 1, 1280], getitem_180, getitem_181, torch.int8)
	        dequantize_affine_180: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_90, [1, 1, 1280], getitem_180, getitem_181, torch.int8);  quantize_affine_90 = getitem_180 = getitem_181 = None
	        dequantize_affine_181: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_90: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_180, dequantize_affine_181, model_audio_tower_layers_15_self_attn_q_proj_bias);  dequantize_affine_180 = dequantize_affine_181 = model_audio_tower_layers_15_self_attn_q_proj_bias = None
	        mul_560: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_90, 0.125);  linear_90 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_45: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_560, [sym_size_int_2, 1500, 20, 64]);  mul_560 = None
	        transpose_60: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_45, 1, 2);  view_45 = None
	        contiguous_60: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_60);  transpose_60 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_91 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_30, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_182: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_91[0]
	        getitem_183: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_91[1];  choose_qparams_affine_default_91 = None
	        quantize_affine_91: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_30, [1, 1, 1280], getitem_182, getitem_183, torch.int8)
	        dequantize_affine_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_91, [1, 1, 1280], getitem_182, getitem_183, torch.int8);  quantize_affine_91 = getitem_182 = getitem_183 = None
	        dequantize_affine_183: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_91: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_182, dequantize_affine_183);  dequantize_affine_182 = dequantize_affine_183 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_46: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_91, [sym_size_int_2, -1, 20, 64]);  linear_91 = None
	        transpose_61: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_46, 1, 2);  view_46 = None
	        contiguous_61: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_61);  transpose_61 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_92 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_30, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_184: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_92[0]
	        getitem_185: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_92[1];  choose_qparams_affine_default_92 = None
	        quantize_affine_92: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_30, [1, 1, 1280], getitem_184, getitem_185, torch.int8);  layer_norm_30 = None
	        dequantize_affine_184: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_92, [1, 1, 1280], getitem_184, getitem_185, torch.int8);  quantize_affine_92 = getitem_184 = getitem_185 = None
	        dequantize_affine_185: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_92: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_184, dequantize_affine_185, model_audio_tower_layers_15_self_attn_v_proj_bias);  dequantize_affine_184 = dequantize_affine_185 = model_audio_tower_layers_15_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_47: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_92, [sym_size_int_2, -1, 20, 64]);  linear_92 = None
	        transpose_62: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_47, 1, 2);  view_47 = None
	        contiguous_62: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_62);  transpose_62 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_15: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_60, contiguous_61, contiguous_62, scale = 1.0);  contiguous_60 = contiguous_61 = contiguous_62 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_63: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_15, 1, 2);  scaled_dot_product_attention_15 = None
	        contiguous_63: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_63);  transpose_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_15: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_63, [sym_size_int_2, 1500, -1]);  contiguous_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_93 = torch.ops.torchao.choose_qparams_affine.default(reshape_15, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_186: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_93[0]
	        getitem_187: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_93[1];  choose_qparams_affine_default_93 = None
	        quantize_affine_93: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_15, [1, 1, 1280], getitem_186, getitem_187, torch.int8);  reshape_15 = None
	        dequantize_affine_186: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_93, [1, 1, 1280], getitem_186, getitem_187, torch.int8);  quantize_affine_93 = getitem_186 = getitem_187 = None
	        dequantize_affine_187: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_93: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_186, dequantize_affine_187, model_audio_tower_layers_15_self_attn_out_proj_bias);  dequantize_affine_186 = dequantize_affine_187 = model_audio_tower_layers_15_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_46: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_93, 0.0, False);  linear_93 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_219: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_210, dropout_46);  add_210 = dropout_46 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_31: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_219, [1280], model_audio_tower_layers_15_final_layer_norm_weight, model_audio_tower_layers_15_final_layer_norm_bias);  model_audio_tower_layers_15_final_layer_norm_weight = model_audio_tower_layers_15_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_94 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_31, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_188: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_94[0]
	        getitem_189: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_94[1];  choose_qparams_affine_default_94 = None
	        quantize_affine_94: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_31, [1, 1, 1280], getitem_188, getitem_189, torch.int8);  layer_norm_31 = None
	        dequantize_affine_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_94, [1, 1, 1280], getitem_188, getitem_189, torch.int8);  quantize_affine_94 = getitem_188 = getitem_189 = None
	        dequantize_affine_189: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_15_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_15_fc1_parametrizations_weight_original1, model_audio_tower_layers_15_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_15_fc1_parametrizations_weight_original0 = model_audio_tower_layers_15_fc1_parametrizations_weight_original1 = model_audio_tower_layers_15_fc1_parametrizations_weight_original2 = None
	        linear_94: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_188, dequantize_affine_189, model_audio_tower_layers_15_fc1_bias);  dequantize_affine_188 = dequantize_affine_189 = model_audio_tower_layers_15_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_17: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_94);  linear_94 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_47: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_17, 0.0, False);  gelu_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_95 = torch.ops.torchao.choose_qparams_affine.default(dropout_47, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_190: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_95[0]
	        getitem_191: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_95[1];  choose_qparams_affine_default_95 = None
	        quantize_affine_95: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_47, [1, 1, 5120], getitem_190, getitem_191, torch.int8);  dropout_47 = None
	        dequantize_affine_190: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_95, [1, 1, 5120], getitem_190, getitem_191, torch.int8);  quantize_affine_95 = getitem_190 = getitem_191 = None
	        dequantize_affine_191: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_15_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_15_fc2_parametrizations_weight_original1, model_audio_tower_layers_15_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_15_fc2_parametrizations_weight_original0 = model_audio_tower_layers_15_fc2_parametrizations_weight_original1 = model_audio_tower_layers_15_fc2_parametrizations_weight_original2 = None
	        linear_95: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_190, dequantize_affine_191, model_audio_tower_layers_15_fc2_bias);  dequantize_affine_190 = dequantize_affine_191 = model_audio_tower_layers_15_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_48: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_95, 0.0, False);  linear_95 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_224: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_219, dropout_48);  add_219 = dropout_48 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_224, [1280], model_audio_tower_layers_16_self_attn_layer_norm_weight, model_audio_tower_layers_16_self_attn_layer_norm_bias);  model_audio_tower_layers_16_self_attn_layer_norm_weight = model_audio_tower_layers_16_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_96 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_32, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_192: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_96[0]
	        getitem_193: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_96[1];  choose_qparams_affine_default_96 = None
	        quantize_affine_96: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_32, [1, 1, 1280], getitem_192, getitem_193, torch.int8)
	        dequantize_affine_192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_96, [1, 1, 1280], getitem_192, getitem_193, torch.int8);  quantize_affine_96 = getitem_192 = getitem_193 = None
	        dequantize_affine_193: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_96: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_192, dequantize_affine_193, model_audio_tower_layers_16_self_attn_q_proj_bias);  dequantize_affine_192 = dequantize_affine_193 = model_audio_tower_layers_16_self_attn_q_proj_bias = None
	        mul_597: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_96, 0.125);  linear_96 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_48: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_597, [sym_size_int_2, 1500, 20, 64]);  mul_597 = None
	        transpose_64: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_48, 1, 2);  view_48 = None
	        contiguous_64: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_64);  transpose_64 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_97 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_32, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_194: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_97[0]
	        getitem_195: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_97[1];  choose_qparams_affine_default_97 = None
	        quantize_affine_97: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_32, [1, 1, 1280], getitem_194, getitem_195, torch.int8)
	        dequantize_affine_194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_97, [1, 1, 1280], getitem_194, getitem_195, torch.int8);  quantize_affine_97 = getitem_194 = getitem_195 = None
	        dequantize_affine_195: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_97: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_194, dequantize_affine_195);  dequantize_affine_194 = dequantize_affine_195 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_49: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_97, [sym_size_int_2, -1, 20, 64]);  linear_97 = None
	        transpose_65: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_49, 1, 2);  view_49 = None
	        contiguous_65: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_65);  transpose_65 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_98 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_32, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_196: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_98[0]
	        getitem_197: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_98[1];  choose_qparams_affine_default_98 = None
	        quantize_affine_98: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_32, [1, 1, 1280], getitem_196, getitem_197, torch.int8);  layer_norm_32 = None
	        dequantize_affine_196: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_98, [1, 1, 1280], getitem_196, getitem_197, torch.int8);  quantize_affine_98 = getitem_196 = getitem_197 = None
	        dequantize_affine_197: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_196, dequantize_affine_197, model_audio_tower_layers_16_self_attn_v_proj_bias);  dequantize_affine_196 = dequantize_affine_197 = model_audio_tower_layers_16_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_50: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_98, [sym_size_int_2, -1, 20, 64]);  linear_98 = None
	        transpose_66: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_50, 1, 2);  view_50 = None
	        contiguous_66: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_66);  transpose_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_16: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_64, contiguous_65, contiguous_66, scale = 1.0);  contiguous_64 = contiguous_65 = contiguous_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_67: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_16, 1, 2);  scaled_dot_product_attention_16 = None
	        contiguous_67: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_67);  transpose_67 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_16: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_67, [sym_size_int_2, 1500, -1]);  contiguous_67 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_99 = torch.ops.torchao.choose_qparams_affine.default(reshape_16, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_198: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_99[0]
	        getitem_199: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_99[1];  choose_qparams_affine_default_99 = None
	        quantize_affine_99: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_16, [1, 1, 1280], getitem_198, getitem_199, torch.int8);  reshape_16 = None
	        dequantize_affine_198: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_99, [1, 1, 1280], getitem_198, getitem_199, torch.int8);  quantize_affine_99 = getitem_198 = getitem_199 = None
	        dequantize_affine_199: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_99: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_198, dequantize_affine_199, model_audio_tower_layers_16_self_attn_out_proj_bias);  dequantize_affine_198 = dequantize_affine_199 = model_audio_tower_layers_16_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_49: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_99, 0.0, False);  linear_99 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_233: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_224, dropout_49);  add_224 = dropout_49 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_33: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_233, [1280], model_audio_tower_layers_16_final_layer_norm_weight, model_audio_tower_layers_16_final_layer_norm_bias);  model_audio_tower_layers_16_final_layer_norm_weight = model_audio_tower_layers_16_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_100 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_33, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_200: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_100[0]
	        getitem_201: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_100[1];  choose_qparams_affine_default_100 = None
	        quantize_affine_100: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_33, [1, 1, 1280], getitem_200, getitem_201, torch.int8);  layer_norm_33 = None
	        dequantize_affine_200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_100, [1, 1, 1280], getitem_200, getitem_201, torch.int8);  quantize_affine_100 = getitem_200 = getitem_201 = None
	        dequantize_affine_201: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_16_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_16_fc1_parametrizations_weight_original1, model_audio_tower_layers_16_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_16_fc1_parametrizations_weight_original0 = model_audio_tower_layers_16_fc1_parametrizations_weight_original1 = model_audio_tower_layers_16_fc1_parametrizations_weight_original2 = None
	        linear_100: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_200, dequantize_affine_201, model_audio_tower_layers_16_fc1_bias);  dequantize_affine_200 = dequantize_affine_201 = model_audio_tower_layers_16_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_18: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_100);  linear_100 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_50: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_18, 0.0, False);  gelu_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_101 = torch.ops.torchao.choose_qparams_affine.default(dropout_50, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_202: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_101[0]
	        getitem_203: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_101[1];  choose_qparams_affine_default_101 = None
	        quantize_affine_101: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_50, [1, 1, 5120], getitem_202, getitem_203, torch.int8);  dropout_50 = None
	        dequantize_affine_202: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_101, [1, 1, 5120], getitem_202, getitem_203, torch.int8);  quantize_affine_101 = getitem_202 = getitem_203 = None
	        dequantize_affine_203: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_16_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_16_fc2_parametrizations_weight_original1, model_audio_tower_layers_16_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_16_fc2_parametrizations_weight_original0 = model_audio_tower_layers_16_fc2_parametrizations_weight_original1 = model_audio_tower_layers_16_fc2_parametrizations_weight_original2 = None
	        linear_101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_202, dequantize_affine_203, model_audio_tower_layers_16_fc2_bias);  dequantize_affine_202 = dequantize_affine_203 = model_audio_tower_layers_16_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_51: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_101, 0.0, False);  linear_101 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_238: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_233, dropout_51);  add_233 = dropout_51 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_34: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_238, [1280], model_audio_tower_layers_17_self_attn_layer_norm_weight, model_audio_tower_layers_17_self_attn_layer_norm_bias);  model_audio_tower_layers_17_self_attn_layer_norm_weight = model_audio_tower_layers_17_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_102 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_34, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_204: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_102[0]
	        getitem_205: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_102[1];  choose_qparams_affine_default_102 = None
	        quantize_affine_102: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_34, [1, 1, 1280], getitem_204, getitem_205, torch.int8)
	        dequantize_affine_204: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_102, [1, 1, 1280], getitem_204, getitem_205, torch.int8);  quantize_affine_102 = getitem_204 = getitem_205 = None
	        dequantize_affine_205: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_102: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_204, dequantize_affine_205, model_audio_tower_layers_17_self_attn_q_proj_bias);  dequantize_affine_204 = dequantize_affine_205 = model_audio_tower_layers_17_self_attn_q_proj_bias = None
	        mul_634: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_102, 0.125);  linear_102 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_51: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_634, [sym_size_int_2, 1500, 20, 64]);  mul_634 = None
	        transpose_68: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_51, 1, 2);  view_51 = None
	        contiguous_68: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_68);  transpose_68 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_103 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_34, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_206: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_103[0]
	        getitem_207: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_103[1];  choose_qparams_affine_default_103 = None
	        quantize_affine_103: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_34, [1, 1, 1280], getitem_206, getitem_207, torch.int8)
	        dequantize_affine_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_103, [1, 1, 1280], getitem_206, getitem_207, torch.int8);  quantize_affine_103 = getitem_206 = getitem_207 = None
	        dequantize_affine_207: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_103: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_206, dequantize_affine_207);  dequantize_affine_206 = dequantize_affine_207 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_52: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_103, [sym_size_int_2, -1, 20, 64]);  linear_103 = None
	        transpose_69: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_52, 1, 2);  view_52 = None
	        contiguous_69: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_69);  transpose_69 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_104 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_34, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_208: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_104[0]
	        getitem_209: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_104[1];  choose_qparams_affine_default_104 = None
	        quantize_affine_104: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_34, [1, 1, 1280], getitem_208, getitem_209, torch.int8);  layer_norm_34 = None
	        dequantize_affine_208: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_104, [1, 1, 1280], getitem_208, getitem_209, torch.int8);  quantize_affine_104 = getitem_208 = getitem_209 = None
	        dequantize_affine_209: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_104: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_208, dequantize_affine_209, model_audio_tower_layers_17_self_attn_v_proj_bias);  dequantize_affine_208 = dequantize_affine_209 = model_audio_tower_layers_17_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_53: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_104, [sym_size_int_2, -1, 20, 64]);  linear_104 = None
	        transpose_70: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_53, 1, 2);  view_53 = None
	        contiguous_70: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_70);  transpose_70 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_17: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_68, contiguous_69, contiguous_70, scale = 1.0);  contiguous_68 = contiguous_69 = contiguous_70 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_71: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_17, 1, 2);  scaled_dot_product_attention_17 = None
	        contiguous_71: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_71);  transpose_71 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_17: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_71, [sym_size_int_2, 1500, -1]);  contiguous_71 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_105 = torch.ops.torchao.choose_qparams_affine.default(reshape_17, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_210: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_105[0]
	        getitem_211: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_105[1];  choose_qparams_affine_default_105 = None
	        quantize_affine_105: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_17, [1, 1, 1280], getitem_210, getitem_211, torch.int8);  reshape_17 = None
	        dequantize_affine_210: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_105, [1, 1, 1280], getitem_210, getitem_211, torch.int8);  quantize_affine_105 = getitem_210 = getitem_211 = None
	        dequantize_affine_211: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_105: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_210, dequantize_affine_211, model_audio_tower_layers_17_self_attn_out_proj_bias);  dequantize_affine_210 = dequantize_affine_211 = model_audio_tower_layers_17_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_52: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_105, 0.0, False);  linear_105 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_247: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_238, dropout_52);  add_238 = dropout_52 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_35: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_247, [1280], model_audio_tower_layers_17_final_layer_norm_weight, model_audio_tower_layers_17_final_layer_norm_bias);  model_audio_tower_layers_17_final_layer_norm_weight = model_audio_tower_layers_17_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_106 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_35, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_212: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_106[0]
	        getitem_213: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_106[1];  choose_qparams_affine_default_106 = None
	        quantize_affine_106: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_35, [1, 1, 1280], getitem_212, getitem_213, torch.int8);  layer_norm_35 = None
	        dequantize_affine_212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_106, [1, 1, 1280], getitem_212, getitem_213, torch.int8);  quantize_affine_106 = getitem_212 = getitem_213 = None
	        dequantize_affine_213: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_17_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_17_fc1_parametrizations_weight_original1, model_audio_tower_layers_17_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_17_fc1_parametrizations_weight_original0 = model_audio_tower_layers_17_fc1_parametrizations_weight_original1 = model_audio_tower_layers_17_fc1_parametrizations_weight_original2 = None
	        linear_106: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_212, dequantize_affine_213, model_audio_tower_layers_17_fc1_bias);  dequantize_affine_212 = dequantize_affine_213 = model_audio_tower_layers_17_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_19: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_106);  linear_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_53: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_19, 0.0, False);  gelu_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_107 = torch.ops.torchao.choose_qparams_affine.default(dropout_53, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_214: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_107[0]
	        getitem_215: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_107[1];  choose_qparams_affine_default_107 = None
	        quantize_affine_107: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_53, [1, 1, 5120], getitem_214, getitem_215, torch.int8);  dropout_53 = None
	        dequantize_affine_214: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_107, [1, 1, 5120], getitem_214, getitem_215, torch.int8);  quantize_affine_107 = getitem_214 = getitem_215 = None
	        dequantize_affine_215: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_17_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_17_fc2_parametrizations_weight_original1, model_audio_tower_layers_17_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_17_fc2_parametrizations_weight_original0 = model_audio_tower_layers_17_fc2_parametrizations_weight_original1 = model_audio_tower_layers_17_fc2_parametrizations_weight_original2 = None
	        linear_107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_214, dequantize_affine_215, model_audio_tower_layers_17_fc2_bias);  dequantize_affine_214 = dequantize_affine_215 = model_audio_tower_layers_17_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_54: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_107, 0.0, False);  linear_107 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_252: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_247, dropout_54);  add_247 = dropout_54 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_36: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_252, [1280], model_audio_tower_layers_18_self_attn_layer_norm_weight, model_audio_tower_layers_18_self_attn_layer_norm_bias);  model_audio_tower_layers_18_self_attn_layer_norm_weight = model_audio_tower_layers_18_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_108 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_36, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_216: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_108[0]
	        getitem_217: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_108[1];  choose_qparams_affine_default_108 = None
	        quantize_affine_108: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_36, [1, 1, 1280], getitem_216, getitem_217, torch.int8)
	        dequantize_affine_216: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_108, [1, 1, 1280], getitem_216, getitem_217, torch.int8);  quantize_affine_108 = getitem_216 = getitem_217 = None
	        dequantize_affine_217: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_108: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_216, dequantize_affine_217, model_audio_tower_layers_18_self_attn_q_proj_bias);  dequantize_affine_216 = dequantize_affine_217 = model_audio_tower_layers_18_self_attn_q_proj_bias = None
	        mul_671: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_108, 0.125);  linear_108 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_54: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_671, [sym_size_int_2, 1500, 20, 64]);  mul_671 = None
	        transpose_72: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_54, 1, 2);  view_54 = None
	        contiguous_72: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_72);  transpose_72 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_109 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_36, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_218: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_109[0]
	        getitem_219: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_109[1];  choose_qparams_affine_default_109 = None
	        quantize_affine_109: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_36, [1, 1, 1280], getitem_218, getitem_219, torch.int8)
	        dequantize_affine_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_109, [1, 1, 1280], getitem_218, getitem_219, torch.int8);  quantize_affine_109 = getitem_218 = getitem_219 = None
	        dequantize_affine_219: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_109: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_218, dequantize_affine_219);  dequantize_affine_218 = dequantize_affine_219 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_55: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_109, [sym_size_int_2, -1, 20, 64]);  linear_109 = None
	        transpose_73: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_55, 1, 2);  view_55 = None
	        contiguous_73: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_73);  transpose_73 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_110 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_36, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_220: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_110[0]
	        getitem_221: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_110[1];  choose_qparams_affine_default_110 = None
	        quantize_affine_110: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_36, [1, 1, 1280], getitem_220, getitem_221, torch.int8);  layer_norm_36 = None
	        dequantize_affine_220: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_110, [1, 1, 1280], getitem_220, getitem_221, torch.int8);  quantize_affine_110 = getitem_220 = getitem_221 = None
	        dequantize_affine_221: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_220, dequantize_affine_221, model_audio_tower_layers_18_self_attn_v_proj_bias);  dequantize_affine_220 = dequantize_affine_221 = model_audio_tower_layers_18_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_56: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_110, [sym_size_int_2, -1, 20, 64]);  linear_110 = None
	        transpose_74: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_56, 1, 2);  view_56 = None
	        contiguous_74: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_74);  transpose_74 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_18: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_72, contiguous_73, contiguous_74, scale = 1.0);  contiguous_72 = contiguous_73 = contiguous_74 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_75: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_18, 1, 2);  scaled_dot_product_attention_18 = None
	        contiguous_75: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_75);  transpose_75 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_18: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_75, [sym_size_int_2, 1500, -1]);  contiguous_75 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_111 = torch.ops.torchao.choose_qparams_affine.default(reshape_18, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_222: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_111[0]
	        getitem_223: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_111[1];  choose_qparams_affine_default_111 = None
	        quantize_affine_111: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_18, [1, 1, 1280], getitem_222, getitem_223, torch.int8);  reshape_18 = None
	        dequantize_affine_222: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_111, [1, 1, 1280], getitem_222, getitem_223, torch.int8);  quantize_affine_111 = getitem_222 = getitem_223 = None
	        dequantize_affine_223: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_111: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_222, dequantize_affine_223, model_audio_tower_layers_18_self_attn_out_proj_bias);  dequantize_affine_222 = dequantize_affine_223 = model_audio_tower_layers_18_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_55: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_111, 0.0, False);  linear_111 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_261: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_252, dropout_55);  add_252 = dropout_55 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_37: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_261, [1280], model_audio_tower_layers_18_final_layer_norm_weight, model_audio_tower_layers_18_final_layer_norm_bias);  model_audio_tower_layers_18_final_layer_norm_weight = model_audio_tower_layers_18_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_112 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_37, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_224: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_112[0]
	        getitem_225: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_112[1];  choose_qparams_affine_default_112 = None
	        quantize_affine_112: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_37, [1, 1, 1280], getitem_224, getitem_225, torch.int8);  layer_norm_37 = None
	        dequantize_affine_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_112, [1, 1, 1280], getitem_224, getitem_225, torch.int8);  quantize_affine_112 = getitem_224 = getitem_225 = None
	        dequantize_affine_225: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_18_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_18_fc1_parametrizations_weight_original1, model_audio_tower_layers_18_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_18_fc1_parametrizations_weight_original0 = model_audio_tower_layers_18_fc1_parametrizations_weight_original1 = model_audio_tower_layers_18_fc1_parametrizations_weight_original2 = None
	        linear_112: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_224, dequantize_affine_225, model_audio_tower_layers_18_fc1_bias);  dequantize_affine_224 = dequantize_affine_225 = model_audio_tower_layers_18_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_20: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_112);  linear_112 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_56: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_20, 0.0, False);  gelu_20 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_113 = torch.ops.torchao.choose_qparams_affine.default(dropout_56, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_226: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_113[0]
	        getitem_227: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_113[1];  choose_qparams_affine_default_113 = None
	        quantize_affine_113: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_56, [1, 1, 5120], getitem_226, getitem_227, torch.int8);  dropout_56 = None
	        dequantize_affine_226: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_113, [1, 1, 5120], getitem_226, getitem_227, torch.int8);  quantize_affine_113 = getitem_226 = getitem_227 = None
	        dequantize_affine_227: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_18_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_18_fc2_parametrizations_weight_original1, model_audio_tower_layers_18_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_18_fc2_parametrizations_weight_original0 = model_audio_tower_layers_18_fc2_parametrizations_weight_original1 = model_audio_tower_layers_18_fc2_parametrizations_weight_original2 = None
	        linear_113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_226, dequantize_affine_227, model_audio_tower_layers_18_fc2_bias);  dequantize_affine_226 = dequantize_affine_227 = model_audio_tower_layers_18_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_57: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_113, 0.0, False);  linear_113 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_266: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_261, dropout_57);  add_261 = dropout_57 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_266, [1280], model_audio_tower_layers_19_self_attn_layer_norm_weight, model_audio_tower_layers_19_self_attn_layer_norm_bias);  model_audio_tower_layers_19_self_attn_layer_norm_weight = model_audio_tower_layers_19_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_114 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_38, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_228: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_114[0]
	        getitem_229: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_114[1];  choose_qparams_affine_default_114 = None
	        quantize_affine_114: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_38, [1, 1, 1280], getitem_228, getitem_229, torch.int8)
	        dequantize_affine_228: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_114, [1, 1, 1280], getitem_228, getitem_229, torch.int8);  quantize_affine_114 = getitem_228 = getitem_229 = None
	        dequantize_affine_229: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_228, dequantize_affine_229, model_audio_tower_layers_19_self_attn_q_proj_bias);  dequantize_affine_228 = dequantize_affine_229 = model_audio_tower_layers_19_self_attn_q_proj_bias = None
	        mul_708: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_114, 0.125);  linear_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_57: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_708, [sym_size_int_2, 1500, 20, 64]);  mul_708 = None
	        transpose_76: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_57, 1, 2);  view_57 = None
	        contiguous_76: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_76);  transpose_76 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_115 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_38, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_230: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_115[0]
	        getitem_231: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_115[1];  choose_qparams_affine_default_115 = None
	        quantize_affine_115: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_38, [1, 1, 1280], getitem_230, getitem_231, torch.int8)
	        dequantize_affine_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_115, [1, 1, 1280], getitem_230, getitem_231, torch.int8);  quantize_affine_115 = getitem_230 = getitem_231 = None
	        dequantize_affine_231: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_115: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_230, dequantize_affine_231);  dequantize_affine_230 = dequantize_affine_231 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_58: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_115, [sym_size_int_2, -1, 20, 64]);  linear_115 = None
	        transpose_77: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_58, 1, 2);  view_58 = None
	        contiguous_77: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_77);  transpose_77 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_116 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_38, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_232: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_116[0]
	        getitem_233: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_116[1];  choose_qparams_affine_default_116 = None
	        quantize_affine_116: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_38, [1, 1, 1280], getitem_232, getitem_233, torch.int8);  layer_norm_38 = None
	        dequantize_affine_232: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_116, [1, 1, 1280], getitem_232, getitem_233, torch.int8);  quantize_affine_116 = getitem_232 = getitem_233 = None
	        dequantize_affine_233: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_232, dequantize_affine_233, model_audio_tower_layers_19_self_attn_v_proj_bias);  dequantize_affine_232 = dequantize_affine_233 = model_audio_tower_layers_19_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_59: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_116, [sym_size_int_2, -1, 20, 64]);  linear_116 = None
	        transpose_78: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_59, 1, 2);  view_59 = None
	        contiguous_78: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_78);  transpose_78 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_19: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_76, contiguous_77, contiguous_78, scale = 1.0);  contiguous_76 = contiguous_77 = contiguous_78 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_79: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_19, 1, 2);  scaled_dot_product_attention_19 = None
	        contiguous_79: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_79);  transpose_79 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_19: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_79, [sym_size_int_2, 1500, -1]);  contiguous_79 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_117 = torch.ops.torchao.choose_qparams_affine.default(reshape_19, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_234: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_117[0]
	        getitem_235: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_117[1];  choose_qparams_affine_default_117 = None
	        quantize_affine_117: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_19, [1, 1, 1280], getitem_234, getitem_235, torch.int8);  reshape_19 = None
	        dequantize_affine_234: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_117, [1, 1, 1280], getitem_234, getitem_235, torch.int8);  quantize_affine_117 = getitem_234 = getitem_235 = None
	        dequantize_affine_235: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_234, dequantize_affine_235, model_audio_tower_layers_19_self_attn_out_proj_bias);  dequantize_affine_234 = dequantize_affine_235 = model_audio_tower_layers_19_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_58: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_117, 0.0, False);  linear_117 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_275: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_266, dropout_58);  add_266 = dropout_58 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_39: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_275, [1280], model_audio_tower_layers_19_final_layer_norm_weight, model_audio_tower_layers_19_final_layer_norm_bias);  model_audio_tower_layers_19_final_layer_norm_weight = model_audio_tower_layers_19_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_118 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_39, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_236: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_118[0]
	        getitem_237: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_118[1];  choose_qparams_affine_default_118 = None
	        quantize_affine_118: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_39, [1, 1, 1280], getitem_236, getitem_237, torch.int8);  layer_norm_39 = None
	        dequantize_affine_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_118, [1, 1, 1280], getitem_236, getitem_237, torch.int8);  quantize_affine_118 = getitem_236 = getitem_237 = None
	        dequantize_affine_237: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_19_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_19_fc1_parametrizations_weight_original1, model_audio_tower_layers_19_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_19_fc1_parametrizations_weight_original0 = model_audio_tower_layers_19_fc1_parametrizations_weight_original1 = model_audio_tower_layers_19_fc1_parametrizations_weight_original2 = None
	        linear_118: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_236, dequantize_affine_237, model_audio_tower_layers_19_fc1_bias);  dequantize_affine_236 = dequantize_affine_237 = model_audio_tower_layers_19_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_21: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_118);  linear_118 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_59: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_21, 0.0, False);  gelu_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_119 = torch.ops.torchao.choose_qparams_affine.default(dropout_59, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_238: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_119[0]
	        getitem_239: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_119[1];  choose_qparams_affine_default_119 = None
	        quantize_affine_119: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_59, [1, 1, 5120], getitem_238, getitem_239, torch.int8);  dropout_59 = None
	        dequantize_affine_238: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_119, [1, 1, 5120], getitem_238, getitem_239, torch.int8);  quantize_affine_119 = getitem_238 = getitem_239 = None
	        dequantize_affine_239: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_19_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_19_fc2_parametrizations_weight_original1, model_audio_tower_layers_19_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_19_fc2_parametrizations_weight_original0 = model_audio_tower_layers_19_fc2_parametrizations_weight_original1 = model_audio_tower_layers_19_fc2_parametrizations_weight_original2 = None
	        linear_119: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_238, dequantize_affine_239, model_audio_tower_layers_19_fc2_bias);  dequantize_affine_238 = dequantize_affine_239 = model_audio_tower_layers_19_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_60: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_119, 0.0, False);  linear_119 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_280: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_275, dropout_60);  add_275 = dropout_60 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_40: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_280, [1280], model_audio_tower_layers_20_self_attn_layer_norm_weight, model_audio_tower_layers_20_self_attn_layer_norm_bias);  model_audio_tower_layers_20_self_attn_layer_norm_weight = model_audio_tower_layers_20_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_120 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_40, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_240: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_120[0]
	        getitem_241: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_120[1];  choose_qparams_affine_default_120 = None
	        quantize_affine_120: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_40, [1, 1, 1280], getitem_240, getitem_241, torch.int8)
	        dequantize_affine_240: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_120, [1, 1, 1280], getitem_240, getitem_241, torch.int8);  quantize_affine_120 = getitem_240 = getitem_241 = None
	        dequantize_affine_241: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_120: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_240, dequantize_affine_241, model_audio_tower_layers_20_self_attn_q_proj_bias);  dequantize_affine_240 = dequantize_affine_241 = model_audio_tower_layers_20_self_attn_q_proj_bias = None
	        mul_745: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_120, 0.125);  linear_120 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_60: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_745, [sym_size_int_2, 1500, 20, 64]);  mul_745 = None
	        transpose_80: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_60, 1, 2);  view_60 = None
	        contiguous_80: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_80);  transpose_80 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_121 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_40, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_242: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_121[0]
	        getitem_243: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_121[1];  choose_qparams_affine_default_121 = None
	        quantize_affine_121: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_40, [1, 1, 1280], getitem_242, getitem_243, torch.int8)
	        dequantize_affine_242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_121, [1, 1, 1280], getitem_242, getitem_243, torch.int8);  quantize_affine_121 = getitem_242 = getitem_243 = None
	        dequantize_affine_243: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_121: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_242, dequantize_affine_243);  dequantize_affine_242 = dequantize_affine_243 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_61: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_121, [sym_size_int_2, -1, 20, 64]);  linear_121 = None
	        transpose_81: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_61, 1, 2);  view_61 = None
	        contiguous_81: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_81);  transpose_81 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_122 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_40, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_244: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_122[0]
	        getitem_245: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_122[1];  choose_qparams_affine_default_122 = None
	        quantize_affine_122: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_40, [1, 1, 1280], getitem_244, getitem_245, torch.int8);  layer_norm_40 = None
	        dequantize_affine_244: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_122, [1, 1, 1280], getitem_244, getitem_245, torch.int8);  quantize_affine_122 = getitem_244 = getitem_245 = None
	        dequantize_affine_245: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_122: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_244, dequantize_affine_245, model_audio_tower_layers_20_self_attn_v_proj_bias);  dequantize_affine_244 = dequantize_affine_245 = model_audio_tower_layers_20_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_62: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_122, [sym_size_int_2, -1, 20, 64]);  linear_122 = None
	        transpose_82: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_62, 1, 2);  view_62 = None
	        contiguous_82: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_82);  transpose_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_20: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_80, contiguous_81, contiguous_82, scale = 1.0);  contiguous_80 = contiguous_81 = contiguous_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_83: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_20, 1, 2);  scaled_dot_product_attention_20 = None
	        contiguous_83: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_83);  transpose_83 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_83, [sym_size_int_2, 1500, -1]);  contiguous_83 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_123 = torch.ops.torchao.choose_qparams_affine.default(reshape_20, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_246: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_123[0]
	        getitem_247: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_123[1];  choose_qparams_affine_default_123 = None
	        quantize_affine_123: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_20, [1, 1, 1280], getitem_246, getitem_247, torch.int8);  reshape_20 = None
	        dequantize_affine_246: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_123, [1, 1, 1280], getitem_246, getitem_247, torch.int8);  quantize_affine_123 = getitem_246 = getitem_247 = None
	        dequantize_affine_247: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_246, dequantize_affine_247, model_audio_tower_layers_20_self_attn_out_proj_bias);  dequantize_affine_246 = dequantize_affine_247 = model_audio_tower_layers_20_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_61: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_123, 0.0, False);  linear_123 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_289: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_280, dropout_61);  add_280 = dropout_61 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_41: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_289, [1280], model_audio_tower_layers_20_final_layer_norm_weight, model_audio_tower_layers_20_final_layer_norm_bias);  model_audio_tower_layers_20_final_layer_norm_weight = model_audio_tower_layers_20_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_124 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_41, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_248: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_124[0]
	        getitem_249: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_124[1];  choose_qparams_affine_default_124 = None
	        quantize_affine_124: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_41, [1, 1, 1280], getitem_248, getitem_249, torch.int8);  layer_norm_41 = None
	        dequantize_affine_248: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_124, [1, 1, 1280], getitem_248, getitem_249, torch.int8);  quantize_affine_124 = getitem_248 = getitem_249 = None
	        dequantize_affine_249: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_20_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_20_fc1_parametrizations_weight_original1, model_audio_tower_layers_20_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_20_fc1_parametrizations_weight_original0 = model_audio_tower_layers_20_fc1_parametrizations_weight_original1 = model_audio_tower_layers_20_fc1_parametrizations_weight_original2 = None
	        linear_124: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_248, dequantize_affine_249, model_audio_tower_layers_20_fc1_bias);  dequantize_affine_248 = dequantize_affine_249 = model_audio_tower_layers_20_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_22: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_124);  linear_124 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_62: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_22, 0.0, False);  gelu_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_125 = torch.ops.torchao.choose_qparams_affine.default(dropout_62, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_250: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_125[0]
	        getitem_251: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_125[1];  choose_qparams_affine_default_125 = None
	        quantize_affine_125: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_62, [1, 1, 5120], getitem_250, getitem_251, torch.int8);  dropout_62 = None
	        dequantize_affine_250: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_125, [1, 1, 5120], getitem_250, getitem_251, torch.int8);  quantize_affine_125 = getitem_250 = getitem_251 = None
	        dequantize_affine_251: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_20_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_20_fc2_parametrizations_weight_original1, model_audio_tower_layers_20_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_20_fc2_parametrizations_weight_original0 = model_audio_tower_layers_20_fc2_parametrizations_weight_original1 = model_audio_tower_layers_20_fc2_parametrizations_weight_original2 = None
	        linear_125: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_250, dequantize_affine_251, model_audio_tower_layers_20_fc2_bias);  dequantize_affine_250 = dequantize_affine_251 = model_audio_tower_layers_20_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_63: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_125, 0.0, False);  linear_125 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_294: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_289, dropout_63);  add_289 = dropout_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_42: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_294, [1280], model_audio_tower_layers_21_self_attn_layer_norm_weight, model_audio_tower_layers_21_self_attn_layer_norm_bias);  model_audio_tower_layers_21_self_attn_layer_norm_weight = model_audio_tower_layers_21_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_126 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_42, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_252: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_126[0]
	        getitem_253: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_126[1];  choose_qparams_affine_default_126 = None
	        quantize_affine_126: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_42, [1, 1, 1280], getitem_252, getitem_253, torch.int8)
	        dequantize_affine_252: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_126, [1, 1, 1280], getitem_252, getitem_253, torch.int8);  quantize_affine_126 = getitem_252 = getitem_253 = None
	        dequantize_affine_253: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_126: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_252, dequantize_affine_253, model_audio_tower_layers_21_self_attn_q_proj_bias);  dequantize_affine_252 = dequantize_affine_253 = model_audio_tower_layers_21_self_attn_q_proj_bias = None
	        mul_782: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_126, 0.125);  linear_126 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_63: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_782, [sym_size_int_2, 1500, 20, 64]);  mul_782 = None
	        transpose_84: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_63, 1, 2);  view_63 = None
	        contiguous_84: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_84);  transpose_84 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_127 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_42, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_254: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_127[0]
	        getitem_255: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_127[1];  choose_qparams_affine_default_127 = None
	        quantize_affine_127: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_42, [1, 1, 1280], getitem_254, getitem_255, torch.int8)
	        dequantize_affine_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_127, [1, 1, 1280], getitem_254, getitem_255, torch.int8);  quantize_affine_127 = getitem_254 = getitem_255 = None
	        dequantize_affine_255: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_254, dequantize_affine_255);  dequantize_affine_254 = dequantize_affine_255 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_64: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_127, [sym_size_int_2, -1, 20, 64]);  linear_127 = None
	        transpose_85: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_64, 1, 2);  view_64 = None
	        contiguous_85: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_85);  transpose_85 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_128 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_42, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_256: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_128[0]
	        getitem_257: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_128[1];  choose_qparams_affine_default_128 = None
	        quantize_affine_128: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_42, [1, 1, 1280], getitem_256, getitem_257, torch.int8);  layer_norm_42 = None
	        dequantize_affine_256: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_128, [1, 1, 1280], getitem_256, getitem_257, torch.int8);  quantize_affine_128 = getitem_256 = getitem_257 = None
	        dequantize_affine_257: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_256, dequantize_affine_257, model_audio_tower_layers_21_self_attn_v_proj_bias);  dequantize_affine_256 = dequantize_affine_257 = model_audio_tower_layers_21_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_65: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_128, [sym_size_int_2, -1, 20, 64]);  linear_128 = None
	        transpose_86: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_65, 1, 2);  view_65 = None
	        contiguous_86: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_86);  transpose_86 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_21: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_84, contiguous_85, contiguous_86, scale = 1.0);  contiguous_84 = contiguous_85 = contiguous_86 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_87: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_21, 1, 2);  scaled_dot_product_attention_21 = None
	        contiguous_87: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_87);  transpose_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_87, [sym_size_int_2, 1500, -1]);  contiguous_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_129 = torch.ops.torchao.choose_qparams_affine.default(reshape_21, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_258: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_129[0]
	        getitem_259: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_129[1];  choose_qparams_affine_default_129 = None
	        quantize_affine_129: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_21, [1, 1, 1280], getitem_258, getitem_259, torch.int8);  reshape_21 = None
	        dequantize_affine_258: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_129, [1, 1, 1280], getitem_258, getitem_259, torch.int8);  quantize_affine_129 = getitem_258 = getitem_259 = None
	        dequantize_affine_259: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_129: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_258, dequantize_affine_259, model_audio_tower_layers_21_self_attn_out_proj_bias);  dequantize_affine_258 = dequantize_affine_259 = model_audio_tower_layers_21_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_64: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_129, 0.0, False);  linear_129 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_303: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_294, dropout_64);  add_294 = dropout_64 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_43: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_303, [1280], model_audio_tower_layers_21_final_layer_norm_weight, model_audio_tower_layers_21_final_layer_norm_bias);  model_audio_tower_layers_21_final_layer_norm_weight = model_audio_tower_layers_21_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_130 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_43, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_260: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_130[0]
	        getitem_261: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_130[1];  choose_qparams_affine_default_130 = None
	        quantize_affine_130: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_43, [1, 1, 1280], getitem_260, getitem_261, torch.int8);  layer_norm_43 = None
	        dequantize_affine_260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_130, [1, 1, 1280], getitem_260, getitem_261, torch.int8);  quantize_affine_130 = getitem_260 = getitem_261 = None
	        dequantize_affine_261: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_21_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_21_fc1_parametrizations_weight_original1, model_audio_tower_layers_21_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_21_fc1_parametrizations_weight_original0 = model_audio_tower_layers_21_fc1_parametrizations_weight_original1 = model_audio_tower_layers_21_fc1_parametrizations_weight_original2 = None
	        linear_130: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_260, dequantize_affine_261, model_audio_tower_layers_21_fc1_bias);  dequantize_affine_260 = dequantize_affine_261 = model_audio_tower_layers_21_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_23: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_130);  linear_130 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_65: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_23, 0.0, False);  gelu_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_131 = torch.ops.torchao.choose_qparams_affine.default(dropout_65, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_262: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_131[0]
	        getitem_263: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_131[1];  choose_qparams_affine_default_131 = None
	        quantize_affine_131: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_65, [1, 1, 5120], getitem_262, getitem_263, torch.int8);  dropout_65 = None
	        dequantize_affine_262: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_131, [1, 1, 5120], getitem_262, getitem_263, torch.int8);  quantize_affine_131 = getitem_262 = getitem_263 = None
	        dequantize_affine_263: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_21_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_21_fc2_parametrizations_weight_original1, model_audio_tower_layers_21_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_21_fc2_parametrizations_weight_original0 = model_audio_tower_layers_21_fc2_parametrizations_weight_original1 = model_audio_tower_layers_21_fc2_parametrizations_weight_original2 = None
	        linear_131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_262, dequantize_affine_263, model_audio_tower_layers_21_fc2_bias);  dequantize_affine_262 = dequantize_affine_263 = model_audio_tower_layers_21_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_66: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_131, 0.0, False);  linear_131 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_308: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_303, dropout_66);  add_303 = dropout_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_308, [1280], model_audio_tower_layers_22_self_attn_layer_norm_weight, model_audio_tower_layers_22_self_attn_layer_norm_bias);  model_audio_tower_layers_22_self_attn_layer_norm_weight = model_audio_tower_layers_22_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_132 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_44, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_264: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_132[0]
	        getitem_265: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_132[1];  choose_qparams_affine_default_132 = None
	        quantize_affine_132: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_44, [1, 1, 1280], getitem_264, getitem_265, torch.int8)
	        dequantize_affine_264: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_132, [1, 1, 1280], getitem_264, getitem_265, torch.int8);  quantize_affine_132 = getitem_264 = getitem_265 = None
	        dequantize_affine_265: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_132: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_264, dequantize_affine_265, model_audio_tower_layers_22_self_attn_q_proj_bias);  dequantize_affine_264 = dequantize_affine_265 = model_audio_tower_layers_22_self_attn_q_proj_bias = None
	        mul_819: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_132, 0.125);  linear_132 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_66: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_819, [sym_size_int_2, 1500, 20, 64]);  mul_819 = None
	        transpose_88: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_66, 1, 2);  view_66 = None
	        contiguous_88: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_88);  transpose_88 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_133 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_44, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_266: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_133[0]
	        getitem_267: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_133[1];  choose_qparams_affine_default_133 = None
	        quantize_affine_133: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_44, [1, 1, 1280], getitem_266, getitem_267, torch.int8)
	        dequantize_affine_266: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_133, [1, 1, 1280], getitem_266, getitem_267, torch.int8);  quantize_affine_133 = getitem_266 = getitem_267 = None
	        dequantize_affine_267: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_133: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_266, dequantize_affine_267);  dequantize_affine_266 = dequantize_affine_267 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_67: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_133, [sym_size_int_2, -1, 20, 64]);  linear_133 = None
	        transpose_89: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_67, 1, 2);  view_67 = None
	        contiguous_89: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_89);  transpose_89 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_134 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_44, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_268: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_134[0]
	        getitem_269: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_134[1];  choose_qparams_affine_default_134 = None
	        quantize_affine_134: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_44, [1, 1, 1280], getitem_268, getitem_269, torch.int8);  layer_norm_44 = None
	        dequantize_affine_268: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_134, [1, 1, 1280], getitem_268, getitem_269, torch.int8);  quantize_affine_134 = getitem_268 = getitem_269 = None
	        dequantize_affine_269: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_268, dequantize_affine_269, model_audio_tower_layers_22_self_attn_v_proj_bias);  dequantize_affine_268 = dequantize_affine_269 = model_audio_tower_layers_22_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_68: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_134, [sym_size_int_2, -1, 20, 64]);  linear_134 = None
	        transpose_90: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_68, 1, 2);  view_68 = None
	        contiguous_90: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_90);  transpose_90 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_22: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_88, contiguous_89, contiguous_90, scale = 1.0);  contiguous_88 = contiguous_89 = contiguous_90 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_91: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_22, 1, 2);  scaled_dot_product_attention_22 = None
	        contiguous_91: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_91);  transpose_91 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_22: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_91, [sym_size_int_2, 1500, -1]);  contiguous_91 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_135 = torch.ops.torchao.choose_qparams_affine.default(reshape_22, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_270: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_135[0]
	        getitem_271: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_135[1];  choose_qparams_affine_default_135 = None
	        quantize_affine_135: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_22, [1, 1, 1280], getitem_270, getitem_271, torch.int8);  reshape_22 = None
	        dequantize_affine_270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_135, [1, 1, 1280], getitem_270, getitem_271, torch.int8);  quantize_affine_135 = getitem_270 = getitem_271 = None
	        dequantize_affine_271: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_135: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_270, dequantize_affine_271, model_audio_tower_layers_22_self_attn_out_proj_bias);  dequantize_affine_270 = dequantize_affine_271 = model_audio_tower_layers_22_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_67: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_135, 0.0, False);  linear_135 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_317: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_308, dropout_67);  add_308 = dropout_67 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_45: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_317, [1280], model_audio_tower_layers_22_final_layer_norm_weight, model_audio_tower_layers_22_final_layer_norm_bias);  model_audio_tower_layers_22_final_layer_norm_weight = model_audio_tower_layers_22_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_136 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_45, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_272: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_136[0]
	        getitem_273: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_136[1];  choose_qparams_affine_default_136 = None
	        quantize_affine_136: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_45, [1, 1, 1280], getitem_272, getitem_273, torch.int8);  layer_norm_45 = None
	        dequantize_affine_272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_136, [1, 1, 1280], getitem_272, getitem_273, torch.int8);  quantize_affine_136 = getitem_272 = getitem_273 = None
	        dequantize_affine_273: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_22_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_22_fc1_parametrizations_weight_original1, model_audio_tower_layers_22_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_22_fc1_parametrizations_weight_original0 = model_audio_tower_layers_22_fc1_parametrizations_weight_original1 = model_audio_tower_layers_22_fc1_parametrizations_weight_original2 = None
	        linear_136: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_272, dequantize_affine_273, model_audio_tower_layers_22_fc1_bias);  dequantize_affine_272 = dequantize_affine_273 = model_audio_tower_layers_22_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_24: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_136);  linear_136 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_68: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_24, 0.0, False);  gelu_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_137 = torch.ops.torchao.choose_qparams_affine.default(dropout_68, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_274: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_137[0]
	        getitem_275: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_137[1];  choose_qparams_affine_default_137 = None
	        quantize_affine_137: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_68, [1, 1, 5120], getitem_274, getitem_275, torch.int8);  dropout_68 = None
	        dequantize_affine_274: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_137, [1, 1, 5120], getitem_274, getitem_275, torch.int8);  quantize_affine_137 = getitem_274 = getitem_275 = None
	        dequantize_affine_275: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_22_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_22_fc2_parametrizations_weight_original1, model_audio_tower_layers_22_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_22_fc2_parametrizations_weight_original0 = model_audio_tower_layers_22_fc2_parametrizations_weight_original1 = model_audio_tower_layers_22_fc2_parametrizations_weight_original2 = None
	        linear_137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_274, dequantize_affine_275, model_audio_tower_layers_22_fc2_bias);  dequantize_affine_274 = dequantize_affine_275 = model_audio_tower_layers_22_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_69: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_137, 0.0, False);  linear_137 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_322: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_317, dropout_69);  add_317 = dropout_69 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_46: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_322, [1280], model_audio_tower_layers_23_self_attn_layer_norm_weight, model_audio_tower_layers_23_self_attn_layer_norm_bias);  model_audio_tower_layers_23_self_attn_layer_norm_weight = model_audio_tower_layers_23_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_138 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_46, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_276: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_138[0]
	        getitem_277: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_138[1];  choose_qparams_affine_default_138 = None
	        quantize_affine_138: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_46, [1, 1, 1280], getitem_276, getitem_277, torch.int8)
	        dequantize_affine_276: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_138, [1, 1, 1280], getitem_276, getitem_277, torch.int8);  quantize_affine_138 = getitem_276 = getitem_277 = None
	        dequantize_affine_277: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_138: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_276, dequantize_affine_277, model_audio_tower_layers_23_self_attn_q_proj_bias);  dequantize_affine_276 = dequantize_affine_277 = model_audio_tower_layers_23_self_attn_q_proj_bias = None
	        mul_856: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_138, 0.125);  linear_138 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_69: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_856, [sym_size_int_2, 1500, 20, 64]);  mul_856 = None
	        transpose_92: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_69, 1, 2);  view_69 = None
	        contiguous_92: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_92);  transpose_92 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_139 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_46, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_278: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_139[0]
	        getitem_279: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_139[1];  choose_qparams_affine_default_139 = None
	        quantize_affine_139: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_46, [1, 1, 1280], getitem_278, getitem_279, torch.int8)
	        dequantize_affine_278: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_139, [1, 1, 1280], getitem_278, getitem_279, torch.int8);  quantize_affine_139 = getitem_278 = getitem_279 = None
	        dequantize_affine_279: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_139: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_278, dequantize_affine_279);  dequantize_affine_278 = dequantize_affine_279 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_70: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_139, [sym_size_int_2, -1, 20, 64]);  linear_139 = None
	        transpose_93: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_70, 1, 2);  view_70 = None
	        contiguous_93: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_93);  transpose_93 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_140 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_46, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_280: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_140[0]
	        getitem_281: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_140[1];  choose_qparams_affine_default_140 = None
	        quantize_affine_140: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_46, [1, 1, 1280], getitem_280, getitem_281, torch.int8);  layer_norm_46 = None
	        dequantize_affine_280: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_140, [1, 1, 1280], getitem_280, getitem_281, torch.int8);  quantize_affine_140 = getitem_280 = getitem_281 = None
	        dequantize_affine_281: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_280, dequantize_affine_281, model_audio_tower_layers_23_self_attn_v_proj_bias);  dequantize_affine_280 = dequantize_affine_281 = model_audio_tower_layers_23_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_71: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_140, [sym_size_int_2, -1, 20, 64]);  linear_140 = None
	        transpose_94: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_71, 1, 2);  view_71 = None
	        contiguous_94: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_94);  transpose_94 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_23: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_92, contiguous_93, contiguous_94, scale = 1.0);  contiguous_92 = contiguous_93 = contiguous_94 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_95: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_23, 1, 2);  scaled_dot_product_attention_23 = None
	        contiguous_95: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_95);  transpose_95 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_23: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_95, [sym_size_int_2, 1500, -1]);  contiguous_95 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_141 = torch.ops.torchao.choose_qparams_affine.default(reshape_23, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_282: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_141[0]
	        getitem_283: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_141[1];  choose_qparams_affine_default_141 = None
	        quantize_affine_141: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_23, [1, 1, 1280], getitem_282, getitem_283, torch.int8);  reshape_23 = None
	        dequantize_affine_282: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_141, [1, 1, 1280], getitem_282, getitem_283, torch.int8);  quantize_affine_141 = getitem_282 = getitem_283 = None
	        dequantize_affine_283: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_141: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_282, dequantize_affine_283, model_audio_tower_layers_23_self_attn_out_proj_bias);  dequantize_affine_282 = dequantize_affine_283 = model_audio_tower_layers_23_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_70: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_141, 0.0, False);  linear_141 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_331: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_322, dropout_70);  add_322 = dropout_70 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_47: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_331, [1280], model_audio_tower_layers_23_final_layer_norm_weight, model_audio_tower_layers_23_final_layer_norm_bias);  model_audio_tower_layers_23_final_layer_norm_weight = model_audio_tower_layers_23_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_142 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_47, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_284: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_142[0]
	        getitem_285: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_142[1];  choose_qparams_affine_default_142 = None
	        quantize_affine_142: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_47, [1, 1, 1280], getitem_284, getitem_285, torch.int8);  layer_norm_47 = None
	        dequantize_affine_284: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_142, [1, 1, 1280], getitem_284, getitem_285, torch.int8);  quantize_affine_142 = getitem_284 = getitem_285 = None
	        dequantize_affine_285: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_23_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_23_fc1_parametrizations_weight_original1, model_audio_tower_layers_23_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_23_fc1_parametrizations_weight_original0 = model_audio_tower_layers_23_fc1_parametrizations_weight_original1 = model_audio_tower_layers_23_fc1_parametrizations_weight_original2 = None
	        linear_142: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_284, dequantize_affine_285, model_audio_tower_layers_23_fc1_bias);  dequantize_affine_284 = dequantize_affine_285 = model_audio_tower_layers_23_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_25: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_142);  linear_142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_71: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_25, 0.0, False);  gelu_25 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_143 = torch.ops.torchao.choose_qparams_affine.default(dropout_71, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_286: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_143[0]
	        getitem_287: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_143[1];  choose_qparams_affine_default_143 = None
	        quantize_affine_143: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_71, [1, 1, 5120], getitem_286, getitem_287, torch.int8);  dropout_71 = None
	        dequantize_affine_286: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_143, [1, 1, 5120], getitem_286, getitem_287, torch.int8);  quantize_affine_143 = getitem_286 = getitem_287 = None
	        dequantize_affine_287: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_23_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_23_fc2_parametrizations_weight_original1, model_audio_tower_layers_23_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_23_fc2_parametrizations_weight_original0 = model_audio_tower_layers_23_fc2_parametrizations_weight_original1 = model_audio_tower_layers_23_fc2_parametrizations_weight_original2 = None
	        linear_143: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_286, dequantize_affine_287, model_audio_tower_layers_23_fc2_bias);  dequantize_affine_286 = dequantize_affine_287 = model_audio_tower_layers_23_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_72: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_143, 0.0, False);  linear_143 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_336: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_331, dropout_72);  add_331 = dropout_72 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_48: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_336, [1280], model_audio_tower_layers_24_self_attn_layer_norm_weight, model_audio_tower_layers_24_self_attn_layer_norm_bias);  model_audio_tower_layers_24_self_attn_layer_norm_weight = model_audio_tower_layers_24_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_144 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_48, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_288: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_144[0]
	        getitem_289: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_144[1];  choose_qparams_affine_default_144 = None
	        quantize_affine_144: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_48, [1, 1, 1280], getitem_288, getitem_289, torch.int8)
	        dequantize_affine_288: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_144, [1, 1, 1280], getitem_288, getitem_289, torch.int8);  quantize_affine_144 = getitem_288 = getitem_289 = None
	        dequantize_affine_289: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_144: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_288, dequantize_affine_289, model_audio_tower_layers_24_self_attn_q_proj_bias);  dequantize_affine_288 = dequantize_affine_289 = model_audio_tower_layers_24_self_attn_q_proj_bias = None
	        mul_893: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_144, 0.125);  linear_144 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_72: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_893, [sym_size_int_2, 1500, 20, 64]);  mul_893 = None
	        transpose_96: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_72, 1, 2);  view_72 = None
	        contiguous_96: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_96);  transpose_96 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_145 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_48, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_290: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_145[0]
	        getitem_291: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_145[1];  choose_qparams_affine_default_145 = None
	        quantize_affine_145: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_48, [1, 1, 1280], getitem_290, getitem_291, torch.int8)
	        dequantize_affine_290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_145, [1, 1, 1280], getitem_290, getitem_291, torch.int8);  quantize_affine_145 = getitem_290 = getitem_291 = None
	        dequantize_affine_291: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_145: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_290, dequantize_affine_291);  dequantize_affine_290 = dequantize_affine_291 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_73: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_145, [sym_size_int_2, -1, 20, 64]);  linear_145 = None
	        transpose_97: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_73, 1, 2);  view_73 = None
	        contiguous_97: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_97);  transpose_97 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_146 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_48, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_292: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_146[0]
	        getitem_293: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_146[1];  choose_qparams_affine_default_146 = None
	        quantize_affine_146: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_48, [1, 1, 1280], getitem_292, getitem_293, torch.int8);  layer_norm_48 = None
	        dequantize_affine_292: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_146, [1, 1, 1280], getitem_292, getitem_293, torch.int8);  quantize_affine_146 = getitem_292 = getitem_293 = None
	        dequantize_affine_293: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_292, dequantize_affine_293, model_audio_tower_layers_24_self_attn_v_proj_bias);  dequantize_affine_292 = dequantize_affine_293 = model_audio_tower_layers_24_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_74: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_146, [sym_size_int_2, -1, 20, 64]);  linear_146 = None
	        transpose_98: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_74, 1, 2);  view_74 = None
	        contiguous_98: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_98);  transpose_98 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_24: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_96, contiguous_97, contiguous_98, scale = 1.0);  contiguous_96 = contiguous_97 = contiguous_98 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_99: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_24, 1, 2);  scaled_dot_product_attention_24 = None
	        contiguous_99: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_99);  transpose_99 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_24: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_99, [sym_size_int_2, 1500, -1]);  contiguous_99 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_147 = torch.ops.torchao.choose_qparams_affine.default(reshape_24, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_294: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_147[0]
	        getitem_295: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_147[1];  choose_qparams_affine_default_147 = None
	        quantize_affine_147: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_24, [1, 1, 1280], getitem_294, getitem_295, torch.int8);  reshape_24 = None
	        dequantize_affine_294: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_147, [1, 1, 1280], getitem_294, getitem_295, torch.int8);  quantize_affine_147 = getitem_294 = getitem_295 = None
	        dequantize_affine_295: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_147: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_294, dequantize_affine_295, model_audio_tower_layers_24_self_attn_out_proj_bias);  dequantize_affine_294 = dequantize_affine_295 = model_audio_tower_layers_24_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_73: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_147, 0.0, False);  linear_147 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_345: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_336, dropout_73);  add_336 = dropout_73 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_49: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_345, [1280], model_audio_tower_layers_24_final_layer_norm_weight, model_audio_tower_layers_24_final_layer_norm_bias);  model_audio_tower_layers_24_final_layer_norm_weight = model_audio_tower_layers_24_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_148 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_49, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_296: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_148[0]
	        getitem_297: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_148[1];  choose_qparams_affine_default_148 = None
	        quantize_affine_148: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_49, [1, 1, 1280], getitem_296, getitem_297, torch.int8);  layer_norm_49 = None
	        dequantize_affine_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_148, [1, 1, 1280], getitem_296, getitem_297, torch.int8);  quantize_affine_148 = getitem_296 = getitem_297 = None
	        dequantize_affine_297: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_24_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_24_fc1_parametrizations_weight_original1, model_audio_tower_layers_24_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_24_fc1_parametrizations_weight_original0 = model_audio_tower_layers_24_fc1_parametrizations_weight_original1 = model_audio_tower_layers_24_fc1_parametrizations_weight_original2 = None
	        linear_148: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_296, dequantize_affine_297, model_audio_tower_layers_24_fc1_bias);  dequantize_affine_296 = dequantize_affine_297 = model_audio_tower_layers_24_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_26: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_148);  linear_148 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_74: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_26, 0.0, False);  gelu_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_149 = torch.ops.torchao.choose_qparams_affine.default(dropout_74, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_298: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_149[0]
	        getitem_299: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_149[1];  choose_qparams_affine_default_149 = None
	        quantize_affine_149: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_74, [1, 1, 5120], getitem_298, getitem_299, torch.int8);  dropout_74 = None
	        dequantize_affine_298: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_149, [1, 1, 5120], getitem_298, getitem_299, torch.int8);  quantize_affine_149 = getitem_298 = getitem_299 = None
	        dequantize_affine_299: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_24_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_24_fc2_parametrizations_weight_original1, model_audio_tower_layers_24_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_24_fc2_parametrizations_weight_original0 = model_audio_tower_layers_24_fc2_parametrizations_weight_original1 = model_audio_tower_layers_24_fc2_parametrizations_weight_original2 = None
	        linear_149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_298, dequantize_affine_299, model_audio_tower_layers_24_fc2_bias);  dequantize_affine_298 = dequantize_affine_299 = model_audio_tower_layers_24_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_75: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_149, 0.0, False);  linear_149 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_350: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_345, dropout_75);  add_345 = dropout_75 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_350, [1280], model_audio_tower_layers_25_self_attn_layer_norm_weight, model_audio_tower_layers_25_self_attn_layer_norm_bias);  model_audio_tower_layers_25_self_attn_layer_norm_weight = model_audio_tower_layers_25_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_150 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_50, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_300: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_150[0]
	        getitem_301: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_150[1];  choose_qparams_affine_default_150 = None
	        quantize_affine_150: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_50, [1, 1, 1280], getitem_300, getitem_301, torch.int8)
	        dequantize_affine_300: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_150, [1, 1, 1280], getitem_300, getitem_301, torch.int8);  quantize_affine_150 = getitem_300 = getitem_301 = None
	        dequantize_affine_301: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_150: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_300, dequantize_affine_301, model_audio_tower_layers_25_self_attn_q_proj_bias);  dequantize_affine_300 = dequantize_affine_301 = model_audio_tower_layers_25_self_attn_q_proj_bias = None
	        mul_930: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_150, 0.125);  linear_150 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_75: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_930, [sym_size_int_2, 1500, 20, 64]);  mul_930 = None
	        transpose_100: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_75, 1, 2);  view_75 = None
	        contiguous_100: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_100);  transpose_100 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_151 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_50, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_302: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_151[0]
	        getitem_303: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_151[1];  choose_qparams_affine_default_151 = None
	        quantize_affine_151: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_50, [1, 1, 1280], getitem_302, getitem_303, torch.int8)
	        dequantize_affine_302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_151, [1, 1, 1280], getitem_302, getitem_303, torch.int8);  quantize_affine_151 = getitem_302 = getitem_303 = None
	        dequantize_affine_303: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_151: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_302, dequantize_affine_303);  dequantize_affine_302 = dequantize_affine_303 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_76: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_151, [sym_size_int_2, -1, 20, 64]);  linear_151 = None
	        transpose_101: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_76, 1, 2);  view_76 = None
	        contiguous_101: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_101);  transpose_101 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_152 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_50, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_304: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_152[0]
	        getitem_305: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_152[1];  choose_qparams_affine_default_152 = None
	        quantize_affine_152: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_50, [1, 1, 1280], getitem_304, getitem_305, torch.int8);  layer_norm_50 = None
	        dequantize_affine_304: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_152, [1, 1, 1280], getitem_304, getitem_305, torch.int8);  quantize_affine_152 = getitem_304 = getitem_305 = None
	        dequantize_affine_305: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_304, dequantize_affine_305, model_audio_tower_layers_25_self_attn_v_proj_bias);  dequantize_affine_304 = dequantize_affine_305 = model_audio_tower_layers_25_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_77: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_152, [sym_size_int_2, -1, 20, 64]);  linear_152 = None
	        transpose_102: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_77, 1, 2);  view_77 = None
	        contiguous_102: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_102);  transpose_102 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_25: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_100, contiguous_101, contiguous_102, scale = 1.0);  contiguous_100 = contiguous_101 = contiguous_102 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_103: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_25, 1, 2);  scaled_dot_product_attention_25 = None
	        contiguous_103: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_103);  transpose_103 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_25: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_103, [sym_size_int_2, 1500, -1]);  contiguous_103 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_153 = torch.ops.torchao.choose_qparams_affine.default(reshape_25, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_306: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_153[0]
	        getitem_307: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_153[1];  choose_qparams_affine_default_153 = None
	        quantize_affine_153: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_25, [1, 1, 1280], getitem_306, getitem_307, torch.int8);  reshape_25 = None
	        dequantize_affine_306: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_153, [1, 1, 1280], getitem_306, getitem_307, torch.int8);  quantize_affine_153 = getitem_306 = getitem_307 = None
	        dequantize_affine_307: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_153: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_306, dequantize_affine_307, model_audio_tower_layers_25_self_attn_out_proj_bias);  dequantize_affine_306 = dequantize_affine_307 = model_audio_tower_layers_25_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_76: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_153, 0.0, False);  linear_153 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_359: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_350, dropout_76);  add_350 = dropout_76 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_51: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_359, [1280], model_audio_tower_layers_25_final_layer_norm_weight, model_audio_tower_layers_25_final_layer_norm_bias);  model_audio_tower_layers_25_final_layer_norm_weight = model_audio_tower_layers_25_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_154 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_51, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_308: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_154[0]
	        getitem_309: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_154[1];  choose_qparams_affine_default_154 = None
	        quantize_affine_154: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_51, [1, 1, 1280], getitem_308, getitem_309, torch.int8);  layer_norm_51 = None
	        dequantize_affine_308: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_154, [1, 1, 1280], getitem_308, getitem_309, torch.int8);  quantize_affine_154 = getitem_308 = getitem_309 = None
	        dequantize_affine_309: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_25_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_25_fc1_parametrizations_weight_original1, model_audio_tower_layers_25_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_25_fc1_parametrizations_weight_original0 = model_audio_tower_layers_25_fc1_parametrizations_weight_original1 = model_audio_tower_layers_25_fc1_parametrizations_weight_original2 = None
	        linear_154: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_308, dequantize_affine_309, model_audio_tower_layers_25_fc1_bias);  dequantize_affine_308 = dequantize_affine_309 = model_audio_tower_layers_25_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_27: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_154);  linear_154 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_77: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_27, 0.0, False);  gelu_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_155 = torch.ops.torchao.choose_qparams_affine.default(dropout_77, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_310: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_155[0]
	        getitem_311: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_155[1];  choose_qparams_affine_default_155 = None
	        quantize_affine_155: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_77, [1, 1, 5120], getitem_310, getitem_311, torch.int8);  dropout_77 = None
	        dequantize_affine_310: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_155, [1, 1, 5120], getitem_310, getitem_311, torch.int8);  quantize_affine_155 = getitem_310 = getitem_311 = None
	        dequantize_affine_311: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_25_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_25_fc2_parametrizations_weight_original1, model_audio_tower_layers_25_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_25_fc2_parametrizations_weight_original0 = model_audio_tower_layers_25_fc2_parametrizations_weight_original1 = model_audio_tower_layers_25_fc2_parametrizations_weight_original2 = None
	        linear_155: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_310, dequantize_affine_311, model_audio_tower_layers_25_fc2_bias);  dequantize_affine_310 = dequantize_affine_311 = model_audio_tower_layers_25_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_78: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_155, 0.0, False);  linear_155 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_364: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_359, dropout_78);  add_359 = dropout_78 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_52: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_364, [1280], model_audio_tower_layers_26_self_attn_layer_norm_weight, model_audio_tower_layers_26_self_attn_layer_norm_bias);  model_audio_tower_layers_26_self_attn_layer_norm_weight = model_audio_tower_layers_26_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_156 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_52, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_312: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_156[0]
	        getitem_313: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_156[1];  choose_qparams_affine_default_156 = None
	        quantize_affine_156: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_52, [1, 1, 1280], getitem_312, getitem_313, torch.int8)
	        dequantize_affine_312: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_156, [1, 1, 1280], getitem_312, getitem_313, torch.int8);  quantize_affine_156 = getitem_312 = getitem_313 = None
	        dequantize_affine_313: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_156: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_312, dequantize_affine_313, model_audio_tower_layers_26_self_attn_q_proj_bias);  dequantize_affine_312 = dequantize_affine_313 = model_audio_tower_layers_26_self_attn_q_proj_bias = None
	        mul_967: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_156, 0.125);  linear_156 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_78: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_967, [sym_size_int_2, 1500, 20, 64]);  mul_967 = None
	        transpose_104: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_78, 1, 2);  view_78 = None
	        contiguous_104: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_104);  transpose_104 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_157 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_52, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_314: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_157[0]
	        getitem_315: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_157[1];  choose_qparams_affine_default_157 = None
	        quantize_affine_157: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_52, [1, 1, 1280], getitem_314, getitem_315, torch.int8)
	        dequantize_affine_314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_157, [1, 1, 1280], getitem_314, getitem_315, torch.int8);  quantize_affine_157 = getitem_314 = getitem_315 = None
	        dequantize_affine_315: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_157: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_314, dequantize_affine_315);  dequantize_affine_314 = dequantize_affine_315 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_79: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_157, [sym_size_int_2, -1, 20, 64]);  linear_157 = None
	        transpose_105: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_79, 1, 2);  view_79 = None
	        contiguous_105: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_105);  transpose_105 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_158 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_52, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_316: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_158[0]
	        getitem_317: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_158[1];  choose_qparams_affine_default_158 = None
	        quantize_affine_158: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_52, [1, 1, 1280], getitem_316, getitem_317, torch.int8);  layer_norm_52 = None
	        dequantize_affine_316: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_158, [1, 1, 1280], getitem_316, getitem_317, torch.int8);  quantize_affine_158 = getitem_316 = getitem_317 = None
	        dequantize_affine_317: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_316, dequantize_affine_317, model_audio_tower_layers_26_self_attn_v_proj_bias);  dequantize_affine_316 = dequantize_affine_317 = model_audio_tower_layers_26_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_80: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_158, [sym_size_int_2, -1, 20, 64]);  linear_158 = None
	        transpose_106: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_80, 1, 2);  view_80 = None
	        contiguous_106: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_106);  transpose_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_26: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_104, contiguous_105, contiguous_106, scale = 1.0);  contiguous_104 = contiguous_105 = contiguous_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_107: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_26, 1, 2);  scaled_dot_product_attention_26 = None
	        contiguous_107: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_107);  transpose_107 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_107, [sym_size_int_2, 1500, -1]);  contiguous_107 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_159 = torch.ops.torchao.choose_qparams_affine.default(reshape_26, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_318: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_159[0]
	        getitem_319: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_159[1];  choose_qparams_affine_default_159 = None
	        quantize_affine_159: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_26, [1, 1, 1280], getitem_318, getitem_319, torch.int8);  reshape_26 = None
	        dequantize_affine_318: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_159, [1, 1, 1280], getitem_318, getitem_319, torch.int8);  quantize_affine_159 = getitem_318 = getitem_319 = None
	        dequantize_affine_319: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_159: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_318, dequantize_affine_319, model_audio_tower_layers_26_self_attn_out_proj_bias);  dequantize_affine_318 = dequantize_affine_319 = model_audio_tower_layers_26_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_79: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_159, 0.0, False);  linear_159 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_373: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_364, dropout_79);  add_364 = dropout_79 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_53: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_373, [1280], model_audio_tower_layers_26_final_layer_norm_weight, model_audio_tower_layers_26_final_layer_norm_bias);  model_audio_tower_layers_26_final_layer_norm_weight = model_audio_tower_layers_26_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_160 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_53, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_320: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_160[0]
	        getitem_321: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_160[1];  choose_qparams_affine_default_160 = None
	        quantize_affine_160: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_53, [1, 1, 1280], getitem_320, getitem_321, torch.int8);  layer_norm_53 = None
	        dequantize_affine_320: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_160, [1, 1, 1280], getitem_320, getitem_321, torch.int8);  quantize_affine_160 = getitem_320 = getitem_321 = None
	        dequantize_affine_321: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_26_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_26_fc1_parametrizations_weight_original1, model_audio_tower_layers_26_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_26_fc1_parametrizations_weight_original0 = model_audio_tower_layers_26_fc1_parametrizations_weight_original1 = model_audio_tower_layers_26_fc1_parametrizations_weight_original2 = None
	        linear_160: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_320, dequantize_affine_321, model_audio_tower_layers_26_fc1_bias);  dequantize_affine_320 = dequantize_affine_321 = model_audio_tower_layers_26_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_28: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_160);  linear_160 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_80: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_28, 0.0, False);  gelu_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_161 = torch.ops.torchao.choose_qparams_affine.default(dropout_80, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_322: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_161[0]
	        getitem_323: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_161[1];  choose_qparams_affine_default_161 = None
	        quantize_affine_161: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_80, [1, 1, 5120], getitem_322, getitem_323, torch.int8);  dropout_80 = None
	        dequantize_affine_322: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_161, [1, 1, 5120], getitem_322, getitem_323, torch.int8);  quantize_affine_161 = getitem_322 = getitem_323 = None
	        dequantize_affine_323: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_26_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_26_fc2_parametrizations_weight_original1, model_audio_tower_layers_26_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_26_fc2_parametrizations_weight_original0 = model_audio_tower_layers_26_fc2_parametrizations_weight_original1 = model_audio_tower_layers_26_fc2_parametrizations_weight_original2 = None
	        linear_161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_322, dequantize_affine_323, model_audio_tower_layers_26_fc2_bias);  dequantize_affine_322 = dequantize_affine_323 = model_audio_tower_layers_26_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_81: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_161, 0.0, False);  linear_161 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_378: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_373, dropout_81);  add_373 = dropout_81 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_54: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_378, [1280], model_audio_tower_layers_27_self_attn_layer_norm_weight, model_audio_tower_layers_27_self_attn_layer_norm_bias);  model_audio_tower_layers_27_self_attn_layer_norm_weight = model_audio_tower_layers_27_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_162 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_54, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_324: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_162[0]
	        getitem_325: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_162[1];  choose_qparams_affine_default_162 = None
	        quantize_affine_162: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_54, [1, 1, 1280], getitem_324, getitem_325, torch.int8)
	        dequantize_affine_324: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_162, [1, 1, 1280], getitem_324, getitem_325, torch.int8);  quantize_affine_162 = getitem_324 = getitem_325 = None
	        dequantize_affine_325: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_162: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_324, dequantize_affine_325, model_audio_tower_layers_27_self_attn_q_proj_bias);  dequantize_affine_324 = dequantize_affine_325 = model_audio_tower_layers_27_self_attn_q_proj_bias = None
	        mul_1004: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_162, 0.125);  linear_162 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_81: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_1004, [sym_size_int_2, 1500, 20, 64]);  mul_1004 = None
	        transpose_108: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_81, 1, 2);  view_81 = None
	        contiguous_108: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_108);  transpose_108 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_163 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_54, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_326: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_163[0]
	        getitem_327: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_163[1];  choose_qparams_affine_default_163 = None
	        quantize_affine_163: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_54, [1, 1, 1280], getitem_326, getitem_327, torch.int8)
	        dequantize_affine_326: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_163, [1, 1, 1280], getitem_326, getitem_327, torch.int8);  quantize_affine_163 = getitem_326 = getitem_327 = None
	        dequantize_affine_327: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_163: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_326, dequantize_affine_327);  dequantize_affine_326 = dequantize_affine_327 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_82: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_163, [sym_size_int_2, -1, 20, 64]);  linear_163 = None
	        transpose_109: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_82, 1, 2);  view_82 = None
	        contiguous_109: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_109);  transpose_109 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_164 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_54, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_328: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_164[0]
	        getitem_329: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_164[1];  choose_qparams_affine_default_164 = None
	        quantize_affine_164: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_54, [1, 1, 1280], getitem_328, getitem_329, torch.int8);  layer_norm_54 = None
	        dequantize_affine_328: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_164, [1, 1, 1280], getitem_328, getitem_329, torch.int8);  quantize_affine_164 = getitem_328 = getitem_329 = None
	        dequantize_affine_329: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_328, dequantize_affine_329, model_audio_tower_layers_27_self_attn_v_proj_bias);  dequantize_affine_328 = dequantize_affine_329 = model_audio_tower_layers_27_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_83: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_164, [sym_size_int_2, -1, 20, 64]);  linear_164 = None
	        transpose_110: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_83, 1, 2);  view_83 = None
	        contiguous_110: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_110);  transpose_110 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_27: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_108, contiguous_109, contiguous_110, scale = 1.0);  contiguous_108 = contiguous_109 = contiguous_110 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_111: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_27, 1, 2);  scaled_dot_product_attention_27 = None
	        contiguous_111: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_111);  transpose_111 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_27: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_111, [sym_size_int_2, 1500, -1]);  contiguous_111 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_165 = torch.ops.torchao.choose_qparams_affine.default(reshape_27, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_330: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_165[0]
	        getitem_331: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_165[1];  choose_qparams_affine_default_165 = None
	        quantize_affine_165: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_27, [1, 1, 1280], getitem_330, getitem_331, torch.int8);  reshape_27 = None
	        dequantize_affine_330: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_165, [1, 1, 1280], getitem_330, getitem_331, torch.int8);  quantize_affine_165 = getitem_330 = getitem_331 = None
	        dequantize_affine_331: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_165: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_330, dequantize_affine_331, model_audio_tower_layers_27_self_attn_out_proj_bias);  dequantize_affine_330 = dequantize_affine_331 = model_audio_tower_layers_27_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_82: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_165, 0.0, False);  linear_165 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_387: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_378, dropout_82);  add_378 = dropout_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_55: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_387, [1280], model_audio_tower_layers_27_final_layer_norm_weight, model_audio_tower_layers_27_final_layer_norm_bias);  model_audio_tower_layers_27_final_layer_norm_weight = model_audio_tower_layers_27_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_166 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_55, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_332: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_166[0]
	        getitem_333: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_166[1];  choose_qparams_affine_default_166 = None
	        quantize_affine_166: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_55, [1, 1, 1280], getitem_332, getitem_333, torch.int8);  layer_norm_55 = None
	        dequantize_affine_332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_166, [1, 1, 1280], getitem_332, getitem_333, torch.int8);  quantize_affine_166 = getitem_332 = getitem_333 = None
	        dequantize_affine_333: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_27_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_27_fc1_parametrizations_weight_original1, model_audio_tower_layers_27_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_27_fc1_parametrizations_weight_original0 = model_audio_tower_layers_27_fc1_parametrizations_weight_original1 = model_audio_tower_layers_27_fc1_parametrizations_weight_original2 = None
	        linear_166: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_332, dequantize_affine_333, model_audio_tower_layers_27_fc1_bias);  dequantize_affine_332 = dequantize_affine_333 = model_audio_tower_layers_27_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_29: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_166);  linear_166 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_83: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_29, 0.0, False);  gelu_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_167 = torch.ops.torchao.choose_qparams_affine.default(dropout_83, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_334: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_167[0]
	        getitem_335: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_167[1];  choose_qparams_affine_default_167 = None
	        quantize_affine_167: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_83, [1, 1, 5120], getitem_334, getitem_335, torch.int8);  dropout_83 = None
	        dequantize_affine_334: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_167, [1, 1, 5120], getitem_334, getitem_335, torch.int8);  quantize_affine_167 = getitem_334 = getitem_335 = None
	        dequantize_affine_335: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_27_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_27_fc2_parametrizations_weight_original1, model_audio_tower_layers_27_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_27_fc2_parametrizations_weight_original0 = model_audio_tower_layers_27_fc2_parametrizations_weight_original1 = model_audio_tower_layers_27_fc2_parametrizations_weight_original2 = None
	        linear_167: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_334, dequantize_affine_335, model_audio_tower_layers_27_fc2_bias);  dequantize_affine_334 = dequantize_affine_335 = model_audio_tower_layers_27_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_84: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_167, 0.0, False);  linear_167 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_392: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_387, dropout_84);  add_387 = dropout_84 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_392, [1280], model_audio_tower_layers_28_self_attn_layer_norm_weight, model_audio_tower_layers_28_self_attn_layer_norm_bias);  model_audio_tower_layers_28_self_attn_layer_norm_weight = model_audio_tower_layers_28_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_168 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_56, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_336: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_168[0]
	        getitem_337: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_168[1];  choose_qparams_affine_default_168 = None
	        quantize_affine_168: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_56, [1, 1, 1280], getitem_336, getitem_337, torch.int8)
	        dequantize_affine_336: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_168, [1, 1, 1280], getitem_336, getitem_337, torch.int8);  quantize_affine_168 = getitem_336 = getitem_337 = None
	        dequantize_affine_337: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_168: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_336, dequantize_affine_337, model_audio_tower_layers_28_self_attn_q_proj_bias);  dequantize_affine_336 = dequantize_affine_337 = model_audio_tower_layers_28_self_attn_q_proj_bias = None
	        mul_1041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_168, 0.125);  linear_168 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_84: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_1041, [sym_size_int_2, 1500, 20, 64]);  mul_1041 = None
	        transpose_112: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_84, 1, 2);  view_84 = None
	        contiguous_112: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_112);  transpose_112 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_169 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_56, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_338: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_169[0]
	        getitem_339: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_169[1];  choose_qparams_affine_default_169 = None
	        quantize_affine_169: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_56, [1, 1, 1280], getitem_338, getitem_339, torch.int8)
	        dequantize_affine_338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_169, [1, 1, 1280], getitem_338, getitem_339, torch.int8);  quantize_affine_169 = getitem_338 = getitem_339 = None
	        dequantize_affine_339: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_169: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_338, dequantize_affine_339);  dequantize_affine_338 = dequantize_affine_339 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_85: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_169, [sym_size_int_2, -1, 20, 64]);  linear_169 = None
	        transpose_113: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_85, 1, 2);  view_85 = None
	        contiguous_113: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_113);  transpose_113 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_170 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_56, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_340: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_170[0]
	        getitem_341: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_170[1];  choose_qparams_affine_default_170 = None
	        quantize_affine_170: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_56, [1, 1, 1280], getitem_340, getitem_341, torch.int8);  layer_norm_56 = None
	        dequantize_affine_340: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_170, [1, 1, 1280], getitem_340, getitem_341, torch.int8);  quantize_affine_170 = getitem_340 = getitem_341 = None
	        dequantize_affine_341: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_340, dequantize_affine_341, model_audio_tower_layers_28_self_attn_v_proj_bias);  dequantize_affine_340 = dequantize_affine_341 = model_audio_tower_layers_28_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_86: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_170, [sym_size_int_2, -1, 20, 64]);  linear_170 = None
	        transpose_114: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_86, 1, 2);  view_86 = None
	        contiguous_114: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_114);  transpose_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_28: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_112, contiguous_113, contiguous_114, scale = 1.0);  contiguous_112 = contiguous_113 = contiguous_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_115: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_28, 1, 2);  scaled_dot_product_attention_28 = None
	        contiguous_115: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_115);  transpose_115 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_28: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_115, [sym_size_int_2, 1500, -1]);  contiguous_115 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_171 = torch.ops.torchao.choose_qparams_affine.default(reshape_28, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_342: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_171[0]
	        getitem_343: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_171[1];  choose_qparams_affine_default_171 = None
	        quantize_affine_171: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_28, [1, 1, 1280], getitem_342, getitem_343, torch.int8);  reshape_28 = None
	        dequantize_affine_342: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_171, [1, 1, 1280], getitem_342, getitem_343, torch.int8);  quantize_affine_171 = getitem_342 = getitem_343 = None
	        dequantize_affine_343: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_171: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_342, dequantize_affine_343, model_audio_tower_layers_28_self_attn_out_proj_bias);  dequantize_affine_342 = dequantize_affine_343 = model_audio_tower_layers_28_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_85: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_171, 0.0, False);  linear_171 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_401: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_392, dropout_85);  add_392 = dropout_85 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_57: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_401, [1280], model_audio_tower_layers_28_final_layer_norm_weight, model_audio_tower_layers_28_final_layer_norm_bias);  model_audio_tower_layers_28_final_layer_norm_weight = model_audio_tower_layers_28_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_172 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_57, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_344: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_172[0]
	        getitem_345: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_172[1];  choose_qparams_affine_default_172 = None
	        quantize_affine_172: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_57, [1, 1, 1280], getitem_344, getitem_345, torch.int8);  layer_norm_57 = None
	        dequantize_affine_344: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_172, [1, 1, 1280], getitem_344, getitem_345, torch.int8);  quantize_affine_172 = getitem_344 = getitem_345 = None
	        dequantize_affine_345: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_28_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_28_fc1_parametrizations_weight_original1, model_audio_tower_layers_28_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_28_fc1_parametrizations_weight_original0 = model_audio_tower_layers_28_fc1_parametrizations_weight_original1 = model_audio_tower_layers_28_fc1_parametrizations_weight_original2 = None
	        linear_172: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_344, dequantize_affine_345, model_audio_tower_layers_28_fc1_bias);  dequantize_affine_344 = dequantize_affine_345 = model_audio_tower_layers_28_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_30: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_172);  linear_172 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_86: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_30, 0.0, False);  gelu_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_173 = torch.ops.torchao.choose_qparams_affine.default(dropout_86, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_346: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_173[0]
	        getitem_347: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_173[1];  choose_qparams_affine_default_173 = None
	        quantize_affine_173: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_86, [1, 1, 5120], getitem_346, getitem_347, torch.int8);  dropout_86 = None
	        dequantize_affine_346: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_173, [1, 1, 5120], getitem_346, getitem_347, torch.int8);  quantize_affine_173 = getitem_346 = getitem_347 = None
	        dequantize_affine_347: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_28_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_28_fc2_parametrizations_weight_original1, model_audio_tower_layers_28_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_28_fc2_parametrizations_weight_original0 = model_audio_tower_layers_28_fc2_parametrizations_weight_original1 = model_audio_tower_layers_28_fc2_parametrizations_weight_original2 = None
	        linear_173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_346, dequantize_affine_347, model_audio_tower_layers_28_fc2_bias);  dequantize_affine_346 = dequantize_affine_347 = model_audio_tower_layers_28_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_87: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_173, 0.0, False);  linear_173 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_406: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_401, dropout_87);  add_401 = dropout_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_58: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_406, [1280], model_audio_tower_layers_29_self_attn_layer_norm_weight, model_audio_tower_layers_29_self_attn_layer_norm_bias);  model_audio_tower_layers_29_self_attn_layer_norm_weight = model_audio_tower_layers_29_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_174 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_58, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_348: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_174[0]
	        getitem_349: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_174[1];  choose_qparams_affine_default_174 = None
	        quantize_affine_174: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_58, [1, 1, 1280], getitem_348, getitem_349, torch.int8)
	        dequantize_affine_348: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_174, [1, 1, 1280], getitem_348, getitem_349, torch.int8);  quantize_affine_174 = getitem_348 = getitem_349 = None
	        dequantize_affine_349: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_348, dequantize_affine_349, model_audio_tower_layers_29_self_attn_q_proj_bias);  dequantize_affine_348 = dequantize_affine_349 = model_audio_tower_layers_29_self_attn_q_proj_bias = None
	        mul_1078: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_174, 0.125);  linear_174 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_87: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_1078, [sym_size_int_2, 1500, 20, 64]);  mul_1078 = None
	        transpose_116: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_87, 1, 2);  view_87 = None
	        contiguous_116: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_116);  transpose_116 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_175 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_58, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_350: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_175[0]
	        getitem_351: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_175[1];  choose_qparams_affine_default_175 = None
	        quantize_affine_175: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_58, [1, 1, 1280], getitem_350, getitem_351, torch.int8)
	        dequantize_affine_350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_175, [1, 1, 1280], getitem_350, getitem_351, torch.int8);  quantize_affine_175 = getitem_350 = getitem_351 = None
	        dequantize_affine_351: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_175: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_350, dequantize_affine_351);  dequantize_affine_350 = dequantize_affine_351 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_88: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_175, [sym_size_int_2, -1, 20, 64]);  linear_175 = None
	        transpose_117: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_88, 1, 2);  view_88 = None
	        contiguous_117: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_117);  transpose_117 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_176 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_58, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_352: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_176[0]
	        getitem_353: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_176[1];  choose_qparams_affine_default_176 = None
	        quantize_affine_176: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_58, [1, 1, 1280], getitem_352, getitem_353, torch.int8);  layer_norm_58 = None
	        dequantize_affine_352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_176, [1, 1, 1280], getitem_352, getitem_353, torch.int8);  quantize_affine_176 = getitem_352 = getitem_353 = None
	        dequantize_affine_353: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_352, dequantize_affine_353, model_audio_tower_layers_29_self_attn_v_proj_bias);  dequantize_affine_352 = dequantize_affine_353 = model_audio_tower_layers_29_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_89: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_176, [sym_size_int_2, -1, 20, 64]);  linear_176 = None
	        transpose_118: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_89, 1, 2);  view_89 = None
	        contiguous_118: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_118);  transpose_118 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_29: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_116, contiguous_117, contiguous_118, scale = 1.0);  contiguous_116 = contiguous_117 = contiguous_118 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_119: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_29, 1, 2);  scaled_dot_product_attention_29 = None
	        contiguous_119: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_119);  transpose_119 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_29: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_119, [sym_size_int_2, 1500, -1]);  contiguous_119 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_177 = torch.ops.torchao.choose_qparams_affine.default(reshape_29, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_354: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_177[0]
	        getitem_355: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_177[1];  choose_qparams_affine_default_177 = None
	        quantize_affine_177: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_29, [1, 1, 1280], getitem_354, getitem_355, torch.int8);  reshape_29 = None
	        dequantize_affine_354: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_177, [1, 1, 1280], getitem_354, getitem_355, torch.int8);  quantize_affine_177 = getitem_354 = getitem_355 = None
	        dequantize_affine_355: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_177: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_354, dequantize_affine_355, model_audio_tower_layers_29_self_attn_out_proj_bias);  dequantize_affine_354 = dequantize_affine_355 = model_audio_tower_layers_29_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_88: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_177, 0.0, False);  linear_177 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_415: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_406, dropout_88);  add_406 = dropout_88 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_59: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_415, [1280], model_audio_tower_layers_29_final_layer_norm_weight, model_audio_tower_layers_29_final_layer_norm_bias);  model_audio_tower_layers_29_final_layer_norm_weight = model_audio_tower_layers_29_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_178 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_59, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_356: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_178[0]
	        getitem_357: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_178[1];  choose_qparams_affine_default_178 = None
	        quantize_affine_178: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_59, [1, 1, 1280], getitem_356, getitem_357, torch.int8);  layer_norm_59 = None
	        dequantize_affine_356: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_178, [1, 1, 1280], getitem_356, getitem_357, torch.int8);  quantize_affine_178 = getitem_356 = getitem_357 = None
	        dequantize_affine_357: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_29_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_29_fc1_parametrizations_weight_original1, model_audio_tower_layers_29_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_29_fc1_parametrizations_weight_original0 = model_audio_tower_layers_29_fc1_parametrizations_weight_original1 = model_audio_tower_layers_29_fc1_parametrizations_weight_original2 = None
	        linear_178: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_356, dequantize_affine_357, model_audio_tower_layers_29_fc1_bias);  dequantize_affine_356 = dequantize_affine_357 = model_audio_tower_layers_29_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_31: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_178);  linear_178 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_89: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_31, 0.0, False);  gelu_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_179 = torch.ops.torchao.choose_qparams_affine.default(dropout_89, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_358: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_179[0]
	        getitem_359: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_179[1];  choose_qparams_affine_default_179 = None
	        quantize_affine_179: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_89, [1, 1, 5120], getitem_358, getitem_359, torch.int8);  dropout_89 = None
	        dequantize_affine_358: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_179, [1, 1, 5120], getitem_358, getitem_359, torch.int8);  quantize_affine_179 = getitem_358 = getitem_359 = None
	        dequantize_affine_359: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_29_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_29_fc2_parametrizations_weight_original1, model_audio_tower_layers_29_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_29_fc2_parametrizations_weight_original0 = model_audio_tower_layers_29_fc2_parametrizations_weight_original1 = model_audio_tower_layers_29_fc2_parametrizations_weight_original2 = None
	        linear_179: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_358, dequantize_affine_359, model_audio_tower_layers_29_fc2_bias);  dequantize_affine_358 = dequantize_affine_359 = model_audio_tower_layers_29_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_90: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_179, 0.0, False);  linear_179 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_420: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_415, dropout_90);  add_415 = dropout_90 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_60: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_420, [1280], model_audio_tower_layers_30_self_attn_layer_norm_weight, model_audio_tower_layers_30_self_attn_layer_norm_bias);  model_audio_tower_layers_30_self_attn_layer_norm_weight = model_audio_tower_layers_30_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_180 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_60, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_360: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_180[0]
	        getitem_361: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_180[1];  choose_qparams_affine_default_180 = None
	        quantize_affine_180: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_60, [1, 1, 1280], getitem_360, getitem_361, torch.int8)
	        dequantize_affine_360: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_180, [1, 1, 1280], getitem_360, getitem_361, torch.int8);  quantize_affine_180 = getitem_360 = getitem_361 = None
	        dequantize_affine_361: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_180: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_360, dequantize_affine_361, model_audio_tower_layers_30_self_attn_q_proj_bias);  dequantize_affine_360 = dequantize_affine_361 = model_audio_tower_layers_30_self_attn_q_proj_bias = None
	        mul_1115: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_180, 0.125);  linear_180 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_90: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_1115, [sym_size_int_2, 1500, 20, 64]);  mul_1115 = None
	        transpose_120: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_90, 1, 2);  view_90 = None
	        contiguous_120: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_120);  transpose_120 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_181 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_60, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_362: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_181[0]
	        getitem_363: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_181[1];  choose_qparams_affine_default_181 = None
	        quantize_affine_181: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_60, [1, 1, 1280], getitem_362, getitem_363, torch.int8)
	        dequantize_affine_362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_181, [1, 1, 1280], getitem_362, getitem_363, torch.int8);  quantize_affine_181 = getitem_362 = getitem_363 = None
	        dequantize_affine_363: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_362, dequantize_affine_363);  dequantize_affine_362 = dequantize_affine_363 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_91: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_181, [sym_size_int_2, -1, 20, 64]);  linear_181 = None
	        transpose_121: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_91, 1, 2);  view_91 = None
	        contiguous_121: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_121);  transpose_121 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_182 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_60, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_364: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_182[0]
	        getitem_365: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_182[1];  choose_qparams_affine_default_182 = None
	        quantize_affine_182: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_60, [1, 1, 1280], getitem_364, getitem_365, torch.int8);  layer_norm_60 = None
	        dequantize_affine_364: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_182, [1, 1, 1280], getitem_364, getitem_365, torch.int8);  quantize_affine_182 = getitem_364 = getitem_365 = None
	        dequantize_affine_365: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_364, dequantize_affine_365, model_audio_tower_layers_30_self_attn_v_proj_bias);  dequantize_affine_364 = dequantize_affine_365 = model_audio_tower_layers_30_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_92: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_182, [sym_size_int_2, -1, 20, 64]);  linear_182 = None
	        transpose_122: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_92, 1, 2);  view_92 = None
	        contiguous_122: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_122);  transpose_122 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_30: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_120, contiguous_121, contiguous_122, scale = 1.0);  contiguous_120 = contiguous_121 = contiguous_122 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_123: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_30, 1, 2);  scaled_dot_product_attention_30 = None
	        contiguous_123: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_123);  transpose_123 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_123, [sym_size_int_2, 1500, -1]);  contiguous_123 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_183 = torch.ops.torchao.choose_qparams_affine.default(reshape_30, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_366: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_183[0]
	        getitem_367: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_183[1];  choose_qparams_affine_default_183 = None
	        quantize_affine_183: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_30, [1, 1, 1280], getitem_366, getitem_367, torch.int8);  reshape_30 = None
	        dequantize_affine_366: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_183, [1, 1, 1280], getitem_366, getitem_367, torch.int8);  quantize_affine_183 = getitem_366 = getitem_367 = None
	        dequantize_affine_367: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_183: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_366, dequantize_affine_367, model_audio_tower_layers_30_self_attn_out_proj_bias);  dequantize_affine_366 = dequantize_affine_367 = model_audio_tower_layers_30_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_91: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_183, 0.0, False);  linear_183 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_429: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_420, dropout_91);  add_420 = dropout_91 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_61: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_429, [1280], model_audio_tower_layers_30_final_layer_norm_weight, model_audio_tower_layers_30_final_layer_norm_bias);  model_audio_tower_layers_30_final_layer_norm_weight = model_audio_tower_layers_30_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_184 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_61, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_368: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_184[0]
	        getitem_369: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_184[1];  choose_qparams_affine_default_184 = None
	        quantize_affine_184: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_61, [1, 1, 1280], getitem_368, getitem_369, torch.int8);  layer_norm_61 = None
	        dequantize_affine_368: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_184, [1, 1, 1280], getitem_368, getitem_369, torch.int8);  quantize_affine_184 = getitem_368 = getitem_369 = None
	        dequantize_affine_369: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_30_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_30_fc1_parametrizations_weight_original1, model_audio_tower_layers_30_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_30_fc1_parametrizations_weight_original0 = model_audio_tower_layers_30_fc1_parametrizations_weight_original1 = model_audio_tower_layers_30_fc1_parametrizations_weight_original2 = None
	        linear_184: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_368, dequantize_affine_369, model_audio_tower_layers_30_fc1_bias);  dequantize_affine_368 = dequantize_affine_369 = model_audio_tower_layers_30_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_32: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_184);  linear_184 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_92: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_32, 0.0, False);  gelu_32 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_185 = torch.ops.torchao.choose_qparams_affine.default(dropout_92, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_370: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_185[0]
	        getitem_371: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_185[1];  choose_qparams_affine_default_185 = None
	        quantize_affine_185: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_92, [1, 1, 5120], getitem_370, getitem_371, torch.int8);  dropout_92 = None
	        dequantize_affine_370: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_185, [1, 1, 5120], getitem_370, getitem_371, torch.int8);  quantize_affine_185 = getitem_370 = getitem_371 = None
	        dequantize_affine_371: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_30_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_30_fc2_parametrizations_weight_original1, model_audio_tower_layers_30_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_30_fc2_parametrizations_weight_original0 = model_audio_tower_layers_30_fc2_parametrizations_weight_original1 = model_audio_tower_layers_30_fc2_parametrizations_weight_original2 = None
	        linear_185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_370, dequantize_affine_371, model_audio_tower_layers_30_fc2_bias);  dequantize_affine_370 = dequantize_affine_371 = model_audio_tower_layers_30_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_93: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_185, 0.0, False);  linear_185 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_434: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_429, dropout_93);  add_429 = dropout_93 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_434, [1280], model_audio_tower_layers_31_self_attn_layer_norm_weight, model_audio_tower_layers_31_self_attn_layer_norm_bias);  model_audio_tower_layers_31_self_attn_layer_norm_weight = model_audio_tower_layers_31_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_186 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_62, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_372: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_186[0]
	        getitem_373: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_186[1];  choose_qparams_affine_default_186 = None
	        quantize_affine_186: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_62, [1, 1, 1280], getitem_372, getitem_373, torch.int8)
	        dequantize_affine_372: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_186, [1, 1, 1280], getitem_372, getitem_373, torch.int8);  quantize_affine_186 = getitem_372 = getitem_373 = None
	        dequantize_affine_373: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_186: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_372, dequantize_affine_373, model_audio_tower_layers_31_self_attn_q_proj_bias);  dequantize_affine_372 = dequantize_affine_373 = model_audio_tower_layers_31_self_attn_q_proj_bias = None
	        mul_1152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_186, 0.125);  linear_186 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_93: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_1152, [sym_size_int_2, 1500, 20, 64]);  mul_1152 = None
	        transpose_124: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_93, 1, 2);  view_93 = None
	        contiguous_124: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_124);  transpose_124 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_187 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_62, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_374: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_187[0]
	        getitem_375: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_187[1];  choose_qparams_affine_default_187 = None
	        quantize_affine_187: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_62, [1, 1, 1280], getitem_374, getitem_375, torch.int8)
	        dequantize_affine_374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_187, [1, 1, 1280], getitem_374, getitem_375, torch.int8);  quantize_affine_187 = getitem_374 = getitem_375 = None
	        dequantize_affine_375: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_187: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_374, dequantize_affine_375);  dequantize_affine_374 = dequantize_affine_375 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_94: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_187, [sym_size_int_2, -1, 20, 64]);  linear_187 = None
	        transpose_125: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_94, 1, 2);  view_94 = None
	        contiguous_125: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_125);  transpose_125 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_188 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_62, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_376: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_188[0]
	        getitem_377: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_188[1];  choose_qparams_affine_default_188 = None
	        quantize_affine_188: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_62, [1, 1, 1280], getitem_376, getitem_377, torch.int8);  layer_norm_62 = None
	        dequantize_affine_376: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_188, [1, 1, 1280], getitem_376, getitem_377, torch.int8);  quantize_affine_188 = getitem_376 = getitem_377 = None
	        dequantize_affine_377: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_376, dequantize_affine_377, model_audio_tower_layers_31_self_attn_v_proj_bias);  dequantize_affine_376 = dequantize_affine_377 = model_audio_tower_layers_31_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_95: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_188, [sym_size_int_2, -1, 20, 64]);  linear_188 = None
	        transpose_126: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_95, 1, 2);  view_95 = None
	        contiguous_126: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_126);  transpose_126 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_31: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_124, contiguous_125, contiguous_126, scale = 1.0);  contiguous_124 = contiguous_125 = contiguous_126 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_127: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_31, 1, 2);  scaled_dot_product_attention_31 = None
	        contiguous_127: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_127);  transpose_127 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_31: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_127, [sym_size_int_2, 1500, -1]);  contiguous_127 = sym_size_int_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_189 = torch.ops.torchao.choose_qparams_affine.default(reshape_31, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_378: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_189[0]
	        getitem_379: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_189[1];  choose_qparams_affine_default_189 = None
	        quantize_affine_189: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_31, [1, 1, 1280], getitem_378, getitem_379, torch.int8);  reshape_31 = None
	        dequantize_affine_378: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_189, [1, 1, 1280], getitem_378, getitem_379, torch.int8);  quantize_affine_189 = getitem_378 = getitem_379 = None
	        dequantize_affine_379: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_378, dequantize_affine_379, model_audio_tower_layers_31_self_attn_out_proj_bias);  dequantize_affine_378 = dequantize_affine_379 = model_audio_tower_layers_31_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_94: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_189, 0.0, False);  linear_189 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_443: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_434, dropout_94);  add_434 = dropout_94 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_63: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_443, [1280], model_audio_tower_layers_31_final_layer_norm_weight, model_audio_tower_layers_31_final_layer_norm_bias);  model_audio_tower_layers_31_final_layer_norm_weight = model_audio_tower_layers_31_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_190 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_63, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_380: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_190[0]
	        getitem_381: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_190[1];  choose_qparams_affine_default_190 = None
	        quantize_affine_190: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_63, [1, 1, 1280], getitem_380, getitem_381, torch.int8);  layer_norm_63 = None
	        dequantize_affine_380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_190, [1, 1, 1280], getitem_380, getitem_381, torch.int8);  quantize_affine_190 = getitem_380 = getitem_381 = None
	        dequantize_affine_381: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_31_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_31_fc1_parametrizations_weight_original1, model_audio_tower_layers_31_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_31_fc1_parametrizations_weight_original0 = model_audio_tower_layers_31_fc1_parametrizations_weight_original1 = model_audio_tower_layers_31_fc1_parametrizations_weight_original2 = None
	        linear_190: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_380, dequantize_affine_381, model_audio_tower_layers_31_fc1_bias);  dequantize_affine_380 = dequantize_affine_381 = model_audio_tower_layers_31_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_33: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_190);  linear_190 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_95: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_33, 0.0, False);  gelu_33 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_191 = torch.ops.torchao.choose_qparams_affine.default(dropout_95, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_382: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_191[0]
	        getitem_383: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_191[1];  choose_qparams_affine_default_191 = None
	        quantize_affine_191: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_95, [1, 1, 5120], getitem_382, getitem_383, torch.int8);  dropout_95 = None
	        dequantize_affine_382: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_191, [1, 1, 5120], getitem_382, getitem_383, torch.int8);  quantize_affine_191 = getitem_382 = getitem_383 = None
	        dequantize_affine_383: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_31_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_31_fc2_parametrizations_weight_original1, model_audio_tower_layers_31_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_31_fc2_parametrizations_weight_original0 = model_audio_tower_layers_31_fc2_parametrizations_weight_original1 = model_audio_tower_layers_31_fc2_parametrizations_weight_original2 = None
	        linear_191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_382, dequantize_affine_383, model_audio_tower_layers_31_fc2_bias);  dequantize_affine_382 = dequantize_affine_383 = model_audio_tower_layers_31_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_96: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_191, 0.0, False);  linear_191 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_448: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_443, dropout_96);  add_443 = dropout_96 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:365 in forward, code: hidden_states = self.layer_norm(hidden_states)
	        layer_norm_64: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_448, [1280], model_audio_tower_layer_norm_weight, model_audio_tower_layer_norm_bias);  add_448 = model_audio_tower_layer_norm_weight = model_audio_tower_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:451 in get_audio_embeds, code: audio_hidden_states = audio_hidden_states.reshape(-1, self.config.audio_config.intermediate_size)
	        reshape_32: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(layer_norm_64, [-1, 5120]);  layer_norm_64 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:389 in forward, code: hidden_states = self.linear_1(audio_features)
	        choose_qparams_affine_default_192 = torch.ops.torchao.choose_qparams_affine.default(reshape_32, 'ASYMMETRIC', [1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_384: "f32[375*s6][1]cuda:0" = choose_qparams_affine_default_192[0]
	        getitem_385: "i8[375*s6][1]cuda:0" = choose_qparams_affine_default_192[1];  choose_qparams_affine_default_192 = None
	        quantize_affine_192: "i8[375*s6, 5120][5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_32, [1, 5120], getitem_384, getitem_385, torch.int8);  reshape_32 = None
	        dequantize_affine_384: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_192, [1, 5120], getitem_384, getitem_385, torch.int8);  quantize_affine_192 = getitem_384 = getitem_385 = None
	        dequantize_affine_385: "f32[3072, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_multi_modal_projector_linear_1_parametrizations_weight_original0, [1, 32], model_multi_modal_projector_linear_1_parametrizations_weight_original1, model_multi_modal_projector_linear_1_parametrizations_weight_original2, torch.int8, -8, 7);  model_multi_modal_projector_linear_1_parametrizations_weight_original0 = model_multi_modal_projector_linear_1_parametrizations_weight_original1 = model_multi_modal_projector_linear_1_parametrizations_weight_original2 = None
	        linear_192: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_384, dequantize_affine_385);  dequantize_affine_384 = dequantize_affine_385 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_34: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.gelu.default(linear_192);  linear_192 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:391 in forward, code: hidden_states = self.linear_2(hidden_states)
	        choose_qparams_affine_default_193 = torch.ops.torchao.choose_qparams_affine.default(gelu_34, 'ASYMMETRIC', [1, 3072], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_386: "f32[375*s6][1]cuda:0" = choose_qparams_affine_default_193[0]
	        getitem_387: "i8[375*s6][1]cuda:0" = choose_qparams_affine_default_193[1];  choose_qparams_affine_default_193 = None
	        quantize_affine_193: "i8[375*s6, 3072][3072, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(gelu_34, [1, 3072], getitem_386, getitem_387, torch.int8);  gelu_34 = None
	        dequantize_affine_386: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_193, [1, 3072], getitem_386, getitem_387, torch.int8);  quantize_affine_193 = getitem_386 = getitem_387 = None
	        dequantize_affine_387: "f32[3072, 3072][3072, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_multi_modal_projector_linear_2_parametrizations_weight_original0, [1, 32], model_multi_modal_projector_linear_2_parametrizations_weight_original1, model_multi_modal_projector_linear_2_parametrizations_weight_original2, torch.int8, -8, 7);  model_multi_modal_projector_linear_2_parametrizations_weight_original0 = model_multi_modal_projector_linear_2_parametrizations_weight_original1 = model_multi_modal_projector_linear_2_parametrizations_weight_original2 = None
	        linear_193: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_386, dequantize_affine_387);  dequantize_affine_386 = dequantize_affine_387 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py:83 in forward, code: return audio_embeds.unsqueeze(0)
	        unsqueeze: "f32[1, 375*s6, 3072][1152000*s6, 3072, 1]cuda:0" = torch.ops.aten.unsqueeze.default(linear_193, 0);  linear_193 = None
	        return (unsqueeze,)
	        
	
	 # graph id: 140612625281824
V0910 09:41:48.456000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "dd7bc90783eb27b9286b7d6f811d4f88"}
	{
	"name": "_recursive_pre_grad_passes",
	"ts": 1757522508456239.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:41:49.005000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "04e5a19e596a3b286e7b7002208f1b1c"}
	{
	"name": "_recursive_pre_grad_passes",
	"ts": 1757522509005007.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:41:49.151000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_inductor/compile_fx.py:2353] {"artifact": {"name": "after_pre_grad_graph", "encoding": "string"}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2511, "name": "compile_fx", "filename": 19, "loc": "model_ = run_pre_grad_passes(model_, example_inputs_)"}, {"line": 2353, "name": "run_pre_grad_passes", "filename": 19, "loc": "trace_structured("}], "has_payload": "52d2886661c5d8302c82502bddf2438b"}
	class GraphModule(torch.nn.Module):
	    def forward(self, input_features: "f32[s6, 128, 3000][384000, 3000, 1]cuda:0"):
	        # No stacktrace found for following nodes
	        model_audio_tower_embed_positions_weight: "f32[1500, 1280][1280, 1]cuda:0" = self.model.audio_tower.embed_positions.weight
	        model_audio_tower_conv1_weight: "f32[1280, 128, 3][384, 3, 1]cuda:0" = self.model.audio_tower.conv1.weight
	        model_audio_tower_conv1_bias: "f32[1280][1]cuda:0" = self.model.audio_tower.conv1.bias
	        model_audio_tower_conv2_weight: "f32[1280, 1280, 3][3840, 3, 1]cuda:0" = self.model.audio_tower.conv2.weight
	        model_audio_tower_conv2_bias: "f32[1280][1]cuda:0" = self.model.audio_tower.conv2.bias
	        model_audio_tower_layers_0_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn_layer_norm.weight
	        model_audio_tower_layers_0_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn_layer_norm.bias
	        model_audio_tower_layers_0_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.bias
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.bias
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.bias
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").final_layer_norm.weight
	        model_audio_tower_layers_0_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").final_layer_norm.bias
	        model_audio_tower_layers_0_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.bias
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_0_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.bias
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn_layer_norm.weight
	        model_audio_tower_layers_1_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn_layer_norm.bias
	        model_audio_tower_layers_1_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.bias
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.bias
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.bias
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").final_layer_norm.weight
	        model_audio_tower_layers_1_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").final_layer_norm.bias
	        model_audio_tower_layers_1_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.bias
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_1_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.bias
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn_layer_norm.weight
	        model_audio_tower_layers_2_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn_layer_norm.bias
	        model_audio_tower_layers_2_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.bias
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.bias
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.bias
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").final_layer_norm.weight
	        model_audio_tower_layers_2_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").final_layer_norm.bias
	        model_audio_tower_layers_2_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.bias
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_2_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.bias
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn_layer_norm.weight
	        model_audio_tower_layers_3_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn_layer_norm.bias
	        model_audio_tower_layers_3_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.bias
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.bias
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.bias
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").final_layer_norm.weight
	        model_audio_tower_layers_3_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").final_layer_norm.bias
	        model_audio_tower_layers_3_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.bias
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_3_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.bias
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn_layer_norm.weight
	        model_audio_tower_layers_4_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn_layer_norm.bias
	        model_audio_tower_layers_4_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.bias
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.bias
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.bias
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").final_layer_norm.weight
	        model_audio_tower_layers_4_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").final_layer_norm.bias
	        model_audio_tower_layers_4_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.bias
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_4_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.bias
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn_layer_norm.weight
	        model_audio_tower_layers_5_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn_layer_norm.bias
	        model_audio_tower_layers_5_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.bias
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.bias
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.bias
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").final_layer_norm.weight
	        model_audio_tower_layers_5_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").final_layer_norm.bias
	        model_audio_tower_layers_5_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.bias
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_5_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.bias
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn_layer_norm.weight
	        model_audio_tower_layers_6_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn_layer_norm.bias
	        model_audio_tower_layers_6_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.bias
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.bias
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.bias
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").final_layer_norm.weight
	        model_audio_tower_layers_6_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").final_layer_norm.bias
	        model_audio_tower_layers_6_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.bias
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_6_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.bias
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn_layer_norm.weight
	        model_audio_tower_layers_7_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn_layer_norm.bias
	        model_audio_tower_layers_7_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.bias
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.bias
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.bias
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").final_layer_norm.weight
	        model_audio_tower_layers_7_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").final_layer_norm.bias
	        model_audio_tower_layers_7_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.bias
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_7_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.bias
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn_layer_norm.weight
	        model_audio_tower_layers_8_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn_layer_norm.bias
	        model_audio_tower_layers_8_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.bias
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.bias
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.bias
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").final_layer_norm.weight
	        model_audio_tower_layers_8_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").final_layer_norm.bias
	        model_audio_tower_layers_8_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.bias
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_8_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.bias
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn_layer_norm.weight
	        model_audio_tower_layers_9_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn_layer_norm.bias
	        model_audio_tower_layers_9_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.bias
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.bias
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.bias
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").final_layer_norm.weight
	        model_audio_tower_layers_9_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").final_layer_norm.bias
	        model_audio_tower_layers_9_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.bias
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_9_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.bias
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn_layer_norm.weight
	        model_audio_tower_layers_10_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn_layer_norm.bias
	        model_audio_tower_layers_10_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.bias
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.bias
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.bias
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").final_layer_norm.weight
	        model_audio_tower_layers_10_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").final_layer_norm.bias
	        model_audio_tower_layers_10_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.bias
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_10_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.bias
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn_layer_norm.weight
	        model_audio_tower_layers_11_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn_layer_norm.bias
	        model_audio_tower_layers_11_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.bias
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.bias
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.bias
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").final_layer_norm.weight
	        model_audio_tower_layers_11_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").final_layer_norm.bias
	        model_audio_tower_layers_11_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.bias
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_11_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.bias
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn_layer_norm.weight
	        model_audio_tower_layers_12_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn_layer_norm.bias
	        model_audio_tower_layers_12_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.bias
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.bias
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.bias
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").final_layer_norm.weight
	        model_audio_tower_layers_12_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").final_layer_norm.bias
	        model_audio_tower_layers_12_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.bias
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_12_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.bias
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn_layer_norm.weight
	        model_audio_tower_layers_13_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn_layer_norm.bias
	        model_audio_tower_layers_13_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.bias
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.bias
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.bias
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").final_layer_norm.weight
	        model_audio_tower_layers_13_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").final_layer_norm.bias
	        model_audio_tower_layers_13_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.bias
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_13_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.bias
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn_layer_norm.weight
	        model_audio_tower_layers_14_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn_layer_norm.bias
	        model_audio_tower_layers_14_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.bias
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.bias
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.bias
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").final_layer_norm.weight
	        model_audio_tower_layers_14_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").final_layer_norm.bias
	        model_audio_tower_layers_14_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.bias
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_14_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.bias
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn_layer_norm.weight
	        model_audio_tower_layers_15_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn_layer_norm.bias
	        model_audio_tower_layers_15_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.bias
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.bias
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.bias
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").final_layer_norm.weight
	        model_audio_tower_layers_15_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").final_layer_norm.bias
	        model_audio_tower_layers_15_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.bias
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_15_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.bias
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn_layer_norm.weight
	        model_audio_tower_layers_16_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn_layer_norm.bias
	        model_audio_tower_layers_16_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.bias
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.bias
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.bias
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").final_layer_norm.weight
	        model_audio_tower_layers_16_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").final_layer_norm.bias
	        model_audio_tower_layers_16_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.bias
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_16_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.bias
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn_layer_norm.weight
	        model_audio_tower_layers_17_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn_layer_norm.bias
	        model_audio_tower_layers_17_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.bias
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.bias
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.bias
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").final_layer_norm.weight
	        model_audio_tower_layers_17_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").final_layer_norm.bias
	        model_audio_tower_layers_17_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.bias
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_17_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.bias
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn_layer_norm.weight
	        model_audio_tower_layers_18_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn_layer_norm.bias
	        model_audio_tower_layers_18_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.bias
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.bias
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.bias
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").final_layer_norm.weight
	        model_audio_tower_layers_18_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").final_layer_norm.bias
	        model_audio_tower_layers_18_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.bias
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_18_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.bias
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn_layer_norm.weight
	        model_audio_tower_layers_19_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn_layer_norm.bias
	        model_audio_tower_layers_19_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.bias
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.bias
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.bias
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").final_layer_norm.weight
	        model_audio_tower_layers_19_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").final_layer_norm.bias
	        model_audio_tower_layers_19_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.bias
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_19_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.bias
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn_layer_norm.weight
	        model_audio_tower_layers_20_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn_layer_norm.bias
	        model_audio_tower_layers_20_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.bias
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.bias
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.bias
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").final_layer_norm.weight
	        model_audio_tower_layers_20_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").final_layer_norm.bias
	        model_audio_tower_layers_20_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.bias
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_20_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.bias
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn_layer_norm.weight
	        model_audio_tower_layers_21_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn_layer_norm.bias
	        model_audio_tower_layers_21_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.bias
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.bias
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.bias
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").final_layer_norm.weight
	        model_audio_tower_layers_21_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").final_layer_norm.bias
	        model_audio_tower_layers_21_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.bias
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_21_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.bias
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn_layer_norm.weight
	        model_audio_tower_layers_22_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn_layer_norm.bias
	        model_audio_tower_layers_22_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.bias
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.bias
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.bias
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").final_layer_norm.weight
	        model_audio_tower_layers_22_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").final_layer_norm.bias
	        model_audio_tower_layers_22_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.bias
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_22_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.bias
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn_layer_norm.weight
	        model_audio_tower_layers_23_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn_layer_norm.bias
	        model_audio_tower_layers_23_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.bias
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.bias
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.bias
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").final_layer_norm.weight
	        model_audio_tower_layers_23_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").final_layer_norm.bias
	        model_audio_tower_layers_23_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.bias
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_23_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.bias
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn_layer_norm.weight
	        model_audio_tower_layers_24_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn_layer_norm.bias
	        model_audio_tower_layers_24_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.bias
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.bias
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.bias
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").final_layer_norm.weight
	        model_audio_tower_layers_24_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").final_layer_norm.bias
	        model_audio_tower_layers_24_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.bias
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_24_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.bias
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn_layer_norm.weight
	        model_audio_tower_layers_25_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn_layer_norm.bias
	        model_audio_tower_layers_25_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.bias
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.bias
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.bias
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").final_layer_norm.weight
	        model_audio_tower_layers_25_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").final_layer_norm.bias
	        model_audio_tower_layers_25_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.bias
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_25_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.bias
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn_layer_norm.weight
	        model_audio_tower_layers_26_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn_layer_norm.bias
	        model_audio_tower_layers_26_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.bias
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.bias
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.bias
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").final_layer_norm.weight
	        model_audio_tower_layers_26_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").final_layer_norm.bias
	        model_audio_tower_layers_26_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.bias
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_26_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.bias
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn_layer_norm.weight
	        model_audio_tower_layers_27_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn_layer_norm.bias
	        model_audio_tower_layers_27_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.bias
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.bias
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.bias
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").final_layer_norm.weight
	        model_audio_tower_layers_27_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").final_layer_norm.bias
	        model_audio_tower_layers_27_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.bias
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_27_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.bias
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn_layer_norm.weight
	        model_audio_tower_layers_28_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn_layer_norm.bias
	        model_audio_tower_layers_28_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.bias
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.bias
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.bias
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").final_layer_norm.weight
	        model_audio_tower_layers_28_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").final_layer_norm.bias
	        model_audio_tower_layers_28_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.bias
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_28_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.bias
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn_layer_norm.weight
	        model_audio_tower_layers_29_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn_layer_norm.bias
	        model_audio_tower_layers_29_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.bias
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.bias
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.bias
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").final_layer_norm.weight
	        model_audio_tower_layers_29_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").final_layer_norm.bias
	        model_audio_tower_layers_29_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.bias
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_29_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.bias
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn_layer_norm.weight
	        model_audio_tower_layers_30_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn_layer_norm.bias
	        model_audio_tower_layers_30_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.bias
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.bias
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.bias
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").final_layer_norm.weight
	        model_audio_tower_layers_30_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").final_layer_norm.bias
	        model_audio_tower_layers_30_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.bias
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_30_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.bias
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn_layer_norm.weight
	        model_audio_tower_layers_31_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn_layer_norm.bias
	        model_audio_tower_layers_31_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.bias
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.bias
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.bias
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").final_layer_norm.weight
	        model_audio_tower_layers_31_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").final_layer_norm.bias
	        model_audio_tower_layers_31_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.bias
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_31_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.bias
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original2
	        model_audio_tower_layer_norm_weight: "f32[1280][1]cuda:0" = self.model.audio_tower.layer_norm.weight
	        model_audio_tower_layer_norm_bias: "f32[1280][1]cuda:0" = self.model.audio_tower.layer_norm.bias
	        model_multi_modal_projector_linear_1_parametrizations_weight_original0: "i8[3072, 5120][5120, 1]cuda:0" = self.model.multi_modal_projector.linear_1.parametrizations.weight.original0
	        model_multi_modal_projector_linear_1_parametrizations_weight_original1: "f32[3072, 160][160, 1]cuda:0" = self.model.multi_modal_projector.linear_1.parametrizations.weight.original1
	        model_multi_modal_projector_linear_1_parametrizations_weight_original2: "i8[3072, 160][160, 1]cuda:0" = self.model.multi_modal_projector.linear_1.parametrizations.weight.original2
	        model_multi_modal_projector_linear_2_parametrizations_weight_original0: "i8[3072, 3072][3072, 1]cuda:0" = self.model.multi_modal_projector.linear_2.parametrizations.weight.original0
	        model_multi_modal_projector_linear_2_parametrizations_weight_original1: "f32[3072, 96][96, 1]cuda:0" = self.model.multi_modal_projector.linear_2.parametrizations.weight.original1
	        model_multi_modal_projector_linear_2_parametrizations_weight_original2: "i8[3072, 96][96, 1]cuda:0" = self.model.multi_modal_projector.linear_2.parametrizations.weight.original2
	        
	         # 
	        sym_size_int_2: "Sym(s6)" = torch.ops.aten.sym_size.int(input_features, 0)
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:348 in forward, code: input_features = input_features.to(dtype=self.conv1.weight.dtype, device=self.conv1.weight.device)
	        _assert_tensor_metadata_default = torch.ops.aten._assert_tensor_metadata.default(input_features, dtype = torch.float32, device = 'cuda:0', layout = torch.strided);  _assert_tensor_metadata_default = None
	        to: "f32[s6, 128, 3000][384000, 3000, 1]cuda:0" = torch.ops.aten.to.device(input_features, 'cuda:0', torch.float32);  input_features = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:349 in forward, code: inputs_embeds = nn.functional.gelu(self.conv1(input_features))
	        conv1d: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.conv1d.default(to, model_audio_tower_conv1_weight, model_audio_tower_conv1_bias, [1], [1]);  to = model_audio_tower_conv1_weight = model_audio_tower_conv1_bias = None
	        gelu: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.gelu.default(conv1d);  conv1d = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:350 in forward, code: inputs_embeds = nn.functional.gelu(self.conv2(inputs_embeds))
	        conv1d_1: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.conv1d.default(gelu, model_audio_tower_conv2_weight, model_audio_tower_conv2_bias, [2], [1]);  gelu = model_audio_tower_conv2_weight = model_audio_tower_conv2_bias = None
	        gelu_1: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.gelu.default(conv1d_1);  conv1d_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:351 in forward, code: inputs_embeds = inputs_embeds.permute(0, 2, 1)
	        permute: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.permute.default(gelu_1, [0, 2, 1]);  gelu_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:354 in forward, code: hidden_states = (inputs_embeds + embed_pos).to(inputs_embeds.dtype)
	        add: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(permute, model_audio_tower_embed_positions_weight);  permute = model_audio_tower_embed_positions_weight = None
	        _assert_tensor_metadata_default_1 = torch.ops.aten._assert_tensor_metadata.default(add, dtype = torch.float32, device = 'cuda:0', layout = torch.strided);  _assert_tensor_metadata_default_1 = None
	        to_1: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.to.dtype(add, torch.float32);  add = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:355 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.dropout.default(to_1, 0.0, False);  to_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(dropout, [1280], model_audio_tower_layers_0_self_attn_layer_norm_weight, model_audio_tower_layers_0_self_attn_layer_norm_bias);  model_audio_tower_layers_0_self_attn_layer_norm_weight = model_audio_tower_layers_0_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default = torch.ops.torchao.choose_qparams_affine.default(layer_norm, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default[0]
	        getitem_1: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default[1];  choose_qparams_affine_default = None
	        quantize_affine: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm, [1, 1, 1280], getitem, getitem_1, torch.int8)
	        dequantize_affine: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine, [1, 1, 1280], getitem, getitem_1, torch.int8);  quantize_affine = getitem = getitem_1 = None
	        dequantize_affine_1: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine, dequantize_affine_1, model_audio_tower_layers_0_self_attn_q_proj_bias);  dequantize_affine = dequantize_affine_1 = model_audio_tower_layers_0_self_attn_q_proj_bias = None
	        mul_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear, 0.125);  linear = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_5, [sym_size_int_2, 1500, 20, 64]);  mul_5 = None
	        transpose: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view, 1, 2);  view = None
	        contiguous: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose);  transpose = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_1 = torch.ops.torchao.choose_qparams_affine.default(layer_norm, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_2: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_1[0]
	        getitem_3: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_1[1];  choose_qparams_affine_default_1 = None
	        quantize_affine_1: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm, [1, 1, 1280], getitem_2, getitem_3, torch.int8)
	        dequantize_affine_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_1, [1, 1, 1280], getitem_2, getitem_3, torch.int8);  quantize_affine_1 = getitem_2 = getitem_3 = None
	        dequantize_affine_3: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_2, dequantize_affine_3);  dequantize_affine_2 = dequantize_affine_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_1, [sym_size_int_2, -1, 20, 64]);  linear_1 = None
	        transpose_1: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_1, 1, 2);  view_1 = None
	        contiguous_1: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_1);  transpose_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_2 = torch.ops.torchao.choose_qparams_affine.default(layer_norm, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_4: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_2[0]
	        getitem_5: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_2[1];  choose_qparams_affine_default_2 = None
	        quantize_affine_2: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm, [1, 1, 1280], getitem_4, getitem_5, torch.int8);  layer_norm = None
	        dequantize_affine_4: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_2, [1, 1, 1280], getitem_4, getitem_5, torch.int8);  quantize_affine_2 = getitem_4 = getitem_5 = None
	        dequantize_affine_5: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_4, dequantize_affine_5, model_audio_tower_layers_0_self_attn_v_proj_bias);  dequantize_affine_4 = dequantize_affine_5 = model_audio_tower_layers_0_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_2, [sym_size_int_2, -1, 20, 64]);  linear_2 = None
	        transpose_2: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_2, 1, 2);  view_2 = None
	        contiguous_2: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_2);  transpose_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous, contiguous_1, contiguous_2, scale = 1.0);  contiguous = contiguous_1 = contiguous_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_3: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention, 1, 2);  scaled_dot_product_attention = None
	        contiguous_3: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_3);  transpose_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_3, [sym_size_int_2, 1500, -1]);  contiguous_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_3 = torch.ops.torchao.choose_qparams_affine.default(reshape, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_6: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_3[0]
	        getitem_7: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_3[1];  choose_qparams_affine_default_3 = None
	        quantize_affine_3: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape, [1, 1, 1280], getitem_6, getitem_7, torch.int8);  reshape = None
	        dequantize_affine_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_3, [1, 1, 1280], getitem_6, getitem_7, torch.int8);  quantize_affine_3 = getitem_6 = getitem_7 = None
	        dequantize_affine_7: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_3: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_6, dequantize_affine_7, model_audio_tower_layers_0_self_attn_out_proj_bias);  dequantize_affine_6 = dequantize_affine_7 = model_audio_tower_layers_0_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_3, 0.0, False);  linear_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_9: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(dropout, dropout_1);  dropout = dropout_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_9, [1280], model_audio_tower_layers_0_final_layer_norm_weight, model_audio_tower_layers_0_final_layer_norm_bias);  model_audio_tower_layers_0_final_layer_norm_weight = model_audio_tower_layers_0_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_4 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_1, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_8: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_4[0]
	        getitem_9: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_4[1];  choose_qparams_affine_default_4 = None
	        quantize_affine_4: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_1, [1, 1, 1280], getitem_8, getitem_9, torch.int8);  layer_norm_1 = None
	        dequantize_affine_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_4, [1, 1, 1280], getitem_8, getitem_9, torch.int8);  quantize_affine_4 = getitem_8 = getitem_9 = None
	        dequantize_affine_9: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_0_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_0_fc1_parametrizations_weight_original1, model_audio_tower_layers_0_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_0_fc1_parametrizations_weight_original0 = model_audio_tower_layers_0_fc1_parametrizations_weight_original1 = model_audio_tower_layers_0_fc1_parametrizations_weight_original2 = None
	        linear_4: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_8, dequantize_affine_9, model_audio_tower_layers_0_fc1_bias);  dequantize_affine_8 = dequantize_affine_9 = model_audio_tower_layers_0_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_2: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_4);  linear_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_2: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_2, 0.0, False);  gelu_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_5 = torch.ops.torchao.choose_qparams_affine.default(dropout_2, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_10: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_5[0]
	        getitem_11: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_5[1];  choose_qparams_affine_default_5 = None
	        quantize_affine_5: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_2, [1, 1, 5120], getitem_10, getitem_11, torch.int8);  dropout_2 = None
	        dequantize_affine_10: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_5, [1, 1, 5120], getitem_10, getitem_11, torch.int8);  quantize_affine_5 = getitem_10 = getitem_11 = None
	        dequantize_affine_11: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_0_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_0_fc2_parametrizations_weight_original1, model_audio_tower_layers_0_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_0_fc2_parametrizations_weight_original0 = model_audio_tower_layers_0_fc2_parametrizations_weight_original1 = model_audio_tower_layers_0_fc2_parametrizations_weight_original2 = None
	        linear_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_10, dequantize_affine_11, model_audio_tower_layers_0_fc2_bias);  dequantize_affine_10 = dequantize_affine_11 = model_audio_tower_layers_0_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_3: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_5, 0.0, False);  linear_5 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_14: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_9, dropout_3);  add_9 = dropout_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_14, [1280], model_audio_tower_layers_1_self_attn_layer_norm_weight, model_audio_tower_layers_1_self_attn_layer_norm_bias);  model_audio_tower_layers_1_self_attn_layer_norm_weight = model_audio_tower_layers_1_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_6 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_2, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_12: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_6[0]
	        getitem_13: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_6[1];  choose_qparams_affine_default_6 = None
	        quantize_affine_6: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_2, [1, 1, 1280], getitem_12, getitem_13, torch.int8)
	        dequantize_affine_12: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_6, [1, 1, 1280], getitem_12, getitem_13, torch.int8);  quantize_affine_6 = getitem_12 = getitem_13 = None
	        dequantize_affine_13: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_12, dequantize_affine_13, model_audio_tower_layers_1_self_attn_q_proj_bias);  dequantize_affine_12 = dequantize_affine_13 = model_audio_tower_layers_1_self_attn_q_proj_bias = None
	        mul_42: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_6, 0.125);  linear_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_3: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_42, [sym_size_int_2, 1500, 20, 64]);  mul_42 = None
	        transpose_4: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_3, 1, 2);  view_3 = None
	        contiguous_4: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_4);  transpose_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_7 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_2, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_14: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_7[0]
	        getitem_15: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_7[1];  choose_qparams_affine_default_7 = None
	        quantize_affine_7: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_2, [1, 1, 1280], getitem_14, getitem_15, torch.int8)
	        dequantize_affine_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_7, [1, 1, 1280], getitem_14, getitem_15, torch.int8);  quantize_affine_7 = getitem_14 = getitem_15 = None
	        dequantize_affine_15: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_7: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_14, dequantize_affine_15);  dequantize_affine_14 = dequantize_affine_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_4: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_7, [sym_size_int_2, -1, 20, 64]);  linear_7 = None
	        transpose_5: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_4, 1, 2);  view_4 = None
	        contiguous_5: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_5);  transpose_5 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_8 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_2, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_16: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_8[0]
	        getitem_17: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_8[1];  choose_qparams_affine_default_8 = None
	        quantize_affine_8: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_2, [1, 1, 1280], getitem_16, getitem_17, torch.int8);  layer_norm_2 = None
	        dequantize_affine_16: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_8, [1, 1, 1280], getitem_16, getitem_17, torch.int8);  quantize_affine_8 = getitem_16 = getitem_17 = None
	        dequantize_affine_17: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_16, dequantize_affine_17, model_audio_tower_layers_1_self_attn_v_proj_bias);  dequantize_affine_16 = dequantize_affine_17 = model_audio_tower_layers_1_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_5: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_8, [sym_size_int_2, -1, 20, 64]);  linear_8 = None
	        transpose_6: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_5, 1, 2);  view_5 = None
	        contiguous_6: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_6);  transpose_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_1: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_4, contiguous_5, contiguous_6, scale = 1.0);  contiguous_4 = contiguous_5 = contiguous_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_7: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_1, 1, 2);  scaled_dot_product_attention_1 = None
	        contiguous_7: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_7);  transpose_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_7, [sym_size_int_2, 1500, -1]);  contiguous_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_9 = torch.ops.torchao.choose_qparams_affine.default(reshape_1, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_18: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_9[0]
	        getitem_19: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_9[1];  choose_qparams_affine_default_9 = None
	        quantize_affine_9: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_1, [1, 1, 1280], getitem_18, getitem_19, torch.int8);  reshape_1 = None
	        dequantize_affine_18: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_9, [1, 1, 1280], getitem_18, getitem_19, torch.int8);  quantize_affine_9 = getitem_18 = getitem_19 = None
	        dequantize_affine_19: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_9: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_18, dequantize_affine_19, model_audio_tower_layers_1_self_attn_out_proj_bias);  dequantize_affine_18 = dequantize_affine_19 = model_audio_tower_layers_1_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_4: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_9, 0.0, False);  linear_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_23: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_14, dropout_4);  add_14 = dropout_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_3: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_23, [1280], model_audio_tower_layers_1_final_layer_norm_weight, model_audio_tower_layers_1_final_layer_norm_bias);  model_audio_tower_layers_1_final_layer_norm_weight = model_audio_tower_layers_1_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_10 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_3, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_20: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_10[0]
	        getitem_21: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_10[1];  choose_qparams_affine_default_10 = None
	        quantize_affine_10: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_3, [1, 1, 1280], getitem_20, getitem_21, torch.int8);  layer_norm_3 = None
	        dequantize_affine_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_10, [1, 1, 1280], getitem_20, getitem_21, torch.int8);  quantize_affine_10 = getitem_20 = getitem_21 = None
	        dequantize_affine_21: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_1_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_1_fc1_parametrizations_weight_original1, model_audio_tower_layers_1_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_1_fc1_parametrizations_weight_original0 = model_audio_tower_layers_1_fc1_parametrizations_weight_original1 = model_audio_tower_layers_1_fc1_parametrizations_weight_original2 = None
	        linear_10: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_20, dequantize_affine_21, model_audio_tower_layers_1_fc1_bias);  dequantize_affine_20 = dequantize_affine_21 = model_audio_tower_layers_1_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_3: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_10);  linear_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_5: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_3, 0.0, False);  gelu_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_11 = torch.ops.torchao.choose_qparams_affine.default(dropout_5, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_22: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_11[0]
	        getitem_23: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_11[1];  choose_qparams_affine_default_11 = None
	        quantize_affine_11: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_5, [1, 1, 5120], getitem_22, getitem_23, torch.int8);  dropout_5 = None
	        dequantize_affine_22: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_11, [1, 1, 5120], getitem_22, getitem_23, torch.int8);  quantize_affine_11 = getitem_22 = getitem_23 = None
	        dequantize_affine_23: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_1_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_1_fc2_parametrizations_weight_original1, model_audio_tower_layers_1_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_1_fc2_parametrizations_weight_original0 = model_audio_tower_layers_1_fc2_parametrizations_weight_original1 = model_audio_tower_layers_1_fc2_parametrizations_weight_original2 = None
	        linear_11: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_22, dequantize_affine_23, model_audio_tower_layers_1_fc2_bias);  dequantize_affine_22 = dequantize_affine_23 = model_audio_tower_layers_1_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_11, 0.0, False);  linear_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_28: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_23, dropout_6);  add_23 = dropout_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_4: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_28, [1280], model_audio_tower_layers_2_self_attn_layer_norm_weight, model_audio_tower_layers_2_self_attn_layer_norm_bias);  model_audio_tower_layers_2_self_attn_layer_norm_weight = model_audio_tower_layers_2_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_12 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_4, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_24: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_12[0]
	        getitem_25: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_12[1];  choose_qparams_affine_default_12 = None
	        quantize_affine_12: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_4, [1, 1, 1280], getitem_24, getitem_25, torch.int8)
	        dequantize_affine_24: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_12, [1, 1, 1280], getitem_24, getitem_25, torch.int8);  quantize_affine_12 = getitem_24 = getitem_25 = None
	        dequantize_affine_25: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_12: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_24, dequantize_affine_25, model_audio_tower_layers_2_self_attn_q_proj_bias);  dequantize_affine_24 = dequantize_affine_25 = model_audio_tower_layers_2_self_attn_q_proj_bias = None
	        mul_79: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_12, 0.125);  linear_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_6: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_79, [sym_size_int_2, 1500, 20, 64]);  mul_79 = None
	        transpose_8: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_6, 1, 2);  view_6 = None
	        contiguous_8: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_8);  transpose_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_13 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_4, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_26: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_13[0]
	        getitem_27: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_13[1];  choose_qparams_affine_default_13 = None
	        quantize_affine_13: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_4, [1, 1, 1280], getitem_26, getitem_27, torch.int8)
	        dequantize_affine_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_13, [1, 1, 1280], getitem_26, getitem_27, torch.int8);  quantize_affine_13 = getitem_26 = getitem_27 = None
	        dequantize_affine_27: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_13: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_26, dequantize_affine_27);  dequantize_affine_26 = dequantize_affine_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_7: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_13, [sym_size_int_2, -1, 20, 64]);  linear_13 = None
	        transpose_9: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_7, 1, 2);  view_7 = None
	        contiguous_9: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_9);  transpose_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_14 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_4, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_28: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_14[0]
	        getitem_29: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_14[1];  choose_qparams_affine_default_14 = None
	        quantize_affine_14: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_4, [1, 1, 1280], getitem_28, getitem_29, torch.int8);  layer_norm_4 = None
	        dequantize_affine_28: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_14, [1, 1, 1280], getitem_28, getitem_29, torch.int8);  quantize_affine_14 = getitem_28 = getitem_29 = None
	        dequantize_affine_29: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_28, dequantize_affine_29, model_audio_tower_layers_2_self_attn_v_proj_bias);  dequantize_affine_28 = dequantize_affine_29 = model_audio_tower_layers_2_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_8: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_14, [sym_size_int_2, -1, 20, 64]);  linear_14 = None
	        transpose_10: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_8, 1, 2);  view_8 = None
	        contiguous_10: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_10);  transpose_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_2: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_8, contiguous_9, contiguous_10, scale = 1.0);  contiguous_8 = contiguous_9 = contiguous_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_11: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_2, 1, 2);  scaled_dot_product_attention_2 = None
	        contiguous_11: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_11);  transpose_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_11, [sym_size_int_2, 1500, -1]);  contiguous_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_15 = torch.ops.torchao.choose_qparams_affine.default(reshape_2, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_30: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_15[0]
	        getitem_31: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_15[1];  choose_qparams_affine_default_15 = None
	        quantize_affine_15: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_2, [1, 1, 1280], getitem_30, getitem_31, torch.int8);  reshape_2 = None
	        dequantize_affine_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_15, [1, 1, 1280], getitem_30, getitem_31, torch.int8);  quantize_affine_15 = getitem_30 = getitem_31 = None
	        dequantize_affine_31: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_15: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_30, dequantize_affine_31, model_audio_tower_layers_2_self_attn_out_proj_bias);  dequantize_affine_30 = dequantize_affine_31 = model_audio_tower_layers_2_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_7: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_15, 0.0, False);  linear_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_37: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_28, dropout_7);  add_28 = dropout_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_37, [1280], model_audio_tower_layers_2_final_layer_norm_weight, model_audio_tower_layers_2_final_layer_norm_bias);  model_audio_tower_layers_2_final_layer_norm_weight = model_audio_tower_layers_2_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_16 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_5, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_32: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_16[0]
	        getitem_33: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_16[1];  choose_qparams_affine_default_16 = None
	        quantize_affine_16: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_5, [1, 1, 1280], getitem_32, getitem_33, torch.int8);  layer_norm_5 = None
	        dequantize_affine_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_16, [1, 1, 1280], getitem_32, getitem_33, torch.int8);  quantize_affine_16 = getitem_32 = getitem_33 = None
	        dequantize_affine_33: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_2_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_2_fc1_parametrizations_weight_original1, model_audio_tower_layers_2_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_2_fc1_parametrizations_weight_original0 = model_audio_tower_layers_2_fc1_parametrizations_weight_original1 = model_audio_tower_layers_2_fc1_parametrizations_weight_original2 = None
	        linear_16: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_32, dequantize_affine_33, model_audio_tower_layers_2_fc1_bias);  dequantize_affine_32 = dequantize_affine_33 = model_audio_tower_layers_2_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_4: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_16);  linear_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_8: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_4, 0.0, False);  gelu_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_17 = torch.ops.torchao.choose_qparams_affine.default(dropout_8, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_34: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_17[0]
	        getitem_35: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_17[1];  choose_qparams_affine_default_17 = None
	        quantize_affine_17: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_8, [1, 1, 5120], getitem_34, getitem_35, torch.int8);  dropout_8 = None
	        dequantize_affine_34: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_17, [1, 1, 5120], getitem_34, getitem_35, torch.int8);  quantize_affine_17 = getitem_34 = getitem_35 = None
	        dequantize_affine_35: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_2_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_2_fc2_parametrizations_weight_original1, model_audio_tower_layers_2_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_2_fc2_parametrizations_weight_original0 = model_audio_tower_layers_2_fc2_parametrizations_weight_original1 = model_audio_tower_layers_2_fc2_parametrizations_weight_original2 = None
	        linear_17: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_34, dequantize_affine_35, model_audio_tower_layers_2_fc2_bias);  dequantize_affine_34 = dequantize_affine_35 = model_audio_tower_layers_2_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_9: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_17, 0.0, False);  linear_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_42: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_37, dropout_9);  add_37 = dropout_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_42, [1280], model_audio_tower_layers_3_self_attn_layer_norm_weight, model_audio_tower_layers_3_self_attn_layer_norm_bias);  model_audio_tower_layers_3_self_attn_layer_norm_weight = model_audio_tower_layers_3_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_18 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_6, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_36: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_18[0]
	        getitem_37: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_18[1];  choose_qparams_affine_default_18 = None
	        quantize_affine_18: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_6, [1, 1, 1280], getitem_36, getitem_37, torch.int8)
	        dequantize_affine_36: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_18, [1, 1, 1280], getitem_36, getitem_37, torch.int8);  quantize_affine_18 = getitem_36 = getitem_37 = None
	        dequantize_affine_37: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_18: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_36, dequantize_affine_37, model_audio_tower_layers_3_self_attn_q_proj_bias);  dequantize_affine_36 = dequantize_affine_37 = model_audio_tower_layers_3_self_attn_q_proj_bias = None
	        mul_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_18, 0.125);  linear_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_9: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_116, [sym_size_int_2, 1500, 20, 64]);  mul_116 = None
	        transpose_12: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_9, 1, 2);  view_9 = None
	        contiguous_12: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_12);  transpose_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_19 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_6, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_38: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_19[0]
	        getitem_39: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_19[1];  choose_qparams_affine_default_19 = None
	        quantize_affine_19: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_6, [1, 1, 1280], getitem_38, getitem_39, torch.int8)
	        dequantize_affine_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_19, [1, 1, 1280], getitem_38, getitem_39, torch.int8);  quantize_affine_19 = getitem_38 = getitem_39 = None
	        dequantize_affine_39: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_19: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_38, dequantize_affine_39);  dequantize_affine_38 = dequantize_affine_39 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_10: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_19, [sym_size_int_2, -1, 20, 64]);  linear_19 = None
	        transpose_13: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_10, 1, 2);  view_10 = None
	        contiguous_13: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_13);  transpose_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_20 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_6, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_40: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_20[0]
	        getitem_41: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_20[1];  choose_qparams_affine_default_20 = None
	        quantize_affine_20: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_6, [1, 1, 1280], getitem_40, getitem_41, torch.int8);  layer_norm_6 = None
	        dequantize_affine_40: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_20, [1, 1, 1280], getitem_40, getitem_41, torch.int8);  quantize_affine_20 = getitem_40 = getitem_41 = None
	        dequantize_affine_41: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_40, dequantize_affine_41, model_audio_tower_layers_3_self_attn_v_proj_bias);  dequantize_affine_40 = dequantize_affine_41 = model_audio_tower_layers_3_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_11: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_20, [sym_size_int_2, -1, 20, 64]);  linear_20 = None
	        transpose_14: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_11, 1, 2);  view_11 = None
	        contiguous_14: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_14);  transpose_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_3: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_12, contiguous_13, contiguous_14, scale = 1.0);  contiguous_12 = contiguous_13 = contiguous_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_15: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_3, 1, 2);  scaled_dot_product_attention_3 = None
	        contiguous_15: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_15);  transpose_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_3: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_15, [sym_size_int_2, 1500, -1]);  contiguous_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_21 = torch.ops.torchao.choose_qparams_affine.default(reshape_3, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_42: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_21[0]
	        getitem_43: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_21[1];  choose_qparams_affine_default_21 = None
	        quantize_affine_21: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_3, [1, 1, 1280], getitem_42, getitem_43, torch.int8);  reshape_3 = None
	        dequantize_affine_42: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_21, [1, 1, 1280], getitem_42, getitem_43, torch.int8);  quantize_affine_21 = getitem_42 = getitem_43 = None
	        dequantize_affine_43: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_42, dequantize_affine_43, model_audio_tower_layers_3_self_attn_out_proj_bias);  dequantize_affine_42 = dequantize_affine_43 = model_audio_tower_layers_3_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_10: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_21, 0.0, False);  linear_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_51: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_42, dropout_10);  add_42 = dropout_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_7: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_51, [1280], model_audio_tower_layers_3_final_layer_norm_weight, model_audio_tower_layers_3_final_layer_norm_bias);  model_audio_tower_layers_3_final_layer_norm_weight = model_audio_tower_layers_3_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_22 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_7, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_44: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_22[0]
	        getitem_45: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_22[1];  choose_qparams_affine_default_22 = None
	        quantize_affine_22: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_7, [1, 1, 1280], getitem_44, getitem_45, torch.int8);  layer_norm_7 = None
	        dequantize_affine_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_22, [1, 1, 1280], getitem_44, getitem_45, torch.int8);  quantize_affine_22 = getitem_44 = getitem_45 = None
	        dequantize_affine_45: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_3_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_3_fc1_parametrizations_weight_original1, model_audio_tower_layers_3_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_3_fc1_parametrizations_weight_original0 = model_audio_tower_layers_3_fc1_parametrizations_weight_original1 = model_audio_tower_layers_3_fc1_parametrizations_weight_original2 = None
	        linear_22: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_44, dequantize_affine_45, model_audio_tower_layers_3_fc1_bias);  dequantize_affine_44 = dequantize_affine_45 = model_audio_tower_layers_3_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_5: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_22);  linear_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_11: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_5, 0.0, False);  gelu_5 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_23 = torch.ops.torchao.choose_qparams_affine.default(dropout_11, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_46: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_23[0]
	        getitem_47: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_23[1];  choose_qparams_affine_default_23 = None
	        quantize_affine_23: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_11, [1, 1, 5120], getitem_46, getitem_47, torch.int8);  dropout_11 = None
	        dequantize_affine_46: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_23, [1, 1, 5120], getitem_46, getitem_47, torch.int8);  quantize_affine_23 = getitem_46 = getitem_47 = None
	        dequantize_affine_47: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_3_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_3_fc2_parametrizations_weight_original1, model_audio_tower_layers_3_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_3_fc2_parametrizations_weight_original0 = model_audio_tower_layers_3_fc2_parametrizations_weight_original1 = model_audio_tower_layers_3_fc2_parametrizations_weight_original2 = None
	        linear_23: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_46, dequantize_affine_47, model_audio_tower_layers_3_fc2_bias);  dequantize_affine_46 = dequantize_affine_47 = model_audio_tower_layers_3_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_12: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_23, 0.0, False);  linear_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_56: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_51, dropout_12);  add_51 = dropout_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_56, [1280], model_audio_tower_layers_4_self_attn_layer_norm_weight, model_audio_tower_layers_4_self_attn_layer_norm_bias);  model_audio_tower_layers_4_self_attn_layer_norm_weight = model_audio_tower_layers_4_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_24 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_8, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_48: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_24[0]
	        getitem_49: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_24[1];  choose_qparams_affine_default_24 = None
	        quantize_affine_24: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_8, [1, 1, 1280], getitem_48, getitem_49, torch.int8)
	        dequantize_affine_48: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_24, [1, 1, 1280], getitem_48, getitem_49, torch.int8);  quantize_affine_24 = getitem_48 = getitem_49 = None
	        dequantize_affine_49: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_24: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_48, dequantize_affine_49, model_audio_tower_layers_4_self_attn_q_proj_bias);  dequantize_affine_48 = dequantize_affine_49 = model_audio_tower_layers_4_self_attn_q_proj_bias = None
	        mul_153: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_24, 0.125);  linear_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_12: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_153, [sym_size_int_2, 1500, 20, 64]);  mul_153 = None
	        transpose_16: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_12, 1, 2);  view_12 = None
	        contiguous_16: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_16);  transpose_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_25 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_8, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_50: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_25[0]
	        getitem_51: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_25[1];  choose_qparams_affine_default_25 = None
	        quantize_affine_25: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_8, [1, 1, 1280], getitem_50, getitem_51, torch.int8)
	        dequantize_affine_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_25, [1, 1, 1280], getitem_50, getitem_51, torch.int8);  quantize_affine_25 = getitem_50 = getitem_51 = None
	        dequantize_affine_51: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_25: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_50, dequantize_affine_51);  dequantize_affine_50 = dequantize_affine_51 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_13: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_25, [sym_size_int_2, -1, 20, 64]);  linear_25 = None
	        transpose_17: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_13, 1, 2);  view_13 = None
	        contiguous_17: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_17);  transpose_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_26 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_8, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_52: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_26[0]
	        getitem_53: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_26[1];  choose_qparams_affine_default_26 = None
	        quantize_affine_26: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_8, [1, 1, 1280], getitem_52, getitem_53, torch.int8);  layer_norm_8 = None
	        dequantize_affine_52: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_26, [1, 1, 1280], getitem_52, getitem_53, torch.int8);  quantize_affine_26 = getitem_52 = getitem_53 = None
	        dequantize_affine_53: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_52, dequantize_affine_53, model_audio_tower_layers_4_self_attn_v_proj_bias);  dequantize_affine_52 = dequantize_affine_53 = model_audio_tower_layers_4_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_14: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_26, [sym_size_int_2, -1, 20, 64]);  linear_26 = None
	        transpose_18: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_14, 1, 2);  view_14 = None
	        contiguous_18: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_18);  transpose_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_4: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_16, contiguous_17, contiguous_18, scale = 1.0);  contiguous_16 = contiguous_17 = contiguous_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_19: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_4, 1, 2);  scaled_dot_product_attention_4 = None
	        contiguous_19: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_19);  transpose_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_4: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_19, [sym_size_int_2, 1500, -1]);  contiguous_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_27 = torch.ops.torchao.choose_qparams_affine.default(reshape_4, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_54: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_27[0]
	        getitem_55: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_27[1];  choose_qparams_affine_default_27 = None
	        quantize_affine_27: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_4, [1, 1, 1280], getitem_54, getitem_55, torch.int8);  reshape_4 = None
	        dequantize_affine_54: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_27, [1, 1, 1280], getitem_54, getitem_55, torch.int8);  quantize_affine_27 = getitem_54 = getitem_55 = None
	        dequantize_affine_55: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_27: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_54, dequantize_affine_55, model_audio_tower_layers_4_self_attn_out_proj_bias);  dequantize_affine_54 = dequantize_affine_55 = model_audio_tower_layers_4_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_13: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_27, 0.0, False);  linear_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_65: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_56, dropout_13);  add_56 = dropout_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_9: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_65, [1280], model_audio_tower_layers_4_final_layer_norm_weight, model_audio_tower_layers_4_final_layer_norm_bias);  model_audio_tower_layers_4_final_layer_norm_weight = model_audio_tower_layers_4_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_28 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_9, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_56: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_28[0]
	        getitem_57: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_28[1];  choose_qparams_affine_default_28 = None
	        quantize_affine_28: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_9, [1, 1, 1280], getitem_56, getitem_57, torch.int8);  layer_norm_9 = None
	        dequantize_affine_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_28, [1, 1, 1280], getitem_56, getitem_57, torch.int8);  quantize_affine_28 = getitem_56 = getitem_57 = None
	        dequantize_affine_57: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_4_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_4_fc1_parametrizations_weight_original1, model_audio_tower_layers_4_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_4_fc1_parametrizations_weight_original0 = model_audio_tower_layers_4_fc1_parametrizations_weight_original1 = model_audio_tower_layers_4_fc1_parametrizations_weight_original2 = None
	        linear_28: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_56, dequantize_affine_57, model_audio_tower_layers_4_fc1_bias);  dequantize_affine_56 = dequantize_affine_57 = model_audio_tower_layers_4_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_6: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_28);  linear_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_14: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_6, 0.0, False);  gelu_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_29 = torch.ops.torchao.choose_qparams_affine.default(dropout_14, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_58: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_29[0]
	        getitem_59: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_29[1];  choose_qparams_affine_default_29 = None
	        quantize_affine_29: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_14, [1, 1, 5120], getitem_58, getitem_59, torch.int8);  dropout_14 = None
	        dequantize_affine_58: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_29, [1, 1, 5120], getitem_58, getitem_59, torch.int8);  quantize_affine_29 = getitem_58 = getitem_59 = None
	        dequantize_affine_59: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_4_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_4_fc2_parametrizations_weight_original1, model_audio_tower_layers_4_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_4_fc2_parametrizations_weight_original0 = model_audio_tower_layers_4_fc2_parametrizations_weight_original1 = model_audio_tower_layers_4_fc2_parametrizations_weight_original2 = None
	        linear_29: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_58, dequantize_affine_59, model_audio_tower_layers_4_fc2_bias);  dequantize_affine_58 = dequantize_affine_59 = model_audio_tower_layers_4_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_15: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_29, 0.0, False);  linear_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_70: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_65, dropout_15);  add_65 = dropout_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_10: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_70, [1280], model_audio_tower_layers_5_self_attn_layer_norm_weight, model_audio_tower_layers_5_self_attn_layer_norm_bias);  model_audio_tower_layers_5_self_attn_layer_norm_weight = model_audio_tower_layers_5_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_30 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_10, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_60: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_30[0]
	        getitem_61: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_30[1];  choose_qparams_affine_default_30 = None
	        quantize_affine_30: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_10, [1, 1, 1280], getitem_60, getitem_61, torch.int8)
	        dequantize_affine_60: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_30, [1, 1, 1280], getitem_60, getitem_61, torch.int8);  quantize_affine_30 = getitem_60 = getitem_61 = None
	        dequantize_affine_61: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_60, dequantize_affine_61, model_audio_tower_layers_5_self_attn_q_proj_bias);  dequantize_affine_60 = dequantize_affine_61 = model_audio_tower_layers_5_self_attn_q_proj_bias = None
	        mul_190: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_30, 0.125);  linear_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_15: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_190, [sym_size_int_2, 1500, 20, 64]);  mul_190 = None
	        transpose_20: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_15, 1, 2);  view_15 = None
	        contiguous_20: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_20);  transpose_20 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_31 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_10, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_62: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_31[0]
	        getitem_63: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_31[1];  choose_qparams_affine_default_31 = None
	        quantize_affine_31: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_10, [1, 1, 1280], getitem_62, getitem_63, torch.int8)
	        dequantize_affine_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_31, [1, 1, 1280], getitem_62, getitem_63, torch.int8);  quantize_affine_31 = getitem_62 = getitem_63 = None
	        dequantize_affine_63: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_31: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_62, dequantize_affine_63);  dequantize_affine_62 = dequantize_affine_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_16: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_31, [sym_size_int_2, -1, 20, 64]);  linear_31 = None
	        transpose_21: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_16, 1, 2);  view_16 = None
	        contiguous_21: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_21);  transpose_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_32 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_10, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_64: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_32[0]
	        getitem_65: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_32[1];  choose_qparams_affine_default_32 = None
	        quantize_affine_32: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_10, [1, 1, 1280], getitem_64, getitem_65, torch.int8);  layer_norm_10 = None
	        dequantize_affine_64: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_32, [1, 1, 1280], getitem_64, getitem_65, torch.int8);  quantize_affine_32 = getitem_64 = getitem_65 = None
	        dequantize_affine_65: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_64, dequantize_affine_65, model_audio_tower_layers_5_self_attn_v_proj_bias);  dequantize_affine_64 = dequantize_affine_65 = model_audio_tower_layers_5_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_17: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_32, [sym_size_int_2, -1, 20, 64]);  linear_32 = None
	        transpose_22: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_17, 1, 2);  view_17 = None
	        contiguous_22: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_22);  transpose_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_5: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_20, contiguous_21, contiguous_22, scale = 1.0);  contiguous_20 = contiguous_21 = contiguous_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_23: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_5, 1, 2);  scaled_dot_product_attention_5 = None
	        contiguous_23: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_23);  transpose_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_23, [sym_size_int_2, 1500, -1]);  contiguous_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_33 = torch.ops.torchao.choose_qparams_affine.default(reshape_5, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_66: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_33[0]
	        getitem_67: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_33[1];  choose_qparams_affine_default_33 = None
	        quantize_affine_33: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_5, [1, 1, 1280], getitem_66, getitem_67, torch.int8);  reshape_5 = None
	        dequantize_affine_66: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_33, [1, 1, 1280], getitem_66, getitem_67, torch.int8);  quantize_affine_33 = getitem_66 = getitem_67 = None
	        dequantize_affine_67: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_33: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_66, dequantize_affine_67, model_audio_tower_layers_5_self_attn_out_proj_bias);  dequantize_affine_66 = dequantize_affine_67 = model_audio_tower_layers_5_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_16: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_33, 0.0, False);  linear_33 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_79: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_70, dropout_16);  add_70 = dropout_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_11: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_79, [1280], model_audio_tower_layers_5_final_layer_norm_weight, model_audio_tower_layers_5_final_layer_norm_bias);  model_audio_tower_layers_5_final_layer_norm_weight = model_audio_tower_layers_5_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_34 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_11, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_68: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_34[0]
	        getitem_69: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_34[1];  choose_qparams_affine_default_34 = None
	        quantize_affine_34: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_11, [1, 1, 1280], getitem_68, getitem_69, torch.int8);  layer_norm_11 = None
	        dequantize_affine_68: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_34, [1, 1, 1280], getitem_68, getitem_69, torch.int8);  quantize_affine_34 = getitem_68 = getitem_69 = None
	        dequantize_affine_69: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_5_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_5_fc1_parametrizations_weight_original1, model_audio_tower_layers_5_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_5_fc1_parametrizations_weight_original0 = model_audio_tower_layers_5_fc1_parametrizations_weight_original1 = model_audio_tower_layers_5_fc1_parametrizations_weight_original2 = None
	        linear_34: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_68, dequantize_affine_69, model_audio_tower_layers_5_fc1_bias);  dequantize_affine_68 = dequantize_affine_69 = model_audio_tower_layers_5_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_7: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_34);  linear_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_17: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_7, 0.0, False);  gelu_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_35 = torch.ops.torchao.choose_qparams_affine.default(dropout_17, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_70: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_35[0]
	        getitem_71: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_35[1];  choose_qparams_affine_default_35 = None
	        quantize_affine_35: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_17, [1, 1, 5120], getitem_70, getitem_71, torch.int8);  dropout_17 = None
	        dequantize_affine_70: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_35, [1, 1, 5120], getitem_70, getitem_71, torch.int8);  quantize_affine_35 = getitem_70 = getitem_71 = None
	        dequantize_affine_71: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_5_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_5_fc2_parametrizations_weight_original1, model_audio_tower_layers_5_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_5_fc2_parametrizations_weight_original0 = model_audio_tower_layers_5_fc2_parametrizations_weight_original1 = model_audio_tower_layers_5_fc2_parametrizations_weight_original2 = None
	        linear_35: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_70, dequantize_affine_71, model_audio_tower_layers_5_fc2_bias);  dequantize_affine_70 = dequantize_affine_71 = model_audio_tower_layers_5_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_18: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_35, 0.0, False);  linear_35 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_84: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_79, dropout_18);  add_79 = dropout_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_12: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_84, [1280], model_audio_tower_layers_6_self_attn_layer_norm_weight, model_audio_tower_layers_6_self_attn_layer_norm_bias);  model_audio_tower_layers_6_self_attn_layer_norm_weight = model_audio_tower_layers_6_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_36 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_12, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_72: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_36[0]
	        getitem_73: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_36[1];  choose_qparams_affine_default_36 = None
	        quantize_affine_36: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_12, [1, 1, 1280], getitem_72, getitem_73, torch.int8)
	        dequantize_affine_72: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_36, [1, 1, 1280], getitem_72, getitem_73, torch.int8);  quantize_affine_36 = getitem_72 = getitem_73 = None
	        dequantize_affine_73: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_36: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_72, dequantize_affine_73, model_audio_tower_layers_6_self_attn_q_proj_bias);  dequantize_affine_72 = dequantize_affine_73 = model_audio_tower_layers_6_self_attn_q_proj_bias = None
	        mul_227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_36, 0.125);  linear_36 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_18: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_227, [sym_size_int_2, 1500, 20, 64]);  mul_227 = None
	        transpose_24: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_18, 1, 2);  view_18 = None
	        contiguous_24: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_24);  transpose_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_37 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_12, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_74: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_37[0]
	        getitem_75: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_37[1];  choose_qparams_affine_default_37 = None
	        quantize_affine_37: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_12, [1, 1, 1280], getitem_74, getitem_75, torch.int8)
	        dequantize_affine_74: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_37, [1, 1, 1280], getitem_74, getitem_75, torch.int8);  quantize_affine_37 = getitem_74 = getitem_75 = None
	        dequantize_affine_75: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_37: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_74, dequantize_affine_75);  dequantize_affine_74 = dequantize_affine_75 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_19: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_37, [sym_size_int_2, -1, 20, 64]);  linear_37 = None
	        transpose_25: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_19, 1, 2);  view_19 = None
	        contiguous_25: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_25);  transpose_25 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_38 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_12, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_76: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_38[0]
	        getitem_77: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_38[1];  choose_qparams_affine_default_38 = None
	        quantize_affine_38: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_12, [1, 1, 1280], getitem_76, getitem_77, torch.int8);  layer_norm_12 = None
	        dequantize_affine_76: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_38, [1, 1, 1280], getitem_76, getitem_77, torch.int8);  quantize_affine_38 = getitem_76 = getitem_77 = None
	        dequantize_affine_77: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_76, dequantize_affine_77, model_audio_tower_layers_6_self_attn_v_proj_bias);  dequantize_affine_76 = dequantize_affine_77 = model_audio_tower_layers_6_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_20: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_38, [sym_size_int_2, -1, 20, 64]);  linear_38 = None
	        transpose_26: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_20, 1, 2);  view_20 = None
	        contiguous_26: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_26);  transpose_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_6: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_24, contiguous_25, contiguous_26, scale = 1.0);  contiguous_24 = contiguous_25 = contiguous_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_27: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_6, 1, 2);  scaled_dot_product_attention_6 = None
	        contiguous_27: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_27);  transpose_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_27, [sym_size_int_2, 1500, -1]);  contiguous_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_39 = torch.ops.torchao.choose_qparams_affine.default(reshape_6, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_78: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_39[0]
	        getitem_79: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_39[1];  choose_qparams_affine_default_39 = None
	        quantize_affine_39: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_6, [1, 1, 1280], getitem_78, getitem_79, torch.int8);  reshape_6 = None
	        dequantize_affine_78: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_39, [1, 1, 1280], getitem_78, getitem_79, torch.int8);  quantize_affine_39 = getitem_78 = getitem_79 = None
	        dequantize_affine_79: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_39: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_78, dequantize_affine_79, model_audio_tower_layers_6_self_attn_out_proj_bias);  dequantize_affine_78 = dequantize_affine_79 = model_audio_tower_layers_6_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_19: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_39, 0.0, False);  linear_39 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_93: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_84, dropout_19);  add_84 = dropout_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_13: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_93, [1280], model_audio_tower_layers_6_final_layer_norm_weight, model_audio_tower_layers_6_final_layer_norm_bias);  model_audio_tower_layers_6_final_layer_norm_weight = model_audio_tower_layers_6_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_40 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_13, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_80: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_40[0]
	        getitem_81: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_40[1];  choose_qparams_affine_default_40 = None
	        quantize_affine_40: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_13, [1, 1, 1280], getitem_80, getitem_81, torch.int8);  layer_norm_13 = None
	        dequantize_affine_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_40, [1, 1, 1280], getitem_80, getitem_81, torch.int8);  quantize_affine_40 = getitem_80 = getitem_81 = None
	        dequantize_affine_81: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_6_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_6_fc1_parametrizations_weight_original1, model_audio_tower_layers_6_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_6_fc1_parametrizations_weight_original0 = model_audio_tower_layers_6_fc1_parametrizations_weight_original1 = model_audio_tower_layers_6_fc1_parametrizations_weight_original2 = None
	        linear_40: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_80, dequantize_affine_81, model_audio_tower_layers_6_fc1_bias);  dequantize_affine_80 = dequantize_affine_81 = model_audio_tower_layers_6_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_8: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_40);  linear_40 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_20: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_8, 0.0, False);  gelu_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_41 = torch.ops.torchao.choose_qparams_affine.default(dropout_20, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_82: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_41[0]
	        getitem_83: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_41[1];  choose_qparams_affine_default_41 = None
	        quantize_affine_41: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_20, [1, 1, 5120], getitem_82, getitem_83, torch.int8);  dropout_20 = None
	        dequantize_affine_82: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_41, [1, 1, 5120], getitem_82, getitem_83, torch.int8);  quantize_affine_41 = getitem_82 = getitem_83 = None
	        dequantize_affine_83: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_6_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_6_fc2_parametrizations_weight_original1, model_audio_tower_layers_6_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_6_fc2_parametrizations_weight_original0 = model_audio_tower_layers_6_fc2_parametrizations_weight_original1 = model_audio_tower_layers_6_fc2_parametrizations_weight_original2 = None
	        linear_41: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_82, dequantize_affine_83, model_audio_tower_layers_6_fc2_bias);  dequantize_affine_82 = dequantize_affine_83 = model_audio_tower_layers_6_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_41, 0.0, False);  linear_41 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_98: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_93, dropout_21);  add_93 = dropout_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_98, [1280], model_audio_tower_layers_7_self_attn_layer_norm_weight, model_audio_tower_layers_7_self_attn_layer_norm_bias);  model_audio_tower_layers_7_self_attn_layer_norm_weight = model_audio_tower_layers_7_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_42 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_14, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_84: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_42[0]
	        getitem_85: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_42[1];  choose_qparams_affine_default_42 = None
	        quantize_affine_42: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_14, [1, 1, 1280], getitem_84, getitem_85, torch.int8)
	        dequantize_affine_84: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_42, [1, 1, 1280], getitem_84, getitem_85, torch.int8);  quantize_affine_42 = getitem_84 = getitem_85 = None
	        dequantize_affine_85: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_42: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_84, dequantize_affine_85, model_audio_tower_layers_7_self_attn_q_proj_bias);  dequantize_affine_84 = dequantize_affine_85 = model_audio_tower_layers_7_self_attn_q_proj_bias = None
	        mul_264: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_42, 0.125);  linear_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_21: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_264, [sym_size_int_2, 1500, 20, 64]);  mul_264 = None
	        transpose_28: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_21, 1, 2);  view_21 = None
	        contiguous_28: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_28);  transpose_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_43 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_14, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_86: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_43[0]
	        getitem_87: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_43[1];  choose_qparams_affine_default_43 = None
	        quantize_affine_43: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_14, [1, 1, 1280], getitem_86, getitem_87, torch.int8)
	        dequantize_affine_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_43, [1, 1, 1280], getitem_86, getitem_87, torch.int8);  quantize_affine_43 = getitem_86 = getitem_87 = None
	        dequantize_affine_87: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_43: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_86, dequantize_affine_87);  dequantize_affine_86 = dequantize_affine_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_22: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_43, [sym_size_int_2, -1, 20, 64]);  linear_43 = None
	        transpose_29: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_22, 1, 2);  view_22 = None
	        contiguous_29: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_29);  transpose_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_44 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_14, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_88: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_44[0]
	        getitem_89: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_44[1];  choose_qparams_affine_default_44 = None
	        quantize_affine_44: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_14, [1, 1, 1280], getitem_88, getitem_89, torch.int8);  layer_norm_14 = None
	        dequantize_affine_88: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_44, [1, 1, 1280], getitem_88, getitem_89, torch.int8);  quantize_affine_44 = getitem_88 = getitem_89 = None
	        dequantize_affine_89: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_88, dequantize_affine_89, model_audio_tower_layers_7_self_attn_v_proj_bias);  dequantize_affine_88 = dequantize_affine_89 = model_audio_tower_layers_7_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_23: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_44, [sym_size_int_2, -1, 20, 64]);  linear_44 = None
	        transpose_30: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_23, 1, 2);  view_23 = None
	        contiguous_30: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_30);  transpose_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_7: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_28, contiguous_29, contiguous_30, scale = 1.0);  contiguous_28 = contiguous_29 = contiguous_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_31: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_7, 1, 2);  scaled_dot_product_attention_7 = None
	        contiguous_31: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_31);  transpose_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_7: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_31, [sym_size_int_2, 1500, -1]);  contiguous_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_45 = torch.ops.torchao.choose_qparams_affine.default(reshape_7, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_90: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_45[0]
	        getitem_91: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_45[1];  choose_qparams_affine_default_45 = None
	        quantize_affine_45: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_7, [1, 1, 1280], getitem_90, getitem_91, torch.int8);  reshape_7 = None
	        dequantize_affine_90: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_45, [1, 1, 1280], getitem_90, getitem_91, torch.int8);  quantize_affine_45 = getitem_90 = getitem_91 = None
	        dequantize_affine_91: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_45: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_90, dequantize_affine_91, model_audio_tower_layers_7_self_attn_out_proj_bias);  dequantize_affine_90 = dequantize_affine_91 = model_audio_tower_layers_7_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_22: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_45, 0.0, False);  linear_45 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_107: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_98, dropout_22);  add_98 = dropout_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_15: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_107, [1280], model_audio_tower_layers_7_final_layer_norm_weight, model_audio_tower_layers_7_final_layer_norm_bias);  model_audio_tower_layers_7_final_layer_norm_weight = model_audio_tower_layers_7_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_46 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_15, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_92: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_46[0]
	        getitem_93: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_46[1];  choose_qparams_affine_default_46 = None
	        quantize_affine_46: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_15, [1, 1, 1280], getitem_92, getitem_93, torch.int8);  layer_norm_15 = None
	        dequantize_affine_92: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_46, [1, 1, 1280], getitem_92, getitem_93, torch.int8);  quantize_affine_46 = getitem_92 = getitem_93 = None
	        dequantize_affine_93: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_7_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_7_fc1_parametrizations_weight_original1, model_audio_tower_layers_7_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_7_fc1_parametrizations_weight_original0 = model_audio_tower_layers_7_fc1_parametrizations_weight_original1 = model_audio_tower_layers_7_fc1_parametrizations_weight_original2 = None
	        linear_46: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_92, dequantize_affine_93, model_audio_tower_layers_7_fc1_bias);  dequantize_affine_92 = dequantize_affine_93 = model_audio_tower_layers_7_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_9: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_46);  linear_46 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_23: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_9, 0.0, False);  gelu_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_47 = torch.ops.torchao.choose_qparams_affine.default(dropout_23, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_94: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_47[0]
	        getitem_95: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_47[1];  choose_qparams_affine_default_47 = None
	        quantize_affine_47: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_23, [1, 1, 5120], getitem_94, getitem_95, torch.int8);  dropout_23 = None
	        dequantize_affine_94: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_47, [1, 1, 5120], getitem_94, getitem_95, torch.int8);  quantize_affine_47 = getitem_94 = getitem_95 = None
	        dequantize_affine_95: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_7_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_7_fc2_parametrizations_weight_original1, model_audio_tower_layers_7_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_7_fc2_parametrizations_weight_original0 = model_audio_tower_layers_7_fc2_parametrizations_weight_original1 = model_audio_tower_layers_7_fc2_parametrizations_weight_original2 = None
	        linear_47: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_94, dequantize_affine_95, model_audio_tower_layers_7_fc2_bias);  dequantize_affine_94 = dequantize_affine_95 = model_audio_tower_layers_7_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_24: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_47, 0.0, False);  linear_47 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_112: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_107, dropout_24);  add_107 = dropout_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_16: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_112, [1280], model_audio_tower_layers_8_self_attn_layer_norm_weight, model_audio_tower_layers_8_self_attn_layer_norm_bias);  model_audio_tower_layers_8_self_attn_layer_norm_weight = model_audio_tower_layers_8_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_48 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_16, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_96: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_48[0]
	        getitem_97: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_48[1];  choose_qparams_affine_default_48 = None
	        quantize_affine_48: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_16, [1, 1, 1280], getitem_96, getitem_97, torch.int8)
	        dequantize_affine_96: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_48, [1, 1, 1280], getitem_96, getitem_97, torch.int8);  quantize_affine_48 = getitem_96 = getitem_97 = None
	        dequantize_affine_97: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_48: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_96, dequantize_affine_97, model_audio_tower_layers_8_self_attn_q_proj_bias);  dequantize_affine_96 = dequantize_affine_97 = model_audio_tower_layers_8_self_attn_q_proj_bias = None
	        mul_301: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_48, 0.125);  linear_48 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_24: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_301, [sym_size_int_2, 1500, 20, 64]);  mul_301 = None
	        transpose_32: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_24, 1, 2);  view_24 = None
	        contiguous_32: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_32);  transpose_32 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_49 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_16, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_98: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_49[0]
	        getitem_99: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_49[1];  choose_qparams_affine_default_49 = None
	        quantize_affine_49: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_16, [1, 1, 1280], getitem_98, getitem_99, torch.int8)
	        dequantize_affine_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_49, [1, 1, 1280], getitem_98, getitem_99, torch.int8);  quantize_affine_49 = getitem_98 = getitem_99 = None
	        dequantize_affine_99: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_49: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_98, dequantize_affine_99);  dequantize_affine_98 = dequantize_affine_99 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_25: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_49, [sym_size_int_2, -1, 20, 64]);  linear_49 = None
	        transpose_33: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_25, 1, 2);  view_25 = None
	        contiguous_33: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_33);  transpose_33 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_50 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_16, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_100: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_50[0]
	        getitem_101: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_50[1];  choose_qparams_affine_default_50 = None
	        quantize_affine_50: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_16, [1, 1, 1280], getitem_100, getitem_101, torch.int8);  layer_norm_16 = None
	        dequantize_affine_100: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_50, [1, 1, 1280], getitem_100, getitem_101, torch.int8);  quantize_affine_50 = getitem_100 = getitem_101 = None
	        dequantize_affine_101: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_100, dequantize_affine_101, model_audio_tower_layers_8_self_attn_v_proj_bias);  dequantize_affine_100 = dequantize_affine_101 = model_audio_tower_layers_8_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_26: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_50, [sym_size_int_2, -1, 20, 64]);  linear_50 = None
	        transpose_34: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_26, 1, 2);  view_26 = None
	        contiguous_34: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_34);  transpose_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_8: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_32, contiguous_33, contiguous_34, scale = 1.0);  contiguous_32 = contiguous_33 = contiguous_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_35: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_8, 1, 2);  scaled_dot_product_attention_8 = None
	        contiguous_35: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_35);  transpose_35 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_35, [sym_size_int_2, 1500, -1]);  contiguous_35 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_51 = torch.ops.torchao.choose_qparams_affine.default(reshape_8, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_102: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_51[0]
	        getitem_103: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_51[1];  choose_qparams_affine_default_51 = None
	        quantize_affine_51: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_8, [1, 1, 1280], getitem_102, getitem_103, torch.int8);  reshape_8 = None
	        dequantize_affine_102: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_51, [1, 1, 1280], getitem_102, getitem_103, torch.int8);  quantize_affine_51 = getitem_102 = getitem_103 = None
	        dequantize_affine_103: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_51: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_102, dequantize_affine_103, model_audio_tower_layers_8_self_attn_out_proj_bias);  dequantize_affine_102 = dequantize_affine_103 = model_audio_tower_layers_8_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_25: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_51, 0.0, False);  linear_51 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_121: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_112, dropout_25);  add_112 = dropout_25 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_17: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_121, [1280], model_audio_tower_layers_8_final_layer_norm_weight, model_audio_tower_layers_8_final_layer_norm_bias);  model_audio_tower_layers_8_final_layer_norm_weight = model_audio_tower_layers_8_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_52 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_17, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_104: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_52[0]
	        getitem_105: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_52[1];  choose_qparams_affine_default_52 = None
	        quantize_affine_52: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_17, [1, 1, 1280], getitem_104, getitem_105, torch.int8);  layer_norm_17 = None
	        dequantize_affine_104: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_52, [1, 1, 1280], getitem_104, getitem_105, torch.int8);  quantize_affine_52 = getitem_104 = getitem_105 = None
	        dequantize_affine_105: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_8_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_8_fc1_parametrizations_weight_original1, model_audio_tower_layers_8_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_8_fc1_parametrizations_weight_original0 = model_audio_tower_layers_8_fc1_parametrizations_weight_original1 = model_audio_tower_layers_8_fc1_parametrizations_weight_original2 = None
	        linear_52: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_104, dequantize_affine_105, model_audio_tower_layers_8_fc1_bias);  dequantize_affine_104 = dequantize_affine_105 = model_audio_tower_layers_8_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_10: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_52);  linear_52 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_26: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_10, 0.0, False);  gelu_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_53 = torch.ops.torchao.choose_qparams_affine.default(dropout_26, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_106: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_53[0]
	        getitem_107: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_53[1];  choose_qparams_affine_default_53 = None
	        quantize_affine_53: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_26, [1, 1, 5120], getitem_106, getitem_107, torch.int8);  dropout_26 = None
	        dequantize_affine_106: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_53, [1, 1, 5120], getitem_106, getitem_107, torch.int8);  quantize_affine_53 = getitem_106 = getitem_107 = None
	        dequantize_affine_107: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_8_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_8_fc2_parametrizations_weight_original1, model_audio_tower_layers_8_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_8_fc2_parametrizations_weight_original0 = model_audio_tower_layers_8_fc2_parametrizations_weight_original1 = model_audio_tower_layers_8_fc2_parametrizations_weight_original2 = None
	        linear_53: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_106, dequantize_affine_107, model_audio_tower_layers_8_fc2_bias);  dequantize_affine_106 = dequantize_affine_107 = model_audio_tower_layers_8_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_27: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_53, 0.0, False);  linear_53 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_126: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_121, dropout_27);  add_121 = dropout_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_18: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_126, [1280], model_audio_tower_layers_9_self_attn_layer_norm_weight, model_audio_tower_layers_9_self_attn_layer_norm_bias);  model_audio_tower_layers_9_self_attn_layer_norm_weight = model_audio_tower_layers_9_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_54 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_18, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_108: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_54[0]
	        getitem_109: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_54[1];  choose_qparams_affine_default_54 = None
	        quantize_affine_54: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_18, [1, 1, 1280], getitem_108, getitem_109, torch.int8)
	        dequantize_affine_108: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_54, [1, 1, 1280], getitem_108, getitem_109, torch.int8);  quantize_affine_54 = getitem_108 = getitem_109 = None
	        dequantize_affine_109: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_54: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_108, dequantize_affine_109, model_audio_tower_layers_9_self_attn_q_proj_bias);  dequantize_affine_108 = dequantize_affine_109 = model_audio_tower_layers_9_self_attn_q_proj_bias = None
	        mul_338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_54, 0.125);  linear_54 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_27: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_338, [sym_size_int_2, 1500, 20, 64]);  mul_338 = None
	        transpose_36: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_27, 1, 2);  view_27 = None
	        contiguous_36: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_36);  transpose_36 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_55 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_18, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_110: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_55[0]
	        getitem_111: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_55[1];  choose_qparams_affine_default_55 = None
	        quantize_affine_55: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_18, [1, 1, 1280], getitem_110, getitem_111, torch.int8)
	        dequantize_affine_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_55, [1, 1, 1280], getitem_110, getitem_111, torch.int8);  quantize_affine_55 = getitem_110 = getitem_111 = None
	        dequantize_affine_111: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_55: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_110, dequantize_affine_111);  dequantize_affine_110 = dequantize_affine_111 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_28: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_55, [sym_size_int_2, -1, 20, 64]);  linear_55 = None
	        transpose_37: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_28, 1, 2);  view_28 = None
	        contiguous_37: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_37);  transpose_37 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_56 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_18, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_112: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_56[0]
	        getitem_113: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_56[1];  choose_qparams_affine_default_56 = None
	        quantize_affine_56: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_18, [1, 1, 1280], getitem_112, getitem_113, torch.int8);  layer_norm_18 = None
	        dequantize_affine_112: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_56, [1, 1, 1280], getitem_112, getitem_113, torch.int8);  quantize_affine_56 = getitem_112 = getitem_113 = None
	        dequantize_affine_113: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_112, dequantize_affine_113, model_audio_tower_layers_9_self_attn_v_proj_bias);  dequantize_affine_112 = dequantize_affine_113 = model_audio_tower_layers_9_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_29: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_56, [sym_size_int_2, -1, 20, 64]);  linear_56 = None
	        transpose_38: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_29, 1, 2);  view_29 = None
	        contiguous_38: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_38);  transpose_38 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_9: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_36, contiguous_37, contiguous_38, scale = 1.0);  contiguous_36 = contiguous_37 = contiguous_38 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_39: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_9, 1, 2);  scaled_dot_product_attention_9 = None
	        contiguous_39: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_39);  transpose_39 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_9: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_39, [sym_size_int_2, 1500, -1]);  contiguous_39 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_57 = torch.ops.torchao.choose_qparams_affine.default(reshape_9, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_114: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_57[0]
	        getitem_115: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_57[1];  choose_qparams_affine_default_57 = None
	        quantize_affine_57: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_9, [1, 1, 1280], getitem_114, getitem_115, torch.int8);  reshape_9 = None
	        dequantize_affine_114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_57, [1, 1, 1280], getitem_114, getitem_115, torch.int8);  quantize_affine_57 = getitem_114 = getitem_115 = None
	        dequantize_affine_115: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_57: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_114, dequantize_affine_115, model_audio_tower_layers_9_self_attn_out_proj_bias);  dequantize_affine_114 = dequantize_affine_115 = model_audio_tower_layers_9_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_28: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_57, 0.0, False);  linear_57 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_135: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_126, dropout_28);  add_126 = dropout_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_19: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_135, [1280], model_audio_tower_layers_9_final_layer_norm_weight, model_audio_tower_layers_9_final_layer_norm_bias);  model_audio_tower_layers_9_final_layer_norm_weight = model_audio_tower_layers_9_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_58 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_19, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_116: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_58[0]
	        getitem_117: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_58[1];  choose_qparams_affine_default_58 = None
	        quantize_affine_58: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_19, [1, 1, 1280], getitem_116, getitem_117, torch.int8);  layer_norm_19 = None
	        dequantize_affine_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_58, [1, 1, 1280], getitem_116, getitem_117, torch.int8);  quantize_affine_58 = getitem_116 = getitem_117 = None
	        dequantize_affine_117: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_9_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_9_fc1_parametrizations_weight_original1, model_audio_tower_layers_9_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_9_fc1_parametrizations_weight_original0 = model_audio_tower_layers_9_fc1_parametrizations_weight_original1 = model_audio_tower_layers_9_fc1_parametrizations_weight_original2 = None
	        linear_58: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_116, dequantize_affine_117, model_audio_tower_layers_9_fc1_bias);  dequantize_affine_116 = dequantize_affine_117 = model_audio_tower_layers_9_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_11: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_58);  linear_58 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_29: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_11, 0.0, False);  gelu_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_59 = torch.ops.torchao.choose_qparams_affine.default(dropout_29, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_118: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_59[0]
	        getitem_119: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_59[1];  choose_qparams_affine_default_59 = None
	        quantize_affine_59: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_29, [1, 1, 5120], getitem_118, getitem_119, torch.int8);  dropout_29 = None
	        dequantize_affine_118: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_59, [1, 1, 5120], getitem_118, getitem_119, torch.int8);  quantize_affine_59 = getitem_118 = getitem_119 = None
	        dequantize_affine_119: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_9_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_9_fc2_parametrizations_weight_original1, model_audio_tower_layers_9_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_9_fc2_parametrizations_weight_original0 = model_audio_tower_layers_9_fc2_parametrizations_weight_original1 = model_audio_tower_layers_9_fc2_parametrizations_weight_original2 = None
	        linear_59: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_118, dequantize_affine_119, model_audio_tower_layers_9_fc2_bias);  dequantize_affine_118 = dequantize_affine_119 = model_audio_tower_layers_9_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_59, 0.0, False);  linear_59 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_140: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_135, dropout_30);  add_135 = dropout_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_140, [1280], model_audio_tower_layers_10_self_attn_layer_norm_weight, model_audio_tower_layers_10_self_attn_layer_norm_bias);  model_audio_tower_layers_10_self_attn_layer_norm_weight = model_audio_tower_layers_10_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_60 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_20, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_120: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_60[0]
	        getitem_121: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_60[1];  choose_qparams_affine_default_60 = None
	        quantize_affine_60: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_20, [1, 1, 1280], getitem_120, getitem_121, torch.int8)
	        dequantize_affine_120: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_60, [1, 1, 1280], getitem_120, getitem_121, torch.int8);  quantize_affine_60 = getitem_120 = getitem_121 = None
	        dequantize_affine_121: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_60: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_120, dequantize_affine_121, model_audio_tower_layers_10_self_attn_q_proj_bias);  dequantize_affine_120 = dequantize_affine_121 = model_audio_tower_layers_10_self_attn_q_proj_bias = None
	        mul_375: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_60, 0.125);  linear_60 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_30: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_375, [sym_size_int_2, 1500, 20, 64]);  mul_375 = None
	        transpose_40: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_30, 1, 2);  view_30 = None
	        contiguous_40: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_40);  transpose_40 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_61 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_20, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_122: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_61[0]
	        getitem_123: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_61[1];  choose_qparams_affine_default_61 = None
	        quantize_affine_61: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_20, [1, 1, 1280], getitem_122, getitem_123, torch.int8)
	        dequantize_affine_122: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_61, [1, 1, 1280], getitem_122, getitem_123, torch.int8);  quantize_affine_61 = getitem_122 = getitem_123 = None
	        dequantize_affine_123: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_61: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_122, dequantize_affine_123);  dequantize_affine_122 = dequantize_affine_123 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_31: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_61, [sym_size_int_2, -1, 20, 64]);  linear_61 = None
	        transpose_41: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_31, 1, 2);  view_31 = None
	        contiguous_41: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_41);  transpose_41 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_62 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_20, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_124: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_62[0]
	        getitem_125: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_62[1];  choose_qparams_affine_default_62 = None
	        quantize_affine_62: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_20, [1, 1, 1280], getitem_124, getitem_125, torch.int8);  layer_norm_20 = None
	        dequantize_affine_124: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_62, [1, 1, 1280], getitem_124, getitem_125, torch.int8);  quantize_affine_62 = getitem_124 = getitem_125 = None
	        dequantize_affine_125: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_124, dequantize_affine_125, model_audio_tower_layers_10_self_attn_v_proj_bias);  dequantize_affine_124 = dequantize_affine_125 = model_audio_tower_layers_10_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_32: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_62, [sym_size_int_2, -1, 20, 64]);  linear_62 = None
	        transpose_42: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_32, 1, 2);  view_32 = None
	        contiguous_42: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_42);  transpose_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_10: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_40, contiguous_41, contiguous_42, scale = 1.0);  contiguous_40 = contiguous_41 = contiguous_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_43: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_10, 1, 2);  scaled_dot_product_attention_10 = None
	        contiguous_43: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_43);  transpose_43 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_10: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_43, [sym_size_int_2, 1500, -1]);  contiguous_43 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_63 = torch.ops.torchao.choose_qparams_affine.default(reshape_10, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_126: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_63[0]
	        getitem_127: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_63[1];  choose_qparams_affine_default_63 = None
	        quantize_affine_63: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_10, [1, 1, 1280], getitem_126, getitem_127, torch.int8);  reshape_10 = None
	        dequantize_affine_126: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_63, [1, 1, 1280], getitem_126, getitem_127, torch.int8);  quantize_affine_63 = getitem_126 = getitem_127 = None
	        dequantize_affine_127: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_63: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_126, dequantize_affine_127, model_audio_tower_layers_10_self_attn_out_proj_bias);  dequantize_affine_126 = dequantize_affine_127 = model_audio_tower_layers_10_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_31: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_63, 0.0, False);  linear_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_149: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_140, dropout_31);  add_140 = dropout_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_149, [1280], model_audio_tower_layers_10_final_layer_norm_weight, model_audio_tower_layers_10_final_layer_norm_bias);  model_audio_tower_layers_10_final_layer_norm_weight = model_audio_tower_layers_10_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_64 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_21, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_128: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_64[0]
	        getitem_129: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_64[1];  choose_qparams_affine_default_64 = None
	        quantize_affine_64: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_21, [1, 1, 1280], getitem_128, getitem_129, torch.int8);  layer_norm_21 = None
	        dequantize_affine_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_64, [1, 1, 1280], getitem_128, getitem_129, torch.int8);  quantize_affine_64 = getitem_128 = getitem_129 = None
	        dequantize_affine_129: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_10_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_10_fc1_parametrizations_weight_original1, model_audio_tower_layers_10_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_10_fc1_parametrizations_weight_original0 = model_audio_tower_layers_10_fc1_parametrizations_weight_original1 = model_audio_tower_layers_10_fc1_parametrizations_weight_original2 = None
	        linear_64: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_128, dequantize_affine_129, model_audio_tower_layers_10_fc1_bias);  dequantize_affine_128 = dequantize_affine_129 = model_audio_tower_layers_10_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_12: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_64);  linear_64 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_32: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_12, 0.0, False);  gelu_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_65 = torch.ops.torchao.choose_qparams_affine.default(dropout_32, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_130: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_65[0]
	        getitem_131: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_65[1];  choose_qparams_affine_default_65 = None
	        quantize_affine_65: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_32, [1, 1, 5120], getitem_130, getitem_131, torch.int8);  dropout_32 = None
	        dequantize_affine_130: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_65, [1, 1, 5120], getitem_130, getitem_131, torch.int8);  quantize_affine_65 = getitem_130 = getitem_131 = None
	        dequantize_affine_131: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_10_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_10_fc2_parametrizations_weight_original1, model_audio_tower_layers_10_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_10_fc2_parametrizations_weight_original0 = model_audio_tower_layers_10_fc2_parametrizations_weight_original1 = model_audio_tower_layers_10_fc2_parametrizations_weight_original2 = None
	        linear_65: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_130, dequantize_affine_131, model_audio_tower_layers_10_fc2_bias);  dequantize_affine_130 = dequantize_affine_131 = model_audio_tower_layers_10_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_33: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_65, 0.0, False);  linear_65 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_154: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_149, dropout_33);  add_149 = dropout_33 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_22: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_154, [1280], model_audio_tower_layers_11_self_attn_layer_norm_weight, model_audio_tower_layers_11_self_attn_layer_norm_bias);  model_audio_tower_layers_11_self_attn_layer_norm_weight = model_audio_tower_layers_11_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_66 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_22, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_132: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_66[0]
	        getitem_133: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_66[1];  choose_qparams_affine_default_66 = None
	        quantize_affine_66: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_22, [1, 1, 1280], getitem_132, getitem_133, torch.int8)
	        dequantize_affine_132: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_66, [1, 1, 1280], getitem_132, getitem_133, torch.int8);  quantize_affine_66 = getitem_132 = getitem_133 = None
	        dequantize_affine_133: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_66: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_132, dequantize_affine_133, model_audio_tower_layers_11_self_attn_q_proj_bias);  dequantize_affine_132 = dequantize_affine_133 = model_audio_tower_layers_11_self_attn_q_proj_bias = None
	        mul_412: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_66, 0.125);  linear_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_33: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_412, [sym_size_int_2, 1500, 20, 64]);  mul_412 = None
	        transpose_44: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_33, 1, 2);  view_33 = None
	        contiguous_44: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_44);  transpose_44 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_67 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_22, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_134: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_67[0]
	        getitem_135: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_67[1];  choose_qparams_affine_default_67 = None
	        quantize_affine_67: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_22, [1, 1, 1280], getitem_134, getitem_135, torch.int8)
	        dequantize_affine_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_67, [1, 1, 1280], getitem_134, getitem_135, torch.int8);  quantize_affine_67 = getitem_134 = getitem_135 = None
	        dequantize_affine_135: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_67: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_134, dequantize_affine_135);  dequantize_affine_134 = dequantize_affine_135 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_34: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_67, [sym_size_int_2, -1, 20, 64]);  linear_67 = None
	        transpose_45: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_34, 1, 2);  view_34 = None
	        contiguous_45: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_45);  transpose_45 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_68 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_22, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_136: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_68[0]
	        getitem_137: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_68[1];  choose_qparams_affine_default_68 = None
	        quantize_affine_68: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_22, [1, 1, 1280], getitem_136, getitem_137, torch.int8);  layer_norm_22 = None
	        dequantize_affine_136: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_68, [1, 1, 1280], getitem_136, getitem_137, torch.int8);  quantize_affine_68 = getitem_136 = getitem_137 = None
	        dequantize_affine_137: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_68: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_136, dequantize_affine_137, model_audio_tower_layers_11_self_attn_v_proj_bias);  dequantize_affine_136 = dequantize_affine_137 = model_audio_tower_layers_11_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_35: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_68, [sym_size_int_2, -1, 20, 64]);  linear_68 = None
	        transpose_46: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_35, 1, 2);  view_35 = None
	        contiguous_46: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_46);  transpose_46 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_11: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_44, contiguous_45, contiguous_46, scale = 1.0);  contiguous_44 = contiguous_45 = contiguous_46 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_47: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_11, 1, 2);  scaled_dot_product_attention_11 = None
	        contiguous_47: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_47);  transpose_47 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_11: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_47, [sym_size_int_2, 1500, -1]);  contiguous_47 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_69 = torch.ops.torchao.choose_qparams_affine.default(reshape_11, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_138: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_69[0]
	        getitem_139: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_69[1];  choose_qparams_affine_default_69 = None
	        quantize_affine_69: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_11, [1, 1, 1280], getitem_138, getitem_139, torch.int8);  reshape_11 = None
	        dequantize_affine_138: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_69, [1, 1, 1280], getitem_138, getitem_139, torch.int8);  quantize_affine_69 = getitem_138 = getitem_139 = None
	        dequantize_affine_139: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_69: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_138, dequantize_affine_139, model_audio_tower_layers_11_self_attn_out_proj_bias);  dequantize_affine_138 = dequantize_affine_139 = model_audio_tower_layers_11_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_34: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_69, 0.0, False);  linear_69 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_163: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_154, dropout_34);  add_154 = dropout_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_23: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_163, [1280], model_audio_tower_layers_11_final_layer_norm_weight, model_audio_tower_layers_11_final_layer_norm_bias);  model_audio_tower_layers_11_final_layer_norm_weight = model_audio_tower_layers_11_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_70 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_23, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_140: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_70[0]
	        getitem_141: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_70[1];  choose_qparams_affine_default_70 = None
	        quantize_affine_70: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_23, [1, 1, 1280], getitem_140, getitem_141, torch.int8);  layer_norm_23 = None
	        dequantize_affine_140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_70, [1, 1, 1280], getitem_140, getitem_141, torch.int8);  quantize_affine_70 = getitem_140 = getitem_141 = None
	        dequantize_affine_141: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_11_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_11_fc1_parametrizations_weight_original1, model_audio_tower_layers_11_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_11_fc1_parametrizations_weight_original0 = model_audio_tower_layers_11_fc1_parametrizations_weight_original1 = model_audio_tower_layers_11_fc1_parametrizations_weight_original2 = None
	        linear_70: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_140, dequantize_affine_141, model_audio_tower_layers_11_fc1_bias);  dequantize_affine_140 = dequantize_affine_141 = model_audio_tower_layers_11_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_13: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_70);  linear_70 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_35: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_13, 0.0, False);  gelu_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_71 = torch.ops.torchao.choose_qparams_affine.default(dropout_35, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_142: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_71[0]
	        getitem_143: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_71[1];  choose_qparams_affine_default_71 = None
	        quantize_affine_71: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_35, [1, 1, 5120], getitem_142, getitem_143, torch.int8);  dropout_35 = None
	        dequantize_affine_142: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_71, [1, 1, 5120], getitem_142, getitem_143, torch.int8);  quantize_affine_71 = getitem_142 = getitem_143 = None
	        dequantize_affine_143: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_11_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_11_fc2_parametrizations_weight_original1, model_audio_tower_layers_11_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_11_fc2_parametrizations_weight_original0 = model_audio_tower_layers_11_fc2_parametrizations_weight_original1 = model_audio_tower_layers_11_fc2_parametrizations_weight_original2 = None
	        linear_71: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_142, dequantize_affine_143, model_audio_tower_layers_11_fc2_bias);  dequantize_affine_142 = dequantize_affine_143 = model_audio_tower_layers_11_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_36: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_71, 0.0, False);  linear_71 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_168: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_163, dropout_36);  add_163 = dropout_36 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_24: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_168, [1280], model_audio_tower_layers_12_self_attn_layer_norm_weight, model_audio_tower_layers_12_self_attn_layer_norm_bias);  model_audio_tower_layers_12_self_attn_layer_norm_weight = model_audio_tower_layers_12_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_72 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_24, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_144: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_72[0]
	        getitem_145: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_72[1];  choose_qparams_affine_default_72 = None
	        quantize_affine_72: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_24, [1, 1, 1280], getitem_144, getitem_145, torch.int8)
	        dequantize_affine_144: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_72, [1, 1, 1280], getitem_144, getitem_145, torch.int8);  quantize_affine_72 = getitem_144 = getitem_145 = None
	        dequantize_affine_145: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_72: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_144, dequantize_affine_145, model_audio_tower_layers_12_self_attn_q_proj_bias);  dequantize_affine_144 = dequantize_affine_145 = model_audio_tower_layers_12_self_attn_q_proj_bias = None
	        mul_449: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_72, 0.125);  linear_72 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_36: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_449, [sym_size_int_2, 1500, 20, 64]);  mul_449 = None
	        transpose_48: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_36, 1, 2);  view_36 = None
	        contiguous_48: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_48);  transpose_48 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_73 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_24, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_146: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_73[0]
	        getitem_147: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_73[1];  choose_qparams_affine_default_73 = None
	        quantize_affine_73: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_24, [1, 1, 1280], getitem_146, getitem_147, torch.int8)
	        dequantize_affine_146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_73, [1, 1, 1280], getitem_146, getitem_147, torch.int8);  quantize_affine_73 = getitem_146 = getitem_147 = None
	        dequantize_affine_147: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_73: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_146, dequantize_affine_147);  dequantize_affine_146 = dequantize_affine_147 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_37: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_73, [sym_size_int_2, -1, 20, 64]);  linear_73 = None
	        transpose_49: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_37, 1, 2);  view_37 = None
	        contiguous_49: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_49);  transpose_49 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_74 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_24, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_148: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_74[0]
	        getitem_149: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_74[1];  choose_qparams_affine_default_74 = None
	        quantize_affine_74: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_24, [1, 1, 1280], getitem_148, getitem_149, torch.int8);  layer_norm_24 = None
	        dequantize_affine_148: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_74, [1, 1, 1280], getitem_148, getitem_149, torch.int8);  quantize_affine_74 = getitem_148 = getitem_149 = None
	        dequantize_affine_149: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_74: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_148, dequantize_affine_149, model_audio_tower_layers_12_self_attn_v_proj_bias);  dequantize_affine_148 = dequantize_affine_149 = model_audio_tower_layers_12_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_38: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_74, [sym_size_int_2, -1, 20, 64]);  linear_74 = None
	        transpose_50: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_38, 1, 2);  view_38 = None
	        contiguous_50: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_50);  transpose_50 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_12: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_48, contiguous_49, contiguous_50, scale = 1.0);  contiguous_48 = contiguous_49 = contiguous_50 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_51: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_12, 1, 2);  scaled_dot_product_attention_12 = None
	        contiguous_51: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_51);  transpose_51 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_12: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_51, [sym_size_int_2, 1500, -1]);  contiguous_51 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_75 = torch.ops.torchao.choose_qparams_affine.default(reshape_12, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_150: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_75[0]
	        getitem_151: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_75[1];  choose_qparams_affine_default_75 = None
	        quantize_affine_75: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_12, [1, 1, 1280], getitem_150, getitem_151, torch.int8);  reshape_12 = None
	        dequantize_affine_150: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_75, [1, 1, 1280], getitem_150, getitem_151, torch.int8);  quantize_affine_75 = getitem_150 = getitem_151 = None
	        dequantize_affine_151: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_75: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_150, dequantize_affine_151, model_audio_tower_layers_12_self_attn_out_proj_bias);  dequantize_affine_150 = dequantize_affine_151 = model_audio_tower_layers_12_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_37: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_75, 0.0, False);  linear_75 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_177: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_168, dropout_37);  add_168 = dropout_37 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_25: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_177, [1280], model_audio_tower_layers_12_final_layer_norm_weight, model_audio_tower_layers_12_final_layer_norm_bias);  model_audio_tower_layers_12_final_layer_norm_weight = model_audio_tower_layers_12_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_76 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_25, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_152: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_76[0]
	        getitem_153: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_76[1];  choose_qparams_affine_default_76 = None
	        quantize_affine_76: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_25, [1, 1, 1280], getitem_152, getitem_153, torch.int8);  layer_norm_25 = None
	        dequantize_affine_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_76, [1, 1, 1280], getitem_152, getitem_153, torch.int8);  quantize_affine_76 = getitem_152 = getitem_153 = None
	        dequantize_affine_153: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_12_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_12_fc1_parametrizations_weight_original1, model_audio_tower_layers_12_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_12_fc1_parametrizations_weight_original0 = model_audio_tower_layers_12_fc1_parametrizations_weight_original1 = model_audio_tower_layers_12_fc1_parametrizations_weight_original2 = None
	        linear_76: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_152, dequantize_affine_153, model_audio_tower_layers_12_fc1_bias);  dequantize_affine_152 = dequantize_affine_153 = model_audio_tower_layers_12_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_14: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_76);  linear_76 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_38: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_14, 0.0, False);  gelu_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_77 = torch.ops.torchao.choose_qparams_affine.default(dropout_38, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_154: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_77[0]
	        getitem_155: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_77[1];  choose_qparams_affine_default_77 = None
	        quantize_affine_77: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_38, [1, 1, 5120], getitem_154, getitem_155, torch.int8);  dropout_38 = None
	        dequantize_affine_154: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_77, [1, 1, 5120], getitem_154, getitem_155, torch.int8);  quantize_affine_77 = getitem_154 = getitem_155 = None
	        dequantize_affine_155: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_12_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_12_fc2_parametrizations_weight_original1, model_audio_tower_layers_12_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_12_fc2_parametrizations_weight_original0 = model_audio_tower_layers_12_fc2_parametrizations_weight_original1 = model_audio_tower_layers_12_fc2_parametrizations_weight_original2 = None
	        linear_77: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_154, dequantize_affine_155, model_audio_tower_layers_12_fc2_bias);  dequantize_affine_154 = dequantize_affine_155 = model_audio_tower_layers_12_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_39: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_77, 0.0, False);  linear_77 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_182: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_177, dropout_39);  add_177 = dropout_39 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_182, [1280], model_audio_tower_layers_13_self_attn_layer_norm_weight, model_audio_tower_layers_13_self_attn_layer_norm_bias);  model_audio_tower_layers_13_self_attn_layer_norm_weight = model_audio_tower_layers_13_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_78 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_26, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_156: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_78[0]
	        getitem_157: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_78[1];  choose_qparams_affine_default_78 = None
	        quantize_affine_78: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_26, [1, 1, 1280], getitem_156, getitem_157, torch.int8)
	        dequantize_affine_156: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_78, [1, 1, 1280], getitem_156, getitem_157, torch.int8);  quantize_affine_78 = getitem_156 = getitem_157 = None
	        dequantize_affine_157: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_78: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_156, dequantize_affine_157, model_audio_tower_layers_13_self_attn_q_proj_bias);  dequantize_affine_156 = dequantize_affine_157 = model_audio_tower_layers_13_self_attn_q_proj_bias = None
	        mul_486: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_78, 0.125);  linear_78 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_39: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_486, [sym_size_int_2, 1500, 20, 64]);  mul_486 = None
	        transpose_52: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_39, 1, 2);  view_39 = None
	        contiguous_52: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_52);  transpose_52 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_79 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_26, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_158: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_79[0]
	        getitem_159: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_79[1];  choose_qparams_affine_default_79 = None
	        quantize_affine_79: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_26, [1, 1, 1280], getitem_158, getitem_159, torch.int8)
	        dequantize_affine_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_79, [1, 1, 1280], getitem_158, getitem_159, torch.int8);  quantize_affine_79 = getitem_158 = getitem_159 = None
	        dequantize_affine_159: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_79: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_158, dequantize_affine_159);  dequantize_affine_158 = dequantize_affine_159 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_40: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_79, [sym_size_int_2, -1, 20, 64]);  linear_79 = None
	        transpose_53: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_40, 1, 2);  view_40 = None
	        contiguous_53: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_53);  transpose_53 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_80 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_26, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_160: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_80[0]
	        getitem_161: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_80[1];  choose_qparams_affine_default_80 = None
	        quantize_affine_80: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_26, [1, 1, 1280], getitem_160, getitem_161, torch.int8);  layer_norm_26 = None
	        dequantize_affine_160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_80, [1, 1, 1280], getitem_160, getitem_161, torch.int8);  quantize_affine_80 = getitem_160 = getitem_161 = None
	        dequantize_affine_161: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_160, dequantize_affine_161, model_audio_tower_layers_13_self_attn_v_proj_bias);  dequantize_affine_160 = dequantize_affine_161 = model_audio_tower_layers_13_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_41: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_80, [sym_size_int_2, -1, 20, 64]);  linear_80 = None
	        transpose_54: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_41, 1, 2);  view_41 = None
	        contiguous_54: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_54);  transpose_54 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_13: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_52, contiguous_53, contiguous_54, scale = 1.0);  contiguous_52 = contiguous_53 = contiguous_54 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_55: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_13, 1, 2);  scaled_dot_product_attention_13 = None
	        contiguous_55: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_55);  transpose_55 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_13: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_55, [sym_size_int_2, 1500, -1]);  contiguous_55 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_81 = torch.ops.torchao.choose_qparams_affine.default(reshape_13, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_162: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_81[0]
	        getitem_163: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_81[1];  choose_qparams_affine_default_81 = None
	        quantize_affine_81: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_13, [1, 1, 1280], getitem_162, getitem_163, torch.int8);  reshape_13 = None
	        dequantize_affine_162: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_81, [1, 1, 1280], getitem_162, getitem_163, torch.int8);  quantize_affine_81 = getitem_162 = getitem_163 = None
	        dequantize_affine_163: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_81: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_162, dequantize_affine_163, model_audio_tower_layers_13_self_attn_out_proj_bias);  dequantize_affine_162 = dequantize_affine_163 = model_audio_tower_layers_13_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_40: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_81, 0.0, False);  linear_81 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_191: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_182, dropout_40);  add_182 = dropout_40 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_27: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_191, [1280], model_audio_tower_layers_13_final_layer_norm_weight, model_audio_tower_layers_13_final_layer_norm_bias);  model_audio_tower_layers_13_final_layer_norm_weight = model_audio_tower_layers_13_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_82 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_27, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_164: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_82[0]
	        getitem_165: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_82[1];  choose_qparams_affine_default_82 = None
	        quantize_affine_82: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_27, [1, 1, 1280], getitem_164, getitem_165, torch.int8);  layer_norm_27 = None
	        dequantize_affine_164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_82, [1, 1, 1280], getitem_164, getitem_165, torch.int8);  quantize_affine_82 = getitem_164 = getitem_165 = None
	        dequantize_affine_165: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_13_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_13_fc1_parametrizations_weight_original1, model_audio_tower_layers_13_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_13_fc1_parametrizations_weight_original0 = model_audio_tower_layers_13_fc1_parametrizations_weight_original1 = model_audio_tower_layers_13_fc1_parametrizations_weight_original2 = None
	        linear_82: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_164, dequantize_affine_165, model_audio_tower_layers_13_fc1_bias);  dequantize_affine_164 = dequantize_affine_165 = model_audio_tower_layers_13_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_15: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_82);  linear_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_41: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_15, 0.0, False);  gelu_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_83 = torch.ops.torchao.choose_qparams_affine.default(dropout_41, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_166: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_83[0]
	        getitem_167: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_83[1];  choose_qparams_affine_default_83 = None
	        quantize_affine_83: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_41, [1, 1, 5120], getitem_166, getitem_167, torch.int8);  dropout_41 = None
	        dequantize_affine_166: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_83, [1, 1, 5120], getitem_166, getitem_167, torch.int8);  quantize_affine_83 = getitem_166 = getitem_167 = None
	        dequantize_affine_167: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_13_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_13_fc2_parametrizations_weight_original1, model_audio_tower_layers_13_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_13_fc2_parametrizations_weight_original0 = model_audio_tower_layers_13_fc2_parametrizations_weight_original1 = model_audio_tower_layers_13_fc2_parametrizations_weight_original2 = None
	        linear_83: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_166, dequantize_affine_167, model_audio_tower_layers_13_fc2_bias);  dequantize_affine_166 = dequantize_affine_167 = model_audio_tower_layers_13_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_42: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_83, 0.0, False);  linear_83 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_196: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_191, dropout_42);  add_191 = dropout_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_28: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_196, [1280], model_audio_tower_layers_14_self_attn_layer_norm_weight, model_audio_tower_layers_14_self_attn_layer_norm_bias);  model_audio_tower_layers_14_self_attn_layer_norm_weight = model_audio_tower_layers_14_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_84 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_28, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_168: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_84[0]
	        getitem_169: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_84[1];  choose_qparams_affine_default_84 = None
	        quantize_affine_84: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_28, [1, 1, 1280], getitem_168, getitem_169, torch.int8)
	        dequantize_affine_168: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_84, [1, 1, 1280], getitem_168, getitem_169, torch.int8);  quantize_affine_84 = getitem_168 = getitem_169 = None
	        dequantize_affine_169: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_84: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_168, dequantize_affine_169, model_audio_tower_layers_14_self_attn_q_proj_bias);  dequantize_affine_168 = dequantize_affine_169 = model_audio_tower_layers_14_self_attn_q_proj_bias = None
	        mul_523: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_84, 0.125);  linear_84 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_42: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_523, [sym_size_int_2, 1500, 20, 64]);  mul_523 = None
	        transpose_56: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_42, 1, 2);  view_42 = None
	        contiguous_56: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_56);  transpose_56 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_85 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_28, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_170: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_85[0]
	        getitem_171: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_85[1];  choose_qparams_affine_default_85 = None
	        quantize_affine_85: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_28, [1, 1, 1280], getitem_170, getitem_171, torch.int8)
	        dequantize_affine_170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_85, [1, 1, 1280], getitem_170, getitem_171, torch.int8);  quantize_affine_85 = getitem_170 = getitem_171 = None
	        dequantize_affine_171: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_85: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_170, dequantize_affine_171);  dequantize_affine_170 = dequantize_affine_171 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_43: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_85, [sym_size_int_2, -1, 20, 64]);  linear_85 = None
	        transpose_57: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_43, 1, 2);  view_43 = None
	        contiguous_57: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_57);  transpose_57 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_86 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_28, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_172: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_86[0]
	        getitem_173: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_86[1];  choose_qparams_affine_default_86 = None
	        quantize_affine_86: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_28, [1, 1, 1280], getitem_172, getitem_173, torch.int8);  layer_norm_28 = None
	        dequantize_affine_172: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_86, [1, 1, 1280], getitem_172, getitem_173, torch.int8);  quantize_affine_86 = getitem_172 = getitem_173 = None
	        dequantize_affine_173: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_172, dequantize_affine_173, model_audio_tower_layers_14_self_attn_v_proj_bias);  dequantize_affine_172 = dequantize_affine_173 = model_audio_tower_layers_14_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_44: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_86, [sym_size_int_2, -1, 20, 64]);  linear_86 = None
	        transpose_58: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_44, 1, 2);  view_44 = None
	        contiguous_58: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_58);  transpose_58 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_14: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_56, contiguous_57, contiguous_58, scale = 1.0);  contiguous_56 = contiguous_57 = contiguous_58 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_59: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_14, 1, 2);  scaled_dot_product_attention_14 = None
	        contiguous_59: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_59);  transpose_59 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_59, [sym_size_int_2, 1500, -1]);  contiguous_59 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_87 = torch.ops.torchao.choose_qparams_affine.default(reshape_14, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_174: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_87[0]
	        getitem_175: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_87[1];  choose_qparams_affine_default_87 = None
	        quantize_affine_87: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_14, [1, 1, 1280], getitem_174, getitem_175, torch.int8);  reshape_14 = None
	        dequantize_affine_174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_87, [1, 1, 1280], getitem_174, getitem_175, torch.int8);  quantize_affine_87 = getitem_174 = getitem_175 = None
	        dequantize_affine_175: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_87: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_174, dequantize_affine_175, model_audio_tower_layers_14_self_attn_out_proj_bias);  dequantize_affine_174 = dequantize_affine_175 = model_audio_tower_layers_14_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_43: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_87, 0.0, False);  linear_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_205: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_196, dropout_43);  add_196 = dropout_43 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_29: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_205, [1280], model_audio_tower_layers_14_final_layer_norm_weight, model_audio_tower_layers_14_final_layer_norm_bias);  model_audio_tower_layers_14_final_layer_norm_weight = model_audio_tower_layers_14_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_88 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_29, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_176: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_88[0]
	        getitem_177: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_88[1];  choose_qparams_affine_default_88 = None
	        quantize_affine_88: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_29, [1, 1, 1280], getitem_176, getitem_177, torch.int8);  layer_norm_29 = None
	        dequantize_affine_176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_88, [1, 1, 1280], getitem_176, getitem_177, torch.int8);  quantize_affine_88 = getitem_176 = getitem_177 = None
	        dequantize_affine_177: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_14_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_14_fc1_parametrizations_weight_original1, model_audio_tower_layers_14_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_14_fc1_parametrizations_weight_original0 = model_audio_tower_layers_14_fc1_parametrizations_weight_original1 = model_audio_tower_layers_14_fc1_parametrizations_weight_original2 = None
	        linear_88: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_176, dequantize_affine_177, model_audio_tower_layers_14_fc1_bias);  dequantize_affine_176 = dequantize_affine_177 = model_audio_tower_layers_14_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_16: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_88);  linear_88 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_44: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_16, 0.0, False);  gelu_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_89 = torch.ops.torchao.choose_qparams_affine.default(dropout_44, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_178: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_89[0]
	        getitem_179: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_89[1];  choose_qparams_affine_default_89 = None
	        quantize_affine_89: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_44, [1, 1, 5120], getitem_178, getitem_179, torch.int8);  dropout_44 = None
	        dequantize_affine_178: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_89, [1, 1, 5120], getitem_178, getitem_179, torch.int8);  quantize_affine_89 = getitem_178 = getitem_179 = None
	        dequantize_affine_179: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_14_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_14_fc2_parametrizations_weight_original1, model_audio_tower_layers_14_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_14_fc2_parametrizations_weight_original0 = model_audio_tower_layers_14_fc2_parametrizations_weight_original1 = model_audio_tower_layers_14_fc2_parametrizations_weight_original2 = None
	        linear_89: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_178, dequantize_affine_179, model_audio_tower_layers_14_fc2_bias);  dequantize_affine_178 = dequantize_affine_179 = model_audio_tower_layers_14_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_45: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_89, 0.0, False);  linear_89 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_210: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_205, dropout_45);  add_205 = dropout_45 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_210, [1280], model_audio_tower_layers_15_self_attn_layer_norm_weight, model_audio_tower_layers_15_self_attn_layer_norm_bias);  model_audio_tower_layers_15_self_attn_layer_norm_weight = model_audio_tower_layers_15_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_90 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_30, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_180: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_90[0]
	        getitem_181: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_90[1];  choose_qparams_affine_default_90 = None
	        quantize_affine_90: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_30, [1, 1, 1280], getitem_180, getitem_181, torch.int8)
	        dequantize_affine_180: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_90, [1, 1, 1280], getitem_180, getitem_181, torch.int8);  quantize_affine_90 = getitem_180 = getitem_181 = None
	        dequantize_affine_181: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_90: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_180, dequantize_affine_181, model_audio_tower_layers_15_self_attn_q_proj_bias);  dequantize_affine_180 = dequantize_affine_181 = model_audio_tower_layers_15_self_attn_q_proj_bias = None
	        mul_560: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_90, 0.125);  linear_90 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_45: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_560, [sym_size_int_2, 1500, 20, 64]);  mul_560 = None
	        transpose_60: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_45, 1, 2);  view_45 = None
	        contiguous_60: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_60);  transpose_60 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_91 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_30, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_182: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_91[0]
	        getitem_183: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_91[1];  choose_qparams_affine_default_91 = None
	        quantize_affine_91: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_30, [1, 1, 1280], getitem_182, getitem_183, torch.int8)
	        dequantize_affine_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_91, [1, 1, 1280], getitem_182, getitem_183, torch.int8);  quantize_affine_91 = getitem_182 = getitem_183 = None
	        dequantize_affine_183: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_91: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_182, dequantize_affine_183);  dequantize_affine_182 = dequantize_affine_183 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_46: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_91, [sym_size_int_2, -1, 20, 64]);  linear_91 = None
	        transpose_61: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_46, 1, 2);  view_46 = None
	        contiguous_61: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_61);  transpose_61 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_92 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_30, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_184: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_92[0]
	        getitem_185: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_92[1];  choose_qparams_affine_default_92 = None
	        quantize_affine_92: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_30, [1, 1, 1280], getitem_184, getitem_185, torch.int8);  layer_norm_30 = None
	        dequantize_affine_184: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_92, [1, 1, 1280], getitem_184, getitem_185, torch.int8);  quantize_affine_92 = getitem_184 = getitem_185 = None
	        dequantize_affine_185: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_92: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_184, dequantize_affine_185, model_audio_tower_layers_15_self_attn_v_proj_bias);  dequantize_affine_184 = dequantize_affine_185 = model_audio_tower_layers_15_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_47: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_92, [sym_size_int_2, -1, 20, 64]);  linear_92 = None
	        transpose_62: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_47, 1, 2);  view_47 = None
	        contiguous_62: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_62);  transpose_62 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_15: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_60, contiguous_61, contiguous_62, scale = 1.0);  contiguous_60 = contiguous_61 = contiguous_62 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_63: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_15, 1, 2);  scaled_dot_product_attention_15 = None
	        contiguous_63: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_63);  transpose_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_15: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_63, [sym_size_int_2, 1500, -1]);  contiguous_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_93 = torch.ops.torchao.choose_qparams_affine.default(reshape_15, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_186: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_93[0]
	        getitem_187: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_93[1];  choose_qparams_affine_default_93 = None
	        quantize_affine_93: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_15, [1, 1, 1280], getitem_186, getitem_187, torch.int8);  reshape_15 = None
	        dequantize_affine_186: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_93, [1, 1, 1280], getitem_186, getitem_187, torch.int8);  quantize_affine_93 = getitem_186 = getitem_187 = None
	        dequantize_affine_187: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_93: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_186, dequantize_affine_187, model_audio_tower_layers_15_self_attn_out_proj_bias);  dequantize_affine_186 = dequantize_affine_187 = model_audio_tower_layers_15_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_46: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_93, 0.0, False);  linear_93 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_219: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_210, dropout_46);  add_210 = dropout_46 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_31: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_219, [1280], model_audio_tower_layers_15_final_layer_norm_weight, model_audio_tower_layers_15_final_layer_norm_bias);  model_audio_tower_layers_15_final_layer_norm_weight = model_audio_tower_layers_15_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_94 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_31, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_188: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_94[0]
	        getitem_189: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_94[1];  choose_qparams_affine_default_94 = None
	        quantize_affine_94: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_31, [1, 1, 1280], getitem_188, getitem_189, torch.int8);  layer_norm_31 = None
	        dequantize_affine_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_94, [1, 1, 1280], getitem_188, getitem_189, torch.int8);  quantize_affine_94 = getitem_188 = getitem_189 = None
	        dequantize_affine_189: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_15_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_15_fc1_parametrizations_weight_original1, model_audio_tower_layers_15_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_15_fc1_parametrizations_weight_original0 = model_audio_tower_layers_15_fc1_parametrizations_weight_original1 = model_audio_tower_layers_15_fc1_parametrizations_weight_original2 = None
	        linear_94: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_188, dequantize_affine_189, model_audio_tower_layers_15_fc1_bias);  dequantize_affine_188 = dequantize_affine_189 = model_audio_tower_layers_15_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_17: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_94);  linear_94 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_47: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_17, 0.0, False);  gelu_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_95 = torch.ops.torchao.choose_qparams_affine.default(dropout_47, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_190: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_95[0]
	        getitem_191: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_95[1];  choose_qparams_affine_default_95 = None
	        quantize_affine_95: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_47, [1, 1, 5120], getitem_190, getitem_191, torch.int8);  dropout_47 = None
	        dequantize_affine_190: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_95, [1, 1, 5120], getitem_190, getitem_191, torch.int8);  quantize_affine_95 = getitem_190 = getitem_191 = None
	        dequantize_affine_191: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_15_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_15_fc2_parametrizations_weight_original1, model_audio_tower_layers_15_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_15_fc2_parametrizations_weight_original0 = model_audio_tower_layers_15_fc2_parametrizations_weight_original1 = model_audio_tower_layers_15_fc2_parametrizations_weight_original2 = None
	        linear_95: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_190, dequantize_affine_191, model_audio_tower_layers_15_fc2_bias);  dequantize_affine_190 = dequantize_affine_191 = model_audio_tower_layers_15_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_48: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_95, 0.0, False);  linear_95 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_224: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_219, dropout_48);  add_219 = dropout_48 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_224, [1280], model_audio_tower_layers_16_self_attn_layer_norm_weight, model_audio_tower_layers_16_self_attn_layer_norm_bias);  model_audio_tower_layers_16_self_attn_layer_norm_weight = model_audio_tower_layers_16_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_96 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_32, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_192: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_96[0]
	        getitem_193: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_96[1];  choose_qparams_affine_default_96 = None
	        quantize_affine_96: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_32, [1, 1, 1280], getitem_192, getitem_193, torch.int8)
	        dequantize_affine_192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_96, [1, 1, 1280], getitem_192, getitem_193, torch.int8);  quantize_affine_96 = getitem_192 = getitem_193 = None
	        dequantize_affine_193: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_96: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_192, dequantize_affine_193, model_audio_tower_layers_16_self_attn_q_proj_bias);  dequantize_affine_192 = dequantize_affine_193 = model_audio_tower_layers_16_self_attn_q_proj_bias = None
	        mul_597: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_96, 0.125);  linear_96 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_48: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_597, [sym_size_int_2, 1500, 20, 64]);  mul_597 = None
	        transpose_64: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_48, 1, 2);  view_48 = None
	        contiguous_64: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_64);  transpose_64 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_97 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_32, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_194: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_97[0]
	        getitem_195: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_97[1];  choose_qparams_affine_default_97 = None
	        quantize_affine_97: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_32, [1, 1, 1280], getitem_194, getitem_195, torch.int8)
	        dequantize_affine_194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_97, [1, 1, 1280], getitem_194, getitem_195, torch.int8);  quantize_affine_97 = getitem_194 = getitem_195 = None
	        dequantize_affine_195: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_97: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_194, dequantize_affine_195);  dequantize_affine_194 = dequantize_affine_195 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_49: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_97, [sym_size_int_2, -1, 20, 64]);  linear_97 = None
	        transpose_65: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_49, 1, 2);  view_49 = None
	        contiguous_65: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_65);  transpose_65 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_98 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_32, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_196: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_98[0]
	        getitem_197: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_98[1];  choose_qparams_affine_default_98 = None
	        quantize_affine_98: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_32, [1, 1, 1280], getitem_196, getitem_197, torch.int8);  layer_norm_32 = None
	        dequantize_affine_196: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_98, [1, 1, 1280], getitem_196, getitem_197, torch.int8);  quantize_affine_98 = getitem_196 = getitem_197 = None
	        dequantize_affine_197: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_196, dequantize_affine_197, model_audio_tower_layers_16_self_attn_v_proj_bias);  dequantize_affine_196 = dequantize_affine_197 = model_audio_tower_layers_16_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_50: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_98, [sym_size_int_2, -1, 20, 64]);  linear_98 = None
	        transpose_66: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_50, 1, 2);  view_50 = None
	        contiguous_66: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_66);  transpose_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_16: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_64, contiguous_65, contiguous_66, scale = 1.0);  contiguous_64 = contiguous_65 = contiguous_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_67: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_16, 1, 2);  scaled_dot_product_attention_16 = None
	        contiguous_67: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_67);  transpose_67 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_16: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_67, [sym_size_int_2, 1500, -1]);  contiguous_67 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_99 = torch.ops.torchao.choose_qparams_affine.default(reshape_16, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_198: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_99[0]
	        getitem_199: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_99[1];  choose_qparams_affine_default_99 = None
	        quantize_affine_99: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_16, [1, 1, 1280], getitem_198, getitem_199, torch.int8);  reshape_16 = None
	        dequantize_affine_198: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_99, [1, 1, 1280], getitem_198, getitem_199, torch.int8);  quantize_affine_99 = getitem_198 = getitem_199 = None
	        dequantize_affine_199: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_99: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_198, dequantize_affine_199, model_audio_tower_layers_16_self_attn_out_proj_bias);  dequantize_affine_198 = dequantize_affine_199 = model_audio_tower_layers_16_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_49: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_99, 0.0, False);  linear_99 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_233: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_224, dropout_49);  add_224 = dropout_49 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_33: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_233, [1280], model_audio_tower_layers_16_final_layer_norm_weight, model_audio_tower_layers_16_final_layer_norm_bias);  model_audio_tower_layers_16_final_layer_norm_weight = model_audio_tower_layers_16_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_100 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_33, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_200: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_100[0]
	        getitem_201: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_100[1];  choose_qparams_affine_default_100 = None
	        quantize_affine_100: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_33, [1, 1, 1280], getitem_200, getitem_201, torch.int8);  layer_norm_33 = None
	        dequantize_affine_200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_100, [1, 1, 1280], getitem_200, getitem_201, torch.int8);  quantize_affine_100 = getitem_200 = getitem_201 = None
	        dequantize_affine_201: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_16_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_16_fc1_parametrizations_weight_original1, model_audio_tower_layers_16_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_16_fc1_parametrizations_weight_original0 = model_audio_tower_layers_16_fc1_parametrizations_weight_original1 = model_audio_tower_layers_16_fc1_parametrizations_weight_original2 = None
	        linear_100: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_200, dequantize_affine_201, model_audio_tower_layers_16_fc1_bias);  dequantize_affine_200 = dequantize_affine_201 = model_audio_tower_layers_16_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_18: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_100);  linear_100 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_50: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_18, 0.0, False);  gelu_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_101 = torch.ops.torchao.choose_qparams_affine.default(dropout_50, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_202: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_101[0]
	        getitem_203: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_101[1];  choose_qparams_affine_default_101 = None
	        quantize_affine_101: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_50, [1, 1, 5120], getitem_202, getitem_203, torch.int8);  dropout_50 = None
	        dequantize_affine_202: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_101, [1, 1, 5120], getitem_202, getitem_203, torch.int8);  quantize_affine_101 = getitem_202 = getitem_203 = None
	        dequantize_affine_203: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_16_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_16_fc2_parametrizations_weight_original1, model_audio_tower_layers_16_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_16_fc2_parametrizations_weight_original0 = model_audio_tower_layers_16_fc2_parametrizations_weight_original1 = model_audio_tower_layers_16_fc2_parametrizations_weight_original2 = None
	        linear_101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_202, dequantize_affine_203, model_audio_tower_layers_16_fc2_bias);  dequantize_affine_202 = dequantize_affine_203 = model_audio_tower_layers_16_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_51: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_101, 0.0, False);  linear_101 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_238: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_233, dropout_51);  add_233 = dropout_51 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_34: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_238, [1280], model_audio_tower_layers_17_self_attn_layer_norm_weight, model_audio_tower_layers_17_self_attn_layer_norm_bias);  model_audio_tower_layers_17_self_attn_layer_norm_weight = model_audio_tower_layers_17_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_102 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_34, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_204: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_102[0]
	        getitem_205: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_102[1];  choose_qparams_affine_default_102 = None
	        quantize_affine_102: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_34, [1, 1, 1280], getitem_204, getitem_205, torch.int8)
	        dequantize_affine_204: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_102, [1, 1, 1280], getitem_204, getitem_205, torch.int8);  quantize_affine_102 = getitem_204 = getitem_205 = None
	        dequantize_affine_205: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_102: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_204, dequantize_affine_205, model_audio_tower_layers_17_self_attn_q_proj_bias);  dequantize_affine_204 = dequantize_affine_205 = model_audio_tower_layers_17_self_attn_q_proj_bias = None
	        mul_634: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_102, 0.125);  linear_102 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_51: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_634, [sym_size_int_2, 1500, 20, 64]);  mul_634 = None
	        transpose_68: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_51, 1, 2);  view_51 = None
	        contiguous_68: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_68);  transpose_68 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_103 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_34, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_206: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_103[0]
	        getitem_207: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_103[1];  choose_qparams_affine_default_103 = None
	        quantize_affine_103: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_34, [1, 1, 1280], getitem_206, getitem_207, torch.int8)
	        dequantize_affine_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_103, [1, 1, 1280], getitem_206, getitem_207, torch.int8);  quantize_affine_103 = getitem_206 = getitem_207 = None
	        dequantize_affine_207: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_103: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_206, dequantize_affine_207);  dequantize_affine_206 = dequantize_affine_207 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_52: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_103, [sym_size_int_2, -1, 20, 64]);  linear_103 = None
	        transpose_69: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_52, 1, 2);  view_52 = None
	        contiguous_69: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_69);  transpose_69 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_104 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_34, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_208: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_104[0]
	        getitem_209: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_104[1];  choose_qparams_affine_default_104 = None
	        quantize_affine_104: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_34, [1, 1, 1280], getitem_208, getitem_209, torch.int8);  layer_norm_34 = None
	        dequantize_affine_208: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_104, [1, 1, 1280], getitem_208, getitem_209, torch.int8);  quantize_affine_104 = getitem_208 = getitem_209 = None
	        dequantize_affine_209: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_104: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_208, dequantize_affine_209, model_audio_tower_layers_17_self_attn_v_proj_bias);  dequantize_affine_208 = dequantize_affine_209 = model_audio_tower_layers_17_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_53: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_104, [sym_size_int_2, -1, 20, 64]);  linear_104 = None
	        transpose_70: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_53, 1, 2);  view_53 = None
	        contiguous_70: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_70);  transpose_70 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_17: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_68, contiguous_69, contiguous_70, scale = 1.0);  contiguous_68 = contiguous_69 = contiguous_70 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_71: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_17, 1, 2);  scaled_dot_product_attention_17 = None
	        contiguous_71: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_71);  transpose_71 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_17: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_71, [sym_size_int_2, 1500, -1]);  contiguous_71 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_105 = torch.ops.torchao.choose_qparams_affine.default(reshape_17, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_210: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_105[0]
	        getitem_211: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_105[1];  choose_qparams_affine_default_105 = None
	        quantize_affine_105: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_17, [1, 1, 1280], getitem_210, getitem_211, torch.int8);  reshape_17 = None
	        dequantize_affine_210: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_105, [1, 1, 1280], getitem_210, getitem_211, torch.int8);  quantize_affine_105 = getitem_210 = getitem_211 = None
	        dequantize_affine_211: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_105: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_210, dequantize_affine_211, model_audio_tower_layers_17_self_attn_out_proj_bias);  dequantize_affine_210 = dequantize_affine_211 = model_audio_tower_layers_17_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_52: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_105, 0.0, False);  linear_105 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_247: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_238, dropout_52);  add_238 = dropout_52 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_35: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_247, [1280], model_audio_tower_layers_17_final_layer_norm_weight, model_audio_tower_layers_17_final_layer_norm_bias);  model_audio_tower_layers_17_final_layer_norm_weight = model_audio_tower_layers_17_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_106 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_35, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_212: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_106[0]
	        getitem_213: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_106[1];  choose_qparams_affine_default_106 = None
	        quantize_affine_106: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_35, [1, 1, 1280], getitem_212, getitem_213, torch.int8);  layer_norm_35 = None
	        dequantize_affine_212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_106, [1, 1, 1280], getitem_212, getitem_213, torch.int8);  quantize_affine_106 = getitem_212 = getitem_213 = None
	        dequantize_affine_213: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_17_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_17_fc1_parametrizations_weight_original1, model_audio_tower_layers_17_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_17_fc1_parametrizations_weight_original0 = model_audio_tower_layers_17_fc1_parametrizations_weight_original1 = model_audio_tower_layers_17_fc1_parametrizations_weight_original2 = None
	        linear_106: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_212, dequantize_affine_213, model_audio_tower_layers_17_fc1_bias);  dequantize_affine_212 = dequantize_affine_213 = model_audio_tower_layers_17_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_19: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_106);  linear_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_53: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_19, 0.0, False);  gelu_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_107 = torch.ops.torchao.choose_qparams_affine.default(dropout_53, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_214: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_107[0]
	        getitem_215: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_107[1];  choose_qparams_affine_default_107 = None
	        quantize_affine_107: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_53, [1, 1, 5120], getitem_214, getitem_215, torch.int8);  dropout_53 = None
	        dequantize_affine_214: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_107, [1, 1, 5120], getitem_214, getitem_215, torch.int8);  quantize_affine_107 = getitem_214 = getitem_215 = None
	        dequantize_affine_215: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_17_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_17_fc2_parametrizations_weight_original1, model_audio_tower_layers_17_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_17_fc2_parametrizations_weight_original0 = model_audio_tower_layers_17_fc2_parametrizations_weight_original1 = model_audio_tower_layers_17_fc2_parametrizations_weight_original2 = None
	        linear_107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_214, dequantize_affine_215, model_audio_tower_layers_17_fc2_bias);  dequantize_affine_214 = dequantize_affine_215 = model_audio_tower_layers_17_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_54: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_107, 0.0, False);  linear_107 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_252: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_247, dropout_54);  add_247 = dropout_54 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_36: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_252, [1280], model_audio_tower_layers_18_self_attn_layer_norm_weight, model_audio_tower_layers_18_self_attn_layer_norm_bias);  model_audio_tower_layers_18_self_attn_layer_norm_weight = model_audio_tower_layers_18_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_108 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_36, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_216: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_108[0]
	        getitem_217: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_108[1];  choose_qparams_affine_default_108 = None
	        quantize_affine_108: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_36, [1, 1, 1280], getitem_216, getitem_217, torch.int8)
	        dequantize_affine_216: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_108, [1, 1, 1280], getitem_216, getitem_217, torch.int8);  quantize_affine_108 = getitem_216 = getitem_217 = None
	        dequantize_affine_217: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_108: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_216, dequantize_affine_217, model_audio_tower_layers_18_self_attn_q_proj_bias);  dequantize_affine_216 = dequantize_affine_217 = model_audio_tower_layers_18_self_attn_q_proj_bias = None
	        mul_671: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_108, 0.125);  linear_108 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_54: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_671, [sym_size_int_2, 1500, 20, 64]);  mul_671 = None
	        transpose_72: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_54, 1, 2);  view_54 = None
	        contiguous_72: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_72);  transpose_72 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_109 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_36, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_218: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_109[0]
	        getitem_219: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_109[1];  choose_qparams_affine_default_109 = None
	        quantize_affine_109: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_36, [1, 1, 1280], getitem_218, getitem_219, torch.int8)
	        dequantize_affine_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_109, [1, 1, 1280], getitem_218, getitem_219, torch.int8);  quantize_affine_109 = getitem_218 = getitem_219 = None
	        dequantize_affine_219: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_109: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_218, dequantize_affine_219);  dequantize_affine_218 = dequantize_affine_219 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_55: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_109, [sym_size_int_2, -1, 20, 64]);  linear_109 = None
	        transpose_73: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_55, 1, 2);  view_55 = None
	        contiguous_73: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_73);  transpose_73 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_110 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_36, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_220: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_110[0]
	        getitem_221: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_110[1];  choose_qparams_affine_default_110 = None
	        quantize_affine_110: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_36, [1, 1, 1280], getitem_220, getitem_221, torch.int8);  layer_norm_36 = None
	        dequantize_affine_220: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_110, [1, 1, 1280], getitem_220, getitem_221, torch.int8);  quantize_affine_110 = getitem_220 = getitem_221 = None
	        dequantize_affine_221: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_220, dequantize_affine_221, model_audio_tower_layers_18_self_attn_v_proj_bias);  dequantize_affine_220 = dequantize_affine_221 = model_audio_tower_layers_18_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_56: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_110, [sym_size_int_2, -1, 20, 64]);  linear_110 = None
	        transpose_74: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_56, 1, 2);  view_56 = None
	        contiguous_74: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_74);  transpose_74 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_18: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_72, contiguous_73, contiguous_74, scale = 1.0);  contiguous_72 = contiguous_73 = contiguous_74 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_75: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_18, 1, 2);  scaled_dot_product_attention_18 = None
	        contiguous_75: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_75);  transpose_75 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_18: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_75, [sym_size_int_2, 1500, -1]);  contiguous_75 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_111 = torch.ops.torchao.choose_qparams_affine.default(reshape_18, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_222: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_111[0]
	        getitem_223: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_111[1];  choose_qparams_affine_default_111 = None
	        quantize_affine_111: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_18, [1, 1, 1280], getitem_222, getitem_223, torch.int8);  reshape_18 = None
	        dequantize_affine_222: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_111, [1, 1, 1280], getitem_222, getitem_223, torch.int8);  quantize_affine_111 = getitem_222 = getitem_223 = None
	        dequantize_affine_223: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_111: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_222, dequantize_affine_223, model_audio_tower_layers_18_self_attn_out_proj_bias);  dequantize_affine_222 = dequantize_affine_223 = model_audio_tower_layers_18_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_55: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_111, 0.0, False);  linear_111 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_261: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_252, dropout_55);  add_252 = dropout_55 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_37: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_261, [1280], model_audio_tower_layers_18_final_layer_norm_weight, model_audio_tower_layers_18_final_layer_norm_bias);  model_audio_tower_layers_18_final_layer_norm_weight = model_audio_tower_layers_18_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_112 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_37, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_224: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_112[0]
	        getitem_225: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_112[1];  choose_qparams_affine_default_112 = None
	        quantize_affine_112: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_37, [1, 1, 1280], getitem_224, getitem_225, torch.int8);  layer_norm_37 = None
	        dequantize_affine_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_112, [1, 1, 1280], getitem_224, getitem_225, torch.int8);  quantize_affine_112 = getitem_224 = getitem_225 = None
	        dequantize_affine_225: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_18_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_18_fc1_parametrizations_weight_original1, model_audio_tower_layers_18_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_18_fc1_parametrizations_weight_original0 = model_audio_tower_layers_18_fc1_parametrizations_weight_original1 = model_audio_tower_layers_18_fc1_parametrizations_weight_original2 = None
	        linear_112: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_224, dequantize_affine_225, model_audio_tower_layers_18_fc1_bias);  dequantize_affine_224 = dequantize_affine_225 = model_audio_tower_layers_18_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_20: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_112);  linear_112 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_56: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_20, 0.0, False);  gelu_20 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_113 = torch.ops.torchao.choose_qparams_affine.default(dropout_56, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_226: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_113[0]
	        getitem_227: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_113[1];  choose_qparams_affine_default_113 = None
	        quantize_affine_113: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_56, [1, 1, 5120], getitem_226, getitem_227, torch.int8);  dropout_56 = None
	        dequantize_affine_226: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_113, [1, 1, 5120], getitem_226, getitem_227, torch.int8);  quantize_affine_113 = getitem_226 = getitem_227 = None
	        dequantize_affine_227: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_18_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_18_fc2_parametrizations_weight_original1, model_audio_tower_layers_18_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_18_fc2_parametrizations_weight_original0 = model_audio_tower_layers_18_fc2_parametrizations_weight_original1 = model_audio_tower_layers_18_fc2_parametrizations_weight_original2 = None
	        linear_113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_226, dequantize_affine_227, model_audio_tower_layers_18_fc2_bias);  dequantize_affine_226 = dequantize_affine_227 = model_audio_tower_layers_18_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_57: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_113, 0.0, False);  linear_113 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_266: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_261, dropout_57);  add_261 = dropout_57 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_266, [1280], model_audio_tower_layers_19_self_attn_layer_norm_weight, model_audio_tower_layers_19_self_attn_layer_norm_bias);  model_audio_tower_layers_19_self_attn_layer_norm_weight = model_audio_tower_layers_19_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_114 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_38, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_228: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_114[0]
	        getitem_229: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_114[1];  choose_qparams_affine_default_114 = None
	        quantize_affine_114: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_38, [1, 1, 1280], getitem_228, getitem_229, torch.int8)
	        dequantize_affine_228: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_114, [1, 1, 1280], getitem_228, getitem_229, torch.int8);  quantize_affine_114 = getitem_228 = getitem_229 = None
	        dequantize_affine_229: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_228, dequantize_affine_229, model_audio_tower_layers_19_self_attn_q_proj_bias);  dequantize_affine_228 = dequantize_affine_229 = model_audio_tower_layers_19_self_attn_q_proj_bias = None
	        mul_708: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_114, 0.125);  linear_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_57: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_708, [sym_size_int_2, 1500, 20, 64]);  mul_708 = None
	        transpose_76: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_57, 1, 2);  view_57 = None
	        contiguous_76: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_76);  transpose_76 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_115 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_38, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_230: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_115[0]
	        getitem_231: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_115[1];  choose_qparams_affine_default_115 = None
	        quantize_affine_115: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_38, [1, 1, 1280], getitem_230, getitem_231, torch.int8)
	        dequantize_affine_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_115, [1, 1, 1280], getitem_230, getitem_231, torch.int8);  quantize_affine_115 = getitem_230 = getitem_231 = None
	        dequantize_affine_231: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_115: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_230, dequantize_affine_231);  dequantize_affine_230 = dequantize_affine_231 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_58: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_115, [sym_size_int_2, -1, 20, 64]);  linear_115 = None
	        transpose_77: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_58, 1, 2);  view_58 = None
	        contiguous_77: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_77);  transpose_77 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_116 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_38, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_232: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_116[0]
	        getitem_233: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_116[1];  choose_qparams_affine_default_116 = None
	        quantize_affine_116: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_38, [1, 1, 1280], getitem_232, getitem_233, torch.int8);  layer_norm_38 = None
	        dequantize_affine_232: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_116, [1, 1, 1280], getitem_232, getitem_233, torch.int8);  quantize_affine_116 = getitem_232 = getitem_233 = None
	        dequantize_affine_233: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_232, dequantize_affine_233, model_audio_tower_layers_19_self_attn_v_proj_bias);  dequantize_affine_232 = dequantize_affine_233 = model_audio_tower_layers_19_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_59: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_116, [sym_size_int_2, -1, 20, 64]);  linear_116 = None
	        transpose_78: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_59, 1, 2);  view_59 = None
	        contiguous_78: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_78);  transpose_78 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_19: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_76, contiguous_77, contiguous_78, scale = 1.0);  contiguous_76 = contiguous_77 = contiguous_78 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_79: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_19, 1, 2);  scaled_dot_product_attention_19 = None
	        contiguous_79: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_79);  transpose_79 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_19: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_79, [sym_size_int_2, 1500, -1]);  contiguous_79 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_117 = torch.ops.torchao.choose_qparams_affine.default(reshape_19, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_234: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_117[0]
	        getitem_235: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_117[1];  choose_qparams_affine_default_117 = None
	        quantize_affine_117: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_19, [1, 1, 1280], getitem_234, getitem_235, torch.int8);  reshape_19 = None
	        dequantize_affine_234: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_117, [1, 1, 1280], getitem_234, getitem_235, torch.int8);  quantize_affine_117 = getitem_234 = getitem_235 = None
	        dequantize_affine_235: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_234, dequantize_affine_235, model_audio_tower_layers_19_self_attn_out_proj_bias);  dequantize_affine_234 = dequantize_affine_235 = model_audio_tower_layers_19_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_58: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_117, 0.0, False);  linear_117 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_275: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_266, dropout_58);  add_266 = dropout_58 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_39: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_275, [1280], model_audio_tower_layers_19_final_layer_norm_weight, model_audio_tower_layers_19_final_layer_norm_bias);  model_audio_tower_layers_19_final_layer_norm_weight = model_audio_tower_layers_19_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_118 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_39, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_236: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_118[0]
	        getitem_237: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_118[1];  choose_qparams_affine_default_118 = None
	        quantize_affine_118: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_39, [1, 1, 1280], getitem_236, getitem_237, torch.int8);  layer_norm_39 = None
	        dequantize_affine_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_118, [1, 1, 1280], getitem_236, getitem_237, torch.int8);  quantize_affine_118 = getitem_236 = getitem_237 = None
	        dequantize_affine_237: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_19_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_19_fc1_parametrizations_weight_original1, model_audio_tower_layers_19_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_19_fc1_parametrizations_weight_original0 = model_audio_tower_layers_19_fc1_parametrizations_weight_original1 = model_audio_tower_layers_19_fc1_parametrizations_weight_original2 = None
	        linear_118: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_236, dequantize_affine_237, model_audio_tower_layers_19_fc1_bias);  dequantize_affine_236 = dequantize_affine_237 = model_audio_tower_layers_19_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_21: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_118);  linear_118 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_59: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_21, 0.0, False);  gelu_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_119 = torch.ops.torchao.choose_qparams_affine.default(dropout_59, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_238: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_119[0]
	        getitem_239: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_119[1];  choose_qparams_affine_default_119 = None
	        quantize_affine_119: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_59, [1, 1, 5120], getitem_238, getitem_239, torch.int8);  dropout_59 = None
	        dequantize_affine_238: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_119, [1, 1, 5120], getitem_238, getitem_239, torch.int8);  quantize_affine_119 = getitem_238 = getitem_239 = None
	        dequantize_affine_239: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_19_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_19_fc2_parametrizations_weight_original1, model_audio_tower_layers_19_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_19_fc2_parametrizations_weight_original0 = model_audio_tower_layers_19_fc2_parametrizations_weight_original1 = model_audio_tower_layers_19_fc2_parametrizations_weight_original2 = None
	        linear_119: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_238, dequantize_affine_239, model_audio_tower_layers_19_fc2_bias);  dequantize_affine_238 = dequantize_affine_239 = model_audio_tower_layers_19_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_60: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_119, 0.0, False);  linear_119 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_280: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_275, dropout_60);  add_275 = dropout_60 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_40: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_280, [1280], model_audio_tower_layers_20_self_attn_layer_norm_weight, model_audio_tower_layers_20_self_attn_layer_norm_bias);  model_audio_tower_layers_20_self_attn_layer_norm_weight = model_audio_tower_layers_20_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_120 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_40, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_240: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_120[0]
	        getitem_241: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_120[1];  choose_qparams_affine_default_120 = None
	        quantize_affine_120: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_40, [1, 1, 1280], getitem_240, getitem_241, torch.int8)
	        dequantize_affine_240: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_120, [1, 1, 1280], getitem_240, getitem_241, torch.int8);  quantize_affine_120 = getitem_240 = getitem_241 = None
	        dequantize_affine_241: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_120: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_240, dequantize_affine_241, model_audio_tower_layers_20_self_attn_q_proj_bias);  dequantize_affine_240 = dequantize_affine_241 = model_audio_tower_layers_20_self_attn_q_proj_bias = None
	        mul_745: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_120, 0.125);  linear_120 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_60: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_745, [sym_size_int_2, 1500, 20, 64]);  mul_745 = None
	        transpose_80: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_60, 1, 2);  view_60 = None
	        contiguous_80: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_80);  transpose_80 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_121 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_40, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_242: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_121[0]
	        getitem_243: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_121[1];  choose_qparams_affine_default_121 = None
	        quantize_affine_121: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_40, [1, 1, 1280], getitem_242, getitem_243, torch.int8)
	        dequantize_affine_242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_121, [1, 1, 1280], getitem_242, getitem_243, torch.int8);  quantize_affine_121 = getitem_242 = getitem_243 = None
	        dequantize_affine_243: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_121: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_242, dequantize_affine_243);  dequantize_affine_242 = dequantize_affine_243 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_61: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_121, [sym_size_int_2, -1, 20, 64]);  linear_121 = None
	        transpose_81: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_61, 1, 2);  view_61 = None
	        contiguous_81: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_81);  transpose_81 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_122 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_40, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_244: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_122[0]
	        getitem_245: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_122[1];  choose_qparams_affine_default_122 = None
	        quantize_affine_122: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_40, [1, 1, 1280], getitem_244, getitem_245, torch.int8);  layer_norm_40 = None
	        dequantize_affine_244: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_122, [1, 1, 1280], getitem_244, getitem_245, torch.int8);  quantize_affine_122 = getitem_244 = getitem_245 = None
	        dequantize_affine_245: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_122: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_244, dequantize_affine_245, model_audio_tower_layers_20_self_attn_v_proj_bias);  dequantize_affine_244 = dequantize_affine_245 = model_audio_tower_layers_20_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_62: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_122, [sym_size_int_2, -1, 20, 64]);  linear_122 = None
	        transpose_82: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_62, 1, 2);  view_62 = None
	        contiguous_82: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_82);  transpose_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_20: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_80, contiguous_81, contiguous_82, scale = 1.0);  contiguous_80 = contiguous_81 = contiguous_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_83: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_20, 1, 2);  scaled_dot_product_attention_20 = None
	        contiguous_83: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_83);  transpose_83 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_83, [sym_size_int_2, 1500, -1]);  contiguous_83 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_123 = torch.ops.torchao.choose_qparams_affine.default(reshape_20, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_246: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_123[0]
	        getitem_247: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_123[1];  choose_qparams_affine_default_123 = None
	        quantize_affine_123: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_20, [1, 1, 1280], getitem_246, getitem_247, torch.int8);  reshape_20 = None
	        dequantize_affine_246: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_123, [1, 1, 1280], getitem_246, getitem_247, torch.int8);  quantize_affine_123 = getitem_246 = getitem_247 = None
	        dequantize_affine_247: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_246, dequantize_affine_247, model_audio_tower_layers_20_self_attn_out_proj_bias);  dequantize_affine_246 = dequantize_affine_247 = model_audio_tower_layers_20_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_61: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_123, 0.0, False);  linear_123 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_289: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_280, dropout_61);  add_280 = dropout_61 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_41: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_289, [1280], model_audio_tower_layers_20_final_layer_norm_weight, model_audio_tower_layers_20_final_layer_norm_bias);  model_audio_tower_layers_20_final_layer_norm_weight = model_audio_tower_layers_20_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_124 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_41, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_248: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_124[0]
	        getitem_249: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_124[1];  choose_qparams_affine_default_124 = None
	        quantize_affine_124: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_41, [1, 1, 1280], getitem_248, getitem_249, torch.int8);  layer_norm_41 = None
	        dequantize_affine_248: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_124, [1, 1, 1280], getitem_248, getitem_249, torch.int8);  quantize_affine_124 = getitem_248 = getitem_249 = None
	        dequantize_affine_249: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_20_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_20_fc1_parametrizations_weight_original1, model_audio_tower_layers_20_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_20_fc1_parametrizations_weight_original0 = model_audio_tower_layers_20_fc1_parametrizations_weight_original1 = model_audio_tower_layers_20_fc1_parametrizations_weight_original2 = None
	        linear_124: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_248, dequantize_affine_249, model_audio_tower_layers_20_fc1_bias);  dequantize_affine_248 = dequantize_affine_249 = model_audio_tower_layers_20_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_22: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_124);  linear_124 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_62: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_22, 0.0, False);  gelu_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_125 = torch.ops.torchao.choose_qparams_affine.default(dropout_62, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_250: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_125[0]
	        getitem_251: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_125[1];  choose_qparams_affine_default_125 = None
	        quantize_affine_125: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_62, [1, 1, 5120], getitem_250, getitem_251, torch.int8);  dropout_62 = None
	        dequantize_affine_250: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_125, [1, 1, 5120], getitem_250, getitem_251, torch.int8);  quantize_affine_125 = getitem_250 = getitem_251 = None
	        dequantize_affine_251: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_20_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_20_fc2_parametrizations_weight_original1, model_audio_tower_layers_20_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_20_fc2_parametrizations_weight_original0 = model_audio_tower_layers_20_fc2_parametrizations_weight_original1 = model_audio_tower_layers_20_fc2_parametrizations_weight_original2 = None
	        linear_125: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_250, dequantize_affine_251, model_audio_tower_layers_20_fc2_bias);  dequantize_affine_250 = dequantize_affine_251 = model_audio_tower_layers_20_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_63: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_125, 0.0, False);  linear_125 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_294: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_289, dropout_63);  add_289 = dropout_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_42: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_294, [1280], model_audio_tower_layers_21_self_attn_layer_norm_weight, model_audio_tower_layers_21_self_attn_layer_norm_bias);  model_audio_tower_layers_21_self_attn_layer_norm_weight = model_audio_tower_layers_21_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_126 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_42, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_252: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_126[0]
	        getitem_253: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_126[1];  choose_qparams_affine_default_126 = None
	        quantize_affine_126: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_42, [1, 1, 1280], getitem_252, getitem_253, torch.int8)
	        dequantize_affine_252: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_126, [1, 1, 1280], getitem_252, getitem_253, torch.int8);  quantize_affine_126 = getitem_252 = getitem_253 = None
	        dequantize_affine_253: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_126: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_252, dequantize_affine_253, model_audio_tower_layers_21_self_attn_q_proj_bias);  dequantize_affine_252 = dequantize_affine_253 = model_audio_tower_layers_21_self_attn_q_proj_bias = None
	        mul_782: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_126, 0.125);  linear_126 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_63: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_782, [sym_size_int_2, 1500, 20, 64]);  mul_782 = None
	        transpose_84: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_63, 1, 2);  view_63 = None
	        contiguous_84: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_84);  transpose_84 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_127 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_42, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_254: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_127[0]
	        getitem_255: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_127[1];  choose_qparams_affine_default_127 = None
	        quantize_affine_127: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_42, [1, 1, 1280], getitem_254, getitem_255, torch.int8)
	        dequantize_affine_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_127, [1, 1, 1280], getitem_254, getitem_255, torch.int8);  quantize_affine_127 = getitem_254 = getitem_255 = None
	        dequantize_affine_255: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_254, dequantize_affine_255);  dequantize_affine_254 = dequantize_affine_255 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_64: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_127, [sym_size_int_2, -1, 20, 64]);  linear_127 = None
	        transpose_85: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_64, 1, 2);  view_64 = None
	        contiguous_85: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_85);  transpose_85 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_128 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_42, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_256: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_128[0]
	        getitem_257: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_128[1];  choose_qparams_affine_default_128 = None
	        quantize_affine_128: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_42, [1, 1, 1280], getitem_256, getitem_257, torch.int8);  layer_norm_42 = None
	        dequantize_affine_256: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_128, [1, 1, 1280], getitem_256, getitem_257, torch.int8);  quantize_affine_128 = getitem_256 = getitem_257 = None
	        dequantize_affine_257: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_256, dequantize_affine_257, model_audio_tower_layers_21_self_attn_v_proj_bias);  dequantize_affine_256 = dequantize_affine_257 = model_audio_tower_layers_21_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_65: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_128, [sym_size_int_2, -1, 20, 64]);  linear_128 = None
	        transpose_86: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_65, 1, 2);  view_65 = None
	        contiguous_86: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_86);  transpose_86 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_21: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_84, contiguous_85, contiguous_86, scale = 1.0);  contiguous_84 = contiguous_85 = contiguous_86 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_87: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_21, 1, 2);  scaled_dot_product_attention_21 = None
	        contiguous_87: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_87);  transpose_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_87, [sym_size_int_2, 1500, -1]);  contiguous_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_129 = torch.ops.torchao.choose_qparams_affine.default(reshape_21, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_258: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_129[0]
	        getitem_259: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_129[1];  choose_qparams_affine_default_129 = None
	        quantize_affine_129: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_21, [1, 1, 1280], getitem_258, getitem_259, torch.int8);  reshape_21 = None
	        dequantize_affine_258: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_129, [1, 1, 1280], getitem_258, getitem_259, torch.int8);  quantize_affine_129 = getitem_258 = getitem_259 = None
	        dequantize_affine_259: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_129: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_258, dequantize_affine_259, model_audio_tower_layers_21_self_attn_out_proj_bias);  dequantize_affine_258 = dequantize_affine_259 = model_audio_tower_layers_21_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_64: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_129, 0.0, False);  linear_129 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_303: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_294, dropout_64);  add_294 = dropout_64 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_43: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_303, [1280], model_audio_tower_layers_21_final_layer_norm_weight, model_audio_tower_layers_21_final_layer_norm_bias);  model_audio_tower_layers_21_final_layer_norm_weight = model_audio_tower_layers_21_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_130 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_43, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_260: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_130[0]
	        getitem_261: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_130[1];  choose_qparams_affine_default_130 = None
	        quantize_affine_130: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_43, [1, 1, 1280], getitem_260, getitem_261, torch.int8);  layer_norm_43 = None
	        dequantize_affine_260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_130, [1, 1, 1280], getitem_260, getitem_261, torch.int8);  quantize_affine_130 = getitem_260 = getitem_261 = None
	        dequantize_affine_261: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_21_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_21_fc1_parametrizations_weight_original1, model_audio_tower_layers_21_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_21_fc1_parametrizations_weight_original0 = model_audio_tower_layers_21_fc1_parametrizations_weight_original1 = model_audio_tower_layers_21_fc1_parametrizations_weight_original2 = None
	        linear_130: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_260, dequantize_affine_261, model_audio_tower_layers_21_fc1_bias);  dequantize_affine_260 = dequantize_affine_261 = model_audio_tower_layers_21_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_23: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_130);  linear_130 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_65: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_23, 0.0, False);  gelu_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_131 = torch.ops.torchao.choose_qparams_affine.default(dropout_65, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_262: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_131[0]
	        getitem_263: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_131[1];  choose_qparams_affine_default_131 = None
	        quantize_affine_131: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_65, [1, 1, 5120], getitem_262, getitem_263, torch.int8);  dropout_65 = None
	        dequantize_affine_262: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_131, [1, 1, 5120], getitem_262, getitem_263, torch.int8);  quantize_affine_131 = getitem_262 = getitem_263 = None
	        dequantize_affine_263: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_21_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_21_fc2_parametrizations_weight_original1, model_audio_tower_layers_21_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_21_fc2_parametrizations_weight_original0 = model_audio_tower_layers_21_fc2_parametrizations_weight_original1 = model_audio_tower_layers_21_fc2_parametrizations_weight_original2 = None
	        linear_131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_262, dequantize_affine_263, model_audio_tower_layers_21_fc2_bias);  dequantize_affine_262 = dequantize_affine_263 = model_audio_tower_layers_21_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_66: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_131, 0.0, False);  linear_131 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_308: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_303, dropout_66);  add_303 = dropout_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_308, [1280], model_audio_tower_layers_22_self_attn_layer_norm_weight, model_audio_tower_layers_22_self_attn_layer_norm_bias);  model_audio_tower_layers_22_self_attn_layer_norm_weight = model_audio_tower_layers_22_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_132 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_44, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_264: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_132[0]
	        getitem_265: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_132[1];  choose_qparams_affine_default_132 = None
	        quantize_affine_132: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_44, [1, 1, 1280], getitem_264, getitem_265, torch.int8)
	        dequantize_affine_264: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_132, [1, 1, 1280], getitem_264, getitem_265, torch.int8);  quantize_affine_132 = getitem_264 = getitem_265 = None
	        dequantize_affine_265: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_132: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_264, dequantize_affine_265, model_audio_tower_layers_22_self_attn_q_proj_bias);  dequantize_affine_264 = dequantize_affine_265 = model_audio_tower_layers_22_self_attn_q_proj_bias = None
	        mul_819: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_132, 0.125);  linear_132 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_66: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_819, [sym_size_int_2, 1500, 20, 64]);  mul_819 = None
	        transpose_88: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_66, 1, 2);  view_66 = None
	        contiguous_88: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_88);  transpose_88 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_133 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_44, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_266: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_133[0]
	        getitem_267: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_133[1];  choose_qparams_affine_default_133 = None
	        quantize_affine_133: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_44, [1, 1, 1280], getitem_266, getitem_267, torch.int8)
	        dequantize_affine_266: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_133, [1, 1, 1280], getitem_266, getitem_267, torch.int8);  quantize_affine_133 = getitem_266 = getitem_267 = None
	        dequantize_affine_267: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_133: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_266, dequantize_affine_267);  dequantize_affine_266 = dequantize_affine_267 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_67: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_133, [sym_size_int_2, -1, 20, 64]);  linear_133 = None
	        transpose_89: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_67, 1, 2);  view_67 = None
	        contiguous_89: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_89);  transpose_89 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_134 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_44, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_268: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_134[0]
	        getitem_269: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_134[1];  choose_qparams_affine_default_134 = None
	        quantize_affine_134: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_44, [1, 1, 1280], getitem_268, getitem_269, torch.int8);  layer_norm_44 = None
	        dequantize_affine_268: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_134, [1, 1, 1280], getitem_268, getitem_269, torch.int8);  quantize_affine_134 = getitem_268 = getitem_269 = None
	        dequantize_affine_269: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_268, dequantize_affine_269, model_audio_tower_layers_22_self_attn_v_proj_bias);  dequantize_affine_268 = dequantize_affine_269 = model_audio_tower_layers_22_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_68: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_134, [sym_size_int_2, -1, 20, 64]);  linear_134 = None
	        transpose_90: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_68, 1, 2);  view_68 = None
	        contiguous_90: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_90);  transpose_90 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_22: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_88, contiguous_89, contiguous_90, scale = 1.0);  contiguous_88 = contiguous_89 = contiguous_90 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_91: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_22, 1, 2);  scaled_dot_product_attention_22 = None
	        contiguous_91: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_91);  transpose_91 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_22: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_91, [sym_size_int_2, 1500, -1]);  contiguous_91 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_135 = torch.ops.torchao.choose_qparams_affine.default(reshape_22, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_270: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_135[0]
	        getitem_271: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_135[1];  choose_qparams_affine_default_135 = None
	        quantize_affine_135: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_22, [1, 1, 1280], getitem_270, getitem_271, torch.int8);  reshape_22 = None
	        dequantize_affine_270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_135, [1, 1, 1280], getitem_270, getitem_271, torch.int8);  quantize_affine_135 = getitem_270 = getitem_271 = None
	        dequantize_affine_271: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_135: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_270, dequantize_affine_271, model_audio_tower_layers_22_self_attn_out_proj_bias);  dequantize_affine_270 = dequantize_affine_271 = model_audio_tower_layers_22_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_67: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_135, 0.0, False);  linear_135 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_317: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_308, dropout_67);  add_308 = dropout_67 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_45: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_317, [1280], model_audio_tower_layers_22_final_layer_norm_weight, model_audio_tower_layers_22_final_layer_norm_bias);  model_audio_tower_layers_22_final_layer_norm_weight = model_audio_tower_layers_22_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_136 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_45, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_272: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_136[0]
	        getitem_273: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_136[1];  choose_qparams_affine_default_136 = None
	        quantize_affine_136: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_45, [1, 1, 1280], getitem_272, getitem_273, torch.int8);  layer_norm_45 = None
	        dequantize_affine_272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_136, [1, 1, 1280], getitem_272, getitem_273, torch.int8);  quantize_affine_136 = getitem_272 = getitem_273 = None
	        dequantize_affine_273: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_22_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_22_fc1_parametrizations_weight_original1, model_audio_tower_layers_22_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_22_fc1_parametrizations_weight_original0 = model_audio_tower_layers_22_fc1_parametrizations_weight_original1 = model_audio_tower_layers_22_fc1_parametrizations_weight_original2 = None
	        linear_136: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_272, dequantize_affine_273, model_audio_tower_layers_22_fc1_bias);  dequantize_affine_272 = dequantize_affine_273 = model_audio_tower_layers_22_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_24: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_136);  linear_136 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_68: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_24, 0.0, False);  gelu_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_137 = torch.ops.torchao.choose_qparams_affine.default(dropout_68, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_274: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_137[0]
	        getitem_275: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_137[1];  choose_qparams_affine_default_137 = None
	        quantize_affine_137: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_68, [1, 1, 5120], getitem_274, getitem_275, torch.int8);  dropout_68 = None
	        dequantize_affine_274: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_137, [1, 1, 5120], getitem_274, getitem_275, torch.int8);  quantize_affine_137 = getitem_274 = getitem_275 = None
	        dequantize_affine_275: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_22_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_22_fc2_parametrizations_weight_original1, model_audio_tower_layers_22_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_22_fc2_parametrizations_weight_original0 = model_audio_tower_layers_22_fc2_parametrizations_weight_original1 = model_audio_tower_layers_22_fc2_parametrizations_weight_original2 = None
	        linear_137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_274, dequantize_affine_275, model_audio_tower_layers_22_fc2_bias);  dequantize_affine_274 = dequantize_affine_275 = model_audio_tower_layers_22_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_69: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_137, 0.0, False);  linear_137 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_322: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_317, dropout_69);  add_317 = dropout_69 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_46: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_322, [1280], model_audio_tower_layers_23_self_attn_layer_norm_weight, model_audio_tower_layers_23_self_attn_layer_norm_bias);  model_audio_tower_layers_23_self_attn_layer_norm_weight = model_audio_tower_layers_23_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_138 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_46, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_276: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_138[0]
	        getitem_277: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_138[1];  choose_qparams_affine_default_138 = None
	        quantize_affine_138: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_46, [1, 1, 1280], getitem_276, getitem_277, torch.int8)
	        dequantize_affine_276: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_138, [1, 1, 1280], getitem_276, getitem_277, torch.int8);  quantize_affine_138 = getitem_276 = getitem_277 = None
	        dequantize_affine_277: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_138: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_276, dequantize_affine_277, model_audio_tower_layers_23_self_attn_q_proj_bias);  dequantize_affine_276 = dequantize_affine_277 = model_audio_tower_layers_23_self_attn_q_proj_bias = None
	        mul_856: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_138, 0.125);  linear_138 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_69: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_856, [sym_size_int_2, 1500, 20, 64]);  mul_856 = None
	        transpose_92: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_69, 1, 2);  view_69 = None
	        contiguous_92: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_92);  transpose_92 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_139 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_46, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_278: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_139[0]
	        getitem_279: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_139[1];  choose_qparams_affine_default_139 = None
	        quantize_affine_139: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_46, [1, 1, 1280], getitem_278, getitem_279, torch.int8)
	        dequantize_affine_278: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_139, [1, 1, 1280], getitem_278, getitem_279, torch.int8);  quantize_affine_139 = getitem_278 = getitem_279 = None
	        dequantize_affine_279: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_139: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_278, dequantize_affine_279);  dequantize_affine_278 = dequantize_affine_279 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_70: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_139, [sym_size_int_2, -1, 20, 64]);  linear_139 = None
	        transpose_93: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_70, 1, 2);  view_70 = None
	        contiguous_93: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_93);  transpose_93 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_140 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_46, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_280: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_140[0]
	        getitem_281: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_140[1];  choose_qparams_affine_default_140 = None
	        quantize_affine_140: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_46, [1, 1, 1280], getitem_280, getitem_281, torch.int8);  layer_norm_46 = None
	        dequantize_affine_280: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_140, [1, 1, 1280], getitem_280, getitem_281, torch.int8);  quantize_affine_140 = getitem_280 = getitem_281 = None
	        dequantize_affine_281: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_280, dequantize_affine_281, model_audio_tower_layers_23_self_attn_v_proj_bias);  dequantize_affine_280 = dequantize_affine_281 = model_audio_tower_layers_23_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_71: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_140, [sym_size_int_2, -1, 20, 64]);  linear_140 = None
	        transpose_94: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_71, 1, 2);  view_71 = None
	        contiguous_94: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_94);  transpose_94 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_23: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_92, contiguous_93, contiguous_94, scale = 1.0);  contiguous_92 = contiguous_93 = contiguous_94 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_95: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_23, 1, 2);  scaled_dot_product_attention_23 = None
	        contiguous_95: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_95);  transpose_95 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_23: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_95, [sym_size_int_2, 1500, -1]);  contiguous_95 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_141 = torch.ops.torchao.choose_qparams_affine.default(reshape_23, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_282: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_141[0]
	        getitem_283: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_141[1];  choose_qparams_affine_default_141 = None
	        quantize_affine_141: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_23, [1, 1, 1280], getitem_282, getitem_283, torch.int8);  reshape_23 = None
	        dequantize_affine_282: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_141, [1, 1, 1280], getitem_282, getitem_283, torch.int8);  quantize_affine_141 = getitem_282 = getitem_283 = None
	        dequantize_affine_283: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_141: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_282, dequantize_affine_283, model_audio_tower_layers_23_self_attn_out_proj_bias);  dequantize_affine_282 = dequantize_affine_283 = model_audio_tower_layers_23_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_70: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_141, 0.0, False);  linear_141 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_331: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_322, dropout_70);  add_322 = dropout_70 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_47: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_331, [1280], model_audio_tower_layers_23_final_layer_norm_weight, model_audio_tower_layers_23_final_layer_norm_bias);  model_audio_tower_layers_23_final_layer_norm_weight = model_audio_tower_layers_23_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_142 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_47, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_284: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_142[0]
	        getitem_285: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_142[1];  choose_qparams_affine_default_142 = None
	        quantize_affine_142: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_47, [1, 1, 1280], getitem_284, getitem_285, torch.int8);  layer_norm_47 = None
	        dequantize_affine_284: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_142, [1, 1, 1280], getitem_284, getitem_285, torch.int8);  quantize_affine_142 = getitem_284 = getitem_285 = None
	        dequantize_affine_285: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_23_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_23_fc1_parametrizations_weight_original1, model_audio_tower_layers_23_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_23_fc1_parametrizations_weight_original0 = model_audio_tower_layers_23_fc1_parametrizations_weight_original1 = model_audio_tower_layers_23_fc1_parametrizations_weight_original2 = None
	        linear_142: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_284, dequantize_affine_285, model_audio_tower_layers_23_fc1_bias);  dequantize_affine_284 = dequantize_affine_285 = model_audio_tower_layers_23_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_25: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_142);  linear_142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_71: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_25, 0.0, False);  gelu_25 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_143 = torch.ops.torchao.choose_qparams_affine.default(dropout_71, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_286: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_143[0]
	        getitem_287: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_143[1];  choose_qparams_affine_default_143 = None
	        quantize_affine_143: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_71, [1, 1, 5120], getitem_286, getitem_287, torch.int8);  dropout_71 = None
	        dequantize_affine_286: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_143, [1, 1, 5120], getitem_286, getitem_287, torch.int8);  quantize_affine_143 = getitem_286 = getitem_287 = None
	        dequantize_affine_287: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_23_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_23_fc2_parametrizations_weight_original1, model_audio_tower_layers_23_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_23_fc2_parametrizations_weight_original0 = model_audio_tower_layers_23_fc2_parametrizations_weight_original1 = model_audio_tower_layers_23_fc2_parametrizations_weight_original2 = None
	        linear_143: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_286, dequantize_affine_287, model_audio_tower_layers_23_fc2_bias);  dequantize_affine_286 = dequantize_affine_287 = model_audio_tower_layers_23_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_72: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_143, 0.0, False);  linear_143 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_336: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_331, dropout_72);  add_331 = dropout_72 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_48: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_336, [1280], model_audio_tower_layers_24_self_attn_layer_norm_weight, model_audio_tower_layers_24_self_attn_layer_norm_bias);  model_audio_tower_layers_24_self_attn_layer_norm_weight = model_audio_tower_layers_24_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_144 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_48, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_288: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_144[0]
	        getitem_289: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_144[1];  choose_qparams_affine_default_144 = None
	        quantize_affine_144: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_48, [1, 1, 1280], getitem_288, getitem_289, torch.int8)
	        dequantize_affine_288: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_144, [1, 1, 1280], getitem_288, getitem_289, torch.int8);  quantize_affine_144 = getitem_288 = getitem_289 = None
	        dequantize_affine_289: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_144: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_288, dequantize_affine_289, model_audio_tower_layers_24_self_attn_q_proj_bias);  dequantize_affine_288 = dequantize_affine_289 = model_audio_tower_layers_24_self_attn_q_proj_bias = None
	        mul_893: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_144, 0.125);  linear_144 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_72: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_893, [sym_size_int_2, 1500, 20, 64]);  mul_893 = None
	        transpose_96: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_72, 1, 2);  view_72 = None
	        contiguous_96: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_96);  transpose_96 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_145 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_48, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_290: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_145[0]
	        getitem_291: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_145[1];  choose_qparams_affine_default_145 = None
	        quantize_affine_145: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_48, [1, 1, 1280], getitem_290, getitem_291, torch.int8)
	        dequantize_affine_290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_145, [1, 1, 1280], getitem_290, getitem_291, torch.int8);  quantize_affine_145 = getitem_290 = getitem_291 = None
	        dequantize_affine_291: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_145: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_290, dequantize_affine_291);  dequantize_affine_290 = dequantize_affine_291 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_73: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_145, [sym_size_int_2, -1, 20, 64]);  linear_145 = None
	        transpose_97: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_73, 1, 2);  view_73 = None
	        contiguous_97: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_97);  transpose_97 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_146 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_48, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_292: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_146[0]
	        getitem_293: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_146[1];  choose_qparams_affine_default_146 = None
	        quantize_affine_146: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_48, [1, 1, 1280], getitem_292, getitem_293, torch.int8);  layer_norm_48 = None
	        dequantize_affine_292: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_146, [1, 1, 1280], getitem_292, getitem_293, torch.int8);  quantize_affine_146 = getitem_292 = getitem_293 = None
	        dequantize_affine_293: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_292, dequantize_affine_293, model_audio_tower_layers_24_self_attn_v_proj_bias);  dequantize_affine_292 = dequantize_affine_293 = model_audio_tower_layers_24_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_74: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_146, [sym_size_int_2, -1, 20, 64]);  linear_146 = None
	        transpose_98: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_74, 1, 2);  view_74 = None
	        contiguous_98: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_98);  transpose_98 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_24: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_96, contiguous_97, contiguous_98, scale = 1.0);  contiguous_96 = contiguous_97 = contiguous_98 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_99: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_24, 1, 2);  scaled_dot_product_attention_24 = None
	        contiguous_99: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_99);  transpose_99 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_24: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_99, [sym_size_int_2, 1500, -1]);  contiguous_99 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_147 = torch.ops.torchao.choose_qparams_affine.default(reshape_24, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_294: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_147[0]
	        getitem_295: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_147[1];  choose_qparams_affine_default_147 = None
	        quantize_affine_147: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_24, [1, 1, 1280], getitem_294, getitem_295, torch.int8);  reshape_24 = None
	        dequantize_affine_294: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_147, [1, 1, 1280], getitem_294, getitem_295, torch.int8);  quantize_affine_147 = getitem_294 = getitem_295 = None
	        dequantize_affine_295: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_147: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_294, dequantize_affine_295, model_audio_tower_layers_24_self_attn_out_proj_bias);  dequantize_affine_294 = dequantize_affine_295 = model_audio_tower_layers_24_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_73: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_147, 0.0, False);  linear_147 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_345: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_336, dropout_73);  add_336 = dropout_73 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_49: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_345, [1280], model_audio_tower_layers_24_final_layer_norm_weight, model_audio_tower_layers_24_final_layer_norm_bias);  model_audio_tower_layers_24_final_layer_norm_weight = model_audio_tower_layers_24_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_148 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_49, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_296: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_148[0]
	        getitem_297: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_148[1];  choose_qparams_affine_default_148 = None
	        quantize_affine_148: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_49, [1, 1, 1280], getitem_296, getitem_297, torch.int8);  layer_norm_49 = None
	        dequantize_affine_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_148, [1, 1, 1280], getitem_296, getitem_297, torch.int8);  quantize_affine_148 = getitem_296 = getitem_297 = None
	        dequantize_affine_297: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_24_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_24_fc1_parametrizations_weight_original1, model_audio_tower_layers_24_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_24_fc1_parametrizations_weight_original0 = model_audio_tower_layers_24_fc1_parametrizations_weight_original1 = model_audio_tower_layers_24_fc1_parametrizations_weight_original2 = None
	        linear_148: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_296, dequantize_affine_297, model_audio_tower_layers_24_fc1_bias);  dequantize_affine_296 = dequantize_affine_297 = model_audio_tower_layers_24_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_26: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_148);  linear_148 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_74: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_26, 0.0, False);  gelu_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_149 = torch.ops.torchao.choose_qparams_affine.default(dropout_74, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_298: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_149[0]
	        getitem_299: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_149[1];  choose_qparams_affine_default_149 = None
	        quantize_affine_149: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_74, [1, 1, 5120], getitem_298, getitem_299, torch.int8);  dropout_74 = None
	        dequantize_affine_298: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_149, [1, 1, 5120], getitem_298, getitem_299, torch.int8);  quantize_affine_149 = getitem_298 = getitem_299 = None
	        dequantize_affine_299: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_24_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_24_fc2_parametrizations_weight_original1, model_audio_tower_layers_24_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_24_fc2_parametrizations_weight_original0 = model_audio_tower_layers_24_fc2_parametrizations_weight_original1 = model_audio_tower_layers_24_fc2_parametrizations_weight_original2 = None
	        linear_149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_298, dequantize_affine_299, model_audio_tower_layers_24_fc2_bias);  dequantize_affine_298 = dequantize_affine_299 = model_audio_tower_layers_24_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_75: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_149, 0.0, False);  linear_149 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_350: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_345, dropout_75);  add_345 = dropout_75 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_350, [1280], model_audio_tower_layers_25_self_attn_layer_norm_weight, model_audio_tower_layers_25_self_attn_layer_norm_bias);  model_audio_tower_layers_25_self_attn_layer_norm_weight = model_audio_tower_layers_25_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_150 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_50, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_300: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_150[0]
	        getitem_301: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_150[1];  choose_qparams_affine_default_150 = None
	        quantize_affine_150: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_50, [1, 1, 1280], getitem_300, getitem_301, torch.int8)
	        dequantize_affine_300: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_150, [1, 1, 1280], getitem_300, getitem_301, torch.int8);  quantize_affine_150 = getitem_300 = getitem_301 = None
	        dequantize_affine_301: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_150: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_300, dequantize_affine_301, model_audio_tower_layers_25_self_attn_q_proj_bias);  dequantize_affine_300 = dequantize_affine_301 = model_audio_tower_layers_25_self_attn_q_proj_bias = None
	        mul_930: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_150, 0.125);  linear_150 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_75: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_930, [sym_size_int_2, 1500, 20, 64]);  mul_930 = None
	        transpose_100: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_75, 1, 2);  view_75 = None
	        contiguous_100: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_100);  transpose_100 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_151 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_50, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_302: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_151[0]
	        getitem_303: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_151[1];  choose_qparams_affine_default_151 = None
	        quantize_affine_151: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_50, [1, 1, 1280], getitem_302, getitem_303, torch.int8)
	        dequantize_affine_302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_151, [1, 1, 1280], getitem_302, getitem_303, torch.int8);  quantize_affine_151 = getitem_302 = getitem_303 = None
	        dequantize_affine_303: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_151: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_302, dequantize_affine_303);  dequantize_affine_302 = dequantize_affine_303 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_76: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_151, [sym_size_int_2, -1, 20, 64]);  linear_151 = None
	        transpose_101: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_76, 1, 2);  view_76 = None
	        contiguous_101: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_101);  transpose_101 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_152 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_50, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_304: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_152[0]
	        getitem_305: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_152[1];  choose_qparams_affine_default_152 = None
	        quantize_affine_152: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_50, [1, 1, 1280], getitem_304, getitem_305, torch.int8);  layer_norm_50 = None
	        dequantize_affine_304: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_152, [1, 1, 1280], getitem_304, getitem_305, torch.int8);  quantize_affine_152 = getitem_304 = getitem_305 = None
	        dequantize_affine_305: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_304, dequantize_affine_305, model_audio_tower_layers_25_self_attn_v_proj_bias);  dequantize_affine_304 = dequantize_affine_305 = model_audio_tower_layers_25_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_77: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_152, [sym_size_int_2, -1, 20, 64]);  linear_152 = None
	        transpose_102: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_77, 1, 2);  view_77 = None
	        contiguous_102: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_102);  transpose_102 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_25: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_100, contiguous_101, contiguous_102, scale = 1.0);  contiguous_100 = contiguous_101 = contiguous_102 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_103: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_25, 1, 2);  scaled_dot_product_attention_25 = None
	        contiguous_103: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_103);  transpose_103 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_25: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_103, [sym_size_int_2, 1500, -1]);  contiguous_103 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_153 = torch.ops.torchao.choose_qparams_affine.default(reshape_25, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_306: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_153[0]
	        getitem_307: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_153[1];  choose_qparams_affine_default_153 = None
	        quantize_affine_153: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_25, [1, 1, 1280], getitem_306, getitem_307, torch.int8);  reshape_25 = None
	        dequantize_affine_306: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_153, [1, 1, 1280], getitem_306, getitem_307, torch.int8);  quantize_affine_153 = getitem_306 = getitem_307 = None
	        dequantize_affine_307: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_153: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_306, dequantize_affine_307, model_audio_tower_layers_25_self_attn_out_proj_bias);  dequantize_affine_306 = dequantize_affine_307 = model_audio_tower_layers_25_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_76: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_153, 0.0, False);  linear_153 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_359: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_350, dropout_76);  add_350 = dropout_76 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_51: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_359, [1280], model_audio_tower_layers_25_final_layer_norm_weight, model_audio_tower_layers_25_final_layer_norm_bias);  model_audio_tower_layers_25_final_layer_norm_weight = model_audio_tower_layers_25_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_154 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_51, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_308: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_154[0]
	        getitem_309: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_154[1];  choose_qparams_affine_default_154 = None
	        quantize_affine_154: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_51, [1, 1, 1280], getitem_308, getitem_309, torch.int8);  layer_norm_51 = None
	        dequantize_affine_308: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_154, [1, 1, 1280], getitem_308, getitem_309, torch.int8);  quantize_affine_154 = getitem_308 = getitem_309 = None
	        dequantize_affine_309: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_25_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_25_fc1_parametrizations_weight_original1, model_audio_tower_layers_25_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_25_fc1_parametrizations_weight_original0 = model_audio_tower_layers_25_fc1_parametrizations_weight_original1 = model_audio_tower_layers_25_fc1_parametrizations_weight_original2 = None
	        linear_154: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_308, dequantize_affine_309, model_audio_tower_layers_25_fc1_bias);  dequantize_affine_308 = dequantize_affine_309 = model_audio_tower_layers_25_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_27: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_154);  linear_154 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_77: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_27, 0.0, False);  gelu_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_155 = torch.ops.torchao.choose_qparams_affine.default(dropout_77, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_310: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_155[0]
	        getitem_311: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_155[1];  choose_qparams_affine_default_155 = None
	        quantize_affine_155: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_77, [1, 1, 5120], getitem_310, getitem_311, torch.int8);  dropout_77 = None
	        dequantize_affine_310: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_155, [1, 1, 5120], getitem_310, getitem_311, torch.int8);  quantize_affine_155 = getitem_310 = getitem_311 = None
	        dequantize_affine_311: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_25_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_25_fc2_parametrizations_weight_original1, model_audio_tower_layers_25_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_25_fc2_parametrizations_weight_original0 = model_audio_tower_layers_25_fc2_parametrizations_weight_original1 = model_audio_tower_layers_25_fc2_parametrizations_weight_original2 = None
	        linear_155: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_310, dequantize_affine_311, model_audio_tower_layers_25_fc2_bias);  dequantize_affine_310 = dequantize_affine_311 = model_audio_tower_layers_25_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_78: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_155, 0.0, False);  linear_155 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_364: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_359, dropout_78);  add_359 = dropout_78 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_52: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_364, [1280], model_audio_tower_layers_26_self_attn_layer_norm_weight, model_audio_tower_layers_26_self_attn_layer_norm_bias);  model_audio_tower_layers_26_self_attn_layer_norm_weight = model_audio_tower_layers_26_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_156 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_52, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_312: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_156[0]
	        getitem_313: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_156[1];  choose_qparams_affine_default_156 = None
	        quantize_affine_156: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_52, [1, 1, 1280], getitem_312, getitem_313, torch.int8)
	        dequantize_affine_312: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_156, [1, 1, 1280], getitem_312, getitem_313, torch.int8);  quantize_affine_156 = getitem_312 = getitem_313 = None
	        dequantize_affine_313: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_156: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_312, dequantize_affine_313, model_audio_tower_layers_26_self_attn_q_proj_bias);  dequantize_affine_312 = dequantize_affine_313 = model_audio_tower_layers_26_self_attn_q_proj_bias = None
	        mul_967: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_156, 0.125);  linear_156 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_78: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_967, [sym_size_int_2, 1500, 20, 64]);  mul_967 = None
	        transpose_104: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_78, 1, 2);  view_78 = None
	        contiguous_104: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_104);  transpose_104 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_157 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_52, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_314: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_157[0]
	        getitem_315: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_157[1];  choose_qparams_affine_default_157 = None
	        quantize_affine_157: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_52, [1, 1, 1280], getitem_314, getitem_315, torch.int8)
	        dequantize_affine_314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_157, [1, 1, 1280], getitem_314, getitem_315, torch.int8);  quantize_affine_157 = getitem_314 = getitem_315 = None
	        dequantize_affine_315: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_157: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_314, dequantize_affine_315);  dequantize_affine_314 = dequantize_affine_315 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_79: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_157, [sym_size_int_2, -1, 20, 64]);  linear_157 = None
	        transpose_105: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_79, 1, 2);  view_79 = None
	        contiguous_105: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_105);  transpose_105 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_158 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_52, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_316: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_158[0]
	        getitem_317: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_158[1];  choose_qparams_affine_default_158 = None
	        quantize_affine_158: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_52, [1, 1, 1280], getitem_316, getitem_317, torch.int8);  layer_norm_52 = None
	        dequantize_affine_316: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_158, [1, 1, 1280], getitem_316, getitem_317, torch.int8);  quantize_affine_158 = getitem_316 = getitem_317 = None
	        dequantize_affine_317: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_316, dequantize_affine_317, model_audio_tower_layers_26_self_attn_v_proj_bias);  dequantize_affine_316 = dequantize_affine_317 = model_audio_tower_layers_26_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_80: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_158, [sym_size_int_2, -1, 20, 64]);  linear_158 = None
	        transpose_106: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_80, 1, 2);  view_80 = None
	        contiguous_106: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_106);  transpose_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_26: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_104, contiguous_105, contiguous_106, scale = 1.0);  contiguous_104 = contiguous_105 = contiguous_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_107: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_26, 1, 2);  scaled_dot_product_attention_26 = None
	        contiguous_107: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_107);  transpose_107 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_107, [sym_size_int_2, 1500, -1]);  contiguous_107 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_159 = torch.ops.torchao.choose_qparams_affine.default(reshape_26, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_318: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_159[0]
	        getitem_319: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_159[1];  choose_qparams_affine_default_159 = None
	        quantize_affine_159: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_26, [1, 1, 1280], getitem_318, getitem_319, torch.int8);  reshape_26 = None
	        dequantize_affine_318: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_159, [1, 1, 1280], getitem_318, getitem_319, torch.int8);  quantize_affine_159 = getitem_318 = getitem_319 = None
	        dequantize_affine_319: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_159: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_318, dequantize_affine_319, model_audio_tower_layers_26_self_attn_out_proj_bias);  dequantize_affine_318 = dequantize_affine_319 = model_audio_tower_layers_26_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_79: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_159, 0.0, False);  linear_159 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_373: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_364, dropout_79);  add_364 = dropout_79 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_53: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_373, [1280], model_audio_tower_layers_26_final_layer_norm_weight, model_audio_tower_layers_26_final_layer_norm_bias);  model_audio_tower_layers_26_final_layer_norm_weight = model_audio_tower_layers_26_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_160 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_53, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_320: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_160[0]
	        getitem_321: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_160[1];  choose_qparams_affine_default_160 = None
	        quantize_affine_160: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_53, [1, 1, 1280], getitem_320, getitem_321, torch.int8);  layer_norm_53 = None
	        dequantize_affine_320: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_160, [1, 1, 1280], getitem_320, getitem_321, torch.int8);  quantize_affine_160 = getitem_320 = getitem_321 = None
	        dequantize_affine_321: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_26_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_26_fc1_parametrizations_weight_original1, model_audio_tower_layers_26_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_26_fc1_parametrizations_weight_original0 = model_audio_tower_layers_26_fc1_parametrizations_weight_original1 = model_audio_tower_layers_26_fc1_parametrizations_weight_original2 = None
	        linear_160: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_320, dequantize_affine_321, model_audio_tower_layers_26_fc1_bias);  dequantize_affine_320 = dequantize_affine_321 = model_audio_tower_layers_26_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_28: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_160);  linear_160 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_80: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_28, 0.0, False);  gelu_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_161 = torch.ops.torchao.choose_qparams_affine.default(dropout_80, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_322: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_161[0]
	        getitem_323: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_161[1];  choose_qparams_affine_default_161 = None
	        quantize_affine_161: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_80, [1, 1, 5120], getitem_322, getitem_323, torch.int8);  dropout_80 = None
	        dequantize_affine_322: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_161, [1, 1, 5120], getitem_322, getitem_323, torch.int8);  quantize_affine_161 = getitem_322 = getitem_323 = None
	        dequantize_affine_323: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_26_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_26_fc2_parametrizations_weight_original1, model_audio_tower_layers_26_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_26_fc2_parametrizations_weight_original0 = model_audio_tower_layers_26_fc2_parametrizations_weight_original1 = model_audio_tower_layers_26_fc2_parametrizations_weight_original2 = None
	        linear_161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_322, dequantize_affine_323, model_audio_tower_layers_26_fc2_bias);  dequantize_affine_322 = dequantize_affine_323 = model_audio_tower_layers_26_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_81: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_161, 0.0, False);  linear_161 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_378: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_373, dropout_81);  add_373 = dropout_81 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_54: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_378, [1280], model_audio_tower_layers_27_self_attn_layer_norm_weight, model_audio_tower_layers_27_self_attn_layer_norm_bias);  model_audio_tower_layers_27_self_attn_layer_norm_weight = model_audio_tower_layers_27_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_162 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_54, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_324: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_162[0]
	        getitem_325: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_162[1];  choose_qparams_affine_default_162 = None
	        quantize_affine_162: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_54, [1, 1, 1280], getitem_324, getitem_325, torch.int8)
	        dequantize_affine_324: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_162, [1, 1, 1280], getitem_324, getitem_325, torch.int8);  quantize_affine_162 = getitem_324 = getitem_325 = None
	        dequantize_affine_325: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_162: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_324, dequantize_affine_325, model_audio_tower_layers_27_self_attn_q_proj_bias);  dequantize_affine_324 = dequantize_affine_325 = model_audio_tower_layers_27_self_attn_q_proj_bias = None
	        mul_1004: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_162, 0.125);  linear_162 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_81: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_1004, [sym_size_int_2, 1500, 20, 64]);  mul_1004 = None
	        transpose_108: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_81, 1, 2);  view_81 = None
	        contiguous_108: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_108);  transpose_108 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_163 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_54, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_326: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_163[0]
	        getitem_327: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_163[1];  choose_qparams_affine_default_163 = None
	        quantize_affine_163: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_54, [1, 1, 1280], getitem_326, getitem_327, torch.int8)
	        dequantize_affine_326: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_163, [1, 1, 1280], getitem_326, getitem_327, torch.int8);  quantize_affine_163 = getitem_326 = getitem_327 = None
	        dequantize_affine_327: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_163: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_326, dequantize_affine_327);  dequantize_affine_326 = dequantize_affine_327 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_82: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_163, [sym_size_int_2, -1, 20, 64]);  linear_163 = None
	        transpose_109: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_82, 1, 2);  view_82 = None
	        contiguous_109: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_109);  transpose_109 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_164 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_54, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_328: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_164[0]
	        getitem_329: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_164[1];  choose_qparams_affine_default_164 = None
	        quantize_affine_164: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_54, [1, 1, 1280], getitem_328, getitem_329, torch.int8);  layer_norm_54 = None
	        dequantize_affine_328: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_164, [1, 1, 1280], getitem_328, getitem_329, torch.int8);  quantize_affine_164 = getitem_328 = getitem_329 = None
	        dequantize_affine_329: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_328, dequantize_affine_329, model_audio_tower_layers_27_self_attn_v_proj_bias);  dequantize_affine_328 = dequantize_affine_329 = model_audio_tower_layers_27_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_83: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_164, [sym_size_int_2, -1, 20, 64]);  linear_164 = None
	        transpose_110: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_83, 1, 2);  view_83 = None
	        contiguous_110: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_110);  transpose_110 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_27: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_108, contiguous_109, contiguous_110, scale = 1.0);  contiguous_108 = contiguous_109 = contiguous_110 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_111: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_27, 1, 2);  scaled_dot_product_attention_27 = None
	        contiguous_111: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_111);  transpose_111 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_27: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_111, [sym_size_int_2, 1500, -1]);  contiguous_111 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_165 = torch.ops.torchao.choose_qparams_affine.default(reshape_27, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_330: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_165[0]
	        getitem_331: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_165[1];  choose_qparams_affine_default_165 = None
	        quantize_affine_165: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_27, [1, 1, 1280], getitem_330, getitem_331, torch.int8);  reshape_27 = None
	        dequantize_affine_330: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_165, [1, 1, 1280], getitem_330, getitem_331, torch.int8);  quantize_affine_165 = getitem_330 = getitem_331 = None
	        dequantize_affine_331: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_165: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_330, dequantize_affine_331, model_audio_tower_layers_27_self_attn_out_proj_bias);  dequantize_affine_330 = dequantize_affine_331 = model_audio_tower_layers_27_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_82: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_165, 0.0, False);  linear_165 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_387: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_378, dropout_82);  add_378 = dropout_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_55: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_387, [1280], model_audio_tower_layers_27_final_layer_norm_weight, model_audio_tower_layers_27_final_layer_norm_bias);  model_audio_tower_layers_27_final_layer_norm_weight = model_audio_tower_layers_27_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_166 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_55, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_332: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_166[0]
	        getitem_333: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_166[1];  choose_qparams_affine_default_166 = None
	        quantize_affine_166: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_55, [1, 1, 1280], getitem_332, getitem_333, torch.int8);  layer_norm_55 = None
	        dequantize_affine_332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_166, [1, 1, 1280], getitem_332, getitem_333, torch.int8);  quantize_affine_166 = getitem_332 = getitem_333 = None
	        dequantize_affine_333: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_27_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_27_fc1_parametrizations_weight_original1, model_audio_tower_layers_27_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_27_fc1_parametrizations_weight_original0 = model_audio_tower_layers_27_fc1_parametrizations_weight_original1 = model_audio_tower_layers_27_fc1_parametrizations_weight_original2 = None
	        linear_166: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_332, dequantize_affine_333, model_audio_tower_layers_27_fc1_bias);  dequantize_affine_332 = dequantize_affine_333 = model_audio_tower_layers_27_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_29: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_166);  linear_166 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_83: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_29, 0.0, False);  gelu_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_167 = torch.ops.torchao.choose_qparams_affine.default(dropout_83, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_334: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_167[0]
	        getitem_335: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_167[1];  choose_qparams_affine_default_167 = None
	        quantize_affine_167: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_83, [1, 1, 5120], getitem_334, getitem_335, torch.int8);  dropout_83 = None
	        dequantize_affine_334: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_167, [1, 1, 5120], getitem_334, getitem_335, torch.int8);  quantize_affine_167 = getitem_334 = getitem_335 = None
	        dequantize_affine_335: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_27_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_27_fc2_parametrizations_weight_original1, model_audio_tower_layers_27_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_27_fc2_parametrizations_weight_original0 = model_audio_tower_layers_27_fc2_parametrizations_weight_original1 = model_audio_tower_layers_27_fc2_parametrizations_weight_original2 = None
	        linear_167: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_334, dequantize_affine_335, model_audio_tower_layers_27_fc2_bias);  dequantize_affine_334 = dequantize_affine_335 = model_audio_tower_layers_27_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_84: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_167, 0.0, False);  linear_167 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_392: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_387, dropout_84);  add_387 = dropout_84 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_392, [1280], model_audio_tower_layers_28_self_attn_layer_norm_weight, model_audio_tower_layers_28_self_attn_layer_norm_bias);  model_audio_tower_layers_28_self_attn_layer_norm_weight = model_audio_tower_layers_28_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_168 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_56, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_336: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_168[0]
	        getitem_337: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_168[1];  choose_qparams_affine_default_168 = None
	        quantize_affine_168: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_56, [1, 1, 1280], getitem_336, getitem_337, torch.int8)
	        dequantize_affine_336: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_168, [1, 1, 1280], getitem_336, getitem_337, torch.int8);  quantize_affine_168 = getitem_336 = getitem_337 = None
	        dequantize_affine_337: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_168: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_336, dequantize_affine_337, model_audio_tower_layers_28_self_attn_q_proj_bias);  dequantize_affine_336 = dequantize_affine_337 = model_audio_tower_layers_28_self_attn_q_proj_bias = None
	        mul_1041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_168, 0.125);  linear_168 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_84: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_1041, [sym_size_int_2, 1500, 20, 64]);  mul_1041 = None
	        transpose_112: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_84, 1, 2);  view_84 = None
	        contiguous_112: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_112);  transpose_112 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_169 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_56, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_338: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_169[0]
	        getitem_339: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_169[1];  choose_qparams_affine_default_169 = None
	        quantize_affine_169: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_56, [1, 1, 1280], getitem_338, getitem_339, torch.int8)
	        dequantize_affine_338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_169, [1, 1, 1280], getitem_338, getitem_339, torch.int8);  quantize_affine_169 = getitem_338 = getitem_339 = None
	        dequantize_affine_339: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_169: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_338, dequantize_affine_339);  dequantize_affine_338 = dequantize_affine_339 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_85: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_169, [sym_size_int_2, -1, 20, 64]);  linear_169 = None
	        transpose_113: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_85, 1, 2);  view_85 = None
	        contiguous_113: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_113);  transpose_113 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_170 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_56, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_340: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_170[0]
	        getitem_341: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_170[1];  choose_qparams_affine_default_170 = None
	        quantize_affine_170: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_56, [1, 1, 1280], getitem_340, getitem_341, torch.int8);  layer_norm_56 = None
	        dequantize_affine_340: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_170, [1, 1, 1280], getitem_340, getitem_341, torch.int8);  quantize_affine_170 = getitem_340 = getitem_341 = None
	        dequantize_affine_341: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_340, dequantize_affine_341, model_audio_tower_layers_28_self_attn_v_proj_bias);  dequantize_affine_340 = dequantize_affine_341 = model_audio_tower_layers_28_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_86: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_170, [sym_size_int_2, -1, 20, 64]);  linear_170 = None
	        transpose_114: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_86, 1, 2);  view_86 = None
	        contiguous_114: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_114);  transpose_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_28: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_112, contiguous_113, contiguous_114, scale = 1.0);  contiguous_112 = contiguous_113 = contiguous_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_115: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_28, 1, 2);  scaled_dot_product_attention_28 = None
	        contiguous_115: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_115);  transpose_115 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_28: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_115, [sym_size_int_2, 1500, -1]);  contiguous_115 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_171 = torch.ops.torchao.choose_qparams_affine.default(reshape_28, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_342: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_171[0]
	        getitem_343: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_171[1];  choose_qparams_affine_default_171 = None
	        quantize_affine_171: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_28, [1, 1, 1280], getitem_342, getitem_343, torch.int8);  reshape_28 = None
	        dequantize_affine_342: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_171, [1, 1, 1280], getitem_342, getitem_343, torch.int8);  quantize_affine_171 = getitem_342 = getitem_343 = None
	        dequantize_affine_343: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_171: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_342, dequantize_affine_343, model_audio_tower_layers_28_self_attn_out_proj_bias);  dequantize_affine_342 = dequantize_affine_343 = model_audio_tower_layers_28_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_85: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_171, 0.0, False);  linear_171 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_401: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_392, dropout_85);  add_392 = dropout_85 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_57: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_401, [1280], model_audio_tower_layers_28_final_layer_norm_weight, model_audio_tower_layers_28_final_layer_norm_bias);  model_audio_tower_layers_28_final_layer_norm_weight = model_audio_tower_layers_28_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_172 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_57, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_344: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_172[0]
	        getitem_345: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_172[1];  choose_qparams_affine_default_172 = None
	        quantize_affine_172: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_57, [1, 1, 1280], getitem_344, getitem_345, torch.int8);  layer_norm_57 = None
	        dequantize_affine_344: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_172, [1, 1, 1280], getitem_344, getitem_345, torch.int8);  quantize_affine_172 = getitem_344 = getitem_345 = None
	        dequantize_affine_345: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_28_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_28_fc1_parametrizations_weight_original1, model_audio_tower_layers_28_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_28_fc1_parametrizations_weight_original0 = model_audio_tower_layers_28_fc1_parametrizations_weight_original1 = model_audio_tower_layers_28_fc1_parametrizations_weight_original2 = None
	        linear_172: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_344, dequantize_affine_345, model_audio_tower_layers_28_fc1_bias);  dequantize_affine_344 = dequantize_affine_345 = model_audio_tower_layers_28_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_30: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_172);  linear_172 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_86: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_30, 0.0, False);  gelu_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_173 = torch.ops.torchao.choose_qparams_affine.default(dropout_86, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_346: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_173[0]
	        getitem_347: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_173[1];  choose_qparams_affine_default_173 = None
	        quantize_affine_173: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_86, [1, 1, 5120], getitem_346, getitem_347, torch.int8);  dropout_86 = None
	        dequantize_affine_346: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_173, [1, 1, 5120], getitem_346, getitem_347, torch.int8);  quantize_affine_173 = getitem_346 = getitem_347 = None
	        dequantize_affine_347: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_28_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_28_fc2_parametrizations_weight_original1, model_audio_tower_layers_28_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_28_fc2_parametrizations_weight_original0 = model_audio_tower_layers_28_fc2_parametrizations_weight_original1 = model_audio_tower_layers_28_fc2_parametrizations_weight_original2 = None
	        linear_173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_346, dequantize_affine_347, model_audio_tower_layers_28_fc2_bias);  dequantize_affine_346 = dequantize_affine_347 = model_audio_tower_layers_28_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_87: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_173, 0.0, False);  linear_173 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_406: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_401, dropout_87);  add_401 = dropout_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_58: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_406, [1280], model_audio_tower_layers_29_self_attn_layer_norm_weight, model_audio_tower_layers_29_self_attn_layer_norm_bias);  model_audio_tower_layers_29_self_attn_layer_norm_weight = model_audio_tower_layers_29_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_174 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_58, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_348: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_174[0]
	        getitem_349: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_174[1];  choose_qparams_affine_default_174 = None
	        quantize_affine_174: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_58, [1, 1, 1280], getitem_348, getitem_349, torch.int8)
	        dequantize_affine_348: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_174, [1, 1, 1280], getitem_348, getitem_349, torch.int8);  quantize_affine_174 = getitem_348 = getitem_349 = None
	        dequantize_affine_349: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_348, dequantize_affine_349, model_audio_tower_layers_29_self_attn_q_proj_bias);  dequantize_affine_348 = dequantize_affine_349 = model_audio_tower_layers_29_self_attn_q_proj_bias = None
	        mul_1078: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_174, 0.125);  linear_174 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_87: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_1078, [sym_size_int_2, 1500, 20, 64]);  mul_1078 = None
	        transpose_116: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_87, 1, 2);  view_87 = None
	        contiguous_116: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_116);  transpose_116 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_175 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_58, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_350: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_175[0]
	        getitem_351: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_175[1];  choose_qparams_affine_default_175 = None
	        quantize_affine_175: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_58, [1, 1, 1280], getitem_350, getitem_351, torch.int8)
	        dequantize_affine_350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_175, [1, 1, 1280], getitem_350, getitem_351, torch.int8);  quantize_affine_175 = getitem_350 = getitem_351 = None
	        dequantize_affine_351: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_175: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_350, dequantize_affine_351);  dequantize_affine_350 = dequantize_affine_351 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_88: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_175, [sym_size_int_2, -1, 20, 64]);  linear_175 = None
	        transpose_117: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_88, 1, 2);  view_88 = None
	        contiguous_117: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_117);  transpose_117 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_176 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_58, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_352: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_176[0]
	        getitem_353: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_176[1];  choose_qparams_affine_default_176 = None
	        quantize_affine_176: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_58, [1, 1, 1280], getitem_352, getitem_353, torch.int8);  layer_norm_58 = None
	        dequantize_affine_352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_176, [1, 1, 1280], getitem_352, getitem_353, torch.int8);  quantize_affine_176 = getitem_352 = getitem_353 = None
	        dequantize_affine_353: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_352, dequantize_affine_353, model_audio_tower_layers_29_self_attn_v_proj_bias);  dequantize_affine_352 = dequantize_affine_353 = model_audio_tower_layers_29_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_89: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_176, [sym_size_int_2, -1, 20, 64]);  linear_176 = None
	        transpose_118: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_89, 1, 2);  view_89 = None
	        contiguous_118: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_118);  transpose_118 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_29: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_116, contiguous_117, contiguous_118, scale = 1.0);  contiguous_116 = contiguous_117 = contiguous_118 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_119: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_29, 1, 2);  scaled_dot_product_attention_29 = None
	        contiguous_119: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_119);  transpose_119 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_29: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_119, [sym_size_int_2, 1500, -1]);  contiguous_119 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_177 = torch.ops.torchao.choose_qparams_affine.default(reshape_29, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_354: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_177[0]
	        getitem_355: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_177[1];  choose_qparams_affine_default_177 = None
	        quantize_affine_177: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_29, [1, 1, 1280], getitem_354, getitem_355, torch.int8);  reshape_29 = None
	        dequantize_affine_354: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_177, [1, 1, 1280], getitem_354, getitem_355, torch.int8);  quantize_affine_177 = getitem_354 = getitem_355 = None
	        dequantize_affine_355: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_177: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_354, dequantize_affine_355, model_audio_tower_layers_29_self_attn_out_proj_bias);  dequantize_affine_354 = dequantize_affine_355 = model_audio_tower_layers_29_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_88: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_177, 0.0, False);  linear_177 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_415: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_406, dropout_88);  add_406 = dropout_88 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_59: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_415, [1280], model_audio_tower_layers_29_final_layer_norm_weight, model_audio_tower_layers_29_final_layer_norm_bias);  model_audio_tower_layers_29_final_layer_norm_weight = model_audio_tower_layers_29_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_178 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_59, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_356: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_178[0]
	        getitem_357: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_178[1];  choose_qparams_affine_default_178 = None
	        quantize_affine_178: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_59, [1, 1, 1280], getitem_356, getitem_357, torch.int8);  layer_norm_59 = None
	        dequantize_affine_356: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_178, [1, 1, 1280], getitem_356, getitem_357, torch.int8);  quantize_affine_178 = getitem_356 = getitem_357 = None
	        dequantize_affine_357: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_29_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_29_fc1_parametrizations_weight_original1, model_audio_tower_layers_29_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_29_fc1_parametrizations_weight_original0 = model_audio_tower_layers_29_fc1_parametrizations_weight_original1 = model_audio_tower_layers_29_fc1_parametrizations_weight_original2 = None
	        linear_178: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_356, dequantize_affine_357, model_audio_tower_layers_29_fc1_bias);  dequantize_affine_356 = dequantize_affine_357 = model_audio_tower_layers_29_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_31: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_178);  linear_178 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_89: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_31, 0.0, False);  gelu_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_179 = torch.ops.torchao.choose_qparams_affine.default(dropout_89, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_358: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_179[0]
	        getitem_359: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_179[1];  choose_qparams_affine_default_179 = None
	        quantize_affine_179: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_89, [1, 1, 5120], getitem_358, getitem_359, torch.int8);  dropout_89 = None
	        dequantize_affine_358: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_179, [1, 1, 5120], getitem_358, getitem_359, torch.int8);  quantize_affine_179 = getitem_358 = getitem_359 = None
	        dequantize_affine_359: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_29_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_29_fc2_parametrizations_weight_original1, model_audio_tower_layers_29_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_29_fc2_parametrizations_weight_original0 = model_audio_tower_layers_29_fc2_parametrizations_weight_original1 = model_audio_tower_layers_29_fc2_parametrizations_weight_original2 = None
	        linear_179: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_358, dequantize_affine_359, model_audio_tower_layers_29_fc2_bias);  dequantize_affine_358 = dequantize_affine_359 = model_audio_tower_layers_29_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_90: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_179, 0.0, False);  linear_179 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_420: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_415, dropout_90);  add_415 = dropout_90 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_60: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_420, [1280], model_audio_tower_layers_30_self_attn_layer_norm_weight, model_audio_tower_layers_30_self_attn_layer_norm_bias);  model_audio_tower_layers_30_self_attn_layer_norm_weight = model_audio_tower_layers_30_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_180 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_60, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_360: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_180[0]
	        getitem_361: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_180[1];  choose_qparams_affine_default_180 = None
	        quantize_affine_180: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_60, [1, 1, 1280], getitem_360, getitem_361, torch.int8)
	        dequantize_affine_360: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_180, [1, 1, 1280], getitem_360, getitem_361, torch.int8);  quantize_affine_180 = getitem_360 = getitem_361 = None
	        dequantize_affine_361: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_180: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_360, dequantize_affine_361, model_audio_tower_layers_30_self_attn_q_proj_bias);  dequantize_affine_360 = dequantize_affine_361 = model_audio_tower_layers_30_self_attn_q_proj_bias = None
	        mul_1115: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_180, 0.125);  linear_180 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_90: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_1115, [sym_size_int_2, 1500, 20, 64]);  mul_1115 = None
	        transpose_120: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_90, 1, 2);  view_90 = None
	        contiguous_120: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_120);  transpose_120 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_181 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_60, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_362: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_181[0]
	        getitem_363: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_181[1];  choose_qparams_affine_default_181 = None
	        quantize_affine_181: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_60, [1, 1, 1280], getitem_362, getitem_363, torch.int8)
	        dequantize_affine_362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_181, [1, 1, 1280], getitem_362, getitem_363, torch.int8);  quantize_affine_181 = getitem_362 = getitem_363 = None
	        dequantize_affine_363: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_362, dequantize_affine_363);  dequantize_affine_362 = dequantize_affine_363 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_91: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_181, [sym_size_int_2, -1, 20, 64]);  linear_181 = None
	        transpose_121: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_91, 1, 2);  view_91 = None
	        contiguous_121: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_121);  transpose_121 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_182 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_60, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_364: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_182[0]
	        getitem_365: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_182[1];  choose_qparams_affine_default_182 = None
	        quantize_affine_182: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_60, [1, 1, 1280], getitem_364, getitem_365, torch.int8);  layer_norm_60 = None
	        dequantize_affine_364: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_182, [1, 1, 1280], getitem_364, getitem_365, torch.int8);  quantize_affine_182 = getitem_364 = getitem_365 = None
	        dequantize_affine_365: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_364, dequantize_affine_365, model_audio_tower_layers_30_self_attn_v_proj_bias);  dequantize_affine_364 = dequantize_affine_365 = model_audio_tower_layers_30_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_92: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_182, [sym_size_int_2, -1, 20, 64]);  linear_182 = None
	        transpose_122: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_92, 1, 2);  view_92 = None
	        contiguous_122: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_122);  transpose_122 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_30: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_120, contiguous_121, contiguous_122, scale = 1.0);  contiguous_120 = contiguous_121 = contiguous_122 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_123: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_30, 1, 2);  scaled_dot_product_attention_30 = None
	        contiguous_123: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_123);  transpose_123 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_123, [sym_size_int_2, 1500, -1]);  contiguous_123 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_183 = torch.ops.torchao.choose_qparams_affine.default(reshape_30, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_366: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_183[0]
	        getitem_367: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_183[1];  choose_qparams_affine_default_183 = None
	        quantize_affine_183: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_30, [1, 1, 1280], getitem_366, getitem_367, torch.int8);  reshape_30 = None
	        dequantize_affine_366: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_183, [1, 1, 1280], getitem_366, getitem_367, torch.int8);  quantize_affine_183 = getitem_366 = getitem_367 = None
	        dequantize_affine_367: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_183: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_366, dequantize_affine_367, model_audio_tower_layers_30_self_attn_out_proj_bias);  dequantize_affine_366 = dequantize_affine_367 = model_audio_tower_layers_30_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_91: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_183, 0.0, False);  linear_183 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_429: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_420, dropout_91);  add_420 = dropout_91 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_61: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_429, [1280], model_audio_tower_layers_30_final_layer_norm_weight, model_audio_tower_layers_30_final_layer_norm_bias);  model_audio_tower_layers_30_final_layer_norm_weight = model_audio_tower_layers_30_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_184 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_61, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_368: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_184[0]
	        getitem_369: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_184[1];  choose_qparams_affine_default_184 = None
	        quantize_affine_184: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_61, [1, 1, 1280], getitem_368, getitem_369, torch.int8);  layer_norm_61 = None
	        dequantize_affine_368: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_184, [1, 1, 1280], getitem_368, getitem_369, torch.int8);  quantize_affine_184 = getitem_368 = getitem_369 = None
	        dequantize_affine_369: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_30_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_30_fc1_parametrizations_weight_original1, model_audio_tower_layers_30_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_30_fc1_parametrizations_weight_original0 = model_audio_tower_layers_30_fc1_parametrizations_weight_original1 = model_audio_tower_layers_30_fc1_parametrizations_weight_original2 = None
	        linear_184: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_368, dequantize_affine_369, model_audio_tower_layers_30_fc1_bias);  dequantize_affine_368 = dequantize_affine_369 = model_audio_tower_layers_30_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_32: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_184);  linear_184 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_92: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_32, 0.0, False);  gelu_32 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_185 = torch.ops.torchao.choose_qparams_affine.default(dropout_92, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_370: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_185[0]
	        getitem_371: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_185[1];  choose_qparams_affine_default_185 = None
	        quantize_affine_185: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_92, [1, 1, 5120], getitem_370, getitem_371, torch.int8);  dropout_92 = None
	        dequantize_affine_370: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_185, [1, 1, 5120], getitem_370, getitem_371, torch.int8);  quantize_affine_185 = getitem_370 = getitem_371 = None
	        dequantize_affine_371: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_30_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_30_fc2_parametrizations_weight_original1, model_audio_tower_layers_30_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_30_fc2_parametrizations_weight_original0 = model_audio_tower_layers_30_fc2_parametrizations_weight_original1 = model_audio_tower_layers_30_fc2_parametrizations_weight_original2 = None
	        linear_185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_370, dequantize_affine_371, model_audio_tower_layers_30_fc2_bias);  dequantize_affine_370 = dequantize_affine_371 = model_audio_tower_layers_30_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_93: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_185, 0.0, False);  linear_185 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_434: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_429, dropout_93);  add_429 = dropout_93 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        layer_norm_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_434, [1280], model_audio_tower_layers_31_self_attn_layer_norm_weight, model_audio_tower_layers_31_self_attn_layer_norm_bias);  model_audio_tower_layers_31_self_attn_layer_norm_weight = model_audio_tower_layers_31_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        choose_qparams_affine_default_186 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_62, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_372: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_186[0]
	        getitem_373: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_186[1];  choose_qparams_affine_default_186 = None
	        quantize_affine_186: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_62, [1, 1, 1280], getitem_372, getitem_373, torch.int8)
	        dequantize_affine_372: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_186, [1, 1, 1280], getitem_372, getitem_373, torch.int8);  quantize_affine_186 = getitem_372 = getitem_373 = None
	        dequantize_affine_373: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1, model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0 = model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1 = model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2 = None
	        linear_186: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_372, dequantize_affine_373, model_audio_tower_layers_31_self_attn_q_proj_bias);  dequantize_affine_372 = dequantize_affine_373 = model_audio_tower_layers_31_self_attn_q_proj_bias = None
	        mul_1152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(linear_186, 0.125);  linear_186 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_93: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_1152, [sym_size_int_2, 1500, 20, 64]);  mul_1152 = None
	        transpose_124: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_93, 1, 2);  view_93 = None
	        contiguous_124: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_124);  transpose_124 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_187 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_62, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_374: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_187[0]
	        getitem_375: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_187[1];  choose_qparams_affine_default_187 = None
	        quantize_affine_187: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_62, [1, 1, 1280], getitem_374, getitem_375, torch.int8)
	        dequantize_affine_374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_187, [1, 1, 1280], getitem_374, getitem_375, torch.int8);  quantize_affine_187 = getitem_374 = getitem_375 = None
	        dequantize_affine_375: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1, model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0 = model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1 = model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2 = None
	        linear_187: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_374, dequantize_affine_375);  dequantize_affine_374 = dequantize_affine_375 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_94: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_187, [sym_size_int_2, -1, 20, 64]);  linear_187 = None
	        transpose_125: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_94, 1, 2);  view_94 = None
	        contiguous_125: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_125);  transpose_125 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        choose_qparams_affine_default_188 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_62, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_376: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_188[0]
	        getitem_377: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_188[1];  choose_qparams_affine_default_188 = None
	        quantize_affine_188: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_62, [1, 1, 1280], getitem_376, getitem_377, torch.int8);  layer_norm_62 = None
	        dequantize_affine_376: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_188, [1, 1, 1280], getitem_376, getitem_377, torch.int8);  quantize_affine_188 = getitem_376 = getitem_377 = None
	        dequantize_affine_377: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1, model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0 = model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1 = model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2 = None
	        linear_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_376, dequantize_affine_377, model_audio_tower_layers_31_self_attn_v_proj_bias);  dequantize_affine_376 = dequantize_affine_377 = model_audio_tower_layers_31_self_attn_v_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_95: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(linear_188, [sym_size_int_2, -1, 20, 64]);  linear_188 = None
	        transpose_126: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.transpose.int(view_95, 1, 2);  view_95 = None
	        contiguous_126: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_126);  transpose_126 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        scaled_dot_product_attention_31: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.scaled_dot_product_attention.default(contiguous_124, contiguous_125, contiguous_126, scale = 1.0);  contiguous_124 = contiguous_125 = contiguous_126 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        transpose_127: "f32[s6, 1500, 20, 64][1920000, 64, 96000, 1]cuda:0" = torch.ops.aten.transpose.int(scaled_dot_product_attention_31, 1, 2);  scaled_dot_product_attention_31 = None
	        contiguous_127: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.contiguous.default(transpose_127);  transpose_127 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        reshape_31: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(contiguous_127, [sym_size_int_2, 1500, -1]);  contiguous_127 = sym_size_int_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        choose_qparams_affine_default_189 = torch.ops.torchao.choose_qparams_affine.default(reshape_31, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_378: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_189[0]
	        getitem_379: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_189[1];  choose_qparams_affine_default_189 = None
	        quantize_affine_189: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_31, [1, 1, 1280], getitem_378, getitem_379, torch.int8);  reshape_31 = None
	        dequantize_affine_378: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_189, [1, 1, 1280], getitem_378, getitem_379, torch.int8);  quantize_affine_189 = getitem_378 = getitem_379 = None
	        dequantize_affine_379: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1, model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0 = model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1 = model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2 = None
	        linear_189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_378, dequantize_affine_379, model_audio_tower_layers_31_self_attn_out_proj_bias);  dequantize_affine_378 = dequantize_affine_379 = model_audio_tower_layers_31_self_attn_out_proj_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_94: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_189, 0.0, False);  linear_189 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_443: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_434, dropout_94);  add_434 = dropout_94 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        layer_norm_63: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_443, [1280], model_audio_tower_layers_31_final_layer_norm_weight, model_audio_tower_layers_31_final_layer_norm_bias);  model_audio_tower_layers_31_final_layer_norm_weight = model_audio_tower_layers_31_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        choose_qparams_affine_default_190 = torch.ops.torchao.choose_qparams_affine.default(layer_norm_63, 'ASYMMETRIC', [1, 1, 1280], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_380: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_190[0]
	        getitem_381: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_190[1];  choose_qparams_affine_default_190 = None
	        quantize_affine_190: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(layer_norm_63, [1, 1, 1280], getitem_380, getitem_381, torch.int8);  layer_norm_63 = None
	        dequantize_affine_380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_190, [1, 1, 1280], getitem_380, getitem_381, torch.int8);  quantize_affine_190 = getitem_380 = getitem_381 = None
	        dequantize_affine_381: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_31_fc1_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_31_fc1_parametrizations_weight_original1, model_audio_tower_layers_31_fc1_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_31_fc1_parametrizations_weight_original0 = model_audio_tower_layers_31_fc1_parametrizations_weight_original1 = model_audio_tower_layers_31_fc1_parametrizations_weight_original2 = None
	        linear_190: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_380, dequantize_affine_381, model_audio_tower_layers_31_fc1_bias);  dequantize_affine_380 = dequantize_affine_381 = model_audio_tower_layers_31_fc1_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_33: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.gelu.default(linear_190);  linear_190 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        dropout_95: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.dropout.default(gelu_33, 0.0, False);  gelu_33 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        choose_qparams_affine_default_191 = torch.ops.torchao.choose_qparams_affine.default(dropout_95, 'ASYMMETRIC', [1, 1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_382: "f32[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_191[0]
	        getitem_383: "i8[s6, 1500][1500, 1]cuda:0" = choose_qparams_affine_default_191[1];  choose_qparams_affine_default_191 = None
	        quantize_affine_191: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(dropout_95, [1, 1, 5120], getitem_382, getitem_383, torch.int8);  dropout_95 = None
	        dequantize_affine_382: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_191, [1, 1, 5120], getitem_382, getitem_383, torch.int8);  quantize_affine_191 = getitem_382 = getitem_383 = None
	        dequantize_affine_383: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_audio_tower_layers_31_fc2_parametrizations_weight_original0, [1, 32], model_audio_tower_layers_31_fc2_parametrizations_weight_original1, model_audio_tower_layers_31_fc2_parametrizations_weight_original2, torch.int8, -8, 7);  model_audio_tower_layers_31_fc2_parametrizations_weight_original0 = model_audio_tower_layers_31_fc2_parametrizations_weight_original1 = model_audio_tower_layers_31_fc2_parametrizations_weight_original2 = None
	        linear_191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_382, dequantize_affine_383, model_audio_tower_layers_31_fc2_bias);  dequantize_affine_382 = dequantize_affine_383 = model_audio_tower_layers_31_fc2_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        dropout_96: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.dropout.default(linear_191, 0.0, False);  linear_191 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_448: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_443, dropout_96);  add_443 = dropout_96 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:365 in forward, code: hidden_states = self.layer_norm(hidden_states)
	        layer_norm_64: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.layer_norm.default(add_448, [1280], model_audio_tower_layer_norm_weight, model_audio_tower_layer_norm_bias);  add_448 = model_audio_tower_layer_norm_weight = model_audio_tower_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:451 in get_audio_embeds, code: audio_hidden_states = audio_hidden_states.reshape(-1, self.config.audio_config.intermediate_size)
	        reshape_32: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(layer_norm_64, [-1, 5120]);  layer_norm_64 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:389 in forward, code: hidden_states = self.linear_1(audio_features)
	        choose_qparams_affine_default_192 = torch.ops.torchao.choose_qparams_affine.default(reshape_32, 'ASYMMETRIC', [1, 5120], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_384: "f32[375*s6][1]cuda:0" = choose_qparams_affine_default_192[0]
	        getitem_385: "i8[375*s6][1]cuda:0" = choose_qparams_affine_default_192[1];  choose_qparams_affine_default_192 = None
	        quantize_affine_192: "i8[375*s6, 5120][5120, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(reshape_32, [1, 5120], getitem_384, getitem_385, torch.int8);  reshape_32 = None
	        dequantize_affine_384: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_192, [1, 5120], getitem_384, getitem_385, torch.int8);  quantize_affine_192 = getitem_384 = getitem_385 = None
	        dequantize_affine_385: "f32[3072, 5120][5120, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_multi_modal_projector_linear_1_parametrizations_weight_original0, [1, 32], model_multi_modal_projector_linear_1_parametrizations_weight_original1, model_multi_modal_projector_linear_1_parametrizations_weight_original2, torch.int8, -8, 7);  model_multi_modal_projector_linear_1_parametrizations_weight_original0 = model_multi_modal_projector_linear_1_parametrizations_weight_original1 = model_multi_modal_projector_linear_1_parametrizations_weight_original2 = None
	        linear_192: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_384, dequantize_affine_385);  dequantize_affine_384 = dequantize_affine_385 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        gelu_34: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.gelu.default(linear_192);  linear_192 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:391 in forward, code: hidden_states = self.linear_2(hidden_states)
	        choose_qparams_affine_default_193 = torch.ops.torchao.choose_qparams_affine.default(gelu_34, 'ASYMMETRIC', [1, 3072], torch.int8, None, None, 1.1920928955078125e-07, torch.float32, torch.int8)
	        getitem_386: "f32[375*s6][1]cuda:0" = choose_qparams_affine_default_193[0]
	        getitem_387: "i8[375*s6][1]cuda:0" = choose_qparams_affine_default_193[1];  choose_qparams_affine_default_193 = None
	        quantize_affine_193: "i8[375*s6, 3072][3072, 1]cuda:0" = torch.ops.torchao.quantize_affine.default(gelu_34, [1, 3072], getitem_386, getitem_387, torch.int8);  gelu_34 = None
	        dequantize_affine_386: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(quantize_affine_193, [1, 3072], getitem_386, getitem_387, torch.int8);  quantize_affine_193 = getitem_386 = getitem_387 = None
	        dequantize_affine_387: "f32[3072, 3072][3072, 1]cuda:0" = torch.ops.torchao.dequantize_affine.default(model_multi_modal_projector_linear_2_parametrizations_weight_original0, [1, 32], model_multi_modal_projector_linear_2_parametrizations_weight_original1, model_multi_modal_projector_linear_2_parametrizations_weight_original2, torch.int8, -8, 7);  model_multi_modal_projector_linear_2_parametrizations_weight_original0 = model_multi_modal_projector_linear_2_parametrizations_weight_original1 = model_multi_modal_projector_linear_2_parametrizations_weight_original2 = None
	        linear_193: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.linear.default(dequantize_affine_386, dequantize_affine_387);  dequantize_affine_386 = dequantize_affine_387 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py:83 in forward, code: return audio_embeds.unsqueeze(0)
	        unsqueeze: "f32[1, 375*s6, 3072][1152000*s6, 3072, 1]cuda:0" = torch.ops.aten.unsqueeze.default(linear_193, 0);  linear_193 = None
	        return (unsqueeze,)
	        
	
	 # graph id: 140612625281824
V0910 09:41:49.330000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e6725e2980cf67cbb5065c00c3c5735d"}
	{
	"name": "create_aot_dispatcher_function",
	"ts": 1757522509330215.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:41:49.340000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "68b7e4e662f6f3138569c683fb0deb42"}
	{
	"name": "aot_collect_metadata",
	"ts": 1757522509340186.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:41:49.907000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_functorch/aot_autograd.py", 20]}
V0910 09:41:49.908000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_functorch/_aot_autograd/collect_metadata_analysis.py", 21]}
V0910 09:41:49.909000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_functorch/_aot_autograd/utils.py", 22]}
V0910 09:41:49.910000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_functorch/_aot_autograd/graph_capture_wrappers.py", 23]}
V0910 09:41:49.911000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/fx/interpreter.py", 24]}
V0910 09:41:49.912000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_subclasses/functional_tensor.py", 25]}
V0910 09:41:49.912000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_subclasses/fake_impls.py", 26]}
V0910 09:41:49.913000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_compile.py", 27]}
V0910 09:41:49.914000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_dynamo/eval_frame.py", 28]}
V0910 09:41:49.915000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_refs/__init__.py", 29]}
V0910 09:41:49.916000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/fx/experimental/sym_node.py", 30]}
V0910 09:41:49.917000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/fx/experimental/symbolic_shapes.py:7190] {"guard_added_fast": {"expr": "Ne(s6, 1)", "user_stack": [], "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2604, "name": "compile_fx", "filename": 19, "loc": "gm, graph_signature = aot_export_module("}, {"line": 1444, "name": "aot_export_module", "filename": 20, "loc": "fx_g, metadata, in_spec, out_spec = _aot_export_function("}, {"line": 1694, "name": "_aot_export_function", "filename": 20, "loc": "aot_state = create_aot_state("}, {"line": 567, "name": "create_aot_state", "filename": 20, "loc": "fw_metadata = run_functionalized_fw_and_collect_metadata("}, {"line": 207, "name": "inner", "filename": 21, "loc": "flat_f_outs = f(*flat_f_args)"}, {"line": 187, "name": "flat_fn", "filename": 22, "loc": "tree_out = fn(*args, **kwargs)"}, {"line": 1350, "name": "functional_call", "filename": 23, "loc": "out = PropagateUnbackedSymInts(mod).run("}, {"line": 174, "name": "run", "filename": 24, "loc": "self.env[node] = self.run_node(node)"}, {"line": 7870, "name": "run_node", "filename": 14, "loc": "result = super().run_node(n)"}, {"line": 256, "name": "run_node", "filename": 24, "loc": "return getattr(self, n.op)(n.target, args, kwargs)"}, {"line": 336, "name": "call_function", "filename": 24, "loc": "return target(*args, **kwargs)"}, {"line": 841, "name": "__call__", "filename": 9, "loc": "return self._op(*args, **kwargs)"}, {"line": 511, "name": "__torch_dispatch__", "filename": 25, "loc": "outs_unwrapped = func._op_dk("}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1511, "name": "_cached_dispatch_impl", "filename": 8, "loc": "output = self._dispatch_impl(func, types, args, kwargs)"}, {"line": 2611, "name": "_dispatch_impl", "filename": 8, "loc": "return maybe_propagate_real_tensors(fast_impl(self, *args, **kwargs))"}, {"line": 1296, "name": "fast_binary_impl", "filename": 26, "loc": "return slow(\"no contiguity match\")"}, {"line": 1165, "name": "slow", "filename": 26, "loc": "return slow_ref(*args, **kwargs)"}, {"line": 309, "name": "_fn", "filename": 10, "loc": "result = fn(*args, **kwargs)"}, {"line": 53, "name": "inner", "filename": 27, "loc": "return disable_fn(*args, **kwargs)"}, {"line": 1044, "name": "_fn", "filename": 28, "loc": "return fn(*args, **kwargs)"}, {"line": 149, "name": "_fn", "filename": 10, "loc": "result = fn(**bound.arguments)"}, {"line": 1141, "name": "add", "filename": 29, "loc": "a, b = _maybe_broadcast(a, b)"}, {"line": 470, "name": "_maybe_broadcast", "filename": 29, "loc": "return tuple(__maybe_broadcast(x, common_shape) for x in args)"}, {"line": 470, "name": "<genexpr>", "filename": 29, "loc": "return tuple(__maybe_broadcast(x, common_shape) for x in args)"}, {"line": 462, "name": "__maybe_broadcast", "filename": 29, "loc": "return x.expand(common_shape)"}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1511, "name": "_cached_dispatch_impl", "filename": 8, "loc": "output = self._dispatch_impl(func, types, args, kwargs)"}, {"line": 2635, "name": "_dispatch_impl", "filename": 8, "loc": "decomposition_table[func](*args, **kwargs)"}, {"line": 3061, "name": "expand", "filename": 29, "loc": "return prims.broadcast_in_dim("}, {"line": 841, "name": "__call__", "filename": 9, "loc": "return self._op(*args, **kwargs)"}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1511, "name": "_cached_dispatch_impl", "filename": 8, "loc": "output = self._dispatch_impl(func, types, args, kwargs)"}, {"line": 2657, "name": "_dispatch_impl", "filename": 8, "loc": "func.prim_meta_impl(*args, **kwargs)"}, {"line": 1323, "name": "_broadcast_in_dim_meta", "filename": 12, "loc": "return a.as_strided(shape, new_strides, a.storage_offset())"}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1533, "name": "_cached_dispatch_impl", "filename": 8, "loc": "entry = self._make_cache_entry(state, key, func, args, kwargs, output)"}, {"line": 1912, "name": "_make_cache_entry", "filename": 8, "loc": "output_info = self._get_output_info_for_cache_entry("}, {"line": 1828, "name": "_get_output_info_for_cache_entry", "filename": 8, "loc": "synth_output = self._output_from_cache_entry("}, {"line": 2013, "name": "_output_from_cache_entry", "filename": 8, "loc": "return self._get_output_tensor_from_cache_entry("}, {"line": 1988, "name": "_get_output_tensor_from_cache_entry", "filename": 8, "loc": "empty.set_(storage, storage_offset, shape, stride)"}, {"line": 607, "name": "guard_or_false", "filename": 30, "loc": "return guard_or_false(SymBool(self))"}, {"line": 1406, "name": "guard_or_false", "filename": 14, "loc": "return _guard_or(a, False)"}, {"line": 1396, "name": "_guard_or", "filename": 14, "loc": "r = sym_node.shape_env.evaluate_sym_node("}, {"line": 7239, "name": "evaluate_sym_node", "filename": 14, "loc": "return self.evaluate_expr("}, {"line": 7339, "name": "evaluate_expr", "filename": 14, "loc": "return self._inner_evaluate_expr("}, {"line": 272, "name": "wrapper", "filename": 15, "loc": "return retlog(fn(*args, **kwargs))"}, {"line": 7362, "name": "_inner_evaluate_expr", "filename": 14, "loc": "return self._evaluate_expr("}, {"line": 7644, "name": "_evaluate_expr", "filename": 14, "loc": "self._log_guard(\"eval [guard suppressed]\", g, forcing_spec=forcing_spec)"}, {"line": 7190, "name": "_log_guard", "filename": 14, "loc": "trace_structured("}, {"line": 1346, "name": "trace_structured", "filename": 16, "loc": "record[name] = metadata_fn()"}]}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2604, "name": "compile_fx", "filename": 19, "loc": "gm, graph_signature = aot_export_module("}, {"line": 1444, "name": "aot_export_module", "filename": 20, "loc": "fx_g, metadata, in_spec, out_spec = _aot_export_function("}, {"line": 1694, "name": "_aot_export_function", "filename": 20, "loc": "aot_state = create_aot_state("}, {"line": 567, "name": "create_aot_state", "filename": 20, "loc": "fw_metadata = run_functionalized_fw_and_collect_metadata("}, {"line": 207, "name": "inner", "filename": 21, "loc": "flat_f_outs = f(*flat_f_args)"}, {"line": 187, "name": "flat_fn", "filename": 22, "loc": "tree_out = fn(*args, **kwargs)"}, {"line": 1350, "name": "functional_call", "filename": 23, "loc": "out = PropagateUnbackedSymInts(mod).run("}, {"line": 174, "name": "run", "filename": 24, "loc": "self.env[node] = self.run_node(node)"}, {"line": 7870, "name": "run_node", "filename": 14, "loc": "result = super().run_node(n)"}, {"line": 256, "name": "run_node", "filename": 24, "loc": "return getattr(self, n.op)(n.target, args, kwargs)"}, {"line": 336, "name": "call_function", "filename": 24, "loc": "return target(*args, **kwargs)"}, {"line": 841, "name": "__call__", "filename": 9, "loc": "return self._op(*args, **kwargs)"}, {"line": 511, "name": "__torch_dispatch__", "filename": 25, "loc": "outs_unwrapped = func._op_dk("}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1511, "name": "_cached_dispatch_impl", "filename": 8, "loc": "output = self._dispatch_impl(func, types, args, kwargs)"}, {"line": 2611, "name": "_dispatch_impl", "filename": 8, "loc": "return maybe_propagate_real_tensors(fast_impl(self, *args, **kwargs))"}, {"line": 1296, "name": "fast_binary_impl", "filename": 26, "loc": "return slow(\"no contiguity match\")"}, {"line": 1165, "name": "slow", "filename": 26, "loc": "return slow_ref(*args, **kwargs)"}, {"line": 309, "name": "_fn", "filename": 10, "loc": "result = fn(*args, **kwargs)"}, {"line": 53, "name": "inner", "filename": 27, "loc": "return disable_fn(*args, **kwargs)"}, {"line": 1044, "name": "_fn", "filename": 28, "loc": "return fn(*args, **kwargs)"}, {"line": 149, "name": "_fn", "filename": 10, "loc": "result = fn(**bound.arguments)"}, {"line": 1141, "name": "add", "filename": 29, "loc": "a, b = _maybe_broadcast(a, b)"}, {"line": 470, "name": "_maybe_broadcast", "filename": 29, "loc": "return tuple(__maybe_broadcast(x, common_shape) for x in args)"}, {"line": 470, "name": "<genexpr>", "filename": 29, "loc": "return tuple(__maybe_broadcast(x, common_shape) for x in args)"}, {"line": 462, "name": "__maybe_broadcast", "filename": 29, "loc": "return x.expand(common_shape)"}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1511, "name": "_cached_dispatch_impl", "filename": 8, "loc": "output = self._dispatch_impl(func, types, args, kwargs)"}, {"line": 2635, "name": "_dispatch_impl", "filename": 8, "loc": "decomposition_table[func](*args, **kwargs)"}, {"line": 3061, "name": "expand", "filename": 29, "loc": "return prims.broadcast_in_dim("}, {"line": 841, "name": "__call__", "filename": 9, "loc": "return self._op(*args, **kwargs)"}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1511, "name": "_cached_dispatch_impl", "filename": 8, "loc": "output = self._dispatch_impl(func, types, args, kwargs)"}, {"line": 2657, "name": "_dispatch_impl", "filename": 8, "loc": "func.prim_meta_impl(*args, **kwargs)"}, {"line": 1323, "name": "_broadcast_in_dim_meta", "filename": 12, "loc": "return a.as_strided(shape, new_strides, a.storage_offset())"}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1533, "name": "_cached_dispatch_impl", "filename": 8, "loc": "entry = self._make_cache_entry(state, key, func, args, kwargs, output)"}, {"line": 1912, "name": "_make_cache_entry", "filename": 8, "loc": "output_info = self._get_output_info_for_cache_entry("}, {"line": 1828, "name": "_get_output_info_for_cache_entry", "filename": 8, "loc": "synth_output = self._output_from_cache_entry("}, {"line": 2013, "name": "_output_from_cache_entry", "filename": 8, "loc": "return self._get_output_tensor_from_cache_entry("}, {"line": 1988, "name": "_get_output_tensor_from_cache_entry", "filename": 8, "loc": "empty.set_(storage, storage_offset, shape, stride)"}, {"line": 607, "name": "guard_or_false", "filename": 30, "loc": "return guard_or_false(SymBool(self))"}, {"line": 1406, "name": "guard_or_false", "filename": 14, "loc": "return _guard_or(a, False)"}, {"line": 1396, "name": "_guard_or", "filename": 14, "loc": "r = sym_node.shape_env.evaluate_sym_node("}, {"line": 7239, "name": "evaluate_sym_node", "filename": 14, "loc": "return self.evaluate_expr("}, {"line": 7339, "name": "evaluate_expr", "filename": 14, "loc": "return self._inner_evaluate_expr("}, {"line": 272, "name": "wrapper", "filename": 15, "loc": "return retlog(fn(*args, **kwargs))"}, {"line": 7362, "name": "_inner_evaluate_expr", "filename": 14, "loc": "return self._evaluate_expr("}, {"line": 7644, "name": "_evaluate_expr", "filename": 14, "loc": "self._log_guard(\"eval [guard suppressed]\", g, forcing_spec=forcing_spec)"}, {"line": 7190, "name": "_log_guard", "filename": 14, "loc": "trace_structured("}]}
V0910 09:41:49.920000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/fx/experimental/symbolic_shapes.py:7190] {"guard_added_fast": {"expr": "Ne(s6, 1)", "user_stack": [], "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2604, "name": "compile_fx", "filename": 19, "loc": "gm, graph_signature = aot_export_module("}, {"line": 1444, "name": "aot_export_module", "filename": 20, "loc": "fx_g, metadata, in_spec, out_spec = _aot_export_function("}, {"line": 1694, "name": "_aot_export_function", "filename": 20, "loc": "aot_state = create_aot_state("}, {"line": 567, "name": "create_aot_state", "filename": 20, "loc": "fw_metadata = run_functionalized_fw_and_collect_metadata("}, {"line": 207, "name": "inner", "filename": 21, "loc": "flat_f_outs = f(*flat_f_args)"}, {"line": 187, "name": "flat_fn", "filename": 22, "loc": "tree_out = fn(*args, **kwargs)"}, {"line": 1350, "name": "functional_call", "filename": 23, "loc": "out = PropagateUnbackedSymInts(mod).run("}, {"line": 174, "name": "run", "filename": 24, "loc": "self.env[node] = self.run_node(node)"}, {"line": 7870, "name": "run_node", "filename": 14, "loc": "result = super().run_node(n)"}, {"line": 256, "name": "run_node", "filename": 24, "loc": "return getattr(self, n.op)(n.target, args, kwargs)"}, {"line": 336, "name": "call_function", "filename": 24, "loc": "return target(*args, **kwargs)"}, {"line": 841, "name": "__call__", "filename": 9, "loc": "return self._op(*args, **kwargs)"}, {"line": 511, "name": "__torch_dispatch__", "filename": 25, "loc": "outs_unwrapped = func._op_dk("}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1511, "name": "_cached_dispatch_impl", "filename": 8, "loc": "output = self._dispatch_impl(func, types, args, kwargs)"}, {"line": 2611, "name": "_dispatch_impl", "filename": 8, "loc": "return maybe_propagate_real_tensors(fast_impl(self, *args, **kwargs))"}, {"line": 1296, "name": "fast_binary_impl", "filename": 26, "loc": "return slow(\"no contiguity match\")"}, {"line": 1165, "name": "slow", "filename": 26, "loc": "return slow_ref(*args, **kwargs)"}, {"line": 309, "name": "_fn", "filename": 10, "loc": "result = fn(*args, **kwargs)"}, {"line": 53, "name": "inner", "filename": 27, "loc": "return disable_fn(*args, **kwargs)"}, {"line": 1044, "name": "_fn", "filename": 28, "loc": "return fn(*args, **kwargs)"}, {"line": 149, "name": "_fn", "filename": 10, "loc": "result = fn(**bound.arguments)"}, {"line": 1141, "name": "add", "filename": 29, "loc": "a, b = _maybe_broadcast(a, b)"}, {"line": 470, "name": "_maybe_broadcast", "filename": 29, "loc": "return tuple(__maybe_broadcast(x, common_shape) for x in args)"}, {"line": 470, "name": "<genexpr>", "filename": 29, "loc": "return tuple(__maybe_broadcast(x, common_shape) for x in args)"}, {"line": 462, "name": "__maybe_broadcast", "filename": 29, "loc": "return x.expand(common_shape)"}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1511, "name": "_cached_dispatch_impl", "filename": 8, "loc": "output = self._dispatch_impl(func, types, args, kwargs)"}, {"line": 2635, "name": "_dispatch_impl", "filename": 8, "loc": "decomposition_table[func](*args, **kwargs)"}, {"line": 3061, "name": "expand", "filename": 29, "loc": "return prims.broadcast_in_dim("}, {"line": 841, "name": "__call__", "filename": 9, "loc": "return self._op(*args, **kwargs)"}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1511, "name": "_cached_dispatch_impl", "filename": 8, "loc": "output = self._dispatch_impl(func, types, args, kwargs)"}, {"line": 2657, "name": "_dispatch_impl", "filename": 8, "loc": "func.prim_meta_impl(*args, **kwargs)"}, {"line": 1323, "name": "_broadcast_in_dim_meta", "filename": 12, "loc": "return a.as_strided(shape, new_strides, a.storage_offset())"}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1533, "name": "_cached_dispatch_impl", "filename": 8, "loc": "entry = self._make_cache_entry(state, key, func, args, kwargs, output)"}, {"line": 1912, "name": "_make_cache_entry", "filename": 8, "loc": "output_info = self._get_output_info_for_cache_entry("}, {"line": 1828, "name": "_get_output_info_for_cache_entry", "filename": 8, "loc": "synth_output = self._output_from_cache_entry("}, {"line": 2013, "name": "_output_from_cache_entry", "filename": 8, "loc": "return self._get_output_tensor_from_cache_entry("}, {"line": 1988, "name": "_get_output_tensor_from_cache_entry", "filename": 8, "loc": "empty.set_(storage, storage_offset, shape, stride)"}, {"line": 538, "name": "guard_bool", "filename": 30, "loc": "r = self.evaluate()"}, {"line": 512, "name": "evaluate", "filename": 30, "loc": "return self.shape_env.evaluate_sym_node(self, size_oblivious)"}, {"line": 7239, "name": "evaluate_sym_node", "filename": 14, "loc": "return self.evaluate_expr("}, {"line": 7339, "name": "evaluate_expr", "filename": 14, "loc": "return self._inner_evaluate_expr("}, {"line": 272, "name": "wrapper", "filename": 15, "loc": "return retlog(fn(*args, **kwargs))"}, {"line": 7362, "name": "_inner_evaluate_expr", "filename": 14, "loc": "return self._evaluate_expr("}, {"line": 7644, "name": "_evaluate_expr", "filename": 14, "loc": "self._log_guard(\"eval [guard suppressed]\", g, forcing_spec=forcing_spec)"}, {"line": 7190, "name": "_log_guard", "filename": 14, "loc": "trace_structured("}, {"line": 1346, "name": "trace_structured", "filename": 16, "loc": "record[name] = metadata_fn()"}]}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2604, "name": "compile_fx", "filename": 19, "loc": "gm, graph_signature = aot_export_module("}, {"line": 1444, "name": "aot_export_module", "filename": 20, "loc": "fx_g, metadata, in_spec, out_spec = _aot_export_function("}, {"line": 1694, "name": "_aot_export_function", "filename": 20, "loc": "aot_state = create_aot_state("}, {"line": 567, "name": "create_aot_state", "filename": 20, "loc": "fw_metadata = run_functionalized_fw_and_collect_metadata("}, {"line": 207, "name": "inner", "filename": 21, "loc": "flat_f_outs = f(*flat_f_args)"}, {"line": 187, "name": "flat_fn", "filename": 22, "loc": "tree_out = fn(*args, **kwargs)"}, {"line": 1350, "name": "functional_call", "filename": 23, "loc": "out = PropagateUnbackedSymInts(mod).run("}, {"line": 174, "name": "run", "filename": 24, "loc": "self.env[node] = self.run_node(node)"}, {"line": 7870, "name": "run_node", "filename": 14, "loc": "result = super().run_node(n)"}, {"line": 256, "name": "run_node", "filename": 24, "loc": "return getattr(self, n.op)(n.target, args, kwargs)"}, {"line": 336, "name": "call_function", "filename": 24, "loc": "return target(*args, **kwargs)"}, {"line": 841, "name": "__call__", "filename": 9, "loc": "return self._op(*args, **kwargs)"}, {"line": 511, "name": "__torch_dispatch__", "filename": 25, "loc": "outs_unwrapped = func._op_dk("}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1511, "name": "_cached_dispatch_impl", "filename": 8, "loc": "output = self._dispatch_impl(func, types, args, kwargs)"}, {"line": 2611, "name": "_dispatch_impl", "filename": 8, "loc": "return maybe_propagate_real_tensors(fast_impl(self, *args, **kwargs))"}, {"line": 1296, "name": "fast_binary_impl", "filename": 26, "loc": "return slow(\"no contiguity match\")"}, {"line": 1165, "name": "slow", "filename": 26, "loc": "return slow_ref(*args, **kwargs)"}, {"line": 309, "name": "_fn", "filename": 10, "loc": "result = fn(*args, **kwargs)"}, {"line": 53, "name": "inner", "filename": 27, "loc": "return disable_fn(*args, **kwargs)"}, {"line": 1044, "name": "_fn", "filename": 28, "loc": "return fn(*args, **kwargs)"}, {"line": 149, "name": "_fn", "filename": 10, "loc": "result = fn(**bound.arguments)"}, {"line": 1141, "name": "add", "filename": 29, "loc": "a, b = _maybe_broadcast(a, b)"}, {"line": 470, "name": "_maybe_broadcast", "filename": 29, "loc": "return tuple(__maybe_broadcast(x, common_shape) for x in args)"}, {"line": 470, "name": "<genexpr>", "filename": 29, "loc": "return tuple(__maybe_broadcast(x, common_shape) for x in args)"}, {"line": 462, "name": "__maybe_broadcast", "filename": 29, "loc": "return x.expand(common_shape)"}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1511, "name": "_cached_dispatch_impl", "filename": 8, "loc": "output = self._dispatch_impl(func, types, args, kwargs)"}, {"line": 2635, "name": "_dispatch_impl", "filename": 8, "loc": "decomposition_table[func](*args, **kwargs)"}, {"line": 3061, "name": "expand", "filename": 29, "loc": "return prims.broadcast_in_dim("}, {"line": 841, "name": "__call__", "filename": 9, "loc": "return self._op(*args, **kwargs)"}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1511, "name": "_cached_dispatch_impl", "filename": 8, "loc": "output = self._dispatch_impl(func, types, args, kwargs)"}, {"line": 2657, "name": "_dispatch_impl", "filename": 8, "loc": "func.prim_meta_impl(*args, **kwargs)"}, {"line": 1323, "name": "_broadcast_in_dim_meta", "filename": 12, "loc": "return a.as_strided(shape, new_strides, a.storage_offset())"}, {"line": 28, "name": "wrapper", "filename": 7, "loc": "return fn(*args, **kwargs)"}, {"line": 1376, "name": "__torch_dispatch__", "filename": 8, "loc": "return self.dispatch(func, types, args, kwargs)"}, {"line": 2092, "name": "dispatch", "filename": 8, "loc": "return self._cached_dispatch_impl(func, types, args, kwargs)"}, {"line": 1533, "name": "_cached_dispatch_impl", "filename": 8, "loc": "entry = self._make_cache_entry(state, key, func, args, kwargs, output)"}, {"line": 1912, "name": "_make_cache_entry", "filename": 8, "loc": "output_info = self._get_output_info_for_cache_entry("}, {"line": 1828, "name": "_get_output_info_for_cache_entry", "filename": 8, "loc": "synth_output = self._output_from_cache_entry("}, {"line": 2013, "name": "_output_from_cache_entry", "filename": 8, "loc": "return self._get_output_tensor_from_cache_entry("}, {"line": 1988, "name": "_get_output_tensor_from_cache_entry", "filename": 8, "loc": "empty.set_(storage, storage_offset, shape, stride)"}, {"line": 538, "name": "guard_bool", "filename": 30, "loc": "r = self.evaluate()"}, {"line": 512, "name": "evaluate", "filename": 30, "loc": "return self.shape_env.evaluate_sym_node(self, size_oblivious)"}, {"line": 7239, "name": "evaluate_sym_node", "filename": 14, "loc": "return self.evaluate_expr("}, {"line": 7339, "name": "evaluate_expr", "filename": 14, "loc": "return self._inner_evaluate_expr("}, {"line": 272, "name": "wrapper", "filename": 15, "loc": "return retlog(fn(*args, **kwargs))"}, {"line": 7362, "name": "_inner_evaluate_expr", "filename": 14, "loc": "return self._evaluate_expr("}, {"line": 7644, "name": "_evaluate_expr", "filename": 14, "loc": "self._log_guard(\"eval [guard suppressed]\", g, forcing_spec=forcing_spec)"}, {"line": 7190, "name": "_log_guard", "filename": 14, "loc": "trace_structured("}]}
V0910 09:41:50.260000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/fx/experimental/symbolic_shapes.py:7190] {"guard_added_fast": {"expr": "1920000 < 1920000*s6", "user_stack": [], "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2604, "name": "compile_fx", "filename": 19, "loc": "gm, graph_signature = aot_export_module("}, {"line": 1444, "name": "aot_export_module", "filename": 20, "loc": "fx_g, metadata, in_spec, out_spec = _aot_export_function("}, {"line": 1694, "name": "_aot_export_function", "filename": 20, "loc": "aot_state = create_aot_state("}, {"line": 567, "name": "create_aot_state", "filename": 20, "loc": "fw_metadata = run_functionalized_fw_and_collect_metadata("}, {"line": 207, "name": "inner", "filename": 21, "loc": "flat_f_outs = f(*flat_f_args)"}, {"line": 187, "name": "flat_fn", "filename": 22, "loc": "tree_out = fn(*args, **kwargs)"}, {"line": 1350, "name": "functional_call", "filename": 23, "loc": "out = PropagateUnbackedSymInts(mod).run("}, {"line": 174, "name": "run", "filename": 24, "loc": "self.env[node] = self.run_node(node)"}, {"line": 7870, "name": "run_node", "filename": 14, "loc": "result = super().run_node(n)"}, {"line": 256, "name": "run_node", "filename": 24, "loc": "return getattr(self, n.op)(n.target, args, kwargs)"}, {"line": 336, "name": "call_function", "filename": 24, "loc": "return target(*args, **kwargs)"}, {"line": 841, "name": "__call__", "filename": 9, "loc": "return self._op(*args, **kwargs)"}, {"line": 307, "name": "_fn", "filename": 10, "loc": "result = fn(*args, is_out=(out is not None), **kwargs)  # type: ignore[arg-type]"}, {"line": 4539, "name": "matmul", "filename": 11, "loc": "output = torch.ops.aten._unsafe_view(t1_folded.mm(t2), output_shape)"}, {"line": 1255, "name": "__call__", "filename": 9, "loc": "return self._op(*args, **kwargs)"}, {"line": 511, "name": "__torch_dispatch__", "filename": 25, "loc": "outs_unwrapped = func._op_dk("}, {"line": 613, "name": "guard_or_true", "filename": 30, "loc": "return guard_or_true(SymBool(self))"}, {"line": 1413, "name": "guard_or_true", "filename": 14, "loc": "return _guard_or(a, True)"}, {"line": 1396, "name": "_guard_or", "filename": 14, "loc": "r = sym_node.shape_env.evaluate_sym_node("}, {"line": 7239, "name": "evaluate_sym_node", "filename": 14, "loc": "return self.evaluate_expr("}, {"line": 7339, "name": "evaluate_expr", "filename": 14, "loc": "return self._inner_evaluate_expr("}, {"line": 272, "name": "wrapper", "filename": 15, "loc": "return retlog(fn(*args, **kwargs))"}, {"line": 7362, "name": "_inner_evaluate_expr", "filename": 14, "loc": "return self._evaluate_expr("}, {"line": 7615, "name": "_evaluate_expr", "filename": 14, "loc": "self._log_guard(\"eval\", g, forcing_spec=forcing_spec)"}, {"line": 7190, "name": "_log_guard", "filename": 14, "loc": "trace_structured("}, {"line": 1346, "name": "trace_structured", "filename": 16, "loc": "record[name] = metadata_fn()"}]}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2604, "name": "compile_fx", "filename": 19, "loc": "gm, graph_signature = aot_export_module("}, {"line": 1444, "name": "aot_export_module", "filename": 20, "loc": "fx_g, metadata, in_spec, out_spec = _aot_export_function("}, {"line": 1694, "name": "_aot_export_function", "filename": 20, "loc": "aot_state = create_aot_state("}, {"line": 567, "name": "create_aot_state", "filename": 20, "loc": "fw_metadata = run_functionalized_fw_and_collect_metadata("}, {"line": 207, "name": "inner", "filename": 21, "loc": "flat_f_outs = f(*flat_f_args)"}, {"line": 187, "name": "flat_fn", "filename": 22, "loc": "tree_out = fn(*args, **kwargs)"}, {"line": 1350, "name": "functional_call", "filename": 23, "loc": "out = PropagateUnbackedSymInts(mod).run("}, {"line": 174, "name": "run", "filename": 24, "loc": "self.env[node] = self.run_node(node)"}, {"line": 7870, "name": "run_node", "filename": 14, "loc": "result = super().run_node(n)"}, {"line": 256, "name": "run_node", "filename": 24, "loc": "return getattr(self, n.op)(n.target, args, kwargs)"}, {"line": 336, "name": "call_function", "filename": 24, "loc": "return target(*args, **kwargs)"}, {"line": 841, "name": "__call__", "filename": 9, "loc": "return self._op(*args, **kwargs)"}, {"line": 307, "name": "_fn", "filename": 10, "loc": "result = fn(*args, is_out=(out is not None), **kwargs)  # type: ignore[arg-type]"}, {"line": 4539, "name": "matmul", "filename": 11, "loc": "output = torch.ops.aten._unsafe_view(t1_folded.mm(t2), output_shape)"}, {"line": 1255, "name": "__call__", "filename": 9, "loc": "return self._op(*args, **kwargs)"}, {"line": 511, "name": "__torch_dispatch__", "filename": 25, "loc": "outs_unwrapped = func._op_dk("}, {"line": 613, "name": "guard_or_true", "filename": 30, "loc": "return guard_or_true(SymBool(self))"}, {"line": 1413, "name": "guard_or_true", "filename": 14, "loc": "return _guard_or(a, True)"}, {"line": 1396, "name": "_guard_or", "filename": 14, "loc": "r = sym_node.shape_env.evaluate_sym_node("}, {"line": 7239, "name": "evaluate_sym_node", "filename": 14, "loc": "return self.evaluate_expr("}, {"line": 7339, "name": "evaluate_expr", "filename": 14, "loc": "return self._inner_evaluate_expr("}, {"line": 272, "name": "wrapper", "filename": 15, "loc": "return retlog(fn(*args, **kwargs))"}, {"line": 7362, "name": "_inner_evaluate_expr", "filename": 14, "loc": "return self._evaluate_expr("}, {"line": 7615, "name": "_evaluate_expr", "filename": 14, "loc": "self._log_guard(\"eval\", g, forcing_spec=forcing_spec)"}, {"line": 7190, "name": "_log_guard", "filename": 14, "loc": "trace_structured("}]}
V0910 09:42:02.682000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "43ff592b18a5c102aed45afa11fd7596"}
	{
	"name": "aot_collect_metadata",
	"ts": 1757522522682321.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:37.575000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_functorch/_aot_autograd/graph_compile.py", 31]}
V0910 09:42:37.576000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_functorch/_aot_autograd/graph_capture.py", 32]}
V0910 09:42:37.628000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_functorch/_aot_autograd/graph_capture.py:301] {"artifact": {"name": "aot_forward_graph_fw_metadata", "encoding": "string"}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2604, "name": "compile_fx", "filename": 19, "loc": "gm, graph_signature = aot_export_module("}, {"line": 1444, "name": "aot_export_module", "filename": 20, "loc": "fx_g, metadata, in_spec, out_spec = _aot_export_function("}, {"line": 1703, "name": "_aot_export_function", "filename": 20, "loc": "aot_graph_capture = aot_stage1_graph_capture(aot_state, flat_fn)"}, {"line": 171, "name": "aot_stage1_graph_capture", "filename": 31, "loc": "aot_dispatch_base_graph(  # type: ignore[assignment]"}, {"line": 301, "name": "aot_dispatch_base_graph", "filename": 32, "loc": "trace_structured("}], "has_payload": "764b6af38215c91927980604d59e1e70"}
	ViewAndMutationMeta(input_info=[InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=True,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False),
	                               InputAliasInfo(is_leaf=True,
	                                              mutates_data=False,
	                                              mutates_metadata=False,
	                                              mutations_hidden_from_autograd=True,
	                                              mutations_under_no_grad_or_inference_mode=False,
	                                              mutation_inductor_storage_resize=False,
	                                              mutates_storage_metadata=False,
	                                              requires_grad=False,
	                                              keep_input_mutations=False)],
	                    output_info=[OutputAliasInfo(output_type=<OutputType.non_alias: 1>,
	                                                raw_type=<class 'torch._subclasses.functional_tensor.FunctionalTensor'>,
	                                                base_idx=None,
	                                                dynamic_dims={1},
	                                                requires_grad=False,
	                                                functional_tensor=None)],
	                    num_intermediate_bases=0,
	                    keep_input_mutations=False,
	                    traced_tangents=[],
	                    traced_tangents_descs=[],
	                    subclass_inp_meta=[PlainTensorMeta(unwrapped_idx=0,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=1,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=2,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=3,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=4,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=5,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=6,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=7,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=8,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=9,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=10,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=11,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=12,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=13,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=14,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=15,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=16,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=17,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=18,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=19,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=20,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=21,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=22,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=23,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=24,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=25,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=26,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=27,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=28,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=29,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=30,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=31,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=32,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=33,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=34,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=35,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=36,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=37,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=38,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=39,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=40,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=41,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=42,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=43,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=44,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=45,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=46,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=47,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=48,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=49,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=50,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=51,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=52,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=53,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=54,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=55,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=56,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=57,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=58,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=59,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=60,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=61,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=62,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=63,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=64,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=65,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=66,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=67,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=68,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=69,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=70,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=71,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=72,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=73,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=74,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=75,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=76,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=77,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=78,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=79,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=80,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=81,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=82,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=83,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=84,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=85,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=86,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=87,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=88,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=89,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=90,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=91,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=92,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=93,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=94,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=95,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=96,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=97,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=98,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=99,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=100,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=101,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=102,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=103,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=104,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=105,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=106,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=107,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=108,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=109,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=110,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=111,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=112,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=113,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=114,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=115,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=116,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=117,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=118,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=119,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=120,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=121,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=122,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=123,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=124,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=125,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=126,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=127,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=128,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=129,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=130,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=131,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=132,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=133,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=134,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=135,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=136,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=137,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=138,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=139,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=140,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=141,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=142,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=143,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=144,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=145,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=146,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=147,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=148,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=149,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=150,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=151,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=152,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=153,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=154,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=155,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=156,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=157,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=158,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=159,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=160,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=161,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=162,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=163,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=164,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=165,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=166,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=167,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=168,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=169,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=170,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=171,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=172,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=173,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=174,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=175,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=176,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=177,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=178,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=179,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=180,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=181,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=182,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=183,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=184,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=185,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=186,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=187,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=188,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=189,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=190,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=191,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=192,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=193,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=194,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=195,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=196,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=197,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=198,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=199,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=200,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=201,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=202,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=203,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=204,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=205,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=206,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=207,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=208,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=209,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=210,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=211,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=212,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=213,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=214,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=215,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=216,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=217,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=218,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=219,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=220,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=221,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=222,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=223,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=224,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=225,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=226,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=227,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=228,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=229,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=230,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=231,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=232,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=233,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=234,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=235,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=236,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=237,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=238,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=239,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=240,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=241,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=242,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=243,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=244,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=245,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=246,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=247,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=248,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=249,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=250,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=251,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=252,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=253,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=254,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=255,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=256,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=257,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=258,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=259,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=260,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=261,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=262,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=263,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=264,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=265,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=266,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=267,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=268,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=269,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=270,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=271,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=272,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=273,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=274,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=275,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=276,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=277,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=278,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=279,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=280,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=281,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=282,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=283,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=284,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=285,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=286,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=287,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=288,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=289,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=290,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=291,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=292,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=293,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=294,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=295,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=296,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=297,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=298,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=299,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=300,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=301,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=302,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=303,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=304,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=305,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=306,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=307,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=308,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=309,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=310,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=311,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=312,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=313,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=314,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=315,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=316,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=317,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=318,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=319,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=320,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=321,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=322,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=323,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=324,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=325,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=326,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=327,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=328,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=329,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=330,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=331,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=332,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=333,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=334,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=335,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=336,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=337,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=338,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=339,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=340,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=341,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=342,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=343,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=344,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=345,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=346,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=347,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=348,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=349,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=350,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=351,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=352,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=353,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=354,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=355,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=356,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=357,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=358,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=359,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=360,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=361,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=362,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=363,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=364,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=365,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=366,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=367,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=368,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=369,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=370,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=371,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=372,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=373,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=374,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=375,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=376,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=377,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=378,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=379,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=380,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=381,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=382,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=383,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=384,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=385,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=386,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=387,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=388,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=389,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=390,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=391,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=392,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=393,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=394,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=395,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=396,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=397,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=398,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=399,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=400,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=401,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=402,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=403,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=404,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=405,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=406,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=407,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=408,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=409,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=410,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=411,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=412,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=413,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=414,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=415,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=416,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=417,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=418,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=419,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=420,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=421,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=422,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=423,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=424,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=425,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=426,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=427,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=428,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=429,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=430,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=431,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=432,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=433,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=434,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=435,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=436,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=437,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=438,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=439,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=440,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=441,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=442,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=443,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=444,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=445,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=446,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=447,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=448,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=449,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=450,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=451,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=452,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=453,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=454,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=455,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=456,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=457,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=458,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=459,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=460,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=461,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=462,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=463,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=464,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=465,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=466,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=467,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=468,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=469,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=470,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=471,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=472,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=473,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=474,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=475,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=476,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=477,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=478,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=479,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=480,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=481,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=482,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=483,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=484,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=485,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=486,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=487,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=488,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=489,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=490,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=491,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=492,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=493,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=494,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=495,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=496,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=497,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=498,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=499,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=500,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=501,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=502,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=503,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=504,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=505,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=506,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=507,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=508,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=509,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=510,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=511,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=512,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=513,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=514,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=515,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=516,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=517,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=518,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=519,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=520,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=521,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=522,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=523,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=524,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=525,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=526,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=527,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=528,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=529,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=530,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=531,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=532,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=533,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=534,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=535,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=536,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=537,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=538,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=539,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=540,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=541,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=542,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=543,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=544,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=545,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=546,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=547,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=548,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=549,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=550,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=551,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=552,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=553,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=554,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=555,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=556,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=557,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=558,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=559,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=560,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=561,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=562,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=563,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=564,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=565,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=566,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=567,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=568,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=569,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=570,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=571,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=572,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=573,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=574,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=575,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=576,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=577,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=578,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=579,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=580,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=581,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=582,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=583,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=584,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=585,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=586,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=587,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=588,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=589,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=590,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=591,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=592,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=593,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=594,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=595,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=596,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=597,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=598,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=599,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=600,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=601,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=602,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=603,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=604,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=605,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=606,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=607,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=608,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=609,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=610,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=611,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=612,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=613,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=614,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=615,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=616,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=617,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=618,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=619,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=620,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=621,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=622,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=623,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=624,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=625,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=626,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=627,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=628,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=629,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=630,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=631,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=632,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=633,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=634,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=635,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=636,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=637,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=638,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=639,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=640,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=641,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=642,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=643,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=644,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=645,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=646,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=647,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=648,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=649,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=650,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=651,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=652,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=653,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=654,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=655,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=656,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=657,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=658,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=659,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=660,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=661,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=662,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=663,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=664,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=665,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=666,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=667,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=668,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=669,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=670,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=671,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=672,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=673,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=674,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=675,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=676,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=677,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=678,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=679,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=680,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=681,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=682,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=683,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=684,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=685,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=686,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=687,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=688,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=689,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=690,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=691,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=692,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=693,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=694,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=695,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=696,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=697,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=698,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=699,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=700,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=701,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=702,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=703,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=704,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=705,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=706,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=707,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=708,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=709,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=710,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=711,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=712,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=713,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=714,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=715,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=716,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=717,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=718,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=719,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=720,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=721,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=722,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=723,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=724,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=725,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=726,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=727,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=728,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=729,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=730,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=731,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=732,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=733,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=734,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=735,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=736,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=737,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=738,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=739,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=740,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=741,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=742,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=743,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=744,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=745,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=746,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=747,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=748,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=749,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=750,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=751,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=752,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=753,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=754,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=755,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=756,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=757,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=758,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=759,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=760,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=761,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=762,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=763,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=764,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=765,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=766,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=767,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=768,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=769,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=770,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=771,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=772,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=773,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=774,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=775,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=776,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=777,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=778,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=779,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=780,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=781,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=782,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=783,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=784,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=785,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=786,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=787,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=788,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=789,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=790,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=791,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=792,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=793,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=794,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=795,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=796,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=797,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=798,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=799,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=800,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=801,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=802,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=803,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=804,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=805,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=806,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=807,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=808,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=809,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=810,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=811,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=812,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=813,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=814,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=815,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=816,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=817,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=818,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=819,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=820,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=821,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=822,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=823,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=824,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=825,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=826,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=827,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=828,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=829,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=830,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=831,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=832,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=833,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=834,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=835,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=836,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=837,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=838,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=839,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=840,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=841,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=842,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=843,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=844,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=845,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=846,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=847,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=848,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=849,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=850,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=851,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=852,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=853,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=854,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=855,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=856,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=857,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=858,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=859,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=860,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=861,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=862,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=863,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=864,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=865,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=866,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=867,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=868,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=869,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=870,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=871,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=872,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=873,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=874,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=875,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=876,
	                                                      memory_format=None),
	                                      PlainTensorMeta(unwrapped_idx=877,
	                                                      memory_format=None)],
	                    subclass_fw_graph_out_meta=[PlainTensorMeta(unwrapped_idx=0,
	                                                               memory_format=None)],
	                    subclass_tangent_meta=[],
	                    is_train=False,
	                    traced_tangent_metas=None,
	                    num_symints_saved_for_bw=None,
	                    grad_enabled_mutation=None,
	                    deterministic=False,
	                    static_input_indices=[],
	                    tokens={},
	                    indices_of_inputs_that_requires_grad_with_mutations_in_bw=[],
	                    bw_donated_idxs=None,
	                    num_backward_tokens=0,
	                    num_graphsafe_rng_states=0,
	                    graphsafe_rng_state_index=None)
V0910 09:42:38.249000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_functorch/_aot_autograd/graph_capture.py:319] {"aot_inference_graph": {}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2604, "name": "compile_fx", "filename": 19, "loc": "gm, graph_signature = aot_export_module("}, {"line": 1444, "name": "aot_export_module", "filename": 20, "loc": "fx_g, metadata, in_spec, out_spec = _aot_export_function("}, {"line": 1703, "name": "_aot_export_function", "filename": 20, "loc": "aot_graph_capture = aot_stage1_graph_capture(aot_state, flat_fn)"}, {"line": 171, "name": "aot_stage1_graph_capture", "filename": 31, "loc": "aot_dispatch_base_graph(  # type: ignore[assignment]"}, {"line": 319, "name": "aot_dispatch_base_graph", "filename": 32, "loc": "trace_structured("}], "has_payload": "cc9f789227240a06e4fe4a9790902286"}
	class <lambda>(torch.nn.Module):
	    def forward(
	        self,
	        arg0_1: "f32[1500, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=0)
	        arg1_1: "f32[1280, 128, 3][384, 3, 1]cuda:0",  # PlainAOTInput(idx=1)
	        arg2_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=2)
	        arg3_1: "f32[1280, 1280, 3][3840, 3, 1]cuda:0",  # PlainAOTInput(idx=3)
	        arg4_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=4)
	        arg5_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=5)
	        arg6_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=6)
	        arg7_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=7)
	        arg8_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=8)
	        arg9_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=9)
	        arg10_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=10)
	        arg11_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=11)
	        arg12_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=12)
	        arg13_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=13)
	        arg14_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=14)
	        arg15_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=15)
	        arg16_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=16)
	        arg17_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=17)
	        arg18_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=18)
	        arg19_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=19)
	        arg20_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=20)
	        arg21_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=21)
	        arg22_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=22)
	        arg23_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=23)
	        arg24_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=24)
	        arg25_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=25)
	        arg26_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=26)
	        arg27_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=27)
	        arg28_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=28)
	        arg29_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=29)
	        arg30_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=30)
	        arg31_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=31)
	        arg32_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=32)
	        arg33_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=33)
	        arg34_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=34)
	        arg35_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=35)
	        arg36_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=36)
	        arg37_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=37)
	        arg38_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=38)
	        arg39_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=39)
	        arg40_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=40)
	        arg41_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=41)
	        arg42_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=42)
	        arg43_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=43)
	        arg44_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=44)
	        arg45_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=45)
	        arg46_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=46)
	        arg47_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=47)
	        arg48_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=48)
	        arg49_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=49)
	        arg50_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=50)
	        arg51_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=51)
	        arg52_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=52)
	        arg53_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=53)
	        arg54_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=54)
	        arg55_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=55)
	        arg56_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=56)
	        arg57_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=57)
	        arg58_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=58)
	        arg59_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=59)
	        arg60_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=60)
	        arg61_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=61)
	        arg62_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=62)
	        arg63_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=63)
	        arg64_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=64)
	        arg65_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=65)
	        arg66_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=66)
	        arg67_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=67)
	        arg68_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=68)
	        arg69_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=69)
	        arg70_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=70)
	        arg71_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=71)
	        arg72_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=72)
	        arg73_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=73)
	        arg74_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=74)
	        arg75_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=75)
	        arg76_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=76)
	        arg77_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=77)
	        arg78_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=78)
	        arg79_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=79)
	        arg80_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=80)
	        arg81_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=81)
	        arg82_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=82)
	        arg83_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=83)
	        arg84_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=84)
	        arg85_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=85)
	        arg86_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=86)
	        arg87_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=87)
	        arg88_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=88)
	        arg89_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=89)
	        arg90_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=90)
	        arg91_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=91)
	        arg92_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=92)
	        arg93_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=93)
	        arg94_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=94)
	        arg95_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=95)
	        arg96_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=96)
	        arg97_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=97)
	        arg98_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=98)
	        arg99_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=99)
	        arg100_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=100)
	        arg101_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=101)
	        arg102_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=102)
	        arg103_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=103)
	        arg104_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=104)
	        arg105_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=105)
	        arg106_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=106)
	        arg107_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=107)
	        arg108_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=108)
	        arg109_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=109)
	        arg110_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=110)
	        arg111_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=111)
	        arg112_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=112)
	        arg113_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=113)
	        arg114_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=114)
	        arg115_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=115)
	        arg116_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=116)
	        arg117_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=117)
	        arg118_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=118)
	        arg119_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=119)
	        arg120_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=120)
	        arg121_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=121)
	        arg122_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=122)
	        arg123_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=123)
	        arg124_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=124)
	        arg125_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=125)
	        arg126_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=126)
	        arg127_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=127)
	        arg128_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=128)
	        arg129_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=129)
	        arg130_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=130)
	        arg131_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=131)
	        arg132_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=132)
	        arg133_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=133)
	        arg134_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=134)
	        arg135_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=135)
	        arg136_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=136)
	        arg137_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=137)
	        arg138_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=138)
	        arg139_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=139)
	        arg140_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=140)
	        arg141_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=141)
	        arg142_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=142)
	        arg143_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=143)
	        arg144_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=144)
	        arg145_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=145)
	        arg146_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=146)
	        arg147_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=147)
	        arg148_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=148)
	        arg149_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=149)
	        arg150_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=150)
	        arg151_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=151)
	        arg152_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=152)
	        arg153_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=153)
	        arg154_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=154)
	        arg155_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=155)
	        arg156_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=156)
	        arg157_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=157)
	        arg158_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=158)
	        arg159_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=159)
	        arg160_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=160)
	        arg161_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=161)
	        arg162_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=162)
	        arg163_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=163)
	        arg164_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=164)
	        arg165_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=165)
	        arg166_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=166)
	        arg167_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=167)
	        arg168_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=168)
	        arg169_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=169)
	        arg170_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=170)
	        arg171_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=171)
	        arg172_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=172)
	        arg173_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=173)
	        arg174_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=174)
	        arg175_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=175)
	        arg176_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=176)
	        arg177_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=177)
	        arg178_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=178)
	        arg179_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=179)
	        arg180_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=180)
	        arg181_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=181)
	        arg182_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=182)
	        arg183_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=183)
	        arg184_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=184)
	        arg185_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=185)
	        arg186_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=186)
	        arg187_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=187)
	        arg188_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=188)
	        arg189_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=189)
	        arg190_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=190)
	        arg191_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=191)
	        arg192_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=192)
	        arg193_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=193)
	        arg194_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=194)
	        arg195_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=195)
	        arg196_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=196)
	        arg197_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=197)
	        arg198_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=198)
	        arg199_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=199)
	        arg200_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=200)
	        arg201_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=201)
	        arg202_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=202)
	        arg203_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=203)
	        arg204_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=204)
	        arg205_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=205)
	        arg206_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=206)
	        arg207_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=207)
	        arg208_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=208)
	        arg209_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=209)
	        arg210_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=210)
	        arg211_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=211)
	        arg212_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=212)
	        arg213_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=213)
	        arg214_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=214)
	        arg215_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=215)
	        arg216_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=216)
	        arg217_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=217)
	        arg218_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=218)
	        arg219_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=219)
	        arg220_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=220)
	        arg221_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=221)
	        arg222_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=222)
	        arg223_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=223)
	        arg224_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=224)
	        arg225_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=225)
	        arg226_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=226)
	        arg227_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=227)
	        arg228_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=228)
	        arg229_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=229)
	        arg230_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=230)
	        arg231_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=231)
	        arg232_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=232)
	        arg233_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=233)
	        arg234_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=234)
	        arg235_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=235)
	        arg236_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=236)
	        arg237_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=237)
	        arg238_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=238)
	        arg239_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=239)
	        arg240_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=240)
	        arg241_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=241)
	        arg242_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=242)
	        arg243_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=243)
	        arg244_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=244)
	        arg245_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=245)
	        arg246_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=246)
	        arg247_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=247)
	        arg248_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=248)
	        arg249_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=249)
	        arg250_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=250)
	        arg251_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=251)
	        arg252_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=252)
	        arg253_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=253)
	        arg254_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=254)
	        arg255_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=255)
	        arg256_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=256)
	        arg257_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=257)
	        arg258_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=258)
	        arg259_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=259)
	        arg260_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=260)
	        arg261_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=261)
	        arg262_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=262)
	        arg263_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=263)
	        arg264_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=264)
	        arg265_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=265)
	        arg266_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=266)
	        arg267_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=267)
	        arg268_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=268)
	        arg269_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=269)
	        arg270_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=270)
	        arg271_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=271)
	        arg272_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=272)
	        arg273_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=273)
	        arg274_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=274)
	        arg275_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=275)
	        arg276_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=276)
	        arg277_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=277)
	        arg278_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=278)
	        arg279_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=279)
	        arg280_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=280)
	        arg281_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=281)
	        arg282_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=282)
	        arg283_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=283)
	        arg284_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=284)
	        arg285_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=285)
	        arg286_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=286)
	        arg287_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=287)
	        arg288_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=288)
	        arg289_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=289)
	        arg290_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=290)
	        arg291_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=291)
	        arg292_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=292)
	        arg293_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=293)
	        arg294_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=294)
	        arg295_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=295)
	        arg296_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=296)
	        arg297_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=297)
	        arg298_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=298)
	        arg299_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=299)
	        arg300_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=300)
	        arg301_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=301)
	        arg302_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=302)
	        arg303_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=303)
	        arg304_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=304)
	        arg305_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=305)
	        arg306_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=306)
	        arg307_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=307)
	        arg308_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=308)
	        arg309_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=309)
	        arg310_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=310)
	        arg311_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=311)
	        arg312_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=312)
	        arg313_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=313)
	        arg314_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=314)
	        arg315_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=315)
	        arg316_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=316)
	        arg317_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=317)
	        arg318_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=318)
	        arg319_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=319)
	        arg320_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=320)
	        arg321_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=321)
	        arg322_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=322)
	        arg323_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=323)
	        arg324_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=324)
	        arg325_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=325)
	        arg326_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=326)
	        arg327_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=327)
	        arg328_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=328)
	        arg329_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=329)
	        arg330_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=330)
	        arg331_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=331)
	        arg332_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=332)
	        arg333_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=333)
	        arg334_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=334)
	        arg335_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=335)
	        arg336_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=336)
	        arg337_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=337)
	        arg338_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=338)
	        arg339_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=339)
	        arg340_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=340)
	        arg341_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=341)
	        arg342_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=342)
	        arg343_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=343)
	        arg344_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=344)
	        arg345_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=345)
	        arg346_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=346)
	        arg347_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=347)
	        arg348_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=348)
	        arg349_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=349)
	        arg350_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=350)
	        arg351_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=351)
	        arg352_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=352)
	        arg353_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=353)
	        arg354_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=354)
	        arg355_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=355)
	        arg356_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=356)
	        arg357_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=357)
	        arg358_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=358)
	        arg359_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=359)
	        arg360_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=360)
	        arg361_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=361)
	        arg362_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=362)
	        arg363_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=363)
	        arg364_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=364)
	        arg365_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=365)
	        arg366_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=366)
	        arg367_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=367)
	        arg368_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=368)
	        arg369_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=369)
	        arg370_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=370)
	        arg371_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=371)
	        arg372_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=372)
	        arg373_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=373)
	        arg374_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=374)
	        arg375_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=375)
	        arg376_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=376)
	        arg377_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=377)
	        arg378_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=378)
	        arg379_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=379)
	        arg380_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=380)
	        arg381_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=381)
	        arg382_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=382)
	        arg383_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=383)
	        arg384_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=384)
	        arg385_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=385)
	        arg386_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=386)
	        arg387_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=387)
	        arg388_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=388)
	        arg389_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=389)
	        arg390_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=390)
	        arg391_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=391)
	        arg392_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=392)
	        arg393_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=393)
	        arg394_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=394)
	        arg395_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=395)
	        arg396_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=396)
	        arg397_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=397)
	        arg398_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=398)
	        arg399_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=399)
	        arg400_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=400)
	        arg401_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=401)
	        arg402_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=402)
	        arg403_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=403)
	        arg404_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=404)
	        arg405_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=405)
	        arg406_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=406)
	        arg407_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=407)
	        arg408_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=408)
	        arg409_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=409)
	        arg410_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=410)
	        arg411_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=411)
	        arg412_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=412)
	        arg413_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=413)
	        arg414_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=414)
	        arg415_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=415)
	        arg416_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=416)
	        arg417_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=417)
	        arg418_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=418)
	        arg419_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=419)
	        arg420_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=420)
	        arg421_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=421)
	        arg422_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=422)
	        arg423_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=423)
	        arg424_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=424)
	        arg425_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=425)
	        arg426_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=426)
	        arg427_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=427)
	        arg428_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=428)
	        arg429_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=429)
	        arg430_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=430)
	        arg431_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=431)
	        arg432_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=432)
	        arg433_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=433)
	        arg434_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=434)
	        arg435_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=435)
	        arg436_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=436)
	        arg437_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=437)
	        arg438_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=438)
	        arg439_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=439)
	        arg440_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=440)
	        arg441_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=441)
	        arg442_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=442)
	        arg443_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=443)
	        arg444_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=444)
	        arg445_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=445)
	        arg446_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=446)
	        arg447_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=447)
	        arg448_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=448)
	        arg449_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=449)
	        arg450_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=450)
	        arg451_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=451)
	        arg452_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=452)
	        arg453_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=453)
	        arg454_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=454)
	        arg455_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=455)
	        arg456_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=456)
	        arg457_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=457)
	        arg458_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=458)
	        arg459_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=459)
	        arg460_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=460)
	        arg461_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=461)
	        arg462_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=462)
	        arg463_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=463)
	        arg464_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=464)
	        arg465_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=465)
	        arg466_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=466)
	        arg467_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=467)
	        arg468_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=468)
	        arg469_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=469)
	        arg470_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=470)
	        arg471_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=471)
	        arg472_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=472)
	        arg473_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=473)
	        arg474_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=474)
	        arg475_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=475)
	        arg476_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=476)
	        arg477_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=477)
	        arg478_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=478)
	        arg479_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=479)
	        arg480_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=480)
	        arg481_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=481)
	        arg482_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=482)
	        arg483_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=483)
	        arg484_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=484)
	        arg485_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=485)
	        arg486_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=486)
	        arg487_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=487)
	        arg488_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=488)
	        arg489_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=489)
	        arg490_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=490)
	        arg491_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=491)
	        arg492_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=492)
	        arg493_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=493)
	        arg494_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=494)
	        arg495_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=495)
	        arg496_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=496)
	        arg497_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=497)
	        arg498_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=498)
	        arg499_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=499)
	        arg500_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=500)
	        arg501_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=501)
	        arg502_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=502)
	        arg503_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=503)
	        arg504_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=504)
	        arg505_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=505)
	        arg506_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=506)
	        arg507_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=507)
	        arg508_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=508)
	        arg509_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=509)
	        arg510_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=510)
	        arg511_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=511)
	        arg512_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=512)
	        arg513_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=513)
	        arg514_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=514)
	        arg515_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=515)
	        arg516_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=516)
	        arg517_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=517)
	        arg518_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=518)
	        arg519_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=519)
	        arg520_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=520)
	        arg521_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=521)
	        arg522_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=522)
	        arg523_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=523)
	        arg524_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=524)
	        arg525_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=525)
	        arg526_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=526)
	        arg527_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=527)
	        arg528_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=528)
	        arg529_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=529)
	        arg530_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=530)
	        arg531_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=531)
	        arg532_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=532)
	        arg533_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=533)
	        arg534_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=534)
	        arg535_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=535)
	        arg536_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=536)
	        arg537_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=537)
	        arg538_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=538)
	        arg539_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=539)
	        arg540_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=540)
	        arg541_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=541)
	        arg542_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=542)
	        arg543_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=543)
	        arg544_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=544)
	        arg545_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=545)
	        arg546_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=546)
	        arg547_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=547)
	        arg548_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=548)
	        arg549_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=549)
	        arg550_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=550)
	        arg551_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=551)
	        arg552_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=552)
	        arg553_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=553)
	        arg554_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=554)
	        arg555_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=555)
	        arg556_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=556)
	        arg557_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=557)
	        arg558_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=558)
	        arg559_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=559)
	        arg560_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=560)
	        arg561_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=561)
	        arg562_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=562)
	        arg563_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=563)
	        arg564_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=564)
	        arg565_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=565)
	        arg566_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=566)
	        arg567_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=567)
	        arg568_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=568)
	        arg569_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=569)
	        arg570_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=570)
	        arg571_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=571)
	        arg572_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=572)
	        arg573_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=573)
	        arg574_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=574)
	        arg575_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=575)
	        arg576_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=576)
	        arg577_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=577)
	        arg578_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=578)
	        arg579_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=579)
	        arg580_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=580)
	        arg581_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=581)
	        arg582_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=582)
	        arg583_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=583)
	        arg584_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=584)
	        arg585_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=585)
	        arg586_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=586)
	        arg587_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=587)
	        arg588_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=588)
	        arg589_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=589)
	        arg590_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=590)
	        arg591_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=591)
	        arg592_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=592)
	        arg593_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=593)
	        arg594_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=594)
	        arg595_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=595)
	        arg596_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=596)
	        arg597_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=597)
	        arg598_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=598)
	        arg599_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=599)
	        arg600_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=600)
	        arg601_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=601)
	        arg602_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=602)
	        arg603_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=603)
	        arg604_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=604)
	        arg605_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=605)
	        arg606_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=606)
	        arg607_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=607)
	        arg608_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=608)
	        arg609_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=609)
	        arg610_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=610)
	        arg611_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=611)
	        arg612_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=612)
	        arg613_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=613)
	        arg614_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=614)
	        arg615_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=615)
	        arg616_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=616)
	        arg617_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=617)
	        arg618_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=618)
	        arg619_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=619)
	        arg620_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=620)
	        arg621_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=621)
	        arg622_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=622)
	        arg623_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=623)
	        arg624_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=624)
	        arg625_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=625)
	        arg626_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=626)
	        arg627_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=627)
	        arg628_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=628)
	        arg629_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=629)
	        arg630_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=630)
	        arg631_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=631)
	        arg632_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=632)
	        arg633_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=633)
	        arg634_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=634)
	        arg635_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=635)
	        arg636_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=636)
	        arg637_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=637)
	        arg638_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=638)
	        arg639_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=639)
	        arg640_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=640)
	        arg641_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=641)
	        arg642_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=642)
	        arg643_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=643)
	        arg644_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=644)
	        arg645_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=645)
	        arg646_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=646)
	        arg647_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=647)
	        arg648_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=648)
	        arg649_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=649)
	        arg650_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=650)
	        arg651_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=651)
	        arg652_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=652)
	        arg653_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=653)
	        arg654_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=654)
	        arg655_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=655)
	        arg656_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=656)
	        arg657_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=657)
	        arg658_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=658)
	        arg659_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=659)
	        arg660_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=660)
	        arg661_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=661)
	        arg662_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=662)
	        arg663_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=663)
	        arg664_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=664)
	        arg665_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=665)
	        arg666_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=666)
	        arg667_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=667)
	        arg668_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=668)
	        arg669_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=669)
	        arg670_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=670)
	        arg671_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=671)
	        arg672_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=672)
	        arg673_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=673)
	        arg674_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=674)
	        arg675_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=675)
	        arg676_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=676)
	        arg677_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=677)
	        arg678_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=678)
	        arg679_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=679)
	        arg680_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=680)
	        arg681_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=681)
	        arg682_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=682)
	        arg683_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=683)
	        arg684_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=684)
	        arg685_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=685)
	        arg686_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=686)
	        arg687_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=687)
	        arg688_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=688)
	        arg689_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=689)
	        arg690_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=690)
	        arg691_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=691)
	        arg692_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=692)
	        arg693_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=693)
	        arg694_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=694)
	        arg695_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=695)
	        arg696_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=696)
	        arg697_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=697)
	        arg698_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=698)
	        arg699_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=699)
	        arg700_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=700)
	        arg701_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=701)
	        arg702_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=702)
	        arg703_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=703)
	        arg704_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=704)
	        arg705_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=705)
	        arg706_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=706)
	        arg707_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=707)
	        arg708_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=708)
	        arg709_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=709)
	        arg710_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=710)
	        arg711_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=711)
	        arg712_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=712)
	        arg713_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=713)
	        arg714_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=714)
	        arg715_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=715)
	        arg716_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=716)
	        arg717_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=717)
	        arg718_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=718)
	        arg719_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=719)
	        arg720_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=720)
	        arg721_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=721)
	        arg722_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=722)
	        arg723_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=723)
	        arg724_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=724)
	        arg725_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=725)
	        arg726_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=726)
	        arg727_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=727)
	        arg728_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=728)
	        arg729_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=729)
	        arg730_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=730)
	        arg731_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=731)
	        arg732_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=732)
	        arg733_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=733)
	        arg734_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=734)
	        arg735_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=735)
	        arg736_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=736)
	        arg737_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=737)
	        arg738_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=738)
	        arg739_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=739)
	        arg740_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=740)
	        arg741_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=741)
	        arg742_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=742)
	        arg743_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=743)
	        arg744_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=744)
	        arg745_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=745)
	        arg746_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=746)
	        arg747_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=747)
	        arg748_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=748)
	        arg749_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=749)
	        arg750_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=750)
	        arg751_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=751)
	        arg752_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=752)
	        arg753_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=753)
	        arg754_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=754)
	        arg755_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=755)
	        arg756_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=756)
	        arg757_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=757)
	        arg758_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=758)
	        arg759_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=759)
	        arg760_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=760)
	        arg761_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=761)
	        arg762_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=762)
	        arg763_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=763)
	        arg764_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=764)
	        arg765_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=765)
	        arg766_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=766)
	        arg767_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=767)
	        arg768_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=768)
	        arg769_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=769)
	        arg770_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=770)
	        arg771_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=771)
	        arg772_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=772)
	        arg773_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=773)
	        arg774_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=774)
	        arg775_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=775)
	        arg776_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=776)
	        arg777_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=777)
	        arg778_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=778)
	        arg779_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=779)
	        arg780_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=780)
	        arg781_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=781)
	        arg782_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=782)
	        arg783_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=783)
	        arg784_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=784)
	        arg785_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=785)
	        arg786_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=786)
	        arg787_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=787)
	        arg788_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=788)
	        arg789_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=789)
	        arg790_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=790)
	        arg791_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=791)
	        arg792_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=792)
	        arg793_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=793)
	        arg794_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=794)
	        arg795_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=795)
	        arg796_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=796)
	        arg797_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=797)
	        arg798_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=798)
	        arg799_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=799)
	        arg800_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=800)
	        arg801_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=801)
	        arg802_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=802)
	        arg803_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=803)
	        arg804_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=804)
	        arg805_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=805)
	        arg806_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=806)
	        arg807_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=807)
	        arg808_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=808)
	        arg809_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=809)
	        arg810_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=810)
	        arg811_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=811)
	        arg812_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=812)
	        arg813_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=813)
	        arg814_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=814)
	        arg815_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=815)
	        arg816_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=816)
	        arg817_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=817)
	        arg818_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=818)
	        arg819_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=819)
	        arg820_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=820)
	        arg821_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=821)
	        arg822_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=822)
	        arg823_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=823)
	        arg824_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=824)
	        arg825_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=825)
	        arg826_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=826)
	        arg827_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=827)
	        arg828_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=828)
	        arg829_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=829)
	        arg830_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=830)
	        arg831_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=831)
	        arg832_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=832)
	        arg833_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=833)
	        arg834_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=834)
	        arg835_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=835)
	        arg836_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=836)
	        arg837_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=837)
	        arg838_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=838)
	        arg839_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=839)
	        arg840_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=840)
	        arg841_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=841)
	        arg842_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=842)
	        arg843_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=843)
	        arg844_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=844)
	        arg845_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=845)
	        arg846_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=846)
	        arg847_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=847)
	        arg848_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=848)
	        arg849_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=849)
	        arg850_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=850)
	        arg851_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=851)
	        arg852_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=852)
	        arg853_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=853)
	        arg854_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=854)
	        arg855_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=855)
	        arg856_1: "i8[1280, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=856)
	        arg857_1: "f32[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=857)
	        arg858_1: "i8[1280, 40][40, 1]cuda:0",  # PlainAOTInput(idx=858)
	        arg859_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=859)
	        arg860_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=860)
	        arg861_1: "f32[5120][1]cuda:0",  # PlainAOTInput(idx=861)
	        arg862_1: "i8[5120, 1280][1280, 1]cuda:0",  # PlainAOTInput(idx=862)
	        arg863_1: "f32[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=863)
	        arg864_1: "i8[5120, 40][40, 1]cuda:0",  # PlainAOTInput(idx=864)
	        arg865_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=865)
	        arg866_1: "i8[1280, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=866)
	        arg867_1: "f32[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=867)
	        arg868_1: "i8[1280, 160][160, 1]cuda:0",  # PlainAOTInput(idx=868)
	        arg869_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=869)
	        arg870_1: "f32[1280][1]cuda:0",  # PlainAOTInput(idx=870)
	        arg871_1: "i8[3072, 5120][5120, 1]cuda:0",  # PlainAOTInput(idx=871)
	        arg872_1: "f32[3072, 160][160, 1]cuda:0",  # PlainAOTInput(idx=872)
	        arg873_1: "i8[3072, 160][160, 1]cuda:0",  # PlainAOTInput(idx=873)
	        arg874_1: "i8[3072, 3072][3072, 1]cuda:0",  # PlainAOTInput(idx=874)
	        arg875_1: "f32[3072, 96][96, 1]cuda:0",  # PlainAOTInput(idx=875)
	        arg876_1: "i8[3072, 96][96, 1]cuda:0",  # PlainAOTInput(idx=876)
	        arg877_1: "f32[s6, 128, 3000][384000, 3000, 1]cuda:0",  # PlainAOTInput(idx=877)
	    ):
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:348 in forward, code: input_features = input_features.to(dtype=self.conv1.weight.dtype, device=self.conv1.weight.device)
	        _assert_tensor_metadata = torch.ops.aten._assert_tensor_metadata.default(arg877_1, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:349 in forward, code: inputs_embeds = nn.functional.gelu(self.conv1(input_features))
	        convolution: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.convolution.default(arg877_1, arg1_1, arg2_1, [1], [1], [1], False, [0], 1);  arg1_1 = arg2_1 = None
	        sym_size_int: "Sym(s6)" = torch.ops.aten.sym_size.int(arg877_1, 0);  arg877_1 = None
	        mul_2: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution, 0.5)
	        mul_3: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution, 0.7071067811865476);  convolution = None
	        erf: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.erf.default(mul_3);  mul_3 = None
	        add_4: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.add.Tensor(erf, 1);  erf = None
	        mul_4: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2, add_4);  mul_2 = add_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:350 in forward, code: inputs_embeds = nn.functional.gelu(self.conv2(inputs_embeds))
	        convolution_1: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.convolution.default(mul_4, arg3_1, arg4_1, [2], [1], [1], False, [0], 1);  mul_4 = arg3_1 = arg4_1 = None
	        mul_9: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution_1, 0.5)
	        mul_10: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution_1, 0.7071067811865476);  convolution_1 = None
	        erf_1: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.erf.default(mul_10);  mul_10 = None
	        add_13: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_1, 1);  erf_1 = None
	        mul_11: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9, add_13);  mul_9 = add_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:351 in forward, code: inputs_embeds = inputs_embeds.permute(0, 2, 1)
	        permute: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.permute.default(mul_11, [0, 2, 1]);  mul_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:354 in forward, code: hidden_states = (inputs_embeds + embed_pos).to(inputs_embeds.dtype)
	        add_22: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(permute, arg0_1);  permute = arg0_1 = None
	        _assert_tensor_metadata_1 = torch.ops.aten._assert_tensor_metadata.default(add_22, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:355 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.clone.default(add_22);  add_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(clone, memory_format = torch.contiguous_format)
	        var_mean = torch.ops.aten.var_mean.correction(clone_1, [2], correction = 0, keepdim = True)
	        getitem: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean[0]
	        getitem_1: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean[1];  var_mean = None
	        add_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem, 1e-05);  getitem = None
	        rsqrt: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_31);  add_31 = None
	        sub_7: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_1, getitem_1);  clone_1 = getitem_1 = None
	        mul_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7, rsqrt);  sub_7 = rsqrt = None
	        mul_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_20, arg5_1);  mul_20 = arg5_1 = None
	        add_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_21, arg6_1);  mul_21 = arg6_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_32, [sym_size_int, 1500, 1280])
	        amin: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view, [2])
	        amax: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view, [2]);  view = None
	        full: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin, full);  amin = full = None
	        full_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax, full_1);  amax = full_1 = None
	        sub_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum, minimum);  maximum = None
	        div: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_18, 255.0);  sub_18 = None
	        clamp_min: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div, 1.1920928955078125e-07);  div = None
	        div_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum, clamp_min);  minimum = None
	        round_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_1);  div_1 = None
	        sub_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_1);  round_1 = None
	        clamp_min_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_24, -128);  sub_24 = None
	        clamp_max: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_1, 127);  clamp_min_1 = None
	        _assert_tensor_metadata_2 = torch.ops.aten._assert_tensor_metadata.default(clamp_min, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_2 = None
	        _assert_tensor_metadata_3 = torch.ops.aten._assert_tensor_metadata.default(clamp_max, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_3 = None
	        convert_element_type: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max, torch.int8);  clamp_max = None
	        view_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_32, [sym_size_int, 1500, 1280])
	        view_2: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min, [sym_size_int, 1500, 1])
	        view_3: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type, [sym_size_int, 1500, 1])
	        reciprocal: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2);  view_2 = None
	        mul_69: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal, 1.0);  reciprocal = None
	        mul_72: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1, mul_69);  view_1 = mul_69 = None
	        round_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_72);  mul_72 = None
	        add_119: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_2, view_3);  round_2 = view_3 = None
	        clamp_min_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_119, -128);  add_119 = None
	        clamp_max_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_2, 127);  clamp_min_2 = None
	        view_4: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_1, [sym_size_int, 1500, 1280]);  clamp_max_1 = None
	        _assert_tensor_metadata_4 = torch.ops.aten._assert_tensor_metadata.default(view_4, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_4 = None
	        convert_element_type_1: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_4, torch.int8);  view_4 = None
	        view_5: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1, [sym_size_int, 1500, 1280]);  convert_element_type_1 = None
	        view_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min, [sym_size_int, 1500, 1]);  clamp_min = None
	        view_7: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type, [sym_size_int, 1500, 1]);  convert_element_type = None
	        _assert_tensor_metadata_5 = torch.ops.aten._assert_tensor_metadata.default(view_5, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_5 = None
	        convert_element_type_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_5, torch.float32);  view_5 = None
	        _assert_tensor_metadata_6 = torch.ops.aten._assert_tensor_metadata.default(view_7, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_6 = None
	        convert_element_type_3: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_7, torch.float32);  view_7 = None
	        sub_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_2, convert_element_type_3);  convert_element_type_2 = convert_element_type_3 = None
	        mul_94: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_44, view_6);  sub_44 = view_6 = None
	        view_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_94, [sym_size_int, 1500, 1280]);  mul_94 = None
	        _assert_tensor_metadata_7 = torch.ops.aten._assert_tensor_metadata.default(view_8, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_7 = None
	        view_9: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg8_1, [1280, 40, 32]);  arg8_1 = None
	        view_10: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg9_1, [1280, 40, 1]);  arg9_1 = None
	        view_11: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg10_1, [1280, 40, 1]);  arg10_1 = None
	        _assert_tensor_metadata_8 = torch.ops.aten._assert_tensor_metadata.default(view_9, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_8 = None
	        convert_element_type_4: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_9, torch.float32);  view_9 = None
	        _assert_tensor_metadata_9 = torch.ops.aten._assert_tensor_metadata.default(view_11, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_9 = None
	        convert_element_type_5: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_11, torch.float32);  view_11 = None
	        sub_48: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_4, convert_element_type_5);  convert_element_type_4 = convert_element_type_5 = None
	        mul_99: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_48, view_10);  sub_48 = view_10 = None
	        view_12: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_99, [1280, 1280]);  mul_99 = None
	        _assert_tensor_metadata_10 = torch.ops.aten._assert_tensor_metadata.default(view_12, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_10 = None
	        mul_104: "Sym(1500*s6)" = sym_size_int * 1500
	        view_13: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_8, [mul_104, 1280]);  view_8 = mul_104 = None
	        permute_1: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_12, [1, 0]);  view_12 = None
	        addmm: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg7_1, view_13, permute_1);  arg7_1 = view_13 = permute_1 = None
	        view_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm, [sym_size_int, 1500, 1280]);  addmm = None
	        mul_111: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_14, 0.125);  view_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_15: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_111, [sym_size_int, 1500, 20, 64]);  mul_111 = None
	        permute_2: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_15, [0, 2, 1, 3]);  view_15 = None
	        clone_2: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_2, memory_format = torch.contiguous_format);  permute_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_16: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_32, [sym_size_int, 1500, 1280])
	        amin_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_16, [2])
	        amax_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_16, [2]);  view_16 = None
	        full_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_1, full_2);  amin_1 = full_2 = None
	        full_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_1, full_3);  amax_1 = full_3 = None
	        sub_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_1, minimum_1);  maximum_1 = None
	        div_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_63, 255.0);  sub_63 = None
	        clamp_min_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_2, 1.1920928955078125e-07);  div_2 = None
	        div_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_1, clamp_min_3);  minimum_1 = None
	        round_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_3);  div_3 = None
	        sub_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_3);  round_3 = None
	        clamp_min_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_69, -128);  sub_69 = None
	        clamp_max_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_4, 127);  clamp_min_4 = None
	        _assert_tensor_metadata_11 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_3, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_11 = None
	        _assert_tensor_metadata_12 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_2, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_12 = None
	        convert_element_type_6: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_2, torch.int8);  clamp_max_2 = None
	        view_17: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_32, [sym_size_int, 1500, 1280])
	        view_18: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_3, [sym_size_int, 1500, 1])
	        view_19: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_6, [sym_size_int, 1500, 1])
	        reciprocal_1: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_18);  view_18 = None
	        mul_165: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_1, 1.0);  reciprocal_1 = None
	        mul_168: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_17, mul_165);  view_17 = mul_165 = None
	        round_4: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_168);  mul_168 = None
	        add_271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_4, view_19);  round_4 = view_19 = None
	        clamp_min_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_271, -128);  add_271 = None
	        clamp_max_3: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_5, 127);  clamp_min_5 = None
	        view_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_3, [sym_size_int, 1500, 1280]);  clamp_max_3 = None
	        _assert_tensor_metadata_13 = torch.ops.aten._assert_tensor_metadata.default(view_20, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_13 = None
	        convert_element_type_7: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_20, torch.int8);  view_20 = None
	        view_21: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_7, [sym_size_int, 1500, 1280]);  convert_element_type_7 = None
	        view_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_3, [sym_size_int, 1500, 1]);  clamp_min_3 = None
	        view_23: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_6, [sym_size_int, 1500, 1]);  convert_element_type_6 = None
	        _assert_tensor_metadata_14 = torch.ops.aten._assert_tensor_metadata.default(view_21, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_14 = None
	        convert_element_type_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_21, torch.float32);  view_21 = None
	        _assert_tensor_metadata_15 = torch.ops.aten._assert_tensor_metadata.default(view_23, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_15 = None
	        convert_element_type_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_23, torch.float32);  view_23 = None
	        sub_89: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_8, convert_element_type_9);  convert_element_type_8 = convert_element_type_9 = None
	        mul_190: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_89, view_22);  sub_89 = view_22 = None
	        view_24: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_190, [sym_size_int, 1500, 1280]);  mul_190 = None
	        _assert_tensor_metadata_16 = torch.ops.aten._assert_tensor_metadata.default(view_24, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_16 = None
	        view_25: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg11_1, [1280, 40, 32]);  arg11_1 = None
	        view_26: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg12_1, [1280, 40, 1]);  arg12_1 = None
	        view_27: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg13_1, [1280, 40, 1]);  arg13_1 = None
	        _assert_tensor_metadata_17 = torch.ops.aten._assert_tensor_metadata.default(view_25, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_17 = None
	        convert_element_type_10: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_25, torch.float32);  view_25 = None
	        _assert_tensor_metadata_18 = torch.ops.aten._assert_tensor_metadata.default(view_27, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_18 = None
	        convert_element_type_11: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_27, torch.float32);  view_27 = None
	        sub_93: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_10, convert_element_type_11);  convert_element_type_10 = convert_element_type_11 = None
	        mul_195: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_93, view_26);  sub_93 = view_26 = None
	        view_28: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_195, [1280, 1280]);  mul_195 = None
	        _assert_tensor_metadata_19 = torch.ops.aten._assert_tensor_metadata.default(view_28, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_19 = None
	        permute_3: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_28, [1, 0]);  view_28 = None
	        mul_198: "Sym(1500*s6)" = sym_size_int * 1500
	        view_29: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_24, [mul_198, 1280]);  view_24 = mul_198 = None
	        mm: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_29, permute_3);  view_29 = permute_3 = None
	        view_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm, [sym_size_int, 1500, 1280]);  mm = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_31: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_30, [sym_size_int, -1, 20, 64]);  view_30 = None
	        permute_4: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_31, [0, 2, 1, 3]);  view_31 = None
	        clone_3: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_4, memory_format = torch.contiguous_format);  permute_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_32, [sym_size_int, 1500, 1280])
	        amin_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_32, [2])
	        amax_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_32, [2]);  view_32 = None
	        full_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_2, full_4);  amin_2 = full_4 = None
	        full_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_2, full_5);  amax_2 = full_5 = None
	        sub_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_2, minimum_2);  maximum_2 = None
	        div_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_107, 255.0);  sub_107 = None
	        clamp_min_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_4, 1.1920928955078125e-07);  div_4 = None
	        div_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_2, clamp_min_6);  minimum_2 = None
	        round_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_5);  div_5 = None
	        sub_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_5);  round_5 = None
	        clamp_min_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_113, -128);  sub_113 = None
	        clamp_max_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_7, 127);  clamp_min_7 = None
	        _assert_tensor_metadata_20 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_6, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_20 = None
	        _assert_tensor_metadata_21 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_4, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_21 = None
	        convert_element_type_12: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_4, torch.int8);  clamp_max_4 = None
	        view_33: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_32, [sym_size_int, 1500, 1280]);  add_32 = None
	        view_34: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_6, [sym_size_int, 1500, 1])
	        view_35: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_12, [sym_size_int, 1500, 1])
	        reciprocal_2: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_34);  view_34 = None
	        mul_264: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_2, 1.0);  reciprocal_2 = None
	        mul_267: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_33, mul_264);  view_33 = mul_264 = None
	        round_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_267);  mul_267 = None
	        add_419: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_6, view_35);  round_6 = view_35 = None
	        clamp_min_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_419, -128);  add_419 = None
	        clamp_max_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_8, 127);  clamp_min_8 = None
	        view_36: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_5, [sym_size_int, 1500, 1280]);  clamp_max_5 = None
	        _assert_tensor_metadata_22 = torch.ops.aten._assert_tensor_metadata.default(view_36, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_22 = None
	        convert_element_type_13: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_36, torch.int8);  view_36 = None
	        view_37: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_13, [sym_size_int, 1500, 1280]);  convert_element_type_13 = None
	        view_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_6, [sym_size_int, 1500, 1]);  clamp_min_6 = None
	        view_39: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_12, [sym_size_int, 1500, 1]);  convert_element_type_12 = None
	        _assert_tensor_metadata_23 = torch.ops.aten._assert_tensor_metadata.default(view_37, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_23 = None
	        convert_element_type_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_37, torch.float32);  view_37 = None
	        _assert_tensor_metadata_24 = torch.ops.aten._assert_tensor_metadata.default(view_39, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_24 = None
	        convert_element_type_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_39, torch.float32);  view_39 = None
	        sub_133: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_14, convert_element_type_15);  convert_element_type_14 = convert_element_type_15 = None
	        mul_289: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_133, view_38);  sub_133 = view_38 = None
	        view_40: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_289, [sym_size_int, 1500, 1280]);  mul_289 = None
	        _assert_tensor_metadata_25 = torch.ops.aten._assert_tensor_metadata.default(view_40, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_25 = None
	        view_41: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg15_1, [1280, 40, 32]);  arg15_1 = None
	        view_42: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg16_1, [1280, 40, 1]);  arg16_1 = None
	        view_43: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg17_1, [1280, 40, 1]);  arg17_1 = None
	        _assert_tensor_metadata_26 = torch.ops.aten._assert_tensor_metadata.default(view_41, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_26 = None
	        convert_element_type_16: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_41, torch.float32);  view_41 = None
	        _assert_tensor_metadata_27 = torch.ops.aten._assert_tensor_metadata.default(view_43, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_27 = None
	        convert_element_type_17: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_43, torch.float32);  view_43 = None
	        sub_137: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_16, convert_element_type_17);  convert_element_type_16 = convert_element_type_17 = None
	        mul_294: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_137, view_42);  sub_137 = view_42 = None
	        view_44: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_294, [1280, 1280]);  mul_294 = None
	        _assert_tensor_metadata_28 = torch.ops.aten._assert_tensor_metadata.default(view_44, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_28 = None
	        mul_299: "Sym(1500*s6)" = sym_size_int * 1500
	        view_45: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_40, [mul_299, 1280]);  view_40 = mul_299 = None
	        permute_5: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_44, [1, 0]);  view_44 = None
	        addmm_1: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg14_1, view_45, permute_5);  arg14_1 = view_45 = permute_5 = None
	        view_46: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_1, [sym_size_int, 1500, 1280]);  addmm_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_47: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_46, [sym_size_int, -1, 20, 64]);  view_46 = None
	        permute_6: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_47, [0, 2, 1, 3]);  view_47 = None
	        clone_4: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_6, memory_format = torch.contiguous_format);  permute_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_2, clone_3, clone_4, None, False, scale = 1.0);  clone_2 = clone_3 = clone_4 = None
	        getitem_2: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention[0];  _scaled_dot_product_efficient_attention = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_7: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_2, [0, 2, 1, 3]);  getitem_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_48: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_7, [sym_size_int, 1500, -1]);  permute_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_49: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_48, [sym_size_int, 1500, 1280])
	        amin_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_49, [2])
	        amax_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_49, [2]);  view_49 = None
	        full_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_3, full_6);  amin_3 = full_6 = None
	        full_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_3, full_7);  amax_3 = full_7 = None
	        sub_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_3, minimum_3);  maximum_3 = None
	        div_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_155, 255.0);  sub_155 = None
	        clamp_min_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_6, 1.1920928955078125e-07);  div_6 = None
	        div_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_3, clamp_min_9);  minimum_3 = None
	        round_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_7);  div_7 = None
	        sub_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_7);  round_7 = None
	        clamp_min_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_161, -128);  sub_161 = None
	        clamp_max_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_10, 127);  clamp_min_10 = None
	        _assert_tensor_metadata_29 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_9, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_29 = None
	        _assert_tensor_metadata_30 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_6, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_30 = None
	        convert_element_type_18: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_6, torch.int8);  clamp_max_6 = None
	        view_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_48, [sym_size_int, 1500, 1280]);  view_48 = None
	        view_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_9, [sym_size_int, 1500, 1])
	        view_52: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_18, [sym_size_int, 1500, 1])
	        reciprocal_3: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_51);  view_51 = None
	        mul_369: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_3, 1.0);  reciprocal_3 = None
	        mul_372: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_50, mul_369);  view_50 = mul_369 = None
	        round_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_372);  mul_372 = None
	        add_583: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_8, view_52);  round_8 = view_52 = None
	        clamp_min_11: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_583, -128);  add_583 = None
	        clamp_max_7: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_11, 127);  clamp_min_11 = None
	        view_53: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_7, [sym_size_int, 1500, 1280]);  clamp_max_7 = None
	        _assert_tensor_metadata_31 = torch.ops.aten._assert_tensor_metadata.default(view_53, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_31 = None
	        convert_element_type_19: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_53, torch.int8);  view_53 = None
	        view_54: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_19, [sym_size_int, 1500, 1280]);  convert_element_type_19 = None
	        view_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_9, [sym_size_int, 1500, 1]);  clamp_min_9 = None
	        view_56: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_18, [sym_size_int, 1500, 1]);  convert_element_type_18 = None
	        _assert_tensor_metadata_32 = torch.ops.aten._assert_tensor_metadata.default(view_54, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_32 = None
	        convert_element_type_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_54, torch.float32);  view_54 = None
	        _assert_tensor_metadata_33 = torch.ops.aten._assert_tensor_metadata.default(view_56, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_33 = None
	        convert_element_type_21: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_56, torch.float32);  view_56 = None
	        sub_181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_20, convert_element_type_21);  convert_element_type_20 = convert_element_type_21 = None
	        mul_394: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_181, view_55);  sub_181 = view_55 = None
	        view_57: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_394, [sym_size_int, 1500, 1280]);  mul_394 = None
	        _assert_tensor_metadata_34 = torch.ops.aten._assert_tensor_metadata.default(view_57, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_34 = None
	        view_58: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg19_1, [1280, 40, 32]);  arg19_1 = None
	        view_59: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg20_1, [1280, 40, 1]);  arg20_1 = None
	        view_60: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg21_1, [1280, 40, 1]);  arg21_1 = None
	        _assert_tensor_metadata_35 = torch.ops.aten._assert_tensor_metadata.default(view_58, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_35 = None
	        convert_element_type_22: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_58, torch.float32);  view_58 = None
	        _assert_tensor_metadata_36 = torch.ops.aten._assert_tensor_metadata.default(view_60, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_36 = None
	        convert_element_type_23: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_60, torch.float32);  view_60 = None
	        sub_185: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_22, convert_element_type_23);  convert_element_type_22 = convert_element_type_23 = None
	        mul_399: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_185, view_59);  sub_185 = view_59 = None
	        view_61: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_399, [1280, 1280]);  mul_399 = None
	        _assert_tensor_metadata_37 = torch.ops.aten._assert_tensor_metadata.default(view_61, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_37 = None
	        mul_404: "Sym(1500*s6)" = sym_size_int * 1500
	        view_62: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_57, [mul_404, 1280]);  view_57 = mul_404 = None
	        permute_8: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_61, [1, 0]);  view_61 = None
	        addmm_2: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg18_1, view_62, permute_8);  arg18_1 = view_62 = permute_8 = None
	        view_63: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_2, [sym_size_int, 1500, 1280]);  addmm_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_63);  view_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_646: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(clone, clone_5);  clone = clone_5 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_646, memory_format = torch.contiguous_format)
	        var_mean_1 = torch.ops.aten.var_mean.correction(clone_6, [2], correction = 0, keepdim = True)
	        getitem_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_1[0]
	        getitem_7: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_1[1];  var_mean_1 = None
	        add_651: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_6, 1e-05);  getitem_6 = None
	        rsqrt_1: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_651);  add_651 = None
	        sub_191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_6, getitem_7);  clone_6 = getitem_7 = None
	        mul_415: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_191, rsqrt_1);  sub_191 = rsqrt_1 = None
	        mul_416: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_415, arg22_1);  mul_415 = arg22_1 = None
	        add_652: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_416, arg23_1);  mul_416 = arg23_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_64: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_652, [sym_size_int, 1500, 1280])
	        amin_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_64, [2])
	        amax_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_64, [2]);  view_64 = None
	        full_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_4, full_8);  amin_4 = full_8 = None
	        full_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_4, full_9);  amax_4 = full_9 = None
	        sub_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_4, minimum_4);  maximum_4 = None
	        div_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_202, 255.0);  sub_202 = None
	        clamp_min_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_8, 1.1920928955078125e-07);  div_8 = None
	        div_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_4, clamp_min_12);  minimum_4 = None
	        round_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_9);  div_9 = None
	        sub_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_9);  round_9 = None
	        clamp_min_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_208, -128);  sub_208 = None
	        clamp_max_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_13, 127);  clamp_min_13 = None
	        _assert_tensor_metadata_38 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_12, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_38 = None
	        _assert_tensor_metadata_39 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_8, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_39 = None
	        convert_element_type_24: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_8, torch.int8);  clamp_max_8 = None
	        view_65: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_652, [sym_size_int, 1500, 1280]);  add_652 = None
	        view_66: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_12, [sym_size_int, 1500, 1])
	        view_67: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_24, [sym_size_int, 1500, 1])
	        reciprocal_4: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_66);  view_66 = None
	        mul_464: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_4, 1.0);  reciprocal_4 = None
	        mul_467: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_65, mul_464);  view_65 = mul_464 = None
	        round_10: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_467);  mul_467 = None
	        add_739: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_10, view_67);  round_10 = view_67 = None
	        clamp_min_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_739, -128);  add_739 = None
	        clamp_max_9: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_14, 127);  clamp_min_14 = None
	        view_68: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_9, [sym_size_int, 1500, 1280]);  clamp_max_9 = None
	        _assert_tensor_metadata_40 = torch.ops.aten._assert_tensor_metadata.default(view_68, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_40 = None
	        convert_element_type_25: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_68, torch.int8);  view_68 = None
	        view_69: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_25, [sym_size_int, 1500, 1280]);  convert_element_type_25 = None
	        view_70: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_12, [sym_size_int, 1500, 1]);  clamp_min_12 = None
	        view_71: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_24, [sym_size_int, 1500, 1]);  convert_element_type_24 = None
	        _assert_tensor_metadata_41 = torch.ops.aten._assert_tensor_metadata.default(view_69, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_41 = None
	        convert_element_type_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_69, torch.float32);  view_69 = None
	        _assert_tensor_metadata_42 = torch.ops.aten._assert_tensor_metadata.default(view_71, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_42 = None
	        convert_element_type_27: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_71, torch.float32);  view_71 = None
	        sub_228: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_26, convert_element_type_27);  convert_element_type_26 = convert_element_type_27 = None
	        mul_489: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_228, view_70);  sub_228 = view_70 = None
	        view_72: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_489, [sym_size_int, 1500, 1280]);  mul_489 = None
	        _assert_tensor_metadata_43 = torch.ops.aten._assert_tensor_metadata.default(view_72, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_43 = None
	        view_73: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg25_1, [5120, 40, 32]);  arg25_1 = None
	        view_74: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg26_1, [5120, 40, 1]);  arg26_1 = None
	        view_75: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg27_1, [5120, 40, 1]);  arg27_1 = None
	        _assert_tensor_metadata_44 = torch.ops.aten._assert_tensor_metadata.default(view_73, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_44 = None
	        convert_element_type_28: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_73, torch.float32);  view_73 = None
	        _assert_tensor_metadata_45 = torch.ops.aten._assert_tensor_metadata.default(view_75, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_45 = None
	        convert_element_type_29: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_75, torch.float32);  view_75 = None
	        sub_232: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_28, convert_element_type_29);  convert_element_type_28 = convert_element_type_29 = None
	        mul_494: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_232, view_74);  sub_232 = view_74 = None
	        view_76: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_494, [5120, 1280]);  mul_494 = None
	        _assert_tensor_metadata_46 = torch.ops.aten._assert_tensor_metadata.default(view_76, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_46 = None
	        mul_499: "Sym(1500*s6)" = sym_size_int * 1500
	        view_77: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_72, [mul_499, 1280]);  view_72 = mul_499 = None
	        permute_9: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_76, [1, 0]);  view_76 = None
	        addmm_3: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg24_1, view_77, permute_9);  arg24_1 = view_77 = permute_9 = None
	        view_78: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_3, [sym_size_int, 1500, 5120]);  addmm_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_506: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_78, 0.5)
	        mul_507: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_78, 0.7071067811865476);  view_78 = None
	        erf_2: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_507);  mul_507 = None
	        add_798: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_2, 1);  erf_2 = None
	        mul_508: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_506, add_798);  mul_506 = add_798 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_7: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_508);  mul_508 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_79: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_7, [sym_size_int, 1500, 5120])
	        amin_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_79, [2])
	        amax_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_79, [2]);  view_79 = None
	        full_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_5, full_10);  amin_5 = full_10 = None
	        full_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_5, full_11);  amax_5 = full_11 = None
	        sub_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_5, minimum_5);  maximum_5 = None
	        div_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_245, 255.0);  sub_245 = None
	        clamp_min_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_10, 1.1920928955078125e-07);  div_10 = None
	        div_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_5, clamp_min_15);  minimum_5 = None
	        round_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_11);  div_11 = None
	        sub_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_11);  round_11 = None
	        clamp_min_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_251, -128);  sub_251 = None
	        clamp_max_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_16, 127);  clamp_min_16 = None
	        _assert_tensor_metadata_47 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_15, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_47 = None
	        _assert_tensor_metadata_48 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_10, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_48 = None
	        convert_element_type_30: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_10, torch.int8);  clamp_max_10 = None
	        view_80: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_7, [sym_size_int, 1500, 5120]);  clone_7 = None
	        view_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_15, [sym_size_int, 1500, 1])
	        view_82: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_30, [sym_size_int, 1500, 1])
	        reciprocal_5: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_81);  view_81 = None
	        mul_554: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_5, 1.0);  reciprocal_5 = None
	        mul_557: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_80, mul_554);  view_80 = mul_554 = None
	        round_12: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_557);  mul_557 = None
	        add_881: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_12, view_82);  round_12 = view_82 = None
	        clamp_min_17: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_881, -128);  add_881 = None
	        clamp_max_11: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_17, 127);  clamp_min_17 = None
	        view_83: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_11, [sym_size_int, 1500, 5120]);  clamp_max_11 = None
	        _assert_tensor_metadata_49 = torch.ops.aten._assert_tensor_metadata.default(view_83, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_49 = None
	        convert_element_type_31: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_83, torch.int8);  view_83 = None
	        view_84: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_31, [sym_size_int, 1500, 5120]);  convert_element_type_31 = None
	        view_85: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_15, [sym_size_int, 1500, 1]);  clamp_min_15 = None
	        view_86: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_30, [sym_size_int, 1500, 1]);  convert_element_type_30 = None
	        _assert_tensor_metadata_50 = torch.ops.aten._assert_tensor_metadata.default(view_84, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_50 = None
	        convert_element_type_32: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_84, torch.float32);  view_84 = None
	        _assert_tensor_metadata_51 = torch.ops.aten._assert_tensor_metadata.default(view_86, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_51 = None
	        convert_element_type_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_86, torch.float32);  view_86 = None
	        sub_271: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_32, convert_element_type_33);  convert_element_type_32 = convert_element_type_33 = None
	        mul_579: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_271, view_85);  sub_271 = view_85 = None
	        view_87: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_579, [sym_size_int, 1500, 5120]);  mul_579 = None
	        _assert_tensor_metadata_52 = torch.ops.aten._assert_tensor_metadata.default(view_87, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_52 = None
	        view_88: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg29_1, [1280, 160, 32]);  arg29_1 = None
	        view_89: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg30_1, [1280, 160, 1]);  arg30_1 = None
	        view_90: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg31_1, [1280, 160, 1]);  arg31_1 = None
	        _assert_tensor_metadata_53 = torch.ops.aten._assert_tensor_metadata.default(view_88, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_53 = None
	        convert_element_type_34: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_88, torch.float32);  view_88 = None
	        _assert_tensor_metadata_54 = torch.ops.aten._assert_tensor_metadata.default(view_90, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_54 = None
	        convert_element_type_35: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_90, torch.float32);  view_90 = None
	        sub_275: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_34, convert_element_type_35);  convert_element_type_34 = convert_element_type_35 = None
	        mul_584: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_275, view_89);  sub_275 = view_89 = None
	        view_91: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_584, [1280, 5120]);  mul_584 = None
	        _assert_tensor_metadata_55 = torch.ops.aten._assert_tensor_metadata.default(view_91, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_55 = None
	        mul_589: "Sym(1500*s6)" = sym_size_int * 1500
	        view_92: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_87, [mul_589, 5120]);  view_87 = mul_589 = None
	        permute_10: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_91, [1, 0]);  view_91 = None
	        addmm_4: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg28_1, view_92, permute_10);  arg28_1 = view_92 = permute_10 = None
	        view_93: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_4, [sym_size_int, 1500, 1280]);  addmm_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_93);  view_93 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_944: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_646, clone_8);  add_646 = clone_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_9: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_944, memory_format = torch.contiguous_format)
	        var_mean_2 = torch.ops.aten.var_mean.correction(clone_9, [2], correction = 0, keepdim = True)
	        getitem_8: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_2[0]
	        getitem_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_2[1];  var_mean_2 = None
	        add_949: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_8, 1e-05);  getitem_8 = None
	        rsqrt_2: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_949);  add_949 = None
	        sub_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_9, getitem_9);  clone_9 = getitem_9 = None
	        mul_600: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_281, rsqrt_2);  sub_281 = rsqrt_2 = None
	        mul_601: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_600, arg32_1);  mul_600 = arg32_1 = None
	        add_950: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_601, arg33_1);  mul_601 = arg33_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_94: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_950, [sym_size_int, 1500, 1280])
	        amin_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_94, [2])
	        amax_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_94, [2]);  view_94 = None
	        full_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_6, full_12);  amin_6 = full_12 = None
	        full_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_6, full_13);  amax_6 = full_13 = None
	        sub_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_6, minimum_6);  maximum_6 = None
	        div_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_292, 255.0);  sub_292 = None
	        clamp_min_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_12, 1.1920928955078125e-07);  div_12 = None
	        div_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_6, clamp_min_18);  minimum_6 = None
	        round_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_13);  div_13 = None
	        sub_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_13);  round_13 = None
	        clamp_min_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_298, -128);  sub_298 = None
	        clamp_max_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_19, 127);  clamp_min_19 = None
	        _assert_tensor_metadata_56 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_18, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_56 = None
	        _assert_tensor_metadata_57 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_12, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_57 = None
	        convert_element_type_36: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_12, torch.int8);  clamp_max_12 = None
	        view_95: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_950, [sym_size_int, 1500, 1280])
	        view_96: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_18, [sym_size_int, 1500, 1])
	        view_97: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_36, [sym_size_int, 1500, 1])
	        reciprocal_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_96);  view_96 = None
	        mul_649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_6, 1.0);  reciprocal_6 = None
	        mul_652: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_95, mul_649);  view_95 = mul_649 = None
	        round_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_652);  mul_652 = None
	        add_1037: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_14, view_97);  round_14 = view_97 = None
	        clamp_min_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1037, -128);  add_1037 = None
	        clamp_max_13: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_20, 127);  clamp_min_20 = None
	        view_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_13, [sym_size_int, 1500, 1280]);  clamp_max_13 = None
	        _assert_tensor_metadata_58 = torch.ops.aten._assert_tensor_metadata.default(view_98, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_58 = None
	        convert_element_type_37: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_98, torch.int8);  view_98 = None
	        view_99: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_37, [sym_size_int, 1500, 1280]);  convert_element_type_37 = None
	        view_100: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_18, [sym_size_int, 1500, 1]);  clamp_min_18 = None
	        view_101: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_36, [sym_size_int, 1500, 1]);  convert_element_type_36 = None
	        _assert_tensor_metadata_59 = torch.ops.aten._assert_tensor_metadata.default(view_99, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_59 = None
	        convert_element_type_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_99, torch.float32);  view_99 = None
	        _assert_tensor_metadata_60 = torch.ops.aten._assert_tensor_metadata.default(view_101, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_60 = None
	        convert_element_type_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_101, torch.float32);  view_101 = None
	        sub_318: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_38, convert_element_type_39);  convert_element_type_38 = convert_element_type_39 = None
	        mul_674: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_318, view_100);  sub_318 = view_100 = None
	        view_102: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_674, [sym_size_int, 1500, 1280]);  mul_674 = None
	        _assert_tensor_metadata_61 = torch.ops.aten._assert_tensor_metadata.default(view_102, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_61 = None
	        view_103: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg35_1, [1280, 40, 32]);  arg35_1 = None
	        view_104: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg36_1, [1280, 40, 1]);  arg36_1 = None
	        view_105: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg37_1, [1280, 40, 1]);  arg37_1 = None
	        _assert_tensor_metadata_62 = torch.ops.aten._assert_tensor_metadata.default(view_103, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_62 = None
	        convert_element_type_40: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_103, torch.float32);  view_103 = None
	        _assert_tensor_metadata_63 = torch.ops.aten._assert_tensor_metadata.default(view_105, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_63 = None
	        convert_element_type_41: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_105, torch.float32);  view_105 = None
	        sub_322: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_40, convert_element_type_41);  convert_element_type_40 = convert_element_type_41 = None
	        mul_679: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_322, view_104);  sub_322 = view_104 = None
	        view_106: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_679, [1280, 1280]);  mul_679 = None
	        _assert_tensor_metadata_64 = torch.ops.aten._assert_tensor_metadata.default(view_106, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_64 = None
	        mul_684: "Sym(1500*s6)" = sym_size_int * 1500
	        view_107: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_102, [mul_684, 1280]);  view_102 = mul_684 = None
	        permute_11: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_106, [1, 0]);  view_106 = None
	        addmm_5: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg34_1, view_107, permute_11);  arg34_1 = view_107 = permute_11 = None
	        view_108: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_5, [sym_size_int, 1500, 1280]);  addmm_5 = None
	        mul_691: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_108, 0.125);  view_108 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_109: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_691, [sym_size_int, 1500, 20, 64]);  mul_691 = None
	        permute_12: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_109, [0, 2, 1, 3]);  view_109 = None
	        clone_10: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_12, memory_format = torch.contiguous_format);  permute_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_950, [sym_size_int, 1500, 1280])
	        amin_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_110, [2])
	        amax_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_110, [2]);  view_110 = None
	        full_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_7, full_14);  amin_7 = full_14 = None
	        full_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_7, full_15);  amax_7 = full_15 = None
	        sub_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_7, minimum_7);  maximum_7 = None
	        div_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_337, 255.0);  sub_337 = None
	        clamp_min_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_14, 1.1920928955078125e-07);  div_14 = None
	        div_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_7, clamp_min_21);  minimum_7 = None
	        round_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_15);  div_15 = None
	        sub_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_15);  round_15 = None
	        clamp_min_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_343, -128);  sub_343 = None
	        clamp_max_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_22, 127);  clamp_min_22 = None
	        _assert_tensor_metadata_65 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_21, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_65 = None
	        _assert_tensor_metadata_66 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_14, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_66 = None
	        convert_element_type_42: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_14, torch.int8);  clamp_max_14 = None
	        view_111: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_950, [sym_size_int, 1500, 1280])
	        view_112: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_21, [sym_size_int, 1500, 1])
	        view_113: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_42, [sym_size_int, 1500, 1])
	        reciprocal_7: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_112);  view_112 = None
	        mul_745: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_7, 1.0);  reciprocal_7 = None
	        mul_748: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_111, mul_745);  view_111 = mul_745 = None
	        round_16: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_748);  mul_748 = None
	        add_1189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_16, view_113);  round_16 = view_113 = None
	        clamp_min_23: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1189, -128);  add_1189 = None
	        clamp_max_15: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_23, 127);  clamp_min_23 = None
	        view_114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_15, [sym_size_int, 1500, 1280]);  clamp_max_15 = None
	        _assert_tensor_metadata_67 = torch.ops.aten._assert_tensor_metadata.default(view_114, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_67 = None
	        convert_element_type_43: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_114, torch.int8);  view_114 = None
	        view_115: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_43, [sym_size_int, 1500, 1280]);  convert_element_type_43 = None
	        view_116: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_21, [sym_size_int, 1500, 1]);  clamp_min_21 = None
	        view_117: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_42, [sym_size_int, 1500, 1]);  convert_element_type_42 = None
	        _assert_tensor_metadata_68 = torch.ops.aten._assert_tensor_metadata.default(view_115, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_68 = None
	        convert_element_type_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_115, torch.float32);  view_115 = None
	        _assert_tensor_metadata_69 = torch.ops.aten._assert_tensor_metadata.default(view_117, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_69 = None
	        convert_element_type_45: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_117, torch.float32);  view_117 = None
	        sub_363: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_44, convert_element_type_45);  convert_element_type_44 = convert_element_type_45 = None
	        mul_770: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_363, view_116);  sub_363 = view_116 = None
	        view_118: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_770, [sym_size_int, 1500, 1280]);  mul_770 = None
	        _assert_tensor_metadata_70 = torch.ops.aten._assert_tensor_metadata.default(view_118, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_70 = None
	        view_119: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg38_1, [1280, 40, 32]);  arg38_1 = None
	        view_120: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg39_1, [1280, 40, 1]);  arg39_1 = None
	        view_121: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg40_1, [1280, 40, 1]);  arg40_1 = None
	        _assert_tensor_metadata_71 = torch.ops.aten._assert_tensor_metadata.default(view_119, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_71 = None
	        convert_element_type_46: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_119, torch.float32);  view_119 = None
	        _assert_tensor_metadata_72 = torch.ops.aten._assert_tensor_metadata.default(view_121, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_72 = None
	        convert_element_type_47: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_121, torch.float32);  view_121 = None
	        sub_367: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_46, convert_element_type_47);  convert_element_type_46 = convert_element_type_47 = None
	        mul_775: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_367, view_120);  sub_367 = view_120 = None
	        view_122: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_775, [1280, 1280]);  mul_775 = None
	        _assert_tensor_metadata_73 = torch.ops.aten._assert_tensor_metadata.default(view_122, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_73 = None
	        permute_13: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_122, [1, 0]);  view_122 = None
	        mul_778: "Sym(1500*s6)" = sym_size_int * 1500
	        view_123: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_118, [mul_778, 1280]);  view_118 = mul_778 = None
	        mm_1: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_123, permute_13);  view_123 = permute_13 = None
	        view_124: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_1, [sym_size_int, 1500, 1280]);  mm_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_125: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_124, [sym_size_int, -1, 20, 64]);  view_124 = None
	        permute_14: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_125, [0, 2, 1, 3]);  view_125 = None
	        clone_11: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_14, memory_format = torch.contiguous_format);  permute_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_126: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_950, [sym_size_int, 1500, 1280])
	        amin_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_126, [2])
	        amax_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_126, [2]);  view_126 = None
	        full_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_8, full_16);  amin_8 = full_16 = None
	        full_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_8, full_17);  amax_8 = full_17 = None
	        sub_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_8, minimum_8);  maximum_8 = None
	        div_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_381, 255.0);  sub_381 = None
	        clamp_min_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_16, 1.1920928955078125e-07);  div_16 = None
	        div_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_8, clamp_min_24);  minimum_8 = None
	        round_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_17);  div_17 = None
	        sub_387: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_17);  round_17 = None
	        clamp_min_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_387, -128);  sub_387 = None
	        clamp_max_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_25, 127);  clamp_min_25 = None
	        _assert_tensor_metadata_74 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_24, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_74 = None
	        _assert_tensor_metadata_75 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_16, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_75 = None
	        convert_element_type_48: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_16, torch.int8);  clamp_max_16 = None
	        view_127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_950, [sym_size_int, 1500, 1280]);  add_950 = None
	        view_128: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_24, [sym_size_int, 1500, 1])
	        view_129: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_48, [sym_size_int, 1500, 1])
	        reciprocal_8: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_128);  view_128 = None
	        mul_844: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_8, 1.0);  reciprocal_8 = None
	        mul_847: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_127, mul_844);  view_127 = mul_844 = None
	        round_18: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_847);  mul_847 = None
	        add_1337: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_18, view_129);  round_18 = view_129 = None
	        clamp_min_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1337, -128);  add_1337 = None
	        clamp_max_17: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_26, 127);  clamp_min_26 = None
	        view_130: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_17, [sym_size_int, 1500, 1280]);  clamp_max_17 = None
	        _assert_tensor_metadata_76 = torch.ops.aten._assert_tensor_metadata.default(view_130, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_76 = None
	        convert_element_type_49: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_130, torch.int8);  view_130 = None
	        view_131: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_49, [sym_size_int, 1500, 1280]);  convert_element_type_49 = None
	        view_132: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_24, [sym_size_int, 1500, 1]);  clamp_min_24 = None
	        view_133: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_48, [sym_size_int, 1500, 1]);  convert_element_type_48 = None
	        _assert_tensor_metadata_77 = torch.ops.aten._assert_tensor_metadata.default(view_131, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_77 = None
	        convert_element_type_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_131, torch.float32);  view_131 = None
	        _assert_tensor_metadata_78 = torch.ops.aten._assert_tensor_metadata.default(view_133, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_78 = None
	        convert_element_type_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_133, torch.float32);  view_133 = None
	        sub_407: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_50, convert_element_type_51);  convert_element_type_50 = convert_element_type_51 = None
	        mul_869: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_407, view_132);  sub_407 = view_132 = None
	        view_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_869, [sym_size_int, 1500, 1280]);  mul_869 = None
	        _assert_tensor_metadata_79 = torch.ops.aten._assert_tensor_metadata.default(view_134, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_79 = None
	        view_135: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg42_1, [1280, 40, 32]);  arg42_1 = None
	        view_136: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg43_1, [1280, 40, 1]);  arg43_1 = None
	        view_137: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg44_1, [1280, 40, 1]);  arg44_1 = None
	        _assert_tensor_metadata_80 = torch.ops.aten._assert_tensor_metadata.default(view_135, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_80 = None
	        convert_element_type_52: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_135, torch.float32);  view_135 = None
	        _assert_tensor_metadata_81 = torch.ops.aten._assert_tensor_metadata.default(view_137, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_81 = None
	        convert_element_type_53: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_137, torch.float32);  view_137 = None
	        sub_411: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_52, convert_element_type_53);  convert_element_type_52 = convert_element_type_53 = None
	        mul_874: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_411, view_136);  sub_411 = view_136 = None
	        view_138: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_874, [1280, 1280]);  mul_874 = None
	        _assert_tensor_metadata_82 = torch.ops.aten._assert_tensor_metadata.default(view_138, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_82 = None
	        mul_879: "Sym(1500*s6)" = sym_size_int * 1500
	        view_139: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_134, [mul_879, 1280]);  view_134 = mul_879 = None
	        permute_15: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_138, [1, 0]);  view_138 = None
	        addmm_6: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg41_1, view_139, permute_15);  arg41_1 = view_139 = permute_15 = None
	        view_140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_6, [sym_size_int, 1500, 1280]);  addmm_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_141: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_140, [sym_size_int, -1, 20, 64]);  view_140 = None
	        permute_16: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_141, [0, 2, 1, 3]);  view_141 = None
	        clone_12: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_16, memory_format = torch.contiguous_format);  permute_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_1 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_10, clone_11, clone_12, None, False, scale = 1.0);  clone_10 = clone_11 = clone_12 = None
	        getitem_10: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_1[0];  _scaled_dot_product_efficient_attention_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_17: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_10, [0, 2, 1, 3]);  getitem_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_17, [sym_size_int, 1500, -1]);  permute_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_143: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_142, [sym_size_int, 1500, 1280])
	        amin_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_143, [2])
	        amax_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_143, [2]);  view_143 = None
	        full_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_9, full_18);  amin_9 = full_18 = None
	        full_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_9, full_19);  amax_9 = full_19 = None
	        sub_429: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_9, minimum_9);  maximum_9 = None
	        div_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_429, 255.0);  sub_429 = None
	        clamp_min_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_18, 1.1920928955078125e-07);  div_18 = None
	        div_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_9, clamp_min_27);  minimum_9 = None
	        round_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_19);  div_19 = None
	        sub_435: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_19);  round_19 = None
	        clamp_min_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_435, -128);  sub_435 = None
	        clamp_max_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_28, 127);  clamp_min_28 = None
	        _assert_tensor_metadata_83 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_27, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_83 = None
	        _assert_tensor_metadata_84 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_18, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_84 = None
	        convert_element_type_54: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_18, torch.int8);  clamp_max_18 = None
	        view_144: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_142, [sym_size_int, 1500, 1280]);  view_142 = None
	        view_145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_27, [sym_size_int, 1500, 1])
	        view_146: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_54, [sym_size_int, 1500, 1])
	        reciprocal_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_145);  view_145 = None
	        mul_949: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_9, 1.0);  reciprocal_9 = None
	        mul_952: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_144, mul_949);  view_144 = mul_949 = None
	        round_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_952);  mul_952 = None
	        add_1501: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_20, view_146);  round_20 = view_146 = None
	        clamp_min_29: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1501, -128);  add_1501 = None
	        clamp_max_19: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_29, 127);  clamp_min_29 = None
	        view_147: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_19, [sym_size_int, 1500, 1280]);  clamp_max_19 = None
	        _assert_tensor_metadata_85 = torch.ops.aten._assert_tensor_metadata.default(view_147, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_85 = None
	        convert_element_type_55: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_147, torch.int8);  view_147 = None
	        view_148: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_55, [sym_size_int, 1500, 1280]);  convert_element_type_55 = None
	        view_149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_27, [sym_size_int, 1500, 1]);  clamp_min_27 = None
	        view_150: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_54, [sym_size_int, 1500, 1]);  convert_element_type_54 = None
	        _assert_tensor_metadata_86 = torch.ops.aten._assert_tensor_metadata.default(view_148, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_86 = None
	        convert_element_type_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_148, torch.float32);  view_148 = None
	        _assert_tensor_metadata_87 = torch.ops.aten._assert_tensor_metadata.default(view_150, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_87 = None
	        convert_element_type_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_150, torch.float32);  view_150 = None
	        sub_455: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_56, convert_element_type_57);  convert_element_type_56 = convert_element_type_57 = None
	        mul_974: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_455, view_149);  sub_455 = view_149 = None
	        view_151: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_974, [sym_size_int, 1500, 1280]);  mul_974 = None
	        _assert_tensor_metadata_88 = torch.ops.aten._assert_tensor_metadata.default(view_151, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_88 = None
	        view_152: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg46_1, [1280, 40, 32]);  arg46_1 = None
	        view_153: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg47_1, [1280, 40, 1]);  arg47_1 = None
	        view_154: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg48_1, [1280, 40, 1]);  arg48_1 = None
	        _assert_tensor_metadata_89 = torch.ops.aten._assert_tensor_metadata.default(view_152, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_89 = None
	        convert_element_type_58: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_152, torch.float32);  view_152 = None
	        _assert_tensor_metadata_90 = torch.ops.aten._assert_tensor_metadata.default(view_154, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_90 = None
	        convert_element_type_59: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_154, torch.float32);  view_154 = None
	        sub_459: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_58, convert_element_type_59);  convert_element_type_58 = convert_element_type_59 = None
	        mul_979: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_459, view_153);  sub_459 = view_153 = None
	        view_155: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_979, [1280, 1280]);  mul_979 = None
	        _assert_tensor_metadata_91 = torch.ops.aten._assert_tensor_metadata.default(view_155, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_91 = None
	        mul_984: "Sym(1500*s6)" = sym_size_int * 1500
	        view_156: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_151, [mul_984, 1280]);  view_151 = mul_984 = None
	        permute_18: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_155, [1, 0]);  view_155 = None
	        addmm_7: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg45_1, view_156, permute_18);  arg45_1 = view_156 = permute_18 = None
	        view_157: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_7, [sym_size_int, 1500, 1280]);  addmm_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_13: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_157);  view_157 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_1564: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_944, clone_13);  add_944 = clone_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_1564, memory_format = torch.contiguous_format)
	        var_mean_3 = torch.ops.aten.var_mean.correction(clone_14, [2], correction = 0, keepdim = True)
	        getitem_14: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_3[0]
	        getitem_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_3[1];  var_mean_3 = None
	        add_1569: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_14, 1e-05);  getitem_14 = None
	        rsqrt_3: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_1569);  add_1569 = None
	        sub_465: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_14, getitem_15);  clone_14 = getitem_15 = None
	        mul_995: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_465, rsqrt_3);  sub_465 = rsqrt_3 = None
	        mul_996: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_995, arg49_1);  mul_995 = arg49_1 = None
	        add_1570: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_996, arg50_1);  mul_996 = arg50_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_1570, [sym_size_int, 1500, 1280])
	        amin_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_158, [2])
	        amax_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_158, [2]);  view_158 = None
	        full_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_10, full_20);  amin_10 = full_20 = None
	        full_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_10, full_21);  amax_10 = full_21 = None
	        sub_476: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_10, minimum_10);  maximum_10 = None
	        div_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_476, 255.0);  sub_476 = None
	        clamp_min_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_20, 1.1920928955078125e-07);  div_20 = None
	        div_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_10, clamp_min_30);  minimum_10 = None
	        round_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_21);  div_21 = None
	        sub_482: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_21);  round_21 = None
	        clamp_min_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_482, -128);  sub_482 = None
	        clamp_max_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_31, 127);  clamp_min_31 = None
	        _assert_tensor_metadata_92 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_30, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_92 = None
	        _assert_tensor_metadata_93 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_20, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_93 = None
	        convert_element_type_60: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_20, torch.int8);  clamp_max_20 = None
	        view_159: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_1570, [sym_size_int, 1500, 1280]);  add_1570 = None
	        view_160: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_30, [sym_size_int, 1500, 1])
	        view_161: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_60, [sym_size_int, 1500, 1])
	        reciprocal_10: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_160);  view_160 = None
	        mul_1044: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_10, 1.0);  reciprocal_10 = None
	        mul_1047: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_159, mul_1044);  view_159 = mul_1044 = None
	        round_22: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1047);  mul_1047 = None
	        add_1657: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_22, view_161);  round_22 = view_161 = None
	        clamp_min_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1657, -128);  add_1657 = None
	        clamp_max_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_32, 127);  clamp_min_32 = None
	        view_162: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_21, [sym_size_int, 1500, 1280]);  clamp_max_21 = None
	        _assert_tensor_metadata_94 = torch.ops.aten._assert_tensor_metadata.default(view_162, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_94 = None
	        convert_element_type_61: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_162, torch.int8);  view_162 = None
	        view_163: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_61, [sym_size_int, 1500, 1280]);  convert_element_type_61 = None
	        view_164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_30, [sym_size_int, 1500, 1]);  clamp_min_30 = None
	        view_165: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_60, [sym_size_int, 1500, 1]);  convert_element_type_60 = None
	        _assert_tensor_metadata_95 = torch.ops.aten._assert_tensor_metadata.default(view_163, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_95 = None
	        convert_element_type_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_163, torch.float32);  view_163 = None
	        _assert_tensor_metadata_96 = torch.ops.aten._assert_tensor_metadata.default(view_165, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_96 = None
	        convert_element_type_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_165, torch.float32);  view_165 = None
	        sub_502: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_62, convert_element_type_63);  convert_element_type_62 = convert_element_type_63 = None
	        mul_1069: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_502, view_164);  sub_502 = view_164 = None
	        view_166: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1069, [sym_size_int, 1500, 1280]);  mul_1069 = None
	        _assert_tensor_metadata_97 = torch.ops.aten._assert_tensor_metadata.default(view_166, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_97 = None
	        view_167: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg52_1, [5120, 40, 32]);  arg52_1 = None
	        view_168: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg53_1, [5120, 40, 1]);  arg53_1 = None
	        view_169: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg54_1, [5120, 40, 1]);  arg54_1 = None
	        _assert_tensor_metadata_98 = torch.ops.aten._assert_tensor_metadata.default(view_167, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_98 = None
	        convert_element_type_64: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_167, torch.float32);  view_167 = None
	        _assert_tensor_metadata_99 = torch.ops.aten._assert_tensor_metadata.default(view_169, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_99 = None
	        convert_element_type_65: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_169, torch.float32);  view_169 = None
	        sub_506: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_64, convert_element_type_65);  convert_element_type_64 = convert_element_type_65 = None
	        mul_1074: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_506, view_168);  sub_506 = view_168 = None
	        view_170: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1074, [5120, 1280]);  mul_1074 = None
	        _assert_tensor_metadata_100 = torch.ops.aten._assert_tensor_metadata.default(view_170, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_100 = None
	        mul_1079: "Sym(1500*s6)" = sym_size_int * 1500
	        view_171: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_166, [mul_1079, 1280]);  view_166 = mul_1079 = None
	        permute_19: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_170, [1, 0]);  view_170 = None
	        addmm_8: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg51_1, view_171, permute_19);  arg51_1 = view_171 = permute_19 = None
	        view_172: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_8, [sym_size_int, 1500, 5120]);  addmm_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_1086: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_172, 0.5)
	        mul_1087: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_172, 0.7071067811865476);  view_172 = None
	        erf_3: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_1087);  mul_1087 = None
	        add_1716: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_3, 1);  erf_3 = None
	        mul_1088: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1086, add_1716);  mul_1086 = add_1716 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_15: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_1088);  mul_1088 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_173: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_15, [sym_size_int, 1500, 5120])
	        amin_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_173, [2])
	        amax_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_173, [2]);  view_173 = None
	        full_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_11, full_22);  amin_11 = full_22 = None
	        full_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_11, full_23);  amax_11 = full_23 = None
	        sub_519: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_11, minimum_11);  maximum_11 = None
	        div_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_519, 255.0);  sub_519 = None
	        clamp_min_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_22, 1.1920928955078125e-07);  div_22 = None
	        div_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_11, clamp_min_33);  minimum_11 = None
	        round_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_23);  div_23 = None
	        sub_525: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_23);  round_23 = None
	        clamp_min_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_525, -128);  sub_525 = None
	        clamp_max_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_34, 127);  clamp_min_34 = None
	        _assert_tensor_metadata_101 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_33, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_101 = None
	        _assert_tensor_metadata_102 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_22, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_102 = None
	        convert_element_type_66: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_22, torch.int8);  clamp_max_22 = None
	        view_174: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_15, [sym_size_int, 1500, 5120]);  clone_15 = None
	        view_175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_33, [sym_size_int, 1500, 1])
	        view_176: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_66, [sym_size_int, 1500, 1])
	        reciprocal_11: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_175);  view_175 = None
	        mul_1134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_11, 1.0);  reciprocal_11 = None
	        mul_1137: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_174, mul_1134);  view_174 = mul_1134 = None
	        round_24: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_1137);  mul_1137 = None
	        add_1799: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_24, view_176);  round_24 = view_176 = None
	        clamp_min_35: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1799, -128);  add_1799 = None
	        clamp_max_23: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_35, 127);  clamp_min_35 = None
	        view_177: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_23, [sym_size_int, 1500, 5120]);  clamp_max_23 = None
	        _assert_tensor_metadata_103 = torch.ops.aten._assert_tensor_metadata.default(view_177, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_103 = None
	        convert_element_type_67: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_177, torch.int8);  view_177 = None
	        view_178: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_67, [sym_size_int, 1500, 5120]);  convert_element_type_67 = None
	        view_179: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_33, [sym_size_int, 1500, 1]);  clamp_min_33 = None
	        view_180: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_66, [sym_size_int, 1500, 1]);  convert_element_type_66 = None
	        _assert_tensor_metadata_104 = torch.ops.aten._assert_tensor_metadata.default(view_178, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_104 = None
	        convert_element_type_68: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_178, torch.float32);  view_178 = None
	        _assert_tensor_metadata_105 = torch.ops.aten._assert_tensor_metadata.default(view_180, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_105 = None
	        convert_element_type_69: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_180, torch.float32);  view_180 = None
	        sub_545: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_68, convert_element_type_69);  convert_element_type_68 = convert_element_type_69 = None
	        mul_1159: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_545, view_179);  sub_545 = view_179 = None
	        view_181: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_1159, [sym_size_int, 1500, 5120]);  mul_1159 = None
	        _assert_tensor_metadata_106 = torch.ops.aten._assert_tensor_metadata.default(view_181, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_106 = None
	        view_182: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg56_1, [1280, 160, 32]);  arg56_1 = None
	        view_183: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg57_1, [1280, 160, 1]);  arg57_1 = None
	        view_184: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg58_1, [1280, 160, 1]);  arg58_1 = None
	        _assert_tensor_metadata_107 = torch.ops.aten._assert_tensor_metadata.default(view_182, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_107 = None
	        convert_element_type_70: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_182, torch.float32);  view_182 = None
	        _assert_tensor_metadata_108 = torch.ops.aten._assert_tensor_metadata.default(view_184, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_108 = None
	        convert_element_type_71: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_184, torch.float32);  view_184 = None
	        sub_549: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_70, convert_element_type_71);  convert_element_type_70 = convert_element_type_71 = None
	        mul_1164: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_549, view_183);  sub_549 = view_183 = None
	        view_185: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_1164, [1280, 5120]);  mul_1164 = None
	        _assert_tensor_metadata_109 = torch.ops.aten._assert_tensor_metadata.default(view_185, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_109 = None
	        mul_1169: "Sym(1500*s6)" = sym_size_int * 1500
	        view_186: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_181, [mul_1169, 5120]);  view_181 = mul_1169 = None
	        permute_20: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_185, [1, 0]);  view_185 = None
	        addmm_9: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg55_1, view_186, permute_20);  arg55_1 = view_186 = permute_20 = None
	        view_187: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_9, [sym_size_int, 1500, 1280]);  addmm_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_16: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_187);  view_187 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_1862: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_1564, clone_16);  add_1564 = clone_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_17: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_1862, memory_format = torch.contiguous_format)
	        var_mean_4 = torch.ops.aten.var_mean.correction(clone_17, [2], correction = 0, keepdim = True)
	        getitem_16: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_4[0]
	        getitem_17: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_4[1];  var_mean_4 = None
	        add_1867: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_16, 1e-05);  getitem_16 = None
	        rsqrt_4: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_1867);  add_1867 = None
	        sub_555: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_17, getitem_17);  clone_17 = getitem_17 = None
	        mul_1180: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_555, rsqrt_4);  sub_555 = rsqrt_4 = None
	        mul_1181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1180, arg59_1);  mul_1180 = arg59_1 = None
	        add_1868: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_1181, arg60_1);  mul_1181 = arg60_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_1868, [sym_size_int, 1500, 1280])
	        amin_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_188, [2])
	        amax_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_188, [2]);  view_188 = None
	        full_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_12, full_24);  amin_12 = full_24 = None
	        full_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_12, full_25);  amax_12 = full_25 = None
	        sub_566: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_12, minimum_12);  maximum_12 = None
	        div_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_566, 255.0);  sub_566 = None
	        clamp_min_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_24, 1.1920928955078125e-07);  div_24 = None
	        div_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_12, clamp_min_36);  minimum_12 = None
	        round_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_25);  div_25 = None
	        sub_572: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_25);  round_25 = None
	        clamp_min_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_572, -128);  sub_572 = None
	        clamp_max_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_37, 127);  clamp_min_37 = None
	        _assert_tensor_metadata_110 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_36, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_110 = None
	        _assert_tensor_metadata_111 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_24, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_111 = None
	        convert_element_type_72: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_24, torch.int8);  clamp_max_24 = None
	        view_189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_1868, [sym_size_int, 1500, 1280])
	        view_190: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_36, [sym_size_int, 1500, 1])
	        view_191: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_72, [sym_size_int, 1500, 1])
	        reciprocal_12: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_190);  view_190 = None
	        mul_1229: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_12, 1.0);  reciprocal_12 = None
	        mul_1232: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_189, mul_1229);  view_189 = mul_1229 = None
	        round_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1232);  mul_1232 = None
	        add_1955: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_26, view_191);  round_26 = view_191 = None
	        clamp_min_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1955, -128);  add_1955 = None
	        clamp_max_25: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_38, 127);  clamp_min_38 = None
	        view_192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_25, [sym_size_int, 1500, 1280]);  clamp_max_25 = None
	        _assert_tensor_metadata_112 = torch.ops.aten._assert_tensor_metadata.default(view_192, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_112 = None
	        convert_element_type_73: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_192, torch.int8);  view_192 = None
	        view_193: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_73, [sym_size_int, 1500, 1280]);  convert_element_type_73 = None
	        view_194: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_36, [sym_size_int, 1500, 1]);  clamp_min_36 = None
	        view_195: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_72, [sym_size_int, 1500, 1]);  convert_element_type_72 = None
	        _assert_tensor_metadata_113 = torch.ops.aten._assert_tensor_metadata.default(view_193, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_113 = None
	        convert_element_type_74: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_193, torch.float32);  view_193 = None
	        _assert_tensor_metadata_114 = torch.ops.aten._assert_tensor_metadata.default(view_195, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_114 = None
	        convert_element_type_75: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_195, torch.float32);  view_195 = None
	        sub_592: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_74, convert_element_type_75);  convert_element_type_74 = convert_element_type_75 = None
	        mul_1254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_592, view_194);  sub_592 = view_194 = None
	        view_196: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1254, [sym_size_int, 1500, 1280]);  mul_1254 = None
	        _assert_tensor_metadata_115 = torch.ops.aten._assert_tensor_metadata.default(view_196, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_115 = None
	        view_197: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg62_1, [1280, 40, 32]);  arg62_1 = None
	        view_198: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg63_1, [1280, 40, 1]);  arg63_1 = None
	        view_199: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg64_1, [1280, 40, 1]);  arg64_1 = None
	        _assert_tensor_metadata_116 = torch.ops.aten._assert_tensor_metadata.default(view_197, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_116 = None
	        convert_element_type_76: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_197, torch.float32);  view_197 = None
	        _assert_tensor_metadata_117 = torch.ops.aten._assert_tensor_metadata.default(view_199, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_117 = None
	        convert_element_type_77: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_199, torch.float32);  view_199 = None
	        sub_596: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_76, convert_element_type_77);  convert_element_type_76 = convert_element_type_77 = None
	        mul_1259: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_596, view_198);  sub_596 = view_198 = None
	        view_200: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1259, [1280, 1280]);  mul_1259 = None
	        _assert_tensor_metadata_118 = torch.ops.aten._assert_tensor_metadata.default(view_200, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_118 = None
	        mul_1264: "Sym(1500*s6)" = sym_size_int * 1500
	        view_201: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_196, [mul_1264, 1280]);  view_196 = mul_1264 = None
	        permute_21: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_200, [1, 0]);  view_200 = None
	        addmm_10: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg61_1, view_201, permute_21);  arg61_1 = view_201 = permute_21 = None
	        view_202: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_10, [sym_size_int, 1500, 1280]);  addmm_10 = None
	        mul_1271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_202, 0.125);  view_202 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_203: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_1271, [sym_size_int, 1500, 20, 64]);  mul_1271 = None
	        permute_22: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_203, [0, 2, 1, 3]);  view_203 = None
	        clone_18: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_22, memory_format = torch.contiguous_format);  permute_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_204: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_1868, [sym_size_int, 1500, 1280])
	        amin_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_204, [2])
	        amax_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_204, [2]);  view_204 = None
	        full_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_13, full_26);  amin_13 = full_26 = None
	        full_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_13, full_27);  amax_13 = full_27 = None
	        sub_611: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_13, minimum_13);  maximum_13 = None
	        div_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_611, 255.0);  sub_611 = None
	        clamp_min_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_26, 1.1920928955078125e-07);  div_26 = None
	        div_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_13, clamp_min_39);  minimum_13 = None
	        round_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_27);  div_27 = None
	        sub_617: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_27);  round_27 = None
	        clamp_min_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_617, -128);  sub_617 = None
	        clamp_max_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_40, 127);  clamp_min_40 = None
	        _assert_tensor_metadata_119 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_39, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_119 = None
	        _assert_tensor_metadata_120 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_26, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_120 = None
	        convert_element_type_78: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_26, torch.int8);  clamp_max_26 = None
	        view_205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_1868, [sym_size_int, 1500, 1280])
	        view_206: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_39, [sym_size_int, 1500, 1])
	        view_207: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_78, [sym_size_int, 1500, 1])
	        reciprocal_13: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_206);  view_206 = None
	        mul_1325: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_13, 1.0);  reciprocal_13 = None
	        mul_1328: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_205, mul_1325);  view_205 = mul_1325 = None
	        round_28: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1328);  mul_1328 = None
	        add_2107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_28, view_207);  round_28 = view_207 = None
	        clamp_min_41: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2107, -128);  add_2107 = None
	        clamp_max_27: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_41, 127);  clamp_min_41 = None
	        view_208: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_27, [sym_size_int, 1500, 1280]);  clamp_max_27 = None
	        _assert_tensor_metadata_121 = torch.ops.aten._assert_tensor_metadata.default(view_208, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_121 = None
	        convert_element_type_79: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_208, torch.int8);  view_208 = None
	        view_209: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_79, [sym_size_int, 1500, 1280]);  convert_element_type_79 = None
	        view_210: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_39, [sym_size_int, 1500, 1]);  clamp_min_39 = None
	        view_211: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_78, [sym_size_int, 1500, 1]);  convert_element_type_78 = None
	        _assert_tensor_metadata_122 = torch.ops.aten._assert_tensor_metadata.default(view_209, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_122 = None
	        convert_element_type_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_209, torch.float32);  view_209 = None
	        _assert_tensor_metadata_123 = torch.ops.aten._assert_tensor_metadata.default(view_211, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_123 = None
	        convert_element_type_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_211, torch.float32);  view_211 = None
	        sub_637: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_80, convert_element_type_81);  convert_element_type_80 = convert_element_type_81 = None
	        mul_1350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_637, view_210);  sub_637 = view_210 = None
	        view_212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1350, [sym_size_int, 1500, 1280]);  mul_1350 = None
	        _assert_tensor_metadata_124 = torch.ops.aten._assert_tensor_metadata.default(view_212, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_124 = None
	        view_213: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg65_1, [1280, 40, 32]);  arg65_1 = None
	        view_214: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg66_1, [1280, 40, 1]);  arg66_1 = None
	        view_215: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg67_1, [1280, 40, 1]);  arg67_1 = None
	        _assert_tensor_metadata_125 = torch.ops.aten._assert_tensor_metadata.default(view_213, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_125 = None
	        convert_element_type_82: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_213, torch.float32);  view_213 = None
	        _assert_tensor_metadata_126 = torch.ops.aten._assert_tensor_metadata.default(view_215, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_126 = None
	        convert_element_type_83: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_215, torch.float32);  view_215 = None
	        sub_641: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_82, convert_element_type_83);  convert_element_type_82 = convert_element_type_83 = None
	        mul_1355: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_641, view_214);  sub_641 = view_214 = None
	        view_216: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1355, [1280, 1280]);  mul_1355 = None
	        _assert_tensor_metadata_127 = torch.ops.aten._assert_tensor_metadata.default(view_216, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_127 = None
	        permute_23: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_216, [1, 0]);  view_216 = None
	        mul_1358: "Sym(1500*s6)" = sym_size_int * 1500
	        view_217: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_212, [mul_1358, 1280]);  view_212 = mul_1358 = None
	        mm_2: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_217, permute_23);  view_217 = permute_23 = None
	        view_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_2, [sym_size_int, 1500, 1280]);  mm_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_219: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_218, [sym_size_int, -1, 20, 64]);  view_218 = None
	        permute_24: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_219, [0, 2, 1, 3]);  view_219 = None
	        clone_19: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_24, memory_format = torch.contiguous_format);  permute_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_220: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_1868, [sym_size_int, 1500, 1280])
	        amin_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_220, [2])
	        amax_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_220, [2]);  view_220 = None
	        full_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_14, full_28);  amin_14 = full_28 = None
	        full_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_14, full_29);  amax_14 = full_29 = None
	        sub_655: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_14, minimum_14);  maximum_14 = None
	        div_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_655, 255.0);  sub_655 = None
	        clamp_min_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_28, 1.1920928955078125e-07);  div_28 = None
	        div_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_14, clamp_min_42);  minimum_14 = None
	        round_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_29);  div_29 = None
	        sub_661: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_29);  round_29 = None
	        clamp_min_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_661, -128);  sub_661 = None
	        clamp_max_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_43, 127);  clamp_min_43 = None
	        _assert_tensor_metadata_128 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_42, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_128 = None
	        _assert_tensor_metadata_129 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_28, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_129 = None
	        convert_element_type_84: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_28, torch.int8);  clamp_max_28 = None
	        view_221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_1868, [sym_size_int, 1500, 1280]);  add_1868 = None
	        view_222: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_42, [sym_size_int, 1500, 1])
	        view_223: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_84, [sym_size_int, 1500, 1])
	        reciprocal_14: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_222);  view_222 = None
	        mul_1424: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_14, 1.0);  reciprocal_14 = None
	        mul_1427: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_221, mul_1424);  view_221 = mul_1424 = None
	        round_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1427);  mul_1427 = None
	        add_2255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_30, view_223);  round_30 = view_223 = None
	        clamp_min_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2255, -128);  add_2255 = None
	        clamp_max_29: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_44, 127);  clamp_min_44 = None
	        view_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_29, [sym_size_int, 1500, 1280]);  clamp_max_29 = None
	        _assert_tensor_metadata_130 = torch.ops.aten._assert_tensor_metadata.default(view_224, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_130 = None
	        convert_element_type_85: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_224, torch.int8);  view_224 = None
	        view_225: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_85, [sym_size_int, 1500, 1280]);  convert_element_type_85 = None
	        view_226: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_42, [sym_size_int, 1500, 1]);  clamp_min_42 = None
	        view_227: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_84, [sym_size_int, 1500, 1]);  convert_element_type_84 = None
	        _assert_tensor_metadata_131 = torch.ops.aten._assert_tensor_metadata.default(view_225, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_131 = None
	        convert_element_type_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_225, torch.float32);  view_225 = None
	        _assert_tensor_metadata_132 = torch.ops.aten._assert_tensor_metadata.default(view_227, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_132 = None
	        convert_element_type_87: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_227, torch.float32);  view_227 = None
	        sub_681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_86, convert_element_type_87);  convert_element_type_86 = convert_element_type_87 = None
	        mul_1449: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_681, view_226);  sub_681 = view_226 = None
	        view_228: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1449, [sym_size_int, 1500, 1280]);  mul_1449 = None
	        _assert_tensor_metadata_133 = torch.ops.aten._assert_tensor_metadata.default(view_228, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_133 = None
	        view_229: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg69_1, [1280, 40, 32]);  arg69_1 = None
	        view_230: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg70_1, [1280, 40, 1]);  arg70_1 = None
	        view_231: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg71_1, [1280, 40, 1]);  arg71_1 = None
	        _assert_tensor_metadata_134 = torch.ops.aten._assert_tensor_metadata.default(view_229, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_134 = None
	        convert_element_type_88: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_229, torch.float32);  view_229 = None
	        _assert_tensor_metadata_135 = torch.ops.aten._assert_tensor_metadata.default(view_231, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_135 = None
	        convert_element_type_89: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_231, torch.float32);  view_231 = None
	        sub_685: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_88, convert_element_type_89);  convert_element_type_88 = convert_element_type_89 = None
	        mul_1454: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_685, view_230);  sub_685 = view_230 = None
	        view_232: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1454, [1280, 1280]);  mul_1454 = None
	        _assert_tensor_metadata_136 = torch.ops.aten._assert_tensor_metadata.default(view_232, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_136 = None
	        mul_1459: "Sym(1500*s6)" = sym_size_int * 1500
	        view_233: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_228, [mul_1459, 1280]);  view_228 = mul_1459 = None
	        permute_25: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_232, [1, 0]);  view_232 = None
	        addmm_11: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg68_1, view_233, permute_25);  arg68_1 = view_233 = permute_25 = None
	        view_234: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_11, [sym_size_int, 1500, 1280]);  addmm_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_235: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_234, [sym_size_int, -1, 20, 64]);  view_234 = None
	        permute_26: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_235, [0, 2, 1, 3]);  view_235 = None
	        clone_20: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_26, memory_format = torch.contiguous_format);  permute_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_2 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_18, clone_19, clone_20, None, False, scale = 1.0);  clone_18 = clone_19 = clone_20 = None
	        getitem_18: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_2[0];  _scaled_dot_product_efficient_attention_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_27: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_18, [0, 2, 1, 3]);  getitem_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_27, [sym_size_int, 1500, -1]);  permute_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_237: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_236, [sym_size_int, 1500, 1280])
	        amin_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_237, [2])
	        amax_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_237, [2]);  view_237 = None
	        full_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_15, full_30);  amin_15 = full_30 = None
	        full_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_15, full_31);  amax_15 = full_31 = None
	        sub_703: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_15, minimum_15);  maximum_15 = None
	        div_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_703, 255.0);  sub_703 = None
	        clamp_min_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_30, 1.1920928955078125e-07);  div_30 = None
	        div_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_15, clamp_min_45);  minimum_15 = None
	        round_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_31);  div_31 = None
	        sub_709: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_31);  round_31 = None
	        clamp_min_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_709, -128);  sub_709 = None
	        clamp_max_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_46, 127);  clamp_min_46 = None
	        _assert_tensor_metadata_137 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_45, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_137 = None
	        _assert_tensor_metadata_138 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_30, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_138 = None
	        convert_element_type_90: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_30, torch.int8);  clamp_max_30 = None
	        view_238: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_236, [sym_size_int, 1500, 1280]);  view_236 = None
	        view_239: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_45, [sym_size_int, 1500, 1])
	        view_240: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_90, [sym_size_int, 1500, 1])
	        reciprocal_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_239);  view_239 = None
	        mul_1529: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_15, 1.0);  reciprocal_15 = None
	        mul_1532: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_238, mul_1529);  view_238 = mul_1529 = None
	        round_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1532);  mul_1532 = None
	        add_2419: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_32, view_240);  round_32 = view_240 = None
	        clamp_min_47: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2419, -128);  add_2419 = None
	        clamp_max_31: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_47, 127);  clamp_min_47 = None
	        view_241: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_31, [sym_size_int, 1500, 1280]);  clamp_max_31 = None
	        _assert_tensor_metadata_139 = torch.ops.aten._assert_tensor_metadata.default(view_241, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_139 = None
	        convert_element_type_91: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_241, torch.int8);  view_241 = None
	        view_242: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_91, [sym_size_int, 1500, 1280]);  convert_element_type_91 = None
	        view_243: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_45, [sym_size_int, 1500, 1]);  clamp_min_45 = None
	        view_244: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_90, [sym_size_int, 1500, 1]);  convert_element_type_90 = None
	        _assert_tensor_metadata_140 = torch.ops.aten._assert_tensor_metadata.default(view_242, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_140 = None
	        convert_element_type_92: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_242, torch.float32);  view_242 = None
	        _assert_tensor_metadata_141 = torch.ops.aten._assert_tensor_metadata.default(view_244, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_141 = None
	        convert_element_type_93: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_244, torch.float32);  view_244 = None
	        sub_729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_92, convert_element_type_93);  convert_element_type_92 = convert_element_type_93 = None
	        mul_1554: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_729, view_243);  sub_729 = view_243 = None
	        view_245: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1554, [sym_size_int, 1500, 1280]);  mul_1554 = None
	        _assert_tensor_metadata_142 = torch.ops.aten._assert_tensor_metadata.default(view_245, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_142 = None
	        view_246: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg73_1, [1280, 40, 32]);  arg73_1 = None
	        view_247: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg74_1, [1280, 40, 1]);  arg74_1 = None
	        view_248: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg75_1, [1280, 40, 1]);  arg75_1 = None
	        _assert_tensor_metadata_143 = torch.ops.aten._assert_tensor_metadata.default(view_246, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_143 = None
	        convert_element_type_94: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_246, torch.float32);  view_246 = None
	        _assert_tensor_metadata_144 = torch.ops.aten._assert_tensor_metadata.default(view_248, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_144 = None
	        convert_element_type_95: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_248, torch.float32);  view_248 = None
	        sub_733: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_94, convert_element_type_95);  convert_element_type_94 = convert_element_type_95 = None
	        mul_1559: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_733, view_247);  sub_733 = view_247 = None
	        view_249: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1559, [1280, 1280]);  mul_1559 = None
	        _assert_tensor_metadata_145 = torch.ops.aten._assert_tensor_metadata.default(view_249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_145 = None
	        mul_1564: "Sym(1500*s6)" = sym_size_int * 1500
	        view_250: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_245, [mul_1564, 1280]);  view_245 = mul_1564 = None
	        permute_28: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_249, [1, 0]);  view_249 = None
	        addmm_12: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg72_1, view_250, permute_28);  arg72_1 = view_250 = permute_28 = None
	        view_251: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_12, [sym_size_int, 1500, 1280]);  addmm_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_251);  view_251 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_2482: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_1862, clone_21);  add_1862 = clone_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_22: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_2482, memory_format = torch.contiguous_format)
	        var_mean_5 = torch.ops.aten.var_mean.correction(clone_22, [2], correction = 0, keepdim = True)
	        getitem_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_5[0]
	        getitem_23: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_5[1];  var_mean_5 = None
	        add_2487: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_22, 1e-05);  getitem_22 = None
	        rsqrt_5: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_2487);  add_2487 = None
	        sub_739: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_22, getitem_23);  clone_22 = getitem_23 = None
	        mul_1575: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_739, rsqrt_5);  sub_739 = rsqrt_5 = None
	        mul_1576: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1575, arg76_1);  mul_1575 = arg76_1 = None
	        add_2488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_1576, arg77_1);  mul_1576 = arg77_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_252: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_2488, [sym_size_int, 1500, 1280])
	        amin_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_252, [2])
	        amax_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_252, [2]);  view_252 = None
	        full_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_16, full_32);  amin_16 = full_32 = None
	        full_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_16, full_33);  amax_16 = full_33 = None
	        sub_750: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_16, minimum_16);  maximum_16 = None
	        div_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_750, 255.0);  sub_750 = None
	        clamp_min_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_32, 1.1920928955078125e-07);  div_32 = None
	        div_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_16, clamp_min_48);  minimum_16 = None
	        round_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_33);  div_33 = None
	        sub_756: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_33);  round_33 = None
	        clamp_min_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_756, -128);  sub_756 = None
	        clamp_max_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_49, 127);  clamp_min_49 = None
	        _assert_tensor_metadata_146 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_48, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_146 = None
	        _assert_tensor_metadata_147 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_32, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_147 = None
	        convert_element_type_96: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_32, torch.int8);  clamp_max_32 = None
	        view_253: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_2488, [sym_size_int, 1500, 1280]);  add_2488 = None
	        view_254: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_48, [sym_size_int, 1500, 1])
	        view_255: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_96, [sym_size_int, 1500, 1])
	        reciprocal_16: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_254);  view_254 = None
	        mul_1624: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_16, 1.0);  reciprocal_16 = None
	        mul_1627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_253, mul_1624);  view_253 = mul_1624 = None
	        round_34: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1627);  mul_1627 = None
	        add_2575: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_34, view_255);  round_34 = view_255 = None
	        clamp_min_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2575, -128);  add_2575 = None
	        clamp_max_33: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_50, 127);  clamp_min_50 = None
	        view_256: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_33, [sym_size_int, 1500, 1280]);  clamp_max_33 = None
	        _assert_tensor_metadata_148 = torch.ops.aten._assert_tensor_metadata.default(view_256, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_148 = None
	        convert_element_type_97: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_256, torch.int8);  view_256 = None
	        view_257: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_97, [sym_size_int, 1500, 1280]);  convert_element_type_97 = None
	        view_258: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_48, [sym_size_int, 1500, 1]);  clamp_min_48 = None
	        view_259: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_96, [sym_size_int, 1500, 1]);  convert_element_type_96 = None
	        _assert_tensor_metadata_149 = torch.ops.aten._assert_tensor_metadata.default(view_257, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_149 = None
	        convert_element_type_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_257, torch.float32);  view_257 = None
	        _assert_tensor_metadata_150 = torch.ops.aten._assert_tensor_metadata.default(view_259, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_150 = None
	        convert_element_type_99: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_259, torch.float32);  view_259 = None
	        sub_776: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_98, convert_element_type_99);  convert_element_type_98 = convert_element_type_99 = None
	        mul_1649: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_776, view_258);  sub_776 = view_258 = None
	        view_260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1649, [sym_size_int, 1500, 1280]);  mul_1649 = None
	        _assert_tensor_metadata_151 = torch.ops.aten._assert_tensor_metadata.default(view_260, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_151 = None
	        view_261: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg79_1, [5120, 40, 32]);  arg79_1 = None
	        view_262: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg80_1, [5120, 40, 1]);  arg80_1 = None
	        view_263: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg81_1, [5120, 40, 1]);  arg81_1 = None
	        _assert_tensor_metadata_152 = torch.ops.aten._assert_tensor_metadata.default(view_261, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_152 = None
	        convert_element_type_100: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_261, torch.float32);  view_261 = None
	        _assert_tensor_metadata_153 = torch.ops.aten._assert_tensor_metadata.default(view_263, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_153 = None
	        convert_element_type_101: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_263, torch.float32);  view_263 = None
	        sub_780: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_100, convert_element_type_101);  convert_element_type_100 = convert_element_type_101 = None
	        mul_1654: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_780, view_262);  sub_780 = view_262 = None
	        view_264: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1654, [5120, 1280]);  mul_1654 = None
	        _assert_tensor_metadata_154 = torch.ops.aten._assert_tensor_metadata.default(view_264, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_154 = None
	        mul_1659: "Sym(1500*s6)" = sym_size_int * 1500
	        view_265: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_260, [mul_1659, 1280]);  view_260 = mul_1659 = None
	        permute_29: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_264, [1, 0]);  view_264 = None
	        addmm_13: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg78_1, view_265, permute_29);  arg78_1 = view_265 = permute_29 = None
	        view_266: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_13, [sym_size_int, 1500, 5120]);  addmm_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_1666: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_266, 0.5)
	        mul_1667: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_266, 0.7071067811865476);  view_266 = None
	        erf_4: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_1667);  mul_1667 = None
	        add_2634: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_4, 1);  erf_4 = None
	        mul_1668: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1666, add_2634);  mul_1666 = add_2634 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_23: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_1668);  mul_1668 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_267: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_23, [sym_size_int, 1500, 5120])
	        amin_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_267, [2])
	        amax_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_267, [2]);  view_267 = None
	        full_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_17, full_34);  amin_17 = full_34 = None
	        full_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_17, full_35);  amax_17 = full_35 = None
	        sub_793: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_17, minimum_17);  maximum_17 = None
	        div_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_793, 255.0);  sub_793 = None
	        clamp_min_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_34, 1.1920928955078125e-07);  div_34 = None
	        div_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_17, clamp_min_51);  minimum_17 = None
	        round_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_35);  div_35 = None
	        sub_799: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_35);  round_35 = None
	        clamp_min_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_799, -128);  sub_799 = None
	        clamp_max_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_52, 127);  clamp_min_52 = None
	        _assert_tensor_metadata_155 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_51, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_155 = None
	        _assert_tensor_metadata_156 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_34, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_156 = None
	        convert_element_type_102: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_34, torch.int8);  clamp_max_34 = None
	        view_268: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_23, [sym_size_int, 1500, 5120]);  clone_23 = None
	        view_269: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_51, [sym_size_int, 1500, 1])
	        view_270: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_102, [sym_size_int, 1500, 1])
	        reciprocal_17: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_269);  view_269 = None
	        mul_1714: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_17, 1.0);  reciprocal_17 = None
	        mul_1717: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_268, mul_1714);  view_268 = mul_1714 = None
	        round_36: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_1717);  mul_1717 = None
	        add_2717: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_36, view_270);  round_36 = view_270 = None
	        clamp_min_53: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2717, -128);  add_2717 = None
	        clamp_max_35: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_53, 127);  clamp_min_53 = None
	        view_271: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_35, [sym_size_int, 1500, 5120]);  clamp_max_35 = None
	        _assert_tensor_metadata_157 = torch.ops.aten._assert_tensor_metadata.default(view_271, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_157 = None
	        convert_element_type_103: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_271, torch.int8);  view_271 = None
	        view_272: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_103, [sym_size_int, 1500, 5120]);  convert_element_type_103 = None
	        view_273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_51, [sym_size_int, 1500, 1]);  clamp_min_51 = None
	        view_274: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_102, [sym_size_int, 1500, 1]);  convert_element_type_102 = None
	        _assert_tensor_metadata_158 = torch.ops.aten._assert_tensor_metadata.default(view_272, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_158 = None
	        convert_element_type_104: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_272, torch.float32);  view_272 = None
	        _assert_tensor_metadata_159 = torch.ops.aten._assert_tensor_metadata.default(view_274, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_159 = None
	        convert_element_type_105: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_274, torch.float32);  view_274 = None
	        sub_819: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_104, convert_element_type_105);  convert_element_type_104 = convert_element_type_105 = None
	        mul_1739: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_819, view_273);  sub_819 = view_273 = None
	        view_275: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_1739, [sym_size_int, 1500, 5120]);  mul_1739 = None
	        _assert_tensor_metadata_160 = torch.ops.aten._assert_tensor_metadata.default(view_275, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_160 = None
	        view_276: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg83_1, [1280, 160, 32]);  arg83_1 = None
	        view_277: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg84_1, [1280, 160, 1]);  arg84_1 = None
	        view_278: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg85_1, [1280, 160, 1]);  arg85_1 = None
	        _assert_tensor_metadata_161 = torch.ops.aten._assert_tensor_metadata.default(view_276, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_161 = None
	        convert_element_type_106: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_276, torch.float32);  view_276 = None
	        _assert_tensor_metadata_162 = torch.ops.aten._assert_tensor_metadata.default(view_278, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_162 = None
	        convert_element_type_107: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_278, torch.float32);  view_278 = None
	        sub_823: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_106, convert_element_type_107);  convert_element_type_106 = convert_element_type_107 = None
	        mul_1744: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_823, view_277);  sub_823 = view_277 = None
	        view_279: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_1744, [1280, 5120]);  mul_1744 = None
	        _assert_tensor_metadata_163 = torch.ops.aten._assert_tensor_metadata.default(view_279, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_163 = None
	        mul_1749: "Sym(1500*s6)" = sym_size_int * 1500
	        view_280: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_275, [mul_1749, 5120]);  view_275 = mul_1749 = None
	        permute_30: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_279, [1, 0]);  view_279 = None
	        addmm_14: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg82_1, view_280, permute_30);  arg82_1 = view_280 = permute_30 = None
	        view_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_14, [sym_size_int, 1500, 1280]);  addmm_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_24: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_281);  view_281 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_2780: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_2482, clone_24);  add_2482 = clone_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_25: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_2780, memory_format = torch.contiguous_format)
	        var_mean_6 = torch.ops.aten.var_mean.correction(clone_25, [2], correction = 0, keepdim = True)
	        getitem_24: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_6[0]
	        getitem_25: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_6[1];  var_mean_6 = None
	        add_2785: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_24, 1e-05);  getitem_24 = None
	        rsqrt_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_2785);  add_2785 = None
	        sub_829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_25, getitem_25);  clone_25 = getitem_25 = None
	        mul_1760: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_829, rsqrt_6);  sub_829 = rsqrt_6 = None
	        mul_1761: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1760, arg86_1);  mul_1760 = arg86_1 = None
	        add_2786: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_1761, arg87_1);  mul_1761 = arg87_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_282: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_2786, [sym_size_int, 1500, 1280])
	        amin_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_282, [2])
	        amax_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_282, [2]);  view_282 = None
	        full_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_18, full_36);  amin_18 = full_36 = None
	        full_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_18, full_37);  amax_18 = full_37 = None
	        sub_840: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_18, minimum_18);  maximum_18 = None
	        div_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_840, 255.0);  sub_840 = None
	        clamp_min_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_36, 1.1920928955078125e-07);  div_36 = None
	        div_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_18, clamp_min_54);  minimum_18 = None
	        round_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_37);  div_37 = None
	        sub_846: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_37);  round_37 = None
	        clamp_min_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_846, -128);  sub_846 = None
	        clamp_max_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_55, 127);  clamp_min_55 = None
	        _assert_tensor_metadata_164 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_54, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_164 = None
	        _assert_tensor_metadata_165 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_36, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_165 = None
	        convert_element_type_108: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_36, torch.int8);  clamp_max_36 = None
	        view_283: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_2786, [sym_size_int, 1500, 1280])
	        view_284: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_54, [sym_size_int, 1500, 1])
	        view_285: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_108, [sym_size_int, 1500, 1])
	        reciprocal_18: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_284);  view_284 = None
	        mul_1809: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_18, 1.0);  reciprocal_18 = None
	        mul_1812: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_283, mul_1809);  view_283 = mul_1809 = None
	        round_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1812);  mul_1812 = None
	        add_2873: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_38, view_285);  round_38 = view_285 = None
	        clamp_min_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2873, -128);  add_2873 = None
	        clamp_max_37: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_56, 127);  clamp_min_56 = None
	        view_286: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_37, [sym_size_int, 1500, 1280]);  clamp_max_37 = None
	        _assert_tensor_metadata_166 = torch.ops.aten._assert_tensor_metadata.default(view_286, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_166 = None
	        convert_element_type_109: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_286, torch.int8);  view_286 = None
	        view_287: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_109, [sym_size_int, 1500, 1280]);  convert_element_type_109 = None
	        view_288: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_54, [sym_size_int, 1500, 1]);  clamp_min_54 = None
	        view_289: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_108, [sym_size_int, 1500, 1]);  convert_element_type_108 = None
	        _assert_tensor_metadata_167 = torch.ops.aten._assert_tensor_metadata.default(view_287, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_167 = None
	        convert_element_type_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_287, torch.float32);  view_287 = None
	        _assert_tensor_metadata_168 = torch.ops.aten._assert_tensor_metadata.default(view_289, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_168 = None
	        convert_element_type_111: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_289, torch.float32);  view_289 = None
	        sub_866: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_110, convert_element_type_111);  convert_element_type_110 = convert_element_type_111 = None
	        mul_1834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_866, view_288);  sub_866 = view_288 = None
	        view_290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1834, [sym_size_int, 1500, 1280]);  mul_1834 = None
	        _assert_tensor_metadata_169 = torch.ops.aten._assert_tensor_metadata.default(view_290, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_169 = None
	        view_291: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg89_1, [1280, 40, 32]);  arg89_1 = None
	        view_292: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg90_1, [1280, 40, 1]);  arg90_1 = None
	        view_293: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg91_1, [1280, 40, 1]);  arg91_1 = None
	        _assert_tensor_metadata_170 = torch.ops.aten._assert_tensor_metadata.default(view_291, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_170 = None
	        convert_element_type_112: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_291, torch.float32);  view_291 = None
	        _assert_tensor_metadata_171 = torch.ops.aten._assert_tensor_metadata.default(view_293, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_171 = None
	        convert_element_type_113: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_293, torch.float32);  view_293 = None
	        sub_870: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_112, convert_element_type_113);  convert_element_type_112 = convert_element_type_113 = None
	        mul_1839: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_870, view_292);  sub_870 = view_292 = None
	        view_294: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1839, [1280, 1280]);  mul_1839 = None
	        _assert_tensor_metadata_172 = torch.ops.aten._assert_tensor_metadata.default(view_294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_172 = None
	        mul_1844: "Sym(1500*s6)" = sym_size_int * 1500
	        view_295: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_290, [mul_1844, 1280]);  view_290 = mul_1844 = None
	        permute_31: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_294, [1, 0]);  view_294 = None
	        addmm_15: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg88_1, view_295, permute_31);  arg88_1 = view_295 = permute_31 = None
	        view_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_15, [sym_size_int, 1500, 1280]);  addmm_15 = None
	        mul_1851: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_296, 0.125);  view_296 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_297: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_1851, [sym_size_int, 1500, 20, 64]);  mul_1851 = None
	        permute_32: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_297, [0, 2, 1, 3]);  view_297 = None
	        clone_26: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_32, memory_format = torch.contiguous_format);  permute_32 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_298: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_2786, [sym_size_int, 1500, 1280])
	        amin_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_298, [2])
	        amax_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_298, [2]);  view_298 = None
	        full_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_19, full_38);  amin_19 = full_38 = None
	        full_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_19, full_39);  amax_19 = full_39 = None
	        sub_885: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_19, minimum_19);  maximum_19 = None
	        div_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_885, 255.0);  sub_885 = None
	        clamp_min_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_38, 1.1920928955078125e-07);  div_38 = None
	        div_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_19, clamp_min_57);  minimum_19 = None
	        round_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_39);  div_39 = None
	        sub_891: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_39);  round_39 = None
	        clamp_min_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_891, -128);  sub_891 = None
	        clamp_max_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_58, 127);  clamp_min_58 = None
	        _assert_tensor_metadata_173 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_57, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_173 = None
	        _assert_tensor_metadata_174 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_38, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_174 = None
	        convert_element_type_114: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_38, torch.int8);  clamp_max_38 = None
	        view_299: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_2786, [sym_size_int, 1500, 1280])
	        view_300: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_57, [sym_size_int, 1500, 1])
	        view_301: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_114, [sym_size_int, 1500, 1])
	        reciprocal_19: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_300);  view_300 = None
	        mul_1905: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_19, 1.0);  reciprocal_19 = None
	        mul_1908: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_299, mul_1905);  view_299 = mul_1905 = None
	        round_40: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1908);  mul_1908 = None
	        add_3025: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_40, view_301);  round_40 = view_301 = None
	        clamp_min_59: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3025, -128);  add_3025 = None
	        clamp_max_39: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_59, 127);  clamp_min_59 = None
	        view_302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_39, [sym_size_int, 1500, 1280]);  clamp_max_39 = None
	        _assert_tensor_metadata_175 = torch.ops.aten._assert_tensor_metadata.default(view_302, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_175 = None
	        convert_element_type_115: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_302, torch.int8);  view_302 = None
	        view_303: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_115, [sym_size_int, 1500, 1280]);  convert_element_type_115 = None
	        view_304: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_57, [sym_size_int, 1500, 1]);  clamp_min_57 = None
	        view_305: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_114, [sym_size_int, 1500, 1]);  convert_element_type_114 = None
	        _assert_tensor_metadata_176 = torch.ops.aten._assert_tensor_metadata.default(view_303, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_176 = None
	        convert_element_type_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_303, torch.float32);  view_303 = None
	        _assert_tensor_metadata_177 = torch.ops.aten._assert_tensor_metadata.default(view_305, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_177 = None
	        convert_element_type_117: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_305, torch.float32);  view_305 = None
	        sub_911: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_116, convert_element_type_117);  convert_element_type_116 = convert_element_type_117 = None
	        mul_1930: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_911, view_304);  sub_911 = view_304 = None
	        view_306: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1930, [sym_size_int, 1500, 1280]);  mul_1930 = None
	        _assert_tensor_metadata_178 = torch.ops.aten._assert_tensor_metadata.default(view_306, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_178 = None
	        view_307: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg92_1, [1280, 40, 32]);  arg92_1 = None
	        view_308: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg93_1, [1280, 40, 1]);  arg93_1 = None
	        view_309: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg94_1, [1280, 40, 1]);  arg94_1 = None
	        _assert_tensor_metadata_179 = torch.ops.aten._assert_tensor_metadata.default(view_307, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_179 = None
	        convert_element_type_118: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_307, torch.float32);  view_307 = None
	        _assert_tensor_metadata_180 = torch.ops.aten._assert_tensor_metadata.default(view_309, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_180 = None
	        convert_element_type_119: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_309, torch.float32);  view_309 = None
	        sub_915: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_118, convert_element_type_119);  convert_element_type_118 = convert_element_type_119 = None
	        mul_1935: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_915, view_308);  sub_915 = view_308 = None
	        view_310: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1935, [1280, 1280]);  mul_1935 = None
	        _assert_tensor_metadata_181 = torch.ops.aten._assert_tensor_metadata.default(view_310, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_181 = None
	        permute_33: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_310, [1, 0]);  view_310 = None
	        mul_1938: "Sym(1500*s6)" = sym_size_int * 1500
	        view_311: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_306, [mul_1938, 1280]);  view_306 = mul_1938 = None
	        mm_3: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_311, permute_33);  view_311 = permute_33 = None
	        view_312: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_3, [sym_size_int, 1500, 1280]);  mm_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_313: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_312, [sym_size_int, -1, 20, 64]);  view_312 = None
	        permute_34: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_313, [0, 2, 1, 3]);  view_313 = None
	        clone_27: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_34, memory_format = torch.contiguous_format);  permute_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_2786, [sym_size_int, 1500, 1280])
	        amin_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_314, [2])
	        amax_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_314, [2]);  view_314 = None
	        full_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_20, full_40);  amin_20 = full_40 = None
	        full_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_20, full_41);  amax_20 = full_41 = None
	        sub_929: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_20, minimum_20);  maximum_20 = None
	        div_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_929, 255.0);  sub_929 = None
	        clamp_min_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_40, 1.1920928955078125e-07);  div_40 = None
	        div_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_20, clamp_min_60);  minimum_20 = None
	        round_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_41);  div_41 = None
	        sub_935: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_41);  round_41 = None
	        clamp_min_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_935, -128);  sub_935 = None
	        clamp_max_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_61, 127);  clamp_min_61 = None
	        _assert_tensor_metadata_182 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_60, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_182 = None
	        _assert_tensor_metadata_183 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_40, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_183 = None
	        convert_element_type_120: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_40, torch.int8);  clamp_max_40 = None
	        view_315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_2786, [sym_size_int, 1500, 1280]);  add_2786 = None
	        view_316: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_60, [sym_size_int, 1500, 1])
	        view_317: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_120, [sym_size_int, 1500, 1])
	        reciprocal_20: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_316);  view_316 = None
	        mul_2004: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_20, 1.0);  reciprocal_20 = None
	        mul_2007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_315, mul_2004);  view_315 = mul_2004 = None
	        round_42: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2007);  mul_2007 = None
	        add_3173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_42, view_317);  round_42 = view_317 = None
	        clamp_min_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3173, -128);  add_3173 = None
	        clamp_max_41: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_62, 127);  clamp_min_62 = None
	        view_318: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_41, [sym_size_int, 1500, 1280]);  clamp_max_41 = None
	        _assert_tensor_metadata_184 = torch.ops.aten._assert_tensor_metadata.default(view_318, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_184 = None
	        convert_element_type_121: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_318, torch.int8);  view_318 = None
	        view_319: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_121, [sym_size_int, 1500, 1280]);  convert_element_type_121 = None
	        view_320: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_60, [sym_size_int, 1500, 1]);  clamp_min_60 = None
	        view_321: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_120, [sym_size_int, 1500, 1]);  convert_element_type_120 = None
	        _assert_tensor_metadata_185 = torch.ops.aten._assert_tensor_metadata.default(view_319, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_185 = None
	        convert_element_type_122: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_319, torch.float32);  view_319 = None
	        _assert_tensor_metadata_186 = torch.ops.aten._assert_tensor_metadata.default(view_321, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_186 = None
	        convert_element_type_123: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_321, torch.float32);  view_321 = None
	        sub_955: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_122, convert_element_type_123);  convert_element_type_122 = convert_element_type_123 = None
	        mul_2029: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_955, view_320);  sub_955 = view_320 = None
	        view_322: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2029, [sym_size_int, 1500, 1280]);  mul_2029 = None
	        _assert_tensor_metadata_187 = torch.ops.aten._assert_tensor_metadata.default(view_322, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_187 = None
	        view_323: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg96_1, [1280, 40, 32]);  arg96_1 = None
	        view_324: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg97_1, [1280, 40, 1]);  arg97_1 = None
	        view_325: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg98_1, [1280, 40, 1]);  arg98_1 = None
	        _assert_tensor_metadata_188 = torch.ops.aten._assert_tensor_metadata.default(view_323, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_188 = None
	        convert_element_type_124: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_323, torch.float32);  view_323 = None
	        _assert_tensor_metadata_189 = torch.ops.aten._assert_tensor_metadata.default(view_325, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_189 = None
	        convert_element_type_125: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_325, torch.float32);  view_325 = None
	        sub_959: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_124, convert_element_type_125);  convert_element_type_124 = convert_element_type_125 = None
	        mul_2034: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_959, view_324);  sub_959 = view_324 = None
	        view_326: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2034, [1280, 1280]);  mul_2034 = None
	        _assert_tensor_metadata_190 = torch.ops.aten._assert_tensor_metadata.default(view_326, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_190 = None
	        mul_2039: "Sym(1500*s6)" = sym_size_int * 1500
	        view_327: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_322, [mul_2039, 1280]);  view_322 = mul_2039 = None
	        permute_35: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_326, [1, 0]);  view_326 = None
	        addmm_16: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg95_1, view_327, permute_35);  arg95_1 = view_327 = permute_35 = None
	        view_328: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_16, [sym_size_int, 1500, 1280]);  addmm_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_329: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_328, [sym_size_int, -1, 20, 64]);  view_328 = None
	        permute_36: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_329, [0, 2, 1, 3]);  view_329 = None
	        clone_28: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_36, memory_format = torch.contiguous_format);  permute_36 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_3 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_26, clone_27, clone_28, None, False, scale = 1.0);  clone_26 = clone_27 = clone_28 = None
	        getitem_26: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_3[0];  _scaled_dot_product_efficient_attention_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_37: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_26, [0, 2, 1, 3]);  getitem_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_330: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_37, [sym_size_int, 1500, -1]);  permute_37 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_331: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_330, [sym_size_int, 1500, 1280])
	        amin_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_331, [2])
	        amax_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_331, [2]);  view_331 = None
	        full_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_21, full_42);  amin_21 = full_42 = None
	        full_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_21, full_43);  amax_21 = full_43 = None
	        sub_977: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_21, minimum_21);  maximum_21 = None
	        div_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_977, 255.0);  sub_977 = None
	        clamp_min_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_42, 1.1920928955078125e-07);  div_42 = None
	        div_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_21, clamp_min_63);  minimum_21 = None
	        round_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_43);  div_43 = None
	        sub_983: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_43);  round_43 = None
	        clamp_min_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_983, -128);  sub_983 = None
	        clamp_max_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_64, 127);  clamp_min_64 = None
	        _assert_tensor_metadata_191 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_63, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_191 = None
	        _assert_tensor_metadata_192 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_42, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_192 = None
	        convert_element_type_126: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_42, torch.int8);  clamp_max_42 = None
	        view_332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_330, [sym_size_int, 1500, 1280]);  view_330 = None
	        view_333: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_63, [sym_size_int, 1500, 1])
	        view_334: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_126, [sym_size_int, 1500, 1])
	        reciprocal_21: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_333);  view_333 = None
	        mul_2109: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_21, 1.0);  reciprocal_21 = None
	        mul_2112: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_332, mul_2109);  view_332 = mul_2109 = None
	        round_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2112);  mul_2112 = None
	        add_3337: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_44, view_334);  round_44 = view_334 = None
	        clamp_min_65: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3337, -128);  add_3337 = None
	        clamp_max_43: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_65, 127);  clamp_min_65 = None
	        view_335: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_43, [sym_size_int, 1500, 1280]);  clamp_max_43 = None
	        _assert_tensor_metadata_193 = torch.ops.aten._assert_tensor_metadata.default(view_335, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_193 = None
	        convert_element_type_127: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_335, torch.int8);  view_335 = None
	        view_336: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_127, [sym_size_int, 1500, 1280]);  convert_element_type_127 = None
	        view_337: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_63, [sym_size_int, 1500, 1]);  clamp_min_63 = None
	        view_338: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_126, [sym_size_int, 1500, 1]);  convert_element_type_126 = None
	        _assert_tensor_metadata_194 = torch.ops.aten._assert_tensor_metadata.default(view_336, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_194 = None
	        convert_element_type_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_336, torch.float32);  view_336 = None
	        _assert_tensor_metadata_195 = torch.ops.aten._assert_tensor_metadata.default(view_338, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_195 = None
	        convert_element_type_129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_338, torch.float32);  view_338 = None
	        sub_1003: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_128, convert_element_type_129);  convert_element_type_128 = convert_element_type_129 = None
	        mul_2134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1003, view_337);  sub_1003 = view_337 = None
	        view_339: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2134, [sym_size_int, 1500, 1280]);  mul_2134 = None
	        _assert_tensor_metadata_196 = torch.ops.aten._assert_tensor_metadata.default(view_339, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_196 = None
	        view_340: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg100_1, [1280, 40, 32]);  arg100_1 = None
	        view_341: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg101_1, [1280, 40, 1]);  arg101_1 = None
	        view_342: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg102_1, [1280, 40, 1]);  arg102_1 = None
	        _assert_tensor_metadata_197 = torch.ops.aten._assert_tensor_metadata.default(view_340, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_197 = None
	        convert_element_type_130: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_340, torch.float32);  view_340 = None
	        _assert_tensor_metadata_198 = torch.ops.aten._assert_tensor_metadata.default(view_342, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_198 = None
	        convert_element_type_131: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_342, torch.float32);  view_342 = None
	        sub_1007: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_130, convert_element_type_131);  convert_element_type_130 = convert_element_type_131 = None
	        mul_2139: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1007, view_341);  sub_1007 = view_341 = None
	        view_343: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2139, [1280, 1280]);  mul_2139 = None
	        _assert_tensor_metadata_199 = torch.ops.aten._assert_tensor_metadata.default(view_343, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_199 = None
	        mul_2144: "Sym(1500*s6)" = sym_size_int * 1500
	        view_344: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_339, [mul_2144, 1280]);  view_339 = mul_2144 = None
	        permute_38: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_343, [1, 0]);  view_343 = None
	        addmm_17: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg99_1, view_344, permute_38);  arg99_1 = view_344 = permute_38 = None
	        view_345: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_17, [sym_size_int, 1500, 1280]);  addmm_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_29: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_345);  view_345 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_3400: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_2780, clone_29);  add_2780 = clone_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_3400, memory_format = torch.contiguous_format)
	        var_mean_7 = torch.ops.aten.var_mean.correction(clone_30, [2], correction = 0, keepdim = True)
	        getitem_30: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_7[0]
	        getitem_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_7[1];  var_mean_7 = None
	        add_3405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_30, 1e-05);  getitem_30 = None
	        rsqrt_7: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_3405);  add_3405 = None
	        sub_1013: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_30, getitem_31);  clone_30 = getitem_31 = None
	        mul_2155: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1013, rsqrt_7);  sub_1013 = rsqrt_7 = None
	        mul_2156: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2155, arg103_1);  mul_2155 = arg103_1 = None
	        add_3406: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2156, arg104_1);  mul_2156 = arg104_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_346: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_3406, [sym_size_int, 1500, 1280])
	        amin_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_346, [2])
	        amax_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_346, [2]);  view_346 = None
	        full_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_22, full_44);  amin_22 = full_44 = None
	        full_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_22, full_45);  amax_22 = full_45 = None
	        sub_1024: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_22, minimum_22);  maximum_22 = None
	        div_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1024, 255.0);  sub_1024 = None
	        clamp_min_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_44, 1.1920928955078125e-07);  div_44 = None
	        div_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_22, clamp_min_66);  minimum_22 = None
	        round_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_45);  div_45 = None
	        sub_1030: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_45);  round_45 = None
	        clamp_min_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1030, -128);  sub_1030 = None
	        clamp_max_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_67, 127);  clamp_min_67 = None
	        _assert_tensor_metadata_200 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_66, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_200 = None
	        _assert_tensor_metadata_201 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_44, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_201 = None
	        convert_element_type_132: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_44, torch.int8);  clamp_max_44 = None
	        view_347: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_3406, [sym_size_int, 1500, 1280]);  add_3406 = None
	        view_348: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_66, [sym_size_int, 1500, 1])
	        view_349: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_132, [sym_size_int, 1500, 1])
	        reciprocal_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_348);  view_348 = None
	        mul_2204: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_22, 1.0);  reciprocal_22 = None
	        mul_2207: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_347, mul_2204);  view_347 = mul_2204 = None
	        round_46: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2207);  mul_2207 = None
	        add_3493: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_46, view_349);  round_46 = view_349 = None
	        clamp_min_68: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3493, -128);  add_3493 = None
	        clamp_max_45: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_68, 127);  clamp_min_68 = None
	        view_350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_45, [sym_size_int, 1500, 1280]);  clamp_max_45 = None
	        _assert_tensor_metadata_202 = torch.ops.aten._assert_tensor_metadata.default(view_350, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_202 = None
	        convert_element_type_133: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_350, torch.int8);  view_350 = None
	        view_351: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_133, [sym_size_int, 1500, 1280]);  convert_element_type_133 = None
	        view_352: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_66, [sym_size_int, 1500, 1]);  clamp_min_66 = None
	        view_353: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_132, [sym_size_int, 1500, 1]);  convert_element_type_132 = None
	        _assert_tensor_metadata_203 = torch.ops.aten._assert_tensor_metadata.default(view_351, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_203 = None
	        convert_element_type_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_351, torch.float32);  view_351 = None
	        _assert_tensor_metadata_204 = torch.ops.aten._assert_tensor_metadata.default(view_353, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_204 = None
	        convert_element_type_135: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_353, torch.float32);  view_353 = None
	        sub_1050: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_134, convert_element_type_135);  convert_element_type_134 = convert_element_type_135 = None
	        mul_2229: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1050, view_352);  sub_1050 = view_352 = None
	        view_354: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2229, [sym_size_int, 1500, 1280]);  mul_2229 = None
	        _assert_tensor_metadata_205 = torch.ops.aten._assert_tensor_metadata.default(view_354, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_205 = None
	        view_355: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg106_1, [5120, 40, 32]);  arg106_1 = None
	        view_356: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg107_1, [5120, 40, 1]);  arg107_1 = None
	        view_357: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg108_1, [5120, 40, 1]);  arg108_1 = None
	        _assert_tensor_metadata_206 = torch.ops.aten._assert_tensor_metadata.default(view_355, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_206 = None
	        convert_element_type_136: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_355, torch.float32);  view_355 = None
	        _assert_tensor_metadata_207 = torch.ops.aten._assert_tensor_metadata.default(view_357, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_207 = None
	        convert_element_type_137: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_357, torch.float32);  view_357 = None
	        sub_1054: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_136, convert_element_type_137);  convert_element_type_136 = convert_element_type_137 = None
	        mul_2234: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1054, view_356);  sub_1054 = view_356 = None
	        view_358: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2234, [5120, 1280]);  mul_2234 = None
	        _assert_tensor_metadata_208 = torch.ops.aten._assert_tensor_metadata.default(view_358, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_208 = None
	        mul_2239: "Sym(1500*s6)" = sym_size_int * 1500
	        view_359: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_354, [mul_2239, 1280]);  view_354 = mul_2239 = None
	        permute_39: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_358, [1, 0]);  view_358 = None
	        addmm_18: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg105_1, view_359, permute_39);  arg105_1 = view_359 = permute_39 = None
	        view_360: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_18, [sym_size_int, 1500, 5120]);  addmm_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_2246: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_360, 0.5)
	        mul_2247: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_360, 0.7071067811865476);  view_360 = None
	        erf_5: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_2247);  mul_2247 = None
	        add_3552: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_5, 1);  erf_5 = None
	        mul_2248: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2246, add_3552);  mul_2246 = add_3552 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_31: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_2248);  mul_2248 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_361: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_31, [sym_size_int, 1500, 5120])
	        amin_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_361, [2])
	        amax_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_361, [2]);  view_361 = None
	        full_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_23, full_46);  amin_23 = full_46 = None
	        full_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_23, full_47);  amax_23 = full_47 = None
	        sub_1067: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_23, minimum_23);  maximum_23 = None
	        div_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1067, 255.0);  sub_1067 = None
	        clamp_min_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_46, 1.1920928955078125e-07);  div_46 = None
	        div_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_23, clamp_min_69);  minimum_23 = None
	        round_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_47);  div_47 = None
	        sub_1073: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_47);  round_47 = None
	        clamp_min_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1073, -128);  sub_1073 = None
	        clamp_max_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_70, 127);  clamp_min_70 = None
	        _assert_tensor_metadata_209 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_69, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_209 = None
	        _assert_tensor_metadata_210 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_46, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_210 = None
	        convert_element_type_138: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_46, torch.int8);  clamp_max_46 = None
	        view_362: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_31, [sym_size_int, 1500, 5120]);  clone_31 = None
	        view_363: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_69, [sym_size_int, 1500, 1])
	        view_364: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_138, [sym_size_int, 1500, 1])
	        reciprocal_23: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_363);  view_363 = None
	        mul_2294: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_23, 1.0);  reciprocal_23 = None
	        mul_2297: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_362, mul_2294);  view_362 = mul_2294 = None
	        round_48: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_2297);  mul_2297 = None
	        add_3635: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_48, view_364);  round_48 = view_364 = None
	        clamp_min_71: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3635, -128);  add_3635 = None
	        clamp_max_47: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_71, 127);  clamp_min_71 = None
	        view_365: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_47, [sym_size_int, 1500, 5120]);  clamp_max_47 = None
	        _assert_tensor_metadata_211 = torch.ops.aten._assert_tensor_metadata.default(view_365, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_211 = None
	        convert_element_type_139: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_365, torch.int8);  view_365 = None
	        view_366: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_139, [sym_size_int, 1500, 5120]);  convert_element_type_139 = None
	        view_367: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_69, [sym_size_int, 1500, 1]);  clamp_min_69 = None
	        view_368: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_138, [sym_size_int, 1500, 1]);  convert_element_type_138 = None
	        _assert_tensor_metadata_212 = torch.ops.aten._assert_tensor_metadata.default(view_366, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_212 = None
	        convert_element_type_140: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_366, torch.float32);  view_366 = None
	        _assert_tensor_metadata_213 = torch.ops.aten._assert_tensor_metadata.default(view_368, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_213 = None
	        convert_element_type_141: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_368, torch.float32);  view_368 = None
	        sub_1093: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_140, convert_element_type_141);  convert_element_type_140 = convert_element_type_141 = None
	        mul_2319: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1093, view_367);  sub_1093 = view_367 = None
	        view_369: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_2319, [sym_size_int, 1500, 5120]);  mul_2319 = None
	        _assert_tensor_metadata_214 = torch.ops.aten._assert_tensor_metadata.default(view_369, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_214 = None
	        view_370: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg110_1, [1280, 160, 32]);  arg110_1 = None
	        view_371: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg111_1, [1280, 160, 1]);  arg111_1 = None
	        view_372: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg112_1, [1280, 160, 1]);  arg112_1 = None
	        _assert_tensor_metadata_215 = torch.ops.aten._assert_tensor_metadata.default(view_370, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_215 = None
	        convert_element_type_142: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_370, torch.float32);  view_370 = None
	        _assert_tensor_metadata_216 = torch.ops.aten._assert_tensor_metadata.default(view_372, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_216 = None
	        convert_element_type_143: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_372, torch.float32);  view_372 = None
	        sub_1097: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_142, convert_element_type_143);  convert_element_type_142 = convert_element_type_143 = None
	        mul_2324: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1097, view_371);  sub_1097 = view_371 = None
	        view_373: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_2324, [1280, 5120]);  mul_2324 = None
	        _assert_tensor_metadata_217 = torch.ops.aten._assert_tensor_metadata.default(view_373, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_217 = None
	        mul_2329: "Sym(1500*s6)" = sym_size_int * 1500
	        view_374: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_369, [mul_2329, 5120]);  view_369 = mul_2329 = None
	        permute_40: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_373, [1, 0]);  view_373 = None
	        addmm_19: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg109_1, view_374, permute_40);  arg109_1 = view_374 = permute_40 = None
	        view_375: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_19, [sym_size_int, 1500, 1280]);  addmm_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_375);  view_375 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_3698: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_3400, clone_32);  add_3400 = clone_32 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_33: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_3698, memory_format = torch.contiguous_format)
	        var_mean_8 = torch.ops.aten.var_mean.correction(clone_33, [2], correction = 0, keepdim = True)
	        getitem_32: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_8[0]
	        getitem_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_8[1];  var_mean_8 = None
	        add_3703: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_32, 1e-05);  getitem_32 = None
	        rsqrt_8: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_3703);  add_3703 = None
	        sub_1103: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_33, getitem_33);  clone_33 = getitem_33 = None
	        mul_2340: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1103, rsqrt_8);  sub_1103 = rsqrt_8 = None
	        mul_2341: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2340, arg113_1);  mul_2340 = arg113_1 = None
	        add_3704: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2341, arg114_1);  mul_2341 = arg114_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_376: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_3704, [sym_size_int, 1500, 1280])
	        amin_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_376, [2])
	        amax_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_376, [2]);  view_376 = None
	        full_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_24, full_48);  amin_24 = full_48 = None
	        full_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_24, full_49);  amax_24 = full_49 = None
	        sub_1114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_24, minimum_24);  maximum_24 = None
	        div_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1114, 255.0);  sub_1114 = None
	        clamp_min_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_48, 1.1920928955078125e-07);  div_48 = None
	        div_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_24, clamp_min_72);  minimum_24 = None
	        round_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_49);  div_49 = None
	        sub_1120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_49);  round_49 = None
	        clamp_min_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1120, -128);  sub_1120 = None
	        clamp_max_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_73, 127);  clamp_min_73 = None
	        _assert_tensor_metadata_218 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_72, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_218 = None
	        _assert_tensor_metadata_219 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_48, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_219 = None
	        convert_element_type_144: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_48, torch.int8);  clamp_max_48 = None
	        view_377: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_3704, [sym_size_int, 1500, 1280])
	        view_378: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_72, [sym_size_int, 1500, 1])
	        view_379: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_144, [sym_size_int, 1500, 1])
	        reciprocal_24: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_378);  view_378 = None
	        mul_2389: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_24, 1.0);  reciprocal_24 = None
	        mul_2392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_377, mul_2389);  view_377 = mul_2389 = None
	        round_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2392);  mul_2392 = None
	        add_3791: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_50, view_379);  round_50 = view_379 = None
	        clamp_min_74: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3791, -128);  add_3791 = None
	        clamp_max_49: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_74, 127);  clamp_min_74 = None
	        view_380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_49, [sym_size_int, 1500, 1280]);  clamp_max_49 = None
	        _assert_tensor_metadata_220 = torch.ops.aten._assert_tensor_metadata.default(view_380, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_220 = None
	        convert_element_type_145: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_380, torch.int8);  view_380 = None
	        view_381: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_145, [sym_size_int, 1500, 1280]);  convert_element_type_145 = None
	        view_382: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_72, [sym_size_int, 1500, 1]);  clamp_min_72 = None
	        view_383: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_144, [sym_size_int, 1500, 1]);  convert_element_type_144 = None
	        _assert_tensor_metadata_221 = torch.ops.aten._assert_tensor_metadata.default(view_381, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_221 = None
	        convert_element_type_146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_381, torch.float32);  view_381 = None
	        _assert_tensor_metadata_222 = torch.ops.aten._assert_tensor_metadata.default(view_383, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_222 = None
	        convert_element_type_147: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_383, torch.float32);  view_383 = None
	        sub_1140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_146, convert_element_type_147);  convert_element_type_146 = convert_element_type_147 = None
	        mul_2414: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1140, view_382);  sub_1140 = view_382 = None
	        view_384: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2414, [sym_size_int, 1500, 1280]);  mul_2414 = None
	        _assert_tensor_metadata_223 = torch.ops.aten._assert_tensor_metadata.default(view_384, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_223 = None
	        view_385: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg116_1, [1280, 40, 32]);  arg116_1 = None
	        view_386: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg117_1, [1280, 40, 1]);  arg117_1 = None
	        view_387: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg118_1, [1280, 40, 1]);  arg118_1 = None
	        _assert_tensor_metadata_224 = torch.ops.aten._assert_tensor_metadata.default(view_385, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_224 = None
	        convert_element_type_148: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_385, torch.float32);  view_385 = None
	        _assert_tensor_metadata_225 = torch.ops.aten._assert_tensor_metadata.default(view_387, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_225 = None
	        convert_element_type_149: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_387, torch.float32);  view_387 = None
	        sub_1144: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_148, convert_element_type_149);  convert_element_type_148 = convert_element_type_149 = None
	        mul_2419: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1144, view_386);  sub_1144 = view_386 = None
	        view_388: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2419, [1280, 1280]);  mul_2419 = None
	        _assert_tensor_metadata_226 = torch.ops.aten._assert_tensor_metadata.default(view_388, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_226 = None
	        mul_2424: "Sym(1500*s6)" = sym_size_int * 1500
	        view_389: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_384, [mul_2424, 1280]);  view_384 = mul_2424 = None
	        permute_41: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_388, [1, 0]);  view_388 = None
	        addmm_20: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg115_1, view_389, permute_41);  arg115_1 = view_389 = permute_41 = None
	        view_390: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_20, [sym_size_int, 1500, 1280]);  addmm_20 = None
	        mul_2431: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_390, 0.125);  view_390 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_391: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_2431, [sym_size_int, 1500, 20, 64]);  mul_2431 = None
	        permute_42: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_391, [0, 2, 1, 3]);  view_391 = None
	        clone_34: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_42, memory_format = torch.contiguous_format);  permute_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_3704, [sym_size_int, 1500, 1280])
	        amin_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_392, [2])
	        amax_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_392, [2]);  view_392 = None
	        full_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_25, full_50);  amin_25 = full_50 = None
	        full_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_25, full_51);  amax_25 = full_51 = None
	        sub_1159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_25, minimum_25);  maximum_25 = None
	        div_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1159, 255.0);  sub_1159 = None
	        clamp_min_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_50, 1.1920928955078125e-07);  div_50 = None
	        div_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_25, clamp_min_75);  minimum_25 = None
	        round_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_51);  div_51 = None
	        sub_1165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_51);  round_51 = None
	        clamp_min_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1165, -128);  sub_1165 = None
	        clamp_max_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_76, 127);  clamp_min_76 = None
	        _assert_tensor_metadata_227 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_75, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_227 = None
	        _assert_tensor_metadata_228 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_50, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_228 = None
	        convert_element_type_150: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_50, torch.int8);  clamp_max_50 = None
	        view_393: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_3704, [sym_size_int, 1500, 1280])
	        view_394: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_75, [sym_size_int, 1500, 1])
	        view_395: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_150, [sym_size_int, 1500, 1])
	        reciprocal_25: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_394);  view_394 = None
	        mul_2485: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_25, 1.0);  reciprocal_25 = None
	        mul_2488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_393, mul_2485);  view_393 = mul_2485 = None
	        round_52: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2488);  mul_2488 = None
	        add_3943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_52, view_395);  round_52 = view_395 = None
	        clamp_min_77: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3943, -128);  add_3943 = None
	        clamp_max_51: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_77, 127);  clamp_min_77 = None
	        view_396: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_51, [sym_size_int, 1500, 1280]);  clamp_max_51 = None
	        _assert_tensor_metadata_229 = torch.ops.aten._assert_tensor_metadata.default(view_396, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_229 = None
	        convert_element_type_151: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_396, torch.int8);  view_396 = None
	        view_397: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_151, [sym_size_int, 1500, 1280]);  convert_element_type_151 = None
	        view_398: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_75, [sym_size_int, 1500, 1]);  clamp_min_75 = None
	        view_399: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_150, [sym_size_int, 1500, 1]);  convert_element_type_150 = None
	        _assert_tensor_metadata_230 = torch.ops.aten._assert_tensor_metadata.default(view_397, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_230 = None
	        convert_element_type_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_397, torch.float32);  view_397 = None
	        _assert_tensor_metadata_231 = torch.ops.aten._assert_tensor_metadata.default(view_399, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_231 = None
	        convert_element_type_153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_399, torch.float32);  view_399 = None
	        sub_1185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_152, convert_element_type_153);  convert_element_type_152 = convert_element_type_153 = None
	        mul_2510: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1185, view_398);  sub_1185 = view_398 = None
	        view_400: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2510, [sym_size_int, 1500, 1280]);  mul_2510 = None
	        _assert_tensor_metadata_232 = torch.ops.aten._assert_tensor_metadata.default(view_400, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_232 = None
	        view_401: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg119_1, [1280, 40, 32]);  arg119_1 = None
	        view_402: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg120_1, [1280, 40, 1]);  arg120_1 = None
	        view_403: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg121_1, [1280, 40, 1]);  arg121_1 = None
	        _assert_tensor_metadata_233 = torch.ops.aten._assert_tensor_metadata.default(view_401, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_233 = None
	        convert_element_type_154: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_401, torch.float32);  view_401 = None
	        _assert_tensor_metadata_234 = torch.ops.aten._assert_tensor_metadata.default(view_403, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_234 = None
	        convert_element_type_155: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_403, torch.float32);  view_403 = None
	        sub_1189: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_154, convert_element_type_155);  convert_element_type_154 = convert_element_type_155 = None
	        mul_2515: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1189, view_402);  sub_1189 = view_402 = None
	        view_404: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2515, [1280, 1280]);  mul_2515 = None
	        _assert_tensor_metadata_235 = torch.ops.aten._assert_tensor_metadata.default(view_404, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_235 = None
	        permute_43: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_404, [1, 0]);  view_404 = None
	        mul_2518: "Sym(1500*s6)" = sym_size_int * 1500
	        view_405: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_400, [mul_2518, 1280]);  view_400 = mul_2518 = None
	        mm_4: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_405, permute_43);  view_405 = permute_43 = None
	        view_406: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_4, [sym_size_int, 1500, 1280]);  mm_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_407: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_406, [sym_size_int, -1, 20, 64]);  view_406 = None
	        permute_44: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_407, [0, 2, 1, 3]);  view_407 = None
	        clone_35: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_44, memory_format = torch.contiguous_format);  permute_44 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_408: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_3704, [sym_size_int, 1500, 1280])
	        amin_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_408, [2])
	        amax_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_408, [2]);  view_408 = None
	        full_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_26, full_52);  amin_26 = full_52 = None
	        full_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_26, full_53);  amax_26 = full_53 = None
	        sub_1203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_26, minimum_26);  maximum_26 = None
	        div_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1203, 255.0);  sub_1203 = None
	        clamp_min_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_52, 1.1920928955078125e-07);  div_52 = None
	        div_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_26, clamp_min_78);  minimum_26 = None
	        round_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_53);  div_53 = None
	        sub_1209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_53);  round_53 = None
	        clamp_min_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1209, -128);  sub_1209 = None
	        clamp_max_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_79, 127);  clamp_min_79 = None
	        _assert_tensor_metadata_236 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_78, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_236 = None
	        _assert_tensor_metadata_237 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_52, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_237 = None
	        convert_element_type_156: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_52, torch.int8);  clamp_max_52 = None
	        view_409: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_3704, [sym_size_int, 1500, 1280]);  add_3704 = None
	        view_410: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_78, [sym_size_int, 1500, 1])
	        view_411: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_156, [sym_size_int, 1500, 1])
	        reciprocal_26: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_410);  view_410 = None
	        mul_2584: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_26, 1.0);  reciprocal_26 = None
	        mul_2587: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_409, mul_2584);  view_409 = mul_2584 = None
	        round_54: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2587);  mul_2587 = None
	        add_4091: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_54, view_411);  round_54 = view_411 = None
	        clamp_min_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4091, -128);  add_4091 = None
	        clamp_max_53: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_80, 127);  clamp_min_80 = None
	        view_412: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_53, [sym_size_int, 1500, 1280]);  clamp_max_53 = None
	        _assert_tensor_metadata_238 = torch.ops.aten._assert_tensor_metadata.default(view_412, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_238 = None
	        convert_element_type_157: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_412, torch.int8);  view_412 = None
	        view_413: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_157, [sym_size_int, 1500, 1280]);  convert_element_type_157 = None
	        view_414: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_78, [sym_size_int, 1500, 1]);  clamp_min_78 = None
	        view_415: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_156, [sym_size_int, 1500, 1]);  convert_element_type_156 = None
	        _assert_tensor_metadata_239 = torch.ops.aten._assert_tensor_metadata.default(view_413, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_239 = None
	        convert_element_type_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_413, torch.float32);  view_413 = None
	        _assert_tensor_metadata_240 = torch.ops.aten._assert_tensor_metadata.default(view_415, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_240 = None
	        convert_element_type_159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_415, torch.float32);  view_415 = None
	        sub_1229: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_158, convert_element_type_159);  convert_element_type_158 = convert_element_type_159 = None
	        mul_2609: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1229, view_414);  sub_1229 = view_414 = None
	        view_416: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2609, [sym_size_int, 1500, 1280]);  mul_2609 = None
	        _assert_tensor_metadata_241 = torch.ops.aten._assert_tensor_metadata.default(view_416, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_241 = None
	        view_417: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg123_1, [1280, 40, 32]);  arg123_1 = None
	        view_418: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg124_1, [1280, 40, 1]);  arg124_1 = None
	        view_419: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg125_1, [1280, 40, 1]);  arg125_1 = None
	        _assert_tensor_metadata_242 = torch.ops.aten._assert_tensor_metadata.default(view_417, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_242 = None
	        convert_element_type_160: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_417, torch.float32);  view_417 = None
	        _assert_tensor_metadata_243 = torch.ops.aten._assert_tensor_metadata.default(view_419, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_243 = None
	        convert_element_type_161: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_419, torch.float32);  view_419 = None
	        sub_1233: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_160, convert_element_type_161);  convert_element_type_160 = convert_element_type_161 = None
	        mul_2614: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1233, view_418);  sub_1233 = view_418 = None
	        view_420: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2614, [1280, 1280]);  mul_2614 = None
	        _assert_tensor_metadata_244 = torch.ops.aten._assert_tensor_metadata.default(view_420, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_244 = None
	        mul_2619: "Sym(1500*s6)" = sym_size_int * 1500
	        view_421: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_416, [mul_2619, 1280]);  view_416 = mul_2619 = None
	        permute_45: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_420, [1, 0]);  view_420 = None
	        addmm_21: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg122_1, view_421, permute_45);  arg122_1 = view_421 = permute_45 = None
	        view_422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_21, [sym_size_int, 1500, 1280]);  addmm_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_423: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_422, [sym_size_int, -1, 20, 64]);  view_422 = None
	        permute_46: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_423, [0, 2, 1, 3]);  view_423 = None
	        clone_36: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_46, memory_format = torch.contiguous_format);  permute_46 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_4 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_34, clone_35, clone_36, None, False, scale = 1.0);  clone_34 = clone_35 = clone_36 = None
	        getitem_34: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_4[0];  _scaled_dot_product_efficient_attention_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_47: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_34, [0, 2, 1, 3]);  getitem_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_424: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_47, [sym_size_int, 1500, -1]);  permute_47 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_425: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_424, [sym_size_int, 1500, 1280])
	        amin_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_425, [2])
	        amax_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_425, [2]);  view_425 = None
	        full_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_27, full_54);  amin_27 = full_54 = None
	        full_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_27, full_55);  amax_27 = full_55 = None
	        sub_1251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_27, minimum_27);  maximum_27 = None
	        div_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1251, 255.0);  sub_1251 = None
	        clamp_min_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_54, 1.1920928955078125e-07);  div_54 = None
	        div_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_27, clamp_min_81);  minimum_27 = None
	        round_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_55);  div_55 = None
	        sub_1257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_55);  round_55 = None
	        clamp_min_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1257, -128);  sub_1257 = None
	        clamp_max_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_82, 127);  clamp_min_82 = None
	        _assert_tensor_metadata_245 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_81, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_245 = None
	        _assert_tensor_metadata_246 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_54, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_246 = None
	        convert_element_type_162: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_54, torch.int8);  clamp_max_54 = None
	        view_426: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_424, [sym_size_int, 1500, 1280]);  view_424 = None
	        view_427: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_81, [sym_size_int, 1500, 1])
	        view_428: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_162, [sym_size_int, 1500, 1])
	        reciprocal_27: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_427);  view_427 = None
	        mul_2689: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_27, 1.0);  reciprocal_27 = None
	        mul_2692: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_426, mul_2689);  view_426 = mul_2689 = None
	        round_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2692);  mul_2692 = None
	        add_4255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_56, view_428);  round_56 = view_428 = None
	        clamp_min_83: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4255, -128);  add_4255 = None
	        clamp_max_55: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_83, 127);  clamp_min_83 = None
	        view_429: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_55, [sym_size_int, 1500, 1280]);  clamp_max_55 = None
	        _assert_tensor_metadata_247 = torch.ops.aten._assert_tensor_metadata.default(view_429, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_247 = None
	        convert_element_type_163: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_429, torch.int8);  view_429 = None
	        view_430: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_163, [sym_size_int, 1500, 1280]);  convert_element_type_163 = None
	        view_431: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_81, [sym_size_int, 1500, 1]);  clamp_min_81 = None
	        view_432: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_162, [sym_size_int, 1500, 1]);  convert_element_type_162 = None
	        _assert_tensor_metadata_248 = torch.ops.aten._assert_tensor_metadata.default(view_430, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_248 = None
	        convert_element_type_164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_430, torch.float32);  view_430 = None
	        _assert_tensor_metadata_249 = torch.ops.aten._assert_tensor_metadata.default(view_432, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_249 = None
	        convert_element_type_165: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_432, torch.float32);  view_432 = None
	        sub_1277: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_164, convert_element_type_165);  convert_element_type_164 = convert_element_type_165 = None
	        mul_2714: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1277, view_431);  sub_1277 = view_431 = None
	        view_433: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2714, [sym_size_int, 1500, 1280]);  mul_2714 = None
	        _assert_tensor_metadata_250 = torch.ops.aten._assert_tensor_metadata.default(view_433, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_250 = None
	        view_434: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg127_1, [1280, 40, 32]);  arg127_1 = None
	        view_435: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg128_1, [1280, 40, 1]);  arg128_1 = None
	        view_436: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg129_1, [1280, 40, 1]);  arg129_1 = None
	        _assert_tensor_metadata_251 = torch.ops.aten._assert_tensor_metadata.default(view_434, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_251 = None
	        convert_element_type_166: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_434, torch.float32);  view_434 = None
	        _assert_tensor_metadata_252 = torch.ops.aten._assert_tensor_metadata.default(view_436, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_252 = None
	        convert_element_type_167: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_436, torch.float32);  view_436 = None
	        sub_1281: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_166, convert_element_type_167);  convert_element_type_166 = convert_element_type_167 = None
	        mul_2719: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1281, view_435);  sub_1281 = view_435 = None
	        view_437: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2719, [1280, 1280]);  mul_2719 = None
	        _assert_tensor_metadata_253 = torch.ops.aten._assert_tensor_metadata.default(view_437, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_253 = None
	        mul_2724: "Sym(1500*s6)" = sym_size_int * 1500
	        view_438: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_433, [mul_2724, 1280]);  view_433 = mul_2724 = None
	        permute_48: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_437, [1, 0]);  view_437 = None
	        addmm_22: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg126_1, view_438, permute_48);  arg126_1 = view_438 = permute_48 = None
	        view_439: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_22, [sym_size_int, 1500, 1280]);  addmm_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_37: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_439);  view_439 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_4318: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_3698, clone_37);  add_3698 = clone_37 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_4318, memory_format = torch.contiguous_format)
	        var_mean_9 = torch.ops.aten.var_mean.correction(clone_38, [2], correction = 0, keepdim = True)
	        getitem_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_9[0]
	        getitem_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_9[1];  var_mean_9 = None
	        add_4323: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_38, 1e-05);  getitem_38 = None
	        rsqrt_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_4323);  add_4323 = None
	        sub_1287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_38, getitem_39);  clone_38 = getitem_39 = None
	        mul_2735: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1287, rsqrt_9);  sub_1287 = rsqrt_9 = None
	        mul_2736: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2735, arg130_1);  mul_2735 = arg130_1 = None
	        add_4324: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2736, arg131_1);  mul_2736 = arg131_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_440: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_4324, [sym_size_int, 1500, 1280])
	        amin_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_440, [2])
	        amax_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_440, [2]);  view_440 = None
	        full_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_28, full_56);  amin_28 = full_56 = None
	        full_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_28, full_57);  amax_28 = full_57 = None
	        sub_1298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_28, minimum_28);  maximum_28 = None
	        div_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1298, 255.0);  sub_1298 = None
	        clamp_min_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_56, 1.1920928955078125e-07);  div_56 = None
	        div_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_28, clamp_min_84);  minimum_28 = None
	        round_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_57);  div_57 = None
	        sub_1304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_57);  round_57 = None
	        clamp_min_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1304, -128);  sub_1304 = None
	        clamp_max_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_85, 127);  clamp_min_85 = None
	        _assert_tensor_metadata_254 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_84, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_254 = None
	        _assert_tensor_metadata_255 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_56, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_255 = None
	        convert_element_type_168: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_56, torch.int8);  clamp_max_56 = None
	        view_441: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_4324, [sym_size_int, 1500, 1280]);  add_4324 = None
	        view_442: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_84, [sym_size_int, 1500, 1])
	        view_443: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_168, [sym_size_int, 1500, 1])
	        reciprocal_28: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_442);  view_442 = None
	        mul_2784: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_28, 1.0);  reciprocal_28 = None
	        mul_2787: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_441, mul_2784);  view_441 = mul_2784 = None
	        round_58: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2787);  mul_2787 = None
	        add_4411: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_58, view_443);  round_58 = view_443 = None
	        clamp_min_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4411, -128);  add_4411 = None
	        clamp_max_57: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_86, 127);  clamp_min_86 = None
	        view_444: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_57, [sym_size_int, 1500, 1280]);  clamp_max_57 = None
	        _assert_tensor_metadata_256 = torch.ops.aten._assert_tensor_metadata.default(view_444, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_256 = None
	        convert_element_type_169: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_444, torch.int8);  view_444 = None
	        view_445: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_169, [sym_size_int, 1500, 1280]);  convert_element_type_169 = None
	        view_446: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_84, [sym_size_int, 1500, 1]);  clamp_min_84 = None
	        view_447: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_168, [sym_size_int, 1500, 1]);  convert_element_type_168 = None
	        _assert_tensor_metadata_257 = torch.ops.aten._assert_tensor_metadata.default(view_445, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_257 = None
	        convert_element_type_170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_445, torch.float32);  view_445 = None
	        _assert_tensor_metadata_258 = torch.ops.aten._assert_tensor_metadata.default(view_447, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_258 = None
	        convert_element_type_171: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_447, torch.float32);  view_447 = None
	        sub_1324: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_170, convert_element_type_171);  convert_element_type_170 = convert_element_type_171 = None
	        mul_2809: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1324, view_446);  sub_1324 = view_446 = None
	        view_448: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2809, [sym_size_int, 1500, 1280]);  mul_2809 = None
	        _assert_tensor_metadata_259 = torch.ops.aten._assert_tensor_metadata.default(view_448, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_259 = None
	        view_449: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg133_1, [5120, 40, 32]);  arg133_1 = None
	        view_450: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg134_1, [5120, 40, 1]);  arg134_1 = None
	        view_451: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg135_1, [5120, 40, 1]);  arg135_1 = None
	        _assert_tensor_metadata_260 = torch.ops.aten._assert_tensor_metadata.default(view_449, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_260 = None
	        convert_element_type_172: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_449, torch.float32);  view_449 = None
	        _assert_tensor_metadata_261 = torch.ops.aten._assert_tensor_metadata.default(view_451, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_261 = None
	        convert_element_type_173: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_451, torch.float32);  view_451 = None
	        sub_1328: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_172, convert_element_type_173);  convert_element_type_172 = convert_element_type_173 = None
	        mul_2814: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1328, view_450);  sub_1328 = view_450 = None
	        view_452: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2814, [5120, 1280]);  mul_2814 = None
	        _assert_tensor_metadata_262 = torch.ops.aten._assert_tensor_metadata.default(view_452, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_262 = None
	        mul_2819: "Sym(1500*s6)" = sym_size_int * 1500
	        view_453: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_448, [mul_2819, 1280]);  view_448 = mul_2819 = None
	        permute_49: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_452, [1, 0]);  view_452 = None
	        addmm_23: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg132_1, view_453, permute_49);  arg132_1 = view_453 = permute_49 = None
	        view_454: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_23, [sym_size_int, 1500, 5120]);  addmm_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_2826: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_454, 0.5)
	        mul_2827: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_454, 0.7071067811865476);  view_454 = None
	        erf_6: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_2827);  mul_2827 = None
	        add_4470: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_6, 1);  erf_6 = None
	        mul_2828: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2826, add_4470);  mul_2826 = add_4470 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_39: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_2828);  mul_2828 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_455: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_39, [sym_size_int, 1500, 5120])
	        amin_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_455, [2])
	        amax_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_455, [2]);  view_455 = None
	        full_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_29, full_58);  amin_29 = full_58 = None
	        full_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_29, full_59);  amax_29 = full_59 = None
	        sub_1341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_29, minimum_29);  maximum_29 = None
	        div_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1341, 255.0);  sub_1341 = None
	        clamp_min_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_58, 1.1920928955078125e-07);  div_58 = None
	        div_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_29, clamp_min_87);  minimum_29 = None
	        round_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_59);  div_59 = None
	        sub_1347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_59);  round_59 = None
	        clamp_min_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1347, -128);  sub_1347 = None
	        clamp_max_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_88, 127);  clamp_min_88 = None
	        _assert_tensor_metadata_263 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_87, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_263 = None
	        _assert_tensor_metadata_264 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_58, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_264 = None
	        convert_element_type_174: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_58, torch.int8);  clamp_max_58 = None
	        view_456: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_39, [sym_size_int, 1500, 5120]);  clone_39 = None
	        view_457: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_87, [sym_size_int, 1500, 1])
	        view_458: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_174, [sym_size_int, 1500, 1])
	        reciprocal_29: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_457);  view_457 = None
	        mul_2874: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_29, 1.0);  reciprocal_29 = None
	        mul_2877: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_456, mul_2874);  view_456 = mul_2874 = None
	        round_60: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_2877);  mul_2877 = None
	        add_4553: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_60, view_458);  round_60 = view_458 = None
	        clamp_min_89: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4553, -128);  add_4553 = None
	        clamp_max_59: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_89, 127);  clamp_min_89 = None
	        view_459: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_59, [sym_size_int, 1500, 5120]);  clamp_max_59 = None
	        _assert_tensor_metadata_265 = torch.ops.aten._assert_tensor_metadata.default(view_459, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_265 = None
	        convert_element_type_175: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_459, torch.int8);  view_459 = None
	        view_460: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_175, [sym_size_int, 1500, 5120]);  convert_element_type_175 = None
	        view_461: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_87, [sym_size_int, 1500, 1]);  clamp_min_87 = None
	        view_462: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_174, [sym_size_int, 1500, 1]);  convert_element_type_174 = None
	        _assert_tensor_metadata_266 = torch.ops.aten._assert_tensor_metadata.default(view_460, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_266 = None
	        convert_element_type_176: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_460, torch.float32);  view_460 = None
	        _assert_tensor_metadata_267 = torch.ops.aten._assert_tensor_metadata.default(view_462, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_267 = None
	        convert_element_type_177: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_462, torch.float32);  view_462 = None
	        sub_1367: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_176, convert_element_type_177);  convert_element_type_176 = convert_element_type_177 = None
	        mul_2899: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1367, view_461);  sub_1367 = view_461 = None
	        view_463: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_2899, [sym_size_int, 1500, 5120]);  mul_2899 = None
	        _assert_tensor_metadata_268 = torch.ops.aten._assert_tensor_metadata.default(view_463, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_268 = None
	        view_464: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg137_1, [1280, 160, 32]);  arg137_1 = None
	        view_465: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg138_1, [1280, 160, 1]);  arg138_1 = None
	        view_466: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg139_1, [1280, 160, 1]);  arg139_1 = None
	        _assert_tensor_metadata_269 = torch.ops.aten._assert_tensor_metadata.default(view_464, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_269 = None
	        convert_element_type_178: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_464, torch.float32);  view_464 = None
	        _assert_tensor_metadata_270 = torch.ops.aten._assert_tensor_metadata.default(view_466, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_270 = None
	        convert_element_type_179: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_466, torch.float32);  view_466 = None
	        sub_1371: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_178, convert_element_type_179);  convert_element_type_178 = convert_element_type_179 = None
	        mul_2904: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1371, view_465);  sub_1371 = view_465 = None
	        view_467: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_2904, [1280, 5120]);  mul_2904 = None
	        _assert_tensor_metadata_271 = torch.ops.aten._assert_tensor_metadata.default(view_467, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_271 = None
	        mul_2909: "Sym(1500*s6)" = sym_size_int * 1500
	        view_468: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_463, [mul_2909, 5120]);  view_463 = mul_2909 = None
	        permute_50: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_467, [1, 0]);  view_467 = None
	        addmm_24: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg136_1, view_468, permute_50);  arg136_1 = view_468 = permute_50 = None
	        view_469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_24, [sym_size_int, 1500, 1280]);  addmm_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_40: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_469);  view_469 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_4616: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_4318, clone_40);  add_4318 = clone_40 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_41: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_4616, memory_format = torch.contiguous_format)
	        var_mean_10 = torch.ops.aten.var_mean.correction(clone_41, [2], correction = 0, keepdim = True)
	        getitem_40: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_10[0]
	        getitem_41: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_10[1];  var_mean_10 = None
	        add_4621: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_40, 1e-05);  getitem_40 = None
	        rsqrt_10: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_4621);  add_4621 = None
	        sub_1377: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_41, getitem_41);  clone_41 = getitem_41 = None
	        mul_2920: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1377, rsqrt_10);  sub_1377 = rsqrt_10 = None
	        mul_2921: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2920, arg140_1);  mul_2920 = arg140_1 = None
	        add_4622: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2921, arg141_1);  mul_2921 = arg141_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_470: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_4622, [sym_size_int, 1500, 1280])
	        amin_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_470, [2])
	        amax_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_470, [2]);  view_470 = None
	        full_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_30, full_60);  amin_30 = full_60 = None
	        full_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_30, full_61);  amax_30 = full_61 = None
	        sub_1388: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_30, minimum_30);  maximum_30 = None
	        div_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1388, 255.0);  sub_1388 = None
	        clamp_min_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_60, 1.1920928955078125e-07);  div_60 = None
	        div_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_30, clamp_min_90);  minimum_30 = None
	        round_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_61);  div_61 = None
	        sub_1394: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_61);  round_61 = None
	        clamp_min_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1394, -128);  sub_1394 = None
	        clamp_max_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_91, 127);  clamp_min_91 = None
	        _assert_tensor_metadata_272 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_90, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_272 = None
	        _assert_tensor_metadata_273 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_60, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_273 = None
	        convert_element_type_180: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_60, torch.int8);  clamp_max_60 = None
	        view_471: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_4622, [sym_size_int, 1500, 1280])
	        view_472: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_90, [sym_size_int, 1500, 1])
	        view_473: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_180, [sym_size_int, 1500, 1])
	        reciprocal_30: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_472);  view_472 = None
	        mul_2969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_30, 1.0);  reciprocal_30 = None
	        mul_2972: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_471, mul_2969);  view_471 = mul_2969 = None
	        round_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2972);  mul_2972 = None
	        add_4709: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_62, view_473);  round_62 = view_473 = None
	        clamp_min_92: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4709, -128);  add_4709 = None
	        clamp_max_61: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_92, 127);  clamp_min_92 = None
	        view_474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_61, [sym_size_int, 1500, 1280]);  clamp_max_61 = None
	        _assert_tensor_metadata_274 = torch.ops.aten._assert_tensor_metadata.default(view_474, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_274 = None
	        convert_element_type_181: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_474, torch.int8);  view_474 = None
	        view_475: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_181, [sym_size_int, 1500, 1280]);  convert_element_type_181 = None
	        view_476: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_90, [sym_size_int, 1500, 1]);  clamp_min_90 = None
	        view_477: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_180, [sym_size_int, 1500, 1]);  convert_element_type_180 = None
	        _assert_tensor_metadata_275 = torch.ops.aten._assert_tensor_metadata.default(view_475, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_275 = None
	        convert_element_type_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_475, torch.float32);  view_475 = None
	        _assert_tensor_metadata_276 = torch.ops.aten._assert_tensor_metadata.default(view_477, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_276 = None
	        convert_element_type_183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_477, torch.float32);  view_477 = None
	        sub_1414: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_182, convert_element_type_183);  convert_element_type_182 = convert_element_type_183 = None
	        mul_2994: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1414, view_476);  sub_1414 = view_476 = None
	        view_478: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2994, [sym_size_int, 1500, 1280]);  mul_2994 = None
	        _assert_tensor_metadata_277 = torch.ops.aten._assert_tensor_metadata.default(view_478, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_277 = None
	        view_479: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg143_1, [1280, 40, 32]);  arg143_1 = None
	        view_480: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg144_1, [1280, 40, 1]);  arg144_1 = None
	        view_481: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg145_1, [1280, 40, 1]);  arg145_1 = None
	        _assert_tensor_metadata_278 = torch.ops.aten._assert_tensor_metadata.default(view_479, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_278 = None
	        convert_element_type_184: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_479, torch.float32);  view_479 = None
	        _assert_tensor_metadata_279 = torch.ops.aten._assert_tensor_metadata.default(view_481, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_279 = None
	        convert_element_type_185: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_481, torch.float32);  view_481 = None
	        sub_1418: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_184, convert_element_type_185);  convert_element_type_184 = convert_element_type_185 = None
	        mul_2999: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1418, view_480);  sub_1418 = view_480 = None
	        view_482: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2999, [1280, 1280]);  mul_2999 = None
	        _assert_tensor_metadata_280 = torch.ops.aten._assert_tensor_metadata.default(view_482, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_280 = None
	        mul_3004: "Sym(1500*s6)" = sym_size_int * 1500
	        view_483: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_478, [mul_3004, 1280]);  view_478 = mul_3004 = None
	        permute_51: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_482, [1, 0]);  view_482 = None
	        addmm_25: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg142_1, view_483, permute_51);  arg142_1 = view_483 = permute_51 = None
	        view_484: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_25, [sym_size_int, 1500, 1280]);  addmm_25 = None
	        mul_3011: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_484, 0.125);  view_484 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_485: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_3011, [sym_size_int, 1500, 20, 64]);  mul_3011 = None
	        permute_52: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_485, [0, 2, 1, 3]);  view_485 = None
	        clone_42: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_52, memory_format = torch.contiguous_format);  permute_52 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_486: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_4622, [sym_size_int, 1500, 1280])
	        amin_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_486, [2])
	        amax_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_486, [2]);  view_486 = None
	        full_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_31, full_62);  amin_31 = full_62 = None
	        full_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_31, full_63);  amax_31 = full_63 = None
	        sub_1433: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_31, minimum_31);  maximum_31 = None
	        div_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1433, 255.0);  sub_1433 = None
	        clamp_min_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_62, 1.1920928955078125e-07);  div_62 = None
	        div_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_31, clamp_min_93);  minimum_31 = None
	        round_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_63);  div_63 = None
	        sub_1439: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_63);  round_63 = None
	        clamp_min_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1439, -128);  sub_1439 = None
	        clamp_max_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_94, 127);  clamp_min_94 = None
	        _assert_tensor_metadata_281 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_93, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_281 = None
	        _assert_tensor_metadata_282 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_62, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_282 = None
	        convert_element_type_186: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_62, torch.int8);  clamp_max_62 = None
	        view_487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_4622, [sym_size_int, 1500, 1280])
	        view_488: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_93, [sym_size_int, 1500, 1])
	        view_489: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_186, [sym_size_int, 1500, 1])
	        reciprocal_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_488);  view_488 = None
	        mul_3065: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_31, 1.0);  reciprocal_31 = None
	        mul_3068: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_487, mul_3065);  view_487 = mul_3065 = None
	        round_64: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3068);  mul_3068 = None
	        add_4861: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_64, view_489);  round_64 = view_489 = None
	        clamp_min_95: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4861, -128);  add_4861 = None
	        clamp_max_63: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_95, 127);  clamp_min_95 = None
	        view_490: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_63, [sym_size_int, 1500, 1280]);  clamp_max_63 = None
	        _assert_tensor_metadata_283 = torch.ops.aten._assert_tensor_metadata.default(view_490, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_283 = None
	        convert_element_type_187: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_490, torch.int8);  view_490 = None
	        view_491: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_187, [sym_size_int, 1500, 1280]);  convert_element_type_187 = None
	        view_492: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_93, [sym_size_int, 1500, 1]);  clamp_min_93 = None
	        view_493: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_186, [sym_size_int, 1500, 1]);  convert_element_type_186 = None
	        _assert_tensor_metadata_284 = torch.ops.aten._assert_tensor_metadata.default(view_491, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_284 = None
	        convert_element_type_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_491, torch.float32);  view_491 = None
	        _assert_tensor_metadata_285 = torch.ops.aten._assert_tensor_metadata.default(view_493, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_285 = None
	        convert_element_type_189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_493, torch.float32);  view_493 = None
	        sub_1459: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_188, convert_element_type_189);  convert_element_type_188 = convert_element_type_189 = None
	        mul_3090: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1459, view_492);  sub_1459 = view_492 = None
	        view_494: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3090, [sym_size_int, 1500, 1280]);  mul_3090 = None
	        _assert_tensor_metadata_286 = torch.ops.aten._assert_tensor_metadata.default(view_494, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_286 = None
	        view_495: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg146_1, [1280, 40, 32]);  arg146_1 = None
	        view_496: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg147_1, [1280, 40, 1]);  arg147_1 = None
	        view_497: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg148_1, [1280, 40, 1]);  arg148_1 = None
	        _assert_tensor_metadata_287 = torch.ops.aten._assert_tensor_metadata.default(view_495, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_287 = None
	        convert_element_type_190: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_495, torch.float32);  view_495 = None
	        _assert_tensor_metadata_288 = torch.ops.aten._assert_tensor_metadata.default(view_497, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_288 = None
	        convert_element_type_191: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_497, torch.float32);  view_497 = None
	        sub_1463: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_190, convert_element_type_191);  convert_element_type_190 = convert_element_type_191 = None
	        mul_3095: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1463, view_496);  sub_1463 = view_496 = None
	        view_498: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3095, [1280, 1280]);  mul_3095 = None
	        _assert_tensor_metadata_289 = torch.ops.aten._assert_tensor_metadata.default(view_498, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_289 = None
	        permute_53: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_498, [1, 0]);  view_498 = None
	        mul_3098: "Sym(1500*s6)" = sym_size_int * 1500
	        view_499: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_494, [mul_3098, 1280]);  view_494 = mul_3098 = None
	        mm_5: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_499, permute_53);  view_499 = permute_53 = None
	        view_500: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_5, [sym_size_int, 1500, 1280]);  mm_5 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_501: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_500, [sym_size_int, -1, 20, 64]);  view_500 = None
	        permute_54: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_501, [0, 2, 1, 3]);  view_501 = None
	        clone_43: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_54, memory_format = torch.contiguous_format);  permute_54 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_502: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_4622, [sym_size_int, 1500, 1280])
	        amin_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_502, [2])
	        amax_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_502, [2]);  view_502 = None
	        full_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_32, full_64);  amin_32 = full_64 = None
	        full_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_32, full_65);  amax_32 = full_65 = None
	        sub_1477: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_32, minimum_32);  maximum_32 = None
	        div_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1477, 255.0);  sub_1477 = None
	        clamp_min_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_64, 1.1920928955078125e-07);  div_64 = None
	        div_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_32, clamp_min_96);  minimum_32 = None
	        round_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_65);  div_65 = None
	        sub_1483: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_65);  round_65 = None
	        clamp_min_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1483, -128);  sub_1483 = None
	        clamp_max_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_97, 127);  clamp_min_97 = None
	        _assert_tensor_metadata_290 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_96, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_290 = None
	        _assert_tensor_metadata_291 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_64, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_291 = None
	        convert_element_type_192: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_64, torch.int8);  clamp_max_64 = None
	        view_503: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_4622, [sym_size_int, 1500, 1280]);  add_4622 = None
	        view_504: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_96, [sym_size_int, 1500, 1])
	        view_505: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_192, [sym_size_int, 1500, 1])
	        reciprocal_32: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_504);  view_504 = None
	        mul_3164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_32, 1.0);  reciprocal_32 = None
	        mul_3167: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_503, mul_3164);  view_503 = mul_3164 = None
	        round_66: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3167);  mul_3167 = None
	        add_5009: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_66, view_505);  round_66 = view_505 = None
	        clamp_min_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5009, -128);  add_5009 = None
	        clamp_max_65: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_98, 127);  clamp_min_98 = None
	        view_506: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_65, [sym_size_int, 1500, 1280]);  clamp_max_65 = None
	        _assert_tensor_metadata_292 = torch.ops.aten._assert_tensor_metadata.default(view_506, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_292 = None
	        convert_element_type_193: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_506, torch.int8);  view_506 = None
	        view_507: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_193, [sym_size_int, 1500, 1280]);  convert_element_type_193 = None
	        view_508: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_96, [sym_size_int, 1500, 1]);  clamp_min_96 = None
	        view_509: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_192, [sym_size_int, 1500, 1]);  convert_element_type_192 = None
	        _assert_tensor_metadata_293 = torch.ops.aten._assert_tensor_metadata.default(view_507, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_293 = None
	        convert_element_type_194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_507, torch.float32);  view_507 = None
	        _assert_tensor_metadata_294 = torch.ops.aten._assert_tensor_metadata.default(view_509, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_294 = None
	        convert_element_type_195: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_509, torch.float32);  view_509 = None
	        sub_1503: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_194, convert_element_type_195);  convert_element_type_194 = convert_element_type_195 = None
	        mul_3189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1503, view_508);  sub_1503 = view_508 = None
	        view_510: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3189, [sym_size_int, 1500, 1280]);  mul_3189 = None
	        _assert_tensor_metadata_295 = torch.ops.aten._assert_tensor_metadata.default(view_510, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_295 = None
	        view_511: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg150_1, [1280, 40, 32]);  arg150_1 = None
	        view_512: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg151_1, [1280, 40, 1]);  arg151_1 = None
	        view_513: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg152_1, [1280, 40, 1]);  arg152_1 = None
	        _assert_tensor_metadata_296 = torch.ops.aten._assert_tensor_metadata.default(view_511, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_296 = None
	        convert_element_type_196: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_511, torch.float32);  view_511 = None
	        _assert_tensor_metadata_297 = torch.ops.aten._assert_tensor_metadata.default(view_513, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_297 = None
	        convert_element_type_197: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_513, torch.float32);  view_513 = None
	        sub_1507: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_196, convert_element_type_197);  convert_element_type_196 = convert_element_type_197 = None
	        mul_3194: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1507, view_512);  sub_1507 = view_512 = None
	        view_514: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3194, [1280, 1280]);  mul_3194 = None
	        _assert_tensor_metadata_298 = torch.ops.aten._assert_tensor_metadata.default(view_514, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_298 = None
	        mul_3199: "Sym(1500*s6)" = sym_size_int * 1500
	        view_515: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_510, [mul_3199, 1280]);  view_510 = mul_3199 = None
	        permute_55: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_514, [1, 0]);  view_514 = None
	        addmm_26: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg149_1, view_515, permute_55);  arg149_1 = view_515 = permute_55 = None
	        view_516: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_26, [sym_size_int, 1500, 1280]);  addmm_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_517: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_516, [sym_size_int, -1, 20, 64]);  view_516 = None
	        permute_56: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_517, [0, 2, 1, 3]);  view_517 = None
	        clone_44: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_56, memory_format = torch.contiguous_format);  permute_56 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_5 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_42, clone_43, clone_44, None, False, scale = 1.0);  clone_42 = clone_43 = clone_44 = None
	        getitem_42: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_5[0];  _scaled_dot_product_efficient_attention_5 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_57: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_42, [0, 2, 1, 3]);  getitem_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_57, [sym_size_int, 1500, -1]);  permute_57 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_519: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_518, [sym_size_int, 1500, 1280])
	        amin_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_519, [2])
	        amax_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_519, [2]);  view_519 = None
	        full_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_33, full_66);  amin_33 = full_66 = None
	        full_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_33, full_67);  amax_33 = full_67 = None
	        sub_1525: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_33, minimum_33);  maximum_33 = None
	        div_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1525, 255.0);  sub_1525 = None
	        clamp_min_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_66, 1.1920928955078125e-07);  div_66 = None
	        div_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_33, clamp_min_99);  minimum_33 = None
	        round_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_67);  div_67 = None
	        sub_1531: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_67);  round_67 = None
	        clamp_min_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1531, -128);  sub_1531 = None
	        clamp_max_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_100, 127);  clamp_min_100 = None
	        _assert_tensor_metadata_299 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_99, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_299 = None
	        _assert_tensor_metadata_300 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_66, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_300 = None
	        convert_element_type_198: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_66, torch.int8);  clamp_max_66 = None
	        view_520: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_518, [sym_size_int, 1500, 1280]);  view_518 = None
	        view_521: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_99, [sym_size_int, 1500, 1])
	        view_522: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_198, [sym_size_int, 1500, 1])
	        reciprocal_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_521);  view_521 = None
	        mul_3269: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_33, 1.0);  reciprocal_33 = None
	        mul_3272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_520, mul_3269);  view_520 = mul_3269 = None
	        round_68: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3272);  mul_3272 = None
	        add_5173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_68, view_522);  round_68 = view_522 = None
	        clamp_min_101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5173, -128);  add_5173 = None
	        clamp_max_67: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_101, 127);  clamp_min_101 = None
	        view_523: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_67, [sym_size_int, 1500, 1280]);  clamp_max_67 = None
	        _assert_tensor_metadata_301 = torch.ops.aten._assert_tensor_metadata.default(view_523, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_301 = None
	        convert_element_type_199: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_523, torch.int8);  view_523 = None
	        view_524: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_199, [sym_size_int, 1500, 1280]);  convert_element_type_199 = None
	        view_525: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_99, [sym_size_int, 1500, 1]);  clamp_min_99 = None
	        view_526: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_198, [sym_size_int, 1500, 1]);  convert_element_type_198 = None
	        _assert_tensor_metadata_302 = torch.ops.aten._assert_tensor_metadata.default(view_524, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_302 = None
	        convert_element_type_200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_524, torch.float32);  view_524 = None
	        _assert_tensor_metadata_303 = torch.ops.aten._assert_tensor_metadata.default(view_526, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_303 = None
	        convert_element_type_201: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_526, torch.float32);  view_526 = None
	        sub_1551: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_200, convert_element_type_201);  convert_element_type_200 = convert_element_type_201 = None
	        mul_3294: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1551, view_525);  sub_1551 = view_525 = None
	        view_527: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3294, [sym_size_int, 1500, 1280]);  mul_3294 = None
	        _assert_tensor_metadata_304 = torch.ops.aten._assert_tensor_metadata.default(view_527, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_304 = None
	        view_528: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg154_1, [1280, 40, 32]);  arg154_1 = None
	        view_529: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg155_1, [1280, 40, 1]);  arg155_1 = None
	        view_530: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg156_1, [1280, 40, 1]);  arg156_1 = None
	        _assert_tensor_metadata_305 = torch.ops.aten._assert_tensor_metadata.default(view_528, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_305 = None
	        convert_element_type_202: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_528, torch.float32);  view_528 = None
	        _assert_tensor_metadata_306 = torch.ops.aten._assert_tensor_metadata.default(view_530, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_306 = None
	        convert_element_type_203: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_530, torch.float32);  view_530 = None
	        sub_1555: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_202, convert_element_type_203);  convert_element_type_202 = convert_element_type_203 = None
	        mul_3299: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1555, view_529);  sub_1555 = view_529 = None
	        view_531: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3299, [1280, 1280]);  mul_3299 = None
	        _assert_tensor_metadata_307 = torch.ops.aten._assert_tensor_metadata.default(view_531, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_307 = None
	        mul_3304: "Sym(1500*s6)" = sym_size_int * 1500
	        view_532: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_527, [mul_3304, 1280]);  view_527 = mul_3304 = None
	        permute_58: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_531, [1, 0]);  view_531 = None
	        addmm_27: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg153_1, view_532, permute_58);  arg153_1 = view_532 = permute_58 = None
	        view_533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_27, [sym_size_int, 1500, 1280]);  addmm_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_45: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_533);  view_533 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_5236: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_4616, clone_45);  add_4616 = clone_45 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_46: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_5236, memory_format = torch.contiguous_format)
	        var_mean_11 = torch.ops.aten.var_mean.correction(clone_46, [2], correction = 0, keepdim = True)
	        getitem_46: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_11[0]
	        getitem_47: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_11[1];  var_mean_11 = None
	        add_5241: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_46, 1e-05);  getitem_46 = None
	        rsqrt_11: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_5241);  add_5241 = None
	        sub_1561: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_46, getitem_47);  clone_46 = getitem_47 = None
	        mul_3315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1561, rsqrt_11);  sub_1561 = rsqrt_11 = None
	        mul_3316: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3315, arg157_1);  mul_3315 = arg157_1 = None
	        add_5242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_3316, arg158_1);  mul_3316 = arg158_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_534: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_5242, [sym_size_int, 1500, 1280])
	        amin_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_534, [2])
	        amax_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_534, [2]);  view_534 = None
	        full_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_34, full_68);  amin_34 = full_68 = None
	        full_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_34, full_69);  amax_34 = full_69 = None
	        sub_1572: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_34, minimum_34);  maximum_34 = None
	        div_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1572, 255.0);  sub_1572 = None
	        clamp_min_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_68, 1.1920928955078125e-07);  div_68 = None
	        div_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_34, clamp_min_102);  minimum_34 = None
	        round_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_69);  div_69 = None
	        sub_1578: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_69);  round_69 = None
	        clamp_min_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1578, -128);  sub_1578 = None
	        clamp_max_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_103, 127);  clamp_min_103 = None
	        _assert_tensor_metadata_308 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_102, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_308 = None
	        _assert_tensor_metadata_309 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_68, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_309 = None
	        convert_element_type_204: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_68, torch.int8);  clamp_max_68 = None
	        view_535: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_5242, [sym_size_int, 1500, 1280]);  add_5242 = None
	        view_536: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_102, [sym_size_int, 1500, 1])
	        view_537: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_204, [sym_size_int, 1500, 1])
	        reciprocal_34: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_536);  view_536 = None
	        mul_3364: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_34, 1.0);  reciprocal_34 = None
	        mul_3367: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_535, mul_3364);  view_535 = mul_3364 = None
	        round_70: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3367);  mul_3367 = None
	        add_5329: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_70, view_537);  round_70 = view_537 = None
	        clamp_min_104: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5329, -128);  add_5329 = None
	        clamp_max_69: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_104, 127);  clamp_min_104 = None
	        view_538: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_69, [sym_size_int, 1500, 1280]);  clamp_max_69 = None
	        _assert_tensor_metadata_310 = torch.ops.aten._assert_tensor_metadata.default(view_538, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_310 = None
	        convert_element_type_205: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_538, torch.int8);  view_538 = None
	        view_539: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_205, [sym_size_int, 1500, 1280]);  convert_element_type_205 = None
	        view_540: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_102, [sym_size_int, 1500, 1]);  clamp_min_102 = None
	        view_541: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_204, [sym_size_int, 1500, 1]);  convert_element_type_204 = None
	        _assert_tensor_metadata_311 = torch.ops.aten._assert_tensor_metadata.default(view_539, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_311 = None
	        convert_element_type_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_539, torch.float32);  view_539 = None
	        _assert_tensor_metadata_312 = torch.ops.aten._assert_tensor_metadata.default(view_541, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_312 = None
	        convert_element_type_207: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_541, torch.float32);  view_541 = None
	        sub_1598: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_206, convert_element_type_207);  convert_element_type_206 = convert_element_type_207 = None
	        mul_3389: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1598, view_540);  sub_1598 = view_540 = None
	        view_542: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3389, [sym_size_int, 1500, 1280]);  mul_3389 = None
	        _assert_tensor_metadata_313 = torch.ops.aten._assert_tensor_metadata.default(view_542, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_313 = None
	        view_543: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg160_1, [5120, 40, 32]);  arg160_1 = None
	        view_544: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg161_1, [5120, 40, 1]);  arg161_1 = None
	        view_545: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg162_1, [5120, 40, 1]);  arg162_1 = None
	        _assert_tensor_metadata_314 = torch.ops.aten._assert_tensor_metadata.default(view_543, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_314 = None
	        convert_element_type_208: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_543, torch.float32);  view_543 = None
	        _assert_tensor_metadata_315 = torch.ops.aten._assert_tensor_metadata.default(view_545, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_315 = None
	        convert_element_type_209: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_545, torch.float32);  view_545 = None
	        sub_1602: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_208, convert_element_type_209);  convert_element_type_208 = convert_element_type_209 = None
	        mul_3394: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1602, view_544);  sub_1602 = view_544 = None
	        view_546: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3394, [5120, 1280]);  mul_3394 = None
	        _assert_tensor_metadata_316 = torch.ops.aten._assert_tensor_metadata.default(view_546, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_316 = None
	        mul_3399: "Sym(1500*s6)" = sym_size_int * 1500
	        view_547: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_542, [mul_3399, 1280]);  view_542 = mul_3399 = None
	        permute_59: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_546, [1, 0]);  view_546 = None
	        addmm_28: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg159_1, view_547, permute_59);  arg159_1 = view_547 = permute_59 = None
	        view_548: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_28, [sym_size_int, 1500, 5120]);  addmm_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_3406: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_548, 0.5)
	        mul_3407: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_548, 0.7071067811865476);  view_548 = None
	        erf_7: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_3407);  mul_3407 = None
	        add_5388: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_7, 1);  erf_7 = None
	        mul_3408: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3406, add_5388);  mul_3406 = add_5388 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_47: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_3408);  mul_3408 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_549: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_47, [sym_size_int, 1500, 5120])
	        amin_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_549, [2])
	        amax_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_549, [2]);  view_549 = None
	        full_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_35, full_70);  amin_35 = full_70 = None
	        full_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_35, full_71);  amax_35 = full_71 = None
	        sub_1615: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_35, minimum_35);  maximum_35 = None
	        div_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1615, 255.0);  sub_1615 = None
	        clamp_min_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_70, 1.1920928955078125e-07);  div_70 = None
	        div_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_35, clamp_min_105);  minimum_35 = None
	        round_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_71);  div_71 = None
	        sub_1621: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_71);  round_71 = None
	        clamp_min_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1621, -128);  sub_1621 = None
	        clamp_max_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_106, 127);  clamp_min_106 = None
	        _assert_tensor_metadata_317 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_105, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_317 = None
	        _assert_tensor_metadata_318 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_70, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_318 = None
	        convert_element_type_210: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_70, torch.int8);  clamp_max_70 = None
	        view_550: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_47, [sym_size_int, 1500, 5120]);  clone_47 = None
	        view_551: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_105, [sym_size_int, 1500, 1])
	        view_552: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_210, [sym_size_int, 1500, 1])
	        reciprocal_35: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_551);  view_551 = None
	        mul_3454: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_35, 1.0);  reciprocal_35 = None
	        mul_3457: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_550, mul_3454);  view_550 = mul_3454 = None
	        round_72: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_3457);  mul_3457 = None
	        add_5471: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_72, view_552);  round_72 = view_552 = None
	        clamp_min_107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5471, -128);  add_5471 = None
	        clamp_max_71: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_107, 127);  clamp_min_107 = None
	        view_553: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_71, [sym_size_int, 1500, 5120]);  clamp_max_71 = None
	        _assert_tensor_metadata_319 = torch.ops.aten._assert_tensor_metadata.default(view_553, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_319 = None
	        convert_element_type_211: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_553, torch.int8);  view_553 = None
	        view_554: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_211, [sym_size_int, 1500, 5120]);  convert_element_type_211 = None
	        view_555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_105, [sym_size_int, 1500, 1]);  clamp_min_105 = None
	        view_556: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_210, [sym_size_int, 1500, 1]);  convert_element_type_210 = None
	        _assert_tensor_metadata_320 = torch.ops.aten._assert_tensor_metadata.default(view_554, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_320 = None
	        convert_element_type_212: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_554, torch.float32);  view_554 = None
	        _assert_tensor_metadata_321 = torch.ops.aten._assert_tensor_metadata.default(view_556, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_321 = None
	        convert_element_type_213: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_556, torch.float32);  view_556 = None
	        sub_1641: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_212, convert_element_type_213);  convert_element_type_212 = convert_element_type_213 = None
	        mul_3479: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1641, view_555);  sub_1641 = view_555 = None
	        view_557: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_3479, [sym_size_int, 1500, 5120]);  mul_3479 = None
	        _assert_tensor_metadata_322 = torch.ops.aten._assert_tensor_metadata.default(view_557, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_322 = None
	        view_558: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg164_1, [1280, 160, 32]);  arg164_1 = None
	        view_559: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg165_1, [1280, 160, 1]);  arg165_1 = None
	        view_560: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg166_1, [1280, 160, 1]);  arg166_1 = None
	        _assert_tensor_metadata_323 = torch.ops.aten._assert_tensor_metadata.default(view_558, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_323 = None
	        convert_element_type_214: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_558, torch.float32);  view_558 = None
	        _assert_tensor_metadata_324 = torch.ops.aten._assert_tensor_metadata.default(view_560, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_324 = None
	        convert_element_type_215: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_560, torch.float32);  view_560 = None
	        sub_1645: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_214, convert_element_type_215);  convert_element_type_214 = convert_element_type_215 = None
	        mul_3484: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1645, view_559);  sub_1645 = view_559 = None
	        view_561: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_3484, [1280, 5120]);  mul_3484 = None
	        _assert_tensor_metadata_325 = torch.ops.aten._assert_tensor_metadata.default(view_561, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_325 = None
	        mul_3489: "Sym(1500*s6)" = sym_size_int * 1500
	        view_562: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_557, [mul_3489, 5120]);  view_557 = mul_3489 = None
	        permute_60: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_561, [1, 0]);  view_561 = None
	        addmm_29: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg163_1, view_562, permute_60);  arg163_1 = view_562 = permute_60 = None
	        view_563: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_29, [sym_size_int, 1500, 1280]);  addmm_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_48: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_563);  view_563 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_5534: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_5236, clone_48);  add_5236 = clone_48 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_49: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_5534, memory_format = torch.contiguous_format)
	        var_mean_12 = torch.ops.aten.var_mean.correction(clone_49, [2], correction = 0, keepdim = True)
	        getitem_48: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_12[0]
	        getitem_49: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_12[1];  var_mean_12 = None
	        add_5539: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_48, 1e-05);  getitem_48 = None
	        rsqrt_12: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_5539);  add_5539 = None
	        sub_1651: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_49, getitem_49);  clone_49 = getitem_49 = None
	        mul_3500: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1651, rsqrt_12);  sub_1651 = rsqrt_12 = None
	        mul_3501: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3500, arg167_1);  mul_3500 = arg167_1 = None
	        add_5540: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_3501, arg168_1);  mul_3501 = arg168_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_564: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_5540, [sym_size_int, 1500, 1280])
	        amin_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_564, [2])
	        amax_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_564, [2]);  view_564 = None
	        full_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_36, full_72);  amin_36 = full_72 = None
	        full_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_36, full_73);  amax_36 = full_73 = None
	        sub_1662: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_36, minimum_36);  maximum_36 = None
	        div_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1662, 255.0);  sub_1662 = None
	        clamp_min_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_72, 1.1920928955078125e-07);  div_72 = None
	        div_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_36, clamp_min_108);  minimum_36 = None
	        round_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_73);  div_73 = None
	        sub_1668: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_73);  round_73 = None
	        clamp_min_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1668, -128);  sub_1668 = None
	        clamp_max_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_109, 127);  clamp_min_109 = None
	        _assert_tensor_metadata_326 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_108, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_326 = None
	        _assert_tensor_metadata_327 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_72, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_327 = None
	        convert_element_type_216: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_72, torch.int8);  clamp_max_72 = None
	        view_565: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_5540, [sym_size_int, 1500, 1280])
	        view_566: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_108, [sym_size_int, 1500, 1])
	        view_567: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_216, [sym_size_int, 1500, 1])
	        reciprocal_36: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_566);  view_566 = None
	        mul_3549: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_36, 1.0);  reciprocal_36 = None
	        mul_3552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_565, mul_3549);  view_565 = mul_3549 = None
	        round_74: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3552);  mul_3552 = None
	        add_5627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_74, view_567);  round_74 = view_567 = None
	        clamp_min_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5627, -128);  add_5627 = None
	        clamp_max_73: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_110, 127);  clamp_min_110 = None
	        view_568: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_73, [sym_size_int, 1500, 1280]);  clamp_max_73 = None
	        _assert_tensor_metadata_328 = torch.ops.aten._assert_tensor_metadata.default(view_568, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_328 = None
	        convert_element_type_217: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_568, torch.int8);  view_568 = None
	        view_569: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_217, [sym_size_int, 1500, 1280]);  convert_element_type_217 = None
	        view_570: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_108, [sym_size_int, 1500, 1]);  clamp_min_108 = None
	        view_571: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_216, [sym_size_int, 1500, 1]);  convert_element_type_216 = None
	        _assert_tensor_metadata_329 = torch.ops.aten._assert_tensor_metadata.default(view_569, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_329 = None
	        convert_element_type_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_569, torch.float32);  view_569 = None
	        _assert_tensor_metadata_330 = torch.ops.aten._assert_tensor_metadata.default(view_571, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_330 = None
	        convert_element_type_219: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_571, torch.float32);  view_571 = None
	        sub_1688: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_218, convert_element_type_219);  convert_element_type_218 = convert_element_type_219 = None
	        mul_3574: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1688, view_570);  sub_1688 = view_570 = None
	        view_572: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3574, [sym_size_int, 1500, 1280]);  mul_3574 = None
	        _assert_tensor_metadata_331 = torch.ops.aten._assert_tensor_metadata.default(view_572, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_331 = None
	        view_573: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg170_1, [1280, 40, 32]);  arg170_1 = None
	        view_574: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg171_1, [1280, 40, 1]);  arg171_1 = None
	        view_575: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg172_1, [1280, 40, 1]);  arg172_1 = None
	        _assert_tensor_metadata_332 = torch.ops.aten._assert_tensor_metadata.default(view_573, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_332 = None
	        convert_element_type_220: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_573, torch.float32);  view_573 = None
	        _assert_tensor_metadata_333 = torch.ops.aten._assert_tensor_metadata.default(view_575, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_333 = None
	        convert_element_type_221: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_575, torch.float32);  view_575 = None
	        sub_1692: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_220, convert_element_type_221);  convert_element_type_220 = convert_element_type_221 = None
	        mul_3579: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1692, view_574);  sub_1692 = view_574 = None
	        view_576: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3579, [1280, 1280]);  mul_3579 = None
	        _assert_tensor_metadata_334 = torch.ops.aten._assert_tensor_metadata.default(view_576, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_334 = None
	        mul_3584: "Sym(1500*s6)" = sym_size_int * 1500
	        view_577: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_572, [mul_3584, 1280]);  view_572 = mul_3584 = None
	        permute_61: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_576, [1, 0]);  view_576 = None
	        addmm_30: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg169_1, view_577, permute_61);  arg169_1 = view_577 = permute_61 = None
	        view_578: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_30, [sym_size_int, 1500, 1280]);  addmm_30 = None
	        mul_3591: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_578, 0.125);  view_578 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_579: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_3591, [sym_size_int, 1500, 20, 64]);  mul_3591 = None
	        permute_62: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_579, [0, 2, 1, 3]);  view_579 = None
	        clone_50: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_62, memory_format = torch.contiguous_format);  permute_62 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_580: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_5540, [sym_size_int, 1500, 1280])
	        amin_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_580, [2])
	        amax_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_580, [2]);  view_580 = None
	        full_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_37, full_74);  amin_37 = full_74 = None
	        full_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_37, full_75);  amax_37 = full_75 = None
	        sub_1707: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_37, minimum_37);  maximum_37 = None
	        div_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1707, 255.0);  sub_1707 = None
	        clamp_min_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_74, 1.1920928955078125e-07);  div_74 = None
	        div_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_37, clamp_min_111);  minimum_37 = None
	        round_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_75);  div_75 = None
	        sub_1713: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_75);  round_75 = None
	        clamp_min_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1713, -128);  sub_1713 = None
	        clamp_max_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_112, 127);  clamp_min_112 = None
	        _assert_tensor_metadata_335 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_111, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_335 = None
	        _assert_tensor_metadata_336 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_74, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_336 = None
	        convert_element_type_222: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_74, torch.int8);  clamp_max_74 = None
	        view_581: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_5540, [sym_size_int, 1500, 1280])
	        view_582: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_111, [sym_size_int, 1500, 1])
	        view_583: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_222, [sym_size_int, 1500, 1])
	        reciprocal_37: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_582);  view_582 = None
	        mul_3645: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_37, 1.0);  reciprocal_37 = None
	        mul_3648: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_581, mul_3645);  view_581 = mul_3645 = None
	        round_76: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3648);  mul_3648 = None
	        add_5779: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_76, view_583);  round_76 = view_583 = None
	        clamp_min_113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5779, -128);  add_5779 = None
	        clamp_max_75: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_113, 127);  clamp_min_113 = None
	        view_584: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_75, [sym_size_int, 1500, 1280]);  clamp_max_75 = None
	        _assert_tensor_metadata_337 = torch.ops.aten._assert_tensor_metadata.default(view_584, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_337 = None
	        convert_element_type_223: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_584, torch.int8);  view_584 = None
	        view_585: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_223, [sym_size_int, 1500, 1280]);  convert_element_type_223 = None
	        view_586: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_111, [sym_size_int, 1500, 1]);  clamp_min_111 = None
	        view_587: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_222, [sym_size_int, 1500, 1]);  convert_element_type_222 = None
	        _assert_tensor_metadata_338 = torch.ops.aten._assert_tensor_metadata.default(view_585, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_338 = None
	        convert_element_type_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_585, torch.float32);  view_585 = None
	        _assert_tensor_metadata_339 = torch.ops.aten._assert_tensor_metadata.default(view_587, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_339 = None
	        convert_element_type_225: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_587, torch.float32);  view_587 = None
	        sub_1733: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_224, convert_element_type_225);  convert_element_type_224 = convert_element_type_225 = None
	        mul_3670: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1733, view_586);  sub_1733 = view_586 = None
	        view_588: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3670, [sym_size_int, 1500, 1280]);  mul_3670 = None
	        _assert_tensor_metadata_340 = torch.ops.aten._assert_tensor_metadata.default(view_588, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_340 = None
	        view_589: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg173_1, [1280, 40, 32]);  arg173_1 = None
	        view_590: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg174_1, [1280, 40, 1]);  arg174_1 = None
	        view_591: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg175_1, [1280, 40, 1]);  arg175_1 = None
	        _assert_tensor_metadata_341 = torch.ops.aten._assert_tensor_metadata.default(view_589, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_341 = None
	        convert_element_type_226: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_589, torch.float32);  view_589 = None
	        _assert_tensor_metadata_342 = torch.ops.aten._assert_tensor_metadata.default(view_591, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_342 = None
	        convert_element_type_227: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_591, torch.float32);  view_591 = None
	        sub_1737: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_226, convert_element_type_227);  convert_element_type_226 = convert_element_type_227 = None
	        mul_3675: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1737, view_590);  sub_1737 = view_590 = None
	        view_592: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3675, [1280, 1280]);  mul_3675 = None
	        _assert_tensor_metadata_343 = torch.ops.aten._assert_tensor_metadata.default(view_592, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_343 = None
	        permute_63: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_592, [1, 0]);  view_592 = None
	        mul_3678: "Sym(1500*s6)" = sym_size_int * 1500
	        view_593: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_588, [mul_3678, 1280]);  view_588 = mul_3678 = None
	        mm_6: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_593, permute_63);  view_593 = permute_63 = None
	        view_594: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_6, [sym_size_int, 1500, 1280]);  mm_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_595: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_594, [sym_size_int, -1, 20, 64]);  view_594 = None
	        permute_64: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_595, [0, 2, 1, 3]);  view_595 = None
	        clone_51: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_64, memory_format = torch.contiguous_format);  permute_64 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_596: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_5540, [sym_size_int, 1500, 1280])
	        amin_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_596, [2])
	        amax_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_596, [2]);  view_596 = None
	        full_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_38, full_76);  amin_38 = full_76 = None
	        full_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_38, full_77);  amax_38 = full_77 = None
	        sub_1751: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_38, minimum_38);  maximum_38 = None
	        div_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1751, 255.0);  sub_1751 = None
	        clamp_min_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_76, 1.1920928955078125e-07);  div_76 = None
	        div_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_38, clamp_min_114);  minimum_38 = None
	        round_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_77);  div_77 = None
	        sub_1757: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_77);  round_77 = None
	        clamp_min_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1757, -128);  sub_1757 = None
	        clamp_max_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_115, 127);  clamp_min_115 = None
	        _assert_tensor_metadata_344 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_114, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_344 = None
	        _assert_tensor_metadata_345 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_76, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_345 = None
	        convert_element_type_228: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_76, torch.int8);  clamp_max_76 = None
	        view_597: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_5540, [sym_size_int, 1500, 1280]);  add_5540 = None
	        view_598: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_114, [sym_size_int, 1500, 1])
	        view_599: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_228, [sym_size_int, 1500, 1])
	        reciprocal_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_598);  view_598 = None
	        mul_3744: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_38, 1.0);  reciprocal_38 = None
	        mul_3747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_597, mul_3744);  view_597 = mul_3744 = None
	        round_78: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3747);  mul_3747 = None
	        add_5927: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_78, view_599);  round_78 = view_599 = None
	        clamp_min_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5927, -128);  add_5927 = None
	        clamp_max_77: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_116, 127);  clamp_min_116 = None
	        view_600: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_77, [sym_size_int, 1500, 1280]);  clamp_max_77 = None
	        _assert_tensor_metadata_346 = torch.ops.aten._assert_tensor_metadata.default(view_600, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_346 = None
	        convert_element_type_229: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_600, torch.int8);  view_600 = None
	        view_601: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_229, [sym_size_int, 1500, 1280]);  convert_element_type_229 = None
	        view_602: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_114, [sym_size_int, 1500, 1]);  clamp_min_114 = None
	        view_603: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_228, [sym_size_int, 1500, 1]);  convert_element_type_228 = None
	        _assert_tensor_metadata_347 = torch.ops.aten._assert_tensor_metadata.default(view_601, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_347 = None
	        convert_element_type_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_601, torch.float32);  view_601 = None
	        _assert_tensor_metadata_348 = torch.ops.aten._assert_tensor_metadata.default(view_603, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_348 = None
	        convert_element_type_231: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_603, torch.float32);  view_603 = None
	        sub_1777: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_230, convert_element_type_231);  convert_element_type_230 = convert_element_type_231 = None
	        mul_3769: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1777, view_602);  sub_1777 = view_602 = None
	        view_604: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3769, [sym_size_int, 1500, 1280]);  mul_3769 = None
	        _assert_tensor_metadata_349 = torch.ops.aten._assert_tensor_metadata.default(view_604, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_349 = None
	        view_605: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg177_1, [1280, 40, 32]);  arg177_1 = None
	        view_606: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg178_1, [1280, 40, 1]);  arg178_1 = None
	        view_607: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg179_1, [1280, 40, 1]);  arg179_1 = None
	        _assert_tensor_metadata_350 = torch.ops.aten._assert_tensor_metadata.default(view_605, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_350 = None
	        convert_element_type_232: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_605, torch.float32);  view_605 = None
	        _assert_tensor_metadata_351 = torch.ops.aten._assert_tensor_metadata.default(view_607, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_351 = None
	        convert_element_type_233: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_607, torch.float32);  view_607 = None
	        sub_1781: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_232, convert_element_type_233);  convert_element_type_232 = convert_element_type_233 = None
	        mul_3774: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1781, view_606);  sub_1781 = view_606 = None
	        view_608: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3774, [1280, 1280]);  mul_3774 = None
	        _assert_tensor_metadata_352 = torch.ops.aten._assert_tensor_metadata.default(view_608, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_352 = None
	        mul_3779: "Sym(1500*s6)" = sym_size_int * 1500
	        view_609: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_604, [mul_3779, 1280]);  view_604 = mul_3779 = None
	        permute_65: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_608, [1, 0]);  view_608 = None
	        addmm_31: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg176_1, view_609, permute_65);  arg176_1 = view_609 = permute_65 = None
	        view_610: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_31, [sym_size_int, 1500, 1280]);  addmm_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_611: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_610, [sym_size_int, -1, 20, 64]);  view_610 = None
	        permute_66: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_611, [0, 2, 1, 3]);  view_611 = None
	        clone_52: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_66, memory_format = torch.contiguous_format);  permute_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_6 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_50, clone_51, clone_52, None, False, scale = 1.0);  clone_50 = clone_51 = clone_52 = None
	        getitem_50: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_6[0];  _scaled_dot_product_efficient_attention_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_67: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_50, [0, 2, 1, 3]);  getitem_50 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_67, [sym_size_int, 1500, -1]);  permute_67 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_613: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_612, [sym_size_int, 1500, 1280])
	        amin_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_613, [2])
	        amax_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_613, [2]);  view_613 = None
	        full_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_39, full_78);  amin_39 = full_78 = None
	        full_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_39, full_79);  amax_39 = full_79 = None
	        sub_1799: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_39, minimum_39);  maximum_39 = None
	        div_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1799, 255.0);  sub_1799 = None
	        clamp_min_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_78, 1.1920928955078125e-07);  div_78 = None
	        div_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_39, clamp_min_117);  minimum_39 = None
	        round_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_79);  div_79 = None
	        sub_1805: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_79);  round_79 = None
	        clamp_min_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1805, -128);  sub_1805 = None
	        clamp_max_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_118, 127);  clamp_min_118 = None
	        _assert_tensor_metadata_353 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_117, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_353 = None
	        _assert_tensor_metadata_354 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_78, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_354 = None
	        convert_element_type_234: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_78, torch.int8);  clamp_max_78 = None
	        view_614: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_612, [sym_size_int, 1500, 1280]);  view_612 = None
	        view_615: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_117, [sym_size_int, 1500, 1])
	        view_616: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_234, [sym_size_int, 1500, 1])
	        reciprocal_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_615);  view_615 = None
	        mul_3849: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_39, 1.0);  reciprocal_39 = None
	        mul_3852: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_614, mul_3849);  view_614 = mul_3849 = None
	        round_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3852);  mul_3852 = None
	        add_6091: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_80, view_616);  round_80 = view_616 = None
	        clamp_min_119: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6091, -128);  add_6091 = None
	        clamp_max_79: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_119, 127);  clamp_min_119 = None
	        view_617: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_79, [sym_size_int, 1500, 1280]);  clamp_max_79 = None
	        _assert_tensor_metadata_355 = torch.ops.aten._assert_tensor_metadata.default(view_617, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_355 = None
	        convert_element_type_235: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_617, torch.int8);  view_617 = None
	        view_618: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_235, [sym_size_int, 1500, 1280]);  convert_element_type_235 = None
	        view_619: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_117, [sym_size_int, 1500, 1]);  clamp_min_117 = None
	        view_620: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_234, [sym_size_int, 1500, 1]);  convert_element_type_234 = None
	        _assert_tensor_metadata_356 = torch.ops.aten._assert_tensor_metadata.default(view_618, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_356 = None
	        convert_element_type_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_618, torch.float32);  view_618 = None
	        _assert_tensor_metadata_357 = torch.ops.aten._assert_tensor_metadata.default(view_620, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_357 = None
	        convert_element_type_237: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_620, torch.float32);  view_620 = None
	        sub_1825: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_236, convert_element_type_237);  convert_element_type_236 = convert_element_type_237 = None
	        mul_3874: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1825, view_619);  sub_1825 = view_619 = None
	        view_621: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3874, [sym_size_int, 1500, 1280]);  mul_3874 = None
	        _assert_tensor_metadata_358 = torch.ops.aten._assert_tensor_metadata.default(view_621, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_358 = None
	        view_622: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg181_1, [1280, 40, 32]);  arg181_1 = None
	        view_623: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg182_1, [1280, 40, 1]);  arg182_1 = None
	        view_624: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg183_1, [1280, 40, 1]);  arg183_1 = None
	        _assert_tensor_metadata_359 = torch.ops.aten._assert_tensor_metadata.default(view_622, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_359 = None
	        convert_element_type_238: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_622, torch.float32);  view_622 = None
	        _assert_tensor_metadata_360 = torch.ops.aten._assert_tensor_metadata.default(view_624, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_360 = None
	        convert_element_type_239: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_624, torch.float32);  view_624 = None
	        sub_1829: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_238, convert_element_type_239);  convert_element_type_238 = convert_element_type_239 = None
	        mul_3879: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1829, view_623);  sub_1829 = view_623 = None
	        view_625: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3879, [1280, 1280]);  mul_3879 = None
	        _assert_tensor_metadata_361 = torch.ops.aten._assert_tensor_metadata.default(view_625, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_361 = None
	        mul_3884: "Sym(1500*s6)" = sym_size_int * 1500
	        view_626: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_621, [mul_3884, 1280]);  view_621 = mul_3884 = None
	        permute_68: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_625, [1, 0]);  view_625 = None
	        addmm_32: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg180_1, view_626, permute_68);  arg180_1 = view_626 = permute_68 = None
	        view_627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_32, [sym_size_int, 1500, 1280]);  addmm_32 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_53: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_627);  view_627 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_6154: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_5534, clone_53);  add_5534 = clone_53 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_54: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_6154, memory_format = torch.contiguous_format)
	        var_mean_13 = torch.ops.aten.var_mean.correction(clone_54, [2], correction = 0, keepdim = True)
	        getitem_54: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_13[0]
	        getitem_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_13[1];  var_mean_13 = None
	        add_6159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_54, 1e-05);  getitem_54 = None
	        rsqrt_13: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_6159);  add_6159 = None
	        sub_1835: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_54, getitem_55);  clone_54 = getitem_55 = None
	        mul_3895: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1835, rsqrt_13);  sub_1835 = rsqrt_13 = None
	        mul_3896: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3895, arg184_1);  mul_3895 = arg184_1 = None
	        add_6160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_3896, arg185_1);  mul_3896 = arg185_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_628: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_6160, [sym_size_int, 1500, 1280])
	        amin_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_628, [2])
	        amax_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_628, [2]);  view_628 = None
	        full_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_40, full_80);  amin_40 = full_80 = None
	        full_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_40, full_81);  amax_40 = full_81 = None
	        sub_1846: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_40, minimum_40);  maximum_40 = None
	        div_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1846, 255.0);  sub_1846 = None
	        clamp_min_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_80, 1.1920928955078125e-07);  div_80 = None
	        div_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_40, clamp_min_120);  minimum_40 = None
	        round_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_81);  div_81 = None
	        sub_1852: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_81);  round_81 = None
	        clamp_min_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1852, -128);  sub_1852 = None
	        clamp_max_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_121, 127);  clamp_min_121 = None
	        _assert_tensor_metadata_362 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_120, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_362 = None
	        _assert_tensor_metadata_363 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_80, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_363 = None
	        convert_element_type_240: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_80, torch.int8);  clamp_max_80 = None
	        view_629: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_6160, [sym_size_int, 1500, 1280]);  add_6160 = None
	        view_630: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_120, [sym_size_int, 1500, 1])
	        view_631: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_240, [sym_size_int, 1500, 1])
	        reciprocal_40: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_630);  view_630 = None
	        mul_3944: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_40, 1.0);  reciprocal_40 = None
	        mul_3947: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_629, mul_3944);  view_629 = mul_3944 = None
	        round_82: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3947);  mul_3947 = None
	        add_6247: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_82, view_631);  round_82 = view_631 = None
	        clamp_min_122: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6247, -128);  add_6247 = None
	        clamp_max_81: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_122, 127);  clamp_min_122 = None
	        view_632: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_81, [sym_size_int, 1500, 1280]);  clamp_max_81 = None
	        _assert_tensor_metadata_364 = torch.ops.aten._assert_tensor_metadata.default(view_632, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_364 = None
	        convert_element_type_241: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_632, torch.int8);  view_632 = None
	        view_633: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_241, [sym_size_int, 1500, 1280]);  convert_element_type_241 = None
	        view_634: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_120, [sym_size_int, 1500, 1]);  clamp_min_120 = None
	        view_635: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_240, [sym_size_int, 1500, 1]);  convert_element_type_240 = None
	        _assert_tensor_metadata_365 = torch.ops.aten._assert_tensor_metadata.default(view_633, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_365 = None
	        convert_element_type_242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_633, torch.float32);  view_633 = None
	        _assert_tensor_metadata_366 = torch.ops.aten._assert_tensor_metadata.default(view_635, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_366 = None
	        convert_element_type_243: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_635, torch.float32);  view_635 = None
	        sub_1872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_242, convert_element_type_243);  convert_element_type_242 = convert_element_type_243 = None
	        mul_3969: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1872, view_634);  sub_1872 = view_634 = None
	        view_636: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3969, [sym_size_int, 1500, 1280]);  mul_3969 = None
	        _assert_tensor_metadata_367 = torch.ops.aten._assert_tensor_metadata.default(view_636, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_367 = None
	        view_637: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg187_1, [5120, 40, 32]);  arg187_1 = None
	        view_638: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg188_1, [5120, 40, 1]);  arg188_1 = None
	        view_639: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg189_1, [5120, 40, 1]);  arg189_1 = None
	        _assert_tensor_metadata_368 = torch.ops.aten._assert_tensor_metadata.default(view_637, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_368 = None
	        convert_element_type_244: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_637, torch.float32);  view_637 = None
	        _assert_tensor_metadata_369 = torch.ops.aten._assert_tensor_metadata.default(view_639, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_369 = None
	        convert_element_type_245: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_639, torch.float32);  view_639 = None
	        sub_1876: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_244, convert_element_type_245);  convert_element_type_244 = convert_element_type_245 = None
	        mul_3974: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1876, view_638);  sub_1876 = view_638 = None
	        view_640: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3974, [5120, 1280]);  mul_3974 = None
	        _assert_tensor_metadata_370 = torch.ops.aten._assert_tensor_metadata.default(view_640, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_370 = None
	        mul_3979: "Sym(1500*s6)" = sym_size_int * 1500
	        view_641: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_636, [mul_3979, 1280]);  view_636 = mul_3979 = None
	        permute_69: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_640, [1, 0]);  view_640 = None
	        addmm_33: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg186_1, view_641, permute_69);  arg186_1 = view_641 = permute_69 = None
	        view_642: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_33, [sym_size_int, 1500, 5120]);  addmm_33 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_3986: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_642, 0.5)
	        mul_3987: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_642, 0.7071067811865476);  view_642 = None
	        erf_8: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_3987);  mul_3987 = None
	        add_6306: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_8, 1);  erf_8 = None
	        mul_3988: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3986, add_6306);  mul_3986 = add_6306 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_55: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_3988);  mul_3988 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_643: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_55, [sym_size_int, 1500, 5120])
	        amin_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_643, [2])
	        amax_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_643, [2]);  view_643 = None
	        full_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_41, full_82);  amin_41 = full_82 = None
	        full_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_41, full_83);  amax_41 = full_83 = None
	        sub_1889: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_41, minimum_41);  maximum_41 = None
	        div_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1889, 255.0);  sub_1889 = None
	        clamp_min_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_82, 1.1920928955078125e-07);  div_82 = None
	        div_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_41, clamp_min_123);  minimum_41 = None
	        round_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_83);  div_83 = None
	        sub_1895: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_83);  round_83 = None
	        clamp_min_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1895, -128);  sub_1895 = None
	        clamp_max_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_124, 127);  clamp_min_124 = None
	        _assert_tensor_metadata_371 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_123, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_371 = None
	        _assert_tensor_metadata_372 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_82, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_372 = None
	        convert_element_type_246: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_82, torch.int8);  clamp_max_82 = None
	        view_644: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_55, [sym_size_int, 1500, 5120]);  clone_55 = None
	        view_645: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_123, [sym_size_int, 1500, 1])
	        view_646: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_246, [sym_size_int, 1500, 1])
	        reciprocal_41: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_645);  view_645 = None
	        mul_4034: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_41, 1.0);  reciprocal_41 = None
	        mul_4037: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_644, mul_4034);  view_644 = mul_4034 = None
	        round_84: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_4037);  mul_4037 = None
	        add_6389: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_84, view_646);  round_84 = view_646 = None
	        clamp_min_125: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6389, -128);  add_6389 = None
	        clamp_max_83: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_125, 127);  clamp_min_125 = None
	        view_647: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_83, [sym_size_int, 1500, 5120]);  clamp_max_83 = None
	        _assert_tensor_metadata_373 = torch.ops.aten._assert_tensor_metadata.default(view_647, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_373 = None
	        convert_element_type_247: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_647, torch.int8);  view_647 = None
	        view_648: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_247, [sym_size_int, 1500, 5120]);  convert_element_type_247 = None
	        view_649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_123, [sym_size_int, 1500, 1]);  clamp_min_123 = None
	        view_650: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_246, [sym_size_int, 1500, 1]);  convert_element_type_246 = None
	        _assert_tensor_metadata_374 = torch.ops.aten._assert_tensor_metadata.default(view_648, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_374 = None
	        convert_element_type_248: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_648, torch.float32);  view_648 = None
	        _assert_tensor_metadata_375 = torch.ops.aten._assert_tensor_metadata.default(view_650, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_375 = None
	        convert_element_type_249: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_650, torch.float32);  view_650 = None
	        sub_1915: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_248, convert_element_type_249);  convert_element_type_248 = convert_element_type_249 = None
	        mul_4059: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1915, view_649);  sub_1915 = view_649 = None
	        view_651: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_4059, [sym_size_int, 1500, 5120]);  mul_4059 = None
	        _assert_tensor_metadata_376 = torch.ops.aten._assert_tensor_metadata.default(view_651, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_376 = None
	        view_652: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg191_1, [1280, 160, 32]);  arg191_1 = None
	        view_653: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg192_1, [1280, 160, 1]);  arg192_1 = None
	        view_654: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg193_1, [1280, 160, 1]);  arg193_1 = None
	        _assert_tensor_metadata_377 = torch.ops.aten._assert_tensor_metadata.default(view_652, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_377 = None
	        convert_element_type_250: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_652, torch.float32);  view_652 = None
	        _assert_tensor_metadata_378 = torch.ops.aten._assert_tensor_metadata.default(view_654, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_378 = None
	        convert_element_type_251: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_654, torch.float32);  view_654 = None
	        sub_1919: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_250, convert_element_type_251);  convert_element_type_250 = convert_element_type_251 = None
	        mul_4064: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1919, view_653);  sub_1919 = view_653 = None
	        view_655: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_4064, [1280, 5120]);  mul_4064 = None
	        _assert_tensor_metadata_379 = torch.ops.aten._assert_tensor_metadata.default(view_655, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_379 = None
	        mul_4069: "Sym(1500*s6)" = sym_size_int * 1500
	        view_656: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_651, [mul_4069, 5120]);  view_651 = mul_4069 = None
	        permute_70: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_655, [1, 0]);  view_655 = None
	        addmm_34: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg190_1, view_656, permute_70);  arg190_1 = view_656 = permute_70 = None
	        view_657: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_34, [sym_size_int, 1500, 1280]);  addmm_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_657);  view_657 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_6452: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_6154, clone_56);  add_6154 = clone_56 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_57: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_6452, memory_format = torch.contiguous_format)
	        var_mean_14 = torch.ops.aten.var_mean.correction(clone_57, [2], correction = 0, keepdim = True)
	        getitem_56: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_14[0]
	        getitem_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_14[1];  var_mean_14 = None
	        add_6457: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_56, 1e-05);  getitem_56 = None
	        rsqrt_14: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_6457);  add_6457 = None
	        sub_1925: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_57, getitem_57);  clone_57 = getitem_57 = None
	        mul_4080: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1925, rsqrt_14);  sub_1925 = rsqrt_14 = None
	        mul_4081: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4080, arg194_1);  mul_4080 = arg194_1 = None
	        add_6458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_4081, arg195_1);  mul_4081 = arg195_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_658: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_6458, [sym_size_int, 1500, 1280])
	        amin_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_658, [2])
	        amax_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_658, [2]);  view_658 = None
	        full_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_42, full_84);  amin_42 = full_84 = None
	        full_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_42, full_85);  amax_42 = full_85 = None
	        sub_1936: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_42, minimum_42);  maximum_42 = None
	        div_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1936, 255.0);  sub_1936 = None
	        clamp_min_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_84, 1.1920928955078125e-07);  div_84 = None
	        div_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_42, clamp_min_126);  minimum_42 = None
	        round_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_85);  div_85 = None
	        sub_1942: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_85);  round_85 = None
	        clamp_min_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1942, -128);  sub_1942 = None
	        clamp_max_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_127, 127);  clamp_min_127 = None
	        _assert_tensor_metadata_380 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_126, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_380 = None
	        _assert_tensor_metadata_381 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_84, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_381 = None
	        convert_element_type_252: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_84, torch.int8);  clamp_max_84 = None
	        view_659: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_6458, [sym_size_int, 1500, 1280])
	        view_660: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_126, [sym_size_int, 1500, 1])
	        view_661: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_252, [sym_size_int, 1500, 1])
	        reciprocal_42: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_660);  view_660 = None
	        mul_4129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_42, 1.0);  reciprocal_42 = None
	        mul_4132: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_659, mul_4129);  view_659 = mul_4129 = None
	        round_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4132);  mul_4132 = None
	        add_6545: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_86, view_661);  round_86 = view_661 = None
	        clamp_min_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6545, -128);  add_6545 = None
	        clamp_max_85: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_128, 127);  clamp_min_128 = None
	        view_662: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_85, [sym_size_int, 1500, 1280]);  clamp_max_85 = None
	        _assert_tensor_metadata_382 = torch.ops.aten._assert_tensor_metadata.default(view_662, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_382 = None
	        convert_element_type_253: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_662, torch.int8);  view_662 = None
	        view_663: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_253, [sym_size_int, 1500, 1280]);  convert_element_type_253 = None
	        view_664: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_126, [sym_size_int, 1500, 1]);  clamp_min_126 = None
	        view_665: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_252, [sym_size_int, 1500, 1]);  convert_element_type_252 = None
	        _assert_tensor_metadata_383 = torch.ops.aten._assert_tensor_metadata.default(view_663, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_383 = None
	        convert_element_type_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_663, torch.float32);  view_663 = None
	        _assert_tensor_metadata_384 = torch.ops.aten._assert_tensor_metadata.default(view_665, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_384 = None
	        convert_element_type_255: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_665, torch.float32);  view_665 = None
	        sub_1962: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_254, convert_element_type_255);  convert_element_type_254 = convert_element_type_255 = None
	        mul_4154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1962, view_664);  sub_1962 = view_664 = None
	        view_666: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4154, [sym_size_int, 1500, 1280]);  mul_4154 = None
	        _assert_tensor_metadata_385 = torch.ops.aten._assert_tensor_metadata.default(view_666, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_385 = None
	        view_667: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg197_1, [1280, 40, 32]);  arg197_1 = None
	        view_668: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg198_1, [1280, 40, 1]);  arg198_1 = None
	        view_669: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg199_1, [1280, 40, 1]);  arg199_1 = None
	        _assert_tensor_metadata_386 = torch.ops.aten._assert_tensor_metadata.default(view_667, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_386 = None
	        convert_element_type_256: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_667, torch.float32);  view_667 = None
	        _assert_tensor_metadata_387 = torch.ops.aten._assert_tensor_metadata.default(view_669, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_387 = None
	        convert_element_type_257: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_669, torch.float32);  view_669 = None
	        sub_1966: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_256, convert_element_type_257);  convert_element_type_256 = convert_element_type_257 = None
	        mul_4159: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1966, view_668);  sub_1966 = view_668 = None
	        view_670: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4159, [1280, 1280]);  mul_4159 = None
	        _assert_tensor_metadata_388 = torch.ops.aten._assert_tensor_metadata.default(view_670, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_388 = None
	        mul_4164: "Sym(1500*s6)" = sym_size_int * 1500
	        view_671: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_666, [mul_4164, 1280]);  view_666 = mul_4164 = None
	        permute_71: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_670, [1, 0]);  view_670 = None
	        addmm_35: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg196_1, view_671, permute_71);  arg196_1 = view_671 = permute_71 = None
	        view_672: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_35, [sym_size_int, 1500, 1280]);  addmm_35 = None
	        mul_4171: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_672, 0.125);  view_672 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_673: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_4171, [sym_size_int, 1500, 20, 64]);  mul_4171 = None
	        permute_72: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_673, [0, 2, 1, 3]);  view_673 = None
	        clone_58: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_72, memory_format = torch.contiguous_format);  permute_72 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_674: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_6458, [sym_size_int, 1500, 1280])
	        amin_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_674, [2])
	        amax_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_674, [2]);  view_674 = None
	        full_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_43, full_86);  amin_43 = full_86 = None
	        full_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_43, full_87);  amax_43 = full_87 = None
	        sub_1981: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_43, minimum_43);  maximum_43 = None
	        div_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1981, 255.0);  sub_1981 = None
	        clamp_min_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_86, 1.1920928955078125e-07);  div_86 = None
	        div_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_43, clamp_min_129);  minimum_43 = None
	        round_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_87);  div_87 = None
	        sub_1987: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_87);  round_87 = None
	        clamp_min_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1987, -128);  sub_1987 = None
	        clamp_max_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_130, 127);  clamp_min_130 = None
	        _assert_tensor_metadata_389 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_129, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_389 = None
	        _assert_tensor_metadata_390 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_86, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_390 = None
	        convert_element_type_258: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_86, torch.int8);  clamp_max_86 = None
	        view_675: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_6458, [sym_size_int, 1500, 1280])
	        view_676: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_129, [sym_size_int, 1500, 1])
	        view_677: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_258, [sym_size_int, 1500, 1])
	        reciprocal_43: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_676);  view_676 = None
	        mul_4225: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_43, 1.0);  reciprocal_43 = None
	        mul_4228: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_675, mul_4225);  view_675 = mul_4225 = None
	        round_88: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4228);  mul_4228 = None
	        add_6697: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_88, view_677);  round_88 = view_677 = None
	        clamp_min_131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6697, -128);  add_6697 = None
	        clamp_max_87: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_131, 127);  clamp_min_131 = None
	        view_678: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_87, [sym_size_int, 1500, 1280]);  clamp_max_87 = None
	        _assert_tensor_metadata_391 = torch.ops.aten._assert_tensor_metadata.default(view_678, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_391 = None
	        convert_element_type_259: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_678, torch.int8);  view_678 = None
	        view_679: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_259, [sym_size_int, 1500, 1280]);  convert_element_type_259 = None
	        view_680: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_129, [sym_size_int, 1500, 1]);  clamp_min_129 = None
	        view_681: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_258, [sym_size_int, 1500, 1]);  convert_element_type_258 = None
	        _assert_tensor_metadata_392 = torch.ops.aten._assert_tensor_metadata.default(view_679, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_392 = None
	        convert_element_type_260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_679, torch.float32);  view_679 = None
	        _assert_tensor_metadata_393 = torch.ops.aten._assert_tensor_metadata.default(view_681, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_393 = None
	        convert_element_type_261: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_681, torch.float32);  view_681 = None
	        sub_2007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_260, convert_element_type_261);  convert_element_type_260 = convert_element_type_261 = None
	        mul_4250: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2007, view_680);  sub_2007 = view_680 = None
	        view_682: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4250, [sym_size_int, 1500, 1280]);  mul_4250 = None
	        _assert_tensor_metadata_394 = torch.ops.aten._assert_tensor_metadata.default(view_682, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_394 = None
	        view_683: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg200_1, [1280, 40, 32]);  arg200_1 = None
	        view_684: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg201_1, [1280, 40, 1]);  arg201_1 = None
	        view_685: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg202_1, [1280, 40, 1]);  arg202_1 = None
	        _assert_tensor_metadata_395 = torch.ops.aten._assert_tensor_metadata.default(view_683, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_395 = None
	        convert_element_type_262: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_683, torch.float32);  view_683 = None
	        _assert_tensor_metadata_396 = torch.ops.aten._assert_tensor_metadata.default(view_685, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_396 = None
	        convert_element_type_263: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_685, torch.float32);  view_685 = None
	        sub_2011: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_262, convert_element_type_263);  convert_element_type_262 = convert_element_type_263 = None
	        mul_4255: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2011, view_684);  sub_2011 = view_684 = None
	        view_686: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4255, [1280, 1280]);  mul_4255 = None
	        _assert_tensor_metadata_397 = torch.ops.aten._assert_tensor_metadata.default(view_686, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_397 = None
	        permute_73: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_686, [1, 0]);  view_686 = None
	        mul_4258: "Sym(1500*s6)" = sym_size_int * 1500
	        view_687: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_682, [mul_4258, 1280]);  view_682 = mul_4258 = None
	        mm_7: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_687, permute_73);  view_687 = permute_73 = None
	        view_688: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_7, [sym_size_int, 1500, 1280]);  mm_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_689: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_688, [sym_size_int, -1, 20, 64]);  view_688 = None
	        permute_74: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_689, [0, 2, 1, 3]);  view_689 = None
	        clone_59: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_74, memory_format = torch.contiguous_format);  permute_74 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_690: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_6458, [sym_size_int, 1500, 1280])
	        amin_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_690, [2])
	        amax_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_690, [2]);  view_690 = None
	        full_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_44, full_88);  amin_44 = full_88 = None
	        full_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_44, full_89);  amax_44 = full_89 = None
	        sub_2025: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_44, minimum_44);  maximum_44 = None
	        div_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2025, 255.0);  sub_2025 = None
	        clamp_min_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_88, 1.1920928955078125e-07);  div_88 = None
	        div_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_44, clamp_min_132);  minimum_44 = None
	        round_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_89);  div_89 = None
	        sub_2031: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_89);  round_89 = None
	        clamp_min_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2031, -128);  sub_2031 = None
	        clamp_max_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_133, 127);  clamp_min_133 = None
	        _assert_tensor_metadata_398 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_132, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_398 = None
	        _assert_tensor_metadata_399 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_88, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_399 = None
	        convert_element_type_264: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_88, torch.int8);  clamp_max_88 = None
	        view_691: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_6458, [sym_size_int, 1500, 1280]);  add_6458 = None
	        view_692: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_132, [sym_size_int, 1500, 1])
	        view_693: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_264, [sym_size_int, 1500, 1])
	        reciprocal_44: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_692);  view_692 = None
	        mul_4324: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_44, 1.0);  reciprocal_44 = None
	        mul_4327: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_691, mul_4324);  view_691 = mul_4324 = None
	        round_90: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4327);  mul_4327 = None
	        add_6845: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_90, view_693);  round_90 = view_693 = None
	        clamp_min_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6845, -128);  add_6845 = None
	        clamp_max_89: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_134, 127);  clamp_min_134 = None
	        view_694: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_89, [sym_size_int, 1500, 1280]);  clamp_max_89 = None
	        _assert_tensor_metadata_400 = torch.ops.aten._assert_tensor_metadata.default(view_694, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_400 = None
	        convert_element_type_265: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_694, torch.int8);  view_694 = None
	        view_695: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_265, [sym_size_int, 1500, 1280]);  convert_element_type_265 = None
	        view_696: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_132, [sym_size_int, 1500, 1]);  clamp_min_132 = None
	        view_697: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_264, [sym_size_int, 1500, 1]);  convert_element_type_264 = None
	        _assert_tensor_metadata_401 = torch.ops.aten._assert_tensor_metadata.default(view_695, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_401 = None
	        convert_element_type_266: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_695, torch.float32);  view_695 = None
	        _assert_tensor_metadata_402 = torch.ops.aten._assert_tensor_metadata.default(view_697, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_402 = None
	        convert_element_type_267: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_697, torch.float32);  view_697 = None
	        sub_2051: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_266, convert_element_type_267);  convert_element_type_266 = convert_element_type_267 = None
	        mul_4349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2051, view_696);  sub_2051 = view_696 = None
	        view_698: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4349, [sym_size_int, 1500, 1280]);  mul_4349 = None
	        _assert_tensor_metadata_403 = torch.ops.aten._assert_tensor_metadata.default(view_698, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_403 = None
	        view_699: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg204_1, [1280, 40, 32]);  arg204_1 = None
	        view_700: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg205_1, [1280, 40, 1]);  arg205_1 = None
	        view_701: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg206_1, [1280, 40, 1]);  arg206_1 = None
	        _assert_tensor_metadata_404 = torch.ops.aten._assert_tensor_metadata.default(view_699, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_404 = None
	        convert_element_type_268: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_699, torch.float32);  view_699 = None
	        _assert_tensor_metadata_405 = torch.ops.aten._assert_tensor_metadata.default(view_701, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_405 = None
	        convert_element_type_269: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_701, torch.float32);  view_701 = None
	        sub_2055: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_268, convert_element_type_269);  convert_element_type_268 = convert_element_type_269 = None
	        mul_4354: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2055, view_700);  sub_2055 = view_700 = None
	        view_702: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4354, [1280, 1280]);  mul_4354 = None
	        _assert_tensor_metadata_406 = torch.ops.aten._assert_tensor_metadata.default(view_702, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_406 = None
	        mul_4359: "Sym(1500*s6)" = sym_size_int * 1500
	        view_703: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_698, [mul_4359, 1280]);  view_698 = mul_4359 = None
	        permute_75: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_702, [1, 0]);  view_702 = None
	        addmm_36: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg203_1, view_703, permute_75);  arg203_1 = view_703 = permute_75 = None
	        view_704: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_36, [sym_size_int, 1500, 1280]);  addmm_36 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_705: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_704, [sym_size_int, -1, 20, 64]);  view_704 = None
	        permute_76: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_705, [0, 2, 1, 3]);  view_705 = None
	        clone_60: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_76, memory_format = torch.contiguous_format);  permute_76 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_7 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_58, clone_59, clone_60, None, False, scale = 1.0);  clone_58 = clone_59 = clone_60 = None
	        getitem_58: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_7[0];  _scaled_dot_product_efficient_attention_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_77: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_58, [0, 2, 1, 3]);  getitem_58 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_706: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_77, [sym_size_int, 1500, -1]);  permute_77 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_707: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_706, [sym_size_int, 1500, 1280])
	        amin_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_707, [2])
	        amax_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_707, [2]);  view_707 = None
	        full_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_45, full_90);  amin_45 = full_90 = None
	        full_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_45, full_91);  amax_45 = full_91 = None
	        sub_2073: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_45, minimum_45);  maximum_45 = None
	        div_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2073, 255.0);  sub_2073 = None
	        clamp_min_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_90, 1.1920928955078125e-07);  div_90 = None
	        div_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_45, clamp_min_135);  minimum_45 = None
	        round_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_91);  div_91 = None
	        sub_2079: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_91);  round_91 = None
	        clamp_min_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2079, -128);  sub_2079 = None
	        clamp_max_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_136, 127);  clamp_min_136 = None
	        _assert_tensor_metadata_407 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_135, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_407 = None
	        _assert_tensor_metadata_408 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_90, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_408 = None
	        convert_element_type_270: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_90, torch.int8);  clamp_max_90 = None
	        view_708: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_706, [sym_size_int, 1500, 1280]);  view_706 = None
	        view_709: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_135, [sym_size_int, 1500, 1])
	        view_710: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_270, [sym_size_int, 1500, 1])
	        reciprocal_45: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_709);  view_709 = None
	        mul_4429: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_45, 1.0);  reciprocal_45 = None
	        mul_4432: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_708, mul_4429);  view_708 = mul_4429 = None
	        round_92: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4432);  mul_4432 = None
	        add_7009: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_92, view_710);  round_92 = view_710 = None
	        clamp_min_137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7009, -128);  add_7009 = None
	        clamp_max_91: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_137, 127);  clamp_min_137 = None
	        view_711: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_91, [sym_size_int, 1500, 1280]);  clamp_max_91 = None
	        _assert_tensor_metadata_409 = torch.ops.aten._assert_tensor_metadata.default(view_711, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_409 = None
	        convert_element_type_271: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_711, torch.int8);  view_711 = None
	        view_712: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_271, [sym_size_int, 1500, 1280]);  convert_element_type_271 = None
	        view_713: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_135, [sym_size_int, 1500, 1]);  clamp_min_135 = None
	        view_714: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_270, [sym_size_int, 1500, 1]);  convert_element_type_270 = None
	        _assert_tensor_metadata_410 = torch.ops.aten._assert_tensor_metadata.default(view_712, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_410 = None
	        convert_element_type_272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_712, torch.float32);  view_712 = None
	        _assert_tensor_metadata_411 = torch.ops.aten._assert_tensor_metadata.default(view_714, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_411 = None
	        convert_element_type_273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_714, torch.float32);  view_714 = None
	        sub_2099: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_272, convert_element_type_273);  convert_element_type_272 = convert_element_type_273 = None
	        mul_4454: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2099, view_713);  sub_2099 = view_713 = None
	        view_715: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4454, [sym_size_int, 1500, 1280]);  mul_4454 = None
	        _assert_tensor_metadata_412 = torch.ops.aten._assert_tensor_metadata.default(view_715, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_412 = None
	        view_716: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg208_1, [1280, 40, 32]);  arg208_1 = None
	        view_717: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg209_1, [1280, 40, 1]);  arg209_1 = None
	        view_718: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg210_1, [1280, 40, 1]);  arg210_1 = None
	        _assert_tensor_metadata_413 = torch.ops.aten._assert_tensor_metadata.default(view_716, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_413 = None
	        convert_element_type_274: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_716, torch.float32);  view_716 = None
	        _assert_tensor_metadata_414 = torch.ops.aten._assert_tensor_metadata.default(view_718, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_414 = None
	        convert_element_type_275: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_718, torch.float32);  view_718 = None
	        sub_2103: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_274, convert_element_type_275);  convert_element_type_274 = convert_element_type_275 = None
	        mul_4459: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2103, view_717);  sub_2103 = view_717 = None
	        view_719: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4459, [1280, 1280]);  mul_4459 = None
	        _assert_tensor_metadata_415 = torch.ops.aten._assert_tensor_metadata.default(view_719, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_415 = None
	        mul_4464: "Sym(1500*s6)" = sym_size_int * 1500
	        view_720: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_715, [mul_4464, 1280]);  view_715 = mul_4464 = None
	        permute_78: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_719, [1, 0]);  view_719 = None
	        addmm_37: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg207_1, view_720, permute_78);  arg207_1 = view_720 = permute_78 = None
	        view_721: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_37, [sym_size_int, 1500, 1280]);  addmm_37 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_61: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_721);  view_721 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_7072: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_6452, clone_61);  add_6452 = clone_61 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_7072, memory_format = torch.contiguous_format)
	        var_mean_15 = torch.ops.aten.var_mean.correction(clone_62, [2], correction = 0, keepdim = True)
	        getitem_62: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_15[0]
	        getitem_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_15[1];  var_mean_15 = None
	        add_7077: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_62, 1e-05);  getitem_62 = None
	        rsqrt_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_7077);  add_7077 = None
	        sub_2109: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_62, getitem_63);  clone_62 = getitem_63 = None
	        mul_4475: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2109, rsqrt_15);  sub_2109 = rsqrt_15 = None
	        mul_4476: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4475, arg211_1);  mul_4475 = arg211_1 = None
	        add_7078: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_4476, arg212_1);  mul_4476 = arg212_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_722: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7078, [sym_size_int, 1500, 1280])
	        amin_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_722, [2])
	        amax_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_722, [2]);  view_722 = None
	        full_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_46, full_92);  amin_46 = full_92 = None
	        full_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_46, full_93);  amax_46 = full_93 = None
	        sub_2120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_46, minimum_46);  maximum_46 = None
	        div_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2120, 255.0);  sub_2120 = None
	        clamp_min_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_92, 1.1920928955078125e-07);  div_92 = None
	        div_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_46, clamp_min_138);  minimum_46 = None
	        round_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_93);  div_93 = None
	        sub_2126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_93);  round_93 = None
	        clamp_min_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2126, -128);  sub_2126 = None
	        clamp_max_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_139, 127);  clamp_min_139 = None
	        _assert_tensor_metadata_416 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_138, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_416 = None
	        _assert_tensor_metadata_417 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_92, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_417 = None
	        convert_element_type_276: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_92, torch.int8);  clamp_max_92 = None
	        view_723: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7078, [sym_size_int, 1500, 1280]);  add_7078 = None
	        view_724: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_138, [sym_size_int, 1500, 1])
	        view_725: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_276, [sym_size_int, 1500, 1])
	        reciprocal_46: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_724);  view_724 = None
	        mul_4524: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_46, 1.0);  reciprocal_46 = None
	        mul_4527: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_723, mul_4524);  view_723 = mul_4524 = None
	        round_94: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4527);  mul_4527 = None
	        add_7165: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_94, view_725);  round_94 = view_725 = None
	        clamp_min_140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7165, -128);  add_7165 = None
	        clamp_max_93: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_140, 127);  clamp_min_140 = None
	        view_726: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_93, [sym_size_int, 1500, 1280]);  clamp_max_93 = None
	        _assert_tensor_metadata_418 = torch.ops.aten._assert_tensor_metadata.default(view_726, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_418 = None
	        convert_element_type_277: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_726, torch.int8);  view_726 = None
	        view_727: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_277, [sym_size_int, 1500, 1280]);  convert_element_type_277 = None
	        view_728: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_138, [sym_size_int, 1500, 1]);  clamp_min_138 = None
	        view_729: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_276, [sym_size_int, 1500, 1]);  convert_element_type_276 = None
	        _assert_tensor_metadata_419 = torch.ops.aten._assert_tensor_metadata.default(view_727, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_419 = None
	        convert_element_type_278: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_727, torch.float32);  view_727 = None
	        _assert_tensor_metadata_420 = torch.ops.aten._assert_tensor_metadata.default(view_729, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_420 = None
	        convert_element_type_279: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_729, torch.float32);  view_729 = None
	        sub_2146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_278, convert_element_type_279);  convert_element_type_278 = convert_element_type_279 = None
	        mul_4549: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2146, view_728);  sub_2146 = view_728 = None
	        view_730: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4549, [sym_size_int, 1500, 1280]);  mul_4549 = None
	        _assert_tensor_metadata_421 = torch.ops.aten._assert_tensor_metadata.default(view_730, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_421 = None
	        view_731: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg214_1, [5120, 40, 32]);  arg214_1 = None
	        view_732: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg215_1, [5120, 40, 1]);  arg215_1 = None
	        view_733: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg216_1, [5120, 40, 1]);  arg216_1 = None
	        _assert_tensor_metadata_422 = torch.ops.aten._assert_tensor_metadata.default(view_731, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_422 = None
	        convert_element_type_280: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_731, torch.float32);  view_731 = None
	        _assert_tensor_metadata_423 = torch.ops.aten._assert_tensor_metadata.default(view_733, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_423 = None
	        convert_element_type_281: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_733, torch.float32);  view_733 = None
	        sub_2150: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_280, convert_element_type_281);  convert_element_type_280 = convert_element_type_281 = None
	        mul_4554: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2150, view_732);  sub_2150 = view_732 = None
	        view_734: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4554, [5120, 1280]);  mul_4554 = None
	        _assert_tensor_metadata_424 = torch.ops.aten._assert_tensor_metadata.default(view_734, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_424 = None
	        mul_4559: "Sym(1500*s6)" = sym_size_int * 1500
	        view_735: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_730, [mul_4559, 1280]);  view_730 = mul_4559 = None
	        permute_79: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_734, [1, 0]);  view_734 = None
	        addmm_38: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg213_1, view_735, permute_79);  arg213_1 = view_735 = permute_79 = None
	        view_736: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_38, [sym_size_int, 1500, 5120]);  addmm_38 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_4566: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_736, 0.5)
	        mul_4567: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_736, 0.7071067811865476);  view_736 = None
	        erf_9: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_4567);  mul_4567 = None
	        add_7224: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_9, 1);  erf_9 = None
	        mul_4568: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4566, add_7224);  mul_4566 = add_7224 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_63: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_4568);  mul_4568 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_737: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_63, [sym_size_int, 1500, 5120])
	        amin_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_737, [2])
	        amax_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_737, [2]);  view_737 = None
	        full_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_47, full_94);  amin_47 = full_94 = None
	        full_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_47, full_95);  amax_47 = full_95 = None
	        sub_2163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_47, minimum_47);  maximum_47 = None
	        div_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2163, 255.0);  sub_2163 = None
	        clamp_min_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_94, 1.1920928955078125e-07);  div_94 = None
	        div_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_47, clamp_min_141);  minimum_47 = None
	        round_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_95);  div_95 = None
	        sub_2169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_95);  round_95 = None
	        clamp_min_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2169, -128);  sub_2169 = None
	        clamp_max_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_142, 127);  clamp_min_142 = None
	        _assert_tensor_metadata_425 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_141, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_425 = None
	        _assert_tensor_metadata_426 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_94, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_426 = None
	        convert_element_type_282: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_94, torch.int8);  clamp_max_94 = None
	        view_738: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_63, [sym_size_int, 1500, 5120]);  clone_63 = None
	        view_739: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_141, [sym_size_int, 1500, 1])
	        view_740: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_282, [sym_size_int, 1500, 1])
	        reciprocal_47: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_739);  view_739 = None
	        mul_4614: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_47, 1.0);  reciprocal_47 = None
	        mul_4617: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_738, mul_4614);  view_738 = mul_4614 = None
	        round_96: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_4617);  mul_4617 = None
	        add_7307: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_96, view_740);  round_96 = view_740 = None
	        clamp_min_143: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7307, -128);  add_7307 = None
	        clamp_max_95: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_143, 127);  clamp_min_143 = None
	        view_741: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_95, [sym_size_int, 1500, 5120]);  clamp_max_95 = None
	        _assert_tensor_metadata_427 = torch.ops.aten._assert_tensor_metadata.default(view_741, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_427 = None
	        convert_element_type_283: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_741, torch.int8);  view_741 = None
	        view_742: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_283, [sym_size_int, 1500, 5120]);  convert_element_type_283 = None
	        view_743: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_141, [sym_size_int, 1500, 1]);  clamp_min_141 = None
	        view_744: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_282, [sym_size_int, 1500, 1]);  convert_element_type_282 = None
	        _assert_tensor_metadata_428 = torch.ops.aten._assert_tensor_metadata.default(view_742, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_428 = None
	        convert_element_type_284: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_742, torch.float32);  view_742 = None
	        _assert_tensor_metadata_429 = torch.ops.aten._assert_tensor_metadata.default(view_744, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_429 = None
	        convert_element_type_285: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_744, torch.float32);  view_744 = None
	        sub_2189: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_284, convert_element_type_285);  convert_element_type_284 = convert_element_type_285 = None
	        mul_4639: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2189, view_743);  sub_2189 = view_743 = None
	        view_745: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_4639, [sym_size_int, 1500, 5120]);  mul_4639 = None
	        _assert_tensor_metadata_430 = torch.ops.aten._assert_tensor_metadata.default(view_745, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_430 = None
	        view_746: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg218_1, [1280, 160, 32]);  arg218_1 = None
	        view_747: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg219_1, [1280, 160, 1]);  arg219_1 = None
	        view_748: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg220_1, [1280, 160, 1]);  arg220_1 = None
	        _assert_tensor_metadata_431 = torch.ops.aten._assert_tensor_metadata.default(view_746, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_431 = None
	        convert_element_type_286: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_746, torch.float32);  view_746 = None
	        _assert_tensor_metadata_432 = torch.ops.aten._assert_tensor_metadata.default(view_748, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_432 = None
	        convert_element_type_287: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_748, torch.float32);  view_748 = None
	        sub_2193: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_286, convert_element_type_287);  convert_element_type_286 = convert_element_type_287 = None
	        mul_4644: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2193, view_747);  sub_2193 = view_747 = None
	        view_749: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_4644, [1280, 5120]);  mul_4644 = None
	        _assert_tensor_metadata_433 = torch.ops.aten._assert_tensor_metadata.default(view_749, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_433 = None
	        mul_4649: "Sym(1500*s6)" = sym_size_int * 1500
	        view_750: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_745, [mul_4649, 5120]);  view_745 = mul_4649 = None
	        permute_80: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_749, [1, 0]);  view_749 = None
	        addmm_39: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg217_1, view_750, permute_80);  arg217_1 = view_750 = permute_80 = None
	        view_751: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_39, [sym_size_int, 1500, 1280]);  addmm_39 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_64: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_751);  view_751 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_7370: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_7072, clone_64);  add_7072 = clone_64 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_65: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_7370, memory_format = torch.contiguous_format)
	        var_mean_16 = torch.ops.aten.var_mean.correction(clone_65, [2], correction = 0, keepdim = True)
	        getitem_64: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_16[0]
	        getitem_65: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_16[1];  var_mean_16 = None
	        add_7375: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_64, 1e-05);  getitem_64 = None
	        rsqrt_16: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_7375);  add_7375 = None
	        sub_2199: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_65, getitem_65);  clone_65 = getitem_65 = None
	        mul_4660: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2199, rsqrt_16);  sub_2199 = rsqrt_16 = None
	        mul_4661: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4660, arg221_1);  mul_4660 = arg221_1 = None
	        add_7376: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_4661, arg222_1);  mul_4661 = arg222_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_752: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7376, [sym_size_int, 1500, 1280])
	        amin_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_752, [2])
	        amax_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_752, [2]);  view_752 = None
	        full_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_48, full_96);  amin_48 = full_96 = None
	        full_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_48, full_97);  amax_48 = full_97 = None
	        sub_2210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_48, minimum_48);  maximum_48 = None
	        div_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2210, 255.0);  sub_2210 = None
	        clamp_min_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_96, 1.1920928955078125e-07);  div_96 = None
	        div_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_48, clamp_min_144);  minimum_48 = None
	        round_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_97);  div_97 = None
	        sub_2216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_97);  round_97 = None
	        clamp_min_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2216, -128);  sub_2216 = None
	        clamp_max_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_145, 127);  clamp_min_145 = None
	        _assert_tensor_metadata_434 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_144, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_434 = None
	        _assert_tensor_metadata_435 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_96, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_435 = None
	        convert_element_type_288: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_96, torch.int8);  clamp_max_96 = None
	        view_753: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7376, [sym_size_int, 1500, 1280])
	        view_754: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_144, [sym_size_int, 1500, 1])
	        view_755: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_288, [sym_size_int, 1500, 1])
	        reciprocal_48: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_754);  view_754 = None
	        mul_4709: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_48, 1.0);  reciprocal_48 = None
	        mul_4712: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_753, mul_4709);  view_753 = mul_4709 = None
	        round_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4712);  mul_4712 = None
	        add_7463: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_98, view_755);  round_98 = view_755 = None
	        clamp_min_146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7463, -128);  add_7463 = None
	        clamp_max_97: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_146, 127);  clamp_min_146 = None
	        view_756: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_97, [sym_size_int, 1500, 1280]);  clamp_max_97 = None
	        _assert_tensor_metadata_436 = torch.ops.aten._assert_tensor_metadata.default(view_756, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_436 = None
	        convert_element_type_289: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_756, torch.int8);  view_756 = None
	        view_757: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_289, [sym_size_int, 1500, 1280]);  convert_element_type_289 = None
	        view_758: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_144, [sym_size_int, 1500, 1]);  clamp_min_144 = None
	        view_759: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_288, [sym_size_int, 1500, 1]);  convert_element_type_288 = None
	        _assert_tensor_metadata_437 = torch.ops.aten._assert_tensor_metadata.default(view_757, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_437 = None
	        convert_element_type_290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_757, torch.float32);  view_757 = None
	        _assert_tensor_metadata_438 = torch.ops.aten._assert_tensor_metadata.default(view_759, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_438 = None
	        convert_element_type_291: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_759, torch.float32);  view_759 = None
	        sub_2236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_290, convert_element_type_291);  convert_element_type_290 = convert_element_type_291 = None
	        mul_4734: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2236, view_758);  sub_2236 = view_758 = None
	        view_760: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4734, [sym_size_int, 1500, 1280]);  mul_4734 = None
	        _assert_tensor_metadata_439 = torch.ops.aten._assert_tensor_metadata.default(view_760, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_439 = None
	        view_761: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg224_1, [1280, 40, 32]);  arg224_1 = None
	        view_762: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg225_1, [1280, 40, 1]);  arg225_1 = None
	        view_763: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg226_1, [1280, 40, 1]);  arg226_1 = None
	        _assert_tensor_metadata_440 = torch.ops.aten._assert_tensor_metadata.default(view_761, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_440 = None
	        convert_element_type_292: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_761, torch.float32);  view_761 = None
	        _assert_tensor_metadata_441 = torch.ops.aten._assert_tensor_metadata.default(view_763, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_441 = None
	        convert_element_type_293: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_763, torch.float32);  view_763 = None
	        sub_2240: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_292, convert_element_type_293);  convert_element_type_292 = convert_element_type_293 = None
	        mul_4739: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2240, view_762);  sub_2240 = view_762 = None
	        view_764: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4739, [1280, 1280]);  mul_4739 = None
	        _assert_tensor_metadata_442 = torch.ops.aten._assert_tensor_metadata.default(view_764, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_442 = None
	        mul_4744: "Sym(1500*s6)" = sym_size_int * 1500
	        view_765: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_760, [mul_4744, 1280]);  view_760 = mul_4744 = None
	        permute_81: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_764, [1, 0]);  view_764 = None
	        addmm_40: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg223_1, view_765, permute_81);  arg223_1 = view_765 = permute_81 = None
	        view_766: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_40, [sym_size_int, 1500, 1280]);  addmm_40 = None
	        mul_4751: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_766, 0.125);  view_766 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_767: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_4751, [sym_size_int, 1500, 20, 64]);  mul_4751 = None
	        permute_82: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_767, [0, 2, 1, 3]);  view_767 = None
	        clone_66: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_82, memory_format = torch.contiguous_format);  permute_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_768: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7376, [sym_size_int, 1500, 1280])
	        amin_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_768, [2])
	        amax_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_768, [2]);  view_768 = None
	        full_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_49, full_98);  amin_49 = full_98 = None
	        full_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_49, full_99);  amax_49 = full_99 = None
	        sub_2255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_49, minimum_49);  maximum_49 = None
	        div_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2255, 255.0);  sub_2255 = None
	        clamp_min_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_98, 1.1920928955078125e-07);  div_98 = None
	        div_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_49, clamp_min_147);  minimum_49 = None
	        round_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_99);  div_99 = None
	        sub_2261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_99);  round_99 = None
	        clamp_min_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2261, -128);  sub_2261 = None
	        clamp_max_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_148, 127);  clamp_min_148 = None
	        _assert_tensor_metadata_443 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_147, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_443 = None
	        _assert_tensor_metadata_444 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_98, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_444 = None
	        convert_element_type_294: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_98, torch.int8);  clamp_max_98 = None
	        view_769: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7376, [sym_size_int, 1500, 1280])
	        view_770: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_147, [sym_size_int, 1500, 1])
	        view_771: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_294, [sym_size_int, 1500, 1])
	        reciprocal_49: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_770);  view_770 = None
	        mul_4805: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_49, 1.0);  reciprocal_49 = None
	        mul_4808: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_769, mul_4805);  view_769 = mul_4805 = None
	        round_100: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4808);  mul_4808 = None
	        add_7615: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_100, view_771);  round_100 = view_771 = None
	        clamp_min_149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7615, -128);  add_7615 = None
	        clamp_max_99: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_149, 127);  clamp_min_149 = None
	        view_772: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_99, [sym_size_int, 1500, 1280]);  clamp_max_99 = None
	        _assert_tensor_metadata_445 = torch.ops.aten._assert_tensor_metadata.default(view_772, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_445 = None
	        convert_element_type_295: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_772, torch.int8);  view_772 = None
	        view_773: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_295, [sym_size_int, 1500, 1280]);  convert_element_type_295 = None
	        view_774: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_147, [sym_size_int, 1500, 1]);  clamp_min_147 = None
	        view_775: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_294, [sym_size_int, 1500, 1]);  convert_element_type_294 = None
	        _assert_tensor_metadata_446 = torch.ops.aten._assert_tensor_metadata.default(view_773, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_446 = None
	        convert_element_type_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_773, torch.float32);  view_773 = None
	        _assert_tensor_metadata_447 = torch.ops.aten._assert_tensor_metadata.default(view_775, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_447 = None
	        convert_element_type_297: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_775, torch.float32);  view_775 = None
	        sub_2281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_296, convert_element_type_297);  convert_element_type_296 = convert_element_type_297 = None
	        mul_4830: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2281, view_774);  sub_2281 = view_774 = None
	        view_776: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4830, [sym_size_int, 1500, 1280]);  mul_4830 = None
	        _assert_tensor_metadata_448 = torch.ops.aten._assert_tensor_metadata.default(view_776, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_448 = None
	        view_777: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg227_1, [1280, 40, 32]);  arg227_1 = None
	        view_778: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg228_1, [1280, 40, 1]);  arg228_1 = None
	        view_779: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg229_1, [1280, 40, 1]);  arg229_1 = None
	        _assert_tensor_metadata_449 = torch.ops.aten._assert_tensor_metadata.default(view_777, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_449 = None
	        convert_element_type_298: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_777, torch.float32);  view_777 = None
	        _assert_tensor_metadata_450 = torch.ops.aten._assert_tensor_metadata.default(view_779, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_450 = None
	        convert_element_type_299: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_779, torch.float32);  view_779 = None
	        sub_2285: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_298, convert_element_type_299);  convert_element_type_298 = convert_element_type_299 = None
	        mul_4835: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2285, view_778);  sub_2285 = view_778 = None
	        view_780: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4835, [1280, 1280]);  mul_4835 = None
	        _assert_tensor_metadata_451 = torch.ops.aten._assert_tensor_metadata.default(view_780, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_451 = None
	        permute_83: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_780, [1, 0]);  view_780 = None
	        mul_4838: "Sym(1500*s6)" = sym_size_int * 1500
	        view_781: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_776, [mul_4838, 1280]);  view_776 = mul_4838 = None
	        mm_8: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_781, permute_83);  view_781 = permute_83 = None
	        view_782: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_8, [sym_size_int, 1500, 1280]);  mm_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_783: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_782, [sym_size_int, -1, 20, 64]);  view_782 = None
	        permute_84: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_783, [0, 2, 1, 3]);  view_783 = None
	        clone_67: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_84, memory_format = torch.contiguous_format);  permute_84 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_784: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7376, [sym_size_int, 1500, 1280])
	        amin_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_784, [2])
	        amax_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_784, [2]);  view_784 = None
	        full_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_50, full_100);  amin_50 = full_100 = None
	        full_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_50, full_101);  amax_50 = full_101 = None
	        sub_2299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_50, minimum_50);  maximum_50 = None
	        div_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2299, 255.0);  sub_2299 = None
	        clamp_min_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_100, 1.1920928955078125e-07);  div_100 = None
	        div_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_50, clamp_min_150);  minimum_50 = None
	        round_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_101);  div_101 = None
	        sub_2305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_101);  round_101 = None
	        clamp_min_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2305, -128);  sub_2305 = None
	        clamp_max_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_151, 127);  clamp_min_151 = None
	        _assert_tensor_metadata_452 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_150, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_452 = None
	        _assert_tensor_metadata_453 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_100, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_453 = None
	        convert_element_type_300: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_100, torch.int8);  clamp_max_100 = None
	        view_785: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7376, [sym_size_int, 1500, 1280]);  add_7376 = None
	        view_786: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_150, [sym_size_int, 1500, 1])
	        view_787: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_300, [sym_size_int, 1500, 1])
	        reciprocal_50: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_786);  view_786 = None
	        mul_4904: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_50, 1.0);  reciprocal_50 = None
	        mul_4907: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_785, mul_4904);  view_785 = mul_4904 = None
	        round_102: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4907);  mul_4907 = None
	        add_7763: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_102, view_787);  round_102 = view_787 = None
	        clamp_min_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7763, -128);  add_7763 = None
	        clamp_max_101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_152, 127);  clamp_min_152 = None
	        view_788: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_101, [sym_size_int, 1500, 1280]);  clamp_max_101 = None
	        _assert_tensor_metadata_454 = torch.ops.aten._assert_tensor_metadata.default(view_788, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_454 = None
	        convert_element_type_301: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_788, torch.int8);  view_788 = None
	        view_789: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_301, [sym_size_int, 1500, 1280]);  convert_element_type_301 = None
	        view_790: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_150, [sym_size_int, 1500, 1]);  clamp_min_150 = None
	        view_791: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_300, [sym_size_int, 1500, 1]);  convert_element_type_300 = None
	        _assert_tensor_metadata_455 = torch.ops.aten._assert_tensor_metadata.default(view_789, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_455 = None
	        convert_element_type_302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_789, torch.float32);  view_789 = None
	        _assert_tensor_metadata_456 = torch.ops.aten._assert_tensor_metadata.default(view_791, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_456 = None
	        convert_element_type_303: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_791, torch.float32);  view_791 = None
	        sub_2325: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_302, convert_element_type_303);  convert_element_type_302 = convert_element_type_303 = None
	        mul_4929: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2325, view_790);  sub_2325 = view_790 = None
	        view_792: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4929, [sym_size_int, 1500, 1280]);  mul_4929 = None
	        _assert_tensor_metadata_457 = torch.ops.aten._assert_tensor_metadata.default(view_792, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_457 = None
	        view_793: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg231_1, [1280, 40, 32]);  arg231_1 = None
	        view_794: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg232_1, [1280, 40, 1]);  arg232_1 = None
	        view_795: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg233_1, [1280, 40, 1]);  arg233_1 = None
	        _assert_tensor_metadata_458 = torch.ops.aten._assert_tensor_metadata.default(view_793, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_458 = None
	        convert_element_type_304: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_793, torch.float32);  view_793 = None
	        _assert_tensor_metadata_459 = torch.ops.aten._assert_tensor_metadata.default(view_795, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_459 = None
	        convert_element_type_305: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_795, torch.float32);  view_795 = None
	        sub_2329: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_304, convert_element_type_305);  convert_element_type_304 = convert_element_type_305 = None
	        mul_4934: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2329, view_794);  sub_2329 = view_794 = None
	        view_796: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4934, [1280, 1280]);  mul_4934 = None
	        _assert_tensor_metadata_460 = torch.ops.aten._assert_tensor_metadata.default(view_796, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_460 = None
	        mul_4939: "Sym(1500*s6)" = sym_size_int * 1500
	        view_797: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_792, [mul_4939, 1280]);  view_792 = mul_4939 = None
	        permute_85: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_796, [1, 0]);  view_796 = None
	        addmm_41: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg230_1, view_797, permute_85);  arg230_1 = view_797 = permute_85 = None
	        view_798: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_41, [sym_size_int, 1500, 1280]);  addmm_41 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_799: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_798, [sym_size_int, -1, 20, 64]);  view_798 = None
	        permute_86: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_799, [0, 2, 1, 3]);  view_799 = None
	        clone_68: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_86, memory_format = torch.contiguous_format);  permute_86 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_8 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_66, clone_67, clone_68, None, False, scale = 1.0);  clone_66 = clone_67 = clone_68 = None
	        getitem_66: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_8[0];  _scaled_dot_product_efficient_attention_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_87: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_66, [0, 2, 1, 3]);  getitem_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_800: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_87, [sym_size_int, 1500, -1]);  permute_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_801: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_800, [sym_size_int, 1500, 1280])
	        amin_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_801, [2])
	        amax_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_801, [2]);  view_801 = None
	        full_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_51, full_102);  amin_51 = full_102 = None
	        full_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_51, full_103);  amax_51 = full_103 = None
	        sub_2347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_51, minimum_51);  maximum_51 = None
	        div_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2347, 255.0);  sub_2347 = None
	        clamp_min_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_102, 1.1920928955078125e-07);  div_102 = None
	        div_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_51, clamp_min_153);  minimum_51 = None
	        round_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_103);  div_103 = None
	        sub_2353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_103);  round_103 = None
	        clamp_min_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2353, -128);  sub_2353 = None
	        clamp_max_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_154, 127);  clamp_min_154 = None
	        _assert_tensor_metadata_461 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_153, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_461 = None
	        _assert_tensor_metadata_462 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_102, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_462 = None
	        convert_element_type_306: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_102, torch.int8);  clamp_max_102 = None
	        view_802: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_800, [sym_size_int, 1500, 1280]);  view_800 = None
	        view_803: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_153, [sym_size_int, 1500, 1])
	        view_804: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_306, [sym_size_int, 1500, 1])
	        reciprocal_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_803);  view_803 = None
	        mul_5009: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_51, 1.0);  reciprocal_51 = None
	        mul_5012: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_802, mul_5009);  view_802 = mul_5009 = None
	        round_104: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5012);  mul_5012 = None
	        add_7927: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_104, view_804);  round_104 = view_804 = None
	        clamp_min_155: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7927, -128);  add_7927 = None
	        clamp_max_103: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_155, 127);  clamp_min_155 = None
	        view_805: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_103, [sym_size_int, 1500, 1280]);  clamp_max_103 = None
	        _assert_tensor_metadata_463 = torch.ops.aten._assert_tensor_metadata.default(view_805, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_463 = None
	        convert_element_type_307: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_805, torch.int8);  view_805 = None
	        view_806: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_307, [sym_size_int, 1500, 1280]);  convert_element_type_307 = None
	        view_807: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_153, [sym_size_int, 1500, 1]);  clamp_min_153 = None
	        view_808: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_306, [sym_size_int, 1500, 1]);  convert_element_type_306 = None
	        _assert_tensor_metadata_464 = torch.ops.aten._assert_tensor_metadata.default(view_806, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_464 = None
	        convert_element_type_308: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_806, torch.float32);  view_806 = None
	        _assert_tensor_metadata_465 = torch.ops.aten._assert_tensor_metadata.default(view_808, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_465 = None
	        convert_element_type_309: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_808, torch.float32);  view_808 = None
	        sub_2373: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_308, convert_element_type_309);  convert_element_type_308 = convert_element_type_309 = None
	        mul_5034: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2373, view_807);  sub_2373 = view_807 = None
	        view_809: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5034, [sym_size_int, 1500, 1280]);  mul_5034 = None
	        _assert_tensor_metadata_466 = torch.ops.aten._assert_tensor_metadata.default(view_809, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_466 = None
	        view_810: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg235_1, [1280, 40, 32]);  arg235_1 = None
	        view_811: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg236_1, [1280, 40, 1]);  arg236_1 = None
	        view_812: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg237_1, [1280, 40, 1]);  arg237_1 = None
	        _assert_tensor_metadata_467 = torch.ops.aten._assert_tensor_metadata.default(view_810, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_467 = None
	        convert_element_type_310: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_810, torch.float32);  view_810 = None
	        _assert_tensor_metadata_468 = torch.ops.aten._assert_tensor_metadata.default(view_812, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_468 = None
	        convert_element_type_311: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_812, torch.float32);  view_812 = None
	        sub_2377: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_310, convert_element_type_311);  convert_element_type_310 = convert_element_type_311 = None
	        mul_5039: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2377, view_811);  sub_2377 = view_811 = None
	        view_813: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5039, [1280, 1280]);  mul_5039 = None
	        _assert_tensor_metadata_469 = torch.ops.aten._assert_tensor_metadata.default(view_813, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_469 = None
	        mul_5044: "Sym(1500*s6)" = sym_size_int * 1500
	        view_814: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_809, [mul_5044, 1280]);  view_809 = mul_5044 = None
	        permute_88: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_813, [1, 0]);  view_813 = None
	        addmm_42: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg234_1, view_814, permute_88);  arg234_1 = view_814 = permute_88 = None
	        view_815: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_42, [sym_size_int, 1500, 1280]);  addmm_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_69: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_815);  view_815 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_7990: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_7370, clone_69);  add_7370 = clone_69 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_70: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_7990, memory_format = torch.contiguous_format)
	        var_mean_17 = torch.ops.aten.var_mean.correction(clone_70, [2], correction = 0, keepdim = True)
	        getitem_70: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_17[0]
	        getitem_71: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_17[1];  var_mean_17 = None
	        add_7995: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_70, 1e-05);  getitem_70 = None
	        rsqrt_17: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_7995);  add_7995 = None
	        sub_2383: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_70, getitem_71);  clone_70 = getitem_71 = None
	        mul_5055: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2383, rsqrt_17);  sub_2383 = rsqrt_17 = None
	        mul_5056: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5055, arg238_1);  mul_5055 = arg238_1 = None
	        add_7996: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5056, arg239_1);  mul_5056 = arg239_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_816: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7996, [sym_size_int, 1500, 1280])
	        amin_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_816, [2])
	        amax_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_816, [2]);  view_816 = None
	        full_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_52, full_104);  amin_52 = full_104 = None
	        full_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_52, full_105);  amax_52 = full_105 = None
	        sub_2394: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_52, minimum_52);  maximum_52 = None
	        div_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2394, 255.0);  sub_2394 = None
	        clamp_min_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_104, 1.1920928955078125e-07);  div_104 = None
	        div_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_52, clamp_min_156);  minimum_52 = None
	        round_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_105);  div_105 = None
	        sub_2400: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_105);  round_105 = None
	        clamp_min_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2400, -128);  sub_2400 = None
	        clamp_max_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_157, 127);  clamp_min_157 = None
	        _assert_tensor_metadata_470 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_156, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_470 = None
	        _assert_tensor_metadata_471 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_104, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_471 = None
	        convert_element_type_312: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_104, torch.int8);  clamp_max_104 = None
	        view_817: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7996, [sym_size_int, 1500, 1280]);  add_7996 = None
	        view_818: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_156, [sym_size_int, 1500, 1])
	        view_819: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_312, [sym_size_int, 1500, 1])
	        reciprocal_52: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_818);  view_818 = None
	        mul_5104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_52, 1.0);  reciprocal_52 = None
	        mul_5107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_817, mul_5104);  view_817 = mul_5104 = None
	        round_106: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5107);  mul_5107 = None
	        add_8083: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_106, view_819);  round_106 = view_819 = None
	        clamp_min_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8083, -128);  add_8083 = None
	        clamp_max_105: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_158, 127);  clamp_min_158 = None
	        view_820: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_105, [sym_size_int, 1500, 1280]);  clamp_max_105 = None
	        _assert_tensor_metadata_472 = torch.ops.aten._assert_tensor_metadata.default(view_820, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_472 = None
	        convert_element_type_313: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_820, torch.int8);  view_820 = None
	        view_821: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_313, [sym_size_int, 1500, 1280]);  convert_element_type_313 = None
	        view_822: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_156, [sym_size_int, 1500, 1]);  clamp_min_156 = None
	        view_823: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_312, [sym_size_int, 1500, 1]);  convert_element_type_312 = None
	        _assert_tensor_metadata_473 = torch.ops.aten._assert_tensor_metadata.default(view_821, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_473 = None
	        convert_element_type_314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_821, torch.float32);  view_821 = None
	        _assert_tensor_metadata_474 = torch.ops.aten._assert_tensor_metadata.default(view_823, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_474 = None
	        convert_element_type_315: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_823, torch.float32);  view_823 = None
	        sub_2420: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_314, convert_element_type_315);  convert_element_type_314 = convert_element_type_315 = None
	        mul_5129: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2420, view_822);  sub_2420 = view_822 = None
	        view_824: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5129, [sym_size_int, 1500, 1280]);  mul_5129 = None
	        _assert_tensor_metadata_475 = torch.ops.aten._assert_tensor_metadata.default(view_824, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_475 = None
	        view_825: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg241_1, [5120, 40, 32]);  arg241_1 = None
	        view_826: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg242_1, [5120, 40, 1]);  arg242_1 = None
	        view_827: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg243_1, [5120, 40, 1]);  arg243_1 = None
	        _assert_tensor_metadata_476 = torch.ops.aten._assert_tensor_metadata.default(view_825, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_476 = None
	        convert_element_type_316: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_825, torch.float32);  view_825 = None
	        _assert_tensor_metadata_477 = torch.ops.aten._assert_tensor_metadata.default(view_827, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_477 = None
	        convert_element_type_317: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_827, torch.float32);  view_827 = None
	        sub_2424: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_316, convert_element_type_317);  convert_element_type_316 = convert_element_type_317 = None
	        mul_5134: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2424, view_826);  sub_2424 = view_826 = None
	        view_828: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5134, [5120, 1280]);  mul_5134 = None
	        _assert_tensor_metadata_478 = torch.ops.aten._assert_tensor_metadata.default(view_828, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_478 = None
	        mul_5139: "Sym(1500*s6)" = sym_size_int * 1500
	        view_829: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_824, [mul_5139, 1280]);  view_824 = mul_5139 = None
	        permute_89: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_828, [1, 0]);  view_828 = None
	        addmm_43: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg240_1, view_829, permute_89);  arg240_1 = view_829 = permute_89 = None
	        view_830: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_43, [sym_size_int, 1500, 5120]);  addmm_43 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_5146: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_830, 0.5)
	        mul_5147: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_830, 0.7071067811865476);  view_830 = None
	        erf_10: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_5147);  mul_5147 = None
	        add_8142: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_10, 1);  erf_10 = None
	        mul_5148: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5146, add_8142);  mul_5146 = add_8142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_71: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_5148);  mul_5148 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_831: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_71, [sym_size_int, 1500, 5120])
	        amin_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_831, [2])
	        amax_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_831, [2]);  view_831 = None
	        full_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_53, full_106);  amin_53 = full_106 = None
	        full_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_53, full_107);  amax_53 = full_107 = None
	        sub_2437: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_53, minimum_53);  maximum_53 = None
	        div_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2437, 255.0);  sub_2437 = None
	        clamp_min_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_106, 1.1920928955078125e-07);  div_106 = None
	        div_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_53, clamp_min_159);  minimum_53 = None
	        round_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_107);  div_107 = None
	        sub_2443: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_107);  round_107 = None
	        clamp_min_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2443, -128);  sub_2443 = None
	        clamp_max_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_160, 127);  clamp_min_160 = None
	        _assert_tensor_metadata_479 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_159, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_479 = None
	        _assert_tensor_metadata_480 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_106, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_480 = None
	        convert_element_type_318: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_106, torch.int8);  clamp_max_106 = None
	        view_832: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_71, [sym_size_int, 1500, 5120]);  clone_71 = None
	        view_833: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_159, [sym_size_int, 1500, 1])
	        view_834: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_318, [sym_size_int, 1500, 1])
	        reciprocal_53: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_833);  view_833 = None
	        mul_5194: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_53, 1.0);  reciprocal_53 = None
	        mul_5197: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_832, mul_5194);  view_832 = mul_5194 = None
	        round_108: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_5197);  mul_5197 = None
	        add_8225: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_108, view_834);  round_108 = view_834 = None
	        clamp_min_161: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8225, -128);  add_8225 = None
	        clamp_max_107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_161, 127);  clamp_min_161 = None
	        view_835: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_107, [sym_size_int, 1500, 5120]);  clamp_max_107 = None
	        _assert_tensor_metadata_481 = torch.ops.aten._assert_tensor_metadata.default(view_835, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_481 = None
	        convert_element_type_319: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_835, torch.int8);  view_835 = None
	        view_836: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_319, [sym_size_int, 1500, 5120]);  convert_element_type_319 = None
	        view_837: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_159, [sym_size_int, 1500, 1]);  clamp_min_159 = None
	        view_838: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_318, [sym_size_int, 1500, 1]);  convert_element_type_318 = None
	        _assert_tensor_metadata_482 = torch.ops.aten._assert_tensor_metadata.default(view_836, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_482 = None
	        convert_element_type_320: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_836, torch.float32);  view_836 = None
	        _assert_tensor_metadata_483 = torch.ops.aten._assert_tensor_metadata.default(view_838, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_483 = None
	        convert_element_type_321: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_838, torch.float32);  view_838 = None
	        sub_2463: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_320, convert_element_type_321);  convert_element_type_320 = convert_element_type_321 = None
	        mul_5219: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2463, view_837);  sub_2463 = view_837 = None
	        view_839: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_5219, [sym_size_int, 1500, 5120]);  mul_5219 = None
	        _assert_tensor_metadata_484 = torch.ops.aten._assert_tensor_metadata.default(view_839, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_484 = None
	        view_840: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg245_1, [1280, 160, 32]);  arg245_1 = None
	        view_841: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg246_1, [1280, 160, 1]);  arg246_1 = None
	        view_842: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg247_1, [1280, 160, 1]);  arg247_1 = None
	        _assert_tensor_metadata_485 = torch.ops.aten._assert_tensor_metadata.default(view_840, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_485 = None
	        convert_element_type_322: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_840, torch.float32);  view_840 = None
	        _assert_tensor_metadata_486 = torch.ops.aten._assert_tensor_metadata.default(view_842, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_486 = None
	        convert_element_type_323: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_842, torch.float32);  view_842 = None
	        sub_2467: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_322, convert_element_type_323);  convert_element_type_322 = convert_element_type_323 = None
	        mul_5224: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2467, view_841);  sub_2467 = view_841 = None
	        view_843: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_5224, [1280, 5120]);  mul_5224 = None
	        _assert_tensor_metadata_487 = torch.ops.aten._assert_tensor_metadata.default(view_843, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_487 = None
	        mul_5229: "Sym(1500*s6)" = sym_size_int * 1500
	        view_844: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_839, [mul_5229, 5120]);  view_839 = mul_5229 = None
	        permute_90: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_843, [1, 0]);  view_843 = None
	        addmm_44: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg244_1, view_844, permute_90);  arg244_1 = view_844 = permute_90 = None
	        view_845: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_44, [sym_size_int, 1500, 1280]);  addmm_44 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_72: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_845);  view_845 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_8288: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_7990, clone_72);  add_7990 = clone_72 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_73: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_8288, memory_format = torch.contiguous_format)
	        var_mean_18 = torch.ops.aten.var_mean.correction(clone_73, [2], correction = 0, keepdim = True)
	        getitem_72: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_18[0]
	        getitem_73: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_18[1];  var_mean_18 = None
	        add_8293: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_72, 1e-05);  getitem_72 = None
	        rsqrt_18: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_8293);  add_8293 = None
	        sub_2473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_73, getitem_73);  clone_73 = getitem_73 = None
	        mul_5240: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2473, rsqrt_18);  sub_2473 = rsqrt_18 = None
	        mul_5241: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5240, arg248_1);  mul_5240 = arg248_1 = None
	        add_8294: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5241, arg249_1);  mul_5241 = arg249_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_846: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_8294, [sym_size_int, 1500, 1280])
	        amin_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_846, [2])
	        amax_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_846, [2]);  view_846 = None
	        full_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_54, full_108);  amin_54 = full_108 = None
	        full_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_54, full_109);  amax_54 = full_109 = None
	        sub_2484: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_54, minimum_54);  maximum_54 = None
	        div_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2484, 255.0);  sub_2484 = None
	        clamp_min_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_108, 1.1920928955078125e-07);  div_108 = None
	        div_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_54, clamp_min_162);  minimum_54 = None
	        round_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_109);  div_109 = None
	        sub_2490: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_109);  round_109 = None
	        clamp_min_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2490, -128);  sub_2490 = None
	        clamp_max_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_163, 127);  clamp_min_163 = None
	        _assert_tensor_metadata_488 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_162, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_488 = None
	        _assert_tensor_metadata_489 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_108, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_489 = None
	        convert_element_type_324: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_108, torch.int8);  clamp_max_108 = None
	        view_847: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_8294, [sym_size_int, 1500, 1280])
	        view_848: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_162, [sym_size_int, 1500, 1])
	        view_849: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_324, [sym_size_int, 1500, 1])
	        reciprocal_54: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_848);  view_848 = None
	        mul_5289: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_54, 1.0);  reciprocal_54 = None
	        mul_5292: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_847, mul_5289);  view_847 = mul_5289 = None
	        round_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5292);  mul_5292 = None
	        add_8381: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_110, view_849);  round_110 = view_849 = None
	        clamp_min_164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8381, -128);  add_8381 = None
	        clamp_max_109: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_164, 127);  clamp_min_164 = None
	        view_850: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_109, [sym_size_int, 1500, 1280]);  clamp_max_109 = None
	        _assert_tensor_metadata_490 = torch.ops.aten._assert_tensor_metadata.default(view_850, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_490 = None
	        convert_element_type_325: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_850, torch.int8);  view_850 = None
	        view_851: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_325, [sym_size_int, 1500, 1280]);  convert_element_type_325 = None
	        view_852: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_162, [sym_size_int, 1500, 1]);  clamp_min_162 = None
	        view_853: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_324, [sym_size_int, 1500, 1]);  convert_element_type_324 = None
	        _assert_tensor_metadata_491 = torch.ops.aten._assert_tensor_metadata.default(view_851, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_491 = None
	        convert_element_type_326: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_851, torch.float32);  view_851 = None
	        _assert_tensor_metadata_492 = torch.ops.aten._assert_tensor_metadata.default(view_853, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_492 = None
	        convert_element_type_327: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_853, torch.float32);  view_853 = None
	        sub_2510: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_326, convert_element_type_327);  convert_element_type_326 = convert_element_type_327 = None
	        mul_5314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2510, view_852);  sub_2510 = view_852 = None
	        view_854: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5314, [sym_size_int, 1500, 1280]);  mul_5314 = None
	        _assert_tensor_metadata_493 = torch.ops.aten._assert_tensor_metadata.default(view_854, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_493 = None
	        view_855: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg251_1, [1280, 40, 32]);  arg251_1 = None
	        view_856: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg252_1, [1280, 40, 1]);  arg252_1 = None
	        view_857: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg253_1, [1280, 40, 1]);  arg253_1 = None
	        _assert_tensor_metadata_494 = torch.ops.aten._assert_tensor_metadata.default(view_855, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_494 = None
	        convert_element_type_328: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_855, torch.float32);  view_855 = None
	        _assert_tensor_metadata_495 = torch.ops.aten._assert_tensor_metadata.default(view_857, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_495 = None
	        convert_element_type_329: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_857, torch.float32);  view_857 = None
	        sub_2514: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_328, convert_element_type_329);  convert_element_type_328 = convert_element_type_329 = None
	        mul_5319: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2514, view_856);  sub_2514 = view_856 = None
	        view_858: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5319, [1280, 1280]);  mul_5319 = None
	        _assert_tensor_metadata_496 = torch.ops.aten._assert_tensor_metadata.default(view_858, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_496 = None
	        mul_5324: "Sym(1500*s6)" = sym_size_int * 1500
	        view_859: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_854, [mul_5324, 1280]);  view_854 = mul_5324 = None
	        permute_91: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_858, [1, 0]);  view_858 = None
	        addmm_45: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg250_1, view_859, permute_91);  arg250_1 = view_859 = permute_91 = None
	        view_860: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_45, [sym_size_int, 1500, 1280]);  addmm_45 = None
	        mul_5331: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_860, 0.125);  view_860 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_861: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_5331, [sym_size_int, 1500, 20, 64]);  mul_5331 = None
	        permute_92: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_861, [0, 2, 1, 3]);  view_861 = None
	        clone_74: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_92, memory_format = torch.contiguous_format);  permute_92 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_862: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_8294, [sym_size_int, 1500, 1280])
	        amin_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_862, [2])
	        amax_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_862, [2]);  view_862 = None
	        full_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_55, full_110);  amin_55 = full_110 = None
	        full_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_55, full_111);  amax_55 = full_111 = None
	        sub_2529: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_55, minimum_55);  maximum_55 = None
	        div_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2529, 255.0);  sub_2529 = None
	        clamp_min_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_110, 1.1920928955078125e-07);  div_110 = None
	        div_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_55, clamp_min_165);  minimum_55 = None
	        round_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_111);  div_111 = None
	        sub_2535: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_111);  round_111 = None
	        clamp_min_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2535, -128);  sub_2535 = None
	        clamp_max_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_166, 127);  clamp_min_166 = None
	        _assert_tensor_metadata_497 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_165, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_497 = None
	        _assert_tensor_metadata_498 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_110, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_498 = None
	        convert_element_type_330: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_110, torch.int8);  clamp_max_110 = None
	        view_863: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_8294, [sym_size_int, 1500, 1280])
	        view_864: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_165, [sym_size_int, 1500, 1])
	        view_865: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_330, [sym_size_int, 1500, 1])
	        reciprocal_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_864);  view_864 = None
	        mul_5385: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_55, 1.0);  reciprocal_55 = None
	        mul_5388: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_863, mul_5385);  view_863 = mul_5385 = None
	        round_112: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5388);  mul_5388 = None
	        add_8533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_112, view_865);  round_112 = view_865 = None
	        clamp_min_167: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8533, -128);  add_8533 = None
	        clamp_max_111: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_167, 127);  clamp_min_167 = None
	        view_866: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_111, [sym_size_int, 1500, 1280]);  clamp_max_111 = None
	        _assert_tensor_metadata_499 = torch.ops.aten._assert_tensor_metadata.default(view_866, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_499 = None
	        convert_element_type_331: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_866, torch.int8);  view_866 = None
	        view_867: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_331, [sym_size_int, 1500, 1280]);  convert_element_type_331 = None
	        view_868: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_165, [sym_size_int, 1500, 1]);  clamp_min_165 = None
	        view_869: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_330, [sym_size_int, 1500, 1]);  convert_element_type_330 = None
	        _assert_tensor_metadata_500 = torch.ops.aten._assert_tensor_metadata.default(view_867, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_500 = None
	        convert_element_type_332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_867, torch.float32);  view_867 = None
	        _assert_tensor_metadata_501 = torch.ops.aten._assert_tensor_metadata.default(view_869, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_501 = None
	        convert_element_type_333: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_869, torch.float32);  view_869 = None
	        sub_2555: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_332, convert_element_type_333);  convert_element_type_332 = convert_element_type_333 = None
	        mul_5410: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2555, view_868);  sub_2555 = view_868 = None
	        view_870: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5410, [sym_size_int, 1500, 1280]);  mul_5410 = None
	        _assert_tensor_metadata_502 = torch.ops.aten._assert_tensor_metadata.default(view_870, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_502 = None
	        view_871: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg254_1, [1280, 40, 32]);  arg254_1 = None
	        view_872: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg255_1, [1280, 40, 1]);  arg255_1 = None
	        view_873: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg256_1, [1280, 40, 1]);  arg256_1 = None
	        _assert_tensor_metadata_503 = torch.ops.aten._assert_tensor_metadata.default(view_871, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_503 = None
	        convert_element_type_334: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_871, torch.float32);  view_871 = None
	        _assert_tensor_metadata_504 = torch.ops.aten._assert_tensor_metadata.default(view_873, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_504 = None
	        convert_element_type_335: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_873, torch.float32);  view_873 = None
	        sub_2559: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_334, convert_element_type_335);  convert_element_type_334 = convert_element_type_335 = None
	        mul_5415: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2559, view_872);  sub_2559 = view_872 = None
	        view_874: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5415, [1280, 1280]);  mul_5415 = None
	        _assert_tensor_metadata_505 = torch.ops.aten._assert_tensor_metadata.default(view_874, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_505 = None
	        permute_93: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_874, [1, 0]);  view_874 = None
	        mul_5418: "Sym(1500*s6)" = sym_size_int * 1500
	        view_875: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_870, [mul_5418, 1280]);  view_870 = mul_5418 = None
	        mm_9: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_875, permute_93);  view_875 = permute_93 = None
	        view_876: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_9, [sym_size_int, 1500, 1280]);  mm_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_877: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_876, [sym_size_int, -1, 20, 64]);  view_876 = None
	        permute_94: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_877, [0, 2, 1, 3]);  view_877 = None
	        clone_75: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_94, memory_format = torch.contiguous_format);  permute_94 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_878: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_8294, [sym_size_int, 1500, 1280])
	        amin_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_878, [2])
	        amax_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_878, [2]);  view_878 = None
	        full_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_56, full_112);  amin_56 = full_112 = None
	        full_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_56, full_113);  amax_56 = full_113 = None
	        sub_2573: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_56, minimum_56);  maximum_56 = None
	        div_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2573, 255.0);  sub_2573 = None
	        clamp_min_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_112, 1.1920928955078125e-07);  div_112 = None
	        div_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_56, clamp_min_168);  minimum_56 = None
	        round_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_113);  div_113 = None
	        sub_2579: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_113);  round_113 = None
	        clamp_min_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2579, -128);  sub_2579 = None
	        clamp_max_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_169, 127);  clamp_min_169 = None
	        _assert_tensor_metadata_506 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_168, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_506 = None
	        _assert_tensor_metadata_507 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_112, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_507 = None
	        convert_element_type_336: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_112, torch.int8);  clamp_max_112 = None
	        view_879: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_8294, [sym_size_int, 1500, 1280]);  add_8294 = None
	        view_880: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_168, [sym_size_int, 1500, 1])
	        view_881: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_336, [sym_size_int, 1500, 1])
	        reciprocal_56: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_880);  view_880 = None
	        mul_5484: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_56, 1.0);  reciprocal_56 = None
	        mul_5487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_879, mul_5484);  view_879 = mul_5484 = None
	        round_114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5487);  mul_5487 = None
	        add_8681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_114, view_881);  round_114 = view_881 = None
	        clamp_min_170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8681, -128);  add_8681 = None
	        clamp_max_113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_170, 127);  clamp_min_170 = None
	        view_882: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_113, [sym_size_int, 1500, 1280]);  clamp_max_113 = None
	        _assert_tensor_metadata_508 = torch.ops.aten._assert_tensor_metadata.default(view_882, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_508 = None
	        convert_element_type_337: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_882, torch.int8);  view_882 = None
	        view_883: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_337, [sym_size_int, 1500, 1280]);  convert_element_type_337 = None
	        view_884: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_168, [sym_size_int, 1500, 1]);  clamp_min_168 = None
	        view_885: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_336, [sym_size_int, 1500, 1]);  convert_element_type_336 = None
	        _assert_tensor_metadata_509 = torch.ops.aten._assert_tensor_metadata.default(view_883, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_509 = None
	        convert_element_type_338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_883, torch.float32);  view_883 = None
	        _assert_tensor_metadata_510 = torch.ops.aten._assert_tensor_metadata.default(view_885, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_510 = None
	        convert_element_type_339: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_885, torch.float32);  view_885 = None
	        sub_2599: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_338, convert_element_type_339);  convert_element_type_338 = convert_element_type_339 = None
	        mul_5509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2599, view_884);  sub_2599 = view_884 = None
	        view_886: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5509, [sym_size_int, 1500, 1280]);  mul_5509 = None
	        _assert_tensor_metadata_511 = torch.ops.aten._assert_tensor_metadata.default(view_886, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_511 = None
	        view_887: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg258_1, [1280, 40, 32]);  arg258_1 = None
	        view_888: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg259_1, [1280, 40, 1]);  arg259_1 = None
	        view_889: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg260_1, [1280, 40, 1]);  arg260_1 = None
	        _assert_tensor_metadata_512 = torch.ops.aten._assert_tensor_metadata.default(view_887, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_512 = None
	        convert_element_type_340: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_887, torch.float32);  view_887 = None
	        _assert_tensor_metadata_513 = torch.ops.aten._assert_tensor_metadata.default(view_889, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_513 = None
	        convert_element_type_341: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_889, torch.float32);  view_889 = None
	        sub_2603: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_340, convert_element_type_341);  convert_element_type_340 = convert_element_type_341 = None
	        mul_5514: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2603, view_888);  sub_2603 = view_888 = None
	        view_890: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5514, [1280, 1280]);  mul_5514 = None
	        _assert_tensor_metadata_514 = torch.ops.aten._assert_tensor_metadata.default(view_890, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_514 = None
	        mul_5519: "Sym(1500*s6)" = sym_size_int * 1500
	        view_891: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_886, [mul_5519, 1280]);  view_886 = mul_5519 = None
	        permute_95: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_890, [1, 0]);  view_890 = None
	        addmm_46: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg257_1, view_891, permute_95);  arg257_1 = view_891 = permute_95 = None
	        view_892: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_46, [sym_size_int, 1500, 1280]);  addmm_46 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_893: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_892, [sym_size_int, -1, 20, 64]);  view_892 = None
	        permute_96: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_893, [0, 2, 1, 3]);  view_893 = None
	        clone_76: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_96, memory_format = torch.contiguous_format);  permute_96 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_9 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_74, clone_75, clone_76, None, False, scale = 1.0);  clone_74 = clone_75 = clone_76 = None
	        getitem_74: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_9[0];  _scaled_dot_product_efficient_attention_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_97: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_74, [0, 2, 1, 3]);  getitem_74 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_97, [sym_size_int, 1500, -1]);  permute_97 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_895: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_894, [sym_size_int, 1500, 1280])
	        amin_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_895, [2])
	        amax_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_895, [2]);  view_895 = None
	        full_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_57, full_114);  amin_57 = full_114 = None
	        full_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_57, full_115);  amax_57 = full_115 = None
	        sub_2621: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_57, minimum_57);  maximum_57 = None
	        div_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2621, 255.0);  sub_2621 = None
	        clamp_min_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_114, 1.1920928955078125e-07);  div_114 = None
	        div_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_57, clamp_min_171);  minimum_57 = None
	        round_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_115);  div_115 = None
	        sub_2627: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_115);  round_115 = None
	        clamp_min_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2627, -128);  sub_2627 = None
	        clamp_max_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_172, 127);  clamp_min_172 = None
	        _assert_tensor_metadata_515 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_171, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_515 = None
	        _assert_tensor_metadata_516 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_114, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_516 = None
	        convert_element_type_342: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_114, torch.int8);  clamp_max_114 = None
	        view_896: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_894, [sym_size_int, 1500, 1280]);  view_894 = None
	        view_897: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_171, [sym_size_int, 1500, 1])
	        view_898: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_342, [sym_size_int, 1500, 1])
	        reciprocal_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_897);  view_897 = None
	        mul_5589: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_57, 1.0);  reciprocal_57 = None
	        mul_5592: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_896, mul_5589);  view_896 = mul_5589 = None
	        round_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5592);  mul_5592 = None
	        add_8845: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_116, view_898);  round_116 = view_898 = None
	        clamp_min_173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8845, -128);  add_8845 = None
	        clamp_max_115: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_173, 127);  clamp_min_173 = None
	        view_899: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_115, [sym_size_int, 1500, 1280]);  clamp_max_115 = None
	        _assert_tensor_metadata_517 = torch.ops.aten._assert_tensor_metadata.default(view_899, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_517 = None
	        convert_element_type_343: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_899, torch.int8);  view_899 = None
	        view_900: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_343, [sym_size_int, 1500, 1280]);  convert_element_type_343 = None
	        view_901: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_171, [sym_size_int, 1500, 1]);  clamp_min_171 = None
	        view_902: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_342, [sym_size_int, 1500, 1]);  convert_element_type_342 = None
	        _assert_tensor_metadata_518 = torch.ops.aten._assert_tensor_metadata.default(view_900, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_518 = None
	        convert_element_type_344: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_900, torch.float32);  view_900 = None
	        _assert_tensor_metadata_519 = torch.ops.aten._assert_tensor_metadata.default(view_902, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_519 = None
	        convert_element_type_345: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_902, torch.float32);  view_902 = None
	        sub_2647: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_344, convert_element_type_345);  convert_element_type_344 = convert_element_type_345 = None
	        mul_5614: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2647, view_901);  sub_2647 = view_901 = None
	        view_903: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5614, [sym_size_int, 1500, 1280]);  mul_5614 = None
	        _assert_tensor_metadata_520 = torch.ops.aten._assert_tensor_metadata.default(view_903, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_520 = None
	        view_904: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg262_1, [1280, 40, 32]);  arg262_1 = None
	        view_905: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg263_1, [1280, 40, 1]);  arg263_1 = None
	        view_906: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg264_1, [1280, 40, 1]);  arg264_1 = None
	        _assert_tensor_metadata_521 = torch.ops.aten._assert_tensor_metadata.default(view_904, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_521 = None
	        convert_element_type_346: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_904, torch.float32);  view_904 = None
	        _assert_tensor_metadata_522 = torch.ops.aten._assert_tensor_metadata.default(view_906, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_522 = None
	        convert_element_type_347: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_906, torch.float32);  view_906 = None
	        sub_2651: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_346, convert_element_type_347);  convert_element_type_346 = convert_element_type_347 = None
	        mul_5619: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2651, view_905);  sub_2651 = view_905 = None
	        view_907: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5619, [1280, 1280]);  mul_5619 = None
	        _assert_tensor_metadata_523 = torch.ops.aten._assert_tensor_metadata.default(view_907, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_523 = None
	        mul_5624: "Sym(1500*s6)" = sym_size_int * 1500
	        view_908: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_903, [mul_5624, 1280]);  view_903 = mul_5624 = None
	        permute_98: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_907, [1, 0]);  view_907 = None
	        addmm_47: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg261_1, view_908, permute_98);  arg261_1 = view_908 = permute_98 = None
	        view_909: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_47, [sym_size_int, 1500, 1280]);  addmm_47 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_77: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_909);  view_909 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_8908: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_8288, clone_77);  add_8288 = clone_77 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_78: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_8908, memory_format = torch.contiguous_format)
	        var_mean_19 = torch.ops.aten.var_mean.correction(clone_78, [2], correction = 0, keepdim = True)
	        getitem_78: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_19[0]
	        getitem_79: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_19[1];  var_mean_19 = None
	        add_8913: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_78, 1e-05);  getitem_78 = None
	        rsqrt_19: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_8913);  add_8913 = None
	        sub_2657: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_78, getitem_79);  clone_78 = getitem_79 = None
	        mul_5635: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2657, rsqrt_19);  sub_2657 = rsqrt_19 = None
	        mul_5636: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5635, arg265_1);  mul_5635 = arg265_1 = None
	        add_8914: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5636, arg266_1);  mul_5636 = arg266_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_910: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_8914, [sym_size_int, 1500, 1280])
	        amin_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_910, [2])
	        amax_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_910, [2]);  view_910 = None
	        full_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_58, full_116);  amin_58 = full_116 = None
	        full_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_58, full_117);  amax_58 = full_117 = None
	        sub_2668: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_58, minimum_58);  maximum_58 = None
	        div_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2668, 255.0);  sub_2668 = None
	        clamp_min_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_116, 1.1920928955078125e-07);  div_116 = None
	        div_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_58, clamp_min_174);  minimum_58 = None
	        round_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_117);  div_117 = None
	        sub_2674: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_117);  round_117 = None
	        clamp_min_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2674, -128);  sub_2674 = None
	        clamp_max_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_175, 127);  clamp_min_175 = None
	        _assert_tensor_metadata_524 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_174, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_524 = None
	        _assert_tensor_metadata_525 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_116, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_525 = None
	        convert_element_type_348: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_116, torch.int8);  clamp_max_116 = None
	        view_911: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_8914, [sym_size_int, 1500, 1280]);  add_8914 = None
	        view_912: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_174, [sym_size_int, 1500, 1])
	        view_913: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_348, [sym_size_int, 1500, 1])
	        reciprocal_58: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_912);  view_912 = None
	        mul_5684: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_58, 1.0);  reciprocal_58 = None
	        mul_5687: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_911, mul_5684);  view_911 = mul_5684 = None
	        round_118: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5687);  mul_5687 = None
	        add_9001: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_118, view_913);  round_118 = view_913 = None
	        clamp_min_176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9001, -128);  add_9001 = None
	        clamp_max_117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_176, 127);  clamp_min_176 = None
	        view_914: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_117, [sym_size_int, 1500, 1280]);  clamp_max_117 = None
	        _assert_tensor_metadata_526 = torch.ops.aten._assert_tensor_metadata.default(view_914, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_526 = None
	        convert_element_type_349: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_914, torch.int8);  view_914 = None
	        view_915: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_349, [sym_size_int, 1500, 1280]);  convert_element_type_349 = None
	        view_916: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_174, [sym_size_int, 1500, 1]);  clamp_min_174 = None
	        view_917: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_348, [sym_size_int, 1500, 1]);  convert_element_type_348 = None
	        _assert_tensor_metadata_527 = torch.ops.aten._assert_tensor_metadata.default(view_915, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_527 = None
	        convert_element_type_350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_915, torch.float32);  view_915 = None
	        _assert_tensor_metadata_528 = torch.ops.aten._assert_tensor_metadata.default(view_917, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_528 = None
	        convert_element_type_351: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_917, torch.float32);  view_917 = None
	        sub_2694: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_350, convert_element_type_351);  convert_element_type_350 = convert_element_type_351 = None
	        mul_5709: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2694, view_916);  sub_2694 = view_916 = None
	        view_918: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5709, [sym_size_int, 1500, 1280]);  mul_5709 = None
	        _assert_tensor_metadata_529 = torch.ops.aten._assert_tensor_metadata.default(view_918, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_529 = None
	        view_919: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg268_1, [5120, 40, 32]);  arg268_1 = None
	        view_920: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg269_1, [5120, 40, 1]);  arg269_1 = None
	        view_921: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg270_1, [5120, 40, 1]);  arg270_1 = None
	        _assert_tensor_metadata_530 = torch.ops.aten._assert_tensor_metadata.default(view_919, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_530 = None
	        convert_element_type_352: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_919, torch.float32);  view_919 = None
	        _assert_tensor_metadata_531 = torch.ops.aten._assert_tensor_metadata.default(view_921, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_531 = None
	        convert_element_type_353: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_921, torch.float32);  view_921 = None
	        sub_2698: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_352, convert_element_type_353);  convert_element_type_352 = convert_element_type_353 = None
	        mul_5714: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2698, view_920);  sub_2698 = view_920 = None
	        view_922: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5714, [5120, 1280]);  mul_5714 = None
	        _assert_tensor_metadata_532 = torch.ops.aten._assert_tensor_metadata.default(view_922, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_532 = None
	        mul_5719: "Sym(1500*s6)" = sym_size_int * 1500
	        view_923: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_918, [mul_5719, 1280]);  view_918 = mul_5719 = None
	        permute_99: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_922, [1, 0]);  view_922 = None
	        addmm_48: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg267_1, view_923, permute_99);  arg267_1 = view_923 = permute_99 = None
	        view_924: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_48, [sym_size_int, 1500, 5120]);  addmm_48 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_5726: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_924, 0.5)
	        mul_5727: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_924, 0.7071067811865476);  view_924 = None
	        erf_11: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_5727);  mul_5727 = None
	        add_9060: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_11, 1);  erf_11 = None
	        mul_5728: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5726, add_9060);  mul_5726 = add_9060 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_79: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_5728);  mul_5728 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_925: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_79, [sym_size_int, 1500, 5120])
	        amin_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_925, [2])
	        amax_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_925, [2]);  view_925 = None
	        full_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_59, full_118);  amin_59 = full_118 = None
	        full_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_59, full_119);  amax_59 = full_119 = None
	        sub_2711: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_59, minimum_59);  maximum_59 = None
	        div_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2711, 255.0);  sub_2711 = None
	        clamp_min_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_118, 1.1920928955078125e-07);  div_118 = None
	        div_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_59, clamp_min_177);  minimum_59 = None
	        round_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_119);  div_119 = None
	        sub_2717: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_119);  round_119 = None
	        clamp_min_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2717, -128);  sub_2717 = None
	        clamp_max_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_178, 127);  clamp_min_178 = None
	        _assert_tensor_metadata_533 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_177, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_533 = None
	        _assert_tensor_metadata_534 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_118, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_534 = None
	        convert_element_type_354: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_118, torch.int8);  clamp_max_118 = None
	        view_926: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_79, [sym_size_int, 1500, 5120]);  clone_79 = None
	        view_927: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_177, [sym_size_int, 1500, 1])
	        view_928: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_354, [sym_size_int, 1500, 1])
	        reciprocal_59: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_927);  view_927 = None
	        mul_5774: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_59, 1.0);  reciprocal_59 = None
	        mul_5777: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_926, mul_5774);  view_926 = mul_5774 = None
	        round_120: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_5777);  mul_5777 = None
	        add_9143: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_120, view_928);  round_120 = view_928 = None
	        clamp_min_179: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9143, -128);  add_9143 = None
	        clamp_max_119: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_179, 127);  clamp_min_179 = None
	        view_929: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_119, [sym_size_int, 1500, 5120]);  clamp_max_119 = None
	        _assert_tensor_metadata_535 = torch.ops.aten._assert_tensor_metadata.default(view_929, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_535 = None
	        convert_element_type_355: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_929, torch.int8);  view_929 = None
	        view_930: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_355, [sym_size_int, 1500, 5120]);  convert_element_type_355 = None
	        view_931: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_177, [sym_size_int, 1500, 1]);  clamp_min_177 = None
	        view_932: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_354, [sym_size_int, 1500, 1]);  convert_element_type_354 = None
	        _assert_tensor_metadata_536 = torch.ops.aten._assert_tensor_metadata.default(view_930, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_536 = None
	        convert_element_type_356: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_930, torch.float32);  view_930 = None
	        _assert_tensor_metadata_537 = torch.ops.aten._assert_tensor_metadata.default(view_932, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_537 = None
	        convert_element_type_357: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_932, torch.float32);  view_932 = None
	        sub_2737: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_356, convert_element_type_357);  convert_element_type_356 = convert_element_type_357 = None
	        mul_5799: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2737, view_931);  sub_2737 = view_931 = None
	        view_933: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_5799, [sym_size_int, 1500, 5120]);  mul_5799 = None
	        _assert_tensor_metadata_538 = torch.ops.aten._assert_tensor_metadata.default(view_933, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_538 = None
	        view_934: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg272_1, [1280, 160, 32]);  arg272_1 = None
	        view_935: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg273_1, [1280, 160, 1]);  arg273_1 = None
	        view_936: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg274_1, [1280, 160, 1]);  arg274_1 = None
	        _assert_tensor_metadata_539 = torch.ops.aten._assert_tensor_metadata.default(view_934, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_539 = None
	        convert_element_type_358: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_934, torch.float32);  view_934 = None
	        _assert_tensor_metadata_540 = torch.ops.aten._assert_tensor_metadata.default(view_936, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_540 = None
	        convert_element_type_359: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_936, torch.float32);  view_936 = None
	        sub_2741: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_358, convert_element_type_359);  convert_element_type_358 = convert_element_type_359 = None
	        mul_5804: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2741, view_935);  sub_2741 = view_935 = None
	        view_937: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_5804, [1280, 5120]);  mul_5804 = None
	        _assert_tensor_metadata_541 = torch.ops.aten._assert_tensor_metadata.default(view_937, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_541 = None
	        mul_5809: "Sym(1500*s6)" = sym_size_int * 1500
	        view_938: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_933, [mul_5809, 5120]);  view_933 = mul_5809 = None
	        permute_100: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_937, [1, 0]);  view_937 = None
	        addmm_49: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg271_1, view_938, permute_100);  arg271_1 = view_938 = permute_100 = None
	        view_939: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_49, [sym_size_int, 1500, 1280]);  addmm_49 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_939);  view_939 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_9206: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_8908, clone_80);  add_8908 = clone_80 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_81: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_9206, memory_format = torch.contiguous_format)
	        var_mean_20 = torch.ops.aten.var_mean.correction(clone_81, [2], correction = 0, keepdim = True)
	        getitem_80: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_20[0]
	        getitem_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_20[1];  var_mean_20 = None
	        add_9211: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_80, 1e-05);  getitem_80 = None
	        rsqrt_20: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_9211);  add_9211 = None
	        sub_2747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_81, getitem_81);  clone_81 = getitem_81 = None
	        mul_5820: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2747, rsqrt_20);  sub_2747 = rsqrt_20 = None
	        mul_5821: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5820, arg275_1);  mul_5820 = arg275_1 = None
	        add_9212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5821, arg276_1);  mul_5821 = arg276_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_940: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_9212, [sym_size_int, 1500, 1280])
	        amin_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_940, [2])
	        amax_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_940, [2]);  view_940 = None
	        full_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_60, full_120);  amin_60 = full_120 = None
	        full_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_60, full_121);  amax_60 = full_121 = None
	        sub_2758: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_60, minimum_60);  maximum_60 = None
	        div_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2758, 255.0);  sub_2758 = None
	        clamp_min_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_120, 1.1920928955078125e-07);  div_120 = None
	        div_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_60, clamp_min_180);  minimum_60 = None
	        round_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_121);  div_121 = None
	        sub_2764: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_121);  round_121 = None
	        clamp_min_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2764, -128);  sub_2764 = None
	        clamp_max_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_181, 127);  clamp_min_181 = None
	        _assert_tensor_metadata_542 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_180, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_542 = None
	        _assert_tensor_metadata_543 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_120, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_543 = None
	        convert_element_type_360: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_120, torch.int8);  clamp_max_120 = None
	        view_941: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_9212, [sym_size_int, 1500, 1280])
	        view_942: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_180, [sym_size_int, 1500, 1])
	        view_943: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_360, [sym_size_int, 1500, 1])
	        reciprocal_60: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_942);  view_942 = None
	        mul_5869: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_60, 1.0);  reciprocal_60 = None
	        mul_5872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_941, mul_5869);  view_941 = mul_5869 = None
	        round_122: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5872);  mul_5872 = None
	        add_9299: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_122, view_943);  round_122 = view_943 = None
	        clamp_min_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9299, -128);  add_9299 = None
	        clamp_max_121: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_182, 127);  clamp_min_182 = None
	        view_944: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_121, [sym_size_int, 1500, 1280]);  clamp_max_121 = None
	        _assert_tensor_metadata_544 = torch.ops.aten._assert_tensor_metadata.default(view_944, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_544 = None
	        convert_element_type_361: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_944, torch.int8);  view_944 = None
	        view_945: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_361, [sym_size_int, 1500, 1280]);  convert_element_type_361 = None
	        view_946: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_180, [sym_size_int, 1500, 1]);  clamp_min_180 = None
	        view_947: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_360, [sym_size_int, 1500, 1]);  convert_element_type_360 = None
	        _assert_tensor_metadata_545 = torch.ops.aten._assert_tensor_metadata.default(view_945, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_545 = None
	        convert_element_type_362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_945, torch.float32);  view_945 = None
	        _assert_tensor_metadata_546 = torch.ops.aten._assert_tensor_metadata.default(view_947, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_546 = None
	        convert_element_type_363: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_947, torch.float32);  view_947 = None
	        sub_2784: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_362, convert_element_type_363);  convert_element_type_362 = convert_element_type_363 = None
	        mul_5894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2784, view_946);  sub_2784 = view_946 = None
	        view_948: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5894, [sym_size_int, 1500, 1280]);  mul_5894 = None
	        _assert_tensor_metadata_547 = torch.ops.aten._assert_tensor_metadata.default(view_948, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_547 = None
	        view_949: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg278_1, [1280, 40, 32]);  arg278_1 = None
	        view_950: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg279_1, [1280, 40, 1]);  arg279_1 = None
	        view_951: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg280_1, [1280, 40, 1]);  arg280_1 = None
	        _assert_tensor_metadata_548 = torch.ops.aten._assert_tensor_metadata.default(view_949, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_548 = None
	        convert_element_type_364: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_949, torch.float32);  view_949 = None
	        _assert_tensor_metadata_549 = torch.ops.aten._assert_tensor_metadata.default(view_951, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_549 = None
	        convert_element_type_365: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_951, torch.float32);  view_951 = None
	        sub_2788: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_364, convert_element_type_365);  convert_element_type_364 = convert_element_type_365 = None
	        mul_5899: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2788, view_950);  sub_2788 = view_950 = None
	        view_952: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5899, [1280, 1280]);  mul_5899 = None
	        _assert_tensor_metadata_550 = torch.ops.aten._assert_tensor_metadata.default(view_952, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_550 = None
	        mul_5904: "Sym(1500*s6)" = sym_size_int * 1500
	        view_953: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_948, [mul_5904, 1280]);  view_948 = mul_5904 = None
	        permute_101: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_952, [1, 0]);  view_952 = None
	        addmm_50: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg277_1, view_953, permute_101);  arg277_1 = view_953 = permute_101 = None
	        view_954: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_50, [sym_size_int, 1500, 1280]);  addmm_50 = None
	        mul_5911: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_954, 0.125);  view_954 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_955: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_5911, [sym_size_int, 1500, 20, 64]);  mul_5911 = None
	        permute_102: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_955, [0, 2, 1, 3]);  view_955 = None
	        clone_82: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_102, memory_format = torch.contiguous_format);  permute_102 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_956: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_9212, [sym_size_int, 1500, 1280])
	        amin_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_956, [2])
	        amax_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_956, [2]);  view_956 = None
	        full_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_61, full_122);  amin_61 = full_122 = None
	        full_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_61, full_123);  amax_61 = full_123 = None
	        sub_2803: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_61, minimum_61);  maximum_61 = None
	        div_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2803, 255.0);  sub_2803 = None
	        clamp_min_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_122, 1.1920928955078125e-07);  div_122 = None
	        div_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_61, clamp_min_183);  minimum_61 = None
	        round_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_123);  div_123 = None
	        sub_2809: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_123);  round_123 = None
	        clamp_min_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2809, -128);  sub_2809 = None
	        clamp_max_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_184, 127);  clamp_min_184 = None
	        _assert_tensor_metadata_551 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_183, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_551 = None
	        _assert_tensor_metadata_552 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_122, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_552 = None
	        convert_element_type_366: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_122, torch.int8);  clamp_max_122 = None
	        view_957: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_9212, [sym_size_int, 1500, 1280])
	        view_958: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_183, [sym_size_int, 1500, 1])
	        view_959: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_366, [sym_size_int, 1500, 1])
	        reciprocal_61: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_958);  view_958 = None
	        mul_5965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_61, 1.0);  reciprocal_61 = None
	        mul_5968: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_957, mul_5965);  view_957 = mul_5965 = None
	        round_124: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5968);  mul_5968 = None
	        add_9451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_124, view_959);  round_124 = view_959 = None
	        clamp_min_185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9451, -128);  add_9451 = None
	        clamp_max_123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_185, 127);  clamp_min_185 = None
	        view_960: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_123, [sym_size_int, 1500, 1280]);  clamp_max_123 = None
	        _assert_tensor_metadata_553 = torch.ops.aten._assert_tensor_metadata.default(view_960, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_553 = None
	        convert_element_type_367: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_960, torch.int8);  view_960 = None
	        view_961: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_367, [sym_size_int, 1500, 1280]);  convert_element_type_367 = None
	        view_962: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_183, [sym_size_int, 1500, 1]);  clamp_min_183 = None
	        view_963: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_366, [sym_size_int, 1500, 1]);  convert_element_type_366 = None
	        _assert_tensor_metadata_554 = torch.ops.aten._assert_tensor_metadata.default(view_961, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_554 = None
	        convert_element_type_368: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_961, torch.float32);  view_961 = None
	        _assert_tensor_metadata_555 = torch.ops.aten._assert_tensor_metadata.default(view_963, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_555 = None
	        convert_element_type_369: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_963, torch.float32);  view_963 = None
	        sub_2829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_368, convert_element_type_369);  convert_element_type_368 = convert_element_type_369 = None
	        mul_5990: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2829, view_962);  sub_2829 = view_962 = None
	        view_964: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5990, [sym_size_int, 1500, 1280]);  mul_5990 = None
	        _assert_tensor_metadata_556 = torch.ops.aten._assert_tensor_metadata.default(view_964, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_556 = None
	        view_965: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg281_1, [1280, 40, 32]);  arg281_1 = None
	        view_966: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg282_1, [1280, 40, 1]);  arg282_1 = None
	        view_967: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg283_1, [1280, 40, 1]);  arg283_1 = None
	        _assert_tensor_metadata_557 = torch.ops.aten._assert_tensor_metadata.default(view_965, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_557 = None
	        convert_element_type_370: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_965, torch.float32);  view_965 = None
	        _assert_tensor_metadata_558 = torch.ops.aten._assert_tensor_metadata.default(view_967, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_558 = None
	        convert_element_type_371: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_967, torch.float32);  view_967 = None
	        sub_2833: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_370, convert_element_type_371);  convert_element_type_370 = convert_element_type_371 = None
	        mul_5995: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2833, view_966);  sub_2833 = view_966 = None
	        view_968: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5995, [1280, 1280]);  mul_5995 = None
	        _assert_tensor_metadata_559 = torch.ops.aten._assert_tensor_metadata.default(view_968, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_559 = None
	        permute_103: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_968, [1, 0]);  view_968 = None
	        mul_5998: "Sym(1500*s6)" = sym_size_int * 1500
	        view_969: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_964, [mul_5998, 1280]);  view_964 = mul_5998 = None
	        mm_10: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_969, permute_103);  view_969 = permute_103 = None
	        view_970: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_10, [sym_size_int, 1500, 1280]);  mm_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_971: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_970, [sym_size_int, -1, 20, 64]);  view_970 = None
	        permute_104: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_971, [0, 2, 1, 3]);  view_971 = None
	        clone_83: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_104, memory_format = torch.contiguous_format);  permute_104 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_972: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_9212, [sym_size_int, 1500, 1280])
	        amin_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_972, [2])
	        amax_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_972, [2]);  view_972 = None
	        full_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_62, full_124);  amin_62 = full_124 = None
	        full_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_62, full_125);  amax_62 = full_125 = None
	        sub_2847: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_62, minimum_62);  maximum_62 = None
	        div_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2847, 255.0);  sub_2847 = None
	        clamp_min_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_124, 1.1920928955078125e-07);  div_124 = None
	        div_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_62, clamp_min_186);  minimum_62 = None
	        round_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_125);  div_125 = None
	        sub_2853: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_125);  round_125 = None
	        clamp_min_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2853, -128);  sub_2853 = None
	        clamp_max_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_187, 127);  clamp_min_187 = None
	        _assert_tensor_metadata_560 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_186, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_560 = None
	        _assert_tensor_metadata_561 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_124, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_561 = None
	        convert_element_type_372: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_124, torch.int8);  clamp_max_124 = None
	        view_973: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_9212, [sym_size_int, 1500, 1280]);  add_9212 = None
	        view_974: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_186, [sym_size_int, 1500, 1])
	        view_975: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_372, [sym_size_int, 1500, 1])
	        reciprocal_62: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_974);  view_974 = None
	        mul_6064: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_62, 1.0);  reciprocal_62 = None
	        mul_6067: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_973, mul_6064);  view_973 = mul_6064 = None
	        round_126: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6067);  mul_6067 = None
	        add_9599: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_126, view_975);  round_126 = view_975 = None
	        clamp_min_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9599, -128);  add_9599 = None
	        clamp_max_125: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_188, 127);  clamp_min_188 = None
	        view_976: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_125, [sym_size_int, 1500, 1280]);  clamp_max_125 = None
	        _assert_tensor_metadata_562 = torch.ops.aten._assert_tensor_metadata.default(view_976, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_562 = None
	        convert_element_type_373: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_976, torch.int8);  view_976 = None
	        view_977: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_373, [sym_size_int, 1500, 1280]);  convert_element_type_373 = None
	        view_978: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_186, [sym_size_int, 1500, 1]);  clamp_min_186 = None
	        view_979: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_372, [sym_size_int, 1500, 1]);  convert_element_type_372 = None
	        _assert_tensor_metadata_563 = torch.ops.aten._assert_tensor_metadata.default(view_977, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_563 = None
	        convert_element_type_374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_977, torch.float32);  view_977 = None
	        _assert_tensor_metadata_564 = torch.ops.aten._assert_tensor_metadata.default(view_979, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_564 = None
	        convert_element_type_375: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_979, torch.float32);  view_979 = None
	        sub_2873: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_374, convert_element_type_375);  convert_element_type_374 = convert_element_type_375 = None
	        mul_6089: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2873, view_978);  sub_2873 = view_978 = None
	        view_980: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6089, [sym_size_int, 1500, 1280]);  mul_6089 = None
	        _assert_tensor_metadata_565 = torch.ops.aten._assert_tensor_metadata.default(view_980, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_565 = None
	        view_981: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg285_1, [1280, 40, 32]);  arg285_1 = None
	        view_982: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg286_1, [1280, 40, 1]);  arg286_1 = None
	        view_983: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg287_1, [1280, 40, 1]);  arg287_1 = None
	        _assert_tensor_metadata_566 = torch.ops.aten._assert_tensor_metadata.default(view_981, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_566 = None
	        convert_element_type_376: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_981, torch.float32);  view_981 = None
	        _assert_tensor_metadata_567 = torch.ops.aten._assert_tensor_metadata.default(view_983, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_567 = None
	        convert_element_type_377: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_983, torch.float32);  view_983 = None
	        sub_2877: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_376, convert_element_type_377);  convert_element_type_376 = convert_element_type_377 = None
	        mul_6094: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2877, view_982);  sub_2877 = view_982 = None
	        view_984: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6094, [1280, 1280]);  mul_6094 = None
	        _assert_tensor_metadata_568 = torch.ops.aten._assert_tensor_metadata.default(view_984, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_568 = None
	        mul_6099: "Sym(1500*s6)" = sym_size_int * 1500
	        view_985: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_980, [mul_6099, 1280]);  view_980 = mul_6099 = None
	        permute_105: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_984, [1, 0]);  view_984 = None
	        addmm_51: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg284_1, view_985, permute_105);  arg284_1 = view_985 = permute_105 = None
	        view_986: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_51, [sym_size_int, 1500, 1280]);  addmm_51 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_987: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_986, [sym_size_int, -1, 20, 64]);  view_986 = None
	        permute_106: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_987, [0, 2, 1, 3]);  view_987 = None
	        clone_84: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_106, memory_format = torch.contiguous_format);  permute_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_10 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_82, clone_83, clone_84, None, False, scale = 1.0);  clone_82 = clone_83 = clone_84 = None
	        getitem_82: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_10[0];  _scaled_dot_product_efficient_attention_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_107: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_82, [0, 2, 1, 3]);  getitem_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_988: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_107, [sym_size_int, 1500, -1]);  permute_107 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_989: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_988, [sym_size_int, 1500, 1280])
	        amin_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_989, [2])
	        amax_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_989, [2]);  view_989 = None
	        full_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_63, full_126);  amin_63 = full_126 = None
	        full_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_63, full_127);  amax_63 = full_127 = None
	        sub_2895: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_63, minimum_63);  maximum_63 = None
	        div_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2895, 255.0);  sub_2895 = None
	        clamp_min_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_126, 1.1920928955078125e-07);  div_126 = None
	        div_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_63, clamp_min_189);  minimum_63 = None
	        round_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_127);  div_127 = None
	        sub_2901: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_127);  round_127 = None
	        clamp_min_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2901, -128);  sub_2901 = None
	        clamp_max_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_190, 127);  clamp_min_190 = None
	        _assert_tensor_metadata_569 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_189, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_569 = None
	        _assert_tensor_metadata_570 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_126, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_570 = None
	        convert_element_type_378: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_126, torch.int8);  clamp_max_126 = None
	        view_990: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_988, [sym_size_int, 1500, 1280]);  view_988 = None
	        view_991: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_189, [sym_size_int, 1500, 1])
	        view_992: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_378, [sym_size_int, 1500, 1])
	        reciprocal_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_991);  view_991 = None
	        mul_6169: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_63, 1.0);  reciprocal_63 = None
	        mul_6172: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_990, mul_6169);  view_990 = mul_6169 = None
	        round_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6172);  mul_6172 = None
	        add_9763: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_128, view_992);  round_128 = view_992 = None
	        clamp_min_191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9763, -128);  add_9763 = None
	        clamp_max_127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_191, 127);  clamp_min_191 = None
	        view_993: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_127, [sym_size_int, 1500, 1280]);  clamp_max_127 = None
	        _assert_tensor_metadata_571 = torch.ops.aten._assert_tensor_metadata.default(view_993, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_571 = None
	        convert_element_type_379: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_993, torch.int8);  view_993 = None
	        view_994: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_379, [sym_size_int, 1500, 1280]);  convert_element_type_379 = None
	        view_995: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_189, [sym_size_int, 1500, 1]);  clamp_min_189 = None
	        view_996: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_378, [sym_size_int, 1500, 1]);  convert_element_type_378 = None
	        _assert_tensor_metadata_572 = torch.ops.aten._assert_tensor_metadata.default(view_994, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_572 = None
	        convert_element_type_380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_994, torch.float32);  view_994 = None
	        _assert_tensor_metadata_573 = torch.ops.aten._assert_tensor_metadata.default(view_996, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_573 = None
	        convert_element_type_381: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_996, torch.float32);  view_996 = None
	        sub_2921: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_380, convert_element_type_381);  convert_element_type_380 = convert_element_type_381 = None
	        mul_6194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2921, view_995);  sub_2921 = view_995 = None
	        view_997: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6194, [sym_size_int, 1500, 1280]);  mul_6194 = None
	        _assert_tensor_metadata_574 = torch.ops.aten._assert_tensor_metadata.default(view_997, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_574 = None
	        view_998: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg289_1, [1280, 40, 32]);  arg289_1 = None
	        view_999: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg290_1, [1280, 40, 1]);  arg290_1 = None
	        view_1000: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg291_1, [1280, 40, 1]);  arg291_1 = None
	        _assert_tensor_metadata_575 = torch.ops.aten._assert_tensor_metadata.default(view_998, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_575 = None
	        convert_element_type_382: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_998, torch.float32);  view_998 = None
	        _assert_tensor_metadata_576 = torch.ops.aten._assert_tensor_metadata.default(view_1000, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_576 = None
	        convert_element_type_383: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1000, torch.float32);  view_1000 = None
	        sub_2925: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_382, convert_element_type_383);  convert_element_type_382 = convert_element_type_383 = None
	        mul_6199: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2925, view_999);  sub_2925 = view_999 = None
	        view_1001: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6199, [1280, 1280]);  mul_6199 = None
	        _assert_tensor_metadata_577 = torch.ops.aten._assert_tensor_metadata.default(view_1001, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_577 = None
	        mul_6204: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1002: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_997, [mul_6204, 1280]);  view_997 = mul_6204 = None
	        permute_108: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1001, [1, 0]);  view_1001 = None
	        addmm_52: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg288_1, view_1002, permute_108);  arg288_1 = view_1002 = permute_108 = None
	        view_1003: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_52, [sym_size_int, 1500, 1280]);  addmm_52 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_85: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1003);  view_1003 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_9826: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_9206, clone_85);  add_9206 = clone_85 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_9826, memory_format = torch.contiguous_format)
	        var_mean_21 = torch.ops.aten.var_mean.correction(clone_86, [2], correction = 0, keepdim = True)
	        getitem_86: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_21[0]
	        getitem_87: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_21[1];  var_mean_21 = None
	        add_9831: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_86, 1e-05);  getitem_86 = None
	        rsqrt_21: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_9831);  add_9831 = None
	        sub_2931: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_86, getitem_87);  clone_86 = getitem_87 = None
	        mul_6215: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2931, rsqrt_21);  sub_2931 = rsqrt_21 = None
	        mul_6216: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6215, arg292_1);  mul_6215 = arg292_1 = None
	        add_9832: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6216, arg293_1);  mul_6216 = arg293_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1004: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_9832, [sym_size_int, 1500, 1280])
	        amin_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1004, [2])
	        amax_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1004, [2]);  view_1004 = None
	        full_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_64, full_128);  amin_64 = full_128 = None
	        full_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_64, full_129);  amax_64 = full_129 = None
	        sub_2942: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_64, minimum_64);  maximum_64 = None
	        div_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2942, 255.0);  sub_2942 = None
	        clamp_min_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_128, 1.1920928955078125e-07);  div_128 = None
	        div_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_64, clamp_min_192);  minimum_64 = None
	        round_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_129);  div_129 = None
	        sub_2948: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_129);  round_129 = None
	        clamp_min_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2948, -128);  sub_2948 = None
	        clamp_max_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_193, 127);  clamp_min_193 = None
	        _assert_tensor_metadata_578 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_192, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_578 = None
	        _assert_tensor_metadata_579 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_128, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_579 = None
	        convert_element_type_384: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_128, torch.int8);  clamp_max_128 = None
	        view_1005: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_9832, [sym_size_int, 1500, 1280]);  add_9832 = None
	        view_1006: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_192, [sym_size_int, 1500, 1])
	        view_1007: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_384, [sym_size_int, 1500, 1])
	        reciprocal_64: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1006);  view_1006 = None
	        mul_6264: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_64, 1.0);  reciprocal_64 = None
	        mul_6267: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1005, mul_6264);  view_1005 = mul_6264 = None
	        round_130: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6267);  mul_6267 = None
	        add_9919: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_130, view_1007);  round_130 = view_1007 = None
	        clamp_min_194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9919, -128);  add_9919 = None
	        clamp_max_129: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_194, 127);  clamp_min_194 = None
	        view_1008: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_129, [sym_size_int, 1500, 1280]);  clamp_max_129 = None
	        _assert_tensor_metadata_580 = torch.ops.aten._assert_tensor_metadata.default(view_1008, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_580 = None
	        convert_element_type_385: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1008, torch.int8);  view_1008 = None
	        view_1009: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_385, [sym_size_int, 1500, 1280]);  convert_element_type_385 = None
	        view_1010: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_192, [sym_size_int, 1500, 1]);  clamp_min_192 = None
	        view_1011: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_384, [sym_size_int, 1500, 1]);  convert_element_type_384 = None
	        _assert_tensor_metadata_581 = torch.ops.aten._assert_tensor_metadata.default(view_1009, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_581 = None
	        convert_element_type_386: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1009, torch.float32);  view_1009 = None
	        _assert_tensor_metadata_582 = torch.ops.aten._assert_tensor_metadata.default(view_1011, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_582 = None
	        convert_element_type_387: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1011, torch.float32);  view_1011 = None
	        sub_2968: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_386, convert_element_type_387);  convert_element_type_386 = convert_element_type_387 = None
	        mul_6289: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2968, view_1010);  sub_2968 = view_1010 = None
	        view_1012: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6289, [sym_size_int, 1500, 1280]);  mul_6289 = None
	        _assert_tensor_metadata_583 = torch.ops.aten._assert_tensor_metadata.default(view_1012, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_583 = None
	        view_1013: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg295_1, [5120, 40, 32]);  arg295_1 = None
	        view_1014: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg296_1, [5120, 40, 1]);  arg296_1 = None
	        view_1015: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg297_1, [5120, 40, 1]);  arg297_1 = None
	        _assert_tensor_metadata_584 = torch.ops.aten._assert_tensor_metadata.default(view_1013, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_584 = None
	        convert_element_type_388: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1013, torch.float32);  view_1013 = None
	        _assert_tensor_metadata_585 = torch.ops.aten._assert_tensor_metadata.default(view_1015, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_585 = None
	        convert_element_type_389: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1015, torch.float32);  view_1015 = None
	        sub_2972: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_388, convert_element_type_389);  convert_element_type_388 = convert_element_type_389 = None
	        mul_6294: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2972, view_1014);  sub_2972 = view_1014 = None
	        view_1016: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6294, [5120, 1280]);  mul_6294 = None
	        _assert_tensor_metadata_586 = torch.ops.aten._assert_tensor_metadata.default(view_1016, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_586 = None
	        mul_6299: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1017: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1012, [mul_6299, 1280]);  view_1012 = mul_6299 = None
	        permute_109: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1016, [1, 0]);  view_1016 = None
	        addmm_53: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg294_1, view_1017, permute_109);  arg294_1 = view_1017 = permute_109 = None
	        view_1018: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_53, [sym_size_int, 1500, 5120]);  addmm_53 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_6306: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1018, 0.5)
	        mul_6307: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1018, 0.7071067811865476);  view_1018 = None
	        erf_12: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_6307);  mul_6307 = None
	        add_9978: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_12, 1);  erf_12 = None
	        mul_6308: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6306, add_9978);  mul_6306 = add_9978 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_87: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_6308);  mul_6308 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1019: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_87, [sym_size_int, 1500, 5120])
	        amin_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1019, [2])
	        amax_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1019, [2]);  view_1019 = None
	        full_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_65, full_130);  amin_65 = full_130 = None
	        full_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_65, full_131);  amax_65 = full_131 = None
	        sub_2985: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_65, minimum_65);  maximum_65 = None
	        div_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2985, 255.0);  sub_2985 = None
	        clamp_min_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_130, 1.1920928955078125e-07);  div_130 = None
	        div_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_65, clamp_min_195);  minimum_65 = None
	        round_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_131);  div_131 = None
	        sub_2991: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_131);  round_131 = None
	        clamp_min_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2991, -128);  sub_2991 = None
	        clamp_max_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_196, 127);  clamp_min_196 = None
	        _assert_tensor_metadata_587 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_195, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_587 = None
	        _assert_tensor_metadata_588 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_130, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_588 = None
	        convert_element_type_390: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_130, torch.int8);  clamp_max_130 = None
	        view_1020: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_87, [sym_size_int, 1500, 5120]);  clone_87 = None
	        view_1021: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_195, [sym_size_int, 1500, 1])
	        view_1022: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_390, [sym_size_int, 1500, 1])
	        reciprocal_65: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1021);  view_1021 = None
	        mul_6354: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_65, 1.0);  reciprocal_65 = None
	        mul_6357: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1020, mul_6354);  view_1020 = mul_6354 = None
	        round_132: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_6357);  mul_6357 = None
	        add_10061: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_132, view_1022);  round_132 = view_1022 = None
	        clamp_min_197: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10061, -128);  add_10061 = None
	        clamp_max_131: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_197, 127);  clamp_min_197 = None
	        view_1023: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_131, [sym_size_int, 1500, 5120]);  clamp_max_131 = None
	        _assert_tensor_metadata_589 = torch.ops.aten._assert_tensor_metadata.default(view_1023, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_589 = None
	        convert_element_type_391: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1023, torch.int8);  view_1023 = None
	        view_1024: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_391, [sym_size_int, 1500, 5120]);  convert_element_type_391 = None
	        view_1025: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_195, [sym_size_int, 1500, 1]);  clamp_min_195 = None
	        view_1026: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_390, [sym_size_int, 1500, 1]);  convert_element_type_390 = None
	        _assert_tensor_metadata_590 = torch.ops.aten._assert_tensor_metadata.default(view_1024, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_590 = None
	        convert_element_type_392: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1024, torch.float32);  view_1024 = None
	        _assert_tensor_metadata_591 = torch.ops.aten._assert_tensor_metadata.default(view_1026, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_591 = None
	        convert_element_type_393: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1026, torch.float32);  view_1026 = None
	        sub_3011: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_392, convert_element_type_393);  convert_element_type_392 = convert_element_type_393 = None
	        mul_6379: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3011, view_1025);  sub_3011 = view_1025 = None
	        view_1027: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_6379, [sym_size_int, 1500, 5120]);  mul_6379 = None
	        _assert_tensor_metadata_592 = torch.ops.aten._assert_tensor_metadata.default(view_1027, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_592 = None
	        view_1028: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg299_1, [1280, 160, 32]);  arg299_1 = None
	        view_1029: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg300_1, [1280, 160, 1]);  arg300_1 = None
	        view_1030: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg301_1, [1280, 160, 1]);  arg301_1 = None
	        _assert_tensor_metadata_593 = torch.ops.aten._assert_tensor_metadata.default(view_1028, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_593 = None
	        convert_element_type_394: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1028, torch.float32);  view_1028 = None
	        _assert_tensor_metadata_594 = torch.ops.aten._assert_tensor_metadata.default(view_1030, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_594 = None
	        convert_element_type_395: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1030, torch.float32);  view_1030 = None
	        sub_3015: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_394, convert_element_type_395);  convert_element_type_394 = convert_element_type_395 = None
	        mul_6384: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3015, view_1029);  sub_3015 = view_1029 = None
	        view_1031: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_6384, [1280, 5120]);  mul_6384 = None
	        _assert_tensor_metadata_595 = torch.ops.aten._assert_tensor_metadata.default(view_1031, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_595 = None
	        mul_6389: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1032: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1027, [mul_6389, 5120]);  view_1027 = mul_6389 = None
	        permute_110: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1031, [1, 0]);  view_1031 = None
	        addmm_54: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg298_1, view_1032, permute_110);  arg298_1 = view_1032 = permute_110 = None
	        view_1033: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_54, [sym_size_int, 1500, 1280]);  addmm_54 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_88: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1033);  view_1033 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_10124: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_9826, clone_88);  add_9826 = clone_88 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_89: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_10124, memory_format = torch.contiguous_format)
	        var_mean_22 = torch.ops.aten.var_mean.correction(clone_89, [2], correction = 0, keepdim = True)
	        getitem_88: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_22[0]
	        getitem_89: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_22[1];  var_mean_22 = None
	        add_10129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_88, 1e-05);  getitem_88 = None
	        rsqrt_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_10129);  add_10129 = None
	        sub_3021: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_89, getitem_89);  clone_89 = getitem_89 = None
	        mul_6400: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3021, rsqrt_22);  sub_3021 = rsqrt_22 = None
	        mul_6401: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6400, arg302_1);  mul_6400 = arg302_1 = None
	        add_10130: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6401, arg303_1);  mul_6401 = arg303_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1034: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_10130, [sym_size_int, 1500, 1280])
	        amin_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1034, [2])
	        amax_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1034, [2]);  view_1034 = None
	        full_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_66, full_132);  amin_66 = full_132 = None
	        full_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_66, full_133);  amax_66 = full_133 = None
	        sub_3032: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_66, minimum_66);  maximum_66 = None
	        div_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3032, 255.0);  sub_3032 = None
	        clamp_min_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_132, 1.1920928955078125e-07);  div_132 = None
	        div_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_66, clamp_min_198);  minimum_66 = None
	        round_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_133);  div_133 = None
	        sub_3038: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_133);  round_133 = None
	        clamp_min_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3038, -128);  sub_3038 = None
	        clamp_max_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_199, 127);  clamp_min_199 = None
	        _assert_tensor_metadata_596 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_198, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_596 = None
	        _assert_tensor_metadata_597 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_132, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_597 = None
	        convert_element_type_396: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_132, torch.int8);  clamp_max_132 = None
	        view_1035: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_10130, [sym_size_int, 1500, 1280])
	        view_1036: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_198, [sym_size_int, 1500, 1])
	        view_1037: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_396, [sym_size_int, 1500, 1])
	        reciprocal_66: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1036);  view_1036 = None
	        mul_6449: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_66, 1.0);  reciprocal_66 = None
	        mul_6452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1035, mul_6449);  view_1035 = mul_6449 = None
	        round_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6452);  mul_6452 = None
	        add_10217: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_134, view_1037);  round_134 = view_1037 = None
	        clamp_min_200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10217, -128);  add_10217 = None
	        clamp_max_133: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_200, 127);  clamp_min_200 = None
	        view_1038: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_133, [sym_size_int, 1500, 1280]);  clamp_max_133 = None
	        _assert_tensor_metadata_598 = torch.ops.aten._assert_tensor_metadata.default(view_1038, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_598 = None
	        convert_element_type_397: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1038, torch.int8);  view_1038 = None
	        view_1039: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_397, [sym_size_int, 1500, 1280]);  convert_element_type_397 = None
	        view_1040: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_198, [sym_size_int, 1500, 1]);  clamp_min_198 = None
	        view_1041: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_396, [sym_size_int, 1500, 1]);  convert_element_type_396 = None
	        _assert_tensor_metadata_599 = torch.ops.aten._assert_tensor_metadata.default(view_1039, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_599 = None
	        convert_element_type_398: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1039, torch.float32);  view_1039 = None
	        _assert_tensor_metadata_600 = torch.ops.aten._assert_tensor_metadata.default(view_1041, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_600 = None
	        convert_element_type_399: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1041, torch.float32);  view_1041 = None
	        sub_3058: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_398, convert_element_type_399);  convert_element_type_398 = convert_element_type_399 = None
	        mul_6474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3058, view_1040);  sub_3058 = view_1040 = None
	        view_1042: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6474, [sym_size_int, 1500, 1280]);  mul_6474 = None
	        _assert_tensor_metadata_601 = torch.ops.aten._assert_tensor_metadata.default(view_1042, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_601 = None
	        view_1043: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg305_1, [1280, 40, 32]);  arg305_1 = None
	        view_1044: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg306_1, [1280, 40, 1]);  arg306_1 = None
	        view_1045: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg307_1, [1280, 40, 1]);  arg307_1 = None
	        _assert_tensor_metadata_602 = torch.ops.aten._assert_tensor_metadata.default(view_1043, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_602 = None
	        convert_element_type_400: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1043, torch.float32);  view_1043 = None
	        _assert_tensor_metadata_603 = torch.ops.aten._assert_tensor_metadata.default(view_1045, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_603 = None
	        convert_element_type_401: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1045, torch.float32);  view_1045 = None
	        sub_3062: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_400, convert_element_type_401);  convert_element_type_400 = convert_element_type_401 = None
	        mul_6479: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3062, view_1044);  sub_3062 = view_1044 = None
	        view_1046: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6479, [1280, 1280]);  mul_6479 = None
	        _assert_tensor_metadata_604 = torch.ops.aten._assert_tensor_metadata.default(view_1046, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_604 = None
	        mul_6484: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1047: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1042, [mul_6484, 1280]);  view_1042 = mul_6484 = None
	        permute_111: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1046, [1, 0]);  view_1046 = None
	        addmm_55: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg304_1, view_1047, permute_111);  arg304_1 = view_1047 = permute_111 = None
	        view_1048: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_55, [sym_size_int, 1500, 1280]);  addmm_55 = None
	        mul_6491: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1048, 0.125);  view_1048 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1049: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_6491, [sym_size_int, 1500, 20, 64]);  mul_6491 = None
	        permute_112: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1049, [0, 2, 1, 3]);  view_1049 = None
	        clone_90: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_112, memory_format = torch.contiguous_format);  permute_112 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1050: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_10130, [sym_size_int, 1500, 1280])
	        amin_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1050, [2])
	        amax_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1050, [2]);  view_1050 = None
	        full_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_67, full_134);  amin_67 = full_134 = None
	        full_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_67, full_135);  amax_67 = full_135 = None
	        sub_3077: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_67, minimum_67);  maximum_67 = None
	        div_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3077, 255.0);  sub_3077 = None
	        clamp_min_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_134, 1.1920928955078125e-07);  div_134 = None
	        div_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_67, clamp_min_201);  minimum_67 = None
	        round_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_135);  div_135 = None
	        sub_3083: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_135);  round_135 = None
	        clamp_min_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3083, -128);  sub_3083 = None
	        clamp_max_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_202, 127);  clamp_min_202 = None
	        _assert_tensor_metadata_605 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_201, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_605 = None
	        _assert_tensor_metadata_606 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_134, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_606 = None
	        convert_element_type_402: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_134, torch.int8);  clamp_max_134 = None
	        view_1051: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_10130, [sym_size_int, 1500, 1280])
	        view_1052: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_201, [sym_size_int, 1500, 1])
	        view_1053: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_402, [sym_size_int, 1500, 1])
	        reciprocal_67: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1052);  view_1052 = None
	        mul_6545: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_67, 1.0);  reciprocal_67 = None
	        mul_6548: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1051, mul_6545);  view_1051 = mul_6545 = None
	        round_136: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6548);  mul_6548 = None
	        add_10369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_136, view_1053);  round_136 = view_1053 = None
	        clamp_min_203: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10369, -128);  add_10369 = None
	        clamp_max_135: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_203, 127);  clamp_min_203 = None
	        view_1054: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_135, [sym_size_int, 1500, 1280]);  clamp_max_135 = None
	        _assert_tensor_metadata_607 = torch.ops.aten._assert_tensor_metadata.default(view_1054, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_607 = None
	        convert_element_type_403: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1054, torch.int8);  view_1054 = None
	        view_1055: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_403, [sym_size_int, 1500, 1280]);  convert_element_type_403 = None
	        view_1056: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_201, [sym_size_int, 1500, 1]);  clamp_min_201 = None
	        view_1057: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_402, [sym_size_int, 1500, 1]);  convert_element_type_402 = None
	        _assert_tensor_metadata_608 = torch.ops.aten._assert_tensor_metadata.default(view_1055, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_608 = None
	        convert_element_type_404: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1055, torch.float32);  view_1055 = None
	        _assert_tensor_metadata_609 = torch.ops.aten._assert_tensor_metadata.default(view_1057, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_609 = None
	        convert_element_type_405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1057, torch.float32);  view_1057 = None
	        sub_3103: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_404, convert_element_type_405);  convert_element_type_404 = convert_element_type_405 = None
	        mul_6570: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3103, view_1056);  sub_3103 = view_1056 = None
	        view_1058: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6570, [sym_size_int, 1500, 1280]);  mul_6570 = None
	        _assert_tensor_metadata_610 = torch.ops.aten._assert_tensor_metadata.default(view_1058, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_610 = None
	        view_1059: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg308_1, [1280, 40, 32]);  arg308_1 = None
	        view_1060: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg309_1, [1280, 40, 1]);  arg309_1 = None
	        view_1061: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg310_1, [1280, 40, 1]);  arg310_1 = None
	        _assert_tensor_metadata_611 = torch.ops.aten._assert_tensor_metadata.default(view_1059, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_611 = None
	        convert_element_type_406: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1059, torch.float32);  view_1059 = None
	        _assert_tensor_metadata_612 = torch.ops.aten._assert_tensor_metadata.default(view_1061, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_612 = None
	        convert_element_type_407: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1061, torch.float32);  view_1061 = None
	        sub_3107: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_406, convert_element_type_407);  convert_element_type_406 = convert_element_type_407 = None
	        mul_6575: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3107, view_1060);  sub_3107 = view_1060 = None
	        view_1062: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6575, [1280, 1280]);  mul_6575 = None
	        _assert_tensor_metadata_613 = torch.ops.aten._assert_tensor_metadata.default(view_1062, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_613 = None
	        permute_113: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1062, [1, 0]);  view_1062 = None
	        mul_6578: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1063: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1058, [mul_6578, 1280]);  view_1058 = mul_6578 = None
	        mm_11: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1063, permute_113);  view_1063 = permute_113 = None
	        view_1064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_11, [sym_size_int, 1500, 1280]);  mm_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1065: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1064, [sym_size_int, -1, 20, 64]);  view_1064 = None
	        permute_114: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1065, [0, 2, 1, 3]);  view_1065 = None
	        clone_91: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_114, memory_format = torch.contiguous_format);  permute_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1066: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_10130, [sym_size_int, 1500, 1280])
	        amin_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1066, [2])
	        amax_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1066, [2]);  view_1066 = None
	        full_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_68, full_136);  amin_68 = full_136 = None
	        full_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_68, full_137);  amax_68 = full_137 = None
	        sub_3121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_68, minimum_68);  maximum_68 = None
	        div_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3121, 255.0);  sub_3121 = None
	        clamp_min_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_136, 1.1920928955078125e-07);  div_136 = None
	        div_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_68, clamp_min_204);  minimum_68 = None
	        round_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_137);  div_137 = None
	        sub_3127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_137);  round_137 = None
	        clamp_min_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3127, -128);  sub_3127 = None
	        clamp_max_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_205, 127);  clamp_min_205 = None
	        _assert_tensor_metadata_614 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_204, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_614 = None
	        _assert_tensor_metadata_615 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_136, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_615 = None
	        convert_element_type_408: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_136, torch.int8);  clamp_max_136 = None
	        view_1067: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_10130, [sym_size_int, 1500, 1280]);  add_10130 = None
	        view_1068: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_204, [sym_size_int, 1500, 1])
	        view_1069: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_408, [sym_size_int, 1500, 1])
	        reciprocal_68: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1068);  view_1068 = None
	        mul_6644: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_68, 1.0);  reciprocal_68 = None
	        mul_6647: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1067, mul_6644);  view_1067 = mul_6644 = None
	        round_138: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6647);  mul_6647 = None
	        add_10517: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_138, view_1069);  round_138 = view_1069 = None
	        clamp_min_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10517, -128);  add_10517 = None
	        clamp_max_137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_206, 127);  clamp_min_206 = None
	        view_1070: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_137, [sym_size_int, 1500, 1280]);  clamp_max_137 = None
	        _assert_tensor_metadata_616 = torch.ops.aten._assert_tensor_metadata.default(view_1070, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_616 = None
	        convert_element_type_409: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1070, torch.int8);  view_1070 = None
	        view_1071: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_409, [sym_size_int, 1500, 1280]);  convert_element_type_409 = None
	        view_1072: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_204, [sym_size_int, 1500, 1]);  clamp_min_204 = None
	        view_1073: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_408, [sym_size_int, 1500, 1]);  convert_element_type_408 = None
	        _assert_tensor_metadata_617 = torch.ops.aten._assert_tensor_metadata.default(view_1071, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_617 = None
	        convert_element_type_410: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1071, torch.float32);  view_1071 = None
	        _assert_tensor_metadata_618 = torch.ops.aten._assert_tensor_metadata.default(view_1073, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_618 = None
	        convert_element_type_411: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1073, torch.float32);  view_1073 = None
	        sub_3147: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_410, convert_element_type_411);  convert_element_type_410 = convert_element_type_411 = None
	        mul_6669: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3147, view_1072);  sub_3147 = view_1072 = None
	        view_1074: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6669, [sym_size_int, 1500, 1280]);  mul_6669 = None
	        _assert_tensor_metadata_619 = torch.ops.aten._assert_tensor_metadata.default(view_1074, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_619 = None
	        view_1075: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg312_1, [1280, 40, 32]);  arg312_1 = None
	        view_1076: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg313_1, [1280, 40, 1]);  arg313_1 = None
	        view_1077: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg314_1, [1280, 40, 1]);  arg314_1 = None
	        _assert_tensor_metadata_620 = torch.ops.aten._assert_tensor_metadata.default(view_1075, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_620 = None
	        convert_element_type_412: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1075, torch.float32);  view_1075 = None
	        _assert_tensor_metadata_621 = torch.ops.aten._assert_tensor_metadata.default(view_1077, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_621 = None
	        convert_element_type_413: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1077, torch.float32);  view_1077 = None
	        sub_3151: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_412, convert_element_type_413);  convert_element_type_412 = convert_element_type_413 = None
	        mul_6674: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3151, view_1076);  sub_3151 = view_1076 = None
	        view_1078: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6674, [1280, 1280]);  mul_6674 = None
	        _assert_tensor_metadata_622 = torch.ops.aten._assert_tensor_metadata.default(view_1078, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_622 = None
	        mul_6679: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1079: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1074, [mul_6679, 1280]);  view_1074 = mul_6679 = None
	        permute_115: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1078, [1, 0]);  view_1078 = None
	        addmm_56: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg311_1, view_1079, permute_115);  arg311_1 = view_1079 = permute_115 = None
	        view_1080: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_56, [sym_size_int, 1500, 1280]);  addmm_56 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1081: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1080, [sym_size_int, -1, 20, 64]);  view_1080 = None
	        permute_116: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1081, [0, 2, 1, 3]);  view_1081 = None
	        clone_92: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_116, memory_format = torch.contiguous_format);  permute_116 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_11 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_90, clone_91, clone_92, None, False, scale = 1.0);  clone_90 = clone_91 = clone_92 = None
	        getitem_90: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_11[0];  _scaled_dot_product_efficient_attention_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_117: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_90, [0, 2, 1, 3]);  getitem_90 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1082: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_117, [sym_size_int, 1500, -1]);  permute_117 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1083: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1082, [sym_size_int, 1500, 1280])
	        amin_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1083, [2])
	        amax_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1083, [2]);  view_1083 = None
	        full_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_69, full_138);  amin_69 = full_138 = None
	        full_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_69, full_139);  amax_69 = full_139 = None
	        sub_3169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_69, minimum_69);  maximum_69 = None
	        div_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3169, 255.0);  sub_3169 = None
	        clamp_min_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_138, 1.1920928955078125e-07);  div_138 = None
	        div_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_69, clamp_min_207);  minimum_69 = None
	        round_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_139);  div_139 = None
	        sub_3175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_139);  round_139 = None
	        clamp_min_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3175, -128);  sub_3175 = None
	        clamp_max_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_208, 127);  clamp_min_208 = None
	        _assert_tensor_metadata_623 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_207, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_623 = None
	        _assert_tensor_metadata_624 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_138, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_624 = None
	        convert_element_type_414: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_138, torch.int8);  clamp_max_138 = None
	        view_1084: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1082, [sym_size_int, 1500, 1280]);  view_1082 = None
	        view_1085: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_207, [sym_size_int, 1500, 1])
	        view_1086: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_414, [sym_size_int, 1500, 1])
	        reciprocal_69: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1085);  view_1085 = None
	        mul_6749: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_69, 1.0);  reciprocal_69 = None
	        mul_6752: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1084, mul_6749);  view_1084 = mul_6749 = None
	        round_140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6752);  mul_6752 = None
	        add_10681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_140, view_1086);  round_140 = view_1086 = None
	        clamp_min_209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10681, -128);  add_10681 = None
	        clamp_max_139: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_209, 127);  clamp_min_209 = None
	        view_1087: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_139, [sym_size_int, 1500, 1280]);  clamp_max_139 = None
	        _assert_tensor_metadata_625 = torch.ops.aten._assert_tensor_metadata.default(view_1087, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_625 = None
	        convert_element_type_415: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1087, torch.int8);  view_1087 = None
	        view_1088: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_415, [sym_size_int, 1500, 1280]);  convert_element_type_415 = None
	        view_1089: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_207, [sym_size_int, 1500, 1]);  clamp_min_207 = None
	        view_1090: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_414, [sym_size_int, 1500, 1]);  convert_element_type_414 = None
	        _assert_tensor_metadata_626 = torch.ops.aten._assert_tensor_metadata.default(view_1088, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_626 = None
	        convert_element_type_416: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1088, torch.float32);  view_1088 = None
	        _assert_tensor_metadata_627 = torch.ops.aten._assert_tensor_metadata.default(view_1090, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_627 = None
	        convert_element_type_417: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1090, torch.float32);  view_1090 = None
	        sub_3195: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_416, convert_element_type_417);  convert_element_type_416 = convert_element_type_417 = None
	        mul_6774: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3195, view_1089);  sub_3195 = view_1089 = None
	        view_1091: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6774, [sym_size_int, 1500, 1280]);  mul_6774 = None
	        _assert_tensor_metadata_628 = torch.ops.aten._assert_tensor_metadata.default(view_1091, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_628 = None
	        view_1092: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg316_1, [1280, 40, 32]);  arg316_1 = None
	        view_1093: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg317_1, [1280, 40, 1]);  arg317_1 = None
	        view_1094: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg318_1, [1280, 40, 1]);  arg318_1 = None
	        _assert_tensor_metadata_629 = torch.ops.aten._assert_tensor_metadata.default(view_1092, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_629 = None
	        convert_element_type_418: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1092, torch.float32);  view_1092 = None
	        _assert_tensor_metadata_630 = torch.ops.aten._assert_tensor_metadata.default(view_1094, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_630 = None
	        convert_element_type_419: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1094, torch.float32);  view_1094 = None
	        sub_3199: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_418, convert_element_type_419);  convert_element_type_418 = convert_element_type_419 = None
	        mul_6779: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3199, view_1093);  sub_3199 = view_1093 = None
	        view_1095: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6779, [1280, 1280]);  mul_6779 = None
	        _assert_tensor_metadata_631 = torch.ops.aten._assert_tensor_metadata.default(view_1095, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_631 = None
	        mul_6784: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1096: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1091, [mul_6784, 1280]);  view_1091 = mul_6784 = None
	        permute_118: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1095, [1, 0]);  view_1095 = None
	        addmm_57: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg315_1, view_1096, permute_118);  arg315_1 = view_1096 = permute_118 = None
	        view_1097: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_57, [sym_size_int, 1500, 1280]);  addmm_57 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_93: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1097);  view_1097 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_10744: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_10124, clone_93);  add_10124 = clone_93 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_94: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_10744, memory_format = torch.contiguous_format)
	        var_mean_23 = torch.ops.aten.var_mean.correction(clone_94, [2], correction = 0, keepdim = True)
	        getitem_94: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_23[0]
	        getitem_95: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_23[1];  var_mean_23 = None
	        add_10749: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_94, 1e-05);  getitem_94 = None
	        rsqrt_23: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_10749);  add_10749 = None
	        sub_3205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_94, getitem_95);  clone_94 = getitem_95 = None
	        mul_6795: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3205, rsqrt_23);  sub_3205 = rsqrt_23 = None
	        mul_6796: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6795, arg319_1);  mul_6795 = arg319_1 = None
	        add_10750: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6796, arg320_1);  mul_6796 = arg320_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1098: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_10750, [sym_size_int, 1500, 1280])
	        amin_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1098, [2])
	        amax_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1098, [2]);  view_1098 = None
	        full_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_70, full_140);  amin_70 = full_140 = None
	        full_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_70, full_141);  amax_70 = full_141 = None
	        sub_3216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_70, minimum_70);  maximum_70 = None
	        div_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3216, 255.0);  sub_3216 = None
	        clamp_min_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_140, 1.1920928955078125e-07);  div_140 = None
	        div_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_70, clamp_min_210);  minimum_70 = None
	        round_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_141);  div_141 = None
	        sub_3222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_141);  round_141 = None
	        clamp_min_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3222, -128);  sub_3222 = None
	        clamp_max_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_211, 127);  clamp_min_211 = None
	        _assert_tensor_metadata_632 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_210, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_632 = None
	        _assert_tensor_metadata_633 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_140, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_633 = None
	        convert_element_type_420: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_140, torch.int8);  clamp_max_140 = None
	        view_1099: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_10750, [sym_size_int, 1500, 1280]);  add_10750 = None
	        view_1100: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_210, [sym_size_int, 1500, 1])
	        view_1101: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_420, [sym_size_int, 1500, 1])
	        reciprocal_70: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1100);  view_1100 = None
	        mul_6844: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_70, 1.0);  reciprocal_70 = None
	        mul_6847: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1099, mul_6844);  view_1099 = mul_6844 = None
	        round_142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6847);  mul_6847 = None
	        add_10837: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_142, view_1101);  round_142 = view_1101 = None
	        clamp_min_212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10837, -128);  add_10837 = None
	        clamp_max_141: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_212, 127);  clamp_min_212 = None
	        view_1102: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_141, [sym_size_int, 1500, 1280]);  clamp_max_141 = None
	        _assert_tensor_metadata_634 = torch.ops.aten._assert_tensor_metadata.default(view_1102, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_634 = None
	        convert_element_type_421: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1102, torch.int8);  view_1102 = None
	        view_1103: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_421, [sym_size_int, 1500, 1280]);  convert_element_type_421 = None
	        view_1104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_210, [sym_size_int, 1500, 1]);  clamp_min_210 = None
	        view_1105: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_420, [sym_size_int, 1500, 1]);  convert_element_type_420 = None
	        _assert_tensor_metadata_635 = torch.ops.aten._assert_tensor_metadata.default(view_1103, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_635 = None
	        convert_element_type_422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1103, torch.float32);  view_1103 = None
	        _assert_tensor_metadata_636 = torch.ops.aten._assert_tensor_metadata.default(view_1105, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_636 = None
	        convert_element_type_423: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1105, torch.float32);  view_1105 = None
	        sub_3242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_422, convert_element_type_423);  convert_element_type_422 = convert_element_type_423 = None
	        mul_6869: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3242, view_1104);  sub_3242 = view_1104 = None
	        view_1106: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6869, [sym_size_int, 1500, 1280]);  mul_6869 = None
	        _assert_tensor_metadata_637 = torch.ops.aten._assert_tensor_metadata.default(view_1106, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_637 = None
	        view_1107: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg322_1, [5120, 40, 32]);  arg322_1 = None
	        view_1108: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg323_1, [5120, 40, 1]);  arg323_1 = None
	        view_1109: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg324_1, [5120, 40, 1]);  arg324_1 = None
	        _assert_tensor_metadata_638 = torch.ops.aten._assert_tensor_metadata.default(view_1107, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_638 = None
	        convert_element_type_424: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1107, torch.float32);  view_1107 = None
	        _assert_tensor_metadata_639 = torch.ops.aten._assert_tensor_metadata.default(view_1109, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_639 = None
	        convert_element_type_425: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1109, torch.float32);  view_1109 = None
	        sub_3246: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_424, convert_element_type_425);  convert_element_type_424 = convert_element_type_425 = None
	        mul_6874: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3246, view_1108);  sub_3246 = view_1108 = None
	        view_1110: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6874, [5120, 1280]);  mul_6874 = None
	        _assert_tensor_metadata_640 = torch.ops.aten._assert_tensor_metadata.default(view_1110, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_640 = None
	        mul_6879: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1111: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1106, [mul_6879, 1280]);  view_1106 = mul_6879 = None
	        permute_119: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1110, [1, 0]);  view_1110 = None
	        addmm_58: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg321_1, view_1111, permute_119);  arg321_1 = view_1111 = permute_119 = None
	        view_1112: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_58, [sym_size_int, 1500, 5120]);  addmm_58 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_6886: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1112, 0.5)
	        mul_6887: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1112, 0.7071067811865476);  view_1112 = None
	        erf_13: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_6887);  mul_6887 = None
	        add_10896: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_13, 1);  erf_13 = None
	        mul_6888: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6886, add_10896);  mul_6886 = add_10896 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_95: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_6888);  mul_6888 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1113: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_95, [sym_size_int, 1500, 5120])
	        amin_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1113, [2])
	        amax_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1113, [2]);  view_1113 = None
	        full_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_71, full_142);  amin_71 = full_142 = None
	        full_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_71, full_143);  amax_71 = full_143 = None
	        sub_3259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_71, minimum_71);  maximum_71 = None
	        div_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3259, 255.0);  sub_3259 = None
	        clamp_min_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_142, 1.1920928955078125e-07);  div_142 = None
	        div_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_71, clamp_min_213);  minimum_71 = None
	        round_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_143);  div_143 = None
	        sub_3265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_143);  round_143 = None
	        clamp_min_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3265, -128);  sub_3265 = None
	        clamp_max_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_214, 127);  clamp_min_214 = None
	        _assert_tensor_metadata_641 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_213, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_641 = None
	        _assert_tensor_metadata_642 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_142, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_642 = None
	        convert_element_type_426: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_142, torch.int8);  clamp_max_142 = None
	        view_1114: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_95, [sym_size_int, 1500, 5120]);  clone_95 = None
	        view_1115: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_213, [sym_size_int, 1500, 1])
	        view_1116: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_426, [sym_size_int, 1500, 1])
	        reciprocal_71: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1115);  view_1115 = None
	        mul_6934: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_71, 1.0);  reciprocal_71 = None
	        mul_6937: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1114, mul_6934);  view_1114 = mul_6934 = None
	        round_144: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_6937);  mul_6937 = None
	        add_10979: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_144, view_1116);  round_144 = view_1116 = None
	        clamp_min_215: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10979, -128);  add_10979 = None
	        clamp_max_143: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_215, 127);  clamp_min_215 = None
	        view_1117: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_143, [sym_size_int, 1500, 5120]);  clamp_max_143 = None
	        _assert_tensor_metadata_643 = torch.ops.aten._assert_tensor_metadata.default(view_1117, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_643 = None
	        convert_element_type_427: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1117, torch.int8);  view_1117 = None
	        view_1118: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_427, [sym_size_int, 1500, 5120]);  convert_element_type_427 = None
	        view_1119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_213, [sym_size_int, 1500, 1]);  clamp_min_213 = None
	        view_1120: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_426, [sym_size_int, 1500, 1]);  convert_element_type_426 = None
	        _assert_tensor_metadata_644 = torch.ops.aten._assert_tensor_metadata.default(view_1118, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_644 = None
	        convert_element_type_428: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1118, torch.float32);  view_1118 = None
	        _assert_tensor_metadata_645 = torch.ops.aten._assert_tensor_metadata.default(view_1120, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_645 = None
	        convert_element_type_429: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1120, torch.float32);  view_1120 = None
	        sub_3285: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_428, convert_element_type_429);  convert_element_type_428 = convert_element_type_429 = None
	        mul_6959: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3285, view_1119);  sub_3285 = view_1119 = None
	        view_1121: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_6959, [sym_size_int, 1500, 5120]);  mul_6959 = None
	        _assert_tensor_metadata_646 = torch.ops.aten._assert_tensor_metadata.default(view_1121, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_646 = None
	        view_1122: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg326_1, [1280, 160, 32]);  arg326_1 = None
	        view_1123: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg327_1, [1280, 160, 1]);  arg327_1 = None
	        view_1124: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg328_1, [1280, 160, 1]);  arg328_1 = None
	        _assert_tensor_metadata_647 = torch.ops.aten._assert_tensor_metadata.default(view_1122, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_647 = None
	        convert_element_type_430: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1122, torch.float32);  view_1122 = None
	        _assert_tensor_metadata_648 = torch.ops.aten._assert_tensor_metadata.default(view_1124, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_648 = None
	        convert_element_type_431: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1124, torch.float32);  view_1124 = None
	        sub_3289: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_430, convert_element_type_431);  convert_element_type_430 = convert_element_type_431 = None
	        mul_6964: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3289, view_1123);  sub_3289 = view_1123 = None
	        view_1125: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_6964, [1280, 5120]);  mul_6964 = None
	        _assert_tensor_metadata_649 = torch.ops.aten._assert_tensor_metadata.default(view_1125, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_649 = None
	        mul_6969: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1126: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1121, [mul_6969, 5120]);  view_1121 = mul_6969 = None
	        permute_120: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1125, [1, 0]);  view_1125 = None
	        addmm_59: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg325_1, view_1126, permute_120);  arg325_1 = view_1126 = permute_120 = None
	        view_1127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_59, [sym_size_int, 1500, 1280]);  addmm_59 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_96: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1127);  view_1127 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_11042: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_10744, clone_96);  add_10744 = clone_96 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_97: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_11042, memory_format = torch.contiguous_format)
	        var_mean_24 = torch.ops.aten.var_mean.correction(clone_97, [2], correction = 0, keepdim = True)
	        getitem_96: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_24[0]
	        getitem_97: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_24[1];  var_mean_24 = None
	        add_11047: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_96, 1e-05);  getitem_96 = None
	        rsqrt_24: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_11047);  add_11047 = None
	        sub_3295: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_97, getitem_97);  clone_97 = getitem_97 = None
	        mul_6980: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3295, rsqrt_24);  sub_3295 = rsqrt_24 = None
	        mul_6981: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6980, arg329_1);  mul_6980 = arg329_1 = None
	        add_11048: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6981, arg330_1);  mul_6981 = arg330_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11048, [sym_size_int, 1500, 1280])
	        amin_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1128, [2])
	        amax_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1128, [2]);  view_1128 = None
	        full_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_72, full_144);  amin_72 = full_144 = None
	        full_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_72, full_145);  amax_72 = full_145 = None
	        sub_3306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_72, minimum_72);  maximum_72 = None
	        div_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3306, 255.0);  sub_3306 = None
	        clamp_min_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_144, 1.1920928955078125e-07);  div_144 = None
	        div_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_72, clamp_min_216);  minimum_72 = None
	        round_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_145);  div_145 = None
	        sub_3312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_145);  round_145 = None
	        clamp_min_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3312, -128);  sub_3312 = None
	        clamp_max_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_217, 127);  clamp_min_217 = None
	        _assert_tensor_metadata_650 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_216, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_650 = None
	        _assert_tensor_metadata_651 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_144, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_651 = None
	        convert_element_type_432: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_144, torch.int8);  clamp_max_144 = None
	        view_1129: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11048, [sym_size_int, 1500, 1280])
	        view_1130: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_216, [sym_size_int, 1500, 1])
	        view_1131: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_432, [sym_size_int, 1500, 1])
	        reciprocal_72: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1130);  view_1130 = None
	        mul_7029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_72, 1.0);  reciprocal_72 = None
	        mul_7032: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1129, mul_7029);  view_1129 = mul_7029 = None
	        round_146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7032);  mul_7032 = None
	        add_11135: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_146, view_1131);  round_146 = view_1131 = None
	        clamp_min_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11135, -128);  add_11135 = None
	        clamp_max_145: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_218, 127);  clamp_min_218 = None
	        view_1132: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_145, [sym_size_int, 1500, 1280]);  clamp_max_145 = None
	        _assert_tensor_metadata_652 = torch.ops.aten._assert_tensor_metadata.default(view_1132, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_652 = None
	        convert_element_type_433: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1132, torch.int8);  view_1132 = None
	        view_1133: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_433, [sym_size_int, 1500, 1280]);  convert_element_type_433 = None
	        view_1134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_216, [sym_size_int, 1500, 1]);  clamp_min_216 = None
	        view_1135: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_432, [sym_size_int, 1500, 1]);  convert_element_type_432 = None
	        _assert_tensor_metadata_653 = torch.ops.aten._assert_tensor_metadata.default(view_1133, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_653 = None
	        convert_element_type_434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1133, torch.float32);  view_1133 = None
	        _assert_tensor_metadata_654 = torch.ops.aten._assert_tensor_metadata.default(view_1135, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_654 = None
	        convert_element_type_435: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1135, torch.float32);  view_1135 = None
	        sub_3332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_434, convert_element_type_435);  convert_element_type_434 = convert_element_type_435 = None
	        mul_7054: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3332, view_1134);  sub_3332 = view_1134 = None
	        view_1136: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7054, [sym_size_int, 1500, 1280]);  mul_7054 = None
	        _assert_tensor_metadata_655 = torch.ops.aten._assert_tensor_metadata.default(view_1136, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_655 = None
	        view_1137: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg332_1, [1280, 40, 32]);  arg332_1 = None
	        view_1138: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg333_1, [1280, 40, 1]);  arg333_1 = None
	        view_1139: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg334_1, [1280, 40, 1]);  arg334_1 = None
	        _assert_tensor_metadata_656 = torch.ops.aten._assert_tensor_metadata.default(view_1137, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_656 = None
	        convert_element_type_436: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1137, torch.float32);  view_1137 = None
	        _assert_tensor_metadata_657 = torch.ops.aten._assert_tensor_metadata.default(view_1139, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_657 = None
	        convert_element_type_437: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1139, torch.float32);  view_1139 = None
	        sub_3336: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_436, convert_element_type_437);  convert_element_type_436 = convert_element_type_437 = None
	        mul_7059: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3336, view_1138);  sub_3336 = view_1138 = None
	        view_1140: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7059, [1280, 1280]);  mul_7059 = None
	        _assert_tensor_metadata_658 = torch.ops.aten._assert_tensor_metadata.default(view_1140, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_658 = None
	        mul_7064: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1141: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1136, [mul_7064, 1280]);  view_1136 = mul_7064 = None
	        permute_121: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1140, [1, 0]);  view_1140 = None
	        addmm_60: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg331_1, view_1141, permute_121);  arg331_1 = view_1141 = permute_121 = None
	        view_1142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_60, [sym_size_int, 1500, 1280]);  addmm_60 = None
	        mul_7071: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1142, 0.125);  view_1142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1143: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_7071, [sym_size_int, 1500, 20, 64]);  mul_7071 = None
	        permute_122: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1143, [0, 2, 1, 3]);  view_1143 = None
	        clone_98: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_122, memory_format = torch.contiguous_format);  permute_122 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1144: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11048, [sym_size_int, 1500, 1280])
	        amin_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1144, [2])
	        amax_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1144, [2]);  view_1144 = None
	        full_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_73, full_146);  amin_73 = full_146 = None
	        full_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_73, full_147);  amax_73 = full_147 = None
	        sub_3351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_73, minimum_73);  maximum_73 = None
	        div_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3351, 255.0);  sub_3351 = None
	        clamp_min_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_146, 1.1920928955078125e-07);  div_146 = None
	        div_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_73, clamp_min_219);  minimum_73 = None
	        round_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_147);  div_147 = None
	        sub_3357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_147);  round_147 = None
	        clamp_min_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3357, -128);  sub_3357 = None
	        clamp_max_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_220, 127);  clamp_min_220 = None
	        _assert_tensor_metadata_659 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_219, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_659 = None
	        _assert_tensor_metadata_660 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_146, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_660 = None
	        convert_element_type_438: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_146, torch.int8);  clamp_max_146 = None
	        view_1145: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11048, [sym_size_int, 1500, 1280])
	        view_1146: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_219, [sym_size_int, 1500, 1])
	        view_1147: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_438, [sym_size_int, 1500, 1])
	        reciprocal_73: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1146);  view_1146 = None
	        mul_7125: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_73, 1.0);  reciprocal_73 = None
	        mul_7128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1145, mul_7125);  view_1145 = mul_7125 = None
	        round_148: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7128);  mul_7128 = None
	        add_11287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_148, view_1147);  round_148 = view_1147 = None
	        clamp_min_221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11287, -128);  add_11287 = None
	        clamp_max_147: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_221, 127);  clamp_min_221 = None
	        view_1148: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_147, [sym_size_int, 1500, 1280]);  clamp_max_147 = None
	        _assert_tensor_metadata_661 = torch.ops.aten._assert_tensor_metadata.default(view_1148, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_661 = None
	        convert_element_type_439: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1148, torch.int8);  view_1148 = None
	        view_1149: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_439, [sym_size_int, 1500, 1280]);  convert_element_type_439 = None
	        view_1150: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_219, [sym_size_int, 1500, 1]);  clamp_min_219 = None
	        view_1151: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_438, [sym_size_int, 1500, 1]);  convert_element_type_438 = None
	        _assert_tensor_metadata_662 = torch.ops.aten._assert_tensor_metadata.default(view_1149, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_662 = None
	        convert_element_type_440: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1149, torch.float32);  view_1149 = None
	        _assert_tensor_metadata_663 = torch.ops.aten._assert_tensor_metadata.default(view_1151, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_663 = None
	        convert_element_type_441: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1151, torch.float32);  view_1151 = None
	        sub_3377: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_440, convert_element_type_441);  convert_element_type_440 = convert_element_type_441 = None
	        mul_7150: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3377, view_1150);  sub_3377 = view_1150 = None
	        view_1152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7150, [sym_size_int, 1500, 1280]);  mul_7150 = None
	        _assert_tensor_metadata_664 = torch.ops.aten._assert_tensor_metadata.default(view_1152, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_664 = None
	        view_1153: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg335_1, [1280, 40, 32]);  arg335_1 = None
	        view_1154: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg336_1, [1280, 40, 1]);  arg336_1 = None
	        view_1155: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg337_1, [1280, 40, 1]);  arg337_1 = None
	        _assert_tensor_metadata_665 = torch.ops.aten._assert_tensor_metadata.default(view_1153, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_665 = None
	        convert_element_type_442: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1153, torch.float32);  view_1153 = None
	        _assert_tensor_metadata_666 = torch.ops.aten._assert_tensor_metadata.default(view_1155, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_666 = None
	        convert_element_type_443: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1155, torch.float32);  view_1155 = None
	        sub_3381: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_442, convert_element_type_443);  convert_element_type_442 = convert_element_type_443 = None
	        mul_7155: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3381, view_1154);  sub_3381 = view_1154 = None
	        view_1156: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7155, [1280, 1280]);  mul_7155 = None
	        _assert_tensor_metadata_667 = torch.ops.aten._assert_tensor_metadata.default(view_1156, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_667 = None
	        permute_123: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1156, [1, 0]);  view_1156 = None
	        mul_7158: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1157: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1152, [mul_7158, 1280]);  view_1152 = mul_7158 = None
	        mm_12: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1157, permute_123);  view_1157 = permute_123 = None
	        view_1158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_12, [sym_size_int, 1500, 1280]);  mm_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1159: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1158, [sym_size_int, -1, 20, 64]);  view_1158 = None
	        permute_124: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1159, [0, 2, 1, 3]);  view_1159 = None
	        clone_99: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_124, memory_format = torch.contiguous_format);  permute_124 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11048, [sym_size_int, 1500, 1280])
	        amin_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1160, [2])
	        amax_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1160, [2]);  view_1160 = None
	        full_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_74, full_148);  amin_74 = full_148 = None
	        full_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_74, full_149);  amax_74 = full_149 = None
	        sub_3395: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_74, minimum_74);  maximum_74 = None
	        div_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3395, 255.0);  sub_3395 = None
	        clamp_min_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_148, 1.1920928955078125e-07);  div_148 = None
	        div_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_74, clamp_min_222);  minimum_74 = None
	        round_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_149);  div_149 = None
	        sub_3401: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_149);  round_149 = None
	        clamp_min_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3401, -128);  sub_3401 = None
	        clamp_max_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_223, 127);  clamp_min_223 = None
	        _assert_tensor_metadata_668 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_222, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_668 = None
	        _assert_tensor_metadata_669 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_148, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_669 = None
	        convert_element_type_444: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_148, torch.int8);  clamp_max_148 = None
	        view_1161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11048, [sym_size_int, 1500, 1280]);  add_11048 = None
	        view_1162: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_222, [sym_size_int, 1500, 1])
	        view_1163: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_444, [sym_size_int, 1500, 1])
	        reciprocal_74: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1162);  view_1162 = None
	        mul_7224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_74, 1.0);  reciprocal_74 = None
	        mul_7227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1161, mul_7224);  view_1161 = mul_7224 = None
	        round_150: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7227);  mul_7227 = None
	        add_11435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_150, view_1163);  round_150 = view_1163 = None
	        clamp_min_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11435, -128);  add_11435 = None
	        clamp_max_149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_224, 127);  clamp_min_224 = None
	        view_1164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_149, [sym_size_int, 1500, 1280]);  clamp_max_149 = None
	        _assert_tensor_metadata_670 = torch.ops.aten._assert_tensor_metadata.default(view_1164, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_670 = None
	        convert_element_type_445: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1164, torch.int8);  view_1164 = None
	        view_1165: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_445, [sym_size_int, 1500, 1280]);  convert_element_type_445 = None
	        view_1166: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_222, [sym_size_int, 1500, 1]);  clamp_min_222 = None
	        view_1167: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_444, [sym_size_int, 1500, 1]);  convert_element_type_444 = None
	        _assert_tensor_metadata_671 = torch.ops.aten._assert_tensor_metadata.default(view_1165, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_671 = None
	        convert_element_type_446: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1165, torch.float32);  view_1165 = None
	        _assert_tensor_metadata_672 = torch.ops.aten._assert_tensor_metadata.default(view_1167, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_672 = None
	        convert_element_type_447: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1167, torch.float32);  view_1167 = None
	        sub_3421: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_446, convert_element_type_447);  convert_element_type_446 = convert_element_type_447 = None
	        mul_7249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3421, view_1166);  sub_3421 = view_1166 = None
	        view_1168: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7249, [sym_size_int, 1500, 1280]);  mul_7249 = None
	        _assert_tensor_metadata_673 = torch.ops.aten._assert_tensor_metadata.default(view_1168, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_673 = None
	        view_1169: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg339_1, [1280, 40, 32]);  arg339_1 = None
	        view_1170: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg340_1, [1280, 40, 1]);  arg340_1 = None
	        view_1171: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg341_1, [1280, 40, 1]);  arg341_1 = None
	        _assert_tensor_metadata_674 = torch.ops.aten._assert_tensor_metadata.default(view_1169, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_674 = None
	        convert_element_type_448: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1169, torch.float32);  view_1169 = None
	        _assert_tensor_metadata_675 = torch.ops.aten._assert_tensor_metadata.default(view_1171, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_675 = None
	        convert_element_type_449: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1171, torch.float32);  view_1171 = None
	        sub_3425: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_448, convert_element_type_449);  convert_element_type_448 = convert_element_type_449 = None
	        mul_7254: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3425, view_1170);  sub_3425 = view_1170 = None
	        view_1172: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7254, [1280, 1280]);  mul_7254 = None
	        _assert_tensor_metadata_676 = torch.ops.aten._assert_tensor_metadata.default(view_1172, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_676 = None
	        mul_7259: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1173: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1168, [mul_7259, 1280]);  view_1168 = mul_7259 = None
	        permute_125: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1172, [1, 0]);  view_1172 = None
	        addmm_61: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg338_1, view_1173, permute_125);  arg338_1 = view_1173 = permute_125 = None
	        view_1174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_61, [sym_size_int, 1500, 1280]);  addmm_61 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1175: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1174, [sym_size_int, -1, 20, 64]);  view_1174 = None
	        permute_126: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1175, [0, 2, 1, 3]);  view_1175 = None
	        clone_100: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_126, memory_format = torch.contiguous_format);  permute_126 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_12 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_98, clone_99, clone_100, None, False, scale = 1.0);  clone_98 = clone_99 = clone_100 = None
	        getitem_98: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_12[0];  _scaled_dot_product_efficient_attention_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_127: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_98, [0, 2, 1, 3]);  getitem_98 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_127, [sym_size_int, 1500, -1]);  permute_127 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1177: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1176, [sym_size_int, 1500, 1280])
	        amin_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1177, [2])
	        amax_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1177, [2]);  view_1177 = None
	        full_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_75, full_150);  amin_75 = full_150 = None
	        full_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_75, full_151);  amax_75 = full_151 = None
	        sub_3443: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_75, minimum_75);  maximum_75 = None
	        div_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3443, 255.0);  sub_3443 = None
	        clamp_min_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_150, 1.1920928955078125e-07);  div_150 = None
	        div_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_75, clamp_min_225);  minimum_75 = None
	        round_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_151);  div_151 = None
	        sub_3449: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_151);  round_151 = None
	        clamp_min_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3449, -128);  sub_3449 = None
	        clamp_max_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_226, 127);  clamp_min_226 = None
	        _assert_tensor_metadata_677 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_225, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_677 = None
	        _assert_tensor_metadata_678 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_150, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_678 = None
	        convert_element_type_450: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_150, torch.int8);  clamp_max_150 = None
	        view_1178: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1176, [sym_size_int, 1500, 1280]);  view_1176 = None
	        view_1179: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_225, [sym_size_int, 1500, 1])
	        view_1180: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_450, [sym_size_int, 1500, 1])
	        reciprocal_75: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1179);  view_1179 = None
	        mul_7329: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_75, 1.0);  reciprocal_75 = None
	        mul_7332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1178, mul_7329);  view_1178 = mul_7329 = None
	        round_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7332);  mul_7332 = None
	        add_11599: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_152, view_1180);  round_152 = view_1180 = None
	        clamp_min_227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11599, -128);  add_11599 = None
	        clamp_max_151: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_227, 127);  clamp_min_227 = None
	        view_1181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_151, [sym_size_int, 1500, 1280]);  clamp_max_151 = None
	        _assert_tensor_metadata_679 = torch.ops.aten._assert_tensor_metadata.default(view_1181, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_679 = None
	        convert_element_type_451: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1181, torch.int8);  view_1181 = None
	        view_1182: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_451, [sym_size_int, 1500, 1280]);  convert_element_type_451 = None
	        view_1183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_225, [sym_size_int, 1500, 1]);  clamp_min_225 = None
	        view_1184: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_450, [sym_size_int, 1500, 1]);  convert_element_type_450 = None
	        _assert_tensor_metadata_680 = torch.ops.aten._assert_tensor_metadata.default(view_1182, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_680 = None
	        convert_element_type_452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1182, torch.float32);  view_1182 = None
	        _assert_tensor_metadata_681 = torch.ops.aten._assert_tensor_metadata.default(view_1184, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_681 = None
	        convert_element_type_453: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1184, torch.float32);  view_1184 = None
	        sub_3469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_452, convert_element_type_453);  convert_element_type_452 = convert_element_type_453 = None
	        mul_7354: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3469, view_1183);  sub_3469 = view_1183 = None
	        view_1185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7354, [sym_size_int, 1500, 1280]);  mul_7354 = None
	        _assert_tensor_metadata_682 = torch.ops.aten._assert_tensor_metadata.default(view_1185, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_682 = None
	        view_1186: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg343_1, [1280, 40, 32]);  arg343_1 = None
	        view_1187: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg344_1, [1280, 40, 1]);  arg344_1 = None
	        view_1188: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg345_1, [1280, 40, 1]);  arg345_1 = None
	        _assert_tensor_metadata_683 = torch.ops.aten._assert_tensor_metadata.default(view_1186, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_683 = None
	        convert_element_type_454: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1186, torch.float32);  view_1186 = None
	        _assert_tensor_metadata_684 = torch.ops.aten._assert_tensor_metadata.default(view_1188, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_684 = None
	        convert_element_type_455: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1188, torch.float32);  view_1188 = None
	        sub_3473: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_454, convert_element_type_455);  convert_element_type_454 = convert_element_type_455 = None
	        mul_7359: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3473, view_1187);  sub_3473 = view_1187 = None
	        view_1189: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7359, [1280, 1280]);  mul_7359 = None
	        _assert_tensor_metadata_685 = torch.ops.aten._assert_tensor_metadata.default(view_1189, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_685 = None
	        mul_7364: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1190: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1185, [mul_7364, 1280]);  view_1185 = mul_7364 = None
	        permute_128: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1189, [1, 0]);  view_1189 = None
	        addmm_62: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg342_1, view_1190, permute_128);  arg342_1 = view_1190 = permute_128 = None
	        view_1191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_62, [sym_size_int, 1500, 1280]);  addmm_62 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1191);  view_1191 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_11662: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_11042, clone_101);  add_11042 = clone_101 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_102: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_11662, memory_format = torch.contiguous_format)
	        var_mean_25 = torch.ops.aten.var_mean.correction(clone_102, [2], correction = 0, keepdim = True)
	        getitem_102: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_25[0]
	        getitem_103: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_25[1];  var_mean_25 = None
	        add_11667: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_102, 1e-05);  getitem_102 = None
	        rsqrt_25: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_11667);  add_11667 = None
	        sub_3479: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_102, getitem_103);  clone_102 = getitem_103 = None
	        mul_7375: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3479, rsqrt_25);  sub_3479 = rsqrt_25 = None
	        mul_7376: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7375, arg346_1);  mul_7375 = arg346_1 = None
	        add_11668: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_7376, arg347_1);  mul_7376 = arg347_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11668, [sym_size_int, 1500, 1280])
	        amin_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1192, [2])
	        amax_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1192, [2]);  view_1192 = None
	        full_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_76, full_152);  amin_76 = full_152 = None
	        full_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_76, full_153);  amax_76 = full_153 = None
	        sub_3490: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_76, minimum_76);  maximum_76 = None
	        div_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3490, 255.0);  sub_3490 = None
	        clamp_min_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_152, 1.1920928955078125e-07);  div_152 = None
	        div_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_76, clamp_min_228);  minimum_76 = None
	        round_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_153);  div_153 = None
	        sub_3496: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_153);  round_153 = None
	        clamp_min_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3496, -128);  sub_3496 = None
	        clamp_max_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_229, 127);  clamp_min_229 = None
	        _assert_tensor_metadata_686 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_228, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_686 = None
	        _assert_tensor_metadata_687 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_152, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_687 = None
	        convert_element_type_456: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_152, torch.int8);  clamp_max_152 = None
	        view_1193: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11668, [sym_size_int, 1500, 1280]);  add_11668 = None
	        view_1194: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_228, [sym_size_int, 1500, 1])
	        view_1195: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_456, [sym_size_int, 1500, 1])
	        reciprocal_76: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1194);  view_1194 = None
	        mul_7424: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_76, 1.0);  reciprocal_76 = None
	        mul_7427: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1193, mul_7424);  view_1193 = mul_7424 = None
	        round_154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7427);  mul_7427 = None
	        add_11755: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_154, view_1195);  round_154 = view_1195 = None
	        clamp_min_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11755, -128);  add_11755 = None
	        clamp_max_153: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_230, 127);  clamp_min_230 = None
	        view_1196: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_153, [sym_size_int, 1500, 1280]);  clamp_max_153 = None
	        _assert_tensor_metadata_688 = torch.ops.aten._assert_tensor_metadata.default(view_1196, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_688 = None
	        convert_element_type_457: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1196, torch.int8);  view_1196 = None
	        view_1197: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_457, [sym_size_int, 1500, 1280]);  convert_element_type_457 = None
	        view_1198: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_228, [sym_size_int, 1500, 1]);  clamp_min_228 = None
	        view_1199: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_456, [sym_size_int, 1500, 1]);  convert_element_type_456 = None
	        _assert_tensor_metadata_689 = torch.ops.aten._assert_tensor_metadata.default(view_1197, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_689 = None
	        convert_element_type_458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1197, torch.float32);  view_1197 = None
	        _assert_tensor_metadata_690 = torch.ops.aten._assert_tensor_metadata.default(view_1199, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_690 = None
	        convert_element_type_459: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1199, torch.float32);  view_1199 = None
	        sub_3516: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_458, convert_element_type_459);  convert_element_type_458 = convert_element_type_459 = None
	        mul_7449: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3516, view_1198);  sub_3516 = view_1198 = None
	        view_1200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7449, [sym_size_int, 1500, 1280]);  mul_7449 = None
	        _assert_tensor_metadata_691 = torch.ops.aten._assert_tensor_metadata.default(view_1200, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_691 = None
	        view_1201: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg349_1, [5120, 40, 32]);  arg349_1 = None
	        view_1202: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg350_1, [5120, 40, 1]);  arg350_1 = None
	        view_1203: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg351_1, [5120, 40, 1]);  arg351_1 = None
	        _assert_tensor_metadata_692 = torch.ops.aten._assert_tensor_metadata.default(view_1201, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_692 = None
	        convert_element_type_460: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1201, torch.float32);  view_1201 = None
	        _assert_tensor_metadata_693 = torch.ops.aten._assert_tensor_metadata.default(view_1203, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_693 = None
	        convert_element_type_461: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1203, torch.float32);  view_1203 = None
	        sub_3520: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_460, convert_element_type_461);  convert_element_type_460 = convert_element_type_461 = None
	        mul_7454: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3520, view_1202);  sub_3520 = view_1202 = None
	        view_1204: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7454, [5120, 1280]);  mul_7454 = None
	        _assert_tensor_metadata_694 = torch.ops.aten._assert_tensor_metadata.default(view_1204, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_694 = None
	        mul_7459: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1205: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1200, [mul_7459, 1280]);  view_1200 = mul_7459 = None
	        permute_129: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1204, [1, 0]);  view_1204 = None
	        addmm_63: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg348_1, view_1205, permute_129);  arg348_1 = view_1205 = permute_129 = None
	        view_1206: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_63, [sym_size_int, 1500, 5120]);  addmm_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_7466: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1206, 0.5)
	        mul_7467: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1206, 0.7071067811865476);  view_1206 = None
	        erf_14: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_7467);  mul_7467 = None
	        add_11814: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_14, 1);  erf_14 = None
	        mul_7468: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7466, add_11814);  mul_7466 = add_11814 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_103: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_7468);  mul_7468 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1207: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_103, [sym_size_int, 1500, 5120])
	        amin_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1207, [2])
	        amax_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1207, [2]);  view_1207 = None
	        full_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_77, full_154);  amin_77 = full_154 = None
	        full_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_77, full_155);  amax_77 = full_155 = None
	        sub_3533: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_77, minimum_77);  maximum_77 = None
	        div_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3533, 255.0);  sub_3533 = None
	        clamp_min_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_154, 1.1920928955078125e-07);  div_154 = None
	        div_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_77, clamp_min_231);  minimum_77 = None
	        round_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_155);  div_155 = None
	        sub_3539: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_155);  round_155 = None
	        clamp_min_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3539, -128);  sub_3539 = None
	        clamp_max_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_232, 127);  clamp_min_232 = None
	        _assert_tensor_metadata_695 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_231, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_695 = None
	        _assert_tensor_metadata_696 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_154, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_696 = None
	        convert_element_type_462: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_154, torch.int8);  clamp_max_154 = None
	        view_1208: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_103, [sym_size_int, 1500, 5120]);  clone_103 = None
	        view_1209: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_231, [sym_size_int, 1500, 1])
	        view_1210: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_462, [sym_size_int, 1500, 1])
	        reciprocal_77: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1209);  view_1209 = None
	        mul_7514: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_77, 1.0);  reciprocal_77 = None
	        mul_7517: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1208, mul_7514);  view_1208 = mul_7514 = None
	        round_156: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_7517);  mul_7517 = None
	        add_11897: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_156, view_1210);  round_156 = view_1210 = None
	        clamp_min_233: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11897, -128);  add_11897 = None
	        clamp_max_155: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_233, 127);  clamp_min_233 = None
	        view_1211: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_155, [sym_size_int, 1500, 5120]);  clamp_max_155 = None
	        _assert_tensor_metadata_697 = torch.ops.aten._assert_tensor_metadata.default(view_1211, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_697 = None
	        convert_element_type_463: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1211, torch.int8);  view_1211 = None
	        view_1212: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_463, [sym_size_int, 1500, 5120]);  convert_element_type_463 = None
	        view_1213: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_231, [sym_size_int, 1500, 1]);  clamp_min_231 = None
	        view_1214: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_462, [sym_size_int, 1500, 1]);  convert_element_type_462 = None
	        _assert_tensor_metadata_698 = torch.ops.aten._assert_tensor_metadata.default(view_1212, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_698 = None
	        convert_element_type_464: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1212, torch.float32);  view_1212 = None
	        _assert_tensor_metadata_699 = torch.ops.aten._assert_tensor_metadata.default(view_1214, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_699 = None
	        convert_element_type_465: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1214, torch.float32);  view_1214 = None
	        sub_3559: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_464, convert_element_type_465);  convert_element_type_464 = convert_element_type_465 = None
	        mul_7539: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3559, view_1213);  sub_3559 = view_1213 = None
	        view_1215: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_7539, [sym_size_int, 1500, 5120]);  mul_7539 = None
	        _assert_tensor_metadata_700 = torch.ops.aten._assert_tensor_metadata.default(view_1215, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_700 = None
	        view_1216: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg353_1, [1280, 160, 32]);  arg353_1 = None
	        view_1217: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg354_1, [1280, 160, 1]);  arg354_1 = None
	        view_1218: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg355_1, [1280, 160, 1]);  arg355_1 = None
	        _assert_tensor_metadata_701 = torch.ops.aten._assert_tensor_metadata.default(view_1216, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_701 = None
	        convert_element_type_466: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1216, torch.float32);  view_1216 = None
	        _assert_tensor_metadata_702 = torch.ops.aten._assert_tensor_metadata.default(view_1218, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_702 = None
	        convert_element_type_467: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1218, torch.float32);  view_1218 = None
	        sub_3563: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_466, convert_element_type_467);  convert_element_type_466 = convert_element_type_467 = None
	        mul_7544: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3563, view_1217);  sub_3563 = view_1217 = None
	        view_1219: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_7544, [1280, 5120]);  mul_7544 = None
	        _assert_tensor_metadata_703 = torch.ops.aten._assert_tensor_metadata.default(view_1219, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_703 = None
	        mul_7549: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1220: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1215, [mul_7549, 5120]);  view_1215 = mul_7549 = None
	        permute_130: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1219, [1, 0]);  view_1219 = None
	        addmm_64: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg352_1, view_1220, permute_130);  arg352_1 = view_1220 = permute_130 = None
	        view_1221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_64, [sym_size_int, 1500, 1280]);  addmm_64 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_104: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1221);  view_1221 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_11960: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_11662, clone_104);  add_11662 = clone_104 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_105: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_11960, memory_format = torch.contiguous_format)
	        var_mean_26 = torch.ops.aten.var_mean.correction(clone_105, [2], correction = 0, keepdim = True)
	        getitem_104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_26[0]
	        getitem_105: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_26[1];  var_mean_26 = None
	        add_11965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_104, 1e-05);  getitem_104 = None
	        rsqrt_26: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_11965);  add_11965 = None
	        sub_3569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_105, getitem_105);  clone_105 = getitem_105 = None
	        mul_7560: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3569, rsqrt_26);  sub_3569 = rsqrt_26 = None
	        mul_7561: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7560, arg356_1);  mul_7560 = arg356_1 = None
	        add_11966: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_7561, arg357_1);  mul_7561 = arg357_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1222: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11966, [sym_size_int, 1500, 1280])
	        amin_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1222, [2])
	        amax_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1222, [2]);  view_1222 = None
	        full_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_78, full_156);  amin_78 = full_156 = None
	        full_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_78, full_157);  amax_78 = full_157 = None
	        sub_3580: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_78, minimum_78);  maximum_78 = None
	        div_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3580, 255.0);  sub_3580 = None
	        clamp_min_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_156, 1.1920928955078125e-07);  div_156 = None
	        div_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_78, clamp_min_234);  minimum_78 = None
	        round_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_157);  div_157 = None
	        sub_3586: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_157);  round_157 = None
	        clamp_min_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3586, -128);  sub_3586 = None
	        clamp_max_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_235, 127);  clamp_min_235 = None
	        _assert_tensor_metadata_704 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_234, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_704 = None
	        _assert_tensor_metadata_705 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_156, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_705 = None
	        convert_element_type_468: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_156, torch.int8);  clamp_max_156 = None
	        view_1223: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11966, [sym_size_int, 1500, 1280])
	        view_1224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_234, [sym_size_int, 1500, 1])
	        view_1225: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_468, [sym_size_int, 1500, 1])
	        reciprocal_78: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1224);  view_1224 = None
	        mul_7609: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_78, 1.0);  reciprocal_78 = None
	        mul_7612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1223, mul_7609);  view_1223 = mul_7609 = None
	        round_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7612);  mul_7612 = None
	        add_12053: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_158, view_1225);  round_158 = view_1225 = None
	        clamp_min_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12053, -128);  add_12053 = None
	        clamp_max_157: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_236, 127);  clamp_min_236 = None
	        view_1226: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_157, [sym_size_int, 1500, 1280]);  clamp_max_157 = None
	        _assert_tensor_metadata_706 = torch.ops.aten._assert_tensor_metadata.default(view_1226, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_706 = None
	        convert_element_type_469: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1226, torch.int8);  view_1226 = None
	        view_1227: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_469, [sym_size_int, 1500, 1280]);  convert_element_type_469 = None
	        view_1228: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_234, [sym_size_int, 1500, 1]);  clamp_min_234 = None
	        view_1229: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_468, [sym_size_int, 1500, 1]);  convert_element_type_468 = None
	        _assert_tensor_metadata_707 = torch.ops.aten._assert_tensor_metadata.default(view_1227, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_707 = None
	        convert_element_type_470: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1227, torch.float32);  view_1227 = None
	        _assert_tensor_metadata_708 = torch.ops.aten._assert_tensor_metadata.default(view_1229, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_708 = None
	        convert_element_type_471: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1229, torch.float32);  view_1229 = None
	        sub_3606: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_470, convert_element_type_471);  convert_element_type_470 = convert_element_type_471 = None
	        mul_7634: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3606, view_1228);  sub_3606 = view_1228 = None
	        view_1230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7634, [sym_size_int, 1500, 1280]);  mul_7634 = None
	        _assert_tensor_metadata_709 = torch.ops.aten._assert_tensor_metadata.default(view_1230, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_709 = None
	        view_1231: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg359_1, [1280, 40, 32]);  arg359_1 = None
	        view_1232: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg360_1, [1280, 40, 1]);  arg360_1 = None
	        view_1233: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg361_1, [1280, 40, 1]);  arg361_1 = None
	        _assert_tensor_metadata_710 = torch.ops.aten._assert_tensor_metadata.default(view_1231, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_710 = None
	        convert_element_type_472: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1231, torch.float32);  view_1231 = None
	        _assert_tensor_metadata_711 = torch.ops.aten._assert_tensor_metadata.default(view_1233, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_711 = None
	        convert_element_type_473: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1233, torch.float32);  view_1233 = None
	        sub_3610: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_472, convert_element_type_473);  convert_element_type_472 = convert_element_type_473 = None
	        mul_7639: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3610, view_1232);  sub_3610 = view_1232 = None
	        view_1234: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7639, [1280, 1280]);  mul_7639 = None
	        _assert_tensor_metadata_712 = torch.ops.aten._assert_tensor_metadata.default(view_1234, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_712 = None
	        mul_7644: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1235: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1230, [mul_7644, 1280]);  view_1230 = mul_7644 = None
	        permute_131: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1234, [1, 0]);  view_1234 = None
	        addmm_65: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg358_1, view_1235, permute_131);  arg358_1 = view_1235 = permute_131 = None
	        view_1236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_65, [sym_size_int, 1500, 1280]);  addmm_65 = None
	        mul_7651: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1236, 0.125);  view_1236 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1237: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_7651, [sym_size_int, 1500, 20, 64]);  mul_7651 = None
	        permute_132: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1237, [0, 2, 1, 3]);  view_1237 = None
	        clone_106: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_132, memory_format = torch.contiguous_format);  permute_132 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1238: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11966, [sym_size_int, 1500, 1280])
	        amin_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1238, [2])
	        amax_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1238, [2]);  view_1238 = None
	        full_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_79, full_158);  amin_79 = full_158 = None
	        full_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_79, full_159);  amax_79 = full_159 = None
	        sub_3625: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_79, minimum_79);  maximum_79 = None
	        div_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3625, 255.0);  sub_3625 = None
	        clamp_min_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_158, 1.1920928955078125e-07);  div_158 = None
	        div_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_79, clamp_min_237);  minimum_79 = None
	        round_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_159);  div_159 = None
	        sub_3631: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_159);  round_159 = None
	        clamp_min_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3631, -128);  sub_3631 = None
	        clamp_max_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_238, 127);  clamp_min_238 = None
	        _assert_tensor_metadata_713 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_237, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_713 = None
	        _assert_tensor_metadata_714 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_158, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_714 = None
	        convert_element_type_474: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_158, torch.int8);  clamp_max_158 = None
	        view_1239: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11966, [sym_size_int, 1500, 1280])
	        view_1240: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_237, [sym_size_int, 1500, 1])
	        view_1241: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_474, [sym_size_int, 1500, 1])
	        reciprocal_79: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1240);  view_1240 = None
	        mul_7705: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_79, 1.0);  reciprocal_79 = None
	        mul_7708: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1239, mul_7705);  view_1239 = mul_7705 = None
	        round_160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7708);  mul_7708 = None
	        add_12205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_160, view_1241);  round_160 = view_1241 = None
	        clamp_min_239: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12205, -128);  add_12205 = None
	        clamp_max_159: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_239, 127);  clamp_min_239 = None
	        view_1242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_159, [sym_size_int, 1500, 1280]);  clamp_max_159 = None
	        _assert_tensor_metadata_715 = torch.ops.aten._assert_tensor_metadata.default(view_1242, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_715 = None
	        convert_element_type_475: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1242, torch.int8);  view_1242 = None
	        view_1243: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_475, [sym_size_int, 1500, 1280]);  convert_element_type_475 = None
	        view_1244: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_237, [sym_size_int, 1500, 1]);  clamp_min_237 = None
	        view_1245: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_474, [sym_size_int, 1500, 1]);  convert_element_type_474 = None
	        _assert_tensor_metadata_716 = torch.ops.aten._assert_tensor_metadata.default(view_1243, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_716 = None
	        convert_element_type_476: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1243, torch.float32);  view_1243 = None
	        _assert_tensor_metadata_717 = torch.ops.aten._assert_tensor_metadata.default(view_1245, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_717 = None
	        convert_element_type_477: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1245, torch.float32);  view_1245 = None
	        sub_3651: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_476, convert_element_type_477);  convert_element_type_476 = convert_element_type_477 = None
	        mul_7730: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3651, view_1244);  sub_3651 = view_1244 = None
	        view_1246: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7730, [sym_size_int, 1500, 1280]);  mul_7730 = None
	        _assert_tensor_metadata_718 = torch.ops.aten._assert_tensor_metadata.default(view_1246, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_718 = None
	        view_1247: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg362_1, [1280, 40, 32]);  arg362_1 = None
	        view_1248: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg363_1, [1280, 40, 1]);  arg363_1 = None
	        view_1249: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg364_1, [1280, 40, 1]);  arg364_1 = None
	        _assert_tensor_metadata_719 = torch.ops.aten._assert_tensor_metadata.default(view_1247, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_719 = None
	        convert_element_type_478: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1247, torch.float32);  view_1247 = None
	        _assert_tensor_metadata_720 = torch.ops.aten._assert_tensor_metadata.default(view_1249, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_720 = None
	        convert_element_type_479: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1249, torch.float32);  view_1249 = None
	        sub_3655: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_478, convert_element_type_479);  convert_element_type_478 = convert_element_type_479 = None
	        mul_7735: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3655, view_1248);  sub_3655 = view_1248 = None
	        view_1250: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7735, [1280, 1280]);  mul_7735 = None
	        _assert_tensor_metadata_721 = torch.ops.aten._assert_tensor_metadata.default(view_1250, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_721 = None
	        permute_133: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1250, [1, 0]);  view_1250 = None
	        mul_7738: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1251: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1246, [mul_7738, 1280]);  view_1246 = mul_7738 = None
	        mm_13: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1251, permute_133);  view_1251 = permute_133 = None
	        view_1252: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_13, [sym_size_int, 1500, 1280]);  mm_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1253: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1252, [sym_size_int, -1, 20, 64]);  view_1252 = None
	        permute_134: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1253, [0, 2, 1, 3]);  view_1253 = None
	        clone_107: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_134, memory_format = torch.contiguous_format);  permute_134 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11966, [sym_size_int, 1500, 1280])
	        amin_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1254, [2])
	        amax_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1254, [2]);  view_1254 = None
	        full_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_80, full_160);  amin_80 = full_160 = None
	        full_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_80, full_161);  amax_80 = full_161 = None
	        sub_3669: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_80, minimum_80);  maximum_80 = None
	        div_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3669, 255.0);  sub_3669 = None
	        clamp_min_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_160, 1.1920928955078125e-07);  div_160 = None
	        div_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_80, clamp_min_240);  minimum_80 = None
	        round_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_161);  div_161 = None
	        sub_3675: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_161);  round_161 = None
	        clamp_min_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3675, -128);  sub_3675 = None
	        clamp_max_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_241, 127);  clamp_min_241 = None
	        _assert_tensor_metadata_722 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_240, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_722 = None
	        _assert_tensor_metadata_723 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_160, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_723 = None
	        convert_element_type_480: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_160, torch.int8);  clamp_max_160 = None
	        view_1255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11966, [sym_size_int, 1500, 1280]);  add_11966 = None
	        view_1256: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_240, [sym_size_int, 1500, 1])
	        view_1257: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_480, [sym_size_int, 1500, 1])
	        reciprocal_80: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1256);  view_1256 = None
	        mul_7804: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_80, 1.0);  reciprocal_80 = None
	        mul_7807: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1255, mul_7804);  view_1255 = mul_7804 = None
	        round_162: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7807);  mul_7807 = None
	        add_12353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_162, view_1257);  round_162 = view_1257 = None
	        clamp_min_242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12353, -128);  add_12353 = None
	        clamp_max_161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_242, 127);  clamp_min_242 = None
	        view_1258: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_161, [sym_size_int, 1500, 1280]);  clamp_max_161 = None
	        _assert_tensor_metadata_724 = torch.ops.aten._assert_tensor_metadata.default(view_1258, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_724 = None
	        convert_element_type_481: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1258, torch.int8);  view_1258 = None
	        view_1259: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_481, [sym_size_int, 1500, 1280]);  convert_element_type_481 = None
	        view_1260: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_240, [sym_size_int, 1500, 1]);  clamp_min_240 = None
	        view_1261: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_480, [sym_size_int, 1500, 1]);  convert_element_type_480 = None
	        _assert_tensor_metadata_725 = torch.ops.aten._assert_tensor_metadata.default(view_1259, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_725 = None
	        convert_element_type_482: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1259, torch.float32);  view_1259 = None
	        _assert_tensor_metadata_726 = torch.ops.aten._assert_tensor_metadata.default(view_1261, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_726 = None
	        convert_element_type_483: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1261, torch.float32);  view_1261 = None
	        sub_3695: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_482, convert_element_type_483);  convert_element_type_482 = convert_element_type_483 = None
	        mul_7829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3695, view_1260);  sub_3695 = view_1260 = None
	        view_1262: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7829, [sym_size_int, 1500, 1280]);  mul_7829 = None
	        _assert_tensor_metadata_727 = torch.ops.aten._assert_tensor_metadata.default(view_1262, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_727 = None
	        view_1263: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg366_1, [1280, 40, 32]);  arg366_1 = None
	        view_1264: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg367_1, [1280, 40, 1]);  arg367_1 = None
	        view_1265: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg368_1, [1280, 40, 1]);  arg368_1 = None
	        _assert_tensor_metadata_728 = torch.ops.aten._assert_tensor_metadata.default(view_1263, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_728 = None
	        convert_element_type_484: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1263, torch.float32);  view_1263 = None
	        _assert_tensor_metadata_729 = torch.ops.aten._assert_tensor_metadata.default(view_1265, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_729 = None
	        convert_element_type_485: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1265, torch.float32);  view_1265 = None
	        sub_3699: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_484, convert_element_type_485);  convert_element_type_484 = convert_element_type_485 = None
	        mul_7834: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3699, view_1264);  sub_3699 = view_1264 = None
	        view_1266: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7834, [1280, 1280]);  mul_7834 = None
	        _assert_tensor_metadata_730 = torch.ops.aten._assert_tensor_metadata.default(view_1266, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_730 = None
	        mul_7839: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1267: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1262, [mul_7839, 1280]);  view_1262 = mul_7839 = None
	        permute_135: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1266, [1, 0]);  view_1266 = None
	        addmm_66: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg365_1, view_1267, permute_135);  arg365_1 = view_1267 = permute_135 = None
	        view_1268: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_66, [sym_size_int, 1500, 1280]);  addmm_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1269: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1268, [sym_size_int, -1, 20, 64]);  view_1268 = None
	        permute_136: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1269, [0, 2, 1, 3]);  view_1269 = None
	        clone_108: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_136, memory_format = torch.contiguous_format);  permute_136 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_13 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_106, clone_107, clone_108, None, False, scale = 1.0);  clone_106 = clone_107 = clone_108 = None
	        getitem_106: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_13[0];  _scaled_dot_product_efficient_attention_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_137: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_106, [0, 2, 1, 3]);  getitem_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_137, [sym_size_int, 1500, -1]);  permute_137 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1270, [sym_size_int, 1500, 1280])
	        amin_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1271, [2])
	        amax_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1271, [2]);  view_1271 = None
	        full_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_81, full_162);  amin_81 = full_162 = None
	        full_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_81, full_163);  amax_81 = full_163 = None
	        sub_3717: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_81, minimum_81);  maximum_81 = None
	        div_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3717, 255.0);  sub_3717 = None
	        clamp_min_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_162, 1.1920928955078125e-07);  div_162 = None
	        div_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_81, clamp_min_243);  minimum_81 = None
	        round_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_163);  div_163 = None
	        sub_3723: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_163);  round_163 = None
	        clamp_min_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3723, -128);  sub_3723 = None
	        clamp_max_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_244, 127);  clamp_min_244 = None
	        _assert_tensor_metadata_731 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_243, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_731 = None
	        _assert_tensor_metadata_732 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_162, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_732 = None
	        convert_element_type_486: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_162, torch.int8);  clamp_max_162 = None
	        view_1272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1270, [sym_size_int, 1500, 1280]);  view_1270 = None
	        view_1273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_243, [sym_size_int, 1500, 1])
	        view_1274: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_486, [sym_size_int, 1500, 1])
	        reciprocal_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1273);  view_1273 = None
	        mul_7909: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_81, 1.0);  reciprocal_81 = None
	        mul_7912: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1272, mul_7909);  view_1272 = mul_7909 = None
	        round_164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7912);  mul_7912 = None
	        add_12517: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_164, view_1274);  round_164 = view_1274 = None
	        clamp_min_245: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12517, -128);  add_12517 = None
	        clamp_max_163: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_245, 127);  clamp_min_245 = None
	        view_1275: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_163, [sym_size_int, 1500, 1280]);  clamp_max_163 = None
	        _assert_tensor_metadata_733 = torch.ops.aten._assert_tensor_metadata.default(view_1275, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_733 = None
	        convert_element_type_487: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1275, torch.int8);  view_1275 = None
	        view_1276: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_487, [sym_size_int, 1500, 1280]);  convert_element_type_487 = None
	        view_1277: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_243, [sym_size_int, 1500, 1]);  clamp_min_243 = None
	        view_1278: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_486, [sym_size_int, 1500, 1]);  convert_element_type_486 = None
	        _assert_tensor_metadata_734 = torch.ops.aten._assert_tensor_metadata.default(view_1276, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_734 = None
	        convert_element_type_488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1276, torch.float32);  view_1276 = None
	        _assert_tensor_metadata_735 = torch.ops.aten._assert_tensor_metadata.default(view_1278, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_735 = None
	        convert_element_type_489: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1278, torch.float32);  view_1278 = None
	        sub_3743: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_488, convert_element_type_489);  convert_element_type_488 = convert_element_type_489 = None
	        mul_7934: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3743, view_1277);  sub_3743 = view_1277 = None
	        view_1279: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7934, [sym_size_int, 1500, 1280]);  mul_7934 = None
	        _assert_tensor_metadata_736 = torch.ops.aten._assert_tensor_metadata.default(view_1279, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_736 = None
	        view_1280: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg370_1, [1280, 40, 32]);  arg370_1 = None
	        view_1281: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg371_1, [1280, 40, 1]);  arg371_1 = None
	        view_1282: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg372_1, [1280, 40, 1]);  arg372_1 = None
	        _assert_tensor_metadata_737 = torch.ops.aten._assert_tensor_metadata.default(view_1280, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_737 = None
	        convert_element_type_490: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1280, torch.float32);  view_1280 = None
	        _assert_tensor_metadata_738 = torch.ops.aten._assert_tensor_metadata.default(view_1282, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_738 = None
	        convert_element_type_491: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1282, torch.float32);  view_1282 = None
	        sub_3747: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_490, convert_element_type_491);  convert_element_type_490 = convert_element_type_491 = None
	        mul_7939: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3747, view_1281);  sub_3747 = view_1281 = None
	        view_1283: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7939, [1280, 1280]);  mul_7939 = None
	        _assert_tensor_metadata_739 = torch.ops.aten._assert_tensor_metadata.default(view_1283, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_739 = None
	        mul_7944: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1284: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1279, [mul_7944, 1280]);  view_1279 = mul_7944 = None
	        permute_138: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1283, [1, 0]);  view_1283 = None
	        addmm_67: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg369_1, view_1284, permute_138);  arg369_1 = view_1284 = permute_138 = None
	        view_1285: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_67, [sym_size_int, 1500, 1280]);  addmm_67 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_109: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1285);  view_1285 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_12580: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_11960, clone_109);  add_11960 = clone_109 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_12580, memory_format = torch.contiguous_format)
	        var_mean_27 = torch.ops.aten.var_mean.correction(clone_110, [2], correction = 0, keepdim = True)
	        getitem_110: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_27[0]
	        getitem_111: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_27[1];  var_mean_27 = None
	        add_12585: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_110, 1e-05);  getitem_110 = None
	        rsqrt_27: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_12585);  add_12585 = None
	        sub_3753: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_110, getitem_111);  clone_110 = getitem_111 = None
	        mul_7955: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3753, rsqrt_27);  sub_3753 = rsqrt_27 = None
	        mul_7956: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7955, arg373_1);  mul_7955 = arg373_1 = None
	        add_12586: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_7956, arg374_1);  mul_7956 = arg374_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1286: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_12586, [sym_size_int, 1500, 1280])
	        amin_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1286, [2])
	        amax_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1286, [2]);  view_1286 = None
	        full_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_82, full_164);  amin_82 = full_164 = None
	        full_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_82, full_165);  amax_82 = full_165 = None
	        sub_3764: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_82, minimum_82);  maximum_82 = None
	        div_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3764, 255.0);  sub_3764 = None
	        clamp_min_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_164, 1.1920928955078125e-07);  div_164 = None
	        div_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_82, clamp_min_246);  minimum_82 = None
	        round_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_165);  div_165 = None
	        sub_3770: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_165);  round_165 = None
	        clamp_min_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3770, -128);  sub_3770 = None
	        clamp_max_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_247, 127);  clamp_min_247 = None
	        _assert_tensor_metadata_740 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_246, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_740 = None
	        _assert_tensor_metadata_741 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_164, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_741 = None
	        convert_element_type_492: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_164, torch.int8);  clamp_max_164 = None
	        view_1287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_12586, [sym_size_int, 1500, 1280]);  add_12586 = None
	        view_1288: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_246, [sym_size_int, 1500, 1])
	        view_1289: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_492, [sym_size_int, 1500, 1])
	        reciprocal_82: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1288);  view_1288 = None
	        mul_8004: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_82, 1.0);  reciprocal_82 = None
	        mul_8007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1287, mul_8004);  view_1287 = mul_8004 = None
	        round_166: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8007);  mul_8007 = None
	        add_12673: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_166, view_1289);  round_166 = view_1289 = None
	        clamp_min_248: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12673, -128);  add_12673 = None
	        clamp_max_165: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_248, 127);  clamp_min_248 = None
	        view_1290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_165, [sym_size_int, 1500, 1280]);  clamp_max_165 = None
	        _assert_tensor_metadata_742 = torch.ops.aten._assert_tensor_metadata.default(view_1290, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_742 = None
	        convert_element_type_493: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1290, torch.int8);  view_1290 = None
	        view_1291: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_493, [sym_size_int, 1500, 1280]);  convert_element_type_493 = None
	        view_1292: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_246, [sym_size_int, 1500, 1]);  clamp_min_246 = None
	        view_1293: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_492, [sym_size_int, 1500, 1]);  convert_element_type_492 = None
	        _assert_tensor_metadata_743 = torch.ops.aten._assert_tensor_metadata.default(view_1291, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_743 = None
	        convert_element_type_494: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1291, torch.float32);  view_1291 = None
	        _assert_tensor_metadata_744 = torch.ops.aten._assert_tensor_metadata.default(view_1293, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_744 = None
	        convert_element_type_495: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1293, torch.float32);  view_1293 = None
	        sub_3790: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_494, convert_element_type_495);  convert_element_type_494 = convert_element_type_495 = None
	        mul_8029: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3790, view_1292);  sub_3790 = view_1292 = None
	        view_1294: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8029, [sym_size_int, 1500, 1280]);  mul_8029 = None
	        _assert_tensor_metadata_745 = torch.ops.aten._assert_tensor_metadata.default(view_1294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_745 = None
	        view_1295: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg376_1, [5120, 40, 32]);  arg376_1 = None
	        view_1296: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg377_1, [5120, 40, 1]);  arg377_1 = None
	        view_1297: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg378_1, [5120, 40, 1]);  arg378_1 = None
	        _assert_tensor_metadata_746 = torch.ops.aten._assert_tensor_metadata.default(view_1295, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_746 = None
	        convert_element_type_496: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1295, torch.float32);  view_1295 = None
	        _assert_tensor_metadata_747 = torch.ops.aten._assert_tensor_metadata.default(view_1297, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_747 = None
	        convert_element_type_497: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1297, torch.float32);  view_1297 = None
	        sub_3794: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_496, convert_element_type_497);  convert_element_type_496 = convert_element_type_497 = None
	        mul_8034: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3794, view_1296);  sub_3794 = view_1296 = None
	        view_1298: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8034, [5120, 1280]);  mul_8034 = None
	        _assert_tensor_metadata_748 = torch.ops.aten._assert_tensor_metadata.default(view_1298, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_748 = None
	        mul_8039: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1299: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1294, [mul_8039, 1280]);  view_1294 = mul_8039 = None
	        permute_139: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1298, [1, 0]);  view_1298 = None
	        addmm_68: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg375_1, view_1299, permute_139);  arg375_1 = view_1299 = permute_139 = None
	        view_1300: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_68, [sym_size_int, 1500, 5120]);  addmm_68 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_8046: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1300, 0.5)
	        mul_8047: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1300, 0.7071067811865476);  view_1300 = None
	        erf_15: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_8047);  mul_8047 = None
	        add_12732: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_15, 1);  erf_15 = None
	        mul_8048: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8046, add_12732);  mul_8046 = add_12732 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_111: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_8048);  mul_8048 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1301: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_111, [sym_size_int, 1500, 5120])
	        amin_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1301, [2])
	        amax_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1301, [2]);  view_1301 = None
	        full_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_83, full_166);  amin_83 = full_166 = None
	        full_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_83, full_167);  amax_83 = full_167 = None
	        sub_3807: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_83, minimum_83);  maximum_83 = None
	        div_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3807, 255.0);  sub_3807 = None
	        clamp_min_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_166, 1.1920928955078125e-07);  div_166 = None
	        div_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_83, clamp_min_249);  minimum_83 = None
	        round_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_167);  div_167 = None
	        sub_3813: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_167);  round_167 = None
	        clamp_min_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3813, -128);  sub_3813 = None
	        clamp_max_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_250, 127);  clamp_min_250 = None
	        _assert_tensor_metadata_749 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_749 = None
	        _assert_tensor_metadata_750 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_166, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_750 = None
	        convert_element_type_498: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_166, torch.int8);  clamp_max_166 = None
	        view_1302: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_111, [sym_size_int, 1500, 5120]);  clone_111 = None
	        view_1303: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_249, [sym_size_int, 1500, 1])
	        view_1304: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_498, [sym_size_int, 1500, 1])
	        reciprocal_83: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1303);  view_1303 = None
	        mul_8094: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_83, 1.0);  reciprocal_83 = None
	        mul_8097: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1302, mul_8094);  view_1302 = mul_8094 = None
	        round_168: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_8097);  mul_8097 = None
	        add_12815: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_168, view_1304);  round_168 = view_1304 = None
	        clamp_min_251: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12815, -128);  add_12815 = None
	        clamp_max_167: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_251, 127);  clamp_min_251 = None
	        view_1305: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_167, [sym_size_int, 1500, 5120]);  clamp_max_167 = None
	        _assert_tensor_metadata_751 = torch.ops.aten._assert_tensor_metadata.default(view_1305, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_751 = None
	        convert_element_type_499: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1305, torch.int8);  view_1305 = None
	        view_1306: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_499, [sym_size_int, 1500, 5120]);  convert_element_type_499 = None
	        view_1307: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_249, [sym_size_int, 1500, 1]);  clamp_min_249 = None
	        view_1308: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_498, [sym_size_int, 1500, 1]);  convert_element_type_498 = None
	        _assert_tensor_metadata_752 = torch.ops.aten._assert_tensor_metadata.default(view_1306, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_752 = None
	        convert_element_type_500: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1306, torch.float32);  view_1306 = None
	        _assert_tensor_metadata_753 = torch.ops.aten._assert_tensor_metadata.default(view_1308, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_753 = None
	        convert_element_type_501: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1308, torch.float32);  view_1308 = None
	        sub_3833: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_500, convert_element_type_501);  convert_element_type_500 = convert_element_type_501 = None
	        mul_8119: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3833, view_1307);  sub_3833 = view_1307 = None
	        view_1309: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_8119, [sym_size_int, 1500, 5120]);  mul_8119 = None
	        _assert_tensor_metadata_754 = torch.ops.aten._assert_tensor_metadata.default(view_1309, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_754 = None
	        view_1310: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg380_1, [1280, 160, 32]);  arg380_1 = None
	        view_1311: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg381_1, [1280, 160, 1]);  arg381_1 = None
	        view_1312: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg382_1, [1280, 160, 1]);  arg382_1 = None
	        _assert_tensor_metadata_755 = torch.ops.aten._assert_tensor_metadata.default(view_1310, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_755 = None
	        convert_element_type_502: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1310, torch.float32);  view_1310 = None
	        _assert_tensor_metadata_756 = torch.ops.aten._assert_tensor_metadata.default(view_1312, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_756 = None
	        convert_element_type_503: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1312, torch.float32);  view_1312 = None
	        sub_3837: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_502, convert_element_type_503);  convert_element_type_502 = convert_element_type_503 = None
	        mul_8124: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3837, view_1311);  sub_3837 = view_1311 = None
	        view_1313: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_8124, [1280, 5120]);  mul_8124 = None
	        _assert_tensor_metadata_757 = torch.ops.aten._assert_tensor_metadata.default(view_1313, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_757 = None
	        mul_8129: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1314: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1309, [mul_8129, 5120]);  view_1309 = mul_8129 = None
	        permute_140: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1313, [1, 0]);  view_1313 = None
	        addmm_69: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg379_1, view_1314, permute_140);  arg379_1 = view_1314 = permute_140 = None
	        view_1315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_69, [sym_size_int, 1500, 1280]);  addmm_69 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_112: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1315);  view_1315 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_12878: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_12580, clone_112);  add_12580 = clone_112 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_12878, memory_format = torch.contiguous_format)
	        var_mean_28 = torch.ops.aten.var_mean.correction(clone_113, [2], correction = 0, keepdim = True)
	        getitem_112: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_28[0]
	        getitem_113: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_28[1];  var_mean_28 = None
	        add_12883: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_112, 1e-05);  getitem_112 = None
	        rsqrt_28: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_12883);  add_12883 = None
	        sub_3843: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_113, getitem_113);  clone_113 = getitem_113 = None
	        mul_8140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3843, rsqrt_28);  sub_3843 = rsqrt_28 = None
	        mul_8141: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8140, arg383_1);  mul_8140 = arg383_1 = None
	        add_12884: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_8141, arg384_1);  mul_8141 = arg384_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1316: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_12884, [sym_size_int, 1500, 1280])
	        amin_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1316, [2])
	        amax_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1316, [2]);  view_1316 = None
	        full_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_84, full_168);  amin_84 = full_168 = None
	        full_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_84, full_169);  amax_84 = full_169 = None
	        sub_3854: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_84, minimum_84);  maximum_84 = None
	        div_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3854, 255.0);  sub_3854 = None
	        clamp_min_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_168, 1.1920928955078125e-07);  div_168 = None
	        div_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_84, clamp_min_252);  minimum_84 = None
	        round_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_169);  div_169 = None
	        sub_3860: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_169);  round_169 = None
	        clamp_min_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3860, -128);  sub_3860 = None
	        clamp_max_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_253, 127);  clamp_min_253 = None
	        _assert_tensor_metadata_758 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_252, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_758 = None
	        _assert_tensor_metadata_759 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_168, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_759 = None
	        convert_element_type_504: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_168, torch.int8);  clamp_max_168 = None
	        view_1317: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_12884, [sym_size_int, 1500, 1280])
	        view_1318: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_252, [sym_size_int, 1500, 1])
	        view_1319: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_504, [sym_size_int, 1500, 1])
	        reciprocal_84: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1318);  view_1318 = None
	        mul_8189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_84, 1.0);  reciprocal_84 = None
	        mul_8192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1317, mul_8189);  view_1317 = mul_8189 = None
	        round_170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8192);  mul_8192 = None
	        add_12971: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_170, view_1319);  round_170 = view_1319 = None
	        clamp_min_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12971, -128);  add_12971 = None
	        clamp_max_169: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_254, 127);  clamp_min_254 = None
	        view_1320: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_169, [sym_size_int, 1500, 1280]);  clamp_max_169 = None
	        _assert_tensor_metadata_760 = torch.ops.aten._assert_tensor_metadata.default(view_1320, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_760 = None
	        convert_element_type_505: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1320, torch.int8);  view_1320 = None
	        view_1321: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_505, [sym_size_int, 1500, 1280]);  convert_element_type_505 = None
	        view_1322: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_252, [sym_size_int, 1500, 1]);  clamp_min_252 = None
	        view_1323: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_504, [sym_size_int, 1500, 1]);  convert_element_type_504 = None
	        _assert_tensor_metadata_761 = torch.ops.aten._assert_tensor_metadata.default(view_1321, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_761 = None
	        convert_element_type_506: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1321, torch.float32);  view_1321 = None
	        _assert_tensor_metadata_762 = torch.ops.aten._assert_tensor_metadata.default(view_1323, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_762 = None
	        convert_element_type_507: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1323, torch.float32);  view_1323 = None
	        sub_3880: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_506, convert_element_type_507);  convert_element_type_506 = convert_element_type_507 = None
	        mul_8214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3880, view_1322);  sub_3880 = view_1322 = None
	        view_1324: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8214, [sym_size_int, 1500, 1280]);  mul_8214 = None
	        _assert_tensor_metadata_763 = torch.ops.aten._assert_tensor_metadata.default(view_1324, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_763 = None
	        view_1325: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg386_1, [1280, 40, 32]);  arg386_1 = None
	        view_1326: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg387_1, [1280, 40, 1]);  arg387_1 = None
	        view_1327: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg388_1, [1280, 40, 1]);  arg388_1 = None
	        _assert_tensor_metadata_764 = torch.ops.aten._assert_tensor_metadata.default(view_1325, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_764 = None
	        convert_element_type_508: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1325, torch.float32);  view_1325 = None
	        _assert_tensor_metadata_765 = torch.ops.aten._assert_tensor_metadata.default(view_1327, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_765 = None
	        convert_element_type_509: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1327, torch.float32);  view_1327 = None
	        sub_3884: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_508, convert_element_type_509);  convert_element_type_508 = convert_element_type_509 = None
	        mul_8219: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3884, view_1326);  sub_3884 = view_1326 = None
	        view_1328: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8219, [1280, 1280]);  mul_8219 = None
	        _assert_tensor_metadata_766 = torch.ops.aten._assert_tensor_metadata.default(view_1328, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_766 = None
	        mul_8224: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1329: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1324, [mul_8224, 1280]);  view_1324 = mul_8224 = None
	        permute_141: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1328, [1, 0]);  view_1328 = None
	        addmm_70: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg385_1, view_1329, permute_141);  arg385_1 = view_1329 = permute_141 = None
	        view_1330: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_70, [sym_size_int, 1500, 1280]);  addmm_70 = None
	        mul_8231: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1330, 0.125);  view_1330 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1331: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_8231, [sym_size_int, 1500, 20, 64]);  mul_8231 = None
	        permute_142: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1331, [0, 2, 1, 3]);  view_1331 = None
	        clone_114: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_142, memory_format = torch.contiguous_format);  permute_142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_12884, [sym_size_int, 1500, 1280])
	        amin_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1332, [2])
	        amax_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1332, [2]);  view_1332 = None
	        full_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_85, full_170);  amin_85 = full_170 = None
	        full_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_85, full_171);  amax_85 = full_171 = None
	        sub_3899: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_85, minimum_85);  maximum_85 = None
	        div_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3899, 255.0);  sub_3899 = None
	        clamp_min_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_170, 1.1920928955078125e-07);  div_170 = None
	        div_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_85, clamp_min_255);  minimum_85 = None
	        round_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_171);  div_171 = None
	        sub_3905: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_171);  round_171 = None
	        clamp_min_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3905, -128);  sub_3905 = None
	        clamp_max_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_256, 127);  clamp_min_256 = None
	        _assert_tensor_metadata_767 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_255, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_767 = None
	        _assert_tensor_metadata_768 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_170, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_768 = None
	        convert_element_type_510: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_170, torch.int8);  clamp_max_170 = None
	        view_1333: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_12884, [sym_size_int, 1500, 1280])
	        view_1334: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_255, [sym_size_int, 1500, 1])
	        view_1335: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_510, [sym_size_int, 1500, 1])
	        reciprocal_85: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1334);  view_1334 = None
	        mul_8285: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_85, 1.0);  reciprocal_85 = None
	        mul_8288: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1333, mul_8285);  view_1333 = mul_8285 = None
	        round_172: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8288);  mul_8288 = None
	        add_13123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_172, view_1335);  round_172 = view_1335 = None
	        clamp_min_257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13123, -128);  add_13123 = None
	        clamp_max_171: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_257, 127);  clamp_min_257 = None
	        view_1336: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_171, [sym_size_int, 1500, 1280]);  clamp_max_171 = None
	        _assert_tensor_metadata_769 = torch.ops.aten._assert_tensor_metadata.default(view_1336, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_769 = None
	        convert_element_type_511: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1336, torch.int8);  view_1336 = None
	        view_1337: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_511, [sym_size_int, 1500, 1280]);  convert_element_type_511 = None
	        view_1338: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_255, [sym_size_int, 1500, 1]);  clamp_min_255 = None
	        view_1339: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_510, [sym_size_int, 1500, 1]);  convert_element_type_510 = None
	        _assert_tensor_metadata_770 = torch.ops.aten._assert_tensor_metadata.default(view_1337, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_770 = None
	        convert_element_type_512: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1337, torch.float32);  view_1337 = None
	        _assert_tensor_metadata_771 = torch.ops.aten._assert_tensor_metadata.default(view_1339, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_771 = None
	        convert_element_type_513: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1339, torch.float32);  view_1339 = None
	        sub_3925: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_512, convert_element_type_513);  convert_element_type_512 = convert_element_type_513 = None
	        mul_8310: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3925, view_1338);  sub_3925 = view_1338 = None
	        view_1340: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8310, [sym_size_int, 1500, 1280]);  mul_8310 = None
	        _assert_tensor_metadata_772 = torch.ops.aten._assert_tensor_metadata.default(view_1340, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_772 = None
	        view_1341: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg389_1, [1280, 40, 32]);  arg389_1 = None
	        view_1342: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg390_1, [1280, 40, 1]);  arg390_1 = None
	        view_1343: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg391_1, [1280, 40, 1]);  arg391_1 = None
	        _assert_tensor_metadata_773 = torch.ops.aten._assert_tensor_metadata.default(view_1341, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_773 = None
	        convert_element_type_514: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1341, torch.float32);  view_1341 = None
	        _assert_tensor_metadata_774 = torch.ops.aten._assert_tensor_metadata.default(view_1343, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_774 = None
	        convert_element_type_515: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1343, torch.float32);  view_1343 = None
	        sub_3929: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_514, convert_element_type_515);  convert_element_type_514 = convert_element_type_515 = None
	        mul_8315: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3929, view_1342);  sub_3929 = view_1342 = None
	        view_1344: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8315, [1280, 1280]);  mul_8315 = None
	        _assert_tensor_metadata_775 = torch.ops.aten._assert_tensor_metadata.default(view_1344, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_775 = None
	        permute_143: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1344, [1, 0]);  view_1344 = None
	        mul_8318: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1345: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1340, [mul_8318, 1280]);  view_1340 = mul_8318 = None
	        mm_14: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1345, permute_143);  view_1345 = permute_143 = None
	        view_1346: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_14, [sym_size_int, 1500, 1280]);  mm_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1347: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1346, [sym_size_int, -1, 20, 64]);  view_1346 = None
	        permute_144: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1347, [0, 2, 1, 3]);  view_1347 = None
	        clone_115: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_144, memory_format = torch.contiguous_format);  permute_144 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1348: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_12884, [sym_size_int, 1500, 1280])
	        amin_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1348, [2])
	        amax_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1348, [2]);  view_1348 = None
	        full_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_86, full_172);  amin_86 = full_172 = None
	        full_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_86, full_173);  amax_86 = full_173 = None
	        sub_3943: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_86, minimum_86);  maximum_86 = None
	        div_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3943, 255.0);  sub_3943 = None
	        clamp_min_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_172, 1.1920928955078125e-07);  div_172 = None
	        div_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_86, clamp_min_258);  minimum_86 = None
	        round_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_173);  div_173 = None
	        sub_3949: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_173);  round_173 = None
	        clamp_min_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3949, -128);  sub_3949 = None
	        clamp_max_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_259, 127);  clamp_min_259 = None
	        _assert_tensor_metadata_776 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_258, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_776 = None
	        _assert_tensor_metadata_777 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_172, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_777 = None
	        convert_element_type_516: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_172, torch.int8);  clamp_max_172 = None
	        view_1349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_12884, [sym_size_int, 1500, 1280]);  add_12884 = None
	        view_1350: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_258, [sym_size_int, 1500, 1])
	        view_1351: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_516, [sym_size_int, 1500, 1])
	        reciprocal_86: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1350);  view_1350 = None
	        mul_8384: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_86, 1.0);  reciprocal_86 = None
	        mul_8387: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1349, mul_8384);  view_1349 = mul_8384 = None
	        round_174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8387);  mul_8387 = None
	        add_13271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_174, view_1351);  round_174 = view_1351 = None
	        clamp_min_260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13271, -128);  add_13271 = None
	        clamp_max_173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_260, 127);  clamp_min_260 = None
	        view_1352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_173, [sym_size_int, 1500, 1280]);  clamp_max_173 = None
	        _assert_tensor_metadata_778 = torch.ops.aten._assert_tensor_metadata.default(view_1352, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_778 = None
	        convert_element_type_517: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1352, torch.int8);  view_1352 = None
	        view_1353: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_517, [sym_size_int, 1500, 1280]);  convert_element_type_517 = None
	        view_1354: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_258, [sym_size_int, 1500, 1]);  clamp_min_258 = None
	        view_1355: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_516, [sym_size_int, 1500, 1]);  convert_element_type_516 = None
	        _assert_tensor_metadata_779 = torch.ops.aten._assert_tensor_metadata.default(view_1353, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_779 = None
	        convert_element_type_518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1353, torch.float32);  view_1353 = None
	        _assert_tensor_metadata_780 = torch.ops.aten._assert_tensor_metadata.default(view_1355, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_780 = None
	        convert_element_type_519: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1355, torch.float32);  view_1355 = None
	        sub_3969: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_518, convert_element_type_519);  convert_element_type_518 = convert_element_type_519 = None
	        mul_8409: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3969, view_1354);  sub_3969 = view_1354 = None
	        view_1356: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8409, [sym_size_int, 1500, 1280]);  mul_8409 = None
	        _assert_tensor_metadata_781 = torch.ops.aten._assert_tensor_metadata.default(view_1356, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_781 = None
	        view_1357: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg393_1, [1280, 40, 32]);  arg393_1 = None
	        view_1358: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg394_1, [1280, 40, 1]);  arg394_1 = None
	        view_1359: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg395_1, [1280, 40, 1]);  arg395_1 = None
	        _assert_tensor_metadata_782 = torch.ops.aten._assert_tensor_metadata.default(view_1357, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_782 = None
	        convert_element_type_520: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1357, torch.float32);  view_1357 = None
	        _assert_tensor_metadata_783 = torch.ops.aten._assert_tensor_metadata.default(view_1359, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_783 = None
	        convert_element_type_521: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1359, torch.float32);  view_1359 = None
	        sub_3973: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_520, convert_element_type_521);  convert_element_type_520 = convert_element_type_521 = None
	        mul_8414: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3973, view_1358);  sub_3973 = view_1358 = None
	        view_1360: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8414, [1280, 1280]);  mul_8414 = None
	        _assert_tensor_metadata_784 = torch.ops.aten._assert_tensor_metadata.default(view_1360, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_784 = None
	        mul_8419: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1361: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1356, [mul_8419, 1280]);  view_1356 = mul_8419 = None
	        permute_145: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1360, [1, 0]);  view_1360 = None
	        addmm_71: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg392_1, view_1361, permute_145);  arg392_1 = view_1361 = permute_145 = None
	        view_1362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_71, [sym_size_int, 1500, 1280]);  addmm_71 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1363: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1362, [sym_size_int, -1, 20, 64]);  view_1362 = None
	        permute_146: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1363, [0, 2, 1, 3]);  view_1363 = None
	        clone_116: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_146, memory_format = torch.contiguous_format);  permute_146 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_14 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_114, clone_115, clone_116, None, False, scale = 1.0);  clone_114 = clone_115 = clone_116 = None
	        getitem_114: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_14[0];  _scaled_dot_product_efficient_attention_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_147: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_114, [0, 2, 1, 3]);  getitem_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1364: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_147, [sym_size_int, 1500, -1]);  permute_147 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1365: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1364, [sym_size_int, 1500, 1280])
	        amin_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1365, [2])
	        amax_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1365, [2]);  view_1365 = None
	        full_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_87, full_174);  amin_87 = full_174 = None
	        full_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_87, full_175);  amax_87 = full_175 = None
	        sub_3991: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_87, minimum_87);  maximum_87 = None
	        div_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3991, 255.0);  sub_3991 = None
	        clamp_min_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_174, 1.1920928955078125e-07);  div_174 = None
	        div_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_87, clamp_min_261);  minimum_87 = None
	        round_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_175);  div_175 = None
	        sub_3997: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_175);  round_175 = None
	        clamp_min_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3997, -128);  sub_3997 = None
	        clamp_max_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_262, 127);  clamp_min_262 = None
	        _assert_tensor_metadata_785 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_261, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_785 = None
	        _assert_tensor_metadata_786 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_174, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_786 = None
	        convert_element_type_522: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_174, torch.int8);  clamp_max_174 = None
	        view_1366: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1364, [sym_size_int, 1500, 1280]);  view_1364 = None
	        view_1367: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_261, [sym_size_int, 1500, 1])
	        view_1368: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_522, [sym_size_int, 1500, 1])
	        reciprocal_87: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1367);  view_1367 = None
	        mul_8489: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_87, 1.0);  reciprocal_87 = None
	        mul_8492: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1366, mul_8489);  view_1366 = mul_8489 = None
	        round_176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8492);  mul_8492 = None
	        add_13435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_176, view_1368);  round_176 = view_1368 = None
	        clamp_min_263: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13435, -128);  add_13435 = None
	        clamp_max_175: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_263, 127);  clamp_min_263 = None
	        view_1369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_175, [sym_size_int, 1500, 1280]);  clamp_max_175 = None
	        _assert_tensor_metadata_787 = torch.ops.aten._assert_tensor_metadata.default(view_1369, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_787 = None
	        convert_element_type_523: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1369, torch.int8);  view_1369 = None
	        view_1370: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_523, [sym_size_int, 1500, 1280]);  convert_element_type_523 = None
	        view_1371: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_261, [sym_size_int, 1500, 1]);  clamp_min_261 = None
	        view_1372: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_522, [sym_size_int, 1500, 1]);  convert_element_type_522 = None
	        _assert_tensor_metadata_788 = torch.ops.aten._assert_tensor_metadata.default(view_1370, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_788 = None
	        convert_element_type_524: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1370, torch.float32);  view_1370 = None
	        _assert_tensor_metadata_789 = torch.ops.aten._assert_tensor_metadata.default(view_1372, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_789 = None
	        convert_element_type_525: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1372, torch.float32);  view_1372 = None
	        sub_4017: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_524, convert_element_type_525);  convert_element_type_524 = convert_element_type_525 = None
	        mul_8514: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4017, view_1371);  sub_4017 = view_1371 = None
	        view_1373: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8514, [sym_size_int, 1500, 1280]);  mul_8514 = None
	        _assert_tensor_metadata_790 = torch.ops.aten._assert_tensor_metadata.default(view_1373, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_790 = None
	        view_1374: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg397_1, [1280, 40, 32]);  arg397_1 = None
	        view_1375: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg398_1, [1280, 40, 1]);  arg398_1 = None
	        view_1376: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg399_1, [1280, 40, 1]);  arg399_1 = None
	        _assert_tensor_metadata_791 = torch.ops.aten._assert_tensor_metadata.default(view_1374, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_791 = None
	        convert_element_type_526: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1374, torch.float32);  view_1374 = None
	        _assert_tensor_metadata_792 = torch.ops.aten._assert_tensor_metadata.default(view_1376, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_792 = None
	        convert_element_type_527: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1376, torch.float32);  view_1376 = None
	        sub_4021: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_526, convert_element_type_527);  convert_element_type_526 = convert_element_type_527 = None
	        mul_8519: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4021, view_1375);  sub_4021 = view_1375 = None
	        view_1377: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8519, [1280, 1280]);  mul_8519 = None
	        _assert_tensor_metadata_793 = torch.ops.aten._assert_tensor_metadata.default(view_1377, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_793 = None
	        mul_8524: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1378: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1373, [mul_8524, 1280]);  view_1373 = mul_8524 = None
	        permute_148: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1377, [1, 0]);  view_1377 = None
	        addmm_72: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg396_1, view_1378, permute_148);  arg396_1 = view_1378 = permute_148 = None
	        view_1379: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_72, [sym_size_int, 1500, 1280]);  addmm_72 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1379);  view_1379 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_13498: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_12878, clone_117);  add_12878 = clone_117 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_118: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_13498, memory_format = torch.contiguous_format)
	        var_mean_29 = torch.ops.aten.var_mean.correction(clone_118, [2], correction = 0, keepdim = True)
	        getitem_118: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_29[0]
	        getitem_119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_29[1];  var_mean_29 = None
	        add_13503: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_118, 1e-05);  getitem_118 = None
	        rsqrt_29: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_13503);  add_13503 = None
	        sub_4027: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_118, getitem_119);  clone_118 = getitem_119 = None
	        mul_8535: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4027, rsqrt_29);  sub_4027 = rsqrt_29 = None
	        mul_8536: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8535, arg400_1);  mul_8535 = arg400_1 = None
	        add_13504: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_8536, arg401_1);  mul_8536 = arg401_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_13504, [sym_size_int, 1500, 1280])
	        amin_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1380, [2])
	        amax_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1380, [2]);  view_1380 = None
	        full_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_88, full_176);  amin_88 = full_176 = None
	        full_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_88, full_177);  amax_88 = full_177 = None
	        sub_4038: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_88, minimum_88);  maximum_88 = None
	        div_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4038, 255.0);  sub_4038 = None
	        clamp_min_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_176, 1.1920928955078125e-07);  div_176 = None
	        div_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_88, clamp_min_264);  minimum_88 = None
	        round_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_177);  div_177 = None
	        sub_4044: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_177);  round_177 = None
	        clamp_min_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4044, -128);  sub_4044 = None
	        clamp_max_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_265, 127);  clamp_min_265 = None
	        _assert_tensor_metadata_794 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_264, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_794 = None
	        _assert_tensor_metadata_795 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_176, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_795 = None
	        convert_element_type_528: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_176, torch.int8);  clamp_max_176 = None
	        view_1381: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_13504, [sym_size_int, 1500, 1280]);  add_13504 = None
	        view_1382: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_264, [sym_size_int, 1500, 1])
	        view_1383: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_528, [sym_size_int, 1500, 1])
	        reciprocal_88: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1382);  view_1382 = None
	        mul_8584: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_88, 1.0);  reciprocal_88 = None
	        mul_8587: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1381, mul_8584);  view_1381 = mul_8584 = None
	        round_178: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8587);  mul_8587 = None
	        add_13591: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_178, view_1383);  round_178 = view_1383 = None
	        clamp_min_266: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13591, -128);  add_13591 = None
	        clamp_max_177: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_266, 127);  clamp_min_266 = None
	        view_1384: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_177, [sym_size_int, 1500, 1280]);  clamp_max_177 = None
	        _assert_tensor_metadata_796 = torch.ops.aten._assert_tensor_metadata.default(view_1384, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_796 = None
	        convert_element_type_529: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1384, torch.int8);  view_1384 = None
	        view_1385: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_529, [sym_size_int, 1500, 1280]);  convert_element_type_529 = None
	        view_1386: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_264, [sym_size_int, 1500, 1]);  clamp_min_264 = None
	        view_1387: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_528, [sym_size_int, 1500, 1]);  convert_element_type_528 = None
	        _assert_tensor_metadata_797 = torch.ops.aten._assert_tensor_metadata.default(view_1385, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_797 = None
	        convert_element_type_530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1385, torch.float32);  view_1385 = None
	        _assert_tensor_metadata_798 = torch.ops.aten._assert_tensor_metadata.default(view_1387, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_798 = None
	        convert_element_type_531: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1387, torch.float32);  view_1387 = None
	        sub_4064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_530, convert_element_type_531);  convert_element_type_530 = convert_element_type_531 = None
	        mul_8609: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4064, view_1386);  sub_4064 = view_1386 = None
	        view_1388: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8609, [sym_size_int, 1500, 1280]);  mul_8609 = None
	        _assert_tensor_metadata_799 = torch.ops.aten._assert_tensor_metadata.default(view_1388, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_799 = None
	        view_1389: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg403_1, [5120, 40, 32]);  arg403_1 = None
	        view_1390: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg404_1, [5120, 40, 1]);  arg404_1 = None
	        view_1391: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg405_1, [5120, 40, 1]);  arg405_1 = None
	        _assert_tensor_metadata_800 = torch.ops.aten._assert_tensor_metadata.default(view_1389, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_800 = None
	        convert_element_type_532: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1389, torch.float32);  view_1389 = None
	        _assert_tensor_metadata_801 = torch.ops.aten._assert_tensor_metadata.default(view_1391, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_801 = None
	        convert_element_type_533: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1391, torch.float32);  view_1391 = None
	        sub_4068: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_532, convert_element_type_533);  convert_element_type_532 = convert_element_type_533 = None
	        mul_8614: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4068, view_1390);  sub_4068 = view_1390 = None
	        view_1392: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8614, [5120, 1280]);  mul_8614 = None
	        _assert_tensor_metadata_802 = torch.ops.aten._assert_tensor_metadata.default(view_1392, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_802 = None
	        mul_8619: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1393: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1388, [mul_8619, 1280]);  view_1388 = mul_8619 = None
	        permute_149: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1392, [1, 0]);  view_1392 = None
	        addmm_73: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg402_1, view_1393, permute_149);  arg402_1 = view_1393 = permute_149 = None
	        view_1394: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_73, [sym_size_int, 1500, 5120]);  addmm_73 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_8626: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1394, 0.5)
	        mul_8627: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1394, 0.7071067811865476);  view_1394 = None
	        erf_16: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_8627);  mul_8627 = None
	        add_13650: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_16, 1);  erf_16 = None
	        mul_8628: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8626, add_13650);  mul_8626 = add_13650 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_119: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_8628);  mul_8628 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1395: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_119, [sym_size_int, 1500, 5120])
	        amin_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1395, [2])
	        amax_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1395, [2]);  view_1395 = None
	        full_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_89, full_178);  amin_89 = full_178 = None
	        full_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_89, full_179);  amax_89 = full_179 = None
	        sub_4081: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_89, minimum_89);  maximum_89 = None
	        div_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4081, 255.0);  sub_4081 = None
	        clamp_min_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_178, 1.1920928955078125e-07);  div_178 = None
	        div_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_89, clamp_min_267);  minimum_89 = None
	        round_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_179);  div_179 = None
	        sub_4087: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_179);  round_179 = None
	        clamp_min_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4087, -128);  sub_4087 = None
	        clamp_max_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_268, 127);  clamp_min_268 = None
	        _assert_tensor_metadata_803 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_267, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_803 = None
	        _assert_tensor_metadata_804 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_178, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_804 = None
	        convert_element_type_534: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_178, torch.int8);  clamp_max_178 = None
	        view_1396: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_119, [sym_size_int, 1500, 5120]);  clone_119 = None
	        view_1397: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_267, [sym_size_int, 1500, 1])
	        view_1398: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_534, [sym_size_int, 1500, 1])
	        reciprocal_89: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1397);  view_1397 = None
	        mul_8674: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_89, 1.0);  reciprocal_89 = None
	        mul_8677: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1396, mul_8674);  view_1396 = mul_8674 = None
	        round_180: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_8677);  mul_8677 = None
	        add_13733: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_180, view_1398);  round_180 = view_1398 = None
	        clamp_min_269: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13733, -128);  add_13733 = None
	        clamp_max_179: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_269, 127);  clamp_min_269 = None
	        view_1399: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_179, [sym_size_int, 1500, 5120]);  clamp_max_179 = None
	        _assert_tensor_metadata_805 = torch.ops.aten._assert_tensor_metadata.default(view_1399, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_805 = None
	        convert_element_type_535: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1399, torch.int8);  view_1399 = None
	        view_1400: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_535, [sym_size_int, 1500, 5120]);  convert_element_type_535 = None
	        view_1401: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_267, [sym_size_int, 1500, 1]);  clamp_min_267 = None
	        view_1402: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_534, [sym_size_int, 1500, 1]);  convert_element_type_534 = None
	        _assert_tensor_metadata_806 = torch.ops.aten._assert_tensor_metadata.default(view_1400, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_806 = None
	        convert_element_type_536: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1400, torch.float32);  view_1400 = None
	        _assert_tensor_metadata_807 = torch.ops.aten._assert_tensor_metadata.default(view_1402, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_807 = None
	        convert_element_type_537: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1402, torch.float32);  view_1402 = None
	        sub_4107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_536, convert_element_type_537);  convert_element_type_536 = convert_element_type_537 = None
	        mul_8699: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4107, view_1401);  sub_4107 = view_1401 = None
	        view_1403: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_8699, [sym_size_int, 1500, 5120]);  mul_8699 = None
	        _assert_tensor_metadata_808 = torch.ops.aten._assert_tensor_metadata.default(view_1403, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_808 = None
	        view_1404: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg407_1, [1280, 160, 32]);  arg407_1 = None
	        view_1405: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg408_1, [1280, 160, 1]);  arg408_1 = None
	        view_1406: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg409_1, [1280, 160, 1]);  arg409_1 = None
	        _assert_tensor_metadata_809 = torch.ops.aten._assert_tensor_metadata.default(view_1404, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_809 = None
	        convert_element_type_538: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1404, torch.float32);  view_1404 = None
	        _assert_tensor_metadata_810 = torch.ops.aten._assert_tensor_metadata.default(view_1406, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_810 = None
	        convert_element_type_539: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1406, torch.float32);  view_1406 = None
	        sub_4111: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_538, convert_element_type_539);  convert_element_type_538 = convert_element_type_539 = None
	        mul_8704: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4111, view_1405);  sub_4111 = view_1405 = None
	        view_1407: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_8704, [1280, 5120]);  mul_8704 = None
	        _assert_tensor_metadata_811 = torch.ops.aten._assert_tensor_metadata.default(view_1407, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_811 = None
	        mul_8709: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1408: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1403, [mul_8709, 5120]);  view_1403 = mul_8709 = None
	        permute_150: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1407, [1, 0]);  view_1407 = None
	        addmm_74: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg406_1, view_1408, permute_150);  arg406_1 = view_1408 = permute_150 = None
	        view_1409: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_74, [sym_size_int, 1500, 1280]);  addmm_74 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_120: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1409);  view_1409 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_13796: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_13498, clone_120);  add_13498 = clone_120 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_121: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_13796, memory_format = torch.contiguous_format)
	        var_mean_30 = torch.ops.aten.var_mean.correction(clone_121, [2], correction = 0, keepdim = True)
	        getitem_120: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_30[0]
	        getitem_121: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_30[1];  var_mean_30 = None
	        add_13801: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_120, 1e-05);  getitem_120 = None
	        rsqrt_30: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_13801);  add_13801 = None
	        sub_4117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_121, getitem_121);  clone_121 = getitem_121 = None
	        mul_8720: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4117, rsqrt_30);  sub_4117 = rsqrt_30 = None
	        mul_8721: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8720, arg410_1);  mul_8720 = arg410_1 = None
	        add_13802: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_8721, arg411_1);  mul_8721 = arg411_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1410: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_13802, [sym_size_int, 1500, 1280])
	        amin_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1410, [2])
	        amax_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1410, [2]);  view_1410 = None
	        full_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_90, full_180);  amin_90 = full_180 = None
	        full_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_90, full_181);  amax_90 = full_181 = None
	        sub_4128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_90, minimum_90);  maximum_90 = None
	        div_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4128, 255.0);  sub_4128 = None
	        clamp_min_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_180, 1.1920928955078125e-07);  div_180 = None
	        div_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_90, clamp_min_270);  minimum_90 = None
	        round_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_181);  div_181 = None
	        sub_4134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_181);  round_181 = None
	        clamp_min_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4134, -128);  sub_4134 = None
	        clamp_max_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_271, 127);  clamp_min_271 = None
	        _assert_tensor_metadata_812 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_270, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_812 = None
	        _assert_tensor_metadata_813 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_180, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_813 = None
	        convert_element_type_540: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_180, torch.int8);  clamp_max_180 = None
	        view_1411: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_13802, [sym_size_int, 1500, 1280])
	        view_1412: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_270, [sym_size_int, 1500, 1])
	        view_1413: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_540, [sym_size_int, 1500, 1])
	        reciprocal_90: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1412);  view_1412 = None
	        mul_8769: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_90, 1.0);  reciprocal_90 = None
	        mul_8772: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1411, mul_8769);  view_1411 = mul_8769 = None
	        round_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8772);  mul_8772 = None
	        add_13889: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_182, view_1413);  round_182 = view_1413 = None
	        clamp_min_272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13889, -128);  add_13889 = None
	        clamp_max_181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_272, 127);  clamp_min_272 = None
	        view_1414: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_181, [sym_size_int, 1500, 1280]);  clamp_max_181 = None
	        _assert_tensor_metadata_814 = torch.ops.aten._assert_tensor_metadata.default(view_1414, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_814 = None
	        convert_element_type_541: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1414, torch.int8);  view_1414 = None
	        view_1415: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_541, [sym_size_int, 1500, 1280]);  convert_element_type_541 = None
	        view_1416: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_270, [sym_size_int, 1500, 1]);  clamp_min_270 = None
	        view_1417: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_540, [sym_size_int, 1500, 1]);  convert_element_type_540 = None
	        _assert_tensor_metadata_815 = torch.ops.aten._assert_tensor_metadata.default(view_1415, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_815 = None
	        convert_element_type_542: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1415, torch.float32);  view_1415 = None
	        _assert_tensor_metadata_816 = torch.ops.aten._assert_tensor_metadata.default(view_1417, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_816 = None
	        convert_element_type_543: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1417, torch.float32);  view_1417 = None
	        sub_4154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_542, convert_element_type_543);  convert_element_type_542 = convert_element_type_543 = None
	        mul_8794: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4154, view_1416);  sub_4154 = view_1416 = None
	        view_1418: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8794, [sym_size_int, 1500, 1280]);  mul_8794 = None
	        _assert_tensor_metadata_817 = torch.ops.aten._assert_tensor_metadata.default(view_1418, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_817 = None
	        view_1419: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg413_1, [1280, 40, 32]);  arg413_1 = None
	        view_1420: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg414_1, [1280, 40, 1]);  arg414_1 = None
	        view_1421: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg415_1, [1280, 40, 1]);  arg415_1 = None
	        _assert_tensor_metadata_818 = torch.ops.aten._assert_tensor_metadata.default(view_1419, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_818 = None
	        convert_element_type_544: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1419, torch.float32);  view_1419 = None
	        _assert_tensor_metadata_819 = torch.ops.aten._assert_tensor_metadata.default(view_1421, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_819 = None
	        convert_element_type_545: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1421, torch.float32);  view_1421 = None
	        sub_4158: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_544, convert_element_type_545);  convert_element_type_544 = convert_element_type_545 = None
	        mul_8799: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4158, view_1420);  sub_4158 = view_1420 = None
	        view_1422: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8799, [1280, 1280]);  mul_8799 = None
	        _assert_tensor_metadata_820 = torch.ops.aten._assert_tensor_metadata.default(view_1422, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_820 = None
	        mul_8804: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1423: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1418, [mul_8804, 1280]);  view_1418 = mul_8804 = None
	        permute_151: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1422, [1, 0]);  view_1422 = None
	        addmm_75: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg412_1, view_1423, permute_151);  arg412_1 = view_1423 = permute_151 = None
	        view_1424: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_75, [sym_size_int, 1500, 1280]);  addmm_75 = None
	        mul_8811: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1424, 0.125);  view_1424 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1425: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_8811, [sym_size_int, 1500, 20, 64]);  mul_8811 = None
	        permute_152: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1425, [0, 2, 1, 3]);  view_1425 = None
	        clone_122: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_152, memory_format = torch.contiguous_format);  permute_152 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1426: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_13802, [sym_size_int, 1500, 1280])
	        amin_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1426, [2])
	        amax_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1426, [2]);  view_1426 = None
	        full_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_91, full_182);  amin_91 = full_182 = None
	        full_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_91, full_183);  amax_91 = full_183 = None
	        sub_4173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_91, minimum_91);  maximum_91 = None
	        div_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4173, 255.0);  sub_4173 = None
	        clamp_min_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_182, 1.1920928955078125e-07);  div_182 = None
	        div_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_91, clamp_min_273);  minimum_91 = None
	        round_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_183);  div_183 = None
	        sub_4179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_183);  round_183 = None
	        clamp_min_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4179, -128);  sub_4179 = None
	        clamp_max_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_274, 127);  clamp_min_274 = None
	        _assert_tensor_metadata_821 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_273, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_821 = None
	        _assert_tensor_metadata_822 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_182, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_822 = None
	        convert_element_type_546: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_182, torch.int8);  clamp_max_182 = None
	        view_1427: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_13802, [sym_size_int, 1500, 1280])
	        view_1428: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_273, [sym_size_int, 1500, 1])
	        view_1429: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_546, [sym_size_int, 1500, 1])
	        reciprocal_91: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1428);  view_1428 = None
	        mul_8865: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_91, 1.0);  reciprocal_91 = None
	        mul_8868: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1427, mul_8865);  view_1427 = mul_8865 = None
	        round_184: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8868);  mul_8868 = None
	        add_14041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_184, view_1429);  round_184 = view_1429 = None
	        clamp_min_275: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14041, -128);  add_14041 = None
	        clamp_max_183: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_275, 127);  clamp_min_275 = None
	        view_1430: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_183, [sym_size_int, 1500, 1280]);  clamp_max_183 = None
	        _assert_tensor_metadata_823 = torch.ops.aten._assert_tensor_metadata.default(view_1430, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_823 = None
	        convert_element_type_547: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1430, torch.int8);  view_1430 = None
	        view_1431: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_547, [sym_size_int, 1500, 1280]);  convert_element_type_547 = None
	        view_1432: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_273, [sym_size_int, 1500, 1]);  clamp_min_273 = None
	        view_1433: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_546, [sym_size_int, 1500, 1]);  convert_element_type_546 = None
	        _assert_tensor_metadata_824 = torch.ops.aten._assert_tensor_metadata.default(view_1431, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_824 = None
	        convert_element_type_548: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1431, torch.float32);  view_1431 = None
	        _assert_tensor_metadata_825 = torch.ops.aten._assert_tensor_metadata.default(view_1433, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_825 = None
	        convert_element_type_549: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1433, torch.float32);  view_1433 = None
	        sub_4199: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_548, convert_element_type_549);  convert_element_type_548 = convert_element_type_549 = None
	        mul_8890: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4199, view_1432);  sub_4199 = view_1432 = None
	        view_1434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8890, [sym_size_int, 1500, 1280]);  mul_8890 = None
	        _assert_tensor_metadata_826 = torch.ops.aten._assert_tensor_metadata.default(view_1434, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_826 = None
	        view_1435: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg416_1, [1280, 40, 32]);  arg416_1 = None
	        view_1436: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg417_1, [1280, 40, 1]);  arg417_1 = None
	        view_1437: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg418_1, [1280, 40, 1]);  arg418_1 = None
	        _assert_tensor_metadata_827 = torch.ops.aten._assert_tensor_metadata.default(view_1435, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_827 = None
	        convert_element_type_550: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1435, torch.float32);  view_1435 = None
	        _assert_tensor_metadata_828 = torch.ops.aten._assert_tensor_metadata.default(view_1437, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_828 = None
	        convert_element_type_551: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1437, torch.float32);  view_1437 = None
	        sub_4203: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_550, convert_element_type_551);  convert_element_type_550 = convert_element_type_551 = None
	        mul_8895: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4203, view_1436);  sub_4203 = view_1436 = None
	        view_1438: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8895, [1280, 1280]);  mul_8895 = None
	        _assert_tensor_metadata_829 = torch.ops.aten._assert_tensor_metadata.default(view_1438, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_829 = None
	        permute_153: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1438, [1, 0]);  view_1438 = None
	        mul_8898: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1439: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1434, [mul_8898, 1280]);  view_1434 = mul_8898 = None
	        mm_15: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1439, permute_153);  view_1439 = permute_153 = None
	        view_1440: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_15, [sym_size_int, 1500, 1280]);  mm_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1441: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1440, [sym_size_int, -1, 20, 64]);  view_1440 = None
	        permute_154: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1441, [0, 2, 1, 3]);  view_1441 = None
	        clone_123: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_154, memory_format = torch.contiguous_format);  permute_154 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1442: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_13802, [sym_size_int, 1500, 1280])
	        amin_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1442, [2])
	        amax_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1442, [2]);  view_1442 = None
	        full_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_92, full_184);  amin_92 = full_184 = None
	        full_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_92, full_185);  amax_92 = full_185 = None
	        sub_4217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_92, minimum_92);  maximum_92 = None
	        div_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4217, 255.0);  sub_4217 = None
	        clamp_min_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_184, 1.1920928955078125e-07);  div_184 = None
	        div_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_92, clamp_min_276);  minimum_92 = None
	        round_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_185);  div_185 = None
	        sub_4223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_185);  round_185 = None
	        clamp_min_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4223, -128);  sub_4223 = None
	        clamp_max_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_277, 127);  clamp_min_277 = None
	        _assert_tensor_metadata_830 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_276, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_830 = None
	        _assert_tensor_metadata_831 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_184, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_831 = None
	        convert_element_type_552: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_184, torch.int8);  clamp_max_184 = None
	        view_1443: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_13802, [sym_size_int, 1500, 1280]);  add_13802 = None
	        view_1444: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_276, [sym_size_int, 1500, 1])
	        view_1445: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_552, [sym_size_int, 1500, 1])
	        reciprocal_92: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1444);  view_1444 = None
	        mul_8964: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_92, 1.0);  reciprocal_92 = None
	        mul_8967: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1443, mul_8964);  view_1443 = mul_8964 = None
	        round_186: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8967);  mul_8967 = None
	        add_14189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_186, view_1445);  round_186 = view_1445 = None
	        clamp_min_278: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14189, -128);  add_14189 = None
	        clamp_max_185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_278, 127);  clamp_min_278 = None
	        view_1446: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_185, [sym_size_int, 1500, 1280]);  clamp_max_185 = None
	        _assert_tensor_metadata_832 = torch.ops.aten._assert_tensor_metadata.default(view_1446, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_832 = None
	        convert_element_type_553: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1446, torch.int8);  view_1446 = None
	        view_1447: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_553, [sym_size_int, 1500, 1280]);  convert_element_type_553 = None
	        view_1448: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_276, [sym_size_int, 1500, 1]);  clamp_min_276 = None
	        view_1449: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_552, [sym_size_int, 1500, 1]);  convert_element_type_552 = None
	        _assert_tensor_metadata_833 = torch.ops.aten._assert_tensor_metadata.default(view_1447, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_833 = None
	        convert_element_type_554: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1447, torch.float32);  view_1447 = None
	        _assert_tensor_metadata_834 = torch.ops.aten._assert_tensor_metadata.default(view_1449, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_834 = None
	        convert_element_type_555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1449, torch.float32);  view_1449 = None
	        sub_4243: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_554, convert_element_type_555);  convert_element_type_554 = convert_element_type_555 = None
	        mul_8989: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4243, view_1448);  sub_4243 = view_1448 = None
	        view_1450: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8989, [sym_size_int, 1500, 1280]);  mul_8989 = None
	        _assert_tensor_metadata_835 = torch.ops.aten._assert_tensor_metadata.default(view_1450, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_835 = None
	        view_1451: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg420_1, [1280, 40, 32]);  arg420_1 = None
	        view_1452: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg421_1, [1280, 40, 1]);  arg421_1 = None
	        view_1453: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg422_1, [1280, 40, 1]);  arg422_1 = None
	        _assert_tensor_metadata_836 = torch.ops.aten._assert_tensor_metadata.default(view_1451, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_836 = None
	        convert_element_type_556: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1451, torch.float32);  view_1451 = None
	        _assert_tensor_metadata_837 = torch.ops.aten._assert_tensor_metadata.default(view_1453, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_837 = None
	        convert_element_type_557: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1453, torch.float32);  view_1453 = None
	        sub_4247: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_556, convert_element_type_557);  convert_element_type_556 = convert_element_type_557 = None
	        mul_8994: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4247, view_1452);  sub_4247 = view_1452 = None
	        view_1454: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8994, [1280, 1280]);  mul_8994 = None
	        _assert_tensor_metadata_838 = torch.ops.aten._assert_tensor_metadata.default(view_1454, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_838 = None
	        mul_8999: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1455: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1450, [mul_8999, 1280]);  view_1450 = mul_8999 = None
	        permute_155: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1454, [1, 0]);  view_1454 = None
	        addmm_76: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg419_1, view_1455, permute_155);  arg419_1 = view_1455 = permute_155 = None
	        view_1456: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_76, [sym_size_int, 1500, 1280]);  addmm_76 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1457: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1456, [sym_size_int, -1, 20, 64]);  view_1456 = None
	        permute_156: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1457, [0, 2, 1, 3]);  view_1457 = None
	        clone_124: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_156, memory_format = torch.contiguous_format);  permute_156 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_15 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_122, clone_123, clone_124, None, False, scale = 1.0);  clone_122 = clone_123 = clone_124 = None
	        getitem_122: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_15[0];  _scaled_dot_product_efficient_attention_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_157: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_122, [0, 2, 1, 3]);  getitem_122 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_157, [sym_size_int, 1500, -1]);  permute_157 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1459: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1458, [sym_size_int, 1500, 1280])
	        amin_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1459, [2])
	        amax_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1459, [2]);  view_1459 = None
	        full_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_93, full_186);  amin_93 = full_186 = None
	        full_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_93, full_187);  amax_93 = full_187 = None
	        sub_4265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_93, minimum_93);  maximum_93 = None
	        div_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4265, 255.0);  sub_4265 = None
	        clamp_min_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_186, 1.1920928955078125e-07);  div_186 = None
	        div_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_93, clamp_min_279);  minimum_93 = None
	        round_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_187);  div_187 = None
	        sub_4271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_187);  round_187 = None
	        clamp_min_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4271, -128);  sub_4271 = None
	        clamp_max_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_280, 127);  clamp_min_280 = None
	        _assert_tensor_metadata_839 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_279, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_839 = None
	        _assert_tensor_metadata_840 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_186, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_840 = None
	        convert_element_type_558: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_186, torch.int8);  clamp_max_186 = None
	        view_1460: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1458, [sym_size_int, 1500, 1280]);  view_1458 = None
	        view_1461: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_279, [sym_size_int, 1500, 1])
	        view_1462: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_558, [sym_size_int, 1500, 1])
	        reciprocal_93: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1461);  view_1461 = None
	        mul_9069: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_93, 1.0);  reciprocal_93 = None
	        mul_9072: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1460, mul_9069);  view_1460 = mul_9069 = None
	        round_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9072);  mul_9072 = None
	        add_14353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_188, view_1462);  round_188 = view_1462 = None
	        clamp_min_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14353, -128);  add_14353 = None
	        clamp_max_187: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_281, 127);  clamp_min_281 = None
	        view_1463: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_187, [sym_size_int, 1500, 1280]);  clamp_max_187 = None
	        _assert_tensor_metadata_841 = torch.ops.aten._assert_tensor_metadata.default(view_1463, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_841 = None
	        convert_element_type_559: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1463, torch.int8);  view_1463 = None
	        view_1464: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_559, [sym_size_int, 1500, 1280]);  convert_element_type_559 = None
	        view_1465: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_279, [sym_size_int, 1500, 1]);  clamp_min_279 = None
	        view_1466: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_558, [sym_size_int, 1500, 1]);  convert_element_type_558 = None
	        _assert_tensor_metadata_842 = torch.ops.aten._assert_tensor_metadata.default(view_1464, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_842 = None
	        convert_element_type_560: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1464, torch.float32);  view_1464 = None
	        _assert_tensor_metadata_843 = torch.ops.aten._assert_tensor_metadata.default(view_1466, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_843 = None
	        convert_element_type_561: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1466, torch.float32);  view_1466 = None
	        sub_4291: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_560, convert_element_type_561);  convert_element_type_560 = convert_element_type_561 = None
	        mul_9094: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4291, view_1465);  sub_4291 = view_1465 = None
	        view_1467: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9094, [sym_size_int, 1500, 1280]);  mul_9094 = None
	        _assert_tensor_metadata_844 = torch.ops.aten._assert_tensor_metadata.default(view_1467, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_844 = None
	        view_1468: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg424_1, [1280, 40, 32]);  arg424_1 = None
	        view_1469: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg425_1, [1280, 40, 1]);  arg425_1 = None
	        view_1470: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg426_1, [1280, 40, 1]);  arg426_1 = None
	        _assert_tensor_metadata_845 = torch.ops.aten._assert_tensor_metadata.default(view_1468, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_845 = None
	        convert_element_type_562: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1468, torch.float32);  view_1468 = None
	        _assert_tensor_metadata_846 = torch.ops.aten._assert_tensor_metadata.default(view_1470, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_846 = None
	        convert_element_type_563: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1470, torch.float32);  view_1470 = None
	        sub_4295: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_562, convert_element_type_563);  convert_element_type_562 = convert_element_type_563 = None
	        mul_9099: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4295, view_1469);  sub_4295 = view_1469 = None
	        view_1471: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9099, [1280, 1280]);  mul_9099 = None
	        _assert_tensor_metadata_847 = torch.ops.aten._assert_tensor_metadata.default(view_1471, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_847 = None
	        mul_9104: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1472: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1467, [mul_9104, 1280]);  view_1467 = mul_9104 = None
	        permute_158: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1471, [1, 0]);  view_1471 = None
	        addmm_77: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg423_1, view_1472, permute_158);  arg423_1 = view_1472 = permute_158 = None
	        view_1473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_77, [sym_size_int, 1500, 1280]);  addmm_77 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_125: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1473);  view_1473 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_14416: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_13796, clone_125);  add_13796 = clone_125 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_126: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_14416, memory_format = torch.contiguous_format)
	        var_mean_31 = torch.ops.aten.var_mean.correction(clone_126, [2], correction = 0, keepdim = True)
	        getitem_126: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_31[0]
	        getitem_127: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_31[1];  var_mean_31 = None
	        add_14421: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_126, 1e-05);  getitem_126 = None
	        rsqrt_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_14421);  add_14421 = None
	        sub_4301: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_126, getitem_127);  clone_126 = getitem_127 = None
	        mul_9115: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4301, rsqrt_31);  sub_4301 = rsqrt_31 = None
	        mul_9116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9115, arg427_1);  mul_9115 = arg427_1 = None
	        add_14422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9116, arg428_1);  mul_9116 = arg428_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_14422, [sym_size_int, 1500, 1280])
	        amin_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1474, [2])
	        amax_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1474, [2]);  view_1474 = None
	        full_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_94, full_188);  amin_94 = full_188 = None
	        full_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_94, full_189);  amax_94 = full_189 = None
	        sub_4312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_94, minimum_94);  maximum_94 = None
	        div_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4312, 255.0);  sub_4312 = None
	        clamp_min_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_188, 1.1920928955078125e-07);  div_188 = None
	        div_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_94, clamp_min_282);  minimum_94 = None
	        round_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_189);  div_189 = None
	        sub_4318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_189);  round_189 = None
	        clamp_min_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4318, -128);  sub_4318 = None
	        clamp_max_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_283, 127);  clamp_min_283 = None
	        _assert_tensor_metadata_848 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_282, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_848 = None
	        _assert_tensor_metadata_849 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_188, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_849 = None
	        convert_element_type_564: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_188, torch.int8);  clamp_max_188 = None
	        view_1475: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_14422, [sym_size_int, 1500, 1280]);  add_14422 = None
	        view_1476: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_282, [sym_size_int, 1500, 1])
	        view_1477: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_564, [sym_size_int, 1500, 1])
	        reciprocal_94: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1476);  view_1476 = None
	        mul_9164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_94, 1.0);  reciprocal_94 = None
	        mul_9167: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1475, mul_9164);  view_1475 = mul_9164 = None
	        round_190: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9167);  mul_9167 = None
	        add_14509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_190, view_1477);  round_190 = view_1477 = None
	        clamp_min_284: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14509, -128);  add_14509 = None
	        clamp_max_189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_284, 127);  clamp_min_284 = None
	        view_1478: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_189, [sym_size_int, 1500, 1280]);  clamp_max_189 = None
	        _assert_tensor_metadata_850 = torch.ops.aten._assert_tensor_metadata.default(view_1478, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_850 = None
	        convert_element_type_565: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1478, torch.int8);  view_1478 = None
	        view_1479: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_565, [sym_size_int, 1500, 1280]);  convert_element_type_565 = None
	        view_1480: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_282, [sym_size_int, 1500, 1]);  clamp_min_282 = None
	        view_1481: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_564, [sym_size_int, 1500, 1]);  convert_element_type_564 = None
	        _assert_tensor_metadata_851 = torch.ops.aten._assert_tensor_metadata.default(view_1479, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_851 = None
	        convert_element_type_566: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1479, torch.float32);  view_1479 = None
	        _assert_tensor_metadata_852 = torch.ops.aten._assert_tensor_metadata.default(view_1481, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_852 = None
	        convert_element_type_567: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1481, torch.float32);  view_1481 = None
	        sub_4338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_566, convert_element_type_567);  convert_element_type_566 = convert_element_type_567 = None
	        mul_9189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4338, view_1480);  sub_4338 = view_1480 = None
	        view_1482: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9189, [sym_size_int, 1500, 1280]);  mul_9189 = None
	        _assert_tensor_metadata_853 = torch.ops.aten._assert_tensor_metadata.default(view_1482, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_853 = None
	        view_1483: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg430_1, [5120, 40, 32]);  arg430_1 = None
	        view_1484: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg431_1, [5120, 40, 1]);  arg431_1 = None
	        view_1485: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg432_1, [5120, 40, 1]);  arg432_1 = None
	        _assert_tensor_metadata_854 = torch.ops.aten._assert_tensor_metadata.default(view_1483, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_854 = None
	        convert_element_type_568: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1483, torch.float32);  view_1483 = None
	        _assert_tensor_metadata_855 = torch.ops.aten._assert_tensor_metadata.default(view_1485, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_855 = None
	        convert_element_type_569: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1485, torch.float32);  view_1485 = None
	        sub_4342: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_568, convert_element_type_569);  convert_element_type_568 = convert_element_type_569 = None
	        mul_9194: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4342, view_1484);  sub_4342 = view_1484 = None
	        view_1486: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9194, [5120, 1280]);  mul_9194 = None
	        _assert_tensor_metadata_856 = torch.ops.aten._assert_tensor_metadata.default(view_1486, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_856 = None
	        mul_9199: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1487: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1482, [mul_9199, 1280]);  view_1482 = mul_9199 = None
	        permute_159: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1486, [1, 0]);  view_1486 = None
	        addmm_78: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg429_1, view_1487, permute_159);  arg429_1 = view_1487 = permute_159 = None
	        view_1488: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_78, [sym_size_int, 1500, 5120]);  addmm_78 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_9206: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1488, 0.5)
	        mul_9207: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1488, 0.7071067811865476);  view_1488 = None
	        erf_17: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_9207);  mul_9207 = None
	        add_14568: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_17, 1);  erf_17 = None
	        mul_9208: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9206, add_14568);  mul_9206 = add_14568 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_127: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_9208);  mul_9208 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1489: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_127, [sym_size_int, 1500, 5120])
	        amin_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1489, [2])
	        amax_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1489, [2]);  view_1489 = None
	        full_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_95, full_190);  amin_95 = full_190 = None
	        full_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_95, full_191);  amax_95 = full_191 = None
	        sub_4355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_95, minimum_95);  maximum_95 = None
	        div_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4355, 255.0);  sub_4355 = None
	        clamp_min_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_190, 1.1920928955078125e-07);  div_190 = None
	        div_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_95, clamp_min_285);  minimum_95 = None
	        round_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_191);  div_191 = None
	        sub_4361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_191);  round_191 = None
	        clamp_min_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4361, -128);  sub_4361 = None
	        clamp_max_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_286, 127);  clamp_min_286 = None
	        _assert_tensor_metadata_857 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_285, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_857 = None
	        _assert_tensor_metadata_858 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_190, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_858 = None
	        convert_element_type_570: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_190, torch.int8);  clamp_max_190 = None
	        view_1490: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_127, [sym_size_int, 1500, 5120]);  clone_127 = None
	        view_1491: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_285, [sym_size_int, 1500, 1])
	        view_1492: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_570, [sym_size_int, 1500, 1])
	        reciprocal_95: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1491);  view_1491 = None
	        mul_9254: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_95, 1.0);  reciprocal_95 = None
	        mul_9257: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1490, mul_9254);  view_1490 = mul_9254 = None
	        round_192: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_9257);  mul_9257 = None
	        add_14651: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_192, view_1492);  round_192 = view_1492 = None
	        clamp_min_287: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14651, -128);  add_14651 = None
	        clamp_max_191: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_287, 127);  clamp_min_287 = None
	        view_1493: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_191, [sym_size_int, 1500, 5120]);  clamp_max_191 = None
	        _assert_tensor_metadata_859 = torch.ops.aten._assert_tensor_metadata.default(view_1493, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_859 = None
	        convert_element_type_571: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1493, torch.int8);  view_1493 = None
	        view_1494: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_571, [sym_size_int, 1500, 5120]);  convert_element_type_571 = None
	        view_1495: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_285, [sym_size_int, 1500, 1]);  clamp_min_285 = None
	        view_1496: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_570, [sym_size_int, 1500, 1]);  convert_element_type_570 = None
	        _assert_tensor_metadata_860 = torch.ops.aten._assert_tensor_metadata.default(view_1494, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_860 = None
	        convert_element_type_572: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1494, torch.float32);  view_1494 = None
	        _assert_tensor_metadata_861 = torch.ops.aten._assert_tensor_metadata.default(view_1496, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_861 = None
	        convert_element_type_573: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1496, torch.float32);  view_1496 = None
	        sub_4381: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_572, convert_element_type_573);  convert_element_type_572 = convert_element_type_573 = None
	        mul_9279: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4381, view_1495);  sub_4381 = view_1495 = None
	        view_1497: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_9279, [sym_size_int, 1500, 5120]);  mul_9279 = None
	        _assert_tensor_metadata_862 = torch.ops.aten._assert_tensor_metadata.default(view_1497, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_862 = None
	        view_1498: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg434_1, [1280, 160, 32]);  arg434_1 = None
	        view_1499: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg435_1, [1280, 160, 1]);  arg435_1 = None
	        view_1500: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg436_1, [1280, 160, 1]);  arg436_1 = None
	        _assert_tensor_metadata_863 = torch.ops.aten._assert_tensor_metadata.default(view_1498, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_863 = None
	        convert_element_type_574: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1498, torch.float32);  view_1498 = None
	        _assert_tensor_metadata_864 = torch.ops.aten._assert_tensor_metadata.default(view_1500, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_864 = None
	        convert_element_type_575: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1500, torch.float32);  view_1500 = None
	        sub_4385: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_574, convert_element_type_575);  convert_element_type_574 = convert_element_type_575 = None
	        mul_9284: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4385, view_1499);  sub_4385 = view_1499 = None
	        view_1501: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_9284, [1280, 5120]);  mul_9284 = None
	        _assert_tensor_metadata_865 = torch.ops.aten._assert_tensor_metadata.default(view_1501, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_865 = None
	        mul_9289: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1502: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1497, [mul_9289, 5120]);  view_1497 = mul_9289 = None
	        permute_160: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1501, [1, 0]);  view_1501 = None
	        addmm_79: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg433_1, view_1502, permute_160);  arg433_1 = view_1502 = permute_160 = None
	        view_1503: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_79, [sym_size_int, 1500, 1280]);  addmm_79 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1503);  view_1503 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_14714: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_14416, clone_128);  add_14416 = clone_128 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_129: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_14714, memory_format = torch.contiguous_format)
	        var_mean_32 = torch.ops.aten.var_mean.correction(clone_129, [2], correction = 0, keepdim = True)
	        getitem_128: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_32[0]
	        getitem_129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_32[1];  var_mean_32 = None
	        add_14719: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_128, 1e-05);  getitem_128 = None
	        rsqrt_32: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_14719);  add_14719 = None
	        sub_4391: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_129, getitem_129);  clone_129 = getitem_129 = None
	        mul_9300: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4391, rsqrt_32);  sub_4391 = rsqrt_32 = None
	        mul_9301: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9300, arg437_1);  mul_9300 = arg437_1 = None
	        add_14720: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9301, arg438_1);  mul_9301 = arg438_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1504: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_14720, [sym_size_int, 1500, 1280])
	        amin_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1504, [2])
	        amax_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1504, [2]);  view_1504 = None
	        full_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_96, full_192);  amin_96 = full_192 = None
	        full_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_96, full_193);  amax_96 = full_193 = None
	        sub_4402: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_96, minimum_96);  maximum_96 = None
	        div_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4402, 255.0);  sub_4402 = None
	        clamp_min_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_192, 1.1920928955078125e-07);  div_192 = None
	        div_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_96, clamp_min_288);  minimum_96 = None
	        round_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_193);  div_193 = None
	        sub_4408: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_193);  round_193 = None
	        clamp_min_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4408, -128);  sub_4408 = None
	        clamp_max_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_289, 127);  clamp_min_289 = None
	        _assert_tensor_metadata_866 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_288, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_866 = None
	        _assert_tensor_metadata_867 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_192, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_867 = None
	        convert_element_type_576: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_192, torch.int8);  clamp_max_192 = None
	        view_1505: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_14720, [sym_size_int, 1500, 1280])
	        view_1506: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_288, [sym_size_int, 1500, 1])
	        view_1507: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_576, [sym_size_int, 1500, 1])
	        reciprocal_96: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1506);  view_1506 = None
	        mul_9349: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_96, 1.0);  reciprocal_96 = None
	        mul_9352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1505, mul_9349);  view_1505 = mul_9349 = None
	        round_194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9352);  mul_9352 = None
	        add_14807: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_194, view_1507);  round_194 = view_1507 = None
	        clamp_min_290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14807, -128);  add_14807 = None
	        clamp_max_193: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_290, 127);  clamp_min_290 = None
	        view_1508: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_193, [sym_size_int, 1500, 1280]);  clamp_max_193 = None
	        _assert_tensor_metadata_868 = torch.ops.aten._assert_tensor_metadata.default(view_1508, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_868 = None
	        convert_element_type_577: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1508, torch.int8);  view_1508 = None
	        view_1509: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_577, [sym_size_int, 1500, 1280]);  convert_element_type_577 = None
	        view_1510: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_288, [sym_size_int, 1500, 1]);  clamp_min_288 = None
	        view_1511: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_576, [sym_size_int, 1500, 1]);  convert_element_type_576 = None
	        _assert_tensor_metadata_869 = torch.ops.aten._assert_tensor_metadata.default(view_1509, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_869 = None
	        convert_element_type_578: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1509, torch.float32);  view_1509 = None
	        _assert_tensor_metadata_870 = torch.ops.aten._assert_tensor_metadata.default(view_1511, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_870 = None
	        convert_element_type_579: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1511, torch.float32);  view_1511 = None
	        sub_4428: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_578, convert_element_type_579);  convert_element_type_578 = convert_element_type_579 = None
	        mul_9374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4428, view_1510);  sub_4428 = view_1510 = None
	        view_1512: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9374, [sym_size_int, 1500, 1280]);  mul_9374 = None
	        _assert_tensor_metadata_871 = torch.ops.aten._assert_tensor_metadata.default(view_1512, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_871 = None
	        view_1513: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg440_1, [1280, 40, 32]);  arg440_1 = None
	        view_1514: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg441_1, [1280, 40, 1]);  arg441_1 = None
	        view_1515: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg442_1, [1280, 40, 1]);  arg442_1 = None
	        _assert_tensor_metadata_872 = torch.ops.aten._assert_tensor_metadata.default(view_1513, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_872 = None
	        convert_element_type_580: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1513, torch.float32);  view_1513 = None
	        _assert_tensor_metadata_873 = torch.ops.aten._assert_tensor_metadata.default(view_1515, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_873 = None
	        convert_element_type_581: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1515, torch.float32);  view_1515 = None
	        sub_4432: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_580, convert_element_type_581);  convert_element_type_580 = convert_element_type_581 = None
	        mul_9379: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4432, view_1514);  sub_4432 = view_1514 = None
	        view_1516: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9379, [1280, 1280]);  mul_9379 = None
	        _assert_tensor_metadata_874 = torch.ops.aten._assert_tensor_metadata.default(view_1516, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_874 = None
	        mul_9384: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1517: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1512, [mul_9384, 1280]);  view_1512 = mul_9384 = None
	        permute_161: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1516, [1, 0]);  view_1516 = None
	        addmm_80: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg439_1, view_1517, permute_161);  arg439_1 = view_1517 = permute_161 = None
	        view_1518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_80, [sym_size_int, 1500, 1280]);  addmm_80 = None
	        mul_9391: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1518, 0.125);  view_1518 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1519: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_9391, [sym_size_int, 1500, 20, 64]);  mul_9391 = None
	        permute_162: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1519, [0, 2, 1, 3]);  view_1519 = None
	        clone_130: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_162, memory_format = torch.contiguous_format);  permute_162 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1520: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_14720, [sym_size_int, 1500, 1280])
	        amin_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1520, [2])
	        amax_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1520, [2]);  view_1520 = None
	        full_194: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_97, full_194);  amin_97 = full_194 = None
	        full_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_97, full_195);  amax_97 = full_195 = None
	        sub_4447: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_97, minimum_97);  maximum_97 = None
	        div_194: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4447, 255.0);  sub_4447 = None
	        clamp_min_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_194, 1.1920928955078125e-07);  div_194 = None
	        div_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_97, clamp_min_291);  minimum_97 = None
	        round_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_195);  div_195 = None
	        sub_4453: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_195);  round_195 = None
	        clamp_min_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4453, -128);  sub_4453 = None
	        clamp_max_194: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_292, 127);  clamp_min_292 = None
	        _assert_tensor_metadata_875 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_291, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_875 = None
	        _assert_tensor_metadata_876 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_194, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_876 = None
	        convert_element_type_582: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_194, torch.int8);  clamp_max_194 = None
	        view_1521: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_14720, [sym_size_int, 1500, 1280])
	        view_1522: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_291, [sym_size_int, 1500, 1])
	        view_1523: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_582, [sym_size_int, 1500, 1])
	        reciprocal_97: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1522);  view_1522 = None
	        mul_9445: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_97, 1.0);  reciprocal_97 = None
	        mul_9448: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1521, mul_9445);  view_1521 = mul_9445 = None
	        round_196: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9448);  mul_9448 = None
	        add_14959: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_196, view_1523);  round_196 = view_1523 = None
	        clamp_min_293: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14959, -128);  add_14959 = None
	        clamp_max_195: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_293, 127);  clamp_min_293 = None
	        view_1524: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_195, [sym_size_int, 1500, 1280]);  clamp_max_195 = None
	        _assert_tensor_metadata_877 = torch.ops.aten._assert_tensor_metadata.default(view_1524, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_877 = None
	        convert_element_type_583: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1524, torch.int8);  view_1524 = None
	        view_1525: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_583, [sym_size_int, 1500, 1280]);  convert_element_type_583 = None
	        view_1526: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_291, [sym_size_int, 1500, 1]);  clamp_min_291 = None
	        view_1527: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_582, [sym_size_int, 1500, 1]);  convert_element_type_582 = None
	        _assert_tensor_metadata_878 = torch.ops.aten._assert_tensor_metadata.default(view_1525, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_878 = None
	        convert_element_type_584: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1525, torch.float32);  view_1525 = None
	        _assert_tensor_metadata_879 = torch.ops.aten._assert_tensor_metadata.default(view_1527, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_879 = None
	        convert_element_type_585: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1527, torch.float32);  view_1527 = None
	        sub_4473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_584, convert_element_type_585);  convert_element_type_584 = convert_element_type_585 = None
	        mul_9470: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4473, view_1526);  sub_4473 = view_1526 = None
	        view_1528: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9470, [sym_size_int, 1500, 1280]);  mul_9470 = None
	        _assert_tensor_metadata_880 = torch.ops.aten._assert_tensor_metadata.default(view_1528, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_880 = None
	        view_1529: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg443_1, [1280, 40, 32]);  arg443_1 = None
	        view_1530: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg444_1, [1280, 40, 1]);  arg444_1 = None
	        view_1531: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg445_1, [1280, 40, 1]);  arg445_1 = None
	        _assert_tensor_metadata_881 = torch.ops.aten._assert_tensor_metadata.default(view_1529, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_881 = None
	        convert_element_type_586: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1529, torch.float32);  view_1529 = None
	        _assert_tensor_metadata_882 = torch.ops.aten._assert_tensor_metadata.default(view_1531, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_882 = None
	        convert_element_type_587: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1531, torch.float32);  view_1531 = None
	        sub_4477: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_586, convert_element_type_587);  convert_element_type_586 = convert_element_type_587 = None
	        mul_9475: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4477, view_1530);  sub_4477 = view_1530 = None
	        view_1532: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9475, [1280, 1280]);  mul_9475 = None
	        _assert_tensor_metadata_883 = torch.ops.aten._assert_tensor_metadata.default(view_1532, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_883 = None
	        permute_163: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1532, [1, 0]);  view_1532 = None
	        mul_9478: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1533: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1528, [mul_9478, 1280]);  view_1528 = mul_9478 = None
	        mm_16: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1533, permute_163);  view_1533 = permute_163 = None
	        view_1534: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_16, [sym_size_int, 1500, 1280]);  mm_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1535: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1534, [sym_size_int, -1, 20, 64]);  view_1534 = None
	        permute_164: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1535, [0, 2, 1, 3]);  view_1535 = None
	        clone_131: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_164, memory_format = torch.contiguous_format);  permute_164 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1536: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_14720, [sym_size_int, 1500, 1280])
	        amin_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1536, [2])
	        amax_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1536, [2]);  view_1536 = None
	        full_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_98, full_196);  amin_98 = full_196 = None
	        full_197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_98, full_197);  amax_98 = full_197 = None
	        sub_4491: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_98, minimum_98);  maximum_98 = None
	        div_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4491, 255.0);  sub_4491 = None
	        clamp_min_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_196, 1.1920928955078125e-07);  div_196 = None
	        div_197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_98, clamp_min_294);  minimum_98 = None
	        round_197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_197);  div_197 = None
	        sub_4497: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_197);  round_197 = None
	        clamp_min_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4497, -128);  sub_4497 = None
	        clamp_max_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_295, 127);  clamp_min_295 = None
	        _assert_tensor_metadata_884 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_884 = None
	        _assert_tensor_metadata_885 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_196, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_885 = None
	        convert_element_type_588: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_196, torch.int8);  clamp_max_196 = None
	        view_1537: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_14720, [sym_size_int, 1500, 1280]);  add_14720 = None
	        view_1538: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_294, [sym_size_int, 1500, 1])
	        view_1539: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_588, [sym_size_int, 1500, 1])
	        reciprocal_98: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1538);  view_1538 = None
	        mul_9544: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_98, 1.0);  reciprocal_98 = None
	        mul_9547: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1537, mul_9544);  view_1537 = mul_9544 = None
	        round_198: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9547);  mul_9547 = None
	        add_15107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_198, view_1539);  round_198 = view_1539 = None
	        clamp_min_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15107, -128);  add_15107 = None
	        clamp_max_197: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_296, 127);  clamp_min_296 = None
	        view_1540: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_197, [sym_size_int, 1500, 1280]);  clamp_max_197 = None
	        _assert_tensor_metadata_886 = torch.ops.aten._assert_tensor_metadata.default(view_1540, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_886 = None
	        convert_element_type_589: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1540, torch.int8);  view_1540 = None
	        view_1541: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_589, [sym_size_int, 1500, 1280]);  convert_element_type_589 = None
	        view_1542: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_294, [sym_size_int, 1500, 1]);  clamp_min_294 = None
	        view_1543: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_588, [sym_size_int, 1500, 1]);  convert_element_type_588 = None
	        _assert_tensor_metadata_887 = torch.ops.aten._assert_tensor_metadata.default(view_1541, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_887 = None
	        convert_element_type_590: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1541, torch.float32);  view_1541 = None
	        _assert_tensor_metadata_888 = torch.ops.aten._assert_tensor_metadata.default(view_1543, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_888 = None
	        convert_element_type_591: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1543, torch.float32);  view_1543 = None
	        sub_4517: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_590, convert_element_type_591);  convert_element_type_590 = convert_element_type_591 = None
	        mul_9569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4517, view_1542);  sub_4517 = view_1542 = None
	        view_1544: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9569, [sym_size_int, 1500, 1280]);  mul_9569 = None
	        _assert_tensor_metadata_889 = torch.ops.aten._assert_tensor_metadata.default(view_1544, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_889 = None
	        view_1545: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg447_1, [1280, 40, 32]);  arg447_1 = None
	        view_1546: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg448_1, [1280, 40, 1]);  arg448_1 = None
	        view_1547: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg449_1, [1280, 40, 1]);  arg449_1 = None
	        _assert_tensor_metadata_890 = torch.ops.aten._assert_tensor_metadata.default(view_1545, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_890 = None
	        convert_element_type_592: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1545, torch.float32);  view_1545 = None
	        _assert_tensor_metadata_891 = torch.ops.aten._assert_tensor_metadata.default(view_1547, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_891 = None
	        convert_element_type_593: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1547, torch.float32);  view_1547 = None
	        sub_4521: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_592, convert_element_type_593);  convert_element_type_592 = convert_element_type_593 = None
	        mul_9574: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4521, view_1546);  sub_4521 = view_1546 = None
	        view_1548: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9574, [1280, 1280]);  mul_9574 = None
	        _assert_tensor_metadata_892 = torch.ops.aten._assert_tensor_metadata.default(view_1548, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_892 = None
	        mul_9579: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1549: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1544, [mul_9579, 1280]);  view_1544 = mul_9579 = None
	        permute_165: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1548, [1, 0]);  view_1548 = None
	        addmm_81: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg446_1, view_1549, permute_165);  arg446_1 = view_1549 = permute_165 = None
	        view_1550: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_81, [sym_size_int, 1500, 1280]);  addmm_81 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1551: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1550, [sym_size_int, -1, 20, 64]);  view_1550 = None
	        permute_166: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1551, [0, 2, 1, 3]);  view_1551 = None
	        clone_132: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_166, memory_format = torch.contiguous_format);  permute_166 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_16 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_130, clone_131, clone_132, None, False, scale = 1.0);  clone_130 = clone_131 = clone_132 = None
	        getitem_130: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_16[0];  _scaled_dot_product_efficient_attention_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_167: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_130, [0, 2, 1, 3]);  getitem_130 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_167, [sym_size_int, 1500, -1]);  permute_167 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1553: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1552, [sym_size_int, 1500, 1280])
	        amin_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1553, [2])
	        amax_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1553, [2]);  view_1553 = None
	        full_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_99, full_198);  amin_99 = full_198 = None
	        full_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_99, full_199);  amax_99 = full_199 = None
	        sub_4539: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_99, minimum_99);  maximum_99 = None
	        div_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4539, 255.0);  sub_4539 = None
	        clamp_min_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_198, 1.1920928955078125e-07);  div_198 = None
	        div_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_99, clamp_min_297);  minimum_99 = None
	        round_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_199);  div_199 = None
	        sub_4545: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_199);  round_199 = None
	        clamp_min_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4545, -128);  sub_4545 = None
	        clamp_max_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_298, 127);  clamp_min_298 = None
	        _assert_tensor_metadata_893 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_297, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_893 = None
	        _assert_tensor_metadata_894 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_198, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_894 = None
	        convert_element_type_594: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_198, torch.int8);  clamp_max_198 = None
	        view_1554: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1552, [sym_size_int, 1500, 1280]);  view_1552 = None
	        view_1555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_297, [sym_size_int, 1500, 1])
	        view_1556: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_594, [sym_size_int, 1500, 1])
	        reciprocal_99: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1555);  view_1555 = None
	        mul_9649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_99, 1.0);  reciprocal_99 = None
	        mul_9652: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1554, mul_9649);  view_1554 = mul_9649 = None
	        round_200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9652);  mul_9652 = None
	        add_15271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_200, view_1556);  round_200 = view_1556 = None
	        clamp_min_299: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15271, -128);  add_15271 = None
	        clamp_max_199: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_299, 127);  clamp_min_299 = None
	        view_1557: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_199, [sym_size_int, 1500, 1280]);  clamp_max_199 = None
	        _assert_tensor_metadata_895 = torch.ops.aten._assert_tensor_metadata.default(view_1557, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_895 = None
	        convert_element_type_595: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1557, torch.int8);  view_1557 = None
	        view_1558: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_595, [sym_size_int, 1500, 1280]);  convert_element_type_595 = None
	        view_1559: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_297, [sym_size_int, 1500, 1]);  clamp_min_297 = None
	        view_1560: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_594, [sym_size_int, 1500, 1]);  convert_element_type_594 = None
	        _assert_tensor_metadata_896 = torch.ops.aten._assert_tensor_metadata.default(view_1558, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_896 = None
	        convert_element_type_596: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1558, torch.float32);  view_1558 = None
	        _assert_tensor_metadata_897 = torch.ops.aten._assert_tensor_metadata.default(view_1560, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_897 = None
	        convert_element_type_597: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1560, torch.float32);  view_1560 = None
	        sub_4565: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_596, convert_element_type_597);  convert_element_type_596 = convert_element_type_597 = None
	        mul_9674: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4565, view_1559);  sub_4565 = view_1559 = None
	        view_1561: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9674, [sym_size_int, 1500, 1280]);  mul_9674 = None
	        _assert_tensor_metadata_898 = torch.ops.aten._assert_tensor_metadata.default(view_1561, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_898 = None
	        view_1562: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg451_1, [1280, 40, 32]);  arg451_1 = None
	        view_1563: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg452_1, [1280, 40, 1]);  arg452_1 = None
	        view_1564: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg453_1, [1280, 40, 1]);  arg453_1 = None
	        _assert_tensor_metadata_899 = torch.ops.aten._assert_tensor_metadata.default(view_1562, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_899 = None
	        convert_element_type_598: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1562, torch.float32);  view_1562 = None
	        _assert_tensor_metadata_900 = torch.ops.aten._assert_tensor_metadata.default(view_1564, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_900 = None
	        convert_element_type_599: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1564, torch.float32);  view_1564 = None
	        sub_4569: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_598, convert_element_type_599);  convert_element_type_598 = convert_element_type_599 = None
	        mul_9679: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4569, view_1563);  sub_4569 = view_1563 = None
	        view_1565: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9679, [1280, 1280]);  mul_9679 = None
	        _assert_tensor_metadata_901 = torch.ops.aten._assert_tensor_metadata.default(view_1565, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_901 = None
	        mul_9684: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1566: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1561, [mul_9684, 1280]);  view_1561 = mul_9684 = None
	        permute_168: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1565, [1, 0]);  view_1565 = None
	        addmm_82: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg450_1, view_1566, permute_168);  arg450_1 = view_1566 = permute_168 = None
	        view_1567: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_82, [sym_size_int, 1500, 1280]);  addmm_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_133: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1567);  view_1567 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_15334: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_14714, clone_133);  add_14714 = clone_133 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_15334, memory_format = torch.contiguous_format)
	        var_mean_33 = torch.ops.aten.var_mean.correction(clone_134, [2], correction = 0, keepdim = True)
	        getitem_134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_33[0]
	        getitem_135: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_33[1];  var_mean_33 = None
	        add_15339: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_134, 1e-05);  getitem_134 = None
	        rsqrt_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_15339);  add_15339 = None
	        sub_4575: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_134, getitem_135);  clone_134 = getitem_135 = None
	        mul_9695: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4575, rsqrt_33);  sub_4575 = rsqrt_33 = None
	        mul_9696: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9695, arg454_1);  mul_9695 = arg454_1 = None
	        add_15340: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9696, arg455_1);  mul_9696 = arg455_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1568: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_15340, [sym_size_int, 1500, 1280])
	        amin_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1568, [2])
	        amax_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1568, [2]);  view_1568 = None
	        full_200: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_100, full_200);  amin_100 = full_200 = None
	        full_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_100, full_201);  amax_100 = full_201 = None
	        sub_4586: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_100, minimum_100);  maximum_100 = None
	        div_200: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4586, 255.0);  sub_4586 = None
	        clamp_min_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_200, 1.1920928955078125e-07);  div_200 = None
	        div_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_100, clamp_min_300);  minimum_100 = None
	        round_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_201);  div_201 = None
	        sub_4592: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_201);  round_201 = None
	        clamp_min_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4592, -128);  sub_4592 = None
	        clamp_max_200: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_301, 127);  clamp_min_301 = None
	        _assert_tensor_metadata_902 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_300, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_902 = None
	        _assert_tensor_metadata_903 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_200, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_903 = None
	        convert_element_type_600: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_200, torch.int8);  clamp_max_200 = None
	        view_1569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_15340, [sym_size_int, 1500, 1280]);  add_15340 = None
	        view_1570: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_300, [sym_size_int, 1500, 1])
	        view_1571: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_600, [sym_size_int, 1500, 1])
	        reciprocal_100: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1570);  view_1570 = None
	        mul_9744: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_100, 1.0);  reciprocal_100 = None
	        mul_9747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1569, mul_9744);  view_1569 = mul_9744 = None
	        round_202: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9747);  mul_9747 = None
	        add_15427: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_202, view_1571);  round_202 = view_1571 = None
	        clamp_min_302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15427, -128);  add_15427 = None
	        clamp_max_201: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_302, 127);  clamp_min_302 = None
	        view_1572: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_201, [sym_size_int, 1500, 1280]);  clamp_max_201 = None
	        _assert_tensor_metadata_904 = torch.ops.aten._assert_tensor_metadata.default(view_1572, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_904 = None
	        convert_element_type_601: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1572, torch.int8);  view_1572 = None
	        view_1573: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_601, [sym_size_int, 1500, 1280]);  convert_element_type_601 = None
	        view_1574: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_300, [sym_size_int, 1500, 1]);  clamp_min_300 = None
	        view_1575: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_600, [sym_size_int, 1500, 1]);  convert_element_type_600 = None
	        _assert_tensor_metadata_905 = torch.ops.aten._assert_tensor_metadata.default(view_1573, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_905 = None
	        convert_element_type_602: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1573, torch.float32);  view_1573 = None
	        _assert_tensor_metadata_906 = torch.ops.aten._assert_tensor_metadata.default(view_1575, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_906 = None
	        convert_element_type_603: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1575, torch.float32);  view_1575 = None
	        sub_4612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_602, convert_element_type_603);  convert_element_type_602 = convert_element_type_603 = None
	        mul_9769: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4612, view_1574);  sub_4612 = view_1574 = None
	        view_1576: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9769, [sym_size_int, 1500, 1280]);  mul_9769 = None
	        _assert_tensor_metadata_907 = torch.ops.aten._assert_tensor_metadata.default(view_1576, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_907 = None
	        view_1577: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg457_1, [5120, 40, 32]);  arg457_1 = None
	        view_1578: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg458_1, [5120, 40, 1]);  arg458_1 = None
	        view_1579: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg459_1, [5120, 40, 1]);  arg459_1 = None
	        _assert_tensor_metadata_908 = torch.ops.aten._assert_tensor_metadata.default(view_1577, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_908 = None
	        convert_element_type_604: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1577, torch.float32);  view_1577 = None
	        _assert_tensor_metadata_909 = torch.ops.aten._assert_tensor_metadata.default(view_1579, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_909 = None
	        convert_element_type_605: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1579, torch.float32);  view_1579 = None
	        sub_4616: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_604, convert_element_type_605);  convert_element_type_604 = convert_element_type_605 = None
	        mul_9774: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4616, view_1578);  sub_4616 = view_1578 = None
	        view_1580: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9774, [5120, 1280]);  mul_9774 = None
	        _assert_tensor_metadata_910 = torch.ops.aten._assert_tensor_metadata.default(view_1580, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_910 = None
	        mul_9779: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1581: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1576, [mul_9779, 1280]);  view_1576 = mul_9779 = None
	        permute_169: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1580, [1, 0]);  view_1580 = None
	        addmm_83: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg456_1, view_1581, permute_169);  arg456_1 = view_1581 = permute_169 = None
	        view_1582: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_83, [sym_size_int, 1500, 5120]);  addmm_83 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_9786: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1582, 0.5)
	        mul_9787: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1582, 0.7071067811865476);  view_1582 = None
	        erf_18: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_9787);  mul_9787 = None
	        add_15486: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_18, 1);  erf_18 = None
	        mul_9788: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9786, add_15486);  mul_9786 = add_15486 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_135: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_9788);  mul_9788 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1583: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_135, [sym_size_int, 1500, 5120])
	        amin_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1583, [2])
	        amax_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1583, [2]);  view_1583 = None
	        full_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_101, full_202);  amin_101 = full_202 = None
	        full_203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_101, full_203);  amax_101 = full_203 = None
	        sub_4629: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_101, minimum_101);  maximum_101 = None
	        div_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4629, 255.0);  sub_4629 = None
	        clamp_min_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_202, 1.1920928955078125e-07);  div_202 = None
	        div_203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_101, clamp_min_303);  minimum_101 = None
	        round_203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_203);  div_203 = None
	        sub_4635: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_203);  round_203 = None
	        clamp_min_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4635, -128);  sub_4635 = None
	        clamp_max_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_304, 127);  clamp_min_304 = None
	        _assert_tensor_metadata_911 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_303, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_911 = None
	        _assert_tensor_metadata_912 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_202, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_912 = None
	        convert_element_type_606: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_202, torch.int8);  clamp_max_202 = None
	        view_1584: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_135, [sym_size_int, 1500, 5120]);  clone_135 = None
	        view_1585: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_303, [sym_size_int, 1500, 1])
	        view_1586: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_606, [sym_size_int, 1500, 1])
	        reciprocal_101: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1585);  view_1585 = None
	        mul_9834: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_101, 1.0);  reciprocal_101 = None
	        mul_9837: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1584, mul_9834);  view_1584 = mul_9834 = None
	        round_204: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_9837);  mul_9837 = None
	        add_15569: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_204, view_1586);  round_204 = view_1586 = None
	        clamp_min_305: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15569, -128);  add_15569 = None
	        clamp_max_203: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_305, 127);  clamp_min_305 = None
	        view_1587: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_203, [sym_size_int, 1500, 5120]);  clamp_max_203 = None
	        _assert_tensor_metadata_913 = torch.ops.aten._assert_tensor_metadata.default(view_1587, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_913 = None
	        convert_element_type_607: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1587, torch.int8);  view_1587 = None
	        view_1588: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_607, [sym_size_int, 1500, 5120]);  convert_element_type_607 = None
	        view_1589: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_303, [sym_size_int, 1500, 1]);  clamp_min_303 = None
	        view_1590: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_606, [sym_size_int, 1500, 1]);  convert_element_type_606 = None
	        _assert_tensor_metadata_914 = torch.ops.aten._assert_tensor_metadata.default(view_1588, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_914 = None
	        convert_element_type_608: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1588, torch.float32);  view_1588 = None
	        _assert_tensor_metadata_915 = torch.ops.aten._assert_tensor_metadata.default(view_1590, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_915 = None
	        convert_element_type_609: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1590, torch.float32);  view_1590 = None
	        sub_4655: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_608, convert_element_type_609);  convert_element_type_608 = convert_element_type_609 = None
	        mul_9859: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4655, view_1589);  sub_4655 = view_1589 = None
	        view_1591: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_9859, [sym_size_int, 1500, 5120]);  mul_9859 = None
	        _assert_tensor_metadata_916 = torch.ops.aten._assert_tensor_metadata.default(view_1591, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_916 = None
	        view_1592: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg461_1, [1280, 160, 32]);  arg461_1 = None
	        view_1593: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg462_1, [1280, 160, 1]);  arg462_1 = None
	        view_1594: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg463_1, [1280, 160, 1]);  arg463_1 = None
	        _assert_tensor_metadata_917 = torch.ops.aten._assert_tensor_metadata.default(view_1592, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_917 = None
	        convert_element_type_610: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1592, torch.float32);  view_1592 = None
	        _assert_tensor_metadata_918 = torch.ops.aten._assert_tensor_metadata.default(view_1594, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_918 = None
	        convert_element_type_611: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1594, torch.float32);  view_1594 = None
	        sub_4659: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_610, convert_element_type_611);  convert_element_type_610 = convert_element_type_611 = None
	        mul_9864: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4659, view_1593);  sub_4659 = view_1593 = None
	        view_1595: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_9864, [1280, 5120]);  mul_9864 = None
	        _assert_tensor_metadata_919 = torch.ops.aten._assert_tensor_metadata.default(view_1595, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_919 = None
	        mul_9869: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1596: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1591, [mul_9869, 5120]);  view_1591 = mul_9869 = None
	        permute_170: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1595, [1, 0]);  view_1595 = None
	        addmm_84: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg460_1, view_1596, permute_170);  arg460_1 = view_1596 = permute_170 = None
	        view_1597: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_84, [sym_size_int, 1500, 1280]);  addmm_84 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_136: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1597);  view_1597 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_15632: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_15334, clone_136);  add_15334 = clone_136 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_15632, memory_format = torch.contiguous_format)
	        var_mean_34 = torch.ops.aten.var_mean.correction(clone_137, [2], correction = 0, keepdim = True)
	        getitem_136: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_34[0]
	        getitem_137: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_34[1];  var_mean_34 = None
	        add_15637: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_136, 1e-05);  getitem_136 = None
	        rsqrt_34: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_15637);  add_15637 = None
	        sub_4665: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_137, getitem_137);  clone_137 = getitem_137 = None
	        mul_9880: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4665, rsqrt_34);  sub_4665 = rsqrt_34 = None
	        mul_9881: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9880, arg464_1);  mul_9880 = arg464_1 = None
	        add_15638: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9881, arg465_1);  mul_9881 = arg465_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1598: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_15638, [sym_size_int, 1500, 1280])
	        amin_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1598, [2])
	        amax_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1598, [2]);  view_1598 = None
	        full_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_102, full_204);  amin_102 = full_204 = None
	        full_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_102, full_205);  amax_102 = full_205 = None
	        sub_4676: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_102, minimum_102);  maximum_102 = None
	        div_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4676, 255.0);  sub_4676 = None
	        clamp_min_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_204, 1.1920928955078125e-07);  div_204 = None
	        div_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_102, clamp_min_306);  minimum_102 = None
	        round_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_205);  div_205 = None
	        sub_4682: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_205);  round_205 = None
	        clamp_min_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4682, -128);  sub_4682 = None
	        clamp_max_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_307, 127);  clamp_min_307 = None
	        _assert_tensor_metadata_920 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_306, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_920 = None
	        _assert_tensor_metadata_921 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_204, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_921 = None
	        convert_element_type_612: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_204, torch.int8);  clamp_max_204 = None
	        view_1599: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_15638, [sym_size_int, 1500, 1280])
	        view_1600: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_306, [sym_size_int, 1500, 1])
	        view_1601: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_612, [sym_size_int, 1500, 1])
	        reciprocal_102: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1600);  view_1600 = None
	        mul_9929: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_102, 1.0);  reciprocal_102 = None
	        mul_9932: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1599, mul_9929);  view_1599 = mul_9929 = None
	        round_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9932);  mul_9932 = None
	        add_15725: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_206, view_1601);  round_206 = view_1601 = None
	        clamp_min_308: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15725, -128);  add_15725 = None
	        clamp_max_205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_308, 127);  clamp_min_308 = None
	        view_1602: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_205, [sym_size_int, 1500, 1280]);  clamp_max_205 = None
	        _assert_tensor_metadata_922 = torch.ops.aten._assert_tensor_metadata.default(view_1602, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_922 = None
	        convert_element_type_613: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1602, torch.int8);  view_1602 = None
	        view_1603: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_613, [sym_size_int, 1500, 1280]);  convert_element_type_613 = None
	        view_1604: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_306, [sym_size_int, 1500, 1]);  clamp_min_306 = None
	        view_1605: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_612, [sym_size_int, 1500, 1]);  convert_element_type_612 = None
	        _assert_tensor_metadata_923 = torch.ops.aten._assert_tensor_metadata.default(view_1603, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_923 = None
	        convert_element_type_614: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1603, torch.float32);  view_1603 = None
	        _assert_tensor_metadata_924 = torch.ops.aten._assert_tensor_metadata.default(view_1605, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_924 = None
	        convert_element_type_615: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1605, torch.float32);  view_1605 = None
	        sub_4702: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_614, convert_element_type_615);  convert_element_type_614 = convert_element_type_615 = None
	        mul_9954: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4702, view_1604);  sub_4702 = view_1604 = None
	        view_1606: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9954, [sym_size_int, 1500, 1280]);  mul_9954 = None
	        _assert_tensor_metadata_925 = torch.ops.aten._assert_tensor_metadata.default(view_1606, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_925 = None
	        view_1607: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg467_1, [1280, 40, 32]);  arg467_1 = None
	        view_1608: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg468_1, [1280, 40, 1]);  arg468_1 = None
	        view_1609: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg469_1, [1280, 40, 1]);  arg469_1 = None
	        _assert_tensor_metadata_926 = torch.ops.aten._assert_tensor_metadata.default(view_1607, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_926 = None
	        convert_element_type_616: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1607, torch.float32);  view_1607 = None
	        _assert_tensor_metadata_927 = torch.ops.aten._assert_tensor_metadata.default(view_1609, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_927 = None
	        convert_element_type_617: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1609, torch.float32);  view_1609 = None
	        sub_4706: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_616, convert_element_type_617);  convert_element_type_616 = convert_element_type_617 = None
	        mul_9959: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4706, view_1608);  sub_4706 = view_1608 = None
	        view_1610: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9959, [1280, 1280]);  mul_9959 = None
	        _assert_tensor_metadata_928 = torch.ops.aten._assert_tensor_metadata.default(view_1610, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_928 = None
	        mul_9964: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1611: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1606, [mul_9964, 1280]);  view_1606 = mul_9964 = None
	        permute_171: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1610, [1, 0]);  view_1610 = None
	        addmm_85: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg466_1, view_1611, permute_171);  arg466_1 = view_1611 = permute_171 = None
	        view_1612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_85, [sym_size_int, 1500, 1280]);  addmm_85 = None
	        mul_9971: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1612, 0.125);  view_1612 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1613: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_9971, [sym_size_int, 1500, 20, 64]);  mul_9971 = None
	        permute_172: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1613, [0, 2, 1, 3]);  view_1613 = None
	        clone_138: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_172, memory_format = torch.contiguous_format);  permute_172 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1614: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_15638, [sym_size_int, 1500, 1280])
	        amin_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1614, [2])
	        amax_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1614, [2]);  view_1614 = None
	        full_206: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_103, full_206);  amin_103 = full_206 = None
	        full_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_103, full_207);  amax_103 = full_207 = None
	        sub_4721: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_103, minimum_103);  maximum_103 = None
	        div_206: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4721, 255.0);  sub_4721 = None
	        clamp_min_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_206, 1.1920928955078125e-07);  div_206 = None
	        div_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_103, clamp_min_309);  minimum_103 = None
	        round_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_207);  div_207 = None
	        sub_4727: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_207);  round_207 = None
	        clamp_min_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4727, -128);  sub_4727 = None
	        clamp_max_206: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_310, 127);  clamp_min_310 = None
	        _assert_tensor_metadata_929 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_309, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_929 = None
	        _assert_tensor_metadata_930 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_206, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_930 = None
	        convert_element_type_618: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_206, torch.int8);  clamp_max_206 = None
	        view_1615: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_15638, [sym_size_int, 1500, 1280])
	        view_1616: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_309, [sym_size_int, 1500, 1])
	        view_1617: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_618, [sym_size_int, 1500, 1])
	        reciprocal_103: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1616);  view_1616 = None
	        mul_10025: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_103, 1.0);  reciprocal_103 = None
	        mul_10028: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1615, mul_10025);  view_1615 = mul_10025 = None
	        round_208: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10028);  mul_10028 = None
	        add_15877: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_208, view_1617);  round_208 = view_1617 = None
	        clamp_min_311: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15877, -128);  add_15877 = None
	        clamp_max_207: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_311, 127);  clamp_min_311 = None
	        view_1618: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_207, [sym_size_int, 1500, 1280]);  clamp_max_207 = None
	        _assert_tensor_metadata_931 = torch.ops.aten._assert_tensor_metadata.default(view_1618, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_931 = None
	        convert_element_type_619: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1618, torch.int8);  view_1618 = None
	        view_1619: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_619, [sym_size_int, 1500, 1280]);  convert_element_type_619 = None
	        view_1620: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_309, [sym_size_int, 1500, 1]);  clamp_min_309 = None
	        view_1621: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_618, [sym_size_int, 1500, 1]);  convert_element_type_618 = None
	        _assert_tensor_metadata_932 = torch.ops.aten._assert_tensor_metadata.default(view_1619, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_932 = None
	        convert_element_type_620: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1619, torch.float32);  view_1619 = None
	        _assert_tensor_metadata_933 = torch.ops.aten._assert_tensor_metadata.default(view_1621, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_933 = None
	        convert_element_type_621: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1621, torch.float32);  view_1621 = None
	        sub_4747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_620, convert_element_type_621);  convert_element_type_620 = convert_element_type_621 = None
	        mul_10050: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4747, view_1620);  sub_4747 = view_1620 = None
	        view_1622: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10050, [sym_size_int, 1500, 1280]);  mul_10050 = None
	        _assert_tensor_metadata_934 = torch.ops.aten._assert_tensor_metadata.default(view_1622, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_934 = None
	        view_1623: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg470_1, [1280, 40, 32]);  arg470_1 = None
	        view_1624: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg471_1, [1280, 40, 1]);  arg471_1 = None
	        view_1625: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg472_1, [1280, 40, 1]);  arg472_1 = None
	        _assert_tensor_metadata_935 = torch.ops.aten._assert_tensor_metadata.default(view_1623, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_935 = None
	        convert_element_type_622: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1623, torch.float32);  view_1623 = None
	        _assert_tensor_metadata_936 = torch.ops.aten._assert_tensor_metadata.default(view_1625, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_936 = None
	        convert_element_type_623: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1625, torch.float32);  view_1625 = None
	        sub_4751: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_622, convert_element_type_623);  convert_element_type_622 = convert_element_type_623 = None
	        mul_10055: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4751, view_1624);  sub_4751 = view_1624 = None
	        view_1626: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10055, [1280, 1280]);  mul_10055 = None
	        _assert_tensor_metadata_937 = torch.ops.aten._assert_tensor_metadata.default(view_1626, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_937 = None
	        permute_173: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1626, [1, 0]);  view_1626 = None
	        mul_10058: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1627: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1622, [mul_10058, 1280]);  view_1622 = mul_10058 = None
	        mm_17: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1627, permute_173);  view_1627 = permute_173 = None
	        view_1628: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_17, [sym_size_int, 1500, 1280]);  mm_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1629: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1628, [sym_size_int, -1, 20, 64]);  view_1628 = None
	        permute_174: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1629, [0, 2, 1, 3]);  view_1629 = None
	        clone_139: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_174, memory_format = torch.contiguous_format);  permute_174 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1630: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_15638, [sym_size_int, 1500, 1280])
	        amin_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1630, [2])
	        amax_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1630, [2]);  view_1630 = None
	        full_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_104, full_208);  amin_104 = full_208 = None
	        full_209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_104, full_209);  amax_104 = full_209 = None
	        sub_4765: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_104, minimum_104);  maximum_104 = None
	        div_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4765, 255.0);  sub_4765 = None
	        clamp_min_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_208, 1.1920928955078125e-07);  div_208 = None
	        div_209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_104, clamp_min_312);  minimum_104 = None
	        round_209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_209);  div_209 = None
	        sub_4771: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_209);  round_209 = None
	        clamp_min_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4771, -128);  sub_4771 = None
	        clamp_max_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_313, 127);  clamp_min_313 = None
	        _assert_tensor_metadata_938 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_312, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_938 = None
	        _assert_tensor_metadata_939 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_208, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_939 = None
	        convert_element_type_624: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_208, torch.int8);  clamp_max_208 = None
	        view_1631: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_15638, [sym_size_int, 1500, 1280]);  add_15638 = None
	        view_1632: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_312, [sym_size_int, 1500, 1])
	        view_1633: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_624, [sym_size_int, 1500, 1])
	        reciprocal_104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1632);  view_1632 = None
	        mul_10124: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_104, 1.0);  reciprocal_104 = None
	        mul_10127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1631, mul_10124);  view_1631 = mul_10124 = None
	        round_210: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10127);  mul_10127 = None
	        add_16025: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_210, view_1633);  round_210 = view_1633 = None
	        clamp_min_314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16025, -128);  add_16025 = None
	        clamp_max_209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_314, 127);  clamp_min_314 = None
	        view_1634: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_209, [sym_size_int, 1500, 1280]);  clamp_max_209 = None
	        _assert_tensor_metadata_940 = torch.ops.aten._assert_tensor_metadata.default(view_1634, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_940 = None
	        convert_element_type_625: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1634, torch.int8);  view_1634 = None
	        view_1635: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_625, [sym_size_int, 1500, 1280]);  convert_element_type_625 = None
	        view_1636: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_312, [sym_size_int, 1500, 1]);  clamp_min_312 = None
	        view_1637: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_624, [sym_size_int, 1500, 1]);  convert_element_type_624 = None
	        _assert_tensor_metadata_941 = torch.ops.aten._assert_tensor_metadata.default(view_1635, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_941 = None
	        convert_element_type_626: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1635, torch.float32);  view_1635 = None
	        _assert_tensor_metadata_942 = torch.ops.aten._assert_tensor_metadata.default(view_1637, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_942 = None
	        convert_element_type_627: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1637, torch.float32);  view_1637 = None
	        sub_4791: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_626, convert_element_type_627);  convert_element_type_626 = convert_element_type_627 = None
	        mul_10149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4791, view_1636);  sub_4791 = view_1636 = None
	        view_1638: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10149, [sym_size_int, 1500, 1280]);  mul_10149 = None
	        _assert_tensor_metadata_943 = torch.ops.aten._assert_tensor_metadata.default(view_1638, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_943 = None
	        view_1639: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg474_1, [1280, 40, 32]);  arg474_1 = None
	        view_1640: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg475_1, [1280, 40, 1]);  arg475_1 = None
	        view_1641: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg476_1, [1280, 40, 1]);  arg476_1 = None
	        _assert_tensor_metadata_944 = torch.ops.aten._assert_tensor_metadata.default(view_1639, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_944 = None
	        convert_element_type_628: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1639, torch.float32);  view_1639 = None
	        _assert_tensor_metadata_945 = torch.ops.aten._assert_tensor_metadata.default(view_1641, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_945 = None
	        convert_element_type_629: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1641, torch.float32);  view_1641 = None
	        sub_4795: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_628, convert_element_type_629);  convert_element_type_628 = convert_element_type_629 = None
	        mul_10154: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4795, view_1640);  sub_4795 = view_1640 = None
	        view_1642: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10154, [1280, 1280]);  mul_10154 = None
	        _assert_tensor_metadata_946 = torch.ops.aten._assert_tensor_metadata.default(view_1642, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_946 = None
	        mul_10159: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1643: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1638, [mul_10159, 1280]);  view_1638 = mul_10159 = None
	        permute_175: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1642, [1, 0]);  view_1642 = None
	        addmm_86: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg473_1, view_1643, permute_175);  arg473_1 = view_1643 = permute_175 = None
	        view_1644: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_86, [sym_size_int, 1500, 1280]);  addmm_86 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1645: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1644, [sym_size_int, -1, 20, 64]);  view_1644 = None
	        permute_176: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1645, [0, 2, 1, 3]);  view_1645 = None
	        clone_140: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_176, memory_format = torch.contiguous_format);  permute_176 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_17 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_138, clone_139, clone_140, None, False, scale = 1.0);  clone_138 = clone_139 = clone_140 = None
	        getitem_138: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_17[0];  _scaled_dot_product_efficient_attention_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_177: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_138, [0, 2, 1, 3]);  getitem_138 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1646: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_177, [sym_size_int, 1500, -1]);  permute_177 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1647: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1646, [sym_size_int, 1500, 1280])
	        amin_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1647, [2])
	        amax_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1647, [2]);  view_1647 = None
	        full_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_105, full_210);  amin_105 = full_210 = None
	        full_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_105, full_211);  amax_105 = full_211 = None
	        sub_4813: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_105, minimum_105);  maximum_105 = None
	        div_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4813, 255.0);  sub_4813 = None
	        clamp_min_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_210, 1.1920928955078125e-07);  div_210 = None
	        div_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_105, clamp_min_315);  minimum_105 = None
	        round_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_211);  div_211 = None
	        sub_4819: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_211);  round_211 = None
	        clamp_min_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4819, -128);  sub_4819 = None
	        clamp_max_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_316, 127);  clamp_min_316 = None
	        _assert_tensor_metadata_947 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_315, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_947 = None
	        _assert_tensor_metadata_948 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_210, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_948 = None
	        convert_element_type_630: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_210, torch.int8);  clamp_max_210 = None
	        view_1648: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1646, [sym_size_int, 1500, 1280]);  view_1646 = None
	        view_1649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_315, [sym_size_int, 1500, 1])
	        view_1650: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_630, [sym_size_int, 1500, 1])
	        reciprocal_105: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1649);  view_1649 = None
	        mul_10229: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_105, 1.0);  reciprocal_105 = None
	        mul_10232: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1648, mul_10229);  view_1648 = mul_10229 = None
	        round_212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10232);  mul_10232 = None
	        add_16189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_212, view_1650);  round_212 = view_1650 = None
	        clamp_min_317: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16189, -128);  add_16189 = None
	        clamp_max_211: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_317, 127);  clamp_min_317 = None
	        view_1651: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_211, [sym_size_int, 1500, 1280]);  clamp_max_211 = None
	        _assert_tensor_metadata_949 = torch.ops.aten._assert_tensor_metadata.default(view_1651, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_949 = None
	        convert_element_type_631: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1651, torch.int8);  view_1651 = None
	        view_1652: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_631, [sym_size_int, 1500, 1280]);  convert_element_type_631 = None
	        view_1653: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_315, [sym_size_int, 1500, 1]);  clamp_min_315 = None
	        view_1654: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_630, [sym_size_int, 1500, 1]);  convert_element_type_630 = None
	        _assert_tensor_metadata_950 = torch.ops.aten._assert_tensor_metadata.default(view_1652, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_950 = None
	        convert_element_type_632: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1652, torch.float32);  view_1652 = None
	        _assert_tensor_metadata_951 = torch.ops.aten._assert_tensor_metadata.default(view_1654, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_951 = None
	        convert_element_type_633: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1654, torch.float32);  view_1654 = None
	        sub_4839: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_632, convert_element_type_633);  convert_element_type_632 = convert_element_type_633 = None
	        mul_10254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4839, view_1653);  sub_4839 = view_1653 = None
	        view_1655: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10254, [sym_size_int, 1500, 1280]);  mul_10254 = None
	        _assert_tensor_metadata_952 = torch.ops.aten._assert_tensor_metadata.default(view_1655, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_952 = None
	        view_1656: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg478_1, [1280, 40, 32]);  arg478_1 = None
	        view_1657: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg479_1, [1280, 40, 1]);  arg479_1 = None
	        view_1658: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg480_1, [1280, 40, 1]);  arg480_1 = None
	        _assert_tensor_metadata_953 = torch.ops.aten._assert_tensor_metadata.default(view_1656, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_953 = None
	        convert_element_type_634: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1656, torch.float32);  view_1656 = None
	        _assert_tensor_metadata_954 = torch.ops.aten._assert_tensor_metadata.default(view_1658, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_954 = None
	        convert_element_type_635: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1658, torch.float32);  view_1658 = None
	        sub_4843: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_634, convert_element_type_635);  convert_element_type_634 = convert_element_type_635 = None
	        mul_10259: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4843, view_1657);  sub_4843 = view_1657 = None
	        view_1659: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10259, [1280, 1280]);  mul_10259 = None
	        _assert_tensor_metadata_955 = torch.ops.aten._assert_tensor_metadata.default(view_1659, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_955 = None
	        mul_10264: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1660: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1655, [mul_10264, 1280]);  view_1655 = mul_10264 = None
	        permute_178: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1659, [1, 0]);  view_1659 = None
	        addmm_87: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg477_1, view_1660, permute_178);  arg477_1 = view_1660 = permute_178 = None
	        view_1661: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_87, [sym_size_int, 1500, 1280]);  addmm_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_141: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1661);  view_1661 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_16252: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_15632, clone_141);  add_15632 = clone_141 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_16252, memory_format = torch.contiguous_format)
	        var_mean_35 = torch.ops.aten.var_mean.correction(clone_142, [2], correction = 0, keepdim = True)
	        getitem_142: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_35[0]
	        getitem_143: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_35[1];  var_mean_35 = None
	        add_16257: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_142, 1e-05);  getitem_142 = None
	        rsqrt_35: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_16257);  add_16257 = None
	        sub_4849: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_142, getitem_143);  clone_142 = getitem_143 = None
	        mul_10275: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4849, rsqrt_35);  sub_4849 = rsqrt_35 = None
	        mul_10276: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10275, arg481_1);  mul_10275 = arg481_1 = None
	        add_16258: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_10276, arg482_1);  mul_10276 = arg482_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1662: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_16258, [sym_size_int, 1500, 1280])
	        amin_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1662, [2])
	        amax_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1662, [2]);  view_1662 = None
	        full_212: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_106, full_212);  amin_106 = full_212 = None
	        full_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_106, full_213);  amax_106 = full_213 = None
	        sub_4860: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_106, minimum_106);  maximum_106 = None
	        div_212: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4860, 255.0);  sub_4860 = None
	        clamp_min_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_212, 1.1920928955078125e-07);  div_212 = None
	        div_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_106, clamp_min_318);  minimum_106 = None
	        round_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_213);  div_213 = None
	        sub_4866: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_213);  round_213 = None
	        clamp_min_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4866, -128);  sub_4866 = None
	        clamp_max_212: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_319, 127);  clamp_min_319 = None
	        _assert_tensor_metadata_956 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_318, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_956 = None
	        _assert_tensor_metadata_957 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_212, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_957 = None
	        convert_element_type_636: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_212, torch.int8);  clamp_max_212 = None
	        view_1663: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_16258, [sym_size_int, 1500, 1280]);  add_16258 = None
	        view_1664: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_318, [sym_size_int, 1500, 1])
	        view_1665: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_636, [sym_size_int, 1500, 1])
	        reciprocal_106: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1664);  view_1664 = None
	        mul_10324: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_106, 1.0);  reciprocal_106 = None
	        mul_10327: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1663, mul_10324);  view_1663 = mul_10324 = None
	        round_214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10327);  mul_10327 = None
	        add_16345: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_214, view_1665);  round_214 = view_1665 = None
	        clamp_min_320: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16345, -128);  add_16345 = None
	        clamp_max_213: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_320, 127);  clamp_min_320 = None
	        view_1666: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_213, [sym_size_int, 1500, 1280]);  clamp_max_213 = None
	        _assert_tensor_metadata_958 = torch.ops.aten._assert_tensor_metadata.default(view_1666, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_958 = None
	        convert_element_type_637: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1666, torch.int8);  view_1666 = None
	        view_1667: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_637, [sym_size_int, 1500, 1280]);  convert_element_type_637 = None
	        view_1668: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_318, [sym_size_int, 1500, 1]);  clamp_min_318 = None
	        view_1669: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_636, [sym_size_int, 1500, 1]);  convert_element_type_636 = None
	        _assert_tensor_metadata_959 = torch.ops.aten._assert_tensor_metadata.default(view_1667, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_959 = None
	        convert_element_type_638: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1667, torch.float32);  view_1667 = None
	        _assert_tensor_metadata_960 = torch.ops.aten._assert_tensor_metadata.default(view_1669, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_960 = None
	        convert_element_type_639: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1669, torch.float32);  view_1669 = None
	        sub_4886: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_638, convert_element_type_639);  convert_element_type_638 = convert_element_type_639 = None
	        mul_10349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4886, view_1668);  sub_4886 = view_1668 = None
	        view_1670: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10349, [sym_size_int, 1500, 1280]);  mul_10349 = None
	        _assert_tensor_metadata_961 = torch.ops.aten._assert_tensor_metadata.default(view_1670, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_961 = None
	        view_1671: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg484_1, [5120, 40, 32]);  arg484_1 = None
	        view_1672: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg485_1, [5120, 40, 1]);  arg485_1 = None
	        view_1673: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg486_1, [5120, 40, 1]);  arg486_1 = None
	        _assert_tensor_metadata_962 = torch.ops.aten._assert_tensor_metadata.default(view_1671, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_962 = None
	        convert_element_type_640: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1671, torch.float32);  view_1671 = None
	        _assert_tensor_metadata_963 = torch.ops.aten._assert_tensor_metadata.default(view_1673, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_963 = None
	        convert_element_type_641: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1673, torch.float32);  view_1673 = None
	        sub_4890: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_640, convert_element_type_641);  convert_element_type_640 = convert_element_type_641 = None
	        mul_10354: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4890, view_1672);  sub_4890 = view_1672 = None
	        view_1674: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10354, [5120, 1280]);  mul_10354 = None
	        _assert_tensor_metadata_964 = torch.ops.aten._assert_tensor_metadata.default(view_1674, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_964 = None
	        mul_10359: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1675: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1670, [mul_10359, 1280]);  view_1670 = mul_10359 = None
	        permute_179: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1674, [1, 0]);  view_1674 = None
	        addmm_88: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg483_1, view_1675, permute_179);  arg483_1 = view_1675 = permute_179 = None
	        view_1676: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_88, [sym_size_int, 1500, 5120]);  addmm_88 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_10366: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1676, 0.5)
	        mul_10367: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1676, 0.7071067811865476);  view_1676 = None
	        erf_19: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_10367);  mul_10367 = None
	        add_16404: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_19, 1);  erf_19 = None
	        mul_10368: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10366, add_16404);  mul_10366 = add_16404 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_143: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_10368);  mul_10368 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1677: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_143, [sym_size_int, 1500, 5120])
	        amin_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1677, [2])
	        amax_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1677, [2]);  view_1677 = None
	        full_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_107, full_214);  amin_107 = full_214 = None
	        full_215: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_107, full_215);  amax_107 = full_215 = None
	        sub_4903: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_107, minimum_107);  maximum_107 = None
	        div_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4903, 255.0);  sub_4903 = None
	        clamp_min_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_214, 1.1920928955078125e-07);  div_214 = None
	        div_215: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_107, clamp_min_321);  minimum_107 = None
	        round_215: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_215);  div_215 = None
	        sub_4909: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_215);  round_215 = None
	        clamp_min_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4909, -128);  sub_4909 = None
	        clamp_max_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_322, 127);  clamp_min_322 = None
	        _assert_tensor_metadata_965 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_321, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_965 = None
	        _assert_tensor_metadata_966 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_214, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_966 = None
	        convert_element_type_642: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_214, torch.int8);  clamp_max_214 = None
	        view_1678: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_143, [sym_size_int, 1500, 5120]);  clone_143 = None
	        view_1679: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_321, [sym_size_int, 1500, 1])
	        view_1680: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_642, [sym_size_int, 1500, 1])
	        reciprocal_107: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1679);  view_1679 = None
	        mul_10414: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_107, 1.0);  reciprocal_107 = None
	        mul_10417: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1678, mul_10414);  view_1678 = mul_10414 = None
	        round_216: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_10417);  mul_10417 = None
	        add_16487: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_216, view_1680);  round_216 = view_1680 = None
	        clamp_min_323: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16487, -128);  add_16487 = None
	        clamp_max_215: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_323, 127);  clamp_min_323 = None
	        view_1681: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_215, [sym_size_int, 1500, 5120]);  clamp_max_215 = None
	        _assert_tensor_metadata_967 = torch.ops.aten._assert_tensor_metadata.default(view_1681, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_967 = None
	        convert_element_type_643: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1681, torch.int8);  view_1681 = None
	        view_1682: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_643, [sym_size_int, 1500, 5120]);  convert_element_type_643 = None
	        view_1683: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_321, [sym_size_int, 1500, 1]);  clamp_min_321 = None
	        view_1684: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_642, [sym_size_int, 1500, 1]);  convert_element_type_642 = None
	        _assert_tensor_metadata_968 = torch.ops.aten._assert_tensor_metadata.default(view_1682, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_968 = None
	        convert_element_type_644: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1682, torch.float32);  view_1682 = None
	        _assert_tensor_metadata_969 = torch.ops.aten._assert_tensor_metadata.default(view_1684, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_969 = None
	        convert_element_type_645: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1684, torch.float32);  view_1684 = None
	        sub_4929: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_644, convert_element_type_645);  convert_element_type_644 = convert_element_type_645 = None
	        mul_10439: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4929, view_1683);  sub_4929 = view_1683 = None
	        view_1685: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_10439, [sym_size_int, 1500, 5120]);  mul_10439 = None
	        _assert_tensor_metadata_970 = torch.ops.aten._assert_tensor_metadata.default(view_1685, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_970 = None
	        view_1686: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg488_1, [1280, 160, 32]);  arg488_1 = None
	        view_1687: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg489_1, [1280, 160, 1]);  arg489_1 = None
	        view_1688: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg490_1, [1280, 160, 1]);  arg490_1 = None
	        _assert_tensor_metadata_971 = torch.ops.aten._assert_tensor_metadata.default(view_1686, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_971 = None
	        convert_element_type_646: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1686, torch.float32);  view_1686 = None
	        _assert_tensor_metadata_972 = torch.ops.aten._assert_tensor_metadata.default(view_1688, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_972 = None
	        convert_element_type_647: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1688, torch.float32);  view_1688 = None
	        sub_4933: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_646, convert_element_type_647);  convert_element_type_646 = convert_element_type_647 = None
	        mul_10444: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4933, view_1687);  sub_4933 = view_1687 = None
	        view_1689: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_10444, [1280, 5120]);  mul_10444 = None
	        _assert_tensor_metadata_973 = torch.ops.aten._assert_tensor_metadata.default(view_1689, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_973 = None
	        mul_10449: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1690: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1685, [mul_10449, 5120]);  view_1685 = mul_10449 = None
	        permute_180: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1689, [1, 0]);  view_1689 = None
	        addmm_89: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg487_1, view_1690, permute_180);  arg487_1 = view_1690 = permute_180 = None
	        view_1691: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_89, [sym_size_int, 1500, 1280]);  addmm_89 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_144: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1691);  view_1691 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_16550: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_16252, clone_144);  add_16252 = clone_144 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_145: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_16550, memory_format = torch.contiguous_format)
	        var_mean_36 = torch.ops.aten.var_mean.correction(clone_145, [2], correction = 0, keepdim = True)
	        getitem_144: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_36[0]
	        getitem_145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_36[1];  var_mean_36 = None
	        add_16555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_144, 1e-05);  getitem_144 = None
	        rsqrt_36: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_16555);  add_16555 = None
	        sub_4939: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_145, getitem_145);  clone_145 = getitem_145 = None
	        mul_10460: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4939, rsqrt_36);  sub_4939 = rsqrt_36 = None
	        mul_10461: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10460, arg491_1);  mul_10460 = arg491_1 = None
	        add_16556: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_10461, arg492_1);  mul_10461 = arg492_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1692: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_16556, [sym_size_int, 1500, 1280])
	        amin_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1692, [2])
	        amax_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1692, [2]);  view_1692 = None
	        full_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_108, full_216);  amin_108 = full_216 = None
	        full_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_108, full_217);  amax_108 = full_217 = None
	        sub_4950: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_108, minimum_108);  maximum_108 = None
	        div_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4950, 255.0);  sub_4950 = None
	        clamp_min_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_216, 1.1920928955078125e-07);  div_216 = None
	        div_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_108, clamp_min_324);  minimum_108 = None
	        round_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_217);  div_217 = None
	        sub_4956: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_217);  round_217 = None
	        clamp_min_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4956, -128);  sub_4956 = None
	        clamp_max_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_325, 127);  clamp_min_325 = None
	        _assert_tensor_metadata_974 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_324, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_974 = None
	        _assert_tensor_metadata_975 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_216, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_975 = None
	        convert_element_type_648: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_216, torch.int8);  clamp_max_216 = None
	        view_1693: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_16556, [sym_size_int, 1500, 1280])
	        view_1694: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_324, [sym_size_int, 1500, 1])
	        view_1695: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_648, [sym_size_int, 1500, 1])
	        reciprocal_108: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1694);  view_1694 = None
	        mul_10509: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_108, 1.0);  reciprocal_108 = None
	        mul_10512: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1693, mul_10509);  view_1693 = mul_10509 = None
	        round_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10512);  mul_10512 = None
	        add_16643: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_218, view_1695);  round_218 = view_1695 = None
	        clamp_min_326: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16643, -128);  add_16643 = None
	        clamp_max_217: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_326, 127);  clamp_min_326 = None
	        view_1696: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_217, [sym_size_int, 1500, 1280]);  clamp_max_217 = None
	        _assert_tensor_metadata_976 = torch.ops.aten._assert_tensor_metadata.default(view_1696, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_976 = None
	        convert_element_type_649: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1696, torch.int8);  view_1696 = None
	        view_1697: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_649, [sym_size_int, 1500, 1280]);  convert_element_type_649 = None
	        view_1698: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_324, [sym_size_int, 1500, 1]);  clamp_min_324 = None
	        view_1699: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_648, [sym_size_int, 1500, 1]);  convert_element_type_648 = None
	        _assert_tensor_metadata_977 = torch.ops.aten._assert_tensor_metadata.default(view_1697, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_977 = None
	        convert_element_type_650: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1697, torch.float32);  view_1697 = None
	        _assert_tensor_metadata_978 = torch.ops.aten._assert_tensor_metadata.default(view_1699, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_978 = None
	        convert_element_type_651: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1699, torch.float32);  view_1699 = None
	        sub_4976: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_650, convert_element_type_651);  convert_element_type_650 = convert_element_type_651 = None
	        mul_10534: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4976, view_1698);  sub_4976 = view_1698 = None
	        view_1700: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10534, [sym_size_int, 1500, 1280]);  mul_10534 = None
	        _assert_tensor_metadata_979 = torch.ops.aten._assert_tensor_metadata.default(view_1700, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_979 = None
	        view_1701: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg494_1, [1280, 40, 32]);  arg494_1 = None
	        view_1702: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg495_1, [1280, 40, 1]);  arg495_1 = None
	        view_1703: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg496_1, [1280, 40, 1]);  arg496_1 = None
	        _assert_tensor_metadata_980 = torch.ops.aten._assert_tensor_metadata.default(view_1701, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_980 = None
	        convert_element_type_652: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1701, torch.float32);  view_1701 = None
	        _assert_tensor_metadata_981 = torch.ops.aten._assert_tensor_metadata.default(view_1703, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_981 = None
	        convert_element_type_653: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1703, torch.float32);  view_1703 = None
	        sub_4980: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_652, convert_element_type_653);  convert_element_type_652 = convert_element_type_653 = None
	        mul_10539: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4980, view_1702);  sub_4980 = view_1702 = None
	        view_1704: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10539, [1280, 1280]);  mul_10539 = None
	        _assert_tensor_metadata_982 = torch.ops.aten._assert_tensor_metadata.default(view_1704, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_982 = None
	        mul_10544: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1705: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1700, [mul_10544, 1280]);  view_1700 = mul_10544 = None
	        permute_181: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1704, [1, 0]);  view_1704 = None
	        addmm_90: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg493_1, view_1705, permute_181);  arg493_1 = view_1705 = permute_181 = None
	        view_1706: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_90, [sym_size_int, 1500, 1280]);  addmm_90 = None
	        mul_10551: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1706, 0.125);  view_1706 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1707: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_10551, [sym_size_int, 1500, 20, 64]);  mul_10551 = None
	        permute_182: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1707, [0, 2, 1, 3]);  view_1707 = None
	        clone_146: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_182, memory_format = torch.contiguous_format);  permute_182 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1708: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_16556, [sym_size_int, 1500, 1280])
	        amin_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1708, [2])
	        amax_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1708, [2]);  view_1708 = None
	        full_218: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_109, full_218);  amin_109 = full_218 = None
	        full_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_109, full_219);  amax_109 = full_219 = None
	        sub_4995: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_109, minimum_109);  maximum_109 = None
	        div_218: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4995, 255.0);  sub_4995 = None
	        clamp_min_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_218, 1.1920928955078125e-07);  div_218 = None
	        div_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_109, clamp_min_327);  minimum_109 = None
	        round_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_219);  div_219 = None
	        sub_5001: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_219);  round_219 = None
	        clamp_min_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5001, -128);  sub_5001 = None
	        clamp_max_218: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_328, 127);  clamp_min_328 = None
	        _assert_tensor_metadata_983 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_327, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_983 = None
	        _assert_tensor_metadata_984 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_218, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_984 = None
	        convert_element_type_654: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_218, torch.int8);  clamp_max_218 = None
	        view_1709: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_16556, [sym_size_int, 1500, 1280])
	        view_1710: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_327, [sym_size_int, 1500, 1])
	        view_1711: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_654, [sym_size_int, 1500, 1])
	        reciprocal_109: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1710);  view_1710 = None
	        mul_10605: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_109, 1.0);  reciprocal_109 = None
	        mul_10608: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1709, mul_10605);  view_1709 = mul_10605 = None
	        round_220: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10608);  mul_10608 = None
	        add_16795: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_220, view_1711);  round_220 = view_1711 = None
	        clamp_min_329: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16795, -128);  add_16795 = None
	        clamp_max_219: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_329, 127);  clamp_min_329 = None
	        view_1712: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_219, [sym_size_int, 1500, 1280]);  clamp_max_219 = None
	        _assert_tensor_metadata_985 = torch.ops.aten._assert_tensor_metadata.default(view_1712, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_985 = None
	        convert_element_type_655: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1712, torch.int8);  view_1712 = None
	        view_1713: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_655, [sym_size_int, 1500, 1280]);  convert_element_type_655 = None
	        view_1714: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_327, [sym_size_int, 1500, 1]);  clamp_min_327 = None
	        view_1715: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_654, [sym_size_int, 1500, 1]);  convert_element_type_654 = None
	        _assert_tensor_metadata_986 = torch.ops.aten._assert_tensor_metadata.default(view_1713, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_986 = None
	        convert_element_type_656: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1713, torch.float32);  view_1713 = None
	        _assert_tensor_metadata_987 = torch.ops.aten._assert_tensor_metadata.default(view_1715, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_987 = None
	        convert_element_type_657: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1715, torch.float32);  view_1715 = None
	        sub_5021: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_656, convert_element_type_657);  convert_element_type_656 = convert_element_type_657 = None
	        mul_10630: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5021, view_1714);  sub_5021 = view_1714 = None
	        view_1716: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10630, [sym_size_int, 1500, 1280]);  mul_10630 = None
	        _assert_tensor_metadata_988 = torch.ops.aten._assert_tensor_metadata.default(view_1716, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_988 = None
	        view_1717: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg497_1, [1280, 40, 32]);  arg497_1 = None
	        view_1718: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg498_1, [1280, 40, 1]);  arg498_1 = None
	        view_1719: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg499_1, [1280, 40, 1]);  arg499_1 = None
	        _assert_tensor_metadata_989 = torch.ops.aten._assert_tensor_metadata.default(view_1717, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_989 = None
	        convert_element_type_658: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1717, torch.float32);  view_1717 = None
	        _assert_tensor_metadata_990 = torch.ops.aten._assert_tensor_metadata.default(view_1719, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_990 = None
	        convert_element_type_659: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1719, torch.float32);  view_1719 = None
	        sub_5025: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_658, convert_element_type_659);  convert_element_type_658 = convert_element_type_659 = None
	        mul_10635: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5025, view_1718);  sub_5025 = view_1718 = None
	        view_1720: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10635, [1280, 1280]);  mul_10635 = None
	        _assert_tensor_metadata_991 = torch.ops.aten._assert_tensor_metadata.default(view_1720, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_991 = None
	        permute_183: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1720, [1, 0]);  view_1720 = None
	        mul_10638: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1721: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1716, [mul_10638, 1280]);  view_1716 = mul_10638 = None
	        mm_18: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1721, permute_183);  view_1721 = permute_183 = None
	        view_1722: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_18, [sym_size_int, 1500, 1280]);  mm_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1723: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1722, [sym_size_int, -1, 20, 64]);  view_1722 = None
	        permute_184: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1723, [0, 2, 1, 3]);  view_1723 = None
	        clone_147: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_184, memory_format = torch.contiguous_format);  permute_184 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1724: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_16556, [sym_size_int, 1500, 1280])
	        amin_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1724, [2])
	        amax_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1724, [2]);  view_1724 = None
	        full_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_110, full_220);  amin_110 = full_220 = None
	        full_221: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_110, full_221);  amax_110 = full_221 = None
	        sub_5039: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_110, minimum_110);  maximum_110 = None
	        div_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5039, 255.0);  sub_5039 = None
	        clamp_min_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_220, 1.1920928955078125e-07);  div_220 = None
	        div_221: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_110, clamp_min_330);  minimum_110 = None
	        round_221: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_221);  div_221 = None
	        sub_5045: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_221);  round_221 = None
	        clamp_min_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5045, -128);  sub_5045 = None
	        clamp_max_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_331, 127);  clamp_min_331 = None
	        _assert_tensor_metadata_992 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_330, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_992 = None
	        _assert_tensor_metadata_993 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_220, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_993 = None
	        convert_element_type_660: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_220, torch.int8);  clamp_max_220 = None
	        view_1725: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_16556, [sym_size_int, 1500, 1280]);  add_16556 = None
	        view_1726: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_330, [sym_size_int, 1500, 1])
	        view_1727: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_660, [sym_size_int, 1500, 1])
	        reciprocal_110: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1726);  view_1726 = None
	        mul_10704: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_110, 1.0);  reciprocal_110 = None
	        mul_10707: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1725, mul_10704);  view_1725 = mul_10704 = None
	        round_222: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10707);  mul_10707 = None
	        add_16943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_222, view_1727);  round_222 = view_1727 = None
	        clamp_min_332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16943, -128);  add_16943 = None
	        clamp_max_221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_332, 127);  clamp_min_332 = None
	        view_1728: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_221, [sym_size_int, 1500, 1280]);  clamp_max_221 = None
	        _assert_tensor_metadata_994 = torch.ops.aten._assert_tensor_metadata.default(view_1728, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_994 = None
	        convert_element_type_661: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1728, torch.int8);  view_1728 = None
	        view_1729: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_661, [sym_size_int, 1500, 1280]);  convert_element_type_661 = None
	        view_1730: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_330, [sym_size_int, 1500, 1]);  clamp_min_330 = None
	        view_1731: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_660, [sym_size_int, 1500, 1]);  convert_element_type_660 = None
	        _assert_tensor_metadata_995 = torch.ops.aten._assert_tensor_metadata.default(view_1729, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_995 = None
	        convert_element_type_662: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1729, torch.float32);  view_1729 = None
	        _assert_tensor_metadata_996 = torch.ops.aten._assert_tensor_metadata.default(view_1731, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_996 = None
	        convert_element_type_663: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1731, torch.float32);  view_1731 = None
	        sub_5065: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_662, convert_element_type_663);  convert_element_type_662 = convert_element_type_663 = None
	        mul_10729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5065, view_1730);  sub_5065 = view_1730 = None
	        view_1732: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10729, [sym_size_int, 1500, 1280]);  mul_10729 = None
	        _assert_tensor_metadata_997 = torch.ops.aten._assert_tensor_metadata.default(view_1732, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_997 = None
	        view_1733: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg501_1, [1280, 40, 32]);  arg501_1 = None
	        view_1734: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg502_1, [1280, 40, 1]);  arg502_1 = None
	        view_1735: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg503_1, [1280, 40, 1]);  arg503_1 = None
	        _assert_tensor_metadata_998 = torch.ops.aten._assert_tensor_metadata.default(view_1733, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_998 = None
	        convert_element_type_664: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1733, torch.float32);  view_1733 = None
	        _assert_tensor_metadata_999 = torch.ops.aten._assert_tensor_metadata.default(view_1735, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_999 = None
	        convert_element_type_665: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1735, torch.float32);  view_1735 = None
	        sub_5069: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_664, convert_element_type_665);  convert_element_type_664 = convert_element_type_665 = None
	        mul_10734: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5069, view_1734);  sub_5069 = view_1734 = None
	        view_1736: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10734, [1280, 1280]);  mul_10734 = None
	        _assert_tensor_metadata_1000 = torch.ops.aten._assert_tensor_metadata.default(view_1736, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1000 = None
	        mul_10739: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1737: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1732, [mul_10739, 1280]);  view_1732 = mul_10739 = None
	        permute_185: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1736, [1, 0]);  view_1736 = None
	        addmm_91: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg500_1, view_1737, permute_185);  arg500_1 = view_1737 = permute_185 = None
	        view_1738: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_91, [sym_size_int, 1500, 1280]);  addmm_91 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1739: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1738, [sym_size_int, -1, 20, 64]);  view_1738 = None
	        permute_186: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1739, [0, 2, 1, 3]);  view_1739 = None
	        clone_148: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_186, memory_format = torch.contiguous_format);  permute_186 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_18 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_146, clone_147, clone_148, None, False, scale = 1.0);  clone_146 = clone_147 = clone_148 = None
	        getitem_146: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_18[0];  _scaled_dot_product_efficient_attention_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_187: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_146, [0, 2, 1, 3]);  getitem_146 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1740: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_187, [sym_size_int, 1500, -1]);  permute_187 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1741: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1740, [sym_size_int, 1500, 1280])
	        amin_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1741, [2])
	        amax_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1741, [2]);  view_1741 = None
	        full_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_111, full_222);  amin_111 = full_222 = None
	        full_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_111, full_223);  amax_111 = full_223 = None
	        sub_5087: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_111, minimum_111);  maximum_111 = None
	        div_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5087, 255.0);  sub_5087 = None
	        clamp_min_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_222, 1.1920928955078125e-07);  div_222 = None
	        div_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_111, clamp_min_333);  minimum_111 = None
	        round_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_223);  div_223 = None
	        sub_5093: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_223);  round_223 = None
	        clamp_min_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5093, -128);  sub_5093 = None
	        clamp_max_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_334, 127);  clamp_min_334 = None
	        _assert_tensor_metadata_1001 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_333, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1001 = None
	        _assert_tensor_metadata_1002 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_222, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1002 = None
	        convert_element_type_666: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_222, torch.int8);  clamp_max_222 = None
	        view_1742: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1740, [sym_size_int, 1500, 1280]);  view_1740 = None
	        view_1743: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_333, [sym_size_int, 1500, 1])
	        view_1744: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_666, [sym_size_int, 1500, 1])
	        reciprocal_111: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1743);  view_1743 = None
	        mul_10809: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_111, 1.0);  reciprocal_111 = None
	        mul_10812: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1742, mul_10809);  view_1742 = mul_10809 = None
	        round_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10812);  mul_10812 = None
	        add_17107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_224, view_1744);  round_224 = view_1744 = None
	        clamp_min_335: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17107, -128);  add_17107 = None
	        clamp_max_223: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_335, 127);  clamp_min_335 = None
	        view_1745: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_223, [sym_size_int, 1500, 1280]);  clamp_max_223 = None
	        _assert_tensor_metadata_1003 = torch.ops.aten._assert_tensor_metadata.default(view_1745, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1003 = None
	        convert_element_type_667: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1745, torch.int8);  view_1745 = None
	        view_1746: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_667, [sym_size_int, 1500, 1280]);  convert_element_type_667 = None
	        view_1747: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_333, [sym_size_int, 1500, 1]);  clamp_min_333 = None
	        view_1748: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_666, [sym_size_int, 1500, 1]);  convert_element_type_666 = None
	        _assert_tensor_metadata_1004 = torch.ops.aten._assert_tensor_metadata.default(view_1746, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1004 = None
	        convert_element_type_668: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1746, torch.float32);  view_1746 = None
	        _assert_tensor_metadata_1005 = torch.ops.aten._assert_tensor_metadata.default(view_1748, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1005 = None
	        convert_element_type_669: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1748, torch.float32);  view_1748 = None
	        sub_5113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_668, convert_element_type_669);  convert_element_type_668 = convert_element_type_669 = None
	        mul_10834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5113, view_1747);  sub_5113 = view_1747 = None
	        view_1749: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10834, [sym_size_int, 1500, 1280]);  mul_10834 = None
	        _assert_tensor_metadata_1006 = torch.ops.aten._assert_tensor_metadata.default(view_1749, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1006 = None
	        view_1750: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg505_1, [1280, 40, 32]);  arg505_1 = None
	        view_1751: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg506_1, [1280, 40, 1]);  arg506_1 = None
	        view_1752: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg507_1, [1280, 40, 1]);  arg507_1 = None
	        _assert_tensor_metadata_1007 = torch.ops.aten._assert_tensor_metadata.default(view_1750, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1007 = None
	        convert_element_type_670: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1750, torch.float32);  view_1750 = None
	        _assert_tensor_metadata_1008 = torch.ops.aten._assert_tensor_metadata.default(view_1752, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1008 = None
	        convert_element_type_671: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1752, torch.float32);  view_1752 = None
	        sub_5117: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_670, convert_element_type_671);  convert_element_type_670 = convert_element_type_671 = None
	        mul_10839: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5117, view_1751);  sub_5117 = view_1751 = None
	        view_1753: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10839, [1280, 1280]);  mul_10839 = None
	        _assert_tensor_metadata_1009 = torch.ops.aten._assert_tensor_metadata.default(view_1753, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1009 = None
	        mul_10844: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1754: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1749, [mul_10844, 1280]);  view_1749 = mul_10844 = None
	        permute_188: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1753, [1, 0]);  view_1753 = None
	        addmm_92: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg504_1, view_1754, permute_188);  arg504_1 = view_1754 = permute_188 = None
	        view_1755: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_92, [sym_size_int, 1500, 1280]);  addmm_92 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1755);  view_1755 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_17170: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_16550, clone_149);  add_16550 = clone_149 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_150: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_17170, memory_format = torch.contiguous_format)
	        var_mean_37 = torch.ops.aten.var_mean.correction(clone_150, [2], correction = 0, keepdim = True)
	        getitem_150: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_37[0]
	        getitem_151: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_37[1];  var_mean_37 = None
	        add_17175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_150, 1e-05);  getitem_150 = None
	        rsqrt_37: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_17175);  add_17175 = None
	        sub_5123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_150, getitem_151);  clone_150 = getitem_151 = None
	        mul_10855: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5123, rsqrt_37);  sub_5123 = rsqrt_37 = None
	        mul_10856: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10855, arg508_1);  mul_10855 = arg508_1 = None
	        add_17176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_10856, arg509_1);  mul_10856 = arg509_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1756: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_17176, [sym_size_int, 1500, 1280])
	        amin_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1756, [2])
	        amax_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1756, [2]);  view_1756 = None
	        full_224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_112, full_224);  amin_112 = full_224 = None
	        full_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_112, full_225);  amax_112 = full_225 = None
	        sub_5134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_112, minimum_112);  maximum_112 = None
	        div_224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5134, 255.0);  sub_5134 = None
	        clamp_min_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_224, 1.1920928955078125e-07);  div_224 = None
	        div_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_112, clamp_min_336);  minimum_112 = None
	        round_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_225);  div_225 = None
	        sub_5140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_225);  round_225 = None
	        clamp_min_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5140, -128);  sub_5140 = None
	        clamp_max_224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_337, 127);  clamp_min_337 = None
	        _assert_tensor_metadata_1010 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_336, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1010 = None
	        _assert_tensor_metadata_1011 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_224, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1011 = None
	        convert_element_type_672: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_224, torch.int8);  clamp_max_224 = None
	        view_1757: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_17176, [sym_size_int, 1500, 1280]);  add_17176 = None
	        view_1758: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_336, [sym_size_int, 1500, 1])
	        view_1759: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_672, [sym_size_int, 1500, 1])
	        reciprocal_112: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1758);  view_1758 = None
	        mul_10904: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_112, 1.0);  reciprocal_112 = None
	        mul_10907: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1757, mul_10904);  view_1757 = mul_10904 = None
	        round_226: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10907);  mul_10907 = None
	        add_17263: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_226, view_1759);  round_226 = view_1759 = None
	        clamp_min_338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17263, -128);  add_17263 = None
	        clamp_max_225: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_338, 127);  clamp_min_338 = None
	        view_1760: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_225, [sym_size_int, 1500, 1280]);  clamp_max_225 = None
	        _assert_tensor_metadata_1012 = torch.ops.aten._assert_tensor_metadata.default(view_1760, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1012 = None
	        convert_element_type_673: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1760, torch.int8);  view_1760 = None
	        view_1761: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_673, [sym_size_int, 1500, 1280]);  convert_element_type_673 = None
	        view_1762: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_336, [sym_size_int, 1500, 1]);  clamp_min_336 = None
	        view_1763: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_672, [sym_size_int, 1500, 1]);  convert_element_type_672 = None
	        _assert_tensor_metadata_1013 = torch.ops.aten._assert_tensor_metadata.default(view_1761, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1013 = None
	        convert_element_type_674: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1761, torch.float32);  view_1761 = None
	        _assert_tensor_metadata_1014 = torch.ops.aten._assert_tensor_metadata.default(view_1763, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1014 = None
	        convert_element_type_675: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1763, torch.float32);  view_1763 = None
	        sub_5160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_674, convert_element_type_675);  convert_element_type_674 = convert_element_type_675 = None
	        mul_10929: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5160, view_1762);  sub_5160 = view_1762 = None
	        view_1764: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10929, [sym_size_int, 1500, 1280]);  mul_10929 = None
	        _assert_tensor_metadata_1015 = torch.ops.aten._assert_tensor_metadata.default(view_1764, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1015 = None
	        view_1765: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg511_1, [5120, 40, 32]);  arg511_1 = None
	        view_1766: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg512_1, [5120, 40, 1]);  arg512_1 = None
	        view_1767: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg513_1, [5120, 40, 1]);  arg513_1 = None
	        _assert_tensor_metadata_1016 = torch.ops.aten._assert_tensor_metadata.default(view_1765, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1016 = None
	        convert_element_type_676: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1765, torch.float32);  view_1765 = None
	        _assert_tensor_metadata_1017 = torch.ops.aten._assert_tensor_metadata.default(view_1767, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1017 = None
	        convert_element_type_677: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1767, torch.float32);  view_1767 = None
	        sub_5164: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_676, convert_element_type_677);  convert_element_type_676 = convert_element_type_677 = None
	        mul_10934: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5164, view_1766);  sub_5164 = view_1766 = None
	        view_1768: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10934, [5120, 1280]);  mul_10934 = None
	        _assert_tensor_metadata_1018 = torch.ops.aten._assert_tensor_metadata.default(view_1768, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1018 = None
	        mul_10939: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1769: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1764, [mul_10939, 1280]);  view_1764 = mul_10939 = None
	        permute_189: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1768, [1, 0]);  view_1768 = None
	        addmm_93: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg510_1, view_1769, permute_189);  arg510_1 = view_1769 = permute_189 = None
	        view_1770: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_93, [sym_size_int, 1500, 5120]);  addmm_93 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_10946: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1770, 0.5)
	        mul_10947: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1770, 0.7071067811865476);  view_1770 = None
	        erf_20: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_10947);  mul_10947 = None
	        add_17322: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_20, 1);  erf_20 = None
	        mul_10948: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10946, add_17322);  mul_10946 = add_17322 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_151: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_10948);  mul_10948 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1771: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_151, [sym_size_int, 1500, 5120])
	        amin_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1771, [2])
	        amax_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1771, [2]);  view_1771 = None
	        full_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_113, full_226);  amin_113 = full_226 = None
	        full_227: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_113, full_227);  amax_113 = full_227 = None
	        sub_5177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_113, minimum_113);  maximum_113 = None
	        div_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5177, 255.0);  sub_5177 = None
	        clamp_min_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_226, 1.1920928955078125e-07);  div_226 = None
	        div_227: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_113, clamp_min_339);  minimum_113 = None
	        round_227: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_227);  div_227 = None
	        sub_5183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_227);  round_227 = None
	        clamp_min_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5183, -128);  sub_5183 = None
	        clamp_max_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_340, 127);  clamp_min_340 = None
	        _assert_tensor_metadata_1019 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_339, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1019 = None
	        _assert_tensor_metadata_1020 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_226, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1020 = None
	        convert_element_type_678: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_226, torch.int8);  clamp_max_226 = None
	        view_1772: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_151, [sym_size_int, 1500, 5120]);  clone_151 = None
	        view_1773: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_339, [sym_size_int, 1500, 1])
	        view_1774: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_678, [sym_size_int, 1500, 1])
	        reciprocal_113: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1773);  view_1773 = None
	        mul_10994: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_113, 1.0);  reciprocal_113 = None
	        mul_10997: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1772, mul_10994);  view_1772 = mul_10994 = None
	        round_228: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_10997);  mul_10997 = None
	        add_17405: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_228, view_1774);  round_228 = view_1774 = None
	        clamp_min_341: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17405, -128);  add_17405 = None
	        clamp_max_227: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_341, 127);  clamp_min_341 = None
	        view_1775: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_227, [sym_size_int, 1500, 5120]);  clamp_max_227 = None
	        _assert_tensor_metadata_1021 = torch.ops.aten._assert_tensor_metadata.default(view_1775, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1021 = None
	        convert_element_type_679: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1775, torch.int8);  view_1775 = None
	        view_1776: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_679, [sym_size_int, 1500, 5120]);  convert_element_type_679 = None
	        view_1777: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_339, [sym_size_int, 1500, 1]);  clamp_min_339 = None
	        view_1778: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_678, [sym_size_int, 1500, 1]);  convert_element_type_678 = None
	        _assert_tensor_metadata_1022 = torch.ops.aten._assert_tensor_metadata.default(view_1776, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1022 = None
	        convert_element_type_680: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1776, torch.float32);  view_1776 = None
	        _assert_tensor_metadata_1023 = torch.ops.aten._assert_tensor_metadata.default(view_1778, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1023 = None
	        convert_element_type_681: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1778, torch.float32);  view_1778 = None
	        sub_5203: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_680, convert_element_type_681);  convert_element_type_680 = convert_element_type_681 = None
	        mul_11019: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5203, view_1777);  sub_5203 = view_1777 = None
	        view_1779: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_11019, [sym_size_int, 1500, 5120]);  mul_11019 = None
	        _assert_tensor_metadata_1024 = torch.ops.aten._assert_tensor_metadata.default(view_1779, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1024 = None
	        view_1780: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg515_1, [1280, 160, 32]);  arg515_1 = None
	        view_1781: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg516_1, [1280, 160, 1]);  arg516_1 = None
	        view_1782: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg517_1, [1280, 160, 1]);  arg517_1 = None
	        _assert_tensor_metadata_1025 = torch.ops.aten._assert_tensor_metadata.default(view_1780, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1025 = None
	        convert_element_type_682: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1780, torch.float32);  view_1780 = None
	        _assert_tensor_metadata_1026 = torch.ops.aten._assert_tensor_metadata.default(view_1782, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1026 = None
	        convert_element_type_683: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1782, torch.float32);  view_1782 = None
	        sub_5207: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_682, convert_element_type_683);  convert_element_type_682 = convert_element_type_683 = None
	        mul_11024: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5207, view_1781);  sub_5207 = view_1781 = None
	        view_1783: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_11024, [1280, 5120]);  mul_11024 = None
	        _assert_tensor_metadata_1027 = torch.ops.aten._assert_tensor_metadata.default(view_1783, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1027 = None
	        mul_11029: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1784: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1779, [mul_11029, 5120]);  view_1779 = mul_11029 = None
	        permute_190: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1783, [1, 0]);  view_1783 = None
	        addmm_94: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg514_1, view_1784, permute_190);  arg514_1 = view_1784 = permute_190 = None
	        view_1785: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_94, [sym_size_int, 1500, 1280]);  addmm_94 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1785);  view_1785 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_17468: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_17170, clone_152);  add_17170 = clone_152 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_153: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_17468, memory_format = torch.contiguous_format)
	        var_mean_38 = torch.ops.aten.var_mean.correction(clone_153, [2], correction = 0, keepdim = True)
	        getitem_152: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_38[0]
	        getitem_153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_38[1];  var_mean_38 = None
	        add_17473: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_152, 1e-05);  getitem_152 = None
	        rsqrt_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_17473);  add_17473 = None
	        sub_5213: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_153, getitem_153);  clone_153 = getitem_153 = None
	        mul_11040: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5213, rsqrt_38);  sub_5213 = rsqrt_38 = None
	        mul_11041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11040, arg518_1);  mul_11040 = arg518_1 = None
	        add_17474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_11041, arg519_1);  mul_11041 = arg519_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1786: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_17474, [sym_size_int, 1500, 1280])
	        amin_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1786, [2])
	        amax_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1786, [2]);  view_1786 = None
	        full_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_114, full_228);  amin_114 = full_228 = None
	        full_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_114, full_229);  amax_114 = full_229 = None
	        sub_5224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_114, minimum_114);  maximum_114 = None
	        div_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5224, 255.0);  sub_5224 = None
	        clamp_min_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_228, 1.1920928955078125e-07);  div_228 = None
	        div_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_114, clamp_min_342);  minimum_114 = None
	        round_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_229);  div_229 = None
	        sub_5230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_229);  round_229 = None
	        clamp_min_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5230, -128);  sub_5230 = None
	        clamp_max_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_343, 127);  clamp_min_343 = None
	        _assert_tensor_metadata_1028 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_342, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1028 = None
	        _assert_tensor_metadata_1029 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_228, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1029 = None
	        convert_element_type_684: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_228, torch.int8);  clamp_max_228 = None
	        view_1787: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_17474, [sym_size_int, 1500, 1280])
	        view_1788: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_342, [sym_size_int, 1500, 1])
	        view_1789: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_684, [sym_size_int, 1500, 1])
	        reciprocal_114: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1788);  view_1788 = None
	        mul_11089: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_114, 1.0);  reciprocal_114 = None
	        mul_11092: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1787, mul_11089);  view_1787 = mul_11089 = None
	        round_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11092);  mul_11092 = None
	        add_17561: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_230, view_1789);  round_230 = view_1789 = None
	        clamp_min_344: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17561, -128);  add_17561 = None
	        clamp_max_229: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_344, 127);  clamp_min_344 = None
	        view_1790: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_229, [sym_size_int, 1500, 1280]);  clamp_max_229 = None
	        _assert_tensor_metadata_1030 = torch.ops.aten._assert_tensor_metadata.default(view_1790, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1030 = None
	        convert_element_type_685: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1790, torch.int8);  view_1790 = None
	        view_1791: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_685, [sym_size_int, 1500, 1280]);  convert_element_type_685 = None
	        view_1792: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_342, [sym_size_int, 1500, 1]);  clamp_min_342 = None
	        view_1793: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_684, [sym_size_int, 1500, 1]);  convert_element_type_684 = None
	        _assert_tensor_metadata_1031 = torch.ops.aten._assert_tensor_metadata.default(view_1791, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1031 = None
	        convert_element_type_686: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1791, torch.float32);  view_1791 = None
	        _assert_tensor_metadata_1032 = torch.ops.aten._assert_tensor_metadata.default(view_1793, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1032 = None
	        convert_element_type_687: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1793, torch.float32);  view_1793 = None
	        sub_5250: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_686, convert_element_type_687);  convert_element_type_686 = convert_element_type_687 = None
	        mul_11114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5250, view_1792);  sub_5250 = view_1792 = None
	        view_1794: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11114, [sym_size_int, 1500, 1280]);  mul_11114 = None
	        _assert_tensor_metadata_1033 = torch.ops.aten._assert_tensor_metadata.default(view_1794, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1033 = None
	        view_1795: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg521_1, [1280, 40, 32]);  arg521_1 = None
	        view_1796: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg522_1, [1280, 40, 1]);  arg522_1 = None
	        view_1797: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg523_1, [1280, 40, 1]);  arg523_1 = None
	        _assert_tensor_metadata_1034 = torch.ops.aten._assert_tensor_metadata.default(view_1795, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1034 = None
	        convert_element_type_688: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1795, torch.float32);  view_1795 = None
	        _assert_tensor_metadata_1035 = torch.ops.aten._assert_tensor_metadata.default(view_1797, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1035 = None
	        convert_element_type_689: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1797, torch.float32);  view_1797 = None
	        sub_5254: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_688, convert_element_type_689);  convert_element_type_688 = convert_element_type_689 = None
	        mul_11119: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5254, view_1796);  sub_5254 = view_1796 = None
	        view_1798: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11119, [1280, 1280]);  mul_11119 = None
	        _assert_tensor_metadata_1036 = torch.ops.aten._assert_tensor_metadata.default(view_1798, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1036 = None
	        mul_11124: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1799: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1794, [mul_11124, 1280]);  view_1794 = mul_11124 = None
	        permute_191: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1798, [1, 0]);  view_1798 = None
	        addmm_95: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg520_1, view_1799, permute_191);  arg520_1 = view_1799 = permute_191 = None
	        view_1800: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_95, [sym_size_int, 1500, 1280]);  addmm_95 = None
	        mul_11131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1800, 0.125);  view_1800 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1801: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_11131, [sym_size_int, 1500, 20, 64]);  mul_11131 = None
	        permute_192: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1801, [0, 2, 1, 3]);  view_1801 = None
	        clone_154: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_192, memory_format = torch.contiguous_format);  permute_192 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1802: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_17474, [sym_size_int, 1500, 1280])
	        amin_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1802, [2])
	        amax_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1802, [2]);  view_1802 = None
	        full_230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_115, full_230);  amin_115 = full_230 = None
	        full_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_115, full_231);  amax_115 = full_231 = None
	        sub_5269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_115, minimum_115);  maximum_115 = None
	        div_230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5269, 255.0);  sub_5269 = None
	        clamp_min_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_230, 1.1920928955078125e-07);  div_230 = None
	        div_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_115, clamp_min_345);  minimum_115 = None
	        round_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_231);  div_231 = None
	        sub_5275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_231);  round_231 = None
	        clamp_min_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5275, -128);  sub_5275 = None
	        clamp_max_230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_346, 127);  clamp_min_346 = None
	        _assert_tensor_metadata_1037 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_345, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1037 = None
	        _assert_tensor_metadata_1038 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_230, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1038 = None
	        convert_element_type_690: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_230, torch.int8);  clamp_max_230 = None
	        view_1803: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_17474, [sym_size_int, 1500, 1280])
	        view_1804: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_345, [sym_size_int, 1500, 1])
	        view_1805: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_690, [sym_size_int, 1500, 1])
	        reciprocal_115: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1804);  view_1804 = None
	        mul_11185: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_115, 1.0);  reciprocal_115 = None
	        mul_11188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1803, mul_11185);  view_1803 = mul_11185 = None
	        round_232: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11188);  mul_11188 = None
	        add_17713: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_232, view_1805);  round_232 = view_1805 = None
	        clamp_min_347: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17713, -128);  add_17713 = None
	        clamp_max_231: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_347, 127);  clamp_min_347 = None
	        view_1806: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_231, [sym_size_int, 1500, 1280]);  clamp_max_231 = None
	        _assert_tensor_metadata_1039 = torch.ops.aten._assert_tensor_metadata.default(view_1806, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1039 = None
	        convert_element_type_691: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1806, torch.int8);  view_1806 = None
	        view_1807: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_691, [sym_size_int, 1500, 1280]);  convert_element_type_691 = None
	        view_1808: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_345, [sym_size_int, 1500, 1]);  clamp_min_345 = None
	        view_1809: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_690, [sym_size_int, 1500, 1]);  convert_element_type_690 = None
	        _assert_tensor_metadata_1040 = torch.ops.aten._assert_tensor_metadata.default(view_1807, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1040 = None
	        convert_element_type_692: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1807, torch.float32);  view_1807 = None
	        _assert_tensor_metadata_1041 = torch.ops.aten._assert_tensor_metadata.default(view_1809, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1041 = None
	        convert_element_type_693: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1809, torch.float32);  view_1809 = None
	        sub_5295: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_692, convert_element_type_693);  convert_element_type_692 = convert_element_type_693 = None
	        mul_11210: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5295, view_1808);  sub_5295 = view_1808 = None
	        view_1810: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11210, [sym_size_int, 1500, 1280]);  mul_11210 = None
	        _assert_tensor_metadata_1042 = torch.ops.aten._assert_tensor_metadata.default(view_1810, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1042 = None
	        view_1811: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg524_1, [1280, 40, 32]);  arg524_1 = None
	        view_1812: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg525_1, [1280, 40, 1]);  arg525_1 = None
	        view_1813: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg526_1, [1280, 40, 1]);  arg526_1 = None
	        _assert_tensor_metadata_1043 = torch.ops.aten._assert_tensor_metadata.default(view_1811, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1043 = None
	        convert_element_type_694: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1811, torch.float32);  view_1811 = None
	        _assert_tensor_metadata_1044 = torch.ops.aten._assert_tensor_metadata.default(view_1813, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1044 = None
	        convert_element_type_695: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1813, torch.float32);  view_1813 = None
	        sub_5299: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_694, convert_element_type_695);  convert_element_type_694 = convert_element_type_695 = None
	        mul_11215: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5299, view_1812);  sub_5299 = view_1812 = None
	        view_1814: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11215, [1280, 1280]);  mul_11215 = None
	        _assert_tensor_metadata_1045 = torch.ops.aten._assert_tensor_metadata.default(view_1814, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1045 = None
	        permute_193: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1814, [1, 0]);  view_1814 = None
	        mul_11218: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1815: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1810, [mul_11218, 1280]);  view_1810 = mul_11218 = None
	        mm_19: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1815, permute_193);  view_1815 = permute_193 = None
	        view_1816: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_19, [sym_size_int, 1500, 1280]);  mm_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1817: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1816, [sym_size_int, -1, 20, 64]);  view_1816 = None
	        permute_194: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1817, [0, 2, 1, 3]);  view_1817 = None
	        clone_155: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_194, memory_format = torch.contiguous_format);  permute_194 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1818: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_17474, [sym_size_int, 1500, 1280])
	        amin_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1818, [2])
	        amax_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1818, [2]);  view_1818 = None
	        full_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_116, full_232);  amin_116 = full_232 = None
	        full_233: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_116, full_233);  amax_116 = full_233 = None
	        sub_5313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_116, minimum_116);  maximum_116 = None
	        div_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5313, 255.0);  sub_5313 = None
	        clamp_min_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_232, 1.1920928955078125e-07);  div_232 = None
	        div_233: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_116, clamp_min_348);  minimum_116 = None
	        round_233: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_233);  div_233 = None
	        sub_5319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_233);  round_233 = None
	        clamp_min_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5319, -128);  sub_5319 = None
	        clamp_max_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_349, 127);  clamp_min_349 = None
	        _assert_tensor_metadata_1046 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_348, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1046 = None
	        _assert_tensor_metadata_1047 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_232, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1047 = None
	        convert_element_type_696: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_232, torch.int8);  clamp_max_232 = None
	        view_1819: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_17474, [sym_size_int, 1500, 1280]);  add_17474 = None
	        view_1820: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_348, [sym_size_int, 1500, 1])
	        view_1821: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_696, [sym_size_int, 1500, 1])
	        reciprocal_116: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1820);  view_1820 = None
	        mul_11284: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_116, 1.0);  reciprocal_116 = None
	        mul_11287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1819, mul_11284);  view_1819 = mul_11284 = None
	        round_234: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11287);  mul_11287 = None
	        add_17861: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_234, view_1821);  round_234 = view_1821 = None
	        clamp_min_350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17861, -128);  add_17861 = None
	        clamp_max_233: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_350, 127);  clamp_min_350 = None
	        view_1822: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_233, [sym_size_int, 1500, 1280]);  clamp_max_233 = None
	        _assert_tensor_metadata_1048 = torch.ops.aten._assert_tensor_metadata.default(view_1822, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1048 = None
	        convert_element_type_697: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1822, torch.int8);  view_1822 = None
	        view_1823: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_697, [sym_size_int, 1500, 1280]);  convert_element_type_697 = None
	        view_1824: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_348, [sym_size_int, 1500, 1]);  clamp_min_348 = None
	        view_1825: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_696, [sym_size_int, 1500, 1]);  convert_element_type_696 = None
	        _assert_tensor_metadata_1049 = torch.ops.aten._assert_tensor_metadata.default(view_1823, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1049 = None
	        convert_element_type_698: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1823, torch.float32);  view_1823 = None
	        _assert_tensor_metadata_1050 = torch.ops.aten._assert_tensor_metadata.default(view_1825, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1050 = None
	        convert_element_type_699: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1825, torch.float32);  view_1825 = None
	        sub_5339: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_698, convert_element_type_699);  convert_element_type_698 = convert_element_type_699 = None
	        mul_11309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5339, view_1824);  sub_5339 = view_1824 = None
	        view_1826: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11309, [sym_size_int, 1500, 1280]);  mul_11309 = None
	        _assert_tensor_metadata_1051 = torch.ops.aten._assert_tensor_metadata.default(view_1826, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1051 = None
	        view_1827: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg528_1, [1280, 40, 32]);  arg528_1 = None
	        view_1828: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg529_1, [1280, 40, 1]);  arg529_1 = None
	        view_1829: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg530_1, [1280, 40, 1]);  arg530_1 = None
	        _assert_tensor_metadata_1052 = torch.ops.aten._assert_tensor_metadata.default(view_1827, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1052 = None
	        convert_element_type_700: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1827, torch.float32);  view_1827 = None
	        _assert_tensor_metadata_1053 = torch.ops.aten._assert_tensor_metadata.default(view_1829, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1053 = None
	        convert_element_type_701: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1829, torch.float32);  view_1829 = None
	        sub_5343: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_700, convert_element_type_701);  convert_element_type_700 = convert_element_type_701 = None
	        mul_11314: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5343, view_1828);  sub_5343 = view_1828 = None
	        view_1830: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11314, [1280, 1280]);  mul_11314 = None
	        _assert_tensor_metadata_1054 = torch.ops.aten._assert_tensor_metadata.default(view_1830, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1054 = None
	        mul_11319: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1831: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1826, [mul_11319, 1280]);  view_1826 = mul_11319 = None
	        permute_195: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1830, [1, 0]);  view_1830 = None
	        addmm_96: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg527_1, view_1831, permute_195);  arg527_1 = view_1831 = permute_195 = None
	        view_1832: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_96, [sym_size_int, 1500, 1280]);  addmm_96 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1833: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1832, [sym_size_int, -1, 20, 64]);  view_1832 = None
	        permute_196: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1833, [0, 2, 1, 3]);  view_1833 = None
	        clone_156: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_196, memory_format = torch.contiguous_format);  permute_196 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_19 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_154, clone_155, clone_156, None, False, scale = 1.0);  clone_154 = clone_155 = clone_156 = None
	        getitem_154: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_19[0];  _scaled_dot_product_efficient_attention_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_197: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_154, [0, 2, 1, 3]);  getitem_154 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_197, [sym_size_int, 1500, -1]);  permute_197 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1835: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1834, [sym_size_int, 1500, 1280])
	        amin_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1835, [2])
	        amax_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1835, [2]);  view_1835 = None
	        full_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_117, full_234);  amin_117 = full_234 = None
	        full_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_117, full_235);  amax_117 = full_235 = None
	        sub_5361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_117, minimum_117);  maximum_117 = None
	        div_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5361, 255.0);  sub_5361 = None
	        clamp_min_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_234, 1.1920928955078125e-07);  div_234 = None
	        div_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_117, clamp_min_351);  minimum_117 = None
	        round_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_235);  div_235 = None
	        sub_5367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_235);  round_235 = None
	        clamp_min_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5367, -128);  sub_5367 = None
	        clamp_max_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_352, 127);  clamp_min_352 = None
	        _assert_tensor_metadata_1055 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_351, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1055 = None
	        _assert_tensor_metadata_1056 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_234, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1056 = None
	        convert_element_type_702: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_234, torch.int8);  clamp_max_234 = None
	        view_1836: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1834, [sym_size_int, 1500, 1280]);  view_1834 = None
	        view_1837: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_351, [sym_size_int, 1500, 1])
	        view_1838: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_702, [sym_size_int, 1500, 1])
	        reciprocal_117: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1837);  view_1837 = None
	        mul_11389: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_117, 1.0);  reciprocal_117 = None
	        mul_11392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1836, mul_11389);  view_1836 = mul_11389 = None
	        round_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11392);  mul_11392 = None
	        add_18025: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_236, view_1838);  round_236 = view_1838 = None
	        clamp_min_353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18025, -128);  add_18025 = None
	        clamp_max_235: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_353, 127);  clamp_min_353 = None
	        view_1839: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_235, [sym_size_int, 1500, 1280]);  clamp_max_235 = None
	        _assert_tensor_metadata_1057 = torch.ops.aten._assert_tensor_metadata.default(view_1839, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1057 = None
	        convert_element_type_703: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1839, torch.int8);  view_1839 = None
	        view_1840: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_703, [sym_size_int, 1500, 1280]);  convert_element_type_703 = None
	        view_1841: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_351, [sym_size_int, 1500, 1]);  clamp_min_351 = None
	        view_1842: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_702, [sym_size_int, 1500, 1]);  convert_element_type_702 = None
	        _assert_tensor_metadata_1058 = torch.ops.aten._assert_tensor_metadata.default(view_1840, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1058 = None
	        convert_element_type_704: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1840, torch.float32);  view_1840 = None
	        _assert_tensor_metadata_1059 = torch.ops.aten._assert_tensor_metadata.default(view_1842, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1059 = None
	        convert_element_type_705: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1842, torch.float32);  view_1842 = None
	        sub_5387: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_704, convert_element_type_705);  convert_element_type_704 = convert_element_type_705 = None
	        mul_11414: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5387, view_1841);  sub_5387 = view_1841 = None
	        view_1843: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11414, [sym_size_int, 1500, 1280]);  mul_11414 = None
	        _assert_tensor_metadata_1060 = torch.ops.aten._assert_tensor_metadata.default(view_1843, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1060 = None
	        view_1844: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg532_1, [1280, 40, 32]);  arg532_1 = None
	        view_1845: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg533_1, [1280, 40, 1]);  arg533_1 = None
	        view_1846: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg534_1, [1280, 40, 1]);  arg534_1 = None
	        _assert_tensor_metadata_1061 = torch.ops.aten._assert_tensor_metadata.default(view_1844, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1061 = None
	        convert_element_type_706: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1844, torch.float32);  view_1844 = None
	        _assert_tensor_metadata_1062 = torch.ops.aten._assert_tensor_metadata.default(view_1846, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1062 = None
	        convert_element_type_707: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1846, torch.float32);  view_1846 = None
	        sub_5391: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_706, convert_element_type_707);  convert_element_type_706 = convert_element_type_707 = None
	        mul_11419: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5391, view_1845);  sub_5391 = view_1845 = None
	        view_1847: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11419, [1280, 1280]);  mul_11419 = None
	        _assert_tensor_metadata_1063 = torch.ops.aten._assert_tensor_metadata.default(view_1847, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1063 = None
	        mul_11424: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1848: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1843, [mul_11424, 1280]);  view_1843 = mul_11424 = None
	        permute_198: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1847, [1, 0]);  view_1847 = None
	        addmm_97: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg531_1, view_1848, permute_198);  arg531_1 = view_1848 = permute_198 = None
	        view_1849: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_97, [sym_size_int, 1500, 1280]);  addmm_97 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_157: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1849);  view_1849 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_18088: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_17468, clone_157);  add_17468 = clone_157 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_18088, memory_format = torch.contiguous_format)
	        var_mean_39 = torch.ops.aten.var_mean.correction(clone_158, [2], correction = 0, keepdim = True)
	        getitem_158: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_39[0]
	        getitem_159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_39[1];  var_mean_39 = None
	        add_18093: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_158, 1e-05);  getitem_158 = None
	        rsqrt_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_18093);  add_18093 = None
	        sub_5397: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_158, getitem_159);  clone_158 = getitem_159 = None
	        mul_11435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5397, rsqrt_39);  sub_5397 = rsqrt_39 = None
	        mul_11436: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11435, arg535_1);  mul_11435 = arg535_1 = None
	        add_18094: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_11436, arg536_1);  mul_11436 = arg536_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1850: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_18094, [sym_size_int, 1500, 1280])
	        amin_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1850, [2])
	        amax_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1850, [2]);  view_1850 = None
	        full_236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_118, full_236);  amin_118 = full_236 = None
	        full_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_118, full_237);  amax_118 = full_237 = None
	        sub_5408: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_118, minimum_118);  maximum_118 = None
	        div_236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5408, 255.0);  sub_5408 = None
	        clamp_min_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_236, 1.1920928955078125e-07);  div_236 = None
	        div_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_118, clamp_min_354);  minimum_118 = None
	        round_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_237);  div_237 = None
	        sub_5414: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_237);  round_237 = None
	        clamp_min_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5414, -128);  sub_5414 = None
	        clamp_max_236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_355, 127);  clamp_min_355 = None
	        _assert_tensor_metadata_1064 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_354, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1064 = None
	        _assert_tensor_metadata_1065 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_236, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1065 = None
	        convert_element_type_708: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_236, torch.int8);  clamp_max_236 = None
	        view_1851: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_18094, [sym_size_int, 1500, 1280]);  add_18094 = None
	        view_1852: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_354, [sym_size_int, 1500, 1])
	        view_1853: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_708, [sym_size_int, 1500, 1])
	        reciprocal_118: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1852);  view_1852 = None
	        mul_11484: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_118, 1.0);  reciprocal_118 = None
	        mul_11487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1851, mul_11484);  view_1851 = mul_11484 = None
	        round_238: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11487);  mul_11487 = None
	        add_18181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_238, view_1853);  round_238 = view_1853 = None
	        clamp_min_356: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18181, -128);  add_18181 = None
	        clamp_max_237: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_356, 127);  clamp_min_356 = None
	        view_1854: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_237, [sym_size_int, 1500, 1280]);  clamp_max_237 = None
	        _assert_tensor_metadata_1066 = torch.ops.aten._assert_tensor_metadata.default(view_1854, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1066 = None
	        convert_element_type_709: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1854, torch.int8);  view_1854 = None
	        view_1855: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_709, [sym_size_int, 1500, 1280]);  convert_element_type_709 = None
	        view_1856: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_354, [sym_size_int, 1500, 1]);  clamp_min_354 = None
	        view_1857: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_708, [sym_size_int, 1500, 1]);  convert_element_type_708 = None
	        _assert_tensor_metadata_1067 = torch.ops.aten._assert_tensor_metadata.default(view_1855, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1067 = None
	        convert_element_type_710: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1855, torch.float32);  view_1855 = None
	        _assert_tensor_metadata_1068 = torch.ops.aten._assert_tensor_metadata.default(view_1857, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1068 = None
	        convert_element_type_711: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1857, torch.float32);  view_1857 = None
	        sub_5434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_710, convert_element_type_711);  convert_element_type_710 = convert_element_type_711 = None
	        mul_11509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5434, view_1856);  sub_5434 = view_1856 = None
	        view_1858: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11509, [sym_size_int, 1500, 1280]);  mul_11509 = None
	        _assert_tensor_metadata_1069 = torch.ops.aten._assert_tensor_metadata.default(view_1858, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1069 = None
	        view_1859: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg538_1, [5120, 40, 32]);  arg538_1 = None
	        view_1860: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg539_1, [5120, 40, 1]);  arg539_1 = None
	        view_1861: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg540_1, [5120, 40, 1]);  arg540_1 = None
	        _assert_tensor_metadata_1070 = torch.ops.aten._assert_tensor_metadata.default(view_1859, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1070 = None
	        convert_element_type_712: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1859, torch.float32);  view_1859 = None
	        _assert_tensor_metadata_1071 = torch.ops.aten._assert_tensor_metadata.default(view_1861, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1071 = None
	        convert_element_type_713: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1861, torch.float32);  view_1861 = None
	        sub_5438: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_712, convert_element_type_713);  convert_element_type_712 = convert_element_type_713 = None
	        mul_11514: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5438, view_1860);  sub_5438 = view_1860 = None
	        view_1862: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11514, [5120, 1280]);  mul_11514 = None
	        _assert_tensor_metadata_1072 = torch.ops.aten._assert_tensor_metadata.default(view_1862, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1072 = None
	        mul_11519: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1863: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1858, [mul_11519, 1280]);  view_1858 = mul_11519 = None
	        permute_199: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1862, [1, 0]);  view_1862 = None
	        addmm_98: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg537_1, view_1863, permute_199);  arg537_1 = view_1863 = permute_199 = None
	        view_1864: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_98, [sym_size_int, 1500, 5120]);  addmm_98 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_11526: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1864, 0.5)
	        mul_11527: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1864, 0.7071067811865476);  view_1864 = None
	        erf_21: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_11527);  mul_11527 = None
	        add_18240: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_21, 1);  erf_21 = None
	        mul_11528: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11526, add_18240);  mul_11526 = add_18240 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_159: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_11528);  mul_11528 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1865: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_159, [sym_size_int, 1500, 5120])
	        amin_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1865, [2])
	        amax_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1865, [2]);  view_1865 = None
	        full_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_119, full_238);  amin_119 = full_238 = None
	        full_239: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_119, full_239);  amax_119 = full_239 = None
	        sub_5451: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_119, minimum_119);  maximum_119 = None
	        div_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5451, 255.0);  sub_5451 = None
	        clamp_min_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_238, 1.1920928955078125e-07);  div_238 = None
	        div_239: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_119, clamp_min_357);  minimum_119 = None
	        round_239: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_239);  div_239 = None
	        sub_5457: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_239);  round_239 = None
	        clamp_min_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5457, -128);  sub_5457 = None
	        clamp_max_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_358, 127);  clamp_min_358 = None
	        _assert_tensor_metadata_1073 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_357, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1073 = None
	        _assert_tensor_metadata_1074 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_238, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1074 = None
	        convert_element_type_714: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_238, torch.int8);  clamp_max_238 = None
	        view_1866: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_159, [sym_size_int, 1500, 5120]);  clone_159 = None
	        view_1867: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_357, [sym_size_int, 1500, 1])
	        view_1868: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_714, [sym_size_int, 1500, 1])
	        reciprocal_119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1867);  view_1867 = None
	        mul_11574: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_119, 1.0);  reciprocal_119 = None
	        mul_11577: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1866, mul_11574);  view_1866 = mul_11574 = None
	        round_240: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_11577);  mul_11577 = None
	        add_18323: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_240, view_1868);  round_240 = view_1868 = None
	        clamp_min_359: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18323, -128);  add_18323 = None
	        clamp_max_239: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_359, 127);  clamp_min_359 = None
	        view_1869: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_239, [sym_size_int, 1500, 5120]);  clamp_max_239 = None
	        _assert_tensor_metadata_1075 = torch.ops.aten._assert_tensor_metadata.default(view_1869, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1075 = None
	        convert_element_type_715: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1869, torch.int8);  view_1869 = None
	        view_1870: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_715, [sym_size_int, 1500, 5120]);  convert_element_type_715 = None
	        view_1871: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_357, [sym_size_int, 1500, 1]);  clamp_min_357 = None
	        view_1872: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_714, [sym_size_int, 1500, 1]);  convert_element_type_714 = None
	        _assert_tensor_metadata_1076 = torch.ops.aten._assert_tensor_metadata.default(view_1870, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1076 = None
	        convert_element_type_716: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1870, torch.float32);  view_1870 = None
	        _assert_tensor_metadata_1077 = torch.ops.aten._assert_tensor_metadata.default(view_1872, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1077 = None
	        convert_element_type_717: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1872, torch.float32);  view_1872 = None
	        sub_5477: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_716, convert_element_type_717);  convert_element_type_716 = convert_element_type_717 = None
	        mul_11599: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5477, view_1871);  sub_5477 = view_1871 = None
	        view_1873: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_11599, [sym_size_int, 1500, 5120]);  mul_11599 = None
	        _assert_tensor_metadata_1078 = torch.ops.aten._assert_tensor_metadata.default(view_1873, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1078 = None
	        view_1874: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg542_1, [1280, 160, 32]);  arg542_1 = None
	        view_1875: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg543_1, [1280, 160, 1]);  arg543_1 = None
	        view_1876: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg544_1, [1280, 160, 1]);  arg544_1 = None
	        _assert_tensor_metadata_1079 = torch.ops.aten._assert_tensor_metadata.default(view_1874, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1079 = None
	        convert_element_type_718: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1874, torch.float32);  view_1874 = None
	        _assert_tensor_metadata_1080 = torch.ops.aten._assert_tensor_metadata.default(view_1876, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1080 = None
	        convert_element_type_719: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1876, torch.float32);  view_1876 = None
	        sub_5481: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_718, convert_element_type_719);  convert_element_type_718 = convert_element_type_719 = None
	        mul_11604: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5481, view_1875);  sub_5481 = view_1875 = None
	        view_1877: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_11604, [1280, 5120]);  mul_11604 = None
	        _assert_tensor_metadata_1081 = torch.ops.aten._assert_tensor_metadata.default(view_1877, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1081 = None
	        mul_11609: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1878: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1873, [mul_11609, 5120]);  view_1873 = mul_11609 = None
	        permute_200: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1877, [1, 0]);  view_1877 = None
	        addmm_99: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg541_1, view_1878, permute_200);  arg541_1 = view_1878 = permute_200 = None
	        view_1879: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_99, [sym_size_int, 1500, 1280]);  addmm_99 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1879);  view_1879 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_18386: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_18088, clone_160);  add_18088 = clone_160 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_18386, memory_format = torch.contiguous_format)
	        var_mean_40 = torch.ops.aten.var_mean.correction(clone_161, [2], correction = 0, keepdim = True)
	        getitem_160: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_40[0]
	        getitem_161: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_40[1];  var_mean_40 = None
	        add_18391: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_160, 1e-05);  getitem_160 = None
	        rsqrt_40: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_18391);  add_18391 = None
	        sub_5487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_161, getitem_161);  clone_161 = getitem_161 = None
	        mul_11620: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5487, rsqrt_40);  sub_5487 = rsqrt_40 = None
	        mul_11621: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11620, arg545_1);  mul_11620 = arg545_1 = None
	        add_18392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_11621, arg546_1);  mul_11621 = arg546_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1880: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_18392, [sym_size_int, 1500, 1280])
	        amin_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1880, [2])
	        amax_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1880, [2]);  view_1880 = None
	        full_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_120, full_240);  amin_120 = full_240 = None
	        full_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_120, full_241);  amax_120 = full_241 = None
	        sub_5498: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_120, minimum_120);  maximum_120 = None
	        div_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5498, 255.0);  sub_5498 = None
	        clamp_min_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_240, 1.1920928955078125e-07);  div_240 = None
	        div_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_120, clamp_min_360);  minimum_120 = None
	        round_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_241);  div_241 = None
	        sub_5504: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_241);  round_241 = None
	        clamp_min_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5504, -128);  sub_5504 = None
	        clamp_max_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_361, 127);  clamp_min_361 = None
	        _assert_tensor_metadata_1082 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_360, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1082 = None
	        _assert_tensor_metadata_1083 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_240, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1083 = None
	        convert_element_type_720: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_240, torch.int8);  clamp_max_240 = None
	        view_1881: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_18392, [sym_size_int, 1500, 1280])
	        view_1882: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_360, [sym_size_int, 1500, 1])
	        view_1883: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_720, [sym_size_int, 1500, 1])
	        reciprocal_120: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1882);  view_1882 = None
	        mul_11669: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_120, 1.0);  reciprocal_120 = None
	        mul_11672: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1881, mul_11669);  view_1881 = mul_11669 = None
	        round_242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11672);  mul_11672 = None
	        add_18479: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_242, view_1883);  round_242 = view_1883 = None
	        clamp_min_362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18479, -128);  add_18479 = None
	        clamp_max_241: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_362, 127);  clamp_min_362 = None
	        view_1884: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_241, [sym_size_int, 1500, 1280]);  clamp_max_241 = None
	        _assert_tensor_metadata_1084 = torch.ops.aten._assert_tensor_metadata.default(view_1884, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1084 = None
	        convert_element_type_721: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1884, torch.int8);  view_1884 = None
	        view_1885: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_721, [sym_size_int, 1500, 1280]);  convert_element_type_721 = None
	        view_1886: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_360, [sym_size_int, 1500, 1]);  clamp_min_360 = None
	        view_1887: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_720, [sym_size_int, 1500, 1]);  convert_element_type_720 = None
	        _assert_tensor_metadata_1085 = torch.ops.aten._assert_tensor_metadata.default(view_1885, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1085 = None
	        convert_element_type_722: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1885, torch.float32);  view_1885 = None
	        _assert_tensor_metadata_1086 = torch.ops.aten._assert_tensor_metadata.default(view_1887, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1086 = None
	        convert_element_type_723: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1887, torch.float32);  view_1887 = None
	        sub_5524: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_722, convert_element_type_723);  convert_element_type_722 = convert_element_type_723 = None
	        mul_11694: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5524, view_1886);  sub_5524 = view_1886 = None
	        view_1888: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11694, [sym_size_int, 1500, 1280]);  mul_11694 = None
	        _assert_tensor_metadata_1087 = torch.ops.aten._assert_tensor_metadata.default(view_1888, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1087 = None
	        view_1889: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg548_1, [1280, 40, 32]);  arg548_1 = None
	        view_1890: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg549_1, [1280, 40, 1]);  arg549_1 = None
	        view_1891: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg550_1, [1280, 40, 1]);  arg550_1 = None
	        _assert_tensor_metadata_1088 = torch.ops.aten._assert_tensor_metadata.default(view_1889, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1088 = None
	        convert_element_type_724: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1889, torch.float32);  view_1889 = None
	        _assert_tensor_metadata_1089 = torch.ops.aten._assert_tensor_metadata.default(view_1891, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1089 = None
	        convert_element_type_725: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1891, torch.float32);  view_1891 = None
	        sub_5528: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_724, convert_element_type_725);  convert_element_type_724 = convert_element_type_725 = None
	        mul_11699: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5528, view_1890);  sub_5528 = view_1890 = None
	        view_1892: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11699, [1280, 1280]);  mul_11699 = None
	        _assert_tensor_metadata_1090 = torch.ops.aten._assert_tensor_metadata.default(view_1892, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1090 = None
	        mul_11704: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1893: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1888, [mul_11704, 1280]);  view_1888 = mul_11704 = None
	        permute_201: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1892, [1, 0]);  view_1892 = None
	        addmm_100: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg547_1, view_1893, permute_201);  arg547_1 = view_1893 = permute_201 = None
	        view_1894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_100, [sym_size_int, 1500, 1280]);  addmm_100 = None
	        mul_11711: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1894, 0.125);  view_1894 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1895: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_11711, [sym_size_int, 1500, 20, 64]);  mul_11711 = None
	        permute_202: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1895, [0, 2, 1, 3]);  view_1895 = None
	        clone_162: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_202, memory_format = torch.contiguous_format);  permute_202 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1896: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_18392, [sym_size_int, 1500, 1280])
	        amin_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1896, [2])
	        amax_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1896, [2]);  view_1896 = None
	        full_242: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_121, full_242);  amin_121 = full_242 = None
	        full_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_121, full_243);  amax_121 = full_243 = None
	        sub_5543: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_121, minimum_121);  maximum_121 = None
	        div_242: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5543, 255.0);  sub_5543 = None
	        clamp_min_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_242, 1.1920928955078125e-07);  div_242 = None
	        div_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_121, clamp_min_363);  minimum_121 = None
	        round_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_243);  div_243 = None
	        sub_5549: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_243);  round_243 = None
	        clamp_min_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5549, -128);  sub_5549 = None
	        clamp_max_242: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_364, 127);  clamp_min_364 = None
	        _assert_tensor_metadata_1091 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_363, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1091 = None
	        _assert_tensor_metadata_1092 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_242, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1092 = None
	        convert_element_type_726: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_242, torch.int8);  clamp_max_242 = None
	        view_1897: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_18392, [sym_size_int, 1500, 1280])
	        view_1898: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_363, [sym_size_int, 1500, 1])
	        view_1899: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_726, [sym_size_int, 1500, 1])
	        reciprocal_121: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1898);  view_1898 = None
	        mul_11765: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_121, 1.0);  reciprocal_121 = None
	        mul_11768: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1897, mul_11765);  view_1897 = mul_11765 = None
	        round_244: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11768);  mul_11768 = None
	        add_18631: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_244, view_1899);  round_244 = view_1899 = None
	        clamp_min_365: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18631, -128);  add_18631 = None
	        clamp_max_243: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_365, 127);  clamp_min_365 = None
	        view_1900: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_243, [sym_size_int, 1500, 1280]);  clamp_max_243 = None
	        _assert_tensor_metadata_1093 = torch.ops.aten._assert_tensor_metadata.default(view_1900, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1093 = None
	        convert_element_type_727: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1900, torch.int8);  view_1900 = None
	        view_1901: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_727, [sym_size_int, 1500, 1280]);  convert_element_type_727 = None
	        view_1902: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_363, [sym_size_int, 1500, 1]);  clamp_min_363 = None
	        view_1903: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_726, [sym_size_int, 1500, 1]);  convert_element_type_726 = None
	        _assert_tensor_metadata_1094 = torch.ops.aten._assert_tensor_metadata.default(view_1901, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1094 = None
	        convert_element_type_728: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1901, torch.float32);  view_1901 = None
	        _assert_tensor_metadata_1095 = torch.ops.aten._assert_tensor_metadata.default(view_1903, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1095 = None
	        convert_element_type_729: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1903, torch.float32);  view_1903 = None
	        sub_5569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_728, convert_element_type_729);  convert_element_type_728 = convert_element_type_729 = None
	        mul_11790: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5569, view_1902);  sub_5569 = view_1902 = None
	        view_1904: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11790, [sym_size_int, 1500, 1280]);  mul_11790 = None
	        _assert_tensor_metadata_1096 = torch.ops.aten._assert_tensor_metadata.default(view_1904, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1096 = None
	        view_1905: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg551_1, [1280, 40, 32]);  arg551_1 = None
	        view_1906: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg552_1, [1280, 40, 1]);  arg552_1 = None
	        view_1907: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg553_1, [1280, 40, 1]);  arg553_1 = None
	        _assert_tensor_metadata_1097 = torch.ops.aten._assert_tensor_metadata.default(view_1905, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1097 = None
	        convert_element_type_730: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1905, torch.float32);  view_1905 = None
	        _assert_tensor_metadata_1098 = torch.ops.aten._assert_tensor_metadata.default(view_1907, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1098 = None
	        convert_element_type_731: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1907, torch.float32);  view_1907 = None
	        sub_5573: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_730, convert_element_type_731);  convert_element_type_730 = convert_element_type_731 = None
	        mul_11795: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5573, view_1906);  sub_5573 = view_1906 = None
	        view_1908: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11795, [1280, 1280]);  mul_11795 = None
	        _assert_tensor_metadata_1099 = torch.ops.aten._assert_tensor_metadata.default(view_1908, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1099 = None
	        permute_203: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1908, [1, 0]);  view_1908 = None
	        mul_11798: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1909: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1904, [mul_11798, 1280]);  view_1904 = mul_11798 = None
	        mm_20: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1909, permute_203);  view_1909 = permute_203 = None
	        view_1910: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_20, [sym_size_int, 1500, 1280]);  mm_20 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1911: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1910, [sym_size_int, -1, 20, 64]);  view_1910 = None
	        permute_204: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1911, [0, 2, 1, 3]);  view_1911 = None
	        clone_163: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_204, memory_format = torch.contiguous_format);  permute_204 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1912: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_18392, [sym_size_int, 1500, 1280])
	        amin_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1912, [2])
	        amax_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1912, [2]);  view_1912 = None
	        full_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_122, full_244);  amin_122 = full_244 = None
	        full_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_122, full_245);  amax_122 = full_245 = None
	        sub_5587: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_122, minimum_122);  maximum_122 = None
	        div_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5587, 255.0);  sub_5587 = None
	        clamp_min_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_244, 1.1920928955078125e-07);  div_244 = None
	        div_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_122, clamp_min_366);  minimum_122 = None
	        round_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_245);  div_245 = None
	        sub_5593: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_245);  round_245 = None
	        clamp_min_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5593, -128);  sub_5593 = None
	        clamp_max_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_367, 127);  clamp_min_367 = None
	        _assert_tensor_metadata_1100 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_366, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1100 = None
	        _assert_tensor_metadata_1101 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_244, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1101 = None
	        convert_element_type_732: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_244, torch.int8);  clamp_max_244 = None
	        view_1913: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_18392, [sym_size_int, 1500, 1280]);  add_18392 = None
	        view_1914: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_366, [sym_size_int, 1500, 1])
	        view_1915: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_732, [sym_size_int, 1500, 1])
	        reciprocal_122: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1914);  view_1914 = None
	        mul_11864: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_122, 1.0);  reciprocal_122 = None
	        mul_11867: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1913, mul_11864);  view_1913 = mul_11864 = None
	        round_246: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11867);  mul_11867 = None
	        add_18779: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_246, view_1915);  round_246 = view_1915 = None
	        clamp_min_368: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18779, -128);  add_18779 = None
	        clamp_max_245: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_368, 127);  clamp_min_368 = None
	        view_1916: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_245, [sym_size_int, 1500, 1280]);  clamp_max_245 = None
	        _assert_tensor_metadata_1102 = torch.ops.aten._assert_tensor_metadata.default(view_1916, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1102 = None
	        convert_element_type_733: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1916, torch.int8);  view_1916 = None
	        view_1917: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_733, [sym_size_int, 1500, 1280]);  convert_element_type_733 = None
	        view_1918: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_366, [sym_size_int, 1500, 1]);  clamp_min_366 = None
	        view_1919: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_732, [sym_size_int, 1500, 1]);  convert_element_type_732 = None
	        _assert_tensor_metadata_1103 = torch.ops.aten._assert_tensor_metadata.default(view_1917, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1103 = None
	        convert_element_type_734: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1917, torch.float32);  view_1917 = None
	        _assert_tensor_metadata_1104 = torch.ops.aten._assert_tensor_metadata.default(view_1919, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1104 = None
	        convert_element_type_735: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1919, torch.float32);  view_1919 = None
	        sub_5613: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_734, convert_element_type_735);  convert_element_type_734 = convert_element_type_735 = None
	        mul_11889: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5613, view_1918);  sub_5613 = view_1918 = None
	        view_1920: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11889, [sym_size_int, 1500, 1280]);  mul_11889 = None
	        _assert_tensor_metadata_1105 = torch.ops.aten._assert_tensor_metadata.default(view_1920, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1105 = None
	        view_1921: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg555_1, [1280, 40, 32]);  arg555_1 = None
	        view_1922: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg556_1, [1280, 40, 1]);  arg556_1 = None
	        view_1923: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg557_1, [1280, 40, 1]);  arg557_1 = None
	        _assert_tensor_metadata_1106 = torch.ops.aten._assert_tensor_metadata.default(view_1921, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1106 = None
	        convert_element_type_736: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1921, torch.float32);  view_1921 = None
	        _assert_tensor_metadata_1107 = torch.ops.aten._assert_tensor_metadata.default(view_1923, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1107 = None
	        convert_element_type_737: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1923, torch.float32);  view_1923 = None
	        sub_5617: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_736, convert_element_type_737);  convert_element_type_736 = convert_element_type_737 = None
	        mul_11894: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5617, view_1922);  sub_5617 = view_1922 = None
	        view_1924: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11894, [1280, 1280]);  mul_11894 = None
	        _assert_tensor_metadata_1108 = torch.ops.aten._assert_tensor_metadata.default(view_1924, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1108 = None
	        mul_11899: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1925: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1920, [mul_11899, 1280]);  view_1920 = mul_11899 = None
	        permute_205: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1924, [1, 0]);  view_1924 = None
	        addmm_101: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg554_1, view_1925, permute_205);  arg554_1 = view_1925 = permute_205 = None
	        view_1926: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_101, [sym_size_int, 1500, 1280]);  addmm_101 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1927: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1926, [sym_size_int, -1, 20, 64]);  view_1926 = None
	        permute_206: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1927, [0, 2, 1, 3]);  view_1927 = None
	        clone_164: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_206, memory_format = torch.contiguous_format);  permute_206 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_20 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_162, clone_163, clone_164, None, False, scale = 1.0);  clone_162 = clone_163 = clone_164 = None
	        getitem_162: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_20[0];  _scaled_dot_product_efficient_attention_20 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_207: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_162, [0, 2, 1, 3]);  getitem_162 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1928: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_207, [sym_size_int, 1500, -1]);  permute_207 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1929: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1928, [sym_size_int, 1500, 1280])
	        amin_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1929, [2])
	        amax_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1929, [2]);  view_1929 = None
	        full_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_123, full_246);  amin_123 = full_246 = None
	        full_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_123, full_247);  amax_123 = full_247 = None
	        sub_5635: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_123, minimum_123);  maximum_123 = None
	        div_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5635, 255.0);  sub_5635 = None
	        clamp_min_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_246, 1.1920928955078125e-07);  div_246 = None
	        div_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_123, clamp_min_369);  minimum_123 = None
	        round_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_247);  div_247 = None
	        sub_5641: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_247);  round_247 = None
	        clamp_min_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5641, -128);  sub_5641 = None
	        clamp_max_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_370, 127);  clamp_min_370 = None
	        _assert_tensor_metadata_1109 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_369, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1109 = None
	        _assert_tensor_metadata_1110 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_246, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1110 = None
	        convert_element_type_738: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_246, torch.int8);  clamp_max_246 = None
	        view_1930: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1928, [sym_size_int, 1500, 1280]);  view_1928 = None
	        view_1931: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_369, [sym_size_int, 1500, 1])
	        view_1932: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_738, [sym_size_int, 1500, 1])
	        reciprocal_123: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1931);  view_1931 = None
	        mul_11969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_123, 1.0);  reciprocal_123 = None
	        mul_11972: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1930, mul_11969);  view_1930 = mul_11969 = None
	        round_248: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11972);  mul_11972 = None
	        add_18943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_248, view_1932);  round_248 = view_1932 = None
	        clamp_min_371: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18943, -128);  add_18943 = None
	        clamp_max_247: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_371, 127);  clamp_min_371 = None
	        view_1933: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_247, [sym_size_int, 1500, 1280]);  clamp_max_247 = None
	        _assert_tensor_metadata_1111 = torch.ops.aten._assert_tensor_metadata.default(view_1933, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1111 = None
	        convert_element_type_739: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1933, torch.int8);  view_1933 = None
	        view_1934: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_739, [sym_size_int, 1500, 1280]);  convert_element_type_739 = None
	        view_1935: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_369, [sym_size_int, 1500, 1]);  clamp_min_369 = None
	        view_1936: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_738, [sym_size_int, 1500, 1]);  convert_element_type_738 = None
	        _assert_tensor_metadata_1112 = torch.ops.aten._assert_tensor_metadata.default(view_1934, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1112 = None
	        convert_element_type_740: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1934, torch.float32);  view_1934 = None
	        _assert_tensor_metadata_1113 = torch.ops.aten._assert_tensor_metadata.default(view_1936, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1113 = None
	        convert_element_type_741: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1936, torch.float32);  view_1936 = None
	        sub_5661: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_740, convert_element_type_741);  convert_element_type_740 = convert_element_type_741 = None
	        mul_11994: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5661, view_1935);  sub_5661 = view_1935 = None
	        view_1937: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11994, [sym_size_int, 1500, 1280]);  mul_11994 = None
	        _assert_tensor_metadata_1114 = torch.ops.aten._assert_tensor_metadata.default(view_1937, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1114 = None
	        view_1938: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg559_1, [1280, 40, 32]);  arg559_1 = None
	        view_1939: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg560_1, [1280, 40, 1]);  arg560_1 = None
	        view_1940: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg561_1, [1280, 40, 1]);  arg561_1 = None
	        _assert_tensor_metadata_1115 = torch.ops.aten._assert_tensor_metadata.default(view_1938, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1115 = None
	        convert_element_type_742: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1938, torch.float32);  view_1938 = None
	        _assert_tensor_metadata_1116 = torch.ops.aten._assert_tensor_metadata.default(view_1940, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1116 = None
	        convert_element_type_743: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1940, torch.float32);  view_1940 = None
	        sub_5665: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_742, convert_element_type_743);  convert_element_type_742 = convert_element_type_743 = None
	        mul_11999: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5665, view_1939);  sub_5665 = view_1939 = None
	        view_1941: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11999, [1280, 1280]);  mul_11999 = None
	        _assert_tensor_metadata_1117 = torch.ops.aten._assert_tensor_metadata.default(view_1941, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1117 = None
	        mul_12004: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1942: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1937, [mul_12004, 1280]);  view_1937 = mul_12004 = None
	        permute_208: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1941, [1, 0]);  view_1941 = None
	        addmm_102: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg558_1, view_1942, permute_208);  arg558_1 = view_1942 = permute_208 = None
	        view_1943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_102, [sym_size_int, 1500, 1280]);  addmm_102 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_165: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1943);  view_1943 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_19006: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_18386, clone_165);  add_18386 = clone_165 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_166: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_19006, memory_format = torch.contiguous_format)
	        var_mean_41 = torch.ops.aten.var_mean.correction(clone_166, [2], correction = 0, keepdim = True)
	        getitem_166: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_41[0]
	        getitem_167: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_41[1];  var_mean_41 = None
	        add_19011: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_166, 1e-05);  getitem_166 = None
	        rsqrt_41: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_19011);  add_19011 = None
	        sub_5671: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_166, getitem_167);  clone_166 = getitem_167 = None
	        mul_12015: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5671, rsqrt_41);  sub_5671 = rsqrt_41 = None
	        mul_12016: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12015, arg562_1);  mul_12015 = arg562_1 = None
	        add_19012: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12016, arg563_1);  mul_12016 = arg563_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1944: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19012, [sym_size_int, 1500, 1280])
	        amin_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1944, [2])
	        amax_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1944, [2]);  view_1944 = None
	        full_248: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_124, full_248);  amin_124 = full_248 = None
	        full_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_124, full_249);  amax_124 = full_249 = None
	        sub_5682: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_124, minimum_124);  maximum_124 = None
	        div_248: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5682, 255.0);  sub_5682 = None
	        clamp_min_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_248, 1.1920928955078125e-07);  div_248 = None
	        div_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_124, clamp_min_372);  minimum_124 = None
	        round_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_249);  div_249 = None
	        sub_5688: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_249);  round_249 = None
	        clamp_min_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5688, -128);  sub_5688 = None
	        clamp_max_248: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_373, 127);  clamp_min_373 = None
	        _assert_tensor_metadata_1118 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_372, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1118 = None
	        _assert_tensor_metadata_1119 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_248, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1119 = None
	        convert_element_type_744: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_248, torch.int8);  clamp_max_248 = None
	        view_1945: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19012, [sym_size_int, 1500, 1280]);  add_19012 = None
	        view_1946: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_372, [sym_size_int, 1500, 1])
	        view_1947: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_744, [sym_size_int, 1500, 1])
	        reciprocal_124: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1946);  view_1946 = None
	        mul_12064: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_124, 1.0);  reciprocal_124 = None
	        mul_12067: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1945, mul_12064);  view_1945 = mul_12064 = None
	        round_250: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12067);  mul_12067 = None
	        add_19099: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_250, view_1947);  round_250 = view_1947 = None
	        clamp_min_374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19099, -128);  add_19099 = None
	        clamp_max_249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_374, 127);  clamp_min_374 = None
	        view_1948: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_249, [sym_size_int, 1500, 1280]);  clamp_max_249 = None
	        _assert_tensor_metadata_1120 = torch.ops.aten._assert_tensor_metadata.default(view_1948, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1120 = None
	        convert_element_type_745: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1948, torch.int8);  view_1948 = None
	        view_1949: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_745, [sym_size_int, 1500, 1280]);  convert_element_type_745 = None
	        view_1950: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_372, [sym_size_int, 1500, 1]);  clamp_min_372 = None
	        view_1951: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_744, [sym_size_int, 1500, 1]);  convert_element_type_744 = None
	        _assert_tensor_metadata_1121 = torch.ops.aten._assert_tensor_metadata.default(view_1949, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1121 = None
	        convert_element_type_746: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1949, torch.float32);  view_1949 = None
	        _assert_tensor_metadata_1122 = torch.ops.aten._assert_tensor_metadata.default(view_1951, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1122 = None
	        convert_element_type_747: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1951, torch.float32);  view_1951 = None
	        sub_5708: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_746, convert_element_type_747);  convert_element_type_746 = convert_element_type_747 = None
	        mul_12089: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5708, view_1950);  sub_5708 = view_1950 = None
	        view_1952: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12089, [sym_size_int, 1500, 1280]);  mul_12089 = None
	        _assert_tensor_metadata_1123 = torch.ops.aten._assert_tensor_metadata.default(view_1952, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1123 = None
	        view_1953: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg565_1, [5120, 40, 32]);  arg565_1 = None
	        view_1954: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg566_1, [5120, 40, 1]);  arg566_1 = None
	        view_1955: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg567_1, [5120, 40, 1]);  arg567_1 = None
	        _assert_tensor_metadata_1124 = torch.ops.aten._assert_tensor_metadata.default(view_1953, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1124 = None
	        convert_element_type_748: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1953, torch.float32);  view_1953 = None
	        _assert_tensor_metadata_1125 = torch.ops.aten._assert_tensor_metadata.default(view_1955, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1125 = None
	        convert_element_type_749: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1955, torch.float32);  view_1955 = None
	        sub_5712: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_748, convert_element_type_749);  convert_element_type_748 = convert_element_type_749 = None
	        mul_12094: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5712, view_1954);  sub_5712 = view_1954 = None
	        view_1956: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12094, [5120, 1280]);  mul_12094 = None
	        _assert_tensor_metadata_1126 = torch.ops.aten._assert_tensor_metadata.default(view_1956, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1126 = None
	        mul_12099: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1957: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1952, [mul_12099, 1280]);  view_1952 = mul_12099 = None
	        permute_209: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1956, [1, 0]);  view_1956 = None
	        addmm_103: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg564_1, view_1957, permute_209);  arg564_1 = view_1957 = permute_209 = None
	        view_1958: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_103, [sym_size_int, 1500, 5120]);  addmm_103 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_12106: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1958, 0.5)
	        mul_12107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1958, 0.7071067811865476);  view_1958 = None
	        erf_22: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_12107);  mul_12107 = None
	        add_19158: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_22, 1);  erf_22 = None
	        mul_12108: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12106, add_19158);  mul_12106 = add_19158 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_167: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_12108);  mul_12108 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1959: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_167, [sym_size_int, 1500, 5120])
	        amin_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1959, [2])
	        amax_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1959, [2]);  view_1959 = None
	        full_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_125, full_250);  amin_125 = full_250 = None
	        full_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_125, full_251);  amax_125 = full_251 = None
	        sub_5725: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_125, minimum_125);  maximum_125 = None
	        div_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5725, 255.0);  sub_5725 = None
	        clamp_min_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_250, 1.1920928955078125e-07);  div_250 = None
	        div_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_125, clamp_min_375);  minimum_125 = None
	        round_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_251);  div_251 = None
	        sub_5731: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_251);  round_251 = None
	        clamp_min_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5731, -128);  sub_5731 = None
	        clamp_max_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_376, 127);  clamp_min_376 = None
	        _assert_tensor_metadata_1127 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_375, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1127 = None
	        _assert_tensor_metadata_1128 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_250, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1128 = None
	        convert_element_type_750: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_250, torch.int8);  clamp_max_250 = None
	        view_1960: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_167, [sym_size_int, 1500, 5120]);  clone_167 = None
	        view_1961: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_375, [sym_size_int, 1500, 1])
	        view_1962: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_750, [sym_size_int, 1500, 1])
	        reciprocal_125: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1961);  view_1961 = None
	        mul_12154: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_125, 1.0);  reciprocal_125 = None
	        mul_12157: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1960, mul_12154);  view_1960 = mul_12154 = None
	        round_252: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_12157);  mul_12157 = None
	        add_19241: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_252, view_1962);  round_252 = view_1962 = None
	        clamp_min_377: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19241, -128);  add_19241 = None
	        clamp_max_251: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_377, 127);  clamp_min_377 = None
	        view_1963: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_251, [sym_size_int, 1500, 5120]);  clamp_max_251 = None
	        _assert_tensor_metadata_1129 = torch.ops.aten._assert_tensor_metadata.default(view_1963, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1129 = None
	        convert_element_type_751: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1963, torch.int8);  view_1963 = None
	        view_1964: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_751, [sym_size_int, 1500, 5120]);  convert_element_type_751 = None
	        view_1965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_375, [sym_size_int, 1500, 1]);  clamp_min_375 = None
	        view_1966: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_750, [sym_size_int, 1500, 1]);  convert_element_type_750 = None
	        _assert_tensor_metadata_1130 = torch.ops.aten._assert_tensor_metadata.default(view_1964, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1130 = None
	        convert_element_type_752: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1964, torch.float32);  view_1964 = None
	        _assert_tensor_metadata_1131 = torch.ops.aten._assert_tensor_metadata.default(view_1966, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1131 = None
	        convert_element_type_753: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1966, torch.float32);  view_1966 = None
	        sub_5751: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_752, convert_element_type_753);  convert_element_type_752 = convert_element_type_753 = None
	        mul_12179: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5751, view_1965);  sub_5751 = view_1965 = None
	        view_1967: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_12179, [sym_size_int, 1500, 5120]);  mul_12179 = None
	        _assert_tensor_metadata_1132 = torch.ops.aten._assert_tensor_metadata.default(view_1967, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1132 = None
	        view_1968: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg569_1, [1280, 160, 32]);  arg569_1 = None
	        view_1969: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg570_1, [1280, 160, 1]);  arg570_1 = None
	        view_1970: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg571_1, [1280, 160, 1]);  arg571_1 = None
	        _assert_tensor_metadata_1133 = torch.ops.aten._assert_tensor_metadata.default(view_1968, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1133 = None
	        convert_element_type_754: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1968, torch.float32);  view_1968 = None
	        _assert_tensor_metadata_1134 = torch.ops.aten._assert_tensor_metadata.default(view_1970, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1134 = None
	        convert_element_type_755: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1970, torch.float32);  view_1970 = None
	        sub_5755: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_754, convert_element_type_755);  convert_element_type_754 = convert_element_type_755 = None
	        mul_12184: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5755, view_1969);  sub_5755 = view_1969 = None
	        view_1971: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_12184, [1280, 5120]);  mul_12184 = None
	        _assert_tensor_metadata_1135 = torch.ops.aten._assert_tensor_metadata.default(view_1971, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1135 = None
	        mul_12189: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1972: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1967, [mul_12189, 5120]);  view_1967 = mul_12189 = None
	        permute_210: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1971, [1, 0]);  view_1971 = None
	        addmm_104: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg568_1, view_1972, permute_210);  arg568_1 = view_1972 = permute_210 = None
	        view_1973: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_104, [sym_size_int, 1500, 1280]);  addmm_104 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_168: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1973);  view_1973 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_19304: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_19006, clone_168);  add_19006 = clone_168 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_169: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_19304, memory_format = torch.contiguous_format)
	        var_mean_42 = torch.ops.aten.var_mean.correction(clone_169, [2], correction = 0, keepdim = True)
	        getitem_168: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_42[0]
	        getitem_169: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_42[1];  var_mean_42 = None
	        add_19309: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_168, 1e-05);  getitem_168 = None
	        rsqrt_42: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_19309);  add_19309 = None
	        sub_5761: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_169, getitem_169);  clone_169 = getitem_169 = None
	        mul_12200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5761, rsqrt_42);  sub_5761 = rsqrt_42 = None
	        mul_12201: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12200, arg572_1);  mul_12200 = arg572_1 = None
	        add_19310: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12201, arg573_1);  mul_12201 = arg573_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1974: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19310, [sym_size_int, 1500, 1280])
	        amin_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1974, [2])
	        amax_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1974, [2]);  view_1974 = None
	        full_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_126, full_252);  amin_126 = full_252 = None
	        full_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_126, full_253);  amax_126 = full_253 = None
	        sub_5772: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_126, minimum_126);  maximum_126 = None
	        div_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5772, 255.0);  sub_5772 = None
	        clamp_min_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_252, 1.1920928955078125e-07);  div_252 = None
	        div_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_126, clamp_min_378);  minimum_126 = None
	        round_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_253);  div_253 = None
	        sub_5778: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_253);  round_253 = None
	        clamp_min_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5778, -128);  sub_5778 = None
	        clamp_max_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_379, 127);  clamp_min_379 = None
	        _assert_tensor_metadata_1136 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_378, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1136 = None
	        _assert_tensor_metadata_1137 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_252, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1137 = None
	        convert_element_type_756: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_252, torch.int8);  clamp_max_252 = None
	        view_1975: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19310, [sym_size_int, 1500, 1280])
	        view_1976: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_378, [sym_size_int, 1500, 1])
	        view_1977: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_756, [sym_size_int, 1500, 1])
	        reciprocal_126: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1976);  view_1976 = None
	        mul_12249: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_126, 1.0);  reciprocal_126 = None
	        mul_12252: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1975, mul_12249);  view_1975 = mul_12249 = None
	        round_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12252);  mul_12252 = None
	        add_19397: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_254, view_1977);  round_254 = view_1977 = None
	        clamp_min_380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19397, -128);  add_19397 = None
	        clamp_max_253: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_380, 127);  clamp_min_380 = None
	        view_1978: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_253, [sym_size_int, 1500, 1280]);  clamp_max_253 = None
	        _assert_tensor_metadata_1138 = torch.ops.aten._assert_tensor_metadata.default(view_1978, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1138 = None
	        convert_element_type_757: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1978, torch.int8);  view_1978 = None
	        view_1979: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_757, [sym_size_int, 1500, 1280]);  convert_element_type_757 = None
	        view_1980: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_378, [sym_size_int, 1500, 1]);  clamp_min_378 = None
	        view_1981: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_756, [sym_size_int, 1500, 1]);  convert_element_type_756 = None
	        _assert_tensor_metadata_1139 = torch.ops.aten._assert_tensor_metadata.default(view_1979, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1139 = None
	        convert_element_type_758: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1979, torch.float32);  view_1979 = None
	        _assert_tensor_metadata_1140 = torch.ops.aten._assert_tensor_metadata.default(view_1981, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1140 = None
	        convert_element_type_759: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1981, torch.float32);  view_1981 = None
	        sub_5798: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_758, convert_element_type_759);  convert_element_type_758 = convert_element_type_759 = None
	        mul_12274: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5798, view_1980);  sub_5798 = view_1980 = None
	        view_1982: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12274, [sym_size_int, 1500, 1280]);  mul_12274 = None
	        _assert_tensor_metadata_1141 = torch.ops.aten._assert_tensor_metadata.default(view_1982, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1141 = None
	        view_1983: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg575_1, [1280, 40, 32]);  arg575_1 = None
	        view_1984: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg576_1, [1280, 40, 1]);  arg576_1 = None
	        view_1985: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg577_1, [1280, 40, 1]);  arg577_1 = None
	        _assert_tensor_metadata_1142 = torch.ops.aten._assert_tensor_metadata.default(view_1983, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1142 = None
	        convert_element_type_760: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1983, torch.float32);  view_1983 = None
	        _assert_tensor_metadata_1143 = torch.ops.aten._assert_tensor_metadata.default(view_1985, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1143 = None
	        convert_element_type_761: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1985, torch.float32);  view_1985 = None
	        sub_5802: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_760, convert_element_type_761);  convert_element_type_760 = convert_element_type_761 = None
	        mul_12279: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5802, view_1984);  sub_5802 = view_1984 = None
	        view_1986: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12279, [1280, 1280]);  mul_12279 = None
	        _assert_tensor_metadata_1144 = torch.ops.aten._assert_tensor_metadata.default(view_1986, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1144 = None
	        mul_12284: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1987: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1982, [mul_12284, 1280]);  view_1982 = mul_12284 = None
	        permute_211: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1986, [1, 0]);  view_1986 = None
	        addmm_105: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg574_1, view_1987, permute_211);  arg574_1 = view_1987 = permute_211 = None
	        view_1988: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_105, [sym_size_int, 1500, 1280]);  addmm_105 = None
	        mul_12291: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1988, 0.125);  view_1988 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1989: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_12291, [sym_size_int, 1500, 20, 64]);  mul_12291 = None
	        permute_212: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1989, [0, 2, 1, 3]);  view_1989 = None
	        clone_170: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_212, memory_format = torch.contiguous_format);  permute_212 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1990: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19310, [sym_size_int, 1500, 1280])
	        amin_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1990, [2])
	        amax_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1990, [2]);  view_1990 = None
	        full_254: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_127, full_254);  amin_127 = full_254 = None
	        full_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_127, full_255);  amax_127 = full_255 = None
	        sub_5817: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_127, minimum_127);  maximum_127 = None
	        div_254: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5817, 255.0);  sub_5817 = None
	        clamp_min_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_254, 1.1920928955078125e-07);  div_254 = None
	        div_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_127, clamp_min_381);  minimum_127 = None
	        round_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_255);  div_255 = None
	        sub_5823: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_255);  round_255 = None
	        clamp_min_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5823, -128);  sub_5823 = None
	        clamp_max_254: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_382, 127);  clamp_min_382 = None
	        _assert_tensor_metadata_1145 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_381, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1145 = None
	        _assert_tensor_metadata_1146 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_254, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1146 = None
	        convert_element_type_762: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_254, torch.int8);  clamp_max_254 = None
	        view_1991: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19310, [sym_size_int, 1500, 1280])
	        view_1992: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_381, [sym_size_int, 1500, 1])
	        view_1993: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_762, [sym_size_int, 1500, 1])
	        reciprocal_127: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1992);  view_1992 = None
	        mul_12345: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_127, 1.0);  reciprocal_127 = None
	        mul_12348: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1991, mul_12345);  view_1991 = mul_12345 = None
	        round_256: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12348);  mul_12348 = None
	        add_19549: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_256, view_1993);  round_256 = view_1993 = None
	        clamp_min_383: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19549, -128);  add_19549 = None
	        clamp_max_255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_383, 127);  clamp_min_383 = None
	        view_1994: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_255, [sym_size_int, 1500, 1280]);  clamp_max_255 = None
	        _assert_tensor_metadata_1147 = torch.ops.aten._assert_tensor_metadata.default(view_1994, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1147 = None
	        convert_element_type_763: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1994, torch.int8);  view_1994 = None
	        view_1995: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_763, [sym_size_int, 1500, 1280]);  convert_element_type_763 = None
	        view_1996: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_381, [sym_size_int, 1500, 1]);  clamp_min_381 = None
	        view_1997: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_762, [sym_size_int, 1500, 1]);  convert_element_type_762 = None
	        _assert_tensor_metadata_1148 = torch.ops.aten._assert_tensor_metadata.default(view_1995, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1148 = None
	        convert_element_type_764: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1995, torch.float32);  view_1995 = None
	        _assert_tensor_metadata_1149 = torch.ops.aten._assert_tensor_metadata.default(view_1997, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1149 = None
	        convert_element_type_765: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1997, torch.float32);  view_1997 = None
	        sub_5843: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_764, convert_element_type_765);  convert_element_type_764 = convert_element_type_765 = None
	        mul_12370: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5843, view_1996);  sub_5843 = view_1996 = None
	        view_1998: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12370, [sym_size_int, 1500, 1280]);  mul_12370 = None
	        _assert_tensor_metadata_1150 = torch.ops.aten._assert_tensor_metadata.default(view_1998, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1150 = None
	        view_1999: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg578_1, [1280, 40, 32]);  arg578_1 = None
	        view_2000: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg579_1, [1280, 40, 1]);  arg579_1 = None
	        view_2001: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg580_1, [1280, 40, 1]);  arg580_1 = None
	        _assert_tensor_metadata_1151 = torch.ops.aten._assert_tensor_metadata.default(view_1999, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1151 = None
	        convert_element_type_766: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1999, torch.float32);  view_1999 = None
	        _assert_tensor_metadata_1152 = torch.ops.aten._assert_tensor_metadata.default(view_2001, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1152 = None
	        convert_element_type_767: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2001, torch.float32);  view_2001 = None
	        sub_5847: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_766, convert_element_type_767);  convert_element_type_766 = convert_element_type_767 = None
	        mul_12375: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5847, view_2000);  sub_5847 = view_2000 = None
	        view_2002: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12375, [1280, 1280]);  mul_12375 = None
	        _assert_tensor_metadata_1153 = torch.ops.aten._assert_tensor_metadata.default(view_2002, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1153 = None
	        permute_213: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2002, [1, 0]);  view_2002 = None
	        mul_12378: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2003: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1998, [mul_12378, 1280]);  view_1998 = mul_12378 = None
	        mm_21: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2003, permute_213);  view_2003 = permute_213 = None
	        view_2004: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_21, [sym_size_int, 1500, 1280]);  mm_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2005: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2004, [sym_size_int, -1, 20, 64]);  view_2004 = None
	        permute_214: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2005, [0, 2, 1, 3]);  view_2005 = None
	        clone_171: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_214, memory_format = torch.contiguous_format);  permute_214 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2006: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19310, [sym_size_int, 1500, 1280])
	        amin_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2006, [2])
	        amax_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2006, [2]);  view_2006 = None
	        full_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_128, full_256);  amin_128 = full_256 = None
	        full_257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_128, full_257);  amax_128 = full_257 = None
	        sub_5861: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_128, minimum_128);  maximum_128 = None
	        div_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5861, 255.0);  sub_5861 = None
	        clamp_min_384: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_256, 1.1920928955078125e-07);  div_256 = None
	        div_257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_128, clamp_min_384);  minimum_128 = None
	        round_257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_257);  div_257 = None
	        sub_5867: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_257);  round_257 = None
	        clamp_min_385: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5867, -128);  sub_5867 = None
	        clamp_max_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_385, 127);  clamp_min_385 = None
	        _assert_tensor_metadata_1154 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_384, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1154 = None
	        _assert_tensor_metadata_1155 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_256, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1155 = None
	        convert_element_type_768: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_256, torch.int8);  clamp_max_256 = None
	        view_2007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19310, [sym_size_int, 1500, 1280]);  add_19310 = None
	        view_2008: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_384, [sym_size_int, 1500, 1])
	        view_2009: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_768, [sym_size_int, 1500, 1])
	        reciprocal_128: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2008);  view_2008 = None
	        mul_12444: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_128, 1.0);  reciprocal_128 = None
	        mul_12447: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2007, mul_12444);  view_2007 = mul_12444 = None
	        round_258: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12447);  mul_12447 = None
	        add_19697: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_258, view_2009);  round_258 = view_2009 = None
	        clamp_min_386: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19697, -128);  add_19697 = None
	        clamp_max_257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_386, 127);  clamp_min_386 = None
	        view_2010: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_257, [sym_size_int, 1500, 1280]);  clamp_max_257 = None
	        _assert_tensor_metadata_1156 = torch.ops.aten._assert_tensor_metadata.default(view_2010, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1156 = None
	        convert_element_type_769: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2010, torch.int8);  view_2010 = None
	        view_2011: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_769, [sym_size_int, 1500, 1280]);  convert_element_type_769 = None
	        view_2012: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_384, [sym_size_int, 1500, 1]);  clamp_min_384 = None
	        view_2013: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_768, [sym_size_int, 1500, 1]);  convert_element_type_768 = None
	        _assert_tensor_metadata_1157 = torch.ops.aten._assert_tensor_metadata.default(view_2011, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1157 = None
	        convert_element_type_770: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2011, torch.float32);  view_2011 = None
	        _assert_tensor_metadata_1158 = torch.ops.aten._assert_tensor_metadata.default(view_2013, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1158 = None
	        convert_element_type_771: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2013, torch.float32);  view_2013 = None
	        sub_5887: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_770, convert_element_type_771);  convert_element_type_770 = convert_element_type_771 = None
	        mul_12469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5887, view_2012);  sub_5887 = view_2012 = None
	        view_2014: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12469, [sym_size_int, 1500, 1280]);  mul_12469 = None
	        _assert_tensor_metadata_1159 = torch.ops.aten._assert_tensor_metadata.default(view_2014, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1159 = None
	        view_2015: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg582_1, [1280, 40, 32]);  arg582_1 = None
	        view_2016: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg583_1, [1280, 40, 1]);  arg583_1 = None
	        view_2017: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg584_1, [1280, 40, 1]);  arg584_1 = None
	        _assert_tensor_metadata_1160 = torch.ops.aten._assert_tensor_metadata.default(view_2015, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1160 = None
	        convert_element_type_772: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2015, torch.float32);  view_2015 = None
	        _assert_tensor_metadata_1161 = torch.ops.aten._assert_tensor_metadata.default(view_2017, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1161 = None
	        convert_element_type_773: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2017, torch.float32);  view_2017 = None
	        sub_5891: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_772, convert_element_type_773);  convert_element_type_772 = convert_element_type_773 = None
	        mul_12474: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5891, view_2016);  sub_5891 = view_2016 = None
	        view_2018: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12474, [1280, 1280]);  mul_12474 = None
	        _assert_tensor_metadata_1162 = torch.ops.aten._assert_tensor_metadata.default(view_2018, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1162 = None
	        mul_12479: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2019: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2014, [mul_12479, 1280]);  view_2014 = mul_12479 = None
	        permute_215: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2018, [1, 0]);  view_2018 = None
	        addmm_106: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg581_1, view_2019, permute_215);  arg581_1 = view_2019 = permute_215 = None
	        view_2020: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_106, [sym_size_int, 1500, 1280]);  addmm_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2021: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2020, [sym_size_int, -1, 20, 64]);  view_2020 = None
	        permute_216: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2021, [0, 2, 1, 3]);  view_2021 = None
	        clone_172: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_216, memory_format = torch.contiguous_format);  permute_216 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_21 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_170, clone_171, clone_172, None, False, scale = 1.0);  clone_170 = clone_171 = clone_172 = None
	        getitem_170: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_21[0];  _scaled_dot_product_efficient_attention_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_217: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_170, [0, 2, 1, 3]);  getitem_170 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2022: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_217, [sym_size_int, 1500, -1]);  permute_217 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2023: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2022, [sym_size_int, 1500, 1280])
	        amin_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2023, [2])
	        amax_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2023, [2]);  view_2023 = None
	        full_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_129, full_258);  amin_129 = full_258 = None
	        full_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_129, full_259);  amax_129 = full_259 = None
	        sub_5909: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_129, minimum_129);  maximum_129 = None
	        div_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5909, 255.0);  sub_5909 = None
	        clamp_min_387: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_258, 1.1920928955078125e-07);  div_258 = None
	        div_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_129, clamp_min_387);  minimum_129 = None
	        round_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_259);  div_259 = None
	        sub_5915: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_259);  round_259 = None
	        clamp_min_388: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5915, -128);  sub_5915 = None
	        clamp_max_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_388, 127);  clamp_min_388 = None
	        _assert_tensor_metadata_1163 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_387, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1163 = None
	        _assert_tensor_metadata_1164 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_258, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1164 = None
	        convert_element_type_774: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_258, torch.int8);  clamp_max_258 = None
	        view_2024: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2022, [sym_size_int, 1500, 1280]);  view_2022 = None
	        view_2025: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_387, [sym_size_int, 1500, 1])
	        view_2026: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_774, [sym_size_int, 1500, 1])
	        reciprocal_129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2025);  view_2025 = None
	        mul_12549: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_129, 1.0);  reciprocal_129 = None
	        mul_12552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2024, mul_12549);  view_2024 = mul_12549 = None
	        round_260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12552);  mul_12552 = None
	        add_19861: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_260, view_2026);  round_260 = view_2026 = None
	        clamp_min_389: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19861, -128);  add_19861 = None
	        clamp_max_259: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_389, 127);  clamp_min_389 = None
	        view_2027: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_259, [sym_size_int, 1500, 1280]);  clamp_max_259 = None
	        _assert_tensor_metadata_1165 = torch.ops.aten._assert_tensor_metadata.default(view_2027, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1165 = None
	        convert_element_type_775: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2027, torch.int8);  view_2027 = None
	        view_2028: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_775, [sym_size_int, 1500, 1280]);  convert_element_type_775 = None
	        view_2029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_387, [sym_size_int, 1500, 1]);  clamp_min_387 = None
	        view_2030: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_774, [sym_size_int, 1500, 1]);  convert_element_type_774 = None
	        _assert_tensor_metadata_1166 = torch.ops.aten._assert_tensor_metadata.default(view_2028, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1166 = None
	        convert_element_type_776: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2028, torch.float32);  view_2028 = None
	        _assert_tensor_metadata_1167 = torch.ops.aten._assert_tensor_metadata.default(view_2030, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1167 = None
	        convert_element_type_777: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2030, torch.float32);  view_2030 = None
	        sub_5935: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_776, convert_element_type_777);  convert_element_type_776 = convert_element_type_777 = None
	        mul_12574: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5935, view_2029);  sub_5935 = view_2029 = None
	        view_2031: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12574, [sym_size_int, 1500, 1280]);  mul_12574 = None
	        _assert_tensor_metadata_1168 = torch.ops.aten._assert_tensor_metadata.default(view_2031, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1168 = None
	        view_2032: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg586_1, [1280, 40, 32]);  arg586_1 = None
	        view_2033: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg587_1, [1280, 40, 1]);  arg587_1 = None
	        view_2034: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg588_1, [1280, 40, 1]);  arg588_1 = None
	        _assert_tensor_metadata_1169 = torch.ops.aten._assert_tensor_metadata.default(view_2032, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1169 = None
	        convert_element_type_778: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2032, torch.float32);  view_2032 = None
	        _assert_tensor_metadata_1170 = torch.ops.aten._assert_tensor_metadata.default(view_2034, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1170 = None
	        convert_element_type_779: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2034, torch.float32);  view_2034 = None
	        sub_5939: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_778, convert_element_type_779);  convert_element_type_778 = convert_element_type_779 = None
	        mul_12579: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5939, view_2033);  sub_5939 = view_2033 = None
	        view_2035: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12579, [1280, 1280]);  mul_12579 = None
	        _assert_tensor_metadata_1171 = torch.ops.aten._assert_tensor_metadata.default(view_2035, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1171 = None
	        mul_12584: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2036: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2031, [mul_12584, 1280]);  view_2031 = mul_12584 = None
	        permute_218: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2035, [1, 0]);  view_2035 = None
	        addmm_107: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg585_1, view_2036, permute_218);  arg585_1 = view_2036 = permute_218 = None
	        view_2037: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_107, [sym_size_int, 1500, 1280]);  addmm_107 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2037);  view_2037 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_19924: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_19304, clone_173);  add_19304 = clone_173 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_19924, memory_format = torch.contiguous_format)
	        var_mean_43 = torch.ops.aten.var_mean.correction(clone_174, [2], correction = 0, keepdim = True)
	        getitem_174: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_43[0]
	        getitem_175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_43[1];  var_mean_43 = None
	        add_19929: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_174, 1e-05);  getitem_174 = None
	        rsqrt_43: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_19929);  add_19929 = None
	        sub_5945: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_174, getitem_175);  clone_174 = getitem_175 = None
	        mul_12595: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5945, rsqrt_43);  sub_5945 = rsqrt_43 = None
	        mul_12596: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12595, arg589_1);  mul_12595 = arg589_1 = None
	        add_19930: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12596, arg590_1);  mul_12596 = arg590_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2038: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19930, [sym_size_int, 1500, 1280])
	        amin_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2038, [2])
	        amax_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2038, [2]);  view_2038 = None
	        full_260: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_130, full_260);  amin_130 = full_260 = None
	        full_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_130, full_261);  amax_130 = full_261 = None
	        sub_5956: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_130, minimum_130);  maximum_130 = None
	        div_260: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5956, 255.0);  sub_5956 = None
	        clamp_min_390: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_260, 1.1920928955078125e-07);  div_260 = None
	        div_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_130, clamp_min_390);  minimum_130 = None
	        round_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_261);  div_261 = None
	        sub_5962: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_261);  round_261 = None
	        clamp_min_391: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5962, -128);  sub_5962 = None
	        clamp_max_260: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_391, 127);  clamp_min_391 = None
	        _assert_tensor_metadata_1172 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_390, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1172 = None
	        _assert_tensor_metadata_1173 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_260, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1173 = None
	        convert_element_type_780: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_260, torch.int8);  clamp_max_260 = None
	        view_2039: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19930, [sym_size_int, 1500, 1280]);  add_19930 = None
	        view_2040: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_390, [sym_size_int, 1500, 1])
	        view_2041: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_780, [sym_size_int, 1500, 1])
	        reciprocal_130: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2040);  view_2040 = None
	        mul_12644: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_130, 1.0);  reciprocal_130 = None
	        mul_12647: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2039, mul_12644);  view_2039 = mul_12644 = None
	        round_262: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12647);  mul_12647 = None
	        add_20017: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_262, view_2041);  round_262 = view_2041 = None
	        clamp_min_392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20017, -128);  add_20017 = None
	        clamp_max_261: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_392, 127);  clamp_min_392 = None
	        view_2042: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_261, [sym_size_int, 1500, 1280]);  clamp_max_261 = None
	        _assert_tensor_metadata_1174 = torch.ops.aten._assert_tensor_metadata.default(view_2042, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1174 = None
	        convert_element_type_781: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2042, torch.int8);  view_2042 = None
	        view_2043: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_781, [sym_size_int, 1500, 1280]);  convert_element_type_781 = None
	        view_2044: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_390, [sym_size_int, 1500, 1]);  clamp_min_390 = None
	        view_2045: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_780, [sym_size_int, 1500, 1]);  convert_element_type_780 = None
	        _assert_tensor_metadata_1175 = torch.ops.aten._assert_tensor_metadata.default(view_2043, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1175 = None
	        convert_element_type_782: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2043, torch.float32);  view_2043 = None
	        _assert_tensor_metadata_1176 = torch.ops.aten._assert_tensor_metadata.default(view_2045, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1176 = None
	        convert_element_type_783: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2045, torch.float32);  view_2045 = None
	        sub_5982: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_782, convert_element_type_783);  convert_element_type_782 = convert_element_type_783 = None
	        mul_12669: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5982, view_2044);  sub_5982 = view_2044 = None
	        view_2046: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12669, [sym_size_int, 1500, 1280]);  mul_12669 = None
	        _assert_tensor_metadata_1177 = torch.ops.aten._assert_tensor_metadata.default(view_2046, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1177 = None
	        view_2047: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg592_1, [5120, 40, 32]);  arg592_1 = None
	        view_2048: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg593_1, [5120, 40, 1]);  arg593_1 = None
	        view_2049: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg594_1, [5120, 40, 1]);  arg594_1 = None
	        _assert_tensor_metadata_1178 = torch.ops.aten._assert_tensor_metadata.default(view_2047, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1178 = None
	        convert_element_type_784: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2047, torch.float32);  view_2047 = None
	        _assert_tensor_metadata_1179 = torch.ops.aten._assert_tensor_metadata.default(view_2049, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1179 = None
	        convert_element_type_785: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2049, torch.float32);  view_2049 = None
	        sub_5986: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_784, convert_element_type_785);  convert_element_type_784 = convert_element_type_785 = None
	        mul_12674: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5986, view_2048);  sub_5986 = view_2048 = None
	        view_2050: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12674, [5120, 1280]);  mul_12674 = None
	        _assert_tensor_metadata_1180 = torch.ops.aten._assert_tensor_metadata.default(view_2050, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1180 = None
	        mul_12679: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2051: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2046, [mul_12679, 1280]);  view_2046 = mul_12679 = None
	        permute_219: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2050, [1, 0]);  view_2050 = None
	        addmm_108: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg591_1, view_2051, permute_219);  arg591_1 = view_2051 = permute_219 = None
	        view_2052: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_108, [sym_size_int, 1500, 5120]);  addmm_108 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_12686: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2052, 0.5)
	        mul_12687: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2052, 0.7071067811865476);  view_2052 = None
	        erf_23: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_12687);  mul_12687 = None
	        add_20076: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_23, 1);  erf_23 = None
	        mul_12688: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12686, add_20076);  mul_12686 = add_20076 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_175: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_12688);  mul_12688 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2053: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_175, [sym_size_int, 1500, 5120])
	        amin_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2053, [2])
	        amax_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2053, [2]);  view_2053 = None
	        full_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_131, full_262);  amin_131 = full_262 = None
	        full_263: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_131, full_263);  amax_131 = full_263 = None
	        sub_5999: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_131, minimum_131);  maximum_131 = None
	        div_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5999, 255.0);  sub_5999 = None
	        clamp_min_393: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_262, 1.1920928955078125e-07);  div_262 = None
	        div_263: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_131, clamp_min_393);  minimum_131 = None
	        round_263: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_263);  div_263 = None
	        sub_6005: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_263);  round_263 = None
	        clamp_min_394: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6005, -128);  sub_6005 = None
	        clamp_max_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_394, 127);  clamp_min_394 = None
	        _assert_tensor_metadata_1181 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_393, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1181 = None
	        _assert_tensor_metadata_1182 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_262, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1182 = None
	        convert_element_type_786: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_262, torch.int8);  clamp_max_262 = None
	        view_2054: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_175, [sym_size_int, 1500, 5120]);  clone_175 = None
	        view_2055: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_393, [sym_size_int, 1500, 1])
	        view_2056: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_786, [sym_size_int, 1500, 1])
	        reciprocal_131: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2055);  view_2055 = None
	        mul_12734: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_131, 1.0);  reciprocal_131 = None
	        mul_12737: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2054, mul_12734);  view_2054 = mul_12734 = None
	        round_264: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_12737);  mul_12737 = None
	        add_20159: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_264, view_2056);  round_264 = view_2056 = None
	        clamp_min_395: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20159, -128);  add_20159 = None
	        clamp_max_263: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_395, 127);  clamp_min_395 = None
	        view_2057: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_263, [sym_size_int, 1500, 5120]);  clamp_max_263 = None
	        _assert_tensor_metadata_1183 = torch.ops.aten._assert_tensor_metadata.default(view_2057, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1183 = None
	        convert_element_type_787: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2057, torch.int8);  view_2057 = None
	        view_2058: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_787, [sym_size_int, 1500, 5120]);  convert_element_type_787 = None
	        view_2059: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_393, [sym_size_int, 1500, 1]);  clamp_min_393 = None
	        view_2060: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_786, [sym_size_int, 1500, 1]);  convert_element_type_786 = None
	        _assert_tensor_metadata_1184 = torch.ops.aten._assert_tensor_metadata.default(view_2058, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1184 = None
	        convert_element_type_788: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2058, torch.float32);  view_2058 = None
	        _assert_tensor_metadata_1185 = torch.ops.aten._assert_tensor_metadata.default(view_2060, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1185 = None
	        convert_element_type_789: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2060, torch.float32);  view_2060 = None
	        sub_6025: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_788, convert_element_type_789);  convert_element_type_788 = convert_element_type_789 = None
	        mul_12759: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6025, view_2059);  sub_6025 = view_2059 = None
	        view_2061: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_12759, [sym_size_int, 1500, 5120]);  mul_12759 = None
	        _assert_tensor_metadata_1186 = torch.ops.aten._assert_tensor_metadata.default(view_2061, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1186 = None
	        view_2062: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg596_1, [1280, 160, 32]);  arg596_1 = None
	        view_2063: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg597_1, [1280, 160, 1]);  arg597_1 = None
	        view_2064: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg598_1, [1280, 160, 1]);  arg598_1 = None
	        _assert_tensor_metadata_1187 = torch.ops.aten._assert_tensor_metadata.default(view_2062, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1187 = None
	        convert_element_type_790: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2062, torch.float32);  view_2062 = None
	        _assert_tensor_metadata_1188 = torch.ops.aten._assert_tensor_metadata.default(view_2064, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1188 = None
	        convert_element_type_791: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2064, torch.float32);  view_2064 = None
	        sub_6029: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_790, convert_element_type_791);  convert_element_type_790 = convert_element_type_791 = None
	        mul_12764: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6029, view_2063);  sub_6029 = view_2063 = None
	        view_2065: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_12764, [1280, 5120]);  mul_12764 = None
	        _assert_tensor_metadata_1189 = torch.ops.aten._assert_tensor_metadata.default(view_2065, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1189 = None
	        mul_12769: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2066: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2061, [mul_12769, 5120]);  view_2061 = mul_12769 = None
	        permute_220: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2065, [1, 0]);  view_2065 = None
	        addmm_109: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg595_1, view_2066, permute_220);  arg595_1 = view_2066 = permute_220 = None
	        view_2067: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_109, [sym_size_int, 1500, 1280]);  addmm_109 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2067);  view_2067 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_20222: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_19924, clone_176);  add_19924 = clone_176 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_177: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_20222, memory_format = torch.contiguous_format)
	        var_mean_44 = torch.ops.aten.var_mean.correction(clone_177, [2], correction = 0, keepdim = True)
	        getitem_176: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_44[0]
	        getitem_177: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_44[1];  var_mean_44 = None
	        add_20227: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_176, 1e-05);  getitem_176 = None
	        rsqrt_44: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_20227);  add_20227 = None
	        sub_6035: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_177, getitem_177);  clone_177 = getitem_177 = None
	        mul_12780: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6035, rsqrt_44);  sub_6035 = rsqrt_44 = None
	        mul_12781: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12780, arg599_1);  mul_12780 = arg599_1 = None
	        add_20228: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12781, arg600_1);  mul_12781 = arg600_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2068: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_20228, [sym_size_int, 1500, 1280])
	        amin_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2068, [2])
	        amax_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2068, [2]);  view_2068 = None
	        full_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_132, full_264);  amin_132 = full_264 = None
	        full_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_132, full_265);  amax_132 = full_265 = None
	        sub_6046: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_132, minimum_132);  maximum_132 = None
	        div_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6046, 255.0);  sub_6046 = None
	        clamp_min_396: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_264, 1.1920928955078125e-07);  div_264 = None
	        div_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_132, clamp_min_396);  minimum_132 = None
	        round_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_265);  div_265 = None
	        sub_6052: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_265);  round_265 = None
	        clamp_min_397: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6052, -128);  sub_6052 = None
	        clamp_max_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_397, 127);  clamp_min_397 = None
	        _assert_tensor_metadata_1190 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_396, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1190 = None
	        _assert_tensor_metadata_1191 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_264, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1191 = None
	        convert_element_type_792: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_264, torch.int8);  clamp_max_264 = None
	        view_2069: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_20228, [sym_size_int, 1500, 1280])
	        view_2070: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_396, [sym_size_int, 1500, 1])
	        view_2071: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_792, [sym_size_int, 1500, 1])
	        reciprocal_132: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2070);  view_2070 = None
	        mul_12829: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_132, 1.0);  reciprocal_132 = None
	        mul_12832: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2069, mul_12829);  view_2069 = mul_12829 = None
	        round_266: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12832);  mul_12832 = None
	        add_20315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_266, view_2071);  round_266 = view_2071 = None
	        clamp_min_398: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20315, -128);  add_20315 = None
	        clamp_max_265: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_398, 127);  clamp_min_398 = None
	        view_2072: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_265, [sym_size_int, 1500, 1280]);  clamp_max_265 = None
	        _assert_tensor_metadata_1192 = torch.ops.aten._assert_tensor_metadata.default(view_2072, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1192 = None
	        convert_element_type_793: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2072, torch.int8);  view_2072 = None
	        view_2073: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_793, [sym_size_int, 1500, 1280]);  convert_element_type_793 = None
	        view_2074: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_396, [sym_size_int, 1500, 1]);  clamp_min_396 = None
	        view_2075: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_792, [sym_size_int, 1500, 1]);  convert_element_type_792 = None
	        _assert_tensor_metadata_1193 = torch.ops.aten._assert_tensor_metadata.default(view_2073, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1193 = None
	        convert_element_type_794: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2073, torch.float32);  view_2073 = None
	        _assert_tensor_metadata_1194 = torch.ops.aten._assert_tensor_metadata.default(view_2075, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1194 = None
	        convert_element_type_795: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2075, torch.float32);  view_2075 = None
	        sub_6072: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_794, convert_element_type_795);  convert_element_type_794 = convert_element_type_795 = None
	        mul_12854: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6072, view_2074);  sub_6072 = view_2074 = None
	        view_2076: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12854, [sym_size_int, 1500, 1280]);  mul_12854 = None
	        _assert_tensor_metadata_1195 = torch.ops.aten._assert_tensor_metadata.default(view_2076, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1195 = None
	        view_2077: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg602_1, [1280, 40, 32]);  arg602_1 = None
	        view_2078: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg603_1, [1280, 40, 1]);  arg603_1 = None
	        view_2079: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg604_1, [1280, 40, 1]);  arg604_1 = None
	        _assert_tensor_metadata_1196 = torch.ops.aten._assert_tensor_metadata.default(view_2077, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1196 = None
	        convert_element_type_796: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2077, torch.float32);  view_2077 = None
	        _assert_tensor_metadata_1197 = torch.ops.aten._assert_tensor_metadata.default(view_2079, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1197 = None
	        convert_element_type_797: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2079, torch.float32);  view_2079 = None
	        sub_6076: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_796, convert_element_type_797);  convert_element_type_796 = convert_element_type_797 = None
	        mul_12859: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6076, view_2078);  sub_6076 = view_2078 = None
	        view_2080: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12859, [1280, 1280]);  mul_12859 = None
	        _assert_tensor_metadata_1198 = torch.ops.aten._assert_tensor_metadata.default(view_2080, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1198 = None
	        mul_12864: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2081: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2076, [mul_12864, 1280]);  view_2076 = mul_12864 = None
	        permute_221: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2080, [1, 0]);  view_2080 = None
	        addmm_110: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg601_1, view_2081, permute_221);  arg601_1 = view_2081 = permute_221 = None
	        view_2082: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_110, [sym_size_int, 1500, 1280]);  addmm_110 = None
	        mul_12871: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2082, 0.125);  view_2082 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2083: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_12871, [sym_size_int, 1500, 20, 64]);  mul_12871 = None
	        permute_222: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2083, [0, 2, 1, 3]);  view_2083 = None
	        clone_178: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_222, memory_format = torch.contiguous_format);  permute_222 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2084: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_20228, [sym_size_int, 1500, 1280])
	        amin_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2084, [2])
	        amax_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2084, [2]);  view_2084 = None
	        full_266: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_133, full_266);  amin_133 = full_266 = None
	        full_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_133, full_267);  amax_133 = full_267 = None
	        sub_6091: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_133, minimum_133);  maximum_133 = None
	        div_266: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6091, 255.0);  sub_6091 = None
	        clamp_min_399: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_266, 1.1920928955078125e-07);  div_266 = None
	        div_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_133, clamp_min_399);  minimum_133 = None
	        round_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_267);  div_267 = None
	        sub_6097: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_267);  round_267 = None
	        clamp_min_400: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6097, -128);  sub_6097 = None
	        clamp_max_266: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_400, 127);  clamp_min_400 = None
	        _assert_tensor_metadata_1199 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_399, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1199 = None
	        _assert_tensor_metadata_1200 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_266, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1200 = None
	        convert_element_type_798: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_266, torch.int8);  clamp_max_266 = None
	        view_2085: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_20228, [sym_size_int, 1500, 1280])
	        view_2086: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_399, [sym_size_int, 1500, 1])
	        view_2087: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_798, [sym_size_int, 1500, 1])
	        reciprocal_133: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2086);  view_2086 = None
	        mul_12925: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_133, 1.0);  reciprocal_133 = None
	        mul_12928: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2085, mul_12925);  view_2085 = mul_12925 = None
	        round_268: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12928);  mul_12928 = None
	        add_20467: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_268, view_2087);  round_268 = view_2087 = None
	        clamp_min_401: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20467, -128);  add_20467 = None
	        clamp_max_267: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_401, 127);  clamp_min_401 = None
	        view_2088: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_267, [sym_size_int, 1500, 1280]);  clamp_max_267 = None
	        _assert_tensor_metadata_1201 = torch.ops.aten._assert_tensor_metadata.default(view_2088, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1201 = None
	        convert_element_type_799: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2088, torch.int8);  view_2088 = None
	        view_2089: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_799, [sym_size_int, 1500, 1280]);  convert_element_type_799 = None
	        view_2090: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_399, [sym_size_int, 1500, 1]);  clamp_min_399 = None
	        view_2091: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_798, [sym_size_int, 1500, 1]);  convert_element_type_798 = None
	        _assert_tensor_metadata_1202 = torch.ops.aten._assert_tensor_metadata.default(view_2089, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1202 = None
	        convert_element_type_800: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2089, torch.float32);  view_2089 = None
	        _assert_tensor_metadata_1203 = torch.ops.aten._assert_tensor_metadata.default(view_2091, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1203 = None
	        convert_element_type_801: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2091, torch.float32);  view_2091 = None
	        sub_6117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_800, convert_element_type_801);  convert_element_type_800 = convert_element_type_801 = None
	        mul_12950: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6117, view_2090);  sub_6117 = view_2090 = None
	        view_2092: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12950, [sym_size_int, 1500, 1280]);  mul_12950 = None
	        _assert_tensor_metadata_1204 = torch.ops.aten._assert_tensor_metadata.default(view_2092, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1204 = None
	        view_2093: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg605_1, [1280, 40, 32]);  arg605_1 = None
	        view_2094: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg606_1, [1280, 40, 1]);  arg606_1 = None
	        view_2095: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg607_1, [1280, 40, 1]);  arg607_1 = None
	        _assert_tensor_metadata_1205 = torch.ops.aten._assert_tensor_metadata.default(view_2093, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1205 = None
	        convert_element_type_802: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2093, torch.float32);  view_2093 = None
	        _assert_tensor_metadata_1206 = torch.ops.aten._assert_tensor_metadata.default(view_2095, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1206 = None
	        convert_element_type_803: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2095, torch.float32);  view_2095 = None
	        sub_6121: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_802, convert_element_type_803);  convert_element_type_802 = convert_element_type_803 = None
	        mul_12955: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6121, view_2094);  sub_6121 = view_2094 = None
	        view_2096: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12955, [1280, 1280]);  mul_12955 = None
	        _assert_tensor_metadata_1207 = torch.ops.aten._assert_tensor_metadata.default(view_2096, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1207 = None
	        permute_223: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2096, [1, 0]);  view_2096 = None
	        mul_12958: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2097: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2092, [mul_12958, 1280]);  view_2092 = mul_12958 = None
	        mm_22: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2097, permute_223);  view_2097 = permute_223 = None
	        view_2098: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_22, [sym_size_int, 1500, 1280]);  mm_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2099: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2098, [sym_size_int, -1, 20, 64]);  view_2098 = None
	        permute_224: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2099, [0, 2, 1, 3]);  view_2099 = None
	        clone_179: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_224, memory_format = torch.contiguous_format);  permute_224 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2100: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_20228, [sym_size_int, 1500, 1280])
	        amin_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2100, [2])
	        amax_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2100, [2]);  view_2100 = None
	        full_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_134, full_268);  amin_134 = full_268 = None
	        full_269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_134, full_269);  amax_134 = full_269 = None
	        sub_6135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_134, minimum_134);  maximum_134 = None
	        div_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6135, 255.0);  sub_6135 = None
	        clamp_min_402: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_268, 1.1920928955078125e-07);  div_268 = None
	        div_269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_134, clamp_min_402);  minimum_134 = None
	        round_269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_269);  div_269 = None
	        sub_6141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_269);  round_269 = None
	        clamp_min_403: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6141, -128);  sub_6141 = None
	        clamp_max_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_403, 127);  clamp_min_403 = None
	        _assert_tensor_metadata_1208 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_402, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1208 = None
	        _assert_tensor_metadata_1209 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_268, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1209 = None
	        convert_element_type_804: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_268, torch.int8);  clamp_max_268 = None
	        view_2101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_20228, [sym_size_int, 1500, 1280]);  add_20228 = None
	        view_2102: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_402, [sym_size_int, 1500, 1])
	        view_2103: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_804, [sym_size_int, 1500, 1])
	        reciprocal_134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2102);  view_2102 = None
	        mul_13024: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_134, 1.0);  reciprocal_134 = None
	        mul_13027: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2101, mul_13024);  view_2101 = mul_13024 = None
	        round_270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13027);  mul_13027 = None
	        add_20615: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_270, view_2103);  round_270 = view_2103 = None
	        clamp_min_404: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20615, -128);  add_20615 = None
	        clamp_max_269: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_404, 127);  clamp_min_404 = None
	        view_2104: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_269, [sym_size_int, 1500, 1280]);  clamp_max_269 = None
	        _assert_tensor_metadata_1210 = torch.ops.aten._assert_tensor_metadata.default(view_2104, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1210 = None
	        convert_element_type_805: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2104, torch.int8);  view_2104 = None
	        view_2105: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_805, [sym_size_int, 1500, 1280]);  convert_element_type_805 = None
	        view_2106: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_402, [sym_size_int, 1500, 1]);  clamp_min_402 = None
	        view_2107: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_804, [sym_size_int, 1500, 1]);  convert_element_type_804 = None
	        _assert_tensor_metadata_1211 = torch.ops.aten._assert_tensor_metadata.default(view_2105, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1211 = None
	        convert_element_type_806: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2105, torch.float32);  view_2105 = None
	        _assert_tensor_metadata_1212 = torch.ops.aten._assert_tensor_metadata.default(view_2107, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1212 = None
	        convert_element_type_807: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2107, torch.float32);  view_2107 = None
	        sub_6161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_806, convert_element_type_807);  convert_element_type_806 = convert_element_type_807 = None
	        mul_13049: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6161, view_2106);  sub_6161 = view_2106 = None
	        view_2108: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13049, [sym_size_int, 1500, 1280]);  mul_13049 = None
	        _assert_tensor_metadata_1213 = torch.ops.aten._assert_tensor_metadata.default(view_2108, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1213 = None
	        view_2109: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg609_1, [1280, 40, 32]);  arg609_1 = None
	        view_2110: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg610_1, [1280, 40, 1]);  arg610_1 = None
	        view_2111: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg611_1, [1280, 40, 1]);  arg611_1 = None
	        _assert_tensor_metadata_1214 = torch.ops.aten._assert_tensor_metadata.default(view_2109, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1214 = None
	        convert_element_type_808: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2109, torch.float32);  view_2109 = None
	        _assert_tensor_metadata_1215 = torch.ops.aten._assert_tensor_metadata.default(view_2111, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1215 = None
	        convert_element_type_809: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2111, torch.float32);  view_2111 = None
	        sub_6165: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_808, convert_element_type_809);  convert_element_type_808 = convert_element_type_809 = None
	        mul_13054: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6165, view_2110);  sub_6165 = view_2110 = None
	        view_2112: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13054, [1280, 1280]);  mul_13054 = None
	        _assert_tensor_metadata_1216 = torch.ops.aten._assert_tensor_metadata.default(view_2112, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1216 = None
	        mul_13059: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2113: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2108, [mul_13059, 1280]);  view_2108 = mul_13059 = None
	        permute_225: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2112, [1, 0]);  view_2112 = None
	        addmm_111: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg608_1, view_2113, permute_225);  arg608_1 = view_2113 = permute_225 = None
	        view_2114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_111, [sym_size_int, 1500, 1280]);  addmm_111 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2115: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2114, [sym_size_int, -1, 20, 64]);  view_2114 = None
	        permute_226: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2115, [0, 2, 1, 3]);  view_2115 = None
	        clone_180: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_226, memory_format = torch.contiguous_format);  permute_226 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_22 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_178, clone_179, clone_180, None, False, scale = 1.0);  clone_178 = clone_179 = clone_180 = None
	        getitem_178: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_22[0];  _scaled_dot_product_efficient_attention_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_227: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_178, [0, 2, 1, 3]);  getitem_178 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_227, [sym_size_int, 1500, -1]);  permute_227 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2116, [sym_size_int, 1500, 1280])
	        amin_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2117, [2])
	        amax_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2117, [2]);  view_2117 = None
	        full_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_135, full_270);  amin_135 = full_270 = None
	        full_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_135, full_271);  amax_135 = full_271 = None
	        sub_6183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_135, minimum_135);  maximum_135 = None
	        div_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6183, 255.0);  sub_6183 = None
	        clamp_min_405: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_270, 1.1920928955078125e-07);  div_270 = None
	        div_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_135, clamp_min_405);  minimum_135 = None
	        round_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_271);  div_271 = None
	        sub_6189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_271);  round_271 = None
	        clamp_min_406: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6189, -128);  sub_6189 = None
	        clamp_max_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_406, 127);  clamp_min_406 = None
	        _assert_tensor_metadata_1217 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_405, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1217 = None
	        _assert_tensor_metadata_1218 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_270, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1218 = None
	        convert_element_type_810: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_270, torch.int8);  clamp_max_270 = None
	        view_2118: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2116, [sym_size_int, 1500, 1280]);  view_2116 = None
	        view_2119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_405, [sym_size_int, 1500, 1])
	        view_2120: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_810, [sym_size_int, 1500, 1])
	        reciprocal_135: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2119);  view_2119 = None
	        mul_13129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_135, 1.0);  reciprocal_135 = None
	        mul_13132: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2118, mul_13129);  view_2118 = mul_13129 = None
	        round_272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13132);  mul_13132 = None
	        add_20779: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_272, view_2120);  round_272 = view_2120 = None
	        clamp_min_407: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20779, -128);  add_20779 = None
	        clamp_max_271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_407, 127);  clamp_min_407 = None
	        view_2121: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_271, [sym_size_int, 1500, 1280]);  clamp_max_271 = None
	        _assert_tensor_metadata_1219 = torch.ops.aten._assert_tensor_metadata.default(view_2121, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1219 = None
	        convert_element_type_811: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2121, torch.int8);  view_2121 = None
	        view_2122: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_811, [sym_size_int, 1500, 1280]);  convert_element_type_811 = None
	        view_2123: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_405, [sym_size_int, 1500, 1]);  clamp_min_405 = None
	        view_2124: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_810, [sym_size_int, 1500, 1]);  convert_element_type_810 = None
	        _assert_tensor_metadata_1220 = torch.ops.aten._assert_tensor_metadata.default(view_2122, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1220 = None
	        convert_element_type_812: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2122, torch.float32);  view_2122 = None
	        _assert_tensor_metadata_1221 = torch.ops.aten._assert_tensor_metadata.default(view_2124, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1221 = None
	        convert_element_type_813: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2124, torch.float32);  view_2124 = None
	        sub_6209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_812, convert_element_type_813);  convert_element_type_812 = convert_element_type_813 = None
	        mul_13154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6209, view_2123);  sub_6209 = view_2123 = None
	        view_2125: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13154, [sym_size_int, 1500, 1280]);  mul_13154 = None
	        _assert_tensor_metadata_1222 = torch.ops.aten._assert_tensor_metadata.default(view_2125, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1222 = None
	        view_2126: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg613_1, [1280, 40, 32]);  arg613_1 = None
	        view_2127: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg614_1, [1280, 40, 1]);  arg614_1 = None
	        view_2128: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg615_1, [1280, 40, 1]);  arg615_1 = None
	        _assert_tensor_metadata_1223 = torch.ops.aten._assert_tensor_metadata.default(view_2126, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1223 = None
	        convert_element_type_814: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2126, torch.float32);  view_2126 = None
	        _assert_tensor_metadata_1224 = torch.ops.aten._assert_tensor_metadata.default(view_2128, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1224 = None
	        convert_element_type_815: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2128, torch.float32);  view_2128 = None
	        sub_6213: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_814, convert_element_type_815);  convert_element_type_814 = convert_element_type_815 = None
	        mul_13159: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6213, view_2127);  sub_6213 = view_2127 = None
	        view_2129: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13159, [1280, 1280]);  mul_13159 = None
	        _assert_tensor_metadata_1225 = torch.ops.aten._assert_tensor_metadata.default(view_2129, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1225 = None
	        mul_13164: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2130: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2125, [mul_13164, 1280]);  view_2125 = mul_13164 = None
	        permute_228: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2129, [1, 0]);  view_2129 = None
	        addmm_112: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg612_1, view_2130, permute_228);  arg612_1 = view_2130 = permute_228 = None
	        view_2131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_112, [sym_size_int, 1500, 1280]);  addmm_112 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2131);  view_2131 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_20842: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_20222, clone_181);  add_20222 = clone_181 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_20842, memory_format = torch.contiguous_format)
	        var_mean_45 = torch.ops.aten.var_mean.correction(clone_182, [2], correction = 0, keepdim = True)
	        getitem_182: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_45[0]
	        getitem_183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_45[1];  var_mean_45 = None
	        add_20847: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_182, 1e-05);  getitem_182 = None
	        rsqrt_45: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_20847);  add_20847 = None
	        sub_6219: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_182, getitem_183);  clone_182 = getitem_183 = None
	        mul_13175: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6219, rsqrt_45);  sub_6219 = rsqrt_45 = None
	        mul_13176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13175, arg616_1);  mul_13175 = arg616_1 = None
	        add_20848: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13176, arg617_1);  mul_13176 = arg617_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2132: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_20848, [sym_size_int, 1500, 1280])
	        amin_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2132, [2])
	        amax_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2132, [2]);  view_2132 = None
	        full_272: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_136, full_272);  amin_136 = full_272 = None
	        full_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_136, full_273);  amax_136 = full_273 = None
	        sub_6230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_136, minimum_136);  maximum_136 = None
	        div_272: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6230, 255.0);  sub_6230 = None
	        clamp_min_408: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_272, 1.1920928955078125e-07);  div_272 = None
	        div_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_136, clamp_min_408);  minimum_136 = None
	        round_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_273);  div_273 = None
	        sub_6236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_273);  round_273 = None
	        clamp_min_409: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6236, -128);  sub_6236 = None
	        clamp_max_272: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_409, 127);  clamp_min_409 = None
	        _assert_tensor_metadata_1226 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_408, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1226 = None
	        _assert_tensor_metadata_1227 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_272, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1227 = None
	        convert_element_type_816: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_272, torch.int8);  clamp_max_272 = None
	        view_2133: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_20848, [sym_size_int, 1500, 1280]);  add_20848 = None
	        view_2134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_408, [sym_size_int, 1500, 1])
	        view_2135: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_816, [sym_size_int, 1500, 1])
	        reciprocal_136: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2134);  view_2134 = None
	        mul_13224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_136, 1.0);  reciprocal_136 = None
	        mul_13227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2133, mul_13224);  view_2133 = mul_13224 = None
	        round_274: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13227);  mul_13227 = None
	        add_20935: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_274, view_2135);  round_274 = view_2135 = None
	        clamp_min_410: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20935, -128);  add_20935 = None
	        clamp_max_273: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_410, 127);  clamp_min_410 = None
	        view_2136: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_273, [sym_size_int, 1500, 1280]);  clamp_max_273 = None
	        _assert_tensor_metadata_1228 = torch.ops.aten._assert_tensor_metadata.default(view_2136, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1228 = None
	        convert_element_type_817: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2136, torch.int8);  view_2136 = None
	        view_2137: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_817, [sym_size_int, 1500, 1280]);  convert_element_type_817 = None
	        view_2138: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_408, [sym_size_int, 1500, 1]);  clamp_min_408 = None
	        view_2139: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_816, [sym_size_int, 1500, 1]);  convert_element_type_816 = None
	        _assert_tensor_metadata_1229 = torch.ops.aten._assert_tensor_metadata.default(view_2137, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1229 = None
	        convert_element_type_818: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2137, torch.float32);  view_2137 = None
	        _assert_tensor_metadata_1230 = torch.ops.aten._assert_tensor_metadata.default(view_2139, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1230 = None
	        convert_element_type_819: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2139, torch.float32);  view_2139 = None
	        sub_6256: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_818, convert_element_type_819);  convert_element_type_818 = convert_element_type_819 = None
	        mul_13249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6256, view_2138);  sub_6256 = view_2138 = None
	        view_2140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13249, [sym_size_int, 1500, 1280]);  mul_13249 = None
	        _assert_tensor_metadata_1231 = torch.ops.aten._assert_tensor_metadata.default(view_2140, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1231 = None
	        view_2141: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg619_1, [5120, 40, 32]);  arg619_1 = None
	        view_2142: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg620_1, [5120, 40, 1]);  arg620_1 = None
	        view_2143: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg621_1, [5120, 40, 1]);  arg621_1 = None
	        _assert_tensor_metadata_1232 = torch.ops.aten._assert_tensor_metadata.default(view_2141, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1232 = None
	        convert_element_type_820: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2141, torch.float32);  view_2141 = None
	        _assert_tensor_metadata_1233 = torch.ops.aten._assert_tensor_metadata.default(view_2143, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1233 = None
	        convert_element_type_821: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2143, torch.float32);  view_2143 = None
	        sub_6260: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_820, convert_element_type_821);  convert_element_type_820 = convert_element_type_821 = None
	        mul_13254: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6260, view_2142);  sub_6260 = view_2142 = None
	        view_2144: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13254, [5120, 1280]);  mul_13254 = None
	        _assert_tensor_metadata_1234 = torch.ops.aten._assert_tensor_metadata.default(view_2144, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1234 = None
	        mul_13259: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2145: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2140, [mul_13259, 1280]);  view_2140 = mul_13259 = None
	        permute_229: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2144, [1, 0]);  view_2144 = None
	        addmm_113: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg618_1, view_2145, permute_229);  arg618_1 = view_2145 = permute_229 = None
	        view_2146: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_113, [sym_size_int, 1500, 5120]);  addmm_113 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_13266: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2146, 0.5)
	        mul_13267: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2146, 0.7071067811865476);  view_2146 = None
	        erf_24: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_13267);  mul_13267 = None
	        add_20994: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_24, 1);  erf_24 = None
	        mul_13268: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13266, add_20994);  mul_13266 = add_20994 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_183: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_13268);  mul_13268 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2147: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_183, [sym_size_int, 1500, 5120])
	        amin_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2147, [2])
	        amax_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2147, [2]);  view_2147 = None
	        full_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_137, full_274);  amin_137 = full_274 = None
	        full_275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_137, full_275);  amax_137 = full_275 = None
	        sub_6273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_137, minimum_137);  maximum_137 = None
	        div_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6273, 255.0);  sub_6273 = None
	        clamp_min_411: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_274, 1.1920928955078125e-07);  div_274 = None
	        div_275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_137, clamp_min_411);  minimum_137 = None
	        round_275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_275);  div_275 = None
	        sub_6279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_275);  round_275 = None
	        clamp_min_412: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6279, -128);  sub_6279 = None
	        clamp_max_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_412, 127);  clamp_min_412 = None
	        _assert_tensor_metadata_1235 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_411, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1235 = None
	        _assert_tensor_metadata_1236 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_274, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1236 = None
	        convert_element_type_822: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_274, torch.int8);  clamp_max_274 = None
	        view_2148: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_183, [sym_size_int, 1500, 5120]);  clone_183 = None
	        view_2149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_411, [sym_size_int, 1500, 1])
	        view_2150: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_822, [sym_size_int, 1500, 1])
	        reciprocal_137: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2149);  view_2149 = None
	        mul_13314: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_137, 1.0);  reciprocal_137 = None
	        mul_13317: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2148, mul_13314);  view_2148 = mul_13314 = None
	        round_276: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_13317);  mul_13317 = None
	        add_21077: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_276, view_2150);  round_276 = view_2150 = None
	        clamp_min_413: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21077, -128);  add_21077 = None
	        clamp_max_275: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_413, 127);  clamp_min_413 = None
	        view_2151: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_275, [sym_size_int, 1500, 5120]);  clamp_max_275 = None
	        _assert_tensor_metadata_1237 = torch.ops.aten._assert_tensor_metadata.default(view_2151, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1237 = None
	        convert_element_type_823: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2151, torch.int8);  view_2151 = None
	        view_2152: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_823, [sym_size_int, 1500, 5120]);  convert_element_type_823 = None
	        view_2153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_411, [sym_size_int, 1500, 1]);  clamp_min_411 = None
	        view_2154: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_822, [sym_size_int, 1500, 1]);  convert_element_type_822 = None
	        _assert_tensor_metadata_1238 = torch.ops.aten._assert_tensor_metadata.default(view_2152, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1238 = None
	        convert_element_type_824: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2152, torch.float32);  view_2152 = None
	        _assert_tensor_metadata_1239 = torch.ops.aten._assert_tensor_metadata.default(view_2154, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1239 = None
	        convert_element_type_825: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2154, torch.float32);  view_2154 = None
	        sub_6299: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_824, convert_element_type_825);  convert_element_type_824 = convert_element_type_825 = None
	        mul_13339: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6299, view_2153);  sub_6299 = view_2153 = None
	        view_2155: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_13339, [sym_size_int, 1500, 5120]);  mul_13339 = None
	        _assert_tensor_metadata_1240 = torch.ops.aten._assert_tensor_metadata.default(view_2155, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1240 = None
	        view_2156: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg623_1, [1280, 160, 32]);  arg623_1 = None
	        view_2157: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg624_1, [1280, 160, 1]);  arg624_1 = None
	        view_2158: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg625_1, [1280, 160, 1]);  arg625_1 = None
	        _assert_tensor_metadata_1241 = torch.ops.aten._assert_tensor_metadata.default(view_2156, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1241 = None
	        convert_element_type_826: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2156, torch.float32);  view_2156 = None
	        _assert_tensor_metadata_1242 = torch.ops.aten._assert_tensor_metadata.default(view_2158, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1242 = None
	        convert_element_type_827: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2158, torch.float32);  view_2158 = None
	        sub_6303: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_826, convert_element_type_827);  convert_element_type_826 = convert_element_type_827 = None
	        mul_13344: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6303, view_2157);  sub_6303 = view_2157 = None
	        view_2159: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_13344, [1280, 5120]);  mul_13344 = None
	        _assert_tensor_metadata_1243 = torch.ops.aten._assert_tensor_metadata.default(view_2159, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1243 = None
	        mul_13349: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2160: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2155, [mul_13349, 5120]);  view_2155 = mul_13349 = None
	        permute_230: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2159, [1, 0]);  view_2159 = None
	        addmm_114: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg622_1, view_2160, permute_230);  arg622_1 = view_2160 = permute_230 = None
	        view_2161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_114, [sym_size_int, 1500, 1280]);  addmm_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_184: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2161);  view_2161 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_21140: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_20842, clone_184);  add_20842 = clone_184 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_21140, memory_format = torch.contiguous_format)
	        var_mean_46 = torch.ops.aten.var_mean.correction(clone_185, [2], correction = 0, keepdim = True)
	        getitem_184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_46[0]
	        getitem_185: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_46[1];  var_mean_46 = None
	        add_21145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_184, 1e-05);  getitem_184 = None
	        rsqrt_46: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_21145);  add_21145 = None
	        sub_6309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_185, getitem_185);  clone_185 = getitem_185 = None
	        mul_13360: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6309, rsqrt_46);  sub_6309 = rsqrt_46 = None
	        mul_13361: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13360, arg626_1);  mul_13360 = arg626_1 = None
	        add_21146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13361, arg627_1);  mul_13361 = arg627_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2162: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_21146, [sym_size_int, 1500, 1280])
	        amin_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2162, [2])
	        amax_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2162, [2]);  view_2162 = None
	        full_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_138, full_276);  amin_138 = full_276 = None
	        full_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_138, full_277);  amax_138 = full_277 = None
	        sub_6320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_138, minimum_138);  maximum_138 = None
	        div_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6320, 255.0);  sub_6320 = None
	        clamp_min_414: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_276, 1.1920928955078125e-07);  div_276 = None
	        div_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_138, clamp_min_414);  minimum_138 = None
	        round_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_277);  div_277 = None
	        sub_6326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_277);  round_277 = None
	        clamp_min_415: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6326, -128);  sub_6326 = None
	        clamp_max_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_415, 127);  clamp_min_415 = None
	        _assert_tensor_metadata_1244 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_414, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1244 = None
	        _assert_tensor_metadata_1245 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_276, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1245 = None
	        convert_element_type_828: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_276, torch.int8);  clamp_max_276 = None
	        view_2163: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_21146, [sym_size_int, 1500, 1280])
	        view_2164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_414, [sym_size_int, 1500, 1])
	        view_2165: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_828, [sym_size_int, 1500, 1])
	        reciprocal_138: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2164);  view_2164 = None
	        mul_13409: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_138, 1.0);  reciprocal_138 = None
	        mul_13412: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2163, mul_13409);  view_2163 = mul_13409 = None
	        round_278: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13412);  mul_13412 = None
	        add_21233: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_278, view_2165);  round_278 = view_2165 = None
	        clamp_min_416: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21233, -128);  add_21233 = None
	        clamp_max_277: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_416, 127);  clamp_min_416 = None
	        view_2166: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_277, [sym_size_int, 1500, 1280]);  clamp_max_277 = None
	        _assert_tensor_metadata_1246 = torch.ops.aten._assert_tensor_metadata.default(view_2166, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1246 = None
	        convert_element_type_829: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2166, torch.int8);  view_2166 = None
	        view_2167: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_829, [sym_size_int, 1500, 1280]);  convert_element_type_829 = None
	        view_2168: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_414, [sym_size_int, 1500, 1]);  clamp_min_414 = None
	        view_2169: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_828, [sym_size_int, 1500, 1]);  convert_element_type_828 = None
	        _assert_tensor_metadata_1247 = torch.ops.aten._assert_tensor_metadata.default(view_2167, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1247 = None
	        convert_element_type_830: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2167, torch.float32);  view_2167 = None
	        _assert_tensor_metadata_1248 = torch.ops.aten._assert_tensor_metadata.default(view_2169, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1248 = None
	        convert_element_type_831: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2169, torch.float32);  view_2169 = None
	        sub_6346: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_830, convert_element_type_831);  convert_element_type_830 = convert_element_type_831 = None
	        mul_13434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6346, view_2168);  sub_6346 = view_2168 = None
	        view_2170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13434, [sym_size_int, 1500, 1280]);  mul_13434 = None
	        _assert_tensor_metadata_1249 = torch.ops.aten._assert_tensor_metadata.default(view_2170, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1249 = None
	        view_2171: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg629_1, [1280, 40, 32]);  arg629_1 = None
	        view_2172: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg630_1, [1280, 40, 1]);  arg630_1 = None
	        view_2173: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg631_1, [1280, 40, 1]);  arg631_1 = None
	        _assert_tensor_metadata_1250 = torch.ops.aten._assert_tensor_metadata.default(view_2171, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1250 = None
	        convert_element_type_832: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2171, torch.float32);  view_2171 = None
	        _assert_tensor_metadata_1251 = torch.ops.aten._assert_tensor_metadata.default(view_2173, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1251 = None
	        convert_element_type_833: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2173, torch.float32);  view_2173 = None
	        sub_6350: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_832, convert_element_type_833);  convert_element_type_832 = convert_element_type_833 = None
	        mul_13439: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6350, view_2172);  sub_6350 = view_2172 = None
	        view_2174: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13439, [1280, 1280]);  mul_13439 = None
	        _assert_tensor_metadata_1252 = torch.ops.aten._assert_tensor_metadata.default(view_2174, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1252 = None
	        mul_13444: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2175: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2170, [mul_13444, 1280]);  view_2170 = mul_13444 = None
	        permute_231: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2174, [1, 0]);  view_2174 = None
	        addmm_115: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg628_1, view_2175, permute_231);  arg628_1 = view_2175 = permute_231 = None
	        view_2176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_115, [sym_size_int, 1500, 1280]);  addmm_115 = None
	        mul_13451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2176, 0.125);  view_2176 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2177: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_13451, [sym_size_int, 1500, 20, 64]);  mul_13451 = None
	        permute_232: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2177, [0, 2, 1, 3]);  view_2177 = None
	        clone_186: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_232, memory_format = torch.contiguous_format);  permute_232 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2178: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_21146, [sym_size_int, 1500, 1280])
	        amin_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2178, [2])
	        amax_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2178, [2]);  view_2178 = None
	        full_278: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_139, full_278);  amin_139 = full_278 = None
	        full_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_139, full_279);  amax_139 = full_279 = None
	        sub_6365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_139, minimum_139);  maximum_139 = None
	        div_278: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6365, 255.0);  sub_6365 = None
	        clamp_min_417: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_278, 1.1920928955078125e-07);  div_278 = None
	        div_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_139, clamp_min_417);  minimum_139 = None
	        round_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_279);  div_279 = None
	        sub_6371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_279);  round_279 = None
	        clamp_min_418: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6371, -128);  sub_6371 = None
	        clamp_max_278: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_418, 127);  clamp_min_418 = None
	        _assert_tensor_metadata_1253 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_417, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1253 = None
	        _assert_tensor_metadata_1254 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_278, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1254 = None
	        convert_element_type_834: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_278, torch.int8);  clamp_max_278 = None
	        view_2179: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_21146, [sym_size_int, 1500, 1280])
	        view_2180: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_417, [sym_size_int, 1500, 1])
	        view_2181: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_834, [sym_size_int, 1500, 1])
	        reciprocal_139: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2180);  view_2180 = None
	        mul_13505: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_139, 1.0);  reciprocal_139 = None
	        mul_13508: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2179, mul_13505);  view_2179 = mul_13505 = None
	        round_280: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13508);  mul_13508 = None
	        add_21385: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_280, view_2181);  round_280 = view_2181 = None
	        clamp_min_419: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21385, -128);  add_21385 = None
	        clamp_max_279: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_419, 127);  clamp_min_419 = None
	        view_2182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_279, [sym_size_int, 1500, 1280]);  clamp_max_279 = None
	        _assert_tensor_metadata_1255 = torch.ops.aten._assert_tensor_metadata.default(view_2182, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1255 = None
	        convert_element_type_835: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2182, torch.int8);  view_2182 = None
	        view_2183: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_835, [sym_size_int, 1500, 1280]);  convert_element_type_835 = None
	        view_2184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_417, [sym_size_int, 1500, 1]);  clamp_min_417 = None
	        view_2185: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_834, [sym_size_int, 1500, 1]);  convert_element_type_834 = None
	        _assert_tensor_metadata_1256 = torch.ops.aten._assert_tensor_metadata.default(view_2183, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1256 = None
	        convert_element_type_836: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2183, torch.float32);  view_2183 = None
	        _assert_tensor_metadata_1257 = torch.ops.aten._assert_tensor_metadata.default(view_2185, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1257 = None
	        convert_element_type_837: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2185, torch.float32);  view_2185 = None
	        sub_6391: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_836, convert_element_type_837);  convert_element_type_836 = convert_element_type_837 = None
	        mul_13530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6391, view_2184);  sub_6391 = view_2184 = None
	        view_2186: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13530, [sym_size_int, 1500, 1280]);  mul_13530 = None
	        _assert_tensor_metadata_1258 = torch.ops.aten._assert_tensor_metadata.default(view_2186, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1258 = None
	        view_2187: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg632_1, [1280, 40, 32]);  arg632_1 = None
	        view_2188: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg633_1, [1280, 40, 1]);  arg633_1 = None
	        view_2189: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg634_1, [1280, 40, 1]);  arg634_1 = None
	        _assert_tensor_metadata_1259 = torch.ops.aten._assert_tensor_metadata.default(view_2187, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1259 = None
	        convert_element_type_838: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2187, torch.float32);  view_2187 = None
	        _assert_tensor_metadata_1260 = torch.ops.aten._assert_tensor_metadata.default(view_2189, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1260 = None
	        convert_element_type_839: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2189, torch.float32);  view_2189 = None
	        sub_6395: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_838, convert_element_type_839);  convert_element_type_838 = convert_element_type_839 = None
	        mul_13535: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6395, view_2188);  sub_6395 = view_2188 = None
	        view_2190: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13535, [1280, 1280]);  mul_13535 = None
	        _assert_tensor_metadata_1261 = torch.ops.aten._assert_tensor_metadata.default(view_2190, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1261 = None
	        permute_233: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2190, [1, 0]);  view_2190 = None
	        mul_13538: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2191: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2186, [mul_13538, 1280]);  view_2186 = mul_13538 = None
	        mm_23: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2191, permute_233);  view_2191 = permute_233 = None
	        view_2192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_23, [sym_size_int, 1500, 1280]);  mm_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2193: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2192, [sym_size_int, -1, 20, 64]);  view_2192 = None
	        permute_234: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2193, [0, 2, 1, 3]);  view_2193 = None
	        clone_187: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_234, memory_format = torch.contiguous_format);  permute_234 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_21146, [sym_size_int, 1500, 1280])
	        amin_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2194, [2])
	        amax_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2194, [2]);  view_2194 = None
	        full_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_140, full_280);  amin_140 = full_280 = None
	        full_281: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_140, full_281);  amax_140 = full_281 = None
	        sub_6409: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_140, minimum_140);  maximum_140 = None
	        div_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6409, 255.0);  sub_6409 = None
	        clamp_min_420: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_280, 1.1920928955078125e-07);  div_280 = None
	        div_281: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_140, clamp_min_420);  minimum_140 = None
	        round_281: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_281);  div_281 = None
	        sub_6415: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_281);  round_281 = None
	        clamp_min_421: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6415, -128);  sub_6415 = None
	        clamp_max_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_421, 127);  clamp_min_421 = None
	        _assert_tensor_metadata_1262 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_420, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1262 = None
	        _assert_tensor_metadata_1263 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_280, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1263 = None
	        convert_element_type_840: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_280, torch.int8);  clamp_max_280 = None
	        view_2195: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_21146, [sym_size_int, 1500, 1280]);  add_21146 = None
	        view_2196: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_420, [sym_size_int, 1500, 1])
	        view_2197: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_840, [sym_size_int, 1500, 1])
	        reciprocal_140: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2196);  view_2196 = None
	        mul_13604: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_140, 1.0);  reciprocal_140 = None
	        mul_13607: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2195, mul_13604);  view_2195 = mul_13604 = None
	        round_282: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13607);  mul_13607 = None
	        add_21533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_282, view_2197);  round_282 = view_2197 = None
	        clamp_min_422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21533, -128);  add_21533 = None
	        clamp_max_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_422, 127);  clamp_min_422 = None
	        view_2198: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_281, [sym_size_int, 1500, 1280]);  clamp_max_281 = None
	        _assert_tensor_metadata_1264 = torch.ops.aten._assert_tensor_metadata.default(view_2198, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1264 = None
	        convert_element_type_841: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2198, torch.int8);  view_2198 = None
	        view_2199: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_841, [sym_size_int, 1500, 1280]);  convert_element_type_841 = None
	        view_2200: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_420, [sym_size_int, 1500, 1]);  clamp_min_420 = None
	        view_2201: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_840, [sym_size_int, 1500, 1]);  convert_element_type_840 = None
	        _assert_tensor_metadata_1265 = torch.ops.aten._assert_tensor_metadata.default(view_2199, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1265 = None
	        convert_element_type_842: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2199, torch.float32);  view_2199 = None
	        _assert_tensor_metadata_1266 = torch.ops.aten._assert_tensor_metadata.default(view_2201, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1266 = None
	        convert_element_type_843: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2201, torch.float32);  view_2201 = None
	        sub_6435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_842, convert_element_type_843);  convert_element_type_842 = convert_element_type_843 = None
	        mul_13629: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6435, view_2200);  sub_6435 = view_2200 = None
	        view_2202: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13629, [sym_size_int, 1500, 1280]);  mul_13629 = None
	        _assert_tensor_metadata_1267 = torch.ops.aten._assert_tensor_metadata.default(view_2202, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1267 = None
	        view_2203: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg636_1, [1280, 40, 32]);  arg636_1 = None
	        view_2204: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg637_1, [1280, 40, 1]);  arg637_1 = None
	        view_2205: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg638_1, [1280, 40, 1]);  arg638_1 = None
	        _assert_tensor_metadata_1268 = torch.ops.aten._assert_tensor_metadata.default(view_2203, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1268 = None
	        convert_element_type_844: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2203, torch.float32);  view_2203 = None
	        _assert_tensor_metadata_1269 = torch.ops.aten._assert_tensor_metadata.default(view_2205, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1269 = None
	        convert_element_type_845: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2205, torch.float32);  view_2205 = None
	        sub_6439: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_844, convert_element_type_845);  convert_element_type_844 = convert_element_type_845 = None
	        mul_13634: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6439, view_2204);  sub_6439 = view_2204 = None
	        view_2206: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13634, [1280, 1280]);  mul_13634 = None
	        _assert_tensor_metadata_1270 = torch.ops.aten._assert_tensor_metadata.default(view_2206, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1270 = None
	        mul_13639: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2207: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2202, [mul_13639, 1280]);  view_2202 = mul_13639 = None
	        permute_235: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2206, [1, 0]);  view_2206 = None
	        addmm_116: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg635_1, view_2207, permute_235);  arg635_1 = view_2207 = permute_235 = None
	        view_2208: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_116, [sym_size_int, 1500, 1280]);  addmm_116 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2209: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2208, [sym_size_int, -1, 20, 64]);  view_2208 = None
	        permute_236: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2209, [0, 2, 1, 3]);  view_2209 = None
	        clone_188: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_236, memory_format = torch.contiguous_format);  permute_236 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_23 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_186, clone_187, clone_188, None, False, scale = 1.0);  clone_186 = clone_187 = clone_188 = None
	        getitem_186: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_23[0];  _scaled_dot_product_efficient_attention_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_237: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_186, [0, 2, 1, 3]);  getitem_186 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2210: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_237, [sym_size_int, 1500, -1]);  permute_237 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2211: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2210, [sym_size_int, 1500, 1280])
	        amin_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2211, [2])
	        amax_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2211, [2]);  view_2211 = None
	        full_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_141, full_282);  amin_141 = full_282 = None
	        full_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_141, full_283);  amax_141 = full_283 = None
	        sub_6457: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_141, minimum_141);  maximum_141 = None
	        div_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6457, 255.0);  sub_6457 = None
	        clamp_min_423: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_282, 1.1920928955078125e-07);  div_282 = None
	        div_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_141, clamp_min_423);  minimum_141 = None
	        round_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_283);  div_283 = None
	        sub_6463: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_283);  round_283 = None
	        clamp_min_424: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6463, -128);  sub_6463 = None
	        clamp_max_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_424, 127);  clamp_min_424 = None
	        _assert_tensor_metadata_1271 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_423, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1271 = None
	        _assert_tensor_metadata_1272 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_282, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1272 = None
	        convert_element_type_846: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_282, torch.int8);  clamp_max_282 = None
	        view_2212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2210, [sym_size_int, 1500, 1280]);  view_2210 = None
	        view_2213: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_423, [sym_size_int, 1500, 1])
	        view_2214: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_846, [sym_size_int, 1500, 1])
	        reciprocal_141: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2213);  view_2213 = None
	        mul_13709: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_141, 1.0);  reciprocal_141 = None
	        mul_13712: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2212, mul_13709);  view_2212 = mul_13709 = None
	        round_284: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13712);  mul_13712 = None
	        add_21697: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_284, view_2214);  round_284 = view_2214 = None
	        clamp_min_425: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21697, -128);  add_21697 = None
	        clamp_max_283: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_425, 127);  clamp_min_425 = None
	        view_2215: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_283, [sym_size_int, 1500, 1280]);  clamp_max_283 = None
	        _assert_tensor_metadata_1273 = torch.ops.aten._assert_tensor_metadata.default(view_2215, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1273 = None
	        convert_element_type_847: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2215, torch.int8);  view_2215 = None
	        view_2216: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_847, [sym_size_int, 1500, 1280]);  convert_element_type_847 = None
	        view_2217: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_423, [sym_size_int, 1500, 1]);  clamp_min_423 = None
	        view_2218: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_846, [sym_size_int, 1500, 1]);  convert_element_type_846 = None
	        _assert_tensor_metadata_1274 = torch.ops.aten._assert_tensor_metadata.default(view_2216, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1274 = None
	        convert_element_type_848: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2216, torch.float32);  view_2216 = None
	        _assert_tensor_metadata_1275 = torch.ops.aten._assert_tensor_metadata.default(view_2218, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1275 = None
	        convert_element_type_849: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2218, torch.float32);  view_2218 = None
	        sub_6483: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_848, convert_element_type_849);  convert_element_type_848 = convert_element_type_849 = None
	        mul_13734: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6483, view_2217);  sub_6483 = view_2217 = None
	        view_2219: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13734, [sym_size_int, 1500, 1280]);  mul_13734 = None
	        _assert_tensor_metadata_1276 = torch.ops.aten._assert_tensor_metadata.default(view_2219, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1276 = None
	        view_2220: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg640_1, [1280, 40, 32]);  arg640_1 = None
	        view_2221: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg641_1, [1280, 40, 1]);  arg641_1 = None
	        view_2222: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg642_1, [1280, 40, 1]);  arg642_1 = None
	        _assert_tensor_metadata_1277 = torch.ops.aten._assert_tensor_metadata.default(view_2220, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1277 = None
	        convert_element_type_850: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2220, torch.float32);  view_2220 = None
	        _assert_tensor_metadata_1278 = torch.ops.aten._assert_tensor_metadata.default(view_2222, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1278 = None
	        convert_element_type_851: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2222, torch.float32);  view_2222 = None
	        sub_6487: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_850, convert_element_type_851);  convert_element_type_850 = convert_element_type_851 = None
	        mul_13739: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6487, view_2221);  sub_6487 = view_2221 = None
	        view_2223: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13739, [1280, 1280]);  mul_13739 = None
	        _assert_tensor_metadata_1279 = torch.ops.aten._assert_tensor_metadata.default(view_2223, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1279 = None
	        mul_13744: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2224: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2219, [mul_13744, 1280]);  view_2219 = mul_13744 = None
	        permute_238: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2223, [1, 0]);  view_2223 = None
	        addmm_117: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg639_1, view_2224, permute_238);  arg639_1 = view_2224 = permute_238 = None
	        view_2225: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_117, [sym_size_int, 1500, 1280]);  addmm_117 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2225);  view_2225 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_21760: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_21140, clone_189);  add_21140 = clone_189 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_190: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_21760, memory_format = torch.contiguous_format)
	        var_mean_47 = torch.ops.aten.var_mean.correction(clone_190, [2], correction = 0, keepdim = True)
	        getitem_190: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_47[0]
	        getitem_191: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_47[1];  var_mean_47 = None
	        add_21765: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_190, 1e-05);  getitem_190 = None
	        rsqrt_47: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_21765);  add_21765 = None
	        sub_6493: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_190, getitem_191);  clone_190 = getitem_191 = None
	        mul_13755: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6493, rsqrt_47);  sub_6493 = rsqrt_47 = None
	        mul_13756: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13755, arg643_1);  mul_13755 = arg643_1 = None
	        add_21766: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13756, arg644_1);  mul_13756 = arg644_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2226: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_21766, [sym_size_int, 1500, 1280])
	        amin_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2226, [2])
	        amax_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2226, [2]);  view_2226 = None
	        full_284: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_142, full_284);  amin_142 = full_284 = None
	        full_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_142, full_285);  amax_142 = full_285 = None
	        sub_6504: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_142, minimum_142);  maximum_142 = None
	        div_284: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6504, 255.0);  sub_6504 = None
	        clamp_min_426: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_284, 1.1920928955078125e-07);  div_284 = None
	        div_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_142, clamp_min_426);  minimum_142 = None
	        round_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_285);  div_285 = None
	        sub_6510: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_285);  round_285 = None
	        clamp_min_427: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6510, -128);  sub_6510 = None
	        clamp_max_284: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_427, 127);  clamp_min_427 = None
	        _assert_tensor_metadata_1280 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_426, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1280 = None
	        _assert_tensor_metadata_1281 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_284, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1281 = None
	        convert_element_type_852: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_284, torch.int8);  clamp_max_284 = None
	        view_2227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_21766, [sym_size_int, 1500, 1280]);  add_21766 = None
	        view_2228: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_426, [sym_size_int, 1500, 1])
	        view_2229: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_852, [sym_size_int, 1500, 1])
	        reciprocal_142: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2228);  view_2228 = None
	        mul_13804: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_142, 1.0);  reciprocal_142 = None
	        mul_13807: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2227, mul_13804);  view_2227 = mul_13804 = None
	        round_286: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13807);  mul_13807 = None
	        add_21853: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_286, view_2229);  round_286 = view_2229 = None
	        clamp_min_428: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21853, -128);  add_21853 = None
	        clamp_max_285: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_428, 127);  clamp_min_428 = None
	        view_2230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_285, [sym_size_int, 1500, 1280]);  clamp_max_285 = None
	        _assert_tensor_metadata_1282 = torch.ops.aten._assert_tensor_metadata.default(view_2230, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1282 = None
	        convert_element_type_853: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2230, torch.int8);  view_2230 = None
	        view_2231: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_853, [sym_size_int, 1500, 1280]);  convert_element_type_853 = None
	        view_2232: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_426, [sym_size_int, 1500, 1]);  clamp_min_426 = None
	        view_2233: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_852, [sym_size_int, 1500, 1]);  convert_element_type_852 = None
	        _assert_tensor_metadata_1283 = torch.ops.aten._assert_tensor_metadata.default(view_2231, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1283 = None
	        convert_element_type_854: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2231, torch.float32);  view_2231 = None
	        _assert_tensor_metadata_1284 = torch.ops.aten._assert_tensor_metadata.default(view_2233, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1284 = None
	        convert_element_type_855: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2233, torch.float32);  view_2233 = None
	        sub_6530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_854, convert_element_type_855);  convert_element_type_854 = convert_element_type_855 = None
	        mul_13829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6530, view_2232);  sub_6530 = view_2232 = None
	        view_2234: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13829, [sym_size_int, 1500, 1280]);  mul_13829 = None
	        _assert_tensor_metadata_1285 = torch.ops.aten._assert_tensor_metadata.default(view_2234, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1285 = None
	        view_2235: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg646_1, [5120, 40, 32]);  arg646_1 = None
	        view_2236: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg647_1, [5120, 40, 1]);  arg647_1 = None
	        view_2237: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg648_1, [5120, 40, 1]);  arg648_1 = None
	        _assert_tensor_metadata_1286 = torch.ops.aten._assert_tensor_metadata.default(view_2235, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1286 = None
	        convert_element_type_856: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2235, torch.float32);  view_2235 = None
	        _assert_tensor_metadata_1287 = torch.ops.aten._assert_tensor_metadata.default(view_2237, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1287 = None
	        convert_element_type_857: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2237, torch.float32);  view_2237 = None
	        sub_6534: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_856, convert_element_type_857);  convert_element_type_856 = convert_element_type_857 = None
	        mul_13834: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6534, view_2236);  sub_6534 = view_2236 = None
	        view_2238: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13834, [5120, 1280]);  mul_13834 = None
	        _assert_tensor_metadata_1288 = torch.ops.aten._assert_tensor_metadata.default(view_2238, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1288 = None
	        mul_13839: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2239: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2234, [mul_13839, 1280]);  view_2234 = mul_13839 = None
	        permute_239: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2238, [1, 0]);  view_2238 = None
	        addmm_118: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg645_1, view_2239, permute_239);  arg645_1 = view_2239 = permute_239 = None
	        view_2240: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_118, [sym_size_int, 1500, 5120]);  addmm_118 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_13846: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2240, 0.5)
	        mul_13847: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2240, 0.7071067811865476);  view_2240 = None
	        erf_25: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_13847);  mul_13847 = None
	        add_21912: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_25, 1);  erf_25 = None
	        mul_13848: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13846, add_21912);  mul_13846 = add_21912 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_191: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_13848);  mul_13848 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2241: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_191, [sym_size_int, 1500, 5120])
	        amin_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2241, [2])
	        amax_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2241, [2]);  view_2241 = None
	        full_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_143, full_286);  amin_143 = full_286 = None
	        full_287: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_143, full_287);  amax_143 = full_287 = None
	        sub_6547: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_143, minimum_143);  maximum_143 = None
	        div_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6547, 255.0);  sub_6547 = None
	        clamp_min_429: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_286, 1.1920928955078125e-07);  div_286 = None
	        div_287: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_143, clamp_min_429);  minimum_143 = None
	        round_287: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_287);  div_287 = None
	        sub_6553: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_287);  round_287 = None
	        clamp_min_430: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6553, -128);  sub_6553 = None
	        clamp_max_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_430, 127);  clamp_min_430 = None
	        _assert_tensor_metadata_1289 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_429, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1289 = None
	        _assert_tensor_metadata_1290 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_286, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1290 = None
	        convert_element_type_858: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_286, torch.int8);  clamp_max_286 = None
	        view_2242: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_191, [sym_size_int, 1500, 5120]);  clone_191 = None
	        view_2243: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_429, [sym_size_int, 1500, 1])
	        view_2244: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_858, [sym_size_int, 1500, 1])
	        reciprocal_143: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2243);  view_2243 = None
	        mul_13894: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_143, 1.0);  reciprocal_143 = None
	        mul_13897: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2242, mul_13894);  view_2242 = mul_13894 = None
	        round_288: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_13897);  mul_13897 = None
	        add_21995: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_288, view_2244);  round_288 = view_2244 = None
	        clamp_min_431: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21995, -128);  add_21995 = None
	        clamp_max_287: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_431, 127);  clamp_min_431 = None
	        view_2245: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_287, [sym_size_int, 1500, 5120]);  clamp_max_287 = None
	        _assert_tensor_metadata_1291 = torch.ops.aten._assert_tensor_metadata.default(view_2245, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1291 = None
	        convert_element_type_859: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2245, torch.int8);  view_2245 = None
	        view_2246: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_859, [sym_size_int, 1500, 5120]);  convert_element_type_859 = None
	        view_2247: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_429, [sym_size_int, 1500, 1]);  clamp_min_429 = None
	        view_2248: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_858, [sym_size_int, 1500, 1]);  convert_element_type_858 = None
	        _assert_tensor_metadata_1292 = torch.ops.aten._assert_tensor_metadata.default(view_2246, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1292 = None
	        convert_element_type_860: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2246, torch.float32);  view_2246 = None
	        _assert_tensor_metadata_1293 = torch.ops.aten._assert_tensor_metadata.default(view_2248, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1293 = None
	        convert_element_type_861: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2248, torch.float32);  view_2248 = None
	        sub_6573: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_860, convert_element_type_861);  convert_element_type_860 = convert_element_type_861 = None
	        mul_13919: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6573, view_2247);  sub_6573 = view_2247 = None
	        view_2249: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_13919, [sym_size_int, 1500, 5120]);  mul_13919 = None
	        _assert_tensor_metadata_1294 = torch.ops.aten._assert_tensor_metadata.default(view_2249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1294 = None
	        view_2250: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg650_1, [1280, 160, 32]);  arg650_1 = None
	        view_2251: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg651_1, [1280, 160, 1]);  arg651_1 = None
	        view_2252: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg652_1, [1280, 160, 1]);  arg652_1 = None
	        _assert_tensor_metadata_1295 = torch.ops.aten._assert_tensor_metadata.default(view_2250, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1295 = None
	        convert_element_type_862: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2250, torch.float32);  view_2250 = None
	        _assert_tensor_metadata_1296 = torch.ops.aten._assert_tensor_metadata.default(view_2252, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1296 = None
	        convert_element_type_863: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2252, torch.float32);  view_2252 = None
	        sub_6577: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_862, convert_element_type_863);  convert_element_type_862 = convert_element_type_863 = None
	        mul_13924: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6577, view_2251);  sub_6577 = view_2251 = None
	        view_2253: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_13924, [1280, 5120]);  mul_13924 = None
	        _assert_tensor_metadata_1297 = torch.ops.aten._assert_tensor_metadata.default(view_2253, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1297 = None
	        mul_13929: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2254: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2249, [mul_13929, 5120]);  view_2249 = mul_13929 = None
	        permute_240: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2253, [1, 0]);  view_2253 = None
	        addmm_119: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg649_1, view_2254, permute_240);  arg649_1 = view_2254 = permute_240 = None
	        view_2255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_119, [sym_size_int, 1500, 1280]);  addmm_119 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2255);  view_2255 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_22058: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_21760, clone_192);  add_21760 = clone_192 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_193: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_22058, memory_format = torch.contiguous_format)
	        var_mean_48 = torch.ops.aten.var_mean.correction(clone_193, [2], correction = 0, keepdim = True)
	        getitem_192: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_48[0]
	        getitem_193: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_48[1];  var_mean_48 = None
	        add_22063: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_192, 1e-05);  getitem_192 = None
	        rsqrt_48: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_22063);  add_22063 = None
	        sub_6583: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_193, getitem_193);  clone_193 = getitem_193 = None
	        mul_13940: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6583, rsqrt_48);  sub_6583 = rsqrt_48 = None
	        mul_13941: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13940, arg653_1);  mul_13940 = arg653_1 = None
	        add_22064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13941, arg654_1);  mul_13941 = arg654_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2256: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22064, [sym_size_int, 1500, 1280])
	        amin_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2256, [2])
	        amax_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2256, [2]);  view_2256 = None
	        full_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_144, full_288);  amin_144 = full_288 = None
	        full_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_144, full_289);  amax_144 = full_289 = None
	        sub_6594: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_144, minimum_144);  maximum_144 = None
	        div_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6594, 255.0);  sub_6594 = None
	        clamp_min_432: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_288, 1.1920928955078125e-07);  div_288 = None
	        div_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_144, clamp_min_432);  minimum_144 = None
	        round_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_289);  div_289 = None
	        sub_6600: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_289);  round_289 = None
	        clamp_min_433: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6600, -128);  sub_6600 = None
	        clamp_max_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_433, 127);  clamp_min_433 = None
	        _assert_tensor_metadata_1298 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_432, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1298 = None
	        _assert_tensor_metadata_1299 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_288, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1299 = None
	        convert_element_type_864: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_288, torch.int8);  clamp_max_288 = None
	        view_2257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22064, [sym_size_int, 1500, 1280])
	        view_2258: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_432, [sym_size_int, 1500, 1])
	        view_2259: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_864, [sym_size_int, 1500, 1])
	        reciprocal_144: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2258);  view_2258 = None
	        mul_13989: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_144, 1.0);  reciprocal_144 = None
	        mul_13992: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2257, mul_13989);  view_2257 = mul_13989 = None
	        round_290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13992);  mul_13992 = None
	        add_22151: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_290, view_2259);  round_290 = view_2259 = None
	        clamp_min_434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22151, -128);  add_22151 = None
	        clamp_max_289: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_434, 127);  clamp_min_434 = None
	        view_2260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_289, [sym_size_int, 1500, 1280]);  clamp_max_289 = None
	        _assert_tensor_metadata_1300 = torch.ops.aten._assert_tensor_metadata.default(view_2260, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1300 = None
	        convert_element_type_865: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2260, torch.int8);  view_2260 = None
	        view_2261: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_865, [sym_size_int, 1500, 1280]);  convert_element_type_865 = None
	        view_2262: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_432, [sym_size_int, 1500, 1]);  clamp_min_432 = None
	        view_2263: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_864, [sym_size_int, 1500, 1]);  convert_element_type_864 = None
	        _assert_tensor_metadata_1301 = torch.ops.aten._assert_tensor_metadata.default(view_2261, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1301 = None
	        convert_element_type_866: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2261, torch.float32);  view_2261 = None
	        _assert_tensor_metadata_1302 = torch.ops.aten._assert_tensor_metadata.default(view_2263, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1302 = None
	        convert_element_type_867: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2263, torch.float32);  view_2263 = None
	        sub_6620: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_866, convert_element_type_867);  convert_element_type_866 = convert_element_type_867 = None
	        mul_14014: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6620, view_2262);  sub_6620 = view_2262 = None
	        view_2264: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14014, [sym_size_int, 1500, 1280]);  mul_14014 = None
	        _assert_tensor_metadata_1303 = torch.ops.aten._assert_tensor_metadata.default(view_2264, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1303 = None
	        view_2265: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg656_1, [1280, 40, 32]);  arg656_1 = None
	        view_2266: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg657_1, [1280, 40, 1]);  arg657_1 = None
	        view_2267: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg658_1, [1280, 40, 1]);  arg658_1 = None
	        _assert_tensor_metadata_1304 = torch.ops.aten._assert_tensor_metadata.default(view_2265, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1304 = None
	        convert_element_type_868: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2265, torch.float32);  view_2265 = None
	        _assert_tensor_metadata_1305 = torch.ops.aten._assert_tensor_metadata.default(view_2267, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1305 = None
	        convert_element_type_869: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2267, torch.float32);  view_2267 = None
	        sub_6624: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_868, convert_element_type_869);  convert_element_type_868 = convert_element_type_869 = None
	        mul_14019: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6624, view_2266);  sub_6624 = view_2266 = None
	        view_2268: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14019, [1280, 1280]);  mul_14019 = None
	        _assert_tensor_metadata_1306 = torch.ops.aten._assert_tensor_metadata.default(view_2268, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1306 = None
	        mul_14024: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2269: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2264, [mul_14024, 1280]);  view_2264 = mul_14024 = None
	        permute_241: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2268, [1, 0]);  view_2268 = None
	        addmm_120: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg655_1, view_2269, permute_241);  arg655_1 = view_2269 = permute_241 = None
	        view_2270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_120, [sym_size_int, 1500, 1280]);  addmm_120 = None
	        mul_14031: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2270, 0.125);  view_2270 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2271: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_14031, [sym_size_int, 1500, 20, 64]);  mul_14031 = None
	        permute_242: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2271, [0, 2, 1, 3]);  view_2271 = None
	        clone_194: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_242, memory_format = torch.contiguous_format);  permute_242 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22064, [sym_size_int, 1500, 1280])
	        amin_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2272, [2])
	        amax_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2272, [2]);  view_2272 = None
	        full_290: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_145, full_290);  amin_145 = full_290 = None
	        full_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_145, full_291);  amax_145 = full_291 = None
	        sub_6639: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_145, minimum_145);  maximum_145 = None
	        div_290: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6639, 255.0);  sub_6639 = None
	        clamp_min_435: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_290, 1.1920928955078125e-07);  div_290 = None
	        div_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_145, clamp_min_435);  minimum_145 = None
	        round_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_291);  div_291 = None
	        sub_6645: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_291);  round_291 = None
	        clamp_min_436: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6645, -128);  sub_6645 = None
	        clamp_max_290: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_436, 127);  clamp_min_436 = None
	        _assert_tensor_metadata_1307 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_435, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1307 = None
	        _assert_tensor_metadata_1308 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_290, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1308 = None
	        convert_element_type_870: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_290, torch.int8);  clamp_max_290 = None
	        view_2273: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22064, [sym_size_int, 1500, 1280])
	        view_2274: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_435, [sym_size_int, 1500, 1])
	        view_2275: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_870, [sym_size_int, 1500, 1])
	        reciprocal_145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2274);  view_2274 = None
	        mul_14085: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_145, 1.0);  reciprocal_145 = None
	        mul_14088: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2273, mul_14085);  view_2273 = mul_14085 = None
	        round_292: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14088);  mul_14088 = None
	        add_22303: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_292, view_2275);  round_292 = view_2275 = None
	        clamp_min_437: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22303, -128);  add_22303 = None
	        clamp_max_291: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_437, 127);  clamp_min_437 = None
	        view_2276: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_291, [sym_size_int, 1500, 1280]);  clamp_max_291 = None
	        _assert_tensor_metadata_1309 = torch.ops.aten._assert_tensor_metadata.default(view_2276, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1309 = None
	        convert_element_type_871: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2276, torch.int8);  view_2276 = None
	        view_2277: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_871, [sym_size_int, 1500, 1280]);  convert_element_type_871 = None
	        view_2278: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_435, [sym_size_int, 1500, 1]);  clamp_min_435 = None
	        view_2279: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_870, [sym_size_int, 1500, 1]);  convert_element_type_870 = None
	        _assert_tensor_metadata_1310 = torch.ops.aten._assert_tensor_metadata.default(view_2277, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1310 = None
	        convert_element_type_872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2277, torch.float32);  view_2277 = None
	        _assert_tensor_metadata_1311 = torch.ops.aten._assert_tensor_metadata.default(view_2279, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1311 = None
	        convert_element_type_873: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2279, torch.float32);  view_2279 = None
	        sub_6665: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_872, convert_element_type_873);  convert_element_type_872 = convert_element_type_873 = None
	        mul_14110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6665, view_2278);  sub_6665 = view_2278 = None
	        view_2280: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14110, [sym_size_int, 1500, 1280]);  mul_14110 = None
	        _assert_tensor_metadata_1312 = torch.ops.aten._assert_tensor_metadata.default(view_2280, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1312 = None
	        view_2281: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg659_1, [1280, 40, 32]);  arg659_1 = None
	        view_2282: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg660_1, [1280, 40, 1]);  arg660_1 = None
	        view_2283: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg661_1, [1280, 40, 1]);  arg661_1 = None
	        _assert_tensor_metadata_1313 = torch.ops.aten._assert_tensor_metadata.default(view_2281, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1313 = None
	        convert_element_type_874: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2281, torch.float32);  view_2281 = None
	        _assert_tensor_metadata_1314 = torch.ops.aten._assert_tensor_metadata.default(view_2283, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1314 = None
	        convert_element_type_875: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2283, torch.float32);  view_2283 = None
	        sub_6669: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_874, convert_element_type_875);  convert_element_type_874 = convert_element_type_875 = None
	        mul_14115: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6669, view_2282);  sub_6669 = view_2282 = None
	        view_2284: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14115, [1280, 1280]);  mul_14115 = None
	        _assert_tensor_metadata_1315 = torch.ops.aten._assert_tensor_metadata.default(view_2284, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1315 = None
	        permute_243: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2284, [1, 0]);  view_2284 = None
	        mul_14118: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2285: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2280, [mul_14118, 1280]);  view_2280 = mul_14118 = None
	        mm_24: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2285, permute_243);  view_2285 = permute_243 = None
	        view_2286: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_24, [sym_size_int, 1500, 1280]);  mm_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2287: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2286, [sym_size_int, -1, 20, 64]);  view_2286 = None
	        permute_244: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2287, [0, 2, 1, 3]);  view_2287 = None
	        clone_195: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_244, memory_format = torch.contiguous_format);  permute_244 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2288: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22064, [sym_size_int, 1500, 1280])
	        amin_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2288, [2])
	        amax_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2288, [2]);  view_2288 = None
	        full_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_146, full_292);  amin_146 = full_292 = None
	        full_293: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_146, full_293);  amax_146 = full_293 = None
	        sub_6683: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_146, minimum_146);  maximum_146 = None
	        div_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6683, 255.0);  sub_6683 = None
	        clamp_min_438: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_292, 1.1920928955078125e-07);  div_292 = None
	        div_293: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_146, clamp_min_438);  minimum_146 = None
	        round_293: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_293);  div_293 = None
	        sub_6689: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_293);  round_293 = None
	        clamp_min_439: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6689, -128);  sub_6689 = None
	        clamp_max_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_439, 127);  clamp_min_439 = None
	        _assert_tensor_metadata_1316 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_438, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1316 = None
	        _assert_tensor_metadata_1317 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_292, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1317 = None
	        convert_element_type_876: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_292, torch.int8);  clamp_max_292 = None
	        view_2289: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22064, [sym_size_int, 1500, 1280]);  add_22064 = None
	        view_2290: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_438, [sym_size_int, 1500, 1])
	        view_2291: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_876, [sym_size_int, 1500, 1])
	        reciprocal_146: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2290);  view_2290 = None
	        mul_14184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_146, 1.0);  reciprocal_146 = None
	        mul_14187: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2289, mul_14184);  view_2289 = mul_14184 = None
	        round_294: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14187);  mul_14187 = None
	        add_22451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_294, view_2291);  round_294 = view_2291 = None
	        clamp_min_440: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22451, -128);  add_22451 = None
	        clamp_max_293: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_440, 127);  clamp_min_440 = None
	        view_2292: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_293, [sym_size_int, 1500, 1280]);  clamp_max_293 = None
	        _assert_tensor_metadata_1318 = torch.ops.aten._assert_tensor_metadata.default(view_2292, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1318 = None
	        convert_element_type_877: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2292, torch.int8);  view_2292 = None
	        view_2293: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_877, [sym_size_int, 1500, 1280]);  convert_element_type_877 = None
	        view_2294: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_438, [sym_size_int, 1500, 1]);  clamp_min_438 = None
	        view_2295: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_876, [sym_size_int, 1500, 1]);  convert_element_type_876 = None
	        _assert_tensor_metadata_1319 = torch.ops.aten._assert_tensor_metadata.default(view_2293, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1319 = None
	        convert_element_type_878: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2293, torch.float32);  view_2293 = None
	        _assert_tensor_metadata_1320 = torch.ops.aten._assert_tensor_metadata.default(view_2295, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1320 = None
	        convert_element_type_879: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2295, torch.float32);  view_2295 = None
	        sub_6709: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_878, convert_element_type_879);  convert_element_type_878 = convert_element_type_879 = None
	        mul_14209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6709, view_2294);  sub_6709 = view_2294 = None
	        view_2296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14209, [sym_size_int, 1500, 1280]);  mul_14209 = None
	        _assert_tensor_metadata_1321 = torch.ops.aten._assert_tensor_metadata.default(view_2296, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1321 = None
	        view_2297: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg663_1, [1280, 40, 32]);  arg663_1 = None
	        view_2298: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg664_1, [1280, 40, 1]);  arg664_1 = None
	        view_2299: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg665_1, [1280, 40, 1]);  arg665_1 = None
	        _assert_tensor_metadata_1322 = torch.ops.aten._assert_tensor_metadata.default(view_2297, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1322 = None
	        convert_element_type_880: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2297, torch.float32);  view_2297 = None
	        _assert_tensor_metadata_1323 = torch.ops.aten._assert_tensor_metadata.default(view_2299, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1323 = None
	        convert_element_type_881: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2299, torch.float32);  view_2299 = None
	        sub_6713: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_880, convert_element_type_881);  convert_element_type_880 = convert_element_type_881 = None
	        mul_14214: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6713, view_2298);  sub_6713 = view_2298 = None
	        view_2300: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14214, [1280, 1280]);  mul_14214 = None
	        _assert_tensor_metadata_1324 = torch.ops.aten._assert_tensor_metadata.default(view_2300, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1324 = None
	        mul_14219: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2301: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2296, [mul_14219, 1280]);  view_2296 = mul_14219 = None
	        permute_245: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2300, [1, 0]);  view_2300 = None
	        addmm_121: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg662_1, view_2301, permute_245);  arg662_1 = view_2301 = permute_245 = None
	        view_2302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_121, [sym_size_int, 1500, 1280]);  addmm_121 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2303: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2302, [sym_size_int, -1, 20, 64]);  view_2302 = None
	        permute_246: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2303, [0, 2, 1, 3]);  view_2303 = None
	        clone_196: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_246, memory_format = torch.contiguous_format);  permute_246 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_24 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_194, clone_195, clone_196, None, False, scale = 1.0);  clone_194 = clone_195 = clone_196 = None
	        getitem_194: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_24[0];  _scaled_dot_product_efficient_attention_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_247: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_194, [0, 2, 1, 3]);  getitem_194 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2304: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_247, [sym_size_int, 1500, -1]);  permute_247 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2305: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2304, [sym_size_int, 1500, 1280])
	        amin_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2305, [2])
	        amax_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2305, [2]);  view_2305 = None
	        full_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_147, full_294);  amin_147 = full_294 = None
	        full_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_147, full_295);  amax_147 = full_295 = None
	        sub_6731: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_147, minimum_147);  maximum_147 = None
	        div_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6731, 255.0);  sub_6731 = None
	        clamp_min_441: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_294, 1.1920928955078125e-07);  div_294 = None
	        div_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_147, clamp_min_441);  minimum_147 = None
	        round_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_295);  div_295 = None
	        sub_6737: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_295);  round_295 = None
	        clamp_min_442: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6737, -128);  sub_6737 = None
	        clamp_max_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_442, 127);  clamp_min_442 = None
	        _assert_tensor_metadata_1325 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_441, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1325 = None
	        _assert_tensor_metadata_1326 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1326 = None
	        convert_element_type_882: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_294, torch.int8);  clamp_max_294 = None
	        view_2306: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2304, [sym_size_int, 1500, 1280]);  view_2304 = None
	        view_2307: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_441, [sym_size_int, 1500, 1])
	        view_2308: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_882, [sym_size_int, 1500, 1])
	        reciprocal_147: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2307);  view_2307 = None
	        mul_14289: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_147, 1.0);  reciprocal_147 = None
	        mul_14292: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2306, mul_14289);  view_2306 = mul_14289 = None
	        round_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14292);  mul_14292 = None
	        add_22615: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_296, view_2308);  round_296 = view_2308 = None
	        clamp_min_443: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22615, -128);  add_22615 = None
	        clamp_max_295: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_443, 127);  clamp_min_443 = None
	        view_2309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_295, [sym_size_int, 1500, 1280]);  clamp_max_295 = None
	        _assert_tensor_metadata_1327 = torch.ops.aten._assert_tensor_metadata.default(view_2309, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1327 = None
	        convert_element_type_883: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2309, torch.int8);  view_2309 = None
	        view_2310: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_883, [sym_size_int, 1500, 1280]);  convert_element_type_883 = None
	        view_2311: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_441, [sym_size_int, 1500, 1]);  clamp_min_441 = None
	        view_2312: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_882, [sym_size_int, 1500, 1]);  convert_element_type_882 = None
	        _assert_tensor_metadata_1328 = torch.ops.aten._assert_tensor_metadata.default(view_2310, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1328 = None
	        convert_element_type_884: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2310, torch.float32);  view_2310 = None
	        _assert_tensor_metadata_1329 = torch.ops.aten._assert_tensor_metadata.default(view_2312, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1329 = None
	        convert_element_type_885: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2312, torch.float32);  view_2312 = None
	        sub_6757: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_884, convert_element_type_885);  convert_element_type_884 = convert_element_type_885 = None
	        mul_14314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6757, view_2311);  sub_6757 = view_2311 = None
	        view_2313: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14314, [sym_size_int, 1500, 1280]);  mul_14314 = None
	        _assert_tensor_metadata_1330 = torch.ops.aten._assert_tensor_metadata.default(view_2313, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1330 = None
	        view_2314: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg667_1, [1280, 40, 32]);  arg667_1 = None
	        view_2315: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg668_1, [1280, 40, 1]);  arg668_1 = None
	        view_2316: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg669_1, [1280, 40, 1]);  arg669_1 = None
	        _assert_tensor_metadata_1331 = torch.ops.aten._assert_tensor_metadata.default(view_2314, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1331 = None
	        convert_element_type_886: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2314, torch.float32);  view_2314 = None
	        _assert_tensor_metadata_1332 = torch.ops.aten._assert_tensor_metadata.default(view_2316, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1332 = None
	        convert_element_type_887: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2316, torch.float32);  view_2316 = None
	        sub_6761: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_886, convert_element_type_887);  convert_element_type_886 = convert_element_type_887 = None
	        mul_14319: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6761, view_2315);  sub_6761 = view_2315 = None
	        view_2317: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14319, [1280, 1280]);  mul_14319 = None
	        _assert_tensor_metadata_1333 = torch.ops.aten._assert_tensor_metadata.default(view_2317, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1333 = None
	        mul_14324: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2318: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2313, [mul_14324, 1280]);  view_2313 = mul_14324 = None
	        permute_248: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2317, [1, 0]);  view_2317 = None
	        addmm_122: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg666_1, view_2318, permute_248);  arg666_1 = view_2318 = permute_248 = None
	        view_2319: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_122, [sym_size_int, 1500, 1280]);  addmm_122 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_197: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2319);  view_2319 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_22678: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_22058, clone_197);  add_22058 = clone_197 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_198: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_22678, memory_format = torch.contiguous_format)
	        var_mean_49 = torch.ops.aten.var_mean.correction(clone_198, [2], correction = 0, keepdim = True)
	        getitem_198: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_49[0]
	        getitem_199: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_49[1];  var_mean_49 = None
	        add_22683: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_198, 1e-05);  getitem_198 = None
	        rsqrt_49: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_22683);  add_22683 = None
	        sub_6767: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_198, getitem_199);  clone_198 = getitem_199 = None
	        mul_14335: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6767, rsqrt_49);  sub_6767 = rsqrt_49 = None
	        mul_14336: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14335, arg670_1);  mul_14335 = arg670_1 = None
	        add_22684: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_14336, arg671_1);  mul_14336 = arg671_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2320: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22684, [sym_size_int, 1500, 1280])
	        amin_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2320, [2])
	        amax_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2320, [2]);  view_2320 = None
	        full_296: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_148, full_296);  amin_148 = full_296 = None
	        full_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_148, full_297);  amax_148 = full_297 = None
	        sub_6778: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_148, minimum_148);  maximum_148 = None
	        div_296: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6778, 255.0);  sub_6778 = None
	        clamp_min_444: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_296, 1.1920928955078125e-07);  div_296 = None
	        div_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_148, clamp_min_444);  minimum_148 = None
	        round_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_297);  div_297 = None
	        sub_6784: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_297);  round_297 = None
	        clamp_min_445: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6784, -128);  sub_6784 = None
	        clamp_max_296: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_445, 127);  clamp_min_445 = None
	        _assert_tensor_metadata_1334 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_444, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1334 = None
	        _assert_tensor_metadata_1335 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_296, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1335 = None
	        convert_element_type_888: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_296, torch.int8);  clamp_max_296 = None
	        view_2321: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22684, [sym_size_int, 1500, 1280]);  add_22684 = None
	        view_2322: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_444, [sym_size_int, 1500, 1])
	        view_2323: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_888, [sym_size_int, 1500, 1])
	        reciprocal_148: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2322);  view_2322 = None
	        mul_14384: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_148, 1.0);  reciprocal_148 = None
	        mul_14387: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2321, mul_14384);  view_2321 = mul_14384 = None
	        round_298: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14387);  mul_14387 = None
	        add_22771: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_298, view_2323);  round_298 = view_2323 = None
	        clamp_min_446: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22771, -128);  add_22771 = None
	        clamp_max_297: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_446, 127);  clamp_min_446 = None
	        view_2324: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_297, [sym_size_int, 1500, 1280]);  clamp_max_297 = None
	        _assert_tensor_metadata_1336 = torch.ops.aten._assert_tensor_metadata.default(view_2324, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1336 = None
	        convert_element_type_889: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2324, torch.int8);  view_2324 = None
	        view_2325: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_889, [sym_size_int, 1500, 1280]);  convert_element_type_889 = None
	        view_2326: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_444, [sym_size_int, 1500, 1]);  clamp_min_444 = None
	        view_2327: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_888, [sym_size_int, 1500, 1]);  convert_element_type_888 = None
	        _assert_tensor_metadata_1337 = torch.ops.aten._assert_tensor_metadata.default(view_2325, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1337 = None
	        convert_element_type_890: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2325, torch.float32);  view_2325 = None
	        _assert_tensor_metadata_1338 = torch.ops.aten._assert_tensor_metadata.default(view_2327, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1338 = None
	        convert_element_type_891: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2327, torch.float32);  view_2327 = None
	        sub_6804: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_890, convert_element_type_891);  convert_element_type_890 = convert_element_type_891 = None
	        mul_14409: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6804, view_2326);  sub_6804 = view_2326 = None
	        view_2328: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14409, [sym_size_int, 1500, 1280]);  mul_14409 = None
	        _assert_tensor_metadata_1339 = torch.ops.aten._assert_tensor_metadata.default(view_2328, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1339 = None
	        view_2329: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg673_1, [5120, 40, 32]);  arg673_1 = None
	        view_2330: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg674_1, [5120, 40, 1]);  arg674_1 = None
	        view_2331: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg675_1, [5120, 40, 1]);  arg675_1 = None
	        _assert_tensor_metadata_1340 = torch.ops.aten._assert_tensor_metadata.default(view_2329, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1340 = None
	        convert_element_type_892: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2329, torch.float32);  view_2329 = None
	        _assert_tensor_metadata_1341 = torch.ops.aten._assert_tensor_metadata.default(view_2331, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1341 = None
	        convert_element_type_893: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2331, torch.float32);  view_2331 = None
	        sub_6808: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_892, convert_element_type_893);  convert_element_type_892 = convert_element_type_893 = None
	        mul_14414: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6808, view_2330);  sub_6808 = view_2330 = None
	        view_2332: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14414, [5120, 1280]);  mul_14414 = None
	        _assert_tensor_metadata_1342 = torch.ops.aten._assert_tensor_metadata.default(view_2332, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1342 = None
	        mul_14419: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2333: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2328, [mul_14419, 1280]);  view_2328 = mul_14419 = None
	        permute_249: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2332, [1, 0]);  view_2332 = None
	        addmm_123: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg672_1, view_2333, permute_249);  arg672_1 = view_2333 = permute_249 = None
	        view_2334: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_123, [sym_size_int, 1500, 5120]);  addmm_123 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_14426: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2334, 0.5)
	        mul_14427: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2334, 0.7071067811865476);  view_2334 = None
	        erf_26: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_14427);  mul_14427 = None
	        add_22830: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_26, 1);  erf_26 = None
	        mul_14428: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14426, add_22830);  mul_14426 = add_22830 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_199: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_14428);  mul_14428 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2335: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_199, [sym_size_int, 1500, 5120])
	        amin_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2335, [2])
	        amax_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2335, [2]);  view_2335 = None
	        full_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_149, full_298);  amin_149 = full_298 = None
	        full_299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_149, full_299);  amax_149 = full_299 = None
	        sub_6821: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_149, minimum_149);  maximum_149 = None
	        div_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6821, 255.0);  sub_6821 = None
	        clamp_min_447: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_298, 1.1920928955078125e-07);  div_298 = None
	        div_299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_149, clamp_min_447);  minimum_149 = None
	        round_299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_299);  div_299 = None
	        sub_6827: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_299);  round_299 = None
	        clamp_min_448: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6827, -128);  sub_6827 = None
	        clamp_max_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_448, 127);  clamp_min_448 = None
	        _assert_tensor_metadata_1343 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_447, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1343 = None
	        _assert_tensor_metadata_1344 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_298, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1344 = None
	        convert_element_type_894: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_298, torch.int8);  clamp_max_298 = None
	        view_2336: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_199, [sym_size_int, 1500, 5120]);  clone_199 = None
	        view_2337: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_447, [sym_size_int, 1500, 1])
	        view_2338: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_894, [sym_size_int, 1500, 1])
	        reciprocal_149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2337);  view_2337 = None
	        mul_14474: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_149, 1.0);  reciprocal_149 = None
	        mul_14477: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2336, mul_14474);  view_2336 = mul_14474 = None
	        round_300: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_14477);  mul_14477 = None
	        add_22913: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_300, view_2338);  round_300 = view_2338 = None
	        clamp_min_449: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22913, -128);  add_22913 = None
	        clamp_max_299: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_449, 127);  clamp_min_449 = None
	        view_2339: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_299, [sym_size_int, 1500, 5120]);  clamp_max_299 = None
	        _assert_tensor_metadata_1345 = torch.ops.aten._assert_tensor_metadata.default(view_2339, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1345 = None
	        convert_element_type_895: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2339, torch.int8);  view_2339 = None
	        view_2340: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_895, [sym_size_int, 1500, 5120]);  convert_element_type_895 = None
	        view_2341: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_447, [sym_size_int, 1500, 1]);  clamp_min_447 = None
	        view_2342: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_894, [sym_size_int, 1500, 1]);  convert_element_type_894 = None
	        _assert_tensor_metadata_1346 = torch.ops.aten._assert_tensor_metadata.default(view_2340, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1346 = None
	        convert_element_type_896: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2340, torch.float32);  view_2340 = None
	        _assert_tensor_metadata_1347 = torch.ops.aten._assert_tensor_metadata.default(view_2342, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1347 = None
	        convert_element_type_897: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2342, torch.float32);  view_2342 = None
	        sub_6847: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_896, convert_element_type_897);  convert_element_type_896 = convert_element_type_897 = None
	        mul_14499: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6847, view_2341);  sub_6847 = view_2341 = None
	        view_2343: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_14499, [sym_size_int, 1500, 5120]);  mul_14499 = None
	        _assert_tensor_metadata_1348 = torch.ops.aten._assert_tensor_metadata.default(view_2343, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1348 = None
	        view_2344: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg677_1, [1280, 160, 32]);  arg677_1 = None
	        view_2345: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg678_1, [1280, 160, 1]);  arg678_1 = None
	        view_2346: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg679_1, [1280, 160, 1]);  arg679_1 = None
	        _assert_tensor_metadata_1349 = torch.ops.aten._assert_tensor_metadata.default(view_2344, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1349 = None
	        convert_element_type_898: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2344, torch.float32);  view_2344 = None
	        _assert_tensor_metadata_1350 = torch.ops.aten._assert_tensor_metadata.default(view_2346, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1350 = None
	        convert_element_type_899: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2346, torch.float32);  view_2346 = None
	        sub_6851: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_898, convert_element_type_899);  convert_element_type_898 = convert_element_type_899 = None
	        mul_14504: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6851, view_2345);  sub_6851 = view_2345 = None
	        view_2347: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_14504, [1280, 5120]);  mul_14504 = None
	        _assert_tensor_metadata_1351 = torch.ops.aten._assert_tensor_metadata.default(view_2347, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1351 = None
	        mul_14509: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2348: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2343, [mul_14509, 5120]);  view_2343 = mul_14509 = None
	        permute_250: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2347, [1, 0]);  view_2347 = None
	        addmm_124: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg676_1, view_2348, permute_250);  arg676_1 = view_2348 = permute_250 = None
	        view_2349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_124, [sym_size_int, 1500, 1280]);  addmm_124 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2349);  view_2349 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_22976: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_22678, clone_200);  add_22678 = clone_200 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_201: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_22976, memory_format = torch.contiguous_format)
	        var_mean_50 = torch.ops.aten.var_mean.correction(clone_201, [2], correction = 0, keepdim = True)
	        getitem_200: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_50[0]
	        getitem_201: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_50[1];  var_mean_50 = None
	        add_22981: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_200, 1e-05);  getitem_200 = None
	        rsqrt_50: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_22981);  add_22981 = None
	        sub_6857: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_201, getitem_201);  clone_201 = getitem_201 = None
	        mul_14520: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6857, rsqrt_50);  sub_6857 = rsqrt_50 = None
	        mul_14521: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14520, arg680_1);  mul_14520 = arg680_1 = None
	        add_22982: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_14521, arg681_1);  mul_14521 = arg681_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22982, [sym_size_int, 1500, 1280])
	        amin_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2350, [2])
	        amax_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2350, [2]);  view_2350 = None
	        full_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_150, full_300);  amin_150 = full_300 = None
	        full_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_150, full_301);  amax_150 = full_301 = None
	        sub_6868: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_150, minimum_150);  maximum_150 = None
	        div_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6868, 255.0);  sub_6868 = None
	        clamp_min_450: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_300, 1.1920928955078125e-07);  div_300 = None
	        div_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_150, clamp_min_450);  minimum_150 = None
	        round_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_301);  div_301 = None
	        sub_6874: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_301);  round_301 = None
	        clamp_min_451: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6874, -128);  sub_6874 = None
	        clamp_max_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_451, 127);  clamp_min_451 = None
	        _assert_tensor_metadata_1352 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_450, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1352 = None
	        _assert_tensor_metadata_1353 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_300, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1353 = None
	        convert_element_type_900: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_300, torch.int8);  clamp_max_300 = None
	        view_2351: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22982, [sym_size_int, 1500, 1280])
	        view_2352: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_450, [sym_size_int, 1500, 1])
	        view_2353: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_900, [sym_size_int, 1500, 1])
	        reciprocal_150: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2352);  view_2352 = None
	        mul_14569: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_150, 1.0);  reciprocal_150 = None
	        mul_14572: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2351, mul_14569);  view_2351 = mul_14569 = None
	        round_302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14572);  mul_14572 = None
	        add_23069: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_302, view_2353);  round_302 = view_2353 = None
	        clamp_min_452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23069, -128);  add_23069 = None
	        clamp_max_301: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_452, 127);  clamp_min_452 = None
	        view_2354: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_301, [sym_size_int, 1500, 1280]);  clamp_max_301 = None
	        _assert_tensor_metadata_1354 = torch.ops.aten._assert_tensor_metadata.default(view_2354, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1354 = None
	        convert_element_type_901: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2354, torch.int8);  view_2354 = None
	        view_2355: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_901, [sym_size_int, 1500, 1280]);  convert_element_type_901 = None
	        view_2356: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_450, [sym_size_int, 1500, 1]);  clamp_min_450 = None
	        view_2357: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_900, [sym_size_int, 1500, 1]);  convert_element_type_900 = None
	        _assert_tensor_metadata_1355 = torch.ops.aten._assert_tensor_metadata.default(view_2355, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1355 = None
	        convert_element_type_902: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2355, torch.float32);  view_2355 = None
	        _assert_tensor_metadata_1356 = torch.ops.aten._assert_tensor_metadata.default(view_2357, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1356 = None
	        convert_element_type_903: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2357, torch.float32);  view_2357 = None
	        sub_6894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_902, convert_element_type_903);  convert_element_type_902 = convert_element_type_903 = None
	        mul_14594: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6894, view_2356);  sub_6894 = view_2356 = None
	        view_2358: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14594, [sym_size_int, 1500, 1280]);  mul_14594 = None
	        _assert_tensor_metadata_1357 = torch.ops.aten._assert_tensor_metadata.default(view_2358, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1357 = None
	        view_2359: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg683_1, [1280, 40, 32]);  arg683_1 = None
	        view_2360: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg684_1, [1280, 40, 1]);  arg684_1 = None
	        view_2361: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg685_1, [1280, 40, 1]);  arg685_1 = None
	        _assert_tensor_metadata_1358 = torch.ops.aten._assert_tensor_metadata.default(view_2359, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1358 = None
	        convert_element_type_904: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2359, torch.float32);  view_2359 = None
	        _assert_tensor_metadata_1359 = torch.ops.aten._assert_tensor_metadata.default(view_2361, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1359 = None
	        convert_element_type_905: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2361, torch.float32);  view_2361 = None
	        sub_6898: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_904, convert_element_type_905);  convert_element_type_904 = convert_element_type_905 = None
	        mul_14599: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6898, view_2360);  sub_6898 = view_2360 = None
	        view_2362: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14599, [1280, 1280]);  mul_14599 = None
	        _assert_tensor_metadata_1360 = torch.ops.aten._assert_tensor_metadata.default(view_2362, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1360 = None
	        mul_14604: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2363: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2358, [mul_14604, 1280]);  view_2358 = mul_14604 = None
	        permute_251: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2362, [1, 0]);  view_2362 = None
	        addmm_125: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg682_1, view_2363, permute_251);  arg682_1 = view_2363 = permute_251 = None
	        view_2364: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_125, [sym_size_int, 1500, 1280]);  addmm_125 = None
	        mul_14611: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2364, 0.125);  view_2364 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2365: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_14611, [sym_size_int, 1500, 20, 64]);  mul_14611 = None
	        permute_252: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2365, [0, 2, 1, 3]);  view_2365 = None
	        clone_202: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_252, memory_format = torch.contiguous_format);  permute_252 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2366: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22982, [sym_size_int, 1500, 1280])
	        amin_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2366, [2])
	        amax_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2366, [2]);  view_2366 = None
	        full_302: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_151, full_302);  amin_151 = full_302 = None
	        full_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_151, full_303);  amax_151 = full_303 = None
	        sub_6913: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_151, minimum_151);  maximum_151 = None
	        div_302: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6913, 255.0);  sub_6913 = None
	        clamp_min_453: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_302, 1.1920928955078125e-07);  div_302 = None
	        div_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_151, clamp_min_453);  minimum_151 = None
	        round_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_303);  div_303 = None
	        sub_6919: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_303);  round_303 = None
	        clamp_min_454: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6919, -128);  sub_6919 = None
	        clamp_max_302: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_454, 127);  clamp_min_454 = None
	        _assert_tensor_metadata_1361 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_453, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1361 = None
	        _assert_tensor_metadata_1362 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_302, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1362 = None
	        convert_element_type_906: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_302, torch.int8);  clamp_max_302 = None
	        view_2367: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22982, [sym_size_int, 1500, 1280])
	        view_2368: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_453, [sym_size_int, 1500, 1])
	        view_2369: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_906, [sym_size_int, 1500, 1])
	        reciprocal_151: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2368);  view_2368 = None
	        mul_14665: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_151, 1.0);  reciprocal_151 = None
	        mul_14668: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2367, mul_14665);  view_2367 = mul_14665 = None
	        round_304: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14668);  mul_14668 = None
	        add_23221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_304, view_2369);  round_304 = view_2369 = None
	        clamp_min_455: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23221, -128);  add_23221 = None
	        clamp_max_303: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_455, 127);  clamp_min_455 = None
	        view_2370: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_303, [sym_size_int, 1500, 1280]);  clamp_max_303 = None
	        _assert_tensor_metadata_1363 = torch.ops.aten._assert_tensor_metadata.default(view_2370, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1363 = None
	        convert_element_type_907: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2370, torch.int8);  view_2370 = None
	        view_2371: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_907, [sym_size_int, 1500, 1280]);  convert_element_type_907 = None
	        view_2372: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_453, [sym_size_int, 1500, 1]);  clamp_min_453 = None
	        view_2373: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_906, [sym_size_int, 1500, 1]);  convert_element_type_906 = None
	        _assert_tensor_metadata_1364 = torch.ops.aten._assert_tensor_metadata.default(view_2371, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1364 = None
	        convert_element_type_908: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2371, torch.float32);  view_2371 = None
	        _assert_tensor_metadata_1365 = torch.ops.aten._assert_tensor_metadata.default(view_2373, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1365 = None
	        convert_element_type_909: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2373, torch.float32);  view_2373 = None
	        sub_6939: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_908, convert_element_type_909);  convert_element_type_908 = convert_element_type_909 = None
	        mul_14690: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6939, view_2372);  sub_6939 = view_2372 = None
	        view_2374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14690, [sym_size_int, 1500, 1280]);  mul_14690 = None
	        _assert_tensor_metadata_1366 = torch.ops.aten._assert_tensor_metadata.default(view_2374, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1366 = None
	        view_2375: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg686_1, [1280, 40, 32]);  arg686_1 = None
	        view_2376: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg687_1, [1280, 40, 1]);  arg687_1 = None
	        view_2377: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg688_1, [1280, 40, 1]);  arg688_1 = None
	        _assert_tensor_metadata_1367 = torch.ops.aten._assert_tensor_metadata.default(view_2375, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1367 = None
	        convert_element_type_910: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2375, torch.float32);  view_2375 = None
	        _assert_tensor_metadata_1368 = torch.ops.aten._assert_tensor_metadata.default(view_2377, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1368 = None
	        convert_element_type_911: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2377, torch.float32);  view_2377 = None
	        sub_6943: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_910, convert_element_type_911);  convert_element_type_910 = convert_element_type_911 = None
	        mul_14695: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6943, view_2376);  sub_6943 = view_2376 = None
	        view_2378: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14695, [1280, 1280]);  mul_14695 = None
	        _assert_tensor_metadata_1369 = torch.ops.aten._assert_tensor_metadata.default(view_2378, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1369 = None
	        permute_253: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2378, [1, 0]);  view_2378 = None
	        mul_14698: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2379: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2374, [mul_14698, 1280]);  view_2374 = mul_14698 = None
	        mm_25: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2379, permute_253);  view_2379 = permute_253 = None
	        view_2380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_25, [sym_size_int, 1500, 1280]);  mm_25 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2381: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2380, [sym_size_int, -1, 20, 64]);  view_2380 = None
	        permute_254: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2381, [0, 2, 1, 3]);  view_2381 = None
	        clone_203: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_254, memory_format = torch.contiguous_format);  permute_254 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2382: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22982, [sym_size_int, 1500, 1280])
	        amin_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2382, [2])
	        amax_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2382, [2]);  view_2382 = None
	        full_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_152, full_304);  amin_152 = full_304 = None
	        full_305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_152, full_305);  amax_152 = full_305 = None
	        sub_6957: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_152, minimum_152);  maximum_152 = None
	        div_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6957, 255.0);  sub_6957 = None
	        clamp_min_456: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_304, 1.1920928955078125e-07);  div_304 = None
	        div_305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_152, clamp_min_456);  minimum_152 = None
	        round_305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_305);  div_305 = None
	        sub_6963: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_305);  round_305 = None
	        clamp_min_457: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6963, -128);  sub_6963 = None
	        clamp_max_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_457, 127);  clamp_min_457 = None
	        _assert_tensor_metadata_1370 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_456, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1370 = None
	        _assert_tensor_metadata_1371 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_304, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1371 = None
	        convert_element_type_912: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_304, torch.int8);  clamp_max_304 = None
	        view_2383: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22982, [sym_size_int, 1500, 1280]);  add_22982 = None
	        view_2384: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_456, [sym_size_int, 1500, 1])
	        view_2385: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_912, [sym_size_int, 1500, 1])
	        reciprocal_152: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2384);  view_2384 = None
	        mul_14764: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_152, 1.0);  reciprocal_152 = None
	        mul_14767: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2383, mul_14764);  view_2383 = mul_14764 = None
	        round_306: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14767);  mul_14767 = None
	        add_23369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_306, view_2385);  round_306 = view_2385 = None
	        clamp_min_458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23369, -128);  add_23369 = None
	        clamp_max_305: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_458, 127);  clamp_min_458 = None
	        view_2386: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_305, [sym_size_int, 1500, 1280]);  clamp_max_305 = None
	        _assert_tensor_metadata_1372 = torch.ops.aten._assert_tensor_metadata.default(view_2386, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1372 = None
	        convert_element_type_913: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2386, torch.int8);  view_2386 = None
	        view_2387: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_913, [sym_size_int, 1500, 1280]);  convert_element_type_913 = None
	        view_2388: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_456, [sym_size_int, 1500, 1]);  clamp_min_456 = None
	        view_2389: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_912, [sym_size_int, 1500, 1]);  convert_element_type_912 = None
	        _assert_tensor_metadata_1373 = torch.ops.aten._assert_tensor_metadata.default(view_2387, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1373 = None
	        convert_element_type_914: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2387, torch.float32);  view_2387 = None
	        _assert_tensor_metadata_1374 = torch.ops.aten._assert_tensor_metadata.default(view_2389, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1374 = None
	        convert_element_type_915: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2389, torch.float32);  view_2389 = None
	        sub_6983: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_914, convert_element_type_915);  convert_element_type_914 = convert_element_type_915 = None
	        mul_14789: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6983, view_2388);  sub_6983 = view_2388 = None
	        view_2390: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14789, [sym_size_int, 1500, 1280]);  mul_14789 = None
	        _assert_tensor_metadata_1375 = torch.ops.aten._assert_tensor_metadata.default(view_2390, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1375 = None
	        view_2391: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg690_1, [1280, 40, 32]);  arg690_1 = None
	        view_2392: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg691_1, [1280, 40, 1]);  arg691_1 = None
	        view_2393: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg692_1, [1280, 40, 1]);  arg692_1 = None
	        _assert_tensor_metadata_1376 = torch.ops.aten._assert_tensor_metadata.default(view_2391, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1376 = None
	        convert_element_type_916: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2391, torch.float32);  view_2391 = None
	        _assert_tensor_metadata_1377 = torch.ops.aten._assert_tensor_metadata.default(view_2393, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1377 = None
	        convert_element_type_917: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2393, torch.float32);  view_2393 = None
	        sub_6987: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_916, convert_element_type_917);  convert_element_type_916 = convert_element_type_917 = None
	        mul_14794: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6987, view_2392);  sub_6987 = view_2392 = None
	        view_2394: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14794, [1280, 1280]);  mul_14794 = None
	        _assert_tensor_metadata_1378 = torch.ops.aten._assert_tensor_metadata.default(view_2394, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1378 = None
	        mul_14799: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2395: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2390, [mul_14799, 1280]);  view_2390 = mul_14799 = None
	        permute_255: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2394, [1, 0]);  view_2394 = None
	        addmm_126: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg689_1, view_2395, permute_255);  arg689_1 = view_2395 = permute_255 = None
	        view_2396: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_126, [sym_size_int, 1500, 1280]);  addmm_126 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2397: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2396, [sym_size_int, -1, 20, 64]);  view_2396 = None
	        permute_256: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2397, [0, 2, 1, 3]);  view_2397 = None
	        clone_204: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_256, memory_format = torch.contiguous_format);  permute_256 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_25 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_202, clone_203, clone_204, None, False, scale = 1.0);  clone_202 = clone_203 = clone_204 = None
	        getitem_202: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_25[0];  _scaled_dot_product_efficient_attention_25 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_257: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_202, [0, 2, 1, 3]);  getitem_202 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2398: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_257, [sym_size_int, 1500, -1]);  permute_257 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2399: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2398, [sym_size_int, 1500, 1280])
	        amin_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2399, [2])
	        amax_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2399, [2]);  view_2399 = None
	        full_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_153, full_306);  amin_153 = full_306 = None
	        full_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_153, full_307);  amax_153 = full_307 = None
	        sub_7005: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_153, minimum_153);  maximum_153 = None
	        div_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7005, 255.0);  sub_7005 = None
	        clamp_min_459: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_306, 1.1920928955078125e-07);  div_306 = None
	        div_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_153, clamp_min_459);  minimum_153 = None
	        round_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_307);  div_307 = None
	        sub_7011: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_307);  round_307 = None
	        clamp_min_460: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7011, -128);  sub_7011 = None
	        clamp_max_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_460, 127);  clamp_min_460 = None
	        _assert_tensor_metadata_1379 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_459, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1379 = None
	        _assert_tensor_metadata_1380 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_306, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1380 = None
	        convert_element_type_918: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_306, torch.int8);  clamp_max_306 = None
	        view_2400: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2398, [sym_size_int, 1500, 1280]);  view_2398 = None
	        view_2401: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_459, [sym_size_int, 1500, 1])
	        view_2402: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_918, [sym_size_int, 1500, 1])
	        reciprocal_153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2401);  view_2401 = None
	        mul_14869: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_153, 1.0);  reciprocal_153 = None
	        mul_14872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2400, mul_14869);  view_2400 = mul_14869 = None
	        round_308: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14872);  mul_14872 = None
	        add_23533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_308, view_2402);  round_308 = view_2402 = None
	        clamp_min_461: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23533, -128);  add_23533 = None
	        clamp_max_307: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_461, 127);  clamp_min_461 = None
	        view_2403: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_307, [sym_size_int, 1500, 1280]);  clamp_max_307 = None
	        _assert_tensor_metadata_1381 = torch.ops.aten._assert_tensor_metadata.default(view_2403, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1381 = None
	        convert_element_type_919: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2403, torch.int8);  view_2403 = None
	        view_2404: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_919, [sym_size_int, 1500, 1280]);  convert_element_type_919 = None
	        view_2405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_459, [sym_size_int, 1500, 1]);  clamp_min_459 = None
	        view_2406: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_918, [sym_size_int, 1500, 1]);  convert_element_type_918 = None
	        _assert_tensor_metadata_1382 = torch.ops.aten._assert_tensor_metadata.default(view_2404, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1382 = None
	        convert_element_type_920: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2404, torch.float32);  view_2404 = None
	        _assert_tensor_metadata_1383 = torch.ops.aten._assert_tensor_metadata.default(view_2406, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1383 = None
	        convert_element_type_921: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2406, torch.float32);  view_2406 = None
	        sub_7031: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_920, convert_element_type_921);  convert_element_type_920 = convert_element_type_921 = None
	        mul_14894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7031, view_2405);  sub_7031 = view_2405 = None
	        view_2407: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14894, [sym_size_int, 1500, 1280]);  mul_14894 = None
	        _assert_tensor_metadata_1384 = torch.ops.aten._assert_tensor_metadata.default(view_2407, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1384 = None
	        view_2408: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg694_1, [1280, 40, 32]);  arg694_1 = None
	        view_2409: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg695_1, [1280, 40, 1]);  arg695_1 = None
	        view_2410: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg696_1, [1280, 40, 1]);  arg696_1 = None
	        _assert_tensor_metadata_1385 = torch.ops.aten._assert_tensor_metadata.default(view_2408, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1385 = None
	        convert_element_type_922: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2408, torch.float32);  view_2408 = None
	        _assert_tensor_metadata_1386 = torch.ops.aten._assert_tensor_metadata.default(view_2410, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1386 = None
	        convert_element_type_923: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2410, torch.float32);  view_2410 = None
	        sub_7035: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_922, convert_element_type_923);  convert_element_type_922 = convert_element_type_923 = None
	        mul_14899: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7035, view_2409);  sub_7035 = view_2409 = None
	        view_2411: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14899, [1280, 1280]);  mul_14899 = None
	        _assert_tensor_metadata_1387 = torch.ops.aten._assert_tensor_metadata.default(view_2411, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1387 = None
	        mul_14904: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2412: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2407, [mul_14904, 1280]);  view_2407 = mul_14904 = None
	        permute_258: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2411, [1, 0]);  view_2411 = None
	        addmm_127: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg693_1, view_2412, permute_258);  arg693_1 = view_2412 = permute_258 = None
	        view_2413: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_127, [sym_size_int, 1500, 1280]);  addmm_127 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2413);  view_2413 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_23596: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_22976, clone_205);  add_22976 = clone_205 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_23596, memory_format = torch.contiguous_format)
	        var_mean_51 = torch.ops.aten.var_mean.correction(clone_206, [2], correction = 0, keepdim = True)
	        getitem_206: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_51[0]
	        getitem_207: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_51[1];  var_mean_51 = None
	        add_23601: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_206, 1e-05);  getitem_206 = None
	        rsqrt_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_23601);  add_23601 = None
	        sub_7041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_206, getitem_207);  clone_206 = getitem_207 = None
	        mul_14915: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7041, rsqrt_51);  sub_7041 = rsqrt_51 = None
	        mul_14916: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14915, arg697_1);  mul_14915 = arg697_1 = None
	        add_23602: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_14916, arg698_1);  mul_14916 = arg698_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2414: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_23602, [sym_size_int, 1500, 1280])
	        amin_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2414, [2])
	        amax_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2414, [2]);  view_2414 = None
	        full_308: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_154, full_308);  amin_154 = full_308 = None
	        full_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_154, full_309);  amax_154 = full_309 = None
	        sub_7052: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_154, minimum_154);  maximum_154 = None
	        div_308: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7052, 255.0);  sub_7052 = None
	        clamp_min_462: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_308, 1.1920928955078125e-07);  div_308 = None
	        div_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_154, clamp_min_462);  minimum_154 = None
	        round_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_309);  div_309 = None
	        sub_7058: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_309);  round_309 = None
	        clamp_min_463: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7058, -128);  sub_7058 = None
	        clamp_max_308: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_463, 127);  clamp_min_463 = None
	        _assert_tensor_metadata_1388 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_462, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1388 = None
	        _assert_tensor_metadata_1389 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_308, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1389 = None
	        convert_element_type_924: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_308, torch.int8);  clamp_max_308 = None
	        view_2415: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_23602, [sym_size_int, 1500, 1280]);  add_23602 = None
	        view_2416: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_462, [sym_size_int, 1500, 1])
	        view_2417: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_924, [sym_size_int, 1500, 1])
	        reciprocal_154: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2416);  view_2416 = None
	        mul_14964: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_154, 1.0);  reciprocal_154 = None
	        mul_14967: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2415, mul_14964);  view_2415 = mul_14964 = None
	        round_310: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14967);  mul_14967 = None
	        add_23689: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_310, view_2417);  round_310 = view_2417 = None
	        clamp_min_464: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23689, -128);  add_23689 = None
	        clamp_max_309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_464, 127);  clamp_min_464 = None
	        view_2418: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_309, [sym_size_int, 1500, 1280]);  clamp_max_309 = None
	        _assert_tensor_metadata_1390 = torch.ops.aten._assert_tensor_metadata.default(view_2418, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1390 = None
	        convert_element_type_925: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2418, torch.int8);  view_2418 = None
	        view_2419: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_925, [sym_size_int, 1500, 1280]);  convert_element_type_925 = None
	        view_2420: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_462, [sym_size_int, 1500, 1]);  clamp_min_462 = None
	        view_2421: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_924, [sym_size_int, 1500, 1]);  convert_element_type_924 = None
	        _assert_tensor_metadata_1391 = torch.ops.aten._assert_tensor_metadata.default(view_2419, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1391 = None
	        convert_element_type_926: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2419, torch.float32);  view_2419 = None
	        _assert_tensor_metadata_1392 = torch.ops.aten._assert_tensor_metadata.default(view_2421, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1392 = None
	        convert_element_type_927: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2421, torch.float32);  view_2421 = None
	        sub_7078: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_926, convert_element_type_927);  convert_element_type_926 = convert_element_type_927 = None
	        mul_14989: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7078, view_2420);  sub_7078 = view_2420 = None
	        view_2422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14989, [sym_size_int, 1500, 1280]);  mul_14989 = None
	        _assert_tensor_metadata_1393 = torch.ops.aten._assert_tensor_metadata.default(view_2422, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1393 = None
	        view_2423: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg700_1, [5120, 40, 32]);  arg700_1 = None
	        view_2424: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg701_1, [5120, 40, 1]);  arg701_1 = None
	        view_2425: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg702_1, [5120, 40, 1]);  arg702_1 = None
	        _assert_tensor_metadata_1394 = torch.ops.aten._assert_tensor_metadata.default(view_2423, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1394 = None
	        convert_element_type_928: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2423, torch.float32);  view_2423 = None
	        _assert_tensor_metadata_1395 = torch.ops.aten._assert_tensor_metadata.default(view_2425, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1395 = None
	        convert_element_type_929: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2425, torch.float32);  view_2425 = None
	        sub_7082: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_928, convert_element_type_929);  convert_element_type_928 = convert_element_type_929 = None
	        mul_14994: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7082, view_2424);  sub_7082 = view_2424 = None
	        view_2426: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14994, [5120, 1280]);  mul_14994 = None
	        _assert_tensor_metadata_1396 = torch.ops.aten._assert_tensor_metadata.default(view_2426, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1396 = None
	        mul_14999: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2427: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2422, [mul_14999, 1280]);  view_2422 = mul_14999 = None
	        permute_259: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2426, [1, 0]);  view_2426 = None
	        addmm_128: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg699_1, view_2427, permute_259);  arg699_1 = view_2427 = permute_259 = None
	        view_2428: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_128, [sym_size_int, 1500, 5120]);  addmm_128 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_15006: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2428, 0.5)
	        mul_15007: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2428, 0.7071067811865476);  view_2428 = None
	        erf_27: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_15007);  mul_15007 = None
	        add_23748: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_27, 1);  erf_27 = None
	        mul_15008: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15006, add_23748);  mul_15006 = add_23748 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_207: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_15008);  mul_15008 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2429: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_207, [sym_size_int, 1500, 5120])
	        amin_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2429, [2])
	        amax_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2429, [2]);  view_2429 = None
	        full_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_155, full_310);  amin_155 = full_310 = None
	        full_311: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_155, full_311);  amax_155 = full_311 = None
	        sub_7095: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_155, minimum_155);  maximum_155 = None
	        div_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7095, 255.0);  sub_7095 = None
	        clamp_min_465: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_310, 1.1920928955078125e-07);  div_310 = None
	        div_311: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_155, clamp_min_465);  minimum_155 = None
	        round_311: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_311);  div_311 = None
	        sub_7101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_311);  round_311 = None
	        clamp_min_466: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7101, -128);  sub_7101 = None
	        clamp_max_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_466, 127);  clamp_min_466 = None
	        _assert_tensor_metadata_1397 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_465, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1397 = None
	        _assert_tensor_metadata_1398 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_310, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1398 = None
	        convert_element_type_930: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_310, torch.int8);  clamp_max_310 = None
	        view_2430: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_207, [sym_size_int, 1500, 5120]);  clone_207 = None
	        view_2431: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_465, [sym_size_int, 1500, 1])
	        view_2432: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_930, [sym_size_int, 1500, 1])
	        reciprocal_155: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2431);  view_2431 = None
	        mul_15054: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_155, 1.0);  reciprocal_155 = None
	        mul_15057: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2430, mul_15054);  view_2430 = mul_15054 = None
	        round_312: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_15057);  mul_15057 = None
	        add_23831: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_312, view_2432);  round_312 = view_2432 = None
	        clamp_min_467: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23831, -128);  add_23831 = None
	        clamp_max_311: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_467, 127);  clamp_min_467 = None
	        view_2433: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_311, [sym_size_int, 1500, 5120]);  clamp_max_311 = None
	        _assert_tensor_metadata_1399 = torch.ops.aten._assert_tensor_metadata.default(view_2433, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1399 = None
	        convert_element_type_931: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2433, torch.int8);  view_2433 = None
	        view_2434: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_931, [sym_size_int, 1500, 5120]);  convert_element_type_931 = None
	        view_2435: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_465, [sym_size_int, 1500, 1]);  clamp_min_465 = None
	        view_2436: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_930, [sym_size_int, 1500, 1]);  convert_element_type_930 = None
	        _assert_tensor_metadata_1400 = torch.ops.aten._assert_tensor_metadata.default(view_2434, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1400 = None
	        convert_element_type_932: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2434, torch.float32);  view_2434 = None
	        _assert_tensor_metadata_1401 = torch.ops.aten._assert_tensor_metadata.default(view_2436, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1401 = None
	        convert_element_type_933: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2436, torch.float32);  view_2436 = None
	        sub_7121: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_932, convert_element_type_933);  convert_element_type_932 = convert_element_type_933 = None
	        mul_15079: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7121, view_2435);  sub_7121 = view_2435 = None
	        view_2437: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_15079, [sym_size_int, 1500, 5120]);  mul_15079 = None
	        _assert_tensor_metadata_1402 = torch.ops.aten._assert_tensor_metadata.default(view_2437, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1402 = None
	        view_2438: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg704_1, [1280, 160, 32]);  arg704_1 = None
	        view_2439: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg705_1, [1280, 160, 1]);  arg705_1 = None
	        view_2440: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg706_1, [1280, 160, 1]);  arg706_1 = None
	        _assert_tensor_metadata_1403 = torch.ops.aten._assert_tensor_metadata.default(view_2438, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1403 = None
	        convert_element_type_934: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2438, torch.float32);  view_2438 = None
	        _assert_tensor_metadata_1404 = torch.ops.aten._assert_tensor_metadata.default(view_2440, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1404 = None
	        convert_element_type_935: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2440, torch.float32);  view_2440 = None
	        sub_7125: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_934, convert_element_type_935);  convert_element_type_934 = convert_element_type_935 = None
	        mul_15084: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7125, view_2439);  sub_7125 = view_2439 = None
	        view_2441: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_15084, [1280, 5120]);  mul_15084 = None
	        _assert_tensor_metadata_1405 = torch.ops.aten._assert_tensor_metadata.default(view_2441, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1405 = None
	        mul_15089: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2442: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2437, [mul_15089, 5120]);  view_2437 = mul_15089 = None
	        permute_260: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2441, [1, 0]);  view_2441 = None
	        addmm_129: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg703_1, view_2442, permute_260);  arg703_1 = view_2442 = permute_260 = None
	        view_2443: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_129, [sym_size_int, 1500, 1280]);  addmm_129 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_208: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2443);  view_2443 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_23894: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_23596, clone_208);  add_23596 = clone_208 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_23894, memory_format = torch.contiguous_format)
	        var_mean_52 = torch.ops.aten.var_mean.correction(clone_209, [2], correction = 0, keepdim = True)
	        getitem_208: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_52[0]
	        getitem_209: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_52[1];  var_mean_52 = None
	        add_23899: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_208, 1e-05);  getitem_208 = None
	        rsqrt_52: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_23899);  add_23899 = None
	        sub_7131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_209, getitem_209);  clone_209 = getitem_209 = None
	        mul_15100: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7131, rsqrt_52);  sub_7131 = rsqrt_52 = None
	        mul_15101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15100, arg707_1);  mul_15100 = arg707_1 = None
	        add_23900: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_15101, arg708_1);  mul_15101 = arg708_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2444: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_23900, [sym_size_int, 1500, 1280])
	        amin_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2444, [2])
	        amax_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2444, [2]);  view_2444 = None
	        full_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_156, full_312);  amin_156 = full_312 = None
	        full_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_156, full_313);  amax_156 = full_313 = None
	        sub_7142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_156, minimum_156);  maximum_156 = None
	        div_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7142, 255.0);  sub_7142 = None
	        clamp_min_468: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_312, 1.1920928955078125e-07);  div_312 = None
	        div_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_156, clamp_min_468);  minimum_156 = None
	        round_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_313);  div_313 = None
	        sub_7148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_313);  round_313 = None
	        clamp_min_469: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7148, -128);  sub_7148 = None
	        clamp_max_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_469, 127);  clamp_min_469 = None
	        _assert_tensor_metadata_1406 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_468, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1406 = None
	        _assert_tensor_metadata_1407 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_312, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1407 = None
	        convert_element_type_936: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_312, torch.int8);  clamp_max_312 = None
	        view_2445: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_23900, [sym_size_int, 1500, 1280])
	        view_2446: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_468, [sym_size_int, 1500, 1])
	        view_2447: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_936, [sym_size_int, 1500, 1])
	        reciprocal_156: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2446);  view_2446 = None
	        mul_15149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_156, 1.0);  reciprocal_156 = None
	        mul_15152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2445, mul_15149);  view_2445 = mul_15149 = None
	        round_314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15152);  mul_15152 = None
	        add_23987: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_314, view_2447);  round_314 = view_2447 = None
	        clamp_min_470: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23987, -128);  add_23987 = None
	        clamp_max_313: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_470, 127);  clamp_min_470 = None
	        view_2448: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_313, [sym_size_int, 1500, 1280]);  clamp_max_313 = None
	        _assert_tensor_metadata_1408 = torch.ops.aten._assert_tensor_metadata.default(view_2448, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1408 = None
	        convert_element_type_937: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2448, torch.int8);  view_2448 = None
	        view_2449: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_937, [sym_size_int, 1500, 1280]);  convert_element_type_937 = None
	        view_2450: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_468, [sym_size_int, 1500, 1]);  clamp_min_468 = None
	        view_2451: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_936, [sym_size_int, 1500, 1]);  convert_element_type_936 = None
	        _assert_tensor_metadata_1409 = torch.ops.aten._assert_tensor_metadata.default(view_2449, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1409 = None
	        convert_element_type_938: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2449, torch.float32);  view_2449 = None
	        _assert_tensor_metadata_1410 = torch.ops.aten._assert_tensor_metadata.default(view_2451, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1410 = None
	        convert_element_type_939: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2451, torch.float32);  view_2451 = None
	        sub_7168: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_938, convert_element_type_939);  convert_element_type_938 = convert_element_type_939 = None
	        mul_15174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7168, view_2450);  sub_7168 = view_2450 = None
	        view_2452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15174, [sym_size_int, 1500, 1280]);  mul_15174 = None
	        _assert_tensor_metadata_1411 = torch.ops.aten._assert_tensor_metadata.default(view_2452, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1411 = None
	        view_2453: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg710_1, [1280, 40, 32]);  arg710_1 = None
	        view_2454: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg711_1, [1280, 40, 1]);  arg711_1 = None
	        view_2455: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg712_1, [1280, 40, 1]);  arg712_1 = None
	        _assert_tensor_metadata_1412 = torch.ops.aten._assert_tensor_metadata.default(view_2453, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1412 = None
	        convert_element_type_940: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2453, torch.float32);  view_2453 = None
	        _assert_tensor_metadata_1413 = torch.ops.aten._assert_tensor_metadata.default(view_2455, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1413 = None
	        convert_element_type_941: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2455, torch.float32);  view_2455 = None
	        sub_7172: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_940, convert_element_type_941);  convert_element_type_940 = convert_element_type_941 = None
	        mul_15179: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7172, view_2454);  sub_7172 = view_2454 = None
	        view_2456: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15179, [1280, 1280]);  mul_15179 = None
	        _assert_tensor_metadata_1414 = torch.ops.aten._assert_tensor_metadata.default(view_2456, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1414 = None
	        mul_15184: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2457: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2452, [mul_15184, 1280]);  view_2452 = mul_15184 = None
	        permute_261: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2456, [1, 0]);  view_2456 = None
	        addmm_130: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg709_1, view_2457, permute_261);  arg709_1 = view_2457 = permute_261 = None
	        view_2458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_130, [sym_size_int, 1500, 1280]);  addmm_130 = None
	        mul_15191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2458, 0.125);  view_2458 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2459: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_15191, [sym_size_int, 1500, 20, 64]);  mul_15191 = None
	        permute_262: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2459, [0, 2, 1, 3]);  view_2459 = None
	        clone_210: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_262, memory_format = torch.contiguous_format);  permute_262 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2460: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_23900, [sym_size_int, 1500, 1280])
	        amin_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2460, [2])
	        amax_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2460, [2]);  view_2460 = None
	        full_314: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_157, full_314);  amin_157 = full_314 = None
	        full_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_157, full_315);  amax_157 = full_315 = None
	        sub_7187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_157, minimum_157);  maximum_157 = None
	        div_314: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7187, 255.0);  sub_7187 = None
	        clamp_min_471: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_314, 1.1920928955078125e-07);  div_314 = None
	        div_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_157, clamp_min_471);  minimum_157 = None
	        round_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_315);  div_315 = None
	        sub_7193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_315);  round_315 = None
	        clamp_min_472: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7193, -128);  sub_7193 = None
	        clamp_max_314: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_472, 127);  clamp_min_472 = None
	        _assert_tensor_metadata_1415 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_471, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1415 = None
	        _assert_tensor_metadata_1416 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_314, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1416 = None
	        convert_element_type_942: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_314, torch.int8);  clamp_max_314 = None
	        view_2461: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_23900, [sym_size_int, 1500, 1280])
	        view_2462: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_471, [sym_size_int, 1500, 1])
	        view_2463: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_942, [sym_size_int, 1500, 1])
	        reciprocal_157: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2462);  view_2462 = None
	        mul_15245: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_157, 1.0);  reciprocal_157 = None
	        mul_15248: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2461, mul_15245);  view_2461 = mul_15245 = None
	        round_316: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15248);  mul_15248 = None
	        add_24139: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_316, view_2463);  round_316 = view_2463 = None
	        clamp_min_473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24139, -128);  add_24139 = None
	        clamp_max_315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_473, 127);  clamp_min_473 = None
	        view_2464: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_315, [sym_size_int, 1500, 1280]);  clamp_max_315 = None
	        _assert_tensor_metadata_1417 = torch.ops.aten._assert_tensor_metadata.default(view_2464, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1417 = None
	        convert_element_type_943: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2464, torch.int8);  view_2464 = None
	        view_2465: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_943, [sym_size_int, 1500, 1280]);  convert_element_type_943 = None
	        view_2466: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_471, [sym_size_int, 1500, 1]);  clamp_min_471 = None
	        view_2467: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_942, [sym_size_int, 1500, 1]);  convert_element_type_942 = None
	        _assert_tensor_metadata_1418 = torch.ops.aten._assert_tensor_metadata.default(view_2465, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1418 = None
	        convert_element_type_944: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2465, torch.float32);  view_2465 = None
	        _assert_tensor_metadata_1419 = torch.ops.aten._assert_tensor_metadata.default(view_2467, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1419 = None
	        convert_element_type_945: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2467, torch.float32);  view_2467 = None
	        sub_7213: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_944, convert_element_type_945);  convert_element_type_944 = convert_element_type_945 = None
	        mul_15270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7213, view_2466);  sub_7213 = view_2466 = None
	        view_2468: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15270, [sym_size_int, 1500, 1280]);  mul_15270 = None
	        _assert_tensor_metadata_1420 = torch.ops.aten._assert_tensor_metadata.default(view_2468, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1420 = None
	        view_2469: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg713_1, [1280, 40, 32]);  arg713_1 = None
	        view_2470: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg714_1, [1280, 40, 1]);  arg714_1 = None
	        view_2471: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg715_1, [1280, 40, 1]);  arg715_1 = None
	        _assert_tensor_metadata_1421 = torch.ops.aten._assert_tensor_metadata.default(view_2469, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1421 = None
	        convert_element_type_946: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2469, torch.float32);  view_2469 = None
	        _assert_tensor_metadata_1422 = torch.ops.aten._assert_tensor_metadata.default(view_2471, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1422 = None
	        convert_element_type_947: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2471, torch.float32);  view_2471 = None
	        sub_7217: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_946, convert_element_type_947);  convert_element_type_946 = convert_element_type_947 = None
	        mul_15275: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7217, view_2470);  sub_7217 = view_2470 = None
	        view_2472: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15275, [1280, 1280]);  mul_15275 = None
	        _assert_tensor_metadata_1423 = torch.ops.aten._assert_tensor_metadata.default(view_2472, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1423 = None
	        permute_263: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2472, [1, 0]);  view_2472 = None
	        mul_15278: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2473: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2468, [mul_15278, 1280]);  view_2468 = mul_15278 = None
	        mm_26: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2473, permute_263);  view_2473 = permute_263 = None
	        view_2474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_26, [sym_size_int, 1500, 1280]);  mm_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2475: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2474, [sym_size_int, -1, 20, 64]);  view_2474 = None
	        permute_264: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2475, [0, 2, 1, 3]);  view_2475 = None
	        clone_211: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_264, memory_format = torch.contiguous_format);  permute_264 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2476: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_23900, [sym_size_int, 1500, 1280])
	        amin_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2476, [2])
	        amax_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2476, [2]);  view_2476 = None
	        full_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_158, full_316);  amin_158 = full_316 = None
	        full_317: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_158, full_317);  amax_158 = full_317 = None
	        sub_7231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_158, minimum_158);  maximum_158 = None
	        div_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7231, 255.0);  sub_7231 = None
	        clamp_min_474: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_316, 1.1920928955078125e-07);  div_316 = None
	        div_317: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_158, clamp_min_474);  minimum_158 = None
	        round_317: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_317);  div_317 = None
	        sub_7237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_317);  round_317 = None
	        clamp_min_475: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7237, -128);  sub_7237 = None
	        clamp_max_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_475, 127);  clamp_min_475 = None
	        _assert_tensor_metadata_1424 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_474, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1424 = None
	        _assert_tensor_metadata_1425 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_316, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1425 = None
	        convert_element_type_948: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_316, torch.int8);  clamp_max_316 = None
	        view_2477: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_23900, [sym_size_int, 1500, 1280]);  add_23900 = None
	        view_2478: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_474, [sym_size_int, 1500, 1])
	        view_2479: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_948, [sym_size_int, 1500, 1])
	        reciprocal_158: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2478);  view_2478 = None
	        mul_15344: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_158, 1.0);  reciprocal_158 = None
	        mul_15347: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2477, mul_15344);  view_2477 = mul_15344 = None
	        round_318: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15347);  mul_15347 = None
	        add_24287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_318, view_2479);  round_318 = view_2479 = None
	        clamp_min_476: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24287, -128);  add_24287 = None
	        clamp_max_317: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_476, 127);  clamp_min_476 = None
	        view_2480: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_317, [sym_size_int, 1500, 1280]);  clamp_max_317 = None
	        _assert_tensor_metadata_1426 = torch.ops.aten._assert_tensor_metadata.default(view_2480, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1426 = None
	        convert_element_type_949: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2480, torch.int8);  view_2480 = None
	        view_2481: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_949, [sym_size_int, 1500, 1280]);  convert_element_type_949 = None
	        view_2482: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_474, [sym_size_int, 1500, 1]);  clamp_min_474 = None
	        view_2483: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_948, [sym_size_int, 1500, 1]);  convert_element_type_948 = None
	        _assert_tensor_metadata_1427 = torch.ops.aten._assert_tensor_metadata.default(view_2481, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1427 = None
	        convert_element_type_950: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2481, torch.float32);  view_2481 = None
	        _assert_tensor_metadata_1428 = torch.ops.aten._assert_tensor_metadata.default(view_2483, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1428 = None
	        convert_element_type_951: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2483, torch.float32);  view_2483 = None
	        sub_7257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_950, convert_element_type_951);  convert_element_type_950 = convert_element_type_951 = None
	        mul_15369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7257, view_2482);  sub_7257 = view_2482 = None
	        view_2484: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15369, [sym_size_int, 1500, 1280]);  mul_15369 = None
	        _assert_tensor_metadata_1429 = torch.ops.aten._assert_tensor_metadata.default(view_2484, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1429 = None
	        view_2485: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg717_1, [1280, 40, 32]);  arg717_1 = None
	        view_2486: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg718_1, [1280, 40, 1]);  arg718_1 = None
	        view_2487: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg719_1, [1280, 40, 1]);  arg719_1 = None
	        _assert_tensor_metadata_1430 = torch.ops.aten._assert_tensor_metadata.default(view_2485, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1430 = None
	        convert_element_type_952: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2485, torch.float32);  view_2485 = None
	        _assert_tensor_metadata_1431 = torch.ops.aten._assert_tensor_metadata.default(view_2487, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1431 = None
	        convert_element_type_953: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2487, torch.float32);  view_2487 = None
	        sub_7261: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_952, convert_element_type_953);  convert_element_type_952 = convert_element_type_953 = None
	        mul_15374: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7261, view_2486);  sub_7261 = view_2486 = None
	        view_2488: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15374, [1280, 1280]);  mul_15374 = None
	        _assert_tensor_metadata_1432 = torch.ops.aten._assert_tensor_metadata.default(view_2488, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1432 = None
	        mul_15379: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2489: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2484, [mul_15379, 1280]);  view_2484 = mul_15379 = None
	        permute_265: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2488, [1, 0]);  view_2488 = None
	        addmm_131: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg716_1, view_2489, permute_265);  arg716_1 = view_2489 = permute_265 = None
	        view_2490: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_131, [sym_size_int, 1500, 1280]);  addmm_131 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2491: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2490, [sym_size_int, -1, 20, 64]);  view_2490 = None
	        permute_266: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2491, [0, 2, 1, 3]);  view_2491 = None
	        clone_212: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_266, memory_format = torch.contiguous_format);  permute_266 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_26 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_210, clone_211, clone_212, None, False, scale = 1.0);  clone_210 = clone_211 = clone_212 = None
	        getitem_210: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_26[0];  _scaled_dot_product_efficient_attention_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_267: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_210, [0, 2, 1, 3]);  getitem_210 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2492: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_267, [sym_size_int, 1500, -1]);  permute_267 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2493: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2492, [sym_size_int, 1500, 1280])
	        amin_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2493, [2])
	        amax_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2493, [2]);  view_2493 = None
	        full_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_159, full_318);  amin_159 = full_318 = None
	        full_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_159, full_319);  amax_159 = full_319 = None
	        sub_7279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_159, minimum_159);  maximum_159 = None
	        div_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7279, 255.0);  sub_7279 = None
	        clamp_min_477: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_318, 1.1920928955078125e-07);  div_318 = None
	        div_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_159, clamp_min_477);  minimum_159 = None
	        round_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_319);  div_319 = None
	        sub_7285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_319);  round_319 = None
	        clamp_min_478: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7285, -128);  sub_7285 = None
	        clamp_max_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_478, 127);  clamp_min_478 = None
	        _assert_tensor_metadata_1433 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_477, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1433 = None
	        _assert_tensor_metadata_1434 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_318, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1434 = None
	        convert_element_type_954: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_318, torch.int8);  clamp_max_318 = None
	        view_2494: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2492, [sym_size_int, 1500, 1280]);  view_2492 = None
	        view_2495: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_477, [sym_size_int, 1500, 1])
	        view_2496: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_954, [sym_size_int, 1500, 1])
	        reciprocal_159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2495);  view_2495 = None
	        mul_15449: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_159, 1.0);  reciprocal_159 = None
	        mul_15452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2494, mul_15449);  view_2494 = mul_15449 = None
	        round_320: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15452);  mul_15452 = None
	        add_24451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_320, view_2496);  round_320 = view_2496 = None
	        clamp_min_479: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24451, -128);  add_24451 = None
	        clamp_max_319: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_479, 127);  clamp_min_479 = None
	        view_2497: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_319, [sym_size_int, 1500, 1280]);  clamp_max_319 = None
	        _assert_tensor_metadata_1435 = torch.ops.aten._assert_tensor_metadata.default(view_2497, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1435 = None
	        convert_element_type_955: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2497, torch.int8);  view_2497 = None
	        view_2498: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_955, [sym_size_int, 1500, 1280]);  convert_element_type_955 = None
	        view_2499: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_477, [sym_size_int, 1500, 1]);  clamp_min_477 = None
	        view_2500: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_954, [sym_size_int, 1500, 1]);  convert_element_type_954 = None
	        _assert_tensor_metadata_1436 = torch.ops.aten._assert_tensor_metadata.default(view_2498, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1436 = None
	        convert_element_type_956: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2498, torch.float32);  view_2498 = None
	        _assert_tensor_metadata_1437 = torch.ops.aten._assert_tensor_metadata.default(view_2500, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1437 = None
	        convert_element_type_957: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2500, torch.float32);  view_2500 = None
	        sub_7305: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_956, convert_element_type_957);  convert_element_type_956 = convert_element_type_957 = None
	        mul_15474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7305, view_2499);  sub_7305 = view_2499 = None
	        view_2501: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15474, [sym_size_int, 1500, 1280]);  mul_15474 = None
	        _assert_tensor_metadata_1438 = torch.ops.aten._assert_tensor_metadata.default(view_2501, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1438 = None
	        view_2502: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg721_1, [1280, 40, 32]);  arg721_1 = None
	        view_2503: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg722_1, [1280, 40, 1]);  arg722_1 = None
	        view_2504: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg723_1, [1280, 40, 1]);  arg723_1 = None
	        _assert_tensor_metadata_1439 = torch.ops.aten._assert_tensor_metadata.default(view_2502, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1439 = None
	        convert_element_type_958: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2502, torch.float32);  view_2502 = None
	        _assert_tensor_metadata_1440 = torch.ops.aten._assert_tensor_metadata.default(view_2504, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1440 = None
	        convert_element_type_959: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2504, torch.float32);  view_2504 = None
	        sub_7309: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_958, convert_element_type_959);  convert_element_type_958 = convert_element_type_959 = None
	        mul_15479: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7309, view_2503);  sub_7309 = view_2503 = None
	        view_2505: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15479, [1280, 1280]);  mul_15479 = None
	        _assert_tensor_metadata_1441 = torch.ops.aten._assert_tensor_metadata.default(view_2505, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1441 = None
	        mul_15484: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2506: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2501, [mul_15484, 1280]);  view_2501 = mul_15484 = None
	        permute_268: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2505, [1, 0]);  view_2505 = None
	        addmm_132: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg720_1, view_2506, permute_268);  arg720_1 = view_2506 = permute_268 = None
	        view_2507: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_132, [sym_size_int, 1500, 1280]);  addmm_132 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_213: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2507);  view_2507 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_24514: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_23894, clone_213);  add_23894 = clone_213 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_24514, memory_format = torch.contiguous_format)
	        var_mean_53 = torch.ops.aten.var_mean.correction(clone_214, [2], correction = 0, keepdim = True)
	        getitem_214: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_53[0]
	        getitem_215: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_53[1];  var_mean_53 = None
	        add_24519: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_214, 1e-05);  getitem_214 = None
	        rsqrt_53: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_24519);  add_24519 = None
	        sub_7315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_214, getitem_215);  clone_214 = getitem_215 = None
	        mul_15495: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7315, rsqrt_53);  sub_7315 = rsqrt_53 = None
	        mul_15496: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15495, arg724_1);  mul_15495 = arg724_1 = None
	        add_24520: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_15496, arg725_1);  mul_15496 = arg725_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2508: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_24520, [sym_size_int, 1500, 1280])
	        amin_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2508, [2])
	        amax_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2508, [2]);  view_2508 = None
	        full_320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_160, full_320);  amin_160 = full_320 = None
	        full_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_160, full_321);  amax_160 = full_321 = None
	        sub_7326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_160, minimum_160);  maximum_160 = None
	        div_320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7326, 255.0);  sub_7326 = None
	        clamp_min_480: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_320, 1.1920928955078125e-07);  div_320 = None
	        div_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_160, clamp_min_480);  minimum_160 = None
	        round_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_321);  div_321 = None
	        sub_7332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_321);  round_321 = None
	        clamp_min_481: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7332, -128);  sub_7332 = None
	        clamp_max_320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_481, 127);  clamp_min_481 = None
	        _assert_tensor_metadata_1442 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_480, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1442 = None
	        _assert_tensor_metadata_1443 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_320, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1443 = None
	        convert_element_type_960: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_320, torch.int8);  clamp_max_320 = None
	        view_2509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_24520, [sym_size_int, 1500, 1280]);  add_24520 = None
	        view_2510: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_480, [sym_size_int, 1500, 1])
	        view_2511: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_960, [sym_size_int, 1500, 1])
	        reciprocal_160: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2510);  view_2510 = None
	        mul_15544: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_160, 1.0);  reciprocal_160 = None
	        mul_15547: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2509, mul_15544);  view_2509 = mul_15544 = None
	        round_322: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15547);  mul_15547 = None
	        add_24607: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_322, view_2511);  round_322 = view_2511 = None
	        clamp_min_482: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24607, -128);  add_24607 = None
	        clamp_max_321: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_482, 127);  clamp_min_482 = None
	        view_2512: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_321, [sym_size_int, 1500, 1280]);  clamp_max_321 = None
	        _assert_tensor_metadata_1444 = torch.ops.aten._assert_tensor_metadata.default(view_2512, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1444 = None
	        convert_element_type_961: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2512, torch.int8);  view_2512 = None
	        view_2513: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_961, [sym_size_int, 1500, 1280]);  convert_element_type_961 = None
	        view_2514: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_480, [sym_size_int, 1500, 1]);  clamp_min_480 = None
	        view_2515: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_960, [sym_size_int, 1500, 1]);  convert_element_type_960 = None
	        _assert_tensor_metadata_1445 = torch.ops.aten._assert_tensor_metadata.default(view_2513, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1445 = None
	        convert_element_type_962: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2513, torch.float32);  view_2513 = None
	        _assert_tensor_metadata_1446 = torch.ops.aten._assert_tensor_metadata.default(view_2515, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1446 = None
	        convert_element_type_963: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2515, torch.float32);  view_2515 = None
	        sub_7352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_962, convert_element_type_963);  convert_element_type_962 = convert_element_type_963 = None
	        mul_15569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7352, view_2514);  sub_7352 = view_2514 = None
	        view_2516: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15569, [sym_size_int, 1500, 1280]);  mul_15569 = None
	        _assert_tensor_metadata_1447 = torch.ops.aten._assert_tensor_metadata.default(view_2516, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1447 = None
	        view_2517: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg727_1, [5120, 40, 32]);  arg727_1 = None
	        view_2518: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg728_1, [5120, 40, 1]);  arg728_1 = None
	        view_2519: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg729_1, [5120, 40, 1]);  arg729_1 = None
	        _assert_tensor_metadata_1448 = torch.ops.aten._assert_tensor_metadata.default(view_2517, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1448 = None
	        convert_element_type_964: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2517, torch.float32);  view_2517 = None
	        _assert_tensor_metadata_1449 = torch.ops.aten._assert_tensor_metadata.default(view_2519, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1449 = None
	        convert_element_type_965: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2519, torch.float32);  view_2519 = None
	        sub_7356: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_964, convert_element_type_965);  convert_element_type_964 = convert_element_type_965 = None
	        mul_15574: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7356, view_2518);  sub_7356 = view_2518 = None
	        view_2520: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15574, [5120, 1280]);  mul_15574 = None
	        _assert_tensor_metadata_1450 = torch.ops.aten._assert_tensor_metadata.default(view_2520, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1450 = None
	        mul_15579: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2521: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2516, [mul_15579, 1280]);  view_2516 = mul_15579 = None
	        permute_269: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2520, [1, 0]);  view_2520 = None
	        addmm_133: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg726_1, view_2521, permute_269);  arg726_1 = view_2521 = permute_269 = None
	        view_2522: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_133, [sym_size_int, 1500, 5120]);  addmm_133 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_15586: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2522, 0.5)
	        mul_15587: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2522, 0.7071067811865476);  view_2522 = None
	        erf_28: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_15587);  mul_15587 = None
	        add_24666: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_28, 1);  erf_28 = None
	        mul_15588: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15586, add_24666);  mul_15586 = add_24666 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_215: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_15588);  mul_15588 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2523: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_215, [sym_size_int, 1500, 5120])
	        amin_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2523, [2])
	        amax_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2523, [2]);  view_2523 = None
	        full_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_161, full_322);  amin_161 = full_322 = None
	        full_323: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_161, full_323);  amax_161 = full_323 = None
	        sub_7369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_161, minimum_161);  maximum_161 = None
	        div_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7369, 255.0);  sub_7369 = None
	        clamp_min_483: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_322, 1.1920928955078125e-07);  div_322 = None
	        div_323: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_161, clamp_min_483);  minimum_161 = None
	        round_323: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_323);  div_323 = None
	        sub_7375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_323);  round_323 = None
	        clamp_min_484: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7375, -128);  sub_7375 = None
	        clamp_max_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_484, 127);  clamp_min_484 = None
	        _assert_tensor_metadata_1451 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_483, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1451 = None
	        _assert_tensor_metadata_1452 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_322, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1452 = None
	        convert_element_type_966: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_322, torch.int8);  clamp_max_322 = None
	        view_2524: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_215, [sym_size_int, 1500, 5120]);  clone_215 = None
	        view_2525: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_483, [sym_size_int, 1500, 1])
	        view_2526: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_966, [sym_size_int, 1500, 1])
	        reciprocal_161: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2525);  view_2525 = None
	        mul_15634: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_161, 1.0);  reciprocal_161 = None
	        mul_15637: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2524, mul_15634);  view_2524 = mul_15634 = None
	        round_324: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_15637);  mul_15637 = None
	        add_24749: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_324, view_2526);  round_324 = view_2526 = None
	        clamp_min_485: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24749, -128);  add_24749 = None
	        clamp_max_323: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_485, 127);  clamp_min_485 = None
	        view_2527: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_323, [sym_size_int, 1500, 5120]);  clamp_max_323 = None
	        _assert_tensor_metadata_1453 = torch.ops.aten._assert_tensor_metadata.default(view_2527, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1453 = None
	        convert_element_type_967: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2527, torch.int8);  view_2527 = None
	        view_2528: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_967, [sym_size_int, 1500, 5120]);  convert_element_type_967 = None
	        view_2529: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_483, [sym_size_int, 1500, 1]);  clamp_min_483 = None
	        view_2530: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_966, [sym_size_int, 1500, 1]);  convert_element_type_966 = None
	        _assert_tensor_metadata_1454 = torch.ops.aten._assert_tensor_metadata.default(view_2528, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1454 = None
	        convert_element_type_968: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2528, torch.float32);  view_2528 = None
	        _assert_tensor_metadata_1455 = torch.ops.aten._assert_tensor_metadata.default(view_2530, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1455 = None
	        convert_element_type_969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2530, torch.float32);  view_2530 = None
	        sub_7395: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_968, convert_element_type_969);  convert_element_type_968 = convert_element_type_969 = None
	        mul_15659: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7395, view_2529);  sub_7395 = view_2529 = None
	        view_2531: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_15659, [sym_size_int, 1500, 5120]);  mul_15659 = None
	        _assert_tensor_metadata_1456 = torch.ops.aten._assert_tensor_metadata.default(view_2531, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1456 = None
	        view_2532: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg731_1, [1280, 160, 32]);  arg731_1 = None
	        view_2533: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg732_1, [1280, 160, 1]);  arg732_1 = None
	        view_2534: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg733_1, [1280, 160, 1]);  arg733_1 = None
	        _assert_tensor_metadata_1457 = torch.ops.aten._assert_tensor_metadata.default(view_2532, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1457 = None
	        convert_element_type_970: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2532, torch.float32);  view_2532 = None
	        _assert_tensor_metadata_1458 = torch.ops.aten._assert_tensor_metadata.default(view_2534, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1458 = None
	        convert_element_type_971: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2534, torch.float32);  view_2534 = None
	        sub_7399: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_970, convert_element_type_971);  convert_element_type_970 = convert_element_type_971 = None
	        mul_15664: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7399, view_2533);  sub_7399 = view_2533 = None
	        view_2535: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_15664, [1280, 5120]);  mul_15664 = None
	        _assert_tensor_metadata_1459 = torch.ops.aten._assert_tensor_metadata.default(view_2535, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1459 = None
	        mul_15669: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2536: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2531, [mul_15669, 5120]);  view_2531 = mul_15669 = None
	        permute_270: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2535, [1, 0]);  view_2535 = None
	        addmm_134: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg730_1, view_2536, permute_270);  arg730_1 = view_2536 = permute_270 = None
	        view_2537: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_134, [sym_size_int, 1500, 1280]);  addmm_134 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_216: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2537);  view_2537 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_24812: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_24514, clone_216);  add_24514 = clone_216 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_217: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_24812, memory_format = torch.contiguous_format)
	        var_mean_54 = torch.ops.aten.var_mean.correction(clone_217, [2], correction = 0, keepdim = True)
	        getitem_216: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_54[0]
	        getitem_217: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_54[1];  var_mean_54 = None
	        add_24817: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_216, 1e-05);  getitem_216 = None
	        rsqrt_54: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_24817);  add_24817 = None
	        sub_7405: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_217, getitem_217);  clone_217 = getitem_217 = None
	        mul_15680: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7405, rsqrt_54);  sub_7405 = rsqrt_54 = None
	        mul_15681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15680, arg734_1);  mul_15680 = arg734_1 = None
	        add_24818: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_15681, arg735_1);  mul_15681 = arg735_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2538: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_24818, [sym_size_int, 1500, 1280])
	        amin_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2538, [2])
	        amax_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2538, [2]);  view_2538 = None
	        full_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_162, full_324);  amin_162 = full_324 = None
	        full_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_162, full_325);  amax_162 = full_325 = None
	        sub_7416: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_162, minimum_162);  maximum_162 = None
	        div_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7416, 255.0);  sub_7416 = None
	        clamp_min_486: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_324, 1.1920928955078125e-07);  div_324 = None
	        div_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_162, clamp_min_486);  minimum_162 = None
	        round_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_325);  div_325 = None
	        sub_7422: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_325);  round_325 = None
	        clamp_min_487: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7422, -128);  sub_7422 = None
	        clamp_max_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_487, 127);  clamp_min_487 = None
	        _assert_tensor_metadata_1460 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_486, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1460 = None
	        _assert_tensor_metadata_1461 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_324, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1461 = None
	        convert_element_type_972: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_324, torch.int8);  clamp_max_324 = None
	        view_2539: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_24818, [sym_size_int, 1500, 1280])
	        view_2540: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_486, [sym_size_int, 1500, 1])
	        view_2541: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_972, [sym_size_int, 1500, 1])
	        reciprocal_162: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2540);  view_2540 = None
	        mul_15729: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_162, 1.0);  reciprocal_162 = None
	        mul_15732: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2539, mul_15729);  view_2539 = mul_15729 = None
	        round_326: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15732);  mul_15732 = None
	        add_24905: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_326, view_2541);  round_326 = view_2541 = None
	        clamp_min_488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24905, -128);  add_24905 = None
	        clamp_max_325: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_488, 127);  clamp_min_488 = None
	        view_2542: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_325, [sym_size_int, 1500, 1280]);  clamp_max_325 = None
	        _assert_tensor_metadata_1462 = torch.ops.aten._assert_tensor_metadata.default(view_2542, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1462 = None
	        convert_element_type_973: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2542, torch.int8);  view_2542 = None
	        view_2543: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_973, [sym_size_int, 1500, 1280]);  convert_element_type_973 = None
	        view_2544: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_486, [sym_size_int, 1500, 1]);  clamp_min_486 = None
	        view_2545: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_972, [sym_size_int, 1500, 1]);  convert_element_type_972 = None
	        _assert_tensor_metadata_1463 = torch.ops.aten._assert_tensor_metadata.default(view_2543, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1463 = None
	        convert_element_type_974: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2543, torch.float32);  view_2543 = None
	        _assert_tensor_metadata_1464 = torch.ops.aten._assert_tensor_metadata.default(view_2545, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1464 = None
	        convert_element_type_975: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2545, torch.float32);  view_2545 = None
	        sub_7442: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_974, convert_element_type_975);  convert_element_type_974 = convert_element_type_975 = None
	        mul_15754: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7442, view_2544);  sub_7442 = view_2544 = None
	        view_2546: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15754, [sym_size_int, 1500, 1280]);  mul_15754 = None
	        _assert_tensor_metadata_1465 = torch.ops.aten._assert_tensor_metadata.default(view_2546, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1465 = None
	        view_2547: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg737_1, [1280, 40, 32]);  arg737_1 = None
	        view_2548: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg738_1, [1280, 40, 1]);  arg738_1 = None
	        view_2549: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg739_1, [1280, 40, 1]);  arg739_1 = None
	        _assert_tensor_metadata_1466 = torch.ops.aten._assert_tensor_metadata.default(view_2547, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1466 = None
	        convert_element_type_976: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2547, torch.float32);  view_2547 = None
	        _assert_tensor_metadata_1467 = torch.ops.aten._assert_tensor_metadata.default(view_2549, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1467 = None
	        convert_element_type_977: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2549, torch.float32);  view_2549 = None
	        sub_7446: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_976, convert_element_type_977);  convert_element_type_976 = convert_element_type_977 = None
	        mul_15759: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7446, view_2548);  sub_7446 = view_2548 = None
	        view_2550: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15759, [1280, 1280]);  mul_15759 = None
	        _assert_tensor_metadata_1468 = torch.ops.aten._assert_tensor_metadata.default(view_2550, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1468 = None
	        mul_15764: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2551: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2546, [mul_15764, 1280]);  view_2546 = mul_15764 = None
	        permute_271: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2550, [1, 0]);  view_2550 = None
	        addmm_135: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg736_1, view_2551, permute_271);  arg736_1 = view_2551 = permute_271 = None
	        view_2552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_135, [sym_size_int, 1500, 1280]);  addmm_135 = None
	        mul_15771: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2552, 0.125);  view_2552 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2553: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_15771, [sym_size_int, 1500, 20, 64]);  mul_15771 = None
	        permute_272: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2553, [0, 2, 1, 3]);  view_2553 = None
	        clone_218: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_272, memory_format = torch.contiguous_format);  permute_272 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2554: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_24818, [sym_size_int, 1500, 1280])
	        amin_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2554, [2])
	        amax_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2554, [2]);  view_2554 = None
	        full_326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_163, full_326);  amin_163 = full_326 = None
	        full_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_163, full_327);  amax_163 = full_327 = None
	        sub_7461: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_163, minimum_163);  maximum_163 = None
	        div_326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7461, 255.0);  sub_7461 = None
	        clamp_min_489: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_326, 1.1920928955078125e-07);  div_326 = None
	        div_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_163, clamp_min_489);  minimum_163 = None
	        round_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_327);  div_327 = None
	        sub_7467: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_327);  round_327 = None
	        clamp_min_490: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7467, -128);  sub_7467 = None
	        clamp_max_326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_490, 127);  clamp_min_490 = None
	        _assert_tensor_metadata_1469 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_489, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1469 = None
	        _assert_tensor_metadata_1470 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_326, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1470 = None
	        convert_element_type_978: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_326, torch.int8);  clamp_max_326 = None
	        view_2555: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_24818, [sym_size_int, 1500, 1280])
	        view_2556: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_489, [sym_size_int, 1500, 1])
	        view_2557: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_978, [sym_size_int, 1500, 1])
	        reciprocal_163: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2556);  view_2556 = None
	        mul_15825: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_163, 1.0);  reciprocal_163 = None
	        mul_15828: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2555, mul_15825);  view_2555 = mul_15825 = None
	        round_328: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15828);  mul_15828 = None
	        add_25057: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_328, view_2557);  round_328 = view_2557 = None
	        clamp_min_491: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25057, -128);  add_25057 = None
	        clamp_max_327: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_491, 127);  clamp_min_491 = None
	        view_2558: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_327, [sym_size_int, 1500, 1280]);  clamp_max_327 = None
	        _assert_tensor_metadata_1471 = torch.ops.aten._assert_tensor_metadata.default(view_2558, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1471 = None
	        convert_element_type_979: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2558, torch.int8);  view_2558 = None
	        view_2559: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_979, [sym_size_int, 1500, 1280]);  convert_element_type_979 = None
	        view_2560: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_489, [sym_size_int, 1500, 1]);  clamp_min_489 = None
	        view_2561: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_978, [sym_size_int, 1500, 1]);  convert_element_type_978 = None
	        _assert_tensor_metadata_1472 = torch.ops.aten._assert_tensor_metadata.default(view_2559, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1472 = None
	        convert_element_type_980: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2559, torch.float32);  view_2559 = None
	        _assert_tensor_metadata_1473 = torch.ops.aten._assert_tensor_metadata.default(view_2561, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1473 = None
	        convert_element_type_981: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2561, torch.float32);  view_2561 = None
	        sub_7487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_980, convert_element_type_981);  convert_element_type_980 = convert_element_type_981 = None
	        mul_15850: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7487, view_2560);  sub_7487 = view_2560 = None
	        view_2562: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15850, [sym_size_int, 1500, 1280]);  mul_15850 = None
	        _assert_tensor_metadata_1474 = torch.ops.aten._assert_tensor_metadata.default(view_2562, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1474 = None
	        view_2563: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg740_1, [1280, 40, 32]);  arg740_1 = None
	        view_2564: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg741_1, [1280, 40, 1]);  arg741_1 = None
	        view_2565: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg742_1, [1280, 40, 1]);  arg742_1 = None
	        _assert_tensor_metadata_1475 = torch.ops.aten._assert_tensor_metadata.default(view_2563, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1475 = None
	        convert_element_type_982: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2563, torch.float32);  view_2563 = None
	        _assert_tensor_metadata_1476 = torch.ops.aten._assert_tensor_metadata.default(view_2565, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1476 = None
	        convert_element_type_983: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2565, torch.float32);  view_2565 = None
	        sub_7491: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_982, convert_element_type_983);  convert_element_type_982 = convert_element_type_983 = None
	        mul_15855: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7491, view_2564);  sub_7491 = view_2564 = None
	        view_2566: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15855, [1280, 1280]);  mul_15855 = None
	        _assert_tensor_metadata_1477 = torch.ops.aten._assert_tensor_metadata.default(view_2566, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1477 = None
	        permute_273: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2566, [1, 0]);  view_2566 = None
	        mul_15858: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2567: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2562, [mul_15858, 1280]);  view_2562 = mul_15858 = None
	        mm_27: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2567, permute_273);  view_2567 = permute_273 = None
	        view_2568: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_27, [sym_size_int, 1500, 1280]);  mm_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2569: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2568, [sym_size_int, -1, 20, 64]);  view_2568 = None
	        permute_274: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2569, [0, 2, 1, 3]);  view_2569 = None
	        clone_219: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_274, memory_format = torch.contiguous_format);  permute_274 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2570: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_24818, [sym_size_int, 1500, 1280])
	        amin_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2570, [2])
	        amax_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2570, [2]);  view_2570 = None
	        full_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_164, full_328);  amin_164 = full_328 = None
	        full_329: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_164, full_329);  amax_164 = full_329 = None
	        sub_7505: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_164, minimum_164);  maximum_164 = None
	        div_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7505, 255.0);  sub_7505 = None
	        clamp_min_492: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_328, 1.1920928955078125e-07);  div_328 = None
	        div_329: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_164, clamp_min_492);  minimum_164 = None
	        round_329: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_329);  div_329 = None
	        sub_7511: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_329);  round_329 = None
	        clamp_min_493: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7511, -128);  sub_7511 = None
	        clamp_max_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_493, 127);  clamp_min_493 = None
	        _assert_tensor_metadata_1478 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_492, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1478 = None
	        _assert_tensor_metadata_1479 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_328, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1479 = None
	        convert_element_type_984: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_328, torch.int8);  clamp_max_328 = None
	        view_2571: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_24818, [sym_size_int, 1500, 1280]);  add_24818 = None
	        view_2572: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_492, [sym_size_int, 1500, 1])
	        view_2573: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_984, [sym_size_int, 1500, 1])
	        reciprocal_164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2572);  view_2572 = None
	        mul_15924: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_164, 1.0);  reciprocal_164 = None
	        mul_15927: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2571, mul_15924);  view_2571 = mul_15924 = None
	        round_330: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15927);  mul_15927 = None
	        add_25205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_330, view_2573);  round_330 = view_2573 = None
	        clamp_min_494: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25205, -128);  add_25205 = None
	        clamp_max_329: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_494, 127);  clamp_min_494 = None
	        view_2574: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_329, [sym_size_int, 1500, 1280]);  clamp_max_329 = None
	        _assert_tensor_metadata_1480 = torch.ops.aten._assert_tensor_metadata.default(view_2574, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1480 = None
	        convert_element_type_985: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2574, torch.int8);  view_2574 = None
	        view_2575: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_985, [sym_size_int, 1500, 1280]);  convert_element_type_985 = None
	        view_2576: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_492, [sym_size_int, 1500, 1]);  clamp_min_492 = None
	        view_2577: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_984, [sym_size_int, 1500, 1]);  convert_element_type_984 = None
	        _assert_tensor_metadata_1481 = torch.ops.aten._assert_tensor_metadata.default(view_2575, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1481 = None
	        convert_element_type_986: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2575, torch.float32);  view_2575 = None
	        _assert_tensor_metadata_1482 = torch.ops.aten._assert_tensor_metadata.default(view_2577, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1482 = None
	        convert_element_type_987: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2577, torch.float32);  view_2577 = None
	        sub_7531: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_986, convert_element_type_987);  convert_element_type_986 = convert_element_type_987 = None
	        mul_15949: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7531, view_2576);  sub_7531 = view_2576 = None
	        view_2578: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15949, [sym_size_int, 1500, 1280]);  mul_15949 = None
	        _assert_tensor_metadata_1483 = torch.ops.aten._assert_tensor_metadata.default(view_2578, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1483 = None
	        view_2579: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg744_1, [1280, 40, 32]);  arg744_1 = None
	        view_2580: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg745_1, [1280, 40, 1]);  arg745_1 = None
	        view_2581: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg746_1, [1280, 40, 1]);  arg746_1 = None
	        _assert_tensor_metadata_1484 = torch.ops.aten._assert_tensor_metadata.default(view_2579, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1484 = None
	        convert_element_type_988: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2579, torch.float32);  view_2579 = None
	        _assert_tensor_metadata_1485 = torch.ops.aten._assert_tensor_metadata.default(view_2581, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1485 = None
	        convert_element_type_989: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2581, torch.float32);  view_2581 = None
	        sub_7535: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_988, convert_element_type_989);  convert_element_type_988 = convert_element_type_989 = None
	        mul_15954: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7535, view_2580);  sub_7535 = view_2580 = None
	        view_2582: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15954, [1280, 1280]);  mul_15954 = None
	        _assert_tensor_metadata_1486 = torch.ops.aten._assert_tensor_metadata.default(view_2582, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1486 = None
	        mul_15959: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2583: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2578, [mul_15959, 1280]);  view_2578 = mul_15959 = None
	        permute_275: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2582, [1, 0]);  view_2582 = None
	        addmm_136: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg743_1, view_2583, permute_275);  arg743_1 = view_2583 = permute_275 = None
	        view_2584: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_136, [sym_size_int, 1500, 1280]);  addmm_136 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2585: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2584, [sym_size_int, -1, 20, 64]);  view_2584 = None
	        permute_276: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2585, [0, 2, 1, 3]);  view_2585 = None
	        clone_220: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_276, memory_format = torch.contiguous_format);  permute_276 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_27 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_218, clone_219, clone_220, None, False, scale = 1.0);  clone_218 = clone_219 = clone_220 = None
	        getitem_218: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_27[0];  _scaled_dot_product_efficient_attention_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_277: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_218, [0, 2, 1, 3]);  getitem_218 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2586: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_277, [sym_size_int, 1500, -1]);  permute_277 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2587: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2586, [sym_size_int, 1500, 1280])
	        amin_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2587, [2])
	        amax_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2587, [2]);  view_2587 = None
	        full_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_165, full_330);  amin_165 = full_330 = None
	        full_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_165, full_331);  amax_165 = full_331 = None
	        sub_7553: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_165, minimum_165);  maximum_165 = None
	        div_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7553, 255.0);  sub_7553 = None
	        clamp_min_495: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_330, 1.1920928955078125e-07);  div_330 = None
	        div_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_165, clamp_min_495);  minimum_165 = None
	        round_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_331);  div_331 = None
	        sub_7559: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_331);  round_331 = None
	        clamp_min_496: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7559, -128);  sub_7559 = None
	        clamp_max_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_496, 127);  clamp_min_496 = None
	        _assert_tensor_metadata_1487 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_495, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1487 = None
	        _assert_tensor_metadata_1488 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_330, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1488 = None
	        convert_element_type_990: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_330, torch.int8);  clamp_max_330 = None
	        view_2588: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2586, [sym_size_int, 1500, 1280]);  view_2586 = None
	        view_2589: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_495, [sym_size_int, 1500, 1])
	        view_2590: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_990, [sym_size_int, 1500, 1])
	        reciprocal_165: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2589);  view_2589 = None
	        mul_16029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_165, 1.0);  reciprocal_165 = None
	        mul_16032: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2588, mul_16029);  view_2588 = mul_16029 = None
	        round_332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16032);  mul_16032 = None
	        add_25369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_332, view_2590);  round_332 = view_2590 = None
	        clamp_min_497: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25369, -128);  add_25369 = None
	        clamp_max_331: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_497, 127);  clamp_min_497 = None
	        view_2591: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_331, [sym_size_int, 1500, 1280]);  clamp_max_331 = None
	        _assert_tensor_metadata_1489 = torch.ops.aten._assert_tensor_metadata.default(view_2591, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1489 = None
	        convert_element_type_991: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2591, torch.int8);  view_2591 = None
	        view_2592: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_991, [sym_size_int, 1500, 1280]);  convert_element_type_991 = None
	        view_2593: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_495, [sym_size_int, 1500, 1]);  clamp_min_495 = None
	        view_2594: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_990, [sym_size_int, 1500, 1]);  convert_element_type_990 = None
	        _assert_tensor_metadata_1490 = torch.ops.aten._assert_tensor_metadata.default(view_2592, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1490 = None
	        convert_element_type_992: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2592, torch.float32);  view_2592 = None
	        _assert_tensor_metadata_1491 = torch.ops.aten._assert_tensor_metadata.default(view_2594, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1491 = None
	        convert_element_type_993: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2594, torch.float32);  view_2594 = None
	        sub_7579: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_992, convert_element_type_993);  convert_element_type_992 = convert_element_type_993 = None
	        mul_16054: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7579, view_2593);  sub_7579 = view_2593 = None
	        view_2595: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16054, [sym_size_int, 1500, 1280]);  mul_16054 = None
	        _assert_tensor_metadata_1492 = torch.ops.aten._assert_tensor_metadata.default(view_2595, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1492 = None
	        view_2596: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg748_1, [1280, 40, 32]);  arg748_1 = None
	        view_2597: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg749_1, [1280, 40, 1]);  arg749_1 = None
	        view_2598: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg750_1, [1280, 40, 1]);  arg750_1 = None
	        _assert_tensor_metadata_1493 = torch.ops.aten._assert_tensor_metadata.default(view_2596, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1493 = None
	        convert_element_type_994: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2596, torch.float32);  view_2596 = None
	        _assert_tensor_metadata_1494 = torch.ops.aten._assert_tensor_metadata.default(view_2598, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1494 = None
	        convert_element_type_995: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2598, torch.float32);  view_2598 = None
	        sub_7583: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_994, convert_element_type_995);  convert_element_type_994 = convert_element_type_995 = None
	        mul_16059: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7583, view_2597);  sub_7583 = view_2597 = None
	        view_2599: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16059, [1280, 1280]);  mul_16059 = None
	        _assert_tensor_metadata_1495 = torch.ops.aten._assert_tensor_metadata.default(view_2599, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1495 = None
	        mul_16064: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2600: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2595, [mul_16064, 1280]);  view_2595 = mul_16064 = None
	        permute_278: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2599, [1, 0]);  view_2599 = None
	        addmm_137: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg747_1, view_2600, permute_278);  arg747_1 = view_2600 = permute_278 = None
	        view_2601: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_137, [sym_size_int, 1500, 1280]);  addmm_137 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2601);  view_2601 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_25432: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_24812, clone_221);  add_24812 = clone_221 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_222: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_25432, memory_format = torch.contiguous_format)
	        var_mean_55 = torch.ops.aten.var_mean.correction(clone_222, [2], correction = 0, keepdim = True)
	        getitem_222: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_55[0]
	        getitem_223: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_55[1];  var_mean_55 = None
	        add_25437: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_222, 1e-05);  getitem_222 = None
	        rsqrt_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_25437);  add_25437 = None
	        sub_7589: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_222, getitem_223);  clone_222 = getitem_223 = None
	        mul_16075: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7589, rsqrt_55);  sub_7589 = rsqrt_55 = None
	        mul_16076: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16075, arg751_1);  mul_16075 = arg751_1 = None
	        add_25438: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16076, arg752_1);  mul_16076 = arg752_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2602: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_25438, [sym_size_int, 1500, 1280])
	        amin_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2602, [2])
	        amax_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2602, [2]);  view_2602 = None
	        full_332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_166, full_332);  amin_166 = full_332 = None
	        full_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_166, full_333);  amax_166 = full_333 = None
	        sub_7600: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_166, minimum_166);  maximum_166 = None
	        div_332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7600, 255.0);  sub_7600 = None
	        clamp_min_498: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_332, 1.1920928955078125e-07);  div_332 = None
	        div_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_166, clamp_min_498);  minimum_166 = None
	        round_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_333);  div_333 = None
	        sub_7606: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_333);  round_333 = None
	        clamp_min_499: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7606, -128);  sub_7606 = None
	        clamp_max_332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_499, 127);  clamp_min_499 = None
	        _assert_tensor_metadata_1496 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_498, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1496 = None
	        _assert_tensor_metadata_1497 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_332, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1497 = None
	        convert_element_type_996: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_332, torch.int8);  clamp_max_332 = None
	        view_2603: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_25438, [sym_size_int, 1500, 1280]);  add_25438 = None
	        view_2604: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_498, [sym_size_int, 1500, 1])
	        view_2605: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_996, [sym_size_int, 1500, 1])
	        reciprocal_166: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2604);  view_2604 = None
	        mul_16124: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_166, 1.0);  reciprocal_166 = None
	        mul_16127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2603, mul_16124);  view_2603 = mul_16124 = None
	        round_334: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16127);  mul_16127 = None
	        add_25525: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_334, view_2605);  round_334 = view_2605 = None
	        clamp_min_500: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25525, -128);  add_25525 = None
	        clamp_max_333: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_500, 127);  clamp_min_500 = None
	        view_2606: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_333, [sym_size_int, 1500, 1280]);  clamp_max_333 = None
	        _assert_tensor_metadata_1498 = torch.ops.aten._assert_tensor_metadata.default(view_2606, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1498 = None
	        convert_element_type_997: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2606, torch.int8);  view_2606 = None
	        view_2607: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_997, [sym_size_int, 1500, 1280]);  convert_element_type_997 = None
	        view_2608: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_498, [sym_size_int, 1500, 1]);  clamp_min_498 = None
	        view_2609: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_996, [sym_size_int, 1500, 1]);  convert_element_type_996 = None
	        _assert_tensor_metadata_1499 = torch.ops.aten._assert_tensor_metadata.default(view_2607, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1499 = None
	        convert_element_type_998: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2607, torch.float32);  view_2607 = None
	        _assert_tensor_metadata_1500 = torch.ops.aten._assert_tensor_metadata.default(view_2609, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1500 = None
	        convert_element_type_999: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2609, torch.float32);  view_2609 = None
	        sub_7626: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_998, convert_element_type_999);  convert_element_type_998 = convert_element_type_999 = None
	        mul_16149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7626, view_2608);  sub_7626 = view_2608 = None
	        view_2610: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16149, [sym_size_int, 1500, 1280]);  mul_16149 = None
	        _assert_tensor_metadata_1501 = torch.ops.aten._assert_tensor_metadata.default(view_2610, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1501 = None
	        view_2611: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg754_1, [5120, 40, 32]);  arg754_1 = None
	        view_2612: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg755_1, [5120, 40, 1]);  arg755_1 = None
	        view_2613: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg756_1, [5120, 40, 1]);  arg756_1 = None
	        _assert_tensor_metadata_1502 = torch.ops.aten._assert_tensor_metadata.default(view_2611, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1502 = None
	        convert_element_type_1000: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2611, torch.float32);  view_2611 = None
	        _assert_tensor_metadata_1503 = torch.ops.aten._assert_tensor_metadata.default(view_2613, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1503 = None
	        convert_element_type_1001: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2613, torch.float32);  view_2613 = None
	        sub_7630: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1000, convert_element_type_1001);  convert_element_type_1000 = convert_element_type_1001 = None
	        mul_16154: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7630, view_2612);  sub_7630 = view_2612 = None
	        view_2614: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16154, [5120, 1280]);  mul_16154 = None
	        _assert_tensor_metadata_1504 = torch.ops.aten._assert_tensor_metadata.default(view_2614, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1504 = None
	        mul_16159: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2615: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2610, [mul_16159, 1280]);  view_2610 = mul_16159 = None
	        permute_279: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2614, [1, 0]);  view_2614 = None
	        addmm_138: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg753_1, view_2615, permute_279);  arg753_1 = view_2615 = permute_279 = None
	        view_2616: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_138, [sym_size_int, 1500, 5120]);  addmm_138 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_16166: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2616, 0.5)
	        mul_16167: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2616, 0.7071067811865476);  view_2616 = None
	        erf_29: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_16167);  mul_16167 = None
	        add_25584: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_29, 1);  erf_29 = None
	        mul_16168: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16166, add_25584);  mul_16166 = add_25584 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_223: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_16168);  mul_16168 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2617: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_223, [sym_size_int, 1500, 5120])
	        amin_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2617, [2])
	        amax_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2617, [2]);  view_2617 = None
	        full_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_167, full_334);  amin_167 = full_334 = None
	        full_335: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_167, full_335);  amax_167 = full_335 = None
	        sub_7643: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_167, minimum_167);  maximum_167 = None
	        div_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7643, 255.0);  sub_7643 = None
	        clamp_min_501: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_334, 1.1920928955078125e-07);  div_334 = None
	        div_335: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_167, clamp_min_501);  minimum_167 = None
	        round_335: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_335);  div_335 = None
	        sub_7649: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_335);  round_335 = None
	        clamp_min_502: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7649, -128);  sub_7649 = None
	        clamp_max_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_502, 127);  clamp_min_502 = None
	        _assert_tensor_metadata_1505 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_501, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1505 = None
	        _assert_tensor_metadata_1506 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_334, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1506 = None
	        convert_element_type_1002: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_334, torch.int8);  clamp_max_334 = None
	        view_2618: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_223, [sym_size_int, 1500, 5120]);  clone_223 = None
	        view_2619: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_501, [sym_size_int, 1500, 1])
	        view_2620: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1002, [sym_size_int, 1500, 1])
	        reciprocal_167: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2619);  view_2619 = None
	        mul_16214: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_167, 1.0);  reciprocal_167 = None
	        mul_16217: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2618, mul_16214);  view_2618 = mul_16214 = None
	        round_336: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_16217);  mul_16217 = None
	        add_25667: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_336, view_2620);  round_336 = view_2620 = None
	        clamp_min_503: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25667, -128);  add_25667 = None
	        clamp_max_335: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_503, 127);  clamp_min_503 = None
	        view_2621: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_335, [sym_size_int, 1500, 5120]);  clamp_max_335 = None
	        _assert_tensor_metadata_1507 = torch.ops.aten._assert_tensor_metadata.default(view_2621, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1507 = None
	        convert_element_type_1003: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2621, torch.int8);  view_2621 = None
	        view_2622: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1003, [sym_size_int, 1500, 5120]);  convert_element_type_1003 = None
	        view_2623: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_501, [sym_size_int, 1500, 1]);  clamp_min_501 = None
	        view_2624: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1002, [sym_size_int, 1500, 1]);  convert_element_type_1002 = None
	        _assert_tensor_metadata_1508 = torch.ops.aten._assert_tensor_metadata.default(view_2622, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1508 = None
	        convert_element_type_1004: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2622, torch.float32);  view_2622 = None
	        _assert_tensor_metadata_1509 = torch.ops.aten._assert_tensor_metadata.default(view_2624, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1509 = None
	        convert_element_type_1005: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2624, torch.float32);  view_2624 = None
	        sub_7669: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1004, convert_element_type_1005);  convert_element_type_1004 = convert_element_type_1005 = None
	        mul_16239: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7669, view_2623);  sub_7669 = view_2623 = None
	        view_2625: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_16239, [sym_size_int, 1500, 5120]);  mul_16239 = None
	        _assert_tensor_metadata_1510 = torch.ops.aten._assert_tensor_metadata.default(view_2625, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1510 = None
	        view_2626: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg758_1, [1280, 160, 32]);  arg758_1 = None
	        view_2627: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg759_1, [1280, 160, 1]);  arg759_1 = None
	        view_2628: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg760_1, [1280, 160, 1]);  arg760_1 = None
	        _assert_tensor_metadata_1511 = torch.ops.aten._assert_tensor_metadata.default(view_2626, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1511 = None
	        convert_element_type_1006: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2626, torch.float32);  view_2626 = None
	        _assert_tensor_metadata_1512 = torch.ops.aten._assert_tensor_metadata.default(view_2628, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1512 = None
	        convert_element_type_1007: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2628, torch.float32);  view_2628 = None
	        sub_7673: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1006, convert_element_type_1007);  convert_element_type_1006 = convert_element_type_1007 = None
	        mul_16244: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7673, view_2627);  sub_7673 = view_2627 = None
	        view_2629: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_16244, [1280, 5120]);  mul_16244 = None
	        _assert_tensor_metadata_1513 = torch.ops.aten._assert_tensor_metadata.default(view_2629, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1513 = None
	        mul_16249: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2630: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2625, [mul_16249, 5120]);  view_2625 = mul_16249 = None
	        permute_280: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2629, [1, 0]);  view_2629 = None
	        addmm_139: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg757_1, view_2630, permute_280);  arg757_1 = view_2630 = permute_280 = None
	        view_2631: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_139, [sym_size_int, 1500, 1280]);  addmm_139 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2631);  view_2631 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_25730: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_25432, clone_224);  add_25432 = clone_224 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_225: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_25730, memory_format = torch.contiguous_format)
	        var_mean_56 = torch.ops.aten.var_mean.correction(clone_225, [2], correction = 0, keepdim = True)
	        getitem_224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_56[0]
	        getitem_225: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_56[1];  var_mean_56 = None
	        add_25735: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_224, 1e-05);  getitem_224 = None
	        rsqrt_56: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_25735);  add_25735 = None
	        sub_7679: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_225, getitem_225);  clone_225 = getitem_225 = None
	        mul_16260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7679, rsqrt_56);  sub_7679 = rsqrt_56 = None
	        mul_16261: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16260, arg761_1);  mul_16260 = arg761_1 = None
	        add_25736: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16261, arg762_1);  mul_16261 = arg762_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2632: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_25736, [sym_size_int, 1500, 1280])
	        amin_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2632, [2])
	        amax_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2632, [2]);  view_2632 = None
	        full_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_168, full_336);  amin_168 = full_336 = None
	        full_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_168, full_337);  amax_168 = full_337 = None
	        sub_7690: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_168, minimum_168);  maximum_168 = None
	        div_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7690, 255.0);  sub_7690 = None
	        clamp_min_504: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_336, 1.1920928955078125e-07);  div_336 = None
	        div_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_168, clamp_min_504);  minimum_168 = None
	        round_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_337);  div_337 = None
	        sub_7696: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_337);  round_337 = None
	        clamp_min_505: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7696, -128);  sub_7696 = None
	        clamp_max_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_505, 127);  clamp_min_505 = None
	        _assert_tensor_metadata_1514 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_504, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1514 = None
	        _assert_tensor_metadata_1515 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_336, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1515 = None
	        convert_element_type_1008: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_336, torch.int8);  clamp_max_336 = None
	        view_2633: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_25736, [sym_size_int, 1500, 1280])
	        view_2634: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_504, [sym_size_int, 1500, 1])
	        view_2635: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1008, [sym_size_int, 1500, 1])
	        reciprocal_168: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2634);  view_2634 = None
	        mul_16309: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_168, 1.0);  reciprocal_168 = None
	        mul_16312: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2633, mul_16309);  view_2633 = mul_16309 = None
	        round_338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16312);  mul_16312 = None
	        add_25823: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_338, view_2635);  round_338 = view_2635 = None
	        clamp_min_506: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25823, -128);  add_25823 = None
	        clamp_max_337: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_506, 127);  clamp_min_506 = None
	        view_2636: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_337, [sym_size_int, 1500, 1280]);  clamp_max_337 = None
	        _assert_tensor_metadata_1516 = torch.ops.aten._assert_tensor_metadata.default(view_2636, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1516 = None
	        convert_element_type_1009: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2636, torch.int8);  view_2636 = None
	        view_2637: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1009, [sym_size_int, 1500, 1280]);  convert_element_type_1009 = None
	        view_2638: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_504, [sym_size_int, 1500, 1]);  clamp_min_504 = None
	        view_2639: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1008, [sym_size_int, 1500, 1]);  convert_element_type_1008 = None
	        _assert_tensor_metadata_1517 = torch.ops.aten._assert_tensor_metadata.default(view_2637, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1517 = None
	        convert_element_type_1010: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2637, torch.float32);  view_2637 = None
	        _assert_tensor_metadata_1518 = torch.ops.aten._assert_tensor_metadata.default(view_2639, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1518 = None
	        convert_element_type_1011: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2639, torch.float32);  view_2639 = None
	        sub_7716: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1010, convert_element_type_1011);  convert_element_type_1010 = convert_element_type_1011 = None
	        mul_16334: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7716, view_2638);  sub_7716 = view_2638 = None
	        view_2640: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16334, [sym_size_int, 1500, 1280]);  mul_16334 = None
	        _assert_tensor_metadata_1519 = torch.ops.aten._assert_tensor_metadata.default(view_2640, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1519 = None
	        view_2641: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg764_1, [1280, 40, 32]);  arg764_1 = None
	        view_2642: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg765_1, [1280, 40, 1]);  arg765_1 = None
	        view_2643: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg766_1, [1280, 40, 1]);  arg766_1 = None
	        _assert_tensor_metadata_1520 = torch.ops.aten._assert_tensor_metadata.default(view_2641, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1520 = None
	        convert_element_type_1012: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2641, torch.float32);  view_2641 = None
	        _assert_tensor_metadata_1521 = torch.ops.aten._assert_tensor_metadata.default(view_2643, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1521 = None
	        convert_element_type_1013: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2643, torch.float32);  view_2643 = None
	        sub_7720: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1012, convert_element_type_1013);  convert_element_type_1012 = convert_element_type_1013 = None
	        mul_16339: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7720, view_2642);  sub_7720 = view_2642 = None
	        view_2644: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16339, [1280, 1280]);  mul_16339 = None
	        _assert_tensor_metadata_1522 = torch.ops.aten._assert_tensor_metadata.default(view_2644, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1522 = None
	        mul_16344: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2645: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2640, [mul_16344, 1280]);  view_2640 = mul_16344 = None
	        permute_281: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2644, [1, 0]);  view_2644 = None
	        addmm_140: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg763_1, view_2645, permute_281);  arg763_1 = view_2645 = permute_281 = None
	        view_2646: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_140, [sym_size_int, 1500, 1280]);  addmm_140 = None
	        mul_16351: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2646, 0.125);  view_2646 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2647: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_16351, [sym_size_int, 1500, 20, 64]);  mul_16351 = None
	        permute_282: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2647, [0, 2, 1, 3]);  view_2647 = None
	        clone_226: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_282, memory_format = torch.contiguous_format);  permute_282 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2648: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_25736, [sym_size_int, 1500, 1280])
	        amin_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2648, [2])
	        amax_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2648, [2]);  view_2648 = None
	        full_338: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_169, full_338);  amin_169 = full_338 = None
	        full_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_169, full_339);  amax_169 = full_339 = None
	        sub_7735: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_169, minimum_169);  maximum_169 = None
	        div_338: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7735, 255.0);  sub_7735 = None
	        clamp_min_507: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_338, 1.1920928955078125e-07);  div_338 = None
	        div_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_169, clamp_min_507);  minimum_169 = None
	        round_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_339);  div_339 = None
	        sub_7741: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_339);  round_339 = None
	        clamp_min_508: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7741, -128);  sub_7741 = None
	        clamp_max_338: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_508, 127);  clamp_min_508 = None
	        _assert_tensor_metadata_1523 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_507, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1523 = None
	        _assert_tensor_metadata_1524 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_338, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1524 = None
	        convert_element_type_1014: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_338, torch.int8);  clamp_max_338 = None
	        view_2649: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_25736, [sym_size_int, 1500, 1280])
	        view_2650: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_507, [sym_size_int, 1500, 1])
	        view_2651: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1014, [sym_size_int, 1500, 1])
	        reciprocal_169: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2650);  view_2650 = None
	        mul_16405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_169, 1.0);  reciprocal_169 = None
	        mul_16408: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2649, mul_16405);  view_2649 = mul_16405 = None
	        round_340: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16408);  mul_16408 = None
	        add_25975: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_340, view_2651);  round_340 = view_2651 = None
	        clamp_min_509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25975, -128);  add_25975 = None
	        clamp_max_339: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_509, 127);  clamp_min_509 = None
	        view_2652: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_339, [sym_size_int, 1500, 1280]);  clamp_max_339 = None
	        _assert_tensor_metadata_1525 = torch.ops.aten._assert_tensor_metadata.default(view_2652, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1525 = None
	        convert_element_type_1015: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2652, torch.int8);  view_2652 = None
	        view_2653: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1015, [sym_size_int, 1500, 1280]);  convert_element_type_1015 = None
	        view_2654: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_507, [sym_size_int, 1500, 1]);  clamp_min_507 = None
	        view_2655: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1014, [sym_size_int, 1500, 1]);  convert_element_type_1014 = None
	        _assert_tensor_metadata_1526 = torch.ops.aten._assert_tensor_metadata.default(view_2653, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1526 = None
	        convert_element_type_1016: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2653, torch.float32);  view_2653 = None
	        _assert_tensor_metadata_1527 = torch.ops.aten._assert_tensor_metadata.default(view_2655, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1527 = None
	        convert_element_type_1017: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2655, torch.float32);  view_2655 = None
	        sub_7761: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1016, convert_element_type_1017);  convert_element_type_1016 = convert_element_type_1017 = None
	        mul_16430: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7761, view_2654);  sub_7761 = view_2654 = None
	        view_2656: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16430, [sym_size_int, 1500, 1280]);  mul_16430 = None
	        _assert_tensor_metadata_1528 = torch.ops.aten._assert_tensor_metadata.default(view_2656, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1528 = None
	        view_2657: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg767_1, [1280, 40, 32]);  arg767_1 = None
	        view_2658: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg768_1, [1280, 40, 1]);  arg768_1 = None
	        view_2659: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg769_1, [1280, 40, 1]);  arg769_1 = None
	        _assert_tensor_metadata_1529 = torch.ops.aten._assert_tensor_metadata.default(view_2657, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1529 = None
	        convert_element_type_1018: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2657, torch.float32);  view_2657 = None
	        _assert_tensor_metadata_1530 = torch.ops.aten._assert_tensor_metadata.default(view_2659, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1530 = None
	        convert_element_type_1019: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2659, torch.float32);  view_2659 = None
	        sub_7765: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1018, convert_element_type_1019);  convert_element_type_1018 = convert_element_type_1019 = None
	        mul_16435: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7765, view_2658);  sub_7765 = view_2658 = None
	        view_2660: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16435, [1280, 1280]);  mul_16435 = None
	        _assert_tensor_metadata_1531 = torch.ops.aten._assert_tensor_metadata.default(view_2660, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1531 = None
	        permute_283: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2660, [1, 0]);  view_2660 = None
	        mul_16438: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2661: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2656, [mul_16438, 1280]);  view_2656 = mul_16438 = None
	        mm_28: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2661, permute_283);  view_2661 = permute_283 = None
	        view_2662: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_28, [sym_size_int, 1500, 1280]);  mm_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2663: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2662, [sym_size_int, -1, 20, 64]);  view_2662 = None
	        permute_284: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2663, [0, 2, 1, 3]);  view_2663 = None
	        clone_227: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_284, memory_format = torch.contiguous_format);  permute_284 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2664: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_25736, [sym_size_int, 1500, 1280])
	        amin_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2664, [2])
	        amax_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2664, [2]);  view_2664 = None
	        full_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_170, full_340);  amin_170 = full_340 = None
	        full_341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_170, full_341);  amax_170 = full_341 = None
	        sub_7779: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_170, minimum_170);  maximum_170 = None
	        div_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7779, 255.0);  sub_7779 = None
	        clamp_min_510: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_340, 1.1920928955078125e-07);  div_340 = None
	        div_341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_170, clamp_min_510);  minimum_170 = None
	        round_341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_341);  div_341 = None
	        sub_7785: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_341);  round_341 = None
	        clamp_min_511: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7785, -128);  sub_7785 = None
	        clamp_max_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_511, 127);  clamp_min_511 = None
	        _assert_tensor_metadata_1532 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_510, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1532 = None
	        _assert_tensor_metadata_1533 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_340, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1533 = None
	        convert_element_type_1020: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_340, torch.int8);  clamp_max_340 = None
	        view_2665: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_25736, [sym_size_int, 1500, 1280]);  add_25736 = None
	        view_2666: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_510, [sym_size_int, 1500, 1])
	        view_2667: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1020, [sym_size_int, 1500, 1])
	        reciprocal_170: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2666);  view_2666 = None
	        mul_16504: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_170, 1.0);  reciprocal_170 = None
	        mul_16507: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2665, mul_16504);  view_2665 = mul_16504 = None
	        round_342: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16507);  mul_16507 = None
	        add_26123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_342, view_2667);  round_342 = view_2667 = None
	        clamp_min_512: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26123, -128);  add_26123 = None
	        clamp_max_341: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_512, 127);  clamp_min_512 = None
	        view_2668: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_341, [sym_size_int, 1500, 1280]);  clamp_max_341 = None
	        _assert_tensor_metadata_1534 = torch.ops.aten._assert_tensor_metadata.default(view_2668, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1534 = None
	        convert_element_type_1021: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2668, torch.int8);  view_2668 = None
	        view_2669: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1021, [sym_size_int, 1500, 1280]);  convert_element_type_1021 = None
	        view_2670: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_510, [sym_size_int, 1500, 1]);  clamp_min_510 = None
	        view_2671: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1020, [sym_size_int, 1500, 1]);  convert_element_type_1020 = None
	        _assert_tensor_metadata_1535 = torch.ops.aten._assert_tensor_metadata.default(view_2669, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1535 = None
	        convert_element_type_1022: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2669, torch.float32);  view_2669 = None
	        _assert_tensor_metadata_1536 = torch.ops.aten._assert_tensor_metadata.default(view_2671, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1536 = None
	        convert_element_type_1023: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2671, torch.float32);  view_2671 = None
	        sub_7805: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1022, convert_element_type_1023);  convert_element_type_1022 = convert_element_type_1023 = None
	        mul_16529: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7805, view_2670);  sub_7805 = view_2670 = None
	        view_2672: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16529, [sym_size_int, 1500, 1280]);  mul_16529 = None
	        _assert_tensor_metadata_1537 = torch.ops.aten._assert_tensor_metadata.default(view_2672, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1537 = None
	        view_2673: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg771_1, [1280, 40, 32]);  arg771_1 = None
	        view_2674: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg772_1, [1280, 40, 1]);  arg772_1 = None
	        view_2675: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg773_1, [1280, 40, 1]);  arg773_1 = None
	        _assert_tensor_metadata_1538 = torch.ops.aten._assert_tensor_metadata.default(view_2673, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1538 = None
	        convert_element_type_1024: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2673, torch.float32);  view_2673 = None
	        _assert_tensor_metadata_1539 = torch.ops.aten._assert_tensor_metadata.default(view_2675, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1539 = None
	        convert_element_type_1025: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2675, torch.float32);  view_2675 = None
	        sub_7809: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1024, convert_element_type_1025);  convert_element_type_1024 = convert_element_type_1025 = None
	        mul_16534: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7809, view_2674);  sub_7809 = view_2674 = None
	        view_2676: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16534, [1280, 1280]);  mul_16534 = None
	        _assert_tensor_metadata_1540 = torch.ops.aten._assert_tensor_metadata.default(view_2676, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1540 = None
	        mul_16539: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2677: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2672, [mul_16539, 1280]);  view_2672 = mul_16539 = None
	        permute_285: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2676, [1, 0]);  view_2676 = None
	        addmm_141: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg770_1, view_2677, permute_285);  arg770_1 = view_2677 = permute_285 = None
	        view_2678: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_141, [sym_size_int, 1500, 1280]);  addmm_141 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2679: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2678, [sym_size_int, -1, 20, 64]);  view_2678 = None
	        permute_286: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2679, [0, 2, 1, 3]);  view_2679 = None
	        clone_228: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_286, memory_format = torch.contiguous_format);  permute_286 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_28 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_226, clone_227, clone_228, None, False, scale = 1.0);  clone_226 = clone_227 = clone_228 = None
	        getitem_226: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_28[0];  _scaled_dot_product_efficient_attention_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_287: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_226, [0, 2, 1, 3]);  getitem_226 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2680: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_287, [sym_size_int, 1500, -1]);  permute_287 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2680, [sym_size_int, 1500, 1280])
	        amin_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2681, [2])
	        amax_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2681, [2]);  view_2681 = None
	        full_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_171, full_342);  amin_171 = full_342 = None
	        full_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_171, full_343);  amax_171 = full_343 = None
	        sub_7827: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_171, minimum_171);  maximum_171 = None
	        div_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7827, 255.0);  sub_7827 = None
	        clamp_min_513: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_342, 1.1920928955078125e-07);  div_342 = None
	        div_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_171, clamp_min_513);  minimum_171 = None
	        round_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_343);  div_343 = None
	        sub_7833: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_343);  round_343 = None
	        clamp_min_514: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7833, -128);  sub_7833 = None
	        clamp_max_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_514, 127);  clamp_min_514 = None
	        _assert_tensor_metadata_1541 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_513, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1541 = None
	        _assert_tensor_metadata_1542 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_342, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1542 = None
	        convert_element_type_1026: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_342, torch.int8);  clamp_max_342 = None
	        view_2682: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2680, [sym_size_int, 1500, 1280]);  view_2680 = None
	        view_2683: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_513, [sym_size_int, 1500, 1])
	        view_2684: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1026, [sym_size_int, 1500, 1])
	        reciprocal_171: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2683);  view_2683 = None
	        mul_16609: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_171, 1.0);  reciprocal_171 = None
	        mul_16612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2682, mul_16609);  view_2682 = mul_16609 = None
	        round_344: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16612);  mul_16612 = None
	        add_26287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_344, view_2684);  round_344 = view_2684 = None
	        clamp_min_515: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26287, -128);  add_26287 = None
	        clamp_max_343: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_515, 127);  clamp_min_515 = None
	        view_2685: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_343, [sym_size_int, 1500, 1280]);  clamp_max_343 = None
	        _assert_tensor_metadata_1543 = torch.ops.aten._assert_tensor_metadata.default(view_2685, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1543 = None
	        convert_element_type_1027: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2685, torch.int8);  view_2685 = None
	        view_2686: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1027, [sym_size_int, 1500, 1280]);  convert_element_type_1027 = None
	        view_2687: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_513, [sym_size_int, 1500, 1]);  clamp_min_513 = None
	        view_2688: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1026, [sym_size_int, 1500, 1]);  convert_element_type_1026 = None
	        _assert_tensor_metadata_1544 = torch.ops.aten._assert_tensor_metadata.default(view_2686, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1544 = None
	        convert_element_type_1028: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2686, torch.float32);  view_2686 = None
	        _assert_tensor_metadata_1545 = torch.ops.aten._assert_tensor_metadata.default(view_2688, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1545 = None
	        convert_element_type_1029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2688, torch.float32);  view_2688 = None
	        sub_7853: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1028, convert_element_type_1029);  convert_element_type_1028 = convert_element_type_1029 = None
	        mul_16634: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7853, view_2687);  sub_7853 = view_2687 = None
	        view_2689: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16634, [sym_size_int, 1500, 1280]);  mul_16634 = None
	        _assert_tensor_metadata_1546 = torch.ops.aten._assert_tensor_metadata.default(view_2689, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1546 = None
	        view_2690: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg775_1, [1280, 40, 32]);  arg775_1 = None
	        view_2691: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg776_1, [1280, 40, 1]);  arg776_1 = None
	        view_2692: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg777_1, [1280, 40, 1]);  arg777_1 = None
	        _assert_tensor_metadata_1547 = torch.ops.aten._assert_tensor_metadata.default(view_2690, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1547 = None
	        convert_element_type_1030: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2690, torch.float32);  view_2690 = None
	        _assert_tensor_metadata_1548 = torch.ops.aten._assert_tensor_metadata.default(view_2692, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1548 = None
	        convert_element_type_1031: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2692, torch.float32);  view_2692 = None
	        sub_7857: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1030, convert_element_type_1031);  convert_element_type_1030 = convert_element_type_1031 = None
	        mul_16639: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7857, view_2691);  sub_7857 = view_2691 = None
	        view_2693: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16639, [1280, 1280]);  mul_16639 = None
	        _assert_tensor_metadata_1549 = torch.ops.aten._assert_tensor_metadata.default(view_2693, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1549 = None
	        mul_16644: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2694: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2689, [mul_16644, 1280]);  view_2689 = mul_16644 = None
	        permute_288: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2693, [1, 0]);  view_2693 = None
	        addmm_142: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg774_1, view_2694, permute_288);  arg774_1 = view_2694 = permute_288 = None
	        view_2695: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_142, [sym_size_int, 1500, 1280]);  addmm_142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_229: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2695);  view_2695 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_26350: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_25730, clone_229);  add_25730 = clone_229 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_26350, memory_format = torch.contiguous_format)
	        var_mean_57 = torch.ops.aten.var_mean.correction(clone_230, [2], correction = 0, keepdim = True)
	        getitem_230: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_57[0]
	        getitem_231: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_57[1];  var_mean_57 = None
	        add_26355: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_230, 1e-05);  getitem_230 = None
	        rsqrt_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_26355);  add_26355 = None
	        sub_7863: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_230, getitem_231);  clone_230 = getitem_231 = None
	        mul_16655: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7863, rsqrt_57);  sub_7863 = rsqrt_57 = None
	        mul_16656: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16655, arg778_1);  mul_16655 = arg778_1 = None
	        add_26356: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16656, arg779_1);  mul_16656 = arg779_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2696: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_26356, [sym_size_int, 1500, 1280])
	        amin_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2696, [2])
	        amax_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2696, [2]);  view_2696 = None
	        full_344: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_172, full_344);  amin_172 = full_344 = None
	        full_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_172, full_345);  amax_172 = full_345 = None
	        sub_7874: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_172, minimum_172);  maximum_172 = None
	        div_344: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7874, 255.0);  sub_7874 = None
	        clamp_min_516: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_344, 1.1920928955078125e-07);  div_344 = None
	        div_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_172, clamp_min_516);  minimum_172 = None
	        round_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_345);  div_345 = None
	        sub_7880: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_345);  round_345 = None
	        clamp_min_517: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7880, -128);  sub_7880 = None
	        clamp_max_344: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_517, 127);  clamp_min_517 = None
	        _assert_tensor_metadata_1550 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_516, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1550 = None
	        _assert_tensor_metadata_1551 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_344, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1551 = None
	        convert_element_type_1032: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_344, torch.int8);  clamp_max_344 = None
	        view_2697: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_26356, [sym_size_int, 1500, 1280]);  add_26356 = None
	        view_2698: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_516, [sym_size_int, 1500, 1])
	        view_2699: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1032, [sym_size_int, 1500, 1])
	        reciprocal_172: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2698);  view_2698 = None
	        mul_16704: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_172, 1.0);  reciprocal_172 = None
	        mul_16707: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2697, mul_16704);  view_2697 = mul_16704 = None
	        round_346: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16707);  mul_16707 = None
	        add_26443: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_346, view_2699);  round_346 = view_2699 = None
	        clamp_min_518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26443, -128);  add_26443 = None
	        clamp_max_345: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_518, 127);  clamp_min_518 = None
	        view_2700: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_345, [sym_size_int, 1500, 1280]);  clamp_max_345 = None
	        _assert_tensor_metadata_1552 = torch.ops.aten._assert_tensor_metadata.default(view_2700, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1552 = None
	        convert_element_type_1033: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2700, torch.int8);  view_2700 = None
	        view_2701: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1033, [sym_size_int, 1500, 1280]);  convert_element_type_1033 = None
	        view_2702: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_516, [sym_size_int, 1500, 1]);  clamp_min_516 = None
	        view_2703: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1032, [sym_size_int, 1500, 1]);  convert_element_type_1032 = None
	        _assert_tensor_metadata_1553 = torch.ops.aten._assert_tensor_metadata.default(view_2701, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1553 = None
	        convert_element_type_1034: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2701, torch.float32);  view_2701 = None
	        _assert_tensor_metadata_1554 = torch.ops.aten._assert_tensor_metadata.default(view_2703, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1554 = None
	        convert_element_type_1035: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2703, torch.float32);  view_2703 = None
	        sub_7900: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1034, convert_element_type_1035);  convert_element_type_1034 = convert_element_type_1035 = None
	        mul_16729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7900, view_2702);  sub_7900 = view_2702 = None
	        view_2704: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16729, [sym_size_int, 1500, 1280]);  mul_16729 = None
	        _assert_tensor_metadata_1555 = torch.ops.aten._assert_tensor_metadata.default(view_2704, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1555 = None
	        view_2705: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg781_1, [5120, 40, 32]);  arg781_1 = None
	        view_2706: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg782_1, [5120, 40, 1]);  arg782_1 = None
	        view_2707: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg783_1, [5120, 40, 1]);  arg783_1 = None
	        _assert_tensor_metadata_1556 = torch.ops.aten._assert_tensor_metadata.default(view_2705, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1556 = None
	        convert_element_type_1036: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2705, torch.float32);  view_2705 = None
	        _assert_tensor_metadata_1557 = torch.ops.aten._assert_tensor_metadata.default(view_2707, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1557 = None
	        convert_element_type_1037: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2707, torch.float32);  view_2707 = None
	        sub_7904: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1036, convert_element_type_1037);  convert_element_type_1036 = convert_element_type_1037 = None
	        mul_16734: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7904, view_2706);  sub_7904 = view_2706 = None
	        view_2708: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16734, [5120, 1280]);  mul_16734 = None
	        _assert_tensor_metadata_1558 = torch.ops.aten._assert_tensor_metadata.default(view_2708, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1558 = None
	        mul_16739: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2709: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2704, [mul_16739, 1280]);  view_2704 = mul_16739 = None
	        permute_289: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2708, [1, 0]);  view_2708 = None
	        addmm_143: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg780_1, view_2709, permute_289);  arg780_1 = view_2709 = permute_289 = None
	        view_2710: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_143, [sym_size_int, 1500, 5120]);  addmm_143 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_16746: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2710, 0.5)
	        mul_16747: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2710, 0.7071067811865476);  view_2710 = None
	        erf_30: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_16747);  mul_16747 = None
	        add_26502: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_30, 1);  erf_30 = None
	        mul_16748: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16746, add_26502);  mul_16746 = add_26502 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_231: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_16748);  mul_16748 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2711: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_231, [sym_size_int, 1500, 5120])
	        amin_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2711, [2])
	        amax_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2711, [2]);  view_2711 = None
	        full_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_173, full_346);  amin_173 = full_346 = None
	        full_347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_173, full_347);  amax_173 = full_347 = None
	        sub_7917: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_173, minimum_173);  maximum_173 = None
	        div_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7917, 255.0);  sub_7917 = None
	        clamp_min_519: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_346, 1.1920928955078125e-07);  div_346 = None
	        div_347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_173, clamp_min_519);  minimum_173 = None
	        round_347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_347);  div_347 = None
	        sub_7923: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_347);  round_347 = None
	        clamp_min_520: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7923, -128);  sub_7923 = None
	        clamp_max_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_520, 127);  clamp_min_520 = None
	        _assert_tensor_metadata_1559 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_519, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1559 = None
	        _assert_tensor_metadata_1560 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_346, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1560 = None
	        convert_element_type_1038: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_346, torch.int8);  clamp_max_346 = None
	        view_2712: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_231, [sym_size_int, 1500, 5120]);  clone_231 = None
	        view_2713: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_519, [sym_size_int, 1500, 1])
	        view_2714: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1038, [sym_size_int, 1500, 1])
	        reciprocal_173: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2713);  view_2713 = None
	        mul_16794: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_173, 1.0);  reciprocal_173 = None
	        mul_16797: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2712, mul_16794);  view_2712 = mul_16794 = None
	        round_348: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_16797);  mul_16797 = None
	        add_26585: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_348, view_2714);  round_348 = view_2714 = None
	        clamp_min_521: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26585, -128);  add_26585 = None
	        clamp_max_347: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_521, 127);  clamp_min_521 = None
	        view_2715: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_347, [sym_size_int, 1500, 5120]);  clamp_max_347 = None
	        _assert_tensor_metadata_1561 = torch.ops.aten._assert_tensor_metadata.default(view_2715, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1561 = None
	        convert_element_type_1039: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2715, torch.int8);  view_2715 = None
	        view_2716: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1039, [sym_size_int, 1500, 5120]);  convert_element_type_1039 = None
	        view_2717: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_519, [sym_size_int, 1500, 1]);  clamp_min_519 = None
	        view_2718: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1038, [sym_size_int, 1500, 1]);  convert_element_type_1038 = None
	        _assert_tensor_metadata_1562 = torch.ops.aten._assert_tensor_metadata.default(view_2716, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1562 = None
	        convert_element_type_1040: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2716, torch.float32);  view_2716 = None
	        _assert_tensor_metadata_1563 = torch.ops.aten._assert_tensor_metadata.default(view_2718, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1563 = None
	        convert_element_type_1041: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2718, torch.float32);  view_2718 = None
	        sub_7943: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1040, convert_element_type_1041);  convert_element_type_1040 = convert_element_type_1041 = None
	        mul_16819: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7943, view_2717);  sub_7943 = view_2717 = None
	        view_2719: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_16819, [sym_size_int, 1500, 5120]);  mul_16819 = None
	        _assert_tensor_metadata_1564 = torch.ops.aten._assert_tensor_metadata.default(view_2719, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1564 = None
	        view_2720: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg785_1, [1280, 160, 32]);  arg785_1 = None
	        view_2721: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg786_1, [1280, 160, 1]);  arg786_1 = None
	        view_2722: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg787_1, [1280, 160, 1]);  arg787_1 = None
	        _assert_tensor_metadata_1565 = torch.ops.aten._assert_tensor_metadata.default(view_2720, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1565 = None
	        convert_element_type_1042: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2720, torch.float32);  view_2720 = None
	        _assert_tensor_metadata_1566 = torch.ops.aten._assert_tensor_metadata.default(view_2722, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1566 = None
	        convert_element_type_1043: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2722, torch.float32);  view_2722 = None
	        sub_7947: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1042, convert_element_type_1043);  convert_element_type_1042 = convert_element_type_1043 = None
	        mul_16824: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7947, view_2721);  sub_7947 = view_2721 = None
	        view_2723: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_16824, [1280, 5120]);  mul_16824 = None
	        _assert_tensor_metadata_1567 = torch.ops.aten._assert_tensor_metadata.default(view_2723, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1567 = None
	        mul_16829: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2724: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2719, [mul_16829, 5120]);  view_2719 = mul_16829 = None
	        permute_290: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2723, [1, 0]);  view_2723 = None
	        addmm_144: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg784_1, view_2724, permute_290);  arg784_1 = view_2724 = permute_290 = None
	        view_2725: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_144, [sym_size_int, 1500, 1280]);  addmm_144 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_232: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2725);  view_2725 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_26648: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_26350, clone_232);  add_26350 = clone_232 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_233: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_26648, memory_format = torch.contiguous_format)
	        var_mean_58 = torch.ops.aten.var_mean.correction(clone_233, [2], correction = 0, keepdim = True)
	        getitem_232: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_58[0]
	        getitem_233: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_58[1];  var_mean_58 = None
	        add_26653: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_232, 1e-05);  getitem_232 = None
	        rsqrt_58: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_26653);  add_26653 = None
	        sub_7953: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_233, getitem_233);  clone_233 = getitem_233 = None
	        mul_16840: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7953, rsqrt_58);  sub_7953 = rsqrt_58 = None
	        mul_16841: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16840, arg788_1);  mul_16840 = arg788_1 = None
	        add_26654: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16841, arg789_1);  mul_16841 = arg789_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2726: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_26654, [sym_size_int, 1500, 1280])
	        amin_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2726, [2])
	        amax_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2726, [2]);  view_2726 = None
	        full_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_174, full_348);  amin_174 = full_348 = None
	        full_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_174, full_349);  amax_174 = full_349 = None
	        sub_7964: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_174, minimum_174);  maximum_174 = None
	        div_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7964, 255.0);  sub_7964 = None
	        clamp_min_522: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_348, 1.1920928955078125e-07);  div_348 = None
	        div_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_174, clamp_min_522);  minimum_174 = None
	        round_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_349);  div_349 = None
	        sub_7970: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_349);  round_349 = None
	        clamp_min_523: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7970, -128);  sub_7970 = None
	        clamp_max_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_523, 127);  clamp_min_523 = None
	        _assert_tensor_metadata_1568 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_522, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1568 = None
	        _assert_tensor_metadata_1569 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_348, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1569 = None
	        convert_element_type_1044: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_348, torch.int8);  clamp_max_348 = None
	        view_2727: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_26654, [sym_size_int, 1500, 1280])
	        view_2728: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_522, [sym_size_int, 1500, 1])
	        view_2729: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1044, [sym_size_int, 1500, 1])
	        reciprocal_174: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2728);  view_2728 = None
	        mul_16889: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_174, 1.0);  reciprocal_174 = None
	        mul_16892: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2727, mul_16889);  view_2727 = mul_16889 = None
	        round_350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16892);  mul_16892 = None
	        add_26741: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_350, view_2729);  round_350 = view_2729 = None
	        clamp_min_524: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26741, -128);  add_26741 = None
	        clamp_max_349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_524, 127);  clamp_min_524 = None
	        view_2730: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_349, [sym_size_int, 1500, 1280]);  clamp_max_349 = None
	        _assert_tensor_metadata_1570 = torch.ops.aten._assert_tensor_metadata.default(view_2730, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1570 = None
	        convert_element_type_1045: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2730, torch.int8);  view_2730 = None
	        view_2731: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1045, [sym_size_int, 1500, 1280]);  convert_element_type_1045 = None
	        view_2732: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_522, [sym_size_int, 1500, 1]);  clamp_min_522 = None
	        view_2733: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1044, [sym_size_int, 1500, 1]);  convert_element_type_1044 = None
	        _assert_tensor_metadata_1571 = torch.ops.aten._assert_tensor_metadata.default(view_2731, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1571 = None
	        convert_element_type_1046: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2731, torch.float32);  view_2731 = None
	        _assert_tensor_metadata_1572 = torch.ops.aten._assert_tensor_metadata.default(view_2733, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1572 = None
	        convert_element_type_1047: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2733, torch.float32);  view_2733 = None
	        sub_7990: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1046, convert_element_type_1047);  convert_element_type_1046 = convert_element_type_1047 = None
	        mul_16914: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7990, view_2732);  sub_7990 = view_2732 = None
	        view_2734: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16914, [sym_size_int, 1500, 1280]);  mul_16914 = None
	        _assert_tensor_metadata_1573 = torch.ops.aten._assert_tensor_metadata.default(view_2734, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1573 = None
	        view_2735: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg791_1, [1280, 40, 32]);  arg791_1 = None
	        view_2736: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg792_1, [1280, 40, 1]);  arg792_1 = None
	        view_2737: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg793_1, [1280, 40, 1]);  arg793_1 = None
	        _assert_tensor_metadata_1574 = torch.ops.aten._assert_tensor_metadata.default(view_2735, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1574 = None
	        convert_element_type_1048: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2735, torch.float32);  view_2735 = None
	        _assert_tensor_metadata_1575 = torch.ops.aten._assert_tensor_metadata.default(view_2737, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1575 = None
	        convert_element_type_1049: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2737, torch.float32);  view_2737 = None
	        sub_7994: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1048, convert_element_type_1049);  convert_element_type_1048 = convert_element_type_1049 = None
	        mul_16919: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7994, view_2736);  sub_7994 = view_2736 = None
	        view_2738: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16919, [1280, 1280]);  mul_16919 = None
	        _assert_tensor_metadata_1576 = torch.ops.aten._assert_tensor_metadata.default(view_2738, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1576 = None
	        mul_16924: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2739: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2734, [mul_16924, 1280]);  view_2734 = mul_16924 = None
	        permute_291: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2738, [1, 0]);  view_2738 = None
	        addmm_145: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg790_1, view_2739, permute_291);  arg790_1 = view_2739 = permute_291 = None
	        view_2740: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_145, [sym_size_int, 1500, 1280]);  addmm_145 = None
	        mul_16931: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2740, 0.125);  view_2740 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2741: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_16931, [sym_size_int, 1500, 20, 64]);  mul_16931 = None
	        permute_292: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2741, [0, 2, 1, 3]);  view_2741 = None
	        clone_234: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_292, memory_format = torch.contiguous_format);  permute_292 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2742: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_26654, [sym_size_int, 1500, 1280])
	        amin_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2742, [2])
	        amax_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2742, [2]);  view_2742 = None
	        full_350: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_175, full_350);  amin_175 = full_350 = None
	        full_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_175, full_351);  amax_175 = full_351 = None
	        sub_8009: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_175, minimum_175);  maximum_175 = None
	        div_350: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8009, 255.0);  sub_8009 = None
	        clamp_min_525: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_350, 1.1920928955078125e-07);  div_350 = None
	        div_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_175, clamp_min_525);  minimum_175 = None
	        round_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_351);  div_351 = None
	        sub_8015: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_351);  round_351 = None
	        clamp_min_526: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8015, -128);  sub_8015 = None
	        clamp_max_350: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_526, 127);  clamp_min_526 = None
	        _assert_tensor_metadata_1577 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_525, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1577 = None
	        _assert_tensor_metadata_1578 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_350, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1578 = None
	        convert_element_type_1050: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_350, torch.int8);  clamp_max_350 = None
	        view_2743: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_26654, [sym_size_int, 1500, 1280])
	        view_2744: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_525, [sym_size_int, 1500, 1])
	        view_2745: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1050, [sym_size_int, 1500, 1])
	        reciprocal_175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2744);  view_2744 = None
	        mul_16985: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_175, 1.0);  reciprocal_175 = None
	        mul_16988: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2743, mul_16985);  view_2743 = mul_16985 = None
	        round_352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16988);  mul_16988 = None
	        add_26893: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_352, view_2745);  round_352 = view_2745 = None
	        clamp_min_527: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26893, -128);  add_26893 = None
	        clamp_max_351: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_527, 127);  clamp_min_527 = None
	        view_2746: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_351, [sym_size_int, 1500, 1280]);  clamp_max_351 = None
	        _assert_tensor_metadata_1579 = torch.ops.aten._assert_tensor_metadata.default(view_2746, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1579 = None
	        convert_element_type_1051: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2746, torch.int8);  view_2746 = None
	        view_2747: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1051, [sym_size_int, 1500, 1280]);  convert_element_type_1051 = None
	        view_2748: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_525, [sym_size_int, 1500, 1]);  clamp_min_525 = None
	        view_2749: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1050, [sym_size_int, 1500, 1]);  convert_element_type_1050 = None
	        _assert_tensor_metadata_1580 = torch.ops.aten._assert_tensor_metadata.default(view_2747, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1580 = None
	        convert_element_type_1052: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2747, torch.float32);  view_2747 = None
	        _assert_tensor_metadata_1581 = torch.ops.aten._assert_tensor_metadata.default(view_2749, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1581 = None
	        convert_element_type_1053: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2749, torch.float32);  view_2749 = None
	        sub_8035: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1052, convert_element_type_1053);  convert_element_type_1052 = convert_element_type_1053 = None
	        mul_17010: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8035, view_2748);  sub_8035 = view_2748 = None
	        view_2750: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17010, [sym_size_int, 1500, 1280]);  mul_17010 = None
	        _assert_tensor_metadata_1582 = torch.ops.aten._assert_tensor_metadata.default(view_2750, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1582 = None
	        view_2751: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg794_1, [1280, 40, 32]);  arg794_1 = None
	        view_2752: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg795_1, [1280, 40, 1]);  arg795_1 = None
	        view_2753: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg796_1, [1280, 40, 1]);  arg796_1 = None
	        _assert_tensor_metadata_1583 = torch.ops.aten._assert_tensor_metadata.default(view_2751, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1583 = None
	        convert_element_type_1054: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2751, torch.float32);  view_2751 = None
	        _assert_tensor_metadata_1584 = torch.ops.aten._assert_tensor_metadata.default(view_2753, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1584 = None
	        convert_element_type_1055: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2753, torch.float32);  view_2753 = None
	        sub_8039: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1054, convert_element_type_1055);  convert_element_type_1054 = convert_element_type_1055 = None
	        mul_17015: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8039, view_2752);  sub_8039 = view_2752 = None
	        view_2754: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17015, [1280, 1280]);  mul_17015 = None
	        _assert_tensor_metadata_1585 = torch.ops.aten._assert_tensor_metadata.default(view_2754, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1585 = None
	        permute_293: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2754, [1, 0]);  view_2754 = None
	        mul_17018: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2755: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2750, [mul_17018, 1280]);  view_2750 = mul_17018 = None
	        mm_29: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2755, permute_293);  view_2755 = permute_293 = None
	        view_2756: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_29, [sym_size_int, 1500, 1280]);  mm_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2757: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2756, [sym_size_int, -1, 20, 64]);  view_2756 = None
	        permute_294: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2757, [0, 2, 1, 3]);  view_2757 = None
	        clone_235: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_294, memory_format = torch.contiguous_format);  permute_294 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2758: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_26654, [sym_size_int, 1500, 1280])
	        amin_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2758, [2])
	        amax_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2758, [2]);  view_2758 = None
	        full_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_176, full_352);  amin_176 = full_352 = None
	        full_353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_176, full_353);  amax_176 = full_353 = None
	        sub_8053: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_176, minimum_176);  maximum_176 = None
	        div_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8053, 255.0);  sub_8053 = None
	        clamp_min_528: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_352, 1.1920928955078125e-07);  div_352 = None
	        div_353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_176, clamp_min_528);  minimum_176 = None
	        round_353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_353);  div_353 = None
	        sub_8059: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_353);  round_353 = None
	        clamp_min_529: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8059, -128);  sub_8059 = None
	        clamp_max_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_529, 127);  clamp_min_529 = None
	        _assert_tensor_metadata_1586 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_528, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1586 = None
	        _assert_tensor_metadata_1587 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_352, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1587 = None
	        convert_element_type_1056: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_352, torch.int8);  clamp_max_352 = None
	        view_2759: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_26654, [sym_size_int, 1500, 1280]);  add_26654 = None
	        view_2760: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_528, [sym_size_int, 1500, 1])
	        view_2761: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1056, [sym_size_int, 1500, 1])
	        reciprocal_176: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2760);  view_2760 = None
	        mul_17084: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_176, 1.0);  reciprocal_176 = None
	        mul_17087: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2759, mul_17084);  view_2759 = mul_17084 = None
	        round_354: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17087);  mul_17087 = None
	        add_27041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_354, view_2761);  round_354 = view_2761 = None
	        clamp_min_530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27041, -128);  add_27041 = None
	        clamp_max_353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_530, 127);  clamp_min_530 = None
	        view_2762: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_353, [sym_size_int, 1500, 1280]);  clamp_max_353 = None
	        _assert_tensor_metadata_1588 = torch.ops.aten._assert_tensor_metadata.default(view_2762, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1588 = None
	        convert_element_type_1057: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2762, torch.int8);  view_2762 = None
	        view_2763: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1057, [sym_size_int, 1500, 1280]);  convert_element_type_1057 = None
	        view_2764: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_528, [sym_size_int, 1500, 1]);  clamp_min_528 = None
	        view_2765: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1056, [sym_size_int, 1500, 1]);  convert_element_type_1056 = None
	        _assert_tensor_metadata_1589 = torch.ops.aten._assert_tensor_metadata.default(view_2763, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1589 = None
	        convert_element_type_1058: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2763, torch.float32);  view_2763 = None
	        _assert_tensor_metadata_1590 = torch.ops.aten._assert_tensor_metadata.default(view_2765, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1590 = None
	        convert_element_type_1059: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2765, torch.float32);  view_2765 = None
	        sub_8079: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1058, convert_element_type_1059);  convert_element_type_1058 = convert_element_type_1059 = None
	        mul_17109: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8079, view_2764);  sub_8079 = view_2764 = None
	        view_2766: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17109, [sym_size_int, 1500, 1280]);  mul_17109 = None
	        _assert_tensor_metadata_1591 = torch.ops.aten._assert_tensor_metadata.default(view_2766, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1591 = None
	        view_2767: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg798_1, [1280, 40, 32]);  arg798_1 = None
	        view_2768: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg799_1, [1280, 40, 1]);  arg799_1 = None
	        view_2769: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg800_1, [1280, 40, 1]);  arg800_1 = None
	        _assert_tensor_metadata_1592 = torch.ops.aten._assert_tensor_metadata.default(view_2767, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1592 = None
	        convert_element_type_1060: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2767, torch.float32);  view_2767 = None
	        _assert_tensor_metadata_1593 = torch.ops.aten._assert_tensor_metadata.default(view_2769, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1593 = None
	        convert_element_type_1061: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2769, torch.float32);  view_2769 = None
	        sub_8083: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1060, convert_element_type_1061);  convert_element_type_1060 = convert_element_type_1061 = None
	        mul_17114: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8083, view_2768);  sub_8083 = view_2768 = None
	        view_2770: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17114, [1280, 1280]);  mul_17114 = None
	        _assert_tensor_metadata_1594 = torch.ops.aten._assert_tensor_metadata.default(view_2770, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1594 = None
	        mul_17119: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2771: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2766, [mul_17119, 1280]);  view_2766 = mul_17119 = None
	        permute_295: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2770, [1, 0]);  view_2770 = None
	        addmm_146: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg797_1, view_2771, permute_295);  arg797_1 = view_2771 = permute_295 = None
	        view_2772: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_146, [sym_size_int, 1500, 1280]);  addmm_146 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2773: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2772, [sym_size_int, -1, 20, 64]);  view_2772 = None
	        permute_296: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2773, [0, 2, 1, 3]);  view_2773 = None
	        clone_236: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_296, memory_format = torch.contiguous_format);  permute_296 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_29 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_234, clone_235, clone_236, None, False, scale = 1.0);  clone_234 = clone_235 = clone_236 = None
	        getitem_234: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_29[0];  _scaled_dot_product_efficient_attention_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_297: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_234, [0, 2, 1, 3]);  getitem_234 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2774: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_297, [sym_size_int, 1500, -1]);  permute_297 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2775: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2774, [sym_size_int, 1500, 1280])
	        amin_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2775, [2])
	        amax_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2775, [2]);  view_2775 = None
	        full_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_177, full_354);  amin_177 = full_354 = None
	        full_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_177, full_355);  amax_177 = full_355 = None
	        sub_8101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_177, minimum_177);  maximum_177 = None
	        div_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8101, 255.0);  sub_8101 = None
	        clamp_min_531: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_354, 1.1920928955078125e-07);  div_354 = None
	        div_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_177, clamp_min_531);  minimum_177 = None
	        round_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_355);  div_355 = None
	        sub_8107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_355);  round_355 = None
	        clamp_min_532: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8107, -128);  sub_8107 = None
	        clamp_max_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_532, 127);  clamp_min_532 = None
	        _assert_tensor_metadata_1595 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_531, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1595 = None
	        _assert_tensor_metadata_1596 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_354, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1596 = None
	        convert_element_type_1062: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_354, torch.int8);  clamp_max_354 = None
	        view_2776: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2774, [sym_size_int, 1500, 1280]);  view_2774 = None
	        view_2777: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_531, [sym_size_int, 1500, 1])
	        view_2778: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1062, [sym_size_int, 1500, 1])
	        reciprocal_177: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2777);  view_2777 = None
	        mul_17189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_177, 1.0);  reciprocal_177 = None
	        mul_17192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2776, mul_17189);  view_2776 = mul_17189 = None
	        round_356: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17192);  mul_17192 = None
	        add_27205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_356, view_2778);  round_356 = view_2778 = None
	        clamp_min_533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27205, -128);  add_27205 = None
	        clamp_max_355: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_533, 127);  clamp_min_533 = None
	        view_2779: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_355, [sym_size_int, 1500, 1280]);  clamp_max_355 = None
	        _assert_tensor_metadata_1597 = torch.ops.aten._assert_tensor_metadata.default(view_2779, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1597 = None
	        convert_element_type_1063: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2779, torch.int8);  view_2779 = None
	        view_2780: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1063, [sym_size_int, 1500, 1280]);  convert_element_type_1063 = None
	        view_2781: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_531, [sym_size_int, 1500, 1]);  clamp_min_531 = None
	        view_2782: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1062, [sym_size_int, 1500, 1]);  convert_element_type_1062 = None
	        _assert_tensor_metadata_1598 = torch.ops.aten._assert_tensor_metadata.default(view_2780, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1598 = None
	        convert_element_type_1064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2780, torch.float32);  view_2780 = None
	        _assert_tensor_metadata_1599 = torch.ops.aten._assert_tensor_metadata.default(view_2782, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1599 = None
	        convert_element_type_1065: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2782, torch.float32);  view_2782 = None
	        sub_8127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1064, convert_element_type_1065);  convert_element_type_1064 = convert_element_type_1065 = None
	        mul_17214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8127, view_2781);  sub_8127 = view_2781 = None
	        view_2783: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17214, [sym_size_int, 1500, 1280]);  mul_17214 = None
	        _assert_tensor_metadata_1600 = torch.ops.aten._assert_tensor_metadata.default(view_2783, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1600 = None
	        view_2784: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg802_1, [1280, 40, 32]);  arg802_1 = None
	        view_2785: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg803_1, [1280, 40, 1]);  arg803_1 = None
	        view_2786: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg804_1, [1280, 40, 1]);  arg804_1 = None
	        _assert_tensor_metadata_1601 = torch.ops.aten._assert_tensor_metadata.default(view_2784, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1601 = None
	        convert_element_type_1066: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2784, torch.float32);  view_2784 = None
	        _assert_tensor_metadata_1602 = torch.ops.aten._assert_tensor_metadata.default(view_2786, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1602 = None
	        convert_element_type_1067: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2786, torch.float32);  view_2786 = None
	        sub_8131: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1066, convert_element_type_1067);  convert_element_type_1066 = convert_element_type_1067 = None
	        mul_17219: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8131, view_2785);  sub_8131 = view_2785 = None
	        view_2787: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17219, [1280, 1280]);  mul_17219 = None
	        _assert_tensor_metadata_1603 = torch.ops.aten._assert_tensor_metadata.default(view_2787, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1603 = None
	        mul_17224: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2788: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2783, [mul_17224, 1280]);  view_2783 = mul_17224 = None
	        permute_298: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2787, [1, 0]);  view_2787 = None
	        addmm_147: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg801_1, view_2788, permute_298);  arg801_1 = view_2788 = permute_298 = None
	        view_2789: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_147, [sym_size_int, 1500, 1280]);  addmm_147 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_237: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2789);  view_2789 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_27268: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_26648, clone_237);  add_26648 = clone_237 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_238: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_27268, memory_format = torch.contiguous_format)
	        var_mean_59 = torch.ops.aten.var_mean.correction(clone_238, [2], correction = 0, keepdim = True)
	        getitem_238: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_59[0]
	        getitem_239: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_59[1];  var_mean_59 = None
	        add_27273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_238, 1e-05);  getitem_238 = None
	        rsqrt_59: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_27273);  add_27273 = None
	        sub_8137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_238, getitem_239);  clone_238 = getitem_239 = None
	        mul_17235: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8137, rsqrt_59);  sub_8137 = rsqrt_59 = None
	        mul_17236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17235, arg805_1);  mul_17235 = arg805_1 = None
	        add_27274: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_17236, arg806_1);  mul_17236 = arg806_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2790: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_27274, [sym_size_int, 1500, 1280])
	        amin_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2790, [2])
	        amax_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2790, [2]);  view_2790 = None
	        full_356: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_178, full_356);  amin_178 = full_356 = None
	        full_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_178, full_357);  amax_178 = full_357 = None
	        sub_8148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_178, minimum_178);  maximum_178 = None
	        div_356: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8148, 255.0);  sub_8148 = None
	        clamp_min_534: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_356, 1.1920928955078125e-07);  div_356 = None
	        div_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_178, clamp_min_534);  minimum_178 = None
	        round_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_357);  div_357 = None
	        sub_8154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_357);  round_357 = None
	        clamp_min_535: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8154, -128);  sub_8154 = None
	        clamp_max_356: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_535, 127);  clamp_min_535 = None
	        _assert_tensor_metadata_1604 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_534, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1604 = None
	        _assert_tensor_metadata_1605 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_356, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1605 = None
	        convert_element_type_1068: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_356, torch.int8);  clamp_max_356 = None
	        view_2791: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_27274, [sym_size_int, 1500, 1280]);  add_27274 = None
	        view_2792: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_534, [sym_size_int, 1500, 1])
	        view_2793: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1068, [sym_size_int, 1500, 1])
	        reciprocal_178: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2792);  view_2792 = None
	        mul_17284: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_178, 1.0);  reciprocal_178 = None
	        mul_17287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2791, mul_17284);  view_2791 = mul_17284 = None
	        round_358: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17287);  mul_17287 = None
	        add_27361: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_358, view_2793);  round_358 = view_2793 = None
	        clamp_min_536: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27361, -128);  add_27361 = None
	        clamp_max_357: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_536, 127);  clamp_min_536 = None
	        view_2794: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_357, [sym_size_int, 1500, 1280]);  clamp_max_357 = None
	        _assert_tensor_metadata_1606 = torch.ops.aten._assert_tensor_metadata.default(view_2794, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1606 = None
	        convert_element_type_1069: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2794, torch.int8);  view_2794 = None
	        view_2795: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1069, [sym_size_int, 1500, 1280]);  convert_element_type_1069 = None
	        view_2796: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_534, [sym_size_int, 1500, 1]);  clamp_min_534 = None
	        view_2797: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1068, [sym_size_int, 1500, 1]);  convert_element_type_1068 = None
	        _assert_tensor_metadata_1607 = torch.ops.aten._assert_tensor_metadata.default(view_2795, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1607 = None
	        convert_element_type_1070: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2795, torch.float32);  view_2795 = None
	        _assert_tensor_metadata_1608 = torch.ops.aten._assert_tensor_metadata.default(view_2797, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1608 = None
	        convert_element_type_1071: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2797, torch.float32);  view_2797 = None
	        sub_8174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1070, convert_element_type_1071);  convert_element_type_1070 = convert_element_type_1071 = None
	        mul_17309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8174, view_2796);  sub_8174 = view_2796 = None
	        view_2798: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17309, [sym_size_int, 1500, 1280]);  mul_17309 = None
	        _assert_tensor_metadata_1609 = torch.ops.aten._assert_tensor_metadata.default(view_2798, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1609 = None
	        view_2799: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg808_1, [5120, 40, 32]);  arg808_1 = None
	        view_2800: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg809_1, [5120, 40, 1]);  arg809_1 = None
	        view_2801: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg810_1, [5120, 40, 1]);  arg810_1 = None
	        _assert_tensor_metadata_1610 = torch.ops.aten._assert_tensor_metadata.default(view_2799, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1610 = None
	        convert_element_type_1072: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2799, torch.float32);  view_2799 = None
	        _assert_tensor_metadata_1611 = torch.ops.aten._assert_tensor_metadata.default(view_2801, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1611 = None
	        convert_element_type_1073: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2801, torch.float32);  view_2801 = None
	        sub_8178: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1072, convert_element_type_1073);  convert_element_type_1072 = convert_element_type_1073 = None
	        mul_17314: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8178, view_2800);  sub_8178 = view_2800 = None
	        view_2802: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17314, [5120, 1280]);  mul_17314 = None
	        _assert_tensor_metadata_1612 = torch.ops.aten._assert_tensor_metadata.default(view_2802, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1612 = None
	        mul_17319: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2803: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2798, [mul_17319, 1280]);  view_2798 = mul_17319 = None
	        permute_299: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2802, [1, 0]);  view_2802 = None
	        addmm_148: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg807_1, view_2803, permute_299);  arg807_1 = view_2803 = permute_299 = None
	        view_2804: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_148, [sym_size_int, 1500, 5120]);  addmm_148 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_17326: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2804, 0.5)
	        mul_17327: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2804, 0.7071067811865476);  view_2804 = None
	        erf_31: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_17327);  mul_17327 = None
	        add_27420: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_31, 1);  erf_31 = None
	        mul_17328: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17326, add_27420);  mul_17326 = add_27420 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_239: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_17328);  mul_17328 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2805: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_239, [sym_size_int, 1500, 5120])
	        amin_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2805, [2])
	        amax_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2805, [2]);  view_2805 = None
	        full_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_179, full_358);  amin_179 = full_358 = None
	        full_359: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_179, full_359);  amax_179 = full_359 = None
	        sub_8191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_179, minimum_179);  maximum_179 = None
	        div_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8191, 255.0);  sub_8191 = None
	        clamp_min_537: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_358, 1.1920928955078125e-07);  div_358 = None
	        div_359: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_179, clamp_min_537);  minimum_179 = None
	        round_359: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_359);  div_359 = None
	        sub_8197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_359);  round_359 = None
	        clamp_min_538: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8197, -128);  sub_8197 = None
	        clamp_max_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_538, 127);  clamp_min_538 = None
	        _assert_tensor_metadata_1613 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_537, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1613 = None
	        _assert_tensor_metadata_1614 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_358, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1614 = None
	        convert_element_type_1074: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_358, torch.int8);  clamp_max_358 = None
	        view_2806: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_239, [sym_size_int, 1500, 5120]);  clone_239 = None
	        view_2807: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_537, [sym_size_int, 1500, 1])
	        view_2808: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1074, [sym_size_int, 1500, 1])
	        reciprocal_179: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2807);  view_2807 = None
	        mul_17374: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_179, 1.0);  reciprocal_179 = None
	        mul_17377: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2806, mul_17374);  view_2806 = mul_17374 = None
	        round_360: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_17377);  mul_17377 = None
	        add_27503: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_360, view_2808);  round_360 = view_2808 = None
	        clamp_min_539: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27503, -128);  add_27503 = None
	        clamp_max_359: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_539, 127);  clamp_min_539 = None
	        view_2809: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_359, [sym_size_int, 1500, 5120]);  clamp_max_359 = None
	        _assert_tensor_metadata_1615 = torch.ops.aten._assert_tensor_metadata.default(view_2809, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1615 = None
	        convert_element_type_1075: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2809, torch.int8);  view_2809 = None
	        view_2810: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1075, [sym_size_int, 1500, 5120]);  convert_element_type_1075 = None
	        view_2811: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_537, [sym_size_int, 1500, 1]);  clamp_min_537 = None
	        view_2812: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1074, [sym_size_int, 1500, 1]);  convert_element_type_1074 = None
	        _assert_tensor_metadata_1616 = torch.ops.aten._assert_tensor_metadata.default(view_2810, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1616 = None
	        convert_element_type_1076: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2810, torch.float32);  view_2810 = None
	        _assert_tensor_metadata_1617 = torch.ops.aten._assert_tensor_metadata.default(view_2812, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1617 = None
	        convert_element_type_1077: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2812, torch.float32);  view_2812 = None
	        sub_8217: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1076, convert_element_type_1077);  convert_element_type_1076 = convert_element_type_1077 = None
	        mul_17399: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8217, view_2811);  sub_8217 = view_2811 = None
	        view_2813: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_17399, [sym_size_int, 1500, 5120]);  mul_17399 = None
	        _assert_tensor_metadata_1618 = torch.ops.aten._assert_tensor_metadata.default(view_2813, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1618 = None
	        view_2814: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg812_1, [1280, 160, 32]);  arg812_1 = None
	        view_2815: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg813_1, [1280, 160, 1]);  arg813_1 = None
	        view_2816: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg814_1, [1280, 160, 1]);  arg814_1 = None
	        _assert_tensor_metadata_1619 = torch.ops.aten._assert_tensor_metadata.default(view_2814, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1619 = None
	        convert_element_type_1078: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2814, torch.float32);  view_2814 = None
	        _assert_tensor_metadata_1620 = torch.ops.aten._assert_tensor_metadata.default(view_2816, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1620 = None
	        convert_element_type_1079: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2816, torch.float32);  view_2816 = None
	        sub_8221: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1078, convert_element_type_1079);  convert_element_type_1078 = convert_element_type_1079 = None
	        mul_17404: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8221, view_2815);  sub_8221 = view_2815 = None
	        view_2817: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_17404, [1280, 5120]);  mul_17404 = None
	        _assert_tensor_metadata_1621 = torch.ops.aten._assert_tensor_metadata.default(view_2817, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1621 = None
	        mul_17409: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2818: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2813, [mul_17409, 5120]);  view_2813 = mul_17409 = None
	        permute_300: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2817, [1, 0]);  view_2817 = None
	        addmm_149: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg811_1, view_2818, permute_300);  arg811_1 = view_2818 = permute_300 = None
	        view_2819: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_149, [sym_size_int, 1500, 1280]);  addmm_149 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_240: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2819);  view_2819 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_27566: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_27268, clone_240);  add_27268 = clone_240 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_241: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_27566, memory_format = torch.contiguous_format)
	        var_mean_60 = torch.ops.aten.var_mean.correction(clone_241, [2], correction = 0, keepdim = True)
	        getitem_240: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_60[0]
	        getitem_241: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_60[1];  var_mean_60 = None
	        add_27571: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_240, 1e-05);  getitem_240 = None
	        rsqrt_60: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_27571);  add_27571 = None
	        sub_8227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_241, getitem_241);  clone_241 = getitem_241 = None
	        mul_17420: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8227, rsqrt_60);  sub_8227 = rsqrt_60 = None
	        mul_17421: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17420, arg815_1);  mul_17420 = arg815_1 = None
	        add_27572: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_17421, arg816_1);  mul_17421 = arg816_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2820: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_27572, [sym_size_int, 1500, 1280])
	        amin_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2820, [2])
	        amax_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2820, [2]);  view_2820 = None
	        full_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_180, full_360);  amin_180 = full_360 = None
	        full_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_180, full_361);  amax_180 = full_361 = None
	        sub_8238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_180, minimum_180);  maximum_180 = None
	        div_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8238, 255.0);  sub_8238 = None
	        clamp_min_540: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_360, 1.1920928955078125e-07);  div_360 = None
	        div_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_180, clamp_min_540);  minimum_180 = None
	        round_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_361);  div_361 = None
	        sub_8244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_361);  round_361 = None
	        clamp_min_541: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8244, -128);  sub_8244 = None
	        clamp_max_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_541, 127);  clamp_min_541 = None
	        _assert_tensor_metadata_1622 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_540, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1622 = None
	        _assert_tensor_metadata_1623 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_360, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1623 = None
	        convert_element_type_1080: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_360, torch.int8);  clamp_max_360 = None
	        view_2821: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_27572, [sym_size_int, 1500, 1280])
	        view_2822: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_540, [sym_size_int, 1500, 1])
	        view_2823: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1080, [sym_size_int, 1500, 1])
	        reciprocal_180: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2822);  view_2822 = None
	        mul_17469: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_180, 1.0);  reciprocal_180 = None
	        mul_17472: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2821, mul_17469);  view_2821 = mul_17469 = None
	        round_362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17472);  mul_17472 = None
	        add_27659: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_362, view_2823);  round_362 = view_2823 = None
	        clamp_min_542: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27659, -128);  add_27659 = None
	        clamp_max_361: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_542, 127);  clamp_min_542 = None
	        view_2824: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_361, [sym_size_int, 1500, 1280]);  clamp_max_361 = None
	        _assert_tensor_metadata_1624 = torch.ops.aten._assert_tensor_metadata.default(view_2824, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1624 = None
	        convert_element_type_1081: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2824, torch.int8);  view_2824 = None
	        view_2825: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1081, [sym_size_int, 1500, 1280]);  convert_element_type_1081 = None
	        view_2826: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_540, [sym_size_int, 1500, 1]);  clamp_min_540 = None
	        view_2827: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1080, [sym_size_int, 1500, 1]);  convert_element_type_1080 = None
	        _assert_tensor_metadata_1625 = torch.ops.aten._assert_tensor_metadata.default(view_2825, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1625 = None
	        convert_element_type_1082: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2825, torch.float32);  view_2825 = None
	        _assert_tensor_metadata_1626 = torch.ops.aten._assert_tensor_metadata.default(view_2827, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1626 = None
	        convert_element_type_1083: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2827, torch.float32);  view_2827 = None
	        sub_8264: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1082, convert_element_type_1083);  convert_element_type_1082 = convert_element_type_1083 = None
	        mul_17494: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8264, view_2826);  sub_8264 = view_2826 = None
	        view_2828: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17494, [sym_size_int, 1500, 1280]);  mul_17494 = None
	        _assert_tensor_metadata_1627 = torch.ops.aten._assert_tensor_metadata.default(view_2828, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1627 = None
	        view_2829: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg818_1, [1280, 40, 32]);  arg818_1 = None
	        view_2830: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg819_1, [1280, 40, 1]);  arg819_1 = None
	        view_2831: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg820_1, [1280, 40, 1]);  arg820_1 = None
	        _assert_tensor_metadata_1628 = torch.ops.aten._assert_tensor_metadata.default(view_2829, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1628 = None
	        convert_element_type_1084: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2829, torch.float32);  view_2829 = None
	        _assert_tensor_metadata_1629 = torch.ops.aten._assert_tensor_metadata.default(view_2831, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1629 = None
	        convert_element_type_1085: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2831, torch.float32);  view_2831 = None
	        sub_8268: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1084, convert_element_type_1085);  convert_element_type_1084 = convert_element_type_1085 = None
	        mul_17499: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8268, view_2830);  sub_8268 = view_2830 = None
	        view_2832: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17499, [1280, 1280]);  mul_17499 = None
	        _assert_tensor_metadata_1630 = torch.ops.aten._assert_tensor_metadata.default(view_2832, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1630 = None
	        mul_17504: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2833: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2828, [mul_17504, 1280]);  view_2828 = mul_17504 = None
	        permute_301: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2832, [1, 0]);  view_2832 = None
	        addmm_150: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg817_1, view_2833, permute_301);  arg817_1 = view_2833 = permute_301 = None
	        view_2834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_150, [sym_size_int, 1500, 1280]);  addmm_150 = None
	        mul_17511: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2834, 0.125);  view_2834 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2835: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_17511, [sym_size_int, 1500, 20, 64]);  mul_17511 = None
	        permute_302: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2835, [0, 2, 1, 3]);  view_2835 = None
	        clone_242: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_302, memory_format = torch.contiguous_format);  permute_302 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2836: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_27572, [sym_size_int, 1500, 1280])
	        amin_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2836, [2])
	        amax_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2836, [2]);  view_2836 = None
	        full_362: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_181, full_362);  amin_181 = full_362 = None
	        full_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_181, full_363);  amax_181 = full_363 = None
	        sub_8283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_181, minimum_181);  maximum_181 = None
	        div_362: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8283, 255.0);  sub_8283 = None
	        clamp_min_543: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_362, 1.1920928955078125e-07);  div_362 = None
	        div_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_181, clamp_min_543);  minimum_181 = None
	        round_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_363);  div_363 = None
	        sub_8289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_363);  round_363 = None
	        clamp_min_544: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8289, -128);  sub_8289 = None
	        clamp_max_362: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_544, 127);  clamp_min_544 = None
	        _assert_tensor_metadata_1631 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_543, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1631 = None
	        _assert_tensor_metadata_1632 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_362, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1632 = None
	        convert_element_type_1086: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_362, torch.int8);  clamp_max_362 = None
	        view_2837: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_27572, [sym_size_int, 1500, 1280])
	        view_2838: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_543, [sym_size_int, 1500, 1])
	        view_2839: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1086, [sym_size_int, 1500, 1])
	        reciprocal_181: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2838);  view_2838 = None
	        mul_17565: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_181, 1.0);  reciprocal_181 = None
	        mul_17568: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2837, mul_17565);  view_2837 = mul_17565 = None
	        round_364: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17568);  mul_17568 = None
	        add_27811: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_364, view_2839);  round_364 = view_2839 = None
	        clamp_min_545: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27811, -128);  add_27811 = None
	        clamp_max_363: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_545, 127);  clamp_min_545 = None
	        view_2840: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_363, [sym_size_int, 1500, 1280]);  clamp_max_363 = None
	        _assert_tensor_metadata_1633 = torch.ops.aten._assert_tensor_metadata.default(view_2840, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1633 = None
	        convert_element_type_1087: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2840, torch.int8);  view_2840 = None
	        view_2841: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1087, [sym_size_int, 1500, 1280]);  convert_element_type_1087 = None
	        view_2842: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_543, [sym_size_int, 1500, 1]);  clamp_min_543 = None
	        view_2843: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1086, [sym_size_int, 1500, 1]);  convert_element_type_1086 = None
	        _assert_tensor_metadata_1634 = torch.ops.aten._assert_tensor_metadata.default(view_2841, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1634 = None
	        convert_element_type_1088: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2841, torch.float32);  view_2841 = None
	        _assert_tensor_metadata_1635 = torch.ops.aten._assert_tensor_metadata.default(view_2843, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1635 = None
	        convert_element_type_1089: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2843, torch.float32);  view_2843 = None
	        sub_8309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1088, convert_element_type_1089);  convert_element_type_1088 = convert_element_type_1089 = None
	        mul_17590: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8309, view_2842);  sub_8309 = view_2842 = None
	        view_2844: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17590, [sym_size_int, 1500, 1280]);  mul_17590 = None
	        _assert_tensor_metadata_1636 = torch.ops.aten._assert_tensor_metadata.default(view_2844, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1636 = None
	        view_2845: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg821_1, [1280, 40, 32]);  arg821_1 = None
	        view_2846: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg822_1, [1280, 40, 1]);  arg822_1 = None
	        view_2847: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg823_1, [1280, 40, 1]);  arg823_1 = None
	        _assert_tensor_metadata_1637 = torch.ops.aten._assert_tensor_metadata.default(view_2845, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1637 = None
	        convert_element_type_1090: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2845, torch.float32);  view_2845 = None
	        _assert_tensor_metadata_1638 = torch.ops.aten._assert_tensor_metadata.default(view_2847, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1638 = None
	        convert_element_type_1091: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2847, torch.float32);  view_2847 = None
	        sub_8313: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1090, convert_element_type_1091);  convert_element_type_1090 = convert_element_type_1091 = None
	        mul_17595: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8313, view_2846);  sub_8313 = view_2846 = None
	        view_2848: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17595, [1280, 1280]);  mul_17595 = None
	        _assert_tensor_metadata_1639 = torch.ops.aten._assert_tensor_metadata.default(view_2848, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1639 = None
	        permute_303: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2848, [1, 0]);  view_2848 = None
	        mul_17598: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2849: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2844, [mul_17598, 1280]);  view_2844 = mul_17598 = None
	        mm_30: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2849, permute_303);  view_2849 = permute_303 = None
	        view_2850: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_30, [sym_size_int, 1500, 1280]);  mm_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2851: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2850, [sym_size_int, -1, 20, 64]);  view_2850 = None
	        permute_304: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2851, [0, 2, 1, 3]);  view_2851 = None
	        clone_243: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_304, memory_format = torch.contiguous_format);  permute_304 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2852: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_27572, [sym_size_int, 1500, 1280])
	        amin_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2852, [2])
	        amax_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2852, [2]);  view_2852 = None
	        full_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_182, full_364);  amin_182 = full_364 = None
	        full_365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_182, full_365);  amax_182 = full_365 = None
	        sub_8327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_182, minimum_182);  maximum_182 = None
	        div_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8327, 255.0);  sub_8327 = None
	        clamp_min_546: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_364, 1.1920928955078125e-07);  div_364 = None
	        div_365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_182, clamp_min_546);  minimum_182 = None
	        round_365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_365);  div_365 = None
	        sub_8333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_365);  round_365 = None
	        clamp_min_547: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8333, -128);  sub_8333 = None
	        clamp_max_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_547, 127);  clamp_min_547 = None
	        _assert_tensor_metadata_1640 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_546, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1640 = None
	        _assert_tensor_metadata_1641 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_364, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1641 = None
	        convert_element_type_1092: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_364, torch.int8);  clamp_max_364 = None
	        view_2853: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_27572, [sym_size_int, 1500, 1280]);  add_27572 = None
	        view_2854: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_546, [sym_size_int, 1500, 1])
	        view_2855: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1092, [sym_size_int, 1500, 1])
	        reciprocal_182: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2854);  view_2854 = None
	        mul_17664: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_182, 1.0);  reciprocal_182 = None
	        mul_17667: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2853, mul_17664);  view_2853 = mul_17664 = None
	        round_366: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17667);  mul_17667 = None
	        add_27959: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_366, view_2855);  round_366 = view_2855 = None
	        clamp_min_548: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27959, -128);  add_27959 = None
	        clamp_max_365: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_548, 127);  clamp_min_548 = None
	        view_2856: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_365, [sym_size_int, 1500, 1280]);  clamp_max_365 = None
	        _assert_tensor_metadata_1642 = torch.ops.aten._assert_tensor_metadata.default(view_2856, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1642 = None
	        convert_element_type_1093: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2856, torch.int8);  view_2856 = None
	        view_2857: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1093, [sym_size_int, 1500, 1280]);  convert_element_type_1093 = None
	        view_2858: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_546, [sym_size_int, 1500, 1]);  clamp_min_546 = None
	        view_2859: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1092, [sym_size_int, 1500, 1]);  convert_element_type_1092 = None
	        _assert_tensor_metadata_1643 = torch.ops.aten._assert_tensor_metadata.default(view_2857, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1643 = None
	        convert_element_type_1094: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2857, torch.float32);  view_2857 = None
	        _assert_tensor_metadata_1644 = torch.ops.aten._assert_tensor_metadata.default(view_2859, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1644 = None
	        convert_element_type_1095: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2859, torch.float32);  view_2859 = None
	        sub_8353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1094, convert_element_type_1095);  convert_element_type_1094 = convert_element_type_1095 = None
	        mul_17689: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8353, view_2858);  sub_8353 = view_2858 = None
	        view_2860: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17689, [sym_size_int, 1500, 1280]);  mul_17689 = None
	        _assert_tensor_metadata_1645 = torch.ops.aten._assert_tensor_metadata.default(view_2860, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1645 = None
	        view_2861: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg825_1, [1280, 40, 32]);  arg825_1 = None
	        view_2862: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg826_1, [1280, 40, 1]);  arg826_1 = None
	        view_2863: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg827_1, [1280, 40, 1]);  arg827_1 = None
	        _assert_tensor_metadata_1646 = torch.ops.aten._assert_tensor_metadata.default(view_2861, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1646 = None
	        convert_element_type_1096: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2861, torch.float32);  view_2861 = None
	        _assert_tensor_metadata_1647 = torch.ops.aten._assert_tensor_metadata.default(view_2863, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1647 = None
	        convert_element_type_1097: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2863, torch.float32);  view_2863 = None
	        sub_8357: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1096, convert_element_type_1097);  convert_element_type_1096 = convert_element_type_1097 = None
	        mul_17694: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8357, view_2862);  sub_8357 = view_2862 = None
	        view_2864: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17694, [1280, 1280]);  mul_17694 = None
	        _assert_tensor_metadata_1648 = torch.ops.aten._assert_tensor_metadata.default(view_2864, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1648 = None
	        mul_17699: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2865: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2860, [mul_17699, 1280]);  view_2860 = mul_17699 = None
	        permute_305: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2864, [1, 0]);  view_2864 = None
	        addmm_151: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg824_1, view_2865, permute_305);  arg824_1 = view_2865 = permute_305 = None
	        view_2866: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_151, [sym_size_int, 1500, 1280]);  addmm_151 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2867: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2866, [sym_size_int, -1, 20, 64]);  view_2866 = None
	        permute_306: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2867, [0, 2, 1, 3]);  view_2867 = None
	        clone_244: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_306, memory_format = torch.contiguous_format);  permute_306 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_30 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_242, clone_243, clone_244, None, False, scale = 1.0);  clone_242 = clone_243 = clone_244 = None
	        getitem_242: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_30[0];  _scaled_dot_product_efficient_attention_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_307: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_242, [0, 2, 1, 3]);  getitem_242 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2868: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_307, [sym_size_int, 1500, -1]);  permute_307 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2869: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2868, [sym_size_int, 1500, 1280])
	        amin_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2869, [2])
	        amax_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2869, [2]);  view_2869 = None
	        full_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_183, full_366);  amin_183 = full_366 = None
	        full_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_183, full_367);  amax_183 = full_367 = None
	        sub_8375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_183, minimum_183);  maximum_183 = None
	        div_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8375, 255.0);  sub_8375 = None
	        clamp_min_549: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_366, 1.1920928955078125e-07);  div_366 = None
	        div_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_183, clamp_min_549);  minimum_183 = None
	        round_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_367);  div_367 = None
	        sub_8381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_367);  round_367 = None
	        clamp_min_550: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8381, -128);  sub_8381 = None
	        clamp_max_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_550, 127);  clamp_min_550 = None
	        _assert_tensor_metadata_1649 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_549, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1649 = None
	        _assert_tensor_metadata_1650 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_366, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1650 = None
	        convert_element_type_1098: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_366, torch.int8);  clamp_max_366 = None
	        view_2870: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2868, [sym_size_int, 1500, 1280]);  view_2868 = None
	        view_2871: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_549, [sym_size_int, 1500, 1])
	        view_2872: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1098, [sym_size_int, 1500, 1])
	        reciprocal_183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2871);  view_2871 = None
	        mul_17769: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_183, 1.0);  reciprocal_183 = None
	        mul_17772: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2870, mul_17769);  view_2870 = mul_17769 = None
	        round_368: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17772);  mul_17772 = None
	        add_28123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_368, view_2872);  round_368 = view_2872 = None
	        clamp_min_551: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28123, -128);  add_28123 = None
	        clamp_max_367: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_551, 127);  clamp_min_551 = None
	        view_2873: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_367, [sym_size_int, 1500, 1280]);  clamp_max_367 = None
	        _assert_tensor_metadata_1651 = torch.ops.aten._assert_tensor_metadata.default(view_2873, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1651 = None
	        convert_element_type_1099: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2873, torch.int8);  view_2873 = None
	        view_2874: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1099, [sym_size_int, 1500, 1280]);  convert_element_type_1099 = None
	        view_2875: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_549, [sym_size_int, 1500, 1]);  clamp_min_549 = None
	        view_2876: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1098, [sym_size_int, 1500, 1]);  convert_element_type_1098 = None
	        _assert_tensor_metadata_1652 = torch.ops.aten._assert_tensor_metadata.default(view_2874, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1652 = None
	        convert_element_type_1100: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2874, torch.float32);  view_2874 = None
	        _assert_tensor_metadata_1653 = torch.ops.aten._assert_tensor_metadata.default(view_2876, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1653 = None
	        convert_element_type_1101: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2876, torch.float32);  view_2876 = None
	        sub_8401: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1100, convert_element_type_1101);  convert_element_type_1100 = convert_element_type_1101 = None
	        mul_17794: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8401, view_2875);  sub_8401 = view_2875 = None
	        view_2877: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17794, [sym_size_int, 1500, 1280]);  mul_17794 = None
	        _assert_tensor_metadata_1654 = torch.ops.aten._assert_tensor_metadata.default(view_2877, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1654 = None
	        view_2878: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg829_1, [1280, 40, 32]);  arg829_1 = None
	        view_2879: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg830_1, [1280, 40, 1]);  arg830_1 = None
	        view_2880: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg831_1, [1280, 40, 1]);  arg831_1 = None
	        _assert_tensor_metadata_1655 = torch.ops.aten._assert_tensor_metadata.default(view_2878, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1655 = None
	        convert_element_type_1102: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2878, torch.float32);  view_2878 = None
	        _assert_tensor_metadata_1656 = torch.ops.aten._assert_tensor_metadata.default(view_2880, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1656 = None
	        convert_element_type_1103: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2880, torch.float32);  view_2880 = None
	        sub_8405: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1102, convert_element_type_1103);  convert_element_type_1102 = convert_element_type_1103 = None
	        mul_17799: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8405, view_2879);  sub_8405 = view_2879 = None
	        view_2881: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17799, [1280, 1280]);  mul_17799 = None
	        _assert_tensor_metadata_1657 = torch.ops.aten._assert_tensor_metadata.default(view_2881, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1657 = None
	        mul_17804: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2882: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2877, [mul_17804, 1280]);  view_2877 = mul_17804 = None
	        permute_308: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2881, [1, 0]);  view_2881 = None
	        addmm_152: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg828_1, view_2882, permute_308);  arg828_1 = view_2882 = permute_308 = None
	        view_2883: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_152, [sym_size_int, 1500, 1280]);  addmm_152 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_245: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2883);  view_2883 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_28186: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_27566, clone_245);  add_27566 = clone_245 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_246: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_28186, memory_format = torch.contiguous_format)
	        var_mean_61 = torch.ops.aten.var_mean.correction(clone_246, [2], correction = 0, keepdim = True)
	        getitem_246: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_61[0]
	        getitem_247: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_61[1];  var_mean_61 = None
	        add_28191: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_246, 1e-05);  getitem_246 = None
	        rsqrt_61: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_28191);  add_28191 = None
	        sub_8411: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_246, getitem_247);  clone_246 = getitem_247 = None
	        mul_17815: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8411, rsqrt_61);  sub_8411 = rsqrt_61 = None
	        mul_17816: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17815, arg832_1);  mul_17815 = arg832_1 = None
	        add_28192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_17816, arg833_1);  mul_17816 = arg833_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2884: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_28192, [sym_size_int, 1500, 1280])
	        amin_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2884, [2])
	        amax_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2884, [2]);  view_2884 = None
	        full_368: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_184, full_368);  amin_184 = full_368 = None
	        full_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_184, full_369);  amax_184 = full_369 = None
	        sub_8422: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_184, minimum_184);  maximum_184 = None
	        div_368: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8422, 255.0);  sub_8422 = None
	        clamp_min_552: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_368, 1.1920928955078125e-07);  div_368 = None
	        div_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_184, clamp_min_552);  minimum_184 = None
	        round_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_369);  div_369 = None
	        sub_8428: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_369);  round_369 = None
	        clamp_min_553: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8428, -128);  sub_8428 = None
	        clamp_max_368: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_553, 127);  clamp_min_553 = None
	        _assert_tensor_metadata_1658 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_552, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1658 = None
	        _assert_tensor_metadata_1659 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_368, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1659 = None
	        convert_element_type_1104: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_368, torch.int8);  clamp_max_368 = None
	        view_2885: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_28192, [sym_size_int, 1500, 1280]);  add_28192 = None
	        view_2886: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_552, [sym_size_int, 1500, 1])
	        view_2887: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1104, [sym_size_int, 1500, 1])
	        reciprocal_184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2886);  view_2886 = None
	        mul_17864: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_184, 1.0);  reciprocal_184 = None
	        mul_17867: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2885, mul_17864);  view_2885 = mul_17864 = None
	        round_370: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17867);  mul_17867 = None
	        add_28279: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_370, view_2887);  round_370 = view_2887 = None
	        clamp_min_554: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28279, -128);  add_28279 = None
	        clamp_max_369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_554, 127);  clamp_min_554 = None
	        view_2888: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_369, [sym_size_int, 1500, 1280]);  clamp_max_369 = None
	        _assert_tensor_metadata_1660 = torch.ops.aten._assert_tensor_metadata.default(view_2888, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1660 = None
	        convert_element_type_1105: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2888, torch.int8);  view_2888 = None
	        view_2889: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1105, [sym_size_int, 1500, 1280]);  convert_element_type_1105 = None
	        view_2890: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_552, [sym_size_int, 1500, 1]);  clamp_min_552 = None
	        view_2891: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1104, [sym_size_int, 1500, 1]);  convert_element_type_1104 = None
	        _assert_tensor_metadata_1661 = torch.ops.aten._assert_tensor_metadata.default(view_2889, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1661 = None
	        convert_element_type_1106: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2889, torch.float32);  view_2889 = None
	        _assert_tensor_metadata_1662 = torch.ops.aten._assert_tensor_metadata.default(view_2891, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1662 = None
	        convert_element_type_1107: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2891, torch.float32);  view_2891 = None
	        sub_8448: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1106, convert_element_type_1107);  convert_element_type_1106 = convert_element_type_1107 = None
	        mul_17889: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8448, view_2890);  sub_8448 = view_2890 = None
	        view_2892: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17889, [sym_size_int, 1500, 1280]);  mul_17889 = None
	        _assert_tensor_metadata_1663 = torch.ops.aten._assert_tensor_metadata.default(view_2892, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1663 = None
	        view_2893: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg835_1, [5120, 40, 32]);  arg835_1 = None
	        view_2894: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg836_1, [5120, 40, 1]);  arg836_1 = None
	        view_2895: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg837_1, [5120, 40, 1]);  arg837_1 = None
	        _assert_tensor_metadata_1664 = torch.ops.aten._assert_tensor_metadata.default(view_2893, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1664 = None
	        convert_element_type_1108: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2893, torch.float32);  view_2893 = None
	        _assert_tensor_metadata_1665 = torch.ops.aten._assert_tensor_metadata.default(view_2895, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1665 = None
	        convert_element_type_1109: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2895, torch.float32);  view_2895 = None
	        sub_8452: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1108, convert_element_type_1109);  convert_element_type_1108 = convert_element_type_1109 = None
	        mul_17894: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8452, view_2894);  sub_8452 = view_2894 = None
	        view_2896: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17894, [5120, 1280]);  mul_17894 = None
	        _assert_tensor_metadata_1666 = torch.ops.aten._assert_tensor_metadata.default(view_2896, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1666 = None
	        mul_17899: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2897: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2892, [mul_17899, 1280]);  view_2892 = mul_17899 = None
	        permute_309: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2896, [1, 0]);  view_2896 = None
	        addmm_153: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg834_1, view_2897, permute_309);  arg834_1 = view_2897 = permute_309 = None
	        view_2898: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_153, [sym_size_int, 1500, 5120]);  addmm_153 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_17906: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2898, 0.5)
	        mul_17907: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2898, 0.7071067811865476);  view_2898 = None
	        erf_32: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_17907);  mul_17907 = None
	        add_28338: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_32, 1);  erf_32 = None
	        mul_17908: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17906, add_28338);  mul_17906 = add_28338 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_247: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_17908);  mul_17908 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2899: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_247, [sym_size_int, 1500, 5120])
	        amin_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2899, [2])
	        amax_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2899, [2]);  view_2899 = None
	        full_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_185, full_370);  amin_185 = full_370 = None
	        full_371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_185, full_371);  amax_185 = full_371 = None
	        sub_8465: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_185, minimum_185);  maximum_185 = None
	        div_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8465, 255.0);  sub_8465 = None
	        clamp_min_555: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_370, 1.1920928955078125e-07);  div_370 = None
	        div_371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_185, clamp_min_555);  minimum_185 = None
	        round_371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_371);  div_371 = None
	        sub_8471: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_371);  round_371 = None
	        clamp_min_556: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8471, -128);  sub_8471 = None
	        clamp_max_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_556, 127);  clamp_min_556 = None
	        _assert_tensor_metadata_1667 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_555, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1667 = None
	        _assert_tensor_metadata_1668 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_370, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1668 = None
	        convert_element_type_1110: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_370, torch.int8);  clamp_max_370 = None
	        view_2900: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_247, [sym_size_int, 1500, 5120]);  clone_247 = None
	        view_2901: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_555, [sym_size_int, 1500, 1])
	        view_2902: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1110, [sym_size_int, 1500, 1])
	        reciprocal_185: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2901);  view_2901 = None
	        mul_17954: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_185, 1.0);  reciprocal_185 = None
	        mul_17957: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2900, mul_17954);  view_2900 = mul_17954 = None
	        round_372: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_17957);  mul_17957 = None
	        add_28421: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_372, view_2902);  round_372 = view_2902 = None
	        clamp_min_557: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28421, -128);  add_28421 = None
	        clamp_max_371: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_557, 127);  clamp_min_557 = None
	        view_2903: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_371, [sym_size_int, 1500, 5120]);  clamp_max_371 = None
	        _assert_tensor_metadata_1669 = torch.ops.aten._assert_tensor_metadata.default(view_2903, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1669 = None
	        convert_element_type_1111: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2903, torch.int8);  view_2903 = None
	        view_2904: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1111, [sym_size_int, 1500, 5120]);  convert_element_type_1111 = None
	        view_2905: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_555, [sym_size_int, 1500, 1]);  clamp_min_555 = None
	        view_2906: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1110, [sym_size_int, 1500, 1]);  convert_element_type_1110 = None
	        _assert_tensor_metadata_1670 = torch.ops.aten._assert_tensor_metadata.default(view_2904, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1670 = None
	        convert_element_type_1112: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2904, torch.float32);  view_2904 = None
	        _assert_tensor_metadata_1671 = torch.ops.aten._assert_tensor_metadata.default(view_2906, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1671 = None
	        convert_element_type_1113: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2906, torch.float32);  view_2906 = None
	        sub_8491: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1112, convert_element_type_1113);  convert_element_type_1112 = convert_element_type_1113 = None
	        mul_17979: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8491, view_2905);  sub_8491 = view_2905 = None
	        view_2907: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_17979, [sym_size_int, 1500, 5120]);  mul_17979 = None
	        _assert_tensor_metadata_1672 = torch.ops.aten._assert_tensor_metadata.default(view_2907, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1672 = None
	        view_2908: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg839_1, [1280, 160, 32]);  arg839_1 = None
	        view_2909: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg840_1, [1280, 160, 1]);  arg840_1 = None
	        view_2910: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg841_1, [1280, 160, 1]);  arg841_1 = None
	        _assert_tensor_metadata_1673 = torch.ops.aten._assert_tensor_metadata.default(view_2908, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1673 = None
	        convert_element_type_1114: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2908, torch.float32);  view_2908 = None
	        _assert_tensor_metadata_1674 = torch.ops.aten._assert_tensor_metadata.default(view_2910, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1674 = None
	        convert_element_type_1115: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2910, torch.float32);  view_2910 = None
	        sub_8495: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1114, convert_element_type_1115);  convert_element_type_1114 = convert_element_type_1115 = None
	        mul_17984: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8495, view_2909);  sub_8495 = view_2909 = None
	        view_2911: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_17984, [1280, 5120]);  mul_17984 = None
	        _assert_tensor_metadata_1675 = torch.ops.aten._assert_tensor_metadata.default(view_2911, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1675 = None
	        mul_17989: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2912: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2907, [mul_17989, 5120]);  view_2907 = mul_17989 = None
	        permute_310: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2911, [1, 0]);  view_2911 = None
	        addmm_154: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg838_1, view_2912, permute_310);  arg838_1 = view_2912 = permute_310 = None
	        view_2913: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_154, [sym_size_int, 1500, 1280]);  addmm_154 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_248: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2913);  view_2913 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_28484: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_28186, clone_248);  add_28186 = clone_248 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_28484, memory_format = torch.contiguous_format)
	        var_mean_62 = torch.ops.aten.var_mean.correction(clone_249, [2], correction = 0, keepdim = True)
	        getitem_248: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_62[0]
	        getitem_249: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_62[1];  var_mean_62 = None
	        add_28489: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_248, 1e-05);  getitem_248 = None
	        rsqrt_62: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_28489);  add_28489 = None
	        sub_8501: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_249, getitem_249);  clone_249 = getitem_249 = None
	        mul_18000: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8501, rsqrt_62);  sub_8501 = rsqrt_62 = None
	        mul_18001: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18000, arg842_1);  mul_18000 = arg842_1 = None
	        add_28490: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_18001, arg843_1);  mul_18001 = arg843_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2914: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_28490, [sym_size_int, 1500, 1280])
	        amin_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2914, [2])
	        amax_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2914, [2]);  view_2914 = None
	        full_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_186, full_372);  amin_186 = full_372 = None
	        full_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_186, full_373);  amax_186 = full_373 = None
	        sub_8512: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_186, minimum_186);  maximum_186 = None
	        div_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8512, 255.0);  sub_8512 = None
	        clamp_min_558: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_372, 1.1920928955078125e-07);  div_372 = None
	        div_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_186, clamp_min_558);  minimum_186 = None
	        round_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_373);  div_373 = None
	        sub_8518: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_373);  round_373 = None
	        clamp_min_559: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8518, -128);  sub_8518 = None
	        clamp_max_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_559, 127);  clamp_min_559 = None
	        _assert_tensor_metadata_1676 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_558, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1676 = None
	        _assert_tensor_metadata_1677 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_372, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1677 = None
	        convert_element_type_1116: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_372, torch.int8);  clamp_max_372 = None
	        view_2915: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_28490, [sym_size_int, 1500, 1280])
	        view_2916: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_558, [sym_size_int, 1500, 1])
	        view_2917: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1116, [sym_size_int, 1500, 1])
	        reciprocal_186: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2916);  view_2916 = None
	        mul_18049: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_186, 1.0);  reciprocal_186 = None
	        mul_18052: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2915, mul_18049);  view_2915 = mul_18049 = None
	        round_374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18052);  mul_18052 = None
	        add_28577: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_374, view_2917);  round_374 = view_2917 = None
	        clamp_min_560: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28577, -128);  add_28577 = None
	        clamp_max_373: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_560, 127);  clamp_min_560 = None
	        view_2918: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_373, [sym_size_int, 1500, 1280]);  clamp_max_373 = None
	        _assert_tensor_metadata_1678 = torch.ops.aten._assert_tensor_metadata.default(view_2918, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1678 = None
	        convert_element_type_1117: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2918, torch.int8);  view_2918 = None
	        view_2919: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1117, [sym_size_int, 1500, 1280]);  convert_element_type_1117 = None
	        view_2920: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_558, [sym_size_int, 1500, 1]);  clamp_min_558 = None
	        view_2921: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1116, [sym_size_int, 1500, 1]);  convert_element_type_1116 = None
	        _assert_tensor_metadata_1679 = torch.ops.aten._assert_tensor_metadata.default(view_2919, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1679 = None
	        convert_element_type_1118: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2919, torch.float32);  view_2919 = None
	        _assert_tensor_metadata_1680 = torch.ops.aten._assert_tensor_metadata.default(view_2921, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1680 = None
	        convert_element_type_1119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2921, torch.float32);  view_2921 = None
	        sub_8538: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1118, convert_element_type_1119);  convert_element_type_1118 = convert_element_type_1119 = None
	        mul_18074: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8538, view_2920);  sub_8538 = view_2920 = None
	        view_2922: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18074, [sym_size_int, 1500, 1280]);  mul_18074 = None
	        _assert_tensor_metadata_1681 = torch.ops.aten._assert_tensor_metadata.default(view_2922, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1681 = None
	        view_2923: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg845_1, [1280, 40, 32]);  arg845_1 = None
	        view_2924: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg846_1, [1280, 40, 1]);  arg846_1 = None
	        view_2925: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg847_1, [1280, 40, 1]);  arg847_1 = None
	        _assert_tensor_metadata_1682 = torch.ops.aten._assert_tensor_metadata.default(view_2923, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1682 = None
	        convert_element_type_1120: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2923, torch.float32);  view_2923 = None
	        _assert_tensor_metadata_1683 = torch.ops.aten._assert_tensor_metadata.default(view_2925, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1683 = None
	        convert_element_type_1121: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2925, torch.float32);  view_2925 = None
	        sub_8542: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1120, convert_element_type_1121);  convert_element_type_1120 = convert_element_type_1121 = None
	        mul_18079: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8542, view_2924);  sub_8542 = view_2924 = None
	        view_2926: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18079, [1280, 1280]);  mul_18079 = None
	        _assert_tensor_metadata_1684 = torch.ops.aten._assert_tensor_metadata.default(view_2926, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1684 = None
	        mul_18084: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2927: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2922, [mul_18084, 1280]);  view_2922 = mul_18084 = None
	        permute_311: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2926, [1, 0]);  view_2926 = None
	        addmm_155: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg844_1, view_2927, permute_311);  arg844_1 = view_2927 = permute_311 = None
	        view_2928: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_155, [sym_size_int, 1500, 1280]);  addmm_155 = None
	        mul_18091: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2928, 0.125);  view_2928 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2929: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_18091, [sym_size_int, 1500, 20, 64]);  mul_18091 = None
	        permute_312: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2929, [0, 2, 1, 3]);  view_2929 = None
	        clone_250: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_312, memory_format = torch.contiguous_format);  permute_312 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2930: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_28490, [sym_size_int, 1500, 1280])
	        amin_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2930, [2])
	        amax_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2930, [2]);  view_2930 = None
	        full_374: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_187, full_374);  amin_187 = full_374 = None
	        full_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_187, full_375);  amax_187 = full_375 = None
	        sub_8557: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_187, minimum_187);  maximum_187 = None
	        div_374: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8557, 255.0);  sub_8557 = None
	        clamp_min_561: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_374, 1.1920928955078125e-07);  div_374 = None
	        div_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_187, clamp_min_561);  minimum_187 = None
	        round_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_375);  div_375 = None
	        sub_8563: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_375);  round_375 = None
	        clamp_min_562: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8563, -128);  sub_8563 = None
	        clamp_max_374: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_562, 127);  clamp_min_562 = None
	        _assert_tensor_metadata_1685 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_561, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1685 = None
	        _assert_tensor_metadata_1686 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_374, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1686 = None
	        convert_element_type_1122: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_374, torch.int8);  clamp_max_374 = None
	        view_2931: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_28490, [sym_size_int, 1500, 1280])
	        view_2932: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_561, [sym_size_int, 1500, 1])
	        view_2933: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1122, [sym_size_int, 1500, 1])
	        reciprocal_187: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2932);  view_2932 = None
	        mul_18145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_187, 1.0);  reciprocal_187 = None
	        mul_18148: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2931, mul_18145);  view_2931 = mul_18145 = None
	        round_376: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18148);  mul_18148 = None
	        add_28729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_376, view_2933);  round_376 = view_2933 = None
	        clamp_min_563: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28729, -128);  add_28729 = None
	        clamp_max_375: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_563, 127);  clamp_min_563 = None
	        view_2934: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_375, [sym_size_int, 1500, 1280]);  clamp_max_375 = None
	        _assert_tensor_metadata_1687 = torch.ops.aten._assert_tensor_metadata.default(view_2934, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1687 = None
	        convert_element_type_1123: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2934, torch.int8);  view_2934 = None
	        view_2935: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1123, [sym_size_int, 1500, 1280]);  convert_element_type_1123 = None
	        view_2936: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_561, [sym_size_int, 1500, 1]);  clamp_min_561 = None
	        view_2937: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1122, [sym_size_int, 1500, 1]);  convert_element_type_1122 = None
	        _assert_tensor_metadata_1688 = torch.ops.aten._assert_tensor_metadata.default(view_2935, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1688 = None
	        convert_element_type_1124: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2935, torch.float32);  view_2935 = None
	        _assert_tensor_metadata_1689 = torch.ops.aten._assert_tensor_metadata.default(view_2937, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1689 = None
	        convert_element_type_1125: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2937, torch.float32);  view_2937 = None
	        sub_8583: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1124, convert_element_type_1125);  convert_element_type_1124 = convert_element_type_1125 = None
	        mul_18170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8583, view_2936);  sub_8583 = view_2936 = None
	        view_2938: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18170, [sym_size_int, 1500, 1280]);  mul_18170 = None
	        _assert_tensor_metadata_1690 = torch.ops.aten._assert_tensor_metadata.default(view_2938, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1690 = None
	        view_2939: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg848_1, [1280, 40, 32]);  arg848_1 = None
	        view_2940: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg849_1, [1280, 40, 1]);  arg849_1 = None
	        view_2941: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg850_1, [1280, 40, 1]);  arg850_1 = None
	        _assert_tensor_metadata_1691 = torch.ops.aten._assert_tensor_metadata.default(view_2939, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1691 = None
	        convert_element_type_1126: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2939, torch.float32);  view_2939 = None
	        _assert_tensor_metadata_1692 = torch.ops.aten._assert_tensor_metadata.default(view_2941, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1692 = None
	        convert_element_type_1127: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2941, torch.float32);  view_2941 = None
	        sub_8587: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1126, convert_element_type_1127);  convert_element_type_1126 = convert_element_type_1127 = None
	        mul_18175: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8587, view_2940);  sub_8587 = view_2940 = None
	        view_2942: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18175, [1280, 1280]);  mul_18175 = None
	        _assert_tensor_metadata_1693 = torch.ops.aten._assert_tensor_metadata.default(view_2942, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1693 = None
	        permute_313: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2942, [1, 0]);  view_2942 = None
	        mul_18178: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2943: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2938, [mul_18178, 1280]);  view_2938 = mul_18178 = None
	        mm_31: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2943, permute_313);  view_2943 = permute_313 = None
	        view_2944: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_31, [sym_size_int, 1500, 1280]);  mm_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2945: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2944, [sym_size_int, -1, 20, 64]);  view_2944 = None
	        permute_314: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2945, [0, 2, 1, 3]);  view_2945 = None
	        clone_251: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_314, memory_format = torch.contiguous_format);  permute_314 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2946: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_28490, [sym_size_int, 1500, 1280])
	        amin_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2946, [2])
	        amax_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2946, [2]);  view_2946 = None
	        full_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_188, full_376);  amin_188 = full_376 = None
	        full_377: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_188, full_377);  amax_188 = full_377 = None
	        sub_8601: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_188, minimum_188);  maximum_188 = None
	        div_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8601, 255.0);  sub_8601 = None
	        clamp_min_564: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_376, 1.1920928955078125e-07);  div_376 = None
	        div_377: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_188, clamp_min_564);  minimum_188 = None
	        round_377: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_377);  div_377 = None
	        sub_8607: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_377);  round_377 = None
	        clamp_min_565: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8607, -128);  sub_8607 = None
	        clamp_max_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_565, 127);  clamp_min_565 = None
	        _assert_tensor_metadata_1694 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_564, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1694 = None
	        _assert_tensor_metadata_1695 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_376, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1695 = None
	        convert_element_type_1128: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_376, torch.int8);  clamp_max_376 = None
	        view_2947: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_28490, [sym_size_int, 1500, 1280]);  add_28490 = None
	        view_2948: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_564, [sym_size_int, 1500, 1])
	        view_2949: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1128, [sym_size_int, 1500, 1])
	        reciprocal_188: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2948);  view_2948 = None
	        mul_18244: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_188, 1.0);  reciprocal_188 = None
	        mul_18247: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2947, mul_18244);  view_2947 = mul_18244 = None
	        round_378: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18247);  mul_18247 = None
	        add_28877: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_378, view_2949);  round_378 = view_2949 = None
	        clamp_min_566: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28877, -128);  add_28877 = None
	        clamp_max_377: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_566, 127);  clamp_min_566 = None
	        view_2950: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_377, [sym_size_int, 1500, 1280]);  clamp_max_377 = None
	        _assert_tensor_metadata_1696 = torch.ops.aten._assert_tensor_metadata.default(view_2950, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1696 = None
	        convert_element_type_1129: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2950, torch.int8);  view_2950 = None
	        view_2951: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1129, [sym_size_int, 1500, 1280]);  convert_element_type_1129 = None
	        view_2952: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_564, [sym_size_int, 1500, 1]);  clamp_min_564 = None
	        view_2953: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1128, [sym_size_int, 1500, 1]);  convert_element_type_1128 = None
	        _assert_tensor_metadata_1697 = torch.ops.aten._assert_tensor_metadata.default(view_2951, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1697 = None
	        convert_element_type_1130: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2951, torch.float32);  view_2951 = None
	        _assert_tensor_metadata_1698 = torch.ops.aten._assert_tensor_metadata.default(view_2953, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1698 = None
	        convert_element_type_1131: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2953, torch.float32);  view_2953 = None
	        sub_8627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1130, convert_element_type_1131);  convert_element_type_1130 = convert_element_type_1131 = None
	        mul_18269: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8627, view_2952);  sub_8627 = view_2952 = None
	        view_2954: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18269, [sym_size_int, 1500, 1280]);  mul_18269 = None
	        _assert_tensor_metadata_1699 = torch.ops.aten._assert_tensor_metadata.default(view_2954, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1699 = None
	        view_2955: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg852_1, [1280, 40, 32]);  arg852_1 = None
	        view_2956: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg853_1, [1280, 40, 1]);  arg853_1 = None
	        view_2957: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg854_1, [1280, 40, 1]);  arg854_1 = None
	        _assert_tensor_metadata_1700 = torch.ops.aten._assert_tensor_metadata.default(view_2955, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1700 = None
	        convert_element_type_1132: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2955, torch.float32);  view_2955 = None
	        _assert_tensor_metadata_1701 = torch.ops.aten._assert_tensor_metadata.default(view_2957, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1701 = None
	        convert_element_type_1133: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2957, torch.float32);  view_2957 = None
	        sub_8631: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1132, convert_element_type_1133);  convert_element_type_1132 = convert_element_type_1133 = None
	        mul_18274: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8631, view_2956);  sub_8631 = view_2956 = None
	        view_2958: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18274, [1280, 1280]);  mul_18274 = None
	        _assert_tensor_metadata_1702 = torch.ops.aten._assert_tensor_metadata.default(view_2958, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1702 = None
	        mul_18279: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2959: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2954, [mul_18279, 1280]);  view_2954 = mul_18279 = None
	        permute_315: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2958, [1, 0]);  view_2958 = None
	        addmm_156: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg851_1, view_2959, permute_315);  arg851_1 = view_2959 = permute_315 = None
	        view_2960: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_156, [sym_size_int, 1500, 1280]);  addmm_156 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2961: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2960, [sym_size_int, -1, 20, 64]);  view_2960 = None
	        permute_316: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2961, [0, 2, 1, 3]);  view_2961 = None
	        clone_252: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_316, memory_format = torch.contiguous_format);  permute_316 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_31 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_250, clone_251, clone_252, None, False, scale = 1.0);  clone_250 = clone_251 = clone_252 = None
	        getitem_250: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_31[0];  _scaled_dot_product_efficient_attention_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_317: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_250, [0, 2, 1, 3]);  getitem_250 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2962: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_317, [sym_size_int, 1500, -1]);  permute_317 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2963: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2962, [sym_size_int, 1500, 1280])
	        amin_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2963, [2])
	        amax_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2963, [2]);  view_2963 = None
	        full_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_189, full_378);  amin_189 = full_378 = None
	        full_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_189, full_379);  amax_189 = full_379 = None
	        sub_8649: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_189, minimum_189);  maximum_189 = None
	        div_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8649, 255.0);  sub_8649 = None
	        clamp_min_567: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_378, 1.1920928955078125e-07);  div_378 = None
	        div_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_189, clamp_min_567);  minimum_189 = None
	        round_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_379);  div_379 = None
	        sub_8655: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_379);  round_379 = None
	        clamp_min_568: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8655, -128);  sub_8655 = None
	        clamp_max_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_568, 127);  clamp_min_568 = None
	        _assert_tensor_metadata_1703 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_567, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1703 = None
	        _assert_tensor_metadata_1704 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_378, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1704 = None
	        convert_element_type_1134: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_378, torch.int8);  clamp_max_378 = None
	        view_2964: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2962, [sym_size_int, 1500, 1280]);  view_2962 = None
	        view_2965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_567, [sym_size_int, 1500, 1])
	        view_2966: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1134, [sym_size_int, 1500, 1])
	        reciprocal_189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2965);  view_2965 = None
	        mul_18349: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_189, 1.0);  reciprocal_189 = None
	        mul_18352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2964, mul_18349);  view_2964 = mul_18349 = None
	        round_380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18352);  mul_18352 = None
	        add_29041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_380, view_2966);  round_380 = view_2966 = None
	        clamp_min_569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29041, -128);  add_29041 = None
	        clamp_max_379: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_569, 127);  clamp_min_569 = None
	        view_2967: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_379, [sym_size_int, 1500, 1280]);  clamp_max_379 = None
	        _assert_tensor_metadata_1705 = torch.ops.aten._assert_tensor_metadata.default(view_2967, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1705 = None
	        convert_element_type_1135: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2967, torch.int8);  view_2967 = None
	        view_2968: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1135, [sym_size_int, 1500, 1280]);  convert_element_type_1135 = None
	        view_2969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_567, [sym_size_int, 1500, 1]);  clamp_min_567 = None
	        view_2970: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1134, [sym_size_int, 1500, 1]);  convert_element_type_1134 = None
	        _assert_tensor_metadata_1706 = torch.ops.aten._assert_tensor_metadata.default(view_2968, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1706 = None
	        convert_element_type_1136: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2968, torch.float32);  view_2968 = None
	        _assert_tensor_metadata_1707 = torch.ops.aten._assert_tensor_metadata.default(view_2970, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1707 = None
	        convert_element_type_1137: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2970, torch.float32);  view_2970 = None
	        sub_8675: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1136, convert_element_type_1137);  convert_element_type_1136 = convert_element_type_1137 = None
	        mul_18374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8675, view_2969);  sub_8675 = view_2969 = None
	        view_2971: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18374, [sym_size_int, 1500, 1280]);  mul_18374 = None
	        _assert_tensor_metadata_1708 = torch.ops.aten._assert_tensor_metadata.default(view_2971, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1708 = None
	        view_2972: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg856_1, [1280, 40, 32]);  arg856_1 = None
	        view_2973: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg857_1, [1280, 40, 1]);  arg857_1 = None
	        view_2974: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg858_1, [1280, 40, 1]);  arg858_1 = None
	        _assert_tensor_metadata_1709 = torch.ops.aten._assert_tensor_metadata.default(view_2972, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1709 = None
	        convert_element_type_1138: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2972, torch.float32);  view_2972 = None
	        _assert_tensor_metadata_1710 = torch.ops.aten._assert_tensor_metadata.default(view_2974, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1710 = None
	        convert_element_type_1139: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2974, torch.float32);  view_2974 = None
	        sub_8679: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1138, convert_element_type_1139);  convert_element_type_1138 = convert_element_type_1139 = None
	        mul_18379: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8679, view_2973);  sub_8679 = view_2973 = None
	        view_2975: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18379, [1280, 1280]);  mul_18379 = None
	        _assert_tensor_metadata_1711 = torch.ops.aten._assert_tensor_metadata.default(view_2975, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1711 = None
	        mul_18384: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2976: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2971, [mul_18384, 1280]);  view_2971 = mul_18384 = None
	        permute_318: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2975, [1, 0]);  view_2975 = None
	        addmm_157: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg855_1, view_2976, permute_318);  arg855_1 = view_2976 = permute_318 = None
	        view_2977: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_157, [sym_size_int, 1500, 1280]);  addmm_157 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_253: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2977);  view_2977 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_29104: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_28484, clone_253);  add_28484 = clone_253 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_29104, memory_format = torch.contiguous_format)
	        var_mean_63 = torch.ops.aten.var_mean.correction(clone_254, [2], correction = 0, keepdim = True)
	        getitem_254: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_63[0]
	        getitem_255: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_63[1];  var_mean_63 = None
	        add_29109: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_254, 1e-05);  getitem_254 = None
	        rsqrt_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_29109);  add_29109 = None
	        sub_8685: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_254, getitem_255);  clone_254 = getitem_255 = None
	        mul_18395: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8685, rsqrt_63);  sub_8685 = rsqrt_63 = None
	        mul_18396: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18395, arg859_1);  mul_18395 = arg859_1 = None
	        add_29110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_18396, arg860_1);  mul_18396 = arg860_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2978: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_29110, [sym_size_int, 1500, 1280])
	        amin_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2978, [2])
	        amax_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2978, [2]);  view_2978 = None
	        full_380: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_190, full_380);  amin_190 = full_380 = None
	        full_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_190, full_381);  amax_190 = full_381 = None
	        sub_8696: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_190, minimum_190);  maximum_190 = None
	        div_380: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8696, 255.0);  sub_8696 = None
	        clamp_min_570: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_380, 1.1920928955078125e-07);  div_380 = None
	        div_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_190, clamp_min_570);  minimum_190 = None
	        round_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_381);  div_381 = None
	        sub_8702: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_381);  round_381 = None
	        clamp_min_571: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8702, -128);  sub_8702 = None
	        clamp_max_380: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_571, 127);  clamp_min_571 = None
	        _assert_tensor_metadata_1712 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_570, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1712 = None
	        _assert_tensor_metadata_1713 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_380, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1713 = None
	        convert_element_type_1140: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_380, torch.int8);  clamp_max_380 = None
	        view_2979: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_29110, [sym_size_int, 1500, 1280]);  add_29110 = None
	        view_2980: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_570, [sym_size_int, 1500, 1])
	        view_2981: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1140, [sym_size_int, 1500, 1])
	        reciprocal_190: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2980);  view_2980 = None
	        mul_18444: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_190, 1.0);  reciprocal_190 = None
	        mul_18447: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2979, mul_18444);  view_2979 = mul_18444 = None
	        round_382: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18447);  mul_18447 = None
	        add_29197: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_382, view_2981);  round_382 = view_2981 = None
	        clamp_min_572: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29197, -128);  add_29197 = None
	        clamp_max_381: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_572, 127);  clamp_min_572 = None
	        view_2982: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_381, [sym_size_int, 1500, 1280]);  clamp_max_381 = None
	        _assert_tensor_metadata_1714 = torch.ops.aten._assert_tensor_metadata.default(view_2982, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1714 = None
	        convert_element_type_1141: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2982, torch.int8);  view_2982 = None
	        view_2983: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1141, [sym_size_int, 1500, 1280]);  convert_element_type_1141 = None
	        view_2984: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_570, [sym_size_int, 1500, 1]);  clamp_min_570 = None
	        view_2985: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1140, [sym_size_int, 1500, 1]);  convert_element_type_1140 = None
	        _assert_tensor_metadata_1715 = torch.ops.aten._assert_tensor_metadata.default(view_2983, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1715 = None
	        convert_element_type_1142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2983, torch.float32);  view_2983 = None
	        _assert_tensor_metadata_1716 = torch.ops.aten._assert_tensor_metadata.default(view_2985, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1716 = None
	        convert_element_type_1143: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2985, torch.float32);  view_2985 = None
	        sub_8722: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1142, convert_element_type_1143);  convert_element_type_1142 = convert_element_type_1143 = None
	        mul_18469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8722, view_2984);  sub_8722 = view_2984 = None
	        view_2986: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18469, [sym_size_int, 1500, 1280]);  mul_18469 = None
	        _assert_tensor_metadata_1717 = torch.ops.aten._assert_tensor_metadata.default(view_2986, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1717 = None
	        view_2987: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(arg862_1, [5120, 40, 32]);  arg862_1 = None
	        view_2988: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg863_1, [5120, 40, 1]);  arg863_1 = None
	        view_2989: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(arg864_1, [5120, 40, 1]);  arg864_1 = None
	        _assert_tensor_metadata_1718 = torch.ops.aten._assert_tensor_metadata.default(view_2987, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1718 = None
	        convert_element_type_1144: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2987, torch.float32);  view_2987 = None
	        _assert_tensor_metadata_1719 = torch.ops.aten._assert_tensor_metadata.default(view_2989, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1719 = None
	        convert_element_type_1145: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2989, torch.float32);  view_2989 = None
	        sub_8726: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1144, convert_element_type_1145);  convert_element_type_1144 = convert_element_type_1145 = None
	        mul_18474: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8726, view_2988);  sub_8726 = view_2988 = None
	        view_2990: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18474, [5120, 1280]);  mul_18474 = None
	        _assert_tensor_metadata_1720 = torch.ops.aten._assert_tensor_metadata.default(view_2990, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1720 = None
	        mul_18479: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2991: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2986, [mul_18479, 1280]);  view_2986 = mul_18479 = None
	        permute_319: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2990, [1, 0]);  view_2990 = None
	        addmm_158: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(arg861_1, view_2991, permute_319);  arg861_1 = view_2991 = permute_319 = None
	        view_2992: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_158, [sym_size_int, 1500, 5120]);  addmm_158 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_18486: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2992, 0.5)
	        mul_18487: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2992, 0.7071067811865476);  view_2992 = None
	        erf_33: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_18487);  mul_18487 = None
	        add_29256: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_33, 1);  erf_33 = None
	        mul_18488: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18486, add_29256);  mul_18486 = add_29256 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_255: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_18488);  mul_18488 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2993: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_255, [sym_size_int, 1500, 5120])
	        amin_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2993, [2])
	        amax_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2993, [2]);  view_2993 = None
	        full_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_191, full_382);  amin_191 = full_382 = None
	        full_383: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_191, full_383);  amax_191 = full_383 = None
	        sub_8739: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_191, minimum_191);  maximum_191 = None
	        div_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8739, 255.0);  sub_8739 = None
	        clamp_min_573: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_382, 1.1920928955078125e-07);  div_382 = None
	        div_383: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_191, clamp_min_573);  minimum_191 = None
	        round_383: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_383);  div_383 = None
	        sub_8745: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_383);  round_383 = None
	        clamp_min_574: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8745, -128);  sub_8745 = None
	        clamp_max_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_574, 127);  clamp_min_574 = None
	        _assert_tensor_metadata_1721 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_573, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1721 = None
	        _assert_tensor_metadata_1722 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_382, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1722 = None
	        convert_element_type_1146: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_382, torch.int8);  clamp_max_382 = None
	        view_2994: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_255, [sym_size_int, 1500, 5120]);  clone_255 = None
	        view_2995: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_573, [sym_size_int, 1500, 1])
	        view_2996: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1146, [sym_size_int, 1500, 1])
	        reciprocal_191: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2995);  view_2995 = None
	        mul_18534: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_191, 1.0);  reciprocal_191 = None
	        mul_18537: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2994, mul_18534);  view_2994 = mul_18534 = None
	        round_384: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_18537);  mul_18537 = None
	        add_29339: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_384, view_2996);  round_384 = view_2996 = None
	        clamp_min_575: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29339, -128);  add_29339 = None
	        clamp_max_383: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_575, 127);  clamp_min_575 = None
	        view_2997: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_383, [sym_size_int, 1500, 5120]);  clamp_max_383 = None
	        _assert_tensor_metadata_1723 = torch.ops.aten._assert_tensor_metadata.default(view_2997, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1723 = None
	        convert_element_type_1147: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2997, torch.int8);  view_2997 = None
	        view_2998: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1147, [sym_size_int, 1500, 5120]);  convert_element_type_1147 = None
	        view_2999: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_573, [sym_size_int, 1500, 1]);  clamp_min_573 = None
	        view_3000: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1146, [sym_size_int, 1500, 1]);  convert_element_type_1146 = None
	        _assert_tensor_metadata_1724 = torch.ops.aten._assert_tensor_metadata.default(view_2998, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1724 = None
	        convert_element_type_1148: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2998, torch.float32);  view_2998 = None
	        _assert_tensor_metadata_1725 = torch.ops.aten._assert_tensor_metadata.default(view_3000, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1725 = None
	        convert_element_type_1149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3000, torch.float32);  view_3000 = None
	        sub_8765: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1148, convert_element_type_1149);  convert_element_type_1148 = convert_element_type_1149 = None
	        mul_18559: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8765, view_2999);  sub_8765 = view_2999 = None
	        view_3001: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_18559, [sym_size_int, 1500, 5120]);  mul_18559 = None
	        _assert_tensor_metadata_1726 = torch.ops.aten._assert_tensor_metadata.default(view_3001, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1726 = None
	        view_3002: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg866_1, [1280, 160, 32]);  arg866_1 = None
	        view_3003: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg867_1, [1280, 160, 1]);  arg867_1 = None
	        view_3004: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg868_1, [1280, 160, 1]);  arg868_1 = None
	        _assert_tensor_metadata_1727 = torch.ops.aten._assert_tensor_metadata.default(view_3002, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1727 = None
	        convert_element_type_1150: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3002, torch.float32);  view_3002 = None
	        _assert_tensor_metadata_1728 = torch.ops.aten._assert_tensor_metadata.default(view_3004, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1728 = None
	        convert_element_type_1151: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3004, torch.float32);  view_3004 = None
	        sub_8769: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1150, convert_element_type_1151);  convert_element_type_1150 = convert_element_type_1151 = None
	        mul_18564: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8769, view_3003);  sub_8769 = view_3003 = None
	        view_3005: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_18564, [1280, 5120]);  mul_18564 = None
	        _assert_tensor_metadata_1729 = torch.ops.aten._assert_tensor_metadata.default(view_3005, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1729 = None
	        mul_18569: "Sym(1500*s6)" = sym_size_int * 1500
	        view_3006: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_3001, [mul_18569, 5120]);  view_3001 = mul_18569 = None
	        permute_320: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_3005, [1, 0]);  view_3005 = None
	        addmm_159: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(arg865_1, view_3006, permute_320);  arg865_1 = view_3006 = permute_320 = None
	        view_3007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_159, [sym_size_int, 1500, 1280]);  addmm_159 = sym_size_int = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_256: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_3007);  view_3007 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_29402: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_29104, clone_256);  add_29104 = clone_256 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:365 in forward, code: hidden_states = self.layer_norm(hidden_states)
	        clone_257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_29402, memory_format = torch.contiguous_format);  add_29402 = None
	        var_mean_64 = torch.ops.aten.var_mean.correction(clone_257, [2], correction = 0, keepdim = True)
	        getitem_256: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_64[0]
	        getitem_257: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_64[1];  var_mean_64 = None
	        add_29407: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_256, 1e-05);  getitem_256 = None
	        rsqrt_64: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_29407);  add_29407 = None
	        sub_8775: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_257, getitem_257);  clone_257 = getitem_257 = None
	        mul_18580: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8775, rsqrt_64);  sub_8775 = rsqrt_64 = None
	        mul_18581: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18580, arg869_1);  mul_18580 = arg869_1 = None
	        add_29408: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_18581, arg870_1);  mul_18581 = arg870_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:451 in get_audio_embeds, code: audio_hidden_states = audio_hidden_states.reshape(-1, self.config.audio_config.intermediate_size)
	        view_3008: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(add_29408, [-1, 5120]);  add_29408 = None
	        sym_size_int_193: "Sym(375*s6)" = torch.ops.aten.sym_size.int(view_3008, 0)
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:389 in forward, code: hidden_states = self.linear_1(audio_features)
	        view_3009: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_3008, [sym_size_int_193, 5120])
	        amin_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.amin.default(view_3009, [1])
	        amax_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.amax.default(view_3009, [1]);  view_3009 = None
	        full_384: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.minimum.default(amin_192, full_384);  amin_192 = full_384 = None
	        full_385: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.maximum.default(amax_192, full_385);  amax_192 = full_385 = None
	        sub_8787: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_192, minimum_192);  maximum_192 = None
	        div_384: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(sub_8787, 255.0);  sub_8787 = None
	        clamp_min_576: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(div_384, 1.1920928955078125e-07);  div_384 = None
	        div_385: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(minimum_192, clamp_min_576);  minimum_192 = None
	        round_385: "f32[375*s6][1]cuda:0" = torch.ops.aten.round.default(div_385);  div_385 = None
	        sub_8793: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_385);  round_385 = None
	        clamp_min_577: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8793, -128);  sub_8793 = None
	        clamp_max_384: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_577, 127);  clamp_min_577 = None
	        _assert_tensor_metadata_1730 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_576, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1730 = None
	        _assert_tensor_metadata_1731 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_384, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1731 = None
	        convert_element_type_1152: "i8[375*s6][1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_384, torch.int8);  clamp_max_384 = None
	        view_3010: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_3008, [sym_size_int_193, 5120]);  view_3008 = None
	        view_3011: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_576, [sym_size_int_193, 1])
	        view_3012: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1152, [sym_size_int_193, 1])
	        reciprocal_192: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_3011);  view_3011 = None
	        mul_18613: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_192, 1.0);  reciprocal_192 = None
	        mul_18615: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_3010, mul_18613);  view_3010 = mul_18613 = None
	        round_386: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.round.default(mul_18615);  mul_18615 = None
	        add_29476: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_386, view_3012);  round_386 = view_3012 = None
	        clamp_min_578: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29476, -128);  add_29476 = None
	        clamp_max_385: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_578, 127);  clamp_min_578 = None
	        view_3013: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_385, [sym_size_int_193, 5120]);  clamp_max_385 = None
	        _assert_tensor_metadata_1732 = torch.ops.aten._assert_tensor_metadata.default(view_3013, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1732 = None
	        convert_element_type_1153: "i8[375*s6, 5120][5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3013, torch.int8);  view_3013 = None
	        view_3014: "i8[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1153, [sym_size_int_193, 5120]);  convert_element_type_1153 = None
	        view_3015: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_576, [sym_size_int_193, 1]);  clamp_min_576 = None
	        view_3016: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1152, [sym_size_int_193, 1]);  convert_element_type_1152 = None
	        _assert_tensor_metadata_1733 = torch.ops.aten._assert_tensor_metadata.default(view_3014, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1733 = None
	        convert_element_type_1154: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3014, torch.float32);  view_3014 = None
	        _assert_tensor_metadata_1734 = torch.ops.aten._assert_tensor_metadata.default(view_3016, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1734 = None
	        convert_element_type_1155: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3016, torch.float32);  view_3016 = None
	        sub_8813: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1154, convert_element_type_1155);  convert_element_type_1154 = convert_element_type_1155 = None
	        mul_18634: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8813, view_3015);  sub_8813 = view_3015 = None
	        view_3017: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_18634, [sym_size_int_193, 5120]);  mul_18634 = None
	        _assert_tensor_metadata_1735 = torch.ops.aten._assert_tensor_metadata.default(view_3017, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1735 = None
	        view_3018: "i8[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(arg871_1, [3072, 160, 32]);  arg871_1 = None
	        view_3019: "f32[3072, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg872_1, [3072, 160, 1]);  arg872_1 = None
	        view_3020: "i8[3072, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(arg873_1, [3072, 160, 1]);  arg873_1 = None
	        _assert_tensor_metadata_1736 = torch.ops.aten._assert_tensor_metadata.default(view_3018, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1736 = None
	        convert_element_type_1156: "f32[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3018, torch.float32);  view_3018 = None
	        _assert_tensor_metadata_1737 = torch.ops.aten._assert_tensor_metadata.default(view_3020, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1737 = None
	        convert_element_type_1157: "f32[3072, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3020, torch.float32);  view_3020 = None
	        sub_8817: "f32[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1156, convert_element_type_1157);  convert_element_type_1156 = convert_element_type_1157 = None
	        mul_18639: "f32[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8817, view_3019);  sub_8817 = view_3019 = None
	        view_3021: "f32[3072, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_18639, [3072, 5120]);  mul_18639 = None
	        _assert_tensor_metadata_1738 = torch.ops.aten._assert_tensor_metadata.default(view_3021, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1738 = None
	        permute_321: "f32[5120, 3072][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_3021, [1, 0]);  view_3021 = None
	        mm_32: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mm.default(view_3017, permute_321);  view_3017 = permute_321 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_18642: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(mm_32, 0.5)
	        mul_18643: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(mm_32, 0.7071067811865476);  mm_32 = None
	        erf_34: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.erf.default(mul_18643);  mul_18643 = None
	        add_29516: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_34, 1);  erf_34 = None
	        mul_18644: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18642, add_29516);  mul_18642 = add_29516 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:391 in forward, code: hidden_states = self.linear_2(hidden_states)
	        view_3022: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.view.default(mul_18644, [sym_size_int_193, 3072])
	        amin_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.amin.default(view_3022, [1])
	        amax_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.amax.default(view_3022, [1]);  view_3022 = None
	        full_386: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.minimum.default(amin_193, full_386);  amin_193 = full_386 = None
	        full_387: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.maximum.default(amax_193, full_387);  amax_193 = full_387 = None
	        sub_8827: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_193, minimum_193);  maximum_193 = None
	        div_386: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(sub_8827, 255.0);  sub_8827 = None
	        clamp_min_579: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(div_386, 1.1920928955078125e-07);  div_386 = None
	        div_387: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(minimum_193, clamp_min_579);  minimum_193 = None
	        round_387: "f32[375*s6][1]cuda:0" = torch.ops.aten.round.default(div_387);  div_387 = None
	        sub_8833: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_387);  round_387 = None
	        clamp_min_580: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8833, -128);  sub_8833 = None
	        clamp_max_386: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_580, 127);  clamp_min_580 = None
	        _assert_tensor_metadata_1739 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_579, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1739 = None
	        _assert_tensor_metadata_1740 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_386, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1740 = None
	        convert_element_type_1158: "i8[375*s6][1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_386, torch.int8);  clamp_max_386 = None
	        view_3023: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.view.default(mul_18644, [sym_size_int_193, 3072]);  mul_18644 = None
	        view_3024: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_579, [sym_size_int_193, 1])
	        view_3025: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1158, [sym_size_int_193, 1])
	        reciprocal_193: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_3024);  view_3024 = None
	        mul_18666: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_193, 1.0);  reciprocal_193 = None
	        mul_18668: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_3023, mul_18666);  view_3023 = mul_18666 = None
	        round_388: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.round.default(mul_18668);  mul_18668 = None
	        add_29572: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.add.Tensor(round_388, view_3025);  round_388 = view_3025 = None
	        clamp_min_581: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29572, -128);  add_29572 = None
	        clamp_max_387: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_581, 127);  clamp_min_581 = None
	        view_3026: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_387, [sym_size_int_193, 3072]);  clamp_max_387 = None
	        _assert_tensor_metadata_1741 = torch.ops.aten._assert_tensor_metadata.default(view_3026, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1741 = None
	        convert_element_type_1159: "i8[375*s6, 3072][3072, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3026, torch.int8);  view_3026 = None
	        view_3027: "i8[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1159, [sym_size_int_193, 3072]);  convert_element_type_1159 = None
	        view_3028: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_579, [sym_size_int_193, 1]);  clamp_min_579 = None
	        view_3029: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1158, [sym_size_int_193, 1]);  convert_element_type_1158 = None
	        _assert_tensor_metadata_1742 = torch.ops.aten._assert_tensor_metadata.default(view_3027, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1742 = None
	        convert_element_type_1160: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3027, torch.float32);  view_3027 = None
	        _assert_tensor_metadata_1743 = torch.ops.aten._assert_tensor_metadata.default(view_3029, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1743 = None
	        convert_element_type_1161: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3029, torch.float32);  view_3029 = None
	        sub_8853: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1160, convert_element_type_1161);  convert_element_type_1160 = convert_element_type_1161 = None
	        mul_18687: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8853, view_3028);  sub_8853 = view_3028 = None
	        view_3030: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.view.default(mul_18687, [sym_size_int_193, 3072]);  mul_18687 = sym_size_int_193 = None
	        _assert_tensor_metadata_1744 = torch.ops.aten._assert_tensor_metadata.default(view_3030, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1744 = None
	        view_3031: "i8[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.aten.view.default(arg874_1, [3072, 96, 32]);  arg874_1 = None
	        view_3032: "f32[3072, 96, 1][96, 1, 1]cuda:0" = torch.ops.aten.view.default(arg875_1, [3072, 96, 1]);  arg875_1 = None
	        view_3033: "i8[3072, 96, 1][96, 1, 1]cuda:0" = torch.ops.aten.view.default(arg876_1, [3072, 96, 1]);  arg876_1 = None
	        _assert_tensor_metadata_1745 = torch.ops.aten._assert_tensor_metadata.default(view_3031, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1745 = None
	        convert_element_type_1162: "f32[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3031, torch.float32);  view_3031 = None
	        _assert_tensor_metadata_1746 = torch.ops.aten._assert_tensor_metadata.default(view_3033, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1746 = None
	        convert_element_type_1163: "f32[3072, 96, 1][96, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3033, torch.float32);  view_3033 = None
	        sub_8857: "f32[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1162, convert_element_type_1163);  convert_element_type_1162 = convert_element_type_1163 = None
	        mul_18692: "f32[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8857, view_3032);  sub_8857 = view_3032 = None
	        view_3034: "f32[3072, 3072][3072, 1]cuda:0" = torch.ops.aten.view.default(mul_18692, [3072, 3072]);  mul_18692 = None
	        _assert_tensor_metadata_1747 = torch.ops.aten._assert_tensor_metadata.default(view_3034, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1747 = None
	        permute_322: "f32[3072, 3072][1, 3072]cuda:0" = torch.ops.aten.permute.default(view_3034, [1, 0]);  view_3034 = None
	        mm_33: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mm.default(view_3030, permute_322);  view_3030 = permute_322 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py:83 in forward, code: return audio_embeds.unsqueeze(0)
	        unsqueeze: "f32[1, 375*s6, 3072][1152000*s6, 3072, 1]cuda:0" = torch.ops.aten.unsqueeze.default(mm_33, 0);  mm_33 = None
	        return (
	            unsqueeze,  # PlainAOTOutput(idx=0)
	        )
	        
V0910 09:42:38.262000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "aadf8291ef4c0837de303c0672d83e74"}
	{
	"name": "create_aot_dispatcher_function",
	"ts": 1757522558262103.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:38.451000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d9507a27626930ba243b21354a0ff6b5"}
	{
	"name": "compile_fx.<locals>.fw_compiler_base",
	"ts": 1757522558451050.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:38.453000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_functorch/_aot_autograd/schemas.py", 33]}
V0910 09:42:39.070000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_inductor/compile_fx.py:2185] {"artifact": {"name": "before_joint_graph", "encoding": "string"}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2657, "name": "compile_fx", "filename": 19, "loc": "return inference_compiler(unlifted_gm, example_inputs_)"}, {"line": 1267, "name": "__call__", "filename": 33, "loc": "return self.compiler_fn(gm, example_inputs)"}, {"line": 2543, "name": "fw_compiler_base", "filename": 19, "loc": "return compile_fx_forward("}, {"line": 2185, "name": "compile_fx_forward", "filename": 19, "loc": "trace_structured("}], "has_payload": "74a0c83ed2ef9de26803dfb9dc7933c9"}
	class <lambda>(torch.nn.Module):
	    def forward(self):
	        arg877_1: "f32[s6, 128, 3000][384000, 3000, 1]cuda:0"; 
	    
	        arg877_1, = fx_pytree.tree_flatten_spec([], self._in_spec)
	        # No stacktrace found for following nodes
	        model_audio_tower_embed_positions_weight: "f32[1500, 1280][1280, 1]cuda:0" = self.model.audio_tower.embed_positions.weight
	        model_audio_tower_conv1_weight: "f32[1280, 128, 3][384, 3, 1]cuda:0" = self.model.audio_tower.conv1.weight
	        model_audio_tower_conv1_bias: "f32[1280][1]cuda:0" = self.model.audio_tower.conv1.bias
	        model_audio_tower_conv2_weight: "f32[1280, 1280, 3][3840, 3, 1]cuda:0" = self.model.audio_tower.conv2.weight
	        model_audio_tower_conv2_bias: "f32[1280][1]cuda:0" = self.model.audio_tower.conv2.bias
	        model_audio_tower_layers_0_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn_layer_norm.weight
	        model_audio_tower_layers_0_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn_layer_norm.bias
	        model_audio_tower_layers_0_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.bias
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.bias
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.bias
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").final_layer_norm.weight
	        model_audio_tower_layers_0_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").final_layer_norm.bias
	        model_audio_tower_layers_0_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.bias
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_0_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.bias
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn_layer_norm.weight
	        model_audio_tower_layers_1_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn_layer_norm.bias
	        model_audio_tower_layers_1_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.bias
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.bias
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.bias
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").final_layer_norm.weight
	        model_audio_tower_layers_1_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").final_layer_norm.bias
	        model_audio_tower_layers_1_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.bias
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_1_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.bias
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn_layer_norm.weight
	        model_audio_tower_layers_2_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn_layer_norm.bias
	        model_audio_tower_layers_2_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.bias
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.bias
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.bias
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").final_layer_norm.weight
	        model_audio_tower_layers_2_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").final_layer_norm.bias
	        model_audio_tower_layers_2_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.bias
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_2_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.bias
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn_layer_norm.weight
	        model_audio_tower_layers_3_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn_layer_norm.bias
	        model_audio_tower_layers_3_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.bias
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.bias
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.bias
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").final_layer_norm.weight
	        model_audio_tower_layers_3_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").final_layer_norm.bias
	        model_audio_tower_layers_3_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.bias
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_3_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.bias
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn_layer_norm.weight
	        model_audio_tower_layers_4_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn_layer_norm.bias
	        model_audio_tower_layers_4_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.bias
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.bias
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.bias
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").final_layer_norm.weight
	        model_audio_tower_layers_4_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").final_layer_norm.bias
	        model_audio_tower_layers_4_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.bias
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_4_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.bias
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn_layer_norm.weight
	        model_audio_tower_layers_5_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn_layer_norm.bias
	        model_audio_tower_layers_5_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.bias
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.bias
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.bias
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").final_layer_norm.weight
	        model_audio_tower_layers_5_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").final_layer_norm.bias
	        model_audio_tower_layers_5_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.bias
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_5_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.bias
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn_layer_norm.weight
	        model_audio_tower_layers_6_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn_layer_norm.bias
	        model_audio_tower_layers_6_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.bias
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.bias
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.bias
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").final_layer_norm.weight
	        model_audio_tower_layers_6_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").final_layer_norm.bias
	        model_audio_tower_layers_6_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.bias
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_6_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.bias
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn_layer_norm.weight
	        model_audio_tower_layers_7_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn_layer_norm.bias
	        model_audio_tower_layers_7_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.bias
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.bias
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.bias
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").final_layer_norm.weight
	        model_audio_tower_layers_7_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").final_layer_norm.bias
	        model_audio_tower_layers_7_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.bias
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_7_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.bias
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn_layer_norm.weight
	        model_audio_tower_layers_8_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn_layer_norm.bias
	        model_audio_tower_layers_8_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.bias
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.bias
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.bias
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").final_layer_norm.weight
	        model_audio_tower_layers_8_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").final_layer_norm.bias
	        model_audio_tower_layers_8_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.bias
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_8_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.bias
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn_layer_norm.weight
	        model_audio_tower_layers_9_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn_layer_norm.bias
	        model_audio_tower_layers_9_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.bias
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.bias
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.bias
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").final_layer_norm.weight
	        model_audio_tower_layers_9_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").final_layer_norm.bias
	        model_audio_tower_layers_9_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.bias
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_9_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.bias
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn_layer_norm.weight
	        model_audio_tower_layers_10_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn_layer_norm.bias
	        model_audio_tower_layers_10_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.bias
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.bias
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.bias
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").final_layer_norm.weight
	        model_audio_tower_layers_10_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").final_layer_norm.bias
	        model_audio_tower_layers_10_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.bias
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_10_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.bias
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn_layer_norm.weight
	        model_audio_tower_layers_11_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn_layer_norm.bias
	        model_audio_tower_layers_11_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.bias
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.bias
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.bias
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").final_layer_norm.weight
	        model_audio_tower_layers_11_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").final_layer_norm.bias
	        model_audio_tower_layers_11_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.bias
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_11_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.bias
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn_layer_norm.weight
	        model_audio_tower_layers_12_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn_layer_norm.bias
	        model_audio_tower_layers_12_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.bias
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.bias
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.bias
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").final_layer_norm.weight
	        model_audio_tower_layers_12_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").final_layer_norm.bias
	        model_audio_tower_layers_12_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.bias
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_12_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.bias
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn_layer_norm.weight
	        model_audio_tower_layers_13_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn_layer_norm.bias
	        model_audio_tower_layers_13_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.bias
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.bias
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.bias
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").final_layer_norm.weight
	        model_audio_tower_layers_13_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").final_layer_norm.bias
	        model_audio_tower_layers_13_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.bias
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_13_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.bias
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn_layer_norm.weight
	        model_audio_tower_layers_14_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn_layer_norm.bias
	        model_audio_tower_layers_14_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.bias
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.bias
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.bias
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").final_layer_norm.weight
	        model_audio_tower_layers_14_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").final_layer_norm.bias
	        model_audio_tower_layers_14_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.bias
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_14_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.bias
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn_layer_norm.weight
	        model_audio_tower_layers_15_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn_layer_norm.bias
	        model_audio_tower_layers_15_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.bias
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.bias
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.bias
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").final_layer_norm.weight
	        model_audio_tower_layers_15_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").final_layer_norm.bias
	        model_audio_tower_layers_15_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.bias
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_15_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.bias
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn_layer_norm.weight
	        model_audio_tower_layers_16_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn_layer_norm.bias
	        model_audio_tower_layers_16_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.bias
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.bias
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.bias
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").final_layer_norm.weight
	        model_audio_tower_layers_16_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").final_layer_norm.bias
	        model_audio_tower_layers_16_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.bias
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_16_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.bias
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn_layer_norm.weight
	        model_audio_tower_layers_17_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn_layer_norm.bias
	        model_audio_tower_layers_17_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.bias
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.bias
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.bias
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").final_layer_norm.weight
	        model_audio_tower_layers_17_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").final_layer_norm.bias
	        model_audio_tower_layers_17_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.bias
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_17_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.bias
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn_layer_norm.weight
	        model_audio_tower_layers_18_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn_layer_norm.bias
	        model_audio_tower_layers_18_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.bias
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.bias
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.bias
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").final_layer_norm.weight
	        model_audio_tower_layers_18_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").final_layer_norm.bias
	        model_audio_tower_layers_18_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.bias
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_18_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.bias
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn_layer_norm.weight
	        model_audio_tower_layers_19_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn_layer_norm.bias
	        model_audio_tower_layers_19_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.bias
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.bias
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.bias
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").final_layer_norm.weight
	        model_audio_tower_layers_19_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").final_layer_norm.bias
	        model_audio_tower_layers_19_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.bias
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_19_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.bias
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn_layer_norm.weight
	        model_audio_tower_layers_20_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn_layer_norm.bias
	        model_audio_tower_layers_20_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.bias
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.bias
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.bias
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").final_layer_norm.weight
	        model_audio_tower_layers_20_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").final_layer_norm.bias
	        model_audio_tower_layers_20_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.bias
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_20_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.bias
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn_layer_norm.weight
	        model_audio_tower_layers_21_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn_layer_norm.bias
	        model_audio_tower_layers_21_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.bias
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.bias
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.bias
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").final_layer_norm.weight
	        model_audio_tower_layers_21_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").final_layer_norm.bias
	        model_audio_tower_layers_21_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.bias
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_21_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.bias
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn_layer_norm.weight
	        model_audio_tower_layers_22_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn_layer_norm.bias
	        model_audio_tower_layers_22_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.bias
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.bias
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.bias
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").final_layer_norm.weight
	        model_audio_tower_layers_22_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").final_layer_norm.bias
	        model_audio_tower_layers_22_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.bias
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_22_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.bias
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn_layer_norm.weight
	        model_audio_tower_layers_23_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn_layer_norm.bias
	        model_audio_tower_layers_23_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.bias
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.bias
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.bias
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").final_layer_norm.weight
	        model_audio_tower_layers_23_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").final_layer_norm.bias
	        model_audio_tower_layers_23_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.bias
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_23_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.bias
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn_layer_norm.weight
	        model_audio_tower_layers_24_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn_layer_norm.bias
	        model_audio_tower_layers_24_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.bias
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.bias
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.bias
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").final_layer_norm.weight
	        model_audio_tower_layers_24_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").final_layer_norm.bias
	        model_audio_tower_layers_24_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.bias
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_24_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.bias
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn_layer_norm.weight
	        model_audio_tower_layers_25_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn_layer_norm.bias
	        model_audio_tower_layers_25_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.bias
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.bias
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.bias
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").final_layer_norm.weight
	        model_audio_tower_layers_25_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").final_layer_norm.bias
	        model_audio_tower_layers_25_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.bias
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_25_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.bias
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn_layer_norm.weight
	        model_audio_tower_layers_26_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn_layer_norm.bias
	        model_audio_tower_layers_26_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.bias
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.bias
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.bias
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").final_layer_norm.weight
	        model_audio_tower_layers_26_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").final_layer_norm.bias
	        model_audio_tower_layers_26_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.bias
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_26_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.bias
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn_layer_norm.weight
	        model_audio_tower_layers_27_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn_layer_norm.bias
	        model_audio_tower_layers_27_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.bias
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.bias
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.bias
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").final_layer_norm.weight
	        model_audio_tower_layers_27_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").final_layer_norm.bias
	        model_audio_tower_layers_27_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.bias
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_27_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.bias
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn_layer_norm.weight
	        model_audio_tower_layers_28_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn_layer_norm.bias
	        model_audio_tower_layers_28_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.bias
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.bias
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.bias
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").final_layer_norm.weight
	        model_audio_tower_layers_28_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").final_layer_norm.bias
	        model_audio_tower_layers_28_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.bias
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_28_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.bias
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn_layer_norm.weight
	        model_audio_tower_layers_29_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn_layer_norm.bias
	        model_audio_tower_layers_29_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.bias
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.bias
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.bias
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").final_layer_norm.weight
	        model_audio_tower_layers_29_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").final_layer_norm.bias
	        model_audio_tower_layers_29_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.bias
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_29_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.bias
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn_layer_norm.weight
	        model_audio_tower_layers_30_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn_layer_norm.bias
	        model_audio_tower_layers_30_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.bias
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.bias
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.bias
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").final_layer_norm.weight
	        model_audio_tower_layers_30_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").final_layer_norm.bias
	        model_audio_tower_layers_30_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.bias
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_30_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.bias
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn_layer_norm.weight
	        model_audio_tower_layers_31_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn_layer_norm.bias
	        model_audio_tower_layers_31_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.bias
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.bias
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.bias
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").final_layer_norm.weight
	        model_audio_tower_layers_31_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").final_layer_norm.bias
	        model_audio_tower_layers_31_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.bias
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_31_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.bias
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original2
	        model_audio_tower_layer_norm_weight: "f32[1280][1]cuda:0" = self.model.audio_tower.layer_norm.weight
	        model_audio_tower_layer_norm_bias: "f32[1280][1]cuda:0" = self.model.audio_tower.layer_norm.bias
	        model_multi_modal_projector_linear_1_parametrizations_weight_original0: "i8[3072, 5120][5120, 1]cuda:0" = self.model.multi_modal_projector.linear_1.parametrizations.weight.original0
	        model_multi_modal_projector_linear_1_parametrizations_weight_original1: "f32[3072, 160][160, 1]cuda:0" = self.model.multi_modal_projector.linear_1.parametrizations.weight.original1
	        model_multi_modal_projector_linear_1_parametrizations_weight_original2: "i8[3072, 160][160, 1]cuda:0" = self.model.multi_modal_projector.linear_1.parametrizations.weight.original2
	        model_multi_modal_projector_linear_2_parametrizations_weight_original0: "i8[3072, 3072][3072, 1]cuda:0" = self.model.multi_modal_projector.linear_2.parametrizations.weight.original0
	        model_multi_modal_projector_linear_2_parametrizations_weight_original1: "f32[3072, 96][96, 1]cuda:0" = self.model.multi_modal_projector.linear_2.parametrizations.weight.original1
	        model_multi_modal_projector_linear_2_parametrizations_weight_original2: "i8[3072, 96][96, 1]cuda:0" = self.model.multi_modal_projector.linear_2.parametrizations.weight.original2
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:348 in forward, code: input_features = input_features.to(dtype=self.conv1.weight.dtype, device=self.conv1.weight.device)
	        _assert_tensor_metadata = torch.ops.aten._assert_tensor_metadata.default(arg877_1, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:349 in forward, code: inputs_embeds = nn.functional.gelu(self.conv1(input_features))
	        convolution: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.convolution.default(arg877_1, model_audio_tower_conv1_weight, model_audio_tower_conv1_bias, [1], [1], [1], False, [0], 1);  model_audio_tower_conv1_weight = model_audio_tower_conv1_bias = None
	        sym_size_int: "Sym(s6)" = torch.ops.aten.sym_size.int(arg877_1, 0);  arg877_1 = None
	        mul_2: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution, 0.5)
	        mul_3: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution, 0.7071067811865476);  convolution = None
	        erf: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.erf.default(mul_3);  mul_3 = None
	        add_4: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.add.Tensor(erf, 1);  erf = None
	        mul_4: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2, add_4);  mul_2 = add_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:350 in forward, code: inputs_embeds = nn.functional.gelu(self.conv2(inputs_embeds))
	        convolution_1: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.convolution.default(mul_4, model_audio_tower_conv2_weight, model_audio_tower_conv2_bias, [2], [1], [1], False, [0], 1);  mul_4 = model_audio_tower_conv2_weight = model_audio_tower_conv2_bias = None
	        mul_9: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution_1, 0.5)
	        mul_10: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution_1, 0.7071067811865476);  convolution_1 = None
	        erf_1: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.erf.default(mul_10);  mul_10 = None
	        add_13: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_1, 1);  erf_1 = None
	        mul_11: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9, add_13);  mul_9 = add_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:351 in forward, code: inputs_embeds = inputs_embeds.permute(0, 2, 1)
	        permute: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.permute.default(mul_11, [0, 2, 1]);  mul_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:354 in forward, code: hidden_states = (inputs_embeds + embed_pos).to(inputs_embeds.dtype)
	        add_22: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(permute, model_audio_tower_embed_positions_weight);  permute = model_audio_tower_embed_positions_weight = None
	        _assert_tensor_metadata_1 = torch.ops.aten._assert_tensor_metadata.default(add_22, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:355 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.clone.default(add_22);  add_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(clone, memory_format = torch.contiguous_format)
	        var_mean = torch.ops.aten.var_mean.correction(clone_1, [2], correction = 0, keepdim = True)
	        getitem: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean[0]
	        getitem_1: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean[1];  var_mean = None
	        add_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem, 1e-05);  getitem = None
	        rsqrt: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_31);  add_31 = None
	        sub_7: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_1, getitem_1);  clone_1 = getitem_1 = None
	        mul_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7, rsqrt);  sub_7 = rsqrt = None
	        mul_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_20, model_audio_tower_layers_0_self_attn_layer_norm_weight);  mul_20 = model_audio_tower_layers_0_self_attn_layer_norm_weight = None
	        add_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_21, model_audio_tower_layers_0_self_attn_layer_norm_bias);  mul_21 = model_audio_tower_layers_0_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_32, [sym_size_int, 1500, 1280])
	        amin: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view, [2])
	        amax: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view, [2]);  view = None
	        full: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin, full);  amin = full = None
	        full_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax, full_1);  amax = full_1 = None
	        sub_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum, minimum);  maximum = None
	        div: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_18, 255.0);  sub_18 = None
	        clamp_min: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div, 1.1920928955078125e-07);  div = None
	        div_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum, clamp_min);  minimum = None
	        round_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_1);  div_1 = None
	        sub_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_1);  round_1 = None
	        clamp_min_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_24, -128);  sub_24 = None
	        clamp_max: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_1, 127);  clamp_min_1 = None
	        _assert_tensor_metadata_2 = torch.ops.aten._assert_tensor_metadata.default(clamp_min, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_2 = None
	        _assert_tensor_metadata_3 = torch.ops.aten._assert_tensor_metadata.default(clamp_max, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_3 = None
	        convert_element_type: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max, torch.int8);  clamp_max = None
	        view_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_32, [sym_size_int, 1500, 1280])
	        view_2: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min, [sym_size_int, 1500, 1])
	        view_3: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type, [sym_size_int, 1500, 1])
	        reciprocal: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2);  view_2 = None
	        mul_69: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal, 1.0);  reciprocal = None
	        mul_72: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1, mul_69);  view_1 = mul_69 = None
	        round_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_72);  mul_72 = None
	        add_119: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_2, view_3);  round_2 = view_3 = None
	        clamp_min_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_119, -128);  add_119 = None
	        clamp_max_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_2, 127);  clamp_min_2 = None
	        view_4: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_1, [sym_size_int, 1500, 1280]);  clamp_max_1 = None
	        _assert_tensor_metadata_4 = torch.ops.aten._assert_tensor_metadata.default(view_4, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_4 = None
	        convert_element_type_1: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_4, torch.int8);  view_4 = None
	        view_5: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1, [sym_size_int, 1500, 1280]);  convert_element_type_1 = None
	        view_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min, [sym_size_int, 1500, 1]);  clamp_min = None
	        view_7: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type, [sym_size_int, 1500, 1]);  convert_element_type = None
	        _assert_tensor_metadata_5 = torch.ops.aten._assert_tensor_metadata.default(view_5, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_5 = None
	        convert_element_type_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_5, torch.float32);  view_5 = None
	        _assert_tensor_metadata_6 = torch.ops.aten._assert_tensor_metadata.default(view_7, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_6 = None
	        convert_element_type_3: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_7, torch.float32);  view_7 = None
	        sub_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_2, convert_element_type_3);  convert_element_type_2 = convert_element_type_3 = None
	        mul_94: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_44, view_6);  sub_44 = view_6 = None
	        view_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_94, [sym_size_int, 1500, 1280]);  mul_94 = None
	        _assert_tensor_metadata_7 = torch.ops.aten._assert_tensor_metadata.default(view_8, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_7 = None
	        view_9: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_10: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_11: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_8 = torch.ops.aten._assert_tensor_metadata.default(view_9, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_8 = None
	        convert_element_type_4: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_9, torch.float32);  view_9 = None
	        _assert_tensor_metadata_9 = torch.ops.aten._assert_tensor_metadata.default(view_11, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_9 = None
	        convert_element_type_5: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_11, torch.float32);  view_11 = None
	        sub_48: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_4, convert_element_type_5);  convert_element_type_4 = convert_element_type_5 = None
	        mul_99: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_48, view_10);  sub_48 = view_10 = None
	        view_12: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_99, [1280, 1280]);  mul_99 = None
	        _assert_tensor_metadata_10 = torch.ops.aten._assert_tensor_metadata.default(view_12, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_10 = None
	        mul_104: "Sym(1500*s6)" = sym_size_int * 1500
	        view_13: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_8, [mul_104, 1280]);  view_8 = mul_104 = None
	        permute_1: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_12, [1, 0]);  view_12 = None
	        addmm: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_0_self_attn_q_proj_bias, view_13, permute_1);  model_audio_tower_layers_0_self_attn_q_proj_bias = view_13 = permute_1 = None
	        view_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm, [sym_size_int, 1500, 1280]);  addmm = None
	        mul_111: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_14, 0.125);  view_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_15: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_111, [sym_size_int, 1500, 20, 64]);  mul_111 = None
	        permute_2: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_15, [0, 2, 1, 3]);  view_15 = None
	        clone_2: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_2, memory_format = torch.contiguous_format);  permute_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_16: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_32, [sym_size_int, 1500, 1280])
	        amin_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_16, [2])
	        amax_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_16, [2]);  view_16 = None
	        full_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_1, full_2);  amin_1 = full_2 = None
	        full_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_1, full_3);  amax_1 = full_3 = None
	        sub_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_1, minimum_1);  maximum_1 = None
	        div_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_63, 255.0);  sub_63 = None
	        clamp_min_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_2, 1.1920928955078125e-07);  div_2 = None
	        div_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_1, clamp_min_3);  minimum_1 = None
	        round_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_3);  div_3 = None
	        sub_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_3);  round_3 = None
	        clamp_min_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_69, -128);  sub_69 = None
	        clamp_max_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_4, 127);  clamp_min_4 = None
	        _assert_tensor_metadata_11 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_3, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_11 = None
	        _assert_tensor_metadata_12 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_2, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_12 = None
	        convert_element_type_6: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_2, torch.int8);  clamp_max_2 = None
	        view_17: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_32, [sym_size_int, 1500, 1280])
	        view_18: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_3, [sym_size_int, 1500, 1])
	        view_19: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_6, [sym_size_int, 1500, 1])
	        reciprocal_1: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_18);  view_18 = None
	        mul_165: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_1, 1.0);  reciprocal_1 = None
	        mul_168: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_17, mul_165);  view_17 = mul_165 = None
	        round_4: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_168);  mul_168 = None
	        add_271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_4, view_19);  round_4 = view_19 = None
	        clamp_min_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_271, -128);  add_271 = None
	        clamp_max_3: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_5, 127);  clamp_min_5 = None
	        view_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_3, [sym_size_int, 1500, 1280]);  clamp_max_3 = None
	        _assert_tensor_metadata_13 = torch.ops.aten._assert_tensor_metadata.default(view_20, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_13 = None
	        convert_element_type_7: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_20, torch.int8);  view_20 = None
	        view_21: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_7, [sym_size_int, 1500, 1280]);  convert_element_type_7 = None
	        view_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_3, [sym_size_int, 1500, 1]);  clamp_min_3 = None
	        view_23: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_6, [sym_size_int, 1500, 1]);  convert_element_type_6 = None
	        _assert_tensor_metadata_14 = torch.ops.aten._assert_tensor_metadata.default(view_21, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_14 = None
	        convert_element_type_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_21, torch.float32);  view_21 = None
	        _assert_tensor_metadata_15 = torch.ops.aten._assert_tensor_metadata.default(view_23, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_15 = None
	        convert_element_type_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_23, torch.float32);  view_23 = None
	        sub_89: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_8, convert_element_type_9);  convert_element_type_8 = convert_element_type_9 = None
	        mul_190: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_89, view_22);  sub_89 = view_22 = None
	        view_24: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_190, [sym_size_int, 1500, 1280]);  mul_190 = None
	        _assert_tensor_metadata_16 = torch.ops.aten._assert_tensor_metadata.default(view_24, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_16 = None
	        view_25: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_26: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_27: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_17 = torch.ops.aten._assert_tensor_metadata.default(view_25, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_17 = None
	        convert_element_type_10: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_25, torch.float32);  view_25 = None
	        _assert_tensor_metadata_18 = torch.ops.aten._assert_tensor_metadata.default(view_27, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_18 = None
	        convert_element_type_11: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_27, torch.float32);  view_27 = None
	        sub_93: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_10, convert_element_type_11);  convert_element_type_10 = convert_element_type_11 = None
	        mul_195: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_93, view_26);  sub_93 = view_26 = None
	        view_28: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_195, [1280, 1280]);  mul_195 = None
	        _assert_tensor_metadata_19 = torch.ops.aten._assert_tensor_metadata.default(view_28, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_19 = None
	        permute_3: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_28, [1, 0]);  view_28 = None
	        mul_198: "Sym(1500*s6)" = sym_size_int * 1500
	        view_29: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_24, [mul_198, 1280]);  view_24 = mul_198 = None
	        mm: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_29, permute_3);  view_29 = permute_3 = None
	        view_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm, [sym_size_int, 1500, 1280]);  mm = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_31: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_30, [sym_size_int, -1, 20, 64]);  view_30 = None
	        permute_4: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_31, [0, 2, 1, 3]);  view_31 = None
	        clone_3: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_4, memory_format = torch.contiguous_format);  permute_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_32, [sym_size_int, 1500, 1280])
	        amin_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_32, [2])
	        amax_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_32, [2]);  view_32 = None
	        full_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_2, full_4);  amin_2 = full_4 = None
	        full_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_2, full_5);  amax_2 = full_5 = None
	        sub_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_2, minimum_2);  maximum_2 = None
	        div_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_107, 255.0);  sub_107 = None
	        clamp_min_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_4, 1.1920928955078125e-07);  div_4 = None
	        div_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_2, clamp_min_6);  minimum_2 = None
	        round_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_5);  div_5 = None
	        sub_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_5);  round_5 = None
	        clamp_min_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_113, -128);  sub_113 = None
	        clamp_max_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_7, 127);  clamp_min_7 = None
	        _assert_tensor_metadata_20 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_6, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_20 = None
	        _assert_tensor_metadata_21 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_4, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_21 = None
	        convert_element_type_12: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_4, torch.int8);  clamp_max_4 = None
	        view_33: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_32, [sym_size_int, 1500, 1280]);  add_32 = None
	        view_34: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_6, [sym_size_int, 1500, 1])
	        view_35: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_12, [sym_size_int, 1500, 1])
	        reciprocal_2: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_34);  view_34 = None
	        mul_264: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_2, 1.0);  reciprocal_2 = None
	        mul_267: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_33, mul_264);  view_33 = mul_264 = None
	        round_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_267);  mul_267 = None
	        add_419: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_6, view_35);  round_6 = view_35 = None
	        clamp_min_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_419, -128);  add_419 = None
	        clamp_max_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_8, 127);  clamp_min_8 = None
	        view_36: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_5, [sym_size_int, 1500, 1280]);  clamp_max_5 = None
	        _assert_tensor_metadata_22 = torch.ops.aten._assert_tensor_metadata.default(view_36, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_22 = None
	        convert_element_type_13: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_36, torch.int8);  view_36 = None
	        view_37: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_13, [sym_size_int, 1500, 1280]);  convert_element_type_13 = None
	        view_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_6, [sym_size_int, 1500, 1]);  clamp_min_6 = None
	        view_39: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_12, [sym_size_int, 1500, 1]);  convert_element_type_12 = None
	        _assert_tensor_metadata_23 = torch.ops.aten._assert_tensor_metadata.default(view_37, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_23 = None
	        convert_element_type_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_37, torch.float32);  view_37 = None
	        _assert_tensor_metadata_24 = torch.ops.aten._assert_tensor_metadata.default(view_39, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_24 = None
	        convert_element_type_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_39, torch.float32);  view_39 = None
	        sub_133: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_14, convert_element_type_15);  convert_element_type_14 = convert_element_type_15 = None
	        mul_289: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_133, view_38);  sub_133 = view_38 = None
	        view_40: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_289, [sym_size_int, 1500, 1280]);  mul_289 = None
	        _assert_tensor_metadata_25 = torch.ops.aten._assert_tensor_metadata.default(view_40, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_25 = None
	        view_41: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_42: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_43: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_26 = torch.ops.aten._assert_tensor_metadata.default(view_41, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_26 = None
	        convert_element_type_16: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_41, torch.float32);  view_41 = None
	        _assert_tensor_metadata_27 = torch.ops.aten._assert_tensor_metadata.default(view_43, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_27 = None
	        convert_element_type_17: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_43, torch.float32);  view_43 = None
	        sub_137: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_16, convert_element_type_17);  convert_element_type_16 = convert_element_type_17 = None
	        mul_294: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_137, view_42);  sub_137 = view_42 = None
	        view_44: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_294, [1280, 1280]);  mul_294 = None
	        _assert_tensor_metadata_28 = torch.ops.aten._assert_tensor_metadata.default(view_44, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_28 = None
	        mul_299: "Sym(1500*s6)" = sym_size_int * 1500
	        view_45: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_40, [mul_299, 1280]);  view_40 = mul_299 = None
	        permute_5: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_44, [1, 0]);  view_44 = None
	        addmm_1: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_0_self_attn_v_proj_bias, view_45, permute_5);  model_audio_tower_layers_0_self_attn_v_proj_bias = view_45 = permute_5 = None
	        view_46: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_1, [sym_size_int, 1500, 1280]);  addmm_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_47: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_46, [sym_size_int, -1, 20, 64]);  view_46 = None
	        permute_6: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_47, [0, 2, 1, 3]);  view_47 = None
	        clone_4: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_6, memory_format = torch.contiguous_format);  permute_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_2, clone_3, clone_4, None, False, scale = 1.0);  clone_2 = clone_3 = clone_4 = None
	        getitem_2: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention[0];  _scaled_dot_product_efficient_attention = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_7: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_2, [0, 2, 1, 3]);  getitem_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_48: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_7, [sym_size_int, 1500, -1]);  permute_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_49: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_48, [sym_size_int, 1500, 1280])
	        amin_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_49, [2])
	        amax_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_49, [2]);  view_49 = None
	        full_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_3, full_6);  amin_3 = full_6 = None
	        full_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_3, full_7);  amax_3 = full_7 = None
	        sub_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_3, minimum_3);  maximum_3 = None
	        div_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_155, 255.0);  sub_155 = None
	        clamp_min_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_6, 1.1920928955078125e-07);  div_6 = None
	        div_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_3, clamp_min_9);  minimum_3 = None
	        round_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_7);  div_7 = None
	        sub_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_7);  round_7 = None
	        clamp_min_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_161, -128);  sub_161 = None
	        clamp_max_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_10, 127);  clamp_min_10 = None
	        _assert_tensor_metadata_29 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_9, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_29 = None
	        _assert_tensor_metadata_30 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_6, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_30 = None
	        convert_element_type_18: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_6, torch.int8);  clamp_max_6 = None
	        view_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_48, [sym_size_int, 1500, 1280]);  view_48 = None
	        view_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_9, [sym_size_int, 1500, 1])
	        view_52: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_18, [sym_size_int, 1500, 1])
	        reciprocal_3: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_51);  view_51 = None
	        mul_369: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_3, 1.0);  reciprocal_3 = None
	        mul_372: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_50, mul_369);  view_50 = mul_369 = None
	        round_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_372);  mul_372 = None
	        add_583: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_8, view_52);  round_8 = view_52 = None
	        clamp_min_11: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_583, -128);  add_583 = None
	        clamp_max_7: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_11, 127);  clamp_min_11 = None
	        view_53: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_7, [sym_size_int, 1500, 1280]);  clamp_max_7 = None
	        _assert_tensor_metadata_31 = torch.ops.aten._assert_tensor_metadata.default(view_53, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_31 = None
	        convert_element_type_19: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_53, torch.int8);  view_53 = None
	        view_54: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_19, [sym_size_int, 1500, 1280]);  convert_element_type_19 = None
	        view_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_9, [sym_size_int, 1500, 1]);  clamp_min_9 = None
	        view_56: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_18, [sym_size_int, 1500, 1]);  convert_element_type_18 = None
	        _assert_tensor_metadata_32 = torch.ops.aten._assert_tensor_metadata.default(view_54, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_32 = None
	        convert_element_type_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_54, torch.float32);  view_54 = None
	        _assert_tensor_metadata_33 = torch.ops.aten._assert_tensor_metadata.default(view_56, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_33 = None
	        convert_element_type_21: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_56, torch.float32);  view_56 = None
	        sub_181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_20, convert_element_type_21);  convert_element_type_20 = convert_element_type_21 = None
	        mul_394: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_181, view_55);  sub_181 = view_55 = None
	        view_57: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_394, [sym_size_int, 1500, 1280]);  mul_394 = None
	        _assert_tensor_metadata_34 = torch.ops.aten._assert_tensor_metadata.default(view_57, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_34 = None
	        view_58: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_59: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_60: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_35 = torch.ops.aten._assert_tensor_metadata.default(view_58, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_35 = None
	        convert_element_type_22: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_58, torch.float32);  view_58 = None
	        _assert_tensor_metadata_36 = torch.ops.aten._assert_tensor_metadata.default(view_60, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_36 = None
	        convert_element_type_23: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_60, torch.float32);  view_60 = None
	        sub_185: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_22, convert_element_type_23);  convert_element_type_22 = convert_element_type_23 = None
	        mul_399: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_185, view_59);  sub_185 = view_59 = None
	        view_61: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_399, [1280, 1280]);  mul_399 = None
	        _assert_tensor_metadata_37 = torch.ops.aten._assert_tensor_metadata.default(view_61, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_37 = None
	        mul_404: "Sym(1500*s6)" = sym_size_int * 1500
	        view_62: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_57, [mul_404, 1280]);  view_57 = mul_404 = None
	        permute_8: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_61, [1, 0]);  view_61 = None
	        addmm_2: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_0_self_attn_out_proj_bias, view_62, permute_8);  model_audio_tower_layers_0_self_attn_out_proj_bias = view_62 = permute_8 = None
	        view_63: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_2, [sym_size_int, 1500, 1280]);  addmm_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_63);  view_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_646: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(clone, clone_5);  clone = clone_5 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_646, memory_format = torch.contiguous_format)
	        var_mean_1 = torch.ops.aten.var_mean.correction(clone_6, [2], correction = 0, keepdim = True)
	        getitem_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_1[0]
	        getitem_7: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_1[1];  var_mean_1 = None
	        add_651: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_6, 1e-05);  getitem_6 = None
	        rsqrt_1: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_651);  add_651 = None
	        sub_191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_6, getitem_7);  clone_6 = getitem_7 = None
	        mul_415: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_191, rsqrt_1);  sub_191 = rsqrt_1 = None
	        mul_416: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_415, model_audio_tower_layers_0_final_layer_norm_weight);  mul_415 = model_audio_tower_layers_0_final_layer_norm_weight = None
	        add_652: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_416, model_audio_tower_layers_0_final_layer_norm_bias);  mul_416 = model_audio_tower_layers_0_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_64: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_652, [sym_size_int, 1500, 1280])
	        amin_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_64, [2])
	        amax_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_64, [2]);  view_64 = None
	        full_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_4, full_8);  amin_4 = full_8 = None
	        full_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_4, full_9);  amax_4 = full_9 = None
	        sub_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_4, minimum_4);  maximum_4 = None
	        div_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_202, 255.0);  sub_202 = None
	        clamp_min_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_8, 1.1920928955078125e-07);  div_8 = None
	        div_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_4, clamp_min_12);  minimum_4 = None
	        round_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_9);  div_9 = None
	        sub_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_9);  round_9 = None
	        clamp_min_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_208, -128);  sub_208 = None
	        clamp_max_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_13, 127);  clamp_min_13 = None
	        _assert_tensor_metadata_38 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_12, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_38 = None
	        _assert_tensor_metadata_39 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_8, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_39 = None
	        convert_element_type_24: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_8, torch.int8);  clamp_max_8 = None
	        view_65: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_652, [sym_size_int, 1500, 1280]);  add_652 = None
	        view_66: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_12, [sym_size_int, 1500, 1])
	        view_67: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_24, [sym_size_int, 1500, 1])
	        reciprocal_4: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_66);  view_66 = None
	        mul_464: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_4, 1.0);  reciprocal_4 = None
	        mul_467: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_65, mul_464);  view_65 = mul_464 = None
	        round_10: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_467);  mul_467 = None
	        add_739: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_10, view_67);  round_10 = view_67 = None
	        clamp_min_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_739, -128);  add_739 = None
	        clamp_max_9: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_14, 127);  clamp_min_14 = None
	        view_68: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_9, [sym_size_int, 1500, 1280]);  clamp_max_9 = None
	        _assert_tensor_metadata_40 = torch.ops.aten._assert_tensor_metadata.default(view_68, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_40 = None
	        convert_element_type_25: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_68, torch.int8);  view_68 = None
	        view_69: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_25, [sym_size_int, 1500, 1280]);  convert_element_type_25 = None
	        view_70: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_12, [sym_size_int, 1500, 1]);  clamp_min_12 = None
	        view_71: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_24, [sym_size_int, 1500, 1]);  convert_element_type_24 = None
	        _assert_tensor_metadata_41 = torch.ops.aten._assert_tensor_metadata.default(view_69, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_41 = None
	        convert_element_type_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_69, torch.float32);  view_69 = None
	        _assert_tensor_metadata_42 = torch.ops.aten._assert_tensor_metadata.default(view_71, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_42 = None
	        convert_element_type_27: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_71, torch.float32);  view_71 = None
	        sub_228: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_26, convert_element_type_27);  convert_element_type_26 = convert_element_type_27 = None
	        mul_489: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_228, view_70);  sub_228 = view_70 = None
	        view_72: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_489, [sym_size_int, 1500, 1280]);  mul_489 = None
	        _assert_tensor_metadata_43 = torch.ops.aten._assert_tensor_metadata.default(view_72, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_43 = None
	        view_73: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_0_fc1_parametrizations_weight_original0 = None
	        view_74: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_0_fc1_parametrizations_weight_original1 = None
	        view_75: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_0_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_44 = torch.ops.aten._assert_tensor_metadata.default(view_73, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_44 = None
	        convert_element_type_28: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_73, torch.float32);  view_73 = None
	        _assert_tensor_metadata_45 = torch.ops.aten._assert_tensor_metadata.default(view_75, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_45 = None
	        convert_element_type_29: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_75, torch.float32);  view_75 = None
	        sub_232: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_28, convert_element_type_29);  convert_element_type_28 = convert_element_type_29 = None
	        mul_494: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_232, view_74);  sub_232 = view_74 = None
	        view_76: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_494, [5120, 1280]);  mul_494 = None
	        _assert_tensor_metadata_46 = torch.ops.aten._assert_tensor_metadata.default(view_76, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_46 = None
	        mul_499: "Sym(1500*s6)" = sym_size_int * 1500
	        view_77: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_72, [mul_499, 1280]);  view_72 = mul_499 = None
	        permute_9: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_76, [1, 0]);  view_76 = None
	        addmm_3: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_0_fc1_bias, view_77, permute_9);  model_audio_tower_layers_0_fc1_bias = view_77 = permute_9 = None
	        view_78: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_3, [sym_size_int, 1500, 5120]);  addmm_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_506: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_78, 0.5)
	        mul_507: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_78, 0.7071067811865476);  view_78 = None
	        erf_2: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_507);  mul_507 = None
	        add_798: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_2, 1);  erf_2 = None
	        mul_508: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_506, add_798);  mul_506 = add_798 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_7: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_508);  mul_508 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_79: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_7, [sym_size_int, 1500, 5120])
	        amin_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_79, [2])
	        amax_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_79, [2]);  view_79 = None
	        full_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_5, full_10);  amin_5 = full_10 = None
	        full_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_5, full_11);  amax_5 = full_11 = None
	        sub_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_5, minimum_5);  maximum_5 = None
	        div_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_245, 255.0);  sub_245 = None
	        clamp_min_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_10, 1.1920928955078125e-07);  div_10 = None
	        div_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_5, clamp_min_15);  minimum_5 = None
	        round_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_11);  div_11 = None
	        sub_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_11);  round_11 = None
	        clamp_min_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_251, -128);  sub_251 = None
	        clamp_max_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_16, 127);  clamp_min_16 = None
	        _assert_tensor_metadata_47 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_15, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_47 = None
	        _assert_tensor_metadata_48 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_10, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_48 = None
	        convert_element_type_30: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_10, torch.int8);  clamp_max_10 = None
	        view_80: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_7, [sym_size_int, 1500, 5120]);  clone_7 = None
	        view_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_15, [sym_size_int, 1500, 1])
	        view_82: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_30, [sym_size_int, 1500, 1])
	        reciprocal_5: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_81);  view_81 = None
	        mul_554: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_5, 1.0);  reciprocal_5 = None
	        mul_557: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_80, mul_554);  view_80 = mul_554 = None
	        round_12: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_557);  mul_557 = None
	        add_881: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_12, view_82);  round_12 = view_82 = None
	        clamp_min_17: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_881, -128);  add_881 = None
	        clamp_max_11: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_17, 127);  clamp_min_17 = None
	        view_83: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_11, [sym_size_int, 1500, 5120]);  clamp_max_11 = None
	        _assert_tensor_metadata_49 = torch.ops.aten._assert_tensor_metadata.default(view_83, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_49 = None
	        convert_element_type_31: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_83, torch.int8);  view_83 = None
	        view_84: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_31, [sym_size_int, 1500, 5120]);  convert_element_type_31 = None
	        view_85: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_15, [sym_size_int, 1500, 1]);  clamp_min_15 = None
	        view_86: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_30, [sym_size_int, 1500, 1]);  convert_element_type_30 = None
	        _assert_tensor_metadata_50 = torch.ops.aten._assert_tensor_metadata.default(view_84, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_50 = None
	        convert_element_type_32: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_84, torch.float32);  view_84 = None
	        _assert_tensor_metadata_51 = torch.ops.aten._assert_tensor_metadata.default(view_86, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_51 = None
	        convert_element_type_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_86, torch.float32);  view_86 = None
	        sub_271: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_32, convert_element_type_33);  convert_element_type_32 = convert_element_type_33 = None
	        mul_579: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_271, view_85);  sub_271 = view_85 = None
	        view_87: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_579, [sym_size_int, 1500, 5120]);  mul_579 = None
	        _assert_tensor_metadata_52 = torch.ops.aten._assert_tensor_metadata.default(view_87, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_52 = None
	        view_88: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_0_fc2_parametrizations_weight_original0 = None
	        view_89: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_0_fc2_parametrizations_weight_original1 = None
	        view_90: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_0_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_53 = torch.ops.aten._assert_tensor_metadata.default(view_88, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_53 = None
	        convert_element_type_34: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_88, torch.float32);  view_88 = None
	        _assert_tensor_metadata_54 = torch.ops.aten._assert_tensor_metadata.default(view_90, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_54 = None
	        convert_element_type_35: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_90, torch.float32);  view_90 = None
	        sub_275: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_34, convert_element_type_35);  convert_element_type_34 = convert_element_type_35 = None
	        mul_584: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_275, view_89);  sub_275 = view_89 = None
	        view_91: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_584, [1280, 5120]);  mul_584 = None
	        _assert_tensor_metadata_55 = torch.ops.aten._assert_tensor_metadata.default(view_91, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_55 = None
	        mul_589: "Sym(1500*s6)" = sym_size_int * 1500
	        view_92: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_87, [mul_589, 5120]);  view_87 = mul_589 = None
	        permute_10: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_91, [1, 0]);  view_91 = None
	        addmm_4: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_0_fc2_bias, view_92, permute_10);  model_audio_tower_layers_0_fc2_bias = view_92 = permute_10 = None
	        view_93: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_4, [sym_size_int, 1500, 1280]);  addmm_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_93);  view_93 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_944: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_646, clone_8);  add_646 = clone_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_9: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_944, memory_format = torch.contiguous_format)
	        var_mean_2 = torch.ops.aten.var_mean.correction(clone_9, [2], correction = 0, keepdim = True)
	        getitem_8: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_2[0]
	        getitem_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_2[1];  var_mean_2 = None
	        add_949: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_8, 1e-05);  getitem_8 = None
	        rsqrt_2: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_949);  add_949 = None
	        sub_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_9, getitem_9);  clone_9 = getitem_9 = None
	        mul_600: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_281, rsqrt_2);  sub_281 = rsqrt_2 = None
	        mul_601: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_600, model_audio_tower_layers_1_self_attn_layer_norm_weight);  mul_600 = model_audio_tower_layers_1_self_attn_layer_norm_weight = None
	        add_950: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_601, model_audio_tower_layers_1_self_attn_layer_norm_bias);  mul_601 = model_audio_tower_layers_1_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_94: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_950, [sym_size_int, 1500, 1280])
	        amin_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_94, [2])
	        amax_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_94, [2]);  view_94 = None
	        full_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_6, full_12);  amin_6 = full_12 = None
	        full_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_6, full_13);  amax_6 = full_13 = None
	        sub_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_6, minimum_6);  maximum_6 = None
	        div_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_292, 255.0);  sub_292 = None
	        clamp_min_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_12, 1.1920928955078125e-07);  div_12 = None
	        div_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_6, clamp_min_18);  minimum_6 = None
	        round_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_13);  div_13 = None
	        sub_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_13);  round_13 = None
	        clamp_min_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_298, -128);  sub_298 = None
	        clamp_max_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_19, 127);  clamp_min_19 = None
	        _assert_tensor_metadata_56 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_18, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_56 = None
	        _assert_tensor_metadata_57 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_12, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_57 = None
	        convert_element_type_36: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_12, torch.int8);  clamp_max_12 = None
	        view_95: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_950, [sym_size_int, 1500, 1280])
	        view_96: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_18, [sym_size_int, 1500, 1])
	        view_97: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_36, [sym_size_int, 1500, 1])
	        reciprocal_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_96);  view_96 = None
	        mul_649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_6, 1.0);  reciprocal_6 = None
	        mul_652: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_95, mul_649);  view_95 = mul_649 = None
	        round_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_652);  mul_652 = None
	        add_1037: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_14, view_97);  round_14 = view_97 = None
	        clamp_min_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1037, -128);  add_1037 = None
	        clamp_max_13: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_20, 127);  clamp_min_20 = None
	        view_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_13, [sym_size_int, 1500, 1280]);  clamp_max_13 = None
	        _assert_tensor_metadata_58 = torch.ops.aten._assert_tensor_metadata.default(view_98, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_58 = None
	        convert_element_type_37: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_98, torch.int8);  view_98 = None
	        view_99: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_37, [sym_size_int, 1500, 1280]);  convert_element_type_37 = None
	        view_100: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_18, [sym_size_int, 1500, 1]);  clamp_min_18 = None
	        view_101: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_36, [sym_size_int, 1500, 1]);  convert_element_type_36 = None
	        _assert_tensor_metadata_59 = torch.ops.aten._assert_tensor_metadata.default(view_99, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_59 = None
	        convert_element_type_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_99, torch.float32);  view_99 = None
	        _assert_tensor_metadata_60 = torch.ops.aten._assert_tensor_metadata.default(view_101, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_60 = None
	        convert_element_type_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_101, torch.float32);  view_101 = None
	        sub_318: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_38, convert_element_type_39);  convert_element_type_38 = convert_element_type_39 = None
	        mul_674: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_318, view_100);  sub_318 = view_100 = None
	        view_102: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_674, [sym_size_int, 1500, 1280]);  mul_674 = None
	        _assert_tensor_metadata_61 = torch.ops.aten._assert_tensor_metadata.default(view_102, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_61 = None
	        view_103: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_104: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_105: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_62 = torch.ops.aten._assert_tensor_metadata.default(view_103, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_62 = None
	        convert_element_type_40: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_103, torch.float32);  view_103 = None
	        _assert_tensor_metadata_63 = torch.ops.aten._assert_tensor_metadata.default(view_105, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_63 = None
	        convert_element_type_41: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_105, torch.float32);  view_105 = None
	        sub_322: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_40, convert_element_type_41);  convert_element_type_40 = convert_element_type_41 = None
	        mul_679: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_322, view_104);  sub_322 = view_104 = None
	        view_106: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_679, [1280, 1280]);  mul_679 = None
	        _assert_tensor_metadata_64 = torch.ops.aten._assert_tensor_metadata.default(view_106, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_64 = None
	        mul_684: "Sym(1500*s6)" = sym_size_int * 1500
	        view_107: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_102, [mul_684, 1280]);  view_102 = mul_684 = None
	        permute_11: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_106, [1, 0]);  view_106 = None
	        addmm_5: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_1_self_attn_q_proj_bias, view_107, permute_11);  model_audio_tower_layers_1_self_attn_q_proj_bias = view_107 = permute_11 = None
	        view_108: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_5, [sym_size_int, 1500, 1280]);  addmm_5 = None
	        mul_691: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_108, 0.125);  view_108 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_109: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_691, [sym_size_int, 1500, 20, 64]);  mul_691 = None
	        permute_12: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_109, [0, 2, 1, 3]);  view_109 = None
	        clone_10: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_12, memory_format = torch.contiguous_format);  permute_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_950, [sym_size_int, 1500, 1280])
	        amin_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_110, [2])
	        amax_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_110, [2]);  view_110 = None
	        full_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_7, full_14);  amin_7 = full_14 = None
	        full_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_7, full_15);  amax_7 = full_15 = None
	        sub_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_7, minimum_7);  maximum_7 = None
	        div_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_337, 255.0);  sub_337 = None
	        clamp_min_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_14, 1.1920928955078125e-07);  div_14 = None
	        div_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_7, clamp_min_21);  minimum_7 = None
	        round_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_15);  div_15 = None
	        sub_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_15);  round_15 = None
	        clamp_min_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_343, -128);  sub_343 = None
	        clamp_max_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_22, 127);  clamp_min_22 = None
	        _assert_tensor_metadata_65 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_21, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_65 = None
	        _assert_tensor_metadata_66 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_14, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_66 = None
	        convert_element_type_42: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_14, torch.int8);  clamp_max_14 = None
	        view_111: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_950, [sym_size_int, 1500, 1280])
	        view_112: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_21, [sym_size_int, 1500, 1])
	        view_113: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_42, [sym_size_int, 1500, 1])
	        reciprocal_7: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_112);  view_112 = None
	        mul_745: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_7, 1.0);  reciprocal_7 = None
	        mul_748: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_111, mul_745);  view_111 = mul_745 = None
	        round_16: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_748);  mul_748 = None
	        add_1189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_16, view_113);  round_16 = view_113 = None
	        clamp_min_23: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1189, -128);  add_1189 = None
	        clamp_max_15: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_23, 127);  clamp_min_23 = None
	        view_114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_15, [sym_size_int, 1500, 1280]);  clamp_max_15 = None
	        _assert_tensor_metadata_67 = torch.ops.aten._assert_tensor_metadata.default(view_114, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_67 = None
	        convert_element_type_43: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_114, torch.int8);  view_114 = None
	        view_115: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_43, [sym_size_int, 1500, 1280]);  convert_element_type_43 = None
	        view_116: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_21, [sym_size_int, 1500, 1]);  clamp_min_21 = None
	        view_117: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_42, [sym_size_int, 1500, 1]);  convert_element_type_42 = None
	        _assert_tensor_metadata_68 = torch.ops.aten._assert_tensor_metadata.default(view_115, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_68 = None
	        convert_element_type_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_115, torch.float32);  view_115 = None
	        _assert_tensor_metadata_69 = torch.ops.aten._assert_tensor_metadata.default(view_117, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_69 = None
	        convert_element_type_45: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_117, torch.float32);  view_117 = None
	        sub_363: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_44, convert_element_type_45);  convert_element_type_44 = convert_element_type_45 = None
	        mul_770: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_363, view_116);  sub_363 = view_116 = None
	        view_118: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_770, [sym_size_int, 1500, 1280]);  mul_770 = None
	        _assert_tensor_metadata_70 = torch.ops.aten._assert_tensor_metadata.default(view_118, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_70 = None
	        view_119: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_120: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_121: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_71 = torch.ops.aten._assert_tensor_metadata.default(view_119, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_71 = None
	        convert_element_type_46: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_119, torch.float32);  view_119 = None
	        _assert_tensor_metadata_72 = torch.ops.aten._assert_tensor_metadata.default(view_121, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_72 = None
	        convert_element_type_47: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_121, torch.float32);  view_121 = None
	        sub_367: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_46, convert_element_type_47);  convert_element_type_46 = convert_element_type_47 = None
	        mul_775: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_367, view_120);  sub_367 = view_120 = None
	        view_122: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_775, [1280, 1280]);  mul_775 = None
	        _assert_tensor_metadata_73 = torch.ops.aten._assert_tensor_metadata.default(view_122, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_73 = None
	        permute_13: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_122, [1, 0]);  view_122 = None
	        mul_778: "Sym(1500*s6)" = sym_size_int * 1500
	        view_123: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_118, [mul_778, 1280]);  view_118 = mul_778 = None
	        mm_1: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_123, permute_13);  view_123 = permute_13 = None
	        view_124: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_1, [sym_size_int, 1500, 1280]);  mm_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_125: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_124, [sym_size_int, -1, 20, 64]);  view_124 = None
	        permute_14: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_125, [0, 2, 1, 3]);  view_125 = None
	        clone_11: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_14, memory_format = torch.contiguous_format);  permute_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_126: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_950, [sym_size_int, 1500, 1280])
	        amin_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_126, [2])
	        amax_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_126, [2]);  view_126 = None
	        full_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_8, full_16);  amin_8 = full_16 = None
	        full_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_8, full_17);  amax_8 = full_17 = None
	        sub_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_8, minimum_8);  maximum_8 = None
	        div_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_381, 255.0);  sub_381 = None
	        clamp_min_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_16, 1.1920928955078125e-07);  div_16 = None
	        div_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_8, clamp_min_24);  minimum_8 = None
	        round_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_17);  div_17 = None
	        sub_387: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_17);  round_17 = None
	        clamp_min_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_387, -128);  sub_387 = None
	        clamp_max_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_25, 127);  clamp_min_25 = None
	        _assert_tensor_metadata_74 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_24, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_74 = None
	        _assert_tensor_metadata_75 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_16, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_75 = None
	        convert_element_type_48: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_16, torch.int8);  clamp_max_16 = None
	        view_127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_950, [sym_size_int, 1500, 1280]);  add_950 = None
	        view_128: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_24, [sym_size_int, 1500, 1])
	        view_129: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_48, [sym_size_int, 1500, 1])
	        reciprocal_8: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_128);  view_128 = None
	        mul_844: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_8, 1.0);  reciprocal_8 = None
	        mul_847: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_127, mul_844);  view_127 = mul_844 = None
	        round_18: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_847);  mul_847 = None
	        add_1337: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_18, view_129);  round_18 = view_129 = None
	        clamp_min_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1337, -128);  add_1337 = None
	        clamp_max_17: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_26, 127);  clamp_min_26 = None
	        view_130: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_17, [sym_size_int, 1500, 1280]);  clamp_max_17 = None
	        _assert_tensor_metadata_76 = torch.ops.aten._assert_tensor_metadata.default(view_130, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_76 = None
	        convert_element_type_49: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_130, torch.int8);  view_130 = None
	        view_131: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_49, [sym_size_int, 1500, 1280]);  convert_element_type_49 = None
	        view_132: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_24, [sym_size_int, 1500, 1]);  clamp_min_24 = None
	        view_133: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_48, [sym_size_int, 1500, 1]);  convert_element_type_48 = None
	        _assert_tensor_metadata_77 = torch.ops.aten._assert_tensor_metadata.default(view_131, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_77 = None
	        convert_element_type_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_131, torch.float32);  view_131 = None
	        _assert_tensor_metadata_78 = torch.ops.aten._assert_tensor_metadata.default(view_133, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_78 = None
	        convert_element_type_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_133, torch.float32);  view_133 = None
	        sub_407: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_50, convert_element_type_51);  convert_element_type_50 = convert_element_type_51 = None
	        mul_869: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_407, view_132);  sub_407 = view_132 = None
	        view_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_869, [sym_size_int, 1500, 1280]);  mul_869 = None
	        _assert_tensor_metadata_79 = torch.ops.aten._assert_tensor_metadata.default(view_134, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_79 = None
	        view_135: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_136: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_137: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_80 = torch.ops.aten._assert_tensor_metadata.default(view_135, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_80 = None
	        convert_element_type_52: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_135, torch.float32);  view_135 = None
	        _assert_tensor_metadata_81 = torch.ops.aten._assert_tensor_metadata.default(view_137, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_81 = None
	        convert_element_type_53: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_137, torch.float32);  view_137 = None
	        sub_411: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_52, convert_element_type_53);  convert_element_type_52 = convert_element_type_53 = None
	        mul_874: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_411, view_136);  sub_411 = view_136 = None
	        view_138: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_874, [1280, 1280]);  mul_874 = None
	        _assert_tensor_metadata_82 = torch.ops.aten._assert_tensor_metadata.default(view_138, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_82 = None
	        mul_879: "Sym(1500*s6)" = sym_size_int * 1500
	        view_139: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_134, [mul_879, 1280]);  view_134 = mul_879 = None
	        permute_15: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_138, [1, 0]);  view_138 = None
	        addmm_6: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_1_self_attn_v_proj_bias, view_139, permute_15);  model_audio_tower_layers_1_self_attn_v_proj_bias = view_139 = permute_15 = None
	        view_140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_6, [sym_size_int, 1500, 1280]);  addmm_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_141: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_140, [sym_size_int, -1, 20, 64]);  view_140 = None
	        permute_16: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_141, [0, 2, 1, 3]);  view_141 = None
	        clone_12: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_16, memory_format = torch.contiguous_format);  permute_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_1 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_10, clone_11, clone_12, None, False, scale = 1.0);  clone_10 = clone_11 = clone_12 = None
	        getitem_10: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_1[0];  _scaled_dot_product_efficient_attention_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_17: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_10, [0, 2, 1, 3]);  getitem_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_17, [sym_size_int, 1500, -1]);  permute_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_143: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_142, [sym_size_int, 1500, 1280])
	        amin_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_143, [2])
	        amax_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_143, [2]);  view_143 = None
	        full_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_9, full_18);  amin_9 = full_18 = None
	        full_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_9, full_19);  amax_9 = full_19 = None
	        sub_429: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_9, minimum_9);  maximum_9 = None
	        div_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_429, 255.0);  sub_429 = None
	        clamp_min_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_18, 1.1920928955078125e-07);  div_18 = None
	        div_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_9, clamp_min_27);  minimum_9 = None
	        round_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_19);  div_19 = None
	        sub_435: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_19);  round_19 = None
	        clamp_min_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_435, -128);  sub_435 = None
	        clamp_max_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_28, 127);  clamp_min_28 = None
	        _assert_tensor_metadata_83 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_27, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_83 = None
	        _assert_tensor_metadata_84 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_18, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_84 = None
	        convert_element_type_54: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_18, torch.int8);  clamp_max_18 = None
	        view_144: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_142, [sym_size_int, 1500, 1280]);  view_142 = None
	        view_145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_27, [sym_size_int, 1500, 1])
	        view_146: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_54, [sym_size_int, 1500, 1])
	        reciprocal_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_145);  view_145 = None
	        mul_949: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_9, 1.0);  reciprocal_9 = None
	        mul_952: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_144, mul_949);  view_144 = mul_949 = None
	        round_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_952);  mul_952 = None
	        add_1501: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_20, view_146);  round_20 = view_146 = None
	        clamp_min_29: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1501, -128);  add_1501 = None
	        clamp_max_19: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_29, 127);  clamp_min_29 = None
	        view_147: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_19, [sym_size_int, 1500, 1280]);  clamp_max_19 = None
	        _assert_tensor_metadata_85 = torch.ops.aten._assert_tensor_metadata.default(view_147, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_85 = None
	        convert_element_type_55: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_147, torch.int8);  view_147 = None
	        view_148: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_55, [sym_size_int, 1500, 1280]);  convert_element_type_55 = None
	        view_149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_27, [sym_size_int, 1500, 1]);  clamp_min_27 = None
	        view_150: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_54, [sym_size_int, 1500, 1]);  convert_element_type_54 = None
	        _assert_tensor_metadata_86 = torch.ops.aten._assert_tensor_metadata.default(view_148, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_86 = None
	        convert_element_type_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_148, torch.float32);  view_148 = None
	        _assert_tensor_metadata_87 = torch.ops.aten._assert_tensor_metadata.default(view_150, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_87 = None
	        convert_element_type_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_150, torch.float32);  view_150 = None
	        sub_455: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_56, convert_element_type_57);  convert_element_type_56 = convert_element_type_57 = None
	        mul_974: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_455, view_149);  sub_455 = view_149 = None
	        view_151: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_974, [sym_size_int, 1500, 1280]);  mul_974 = None
	        _assert_tensor_metadata_88 = torch.ops.aten._assert_tensor_metadata.default(view_151, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_88 = None
	        view_152: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_153: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_154: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_89 = torch.ops.aten._assert_tensor_metadata.default(view_152, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_89 = None
	        convert_element_type_58: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_152, torch.float32);  view_152 = None
	        _assert_tensor_metadata_90 = torch.ops.aten._assert_tensor_metadata.default(view_154, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_90 = None
	        convert_element_type_59: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_154, torch.float32);  view_154 = None
	        sub_459: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_58, convert_element_type_59);  convert_element_type_58 = convert_element_type_59 = None
	        mul_979: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_459, view_153);  sub_459 = view_153 = None
	        view_155: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_979, [1280, 1280]);  mul_979 = None
	        _assert_tensor_metadata_91 = torch.ops.aten._assert_tensor_metadata.default(view_155, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_91 = None
	        mul_984: "Sym(1500*s6)" = sym_size_int * 1500
	        view_156: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_151, [mul_984, 1280]);  view_151 = mul_984 = None
	        permute_18: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_155, [1, 0]);  view_155 = None
	        addmm_7: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_1_self_attn_out_proj_bias, view_156, permute_18);  model_audio_tower_layers_1_self_attn_out_proj_bias = view_156 = permute_18 = None
	        view_157: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_7, [sym_size_int, 1500, 1280]);  addmm_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_13: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_157);  view_157 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_1564: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_944, clone_13);  add_944 = clone_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_1564, memory_format = torch.contiguous_format)
	        var_mean_3 = torch.ops.aten.var_mean.correction(clone_14, [2], correction = 0, keepdim = True)
	        getitem_14: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_3[0]
	        getitem_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_3[1];  var_mean_3 = None
	        add_1569: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_14, 1e-05);  getitem_14 = None
	        rsqrt_3: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_1569);  add_1569 = None
	        sub_465: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_14, getitem_15);  clone_14 = getitem_15 = None
	        mul_995: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_465, rsqrt_3);  sub_465 = rsqrt_3 = None
	        mul_996: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_995, model_audio_tower_layers_1_final_layer_norm_weight);  mul_995 = model_audio_tower_layers_1_final_layer_norm_weight = None
	        add_1570: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_996, model_audio_tower_layers_1_final_layer_norm_bias);  mul_996 = model_audio_tower_layers_1_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_1570, [sym_size_int, 1500, 1280])
	        amin_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_158, [2])
	        amax_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_158, [2]);  view_158 = None
	        full_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_10, full_20);  amin_10 = full_20 = None
	        full_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_10, full_21);  amax_10 = full_21 = None
	        sub_476: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_10, minimum_10);  maximum_10 = None
	        div_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_476, 255.0);  sub_476 = None
	        clamp_min_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_20, 1.1920928955078125e-07);  div_20 = None
	        div_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_10, clamp_min_30);  minimum_10 = None
	        round_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_21);  div_21 = None
	        sub_482: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_21);  round_21 = None
	        clamp_min_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_482, -128);  sub_482 = None
	        clamp_max_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_31, 127);  clamp_min_31 = None
	        _assert_tensor_metadata_92 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_30, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_92 = None
	        _assert_tensor_metadata_93 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_20, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_93 = None
	        convert_element_type_60: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_20, torch.int8);  clamp_max_20 = None
	        view_159: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_1570, [sym_size_int, 1500, 1280]);  add_1570 = None
	        view_160: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_30, [sym_size_int, 1500, 1])
	        view_161: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_60, [sym_size_int, 1500, 1])
	        reciprocal_10: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_160);  view_160 = None
	        mul_1044: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_10, 1.0);  reciprocal_10 = None
	        mul_1047: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_159, mul_1044);  view_159 = mul_1044 = None
	        round_22: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1047);  mul_1047 = None
	        add_1657: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_22, view_161);  round_22 = view_161 = None
	        clamp_min_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1657, -128);  add_1657 = None
	        clamp_max_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_32, 127);  clamp_min_32 = None
	        view_162: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_21, [sym_size_int, 1500, 1280]);  clamp_max_21 = None
	        _assert_tensor_metadata_94 = torch.ops.aten._assert_tensor_metadata.default(view_162, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_94 = None
	        convert_element_type_61: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_162, torch.int8);  view_162 = None
	        view_163: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_61, [sym_size_int, 1500, 1280]);  convert_element_type_61 = None
	        view_164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_30, [sym_size_int, 1500, 1]);  clamp_min_30 = None
	        view_165: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_60, [sym_size_int, 1500, 1]);  convert_element_type_60 = None
	        _assert_tensor_metadata_95 = torch.ops.aten._assert_tensor_metadata.default(view_163, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_95 = None
	        convert_element_type_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_163, torch.float32);  view_163 = None
	        _assert_tensor_metadata_96 = torch.ops.aten._assert_tensor_metadata.default(view_165, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_96 = None
	        convert_element_type_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_165, torch.float32);  view_165 = None
	        sub_502: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_62, convert_element_type_63);  convert_element_type_62 = convert_element_type_63 = None
	        mul_1069: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_502, view_164);  sub_502 = view_164 = None
	        view_166: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1069, [sym_size_int, 1500, 1280]);  mul_1069 = None
	        _assert_tensor_metadata_97 = torch.ops.aten._assert_tensor_metadata.default(view_166, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_97 = None
	        view_167: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_1_fc1_parametrizations_weight_original0 = None
	        view_168: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_1_fc1_parametrizations_weight_original1 = None
	        view_169: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_1_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_98 = torch.ops.aten._assert_tensor_metadata.default(view_167, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_98 = None
	        convert_element_type_64: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_167, torch.float32);  view_167 = None
	        _assert_tensor_metadata_99 = torch.ops.aten._assert_tensor_metadata.default(view_169, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_99 = None
	        convert_element_type_65: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_169, torch.float32);  view_169 = None
	        sub_506: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_64, convert_element_type_65);  convert_element_type_64 = convert_element_type_65 = None
	        mul_1074: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_506, view_168);  sub_506 = view_168 = None
	        view_170: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1074, [5120, 1280]);  mul_1074 = None
	        _assert_tensor_metadata_100 = torch.ops.aten._assert_tensor_metadata.default(view_170, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_100 = None
	        mul_1079: "Sym(1500*s6)" = sym_size_int * 1500
	        view_171: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_166, [mul_1079, 1280]);  view_166 = mul_1079 = None
	        permute_19: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_170, [1, 0]);  view_170 = None
	        addmm_8: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_1_fc1_bias, view_171, permute_19);  model_audio_tower_layers_1_fc1_bias = view_171 = permute_19 = None
	        view_172: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_8, [sym_size_int, 1500, 5120]);  addmm_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_1086: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_172, 0.5)
	        mul_1087: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_172, 0.7071067811865476);  view_172 = None
	        erf_3: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_1087);  mul_1087 = None
	        add_1716: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_3, 1);  erf_3 = None
	        mul_1088: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1086, add_1716);  mul_1086 = add_1716 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_15: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_1088);  mul_1088 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_173: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_15, [sym_size_int, 1500, 5120])
	        amin_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_173, [2])
	        amax_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_173, [2]);  view_173 = None
	        full_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_11, full_22);  amin_11 = full_22 = None
	        full_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_11, full_23);  amax_11 = full_23 = None
	        sub_519: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_11, minimum_11);  maximum_11 = None
	        div_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_519, 255.0);  sub_519 = None
	        clamp_min_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_22, 1.1920928955078125e-07);  div_22 = None
	        div_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_11, clamp_min_33);  minimum_11 = None
	        round_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_23);  div_23 = None
	        sub_525: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_23);  round_23 = None
	        clamp_min_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_525, -128);  sub_525 = None
	        clamp_max_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_34, 127);  clamp_min_34 = None
	        _assert_tensor_metadata_101 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_33, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_101 = None
	        _assert_tensor_metadata_102 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_22, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_102 = None
	        convert_element_type_66: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_22, torch.int8);  clamp_max_22 = None
	        view_174: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_15, [sym_size_int, 1500, 5120]);  clone_15 = None
	        view_175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_33, [sym_size_int, 1500, 1])
	        view_176: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_66, [sym_size_int, 1500, 1])
	        reciprocal_11: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_175);  view_175 = None
	        mul_1134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_11, 1.0);  reciprocal_11 = None
	        mul_1137: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_174, mul_1134);  view_174 = mul_1134 = None
	        round_24: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_1137);  mul_1137 = None
	        add_1799: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_24, view_176);  round_24 = view_176 = None
	        clamp_min_35: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1799, -128);  add_1799 = None
	        clamp_max_23: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_35, 127);  clamp_min_35 = None
	        view_177: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_23, [sym_size_int, 1500, 5120]);  clamp_max_23 = None
	        _assert_tensor_metadata_103 = torch.ops.aten._assert_tensor_metadata.default(view_177, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_103 = None
	        convert_element_type_67: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_177, torch.int8);  view_177 = None
	        view_178: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_67, [sym_size_int, 1500, 5120]);  convert_element_type_67 = None
	        view_179: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_33, [sym_size_int, 1500, 1]);  clamp_min_33 = None
	        view_180: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_66, [sym_size_int, 1500, 1]);  convert_element_type_66 = None
	        _assert_tensor_metadata_104 = torch.ops.aten._assert_tensor_metadata.default(view_178, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_104 = None
	        convert_element_type_68: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_178, torch.float32);  view_178 = None
	        _assert_tensor_metadata_105 = torch.ops.aten._assert_tensor_metadata.default(view_180, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_105 = None
	        convert_element_type_69: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_180, torch.float32);  view_180 = None
	        sub_545: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_68, convert_element_type_69);  convert_element_type_68 = convert_element_type_69 = None
	        mul_1159: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_545, view_179);  sub_545 = view_179 = None
	        view_181: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_1159, [sym_size_int, 1500, 5120]);  mul_1159 = None
	        _assert_tensor_metadata_106 = torch.ops.aten._assert_tensor_metadata.default(view_181, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_106 = None
	        view_182: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_1_fc2_parametrizations_weight_original0 = None
	        view_183: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_1_fc2_parametrizations_weight_original1 = None
	        view_184: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_1_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_107 = torch.ops.aten._assert_tensor_metadata.default(view_182, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_107 = None
	        convert_element_type_70: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_182, torch.float32);  view_182 = None
	        _assert_tensor_metadata_108 = torch.ops.aten._assert_tensor_metadata.default(view_184, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_108 = None
	        convert_element_type_71: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_184, torch.float32);  view_184 = None
	        sub_549: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_70, convert_element_type_71);  convert_element_type_70 = convert_element_type_71 = None
	        mul_1164: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_549, view_183);  sub_549 = view_183 = None
	        view_185: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_1164, [1280, 5120]);  mul_1164 = None
	        _assert_tensor_metadata_109 = torch.ops.aten._assert_tensor_metadata.default(view_185, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_109 = None
	        mul_1169: "Sym(1500*s6)" = sym_size_int * 1500
	        view_186: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_181, [mul_1169, 5120]);  view_181 = mul_1169 = None
	        permute_20: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_185, [1, 0]);  view_185 = None
	        addmm_9: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_1_fc2_bias, view_186, permute_20);  model_audio_tower_layers_1_fc2_bias = view_186 = permute_20 = None
	        view_187: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_9, [sym_size_int, 1500, 1280]);  addmm_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_16: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_187);  view_187 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_1862: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_1564, clone_16);  add_1564 = clone_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_17: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_1862, memory_format = torch.contiguous_format)
	        var_mean_4 = torch.ops.aten.var_mean.correction(clone_17, [2], correction = 0, keepdim = True)
	        getitem_16: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_4[0]
	        getitem_17: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_4[1];  var_mean_4 = None
	        add_1867: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_16, 1e-05);  getitem_16 = None
	        rsqrt_4: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_1867);  add_1867 = None
	        sub_555: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_17, getitem_17);  clone_17 = getitem_17 = None
	        mul_1180: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_555, rsqrt_4);  sub_555 = rsqrt_4 = None
	        mul_1181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1180, model_audio_tower_layers_2_self_attn_layer_norm_weight);  mul_1180 = model_audio_tower_layers_2_self_attn_layer_norm_weight = None
	        add_1868: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_1181, model_audio_tower_layers_2_self_attn_layer_norm_bias);  mul_1181 = model_audio_tower_layers_2_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_1868, [sym_size_int, 1500, 1280])
	        amin_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_188, [2])
	        amax_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_188, [2]);  view_188 = None
	        full_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_12, full_24);  amin_12 = full_24 = None
	        full_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_12, full_25);  amax_12 = full_25 = None
	        sub_566: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_12, minimum_12);  maximum_12 = None
	        div_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_566, 255.0);  sub_566 = None
	        clamp_min_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_24, 1.1920928955078125e-07);  div_24 = None
	        div_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_12, clamp_min_36);  minimum_12 = None
	        round_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_25);  div_25 = None
	        sub_572: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_25);  round_25 = None
	        clamp_min_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_572, -128);  sub_572 = None
	        clamp_max_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_37, 127);  clamp_min_37 = None
	        _assert_tensor_metadata_110 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_36, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_110 = None
	        _assert_tensor_metadata_111 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_24, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_111 = None
	        convert_element_type_72: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_24, torch.int8);  clamp_max_24 = None
	        view_189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_1868, [sym_size_int, 1500, 1280])
	        view_190: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_36, [sym_size_int, 1500, 1])
	        view_191: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_72, [sym_size_int, 1500, 1])
	        reciprocal_12: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_190);  view_190 = None
	        mul_1229: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_12, 1.0);  reciprocal_12 = None
	        mul_1232: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_189, mul_1229);  view_189 = mul_1229 = None
	        round_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1232);  mul_1232 = None
	        add_1955: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_26, view_191);  round_26 = view_191 = None
	        clamp_min_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1955, -128);  add_1955 = None
	        clamp_max_25: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_38, 127);  clamp_min_38 = None
	        view_192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_25, [sym_size_int, 1500, 1280]);  clamp_max_25 = None
	        _assert_tensor_metadata_112 = torch.ops.aten._assert_tensor_metadata.default(view_192, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_112 = None
	        convert_element_type_73: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_192, torch.int8);  view_192 = None
	        view_193: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_73, [sym_size_int, 1500, 1280]);  convert_element_type_73 = None
	        view_194: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_36, [sym_size_int, 1500, 1]);  clamp_min_36 = None
	        view_195: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_72, [sym_size_int, 1500, 1]);  convert_element_type_72 = None
	        _assert_tensor_metadata_113 = torch.ops.aten._assert_tensor_metadata.default(view_193, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_113 = None
	        convert_element_type_74: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_193, torch.float32);  view_193 = None
	        _assert_tensor_metadata_114 = torch.ops.aten._assert_tensor_metadata.default(view_195, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_114 = None
	        convert_element_type_75: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_195, torch.float32);  view_195 = None
	        sub_592: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_74, convert_element_type_75);  convert_element_type_74 = convert_element_type_75 = None
	        mul_1254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_592, view_194);  sub_592 = view_194 = None
	        view_196: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1254, [sym_size_int, 1500, 1280]);  mul_1254 = None
	        _assert_tensor_metadata_115 = torch.ops.aten._assert_tensor_metadata.default(view_196, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_115 = None
	        view_197: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_198: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_199: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_116 = torch.ops.aten._assert_tensor_metadata.default(view_197, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_116 = None
	        convert_element_type_76: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_197, torch.float32);  view_197 = None
	        _assert_tensor_metadata_117 = torch.ops.aten._assert_tensor_metadata.default(view_199, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_117 = None
	        convert_element_type_77: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_199, torch.float32);  view_199 = None
	        sub_596: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_76, convert_element_type_77);  convert_element_type_76 = convert_element_type_77 = None
	        mul_1259: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_596, view_198);  sub_596 = view_198 = None
	        view_200: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1259, [1280, 1280]);  mul_1259 = None
	        _assert_tensor_metadata_118 = torch.ops.aten._assert_tensor_metadata.default(view_200, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_118 = None
	        mul_1264: "Sym(1500*s6)" = sym_size_int * 1500
	        view_201: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_196, [mul_1264, 1280]);  view_196 = mul_1264 = None
	        permute_21: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_200, [1, 0]);  view_200 = None
	        addmm_10: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_2_self_attn_q_proj_bias, view_201, permute_21);  model_audio_tower_layers_2_self_attn_q_proj_bias = view_201 = permute_21 = None
	        view_202: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_10, [sym_size_int, 1500, 1280]);  addmm_10 = None
	        mul_1271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_202, 0.125);  view_202 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_203: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_1271, [sym_size_int, 1500, 20, 64]);  mul_1271 = None
	        permute_22: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_203, [0, 2, 1, 3]);  view_203 = None
	        clone_18: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_22, memory_format = torch.contiguous_format);  permute_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_204: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_1868, [sym_size_int, 1500, 1280])
	        amin_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_204, [2])
	        amax_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_204, [2]);  view_204 = None
	        full_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_13, full_26);  amin_13 = full_26 = None
	        full_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_13, full_27);  amax_13 = full_27 = None
	        sub_611: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_13, minimum_13);  maximum_13 = None
	        div_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_611, 255.0);  sub_611 = None
	        clamp_min_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_26, 1.1920928955078125e-07);  div_26 = None
	        div_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_13, clamp_min_39);  minimum_13 = None
	        round_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_27);  div_27 = None
	        sub_617: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_27);  round_27 = None
	        clamp_min_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_617, -128);  sub_617 = None
	        clamp_max_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_40, 127);  clamp_min_40 = None
	        _assert_tensor_metadata_119 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_39, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_119 = None
	        _assert_tensor_metadata_120 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_26, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_120 = None
	        convert_element_type_78: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_26, torch.int8);  clamp_max_26 = None
	        view_205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_1868, [sym_size_int, 1500, 1280])
	        view_206: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_39, [sym_size_int, 1500, 1])
	        view_207: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_78, [sym_size_int, 1500, 1])
	        reciprocal_13: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_206);  view_206 = None
	        mul_1325: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_13, 1.0);  reciprocal_13 = None
	        mul_1328: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_205, mul_1325);  view_205 = mul_1325 = None
	        round_28: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1328);  mul_1328 = None
	        add_2107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_28, view_207);  round_28 = view_207 = None
	        clamp_min_41: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2107, -128);  add_2107 = None
	        clamp_max_27: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_41, 127);  clamp_min_41 = None
	        view_208: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_27, [sym_size_int, 1500, 1280]);  clamp_max_27 = None
	        _assert_tensor_metadata_121 = torch.ops.aten._assert_tensor_metadata.default(view_208, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_121 = None
	        convert_element_type_79: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_208, torch.int8);  view_208 = None
	        view_209: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_79, [sym_size_int, 1500, 1280]);  convert_element_type_79 = None
	        view_210: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_39, [sym_size_int, 1500, 1]);  clamp_min_39 = None
	        view_211: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_78, [sym_size_int, 1500, 1]);  convert_element_type_78 = None
	        _assert_tensor_metadata_122 = torch.ops.aten._assert_tensor_metadata.default(view_209, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_122 = None
	        convert_element_type_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_209, torch.float32);  view_209 = None
	        _assert_tensor_metadata_123 = torch.ops.aten._assert_tensor_metadata.default(view_211, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_123 = None
	        convert_element_type_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_211, torch.float32);  view_211 = None
	        sub_637: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_80, convert_element_type_81);  convert_element_type_80 = convert_element_type_81 = None
	        mul_1350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_637, view_210);  sub_637 = view_210 = None
	        view_212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1350, [sym_size_int, 1500, 1280]);  mul_1350 = None
	        _assert_tensor_metadata_124 = torch.ops.aten._assert_tensor_metadata.default(view_212, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_124 = None
	        view_213: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_214: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_215: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_125 = torch.ops.aten._assert_tensor_metadata.default(view_213, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_125 = None
	        convert_element_type_82: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_213, torch.float32);  view_213 = None
	        _assert_tensor_metadata_126 = torch.ops.aten._assert_tensor_metadata.default(view_215, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_126 = None
	        convert_element_type_83: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_215, torch.float32);  view_215 = None
	        sub_641: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_82, convert_element_type_83);  convert_element_type_82 = convert_element_type_83 = None
	        mul_1355: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_641, view_214);  sub_641 = view_214 = None
	        view_216: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1355, [1280, 1280]);  mul_1355 = None
	        _assert_tensor_metadata_127 = torch.ops.aten._assert_tensor_metadata.default(view_216, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_127 = None
	        permute_23: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_216, [1, 0]);  view_216 = None
	        mul_1358: "Sym(1500*s6)" = sym_size_int * 1500
	        view_217: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_212, [mul_1358, 1280]);  view_212 = mul_1358 = None
	        mm_2: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_217, permute_23);  view_217 = permute_23 = None
	        view_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_2, [sym_size_int, 1500, 1280]);  mm_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_219: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_218, [sym_size_int, -1, 20, 64]);  view_218 = None
	        permute_24: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_219, [0, 2, 1, 3]);  view_219 = None
	        clone_19: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_24, memory_format = torch.contiguous_format);  permute_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_220: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_1868, [sym_size_int, 1500, 1280])
	        amin_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_220, [2])
	        amax_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_220, [2]);  view_220 = None
	        full_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_14, full_28);  amin_14 = full_28 = None
	        full_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_14, full_29);  amax_14 = full_29 = None
	        sub_655: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_14, minimum_14);  maximum_14 = None
	        div_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_655, 255.0);  sub_655 = None
	        clamp_min_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_28, 1.1920928955078125e-07);  div_28 = None
	        div_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_14, clamp_min_42);  minimum_14 = None
	        round_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_29);  div_29 = None
	        sub_661: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_29);  round_29 = None
	        clamp_min_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_661, -128);  sub_661 = None
	        clamp_max_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_43, 127);  clamp_min_43 = None
	        _assert_tensor_metadata_128 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_42, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_128 = None
	        _assert_tensor_metadata_129 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_28, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_129 = None
	        convert_element_type_84: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_28, torch.int8);  clamp_max_28 = None
	        view_221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_1868, [sym_size_int, 1500, 1280]);  add_1868 = None
	        view_222: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_42, [sym_size_int, 1500, 1])
	        view_223: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_84, [sym_size_int, 1500, 1])
	        reciprocal_14: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_222);  view_222 = None
	        mul_1424: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_14, 1.0);  reciprocal_14 = None
	        mul_1427: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_221, mul_1424);  view_221 = mul_1424 = None
	        round_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1427);  mul_1427 = None
	        add_2255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_30, view_223);  round_30 = view_223 = None
	        clamp_min_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2255, -128);  add_2255 = None
	        clamp_max_29: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_44, 127);  clamp_min_44 = None
	        view_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_29, [sym_size_int, 1500, 1280]);  clamp_max_29 = None
	        _assert_tensor_metadata_130 = torch.ops.aten._assert_tensor_metadata.default(view_224, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_130 = None
	        convert_element_type_85: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_224, torch.int8);  view_224 = None
	        view_225: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_85, [sym_size_int, 1500, 1280]);  convert_element_type_85 = None
	        view_226: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_42, [sym_size_int, 1500, 1]);  clamp_min_42 = None
	        view_227: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_84, [sym_size_int, 1500, 1]);  convert_element_type_84 = None
	        _assert_tensor_metadata_131 = torch.ops.aten._assert_tensor_metadata.default(view_225, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_131 = None
	        convert_element_type_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_225, torch.float32);  view_225 = None
	        _assert_tensor_metadata_132 = torch.ops.aten._assert_tensor_metadata.default(view_227, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_132 = None
	        convert_element_type_87: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_227, torch.float32);  view_227 = None
	        sub_681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_86, convert_element_type_87);  convert_element_type_86 = convert_element_type_87 = None
	        mul_1449: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_681, view_226);  sub_681 = view_226 = None
	        view_228: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1449, [sym_size_int, 1500, 1280]);  mul_1449 = None
	        _assert_tensor_metadata_133 = torch.ops.aten._assert_tensor_metadata.default(view_228, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_133 = None
	        view_229: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_230: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_231: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_134 = torch.ops.aten._assert_tensor_metadata.default(view_229, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_134 = None
	        convert_element_type_88: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_229, torch.float32);  view_229 = None
	        _assert_tensor_metadata_135 = torch.ops.aten._assert_tensor_metadata.default(view_231, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_135 = None
	        convert_element_type_89: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_231, torch.float32);  view_231 = None
	        sub_685: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_88, convert_element_type_89);  convert_element_type_88 = convert_element_type_89 = None
	        mul_1454: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_685, view_230);  sub_685 = view_230 = None
	        view_232: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1454, [1280, 1280]);  mul_1454 = None
	        _assert_tensor_metadata_136 = torch.ops.aten._assert_tensor_metadata.default(view_232, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_136 = None
	        mul_1459: "Sym(1500*s6)" = sym_size_int * 1500
	        view_233: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_228, [mul_1459, 1280]);  view_228 = mul_1459 = None
	        permute_25: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_232, [1, 0]);  view_232 = None
	        addmm_11: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_2_self_attn_v_proj_bias, view_233, permute_25);  model_audio_tower_layers_2_self_attn_v_proj_bias = view_233 = permute_25 = None
	        view_234: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_11, [sym_size_int, 1500, 1280]);  addmm_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_235: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_234, [sym_size_int, -1, 20, 64]);  view_234 = None
	        permute_26: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_235, [0, 2, 1, 3]);  view_235 = None
	        clone_20: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_26, memory_format = torch.contiguous_format);  permute_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_2 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_18, clone_19, clone_20, None, False, scale = 1.0);  clone_18 = clone_19 = clone_20 = None
	        getitem_18: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_2[0];  _scaled_dot_product_efficient_attention_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_27: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_18, [0, 2, 1, 3]);  getitem_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_27, [sym_size_int, 1500, -1]);  permute_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_237: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_236, [sym_size_int, 1500, 1280])
	        amin_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_237, [2])
	        amax_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_237, [2]);  view_237 = None
	        full_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_15, full_30);  amin_15 = full_30 = None
	        full_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_15, full_31);  amax_15 = full_31 = None
	        sub_703: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_15, minimum_15);  maximum_15 = None
	        div_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_703, 255.0);  sub_703 = None
	        clamp_min_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_30, 1.1920928955078125e-07);  div_30 = None
	        div_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_15, clamp_min_45);  minimum_15 = None
	        round_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_31);  div_31 = None
	        sub_709: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_31);  round_31 = None
	        clamp_min_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_709, -128);  sub_709 = None
	        clamp_max_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_46, 127);  clamp_min_46 = None
	        _assert_tensor_metadata_137 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_45, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_137 = None
	        _assert_tensor_metadata_138 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_30, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_138 = None
	        convert_element_type_90: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_30, torch.int8);  clamp_max_30 = None
	        view_238: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_236, [sym_size_int, 1500, 1280]);  view_236 = None
	        view_239: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_45, [sym_size_int, 1500, 1])
	        view_240: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_90, [sym_size_int, 1500, 1])
	        reciprocal_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_239);  view_239 = None
	        mul_1529: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_15, 1.0);  reciprocal_15 = None
	        mul_1532: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_238, mul_1529);  view_238 = mul_1529 = None
	        round_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1532);  mul_1532 = None
	        add_2419: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_32, view_240);  round_32 = view_240 = None
	        clamp_min_47: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2419, -128);  add_2419 = None
	        clamp_max_31: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_47, 127);  clamp_min_47 = None
	        view_241: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_31, [sym_size_int, 1500, 1280]);  clamp_max_31 = None
	        _assert_tensor_metadata_139 = torch.ops.aten._assert_tensor_metadata.default(view_241, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_139 = None
	        convert_element_type_91: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_241, torch.int8);  view_241 = None
	        view_242: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_91, [sym_size_int, 1500, 1280]);  convert_element_type_91 = None
	        view_243: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_45, [sym_size_int, 1500, 1]);  clamp_min_45 = None
	        view_244: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_90, [sym_size_int, 1500, 1]);  convert_element_type_90 = None
	        _assert_tensor_metadata_140 = torch.ops.aten._assert_tensor_metadata.default(view_242, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_140 = None
	        convert_element_type_92: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_242, torch.float32);  view_242 = None
	        _assert_tensor_metadata_141 = torch.ops.aten._assert_tensor_metadata.default(view_244, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_141 = None
	        convert_element_type_93: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_244, torch.float32);  view_244 = None
	        sub_729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_92, convert_element_type_93);  convert_element_type_92 = convert_element_type_93 = None
	        mul_1554: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_729, view_243);  sub_729 = view_243 = None
	        view_245: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1554, [sym_size_int, 1500, 1280]);  mul_1554 = None
	        _assert_tensor_metadata_142 = torch.ops.aten._assert_tensor_metadata.default(view_245, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_142 = None
	        view_246: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_247: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_248: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_143 = torch.ops.aten._assert_tensor_metadata.default(view_246, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_143 = None
	        convert_element_type_94: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_246, torch.float32);  view_246 = None
	        _assert_tensor_metadata_144 = torch.ops.aten._assert_tensor_metadata.default(view_248, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_144 = None
	        convert_element_type_95: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_248, torch.float32);  view_248 = None
	        sub_733: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_94, convert_element_type_95);  convert_element_type_94 = convert_element_type_95 = None
	        mul_1559: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_733, view_247);  sub_733 = view_247 = None
	        view_249: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1559, [1280, 1280]);  mul_1559 = None
	        _assert_tensor_metadata_145 = torch.ops.aten._assert_tensor_metadata.default(view_249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_145 = None
	        mul_1564: "Sym(1500*s6)" = sym_size_int * 1500
	        view_250: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_245, [mul_1564, 1280]);  view_245 = mul_1564 = None
	        permute_28: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_249, [1, 0]);  view_249 = None
	        addmm_12: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_2_self_attn_out_proj_bias, view_250, permute_28);  model_audio_tower_layers_2_self_attn_out_proj_bias = view_250 = permute_28 = None
	        view_251: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_12, [sym_size_int, 1500, 1280]);  addmm_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_251);  view_251 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_2482: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_1862, clone_21);  add_1862 = clone_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_22: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_2482, memory_format = torch.contiguous_format)
	        var_mean_5 = torch.ops.aten.var_mean.correction(clone_22, [2], correction = 0, keepdim = True)
	        getitem_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_5[0]
	        getitem_23: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_5[1];  var_mean_5 = None
	        add_2487: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_22, 1e-05);  getitem_22 = None
	        rsqrt_5: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_2487);  add_2487 = None
	        sub_739: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_22, getitem_23);  clone_22 = getitem_23 = None
	        mul_1575: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_739, rsqrt_5);  sub_739 = rsqrt_5 = None
	        mul_1576: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1575, model_audio_tower_layers_2_final_layer_norm_weight);  mul_1575 = model_audio_tower_layers_2_final_layer_norm_weight = None
	        add_2488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_1576, model_audio_tower_layers_2_final_layer_norm_bias);  mul_1576 = model_audio_tower_layers_2_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_252: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_2488, [sym_size_int, 1500, 1280])
	        amin_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_252, [2])
	        amax_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_252, [2]);  view_252 = None
	        full_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_16, full_32);  amin_16 = full_32 = None
	        full_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_16, full_33);  amax_16 = full_33 = None
	        sub_750: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_16, minimum_16);  maximum_16 = None
	        div_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_750, 255.0);  sub_750 = None
	        clamp_min_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_32, 1.1920928955078125e-07);  div_32 = None
	        div_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_16, clamp_min_48);  minimum_16 = None
	        round_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_33);  div_33 = None
	        sub_756: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_33);  round_33 = None
	        clamp_min_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_756, -128);  sub_756 = None
	        clamp_max_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_49, 127);  clamp_min_49 = None
	        _assert_tensor_metadata_146 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_48, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_146 = None
	        _assert_tensor_metadata_147 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_32, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_147 = None
	        convert_element_type_96: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_32, torch.int8);  clamp_max_32 = None
	        view_253: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_2488, [sym_size_int, 1500, 1280]);  add_2488 = None
	        view_254: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_48, [sym_size_int, 1500, 1])
	        view_255: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_96, [sym_size_int, 1500, 1])
	        reciprocal_16: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_254);  view_254 = None
	        mul_1624: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_16, 1.0);  reciprocal_16 = None
	        mul_1627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_253, mul_1624);  view_253 = mul_1624 = None
	        round_34: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1627);  mul_1627 = None
	        add_2575: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_34, view_255);  round_34 = view_255 = None
	        clamp_min_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2575, -128);  add_2575 = None
	        clamp_max_33: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_50, 127);  clamp_min_50 = None
	        view_256: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_33, [sym_size_int, 1500, 1280]);  clamp_max_33 = None
	        _assert_tensor_metadata_148 = torch.ops.aten._assert_tensor_metadata.default(view_256, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_148 = None
	        convert_element_type_97: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_256, torch.int8);  view_256 = None
	        view_257: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_97, [sym_size_int, 1500, 1280]);  convert_element_type_97 = None
	        view_258: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_48, [sym_size_int, 1500, 1]);  clamp_min_48 = None
	        view_259: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_96, [sym_size_int, 1500, 1]);  convert_element_type_96 = None
	        _assert_tensor_metadata_149 = torch.ops.aten._assert_tensor_metadata.default(view_257, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_149 = None
	        convert_element_type_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_257, torch.float32);  view_257 = None
	        _assert_tensor_metadata_150 = torch.ops.aten._assert_tensor_metadata.default(view_259, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_150 = None
	        convert_element_type_99: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_259, torch.float32);  view_259 = None
	        sub_776: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_98, convert_element_type_99);  convert_element_type_98 = convert_element_type_99 = None
	        mul_1649: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_776, view_258);  sub_776 = view_258 = None
	        view_260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1649, [sym_size_int, 1500, 1280]);  mul_1649 = None
	        _assert_tensor_metadata_151 = torch.ops.aten._assert_tensor_metadata.default(view_260, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_151 = None
	        view_261: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_2_fc1_parametrizations_weight_original0 = None
	        view_262: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_2_fc1_parametrizations_weight_original1 = None
	        view_263: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_2_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_152 = torch.ops.aten._assert_tensor_metadata.default(view_261, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_152 = None
	        convert_element_type_100: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_261, torch.float32);  view_261 = None
	        _assert_tensor_metadata_153 = torch.ops.aten._assert_tensor_metadata.default(view_263, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_153 = None
	        convert_element_type_101: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_263, torch.float32);  view_263 = None
	        sub_780: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_100, convert_element_type_101);  convert_element_type_100 = convert_element_type_101 = None
	        mul_1654: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_780, view_262);  sub_780 = view_262 = None
	        view_264: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1654, [5120, 1280]);  mul_1654 = None
	        _assert_tensor_metadata_154 = torch.ops.aten._assert_tensor_metadata.default(view_264, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_154 = None
	        mul_1659: "Sym(1500*s6)" = sym_size_int * 1500
	        view_265: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_260, [mul_1659, 1280]);  view_260 = mul_1659 = None
	        permute_29: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_264, [1, 0]);  view_264 = None
	        addmm_13: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_2_fc1_bias, view_265, permute_29);  model_audio_tower_layers_2_fc1_bias = view_265 = permute_29 = None
	        view_266: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_13, [sym_size_int, 1500, 5120]);  addmm_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_1666: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_266, 0.5)
	        mul_1667: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_266, 0.7071067811865476);  view_266 = None
	        erf_4: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_1667);  mul_1667 = None
	        add_2634: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_4, 1);  erf_4 = None
	        mul_1668: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1666, add_2634);  mul_1666 = add_2634 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_23: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_1668);  mul_1668 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_267: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_23, [sym_size_int, 1500, 5120])
	        amin_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_267, [2])
	        amax_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_267, [2]);  view_267 = None
	        full_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_17, full_34);  amin_17 = full_34 = None
	        full_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_17, full_35);  amax_17 = full_35 = None
	        sub_793: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_17, minimum_17);  maximum_17 = None
	        div_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_793, 255.0);  sub_793 = None
	        clamp_min_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_34, 1.1920928955078125e-07);  div_34 = None
	        div_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_17, clamp_min_51);  minimum_17 = None
	        round_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_35);  div_35 = None
	        sub_799: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_35);  round_35 = None
	        clamp_min_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_799, -128);  sub_799 = None
	        clamp_max_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_52, 127);  clamp_min_52 = None
	        _assert_tensor_metadata_155 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_51, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_155 = None
	        _assert_tensor_metadata_156 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_34, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_156 = None
	        convert_element_type_102: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_34, torch.int8);  clamp_max_34 = None
	        view_268: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_23, [sym_size_int, 1500, 5120]);  clone_23 = None
	        view_269: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_51, [sym_size_int, 1500, 1])
	        view_270: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_102, [sym_size_int, 1500, 1])
	        reciprocal_17: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_269);  view_269 = None
	        mul_1714: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_17, 1.0);  reciprocal_17 = None
	        mul_1717: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_268, mul_1714);  view_268 = mul_1714 = None
	        round_36: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_1717);  mul_1717 = None
	        add_2717: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_36, view_270);  round_36 = view_270 = None
	        clamp_min_53: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2717, -128);  add_2717 = None
	        clamp_max_35: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_53, 127);  clamp_min_53 = None
	        view_271: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_35, [sym_size_int, 1500, 5120]);  clamp_max_35 = None
	        _assert_tensor_metadata_157 = torch.ops.aten._assert_tensor_metadata.default(view_271, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_157 = None
	        convert_element_type_103: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_271, torch.int8);  view_271 = None
	        view_272: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_103, [sym_size_int, 1500, 5120]);  convert_element_type_103 = None
	        view_273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_51, [sym_size_int, 1500, 1]);  clamp_min_51 = None
	        view_274: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_102, [sym_size_int, 1500, 1]);  convert_element_type_102 = None
	        _assert_tensor_metadata_158 = torch.ops.aten._assert_tensor_metadata.default(view_272, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_158 = None
	        convert_element_type_104: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_272, torch.float32);  view_272 = None
	        _assert_tensor_metadata_159 = torch.ops.aten._assert_tensor_metadata.default(view_274, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_159 = None
	        convert_element_type_105: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_274, torch.float32);  view_274 = None
	        sub_819: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_104, convert_element_type_105);  convert_element_type_104 = convert_element_type_105 = None
	        mul_1739: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_819, view_273);  sub_819 = view_273 = None
	        view_275: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_1739, [sym_size_int, 1500, 5120]);  mul_1739 = None
	        _assert_tensor_metadata_160 = torch.ops.aten._assert_tensor_metadata.default(view_275, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_160 = None
	        view_276: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_2_fc2_parametrizations_weight_original0 = None
	        view_277: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_2_fc2_parametrizations_weight_original1 = None
	        view_278: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_2_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_161 = torch.ops.aten._assert_tensor_metadata.default(view_276, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_161 = None
	        convert_element_type_106: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_276, torch.float32);  view_276 = None
	        _assert_tensor_metadata_162 = torch.ops.aten._assert_tensor_metadata.default(view_278, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_162 = None
	        convert_element_type_107: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_278, torch.float32);  view_278 = None
	        sub_823: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_106, convert_element_type_107);  convert_element_type_106 = convert_element_type_107 = None
	        mul_1744: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_823, view_277);  sub_823 = view_277 = None
	        view_279: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_1744, [1280, 5120]);  mul_1744 = None
	        _assert_tensor_metadata_163 = torch.ops.aten._assert_tensor_metadata.default(view_279, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_163 = None
	        mul_1749: "Sym(1500*s6)" = sym_size_int * 1500
	        view_280: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_275, [mul_1749, 5120]);  view_275 = mul_1749 = None
	        permute_30: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_279, [1, 0]);  view_279 = None
	        addmm_14: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_2_fc2_bias, view_280, permute_30);  model_audio_tower_layers_2_fc2_bias = view_280 = permute_30 = None
	        view_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_14, [sym_size_int, 1500, 1280]);  addmm_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_24: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_281);  view_281 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_2780: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_2482, clone_24);  add_2482 = clone_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_25: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_2780, memory_format = torch.contiguous_format)
	        var_mean_6 = torch.ops.aten.var_mean.correction(clone_25, [2], correction = 0, keepdim = True)
	        getitem_24: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_6[0]
	        getitem_25: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_6[1];  var_mean_6 = None
	        add_2785: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_24, 1e-05);  getitem_24 = None
	        rsqrt_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_2785);  add_2785 = None
	        sub_829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_25, getitem_25);  clone_25 = getitem_25 = None
	        mul_1760: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_829, rsqrt_6);  sub_829 = rsqrt_6 = None
	        mul_1761: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1760, model_audio_tower_layers_3_self_attn_layer_norm_weight);  mul_1760 = model_audio_tower_layers_3_self_attn_layer_norm_weight = None
	        add_2786: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_1761, model_audio_tower_layers_3_self_attn_layer_norm_bias);  mul_1761 = model_audio_tower_layers_3_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_282: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_2786, [sym_size_int, 1500, 1280])
	        amin_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_282, [2])
	        amax_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_282, [2]);  view_282 = None
	        full_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_18, full_36);  amin_18 = full_36 = None
	        full_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_18, full_37);  amax_18 = full_37 = None
	        sub_840: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_18, minimum_18);  maximum_18 = None
	        div_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_840, 255.0);  sub_840 = None
	        clamp_min_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_36, 1.1920928955078125e-07);  div_36 = None
	        div_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_18, clamp_min_54);  minimum_18 = None
	        round_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_37);  div_37 = None
	        sub_846: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_37);  round_37 = None
	        clamp_min_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_846, -128);  sub_846 = None
	        clamp_max_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_55, 127);  clamp_min_55 = None
	        _assert_tensor_metadata_164 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_54, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_164 = None
	        _assert_tensor_metadata_165 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_36, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_165 = None
	        convert_element_type_108: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_36, torch.int8);  clamp_max_36 = None
	        view_283: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_2786, [sym_size_int, 1500, 1280])
	        view_284: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_54, [sym_size_int, 1500, 1])
	        view_285: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_108, [sym_size_int, 1500, 1])
	        reciprocal_18: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_284);  view_284 = None
	        mul_1809: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_18, 1.0);  reciprocal_18 = None
	        mul_1812: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_283, mul_1809);  view_283 = mul_1809 = None
	        round_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1812);  mul_1812 = None
	        add_2873: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_38, view_285);  round_38 = view_285 = None
	        clamp_min_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2873, -128);  add_2873 = None
	        clamp_max_37: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_56, 127);  clamp_min_56 = None
	        view_286: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_37, [sym_size_int, 1500, 1280]);  clamp_max_37 = None
	        _assert_tensor_metadata_166 = torch.ops.aten._assert_tensor_metadata.default(view_286, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_166 = None
	        convert_element_type_109: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_286, torch.int8);  view_286 = None
	        view_287: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_109, [sym_size_int, 1500, 1280]);  convert_element_type_109 = None
	        view_288: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_54, [sym_size_int, 1500, 1]);  clamp_min_54 = None
	        view_289: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_108, [sym_size_int, 1500, 1]);  convert_element_type_108 = None
	        _assert_tensor_metadata_167 = torch.ops.aten._assert_tensor_metadata.default(view_287, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_167 = None
	        convert_element_type_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_287, torch.float32);  view_287 = None
	        _assert_tensor_metadata_168 = torch.ops.aten._assert_tensor_metadata.default(view_289, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_168 = None
	        convert_element_type_111: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_289, torch.float32);  view_289 = None
	        sub_866: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_110, convert_element_type_111);  convert_element_type_110 = convert_element_type_111 = None
	        mul_1834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_866, view_288);  sub_866 = view_288 = None
	        view_290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1834, [sym_size_int, 1500, 1280]);  mul_1834 = None
	        _assert_tensor_metadata_169 = torch.ops.aten._assert_tensor_metadata.default(view_290, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_169 = None
	        view_291: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_292: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_293: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_170 = torch.ops.aten._assert_tensor_metadata.default(view_291, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_170 = None
	        convert_element_type_112: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_291, torch.float32);  view_291 = None
	        _assert_tensor_metadata_171 = torch.ops.aten._assert_tensor_metadata.default(view_293, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_171 = None
	        convert_element_type_113: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_293, torch.float32);  view_293 = None
	        sub_870: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_112, convert_element_type_113);  convert_element_type_112 = convert_element_type_113 = None
	        mul_1839: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_870, view_292);  sub_870 = view_292 = None
	        view_294: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1839, [1280, 1280]);  mul_1839 = None
	        _assert_tensor_metadata_172 = torch.ops.aten._assert_tensor_metadata.default(view_294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_172 = None
	        mul_1844: "Sym(1500*s6)" = sym_size_int * 1500
	        view_295: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_290, [mul_1844, 1280]);  view_290 = mul_1844 = None
	        permute_31: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_294, [1, 0]);  view_294 = None
	        addmm_15: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_3_self_attn_q_proj_bias, view_295, permute_31);  model_audio_tower_layers_3_self_attn_q_proj_bias = view_295 = permute_31 = None
	        view_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_15, [sym_size_int, 1500, 1280]);  addmm_15 = None
	        mul_1851: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_296, 0.125);  view_296 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_297: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_1851, [sym_size_int, 1500, 20, 64]);  mul_1851 = None
	        permute_32: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_297, [0, 2, 1, 3]);  view_297 = None
	        clone_26: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_32, memory_format = torch.contiguous_format);  permute_32 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_298: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_2786, [sym_size_int, 1500, 1280])
	        amin_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_298, [2])
	        amax_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_298, [2]);  view_298 = None
	        full_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_19, full_38);  amin_19 = full_38 = None
	        full_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_19, full_39);  amax_19 = full_39 = None
	        sub_885: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_19, minimum_19);  maximum_19 = None
	        div_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_885, 255.0);  sub_885 = None
	        clamp_min_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_38, 1.1920928955078125e-07);  div_38 = None
	        div_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_19, clamp_min_57);  minimum_19 = None
	        round_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_39);  div_39 = None
	        sub_891: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_39);  round_39 = None
	        clamp_min_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_891, -128);  sub_891 = None
	        clamp_max_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_58, 127);  clamp_min_58 = None
	        _assert_tensor_metadata_173 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_57, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_173 = None
	        _assert_tensor_metadata_174 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_38, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_174 = None
	        convert_element_type_114: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_38, torch.int8);  clamp_max_38 = None
	        view_299: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_2786, [sym_size_int, 1500, 1280])
	        view_300: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_57, [sym_size_int, 1500, 1])
	        view_301: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_114, [sym_size_int, 1500, 1])
	        reciprocal_19: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_300);  view_300 = None
	        mul_1905: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_19, 1.0);  reciprocal_19 = None
	        mul_1908: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_299, mul_1905);  view_299 = mul_1905 = None
	        round_40: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1908);  mul_1908 = None
	        add_3025: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_40, view_301);  round_40 = view_301 = None
	        clamp_min_59: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3025, -128);  add_3025 = None
	        clamp_max_39: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_59, 127);  clamp_min_59 = None
	        view_302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_39, [sym_size_int, 1500, 1280]);  clamp_max_39 = None
	        _assert_tensor_metadata_175 = torch.ops.aten._assert_tensor_metadata.default(view_302, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_175 = None
	        convert_element_type_115: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_302, torch.int8);  view_302 = None
	        view_303: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_115, [sym_size_int, 1500, 1280]);  convert_element_type_115 = None
	        view_304: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_57, [sym_size_int, 1500, 1]);  clamp_min_57 = None
	        view_305: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_114, [sym_size_int, 1500, 1]);  convert_element_type_114 = None
	        _assert_tensor_metadata_176 = torch.ops.aten._assert_tensor_metadata.default(view_303, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_176 = None
	        convert_element_type_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_303, torch.float32);  view_303 = None
	        _assert_tensor_metadata_177 = torch.ops.aten._assert_tensor_metadata.default(view_305, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_177 = None
	        convert_element_type_117: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_305, torch.float32);  view_305 = None
	        sub_911: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_116, convert_element_type_117);  convert_element_type_116 = convert_element_type_117 = None
	        mul_1930: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_911, view_304);  sub_911 = view_304 = None
	        view_306: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1930, [sym_size_int, 1500, 1280]);  mul_1930 = None
	        _assert_tensor_metadata_178 = torch.ops.aten._assert_tensor_metadata.default(view_306, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_178 = None
	        view_307: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_308: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_309: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_179 = torch.ops.aten._assert_tensor_metadata.default(view_307, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_179 = None
	        convert_element_type_118: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_307, torch.float32);  view_307 = None
	        _assert_tensor_metadata_180 = torch.ops.aten._assert_tensor_metadata.default(view_309, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_180 = None
	        convert_element_type_119: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_309, torch.float32);  view_309 = None
	        sub_915: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_118, convert_element_type_119);  convert_element_type_118 = convert_element_type_119 = None
	        mul_1935: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_915, view_308);  sub_915 = view_308 = None
	        view_310: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1935, [1280, 1280]);  mul_1935 = None
	        _assert_tensor_metadata_181 = torch.ops.aten._assert_tensor_metadata.default(view_310, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_181 = None
	        permute_33: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_310, [1, 0]);  view_310 = None
	        mul_1938: "Sym(1500*s6)" = sym_size_int * 1500
	        view_311: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_306, [mul_1938, 1280]);  view_306 = mul_1938 = None
	        mm_3: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_311, permute_33);  view_311 = permute_33 = None
	        view_312: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_3, [sym_size_int, 1500, 1280]);  mm_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_313: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_312, [sym_size_int, -1, 20, 64]);  view_312 = None
	        permute_34: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_313, [0, 2, 1, 3]);  view_313 = None
	        clone_27: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_34, memory_format = torch.contiguous_format);  permute_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_2786, [sym_size_int, 1500, 1280])
	        amin_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_314, [2])
	        amax_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_314, [2]);  view_314 = None
	        full_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_20, full_40);  amin_20 = full_40 = None
	        full_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_20, full_41);  amax_20 = full_41 = None
	        sub_929: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_20, minimum_20);  maximum_20 = None
	        div_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_929, 255.0);  sub_929 = None
	        clamp_min_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_40, 1.1920928955078125e-07);  div_40 = None
	        div_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_20, clamp_min_60);  minimum_20 = None
	        round_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_41);  div_41 = None
	        sub_935: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_41);  round_41 = None
	        clamp_min_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_935, -128);  sub_935 = None
	        clamp_max_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_61, 127);  clamp_min_61 = None
	        _assert_tensor_metadata_182 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_60, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_182 = None
	        _assert_tensor_metadata_183 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_40, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_183 = None
	        convert_element_type_120: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_40, torch.int8);  clamp_max_40 = None
	        view_315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_2786, [sym_size_int, 1500, 1280]);  add_2786 = None
	        view_316: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_60, [sym_size_int, 1500, 1])
	        view_317: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_120, [sym_size_int, 1500, 1])
	        reciprocal_20: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_316);  view_316 = None
	        mul_2004: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_20, 1.0);  reciprocal_20 = None
	        mul_2007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_315, mul_2004);  view_315 = mul_2004 = None
	        round_42: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2007);  mul_2007 = None
	        add_3173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_42, view_317);  round_42 = view_317 = None
	        clamp_min_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3173, -128);  add_3173 = None
	        clamp_max_41: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_62, 127);  clamp_min_62 = None
	        view_318: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_41, [sym_size_int, 1500, 1280]);  clamp_max_41 = None
	        _assert_tensor_metadata_184 = torch.ops.aten._assert_tensor_metadata.default(view_318, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_184 = None
	        convert_element_type_121: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_318, torch.int8);  view_318 = None
	        view_319: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_121, [sym_size_int, 1500, 1280]);  convert_element_type_121 = None
	        view_320: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_60, [sym_size_int, 1500, 1]);  clamp_min_60 = None
	        view_321: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_120, [sym_size_int, 1500, 1]);  convert_element_type_120 = None
	        _assert_tensor_metadata_185 = torch.ops.aten._assert_tensor_metadata.default(view_319, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_185 = None
	        convert_element_type_122: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_319, torch.float32);  view_319 = None
	        _assert_tensor_metadata_186 = torch.ops.aten._assert_tensor_metadata.default(view_321, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_186 = None
	        convert_element_type_123: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_321, torch.float32);  view_321 = None
	        sub_955: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_122, convert_element_type_123);  convert_element_type_122 = convert_element_type_123 = None
	        mul_2029: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_955, view_320);  sub_955 = view_320 = None
	        view_322: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2029, [sym_size_int, 1500, 1280]);  mul_2029 = None
	        _assert_tensor_metadata_187 = torch.ops.aten._assert_tensor_metadata.default(view_322, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_187 = None
	        view_323: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_324: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_325: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_188 = torch.ops.aten._assert_tensor_metadata.default(view_323, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_188 = None
	        convert_element_type_124: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_323, torch.float32);  view_323 = None
	        _assert_tensor_metadata_189 = torch.ops.aten._assert_tensor_metadata.default(view_325, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_189 = None
	        convert_element_type_125: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_325, torch.float32);  view_325 = None
	        sub_959: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_124, convert_element_type_125);  convert_element_type_124 = convert_element_type_125 = None
	        mul_2034: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_959, view_324);  sub_959 = view_324 = None
	        view_326: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2034, [1280, 1280]);  mul_2034 = None
	        _assert_tensor_metadata_190 = torch.ops.aten._assert_tensor_metadata.default(view_326, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_190 = None
	        mul_2039: "Sym(1500*s6)" = sym_size_int * 1500
	        view_327: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_322, [mul_2039, 1280]);  view_322 = mul_2039 = None
	        permute_35: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_326, [1, 0]);  view_326 = None
	        addmm_16: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_3_self_attn_v_proj_bias, view_327, permute_35);  model_audio_tower_layers_3_self_attn_v_proj_bias = view_327 = permute_35 = None
	        view_328: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_16, [sym_size_int, 1500, 1280]);  addmm_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_329: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_328, [sym_size_int, -1, 20, 64]);  view_328 = None
	        permute_36: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_329, [0, 2, 1, 3]);  view_329 = None
	        clone_28: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_36, memory_format = torch.contiguous_format);  permute_36 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_3 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_26, clone_27, clone_28, None, False, scale = 1.0);  clone_26 = clone_27 = clone_28 = None
	        getitem_26: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_3[0];  _scaled_dot_product_efficient_attention_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_37: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_26, [0, 2, 1, 3]);  getitem_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_330: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_37, [sym_size_int, 1500, -1]);  permute_37 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_331: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_330, [sym_size_int, 1500, 1280])
	        amin_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_331, [2])
	        amax_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_331, [2]);  view_331 = None
	        full_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_21, full_42);  amin_21 = full_42 = None
	        full_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_21, full_43);  amax_21 = full_43 = None
	        sub_977: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_21, minimum_21);  maximum_21 = None
	        div_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_977, 255.0);  sub_977 = None
	        clamp_min_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_42, 1.1920928955078125e-07);  div_42 = None
	        div_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_21, clamp_min_63);  minimum_21 = None
	        round_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_43);  div_43 = None
	        sub_983: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_43);  round_43 = None
	        clamp_min_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_983, -128);  sub_983 = None
	        clamp_max_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_64, 127);  clamp_min_64 = None
	        _assert_tensor_metadata_191 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_63, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_191 = None
	        _assert_tensor_metadata_192 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_42, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_192 = None
	        convert_element_type_126: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_42, torch.int8);  clamp_max_42 = None
	        view_332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_330, [sym_size_int, 1500, 1280]);  view_330 = None
	        view_333: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_63, [sym_size_int, 1500, 1])
	        view_334: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_126, [sym_size_int, 1500, 1])
	        reciprocal_21: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_333);  view_333 = None
	        mul_2109: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_21, 1.0);  reciprocal_21 = None
	        mul_2112: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_332, mul_2109);  view_332 = mul_2109 = None
	        round_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2112);  mul_2112 = None
	        add_3337: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_44, view_334);  round_44 = view_334 = None
	        clamp_min_65: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3337, -128);  add_3337 = None
	        clamp_max_43: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_65, 127);  clamp_min_65 = None
	        view_335: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_43, [sym_size_int, 1500, 1280]);  clamp_max_43 = None
	        _assert_tensor_metadata_193 = torch.ops.aten._assert_tensor_metadata.default(view_335, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_193 = None
	        convert_element_type_127: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_335, torch.int8);  view_335 = None
	        view_336: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_127, [sym_size_int, 1500, 1280]);  convert_element_type_127 = None
	        view_337: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_63, [sym_size_int, 1500, 1]);  clamp_min_63 = None
	        view_338: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_126, [sym_size_int, 1500, 1]);  convert_element_type_126 = None
	        _assert_tensor_metadata_194 = torch.ops.aten._assert_tensor_metadata.default(view_336, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_194 = None
	        convert_element_type_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_336, torch.float32);  view_336 = None
	        _assert_tensor_metadata_195 = torch.ops.aten._assert_tensor_metadata.default(view_338, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_195 = None
	        convert_element_type_129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_338, torch.float32);  view_338 = None
	        sub_1003: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_128, convert_element_type_129);  convert_element_type_128 = convert_element_type_129 = None
	        mul_2134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1003, view_337);  sub_1003 = view_337 = None
	        view_339: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2134, [sym_size_int, 1500, 1280]);  mul_2134 = None
	        _assert_tensor_metadata_196 = torch.ops.aten._assert_tensor_metadata.default(view_339, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_196 = None
	        view_340: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_341: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_342: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_197 = torch.ops.aten._assert_tensor_metadata.default(view_340, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_197 = None
	        convert_element_type_130: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_340, torch.float32);  view_340 = None
	        _assert_tensor_metadata_198 = torch.ops.aten._assert_tensor_metadata.default(view_342, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_198 = None
	        convert_element_type_131: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_342, torch.float32);  view_342 = None
	        sub_1007: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_130, convert_element_type_131);  convert_element_type_130 = convert_element_type_131 = None
	        mul_2139: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1007, view_341);  sub_1007 = view_341 = None
	        view_343: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2139, [1280, 1280]);  mul_2139 = None
	        _assert_tensor_metadata_199 = torch.ops.aten._assert_tensor_metadata.default(view_343, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_199 = None
	        mul_2144: "Sym(1500*s6)" = sym_size_int * 1500
	        view_344: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_339, [mul_2144, 1280]);  view_339 = mul_2144 = None
	        permute_38: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_343, [1, 0]);  view_343 = None
	        addmm_17: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_3_self_attn_out_proj_bias, view_344, permute_38);  model_audio_tower_layers_3_self_attn_out_proj_bias = view_344 = permute_38 = None
	        view_345: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_17, [sym_size_int, 1500, 1280]);  addmm_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_29: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_345);  view_345 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_3400: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_2780, clone_29);  add_2780 = clone_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_3400, memory_format = torch.contiguous_format)
	        var_mean_7 = torch.ops.aten.var_mean.correction(clone_30, [2], correction = 0, keepdim = True)
	        getitem_30: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_7[0]
	        getitem_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_7[1];  var_mean_7 = None
	        add_3405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_30, 1e-05);  getitem_30 = None
	        rsqrt_7: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_3405);  add_3405 = None
	        sub_1013: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_30, getitem_31);  clone_30 = getitem_31 = None
	        mul_2155: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1013, rsqrt_7);  sub_1013 = rsqrt_7 = None
	        mul_2156: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2155, model_audio_tower_layers_3_final_layer_norm_weight);  mul_2155 = model_audio_tower_layers_3_final_layer_norm_weight = None
	        add_3406: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2156, model_audio_tower_layers_3_final_layer_norm_bias);  mul_2156 = model_audio_tower_layers_3_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_346: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_3406, [sym_size_int, 1500, 1280])
	        amin_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_346, [2])
	        amax_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_346, [2]);  view_346 = None
	        full_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_22, full_44);  amin_22 = full_44 = None
	        full_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_22, full_45);  amax_22 = full_45 = None
	        sub_1024: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_22, minimum_22);  maximum_22 = None
	        div_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1024, 255.0);  sub_1024 = None
	        clamp_min_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_44, 1.1920928955078125e-07);  div_44 = None
	        div_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_22, clamp_min_66);  minimum_22 = None
	        round_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_45);  div_45 = None
	        sub_1030: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_45);  round_45 = None
	        clamp_min_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1030, -128);  sub_1030 = None
	        clamp_max_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_67, 127);  clamp_min_67 = None
	        _assert_tensor_metadata_200 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_66, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_200 = None
	        _assert_tensor_metadata_201 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_44, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_201 = None
	        convert_element_type_132: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_44, torch.int8);  clamp_max_44 = None
	        view_347: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_3406, [sym_size_int, 1500, 1280]);  add_3406 = None
	        view_348: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_66, [sym_size_int, 1500, 1])
	        view_349: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_132, [sym_size_int, 1500, 1])
	        reciprocal_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_348);  view_348 = None
	        mul_2204: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_22, 1.0);  reciprocal_22 = None
	        mul_2207: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_347, mul_2204);  view_347 = mul_2204 = None
	        round_46: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2207);  mul_2207 = None
	        add_3493: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_46, view_349);  round_46 = view_349 = None
	        clamp_min_68: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3493, -128);  add_3493 = None
	        clamp_max_45: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_68, 127);  clamp_min_68 = None
	        view_350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_45, [sym_size_int, 1500, 1280]);  clamp_max_45 = None
	        _assert_tensor_metadata_202 = torch.ops.aten._assert_tensor_metadata.default(view_350, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_202 = None
	        convert_element_type_133: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_350, torch.int8);  view_350 = None
	        view_351: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_133, [sym_size_int, 1500, 1280]);  convert_element_type_133 = None
	        view_352: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_66, [sym_size_int, 1500, 1]);  clamp_min_66 = None
	        view_353: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_132, [sym_size_int, 1500, 1]);  convert_element_type_132 = None
	        _assert_tensor_metadata_203 = torch.ops.aten._assert_tensor_metadata.default(view_351, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_203 = None
	        convert_element_type_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_351, torch.float32);  view_351 = None
	        _assert_tensor_metadata_204 = torch.ops.aten._assert_tensor_metadata.default(view_353, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_204 = None
	        convert_element_type_135: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_353, torch.float32);  view_353 = None
	        sub_1050: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_134, convert_element_type_135);  convert_element_type_134 = convert_element_type_135 = None
	        mul_2229: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1050, view_352);  sub_1050 = view_352 = None
	        view_354: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2229, [sym_size_int, 1500, 1280]);  mul_2229 = None
	        _assert_tensor_metadata_205 = torch.ops.aten._assert_tensor_metadata.default(view_354, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_205 = None
	        view_355: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_3_fc1_parametrizations_weight_original0 = None
	        view_356: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_3_fc1_parametrizations_weight_original1 = None
	        view_357: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_3_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_206 = torch.ops.aten._assert_tensor_metadata.default(view_355, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_206 = None
	        convert_element_type_136: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_355, torch.float32);  view_355 = None
	        _assert_tensor_metadata_207 = torch.ops.aten._assert_tensor_metadata.default(view_357, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_207 = None
	        convert_element_type_137: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_357, torch.float32);  view_357 = None
	        sub_1054: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_136, convert_element_type_137);  convert_element_type_136 = convert_element_type_137 = None
	        mul_2234: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1054, view_356);  sub_1054 = view_356 = None
	        view_358: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2234, [5120, 1280]);  mul_2234 = None
	        _assert_tensor_metadata_208 = torch.ops.aten._assert_tensor_metadata.default(view_358, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_208 = None
	        mul_2239: "Sym(1500*s6)" = sym_size_int * 1500
	        view_359: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_354, [mul_2239, 1280]);  view_354 = mul_2239 = None
	        permute_39: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_358, [1, 0]);  view_358 = None
	        addmm_18: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_3_fc1_bias, view_359, permute_39);  model_audio_tower_layers_3_fc1_bias = view_359 = permute_39 = None
	        view_360: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_18, [sym_size_int, 1500, 5120]);  addmm_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_2246: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_360, 0.5)
	        mul_2247: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_360, 0.7071067811865476);  view_360 = None
	        erf_5: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_2247);  mul_2247 = None
	        add_3552: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_5, 1);  erf_5 = None
	        mul_2248: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2246, add_3552);  mul_2246 = add_3552 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_31: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_2248);  mul_2248 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_361: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_31, [sym_size_int, 1500, 5120])
	        amin_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_361, [2])
	        amax_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_361, [2]);  view_361 = None
	        full_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_23, full_46);  amin_23 = full_46 = None
	        full_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_23, full_47);  amax_23 = full_47 = None
	        sub_1067: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_23, minimum_23);  maximum_23 = None
	        div_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1067, 255.0);  sub_1067 = None
	        clamp_min_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_46, 1.1920928955078125e-07);  div_46 = None
	        div_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_23, clamp_min_69);  minimum_23 = None
	        round_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_47);  div_47 = None
	        sub_1073: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_47);  round_47 = None
	        clamp_min_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1073, -128);  sub_1073 = None
	        clamp_max_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_70, 127);  clamp_min_70 = None
	        _assert_tensor_metadata_209 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_69, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_209 = None
	        _assert_tensor_metadata_210 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_46, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_210 = None
	        convert_element_type_138: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_46, torch.int8);  clamp_max_46 = None
	        view_362: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_31, [sym_size_int, 1500, 5120]);  clone_31 = None
	        view_363: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_69, [sym_size_int, 1500, 1])
	        view_364: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_138, [sym_size_int, 1500, 1])
	        reciprocal_23: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_363);  view_363 = None
	        mul_2294: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_23, 1.0);  reciprocal_23 = None
	        mul_2297: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_362, mul_2294);  view_362 = mul_2294 = None
	        round_48: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_2297);  mul_2297 = None
	        add_3635: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_48, view_364);  round_48 = view_364 = None
	        clamp_min_71: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3635, -128);  add_3635 = None
	        clamp_max_47: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_71, 127);  clamp_min_71 = None
	        view_365: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_47, [sym_size_int, 1500, 5120]);  clamp_max_47 = None
	        _assert_tensor_metadata_211 = torch.ops.aten._assert_tensor_metadata.default(view_365, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_211 = None
	        convert_element_type_139: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_365, torch.int8);  view_365 = None
	        view_366: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_139, [sym_size_int, 1500, 5120]);  convert_element_type_139 = None
	        view_367: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_69, [sym_size_int, 1500, 1]);  clamp_min_69 = None
	        view_368: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_138, [sym_size_int, 1500, 1]);  convert_element_type_138 = None
	        _assert_tensor_metadata_212 = torch.ops.aten._assert_tensor_metadata.default(view_366, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_212 = None
	        convert_element_type_140: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_366, torch.float32);  view_366 = None
	        _assert_tensor_metadata_213 = torch.ops.aten._assert_tensor_metadata.default(view_368, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_213 = None
	        convert_element_type_141: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_368, torch.float32);  view_368 = None
	        sub_1093: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_140, convert_element_type_141);  convert_element_type_140 = convert_element_type_141 = None
	        mul_2319: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1093, view_367);  sub_1093 = view_367 = None
	        view_369: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_2319, [sym_size_int, 1500, 5120]);  mul_2319 = None
	        _assert_tensor_metadata_214 = torch.ops.aten._assert_tensor_metadata.default(view_369, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_214 = None
	        view_370: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_3_fc2_parametrizations_weight_original0 = None
	        view_371: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_3_fc2_parametrizations_weight_original1 = None
	        view_372: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_3_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_215 = torch.ops.aten._assert_tensor_metadata.default(view_370, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_215 = None
	        convert_element_type_142: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_370, torch.float32);  view_370 = None
	        _assert_tensor_metadata_216 = torch.ops.aten._assert_tensor_metadata.default(view_372, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_216 = None
	        convert_element_type_143: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_372, torch.float32);  view_372 = None
	        sub_1097: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_142, convert_element_type_143);  convert_element_type_142 = convert_element_type_143 = None
	        mul_2324: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1097, view_371);  sub_1097 = view_371 = None
	        view_373: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_2324, [1280, 5120]);  mul_2324 = None
	        _assert_tensor_metadata_217 = torch.ops.aten._assert_tensor_metadata.default(view_373, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_217 = None
	        mul_2329: "Sym(1500*s6)" = sym_size_int * 1500
	        view_374: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_369, [mul_2329, 5120]);  view_369 = mul_2329 = None
	        permute_40: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_373, [1, 0]);  view_373 = None
	        addmm_19: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_3_fc2_bias, view_374, permute_40);  model_audio_tower_layers_3_fc2_bias = view_374 = permute_40 = None
	        view_375: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_19, [sym_size_int, 1500, 1280]);  addmm_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_375);  view_375 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_3698: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_3400, clone_32);  add_3400 = clone_32 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_33: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_3698, memory_format = torch.contiguous_format)
	        var_mean_8 = torch.ops.aten.var_mean.correction(clone_33, [2], correction = 0, keepdim = True)
	        getitem_32: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_8[0]
	        getitem_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_8[1];  var_mean_8 = None
	        add_3703: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_32, 1e-05);  getitem_32 = None
	        rsqrt_8: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_3703);  add_3703 = None
	        sub_1103: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_33, getitem_33);  clone_33 = getitem_33 = None
	        mul_2340: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1103, rsqrt_8);  sub_1103 = rsqrt_8 = None
	        mul_2341: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2340, model_audio_tower_layers_4_self_attn_layer_norm_weight);  mul_2340 = model_audio_tower_layers_4_self_attn_layer_norm_weight = None
	        add_3704: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2341, model_audio_tower_layers_4_self_attn_layer_norm_bias);  mul_2341 = model_audio_tower_layers_4_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_376: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_3704, [sym_size_int, 1500, 1280])
	        amin_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_376, [2])
	        amax_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_376, [2]);  view_376 = None
	        full_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_24, full_48);  amin_24 = full_48 = None
	        full_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_24, full_49);  amax_24 = full_49 = None
	        sub_1114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_24, minimum_24);  maximum_24 = None
	        div_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1114, 255.0);  sub_1114 = None
	        clamp_min_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_48, 1.1920928955078125e-07);  div_48 = None
	        div_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_24, clamp_min_72);  minimum_24 = None
	        round_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_49);  div_49 = None
	        sub_1120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_49);  round_49 = None
	        clamp_min_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1120, -128);  sub_1120 = None
	        clamp_max_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_73, 127);  clamp_min_73 = None
	        _assert_tensor_metadata_218 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_72, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_218 = None
	        _assert_tensor_metadata_219 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_48, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_219 = None
	        convert_element_type_144: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_48, torch.int8);  clamp_max_48 = None
	        view_377: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_3704, [sym_size_int, 1500, 1280])
	        view_378: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_72, [sym_size_int, 1500, 1])
	        view_379: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_144, [sym_size_int, 1500, 1])
	        reciprocal_24: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_378);  view_378 = None
	        mul_2389: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_24, 1.0);  reciprocal_24 = None
	        mul_2392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_377, mul_2389);  view_377 = mul_2389 = None
	        round_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2392);  mul_2392 = None
	        add_3791: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_50, view_379);  round_50 = view_379 = None
	        clamp_min_74: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3791, -128);  add_3791 = None
	        clamp_max_49: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_74, 127);  clamp_min_74 = None
	        view_380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_49, [sym_size_int, 1500, 1280]);  clamp_max_49 = None
	        _assert_tensor_metadata_220 = torch.ops.aten._assert_tensor_metadata.default(view_380, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_220 = None
	        convert_element_type_145: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_380, torch.int8);  view_380 = None
	        view_381: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_145, [sym_size_int, 1500, 1280]);  convert_element_type_145 = None
	        view_382: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_72, [sym_size_int, 1500, 1]);  clamp_min_72 = None
	        view_383: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_144, [sym_size_int, 1500, 1]);  convert_element_type_144 = None
	        _assert_tensor_metadata_221 = torch.ops.aten._assert_tensor_metadata.default(view_381, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_221 = None
	        convert_element_type_146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_381, torch.float32);  view_381 = None
	        _assert_tensor_metadata_222 = torch.ops.aten._assert_tensor_metadata.default(view_383, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_222 = None
	        convert_element_type_147: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_383, torch.float32);  view_383 = None
	        sub_1140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_146, convert_element_type_147);  convert_element_type_146 = convert_element_type_147 = None
	        mul_2414: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1140, view_382);  sub_1140 = view_382 = None
	        view_384: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2414, [sym_size_int, 1500, 1280]);  mul_2414 = None
	        _assert_tensor_metadata_223 = torch.ops.aten._assert_tensor_metadata.default(view_384, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_223 = None
	        view_385: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_386: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_387: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_224 = torch.ops.aten._assert_tensor_metadata.default(view_385, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_224 = None
	        convert_element_type_148: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_385, torch.float32);  view_385 = None
	        _assert_tensor_metadata_225 = torch.ops.aten._assert_tensor_metadata.default(view_387, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_225 = None
	        convert_element_type_149: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_387, torch.float32);  view_387 = None
	        sub_1144: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_148, convert_element_type_149);  convert_element_type_148 = convert_element_type_149 = None
	        mul_2419: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1144, view_386);  sub_1144 = view_386 = None
	        view_388: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2419, [1280, 1280]);  mul_2419 = None
	        _assert_tensor_metadata_226 = torch.ops.aten._assert_tensor_metadata.default(view_388, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_226 = None
	        mul_2424: "Sym(1500*s6)" = sym_size_int * 1500
	        view_389: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_384, [mul_2424, 1280]);  view_384 = mul_2424 = None
	        permute_41: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_388, [1, 0]);  view_388 = None
	        addmm_20: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_4_self_attn_q_proj_bias, view_389, permute_41);  model_audio_tower_layers_4_self_attn_q_proj_bias = view_389 = permute_41 = None
	        view_390: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_20, [sym_size_int, 1500, 1280]);  addmm_20 = None
	        mul_2431: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_390, 0.125);  view_390 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_391: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_2431, [sym_size_int, 1500, 20, 64]);  mul_2431 = None
	        permute_42: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_391, [0, 2, 1, 3]);  view_391 = None
	        clone_34: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_42, memory_format = torch.contiguous_format);  permute_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_3704, [sym_size_int, 1500, 1280])
	        amin_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_392, [2])
	        amax_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_392, [2]);  view_392 = None
	        full_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_25, full_50);  amin_25 = full_50 = None
	        full_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_25, full_51);  amax_25 = full_51 = None
	        sub_1159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_25, minimum_25);  maximum_25 = None
	        div_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1159, 255.0);  sub_1159 = None
	        clamp_min_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_50, 1.1920928955078125e-07);  div_50 = None
	        div_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_25, clamp_min_75);  minimum_25 = None
	        round_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_51);  div_51 = None
	        sub_1165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_51);  round_51 = None
	        clamp_min_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1165, -128);  sub_1165 = None
	        clamp_max_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_76, 127);  clamp_min_76 = None
	        _assert_tensor_metadata_227 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_75, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_227 = None
	        _assert_tensor_metadata_228 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_50, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_228 = None
	        convert_element_type_150: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_50, torch.int8);  clamp_max_50 = None
	        view_393: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_3704, [sym_size_int, 1500, 1280])
	        view_394: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_75, [sym_size_int, 1500, 1])
	        view_395: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_150, [sym_size_int, 1500, 1])
	        reciprocal_25: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_394);  view_394 = None
	        mul_2485: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_25, 1.0);  reciprocal_25 = None
	        mul_2488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_393, mul_2485);  view_393 = mul_2485 = None
	        round_52: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2488);  mul_2488 = None
	        add_3943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_52, view_395);  round_52 = view_395 = None
	        clamp_min_77: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3943, -128);  add_3943 = None
	        clamp_max_51: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_77, 127);  clamp_min_77 = None
	        view_396: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_51, [sym_size_int, 1500, 1280]);  clamp_max_51 = None
	        _assert_tensor_metadata_229 = torch.ops.aten._assert_tensor_metadata.default(view_396, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_229 = None
	        convert_element_type_151: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_396, torch.int8);  view_396 = None
	        view_397: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_151, [sym_size_int, 1500, 1280]);  convert_element_type_151 = None
	        view_398: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_75, [sym_size_int, 1500, 1]);  clamp_min_75 = None
	        view_399: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_150, [sym_size_int, 1500, 1]);  convert_element_type_150 = None
	        _assert_tensor_metadata_230 = torch.ops.aten._assert_tensor_metadata.default(view_397, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_230 = None
	        convert_element_type_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_397, torch.float32);  view_397 = None
	        _assert_tensor_metadata_231 = torch.ops.aten._assert_tensor_metadata.default(view_399, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_231 = None
	        convert_element_type_153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_399, torch.float32);  view_399 = None
	        sub_1185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_152, convert_element_type_153);  convert_element_type_152 = convert_element_type_153 = None
	        mul_2510: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1185, view_398);  sub_1185 = view_398 = None
	        view_400: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2510, [sym_size_int, 1500, 1280]);  mul_2510 = None
	        _assert_tensor_metadata_232 = torch.ops.aten._assert_tensor_metadata.default(view_400, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_232 = None
	        view_401: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_402: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_403: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_233 = torch.ops.aten._assert_tensor_metadata.default(view_401, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_233 = None
	        convert_element_type_154: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_401, torch.float32);  view_401 = None
	        _assert_tensor_metadata_234 = torch.ops.aten._assert_tensor_metadata.default(view_403, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_234 = None
	        convert_element_type_155: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_403, torch.float32);  view_403 = None
	        sub_1189: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_154, convert_element_type_155);  convert_element_type_154 = convert_element_type_155 = None
	        mul_2515: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1189, view_402);  sub_1189 = view_402 = None
	        view_404: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2515, [1280, 1280]);  mul_2515 = None
	        _assert_tensor_metadata_235 = torch.ops.aten._assert_tensor_metadata.default(view_404, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_235 = None
	        permute_43: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_404, [1, 0]);  view_404 = None
	        mul_2518: "Sym(1500*s6)" = sym_size_int * 1500
	        view_405: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_400, [mul_2518, 1280]);  view_400 = mul_2518 = None
	        mm_4: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_405, permute_43);  view_405 = permute_43 = None
	        view_406: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_4, [sym_size_int, 1500, 1280]);  mm_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_407: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_406, [sym_size_int, -1, 20, 64]);  view_406 = None
	        permute_44: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_407, [0, 2, 1, 3]);  view_407 = None
	        clone_35: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_44, memory_format = torch.contiguous_format);  permute_44 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_408: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_3704, [sym_size_int, 1500, 1280])
	        amin_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_408, [2])
	        amax_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_408, [2]);  view_408 = None
	        full_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_26, full_52);  amin_26 = full_52 = None
	        full_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_26, full_53);  amax_26 = full_53 = None
	        sub_1203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_26, minimum_26);  maximum_26 = None
	        div_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1203, 255.0);  sub_1203 = None
	        clamp_min_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_52, 1.1920928955078125e-07);  div_52 = None
	        div_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_26, clamp_min_78);  minimum_26 = None
	        round_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_53);  div_53 = None
	        sub_1209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_53);  round_53 = None
	        clamp_min_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1209, -128);  sub_1209 = None
	        clamp_max_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_79, 127);  clamp_min_79 = None
	        _assert_tensor_metadata_236 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_78, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_236 = None
	        _assert_tensor_metadata_237 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_52, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_237 = None
	        convert_element_type_156: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_52, torch.int8);  clamp_max_52 = None
	        view_409: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_3704, [sym_size_int, 1500, 1280]);  add_3704 = None
	        view_410: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_78, [sym_size_int, 1500, 1])
	        view_411: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_156, [sym_size_int, 1500, 1])
	        reciprocal_26: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_410);  view_410 = None
	        mul_2584: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_26, 1.0);  reciprocal_26 = None
	        mul_2587: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_409, mul_2584);  view_409 = mul_2584 = None
	        round_54: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2587);  mul_2587 = None
	        add_4091: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_54, view_411);  round_54 = view_411 = None
	        clamp_min_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4091, -128);  add_4091 = None
	        clamp_max_53: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_80, 127);  clamp_min_80 = None
	        view_412: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_53, [sym_size_int, 1500, 1280]);  clamp_max_53 = None
	        _assert_tensor_metadata_238 = torch.ops.aten._assert_tensor_metadata.default(view_412, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_238 = None
	        convert_element_type_157: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_412, torch.int8);  view_412 = None
	        view_413: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_157, [sym_size_int, 1500, 1280]);  convert_element_type_157 = None
	        view_414: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_78, [sym_size_int, 1500, 1]);  clamp_min_78 = None
	        view_415: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_156, [sym_size_int, 1500, 1]);  convert_element_type_156 = None
	        _assert_tensor_metadata_239 = torch.ops.aten._assert_tensor_metadata.default(view_413, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_239 = None
	        convert_element_type_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_413, torch.float32);  view_413 = None
	        _assert_tensor_metadata_240 = torch.ops.aten._assert_tensor_metadata.default(view_415, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_240 = None
	        convert_element_type_159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_415, torch.float32);  view_415 = None
	        sub_1229: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_158, convert_element_type_159);  convert_element_type_158 = convert_element_type_159 = None
	        mul_2609: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1229, view_414);  sub_1229 = view_414 = None
	        view_416: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2609, [sym_size_int, 1500, 1280]);  mul_2609 = None
	        _assert_tensor_metadata_241 = torch.ops.aten._assert_tensor_metadata.default(view_416, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_241 = None
	        view_417: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_418: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_419: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_242 = torch.ops.aten._assert_tensor_metadata.default(view_417, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_242 = None
	        convert_element_type_160: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_417, torch.float32);  view_417 = None
	        _assert_tensor_metadata_243 = torch.ops.aten._assert_tensor_metadata.default(view_419, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_243 = None
	        convert_element_type_161: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_419, torch.float32);  view_419 = None
	        sub_1233: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_160, convert_element_type_161);  convert_element_type_160 = convert_element_type_161 = None
	        mul_2614: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1233, view_418);  sub_1233 = view_418 = None
	        view_420: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2614, [1280, 1280]);  mul_2614 = None
	        _assert_tensor_metadata_244 = torch.ops.aten._assert_tensor_metadata.default(view_420, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_244 = None
	        mul_2619: "Sym(1500*s6)" = sym_size_int * 1500
	        view_421: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_416, [mul_2619, 1280]);  view_416 = mul_2619 = None
	        permute_45: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_420, [1, 0]);  view_420 = None
	        addmm_21: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_4_self_attn_v_proj_bias, view_421, permute_45);  model_audio_tower_layers_4_self_attn_v_proj_bias = view_421 = permute_45 = None
	        view_422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_21, [sym_size_int, 1500, 1280]);  addmm_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_423: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_422, [sym_size_int, -1, 20, 64]);  view_422 = None
	        permute_46: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_423, [0, 2, 1, 3]);  view_423 = None
	        clone_36: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_46, memory_format = torch.contiguous_format);  permute_46 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_4 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_34, clone_35, clone_36, None, False, scale = 1.0);  clone_34 = clone_35 = clone_36 = None
	        getitem_34: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_4[0];  _scaled_dot_product_efficient_attention_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_47: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_34, [0, 2, 1, 3]);  getitem_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_424: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_47, [sym_size_int, 1500, -1]);  permute_47 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_425: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_424, [sym_size_int, 1500, 1280])
	        amin_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_425, [2])
	        amax_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_425, [2]);  view_425 = None
	        full_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_27, full_54);  amin_27 = full_54 = None
	        full_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_27, full_55);  amax_27 = full_55 = None
	        sub_1251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_27, minimum_27);  maximum_27 = None
	        div_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1251, 255.0);  sub_1251 = None
	        clamp_min_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_54, 1.1920928955078125e-07);  div_54 = None
	        div_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_27, clamp_min_81);  minimum_27 = None
	        round_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_55);  div_55 = None
	        sub_1257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_55);  round_55 = None
	        clamp_min_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1257, -128);  sub_1257 = None
	        clamp_max_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_82, 127);  clamp_min_82 = None
	        _assert_tensor_metadata_245 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_81, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_245 = None
	        _assert_tensor_metadata_246 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_54, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_246 = None
	        convert_element_type_162: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_54, torch.int8);  clamp_max_54 = None
	        view_426: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_424, [sym_size_int, 1500, 1280]);  view_424 = None
	        view_427: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_81, [sym_size_int, 1500, 1])
	        view_428: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_162, [sym_size_int, 1500, 1])
	        reciprocal_27: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_427);  view_427 = None
	        mul_2689: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_27, 1.0);  reciprocal_27 = None
	        mul_2692: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_426, mul_2689);  view_426 = mul_2689 = None
	        round_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2692);  mul_2692 = None
	        add_4255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_56, view_428);  round_56 = view_428 = None
	        clamp_min_83: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4255, -128);  add_4255 = None
	        clamp_max_55: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_83, 127);  clamp_min_83 = None
	        view_429: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_55, [sym_size_int, 1500, 1280]);  clamp_max_55 = None
	        _assert_tensor_metadata_247 = torch.ops.aten._assert_tensor_metadata.default(view_429, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_247 = None
	        convert_element_type_163: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_429, torch.int8);  view_429 = None
	        view_430: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_163, [sym_size_int, 1500, 1280]);  convert_element_type_163 = None
	        view_431: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_81, [sym_size_int, 1500, 1]);  clamp_min_81 = None
	        view_432: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_162, [sym_size_int, 1500, 1]);  convert_element_type_162 = None
	        _assert_tensor_metadata_248 = torch.ops.aten._assert_tensor_metadata.default(view_430, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_248 = None
	        convert_element_type_164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_430, torch.float32);  view_430 = None
	        _assert_tensor_metadata_249 = torch.ops.aten._assert_tensor_metadata.default(view_432, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_249 = None
	        convert_element_type_165: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_432, torch.float32);  view_432 = None
	        sub_1277: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_164, convert_element_type_165);  convert_element_type_164 = convert_element_type_165 = None
	        mul_2714: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1277, view_431);  sub_1277 = view_431 = None
	        view_433: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2714, [sym_size_int, 1500, 1280]);  mul_2714 = None
	        _assert_tensor_metadata_250 = torch.ops.aten._assert_tensor_metadata.default(view_433, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_250 = None
	        view_434: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_435: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_436: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_251 = torch.ops.aten._assert_tensor_metadata.default(view_434, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_251 = None
	        convert_element_type_166: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_434, torch.float32);  view_434 = None
	        _assert_tensor_metadata_252 = torch.ops.aten._assert_tensor_metadata.default(view_436, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_252 = None
	        convert_element_type_167: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_436, torch.float32);  view_436 = None
	        sub_1281: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_166, convert_element_type_167);  convert_element_type_166 = convert_element_type_167 = None
	        mul_2719: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1281, view_435);  sub_1281 = view_435 = None
	        view_437: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2719, [1280, 1280]);  mul_2719 = None
	        _assert_tensor_metadata_253 = torch.ops.aten._assert_tensor_metadata.default(view_437, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_253 = None
	        mul_2724: "Sym(1500*s6)" = sym_size_int * 1500
	        view_438: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_433, [mul_2724, 1280]);  view_433 = mul_2724 = None
	        permute_48: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_437, [1, 0]);  view_437 = None
	        addmm_22: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_4_self_attn_out_proj_bias, view_438, permute_48);  model_audio_tower_layers_4_self_attn_out_proj_bias = view_438 = permute_48 = None
	        view_439: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_22, [sym_size_int, 1500, 1280]);  addmm_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_37: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_439);  view_439 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_4318: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_3698, clone_37);  add_3698 = clone_37 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_4318, memory_format = torch.contiguous_format)
	        var_mean_9 = torch.ops.aten.var_mean.correction(clone_38, [2], correction = 0, keepdim = True)
	        getitem_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_9[0]
	        getitem_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_9[1];  var_mean_9 = None
	        add_4323: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_38, 1e-05);  getitem_38 = None
	        rsqrt_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_4323);  add_4323 = None
	        sub_1287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_38, getitem_39);  clone_38 = getitem_39 = None
	        mul_2735: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1287, rsqrt_9);  sub_1287 = rsqrt_9 = None
	        mul_2736: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2735, model_audio_tower_layers_4_final_layer_norm_weight);  mul_2735 = model_audio_tower_layers_4_final_layer_norm_weight = None
	        add_4324: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2736, model_audio_tower_layers_4_final_layer_norm_bias);  mul_2736 = model_audio_tower_layers_4_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_440: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_4324, [sym_size_int, 1500, 1280])
	        amin_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_440, [2])
	        amax_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_440, [2]);  view_440 = None
	        full_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_28, full_56);  amin_28 = full_56 = None
	        full_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_28, full_57);  amax_28 = full_57 = None
	        sub_1298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_28, minimum_28);  maximum_28 = None
	        div_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1298, 255.0);  sub_1298 = None
	        clamp_min_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_56, 1.1920928955078125e-07);  div_56 = None
	        div_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_28, clamp_min_84);  minimum_28 = None
	        round_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_57);  div_57 = None
	        sub_1304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_57);  round_57 = None
	        clamp_min_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1304, -128);  sub_1304 = None
	        clamp_max_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_85, 127);  clamp_min_85 = None
	        _assert_tensor_metadata_254 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_84, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_254 = None
	        _assert_tensor_metadata_255 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_56, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_255 = None
	        convert_element_type_168: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_56, torch.int8);  clamp_max_56 = None
	        view_441: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_4324, [sym_size_int, 1500, 1280]);  add_4324 = None
	        view_442: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_84, [sym_size_int, 1500, 1])
	        view_443: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_168, [sym_size_int, 1500, 1])
	        reciprocal_28: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_442);  view_442 = None
	        mul_2784: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_28, 1.0);  reciprocal_28 = None
	        mul_2787: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_441, mul_2784);  view_441 = mul_2784 = None
	        round_58: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2787);  mul_2787 = None
	        add_4411: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_58, view_443);  round_58 = view_443 = None
	        clamp_min_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4411, -128);  add_4411 = None
	        clamp_max_57: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_86, 127);  clamp_min_86 = None
	        view_444: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_57, [sym_size_int, 1500, 1280]);  clamp_max_57 = None
	        _assert_tensor_metadata_256 = torch.ops.aten._assert_tensor_metadata.default(view_444, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_256 = None
	        convert_element_type_169: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_444, torch.int8);  view_444 = None
	        view_445: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_169, [sym_size_int, 1500, 1280]);  convert_element_type_169 = None
	        view_446: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_84, [sym_size_int, 1500, 1]);  clamp_min_84 = None
	        view_447: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_168, [sym_size_int, 1500, 1]);  convert_element_type_168 = None
	        _assert_tensor_metadata_257 = torch.ops.aten._assert_tensor_metadata.default(view_445, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_257 = None
	        convert_element_type_170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_445, torch.float32);  view_445 = None
	        _assert_tensor_metadata_258 = torch.ops.aten._assert_tensor_metadata.default(view_447, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_258 = None
	        convert_element_type_171: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_447, torch.float32);  view_447 = None
	        sub_1324: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_170, convert_element_type_171);  convert_element_type_170 = convert_element_type_171 = None
	        mul_2809: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1324, view_446);  sub_1324 = view_446 = None
	        view_448: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2809, [sym_size_int, 1500, 1280]);  mul_2809 = None
	        _assert_tensor_metadata_259 = torch.ops.aten._assert_tensor_metadata.default(view_448, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_259 = None
	        view_449: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_4_fc1_parametrizations_weight_original0 = None
	        view_450: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_4_fc1_parametrizations_weight_original1 = None
	        view_451: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_4_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_260 = torch.ops.aten._assert_tensor_metadata.default(view_449, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_260 = None
	        convert_element_type_172: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_449, torch.float32);  view_449 = None
	        _assert_tensor_metadata_261 = torch.ops.aten._assert_tensor_metadata.default(view_451, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_261 = None
	        convert_element_type_173: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_451, torch.float32);  view_451 = None
	        sub_1328: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_172, convert_element_type_173);  convert_element_type_172 = convert_element_type_173 = None
	        mul_2814: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1328, view_450);  sub_1328 = view_450 = None
	        view_452: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2814, [5120, 1280]);  mul_2814 = None
	        _assert_tensor_metadata_262 = torch.ops.aten._assert_tensor_metadata.default(view_452, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_262 = None
	        mul_2819: "Sym(1500*s6)" = sym_size_int * 1500
	        view_453: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_448, [mul_2819, 1280]);  view_448 = mul_2819 = None
	        permute_49: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_452, [1, 0]);  view_452 = None
	        addmm_23: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_4_fc1_bias, view_453, permute_49);  model_audio_tower_layers_4_fc1_bias = view_453 = permute_49 = None
	        view_454: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_23, [sym_size_int, 1500, 5120]);  addmm_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_2826: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_454, 0.5)
	        mul_2827: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_454, 0.7071067811865476);  view_454 = None
	        erf_6: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_2827);  mul_2827 = None
	        add_4470: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_6, 1);  erf_6 = None
	        mul_2828: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2826, add_4470);  mul_2826 = add_4470 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_39: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_2828);  mul_2828 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_455: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_39, [sym_size_int, 1500, 5120])
	        amin_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_455, [2])
	        amax_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_455, [2]);  view_455 = None
	        full_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_29, full_58);  amin_29 = full_58 = None
	        full_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_29, full_59);  amax_29 = full_59 = None
	        sub_1341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_29, minimum_29);  maximum_29 = None
	        div_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1341, 255.0);  sub_1341 = None
	        clamp_min_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_58, 1.1920928955078125e-07);  div_58 = None
	        div_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_29, clamp_min_87);  minimum_29 = None
	        round_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_59);  div_59 = None
	        sub_1347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_59);  round_59 = None
	        clamp_min_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1347, -128);  sub_1347 = None
	        clamp_max_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_88, 127);  clamp_min_88 = None
	        _assert_tensor_metadata_263 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_87, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_263 = None
	        _assert_tensor_metadata_264 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_58, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_264 = None
	        convert_element_type_174: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_58, torch.int8);  clamp_max_58 = None
	        view_456: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_39, [sym_size_int, 1500, 5120]);  clone_39 = None
	        view_457: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_87, [sym_size_int, 1500, 1])
	        view_458: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_174, [sym_size_int, 1500, 1])
	        reciprocal_29: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_457);  view_457 = None
	        mul_2874: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_29, 1.0);  reciprocal_29 = None
	        mul_2877: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_456, mul_2874);  view_456 = mul_2874 = None
	        round_60: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_2877);  mul_2877 = None
	        add_4553: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_60, view_458);  round_60 = view_458 = None
	        clamp_min_89: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4553, -128);  add_4553 = None
	        clamp_max_59: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_89, 127);  clamp_min_89 = None
	        view_459: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_59, [sym_size_int, 1500, 5120]);  clamp_max_59 = None
	        _assert_tensor_metadata_265 = torch.ops.aten._assert_tensor_metadata.default(view_459, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_265 = None
	        convert_element_type_175: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_459, torch.int8);  view_459 = None
	        view_460: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_175, [sym_size_int, 1500, 5120]);  convert_element_type_175 = None
	        view_461: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_87, [sym_size_int, 1500, 1]);  clamp_min_87 = None
	        view_462: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_174, [sym_size_int, 1500, 1]);  convert_element_type_174 = None
	        _assert_tensor_metadata_266 = torch.ops.aten._assert_tensor_metadata.default(view_460, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_266 = None
	        convert_element_type_176: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_460, torch.float32);  view_460 = None
	        _assert_tensor_metadata_267 = torch.ops.aten._assert_tensor_metadata.default(view_462, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_267 = None
	        convert_element_type_177: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_462, torch.float32);  view_462 = None
	        sub_1367: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_176, convert_element_type_177);  convert_element_type_176 = convert_element_type_177 = None
	        mul_2899: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1367, view_461);  sub_1367 = view_461 = None
	        view_463: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_2899, [sym_size_int, 1500, 5120]);  mul_2899 = None
	        _assert_tensor_metadata_268 = torch.ops.aten._assert_tensor_metadata.default(view_463, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_268 = None
	        view_464: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_4_fc2_parametrizations_weight_original0 = None
	        view_465: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_4_fc2_parametrizations_weight_original1 = None
	        view_466: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_4_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_269 = torch.ops.aten._assert_tensor_metadata.default(view_464, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_269 = None
	        convert_element_type_178: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_464, torch.float32);  view_464 = None
	        _assert_tensor_metadata_270 = torch.ops.aten._assert_tensor_metadata.default(view_466, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_270 = None
	        convert_element_type_179: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_466, torch.float32);  view_466 = None
	        sub_1371: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_178, convert_element_type_179);  convert_element_type_178 = convert_element_type_179 = None
	        mul_2904: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1371, view_465);  sub_1371 = view_465 = None
	        view_467: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_2904, [1280, 5120]);  mul_2904 = None
	        _assert_tensor_metadata_271 = torch.ops.aten._assert_tensor_metadata.default(view_467, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_271 = None
	        mul_2909: "Sym(1500*s6)" = sym_size_int * 1500
	        view_468: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_463, [mul_2909, 5120]);  view_463 = mul_2909 = None
	        permute_50: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_467, [1, 0]);  view_467 = None
	        addmm_24: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_4_fc2_bias, view_468, permute_50);  model_audio_tower_layers_4_fc2_bias = view_468 = permute_50 = None
	        view_469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_24, [sym_size_int, 1500, 1280]);  addmm_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_40: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_469);  view_469 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_4616: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_4318, clone_40);  add_4318 = clone_40 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_41: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_4616, memory_format = torch.contiguous_format)
	        var_mean_10 = torch.ops.aten.var_mean.correction(clone_41, [2], correction = 0, keepdim = True)
	        getitem_40: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_10[0]
	        getitem_41: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_10[1];  var_mean_10 = None
	        add_4621: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_40, 1e-05);  getitem_40 = None
	        rsqrt_10: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_4621);  add_4621 = None
	        sub_1377: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_41, getitem_41);  clone_41 = getitem_41 = None
	        mul_2920: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1377, rsqrt_10);  sub_1377 = rsqrt_10 = None
	        mul_2921: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2920, model_audio_tower_layers_5_self_attn_layer_norm_weight);  mul_2920 = model_audio_tower_layers_5_self_attn_layer_norm_weight = None
	        add_4622: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2921, model_audio_tower_layers_5_self_attn_layer_norm_bias);  mul_2921 = model_audio_tower_layers_5_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_470: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_4622, [sym_size_int, 1500, 1280])
	        amin_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_470, [2])
	        amax_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_470, [2]);  view_470 = None
	        full_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_30, full_60);  amin_30 = full_60 = None
	        full_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_30, full_61);  amax_30 = full_61 = None
	        sub_1388: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_30, minimum_30);  maximum_30 = None
	        div_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1388, 255.0);  sub_1388 = None
	        clamp_min_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_60, 1.1920928955078125e-07);  div_60 = None
	        div_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_30, clamp_min_90);  minimum_30 = None
	        round_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_61);  div_61 = None
	        sub_1394: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_61);  round_61 = None
	        clamp_min_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1394, -128);  sub_1394 = None
	        clamp_max_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_91, 127);  clamp_min_91 = None
	        _assert_tensor_metadata_272 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_90, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_272 = None
	        _assert_tensor_metadata_273 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_60, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_273 = None
	        convert_element_type_180: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_60, torch.int8);  clamp_max_60 = None
	        view_471: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_4622, [sym_size_int, 1500, 1280])
	        view_472: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_90, [sym_size_int, 1500, 1])
	        view_473: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_180, [sym_size_int, 1500, 1])
	        reciprocal_30: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_472);  view_472 = None
	        mul_2969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_30, 1.0);  reciprocal_30 = None
	        mul_2972: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_471, mul_2969);  view_471 = mul_2969 = None
	        round_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2972);  mul_2972 = None
	        add_4709: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_62, view_473);  round_62 = view_473 = None
	        clamp_min_92: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4709, -128);  add_4709 = None
	        clamp_max_61: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_92, 127);  clamp_min_92 = None
	        view_474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_61, [sym_size_int, 1500, 1280]);  clamp_max_61 = None
	        _assert_tensor_metadata_274 = torch.ops.aten._assert_tensor_metadata.default(view_474, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_274 = None
	        convert_element_type_181: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_474, torch.int8);  view_474 = None
	        view_475: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_181, [sym_size_int, 1500, 1280]);  convert_element_type_181 = None
	        view_476: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_90, [sym_size_int, 1500, 1]);  clamp_min_90 = None
	        view_477: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_180, [sym_size_int, 1500, 1]);  convert_element_type_180 = None
	        _assert_tensor_metadata_275 = torch.ops.aten._assert_tensor_metadata.default(view_475, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_275 = None
	        convert_element_type_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_475, torch.float32);  view_475 = None
	        _assert_tensor_metadata_276 = torch.ops.aten._assert_tensor_metadata.default(view_477, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_276 = None
	        convert_element_type_183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_477, torch.float32);  view_477 = None
	        sub_1414: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_182, convert_element_type_183);  convert_element_type_182 = convert_element_type_183 = None
	        mul_2994: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1414, view_476);  sub_1414 = view_476 = None
	        view_478: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2994, [sym_size_int, 1500, 1280]);  mul_2994 = None
	        _assert_tensor_metadata_277 = torch.ops.aten._assert_tensor_metadata.default(view_478, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_277 = None
	        view_479: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_480: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_481: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_278 = torch.ops.aten._assert_tensor_metadata.default(view_479, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_278 = None
	        convert_element_type_184: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_479, torch.float32);  view_479 = None
	        _assert_tensor_metadata_279 = torch.ops.aten._assert_tensor_metadata.default(view_481, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_279 = None
	        convert_element_type_185: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_481, torch.float32);  view_481 = None
	        sub_1418: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_184, convert_element_type_185);  convert_element_type_184 = convert_element_type_185 = None
	        mul_2999: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1418, view_480);  sub_1418 = view_480 = None
	        view_482: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2999, [1280, 1280]);  mul_2999 = None
	        _assert_tensor_metadata_280 = torch.ops.aten._assert_tensor_metadata.default(view_482, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_280 = None
	        mul_3004: "Sym(1500*s6)" = sym_size_int * 1500
	        view_483: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_478, [mul_3004, 1280]);  view_478 = mul_3004 = None
	        permute_51: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_482, [1, 0]);  view_482 = None
	        addmm_25: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_5_self_attn_q_proj_bias, view_483, permute_51);  model_audio_tower_layers_5_self_attn_q_proj_bias = view_483 = permute_51 = None
	        view_484: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_25, [sym_size_int, 1500, 1280]);  addmm_25 = None
	        mul_3011: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_484, 0.125);  view_484 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_485: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_3011, [sym_size_int, 1500, 20, 64]);  mul_3011 = None
	        permute_52: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_485, [0, 2, 1, 3]);  view_485 = None
	        clone_42: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_52, memory_format = torch.contiguous_format);  permute_52 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_486: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_4622, [sym_size_int, 1500, 1280])
	        amin_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_486, [2])
	        amax_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_486, [2]);  view_486 = None
	        full_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_31, full_62);  amin_31 = full_62 = None
	        full_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_31, full_63);  amax_31 = full_63 = None
	        sub_1433: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_31, minimum_31);  maximum_31 = None
	        div_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1433, 255.0);  sub_1433 = None
	        clamp_min_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_62, 1.1920928955078125e-07);  div_62 = None
	        div_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_31, clamp_min_93);  minimum_31 = None
	        round_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_63);  div_63 = None
	        sub_1439: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_63);  round_63 = None
	        clamp_min_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1439, -128);  sub_1439 = None
	        clamp_max_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_94, 127);  clamp_min_94 = None
	        _assert_tensor_metadata_281 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_93, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_281 = None
	        _assert_tensor_metadata_282 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_62, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_282 = None
	        convert_element_type_186: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_62, torch.int8);  clamp_max_62 = None
	        view_487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_4622, [sym_size_int, 1500, 1280])
	        view_488: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_93, [sym_size_int, 1500, 1])
	        view_489: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_186, [sym_size_int, 1500, 1])
	        reciprocal_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_488);  view_488 = None
	        mul_3065: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_31, 1.0);  reciprocal_31 = None
	        mul_3068: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_487, mul_3065);  view_487 = mul_3065 = None
	        round_64: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3068);  mul_3068 = None
	        add_4861: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_64, view_489);  round_64 = view_489 = None
	        clamp_min_95: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4861, -128);  add_4861 = None
	        clamp_max_63: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_95, 127);  clamp_min_95 = None
	        view_490: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_63, [sym_size_int, 1500, 1280]);  clamp_max_63 = None
	        _assert_tensor_metadata_283 = torch.ops.aten._assert_tensor_metadata.default(view_490, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_283 = None
	        convert_element_type_187: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_490, torch.int8);  view_490 = None
	        view_491: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_187, [sym_size_int, 1500, 1280]);  convert_element_type_187 = None
	        view_492: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_93, [sym_size_int, 1500, 1]);  clamp_min_93 = None
	        view_493: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_186, [sym_size_int, 1500, 1]);  convert_element_type_186 = None
	        _assert_tensor_metadata_284 = torch.ops.aten._assert_tensor_metadata.default(view_491, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_284 = None
	        convert_element_type_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_491, torch.float32);  view_491 = None
	        _assert_tensor_metadata_285 = torch.ops.aten._assert_tensor_metadata.default(view_493, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_285 = None
	        convert_element_type_189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_493, torch.float32);  view_493 = None
	        sub_1459: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_188, convert_element_type_189);  convert_element_type_188 = convert_element_type_189 = None
	        mul_3090: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1459, view_492);  sub_1459 = view_492 = None
	        view_494: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3090, [sym_size_int, 1500, 1280]);  mul_3090 = None
	        _assert_tensor_metadata_286 = torch.ops.aten._assert_tensor_metadata.default(view_494, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_286 = None
	        view_495: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_496: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_497: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_287 = torch.ops.aten._assert_tensor_metadata.default(view_495, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_287 = None
	        convert_element_type_190: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_495, torch.float32);  view_495 = None
	        _assert_tensor_metadata_288 = torch.ops.aten._assert_tensor_metadata.default(view_497, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_288 = None
	        convert_element_type_191: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_497, torch.float32);  view_497 = None
	        sub_1463: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_190, convert_element_type_191);  convert_element_type_190 = convert_element_type_191 = None
	        mul_3095: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1463, view_496);  sub_1463 = view_496 = None
	        view_498: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3095, [1280, 1280]);  mul_3095 = None
	        _assert_tensor_metadata_289 = torch.ops.aten._assert_tensor_metadata.default(view_498, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_289 = None
	        permute_53: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_498, [1, 0]);  view_498 = None
	        mul_3098: "Sym(1500*s6)" = sym_size_int * 1500
	        view_499: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_494, [mul_3098, 1280]);  view_494 = mul_3098 = None
	        mm_5: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_499, permute_53);  view_499 = permute_53 = None
	        view_500: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_5, [sym_size_int, 1500, 1280]);  mm_5 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_501: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_500, [sym_size_int, -1, 20, 64]);  view_500 = None
	        permute_54: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_501, [0, 2, 1, 3]);  view_501 = None
	        clone_43: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_54, memory_format = torch.contiguous_format);  permute_54 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_502: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_4622, [sym_size_int, 1500, 1280])
	        amin_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_502, [2])
	        amax_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_502, [2]);  view_502 = None
	        full_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_32, full_64);  amin_32 = full_64 = None
	        full_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_32, full_65);  amax_32 = full_65 = None
	        sub_1477: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_32, minimum_32);  maximum_32 = None
	        div_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1477, 255.0);  sub_1477 = None
	        clamp_min_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_64, 1.1920928955078125e-07);  div_64 = None
	        div_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_32, clamp_min_96);  minimum_32 = None
	        round_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_65);  div_65 = None
	        sub_1483: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_65);  round_65 = None
	        clamp_min_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1483, -128);  sub_1483 = None
	        clamp_max_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_97, 127);  clamp_min_97 = None
	        _assert_tensor_metadata_290 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_96, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_290 = None
	        _assert_tensor_metadata_291 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_64, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_291 = None
	        convert_element_type_192: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_64, torch.int8);  clamp_max_64 = None
	        view_503: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_4622, [sym_size_int, 1500, 1280]);  add_4622 = None
	        view_504: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_96, [sym_size_int, 1500, 1])
	        view_505: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_192, [sym_size_int, 1500, 1])
	        reciprocal_32: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_504);  view_504 = None
	        mul_3164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_32, 1.0);  reciprocal_32 = None
	        mul_3167: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_503, mul_3164);  view_503 = mul_3164 = None
	        round_66: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3167);  mul_3167 = None
	        add_5009: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_66, view_505);  round_66 = view_505 = None
	        clamp_min_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5009, -128);  add_5009 = None
	        clamp_max_65: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_98, 127);  clamp_min_98 = None
	        view_506: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_65, [sym_size_int, 1500, 1280]);  clamp_max_65 = None
	        _assert_tensor_metadata_292 = torch.ops.aten._assert_tensor_metadata.default(view_506, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_292 = None
	        convert_element_type_193: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_506, torch.int8);  view_506 = None
	        view_507: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_193, [sym_size_int, 1500, 1280]);  convert_element_type_193 = None
	        view_508: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_96, [sym_size_int, 1500, 1]);  clamp_min_96 = None
	        view_509: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_192, [sym_size_int, 1500, 1]);  convert_element_type_192 = None
	        _assert_tensor_metadata_293 = torch.ops.aten._assert_tensor_metadata.default(view_507, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_293 = None
	        convert_element_type_194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_507, torch.float32);  view_507 = None
	        _assert_tensor_metadata_294 = torch.ops.aten._assert_tensor_metadata.default(view_509, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_294 = None
	        convert_element_type_195: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_509, torch.float32);  view_509 = None
	        sub_1503: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_194, convert_element_type_195);  convert_element_type_194 = convert_element_type_195 = None
	        mul_3189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1503, view_508);  sub_1503 = view_508 = None
	        view_510: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3189, [sym_size_int, 1500, 1280]);  mul_3189 = None
	        _assert_tensor_metadata_295 = torch.ops.aten._assert_tensor_metadata.default(view_510, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_295 = None
	        view_511: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_512: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_513: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_296 = torch.ops.aten._assert_tensor_metadata.default(view_511, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_296 = None
	        convert_element_type_196: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_511, torch.float32);  view_511 = None
	        _assert_tensor_metadata_297 = torch.ops.aten._assert_tensor_metadata.default(view_513, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_297 = None
	        convert_element_type_197: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_513, torch.float32);  view_513 = None
	        sub_1507: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_196, convert_element_type_197);  convert_element_type_196 = convert_element_type_197 = None
	        mul_3194: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1507, view_512);  sub_1507 = view_512 = None
	        view_514: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3194, [1280, 1280]);  mul_3194 = None
	        _assert_tensor_metadata_298 = torch.ops.aten._assert_tensor_metadata.default(view_514, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_298 = None
	        mul_3199: "Sym(1500*s6)" = sym_size_int * 1500
	        view_515: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_510, [mul_3199, 1280]);  view_510 = mul_3199 = None
	        permute_55: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_514, [1, 0]);  view_514 = None
	        addmm_26: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_5_self_attn_v_proj_bias, view_515, permute_55);  model_audio_tower_layers_5_self_attn_v_proj_bias = view_515 = permute_55 = None
	        view_516: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_26, [sym_size_int, 1500, 1280]);  addmm_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_517: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_516, [sym_size_int, -1, 20, 64]);  view_516 = None
	        permute_56: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_517, [0, 2, 1, 3]);  view_517 = None
	        clone_44: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_56, memory_format = torch.contiguous_format);  permute_56 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_5 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_42, clone_43, clone_44, None, False, scale = 1.0);  clone_42 = clone_43 = clone_44 = None
	        getitem_42: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_5[0];  _scaled_dot_product_efficient_attention_5 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_57: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_42, [0, 2, 1, 3]);  getitem_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_57, [sym_size_int, 1500, -1]);  permute_57 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_519: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_518, [sym_size_int, 1500, 1280])
	        amin_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_519, [2])
	        amax_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_519, [2]);  view_519 = None
	        full_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_33, full_66);  amin_33 = full_66 = None
	        full_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_33, full_67);  amax_33 = full_67 = None
	        sub_1525: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_33, minimum_33);  maximum_33 = None
	        div_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1525, 255.0);  sub_1525 = None
	        clamp_min_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_66, 1.1920928955078125e-07);  div_66 = None
	        div_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_33, clamp_min_99);  minimum_33 = None
	        round_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_67);  div_67 = None
	        sub_1531: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_67);  round_67 = None
	        clamp_min_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1531, -128);  sub_1531 = None
	        clamp_max_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_100, 127);  clamp_min_100 = None
	        _assert_tensor_metadata_299 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_99, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_299 = None
	        _assert_tensor_metadata_300 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_66, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_300 = None
	        convert_element_type_198: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_66, torch.int8);  clamp_max_66 = None
	        view_520: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_518, [sym_size_int, 1500, 1280]);  view_518 = None
	        view_521: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_99, [sym_size_int, 1500, 1])
	        view_522: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_198, [sym_size_int, 1500, 1])
	        reciprocal_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_521);  view_521 = None
	        mul_3269: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_33, 1.0);  reciprocal_33 = None
	        mul_3272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_520, mul_3269);  view_520 = mul_3269 = None
	        round_68: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3272);  mul_3272 = None
	        add_5173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_68, view_522);  round_68 = view_522 = None
	        clamp_min_101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5173, -128);  add_5173 = None
	        clamp_max_67: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_101, 127);  clamp_min_101 = None
	        view_523: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_67, [sym_size_int, 1500, 1280]);  clamp_max_67 = None
	        _assert_tensor_metadata_301 = torch.ops.aten._assert_tensor_metadata.default(view_523, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_301 = None
	        convert_element_type_199: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_523, torch.int8);  view_523 = None
	        view_524: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_199, [sym_size_int, 1500, 1280]);  convert_element_type_199 = None
	        view_525: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_99, [sym_size_int, 1500, 1]);  clamp_min_99 = None
	        view_526: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_198, [sym_size_int, 1500, 1]);  convert_element_type_198 = None
	        _assert_tensor_metadata_302 = torch.ops.aten._assert_tensor_metadata.default(view_524, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_302 = None
	        convert_element_type_200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_524, torch.float32);  view_524 = None
	        _assert_tensor_metadata_303 = torch.ops.aten._assert_tensor_metadata.default(view_526, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_303 = None
	        convert_element_type_201: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_526, torch.float32);  view_526 = None
	        sub_1551: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_200, convert_element_type_201);  convert_element_type_200 = convert_element_type_201 = None
	        mul_3294: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1551, view_525);  sub_1551 = view_525 = None
	        view_527: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3294, [sym_size_int, 1500, 1280]);  mul_3294 = None
	        _assert_tensor_metadata_304 = torch.ops.aten._assert_tensor_metadata.default(view_527, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_304 = None
	        view_528: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_529: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_530: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_305 = torch.ops.aten._assert_tensor_metadata.default(view_528, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_305 = None
	        convert_element_type_202: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_528, torch.float32);  view_528 = None
	        _assert_tensor_metadata_306 = torch.ops.aten._assert_tensor_metadata.default(view_530, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_306 = None
	        convert_element_type_203: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_530, torch.float32);  view_530 = None
	        sub_1555: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_202, convert_element_type_203);  convert_element_type_202 = convert_element_type_203 = None
	        mul_3299: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1555, view_529);  sub_1555 = view_529 = None
	        view_531: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3299, [1280, 1280]);  mul_3299 = None
	        _assert_tensor_metadata_307 = torch.ops.aten._assert_tensor_metadata.default(view_531, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_307 = None
	        mul_3304: "Sym(1500*s6)" = sym_size_int * 1500
	        view_532: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_527, [mul_3304, 1280]);  view_527 = mul_3304 = None
	        permute_58: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_531, [1, 0]);  view_531 = None
	        addmm_27: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_5_self_attn_out_proj_bias, view_532, permute_58);  model_audio_tower_layers_5_self_attn_out_proj_bias = view_532 = permute_58 = None
	        view_533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_27, [sym_size_int, 1500, 1280]);  addmm_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_45: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_533);  view_533 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_5236: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_4616, clone_45);  add_4616 = clone_45 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_46: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_5236, memory_format = torch.contiguous_format)
	        var_mean_11 = torch.ops.aten.var_mean.correction(clone_46, [2], correction = 0, keepdim = True)
	        getitem_46: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_11[0]
	        getitem_47: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_11[1];  var_mean_11 = None
	        add_5241: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_46, 1e-05);  getitem_46 = None
	        rsqrt_11: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_5241);  add_5241 = None
	        sub_1561: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_46, getitem_47);  clone_46 = getitem_47 = None
	        mul_3315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1561, rsqrt_11);  sub_1561 = rsqrt_11 = None
	        mul_3316: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3315, model_audio_tower_layers_5_final_layer_norm_weight);  mul_3315 = model_audio_tower_layers_5_final_layer_norm_weight = None
	        add_5242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_3316, model_audio_tower_layers_5_final_layer_norm_bias);  mul_3316 = model_audio_tower_layers_5_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_534: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_5242, [sym_size_int, 1500, 1280])
	        amin_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_534, [2])
	        amax_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_534, [2]);  view_534 = None
	        full_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_34, full_68);  amin_34 = full_68 = None
	        full_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_34, full_69);  amax_34 = full_69 = None
	        sub_1572: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_34, minimum_34);  maximum_34 = None
	        div_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1572, 255.0);  sub_1572 = None
	        clamp_min_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_68, 1.1920928955078125e-07);  div_68 = None
	        div_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_34, clamp_min_102);  minimum_34 = None
	        round_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_69);  div_69 = None
	        sub_1578: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_69);  round_69 = None
	        clamp_min_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1578, -128);  sub_1578 = None
	        clamp_max_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_103, 127);  clamp_min_103 = None
	        _assert_tensor_metadata_308 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_102, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_308 = None
	        _assert_tensor_metadata_309 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_68, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_309 = None
	        convert_element_type_204: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_68, torch.int8);  clamp_max_68 = None
	        view_535: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_5242, [sym_size_int, 1500, 1280]);  add_5242 = None
	        view_536: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_102, [sym_size_int, 1500, 1])
	        view_537: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_204, [sym_size_int, 1500, 1])
	        reciprocal_34: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_536);  view_536 = None
	        mul_3364: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_34, 1.0);  reciprocal_34 = None
	        mul_3367: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_535, mul_3364);  view_535 = mul_3364 = None
	        round_70: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3367);  mul_3367 = None
	        add_5329: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_70, view_537);  round_70 = view_537 = None
	        clamp_min_104: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5329, -128);  add_5329 = None
	        clamp_max_69: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_104, 127);  clamp_min_104 = None
	        view_538: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_69, [sym_size_int, 1500, 1280]);  clamp_max_69 = None
	        _assert_tensor_metadata_310 = torch.ops.aten._assert_tensor_metadata.default(view_538, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_310 = None
	        convert_element_type_205: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_538, torch.int8);  view_538 = None
	        view_539: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_205, [sym_size_int, 1500, 1280]);  convert_element_type_205 = None
	        view_540: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_102, [sym_size_int, 1500, 1]);  clamp_min_102 = None
	        view_541: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_204, [sym_size_int, 1500, 1]);  convert_element_type_204 = None
	        _assert_tensor_metadata_311 = torch.ops.aten._assert_tensor_metadata.default(view_539, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_311 = None
	        convert_element_type_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_539, torch.float32);  view_539 = None
	        _assert_tensor_metadata_312 = torch.ops.aten._assert_tensor_metadata.default(view_541, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_312 = None
	        convert_element_type_207: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_541, torch.float32);  view_541 = None
	        sub_1598: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_206, convert_element_type_207);  convert_element_type_206 = convert_element_type_207 = None
	        mul_3389: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1598, view_540);  sub_1598 = view_540 = None
	        view_542: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3389, [sym_size_int, 1500, 1280]);  mul_3389 = None
	        _assert_tensor_metadata_313 = torch.ops.aten._assert_tensor_metadata.default(view_542, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_313 = None
	        view_543: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_5_fc1_parametrizations_weight_original0 = None
	        view_544: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_5_fc1_parametrizations_weight_original1 = None
	        view_545: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_5_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_314 = torch.ops.aten._assert_tensor_metadata.default(view_543, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_314 = None
	        convert_element_type_208: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_543, torch.float32);  view_543 = None
	        _assert_tensor_metadata_315 = torch.ops.aten._assert_tensor_metadata.default(view_545, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_315 = None
	        convert_element_type_209: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_545, torch.float32);  view_545 = None
	        sub_1602: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_208, convert_element_type_209);  convert_element_type_208 = convert_element_type_209 = None
	        mul_3394: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1602, view_544);  sub_1602 = view_544 = None
	        view_546: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3394, [5120, 1280]);  mul_3394 = None
	        _assert_tensor_metadata_316 = torch.ops.aten._assert_tensor_metadata.default(view_546, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_316 = None
	        mul_3399: "Sym(1500*s6)" = sym_size_int * 1500
	        view_547: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_542, [mul_3399, 1280]);  view_542 = mul_3399 = None
	        permute_59: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_546, [1, 0]);  view_546 = None
	        addmm_28: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_5_fc1_bias, view_547, permute_59);  model_audio_tower_layers_5_fc1_bias = view_547 = permute_59 = None
	        view_548: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_28, [sym_size_int, 1500, 5120]);  addmm_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_3406: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_548, 0.5)
	        mul_3407: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_548, 0.7071067811865476);  view_548 = None
	        erf_7: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_3407);  mul_3407 = None
	        add_5388: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_7, 1);  erf_7 = None
	        mul_3408: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3406, add_5388);  mul_3406 = add_5388 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_47: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_3408);  mul_3408 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_549: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_47, [sym_size_int, 1500, 5120])
	        amin_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_549, [2])
	        amax_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_549, [2]);  view_549 = None
	        full_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_35, full_70);  amin_35 = full_70 = None
	        full_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_35, full_71);  amax_35 = full_71 = None
	        sub_1615: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_35, minimum_35);  maximum_35 = None
	        div_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1615, 255.0);  sub_1615 = None
	        clamp_min_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_70, 1.1920928955078125e-07);  div_70 = None
	        div_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_35, clamp_min_105);  minimum_35 = None
	        round_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_71);  div_71 = None
	        sub_1621: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_71);  round_71 = None
	        clamp_min_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1621, -128);  sub_1621 = None
	        clamp_max_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_106, 127);  clamp_min_106 = None
	        _assert_tensor_metadata_317 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_105, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_317 = None
	        _assert_tensor_metadata_318 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_70, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_318 = None
	        convert_element_type_210: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_70, torch.int8);  clamp_max_70 = None
	        view_550: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_47, [sym_size_int, 1500, 5120]);  clone_47 = None
	        view_551: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_105, [sym_size_int, 1500, 1])
	        view_552: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_210, [sym_size_int, 1500, 1])
	        reciprocal_35: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_551);  view_551 = None
	        mul_3454: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_35, 1.0);  reciprocal_35 = None
	        mul_3457: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_550, mul_3454);  view_550 = mul_3454 = None
	        round_72: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_3457);  mul_3457 = None
	        add_5471: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_72, view_552);  round_72 = view_552 = None
	        clamp_min_107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5471, -128);  add_5471 = None
	        clamp_max_71: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_107, 127);  clamp_min_107 = None
	        view_553: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_71, [sym_size_int, 1500, 5120]);  clamp_max_71 = None
	        _assert_tensor_metadata_319 = torch.ops.aten._assert_tensor_metadata.default(view_553, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_319 = None
	        convert_element_type_211: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_553, torch.int8);  view_553 = None
	        view_554: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_211, [sym_size_int, 1500, 5120]);  convert_element_type_211 = None
	        view_555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_105, [sym_size_int, 1500, 1]);  clamp_min_105 = None
	        view_556: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_210, [sym_size_int, 1500, 1]);  convert_element_type_210 = None
	        _assert_tensor_metadata_320 = torch.ops.aten._assert_tensor_metadata.default(view_554, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_320 = None
	        convert_element_type_212: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_554, torch.float32);  view_554 = None
	        _assert_tensor_metadata_321 = torch.ops.aten._assert_tensor_metadata.default(view_556, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_321 = None
	        convert_element_type_213: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_556, torch.float32);  view_556 = None
	        sub_1641: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_212, convert_element_type_213);  convert_element_type_212 = convert_element_type_213 = None
	        mul_3479: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1641, view_555);  sub_1641 = view_555 = None
	        view_557: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_3479, [sym_size_int, 1500, 5120]);  mul_3479 = None
	        _assert_tensor_metadata_322 = torch.ops.aten._assert_tensor_metadata.default(view_557, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_322 = None
	        view_558: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_5_fc2_parametrizations_weight_original0 = None
	        view_559: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_5_fc2_parametrizations_weight_original1 = None
	        view_560: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_5_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_323 = torch.ops.aten._assert_tensor_metadata.default(view_558, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_323 = None
	        convert_element_type_214: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_558, torch.float32);  view_558 = None
	        _assert_tensor_metadata_324 = torch.ops.aten._assert_tensor_metadata.default(view_560, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_324 = None
	        convert_element_type_215: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_560, torch.float32);  view_560 = None
	        sub_1645: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_214, convert_element_type_215);  convert_element_type_214 = convert_element_type_215 = None
	        mul_3484: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1645, view_559);  sub_1645 = view_559 = None
	        view_561: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_3484, [1280, 5120]);  mul_3484 = None
	        _assert_tensor_metadata_325 = torch.ops.aten._assert_tensor_metadata.default(view_561, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_325 = None
	        mul_3489: "Sym(1500*s6)" = sym_size_int * 1500
	        view_562: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_557, [mul_3489, 5120]);  view_557 = mul_3489 = None
	        permute_60: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_561, [1, 0]);  view_561 = None
	        addmm_29: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_5_fc2_bias, view_562, permute_60);  model_audio_tower_layers_5_fc2_bias = view_562 = permute_60 = None
	        view_563: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_29, [sym_size_int, 1500, 1280]);  addmm_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_48: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_563);  view_563 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_5534: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_5236, clone_48);  add_5236 = clone_48 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_49: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_5534, memory_format = torch.contiguous_format)
	        var_mean_12 = torch.ops.aten.var_mean.correction(clone_49, [2], correction = 0, keepdim = True)
	        getitem_48: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_12[0]
	        getitem_49: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_12[1];  var_mean_12 = None
	        add_5539: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_48, 1e-05);  getitem_48 = None
	        rsqrt_12: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_5539);  add_5539 = None
	        sub_1651: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_49, getitem_49);  clone_49 = getitem_49 = None
	        mul_3500: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1651, rsqrt_12);  sub_1651 = rsqrt_12 = None
	        mul_3501: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3500, model_audio_tower_layers_6_self_attn_layer_norm_weight);  mul_3500 = model_audio_tower_layers_6_self_attn_layer_norm_weight = None
	        add_5540: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_3501, model_audio_tower_layers_6_self_attn_layer_norm_bias);  mul_3501 = model_audio_tower_layers_6_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_564: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_5540, [sym_size_int, 1500, 1280])
	        amin_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_564, [2])
	        amax_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_564, [2]);  view_564 = None
	        full_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_36, full_72);  amin_36 = full_72 = None
	        full_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_36, full_73);  amax_36 = full_73 = None
	        sub_1662: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_36, minimum_36);  maximum_36 = None
	        div_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1662, 255.0);  sub_1662 = None
	        clamp_min_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_72, 1.1920928955078125e-07);  div_72 = None
	        div_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_36, clamp_min_108);  minimum_36 = None
	        round_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_73);  div_73 = None
	        sub_1668: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_73);  round_73 = None
	        clamp_min_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1668, -128);  sub_1668 = None
	        clamp_max_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_109, 127);  clamp_min_109 = None
	        _assert_tensor_metadata_326 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_108, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_326 = None
	        _assert_tensor_metadata_327 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_72, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_327 = None
	        convert_element_type_216: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_72, torch.int8);  clamp_max_72 = None
	        view_565: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_5540, [sym_size_int, 1500, 1280])
	        view_566: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_108, [sym_size_int, 1500, 1])
	        view_567: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_216, [sym_size_int, 1500, 1])
	        reciprocal_36: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_566);  view_566 = None
	        mul_3549: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_36, 1.0);  reciprocal_36 = None
	        mul_3552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_565, mul_3549);  view_565 = mul_3549 = None
	        round_74: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3552);  mul_3552 = None
	        add_5627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_74, view_567);  round_74 = view_567 = None
	        clamp_min_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5627, -128);  add_5627 = None
	        clamp_max_73: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_110, 127);  clamp_min_110 = None
	        view_568: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_73, [sym_size_int, 1500, 1280]);  clamp_max_73 = None
	        _assert_tensor_metadata_328 = torch.ops.aten._assert_tensor_metadata.default(view_568, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_328 = None
	        convert_element_type_217: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_568, torch.int8);  view_568 = None
	        view_569: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_217, [sym_size_int, 1500, 1280]);  convert_element_type_217 = None
	        view_570: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_108, [sym_size_int, 1500, 1]);  clamp_min_108 = None
	        view_571: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_216, [sym_size_int, 1500, 1]);  convert_element_type_216 = None
	        _assert_tensor_metadata_329 = torch.ops.aten._assert_tensor_metadata.default(view_569, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_329 = None
	        convert_element_type_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_569, torch.float32);  view_569 = None
	        _assert_tensor_metadata_330 = torch.ops.aten._assert_tensor_metadata.default(view_571, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_330 = None
	        convert_element_type_219: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_571, torch.float32);  view_571 = None
	        sub_1688: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_218, convert_element_type_219);  convert_element_type_218 = convert_element_type_219 = None
	        mul_3574: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1688, view_570);  sub_1688 = view_570 = None
	        view_572: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3574, [sym_size_int, 1500, 1280]);  mul_3574 = None
	        _assert_tensor_metadata_331 = torch.ops.aten._assert_tensor_metadata.default(view_572, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_331 = None
	        view_573: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_574: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_575: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_332 = torch.ops.aten._assert_tensor_metadata.default(view_573, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_332 = None
	        convert_element_type_220: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_573, torch.float32);  view_573 = None
	        _assert_tensor_metadata_333 = torch.ops.aten._assert_tensor_metadata.default(view_575, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_333 = None
	        convert_element_type_221: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_575, torch.float32);  view_575 = None
	        sub_1692: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_220, convert_element_type_221);  convert_element_type_220 = convert_element_type_221 = None
	        mul_3579: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1692, view_574);  sub_1692 = view_574 = None
	        view_576: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3579, [1280, 1280]);  mul_3579 = None
	        _assert_tensor_metadata_334 = torch.ops.aten._assert_tensor_metadata.default(view_576, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_334 = None
	        mul_3584: "Sym(1500*s6)" = sym_size_int * 1500
	        view_577: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_572, [mul_3584, 1280]);  view_572 = mul_3584 = None
	        permute_61: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_576, [1, 0]);  view_576 = None
	        addmm_30: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_6_self_attn_q_proj_bias, view_577, permute_61);  model_audio_tower_layers_6_self_attn_q_proj_bias = view_577 = permute_61 = None
	        view_578: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_30, [sym_size_int, 1500, 1280]);  addmm_30 = None
	        mul_3591: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_578, 0.125);  view_578 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_579: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_3591, [sym_size_int, 1500, 20, 64]);  mul_3591 = None
	        permute_62: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_579, [0, 2, 1, 3]);  view_579 = None
	        clone_50: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_62, memory_format = torch.contiguous_format);  permute_62 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_580: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_5540, [sym_size_int, 1500, 1280])
	        amin_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_580, [2])
	        amax_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_580, [2]);  view_580 = None
	        full_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_37, full_74);  amin_37 = full_74 = None
	        full_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_37, full_75);  amax_37 = full_75 = None
	        sub_1707: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_37, minimum_37);  maximum_37 = None
	        div_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1707, 255.0);  sub_1707 = None
	        clamp_min_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_74, 1.1920928955078125e-07);  div_74 = None
	        div_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_37, clamp_min_111);  minimum_37 = None
	        round_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_75);  div_75 = None
	        sub_1713: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_75);  round_75 = None
	        clamp_min_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1713, -128);  sub_1713 = None
	        clamp_max_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_112, 127);  clamp_min_112 = None
	        _assert_tensor_metadata_335 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_111, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_335 = None
	        _assert_tensor_metadata_336 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_74, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_336 = None
	        convert_element_type_222: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_74, torch.int8);  clamp_max_74 = None
	        view_581: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_5540, [sym_size_int, 1500, 1280])
	        view_582: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_111, [sym_size_int, 1500, 1])
	        view_583: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_222, [sym_size_int, 1500, 1])
	        reciprocal_37: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_582);  view_582 = None
	        mul_3645: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_37, 1.0);  reciprocal_37 = None
	        mul_3648: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_581, mul_3645);  view_581 = mul_3645 = None
	        round_76: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3648);  mul_3648 = None
	        add_5779: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_76, view_583);  round_76 = view_583 = None
	        clamp_min_113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5779, -128);  add_5779 = None
	        clamp_max_75: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_113, 127);  clamp_min_113 = None
	        view_584: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_75, [sym_size_int, 1500, 1280]);  clamp_max_75 = None
	        _assert_tensor_metadata_337 = torch.ops.aten._assert_tensor_metadata.default(view_584, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_337 = None
	        convert_element_type_223: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_584, torch.int8);  view_584 = None
	        view_585: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_223, [sym_size_int, 1500, 1280]);  convert_element_type_223 = None
	        view_586: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_111, [sym_size_int, 1500, 1]);  clamp_min_111 = None
	        view_587: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_222, [sym_size_int, 1500, 1]);  convert_element_type_222 = None
	        _assert_tensor_metadata_338 = torch.ops.aten._assert_tensor_metadata.default(view_585, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_338 = None
	        convert_element_type_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_585, torch.float32);  view_585 = None
	        _assert_tensor_metadata_339 = torch.ops.aten._assert_tensor_metadata.default(view_587, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_339 = None
	        convert_element_type_225: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_587, torch.float32);  view_587 = None
	        sub_1733: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_224, convert_element_type_225);  convert_element_type_224 = convert_element_type_225 = None
	        mul_3670: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1733, view_586);  sub_1733 = view_586 = None
	        view_588: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3670, [sym_size_int, 1500, 1280]);  mul_3670 = None
	        _assert_tensor_metadata_340 = torch.ops.aten._assert_tensor_metadata.default(view_588, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_340 = None
	        view_589: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_590: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_591: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_341 = torch.ops.aten._assert_tensor_metadata.default(view_589, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_341 = None
	        convert_element_type_226: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_589, torch.float32);  view_589 = None
	        _assert_tensor_metadata_342 = torch.ops.aten._assert_tensor_metadata.default(view_591, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_342 = None
	        convert_element_type_227: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_591, torch.float32);  view_591 = None
	        sub_1737: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_226, convert_element_type_227);  convert_element_type_226 = convert_element_type_227 = None
	        mul_3675: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1737, view_590);  sub_1737 = view_590 = None
	        view_592: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3675, [1280, 1280]);  mul_3675 = None
	        _assert_tensor_metadata_343 = torch.ops.aten._assert_tensor_metadata.default(view_592, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_343 = None
	        permute_63: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_592, [1, 0]);  view_592 = None
	        mul_3678: "Sym(1500*s6)" = sym_size_int * 1500
	        view_593: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_588, [mul_3678, 1280]);  view_588 = mul_3678 = None
	        mm_6: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_593, permute_63);  view_593 = permute_63 = None
	        view_594: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_6, [sym_size_int, 1500, 1280]);  mm_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_595: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_594, [sym_size_int, -1, 20, 64]);  view_594 = None
	        permute_64: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_595, [0, 2, 1, 3]);  view_595 = None
	        clone_51: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_64, memory_format = torch.contiguous_format);  permute_64 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_596: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_5540, [sym_size_int, 1500, 1280])
	        amin_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_596, [2])
	        amax_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_596, [2]);  view_596 = None
	        full_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_38, full_76);  amin_38 = full_76 = None
	        full_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_38, full_77);  amax_38 = full_77 = None
	        sub_1751: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_38, minimum_38);  maximum_38 = None
	        div_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1751, 255.0);  sub_1751 = None
	        clamp_min_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_76, 1.1920928955078125e-07);  div_76 = None
	        div_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_38, clamp_min_114);  minimum_38 = None
	        round_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_77);  div_77 = None
	        sub_1757: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_77);  round_77 = None
	        clamp_min_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1757, -128);  sub_1757 = None
	        clamp_max_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_115, 127);  clamp_min_115 = None
	        _assert_tensor_metadata_344 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_114, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_344 = None
	        _assert_tensor_metadata_345 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_76, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_345 = None
	        convert_element_type_228: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_76, torch.int8);  clamp_max_76 = None
	        view_597: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_5540, [sym_size_int, 1500, 1280]);  add_5540 = None
	        view_598: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_114, [sym_size_int, 1500, 1])
	        view_599: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_228, [sym_size_int, 1500, 1])
	        reciprocal_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_598);  view_598 = None
	        mul_3744: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_38, 1.0);  reciprocal_38 = None
	        mul_3747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_597, mul_3744);  view_597 = mul_3744 = None
	        round_78: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3747);  mul_3747 = None
	        add_5927: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_78, view_599);  round_78 = view_599 = None
	        clamp_min_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5927, -128);  add_5927 = None
	        clamp_max_77: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_116, 127);  clamp_min_116 = None
	        view_600: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_77, [sym_size_int, 1500, 1280]);  clamp_max_77 = None
	        _assert_tensor_metadata_346 = torch.ops.aten._assert_tensor_metadata.default(view_600, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_346 = None
	        convert_element_type_229: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_600, torch.int8);  view_600 = None
	        view_601: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_229, [sym_size_int, 1500, 1280]);  convert_element_type_229 = None
	        view_602: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_114, [sym_size_int, 1500, 1]);  clamp_min_114 = None
	        view_603: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_228, [sym_size_int, 1500, 1]);  convert_element_type_228 = None
	        _assert_tensor_metadata_347 = torch.ops.aten._assert_tensor_metadata.default(view_601, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_347 = None
	        convert_element_type_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_601, torch.float32);  view_601 = None
	        _assert_tensor_metadata_348 = torch.ops.aten._assert_tensor_metadata.default(view_603, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_348 = None
	        convert_element_type_231: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_603, torch.float32);  view_603 = None
	        sub_1777: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_230, convert_element_type_231);  convert_element_type_230 = convert_element_type_231 = None
	        mul_3769: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1777, view_602);  sub_1777 = view_602 = None
	        view_604: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3769, [sym_size_int, 1500, 1280]);  mul_3769 = None
	        _assert_tensor_metadata_349 = torch.ops.aten._assert_tensor_metadata.default(view_604, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_349 = None
	        view_605: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_606: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_607: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_350 = torch.ops.aten._assert_tensor_metadata.default(view_605, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_350 = None
	        convert_element_type_232: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_605, torch.float32);  view_605 = None
	        _assert_tensor_metadata_351 = torch.ops.aten._assert_tensor_metadata.default(view_607, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_351 = None
	        convert_element_type_233: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_607, torch.float32);  view_607 = None
	        sub_1781: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_232, convert_element_type_233);  convert_element_type_232 = convert_element_type_233 = None
	        mul_3774: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1781, view_606);  sub_1781 = view_606 = None
	        view_608: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3774, [1280, 1280]);  mul_3774 = None
	        _assert_tensor_metadata_352 = torch.ops.aten._assert_tensor_metadata.default(view_608, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_352 = None
	        mul_3779: "Sym(1500*s6)" = sym_size_int * 1500
	        view_609: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_604, [mul_3779, 1280]);  view_604 = mul_3779 = None
	        permute_65: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_608, [1, 0]);  view_608 = None
	        addmm_31: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_6_self_attn_v_proj_bias, view_609, permute_65);  model_audio_tower_layers_6_self_attn_v_proj_bias = view_609 = permute_65 = None
	        view_610: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_31, [sym_size_int, 1500, 1280]);  addmm_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_611: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_610, [sym_size_int, -1, 20, 64]);  view_610 = None
	        permute_66: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_611, [0, 2, 1, 3]);  view_611 = None
	        clone_52: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_66, memory_format = torch.contiguous_format);  permute_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_6 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_50, clone_51, clone_52, None, False, scale = 1.0);  clone_50 = clone_51 = clone_52 = None
	        getitem_50: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_6[0];  _scaled_dot_product_efficient_attention_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_67: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_50, [0, 2, 1, 3]);  getitem_50 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_67, [sym_size_int, 1500, -1]);  permute_67 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_613: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_612, [sym_size_int, 1500, 1280])
	        amin_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_613, [2])
	        amax_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_613, [2]);  view_613 = None
	        full_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_39, full_78);  amin_39 = full_78 = None
	        full_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_39, full_79);  amax_39 = full_79 = None
	        sub_1799: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_39, minimum_39);  maximum_39 = None
	        div_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1799, 255.0);  sub_1799 = None
	        clamp_min_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_78, 1.1920928955078125e-07);  div_78 = None
	        div_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_39, clamp_min_117);  minimum_39 = None
	        round_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_79);  div_79 = None
	        sub_1805: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_79);  round_79 = None
	        clamp_min_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1805, -128);  sub_1805 = None
	        clamp_max_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_118, 127);  clamp_min_118 = None
	        _assert_tensor_metadata_353 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_117, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_353 = None
	        _assert_tensor_metadata_354 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_78, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_354 = None
	        convert_element_type_234: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_78, torch.int8);  clamp_max_78 = None
	        view_614: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_612, [sym_size_int, 1500, 1280]);  view_612 = None
	        view_615: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_117, [sym_size_int, 1500, 1])
	        view_616: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_234, [sym_size_int, 1500, 1])
	        reciprocal_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_615);  view_615 = None
	        mul_3849: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_39, 1.0);  reciprocal_39 = None
	        mul_3852: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_614, mul_3849);  view_614 = mul_3849 = None
	        round_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3852);  mul_3852 = None
	        add_6091: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_80, view_616);  round_80 = view_616 = None
	        clamp_min_119: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6091, -128);  add_6091 = None
	        clamp_max_79: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_119, 127);  clamp_min_119 = None
	        view_617: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_79, [sym_size_int, 1500, 1280]);  clamp_max_79 = None
	        _assert_tensor_metadata_355 = torch.ops.aten._assert_tensor_metadata.default(view_617, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_355 = None
	        convert_element_type_235: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_617, torch.int8);  view_617 = None
	        view_618: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_235, [sym_size_int, 1500, 1280]);  convert_element_type_235 = None
	        view_619: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_117, [sym_size_int, 1500, 1]);  clamp_min_117 = None
	        view_620: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_234, [sym_size_int, 1500, 1]);  convert_element_type_234 = None
	        _assert_tensor_metadata_356 = torch.ops.aten._assert_tensor_metadata.default(view_618, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_356 = None
	        convert_element_type_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_618, torch.float32);  view_618 = None
	        _assert_tensor_metadata_357 = torch.ops.aten._assert_tensor_metadata.default(view_620, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_357 = None
	        convert_element_type_237: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_620, torch.float32);  view_620 = None
	        sub_1825: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_236, convert_element_type_237);  convert_element_type_236 = convert_element_type_237 = None
	        mul_3874: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1825, view_619);  sub_1825 = view_619 = None
	        view_621: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3874, [sym_size_int, 1500, 1280]);  mul_3874 = None
	        _assert_tensor_metadata_358 = torch.ops.aten._assert_tensor_metadata.default(view_621, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_358 = None
	        view_622: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_623: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_624: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_359 = torch.ops.aten._assert_tensor_metadata.default(view_622, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_359 = None
	        convert_element_type_238: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_622, torch.float32);  view_622 = None
	        _assert_tensor_metadata_360 = torch.ops.aten._assert_tensor_metadata.default(view_624, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_360 = None
	        convert_element_type_239: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_624, torch.float32);  view_624 = None
	        sub_1829: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_238, convert_element_type_239);  convert_element_type_238 = convert_element_type_239 = None
	        mul_3879: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1829, view_623);  sub_1829 = view_623 = None
	        view_625: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3879, [1280, 1280]);  mul_3879 = None
	        _assert_tensor_metadata_361 = torch.ops.aten._assert_tensor_metadata.default(view_625, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_361 = None
	        mul_3884: "Sym(1500*s6)" = sym_size_int * 1500
	        view_626: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_621, [mul_3884, 1280]);  view_621 = mul_3884 = None
	        permute_68: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_625, [1, 0]);  view_625 = None
	        addmm_32: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_6_self_attn_out_proj_bias, view_626, permute_68);  model_audio_tower_layers_6_self_attn_out_proj_bias = view_626 = permute_68 = None
	        view_627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_32, [sym_size_int, 1500, 1280]);  addmm_32 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_53: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_627);  view_627 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_6154: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_5534, clone_53);  add_5534 = clone_53 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_54: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_6154, memory_format = torch.contiguous_format)
	        var_mean_13 = torch.ops.aten.var_mean.correction(clone_54, [2], correction = 0, keepdim = True)
	        getitem_54: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_13[0]
	        getitem_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_13[1];  var_mean_13 = None
	        add_6159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_54, 1e-05);  getitem_54 = None
	        rsqrt_13: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_6159);  add_6159 = None
	        sub_1835: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_54, getitem_55);  clone_54 = getitem_55 = None
	        mul_3895: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1835, rsqrt_13);  sub_1835 = rsqrt_13 = None
	        mul_3896: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3895, model_audio_tower_layers_6_final_layer_norm_weight);  mul_3895 = model_audio_tower_layers_6_final_layer_norm_weight = None
	        add_6160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_3896, model_audio_tower_layers_6_final_layer_norm_bias);  mul_3896 = model_audio_tower_layers_6_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_628: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_6160, [sym_size_int, 1500, 1280])
	        amin_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_628, [2])
	        amax_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_628, [2]);  view_628 = None
	        full_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_40, full_80);  amin_40 = full_80 = None
	        full_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_40, full_81);  amax_40 = full_81 = None
	        sub_1846: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_40, minimum_40);  maximum_40 = None
	        div_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1846, 255.0);  sub_1846 = None
	        clamp_min_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_80, 1.1920928955078125e-07);  div_80 = None
	        div_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_40, clamp_min_120);  minimum_40 = None
	        round_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_81);  div_81 = None
	        sub_1852: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_81);  round_81 = None
	        clamp_min_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1852, -128);  sub_1852 = None
	        clamp_max_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_121, 127);  clamp_min_121 = None
	        _assert_tensor_metadata_362 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_120, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_362 = None
	        _assert_tensor_metadata_363 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_80, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_363 = None
	        convert_element_type_240: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_80, torch.int8);  clamp_max_80 = None
	        view_629: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_6160, [sym_size_int, 1500, 1280]);  add_6160 = None
	        view_630: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_120, [sym_size_int, 1500, 1])
	        view_631: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_240, [sym_size_int, 1500, 1])
	        reciprocal_40: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_630);  view_630 = None
	        mul_3944: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_40, 1.0);  reciprocal_40 = None
	        mul_3947: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_629, mul_3944);  view_629 = mul_3944 = None
	        round_82: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3947);  mul_3947 = None
	        add_6247: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_82, view_631);  round_82 = view_631 = None
	        clamp_min_122: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6247, -128);  add_6247 = None
	        clamp_max_81: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_122, 127);  clamp_min_122 = None
	        view_632: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_81, [sym_size_int, 1500, 1280]);  clamp_max_81 = None
	        _assert_tensor_metadata_364 = torch.ops.aten._assert_tensor_metadata.default(view_632, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_364 = None
	        convert_element_type_241: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_632, torch.int8);  view_632 = None
	        view_633: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_241, [sym_size_int, 1500, 1280]);  convert_element_type_241 = None
	        view_634: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_120, [sym_size_int, 1500, 1]);  clamp_min_120 = None
	        view_635: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_240, [sym_size_int, 1500, 1]);  convert_element_type_240 = None
	        _assert_tensor_metadata_365 = torch.ops.aten._assert_tensor_metadata.default(view_633, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_365 = None
	        convert_element_type_242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_633, torch.float32);  view_633 = None
	        _assert_tensor_metadata_366 = torch.ops.aten._assert_tensor_metadata.default(view_635, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_366 = None
	        convert_element_type_243: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_635, torch.float32);  view_635 = None
	        sub_1872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_242, convert_element_type_243);  convert_element_type_242 = convert_element_type_243 = None
	        mul_3969: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1872, view_634);  sub_1872 = view_634 = None
	        view_636: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3969, [sym_size_int, 1500, 1280]);  mul_3969 = None
	        _assert_tensor_metadata_367 = torch.ops.aten._assert_tensor_metadata.default(view_636, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_367 = None
	        view_637: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_6_fc1_parametrizations_weight_original0 = None
	        view_638: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_6_fc1_parametrizations_weight_original1 = None
	        view_639: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_6_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_368 = torch.ops.aten._assert_tensor_metadata.default(view_637, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_368 = None
	        convert_element_type_244: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_637, torch.float32);  view_637 = None
	        _assert_tensor_metadata_369 = torch.ops.aten._assert_tensor_metadata.default(view_639, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_369 = None
	        convert_element_type_245: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_639, torch.float32);  view_639 = None
	        sub_1876: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_244, convert_element_type_245);  convert_element_type_244 = convert_element_type_245 = None
	        mul_3974: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1876, view_638);  sub_1876 = view_638 = None
	        view_640: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3974, [5120, 1280]);  mul_3974 = None
	        _assert_tensor_metadata_370 = torch.ops.aten._assert_tensor_metadata.default(view_640, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_370 = None
	        mul_3979: "Sym(1500*s6)" = sym_size_int * 1500
	        view_641: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_636, [mul_3979, 1280]);  view_636 = mul_3979 = None
	        permute_69: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_640, [1, 0]);  view_640 = None
	        addmm_33: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_6_fc1_bias, view_641, permute_69);  model_audio_tower_layers_6_fc1_bias = view_641 = permute_69 = None
	        view_642: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_33, [sym_size_int, 1500, 5120]);  addmm_33 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_3986: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_642, 0.5)
	        mul_3987: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_642, 0.7071067811865476);  view_642 = None
	        erf_8: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_3987);  mul_3987 = None
	        add_6306: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_8, 1);  erf_8 = None
	        mul_3988: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3986, add_6306);  mul_3986 = add_6306 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_55: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_3988);  mul_3988 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_643: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_55, [sym_size_int, 1500, 5120])
	        amin_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_643, [2])
	        amax_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_643, [2]);  view_643 = None
	        full_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_41, full_82);  amin_41 = full_82 = None
	        full_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_41, full_83);  amax_41 = full_83 = None
	        sub_1889: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_41, minimum_41);  maximum_41 = None
	        div_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1889, 255.0);  sub_1889 = None
	        clamp_min_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_82, 1.1920928955078125e-07);  div_82 = None
	        div_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_41, clamp_min_123);  minimum_41 = None
	        round_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_83);  div_83 = None
	        sub_1895: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_83);  round_83 = None
	        clamp_min_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1895, -128);  sub_1895 = None
	        clamp_max_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_124, 127);  clamp_min_124 = None
	        _assert_tensor_metadata_371 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_123, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_371 = None
	        _assert_tensor_metadata_372 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_82, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_372 = None
	        convert_element_type_246: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_82, torch.int8);  clamp_max_82 = None
	        view_644: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_55, [sym_size_int, 1500, 5120]);  clone_55 = None
	        view_645: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_123, [sym_size_int, 1500, 1])
	        view_646: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_246, [sym_size_int, 1500, 1])
	        reciprocal_41: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_645);  view_645 = None
	        mul_4034: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_41, 1.0);  reciprocal_41 = None
	        mul_4037: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_644, mul_4034);  view_644 = mul_4034 = None
	        round_84: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_4037);  mul_4037 = None
	        add_6389: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_84, view_646);  round_84 = view_646 = None
	        clamp_min_125: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6389, -128);  add_6389 = None
	        clamp_max_83: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_125, 127);  clamp_min_125 = None
	        view_647: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_83, [sym_size_int, 1500, 5120]);  clamp_max_83 = None
	        _assert_tensor_metadata_373 = torch.ops.aten._assert_tensor_metadata.default(view_647, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_373 = None
	        convert_element_type_247: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_647, torch.int8);  view_647 = None
	        view_648: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_247, [sym_size_int, 1500, 5120]);  convert_element_type_247 = None
	        view_649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_123, [sym_size_int, 1500, 1]);  clamp_min_123 = None
	        view_650: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_246, [sym_size_int, 1500, 1]);  convert_element_type_246 = None
	        _assert_tensor_metadata_374 = torch.ops.aten._assert_tensor_metadata.default(view_648, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_374 = None
	        convert_element_type_248: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_648, torch.float32);  view_648 = None
	        _assert_tensor_metadata_375 = torch.ops.aten._assert_tensor_metadata.default(view_650, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_375 = None
	        convert_element_type_249: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_650, torch.float32);  view_650 = None
	        sub_1915: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_248, convert_element_type_249);  convert_element_type_248 = convert_element_type_249 = None
	        mul_4059: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1915, view_649);  sub_1915 = view_649 = None
	        view_651: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_4059, [sym_size_int, 1500, 5120]);  mul_4059 = None
	        _assert_tensor_metadata_376 = torch.ops.aten._assert_tensor_metadata.default(view_651, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_376 = None
	        view_652: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_6_fc2_parametrizations_weight_original0 = None
	        view_653: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_6_fc2_parametrizations_weight_original1 = None
	        view_654: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_6_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_377 = torch.ops.aten._assert_tensor_metadata.default(view_652, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_377 = None
	        convert_element_type_250: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_652, torch.float32);  view_652 = None
	        _assert_tensor_metadata_378 = torch.ops.aten._assert_tensor_metadata.default(view_654, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_378 = None
	        convert_element_type_251: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_654, torch.float32);  view_654 = None
	        sub_1919: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_250, convert_element_type_251);  convert_element_type_250 = convert_element_type_251 = None
	        mul_4064: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1919, view_653);  sub_1919 = view_653 = None
	        view_655: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_4064, [1280, 5120]);  mul_4064 = None
	        _assert_tensor_metadata_379 = torch.ops.aten._assert_tensor_metadata.default(view_655, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_379 = None
	        mul_4069: "Sym(1500*s6)" = sym_size_int * 1500
	        view_656: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_651, [mul_4069, 5120]);  view_651 = mul_4069 = None
	        permute_70: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_655, [1, 0]);  view_655 = None
	        addmm_34: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_6_fc2_bias, view_656, permute_70);  model_audio_tower_layers_6_fc2_bias = view_656 = permute_70 = None
	        view_657: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_34, [sym_size_int, 1500, 1280]);  addmm_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_657);  view_657 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_6452: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_6154, clone_56);  add_6154 = clone_56 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_57: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_6452, memory_format = torch.contiguous_format)
	        var_mean_14 = torch.ops.aten.var_mean.correction(clone_57, [2], correction = 0, keepdim = True)
	        getitem_56: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_14[0]
	        getitem_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_14[1];  var_mean_14 = None
	        add_6457: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_56, 1e-05);  getitem_56 = None
	        rsqrt_14: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_6457);  add_6457 = None
	        sub_1925: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_57, getitem_57);  clone_57 = getitem_57 = None
	        mul_4080: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1925, rsqrt_14);  sub_1925 = rsqrt_14 = None
	        mul_4081: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4080, model_audio_tower_layers_7_self_attn_layer_norm_weight);  mul_4080 = model_audio_tower_layers_7_self_attn_layer_norm_weight = None
	        add_6458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_4081, model_audio_tower_layers_7_self_attn_layer_norm_bias);  mul_4081 = model_audio_tower_layers_7_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_658: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_6458, [sym_size_int, 1500, 1280])
	        amin_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_658, [2])
	        amax_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_658, [2]);  view_658 = None
	        full_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_42, full_84);  amin_42 = full_84 = None
	        full_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_42, full_85);  amax_42 = full_85 = None
	        sub_1936: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_42, minimum_42);  maximum_42 = None
	        div_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1936, 255.0);  sub_1936 = None
	        clamp_min_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_84, 1.1920928955078125e-07);  div_84 = None
	        div_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_42, clamp_min_126);  minimum_42 = None
	        round_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_85);  div_85 = None
	        sub_1942: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_85);  round_85 = None
	        clamp_min_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1942, -128);  sub_1942 = None
	        clamp_max_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_127, 127);  clamp_min_127 = None
	        _assert_tensor_metadata_380 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_126, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_380 = None
	        _assert_tensor_metadata_381 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_84, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_381 = None
	        convert_element_type_252: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_84, torch.int8);  clamp_max_84 = None
	        view_659: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_6458, [sym_size_int, 1500, 1280])
	        view_660: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_126, [sym_size_int, 1500, 1])
	        view_661: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_252, [sym_size_int, 1500, 1])
	        reciprocal_42: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_660);  view_660 = None
	        mul_4129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_42, 1.0);  reciprocal_42 = None
	        mul_4132: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_659, mul_4129);  view_659 = mul_4129 = None
	        round_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4132);  mul_4132 = None
	        add_6545: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_86, view_661);  round_86 = view_661 = None
	        clamp_min_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6545, -128);  add_6545 = None
	        clamp_max_85: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_128, 127);  clamp_min_128 = None
	        view_662: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_85, [sym_size_int, 1500, 1280]);  clamp_max_85 = None
	        _assert_tensor_metadata_382 = torch.ops.aten._assert_tensor_metadata.default(view_662, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_382 = None
	        convert_element_type_253: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_662, torch.int8);  view_662 = None
	        view_663: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_253, [sym_size_int, 1500, 1280]);  convert_element_type_253 = None
	        view_664: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_126, [sym_size_int, 1500, 1]);  clamp_min_126 = None
	        view_665: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_252, [sym_size_int, 1500, 1]);  convert_element_type_252 = None
	        _assert_tensor_metadata_383 = torch.ops.aten._assert_tensor_metadata.default(view_663, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_383 = None
	        convert_element_type_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_663, torch.float32);  view_663 = None
	        _assert_tensor_metadata_384 = torch.ops.aten._assert_tensor_metadata.default(view_665, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_384 = None
	        convert_element_type_255: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_665, torch.float32);  view_665 = None
	        sub_1962: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_254, convert_element_type_255);  convert_element_type_254 = convert_element_type_255 = None
	        mul_4154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1962, view_664);  sub_1962 = view_664 = None
	        view_666: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4154, [sym_size_int, 1500, 1280]);  mul_4154 = None
	        _assert_tensor_metadata_385 = torch.ops.aten._assert_tensor_metadata.default(view_666, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_385 = None
	        view_667: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_668: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_669: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_386 = torch.ops.aten._assert_tensor_metadata.default(view_667, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_386 = None
	        convert_element_type_256: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_667, torch.float32);  view_667 = None
	        _assert_tensor_metadata_387 = torch.ops.aten._assert_tensor_metadata.default(view_669, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_387 = None
	        convert_element_type_257: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_669, torch.float32);  view_669 = None
	        sub_1966: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_256, convert_element_type_257);  convert_element_type_256 = convert_element_type_257 = None
	        mul_4159: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1966, view_668);  sub_1966 = view_668 = None
	        view_670: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4159, [1280, 1280]);  mul_4159 = None
	        _assert_tensor_metadata_388 = torch.ops.aten._assert_tensor_metadata.default(view_670, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_388 = None
	        mul_4164: "Sym(1500*s6)" = sym_size_int * 1500
	        view_671: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_666, [mul_4164, 1280]);  view_666 = mul_4164 = None
	        permute_71: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_670, [1, 0]);  view_670 = None
	        addmm_35: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_7_self_attn_q_proj_bias, view_671, permute_71);  model_audio_tower_layers_7_self_attn_q_proj_bias = view_671 = permute_71 = None
	        view_672: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_35, [sym_size_int, 1500, 1280]);  addmm_35 = None
	        mul_4171: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_672, 0.125);  view_672 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_673: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_4171, [sym_size_int, 1500, 20, 64]);  mul_4171 = None
	        permute_72: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_673, [0, 2, 1, 3]);  view_673 = None
	        clone_58: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_72, memory_format = torch.contiguous_format);  permute_72 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_674: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_6458, [sym_size_int, 1500, 1280])
	        amin_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_674, [2])
	        amax_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_674, [2]);  view_674 = None
	        full_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_43, full_86);  amin_43 = full_86 = None
	        full_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_43, full_87);  amax_43 = full_87 = None
	        sub_1981: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_43, minimum_43);  maximum_43 = None
	        div_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1981, 255.0);  sub_1981 = None
	        clamp_min_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_86, 1.1920928955078125e-07);  div_86 = None
	        div_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_43, clamp_min_129);  minimum_43 = None
	        round_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_87);  div_87 = None
	        sub_1987: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_87);  round_87 = None
	        clamp_min_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1987, -128);  sub_1987 = None
	        clamp_max_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_130, 127);  clamp_min_130 = None
	        _assert_tensor_metadata_389 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_129, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_389 = None
	        _assert_tensor_metadata_390 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_86, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_390 = None
	        convert_element_type_258: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_86, torch.int8);  clamp_max_86 = None
	        view_675: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_6458, [sym_size_int, 1500, 1280])
	        view_676: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_129, [sym_size_int, 1500, 1])
	        view_677: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_258, [sym_size_int, 1500, 1])
	        reciprocal_43: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_676);  view_676 = None
	        mul_4225: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_43, 1.0);  reciprocal_43 = None
	        mul_4228: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_675, mul_4225);  view_675 = mul_4225 = None
	        round_88: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4228);  mul_4228 = None
	        add_6697: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_88, view_677);  round_88 = view_677 = None
	        clamp_min_131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6697, -128);  add_6697 = None
	        clamp_max_87: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_131, 127);  clamp_min_131 = None
	        view_678: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_87, [sym_size_int, 1500, 1280]);  clamp_max_87 = None
	        _assert_tensor_metadata_391 = torch.ops.aten._assert_tensor_metadata.default(view_678, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_391 = None
	        convert_element_type_259: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_678, torch.int8);  view_678 = None
	        view_679: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_259, [sym_size_int, 1500, 1280]);  convert_element_type_259 = None
	        view_680: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_129, [sym_size_int, 1500, 1]);  clamp_min_129 = None
	        view_681: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_258, [sym_size_int, 1500, 1]);  convert_element_type_258 = None
	        _assert_tensor_metadata_392 = torch.ops.aten._assert_tensor_metadata.default(view_679, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_392 = None
	        convert_element_type_260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_679, torch.float32);  view_679 = None
	        _assert_tensor_metadata_393 = torch.ops.aten._assert_tensor_metadata.default(view_681, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_393 = None
	        convert_element_type_261: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_681, torch.float32);  view_681 = None
	        sub_2007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_260, convert_element_type_261);  convert_element_type_260 = convert_element_type_261 = None
	        mul_4250: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2007, view_680);  sub_2007 = view_680 = None
	        view_682: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4250, [sym_size_int, 1500, 1280]);  mul_4250 = None
	        _assert_tensor_metadata_394 = torch.ops.aten._assert_tensor_metadata.default(view_682, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_394 = None
	        view_683: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_684: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_685: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_395 = torch.ops.aten._assert_tensor_metadata.default(view_683, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_395 = None
	        convert_element_type_262: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_683, torch.float32);  view_683 = None
	        _assert_tensor_metadata_396 = torch.ops.aten._assert_tensor_metadata.default(view_685, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_396 = None
	        convert_element_type_263: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_685, torch.float32);  view_685 = None
	        sub_2011: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_262, convert_element_type_263);  convert_element_type_262 = convert_element_type_263 = None
	        mul_4255: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2011, view_684);  sub_2011 = view_684 = None
	        view_686: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4255, [1280, 1280]);  mul_4255 = None
	        _assert_tensor_metadata_397 = torch.ops.aten._assert_tensor_metadata.default(view_686, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_397 = None
	        permute_73: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_686, [1, 0]);  view_686 = None
	        mul_4258: "Sym(1500*s6)" = sym_size_int * 1500
	        view_687: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_682, [mul_4258, 1280]);  view_682 = mul_4258 = None
	        mm_7: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_687, permute_73);  view_687 = permute_73 = None
	        view_688: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_7, [sym_size_int, 1500, 1280]);  mm_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_689: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_688, [sym_size_int, -1, 20, 64]);  view_688 = None
	        permute_74: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_689, [0, 2, 1, 3]);  view_689 = None
	        clone_59: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_74, memory_format = torch.contiguous_format);  permute_74 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_690: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_6458, [sym_size_int, 1500, 1280])
	        amin_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_690, [2])
	        amax_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_690, [2]);  view_690 = None
	        full_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_44, full_88);  amin_44 = full_88 = None
	        full_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_44, full_89);  amax_44 = full_89 = None
	        sub_2025: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_44, minimum_44);  maximum_44 = None
	        div_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2025, 255.0);  sub_2025 = None
	        clamp_min_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_88, 1.1920928955078125e-07);  div_88 = None
	        div_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_44, clamp_min_132);  minimum_44 = None
	        round_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_89);  div_89 = None
	        sub_2031: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_89);  round_89 = None
	        clamp_min_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2031, -128);  sub_2031 = None
	        clamp_max_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_133, 127);  clamp_min_133 = None
	        _assert_tensor_metadata_398 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_132, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_398 = None
	        _assert_tensor_metadata_399 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_88, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_399 = None
	        convert_element_type_264: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_88, torch.int8);  clamp_max_88 = None
	        view_691: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_6458, [sym_size_int, 1500, 1280]);  add_6458 = None
	        view_692: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_132, [sym_size_int, 1500, 1])
	        view_693: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_264, [sym_size_int, 1500, 1])
	        reciprocal_44: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_692);  view_692 = None
	        mul_4324: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_44, 1.0);  reciprocal_44 = None
	        mul_4327: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_691, mul_4324);  view_691 = mul_4324 = None
	        round_90: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4327);  mul_4327 = None
	        add_6845: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_90, view_693);  round_90 = view_693 = None
	        clamp_min_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6845, -128);  add_6845 = None
	        clamp_max_89: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_134, 127);  clamp_min_134 = None
	        view_694: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_89, [sym_size_int, 1500, 1280]);  clamp_max_89 = None
	        _assert_tensor_metadata_400 = torch.ops.aten._assert_tensor_metadata.default(view_694, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_400 = None
	        convert_element_type_265: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_694, torch.int8);  view_694 = None
	        view_695: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_265, [sym_size_int, 1500, 1280]);  convert_element_type_265 = None
	        view_696: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_132, [sym_size_int, 1500, 1]);  clamp_min_132 = None
	        view_697: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_264, [sym_size_int, 1500, 1]);  convert_element_type_264 = None
	        _assert_tensor_metadata_401 = torch.ops.aten._assert_tensor_metadata.default(view_695, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_401 = None
	        convert_element_type_266: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_695, torch.float32);  view_695 = None
	        _assert_tensor_metadata_402 = torch.ops.aten._assert_tensor_metadata.default(view_697, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_402 = None
	        convert_element_type_267: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_697, torch.float32);  view_697 = None
	        sub_2051: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_266, convert_element_type_267);  convert_element_type_266 = convert_element_type_267 = None
	        mul_4349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2051, view_696);  sub_2051 = view_696 = None
	        view_698: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4349, [sym_size_int, 1500, 1280]);  mul_4349 = None
	        _assert_tensor_metadata_403 = torch.ops.aten._assert_tensor_metadata.default(view_698, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_403 = None
	        view_699: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_700: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_701: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_404 = torch.ops.aten._assert_tensor_metadata.default(view_699, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_404 = None
	        convert_element_type_268: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_699, torch.float32);  view_699 = None
	        _assert_tensor_metadata_405 = torch.ops.aten._assert_tensor_metadata.default(view_701, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_405 = None
	        convert_element_type_269: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_701, torch.float32);  view_701 = None
	        sub_2055: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_268, convert_element_type_269);  convert_element_type_268 = convert_element_type_269 = None
	        mul_4354: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2055, view_700);  sub_2055 = view_700 = None
	        view_702: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4354, [1280, 1280]);  mul_4354 = None
	        _assert_tensor_metadata_406 = torch.ops.aten._assert_tensor_metadata.default(view_702, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_406 = None
	        mul_4359: "Sym(1500*s6)" = sym_size_int * 1500
	        view_703: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_698, [mul_4359, 1280]);  view_698 = mul_4359 = None
	        permute_75: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_702, [1, 0]);  view_702 = None
	        addmm_36: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_7_self_attn_v_proj_bias, view_703, permute_75);  model_audio_tower_layers_7_self_attn_v_proj_bias = view_703 = permute_75 = None
	        view_704: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_36, [sym_size_int, 1500, 1280]);  addmm_36 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_705: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_704, [sym_size_int, -1, 20, 64]);  view_704 = None
	        permute_76: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_705, [0, 2, 1, 3]);  view_705 = None
	        clone_60: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_76, memory_format = torch.contiguous_format);  permute_76 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_7 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_58, clone_59, clone_60, None, False, scale = 1.0);  clone_58 = clone_59 = clone_60 = None
	        getitem_58: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_7[0];  _scaled_dot_product_efficient_attention_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_77: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_58, [0, 2, 1, 3]);  getitem_58 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_706: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_77, [sym_size_int, 1500, -1]);  permute_77 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_707: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_706, [sym_size_int, 1500, 1280])
	        amin_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_707, [2])
	        amax_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_707, [2]);  view_707 = None
	        full_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_45, full_90);  amin_45 = full_90 = None
	        full_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_45, full_91);  amax_45 = full_91 = None
	        sub_2073: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_45, minimum_45);  maximum_45 = None
	        div_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2073, 255.0);  sub_2073 = None
	        clamp_min_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_90, 1.1920928955078125e-07);  div_90 = None
	        div_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_45, clamp_min_135);  minimum_45 = None
	        round_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_91);  div_91 = None
	        sub_2079: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_91);  round_91 = None
	        clamp_min_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2079, -128);  sub_2079 = None
	        clamp_max_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_136, 127);  clamp_min_136 = None
	        _assert_tensor_metadata_407 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_135, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_407 = None
	        _assert_tensor_metadata_408 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_90, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_408 = None
	        convert_element_type_270: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_90, torch.int8);  clamp_max_90 = None
	        view_708: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_706, [sym_size_int, 1500, 1280]);  view_706 = None
	        view_709: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_135, [sym_size_int, 1500, 1])
	        view_710: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_270, [sym_size_int, 1500, 1])
	        reciprocal_45: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_709);  view_709 = None
	        mul_4429: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_45, 1.0);  reciprocal_45 = None
	        mul_4432: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_708, mul_4429);  view_708 = mul_4429 = None
	        round_92: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4432);  mul_4432 = None
	        add_7009: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_92, view_710);  round_92 = view_710 = None
	        clamp_min_137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7009, -128);  add_7009 = None
	        clamp_max_91: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_137, 127);  clamp_min_137 = None
	        view_711: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_91, [sym_size_int, 1500, 1280]);  clamp_max_91 = None
	        _assert_tensor_metadata_409 = torch.ops.aten._assert_tensor_metadata.default(view_711, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_409 = None
	        convert_element_type_271: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_711, torch.int8);  view_711 = None
	        view_712: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_271, [sym_size_int, 1500, 1280]);  convert_element_type_271 = None
	        view_713: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_135, [sym_size_int, 1500, 1]);  clamp_min_135 = None
	        view_714: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_270, [sym_size_int, 1500, 1]);  convert_element_type_270 = None
	        _assert_tensor_metadata_410 = torch.ops.aten._assert_tensor_metadata.default(view_712, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_410 = None
	        convert_element_type_272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_712, torch.float32);  view_712 = None
	        _assert_tensor_metadata_411 = torch.ops.aten._assert_tensor_metadata.default(view_714, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_411 = None
	        convert_element_type_273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_714, torch.float32);  view_714 = None
	        sub_2099: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_272, convert_element_type_273);  convert_element_type_272 = convert_element_type_273 = None
	        mul_4454: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2099, view_713);  sub_2099 = view_713 = None
	        view_715: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4454, [sym_size_int, 1500, 1280]);  mul_4454 = None
	        _assert_tensor_metadata_412 = torch.ops.aten._assert_tensor_metadata.default(view_715, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_412 = None
	        view_716: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_717: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_718: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_413 = torch.ops.aten._assert_tensor_metadata.default(view_716, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_413 = None
	        convert_element_type_274: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_716, torch.float32);  view_716 = None
	        _assert_tensor_metadata_414 = torch.ops.aten._assert_tensor_metadata.default(view_718, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_414 = None
	        convert_element_type_275: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_718, torch.float32);  view_718 = None
	        sub_2103: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_274, convert_element_type_275);  convert_element_type_274 = convert_element_type_275 = None
	        mul_4459: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2103, view_717);  sub_2103 = view_717 = None
	        view_719: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4459, [1280, 1280]);  mul_4459 = None
	        _assert_tensor_metadata_415 = torch.ops.aten._assert_tensor_metadata.default(view_719, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_415 = None
	        mul_4464: "Sym(1500*s6)" = sym_size_int * 1500
	        view_720: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_715, [mul_4464, 1280]);  view_715 = mul_4464 = None
	        permute_78: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_719, [1, 0]);  view_719 = None
	        addmm_37: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_7_self_attn_out_proj_bias, view_720, permute_78);  model_audio_tower_layers_7_self_attn_out_proj_bias = view_720 = permute_78 = None
	        view_721: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_37, [sym_size_int, 1500, 1280]);  addmm_37 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_61: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_721);  view_721 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_7072: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_6452, clone_61);  add_6452 = clone_61 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_7072, memory_format = torch.contiguous_format)
	        var_mean_15 = torch.ops.aten.var_mean.correction(clone_62, [2], correction = 0, keepdim = True)
	        getitem_62: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_15[0]
	        getitem_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_15[1];  var_mean_15 = None
	        add_7077: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_62, 1e-05);  getitem_62 = None
	        rsqrt_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_7077);  add_7077 = None
	        sub_2109: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_62, getitem_63);  clone_62 = getitem_63 = None
	        mul_4475: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2109, rsqrt_15);  sub_2109 = rsqrt_15 = None
	        mul_4476: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4475, model_audio_tower_layers_7_final_layer_norm_weight);  mul_4475 = model_audio_tower_layers_7_final_layer_norm_weight = None
	        add_7078: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_4476, model_audio_tower_layers_7_final_layer_norm_bias);  mul_4476 = model_audio_tower_layers_7_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_722: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7078, [sym_size_int, 1500, 1280])
	        amin_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_722, [2])
	        amax_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_722, [2]);  view_722 = None
	        full_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_46, full_92);  amin_46 = full_92 = None
	        full_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_46, full_93);  amax_46 = full_93 = None
	        sub_2120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_46, minimum_46);  maximum_46 = None
	        div_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2120, 255.0);  sub_2120 = None
	        clamp_min_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_92, 1.1920928955078125e-07);  div_92 = None
	        div_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_46, clamp_min_138);  minimum_46 = None
	        round_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_93);  div_93 = None
	        sub_2126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_93);  round_93 = None
	        clamp_min_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2126, -128);  sub_2126 = None
	        clamp_max_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_139, 127);  clamp_min_139 = None
	        _assert_tensor_metadata_416 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_138, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_416 = None
	        _assert_tensor_metadata_417 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_92, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_417 = None
	        convert_element_type_276: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_92, torch.int8);  clamp_max_92 = None
	        view_723: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7078, [sym_size_int, 1500, 1280]);  add_7078 = None
	        view_724: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_138, [sym_size_int, 1500, 1])
	        view_725: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_276, [sym_size_int, 1500, 1])
	        reciprocal_46: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_724);  view_724 = None
	        mul_4524: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_46, 1.0);  reciprocal_46 = None
	        mul_4527: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_723, mul_4524);  view_723 = mul_4524 = None
	        round_94: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4527);  mul_4527 = None
	        add_7165: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_94, view_725);  round_94 = view_725 = None
	        clamp_min_140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7165, -128);  add_7165 = None
	        clamp_max_93: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_140, 127);  clamp_min_140 = None
	        view_726: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_93, [sym_size_int, 1500, 1280]);  clamp_max_93 = None
	        _assert_tensor_metadata_418 = torch.ops.aten._assert_tensor_metadata.default(view_726, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_418 = None
	        convert_element_type_277: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_726, torch.int8);  view_726 = None
	        view_727: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_277, [sym_size_int, 1500, 1280]);  convert_element_type_277 = None
	        view_728: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_138, [sym_size_int, 1500, 1]);  clamp_min_138 = None
	        view_729: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_276, [sym_size_int, 1500, 1]);  convert_element_type_276 = None
	        _assert_tensor_metadata_419 = torch.ops.aten._assert_tensor_metadata.default(view_727, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_419 = None
	        convert_element_type_278: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_727, torch.float32);  view_727 = None
	        _assert_tensor_metadata_420 = torch.ops.aten._assert_tensor_metadata.default(view_729, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_420 = None
	        convert_element_type_279: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_729, torch.float32);  view_729 = None
	        sub_2146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_278, convert_element_type_279);  convert_element_type_278 = convert_element_type_279 = None
	        mul_4549: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2146, view_728);  sub_2146 = view_728 = None
	        view_730: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4549, [sym_size_int, 1500, 1280]);  mul_4549 = None
	        _assert_tensor_metadata_421 = torch.ops.aten._assert_tensor_metadata.default(view_730, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_421 = None
	        view_731: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_7_fc1_parametrizations_weight_original0 = None
	        view_732: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_7_fc1_parametrizations_weight_original1 = None
	        view_733: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_7_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_422 = torch.ops.aten._assert_tensor_metadata.default(view_731, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_422 = None
	        convert_element_type_280: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_731, torch.float32);  view_731 = None
	        _assert_tensor_metadata_423 = torch.ops.aten._assert_tensor_metadata.default(view_733, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_423 = None
	        convert_element_type_281: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_733, torch.float32);  view_733 = None
	        sub_2150: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_280, convert_element_type_281);  convert_element_type_280 = convert_element_type_281 = None
	        mul_4554: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2150, view_732);  sub_2150 = view_732 = None
	        view_734: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4554, [5120, 1280]);  mul_4554 = None
	        _assert_tensor_metadata_424 = torch.ops.aten._assert_tensor_metadata.default(view_734, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_424 = None
	        mul_4559: "Sym(1500*s6)" = sym_size_int * 1500
	        view_735: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_730, [mul_4559, 1280]);  view_730 = mul_4559 = None
	        permute_79: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_734, [1, 0]);  view_734 = None
	        addmm_38: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_7_fc1_bias, view_735, permute_79);  model_audio_tower_layers_7_fc1_bias = view_735 = permute_79 = None
	        view_736: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_38, [sym_size_int, 1500, 5120]);  addmm_38 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_4566: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_736, 0.5)
	        mul_4567: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_736, 0.7071067811865476);  view_736 = None
	        erf_9: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_4567);  mul_4567 = None
	        add_7224: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_9, 1);  erf_9 = None
	        mul_4568: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4566, add_7224);  mul_4566 = add_7224 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_63: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_4568);  mul_4568 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_737: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_63, [sym_size_int, 1500, 5120])
	        amin_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_737, [2])
	        amax_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_737, [2]);  view_737 = None
	        full_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_47, full_94);  amin_47 = full_94 = None
	        full_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_47, full_95);  amax_47 = full_95 = None
	        sub_2163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_47, minimum_47);  maximum_47 = None
	        div_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2163, 255.0);  sub_2163 = None
	        clamp_min_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_94, 1.1920928955078125e-07);  div_94 = None
	        div_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_47, clamp_min_141);  minimum_47 = None
	        round_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_95);  div_95 = None
	        sub_2169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_95);  round_95 = None
	        clamp_min_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2169, -128);  sub_2169 = None
	        clamp_max_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_142, 127);  clamp_min_142 = None
	        _assert_tensor_metadata_425 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_141, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_425 = None
	        _assert_tensor_metadata_426 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_94, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_426 = None
	        convert_element_type_282: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_94, torch.int8);  clamp_max_94 = None
	        view_738: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_63, [sym_size_int, 1500, 5120]);  clone_63 = None
	        view_739: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_141, [sym_size_int, 1500, 1])
	        view_740: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_282, [sym_size_int, 1500, 1])
	        reciprocal_47: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_739);  view_739 = None
	        mul_4614: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_47, 1.0);  reciprocal_47 = None
	        mul_4617: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_738, mul_4614);  view_738 = mul_4614 = None
	        round_96: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_4617);  mul_4617 = None
	        add_7307: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_96, view_740);  round_96 = view_740 = None
	        clamp_min_143: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7307, -128);  add_7307 = None
	        clamp_max_95: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_143, 127);  clamp_min_143 = None
	        view_741: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_95, [sym_size_int, 1500, 5120]);  clamp_max_95 = None
	        _assert_tensor_metadata_427 = torch.ops.aten._assert_tensor_metadata.default(view_741, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_427 = None
	        convert_element_type_283: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_741, torch.int8);  view_741 = None
	        view_742: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_283, [sym_size_int, 1500, 5120]);  convert_element_type_283 = None
	        view_743: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_141, [sym_size_int, 1500, 1]);  clamp_min_141 = None
	        view_744: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_282, [sym_size_int, 1500, 1]);  convert_element_type_282 = None
	        _assert_tensor_metadata_428 = torch.ops.aten._assert_tensor_metadata.default(view_742, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_428 = None
	        convert_element_type_284: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_742, torch.float32);  view_742 = None
	        _assert_tensor_metadata_429 = torch.ops.aten._assert_tensor_metadata.default(view_744, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_429 = None
	        convert_element_type_285: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_744, torch.float32);  view_744 = None
	        sub_2189: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_284, convert_element_type_285);  convert_element_type_284 = convert_element_type_285 = None
	        mul_4639: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2189, view_743);  sub_2189 = view_743 = None
	        view_745: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_4639, [sym_size_int, 1500, 5120]);  mul_4639 = None
	        _assert_tensor_metadata_430 = torch.ops.aten._assert_tensor_metadata.default(view_745, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_430 = None
	        view_746: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_7_fc2_parametrizations_weight_original0 = None
	        view_747: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_7_fc2_parametrizations_weight_original1 = None
	        view_748: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_7_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_431 = torch.ops.aten._assert_tensor_metadata.default(view_746, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_431 = None
	        convert_element_type_286: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_746, torch.float32);  view_746 = None
	        _assert_tensor_metadata_432 = torch.ops.aten._assert_tensor_metadata.default(view_748, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_432 = None
	        convert_element_type_287: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_748, torch.float32);  view_748 = None
	        sub_2193: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_286, convert_element_type_287);  convert_element_type_286 = convert_element_type_287 = None
	        mul_4644: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2193, view_747);  sub_2193 = view_747 = None
	        view_749: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_4644, [1280, 5120]);  mul_4644 = None
	        _assert_tensor_metadata_433 = torch.ops.aten._assert_tensor_metadata.default(view_749, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_433 = None
	        mul_4649: "Sym(1500*s6)" = sym_size_int * 1500
	        view_750: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_745, [mul_4649, 5120]);  view_745 = mul_4649 = None
	        permute_80: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_749, [1, 0]);  view_749 = None
	        addmm_39: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_7_fc2_bias, view_750, permute_80);  model_audio_tower_layers_7_fc2_bias = view_750 = permute_80 = None
	        view_751: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_39, [sym_size_int, 1500, 1280]);  addmm_39 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_64: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_751);  view_751 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_7370: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_7072, clone_64);  add_7072 = clone_64 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_65: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_7370, memory_format = torch.contiguous_format)
	        var_mean_16 = torch.ops.aten.var_mean.correction(clone_65, [2], correction = 0, keepdim = True)
	        getitem_64: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_16[0]
	        getitem_65: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_16[1];  var_mean_16 = None
	        add_7375: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_64, 1e-05);  getitem_64 = None
	        rsqrt_16: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_7375);  add_7375 = None
	        sub_2199: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_65, getitem_65);  clone_65 = getitem_65 = None
	        mul_4660: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2199, rsqrt_16);  sub_2199 = rsqrt_16 = None
	        mul_4661: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4660, model_audio_tower_layers_8_self_attn_layer_norm_weight);  mul_4660 = model_audio_tower_layers_8_self_attn_layer_norm_weight = None
	        add_7376: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_4661, model_audio_tower_layers_8_self_attn_layer_norm_bias);  mul_4661 = model_audio_tower_layers_8_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_752: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7376, [sym_size_int, 1500, 1280])
	        amin_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_752, [2])
	        amax_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_752, [2]);  view_752 = None
	        full_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_48, full_96);  amin_48 = full_96 = None
	        full_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_48, full_97);  amax_48 = full_97 = None
	        sub_2210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_48, minimum_48);  maximum_48 = None
	        div_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2210, 255.0);  sub_2210 = None
	        clamp_min_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_96, 1.1920928955078125e-07);  div_96 = None
	        div_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_48, clamp_min_144);  minimum_48 = None
	        round_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_97);  div_97 = None
	        sub_2216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_97);  round_97 = None
	        clamp_min_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2216, -128);  sub_2216 = None
	        clamp_max_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_145, 127);  clamp_min_145 = None
	        _assert_tensor_metadata_434 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_144, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_434 = None
	        _assert_tensor_metadata_435 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_96, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_435 = None
	        convert_element_type_288: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_96, torch.int8);  clamp_max_96 = None
	        view_753: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7376, [sym_size_int, 1500, 1280])
	        view_754: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_144, [sym_size_int, 1500, 1])
	        view_755: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_288, [sym_size_int, 1500, 1])
	        reciprocal_48: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_754);  view_754 = None
	        mul_4709: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_48, 1.0);  reciprocal_48 = None
	        mul_4712: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_753, mul_4709);  view_753 = mul_4709 = None
	        round_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4712);  mul_4712 = None
	        add_7463: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_98, view_755);  round_98 = view_755 = None
	        clamp_min_146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7463, -128);  add_7463 = None
	        clamp_max_97: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_146, 127);  clamp_min_146 = None
	        view_756: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_97, [sym_size_int, 1500, 1280]);  clamp_max_97 = None
	        _assert_tensor_metadata_436 = torch.ops.aten._assert_tensor_metadata.default(view_756, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_436 = None
	        convert_element_type_289: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_756, torch.int8);  view_756 = None
	        view_757: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_289, [sym_size_int, 1500, 1280]);  convert_element_type_289 = None
	        view_758: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_144, [sym_size_int, 1500, 1]);  clamp_min_144 = None
	        view_759: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_288, [sym_size_int, 1500, 1]);  convert_element_type_288 = None
	        _assert_tensor_metadata_437 = torch.ops.aten._assert_tensor_metadata.default(view_757, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_437 = None
	        convert_element_type_290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_757, torch.float32);  view_757 = None
	        _assert_tensor_metadata_438 = torch.ops.aten._assert_tensor_metadata.default(view_759, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_438 = None
	        convert_element_type_291: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_759, torch.float32);  view_759 = None
	        sub_2236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_290, convert_element_type_291);  convert_element_type_290 = convert_element_type_291 = None
	        mul_4734: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2236, view_758);  sub_2236 = view_758 = None
	        view_760: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4734, [sym_size_int, 1500, 1280]);  mul_4734 = None
	        _assert_tensor_metadata_439 = torch.ops.aten._assert_tensor_metadata.default(view_760, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_439 = None
	        view_761: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_762: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_763: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_440 = torch.ops.aten._assert_tensor_metadata.default(view_761, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_440 = None
	        convert_element_type_292: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_761, torch.float32);  view_761 = None
	        _assert_tensor_metadata_441 = torch.ops.aten._assert_tensor_metadata.default(view_763, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_441 = None
	        convert_element_type_293: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_763, torch.float32);  view_763 = None
	        sub_2240: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_292, convert_element_type_293);  convert_element_type_292 = convert_element_type_293 = None
	        mul_4739: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2240, view_762);  sub_2240 = view_762 = None
	        view_764: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4739, [1280, 1280]);  mul_4739 = None
	        _assert_tensor_metadata_442 = torch.ops.aten._assert_tensor_metadata.default(view_764, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_442 = None
	        mul_4744: "Sym(1500*s6)" = sym_size_int * 1500
	        view_765: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_760, [mul_4744, 1280]);  view_760 = mul_4744 = None
	        permute_81: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_764, [1, 0]);  view_764 = None
	        addmm_40: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_8_self_attn_q_proj_bias, view_765, permute_81);  model_audio_tower_layers_8_self_attn_q_proj_bias = view_765 = permute_81 = None
	        view_766: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_40, [sym_size_int, 1500, 1280]);  addmm_40 = None
	        mul_4751: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_766, 0.125);  view_766 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_767: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_4751, [sym_size_int, 1500, 20, 64]);  mul_4751 = None
	        permute_82: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_767, [0, 2, 1, 3]);  view_767 = None
	        clone_66: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_82, memory_format = torch.contiguous_format);  permute_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_768: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7376, [sym_size_int, 1500, 1280])
	        amin_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_768, [2])
	        amax_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_768, [2]);  view_768 = None
	        full_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_49, full_98);  amin_49 = full_98 = None
	        full_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_49, full_99);  amax_49 = full_99 = None
	        sub_2255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_49, minimum_49);  maximum_49 = None
	        div_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2255, 255.0);  sub_2255 = None
	        clamp_min_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_98, 1.1920928955078125e-07);  div_98 = None
	        div_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_49, clamp_min_147);  minimum_49 = None
	        round_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_99);  div_99 = None
	        sub_2261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_99);  round_99 = None
	        clamp_min_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2261, -128);  sub_2261 = None
	        clamp_max_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_148, 127);  clamp_min_148 = None
	        _assert_tensor_metadata_443 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_147, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_443 = None
	        _assert_tensor_metadata_444 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_98, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_444 = None
	        convert_element_type_294: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_98, torch.int8);  clamp_max_98 = None
	        view_769: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7376, [sym_size_int, 1500, 1280])
	        view_770: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_147, [sym_size_int, 1500, 1])
	        view_771: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_294, [sym_size_int, 1500, 1])
	        reciprocal_49: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_770);  view_770 = None
	        mul_4805: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_49, 1.0);  reciprocal_49 = None
	        mul_4808: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_769, mul_4805);  view_769 = mul_4805 = None
	        round_100: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4808);  mul_4808 = None
	        add_7615: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_100, view_771);  round_100 = view_771 = None
	        clamp_min_149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7615, -128);  add_7615 = None
	        clamp_max_99: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_149, 127);  clamp_min_149 = None
	        view_772: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_99, [sym_size_int, 1500, 1280]);  clamp_max_99 = None
	        _assert_tensor_metadata_445 = torch.ops.aten._assert_tensor_metadata.default(view_772, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_445 = None
	        convert_element_type_295: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_772, torch.int8);  view_772 = None
	        view_773: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_295, [sym_size_int, 1500, 1280]);  convert_element_type_295 = None
	        view_774: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_147, [sym_size_int, 1500, 1]);  clamp_min_147 = None
	        view_775: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_294, [sym_size_int, 1500, 1]);  convert_element_type_294 = None
	        _assert_tensor_metadata_446 = torch.ops.aten._assert_tensor_metadata.default(view_773, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_446 = None
	        convert_element_type_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_773, torch.float32);  view_773 = None
	        _assert_tensor_metadata_447 = torch.ops.aten._assert_tensor_metadata.default(view_775, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_447 = None
	        convert_element_type_297: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_775, torch.float32);  view_775 = None
	        sub_2281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_296, convert_element_type_297);  convert_element_type_296 = convert_element_type_297 = None
	        mul_4830: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2281, view_774);  sub_2281 = view_774 = None
	        view_776: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4830, [sym_size_int, 1500, 1280]);  mul_4830 = None
	        _assert_tensor_metadata_448 = torch.ops.aten._assert_tensor_metadata.default(view_776, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_448 = None
	        view_777: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_778: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_779: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_449 = torch.ops.aten._assert_tensor_metadata.default(view_777, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_449 = None
	        convert_element_type_298: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_777, torch.float32);  view_777 = None
	        _assert_tensor_metadata_450 = torch.ops.aten._assert_tensor_metadata.default(view_779, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_450 = None
	        convert_element_type_299: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_779, torch.float32);  view_779 = None
	        sub_2285: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_298, convert_element_type_299);  convert_element_type_298 = convert_element_type_299 = None
	        mul_4835: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2285, view_778);  sub_2285 = view_778 = None
	        view_780: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4835, [1280, 1280]);  mul_4835 = None
	        _assert_tensor_metadata_451 = torch.ops.aten._assert_tensor_metadata.default(view_780, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_451 = None
	        permute_83: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_780, [1, 0]);  view_780 = None
	        mul_4838: "Sym(1500*s6)" = sym_size_int * 1500
	        view_781: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_776, [mul_4838, 1280]);  view_776 = mul_4838 = None
	        mm_8: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_781, permute_83);  view_781 = permute_83 = None
	        view_782: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_8, [sym_size_int, 1500, 1280]);  mm_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_783: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_782, [sym_size_int, -1, 20, 64]);  view_782 = None
	        permute_84: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_783, [0, 2, 1, 3]);  view_783 = None
	        clone_67: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_84, memory_format = torch.contiguous_format);  permute_84 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_784: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7376, [sym_size_int, 1500, 1280])
	        amin_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_784, [2])
	        amax_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_784, [2]);  view_784 = None
	        full_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_50, full_100);  amin_50 = full_100 = None
	        full_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_50, full_101);  amax_50 = full_101 = None
	        sub_2299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_50, minimum_50);  maximum_50 = None
	        div_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2299, 255.0);  sub_2299 = None
	        clamp_min_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_100, 1.1920928955078125e-07);  div_100 = None
	        div_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_50, clamp_min_150);  minimum_50 = None
	        round_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_101);  div_101 = None
	        sub_2305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_101);  round_101 = None
	        clamp_min_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2305, -128);  sub_2305 = None
	        clamp_max_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_151, 127);  clamp_min_151 = None
	        _assert_tensor_metadata_452 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_150, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_452 = None
	        _assert_tensor_metadata_453 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_100, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_453 = None
	        convert_element_type_300: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_100, torch.int8);  clamp_max_100 = None
	        view_785: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7376, [sym_size_int, 1500, 1280]);  add_7376 = None
	        view_786: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_150, [sym_size_int, 1500, 1])
	        view_787: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_300, [sym_size_int, 1500, 1])
	        reciprocal_50: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_786);  view_786 = None
	        mul_4904: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_50, 1.0);  reciprocal_50 = None
	        mul_4907: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_785, mul_4904);  view_785 = mul_4904 = None
	        round_102: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4907);  mul_4907 = None
	        add_7763: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_102, view_787);  round_102 = view_787 = None
	        clamp_min_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7763, -128);  add_7763 = None
	        clamp_max_101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_152, 127);  clamp_min_152 = None
	        view_788: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_101, [sym_size_int, 1500, 1280]);  clamp_max_101 = None
	        _assert_tensor_metadata_454 = torch.ops.aten._assert_tensor_metadata.default(view_788, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_454 = None
	        convert_element_type_301: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_788, torch.int8);  view_788 = None
	        view_789: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_301, [sym_size_int, 1500, 1280]);  convert_element_type_301 = None
	        view_790: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_150, [sym_size_int, 1500, 1]);  clamp_min_150 = None
	        view_791: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_300, [sym_size_int, 1500, 1]);  convert_element_type_300 = None
	        _assert_tensor_metadata_455 = torch.ops.aten._assert_tensor_metadata.default(view_789, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_455 = None
	        convert_element_type_302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_789, torch.float32);  view_789 = None
	        _assert_tensor_metadata_456 = torch.ops.aten._assert_tensor_metadata.default(view_791, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_456 = None
	        convert_element_type_303: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_791, torch.float32);  view_791 = None
	        sub_2325: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_302, convert_element_type_303);  convert_element_type_302 = convert_element_type_303 = None
	        mul_4929: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2325, view_790);  sub_2325 = view_790 = None
	        view_792: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4929, [sym_size_int, 1500, 1280]);  mul_4929 = None
	        _assert_tensor_metadata_457 = torch.ops.aten._assert_tensor_metadata.default(view_792, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_457 = None
	        view_793: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_794: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_795: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_458 = torch.ops.aten._assert_tensor_metadata.default(view_793, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_458 = None
	        convert_element_type_304: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_793, torch.float32);  view_793 = None
	        _assert_tensor_metadata_459 = torch.ops.aten._assert_tensor_metadata.default(view_795, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_459 = None
	        convert_element_type_305: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_795, torch.float32);  view_795 = None
	        sub_2329: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_304, convert_element_type_305);  convert_element_type_304 = convert_element_type_305 = None
	        mul_4934: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2329, view_794);  sub_2329 = view_794 = None
	        view_796: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4934, [1280, 1280]);  mul_4934 = None
	        _assert_tensor_metadata_460 = torch.ops.aten._assert_tensor_metadata.default(view_796, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_460 = None
	        mul_4939: "Sym(1500*s6)" = sym_size_int * 1500
	        view_797: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_792, [mul_4939, 1280]);  view_792 = mul_4939 = None
	        permute_85: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_796, [1, 0]);  view_796 = None
	        addmm_41: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_8_self_attn_v_proj_bias, view_797, permute_85);  model_audio_tower_layers_8_self_attn_v_proj_bias = view_797 = permute_85 = None
	        view_798: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_41, [sym_size_int, 1500, 1280]);  addmm_41 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_799: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_798, [sym_size_int, -1, 20, 64]);  view_798 = None
	        permute_86: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_799, [0, 2, 1, 3]);  view_799 = None
	        clone_68: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_86, memory_format = torch.contiguous_format);  permute_86 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_8 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_66, clone_67, clone_68, None, False, scale = 1.0);  clone_66 = clone_67 = clone_68 = None
	        getitem_66: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_8[0];  _scaled_dot_product_efficient_attention_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_87: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_66, [0, 2, 1, 3]);  getitem_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_800: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_87, [sym_size_int, 1500, -1]);  permute_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_801: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_800, [sym_size_int, 1500, 1280])
	        amin_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_801, [2])
	        amax_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_801, [2]);  view_801 = None
	        full_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_51, full_102);  amin_51 = full_102 = None
	        full_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_51, full_103);  amax_51 = full_103 = None
	        sub_2347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_51, minimum_51);  maximum_51 = None
	        div_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2347, 255.0);  sub_2347 = None
	        clamp_min_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_102, 1.1920928955078125e-07);  div_102 = None
	        div_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_51, clamp_min_153);  minimum_51 = None
	        round_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_103);  div_103 = None
	        sub_2353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_103);  round_103 = None
	        clamp_min_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2353, -128);  sub_2353 = None
	        clamp_max_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_154, 127);  clamp_min_154 = None
	        _assert_tensor_metadata_461 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_153, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_461 = None
	        _assert_tensor_metadata_462 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_102, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_462 = None
	        convert_element_type_306: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_102, torch.int8);  clamp_max_102 = None
	        view_802: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_800, [sym_size_int, 1500, 1280]);  view_800 = None
	        view_803: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_153, [sym_size_int, 1500, 1])
	        view_804: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_306, [sym_size_int, 1500, 1])
	        reciprocal_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_803);  view_803 = None
	        mul_5009: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_51, 1.0);  reciprocal_51 = None
	        mul_5012: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_802, mul_5009);  view_802 = mul_5009 = None
	        round_104: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5012);  mul_5012 = None
	        add_7927: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_104, view_804);  round_104 = view_804 = None
	        clamp_min_155: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7927, -128);  add_7927 = None
	        clamp_max_103: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_155, 127);  clamp_min_155 = None
	        view_805: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_103, [sym_size_int, 1500, 1280]);  clamp_max_103 = None
	        _assert_tensor_metadata_463 = torch.ops.aten._assert_tensor_metadata.default(view_805, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_463 = None
	        convert_element_type_307: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_805, torch.int8);  view_805 = None
	        view_806: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_307, [sym_size_int, 1500, 1280]);  convert_element_type_307 = None
	        view_807: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_153, [sym_size_int, 1500, 1]);  clamp_min_153 = None
	        view_808: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_306, [sym_size_int, 1500, 1]);  convert_element_type_306 = None
	        _assert_tensor_metadata_464 = torch.ops.aten._assert_tensor_metadata.default(view_806, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_464 = None
	        convert_element_type_308: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_806, torch.float32);  view_806 = None
	        _assert_tensor_metadata_465 = torch.ops.aten._assert_tensor_metadata.default(view_808, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_465 = None
	        convert_element_type_309: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_808, torch.float32);  view_808 = None
	        sub_2373: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_308, convert_element_type_309);  convert_element_type_308 = convert_element_type_309 = None
	        mul_5034: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2373, view_807);  sub_2373 = view_807 = None
	        view_809: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5034, [sym_size_int, 1500, 1280]);  mul_5034 = None
	        _assert_tensor_metadata_466 = torch.ops.aten._assert_tensor_metadata.default(view_809, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_466 = None
	        view_810: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_811: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_812: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_467 = torch.ops.aten._assert_tensor_metadata.default(view_810, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_467 = None
	        convert_element_type_310: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_810, torch.float32);  view_810 = None
	        _assert_tensor_metadata_468 = torch.ops.aten._assert_tensor_metadata.default(view_812, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_468 = None
	        convert_element_type_311: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_812, torch.float32);  view_812 = None
	        sub_2377: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_310, convert_element_type_311);  convert_element_type_310 = convert_element_type_311 = None
	        mul_5039: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2377, view_811);  sub_2377 = view_811 = None
	        view_813: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5039, [1280, 1280]);  mul_5039 = None
	        _assert_tensor_metadata_469 = torch.ops.aten._assert_tensor_metadata.default(view_813, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_469 = None
	        mul_5044: "Sym(1500*s6)" = sym_size_int * 1500
	        view_814: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_809, [mul_5044, 1280]);  view_809 = mul_5044 = None
	        permute_88: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_813, [1, 0]);  view_813 = None
	        addmm_42: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_8_self_attn_out_proj_bias, view_814, permute_88);  model_audio_tower_layers_8_self_attn_out_proj_bias = view_814 = permute_88 = None
	        view_815: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_42, [sym_size_int, 1500, 1280]);  addmm_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_69: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_815);  view_815 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_7990: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_7370, clone_69);  add_7370 = clone_69 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_70: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_7990, memory_format = torch.contiguous_format)
	        var_mean_17 = torch.ops.aten.var_mean.correction(clone_70, [2], correction = 0, keepdim = True)
	        getitem_70: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_17[0]
	        getitem_71: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_17[1];  var_mean_17 = None
	        add_7995: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_70, 1e-05);  getitem_70 = None
	        rsqrt_17: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_7995);  add_7995 = None
	        sub_2383: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_70, getitem_71);  clone_70 = getitem_71 = None
	        mul_5055: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2383, rsqrt_17);  sub_2383 = rsqrt_17 = None
	        mul_5056: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5055, model_audio_tower_layers_8_final_layer_norm_weight);  mul_5055 = model_audio_tower_layers_8_final_layer_norm_weight = None
	        add_7996: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5056, model_audio_tower_layers_8_final_layer_norm_bias);  mul_5056 = model_audio_tower_layers_8_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_816: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7996, [sym_size_int, 1500, 1280])
	        amin_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_816, [2])
	        amax_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_816, [2]);  view_816 = None
	        full_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_52, full_104);  amin_52 = full_104 = None
	        full_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_52, full_105);  amax_52 = full_105 = None
	        sub_2394: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_52, minimum_52);  maximum_52 = None
	        div_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2394, 255.0);  sub_2394 = None
	        clamp_min_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_104, 1.1920928955078125e-07);  div_104 = None
	        div_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_52, clamp_min_156);  minimum_52 = None
	        round_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_105);  div_105 = None
	        sub_2400: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_105);  round_105 = None
	        clamp_min_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2400, -128);  sub_2400 = None
	        clamp_max_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_157, 127);  clamp_min_157 = None
	        _assert_tensor_metadata_470 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_156, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_470 = None
	        _assert_tensor_metadata_471 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_104, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_471 = None
	        convert_element_type_312: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_104, torch.int8);  clamp_max_104 = None
	        view_817: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_7996, [sym_size_int, 1500, 1280]);  add_7996 = None
	        view_818: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_156, [sym_size_int, 1500, 1])
	        view_819: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_312, [sym_size_int, 1500, 1])
	        reciprocal_52: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_818);  view_818 = None
	        mul_5104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_52, 1.0);  reciprocal_52 = None
	        mul_5107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_817, mul_5104);  view_817 = mul_5104 = None
	        round_106: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5107);  mul_5107 = None
	        add_8083: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_106, view_819);  round_106 = view_819 = None
	        clamp_min_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8083, -128);  add_8083 = None
	        clamp_max_105: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_158, 127);  clamp_min_158 = None
	        view_820: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_105, [sym_size_int, 1500, 1280]);  clamp_max_105 = None
	        _assert_tensor_metadata_472 = torch.ops.aten._assert_tensor_metadata.default(view_820, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_472 = None
	        convert_element_type_313: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_820, torch.int8);  view_820 = None
	        view_821: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_313, [sym_size_int, 1500, 1280]);  convert_element_type_313 = None
	        view_822: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_156, [sym_size_int, 1500, 1]);  clamp_min_156 = None
	        view_823: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_312, [sym_size_int, 1500, 1]);  convert_element_type_312 = None
	        _assert_tensor_metadata_473 = torch.ops.aten._assert_tensor_metadata.default(view_821, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_473 = None
	        convert_element_type_314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_821, torch.float32);  view_821 = None
	        _assert_tensor_metadata_474 = torch.ops.aten._assert_tensor_metadata.default(view_823, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_474 = None
	        convert_element_type_315: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_823, torch.float32);  view_823 = None
	        sub_2420: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_314, convert_element_type_315);  convert_element_type_314 = convert_element_type_315 = None
	        mul_5129: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2420, view_822);  sub_2420 = view_822 = None
	        view_824: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5129, [sym_size_int, 1500, 1280]);  mul_5129 = None
	        _assert_tensor_metadata_475 = torch.ops.aten._assert_tensor_metadata.default(view_824, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_475 = None
	        view_825: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_8_fc1_parametrizations_weight_original0 = None
	        view_826: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_8_fc1_parametrizations_weight_original1 = None
	        view_827: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_8_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_476 = torch.ops.aten._assert_tensor_metadata.default(view_825, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_476 = None
	        convert_element_type_316: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_825, torch.float32);  view_825 = None
	        _assert_tensor_metadata_477 = torch.ops.aten._assert_tensor_metadata.default(view_827, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_477 = None
	        convert_element_type_317: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_827, torch.float32);  view_827 = None
	        sub_2424: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_316, convert_element_type_317);  convert_element_type_316 = convert_element_type_317 = None
	        mul_5134: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2424, view_826);  sub_2424 = view_826 = None
	        view_828: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5134, [5120, 1280]);  mul_5134 = None
	        _assert_tensor_metadata_478 = torch.ops.aten._assert_tensor_metadata.default(view_828, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_478 = None
	        mul_5139: "Sym(1500*s6)" = sym_size_int * 1500
	        view_829: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_824, [mul_5139, 1280]);  view_824 = mul_5139 = None
	        permute_89: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_828, [1, 0]);  view_828 = None
	        addmm_43: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_8_fc1_bias, view_829, permute_89);  model_audio_tower_layers_8_fc1_bias = view_829 = permute_89 = None
	        view_830: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_43, [sym_size_int, 1500, 5120]);  addmm_43 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_5146: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_830, 0.5)
	        mul_5147: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_830, 0.7071067811865476);  view_830 = None
	        erf_10: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_5147);  mul_5147 = None
	        add_8142: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_10, 1);  erf_10 = None
	        mul_5148: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5146, add_8142);  mul_5146 = add_8142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_71: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_5148);  mul_5148 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_831: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_71, [sym_size_int, 1500, 5120])
	        amin_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_831, [2])
	        amax_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_831, [2]);  view_831 = None
	        full_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_53, full_106);  amin_53 = full_106 = None
	        full_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_53, full_107);  amax_53 = full_107 = None
	        sub_2437: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_53, minimum_53);  maximum_53 = None
	        div_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2437, 255.0);  sub_2437 = None
	        clamp_min_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_106, 1.1920928955078125e-07);  div_106 = None
	        div_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_53, clamp_min_159);  minimum_53 = None
	        round_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_107);  div_107 = None
	        sub_2443: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_107);  round_107 = None
	        clamp_min_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2443, -128);  sub_2443 = None
	        clamp_max_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_160, 127);  clamp_min_160 = None
	        _assert_tensor_metadata_479 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_159, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_479 = None
	        _assert_tensor_metadata_480 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_106, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_480 = None
	        convert_element_type_318: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_106, torch.int8);  clamp_max_106 = None
	        view_832: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_71, [sym_size_int, 1500, 5120]);  clone_71 = None
	        view_833: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_159, [sym_size_int, 1500, 1])
	        view_834: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_318, [sym_size_int, 1500, 1])
	        reciprocal_53: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_833);  view_833 = None
	        mul_5194: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_53, 1.0);  reciprocal_53 = None
	        mul_5197: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_832, mul_5194);  view_832 = mul_5194 = None
	        round_108: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_5197);  mul_5197 = None
	        add_8225: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_108, view_834);  round_108 = view_834 = None
	        clamp_min_161: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8225, -128);  add_8225 = None
	        clamp_max_107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_161, 127);  clamp_min_161 = None
	        view_835: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_107, [sym_size_int, 1500, 5120]);  clamp_max_107 = None
	        _assert_tensor_metadata_481 = torch.ops.aten._assert_tensor_metadata.default(view_835, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_481 = None
	        convert_element_type_319: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_835, torch.int8);  view_835 = None
	        view_836: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_319, [sym_size_int, 1500, 5120]);  convert_element_type_319 = None
	        view_837: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_159, [sym_size_int, 1500, 1]);  clamp_min_159 = None
	        view_838: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_318, [sym_size_int, 1500, 1]);  convert_element_type_318 = None
	        _assert_tensor_metadata_482 = torch.ops.aten._assert_tensor_metadata.default(view_836, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_482 = None
	        convert_element_type_320: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_836, torch.float32);  view_836 = None
	        _assert_tensor_metadata_483 = torch.ops.aten._assert_tensor_metadata.default(view_838, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_483 = None
	        convert_element_type_321: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_838, torch.float32);  view_838 = None
	        sub_2463: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_320, convert_element_type_321);  convert_element_type_320 = convert_element_type_321 = None
	        mul_5219: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2463, view_837);  sub_2463 = view_837 = None
	        view_839: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_5219, [sym_size_int, 1500, 5120]);  mul_5219 = None
	        _assert_tensor_metadata_484 = torch.ops.aten._assert_tensor_metadata.default(view_839, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_484 = None
	        view_840: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_8_fc2_parametrizations_weight_original0 = None
	        view_841: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_8_fc2_parametrizations_weight_original1 = None
	        view_842: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_8_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_485 = torch.ops.aten._assert_tensor_metadata.default(view_840, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_485 = None
	        convert_element_type_322: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_840, torch.float32);  view_840 = None
	        _assert_tensor_metadata_486 = torch.ops.aten._assert_tensor_metadata.default(view_842, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_486 = None
	        convert_element_type_323: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_842, torch.float32);  view_842 = None
	        sub_2467: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_322, convert_element_type_323);  convert_element_type_322 = convert_element_type_323 = None
	        mul_5224: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2467, view_841);  sub_2467 = view_841 = None
	        view_843: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_5224, [1280, 5120]);  mul_5224 = None
	        _assert_tensor_metadata_487 = torch.ops.aten._assert_tensor_metadata.default(view_843, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_487 = None
	        mul_5229: "Sym(1500*s6)" = sym_size_int * 1500
	        view_844: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_839, [mul_5229, 5120]);  view_839 = mul_5229 = None
	        permute_90: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_843, [1, 0]);  view_843 = None
	        addmm_44: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_8_fc2_bias, view_844, permute_90);  model_audio_tower_layers_8_fc2_bias = view_844 = permute_90 = None
	        view_845: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_44, [sym_size_int, 1500, 1280]);  addmm_44 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_72: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_845);  view_845 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_8288: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_7990, clone_72);  add_7990 = clone_72 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_73: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_8288, memory_format = torch.contiguous_format)
	        var_mean_18 = torch.ops.aten.var_mean.correction(clone_73, [2], correction = 0, keepdim = True)
	        getitem_72: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_18[0]
	        getitem_73: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_18[1];  var_mean_18 = None
	        add_8293: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_72, 1e-05);  getitem_72 = None
	        rsqrt_18: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_8293);  add_8293 = None
	        sub_2473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_73, getitem_73);  clone_73 = getitem_73 = None
	        mul_5240: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2473, rsqrt_18);  sub_2473 = rsqrt_18 = None
	        mul_5241: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5240, model_audio_tower_layers_9_self_attn_layer_norm_weight);  mul_5240 = model_audio_tower_layers_9_self_attn_layer_norm_weight = None
	        add_8294: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5241, model_audio_tower_layers_9_self_attn_layer_norm_bias);  mul_5241 = model_audio_tower_layers_9_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_846: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_8294, [sym_size_int, 1500, 1280])
	        amin_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_846, [2])
	        amax_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_846, [2]);  view_846 = None
	        full_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_54, full_108);  amin_54 = full_108 = None
	        full_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_54, full_109);  amax_54 = full_109 = None
	        sub_2484: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_54, minimum_54);  maximum_54 = None
	        div_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2484, 255.0);  sub_2484 = None
	        clamp_min_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_108, 1.1920928955078125e-07);  div_108 = None
	        div_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_54, clamp_min_162);  minimum_54 = None
	        round_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_109);  div_109 = None
	        sub_2490: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_109);  round_109 = None
	        clamp_min_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2490, -128);  sub_2490 = None
	        clamp_max_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_163, 127);  clamp_min_163 = None
	        _assert_tensor_metadata_488 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_162, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_488 = None
	        _assert_tensor_metadata_489 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_108, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_489 = None
	        convert_element_type_324: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_108, torch.int8);  clamp_max_108 = None
	        view_847: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_8294, [sym_size_int, 1500, 1280])
	        view_848: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_162, [sym_size_int, 1500, 1])
	        view_849: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_324, [sym_size_int, 1500, 1])
	        reciprocal_54: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_848);  view_848 = None
	        mul_5289: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_54, 1.0);  reciprocal_54 = None
	        mul_5292: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_847, mul_5289);  view_847 = mul_5289 = None
	        round_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5292);  mul_5292 = None
	        add_8381: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_110, view_849);  round_110 = view_849 = None
	        clamp_min_164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8381, -128);  add_8381 = None
	        clamp_max_109: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_164, 127);  clamp_min_164 = None
	        view_850: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_109, [sym_size_int, 1500, 1280]);  clamp_max_109 = None
	        _assert_tensor_metadata_490 = torch.ops.aten._assert_tensor_metadata.default(view_850, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_490 = None
	        convert_element_type_325: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_850, torch.int8);  view_850 = None
	        view_851: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_325, [sym_size_int, 1500, 1280]);  convert_element_type_325 = None
	        view_852: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_162, [sym_size_int, 1500, 1]);  clamp_min_162 = None
	        view_853: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_324, [sym_size_int, 1500, 1]);  convert_element_type_324 = None
	        _assert_tensor_metadata_491 = torch.ops.aten._assert_tensor_metadata.default(view_851, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_491 = None
	        convert_element_type_326: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_851, torch.float32);  view_851 = None
	        _assert_tensor_metadata_492 = torch.ops.aten._assert_tensor_metadata.default(view_853, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_492 = None
	        convert_element_type_327: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_853, torch.float32);  view_853 = None
	        sub_2510: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_326, convert_element_type_327);  convert_element_type_326 = convert_element_type_327 = None
	        mul_5314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2510, view_852);  sub_2510 = view_852 = None
	        view_854: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5314, [sym_size_int, 1500, 1280]);  mul_5314 = None
	        _assert_tensor_metadata_493 = torch.ops.aten._assert_tensor_metadata.default(view_854, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_493 = None
	        view_855: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_856: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_857: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_494 = torch.ops.aten._assert_tensor_metadata.default(view_855, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_494 = None
	        convert_element_type_328: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_855, torch.float32);  view_855 = None
	        _assert_tensor_metadata_495 = torch.ops.aten._assert_tensor_metadata.default(view_857, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_495 = None
	        convert_element_type_329: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_857, torch.float32);  view_857 = None
	        sub_2514: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_328, convert_element_type_329);  convert_element_type_328 = convert_element_type_329 = None
	        mul_5319: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2514, view_856);  sub_2514 = view_856 = None
	        view_858: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5319, [1280, 1280]);  mul_5319 = None
	        _assert_tensor_metadata_496 = torch.ops.aten._assert_tensor_metadata.default(view_858, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_496 = None
	        mul_5324: "Sym(1500*s6)" = sym_size_int * 1500
	        view_859: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_854, [mul_5324, 1280]);  view_854 = mul_5324 = None
	        permute_91: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_858, [1, 0]);  view_858 = None
	        addmm_45: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_9_self_attn_q_proj_bias, view_859, permute_91);  model_audio_tower_layers_9_self_attn_q_proj_bias = view_859 = permute_91 = None
	        view_860: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_45, [sym_size_int, 1500, 1280]);  addmm_45 = None
	        mul_5331: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_860, 0.125);  view_860 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_861: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_5331, [sym_size_int, 1500, 20, 64]);  mul_5331 = None
	        permute_92: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_861, [0, 2, 1, 3]);  view_861 = None
	        clone_74: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_92, memory_format = torch.contiguous_format);  permute_92 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_862: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_8294, [sym_size_int, 1500, 1280])
	        amin_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_862, [2])
	        amax_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_862, [2]);  view_862 = None
	        full_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_55, full_110);  amin_55 = full_110 = None
	        full_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_55, full_111);  amax_55 = full_111 = None
	        sub_2529: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_55, minimum_55);  maximum_55 = None
	        div_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2529, 255.0);  sub_2529 = None
	        clamp_min_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_110, 1.1920928955078125e-07);  div_110 = None
	        div_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_55, clamp_min_165);  minimum_55 = None
	        round_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_111);  div_111 = None
	        sub_2535: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_111);  round_111 = None
	        clamp_min_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2535, -128);  sub_2535 = None
	        clamp_max_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_166, 127);  clamp_min_166 = None
	        _assert_tensor_metadata_497 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_165, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_497 = None
	        _assert_tensor_metadata_498 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_110, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_498 = None
	        convert_element_type_330: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_110, torch.int8);  clamp_max_110 = None
	        view_863: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_8294, [sym_size_int, 1500, 1280])
	        view_864: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_165, [sym_size_int, 1500, 1])
	        view_865: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_330, [sym_size_int, 1500, 1])
	        reciprocal_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_864);  view_864 = None
	        mul_5385: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_55, 1.0);  reciprocal_55 = None
	        mul_5388: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_863, mul_5385);  view_863 = mul_5385 = None
	        round_112: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5388);  mul_5388 = None
	        add_8533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_112, view_865);  round_112 = view_865 = None
	        clamp_min_167: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8533, -128);  add_8533 = None
	        clamp_max_111: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_167, 127);  clamp_min_167 = None
	        view_866: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_111, [sym_size_int, 1500, 1280]);  clamp_max_111 = None
	        _assert_tensor_metadata_499 = torch.ops.aten._assert_tensor_metadata.default(view_866, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_499 = None
	        convert_element_type_331: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_866, torch.int8);  view_866 = None
	        view_867: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_331, [sym_size_int, 1500, 1280]);  convert_element_type_331 = None
	        view_868: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_165, [sym_size_int, 1500, 1]);  clamp_min_165 = None
	        view_869: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_330, [sym_size_int, 1500, 1]);  convert_element_type_330 = None
	        _assert_tensor_metadata_500 = torch.ops.aten._assert_tensor_metadata.default(view_867, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_500 = None
	        convert_element_type_332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_867, torch.float32);  view_867 = None
	        _assert_tensor_metadata_501 = torch.ops.aten._assert_tensor_metadata.default(view_869, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_501 = None
	        convert_element_type_333: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_869, torch.float32);  view_869 = None
	        sub_2555: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_332, convert_element_type_333);  convert_element_type_332 = convert_element_type_333 = None
	        mul_5410: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2555, view_868);  sub_2555 = view_868 = None
	        view_870: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5410, [sym_size_int, 1500, 1280]);  mul_5410 = None
	        _assert_tensor_metadata_502 = torch.ops.aten._assert_tensor_metadata.default(view_870, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_502 = None
	        view_871: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_872: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_873: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_503 = torch.ops.aten._assert_tensor_metadata.default(view_871, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_503 = None
	        convert_element_type_334: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_871, torch.float32);  view_871 = None
	        _assert_tensor_metadata_504 = torch.ops.aten._assert_tensor_metadata.default(view_873, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_504 = None
	        convert_element_type_335: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_873, torch.float32);  view_873 = None
	        sub_2559: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_334, convert_element_type_335);  convert_element_type_334 = convert_element_type_335 = None
	        mul_5415: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2559, view_872);  sub_2559 = view_872 = None
	        view_874: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5415, [1280, 1280]);  mul_5415 = None
	        _assert_tensor_metadata_505 = torch.ops.aten._assert_tensor_metadata.default(view_874, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_505 = None
	        permute_93: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_874, [1, 0]);  view_874 = None
	        mul_5418: "Sym(1500*s6)" = sym_size_int * 1500
	        view_875: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_870, [mul_5418, 1280]);  view_870 = mul_5418 = None
	        mm_9: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_875, permute_93);  view_875 = permute_93 = None
	        view_876: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_9, [sym_size_int, 1500, 1280]);  mm_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_877: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_876, [sym_size_int, -1, 20, 64]);  view_876 = None
	        permute_94: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_877, [0, 2, 1, 3]);  view_877 = None
	        clone_75: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_94, memory_format = torch.contiguous_format);  permute_94 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_878: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_8294, [sym_size_int, 1500, 1280])
	        amin_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_878, [2])
	        amax_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_878, [2]);  view_878 = None
	        full_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_56, full_112);  amin_56 = full_112 = None
	        full_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_56, full_113);  amax_56 = full_113 = None
	        sub_2573: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_56, minimum_56);  maximum_56 = None
	        div_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2573, 255.0);  sub_2573 = None
	        clamp_min_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_112, 1.1920928955078125e-07);  div_112 = None
	        div_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_56, clamp_min_168);  minimum_56 = None
	        round_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_113);  div_113 = None
	        sub_2579: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_113);  round_113 = None
	        clamp_min_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2579, -128);  sub_2579 = None
	        clamp_max_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_169, 127);  clamp_min_169 = None
	        _assert_tensor_metadata_506 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_168, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_506 = None
	        _assert_tensor_metadata_507 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_112, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_507 = None
	        convert_element_type_336: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_112, torch.int8);  clamp_max_112 = None
	        view_879: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_8294, [sym_size_int, 1500, 1280]);  add_8294 = None
	        view_880: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_168, [sym_size_int, 1500, 1])
	        view_881: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_336, [sym_size_int, 1500, 1])
	        reciprocal_56: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_880);  view_880 = None
	        mul_5484: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_56, 1.0);  reciprocal_56 = None
	        mul_5487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_879, mul_5484);  view_879 = mul_5484 = None
	        round_114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5487);  mul_5487 = None
	        add_8681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_114, view_881);  round_114 = view_881 = None
	        clamp_min_170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8681, -128);  add_8681 = None
	        clamp_max_113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_170, 127);  clamp_min_170 = None
	        view_882: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_113, [sym_size_int, 1500, 1280]);  clamp_max_113 = None
	        _assert_tensor_metadata_508 = torch.ops.aten._assert_tensor_metadata.default(view_882, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_508 = None
	        convert_element_type_337: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_882, torch.int8);  view_882 = None
	        view_883: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_337, [sym_size_int, 1500, 1280]);  convert_element_type_337 = None
	        view_884: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_168, [sym_size_int, 1500, 1]);  clamp_min_168 = None
	        view_885: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_336, [sym_size_int, 1500, 1]);  convert_element_type_336 = None
	        _assert_tensor_metadata_509 = torch.ops.aten._assert_tensor_metadata.default(view_883, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_509 = None
	        convert_element_type_338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_883, torch.float32);  view_883 = None
	        _assert_tensor_metadata_510 = torch.ops.aten._assert_tensor_metadata.default(view_885, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_510 = None
	        convert_element_type_339: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_885, torch.float32);  view_885 = None
	        sub_2599: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_338, convert_element_type_339);  convert_element_type_338 = convert_element_type_339 = None
	        mul_5509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2599, view_884);  sub_2599 = view_884 = None
	        view_886: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5509, [sym_size_int, 1500, 1280]);  mul_5509 = None
	        _assert_tensor_metadata_511 = torch.ops.aten._assert_tensor_metadata.default(view_886, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_511 = None
	        view_887: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_888: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_889: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_512 = torch.ops.aten._assert_tensor_metadata.default(view_887, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_512 = None
	        convert_element_type_340: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_887, torch.float32);  view_887 = None
	        _assert_tensor_metadata_513 = torch.ops.aten._assert_tensor_metadata.default(view_889, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_513 = None
	        convert_element_type_341: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_889, torch.float32);  view_889 = None
	        sub_2603: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_340, convert_element_type_341);  convert_element_type_340 = convert_element_type_341 = None
	        mul_5514: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2603, view_888);  sub_2603 = view_888 = None
	        view_890: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5514, [1280, 1280]);  mul_5514 = None
	        _assert_tensor_metadata_514 = torch.ops.aten._assert_tensor_metadata.default(view_890, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_514 = None
	        mul_5519: "Sym(1500*s6)" = sym_size_int * 1500
	        view_891: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_886, [mul_5519, 1280]);  view_886 = mul_5519 = None
	        permute_95: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_890, [1, 0]);  view_890 = None
	        addmm_46: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_9_self_attn_v_proj_bias, view_891, permute_95);  model_audio_tower_layers_9_self_attn_v_proj_bias = view_891 = permute_95 = None
	        view_892: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_46, [sym_size_int, 1500, 1280]);  addmm_46 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_893: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_892, [sym_size_int, -1, 20, 64]);  view_892 = None
	        permute_96: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_893, [0, 2, 1, 3]);  view_893 = None
	        clone_76: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_96, memory_format = torch.contiguous_format);  permute_96 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_9 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_74, clone_75, clone_76, None, False, scale = 1.0);  clone_74 = clone_75 = clone_76 = None
	        getitem_74: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_9[0];  _scaled_dot_product_efficient_attention_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_97: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_74, [0, 2, 1, 3]);  getitem_74 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_97, [sym_size_int, 1500, -1]);  permute_97 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_895: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_894, [sym_size_int, 1500, 1280])
	        amin_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_895, [2])
	        amax_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_895, [2]);  view_895 = None
	        full_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_57, full_114);  amin_57 = full_114 = None
	        full_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_57, full_115);  amax_57 = full_115 = None
	        sub_2621: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_57, minimum_57);  maximum_57 = None
	        div_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2621, 255.0);  sub_2621 = None
	        clamp_min_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_114, 1.1920928955078125e-07);  div_114 = None
	        div_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_57, clamp_min_171);  minimum_57 = None
	        round_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_115);  div_115 = None
	        sub_2627: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_115);  round_115 = None
	        clamp_min_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2627, -128);  sub_2627 = None
	        clamp_max_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_172, 127);  clamp_min_172 = None
	        _assert_tensor_metadata_515 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_171, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_515 = None
	        _assert_tensor_metadata_516 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_114, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_516 = None
	        convert_element_type_342: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_114, torch.int8);  clamp_max_114 = None
	        view_896: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_894, [sym_size_int, 1500, 1280]);  view_894 = None
	        view_897: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_171, [sym_size_int, 1500, 1])
	        view_898: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_342, [sym_size_int, 1500, 1])
	        reciprocal_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_897);  view_897 = None
	        mul_5589: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_57, 1.0);  reciprocal_57 = None
	        mul_5592: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_896, mul_5589);  view_896 = mul_5589 = None
	        round_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5592);  mul_5592 = None
	        add_8845: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_116, view_898);  round_116 = view_898 = None
	        clamp_min_173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8845, -128);  add_8845 = None
	        clamp_max_115: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_173, 127);  clamp_min_173 = None
	        view_899: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_115, [sym_size_int, 1500, 1280]);  clamp_max_115 = None
	        _assert_tensor_metadata_517 = torch.ops.aten._assert_tensor_metadata.default(view_899, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_517 = None
	        convert_element_type_343: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_899, torch.int8);  view_899 = None
	        view_900: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_343, [sym_size_int, 1500, 1280]);  convert_element_type_343 = None
	        view_901: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_171, [sym_size_int, 1500, 1]);  clamp_min_171 = None
	        view_902: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_342, [sym_size_int, 1500, 1]);  convert_element_type_342 = None
	        _assert_tensor_metadata_518 = torch.ops.aten._assert_tensor_metadata.default(view_900, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_518 = None
	        convert_element_type_344: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_900, torch.float32);  view_900 = None
	        _assert_tensor_metadata_519 = torch.ops.aten._assert_tensor_metadata.default(view_902, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_519 = None
	        convert_element_type_345: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_902, torch.float32);  view_902 = None
	        sub_2647: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_344, convert_element_type_345);  convert_element_type_344 = convert_element_type_345 = None
	        mul_5614: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2647, view_901);  sub_2647 = view_901 = None
	        view_903: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5614, [sym_size_int, 1500, 1280]);  mul_5614 = None
	        _assert_tensor_metadata_520 = torch.ops.aten._assert_tensor_metadata.default(view_903, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_520 = None
	        view_904: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_905: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_906: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_521 = torch.ops.aten._assert_tensor_metadata.default(view_904, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_521 = None
	        convert_element_type_346: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_904, torch.float32);  view_904 = None
	        _assert_tensor_metadata_522 = torch.ops.aten._assert_tensor_metadata.default(view_906, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_522 = None
	        convert_element_type_347: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_906, torch.float32);  view_906 = None
	        sub_2651: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_346, convert_element_type_347);  convert_element_type_346 = convert_element_type_347 = None
	        mul_5619: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2651, view_905);  sub_2651 = view_905 = None
	        view_907: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5619, [1280, 1280]);  mul_5619 = None
	        _assert_tensor_metadata_523 = torch.ops.aten._assert_tensor_metadata.default(view_907, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_523 = None
	        mul_5624: "Sym(1500*s6)" = sym_size_int * 1500
	        view_908: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_903, [mul_5624, 1280]);  view_903 = mul_5624 = None
	        permute_98: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_907, [1, 0]);  view_907 = None
	        addmm_47: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_9_self_attn_out_proj_bias, view_908, permute_98);  model_audio_tower_layers_9_self_attn_out_proj_bias = view_908 = permute_98 = None
	        view_909: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_47, [sym_size_int, 1500, 1280]);  addmm_47 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_77: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_909);  view_909 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_8908: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_8288, clone_77);  add_8288 = clone_77 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_78: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_8908, memory_format = torch.contiguous_format)
	        var_mean_19 = torch.ops.aten.var_mean.correction(clone_78, [2], correction = 0, keepdim = True)
	        getitem_78: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_19[0]
	        getitem_79: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_19[1];  var_mean_19 = None
	        add_8913: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_78, 1e-05);  getitem_78 = None
	        rsqrt_19: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_8913);  add_8913 = None
	        sub_2657: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_78, getitem_79);  clone_78 = getitem_79 = None
	        mul_5635: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2657, rsqrt_19);  sub_2657 = rsqrt_19 = None
	        mul_5636: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5635, model_audio_tower_layers_9_final_layer_norm_weight);  mul_5635 = model_audio_tower_layers_9_final_layer_norm_weight = None
	        add_8914: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5636, model_audio_tower_layers_9_final_layer_norm_bias);  mul_5636 = model_audio_tower_layers_9_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_910: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_8914, [sym_size_int, 1500, 1280])
	        amin_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_910, [2])
	        amax_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_910, [2]);  view_910 = None
	        full_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_58, full_116);  amin_58 = full_116 = None
	        full_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_58, full_117);  amax_58 = full_117 = None
	        sub_2668: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_58, minimum_58);  maximum_58 = None
	        div_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2668, 255.0);  sub_2668 = None
	        clamp_min_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_116, 1.1920928955078125e-07);  div_116 = None
	        div_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_58, clamp_min_174);  minimum_58 = None
	        round_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_117);  div_117 = None
	        sub_2674: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_117);  round_117 = None
	        clamp_min_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2674, -128);  sub_2674 = None
	        clamp_max_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_175, 127);  clamp_min_175 = None
	        _assert_tensor_metadata_524 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_174, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_524 = None
	        _assert_tensor_metadata_525 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_116, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_525 = None
	        convert_element_type_348: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_116, torch.int8);  clamp_max_116 = None
	        view_911: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_8914, [sym_size_int, 1500, 1280]);  add_8914 = None
	        view_912: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_174, [sym_size_int, 1500, 1])
	        view_913: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_348, [sym_size_int, 1500, 1])
	        reciprocal_58: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_912);  view_912 = None
	        mul_5684: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_58, 1.0);  reciprocal_58 = None
	        mul_5687: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_911, mul_5684);  view_911 = mul_5684 = None
	        round_118: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5687);  mul_5687 = None
	        add_9001: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_118, view_913);  round_118 = view_913 = None
	        clamp_min_176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9001, -128);  add_9001 = None
	        clamp_max_117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_176, 127);  clamp_min_176 = None
	        view_914: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_117, [sym_size_int, 1500, 1280]);  clamp_max_117 = None
	        _assert_tensor_metadata_526 = torch.ops.aten._assert_tensor_metadata.default(view_914, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_526 = None
	        convert_element_type_349: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_914, torch.int8);  view_914 = None
	        view_915: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_349, [sym_size_int, 1500, 1280]);  convert_element_type_349 = None
	        view_916: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_174, [sym_size_int, 1500, 1]);  clamp_min_174 = None
	        view_917: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_348, [sym_size_int, 1500, 1]);  convert_element_type_348 = None
	        _assert_tensor_metadata_527 = torch.ops.aten._assert_tensor_metadata.default(view_915, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_527 = None
	        convert_element_type_350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_915, torch.float32);  view_915 = None
	        _assert_tensor_metadata_528 = torch.ops.aten._assert_tensor_metadata.default(view_917, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_528 = None
	        convert_element_type_351: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_917, torch.float32);  view_917 = None
	        sub_2694: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_350, convert_element_type_351);  convert_element_type_350 = convert_element_type_351 = None
	        mul_5709: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2694, view_916);  sub_2694 = view_916 = None
	        view_918: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5709, [sym_size_int, 1500, 1280]);  mul_5709 = None
	        _assert_tensor_metadata_529 = torch.ops.aten._assert_tensor_metadata.default(view_918, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_529 = None
	        view_919: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_9_fc1_parametrizations_weight_original0 = None
	        view_920: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_9_fc1_parametrizations_weight_original1 = None
	        view_921: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_9_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_530 = torch.ops.aten._assert_tensor_metadata.default(view_919, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_530 = None
	        convert_element_type_352: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_919, torch.float32);  view_919 = None
	        _assert_tensor_metadata_531 = torch.ops.aten._assert_tensor_metadata.default(view_921, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_531 = None
	        convert_element_type_353: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_921, torch.float32);  view_921 = None
	        sub_2698: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_352, convert_element_type_353);  convert_element_type_352 = convert_element_type_353 = None
	        mul_5714: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2698, view_920);  sub_2698 = view_920 = None
	        view_922: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5714, [5120, 1280]);  mul_5714 = None
	        _assert_tensor_metadata_532 = torch.ops.aten._assert_tensor_metadata.default(view_922, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_532 = None
	        mul_5719: "Sym(1500*s6)" = sym_size_int * 1500
	        view_923: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_918, [mul_5719, 1280]);  view_918 = mul_5719 = None
	        permute_99: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_922, [1, 0]);  view_922 = None
	        addmm_48: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_9_fc1_bias, view_923, permute_99);  model_audio_tower_layers_9_fc1_bias = view_923 = permute_99 = None
	        view_924: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_48, [sym_size_int, 1500, 5120]);  addmm_48 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_5726: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_924, 0.5)
	        mul_5727: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_924, 0.7071067811865476);  view_924 = None
	        erf_11: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_5727);  mul_5727 = None
	        add_9060: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_11, 1);  erf_11 = None
	        mul_5728: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5726, add_9060);  mul_5726 = add_9060 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_79: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_5728);  mul_5728 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_925: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_79, [sym_size_int, 1500, 5120])
	        amin_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_925, [2])
	        amax_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_925, [2]);  view_925 = None
	        full_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_59, full_118);  amin_59 = full_118 = None
	        full_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_59, full_119);  amax_59 = full_119 = None
	        sub_2711: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_59, minimum_59);  maximum_59 = None
	        div_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2711, 255.0);  sub_2711 = None
	        clamp_min_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_118, 1.1920928955078125e-07);  div_118 = None
	        div_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_59, clamp_min_177);  minimum_59 = None
	        round_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_119);  div_119 = None
	        sub_2717: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_119);  round_119 = None
	        clamp_min_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2717, -128);  sub_2717 = None
	        clamp_max_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_178, 127);  clamp_min_178 = None
	        _assert_tensor_metadata_533 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_177, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_533 = None
	        _assert_tensor_metadata_534 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_118, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_534 = None
	        convert_element_type_354: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_118, torch.int8);  clamp_max_118 = None
	        view_926: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_79, [sym_size_int, 1500, 5120]);  clone_79 = None
	        view_927: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_177, [sym_size_int, 1500, 1])
	        view_928: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_354, [sym_size_int, 1500, 1])
	        reciprocal_59: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_927);  view_927 = None
	        mul_5774: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_59, 1.0);  reciprocal_59 = None
	        mul_5777: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_926, mul_5774);  view_926 = mul_5774 = None
	        round_120: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_5777);  mul_5777 = None
	        add_9143: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_120, view_928);  round_120 = view_928 = None
	        clamp_min_179: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9143, -128);  add_9143 = None
	        clamp_max_119: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_179, 127);  clamp_min_179 = None
	        view_929: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_119, [sym_size_int, 1500, 5120]);  clamp_max_119 = None
	        _assert_tensor_metadata_535 = torch.ops.aten._assert_tensor_metadata.default(view_929, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_535 = None
	        convert_element_type_355: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_929, torch.int8);  view_929 = None
	        view_930: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_355, [sym_size_int, 1500, 5120]);  convert_element_type_355 = None
	        view_931: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_177, [sym_size_int, 1500, 1]);  clamp_min_177 = None
	        view_932: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_354, [sym_size_int, 1500, 1]);  convert_element_type_354 = None
	        _assert_tensor_metadata_536 = torch.ops.aten._assert_tensor_metadata.default(view_930, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_536 = None
	        convert_element_type_356: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_930, torch.float32);  view_930 = None
	        _assert_tensor_metadata_537 = torch.ops.aten._assert_tensor_metadata.default(view_932, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_537 = None
	        convert_element_type_357: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_932, torch.float32);  view_932 = None
	        sub_2737: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_356, convert_element_type_357);  convert_element_type_356 = convert_element_type_357 = None
	        mul_5799: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2737, view_931);  sub_2737 = view_931 = None
	        view_933: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_5799, [sym_size_int, 1500, 5120]);  mul_5799 = None
	        _assert_tensor_metadata_538 = torch.ops.aten._assert_tensor_metadata.default(view_933, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_538 = None
	        view_934: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_9_fc2_parametrizations_weight_original0 = None
	        view_935: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_9_fc2_parametrizations_weight_original1 = None
	        view_936: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_9_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_539 = torch.ops.aten._assert_tensor_metadata.default(view_934, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_539 = None
	        convert_element_type_358: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_934, torch.float32);  view_934 = None
	        _assert_tensor_metadata_540 = torch.ops.aten._assert_tensor_metadata.default(view_936, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_540 = None
	        convert_element_type_359: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_936, torch.float32);  view_936 = None
	        sub_2741: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_358, convert_element_type_359);  convert_element_type_358 = convert_element_type_359 = None
	        mul_5804: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2741, view_935);  sub_2741 = view_935 = None
	        view_937: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_5804, [1280, 5120]);  mul_5804 = None
	        _assert_tensor_metadata_541 = torch.ops.aten._assert_tensor_metadata.default(view_937, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_541 = None
	        mul_5809: "Sym(1500*s6)" = sym_size_int * 1500
	        view_938: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_933, [mul_5809, 5120]);  view_933 = mul_5809 = None
	        permute_100: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_937, [1, 0]);  view_937 = None
	        addmm_49: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_9_fc2_bias, view_938, permute_100);  model_audio_tower_layers_9_fc2_bias = view_938 = permute_100 = None
	        view_939: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_49, [sym_size_int, 1500, 1280]);  addmm_49 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_939);  view_939 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_9206: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_8908, clone_80);  add_8908 = clone_80 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_81: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_9206, memory_format = torch.contiguous_format)
	        var_mean_20 = torch.ops.aten.var_mean.correction(clone_81, [2], correction = 0, keepdim = True)
	        getitem_80: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_20[0]
	        getitem_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_20[1];  var_mean_20 = None
	        add_9211: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_80, 1e-05);  getitem_80 = None
	        rsqrt_20: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_9211);  add_9211 = None
	        sub_2747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_81, getitem_81);  clone_81 = getitem_81 = None
	        mul_5820: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2747, rsqrt_20);  sub_2747 = rsqrt_20 = None
	        mul_5821: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5820, model_audio_tower_layers_10_self_attn_layer_norm_weight);  mul_5820 = model_audio_tower_layers_10_self_attn_layer_norm_weight = None
	        add_9212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5821, model_audio_tower_layers_10_self_attn_layer_norm_bias);  mul_5821 = model_audio_tower_layers_10_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_940: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_9212, [sym_size_int, 1500, 1280])
	        amin_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_940, [2])
	        amax_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_940, [2]);  view_940 = None
	        full_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_60, full_120);  amin_60 = full_120 = None
	        full_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_60, full_121);  amax_60 = full_121 = None
	        sub_2758: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_60, minimum_60);  maximum_60 = None
	        div_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2758, 255.0);  sub_2758 = None
	        clamp_min_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_120, 1.1920928955078125e-07);  div_120 = None
	        div_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_60, clamp_min_180);  minimum_60 = None
	        round_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_121);  div_121 = None
	        sub_2764: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_121);  round_121 = None
	        clamp_min_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2764, -128);  sub_2764 = None
	        clamp_max_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_181, 127);  clamp_min_181 = None
	        _assert_tensor_metadata_542 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_180, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_542 = None
	        _assert_tensor_metadata_543 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_120, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_543 = None
	        convert_element_type_360: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_120, torch.int8);  clamp_max_120 = None
	        view_941: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_9212, [sym_size_int, 1500, 1280])
	        view_942: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_180, [sym_size_int, 1500, 1])
	        view_943: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_360, [sym_size_int, 1500, 1])
	        reciprocal_60: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_942);  view_942 = None
	        mul_5869: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_60, 1.0);  reciprocal_60 = None
	        mul_5872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_941, mul_5869);  view_941 = mul_5869 = None
	        round_122: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5872);  mul_5872 = None
	        add_9299: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_122, view_943);  round_122 = view_943 = None
	        clamp_min_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9299, -128);  add_9299 = None
	        clamp_max_121: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_182, 127);  clamp_min_182 = None
	        view_944: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_121, [sym_size_int, 1500, 1280]);  clamp_max_121 = None
	        _assert_tensor_metadata_544 = torch.ops.aten._assert_tensor_metadata.default(view_944, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_544 = None
	        convert_element_type_361: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_944, torch.int8);  view_944 = None
	        view_945: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_361, [sym_size_int, 1500, 1280]);  convert_element_type_361 = None
	        view_946: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_180, [sym_size_int, 1500, 1]);  clamp_min_180 = None
	        view_947: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_360, [sym_size_int, 1500, 1]);  convert_element_type_360 = None
	        _assert_tensor_metadata_545 = torch.ops.aten._assert_tensor_metadata.default(view_945, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_545 = None
	        convert_element_type_362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_945, torch.float32);  view_945 = None
	        _assert_tensor_metadata_546 = torch.ops.aten._assert_tensor_metadata.default(view_947, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_546 = None
	        convert_element_type_363: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_947, torch.float32);  view_947 = None
	        sub_2784: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_362, convert_element_type_363);  convert_element_type_362 = convert_element_type_363 = None
	        mul_5894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2784, view_946);  sub_2784 = view_946 = None
	        view_948: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5894, [sym_size_int, 1500, 1280]);  mul_5894 = None
	        _assert_tensor_metadata_547 = torch.ops.aten._assert_tensor_metadata.default(view_948, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_547 = None
	        view_949: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_950: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_951: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_548 = torch.ops.aten._assert_tensor_metadata.default(view_949, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_548 = None
	        convert_element_type_364: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_949, torch.float32);  view_949 = None
	        _assert_tensor_metadata_549 = torch.ops.aten._assert_tensor_metadata.default(view_951, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_549 = None
	        convert_element_type_365: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_951, torch.float32);  view_951 = None
	        sub_2788: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_364, convert_element_type_365);  convert_element_type_364 = convert_element_type_365 = None
	        mul_5899: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2788, view_950);  sub_2788 = view_950 = None
	        view_952: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5899, [1280, 1280]);  mul_5899 = None
	        _assert_tensor_metadata_550 = torch.ops.aten._assert_tensor_metadata.default(view_952, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_550 = None
	        mul_5904: "Sym(1500*s6)" = sym_size_int * 1500
	        view_953: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_948, [mul_5904, 1280]);  view_948 = mul_5904 = None
	        permute_101: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_952, [1, 0]);  view_952 = None
	        addmm_50: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_10_self_attn_q_proj_bias, view_953, permute_101);  model_audio_tower_layers_10_self_attn_q_proj_bias = view_953 = permute_101 = None
	        view_954: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_50, [sym_size_int, 1500, 1280]);  addmm_50 = None
	        mul_5911: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_954, 0.125);  view_954 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_955: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_5911, [sym_size_int, 1500, 20, 64]);  mul_5911 = None
	        permute_102: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_955, [0, 2, 1, 3]);  view_955 = None
	        clone_82: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_102, memory_format = torch.contiguous_format);  permute_102 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_956: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_9212, [sym_size_int, 1500, 1280])
	        amin_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_956, [2])
	        amax_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_956, [2]);  view_956 = None
	        full_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_61, full_122);  amin_61 = full_122 = None
	        full_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_61, full_123);  amax_61 = full_123 = None
	        sub_2803: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_61, minimum_61);  maximum_61 = None
	        div_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2803, 255.0);  sub_2803 = None
	        clamp_min_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_122, 1.1920928955078125e-07);  div_122 = None
	        div_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_61, clamp_min_183);  minimum_61 = None
	        round_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_123);  div_123 = None
	        sub_2809: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_123);  round_123 = None
	        clamp_min_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2809, -128);  sub_2809 = None
	        clamp_max_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_184, 127);  clamp_min_184 = None
	        _assert_tensor_metadata_551 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_183, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_551 = None
	        _assert_tensor_metadata_552 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_122, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_552 = None
	        convert_element_type_366: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_122, torch.int8);  clamp_max_122 = None
	        view_957: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_9212, [sym_size_int, 1500, 1280])
	        view_958: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_183, [sym_size_int, 1500, 1])
	        view_959: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_366, [sym_size_int, 1500, 1])
	        reciprocal_61: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_958);  view_958 = None
	        mul_5965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_61, 1.0);  reciprocal_61 = None
	        mul_5968: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_957, mul_5965);  view_957 = mul_5965 = None
	        round_124: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5968);  mul_5968 = None
	        add_9451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_124, view_959);  round_124 = view_959 = None
	        clamp_min_185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9451, -128);  add_9451 = None
	        clamp_max_123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_185, 127);  clamp_min_185 = None
	        view_960: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_123, [sym_size_int, 1500, 1280]);  clamp_max_123 = None
	        _assert_tensor_metadata_553 = torch.ops.aten._assert_tensor_metadata.default(view_960, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_553 = None
	        convert_element_type_367: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_960, torch.int8);  view_960 = None
	        view_961: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_367, [sym_size_int, 1500, 1280]);  convert_element_type_367 = None
	        view_962: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_183, [sym_size_int, 1500, 1]);  clamp_min_183 = None
	        view_963: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_366, [sym_size_int, 1500, 1]);  convert_element_type_366 = None
	        _assert_tensor_metadata_554 = torch.ops.aten._assert_tensor_metadata.default(view_961, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_554 = None
	        convert_element_type_368: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_961, torch.float32);  view_961 = None
	        _assert_tensor_metadata_555 = torch.ops.aten._assert_tensor_metadata.default(view_963, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_555 = None
	        convert_element_type_369: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_963, torch.float32);  view_963 = None
	        sub_2829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_368, convert_element_type_369);  convert_element_type_368 = convert_element_type_369 = None
	        mul_5990: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2829, view_962);  sub_2829 = view_962 = None
	        view_964: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5990, [sym_size_int, 1500, 1280]);  mul_5990 = None
	        _assert_tensor_metadata_556 = torch.ops.aten._assert_tensor_metadata.default(view_964, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_556 = None
	        view_965: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_966: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_967: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_557 = torch.ops.aten._assert_tensor_metadata.default(view_965, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_557 = None
	        convert_element_type_370: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_965, torch.float32);  view_965 = None
	        _assert_tensor_metadata_558 = torch.ops.aten._assert_tensor_metadata.default(view_967, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_558 = None
	        convert_element_type_371: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_967, torch.float32);  view_967 = None
	        sub_2833: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_370, convert_element_type_371);  convert_element_type_370 = convert_element_type_371 = None
	        mul_5995: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2833, view_966);  sub_2833 = view_966 = None
	        view_968: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5995, [1280, 1280]);  mul_5995 = None
	        _assert_tensor_metadata_559 = torch.ops.aten._assert_tensor_metadata.default(view_968, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_559 = None
	        permute_103: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_968, [1, 0]);  view_968 = None
	        mul_5998: "Sym(1500*s6)" = sym_size_int * 1500
	        view_969: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_964, [mul_5998, 1280]);  view_964 = mul_5998 = None
	        mm_10: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_969, permute_103);  view_969 = permute_103 = None
	        view_970: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_10, [sym_size_int, 1500, 1280]);  mm_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_971: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_970, [sym_size_int, -1, 20, 64]);  view_970 = None
	        permute_104: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_971, [0, 2, 1, 3]);  view_971 = None
	        clone_83: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_104, memory_format = torch.contiguous_format);  permute_104 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_972: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_9212, [sym_size_int, 1500, 1280])
	        amin_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_972, [2])
	        amax_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_972, [2]);  view_972 = None
	        full_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_62, full_124);  amin_62 = full_124 = None
	        full_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_62, full_125);  amax_62 = full_125 = None
	        sub_2847: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_62, minimum_62);  maximum_62 = None
	        div_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2847, 255.0);  sub_2847 = None
	        clamp_min_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_124, 1.1920928955078125e-07);  div_124 = None
	        div_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_62, clamp_min_186);  minimum_62 = None
	        round_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_125);  div_125 = None
	        sub_2853: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_125);  round_125 = None
	        clamp_min_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2853, -128);  sub_2853 = None
	        clamp_max_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_187, 127);  clamp_min_187 = None
	        _assert_tensor_metadata_560 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_186, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_560 = None
	        _assert_tensor_metadata_561 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_124, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_561 = None
	        convert_element_type_372: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_124, torch.int8);  clamp_max_124 = None
	        view_973: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_9212, [sym_size_int, 1500, 1280]);  add_9212 = None
	        view_974: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_186, [sym_size_int, 1500, 1])
	        view_975: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_372, [sym_size_int, 1500, 1])
	        reciprocal_62: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_974);  view_974 = None
	        mul_6064: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_62, 1.0);  reciprocal_62 = None
	        mul_6067: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_973, mul_6064);  view_973 = mul_6064 = None
	        round_126: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6067);  mul_6067 = None
	        add_9599: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_126, view_975);  round_126 = view_975 = None
	        clamp_min_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9599, -128);  add_9599 = None
	        clamp_max_125: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_188, 127);  clamp_min_188 = None
	        view_976: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_125, [sym_size_int, 1500, 1280]);  clamp_max_125 = None
	        _assert_tensor_metadata_562 = torch.ops.aten._assert_tensor_metadata.default(view_976, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_562 = None
	        convert_element_type_373: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_976, torch.int8);  view_976 = None
	        view_977: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_373, [sym_size_int, 1500, 1280]);  convert_element_type_373 = None
	        view_978: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_186, [sym_size_int, 1500, 1]);  clamp_min_186 = None
	        view_979: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_372, [sym_size_int, 1500, 1]);  convert_element_type_372 = None
	        _assert_tensor_metadata_563 = torch.ops.aten._assert_tensor_metadata.default(view_977, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_563 = None
	        convert_element_type_374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_977, torch.float32);  view_977 = None
	        _assert_tensor_metadata_564 = torch.ops.aten._assert_tensor_metadata.default(view_979, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_564 = None
	        convert_element_type_375: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_979, torch.float32);  view_979 = None
	        sub_2873: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_374, convert_element_type_375);  convert_element_type_374 = convert_element_type_375 = None
	        mul_6089: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2873, view_978);  sub_2873 = view_978 = None
	        view_980: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6089, [sym_size_int, 1500, 1280]);  mul_6089 = None
	        _assert_tensor_metadata_565 = torch.ops.aten._assert_tensor_metadata.default(view_980, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_565 = None
	        view_981: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_982: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_983: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_566 = torch.ops.aten._assert_tensor_metadata.default(view_981, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_566 = None
	        convert_element_type_376: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_981, torch.float32);  view_981 = None
	        _assert_tensor_metadata_567 = torch.ops.aten._assert_tensor_metadata.default(view_983, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_567 = None
	        convert_element_type_377: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_983, torch.float32);  view_983 = None
	        sub_2877: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_376, convert_element_type_377);  convert_element_type_376 = convert_element_type_377 = None
	        mul_6094: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2877, view_982);  sub_2877 = view_982 = None
	        view_984: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6094, [1280, 1280]);  mul_6094 = None
	        _assert_tensor_metadata_568 = torch.ops.aten._assert_tensor_metadata.default(view_984, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_568 = None
	        mul_6099: "Sym(1500*s6)" = sym_size_int * 1500
	        view_985: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_980, [mul_6099, 1280]);  view_980 = mul_6099 = None
	        permute_105: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_984, [1, 0]);  view_984 = None
	        addmm_51: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_10_self_attn_v_proj_bias, view_985, permute_105);  model_audio_tower_layers_10_self_attn_v_proj_bias = view_985 = permute_105 = None
	        view_986: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_51, [sym_size_int, 1500, 1280]);  addmm_51 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_987: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_986, [sym_size_int, -1, 20, 64]);  view_986 = None
	        permute_106: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_987, [0, 2, 1, 3]);  view_987 = None
	        clone_84: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_106, memory_format = torch.contiguous_format);  permute_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_10 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_82, clone_83, clone_84, None, False, scale = 1.0);  clone_82 = clone_83 = clone_84 = None
	        getitem_82: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_10[0];  _scaled_dot_product_efficient_attention_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_107: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_82, [0, 2, 1, 3]);  getitem_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_988: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_107, [sym_size_int, 1500, -1]);  permute_107 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_989: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_988, [sym_size_int, 1500, 1280])
	        amin_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_989, [2])
	        amax_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_989, [2]);  view_989 = None
	        full_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_63, full_126);  amin_63 = full_126 = None
	        full_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_63, full_127);  amax_63 = full_127 = None
	        sub_2895: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_63, minimum_63);  maximum_63 = None
	        div_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2895, 255.0);  sub_2895 = None
	        clamp_min_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_126, 1.1920928955078125e-07);  div_126 = None
	        div_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_63, clamp_min_189);  minimum_63 = None
	        round_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_127);  div_127 = None
	        sub_2901: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_127);  round_127 = None
	        clamp_min_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2901, -128);  sub_2901 = None
	        clamp_max_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_190, 127);  clamp_min_190 = None
	        _assert_tensor_metadata_569 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_189, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_569 = None
	        _assert_tensor_metadata_570 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_126, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_570 = None
	        convert_element_type_378: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_126, torch.int8);  clamp_max_126 = None
	        view_990: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_988, [sym_size_int, 1500, 1280]);  view_988 = None
	        view_991: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_189, [sym_size_int, 1500, 1])
	        view_992: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_378, [sym_size_int, 1500, 1])
	        reciprocal_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_991);  view_991 = None
	        mul_6169: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_63, 1.0);  reciprocal_63 = None
	        mul_6172: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_990, mul_6169);  view_990 = mul_6169 = None
	        round_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6172);  mul_6172 = None
	        add_9763: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_128, view_992);  round_128 = view_992 = None
	        clamp_min_191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9763, -128);  add_9763 = None
	        clamp_max_127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_191, 127);  clamp_min_191 = None
	        view_993: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_127, [sym_size_int, 1500, 1280]);  clamp_max_127 = None
	        _assert_tensor_metadata_571 = torch.ops.aten._assert_tensor_metadata.default(view_993, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_571 = None
	        convert_element_type_379: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_993, torch.int8);  view_993 = None
	        view_994: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_379, [sym_size_int, 1500, 1280]);  convert_element_type_379 = None
	        view_995: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_189, [sym_size_int, 1500, 1]);  clamp_min_189 = None
	        view_996: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_378, [sym_size_int, 1500, 1]);  convert_element_type_378 = None
	        _assert_tensor_metadata_572 = torch.ops.aten._assert_tensor_metadata.default(view_994, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_572 = None
	        convert_element_type_380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_994, torch.float32);  view_994 = None
	        _assert_tensor_metadata_573 = torch.ops.aten._assert_tensor_metadata.default(view_996, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_573 = None
	        convert_element_type_381: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_996, torch.float32);  view_996 = None
	        sub_2921: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_380, convert_element_type_381);  convert_element_type_380 = convert_element_type_381 = None
	        mul_6194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2921, view_995);  sub_2921 = view_995 = None
	        view_997: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6194, [sym_size_int, 1500, 1280]);  mul_6194 = None
	        _assert_tensor_metadata_574 = torch.ops.aten._assert_tensor_metadata.default(view_997, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_574 = None
	        view_998: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_999: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1000: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_575 = torch.ops.aten._assert_tensor_metadata.default(view_998, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_575 = None
	        convert_element_type_382: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_998, torch.float32);  view_998 = None
	        _assert_tensor_metadata_576 = torch.ops.aten._assert_tensor_metadata.default(view_1000, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_576 = None
	        convert_element_type_383: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1000, torch.float32);  view_1000 = None
	        sub_2925: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_382, convert_element_type_383);  convert_element_type_382 = convert_element_type_383 = None
	        mul_6199: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2925, view_999);  sub_2925 = view_999 = None
	        view_1001: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6199, [1280, 1280]);  mul_6199 = None
	        _assert_tensor_metadata_577 = torch.ops.aten._assert_tensor_metadata.default(view_1001, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_577 = None
	        mul_6204: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1002: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_997, [mul_6204, 1280]);  view_997 = mul_6204 = None
	        permute_108: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1001, [1, 0]);  view_1001 = None
	        addmm_52: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_10_self_attn_out_proj_bias, view_1002, permute_108);  model_audio_tower_layers_10_self_attn_out_proj_bias = view_1002 = permute_108 = None
	        view_1003: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_52, [sym_size_int, 1500, 1280]);  addmm_52 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_85: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1003);  view_1003 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_9826: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_9206, clone_85);  add_9206 = clone_85 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_9826, memory_format = torch.contiguous_format)
	        var_mean_21 = torch.ops.aten.var_mean.correction(clone_86, [2], correction = 0, keepdim = True)
	        getitem_86: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_21[0]
	        getitem_87: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_21[1];  var_mean_21 = None
	        add_9831: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_86, 1e-05);  getitem_86 = None
	        rsqrt_21: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_9831);  add_9831 = None
	        sub_2931: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_86, getitem_87);  clone_86 = getitem_87 = None
	        mul_6215: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2931, rsqrt_21);  sub_2931 = rsqrt_21 = None
	        mul_6216: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6215, model_audio_tower_layers_10_final_layer_norm_weight);  mul_6215 = model_audio_tower_layers_10_final_layer_norm_weight = None
	        add_9832: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6216, model_audio_tower_layers_10_final_layer_norm_bias);  mul_6216 = model_audio_tower_layers_10_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1004: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_9832, [sym_size_int, 1500, 1280])
	        amin_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1004, [2])
	        amax_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1004, [2]);  view_1004 = None
	        full_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_64, full_128);  amin_64 = full_128 = None
	        full_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_64, full_129);  amax_64 = full_129 = None
	        sub_2942: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_64, minimum_64);  maximum_64 = None
	        div_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2942, 255.0);  sub_2942 = None
	        clamp_min_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_128, 1.1920928955078125e-07);  div_128 = None
	        div_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_64, clamp_min_192);  minimum_64 = None
	        round_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_129);  div_129 = None
	        sub_2948: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_129);  round_129 = None
	        clamp_min_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2948, -128);  sub_2948 = None
	        clamp_max_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_193, 127);  clamp_min_193 = None
	        _assert_tensor_metadata_578 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_192, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_578 = None
	        _assert_tensor_metadata_579 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_128, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_579 = None
	        convert_element_type_384: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_128, torch.int8);  clamp_max_128 = None
	        view_1005: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_9832, [sym_size_int, 1500, 1280]);  add_9832 = None
	        view_1006: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_192, [sym_size_int, 1500, 1])
	        view_1007: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_384, [sym_size_int, 1500, 1])
	        reciprocal_64: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1006);  view_1006 = None
	        mul_6264: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_64, 1.0);  reciprocal_64 = None
	        mul_6267: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1005, mul_6264);  view_1005 = mul_6264 = None
	        round_130: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6267);  mul_6267 = None
	        add_9919: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_130, view_1007);  round_130 = view_1007 = None
	        clamp_min_194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9919, -128);  add_9919 = None
	        clamp_max_129: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_194, 127);  clamp_min_194 = None
	        view_1008: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_129, [sym_size_int, 1500, 1280]);  clamp_max_129 = None
	        _assert_tensor_metadata_580 = torch.ops.aten._assert_tensor_metadata.default(view_1008, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_580 = None
	        convert_element_type_385: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1008, torch.int8);  view_1008 = None
	        view_1009: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_385, [sym_size_int, 1500, 1280]);  convert_element_type_385 = None
	        view_1010: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_192, [sym_size_int, 1500, 1]);  clamp_min_192 = None
	        view_1011: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_384, [sym_size_int, 1500, 1]);  convert_element_type_384 = None
	        _assert_tensor_metadata_581 = torch.ops.aten._assert_tensor_metadata.default(view_1009, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_581 = None
	        convert_element_type_386: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1009, torch.float32);  view_1009 = None
	        _assert_tensor_metadata_582 = torch.ops.aten._assert_tensor_metadata.default(view_1011, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_582 = None
	        convert_element_type_387: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1011, torch.float32);  view_1011 = None
	        sub_2968: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_386, convert_element_type_387);  convert_element_type_386 = convert_element_type_387 = None
	        mul_6289: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2968, view_1010);  sub_2968 = view_1010 = None
	        view_1012: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6289, [sym_size_int, 1500, 1280]);  mul_6289 = None
	        _assert_tensor_metadata_583 = torch.ops.aten._assert_tensor_metadata.default(view_1012, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_583 = None
	        view_1013: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_10_fc1_parametrizations_weight_original0 = None
	        view_1014: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_10_fc1_parametrizations_weight_original1 = None
	        view_1015: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_10_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_584 = torch.ops.aten._assert_tensor_metadata.default(view_1013, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_584 = None
	        convert_element_type_388: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1013, torch.float32);  view_1013 = None
	        _assert_tensor_metadata_585 = torch.ops.aten._assert_tensor_metadata.default(view_1015, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_585 = None
	        convert_element_type_389: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1015, torch.float32);  view_1015 = None
	        sub_2972: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_388, convert_element_type_389);  convert_element_type_388 = convert_element_type_389 = None
	        mul_6294: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2972, view_1014);  sub_2972 = view_1014 = None
	        view_1016: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6294, [5120, 1280]);  mul_6294 = None
	        _assert_tensor_metadata_586 = torch.ops.aten._assert_tensor_metadata.default(view_1016, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_586 = None
	        mul_6299: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1017: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1012, [mul_6299, 1280]);  view_1012 = mul_6299 = None
	        permute_109: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1016, [1, 0]);  view_1016 = None
	        addmm_53: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_10_fc1_bias, view_1017, permute_109);  model_audio_tower_layers_10_fc1_bias = view_1017 = permute_109 = None
	        view_1018: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_53, [sym_size_int, 1500, 5120]);  addmm_53 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_6306: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1018, 0.5)
	        mul_6307: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1018, 0.7071067811865476);  view_1018 = None
	        erf_12: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_6307);  mul_6307 = None
	        add_9978: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_12, 1);  erf_12 = None
	        mul_6308: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6306, add_9978);  mul_6306 = add_9978 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_87: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_6308);  mul_6308 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1019: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_87, [sym_size_int, 1500, 5120])
	        amin_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1019, [2])
	        amax_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1019, [2]);  view_1019 = None
	        full_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_65, full_130);  amin_65 = full_130 = None
	        full_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_65, full_131);  amax_65 = full_131 = None
	        sub_2985: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_65, minimum_65);  maximum_65 = None
	        div_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2985, 255.0);  sub_2985 = None
	        clamp_min_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_130, 1.1920928955078125e-07);  div_130 = None
	        div_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_65, clamp_min_195);  minimum_65 = None
	        round_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_131);  div_131 = None
	        sub_2991: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_131);  round_131 = None
	        clamp_min_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2991, -128);  sub_2991 = None
	        clamp_max_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_196, 127);  clamp_min_196 = None
	        _assert_tensor_metadata_587 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_195, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_587 = None
	        _assert_tensor_metadata_588 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_130, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_588 = None
	        convert_element_type_390: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_130, torch.int8);  clamp_max_130 = None
	        view_1020: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_87, [sym_size_int, 1500, 5120]);  clone_87 = None
	        view_1021: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_195, [sym_size_int, 1500, 1])
	        view_1022: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_390, [sym_size_int, 1500, 1])
	        reciprocal_65: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1021);  view_1021 = None
	        mul_6354: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_65, 1.0);  reciprocal_65 = None
	        mul_6357: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1020, mul_6354);  view_1020 = mul_6354 = None
	        round_132: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_6357);  mul_6357 = None
	        add_10061: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_132, view_1022);  round_132 = view_1022 = None
	        clamp_min_197: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10061, -128);  add_10061 = None
	        clamp_max_131: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_197, 127);  clamp_min_197 = None
	        view_1023: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_131, [sym_size_int, 1500, 5120]);  clamp_max_131 = None
	        _assert_tensor_metadata_589 = torch.ops.aten._assert_tensor_metadata.default(view_1023, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_589 = None
	        convert_element_type_391: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1023, torch.int8);  view_1023 = None
	        view_1024: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_391, [sym_size_int, 1500, 5120]);  convert_element_type_391 = None
	        view_1025: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_195, [sym_size_int, 1500, 1]);  clamp_min_195 = None
	        view_1026: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_390, [sym_size_int, 1500, 1]);  convert_element_type_390 = None
	        _assert_tensor_metadata_590 = torch.ops.aten._assert_tensor_metadata.default(view_1024, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_590 = None
	        convert_element_type_392: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1024, torch.float32);  view_1024 = None
	        _assert_tensor_metadata_591 = torch.ops.aten._assert_tensor_metadata.default(view_1026, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_591 = None
	        convert_element_type_393: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1026, torch.float32);  view_1026 = None
	        sub_3011: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_392, convert_element_type_393);  convert_element_type_392 = convert_element_type_393 = None
	        mul_6379: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3011, view_1025);  sub_3011 = view_1025 = None
	        view_1027: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_6379, [sym_size_int, 1500, 5120]);  mul_6379 = None
	        _assert_tensor_metadata_592 = torch.ops.aten._assert_tensor_metadata.default(view_1027, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_592 = None
	        view_1028: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_10_fc2_parametrizations_weight_original0 = None
	        view_1029: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_10_fc2_parametrizations_weight_original1 = None
	        view_1030: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_10_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_593 = torch.ops.aten._assert_tensor_metadata.default(view_1028, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_593 = None
	        convert_element_type_394: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1028, torch.float32);  view_1028 = None
	        _assert_tensor_metadata_594 = torch.ops.aten._assert_tensor_metadata.default(view_1030, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_594 = None
	        convert_element_type_395: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1030, torch.float32);  view_1030 = None
	        sub_3015: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_394, convert_element_type_395);  convert_element_type_394 = convert_element_type_395 = None
	        mul_6384: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3015, view_1029);  sub_3015 = view_1029 = None
	        view_1031: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_6384, [1280, 5120]);  mul_6384 = None
	        _assert_tensor_metadata_595 = torch.ops.aten._assert_tensor_metadata.default(view_1031, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_595 = None
	        mul_6389: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1032: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1027, [mul_6389, 5120]);  view_1027 = mul_6389 = None
	        permute_110: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1031, [1, 0]);  view_1031 = None
	        addmm_54: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_10_fc2_bias, view_1032, permute_110);  model_audio_tower_layers_10_fc2_bias = view_1032 = permute_110 = None
	        view_1033: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_54, [sym_size_int, 1500, 1280]);  addmm_54 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_88: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1033);  view_1033 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_10124: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_9826, clone_88);  add_9826 = clone_88 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_89: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_10124, memory_format = torch.contiguous_format)
	        var_mean_22 = torch.ops.aten.var_mean.correction(clone_89, [2], correction = 0, keepdim = True)
	        getitem_88: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_22[0]
	        getitem_89: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_22[1];  var_mean_22 = None
	        add_10129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_88, 1e-05);  getitem_88 = None
	        rsqrt_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_10129);  add_10129 = None
	        sub_3021: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_89, getitem_89);  clone_89 = getitem_89 = None
	        mul_6400: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3021, rsqrt_22);  sub_3021 = rsqrt_22 = None
	        mul_6401: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6400, model_audio_tower_layers_11_self_attn_layer_norm_weight);  mul_6400 = model_audio_tower_layers_11_self_attn_layer_norm_weight = None
	        add_10130: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6401, model_audio_tower_layers_11_self_attn_layer_norm_bias);  mul_6401 = model_audio_tower_layers_11_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1034: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_10130, [sym_size_int, 1500, 1280])
	        amin_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1034, [2])
	        amax_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1034, [2]);  view_1034 = None
	        full_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_66, full_132);  amin_66 = full_132 = None
	        full_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_66, full_133);  amax_66 = full_133 = None
	        sub_3032: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_66, minimum_66);  maximum_66 = None
	        div_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3032, 255.0);  sub_3032 = None
	        clamp_min_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_132, 1.1920928955078125e-07);  div_132 = None
	        div_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_66, clamp_min_198);  minimum_66 = None
	        round_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_133);  div_133 = None
	        sub_3038: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_133);  round_133 = None
	        clamp_min_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3038, -128);  sub_3038 = None
	        clamp_max_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_199, 127);  clamp_min_199 = None
	        _assert_tensor_metadata_596 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_198, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_596 = None
	        _assert_tensor_metadata_597 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_132, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_597 = None
	        convert_element_type_396: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_132, torch.int8);  clamp_max_132 = None
	        view_1035: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_10130, [sym_size_int, 1500, 1280])
	        view_1036: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_198, [sym_size_int, 1500, 1])
	        view_1037: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_396, [sym_size_int, 1500, 1])
	        reciprocal_66: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1036);  view_1036 = None
	        mul_6449: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_66, 1.0);  reciprocal_66 = None
	        mul_6452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1035, mul_6449);  view_1035 = mul_6449 = None
	        round_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6452);  mul_6452 = None
	        add_10217: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_134, view_1037);  round_134 = view_1037 = None
	        clamp_min_200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10217, -128);  add_10217 = None
	        clamp_max_133: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_200, 127);  clamp_min_200 = None
	        view_1038: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_133, [sym_size_int, 1500, 1280]);  clamp_max_133 = None
	        _assert_tensor_metadata_598 = torch.ops.aten._assert_tensor_metadata.default(view_1038, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_598 = None
	        convert_element_type_397: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1038, torch.int8);  view_1038 = None
	        view_1039: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_397, [sym_size_int, 1500, 1280]);  convert_element_type_397 = None
	        view_1040: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_198, [sym_size_int, 1500, 1]);  clamp_min_198 = None
	        view_1041: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_396, [sym_size_int, 1500, 1]);  convert_element_type_396 = None
	        _assert_tensor_metadata_599 = torch.ops.aten._assert_tensor_metadata.default(view_1039, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_599 = None
	        convert_element_type_398: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1039, torch.float32);  view_1039 = None
	        _assert_tensor_metadata_600 = torch.ops.aten._assert_tensor_metadata.default(view_1041, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_600 = None
	        convert_element_type_399: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1041, torch.float32);  view_1041 = None
	        sub_3058: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_398, convert_element_type_399);  convert_element_type_398 = convert_element_type_399 = None
	        mul_6474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3058, view_1040);  sub_3058 = view_1040 = None
	        view_1042: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6474, [sym_size_int, 1500, 1280]);  mul_6474 = None
	        _assert_tensor_metadata_601 = torch.ops.aten._assert_tensor_metadata.default(view_1042, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_601 = None
	        view_1043: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1044: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1045: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_602 = torch.ops.aten._assert_tensor_metadata.default(view_1043, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_602 = None
	        convert_element_type_400: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1043, torch.float32);  view_1043 = None
	        _assert_tensor_metadata_603 = torch.ops.aten._assert_tensor_metadata.default(view_1045, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_603 = None
	        convert_element_type_401: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1045, torch.float32);  view_1045 = None
	        sub_3062: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_400, convert_element_type_401);  convert_element_type_400 = convert_element_type_401 = None
	        mul_6479: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3062, view_1044);  sub_3062 = view_1044 = None
	        view_1046: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6479, [1280, 1280]);  mul_6479 = None
	        _assert_tensor_metadata_604 = torch.ops.aten._assert_tensor_metadata.default(view_1046, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_604 = None
	        mul_6484: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1047: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1042, [mul_6484, 1280]);  view_1042 = mul_6484 = None
	        permute_111: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1046, [1, 0]);  view_1046 = None
	        addmm_55: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_11_self_attn_q_proj_bias, view_1047, permute_111);  model_audio_tower_layers_11_self_attn_q_proj_bias = view_1047 = permute_111 = None
	        view_1048: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_55, [sym_size_int, 1500, 1280]);  addmm_55 = None
	        mul_6491: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1048, 0.125);  view_1048 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1049: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_6491, [sym_size_int, 1500, 20, 64]);  mul_6491 = None
	        permute_112: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1049, [0, 2, 1, 3]);  view_1049 = None
	        clone_90: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_112, memory_format = torch.contiguous_format);  permute_112 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1050: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_10130, [sym_size_int, 1500, 1280])
	        amin_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1050, [2])
	        amax_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1050, [2]);  view_1050 = None
	        full_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_67, full_134);  amin_67 = full_134 = None
	        full_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_67, full_135);  amax_67 = full_135 = None
	        sub_3077: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_67, minimum_67);  maximum_67 = None
	        div_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3077, 255.0);  sub_3077 = None
	        clamp_min_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_134, 1.1920928955078125e-07);  div_134 = None
	        div_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_67, clamp_min_201);  minimum_67 = None
	        round_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_135);  div_135 = None
	        sub_3083: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_135);  round_135 = None
	        clamp_min_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3083, -128);  sub_3083 = None
	        clamp_max_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_202, 127);  clamp_min_202 = None
	        _assert_tensor_metadata_605 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_201, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_605 = None
	        _assert_tensor_metadata_606 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_134, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_606 = None
	        convert_element_type_402: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_134, torch.int8);  clamp_max_134 = None
	        view_1051: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_10130, [sym_size_int, 1500, 1280])
	        view_1052: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_201, [sym_size_int, 1500, 1])
	        view_1053: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_402, [sym_size_int, 1500, 1])
	        reciprocal_67: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1052);  view_1052 = None
	        mul_6545: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_67, 1.0);  reciprocal_67 = None
	        mul_6548: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1051, mul_6545);  view_1051 = mul_6545 = None
	        round_136: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6548);  mul_6548 = None
	        add_10369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_136, view_1053);  round_136 = view_1053 = None
	        clamp_min_203: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10369, -128);  add_10369 = None
	        clamp_max_135: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_203, 127);  clamp_min_203 = None
	        view_1054: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_135, [sym_size_int, 1500, 1280]);  clamp_max_135 = None
	        _assert_tensor_metadata_607 = torch.ops.aten._assert_tensor_metadata.default(view_1054, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_607 = None
	        convert_element_type_403: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1054, torch.int8);  view_1054 = None
	        view_1055: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_403, [sym_size_int, 1500, 1280]);  convert_element_type_403 = None
	        view_1056: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_201, [sym_size_int, 1500, 1]);  clamp_min_201 = None
	        view_1057: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_402, [sym_size_int, 1500, 1]);  convert_element_type_402 = None
	        _assert_tensor_metadata_608 = torch.ops.aten._assert_tensor_metadata.default(view_1055, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_608 = None
	        convert_element_type_404: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1055, torch.float32);  view_1055 = None
	        _assert_tensor_metadata_609 = torch.ops.aten._assert_tensor_metadata.default(view_1057, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_609 = None
	        convert_element_type_405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1057, torch.float32);  view_1057 = None
	        sub_3103: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_404, convert_element_type_405);  convert_element_type_404 = convert_element_type_405 = None
	        mul_6570: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3103, view_1056);  sub_3103 = view_1056 = None
	        view_1058: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6570, [sym_size_int, 1500, 1280]);  mul_6570 = None
	        _assert_tensor_metadata_610 = torch.ops.aten._assert_tensor_metadata.default(view_1058, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_610 = None
	        view_1059: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1060: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1061: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_611 = torch.ops.aten._assert_tensor_metadata.default(view_1059, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_611 = None
	        convert_element_type_406: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1059, torch.float32);  view_1059 = None
	        _assert_tensor_metadata_612 = torch.ops.aten._assert_tensor_metadata.default(view_1061, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_612 = None
	        convert_element_type_407: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1061, torch.float32);  view_1061 = None
	        sub_3107: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_406, convert_element_type_407);  convert_element_type_406 = convert_element_type_407 = None
	        mul_6575: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3107, view_1060);  sub_3107 = view_1060 = None
	        view_1062: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6575, [1280, 1280]);  mul_6575 = None
	        _assert_tensor_metadata_613 = torch.ops.aten._assert_tensor_metadata.default(view_1062, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_613 = None
	        permute_113: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1062, [1, 0]);  view_1062 = None
	        mul_6578: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1063: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1058, [mul_6578, 1280]);  view_1058 = mul_6578 = None
	        mm_11: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1063, permute_113);  view_1063 = permute_113 = None
	        view_1064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_11, [sym_size_int, 1500, 1280]);  mm_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1065: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1064, [sym_size_int, -1, 20, 64]);  view_1064 = None
	        permute_114: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1065, [0, 2, 1, 3]);  view_1065 = None
	        clone_91: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_114, memory_format = torch.contiguous_format);  permute_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1066: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_10130, [sym_size_int, 1500, 1280])
	        amin_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1066, [2])
	        amax_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1066, [2]);  view_1066 = None
	        full_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_68, full_136);  amin_68 = full_136 = None
	        full_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_68, full_137);  amax_68 = full_137 = None
	        sub_3121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_68, minimum_68);  maximum_68 = None
	        div_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3121, 255.0);  sub_3121 = None
	        clamp_min_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_136, 1.1920928955078125e-07);  div_136 = None
	        div_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_68, clamp_min_204);  minimum_68 = None
	        round_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_137);  div_137 = None
	        sub_3127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_137);  round_137 = None
	        clamp_min_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3127, -128);  sub_3127 = None
	        clamp_max_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_205, 127);  clamp_min_205 = None
	        _assert_tensor_metadata_614 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_204, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_614 = None
	        _assert_tensor_metadata_615 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_136, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_615 = None
	        convert_element_type_408: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_136, torch.int8);  clamp_max_136 = None
	        view_1067: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_10130, [sym_size_int, 1500, 1280]);  add_10130 = None
	        view_1068: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_204, [sym_size_int, 1500, 1])
	        view_1069: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_408, [sym_size_int, 1500, 1])
	        reciprocal_68: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1068);  view_1068 = None
	        mul_6644: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_68, 1.0);  reciprocal_68 = None
	        mul_6647: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1067, mul_6644);  view_1067 = mul_6644 = None
	        round_138: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6647);  mul_6647 = None
	        add_10517: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_138, view_1069);  round_138 = view_1069 = None
	        clamp_min_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10517, -128);  add_10517 = None
	        clamp_max_137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_206, 127);  clamp_min_206 = None
	        view_1070: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_137, [sym_size_int, 1500, 1280]);  clamp_max_137 = None
	        _assert_tensor_metadata_616 = torch.ops.aten._assert_tensor_metadata.default(view_1070, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_616 = None
	        convert_element_type_409: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1070, torch.int8);  view_1070 = None
	        view_1071: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_409, [sym_size_int, 1500, 1280]);  convert_element_type_409 = None
	        view_1072: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_204, [sym_size_int, 1500, 1]);  clamp_min_204 = None
	        view_1073: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_408, [sym_size_int, 1500, 1]);  convert_element_type_408 = None
	        _assert_tensor_metadata_617 = torch.ops.aten._assert_tensor_metadata.default(view_1071, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_617 = None
	        convert_element_type_410: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1071, torch.float32);  view_1071 = None
	        _assert_tensor_metadata_618 = torch.ops.aten._assert_tensor_metadata.default(view_1073, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_618 = None
	        convert_element_type_411: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1073, torch.float32);  view_1073 = None
	        sub_3147: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_410, convert_element_type_411);  convert_element_type_410 = convert_element_type_411 = None
	        mul_6669: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3147, view_1072);  sub_3147 = view_1072 = None
	        view_1074: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6669, [sym_size_int, 1500, 1280]);  mul_6669 = None
	        _assert_tensor_metadata_619 = torch.ops.aten._assert_tensor_metadata.default(view_1074, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_619 = None
	        view_1075: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1076: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1077: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_620 = torch.ops.aten._assert_tensor_metadata.default(view_1075, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_620 = None
	        convert_element_type_412: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1075, torch.float32);  view_1075 = None
	        _assert_tensor_metadata_621 = torch.ops.aten._assert_tensor_metadata.default(view_1077, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_621 = None
	        convert_element_type_413: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1077, torch.float32);  view_1077 = None
	        sub_3151: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_412, convert_element_type_413);  convert_element_type_412 = convert_element_type_413 = None
	        mul_6674: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3151, view_1076);  sub_3151 = view_1076 = None
	        view_1078: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6674, [1280, 1280]);  mul_6674 = None
	        _assert_tensor_metadata_622 = torch.ops.aten._assert_tensor_metadata.default(view_1078, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_622 = None
	        mul_6679: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1079: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1074, [mul_6679, 1280]);  view_1074 = mul_6679 = None
	        permute_115: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1078, [1, 0]);  view_1078 = None
	        addmm_56: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_11_self_attn_v_proj_bias, view_1079, permute_115);  model_audio_tower_layers_11_self_attn_v_proj_bias = view_1079 = permute_115 = None
	        view_1080: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_56, [sym_size_int, 1500, 1280]);  addmm_56 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1081: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1080, [sym_size_int, -1, 20, 64]);  view_1080 = None
	        permute_116: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1081, [0, 2, 1, 3]);  view_1081 = None
	        clone_92: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_116, memory_format = torch.contiguous_format);  permute_116 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_11 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_90, clone_91, clone_92, None, False, scale = 1.0);  clone_90 = clone_91 = clone_92 = None
	        getitem_90: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_11[0];  _scaled_dot_product_efficient_attention_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_117: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_90, [0, 2, 1, 3]);  getitem_90 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1082: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_117, [sym_size_int, 1500, -1]);  permute_117 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1083: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1082, [sym_size_int, 1500, 1280])
	        amin_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1083, [2])
	        amax_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1083, [2]);  view_1083 = None
	        full_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_69, full_138);  amin_69 = full_138 = None
	        full_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_69, full_139);  amax_69 = full_139 = None
	        sub_3169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_69, minimum_69);  maximum_69 = None
	        div_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3169, 255.0);  sub_3169 = None
	        clamp_min_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_138, 1.1920928955078125e-07);  div_138 = None
	        div_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_69, clamp_min_207);  minimum_69 = None
	        round_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_139);  div_139 = None
	        sub_3175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_139);  round_139 = None
	        clamp_min_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3175, -128);  sub_3175 = None
	        clamp_max_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_208, 127);  clamp_min_208 = None
	        _assert_tensor_metadata_623 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_207, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_623 = None
	        _assert_tensor_metadata_624 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_138, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_624 = None
	        convert_element_type_414: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_138, torch.int8);  clamp_max_138 = None
	        view_1084: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1082, [sym_size_int, 1500, 1280]);  view_1082 = None
	        view_1085: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_207, [sym_size_int, 1500, 1])
	        view_1086: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_414, [sym_size_int, 1500, 1])
	        reciprocal_69: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1085);  view_1085 = None
	        mul_6749: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_69, 1.0);  reciprocal_69 = None
	        mul_6752: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1084, mul_6749);  view_1084 = mul_6749 = None
	        round_140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6752);  mul_6752 = None
	        add_10681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_140, view_1086);  round_140 = view_1086 = None
	        clamp_min_209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10681, -128);  add_10681 = None
	        clamp_max_139: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_209, 127);  clamp_min_209 = None
	        view_1087: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_139, [sym_size_int, 1500, 1280]);  clamp_max_139 = None
	        _assert_tensor_metadata_625 = torch.ops.aten._assert_tensor_metadata.default(view_1087, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_625 = None
	        convert_element_type_415: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1087, torch.int8);  view_1087 = None
	        view_1088: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_415, [sym_size_int, 1500, 1280]);  convert_element_type_415 = None
	        view_1089: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_207, [sym_size_int, 1500, 1]);  clamp_min_207 = None
	        view_1090: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_414, [sym_size_int, 1500, 1]);  convert_element_type_414 = None
	        _assert_tensor_metadata_626 = torch.ops.aten._assert_tensor_metadata.default(view_1088, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_626 = None
	        convert_element_type_416: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1088, torch.float32);  view_1088 = None
	        _assert_tensor_metadata_627 = torch.ops.aten._assert_tensor_metadata.default(view_1090, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_627 = None
	        convert_element_type_417: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1090, torch.float32);  view_1090 = None
	        sub_3195: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_416, convert_element_type_417);  convert_element_type_416 = convert_element_type_417 = None
	        mul_6774: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3195, view_1089);  sub_3195 = view_1089 = None
	        view_1091: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6774, [sym_size_int, 1500, 1280]);  mul_6774 = None
	        _assert_tensor_metadata_628 = torch.ops.aten._assert_tensor_metadata.default(view_1091, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_628 = None
	        view_1092: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1093: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1094: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_629 = torch.ops.aten._assert_tensor_metadata.default(view_1092, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_629 = None
	        convert_element_type_418: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1092, torch.float32);  view_1092 = None
	        _assert_tensor_metadata_630 = torch.ops.aten._assert_tensor_metadata.default(view_1094, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_630 = None
	        convert_element_type_419: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1094, torch.float32);  view_1094 = None
	        sub_3199: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_418, convert_element_type_419);  convert_element_type_418 = convert_element_type_419 = None
	        mul_6779: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3199, view_1093);  sub_3199 = view_1093 = None
	        view_1095: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6779, [1280, 1280]);  mul_6779 = None
	        _assert_tensor_metadata_631 = torch.ops.aten._assert_tensor_metadata.default(view_1095, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_631 = None
	        mul_6784: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1096: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1091, [mul_6784, 1280]);  view_1091 = mul_6784 = None
	        permute_118: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1095, [1, 0]);  view_1095 = None
	        addmm_57: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_11_self_attn_out_proj_bias, view_1096, permute_118);  model_audio_tower_layers_11_self_attn_out_proj_bias = view_1096 = permute_118 = None
	        view_1097: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_57, [sym_size_int, 1500, 1280]);  addmm_57 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_93: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1097);  view_1097 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_10744: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_10124, clone_93);  add_10124 = clone_93 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_94: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_10744, memory_format = torch.contiguous_format)
	        var_mean_23 = torch.ops.aten.var_mean.correction(clone_94, [2], correction = 0, keepdim = True)
	        getitem_94: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_23[0]
	        getitem_95: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_23[1];  var_mean_23 = None
	        add_10749: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_94, 1e-05);  getitem_94 = None
	        rsqrt_23: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_10749);  add_10749 = None
	        sub_3205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_94, getitem_95);  clone_94 = getitem_95 = None
	        mul_6795: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3205, rsqrt_23);  sub_3205 = rsqrt_23 = None
	        mul_6796: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6795, model_audio_tower_layers_11_final_layer_norm_weight);  mul_6795 = model_audio_tower_layers_11_final_layer_norm_weight = None
	        add_10750: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6796, model_audio_tower_layers_11_final_layer_norm_bias);  mul_6796 = model_audio_tower_layers_11_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1098: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_10750, [sym_size_int, 1500, 1280])
	        amin_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1098, [2])
	        amax_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1098, [2]);  view_1098 = None
	        full_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_70, full_140);  amin_70 = full_140 = None
	        full_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_70, full_141);  amax_70 = full_141 = None
	        sub_3216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_70, minimum_70);  maximum_70 = None
	        div_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3216, 255.0);  sub_3216 = None
	        clamp_min_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_140, 1.1920928955078125e-07);  div_140 = None
	        div_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_70, clamp_min_210);  minimum_70 = None
	        round_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_141);  div_141 = None
	        sub_3222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_141);  round_141 = None
	        clamp_min_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3222, -128);  sub_3222 = None
	        clamp_max_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_211, 127);  clamp_min_211 = None
	        _assert_tensor_metadata_632 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_210, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_632 = None
	        _assert_tensor_metadata_633 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_140, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_633 = None
	        convert_element_type_420: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_140, torch.int8);  clamp_max_140 = None
	        view_1099: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_10750, [sym_size_int, 1500, 1280]);  add_10750 = None
	        view_1100: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_210, [sym_size_int, 1500, 1])
	        view_1101: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_420, [sym_size_int, 1500, 1])
	        reciprocal_70: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1100);  view_1100 = None
	        mul_6844: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_70, 1.0);  reciprocal_70 = None
	        mul_6847: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1099, mul_6844);  view_1099 = mul_6844 = None
	        round_142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6847);  mul_6847 = None
	        add_10837: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_142, view_1101);  round_142 = view_1101 = None
	        clamp_min_212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10837, -128);  add_10837 = None
	        clamp_max_141: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_212, 127);  clamp_min_212 = None
	        view_1102: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_141, [sym_size_int, 1500, 1280]);  clamp_max_141 = None
	        _assert_tensor_metadata_634 = torch.ops.aten._assert_tensor_metadata.default(view_1102, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_634 = None
	        convert_element_type_421: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1102, torch.int8);  view_1102 = None
	        view_1103: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_421, [sym_size_int, 1500, 1280]);  convert_element_type_421 = None
	        view_1104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_210, [sym_size_int, 1500, 1]);  clamp_min_210 = None
	        view_1105: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_420, [sym_size_int, 1500, 1]);  convert_element_type_420 = None
	        _assert_tensor_metadata_635 = torch.ops.aten._assert_tensor_metadata.default(view_1103, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_635 = None
	        convert_element_type_422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1103, torch.float32);  view_1103 = None
	        _assert_tensor_metadata_636 = torch.ops.aten._assert_tensor_metadata.default(view_1105, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_636 = None
	        convert_element_type_423: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1105, torch.float32);  view_1105 = None
	        sub_3242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_422, convert_element_type_423);  convert_element_type_422 = convert_element_type_423 = None
	        mul_6869: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3242, view_1104);  sub_3242 = view_1104 = None
	        view_1106: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6869, [sym_size_int, 1500, 1280]);  mul_6869 = None
	        _assert_tensor_metadata_637 = torch.ops.aten._assert_tensor_metadata.default(view_1106, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_637 = None
	        view_1107: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_11_fc1_parametrizations_weight_original0 = None
	        view_1108: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_11_fc1_parametrizations_weight_original1 = None
	        view_1109: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_11_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_638 = torch.ops.aten._assert_tensor_metadata.default(view_1107, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_638 = None
	        convert_element_type_424: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1107, torch.float32);  view_1107 = None
	        _assert_tensor_metadata_639 = torch.ops.aten._assert_tensor_metadata.default(view_1109, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_639 = None
	        convert_element_type_425: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1109, torch.float32);  view_1109 = None
	        sub_3246: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_424, convert_element_type_425);  convert_element_type_424 = convert_element_type_425 = None
	        mul_6874: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3246, view_1108);  sub_3246 = view_1108 = None
	        view_1110: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6874, [5120, 1280]);  mul_6874 = None
	        _assert_tensor_metadata_640 = torch.ops.aten._assert_tensor_metadata.default(view_1110, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_640 = None
	        mul_6879: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1111: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1106, [mul_6879, 1280]);  view_1106 = mul_6879 = None
	        permute_119: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1110, [1, 0]);  view_1110 = None
	        addmm_58: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_11_fc1_bias, view_1111, permute_119);  model_audio_tower_layers_11_fc1_bias = view_1111 = permute_119 = None
	        view_1112: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_58, [sym_size_int, 1500, 5120]);  addmm_58 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_6886: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1112, 0.5)
	        mul_6887: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1112, 0.7071067811865476);  view_1112 = None
	        erf_13: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_6887);  mul_6887 = None
	        add_10896: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_13, 1);  erf_13 = None
	        mul_6888: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6886, add_10896);  mul_6886 = add_10896 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_95: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_6888);  mul_6888 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1113: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_95, [sym_size_int, 1500, 5120])
	        amin_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1113, [2])
	        amax_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1113, [2]);  view_1113 = None
	        full_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_71, full_142);  amin_71 = full_142 = None
	        full_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_71, full_143);  amax_71 = full_143 = None
	        sub_3259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_71, minimum_71);  maximum_71 = None
	        div_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3259, 255.0);  sub_3259 = None
	        clamp_min_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_142, 1.1920928955078125e-07);  div_142 = None
	        div_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_71, clamp_min_213);  minimum_71 = None
	        round_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_143);  div_143 = None
	        sub_3265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_143);  round_143 = None
	        clamp_min_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3265, -128);  sub_3265 = None
	        clamp_max_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_214, 127);  clamp_min_214 = None
	        _assert_tensor_metadata_641 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_213, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_641 = None
	        _assert_tensor_metadata_642 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_142, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_642 = None
	        convert_element_type_426: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_142, torch.int8);  clamp_max_142 = None
	        view_1114: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_95, [sym_size_int, 1500, 5120]);  clone_95 = None
	        view_1115: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_213, [sym_size_int, 1500, 1])
	        view_1116: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_426, [sym_size_int, 1500, 1])
	        reciprocal_71: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1115);  view_1115 = None
	        mul_6934: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_71, 1.0);  reciprocal_71 = None
	        mul_6937: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1114, mul_6934);  view_1114 = mul_6934 = None
	        round_144: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_6937);  mul_6937 = None
	        add_10979: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_144, view_1116);  round_144 = view_1116 = None
	        clamp_min_215: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10979, -128);  add_10979 = None
	        clamp_max_143: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_215, 127);  clamp_min_215 = None
	        view_1117: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_143, [sym_size_int, 1500, 5120]);  clamp_max_143 = None
	        _assert_tensor_metadata_643 = torch.ops.aten._assert_tensor_metadata.default(view_1117, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_643 = None
	        convert_element_type_427: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1117, torch.int8);  view_1117 = None
	        view_1118: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_427, [sym_size_int, 1500, 5120]);  convert_element_type_427 = None
	        view_1119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_213, [sym_size_int, 1500, 1]);  clamp_min_213 = None
	        view_1120: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_426, [sym_size_int, 1500, 1]);  convert_element_type_426 = None
	        _assert_tensor_metadata_644 = torch.ops.aten._assert_tensor_metadata.default(view_1118, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_644 = None
	        convert_element_type_428: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1118, torch.float32);  view_1118 = None
	        _assert_tensor_metadata_645 = torch.ops.aten._assert_tensor_metadata.default(view_1120, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_645 = None
	        convert_element_type_429: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1120, torch.float32);  view_1120 = None
	        sub_3285: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_428, convert_element_type_429);  convert_element_type_428 = convert_element_type_429 = None
	        mul_6959: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3285, view_1119);  sub_3285 = view_1119 = None
	        view_1121: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_6959, [sym_size_int, 1500, 5120]);  mul_6959 = None
	        _assert_tensor_metadata_646 = torch.ops.aten._assert_tensor_metadata.default(view_1121, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_646 = None
	        view_1122: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_11_fc2_parametrizations_weight_original0 = None
	        view_1123: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_11_fc2_parametrizations_weight_original1 = None
	        view_1124: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_11_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_647 = torch.ops.aten._assert_tensor_metadata.default(view_1122, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_647 = None
	        convert_element_type_430: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1122, torch.float32);  view_1122 = None
	        _assert_tensor_metadata_648 = torch.ops.aten._assert_tensor_metadata.default(view_1124, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_648 = None
	        convert_element_type_431: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1124, torch.float32);  view_1124 = None
	        sub_3289: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_430, convert_element_type_431);  convert_element_type_430 = convert_element_type_431 = None
	        mul_6964: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3289, view_1123);  sub_3289 = view_1123 = None
	        view_1125: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_6964, [1280, 5120]);  mul_6964 = None
	        _assert_tensor_metadata_649 = torch.ops.aten._assert_tensor_metadata.default(view_1125, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_649 = None
	        mul_6969: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1126: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1121, [mul_6969, 5120]);  view_1121 = mul_6969 = None
	        permute_120: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1125, [1, 0]);  view_1125 = None
	        addmm_59: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_11_fc2_bias, view_1126, permute_120);  model_audio_tower_layers_11_fc2_bias = view_1126 = permute_120 = None
	        view_1127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_59, [sym_size_int, 1500, 1280]);  addmm_59 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_96: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1127);  view_1127 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_11042: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_10744, clone_96);  add_10744 = clone_96 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_97: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_11042, memory_format = torch.contiguous_format)
	        var_mean_24 = torch.ops.aten.var_mean.correction(clone_97, [2], correction = 0, keepdim = True)
	        getitem_96: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_24[0]
	        getitem_97: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_24[1];  var_mean_24 = None
	        add_11047: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_96, 1e-05);  getitem_96 = None
	        rsqrt_24: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_11047);  add_11047 = None
	        sub_3295: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_97, getitem_97);  clone_97 = getitem_97 = None
	        mul_6980: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3295, rsqrt_24);  sub_3295 = rsqrt_24 = None
	        mul_6981: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6980, model_audio_tower_layers_12_self_attn_layer_norm_weight);  mul_6980 = model_audio_tower_layers_12_self_attn_layer_norm_weight = None
	        add_11048: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6981, model_audio_tower_layers_12_self_attn_layer_norm_bias);  mul_6981 = model_audio_tower_layers_12_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11048, [sym_size_int, 1500, 1280])
	        amin_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1128, [2])
	        amax_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1128, [2]);  view_1128 = None
	        full_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_72, full_144);  amin_72 = full_144 = None
	        full_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_72, full_145);  amax_72 = full_145 = None
	        sub_3306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_72, minimum_72);  maximum_72 = None
	        div_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3306, 255.0);  sub_3306 = None
	        clamp_min_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_144, 1.1920928955078125e-07);  div_144 = None
	        div_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_72, clamp_min_216);  minimum_72 = None
	        round_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_145);  div_145 = None
	        sub_3312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_145);  round_145 = None
	        clamp_min_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3312, -128);  sub_3312 = None
	        clamp_max_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_217, 127);  clamp_min_217 = None
	        _assert_tensor_metadata_650 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_216, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_650 = None
	        _assert_tensor_metadata_651 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_144, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_651 = None
	        convert_element_type_432: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_144, torch.int8);  clamp_max_144 = None
	        view_1129: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11048, [sym_size_int, 1500, 1280])
	        view_1130: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_216, [sym_size_int, 1500, 1])
	        view_1131: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_432, [sym_size_int, 1500, 1])
	        reciprocal_72: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1130);  view_1130 = None
	        mul_7029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_72, 1.0);  reciprocal_72 = None
	        mul_7032: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1129, mul_7029);  view_1129 = mul_7029 = None
	        round_146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7032);  mul_7032 = None
	        add_11135: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_146, view_1131);  round_146 = view_1131 = None
	        clamp_min_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11135, -128);  add_11135 = None
	        clamp_max_145: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_218, 127);  clamp_min_218 = None
	        view_1132: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_145, [sym_size_int, 1500, 1280]);  clamp_max_145 = None
	        _assert_tensor_metadata_652 = torch.ops.aten._assert_tensor_metadata.default(view_1132, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_652 = None
	        convert_element_type_433: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1132, torch.int8);  view_1132 = None
	        view_1133: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_433, [sym_size_int, 1500, 1280]);  convert_element_type_433 = None
	        view_1134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_216, [sym_size_int, 1500, 1]);  clamp_min_216 = None
	        view_1135: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_432, [sym_size_int, 1500, 1]);  convert_element_type_432 = None
	        _assert_tensor_metadata_653 = torch.ops.aten._assert_tensor_metadata.default(view_1133, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_653 = None
	        convert_element_type_434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1133, torch.float32);  view_1133 = None
	        _assert_tensor_metadata_654 = torch.ops.aten._assert_tensor_metadata.default(view_1135, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_654 = None
	        convert_element_type_435: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1135, torch.float32);  view_1135 = None
	        sub_3332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_434, convert_element_type_435);  convert_element_type_434 = convert_element_type_435 = None
	        mul_7054: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3332, view_1134);  sub_3332 = view_1134 = None
	        view_1136: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7054, [sym_size_int, 1500, 1280]);  mul_7054 = None
	        _assert_tensor_metadata_655 = torch.ops.aten._assert_tensor_metadata.default(view_1136, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_655 = None
	        view_1137: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1138: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1139: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_656 = torch.ops.aten._assert_tensor_metadata.default(view_1137, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_656 = None
	        convert_element_type_436: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1137, torch.float32);  view_1137 = None
	        _assert_tensor_metadata_657 = torch.ops.aten._assert_tensor_metadata.default(view_1139, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_657 = None
	        convert_element_type_437: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1139, torch.float32);  view_1139 = None
	        sub_3336: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_436, convert_element_type_437);  convert_element_type_436 = convert_element_type_437 = None
	        mul_7059: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3336, view_1138);  sub_3336 = view_1138 = None
	        view_1140: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7059, [1280, 1280]);  mul_7059 = None
	        _assert_tensor_metadata_658 = torch.ops.aten._assert_tensor_metadata.default(view_1140, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_658 = None
	        mul_7064: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1141: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1136, [mul_7064, 1280]);  view_1136 = mul_7064 = None
	        permute_121: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1140, [1, 0]);  view_1140 = None
	        addmm_60: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_12_self_attn_q_proj_bias, view_1141, permute_121);  model_audio_tower_layers_12_self_attn_q_proj_bias = view_1141 = permute_121 = None
	        view_1142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_60, [sym_size_int, 1500, 1280]);  addmm_60 = None
	        mul_7071: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1142, 0.125);  view_1142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1143: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_7071, [sym_size_int, 1500, 20, 64]);  mul_7071 = None
	        permute_122: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1143, [0, 2, 1, 3]);  view_1143 = None
	        clone_98: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_122, memory_format = torch.contiguous_format);  permute_122 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1144: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11048, [sym_size_int, 1500, 1280])
	        amin_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1144, [2])
	        amax_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1144, [2]);  view_1144 = None
	        full_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_73, full_146);  amin_73 = full_146 = None
	        full_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_73, full_147);  amax_73 = full_147 = None
	        sub_3351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_73, minimum_73);  maximum_73 = None
	        div_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3351, 255.0);  sub_3351 = None
	        clamp_min_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_146, 1.1920928955078125e-07);  div_146 = None
	        div_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_73, clamp_min_219);  minimum_73 = None
	        round_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_147);  div_147 = None
	        sub_3357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_147);  round_147 = None
	        clamp_min_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3357, -128);  sub_3357 = None
	        clamp_max_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_220, 127);  clamp_min_220 = None
	        _assert_tensor_metadata_659 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_219, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_659 = None
	        _assert_tensor_metadata_660 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_146, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_660 = None
	        convert_element_type_438: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_146, torch.int8);  clamp_max_146 = None
	        view_1145: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11048, [sym_size_int, 1500, 1280])
	        view_1146: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_219, [sym_size_int, 1500, 1])
	        view_1147: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_438, [sym_size_int, 1500, 1])
	        reciprocal_73: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1146);  view_1146 = None
	        mul_7125: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_73, 1.0);  reciprocal_73 = None
	        mul_7128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1145, mul_7125);  view_1145 = mul_7125 = None
	        round_148: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7128);  mul_7128 = None
	        add_11287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_148, view_1147);  round_148 = view_1147 = None
	        clamp_min_221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11287, -128);  add_11287 = None
	        clamp_max_147: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_221, 127);  clamp_min_221 = None
	        view_1148: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_147, [sym_size_int, 1500, 1280]);  clamp_max_147 = None
	        _assert_tensor_metadata_661 = torch.ops.aten._assert_tensor_metadata.default(view_1148, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_661 = None
	        convert_element_type_439: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1148, torch.int8);  view_1148 = None
	        view_1149: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_439, [sym_size_int, 1500, 1280]);  convert_element_type_439 = None
	        view_1150: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_219, [sym_size_int, 1500, 1]);  clamp_min_219 = None
	        view_1151: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_438, [sym_size_int, 1500, 1]);  convert_element_type_438 = None
	        _assert_tensor_metadata_662 = torch.ops.aten._assert_tensor_metadata.default(view_1149, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_662 = None
	        convert_element_type_440: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1149, torch.float32);  view_1149 = None
	        _assert_tensor_metadata_663 = torch.ops.aten._assert_tensor_metadata.default(view_1151, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_663 = None
	        convert_element_type_441: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1151, torch.float32);  view_1151 = None
	        sub_3377: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_440, convert_element_type_441);  convert_element_type_440 = convert_element_type_441 = None
	        mul_7150: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3377, view_1150);  sub_3377 = view_1150 = None
	        view_1152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7150, [sym_size_int, 1500, 1280]);  mul_7150 = None
	        _assert_tensor_metadata_664 = torch.ops.aten._assert_tensor_metadata.default(view_1152, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_664 = None
	        view_1153: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1154: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1155: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_665 = torch.ops.aten._assert_tensor_metadata.default(view_1153, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_665 = None
	        convert_element_type_442: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1153, torch.float32);  view_1153 = None
	        _assert_tensor_metadata_666 = torch.ops.aten._assert_tensor_metadata.default(view_1155, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_666 = None
	        convert_element_type_443: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1155, torch.float32);  view_1155 = None
	        sub_3381: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_442, convert_element_type_443);  convert_element_type_442 = convert_element_type_443 = None
	        mul_7155: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3381, view_1154);  sub_3381 = view_1154 = None
	        view_1156: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7155, [1280, 1280]);  mul_7155 = None
	        _assert_tensor_metadata_667 = torch.ops.aten._assert_tensor_metadata.default(view_1156, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_667 = None
	        permute_123: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1156, [1, 0]);  view_1156 = None
	        mul_7158: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1157: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1152, [mul_7158, 1280]);  view_1152 = mul_7158 = None
	        mm_12: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1157, permute_123);  view_1157 = permute_123 = None
	        view_1158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_12, [sym_size_int, 1500, 1280]);  mm_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1159: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1158, [sym_size_int, -1, 20, 64]);  view_1158 = None
	        permute_124: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1159, [0, 2, 1, 3]);  view_1159 = None
	        clone_99: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_124, memory_format = torch.contiguous_format);  permute_124 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11048, [sym_size_int, 1500, 1280])
	        amin_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1160, [2])
	        amax_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1160, [2]);  view_1160 = None
	        full_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_74, full_148);  amin_74 = full_148 = None
	        full_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_74, full_149);  amax_74 = full_149 = None
	        sub_3395: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_74, minimum_74);  maximum_74 = None
	        div_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3395, 255.0);  sub_3395 = None
	        clamp_min_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_148, 1.1920928955078125e-07);  div_148 = None
	        div_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_74, clamp_min_222);  minimum_74 = None
	        round_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_149);  div_149 = None
	        sub_3401: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_149);  round_149 = None
	        clamp_min_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3401, -128);  sub_3401 = None
	        clamp_max_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_223, 127);  clamp_min_223 = None
	        _assert_tensor_metadata_668 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_222, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_668 = None
	        _assert_tensor_metadata_669 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_148, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_669 = None
	        convert_element_type_444: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_148, torch.int8);  clamp_max_148 = None
	        view_1161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11048, [sym_size_int, 1500, 1280]);  add_11048 = None
	        view_1162: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_222, [sym_size_int, 1500, 1])
	        view_1163: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_444, [sym_size_int, 1500, 1])
	        reciprocal_74: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1162);  view_1162 = None
	        mul_7224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_74, 1.0);  reciprocal_74 = None
	        mul_7227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1161, mul_7224);  view_1161 = mul_7224 = None
	        round_150: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7227);  mul_7227 = None
	        add_11435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_150, view_1163);  round_150 = view_1163 = None
	        clamp_min_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11435, -128);  add_11435 = None
	        clamp_max_149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_224, 127);  clamp_min_224 = None
	        view_1164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_149, [sym_size_int, 1500, 1280]);  clamp_max_149 = None
	        _assert_tensor_metadata_670 = torch.ops.aten._assert_tensor_metadata.default(view_1164, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_670 = None
	        convert_element_type_445: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1164, torch.int8);  view_1164 = None
	        view_1165: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_445, [sym_size_int, 1500, 1280]);  convert_element_type_445 = None
	        view_1166: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_222, [sym_size_int, 1500, 1]);  clamp_min_222 = None
	        view_1167: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_444, [sym_size_int, 1500, 1]);  convert_element_type_444 = None
	        _assert_tensor_metadata_671 = torch.ops.aten._assert_tensor_metadata.default(view_1165, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_671 = None
	        convert_element_type_446: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1165, torch.float32);  view_1165 = None
	        _assert_tensor_metadata_672 = torch.ops.aten._assert_tensor_metadata.default(view_1167, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_672 = None
	        convert_element_type_447: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1167, torch.float32);  view_1167 = None
	        sub_3421: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_446, convert_element_type_447);  convert_element_type_446 = convert_element_type_447 = None
	        mul_7249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3421, view_1166);  sub_3421 = view_1166 = None
	        view_1168: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7249, [sym_size_int, 1500, 1280]);  mul_7249 = None
	        _assert_tensor_metadata_673 = torch.ops.aten._assert_tensor_metadata.default(view_1168, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_673 = None
	        view_1169: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1170: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1171: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_674 = torch.ops.aten._assert_tensor_metadata.default(view_1169, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_674 = None
	        convert_element_type_448: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1169, torch.float32);  view_1169 = None
	        _assert_tensor_metadata_675 = torch.ops.aten._assert_tensor_metadata.default(view_1171, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_675 = None
	        convert_element_type_449: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1171, torch.float32);  view_1171 = None
	        sub_3425: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_448, convert_element_type_449);  convert_element_type_448 = convert_element_type_449 = None
	        mul_7254: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3425, view_1170);  sub_3425 = view_1170 = None
	        view_1172: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7254, [1280, 1280]);  mul_7254 = None
	        _assert_tensor_metadata_676 = torch.ops.aten._assert_tensor_metadata.default(view_1172, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_676 = None
	        mul_7259: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1173: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1168, [mul_7259, 1280]);  view_1168 = mul_7259 = None
	        permute_125: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1172, [1, 0]);  view_1172 = None
	        addmm_61: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_12_self_attn_v_proj_bias, view_1173, permute_125);  model_audio_tower_layers_12_self_attn_v_proj_bias = view_1173 = permute_125 = None
	        view_1174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_61, [sym_size_int, 1500, 1280]);  addmm_61 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1175: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1174, [sym_size_int, -1, 20, 64]);  view_1174 = None
	        permute_126: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1175, [0, 2, 1, 3]);  view_1175 = None
	        clone_100: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_126, memory_format = torch.contiguous_format);  permute_126 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_12 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_98, clone_99, clone_100, None, False, scale = 1.0);  clone_98 = clone_99 = clone_100 = None
	        getitem_98: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_12[0];  _scaled_dot_product_efficient_attention_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_127: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_98, [0, 2, 1, 3]);  getitem_98 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_127, [sym_size_int, 1500, -1]);  permute_127 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1177: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1176, [sym_size_int, 1500, 1280])
	        amin_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1177, [2])
	        amax_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1177, [2]);  view_1177 = None
	        full_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_75, full_150);  amin_75 = full_150 = None
	        full_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_75, full_151);  amax_75 = full_151 = None
	        sub_3443: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_75, minimum_75);  maximum_75 = None
	        div_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3443, 255.0);  sub_3443 = None
	        clamp_min_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_150, 1.1920928955078125e-07);  div_150 = None
	        div_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_75, clamp_min_225);  minimum_75 = None
	        round_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_151);  div_151 = None
	        sub_3449: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_151);  round_151 = None
	        clamp_min_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3449, -128);  sub_3449 = None
	        clamp_max_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_226, 127);  clamp_min_226 = None
	        _assert_tensor_metadata_677 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_225, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_677 = None
	        _assert_tensor_metadata_678 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_150, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_678 = None
	        convert_element_type_450: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_150, torch.int8);  clamp_max_150 = None
	        view_1178: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1176, [sym_size_int, 1500, 1280]);  view_1176 = None
	        view_1179: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_225, [sym_size_int, 1500, 1])
	        view_1180: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_450, [sym_size_int, 1500, 1])
	        reciprocal_75: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1179);  view_1179 = None
	        mul_7329: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_75, 1.0);  reciprocal_75 = None
	        mul_7332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1178, mul_7329);  view_1178 = mul_7329 = None
	        round_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7332);  mul_7332 = None
	        add_11599: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_152, view_1180);  round_152 = view_1180 = None
	        clamp_min_227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11599, -128);  add_11599 = None
	        clamp_max_151: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_227, 127);  clamp_min_227 = None
	        view_1181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_151, [sym_size_int, 1500, 1280]);  clamp_max_151 = None
	        _assert_tensor_metadata_679 = torch.ops.aten._assert_tensor_metadata.default(view_1181, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_679 = None
	        convert_element_type_451: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1181, torch.int8);  view_1181 = None
	        view_1182: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_451, [sym_size_int, 1500, 1280]);  convert_element_type_451 = None
	        view_1183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_225, [sym_size_int, 1500, 1]);  clamp_min_225 = None
	        view_1184: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_450, [sym_size_int, 1500, 1]);  convert_element_type_450 = None
	        _assert_tensor_metadata_680 = torch.ops.aten._assert_tensor_metadata.default(view_1182, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_680 = None
	        convert_element_type_452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1182, torch.float32);  view_1182 = None
	        _assert_tensor_metadata_681 = torch.ops.aten._assert_tensor_metadata.default(view_1184, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_681 = None
	        convert_element_type_453: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1184, torch.float32);  view_1184 = None
	        sub_3469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_452, convert_element_type_453);  convert_element_type_452 = convert_element_type_453 = None
	        mul_7354: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3469, view_1183);  sub_3469 = view_1183 = None
	        view_1185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7354, [sym_size_int, 1500, 1280]);  mul_7354 = None
	        _assert_tensor_metadata_682 = torch.ops.aten._assert_tensor_metadata.default(view_1185, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_682 = None
	        view_1186: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1187: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1188: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_683 = torch.ops.aten._assert_tensor_metadata.default(view_1186, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_683 = None
	        convert_element_type_454: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1186, torch.float32);  view_1186 = None
	        _assert_tensor_metadata_684 = torch.ops.aten._assert_tensor_metadata.default(view_1188, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_684 = None
	        convert_element_type_455: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1188, torch.float32);  view_1188 = None
	        sub_3473: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_454, convert_element_type_455);  convert_element_type_454 = convert_element_type_455 = None
	        mul_7359: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3473, view_1187);  sub_3473 = view_1187 = None
	        view_1189: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7359, [1280, 1280]);  mul_7359 = None
	        _assert_tensor_metadata_685 = torch.ops.aten._assert_tensor_metadata.default(view_1189, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_685 = None
	        mul_7364: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1190: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1185, [mul_7364, 1280]);  view_1185 = mul_7364 = None
	        permute_128: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1189, [1, 0]);  view_1189 = None
	        addmm_62: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_12_self_attn_out_proj_bias, view_1190, permute_128);  model_audio_tower_layers_12_self_attn_out_proj_bias = view_1190 = permute_128 = None
	        view_1191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_62, [sym_size_int, 1500, 1280]);  addmm_62 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1191);  view_1191 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_11662: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_11042, clone_101);  add_11042 = clone_101 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_102: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_11662, memory_format = torch.contiguous_format)
	        var_mean_25 = torch.ops.aten.var_mean.correction(clone_102, [2], correction = 0, keepdim = True)
	        getitem_102: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_25[0]
	        getitem_103: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_25[1];  var_mean_25 = None
	        add_11667: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_102, 1e-05);  getitem_102 = None
	        rsqrt_25: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_11667);  add_11667 = None
	        sub_3479: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_102, getitem_103);  clone_102 = getitem_103 = None
	        mul_7375: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3479, rsqrt_25);  sub_3479 = rsqrt_25 = None
	        mul_7376: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7375, model_audio_tower_layers_12_final_layer_norm_weight);  mul_7375 = model_audio_tower_layers_12_final_layer_norm_weight = None
	        add_11668: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_7376, model_audio_tower_layers_12_final_layer_norm_bias);  mul_7376 = model_audio_tower_layers_12_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11668, [sym_size_int, 1500, 1280])
	        amin_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1192, [2])
	        amax_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1192, [2]);  view_1192 = None
	        full_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_76, full_152);  amin_76 = full_152 = None
	        full_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_76, full_153);  amax_76 = full_153 = None
	        sub_3490: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_76, minimum_76);  maximum_76 = None
	        div_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3490, 255.0);  sub_3490 = None
	        clamp_min_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_152, 1.1920928955078125e-07);  div_152 = None
	        div_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_76, clamp_min_228);  minimum_76 = None
	        round_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_153);  div_153 = None
	        sub_3496: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_153);  round_153 = None
	        clamp_min_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3496, -128);  sub_3496 = None
	        clamp_max_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_229, 127);  clamp_min_229 = None
	        _assert_tensor_metadata_686 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_228, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_686 = None
	        _assert_tensor_metadata_687 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_152, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_687 = None
	        convert_element_type_456: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_152, torch.int8);  clamp_max_152 = None
	        view_1193: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11668, [sym_size_int, 1500, 1280]);  add_11668 = None
	        view_1194: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_228, [sym_size_int, 1500, 1])
	        view_1195: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_456, [sym_size_int, 1500, 1])
	        reciprocal_76: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1194);  view_1194 = None
	        mul_7424: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_76, 1.0);  reciprocal_76 = None
	        mul_7427: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1193, mul_7424);  view_1193 = mul_7424 = None
	        round_154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7427);  mul_7427 = None
	        add_11755: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_154, view_1195);  round_154 = view_1195 = None
	        clamp_min_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11755, -128);  add_11755 = None
	        clamp_max_153: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_230, 127);  clamp_min_230 = None
	        view_1196: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_153, [sym_size_int, 1500, 1280]);  clamp_max_153 = None
	        _assert_tensor_metadata_688 = torch.ops.aten._assert_tensor_metadata.default(view_1196, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_688 = None
	        convert_element_type_457: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1196, torch.int8);  view_1196 = None
	        view_1197: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_457, [sym_size_int, 1500, 1280]);  convert_element_type_457 = None
	        view_1198: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_228, [sym_size_int, 1500, 1]);  clamp_min_228 = None
	        view_1199: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_456, [sym_size_int, 1500, 1]);  convert_element_type_456 = None
	        _assert_tensor_metadata_689 = torch.ops.aten._assert_tensor_metadata.default(view_1197, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_689 = None
	        convert_element_type_458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1197, torch.float32);  view_1197 = None
	        _assert_tensor_metadata_690 = torch.ops.aten._assert_tensor_metadata.default(view_1199, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_690 = None
	        convert_element_type_459: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1199, torch.float32);  view_1199 = None
	        sub_3516: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_458, convert_element_type_459);  convert_element_type_458 = convert_element_type_459 = None
	        mul_7449: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3516, view_1198);  sub_3516 = view_1198 = None
	        view_1200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7449, [sym_size_int, 1500, 1280]);  mul_7449 = None
	        _assert_tensor_metadata_691 = torch.ops.aten._assert_tensor_metadata.default(view_1200, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_691 = None
	        view_1201: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_12_fc1_parametrizations_weight_original0 = None
	        view_1202: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_12_fc1_parametrizations_weight_original1 = None
	        view_1203: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_12_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_692 = torch.ops.aten._assert_tensor_metadata.default(view_1201, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_692 = None
	        convert_element_type_460: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1201, torch.float32);  view_1201 = None
	        _assert_tensor_metadata_693 = torch.ops.aten._assert_tensor_metadata.default(view_1203, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_693 = None
	        convert_element_type_461: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1203, torch.float32);  view_1203 = None
	        sub_3520: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_460, convert_element_type_461);  convert_element_type_460 = convert_element_type_461 = None
	        mul_7454: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3520, view_1202);  sub_3520 = view_1202 = None
	        view_1204: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7454, [5120, 1280]);  mul_7454 = None
	        _assert_tensor_metadata_694 = torch.ops.aten._assert_tensor_metadata.default(view_1204, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_694 = None
	        mul_7459: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1205: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1200, [mul_7459, 1280]);  view_1200 = mul_7459 = None
	        permute_129: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1204, [1, 0]);  view_1204 = None
	        addmm_63: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_12_fc1_bias, view_1205, permute_129);  model_audio_tower_layers_12_fc1_bias = view_1205 = permute_129 = None
	        view_1206: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_63, [sym_size_int, 1500, 5120]);  addmm_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_7466: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1206, 0.5)
	        mul_7467: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1206, 0.7071067811865476);  view_1206 = None
	        erf_14: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_7467);  mul_7467 = None
	        add_11814: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_14, 1);  erf_14 = None
	        mul_7468: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7466, add_11814);  mul_7466 = add_11814 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_103: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_7468);  mul_7468 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1207: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_103, [sym_size_int, 1500, 5120])
	        amin_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1207, [2])
	        amax_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1207, [2]);  view_1207 = None
	        full_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_77, full_154);  amin_77 = full_154 = None
	        full_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_77, full_155);  amax_77 = full_155 = None
	        sub_3533: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_77, minimum_77);  maximum_77 = None
	        div_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3533, 255.0);  sub_3533 = None
	        clamp_min_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_154, 1.1920928955078125e-07);  div_154 = None
	        div_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_77, clamp_min_231);  minimum_77 = None
	        round_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_155);  div_155 = None
	        sub_3539: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_155);  round_155 = None
	        clamp_min_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3539, -128);  sub_3539 = None
	        clamp_max_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_232, 127);  clamp_min_232 = None
	        _assert_tensor_metadata_695 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_231, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_695 = None
	        _assert_tensor_metadata_696 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_154, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_696 = None
	        convert_element_type_462: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_154, torch.int8);  clamp_max_154 = None
	        view_1208: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_103, [sym_size_int, 1500, 5120]);  clone_103 = None
	        view_1209: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_231, [sym_size_int, 1500, 1])
	        view_1210: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_462, [sym_size_int, 1500, 1])
	        reciprocal_77: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1209);  view_1209 = None
	        mul_7514: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_77, 1.0);  reciprocal_77 = None
	        mul_7517: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1208, mul_7514);  view_1208 = mul_7514 = None
	        round_156: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_7517);  mul_7517 = None
	        add_11897: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_156, view_1210);  round_156 = view_1210 = None
	        clamp_min_233: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11897, -128);  add_11897 = None
	        clamp_max_155: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_233, 127);  clamp_min_233 = None
	        view_1211: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_155, [sym_size_int, 1500, 5120]);  clamp_max_155 = None
	        _assert_tensor_metadata_697 = torch.ops.aten._assert_tensor_metadata.default(view_1211, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_697 = None
	        convert_element_type_463: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1211, torch.int8);  view_1211 = None
	        view_1212: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_463, [sym_size_int, 1500, 5120]);  convert_element_type_463 = None
	        view_1213: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_231, [sym_size_int, 1500, 1]);  clamp_min_231 = None
	        view_1214: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_462, [sym_size_int, 1500, 1]);  convert_element_type_462 = None
	        _assert_tensor_metadata_698 = torch.ops.aten._assert_tensor_metadata.default(view_1212, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_698 = None
	        convert_element_type_464: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1212, torch.float32);  view_1212 = None
	        _assert_tensor_metadata_699 = torch.ops.aten._assert_tensor_metadata.default(view_1214, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_699 = None
	        convert_element_type_465: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1214, torch.float32);  view_1214 = None
	        sub_3559: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_464, convert_element_type_465);  convert_element_type_464 = convert_element_type_465 = None
	        mul_7539: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3559, view_1213);  sub_3559 = view_1213 = None
	        view_1215: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_7539, [sym_size_int, 1500, 5120]);  mul_7539 = None
	        _assert_tensor_metadata_700 = torch.ops.aten._assert_tensor_metadata.default(view_1215, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_700 = None
	        view_1216: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_12_fc2_parametrizations_weight_original0 = None
	        view_1217: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_12_fc2_parametrizations_weight_original1 = None
	        view_1218: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_12_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_701 = torch.ops.aten._assert_tensor_metadata.default(view_1216, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_701 = None
	        convert_element_type_466: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1216, torch.float32);  view_1216 = None
	        _assert_tensor_metadata_702 = torch.ops.aten._assert_tensor_metadata.default(view_1218, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_702 = None
	        convert_element_type_467: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1218, torch.float32);  view_1218 = None
	        sub_3563: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_466, convert_element_type_467);  convert_element_type_466 = convert_element_type_467 = None
	        mul_7544: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3563, view_1217);  sub_3563 = view_1217 = None
	        view_1219: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_7544, [1280, 5120]);  mul_7544 = None
	        _assert_tensor_metadata_703 = torch.ops.aten._assert_tensor_metadata.default(view_1219, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_703 = None
	        mul_7549: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1220: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1215, [mul_7549, 5120]);  view_1215 = mul_7549 = None
	        permute_130: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1219, [1, 0]);  view_1219 = None
	        addmm_64: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_12_fc2_bias, view_1220, permute_130);  model_audio_tower_layers_12_fc2_bias = view_1220 = permute_130 = None
	        view_1221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_64, [sym_size_int, 1500, 1280]);  addmm_64 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_104: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1221);  view_1221 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_11960: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_11662, clone_104);  add_11662 = clone_104 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_105: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_11960, memory_format = torch.contiguous_format)
	        var_mean_26 = torch.ops.aten.var_mean.correction(clone_105, [2], correction = 0, keepdim = True)
	        getitem_104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_26[0]
	        getitem_105: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_26[1];  var_mean_26 = None
	        add_11965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_104, 1e-05);  getitem_104 = None
	        rsqrt_26: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_11965);  add_11965 = None
	        sub_3569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_105, getitem_105);  clone_105 = getitem_105 = None
	        mul_7560: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3569, rsqrt_26);  sub_3569 = rsqrt_26 = None
	        mul_7561: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7560, model_audio_tower_layers_13_self_attn_layer_norm_weight);  mul_7560 = model_audio_tower_layers_13_self_attn_layer_norm_weight = None
	        add_11966: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_7561, model_audio_tower_layers_13_self_attn_layer_norm_bias);  mul_7561 = model_audio_tower_layers_13_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1222: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11966, [sym_size_int, 1500, 1280])
	        amin_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1222, [2])
	        amax_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1222, [2]);  view_1222 = None
	        full_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_78, full_156);  amin_78 = full_156 = None
	        full_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_78, full_157);  amax_78 = full_157 = None
	        sub_3580: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_78, minimum_78);  maximum_78 = None
	        div_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3580, 255.0);  sub_3580 = None
	        clamp_min_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_156, 1.1920928955078125e-07);  div_156 = None
	        div_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_78, clamp_min_234);  minimum_78 = None
	        round_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_157);  div_157 = None
	        sub_3586: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_157);  round_157 = None
	        clamp_min_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3586, -128);  sub_3586 = None
	        clamp_max_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_235, 127);  clamp_min_235 = None
	        _assert_tensor_metadata_704 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_234, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_704 = None
	        _assert_tensor_metadata_705 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_156, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_705 = None
	        convert_element_type_468: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_156, torch.int8);  clamp_max_156 = None
	        view_1223: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11966, [sym_size_int, 1500, 1280])
	        view_1224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_234, [sym_size_int, 1500, 1])
	        view_1225: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_468, [sym_size_int, 1500, 1])
	        reciprocal_78: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1224);  view_1224 = None
	        mul_7609: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_78, 1.0);  reciprocal_78 = None
	        mul_7612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1223, mul_7609);  view_1223 = mul_7609 = None
	        round_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7612);  mul_7612 = None
	        add_12053: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_158, view_1225);  round_158 = view_1225 = None
	        clamp_min_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12053, -128);  add_12053 = None
	        clamp_max_157: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_236, 127);  clamp_min_236 = None
	        view_1226: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_157, [sym_size_int, 1500, 1280]);  clamp_max_157 = None
	        _assert_tensor_metadata_706 = torch.ops.aten._assert_tensor_metadata.default(view_1226, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_706 = None
	        convert_element_type_469: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1226, torch.int8);  view_1226 = None
	        view_1227: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_469, [sym_size_int, 1500, 1280]);  convert_element_type_469 = None
	        view_1228: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_234, [sym_size_int, 1500, 1]);  clamp_min_234 = None
	        view_1229: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_468, [sym_size_int, 1500, 1]);  convert_element_type_468 = None
	        _assert_tensor_metadata_707 = torch.ops.aten._assert_tensor_metadata.default(view_1227, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_707 = None
	        convert_element_type_470: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1227, torch.float32);  view_1227 = None
	        _assert_tensor_metadata_708 = torch.ops.aten._assert_tensor_metadata.default(view_1229, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_708 = None
	        convert_element_type_471: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1229, torch.float32);  view_1229 = None
	        sub_3606: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_470, convert_element_type_471);  convert_element_type_470 = convert_element_type_471 = None
	        mul_7634: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3606, view_1228);  sub_3606 = view_1228 = None
	        view_1230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7634, [sym_size_int, 1500, 1280]);  mul_7634 = None
	        _assert_tensor_metadata_709 = torch.ops.aten._assert_tensor_metadata.default(view_1230, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_709 = None
	        view_1231: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1232: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1233: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_710 = torch.ops.aten._assert_tensor_metadata.default(view_1231, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_710 = None
	        convert_element_type_472: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1231, torch.float32);  view_1231 = None
	        _assert_tensor_metadata_711 = torch.ops.aten._assert_tensor_metadata.default(view_1233, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_711 = None
	        convert_element_type_473: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1233, torch.float32);  view_1233 = None
	        sub_3610: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_472, convert_element_type_473);  convert_element_type_472 = convert_element_type_473 = None
	        mul_7639: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3610, view_1232);  sub_3610 = view_1232 = None
	        view_1234: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7639, [1280, 1280]);  mul_7639 = None
	        _assert_tensor_metadata_712 = torch.ops.aten._assert_tensor_metadata.default(view_1234, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_712 = None
	        mul_7644: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1235: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1230, [mul_7644, 1280]);  view_1230 = mul_7644 = None
	        permute_131: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1234, [1, 0]);  view_1234 = None
	        addmm_65: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_13_self_attn_q_proj_bias, view_1235, permute_131);  model_audio_tower_layers_13_self_attn_q_proj_bias = view_1235 = permute_131 = None
	        view_1236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_65, [sym_size_int, 1500, 1280]);  addmm_65 = None
	        mul_7651: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1236, 0.125);  view_1236 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1237: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_7651, [sym_size_int, 1500, 20, 64]);  mul_7651 = None
	        permute_132: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1237, [0, 2, 1, 3]);  view_1237 = None
	        clone_106: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_132, memory_format = torch.contiguous_format);  permute_132 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1238: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11966, [sym_size_int, 1500, 1280])
	        amin_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1238, [2])
	        amax_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1238, [2]);  view_1238 = None
	        full_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_79, full_158);  amin_79 = full_158 = None
	        full_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_79, full_159);  amax_79 = full_159 = None
	        sub_3625: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_79, minimum_79);  maximum_79 = None
	        div_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3625, 255.0);  sub_3625 = None
	        clamp_min_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_158, 1.1920928955078125e-07);  div_158 = None
	        div_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_79, clamp_min_237);  minimum_79 = None
	        round_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_159);  div_159 = None
	        sub_3631: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_159);  round_159 = None
	        clamp_min_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3631, -128);  sub_3631 = None
	        clamp_max_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_238, 127);  clamp_min_238 = None
	        _assert_tensor_metadata_713 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_237, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_713 = None
	        _assert_tensor_metadata_714 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_158, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_714 = None
	        convert_element_type_474: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_158, torch.int8);  clamp_max_158 = None
	        view_1239: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11966, [sym_size_int, 1500, 1280])
	        view_1240: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_237, [sym_size_int, 1500, 1])
	        view_1241: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_474, [sym_size_int, 1500, 1])
	        reciprocal_79: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1240);  view_1240 = None
	        mul_7705: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_79, 1.0);  reciprocal_79 = None
	        mul_7708: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1239, mul_7705);  view_1239 = mul_7705 = None
	        round_160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7708);  mul_7708 = None
	        add_12205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_160, view_1241);  round_160 = view_1241 = None
	        clamp_min_239: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12205, -128);  add_12205 = None
	        clamp_max_159: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_239, 127);  clamp_min_239 = None
	        view_1242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_159, [sym_size_int, 1500, 1280]);  clamp_max_159 = None
	        _assert_tensor_metadata_715 = torch.ops.aten._assert_tensor_metadata.default(view_1242, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_715 = None
	        convert_element_type_475: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1242, torch.int8);  view_1242 = None
	        view_1243: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_475, [sym_size_int, 1500, 1280]);  convert_element_type_475 = None
	        view_1244: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_237, [sym_size_int, 1500, 1]);  clamp_min_237 = None
	        view_1245: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_474, [sym_size_int, 1500, 1]);  convert_element_type_474 = None
	        _assert_tensor_metadata_716 = torch.ops.aten._assert_tensor_metadata.default(view_1243, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_716 = None
	        convert_element_type_476: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1243, torch.float32);  view_1243 = None
	        _assert_tensor_metadata_717 = torch.ops.aten._assert_tensor_metadata.default(view_1245, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_717 = None
	        convert_element_type_477: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1245, torch.float32);  view_1245 = None
	        sub_3651: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_476, convert_element_type_477);  convert_element_type_476 = convert_element_type_477 = None
	        mul_7730: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3651, view_1244);  sub_3651 = view_1244 = None
	        view_1246: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7730, [sym_size_int, 1500, 1280]);  mul_7730 = None
	        _assert_tensor_metadata_718 = torch.ops.aten._assert_tensor_metadata.default(view_1246, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_718 = None
	        view_1247: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1248: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1249: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_719 = torch.ops.aten._assert_tensor_metadata.default(view_1247, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_719 = None
	        convert_element_type_478: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1247, torch.float32);  view_1247 = None
	        _assert_tensor_metadata_720 = torch.ops.aten._assert_tensor_metadata.default(view_1249, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_720 = None
	        convert_element_type_479: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1249, torch.float32);  view_1249 = None
	        sub_3655: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_478, convert_element_type_479);  convert_element_type_478 = convert_element_type_479 = None
	        mul_7735: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3655, view_1248);  sub_3655 = view_1248 = None
	        view_1250: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7735, [1280, 1280]);  mul_7735 = None
	        _assert_tensor_metadata_721 = torch.ops.aten._assert_tensor_metadata.default(view_1250, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_721 = None
	        permute_133: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1250, [1, 0]);  view_1250 = None
	        mul_7738: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1251: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1246, [mul_7738, 1280]);  view_1246 = mul_7738 = None
	        mm_13: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1251, permute_133);  view_1251 = permute_133 = None
	        view_1252: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_13, [sym_size_int, 1500, 1280]);  mm_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1253: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1252, [sym_size_int, -1, 20, 64]);  view_1252 = None
	        permute_134: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1253, [0, 2, 1, 3]);  view_1253 = None
	        clone_107: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_134, memory_format = torch.contiguous_format);  permute_134 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11966, [sym_size_int, 1500, 1280])
	        amin_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1254, [2])
	        amax_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1254, [2]);  view_1254 = None
	        full_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_80, full_160);  amin_80 = full_160 = None
	        full_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_80, full_161);  amax_80 = full_161 = None
	        sub_3669: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_80, minimum_80);  maximum_80 = None
	        div_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3669, 255.0);  sub_3669 = None
	        clamp_min_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_160, 1.1920928955078125e-07);  div_160 = None
	        div_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_80, clamp_min_240);  minimum_80 = None
	        round_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_161);  div_161 = None
	        sub_3675: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_161);  round_161 = None
	        clamp_min_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3675, -128);  sub_3675 = None
	        clamp_max_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_241, 127);  clamp_min_241 = None
	        _assert_tensor_metadata_722 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_240, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_722 = None
	        _assert_tensor_metadata_723 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_160, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_723 = None
	        convert_element_type_480: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_160, torch.int8);  clamp_max_160 = None
	        view_1255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_11966, [sym_size_int, 1500, 1280]);  add_11966 = None
	        view_1256: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_240, [sym_size_int, 1500, 1])
	        view_1257: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_480, [sym_size_int, 1500, 1])
	        reciprocal_80: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1256);  view_1256 = None
	        mul_7804: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_80, 1.0);  reciprocal_80 = None
	        mul_7807: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1255, mul_7804);  view_1255 = mul_7804 = None
	        round_162: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7807);  mul_7807 = None
	        add_12353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_162, view_1257);  round_162 = view_1257 = None
	        clamp_min_242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12353, -128);  add_12353 = None
	        clamp_max_161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_242, 127);  clamp_min_242 = None
	        view_1258: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_161, [sym_size_int, 1500, 1280]);  clamp_max_161 = None
	        _assert_tensor_metadata_724 = torch.ops.aten._assert_tensor_metadata.default(view_1258, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_724 = None
	        convert_element_type_481: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1258, torch.int8);  view_1258 = None
	        view_1259: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_481, [sym_size_int, 1500, 1280]);  convert_element_type_481 = None
	        view_1260: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_240, [sym_size_int, 1500, 1]);  clamp_min_240 = None
	        view_1261: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_480, [sym_size_int, 1500, 1]);  convert_element_type_480 = None
	        _assert_tensor_metadata_725 = torch.ops.aten._assert_tensor_metadata.default(view_1259, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_725 = None
	        convert_element_type_482: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1259, torch.float32);  view_1259 = None
	        _assert_tensor_metadata_726 = torch.ops.aten._assert_tensor_metadata.default(view_1261, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_726 = None
	        convert_element_type_483: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1261, torch.float32);  view_1261 = None
	        sub_3695: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_482, convert_element_type_483);  convert_element_type_482 = convert_element_type_483 = None
	        mul_7829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3695, view_1260);  sub_3695 = view_1260 = None
	        view_1262: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7829, [sym_size_int, 1500, 1280]);  mul_7829 = None
	        _assert_tensor_metadata_727 = torch.ops.aten._assert_tensor_metadata.default(view_1262, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_727 = None
	        view_1263: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1264: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1265: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_728 = torch.ops.aten._assert_tensor_metadata.default(view_1263, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_728 = None
	        convert_element_type_484: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1263, torch.float32);  view_1263 = None
	        _assert_tensor_metadata_729 = torch.ops.aten._assert_tensor_metadata.default(view_1265, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_729 = None
	        convert_element_type_485: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1265, torch.float32);  view_1265 = None
	        sub_3699: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_484, convert_element_type_485);  convert_element_type_484 = convert_element_type_485 = None
	        mul_7834: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3699, view_1264);  sub_3699 = view_1264 = None
	        view_1266: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7834, [1280, 1280]);  mul_7834 = None
	        _assert_tensor_metadata_730 = torch.ops.aten._assert_tensor_metadata.default(view_1266, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_730 = None
	        mul_7839: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1267: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1262, [mul_7839, 1280]);  view_1262 = mul_7839 = None
	        permute_135: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1266, [1, 0]);  view_1266 = None
	        addmm_66: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_13_self_attn_v_proj_bias, view_1267, permute_135);  model_audio_tower_layers_13_self_attn_v_proj_bias = view_1267 = permute_135 = None
	        view_1268: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_66, [sym_size_int, 1500, 1280]);  addmm_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1269: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1268, [sym_size_int, -1, 20, 64]);  view_1268 = None
	        permute_136: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1269, [0, 2, 1, 3]);  view_1269 = None
	        clone_108: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_136, memory_format = torch.contiguous_format);  permute_136 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_13 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_106, clone_107, clone_108, None, False, scale = 1.0);  clone_106 = clone_107 = clone_108 = None
	        getitem_106: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_13[0];  _scaled_dot_product_efficient_attention_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_137: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_106, [0, 2, 1, 3]);  getitem_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_137, [sym_size_int, 1500, -1]);  permute_137 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1270, [sym_size_int, 1500, 1280])
	        amin_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1271, [2])
	        amax_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1271, [2]);  view_1271 = None
	        full_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_81, full_162);  amin_81 = full_162 = None
	        full_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_81, full_163);  amax_81 = full_163 = None
	        sub_3717: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_81, minimum_81);  maximum_81 = None
	        div_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3717, 255.0);  sub_3717 = None
	        clamp_min_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_162, 1.1920928955078125e-07);  div_162 = None
	        div_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_81, clamp_min_243);  minimum_81 = None
	        round_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_163);  div_163 = None
	        sub_3723: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_163);  round_163 = None
	        clamp_min_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3723, -128);  sub_3723 = None
	        clamp_max_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_244, 127);  clamp_min_244 = None
	        _assert_tensor_metadata_731 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_243, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_731 = None
	        _assert_tensor_metadata_732 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_162, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_732 = None
	        convert_element_type_486: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_162, torch.int8);  clamp_max_162 = None
	        view_1272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1270, [sym_size_int, 1500, 1280]);  view_1270 = None
	        view_1273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_243, [sym_size_int, 1500, 1])
	        view_1274: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_486, [sym_size_int, 1500, 1])
	        reciprocal_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1273);  view_1273 = None
	        mul_7909: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_81, 1.0);  reciprocal_81 = None
	        mul_7912: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1272, mul_7909);  view_1272 = mul_7909 = None
	        round_164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7912);  mul_7912 = None
	        add_12517: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_164, view_1274);  round_164 = view_1274 = None
	        clamp_min_245: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12517, -128);  add_12517 = None
	        clamp_max_163: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_245, 127);  clamp_min_245 = None
	        view_1275: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_163, [sym_size_int, 1500, 1280]);  clamp_max_163 = None
	        _assert_tensor_metadata_733 = torch.ops.aten._assert_tensor_metadata.default(view_1275, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_733 = None
	        convert_element_type_487: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1275, torch.int8);  view_1275 = None
	        view_1276: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_487, [sym_size_int, 1500, 1280]);  convert_element_type_487 = None
	        view_1277: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_243, [sym_size_int, 1500, 1]);  clamp_min_243 = None
	        view_1278: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_486, [sym_size_int, 1500, 1]);  convert_element_type_486 = None
	        _assert_tensor_metadata_734 = torch.ops.aten._assert_tensor_metadata.default(view_1276, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_734 = None
	        convert_element_type_488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1276, torch.float32);  view_1276 = None
	        _assert_tensor_metadata_735 = torch.ops.aten._assert_tensor_metadata.default(view_1278, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_735 = None
	        convert_element_type_489: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1278, torch.float32);  view_1278 = None
	        sub_3743: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_488, convert_element_type_489);  convert_element_type_488 = convert_element_type_489 = None
	        mul_7934: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3743, view_1277);  sub_3743 = view_1277 = None
	        view_1279: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7934, [sym_size_int, 1500, 1280]);  mul_7934 = None
	        _assert_tensor_metadata_736 = torch.ops.aten._assert_tensor_metadata.default(view_1279, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_736 = None
	        view_1280: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1281: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1282: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_737 = torch.ops.aten._assert_tensor_metadata.default(view_1280, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_737 = None
	        convert_element_type_490: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1280, torch.float32);  view_1280 = None
	        _assert_tensor_metadata_738 = torch.ops.aten._assert_tensor_metadata.default(view_1282, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_738 = None
	        convert_element_type_491: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1282, torch.float32);  view_1282 = None
	        sub_3747: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_490, convert_element_type_491);  convert_element_type_490 = convert_element_type_491 = None
	        mul_7939: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3747, view_1281);  sub_3747 = view_1281 = None
	        view_1283: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7939, [1280, 1280]);  mul_7939 = None
	        _assert_tensor_metadata_739 = torch.ops.aten._assert_tensor_metadata.default(view_1283, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_739 = None
	        mul_7944: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1284: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1279, [mul_7944, 1280]);  view_1279 = mul_7944 = None
	        permute_138: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1283, [1, 0]);  view_1283 = None
	        addmm_67: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_13_self_attn_out_proj_bias, view_1284, permute_138);  model_audio_tower_layers_13_self_attn_out_proj_bias = view_1284 = permute_138 = None
	        view_1285: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_67, [sym_size_int, 1500, 1280]);  addmm_67 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_109: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1285);  view_1285 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_12580: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_11960, clone_109);  add_11960 = clone_109 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_12580, memory_format = torch.contiguous_format)
	        var_mean_27 = torch.ops.aten.var_mean.correction(clone_110, [2], correction = 0, keepdim = True)
	        getitem_110: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_27[0]
	        getitem_111: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_27[1];  var_mean_27 = None
	        add_12585: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_110, 1e-05);  getitem_110 = None
	        rsqrt_27: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_12585);  add_12585 = None
	        sub_3753: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_110, getitem_111);  clone_110 = getitem_111 = None
	        mul_7955: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3753, rsqrt_27);  sub_3753 = rsqrt_27 = None
	        mul_7956: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7955, model_audio_tower_layers_13_final_layer_norm_weight);  mul_7955 = model_audio_tower_layers_13_final_layer_norm_weight = None
	        add_12586: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_7956, model_audio_tower_layers_13_final_layer_norm_bias);  mul_7956 = model_audio_tower_layers_13_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1286: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_12586, [sym_size_int, 1500, 1280])
	        amin_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1286, [2])
	        amax_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1286, [2]);  view_1286 = None
	        full_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_82, full_164);  amin_82 = full_164 = None
	        full_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_82, full_165);  amax_82 = full_165 = None
	        sub_3764: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_82, minimum_82);  maximum_82 = None
	        div_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3764, 255.0);  sub_3764 = None
	        clamp_min_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_164, 1.1920928955078125e-07);  div_164 = None
	        div_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_82, clamp_min_246);  minimum_82 = None
	        round_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_165);  div_165 = None
	        sub_3770: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_165);  round_165 = None
	        clamp_min_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3770, -128);  sub_3770 = None
	        clamp_max_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_247, 127);  clamp_min_247 = None
	        _assert_tensor_metadata_740 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_246, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_740 = None
	        _assert_tensor_metadata_741 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_164, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_741 = None
	        convert_element_type_492: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_164, torch.int8);  clamp_max_164 = None
	        view_1287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_12586, [sym_size_int, 1500, 1280]);  add_12586 = None
	        view_1288: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_246, [sym_size_int, 1500, 1])
	        view_1289: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_492, [sym_size_int, 1500, 1])
	        reciprocal_82: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1288);  view_1288 = None
	        mul_8004: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_82, 1.0);  reciprocal_82 = None
	        mul_8007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1287, mul_8004);  view_1287 = mul_8004 = None
	        round_166: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8007);  mul_8007 = None
	        add_12673: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_166, view_1289);  round_166 = view_1289 = None
	        clamp_min_248: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12673, -128);  add_12673 = None
	        clamp_max_165: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_248, 127);  clamp_min_248 = None
	        view_1290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_165, [sym_size_int, 1500, 1280]);  clamp_max_165 = None
	        _assert_tensor_metadata_742 = torch.ops.aten._assert_tensor_metadata.default(view_1290, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_742 = None
	        convert_element_type_493: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1290, torch.int8);  view_1290 = None
	        view_1291: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_493, [sym_size_int, 1500, 1280]);  convert_element_type_493 = None
	        view_1292: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_246, [sym_size_int, 1500, 1]);  clamp_min_246 = None
	        view_1293: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_492, [sym_size_int, 1500, 1]);  convert_element_type_492 = None
	        _assert_tensor_metadata_743 = torch.ops.aten._assert_tensor_metadata.default(view_1291, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_743 = None
	        convert_element_type_494: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1291, torch.float32);  view_1291 = None
	        _assert_tensor_metadata_744 = torch.ops.aten._assert_tensor_metadata.default(view_1293, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_744 = None
	        convert_element_type_495: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1293, torch.float32);  view_1293 = None
	        sub_3790: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_494, convert_element_type_495);  convert_element_type_494 = convert_element_type_495 = None
	        mul_8029: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3790, view_1292);  sub_3790 = view_1292 = None
	        view_1294: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8029, [sym_size_int, 1500, 1280]);  mul_8029 = None
	        _assert_tensor_metadata_745 = torch.ops.aten._assert_tensor_metadata.default(view_1294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_745 = None
	        view_1295: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_13_fc1_parametrizations_weight_original0 = None
	        view_1296: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_13_fc1_parametrizations_weight_original1 = None
	        view_1297: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_13_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_746 = torch.ops.aten._assert_tensor_metadata.default(view_1295, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_746 = None
	        convert_element_type_496: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1295, torch.float32);  view_1295 = None
	        _assert_tensor_metadata_747 = torch.ops.aten._assert_tensor_metadata.default(view_1297, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_747 = None
	        convert_element_type_497: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1297, torch.float32);  view_1297 = None
	        sub_3794: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_496, convert_element_type_497);  convert_element_type_496 = convert_element_type_497 = None
	        mul_8034: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3794, view_1296);  sub_3794 = view_1296 = None
	        view_1298: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8034, [5120, 1280]);  mul_8034 = None
	        _assert_tensor_metadata_748 = torch.ops.aten._assert_tensor_metadata.default(view_1298, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_748 = None
	        mul_8039: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1299: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1294, [mul_8039, 1280]);  view_1294 = mul_8039 = None
	        permute_139: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1298, [1, 0]);  view_1298 = None
	        addmm_68: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_13_fc1_bias, view_1299, permute_139);  model_audio_tower_layers_13_fc1_bias = view_1299 = permute_139 = None
	        view_1300: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_68, [sym_size_int, 1500, 5120]);  addmm_68 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_8046: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1300, 0.5)
	        mul_8047: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1300, 0.7071067811865476);  view_1300 = None
	        erf_15: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_8047);  mul_8047 = None
	        add_12732: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_15, 1);  erf_15 = None
	        mul_8048: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8046, add_12732);  mul_8046 = add_12732 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_111: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_8048);  mul_8048 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1301: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_111, [sym_size_int, 1500, 5120])
	        amin_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1301, [2])
	        amax_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1301, [2]);  view_1301 = None
	        full_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_83, full_166);  amin_83 = full_166 = None
	        full_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_83, full_167);  amax_83 = full_167 = None
	        sub_3807: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_83, minimum_83);  maximum_83 = None
	        div_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3807, 255.0);  sub_3807 = None
	        clamp_min_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_166, 1.1920928955078125e-07);  div_166 = None
	        div_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_83, clamp_min_249);  minimum_83 = None
	        round_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_167);  div_167 = None
	        sub_3813: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_167);  round_167 = None
	        clamp_min_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3813, -128);  sub_3813 = None
	        clamp_max_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_250, 127);  clamp_min_250 = None
	        _assert_tensor_metadata_749 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_749 = None
	        _assert_tensor_metadata_750 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_166, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_750 = None
	        convert_element_type_498: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_166, torch.int8);  clamp_max_166 = None
	        view_1302: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_111, [sym_size_int, 1500, 5120]);  clone_111 = None
	        view_1303: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_249, [sym_size_int, 1500, 1])
	        view_1304: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_498, [sym_size_int, 1500, 1])
	        reciprocal_83: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1303);  view_1303 = None
	        mul_8094: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_83, 1.0);  reciprocal_83 = None
	        mul_8097: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1302, mul_8094);  view_1302 = mul_8094 = None
	        round_168: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_8097);  mul_8097 = None
	        add_12815: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_168, view_1304);  round_168 = view_1304 = None
	        clamp_min_251: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12815, -128);  add_12815 = None
	        clamp_max_167: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_251, 127);  clamp_min_251 = None
	        view_1305: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_167, [sym_size_int, 1500, 5120]);  clamp_max_167 = None
	        _assert_tensor_metadata_751 = torch.ops.aten._assert_tensor_metadata.default(view_1305, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_751 = None
	        convert_element_type_499: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1305, torch.int8);  view_1305 = None
	        view_1306: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_499, [sym_size_int, 1500, 5120]);  convert_element_type_499 = None
	        view_1307: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_249, [sym_size_int, 1500, 1]);  clamp_min_249 = None
	        view_1308: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_498, [sym_size_int, 1500, 1]);  convert_element_type_498 = None
	        _assert_tensor_metadata_752 = torch.ops.aten._assert_tensor_metadata.default(view_1306, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_752 = None
	        convert_element_type_500: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1306, torch.float32);  view_1306 = None
	        _assert_tensor_metadata_753 = torch.ops.aten._assert_tensor_metadata.default(view_1308, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_753 = None
	        convert_element_type_501: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1308, torch.float32);  view_1308 = None
	        sub_3833: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_500, convert_element_type_501);  convert_element_type_500 = convert_element_type_501 = None
	        mul_8119: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3833, view_1307);  sub_3833 = view_1307 = None
	        view_1309: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_8119, [sym_size_int, 1500, 5120]);  mul_8119 = None
	        _assert_tensor_metadata_754 = torch.ops.aten._assert_tensor_metadata.default(view_1309, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_754 = None
	        view_1310: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_13_fc2_parametrizations_weight_original0 = None
	        view_1311: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_13_fc2_parametrizations_weight_original1 = None
	        view_1312: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_13_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_755 = torch.ops.aten._assert_tensor_metadata.default(view_1310, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_755 = None
	        convert_element_type_502: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1310, torch.float32);  view_1310 = None
	        _assert_tensor_metadata_756 = torch.ops.aten._assert_tensor_metadata.default(view_1312, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_756 = None
	        convert_element_type_503: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1312, torch.float32);  view_1312 = None
	        sub_3837: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_502, convert_element_type_503);  convert_element_type_502 = convert_element_type_503 = None
	        mul_8124: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3837, view_1311);  sub_3837 = view_1311 = None
	        view_1313: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_8124, [1280, 5120]);  mul_8124 = None
	        _assert_tensor_metadata_757 = torch.ops.aten._assert_tensor_metadata.default(view_1313, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_757 = None
	        mul_8129: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1314: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1309, [mul_8129, 5120]);  view_1309 = mul_8129 = None
	        permute_140: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1313, [1, 0]);  view_1313 = None
	        addmm_69: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_13_fc2_bias, view_1314, permute_140);  model_audio_tower_layers_13_fc2_bias = view_1314 = permute_140 = None
	        view_1315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_69, [sym_size_int, 1500, 1280]);  addmm_69 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_112: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1315);  view_1315 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_12878: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_12580, clone_112);  add_12580 = clone_112 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_12878, memory_format = torch.contiguous_format)
	        var_mean_28 = torch.ops.aten.var_mean.correction(clone_113, [2], correction = 0, keepdim = True)
	        getitem_112: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_28[0]
	        getitem_113: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_28[1];  var_mean_28 = None
	        add_12883: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_112, 1e-05);  getitem_112 = None
	        rsqrt_28: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_12883);  add_12883 = None
	        sub_3843: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_113, getitem_113);  clone_113 = getitem_113 = None
	        mul_8140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3843, rsqrt_28);  sub_3843 = rsqrt_28 = None
	        mul_8141: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8140, model_audio_tower_layers_14_self_attn_layer_norm_weight);  mul_8140 = model_audio_tower_layers_14_self_attn_layer_norm_weight = None
	        add_12884: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_8141, model_audio_tower_layers_14_self_attn_layer_norm_bias);  mul_8141 = model_audio_tower_layers_14_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1316: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_12884, [sym_size_int, 1500, 1280])
	        amin_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1316, [2])
	        amax_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1316, [2]);  view_1316 = None
	        full_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_84, full_168);  amin_84 = full_168 = None
	        full_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_84, full_169);  amax_84 = full_169 = None
	        sub_3854: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_84, minimum_84);  maximum_84 = None
	        div_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3854, 255.0);  sub_3854 = None
	        clamp_min_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_168, 1.1920928955078125e-07);  div_168 = None
	        div_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_84, clamp_min_252);  minimum_84 = None
	        round_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_169);  div_169 = None
	        sub_3860: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_169);  round_169 = None
	        clamp_min_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3860, -128);  sub_3860 = None
	        clamp_max_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_253, 127);  clamp_min_253 = None
	        _assert_tensor_metadata_758 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_252, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_758 = None
	        _assert_tensor_metadata_759 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_168, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_759 = None
	        convert_element_type_504: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_168, torch.int8);  clamp_max_168 = None
	        view_1317: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_12884, [sym_size_int, 1500, 1280])
	        view_1318: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_252, [sym_size_int, 1500, 1])
	        view_1319: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_504, [sym_size_int, 1500, 1])
	        reciprocal_84: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1318);  view_1318 = None
	        mul_8189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_84, 1.0);  reciprocal_84 = None
	        mul_8192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1317, mul_8189);  view_1317 = mul_8189 = None
	        round_170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8192);  mul_8192 = None
	        add_12971: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_170, view_1319);  round_170 = view_1319 = None
	        clamp_min_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12971, -128);  add_12971 = None
	        clamp_max_169: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_254, 127);  clamp_min_254 = None
	        view_1320: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_169, [sym_size_int, 1500, 1280]);  clamp_max_169 = None
	        _assert_tensor_metadata_760 = torch.ops.aten._assert_tensor_metadata.default(view_1320, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_760 = None
	        convert_element_type_505: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1320, torch.int8);  view_1320 = None
	        view_1321: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_505, [sym_size_int, 1500, 1280]);  convert_element_type_505 = None
	        view_1322: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_252, [sym_size_int, 1500, 1]);  clamp_min_252 = None
	        view_1323: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_504, [sym_size_int, 1500, 1]);  convert_element_type_504 = None
	        _assert_tensor_metadata_761 = torch.ops.aten._assert_tensor_metadata.default(view_1321, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_761 = None
	        convert_element_type_506: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1321, torch.float32);  view_1321 = None
	        _assert_tensor_metadata_762 = torch.ops.aten._assert_tensor_metadata.default(view_1323, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_762 = None
	        convert_element_type_507: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1323, torch.float32);  view_1323 = None
	        sub_3880: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_506, convert_element_type_507);  convert_element_type_506 = convert_element_type_507 = None
	        mul_8214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3880, view_1322);  sub_3880 = view_1322 = None
	        view_1324: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8214, [sym_size_int, 1500, 1280]);  mul_8214 = None
	        _assert_tensor_metadata_763 = torch.ops.aten._assert_tensor_metadata.default(view_1324, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_763 = None
	        view_1325: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1326: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1327: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_764 = torch.ops.aten._assert_tensor_metadata.default(view_1325, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_764 = None
	        convert_element_type_508: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1325, torch.float32);  view_1325 = None
	        _assert_tensor_metadata_765 = torch.ops.aten._assert_tensor_metadata.default(view_1327, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_765 = None
	        convert_element_type_509: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1327, torch.float32);  view_1327 = None
	        sub_3884: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_508, convert_element_type_509);  convert_element_type_508 = convert_element_type_509 = None
	        mul_8219: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3884, view_1326);  sub_3884 = view_1326 = None
	        view_1328: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8219, [1280, 1280]);  mul_8219 = None
	        _assert_tensor_metadata_766 = torch.ops.aten._assert_tensor_metadata.default(view_1328, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_766 = None
	        mul_8224: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1329: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1324, [mul_8224, 1280]);  view_1324 = mul_8224 = None
	        permute_141: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1328, [1, 0]);  view_1328 = None
	        addmm_70: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_14_self_attn_q_proj_bias, view_1329, permute_141);  model_audio_tower_layers_14_self_attn_q_proj_bias = view_1329 = permute_141 = None
	        view_1330: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_70, [sym_size_int, 1500, 1280]);  addmm_70 = None
	        mul_8231: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1330, 0.125);  view_1330 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1331: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_8231, [sym_size_int, 1500, 20, 64]);  mul_8231 = None
	        permute_142: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1331, [0, 2, 1, 3]);  view_1331 = None
	        clone_114: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_142, memory_format = torch.contiguous_format);  permute_142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_12884, [sym_size_int, 1500, 1280])
	        amin_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1332, [2])
	        amax_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1332, [2]);  view_1332 = None
	        full_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_85, full_170);  amin_85 = full_170 = None
	        full_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_85, full_171);  amax_85 = full_171 = None
	        sub_3899: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_85, minimum_85);  maximum_85 = None
	        div_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3899, 255.0);  sub_3899 = None
	        clamp_min_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_170, 1.1920928955078125e-07);  div_170 = None
	        div_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_85, clamp_min_255);  minimum_85 = None
	        round_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_171);  div_171 = None
	        sub_3905: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_171);  round_171 = None
	        clamp_min_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3905, -128);  sub_3905 = None
	        clamp_max_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_256, 127);  clamp_min_256 = None
	        _assert_tensor_metadata_767 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_255, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_767 = None
	        _assert_tensor_metadata_768 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_170, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_768 = None
	        convert_element_type_510: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_170, torch.int8);  clamp_max_170 = None
	        view_1333: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_12884, [sym_size_int, 1500, 1280])
	        view_1334: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_255, [sym_size_int, 1500, 1])
	        view_1335: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_510, [sym_size_int, 1500, 1])
	        reciprocal_85: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1334);  view_1334 = None
	        mul_8285: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_85, 1.0);  reciprocal_85 = None
	        mul_8288: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1333, mul_8285);  view_1333 = mul_8285 = None
	        round_172: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8288);  mul_8288 = None
	        add_13123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_172, view_1335);  round_172 = view_1335 = None
	        clamp_min_257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13123, -128);  add_13123 = None
	        clamp_max_171: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_257, 127);  clamp_min_257 = None
	        view_1336: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_171, [sym_size_int, 1500, 1280]);  clamp_max_171 = None
	        _assert_tensor_metadata_769 = torch.ops.aten._assert_tensor_metadata.default(view_1336, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_769 = None
	        convert_element_type_511: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1336, torch.int8);  view_1336 = None
	        view_1337: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_511, [sym_size_int, 1500, 1280]);  convert_element_type_511 = None
	        view_1338: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_255, [sym_size_int, 1500, 1]);  clamp_min_255 = None
	        view_1339: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_510, [sym_size_int, 1500, 1]);  convert_element_type_510 = None
	        _assert_tensor_metadata_770 = torch.ops.aten._assert_tensor_metadata.default(view_1337, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_770 = None
	        convert_element_type_512: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1337, torch.float32);  view_1337 = None
	        _assert_tensor_metadata_771 = torch.ops.aten._assert_tensor_metadata.default(view_1339, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_771 = None
	        convert_element_type_513: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1339, torch.float32);  view_1339 = None
	        sub_3925: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_512, convert_element_type_513);  convert_element_type_512 = convert_element_type_513 = None
	        mul_8310: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3925, view_1338);  sub_3925 = view_1338 = None
	        view_1340: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8310, [sym_size_int, 1500, 1280]);  mul_8310 = None
	        _assert_tensor_metadata_772 = torch.ops.aten._assert_tensor_metadata.default(view_1340, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_772 = None
	        view_1341: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1342: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1343: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_773 = torch.ops.aten._assert_tensor_metadata.default(view_1341, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_773 = None
	        convert_element_type_514: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1341, torch.float32);  view_1341 = None
	        _assert_tensor_metadata_774 = torch.ops.aten._assert_tensor_metadata.default(view_1343, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_774 = None
	        convert_element_type_515: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1343, torch.float32);  view_1343 = None
	        sub_3929: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_514, convert_element_type_515);  convert_element_type_514 = convert_element_type_515 = None
	        mul_8315: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3929, view_1342);  sub_3929 = view_1342 = None
	        view_1344: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8315, [1280, 1280]);  mul_8315 = None
	        _assert_tensor_metadata_775 = torch.ops.aten._assert_tensor_metadata.default(view_1344, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_775 = None
	        permute_143: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1344, [1, 0]);  view_1344 = None
	        mul_8318: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1345: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1340, [mul_8318, 1280]);  view_1340 = mul_8318 = None
	        mm_14: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1345, permute_143);  view_1345 = permute_143 = None
	        view_1346: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_14, [sym_size_int, 1500, 1280]);  mm_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1347: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1346, [sym_size_int, -1, 20, 64]);  view_1346 = None
	        permute_144: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1347, [0, 2, 1, 3]);  view_1347 = None
	        clone_115: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_144, memory_format = torch.contiguous_format);  permute_144 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1348: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_12884, [sym_size_int, 1500, 1280])
	        amin_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1348, [2])
	        amax_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1348, [2]);  view_1348 = None
	        full_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_86, full_172);  amin_86 = full_172 = None
	        full_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_86, full_173);  amax_86 = full_173 = None
	        sub_3943: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_86, minimum_86);  maximum_86 = None
	        div_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3943, 255.0);  sub_3943 = None
	        clamp_min_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_172, 1.1920928955078125e-07);  div_172 = None
	        div_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_86, clamp_min_258);  minimum_86 = None
	        round_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_173);  div_173 = None
	        sub_3949: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_173);  round_173 = None
	        clamp_min_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3949, -128);  sub_3949 = None
	        clamp_max_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_259, 127);  clamp_min_259 = None
	        _assert_tensor_metadata_776 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_258, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_776 = None
	        _assert_tensor_metadata_777 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_172, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_777 = None
	        convert_element_type_516: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_172, torch.int8);  clamp_max_172 = None
	        view_1349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_12884, [sym_size_int, 1500, 1280]);  add_12884 = None
	        view_1350: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_258, [sym_size_int, 1500, 1])
	        view_1351: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_516, [sym_size_int, 1500, 1])
	        reciprocal_86: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1350);  view_1350 = None
	        mul_8384: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_86, 1.0);  reciprocal_86 = None
	        mul_8387: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1349, mul_8384);  view_1349 = mul_8384 = None
	        round_174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8387);  mul_8387 = None
	        add_13271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_174, view_1351);  round_174 = view_1351 = None
	        clamp_min_260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13271, -128);  add_13271 = None
	        clamp_max_173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_260, 127);  clamp_min_260 = None
	        view_1352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_173, [sym_size_int, 1500, 1280]);  clamp_max_173 = None
	        _assert_tensor_metadata_778 = torch.ops.aten._assert_tensor_metadata.default(view_1352, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_778 = None
	        convert_element_type_517: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1352, torch.int8);  view_1352 = None
	        view_1353: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_517, [sym_size_int, 1500, 1280]);  convert_element_type_517 = None
	        view_1354: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_258, [sym_size_int, 1500, 1]);  clamp_min_258 = None
	        view_1355: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_516, [sym_size_int, 1500, 1]);  convert_element_type_516 = None
	        _assert_tensor_metadata_779 = torch.ops.aten._assert_tensor_metadata.default(view_1353, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_779 = None
	        convert_element_type_518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1353, torch.float32);  view_1353 = None
	        _assert_tensor_metadata_780 = torch.ops.aten._assert_tensor_metadata.default(view_1355, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_780 = None
	        convert_element_type_519: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1355, torch.float32);  view_1355 = None
	        sub_3969: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_518, convert_element_type_519);  convert_element_type_518 = convert_element_type_519 = None
	        mul_8409: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3969, view_1354);  sub_3969 = view_1354 = None
	        view_1356: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8409, [sym_size_int, 1500, 1280]);  mul_8409 = None
	        _assert_tensor_metadata_781 = torch.ops.aten._assert_tensor_metadata.default(view_1356, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_781 = None
	        view_1357: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1358: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1359: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_782 = torch.ops.aten._assert_tensor_metadata.default(view_1357, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_782 = None
	        convert_element_type_520: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1357, torch.float32);  view_1357 = None
	        _assert_tensor_metadata_783 = torch.ops.aten._assert_tensor_metadata.default(view_1359, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_783 = None
	        convert_element_type_521: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1359, torch.float32);  view_1359 = None
	        sub_3973: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_520, convert_element_type_521);  convert_element_type_520 = convert_element_type_521 = None
	        mul_8414: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3973, view_1358);  sub_3973 = view_1358 = None
	        view_1360: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8414, [1280, 1280]);  mul_8414 = None
	        _assert_tensor_metadata_784 = torch.ops.aten._assert_tensor_metadata.default(view_1360, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_784 = None
	        mul_8419: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1361: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1356, [mul_8419, 1280]);  view_1356 = mul_8419 = None
	        permute_145: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1360, [1, 0]);  view_1360 = None
	        addmm_71: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_14_self_attn_v_proj_bias, view_1361, permute_145);  model_audio_tower_layers_14_self_attn_v_proj_bias = view_1361 = permute_145 = None
	        view_1362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_71, [sym_size_int, 1500, 1280]);  addmm_71 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1363: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1362, [sym_size_int, -1, 20, 64]);  view_1362 = None
	        permute_146: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1363, [0, 2, 1, 3]);  view_1363 = None
	        clone_116: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_146, memory_format = torch.contiguous_format);  permute_146 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_14 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_114, clone_115, clone_116, None, False, scale = 1.0);  clone_114 = clone_115 = clone_116 = None
	        getitem_114: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_14[0];  _scaled_dot_product_efficient_attention_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_147: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_114, [0, 2, 1, 3]);  getitem_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1364: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_147, [sym_size_int, 1500, -1]);  permute_147 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1365: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1364, [sym_size_int, 1500, 1280])
	        amin_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1365, [2])
	        amax_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1365, [2]);  view_1365 = None
	        full_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_87, full_174);  amin_87 = full_174 = None
	        full_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_87, full_175);  amax_87 = full_175 = None
	        sub_3991: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_87, minimum_87);  maximum_87 = None
	        div_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3991, 255.0);  sub_3991 = None
	        clamp_min_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_174, 1.1920928955078125e-07);  div_174 = None
	        div_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_87, clamp_min_261);  minimum_87 = None
	        round_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_175);  div_175 = None
	        sub_3997: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_175);  round_175 = None
	        clamp_min_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3997, -128);  sub_3997 = None
	        clamp_max_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_262, 127);  clamp_min_262 = None
	        _assert_tensor_metadata_785 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_261, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_785 = None
	        _assert_tensor_metadata_786 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_174, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_786 = None
	        convert_element_type_522: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_174, torch.int8);  clamp_max_174 = None
	        view_1366: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1364, [sym_size_int, 1500, 1280]);  view_1364 = None
	        view_1367: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_261, [sym_size_int, 1500, 1])
	        view_1368: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_522, [sym_size_int, 1500, 1])
	        reciprocal_87: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1367);  view_1367 = None
	        mul_8489: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_87, 1.0);  reciprocal_87 = None
	        mul_8492: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1366, mul_8489);  view_1366 = mul_8489 = None
	        round_176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8492);  mul_8492 = None
	        add_13435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_176, view_1368);  round_176 = view_1368 = None
	        clamp_min_263: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13435, -128);  add_13435 = None
	        clamp_max_175: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_263, 127);  clamp_min_263 = None
	        view_1369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_175, [sym_size_int, 1500, 1280]);  clamp_max_175 = None
	        _assert_tensor_metadata_787 = torch.ops.aten._assert_tensor_metadata.default(view_1369, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_787 = None
	        convert_element_type_523: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1369, torch.int8);  view_1369 = None
	        view_1370: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_523, [sym_size_int, 1500, 1280]);  convert_element_type_523 = None
	        view_1371: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_261, [sym_size_int, 1500, 1]);  clamp_min_261 = None
	        view_1372: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_522, [sym_size_int, 1500, 1]);  convert_element_type_522 = None
	        _assert_tensor_metadata_788 = torch.ops.aten._assert_tensor_metadata.default(view_1370, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_788 = None
	        convert_element_type_524: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1370, torch.float32);  view_1370 = None
	        _assert_tensor_metadata_789 = torch.ops.aten._assert_tensor_metadata.default(view_1372, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_789 = None
	        convert_element_type_525: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1372, torch.float32);  view_1372 = None
	        sub_4017: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_524, convert_element_type_525);  convert_element_type_524 = convert_element_type_525 = None
	        mul_8514: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4017, view_1371);  sub_4017 = view_1371 = None
	        view_1373: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8514, [sym_size_int, 1500, 1280]);  mul_8514 = None
	        _assert_tensor_metadata_790 = torch.ops.aten._assert_tensor_metadata.default(view_1373, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_790 = None
	        view_1374: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1375: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1376: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_791 = torch.ops.aten._assert_tensor_metadata.default(view_1374, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_791 = None
	        convert_element_type_526: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1374, torch.float32);  view_1374 = None
	        _assert_tensor_metadata_792 = torch.ops.aten._assert_tensor_metadata.default(view_1376, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_792 = None
	        convert_element_type_527: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1376, torch.float32);  view_1376 = None
	        sub_4021: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_526, convert_element_type_527);  convert_element_type_526 = convert_element_type_527 = None
	        mul_8519: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4021, view_1375);  sub_4021 = view_1375 = None
	        view_1377: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8519, [1280, 1280]);  mul_8519 = None
	        _assert_tensor_metadata_793 = torch.ops.aten._assert_tensor_metadata.default(view_1377, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_793 = None
	        mul_8524: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1378: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1373, [mul_8524, 1280]);  view_1373 = mul_8524 = None
	        permute_148: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1377, [1, 0]);  view_1377 = None
	        addmm_72: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_14_self_attn_out_proj_bias, view_1378, permute_148);  model_audio_tower_layers_14_self_attn_out_proj_bias = view_1378 = permute_148 = None
	        view_1379: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_72, [sym_size_int, 1500, 1280]);  addmm_72 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1379);  view_1379 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_13498: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_12878, clone_117);  add_12878 = clone_117 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_118: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_13498, memory_format = torch.contiguous_format)
	        var_mean_29 = torch.ops.aten.var_mean.correction(clone_118, [2], correction = 0, keepdim = True)
	        getitem_118: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_29[0]
	        getitem_119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_29[1];  var_mean_29 = None
	        add_13503: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_118, 1e-05);  getitem_118 = None
	        rsqrt_29: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_13503);  add_13503 = None
	        sub_4027: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_118, getitem_119);  clone_118 = getitem_119 = None
	        mul_8535: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4027, rsqrt_29);  sub_4027 = rsqrt_29 = None
	        mul_8536: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8535, model_audio_tower_layers_14_final_layer_norm_weight);  mul_8535 = model_audio_tower_layers_14_final_layer_norm_weight = None
	        add_13504: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_8536, model_audio_tower_layers_14_final_layer_norm_bias);  mul_8536 = model_audio_tower_layers_14_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_13504, [sym_size_int, 1500, 1280])
	        amin_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1380, [2])
	        amax_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1380, [2]);  view_1380 = None
	        full_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_88, full_176);  amin_88 = full_176 = None
	        full_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_88, full_177);  amax_88 = full_177 = None
	        sub_4038: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_88, minimum_88);  maximum_88 = None
	        div_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4038, 255.0);  sub_4038 = None
	        clamp_min_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_176, 1.1920928955078125e-07);  div_176 = None
	        div_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_88, clamp_min_264);  minimum_88 = None
	        round_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_177);  div_177 = None
	        sub_4044: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_177);  round_177 = None
	        clamp_min_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4044, -128);  sub_4044 = None
	        clamp_max_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_265, 127);  clamp_min_265 = None
	        _assert_tensor_metadata_794 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_264, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_794 = None
	        _assert_tensor_metadata_795 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_176, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_795 = None
	        convert_element_type_528: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_176, torch.int8);  clamp_max_176 = None
	        view_1381: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_13504, [sym_size_int, 1500, 1280]);  add_13504 = None
	        view_1382: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_264, [sym_size_int, 1500, 1])
	        view_1383: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_528, [sym_size_int, 1500, 1])
	        reciprocal_88: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1382);  view_1382 = None
	        mul_8584: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_88, 1.0);  reciprocal_88 = None
	        mul_8587: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1381, mul_8584);  view_1381 = mul_8584 = None
	        round_178: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8587);  mul_8587 = None
	        add_13591: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_178, view_1383);  round_178 = view_1383 = None
	        clamp_min_266: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13591, -128);  add_13591 = None
	        clamp_max_177: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_266, 127);  clamp_min_266 = None
	        view_1384: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_177, [sym_size_int, 1500, 1280]);  clamp_max_177 = None
	        _assert_tensor_metadata_796 = torch.ops.aten._assert_tensor_metadata.default(view_1384, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_796 = None
	        convert_element_type_529: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1384, torch.int8);  view_1384 = None
	        view_1385: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_529, [sym_size_int, 1500, 1280]);  convert_element_type_529 = None
	        view_1386: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_264, [sym_size_int, 1500, 1]);  clamp_min_264 = None
	        view_1387: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_528, [sym_size_int, 1500, 1]);  convert_element_type_528 = None
	        _assert_tensor_metadata_797 = torch.ops.aten._assert_tensor_metadata.default(view_1385, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_797 = None
	        convert_element_type_530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1385, torch.float32);  view_1385 = None
	        _assert_tensor_metadata_798 = torch.ops.aten._assert_tensor_metadata.default(view_1387, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_798 = None
	        convert_element_type_531: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1387, torch.float32);  view_1387 = None
	        sub_4064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_530, convert_element_type_531);  convert_element_type_530 = convert_element_type_531 = None
	        mul_8609: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4064, view_1386);  sub_4064 = view_1386 = None
	        view_1388: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8609, [sym_size_int, 1500, 1280]);  mul_8609 = None
	        _assert_tensor_metadata_799 = torch.ops.aten._assert_tensor_metadata.default(view_1388, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_799 = None
	        view_1389: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_14_fc1_parametrizations_weight_original0 = None
	        view_1390: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_14_fc1_parametrizations_weight_original1 = None
	        view_1391: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_14_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_800 = torch.ops.aten._assert_tensor_metadata.default(view_1389, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_800 = None
	        convert_element_type_532: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1389, torch.float32);  view_1389 = None
	        _assert_tensor_metadata_801 = torch.ops.aten._assert_tensor_metadata.default(view_1391, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_801 = None
	        convert_element_type_533: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1391, torch.float32);  view_1391 = None
	        sub_4068: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_532, convert_element_type_533);  convert_element_type_532 = convert_element_type_533 = None
	        mul_8614: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4068, view_1390);  sub_4068 = view_1390 = None
	        view_1392: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8614, [5120, 1280]);  mul_8614 = None
	        _assert_tensor_metadata_802 = torch.ops.aten._assert_tensor_metadata.default(view_1392, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_802 = None
	        mul_8619: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1393: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1388, [mul_8619, 1280]);  view_1388 = mul_8619 = None
	        permute_149: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1392, [1, 0]);  view_1392 = None
	        addmm_73: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_14_fc1_bias, view_1393, permute_149);  model_audio_tower_layers_14_fc1_bias = view_1393 = permute_149 = None
	        view_1394: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_73, [sym_size_int, 1500, 5120]);  addmm_73 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_8626: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1394, 0.5)
	        mul_8627: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1394, 0.7071067811865476);  view_1394 = None
	        erf_16: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_8627);  mul_8627 = None
	        add_13650: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_16, 1);  erf_16 = None
	        mul_8628: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8626, add_13650);  mul_8626 = add_13650 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_119: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_8628);  mul_8628 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1395: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_119, [sym_size_int, 1500, 5120])
	        amin_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1395, [2])
	        amax_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1395, [2]);  view_1395 = None
	        full_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_89, full_178);  amin_89 = full_178 = None
	        full_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_89, full_179);  amax_89 = full_179 = None
	        sub_4081: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_89, minimum_89);  maximum_89 = None
	        div_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4081, 255.0);  sub_4081 = None
	        clamp_min_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_178, 1.1920928955078125e-07);  div_178 = None
	        div_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_89, clamp_min_267);  minimum_89 = None
	        round_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_179);  div_179 = None
	        sub_4087: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_179);  round_179 = None
	        clamp_min_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4087, -128);  sub_4087 = None
	        clamp_max_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_268, 127);  clamp_min_268 = None
	        _assert_tensor_metadata_803 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_267, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_803 = None
	        _assert_tensor_metadata_804 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_178, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_804 = None
	        convert_element_type_534: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_178, torch.int8);  clamp_max_178 = None
	        view_1396: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_119, [sym_size_int, 1500, 5120]);  clone_119 = None
	        view_1397: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_267, [sym_size_int, 1500, 1])
	        view_1398: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_534, [sym_size_int, 1500, 1])
	        reciprocal_89: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1397);  view_1397 = None
	        mul_8674: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_89, 1.0);  reciprocal_89 = None
	        mul_8677: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1396, mul_8674);  view_1396 = mul_8674 = None
	        round_180: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_8677);  mul_8677 = None
	        add_13733: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_180, view_1398);  round_180 = view_1398 = None
	        clamp_min_269: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13733, -128);  add_13733 = None
	        clamp_max_179: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_269, 127);  clamp_min_269 = None
	        view_1399: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_179, [sym_size_int, 1500, 5120]);  clamp_max_179 = None
	        _assert_tensor_metadata_805 = torch.ops.aten._assert_tensor_metadata.default(view_1399, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_805 = None
	        convert_element_type_535: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1399, torch.int8);  view_1399 = None
	        view_1400: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_535, [sym_size_int, 1500, 5120]);  convert_element_type_535 = None
	        view_1401: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_267, [sym_size_int, 1500, 1]);  clamp_min_267 = None
	        view_1402: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_534, [sym_size_int, 1500, 1]);  convert_element_type_534 = None
	        _assert_tensor_metadata_806 = torch.ops.aten._assert_tensor_metadata.default(view_1400, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_806 = None
	        convert_element_type_536: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1400, torch.float32);  view_1400 = None
	        _assert_tensor_metadata_807 = torch.ops.aten._assert_tensor_metadata.default(view_1402, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_807 = None
	        convert_element_type_537: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1402, torch.float32);  view_1402 = None
	        sub_4107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_536, convert_element_type_537);  convert_element_type_536 = convert_element_type_537 = None
	        mul_8699: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4107, view_1401);  sub_4107 = view_1401 = None
	        view_1403: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_8699, [sym_size_int, 1500, 5120]);  mul_8699 = None
	        _assert_tensor_metadata_808 = torch.ops.aten._assert_tensor_metadata.default(view_1403, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_808 = None
	        view_1404: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_14_fc2_parametrizations_weight_original0 = None
	        view_1405: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_14_fc2_parametrizations_weight_original1 = None
	        view_1406: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_14_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_809 = torch.ops.aten._assert_tensor_metadata.default(view_1404, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_809 = None
	        convert_element_type_538: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1404, torch.float32);  view_1404 = None
	        _assert_tensor_metadata_810 = torch.ops.aten._assert_tensor_metadata.default(view_1406, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_810 = None
	        convert_element_type_539: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1406, torch.float32);  view_1406 = None
	        sub_4111: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_538, convert_element_type_539);  convert_element_type_538 = convert_element_type_539 = None
	        mul_8704: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4111, view_1405);  sub_4111 = view_1405 = None
	        view_1407: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_8704, [1280, 5120]);  mul_8704 = None
	        _assert_tensor_metadata_811 = torch.ops.aten._assert_tensor_metadata.default(view_1407, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_811 = None
	        mul_8709: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1408: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1403, [mul_8709, 5120]);  view_1403 = mul_8709 = None
	        permute_150: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1407, [1, 0]);  view_1407 = None
	        addmm_74: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_14_fc2_bias, view_1408, permute_150);  model_audio_tower_layers_14_fc2_bias = view_1408 = permute_150 = None
	        view_1409: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_74, [sym_size_int, 1500, 1280]);  addmm_74 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_120: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1409);  view_1409 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_13796: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_13498, clone_120);  add_13498 = clone_120 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_121: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_13796, memory_format = torch.contiguous_format)
	        var_mean_30 = torch.ops.aten.var_mean.correction(clone_121, [2], correction = 0, keepdim = True)
	        getitem_120: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_30[0]
	        getitem_121: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_30[1];  var_mean_30 = None
	        add_13801: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_120, 1e-05);  getitem_120 = None
	        rsqrt_30: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_13801);  add_13801 = None
	        sub_4117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_121, getitem_121);  clone_121 = getitem_121 = None
	        mul_8720: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4117, rsqrt_30);  sub_4117 = rsqrt_30 = None
	        mul_8721: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8720, model_audio_tower_layers_15_self_attn_layer_norm_weight);  mul_8720 = model_audio_tower_layers_15_self_attn_layer_norm_weight = None
	        add_13802: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_8721, model_audio_tower_layers_15_self_attn_layer_norm_bias);  mul_8721 = model_audio_tower_layers_15_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1410: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_13802, [sym_size_int, 1500, 1280])
	        amin_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1410, [2])
	        amax_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1410, [2]);  view_1410 = None
	        full_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_90, full_180);  amin_90 = full_180 = None
	        full_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_90, full_181);  amax_90 = full_181 = None
	        sub_4128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_90, minimum_90);  maximum_90 = None
	        div_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4128, 255.0);  sub_4128 = None
	        clamp_min_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_180, 1.1920928955078125e-07);  div_180 = None
	        div_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_90, clamp_min_270);  minimum_90 = None
	        round_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_181);  div_181 = None
	        sub_4134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_181);  round_181 = None
	        clamp_min_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4134, -128);  sub_4134 = None
	        clamp_max_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_271, 127);  clamp_min_271 = None
	        _assert_tensor_metadata_812 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_270, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_812 = None
	        _assert_tensor_metadata_813 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_180, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_813 = None
	        convert_element_type_540: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_180, torch.int8);  clamp_max_180 = None
	        view_1411: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_13802, [sym_size_int, 1500, 1280])
	        view_1412: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_270, [sym_size_int, 1500, 1])
	        view_1413: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_540, [sym_size_int, 1500, 1])
	        reciprocal_90: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1412);  view_1412 = None
	        mul_8769: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_90, 1.0);  reciprocal_90 = None
	        mul_8772: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1411, mul_8769);  view_1411 = mul_8769 = None
	        round_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8772);  mul_8772 = None
	        add_13889: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_182, view_1413);  round_182 = view_1413 = None
	        clamp_min_272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13889, -128);  add_13889 = None
	        clamp_max_181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_272, 127);  clamp_min_272 = None
	        view_1414: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_181, [sym_size_int, 1500, 1280]);  clamp_max_181 = None
	        _assert_tensor_metadata_814 = torch.ops.aten._assert_tensor_metadata.default(view_1414, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_814 = None
	        convert_element_type_541: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1414, torch.int8);  view_1414 = None
	        view_1415: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_541, [sym_size_int, 1500, 1280]);  convert_element_type_541 = None
	        view_1416: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_270, [sym_size_int, 1500, 1]);  clamp_min_270 = None
	        view_1417: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_540, [sym_size_int, 1500, 1]);  convert_element_type_540 = None
	        _assert_tensor_metadata_815 = torch.ops.aten._assert_tensor_metadata.default(view_1415, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_815 = None
	        convert_element_type_542: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1415, torch.float32);  view_1415 = None
	        _assert_tensor_metadata_816 = torch.ops.aten._assert_tensor_metadata.default(view_1417, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_816 = None
	        convert_element_type_543: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1417, torch.float32);  view_1417 = None
	        sub_4154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_542, convert_element_type_543);  convert_element_type_542 = convert_element_type_543 = None
	        mul_8794: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4154, view_1416);  sub_4154 = view_1416 = None
	        view_1418: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8794, [sym_size_int, 1500, 1280]);  mul_8794 = None
	        _assert_tensor_metadata_817 = torch.ops.aten._assert_tensor_metadata.default(view_1418, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_817 = None
	        view_1419: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1420: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1421: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_818 = torch.ops.aten._assert_tensor_metadata.default(view_1419, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_818 = None
	        convert_element_type_544: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1419, torch.float32);  view_1419 = None
	        _assert_tensor_metadata_819 = torch.ops.aten._assert_tensor_metadata.default(view_1421, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_819 = None
	        convert_element_type_545: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1421, torch.float32);  view_1421 = None
	        sub_4158: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_544, convert_element_type_545);  convert_element_type_544 = convert_element_type_545 = None
	        mul_8799: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4158, view_1420);  sub_4158 = view_1420 = None
	        view_1422: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8799, [1280, 1280]);  mul_8799 = None
	        _assert_tensor_metadata_820 = torch.ops.aten._assert_tensor_metadata.default(view_1422, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_820 = None
	        mul_8804: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1423: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1418, [mul_8804, 1280]);  view_1418 = mul_8804 = None
	        permute_151: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1422, [1, 0]);  view_1422 = None
	        addmm_75: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_15_self_attn_q_proj_bias, view_1423, permute_151);  model_audio_tower_layers_15_self_attn_q_proj_bias = view_1423 = permute_151 = None
	        view_1424: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_75, [sym_size_int, 1500, 1280]);  addmm_75 = None
	        mul_8811: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1424, 0.125);  view_1424 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1425: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_8811, [sym_size_int, 1500, 20, 64]);  mul_8811 = None
	        permute_152: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1425, [0, 2, 1, 3]);  view_1425 = None
	        clone_122: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_152, memory_format = torch.contiguous_format);  permute_152 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1426: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_13802, [sym_size_int, 1500, 1280])
	        amin_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1426, [2])
	        amax_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1426, [2]);  view_1426 = None
	        full_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_91, full_182);  amin_91 = full_182 = None
	        full_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_91, full_183);  amax_91 = full_183 = None
	        sub_4173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_91, minimum_91);  maximum_91 = None
	        div_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4173, 255.0);  sub_4173 = None
	        clamp_min_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_182, 1.1920928955078125e-07);  div_182 = None
	        div_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_91, clamp_min_273);  minimum_91 = None
	        round_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_183);  div_183 = None
	        sub_4179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_183);  round_183 = None
	        clamp_min_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4179, -128);  sub_4179 = None
	        clamp_max_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_274, 127);  clamp_min_274 = None
	        _assert_tensor_metadata_821 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_273, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_821 = None
	        _assert_tensor_metadata_822 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_182, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_822 = None
	        convert_element_type_546: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_182, torch.int8);  clamp_max_182 = None
	        view_1427: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_13802, [sym_size_int, 1500, 1280])
	        view_1428: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_273, [sym_size_int, 1500, 1])
	        view_1429: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_546, [sym_size_int, 1500, 1])
	        reciprocal_91: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1428);  view_1428 = None
	        mul_8865: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_91, 1.0);  reciprocal_91 = None
	        mul_8868: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1427, mul_8865);  view_1427 = mul_8865 = None
	        round_184: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8868);  mul_8868 = None
	        add_14041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_184, view_1429);  round_184 = view_1429 = None
	        clamp_min_275: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14041, -128);  add_14041 = None
	        clamp_max_183: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_275, 127);  clamp_min_275 = None
	        view_1430: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_183, [sym_size_int, 1500, 1280]);  clamp_max_183 = None
	        _assert_tensor_metadata_823 = torch.ops.aten._assert_tensor_metadata.default(view_1430, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_823 = None
	        convert_element_type_547: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1430, torch.int8);  view_1430 = None
	        view_1431: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_547, [sym_size_int, 1500, 1280]);  convert_element_type_547 = None
	        view_1432: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_273, [sym_size_int, 1500, 1]);  clamp_min_273 = None
	        view_1433: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_546, [sym_size_int, 1500, 1]);  convert_element_type_546 = None
	        _assert_tensor_metadata_824 = torch.ops.aten._assert_tensor_metadata.default(view_1431, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_824 = None
	        convert_element_type_548: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1431, torch.float32);  view_1431 = None
	        _assert_tensor_metadata_825 = torch.ops.aten._assert_tensor_metadata.default(view_1433, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_825 = None
	        convert_element_type_549: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1433, torch.float32);  view_1433 = None
	        sub_4199: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_548, convert_element_type_549);  convert_element_type_548 = convert_element_type_549 = None
	        mul_8890: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4199, view_1432);  sub_4199 = view_1432 = None
	        view_1434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8890, [sym_size_int, 1500, 1280]);  mul_8890 = None
	        _assert_tensor_metadata_826 = torch.ops.aten._assert_tensor_metadata.default(view_1434, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_826 = None
	        view_1435: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1436: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1437: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_827 = torch.ops.aten._assert_tensor_metadata.default(view_1435, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_827 = None
	        convert_element_type_550: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1435, torch.float32);  view_1435 = None
	        _assert_tensor_metadata_828 = torch.ops.aten._assert_tensor_metadata.default(view_1437, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_828 = None
	        convert_element_type_551: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1437, torch.float32);  view_1437 = None
	        sub_4203: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_550, convert_element_type_551);  convert_element_type_550 = convert_element_type_551 = None
	        mul_8895: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4203, view_1436);  sub_4203 = view_1436 = None
	        view_1438: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8895, [1280, 1280]);  mul_8895 = None
	        _assert_tensor_metadata_829 = torch.ops.aten._assert_tensor_metadata.default(view_1438, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_829 = None
	        permute_153: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1438, [1, 0]);  view_1438 = None
	        mul_8898: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1439: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1434, [mul_8898, 1280]);  view_1434 = mul_8898 = None
	        mm_15: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1439, permute_153);  view_1439 = permute_153 = None
	        view_1440: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_15, [sym_size_int, 1500, 1280]);  mm_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1441: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1440, [sym_size_int, -1, 20, 64]);  view_1440 = None
	        permute_154: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1441, [0, 2, 1, 3]);  view_1441 = None
	        clone_123: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_154, memory_format = torch.contiguous_format);  permute_154 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1442: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_13802, [sym_size_int, 1500, 1280])
	        amin_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1442, [2])
	        amax_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1442, [2]);  view_1442 = None
	        full_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_92, full_184);  amin_92 = full_184 = None
	        full_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_92, full_185);  amax_92 = full_185 = None
	        sub_4217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_92, minimum_92);  maximum_92 = None
	        div_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4217, 255.0);  sub_4217 = None
	        clamp_min_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_184, 1.1920928955078125e-07);  div_184 = None
	        div_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_92, clamp_min_276);  minimum_92 = None
	        round_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_185);  div_185 = None
	        sub_4223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_185);  round_185 = None
	        clamp_min_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4223, -128);  sub_4223 = None
	        clamp_max_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_277, 127);  clamp_min_277 = None
	        _assert_tensor_metadata_830 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_276, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_830 = None
	        _assert_tensor_metadata_831 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_184, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_831 = None
	        convert_element_type_552: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_184, torch.int8);  clamp_max_184 = None
	        view_1443: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_13802, [sym_size_int, 1500, 1280]);  add_13802 = None
	        view_1444: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_276, [sym_size_int, 1500, 1])
	        view_1445: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_552, [sym_size_int, 1500, 1])
	        reciprocal_92: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1444);  view_1444 = None
	        mul_8964: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_92, 1.0);  reciprocal_92 = None
	        mul_8967: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1443, mul_8964);  view_1443 = mul_8964 = None
	        round_186: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8967);  mul_8967 = None
	        add_14189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_186, view_1445);  round_186 = view_1445 = None
	        clamp_min_278: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14189, -128);  add_14189 = None
	        clamp_max_185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_278, 127);  clamp_min_278 = None
	        view_1446: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_185, [sym_size_int, 1500, 1280]);  clamp_max_185 = None
	        _assert_tensor_metadata_832 = torch.ops.aten._assert_tensor_metadata.default(view_1446, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_832 = None
	        convert_element_type_553: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1446, torch.int8);  view_1446 = None
	        view_1447: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_553, [sym_size_int, 1500, 1280]);  convert_element_type_553 = None
	        view_1448: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_276, [sym_size_int, 1500, 1]);  clamp_min_276 = None
	        view_1449: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_552, [sym_size_int, 1500, 1]);  convert_element_type_552 = None
	        _assert_tensor_metadata_833 = torch.ops.aten._assert_tensor_metadata.default(view_1447, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_833 = None
	        convert_element_type_554: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1447, torch.float32);  view_1447 = None
	        _assert_tensor_metadata_834 = torch.ops.aten._assert_tensor_metadata.default(view_1449, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_834 = None
	        convert_element_type_555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1449, torch.float32);  view_1449 = None
	        sub_4243: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_554, convert_element_type_555);  convert_element_type_554 = convert_element_type_555 = None
	        mul_8989: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4243, view_1448);  sub_4243 = view_1448 = None
	        view_1450: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8989, [sym_size_int, 1500, 1280]);  mul_8989 = None
	        _assert_tensor_metadata_835 = torch.ops.aten._assert_tensor_metadata.default(view_1450, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_835 = None
	        view_1451: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1452: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1453: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_836 = torch.ops.aten._assert_tensor_metadata.default(view_1451, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_836 = None
	        convert_element_type_556: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1451, torch.float32);  view_1451 = None
	        _assert_tensor_metadata_837 = torch.ops.aten._assert_tensor_metadata.default(view_1453, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_837 = None
	        convert_element_type_557: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1453, torch.float32);  view_1453 = None
	        sub_4247: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_556, convert_element_type_557);  convert_element_type_556 = convert_element_type_557 = None
	        mul_8994: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4247, view_1452);  sub_4247 = view_1452 = None
	        view_1454: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8994, [1280, 1280]);  mul_8994 = None
	        _assert_tensor_metadata_838 = torch.ops.aten._assert_tensor_metadata.default(view_1454, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_838 = None
	        mul_8999: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1455: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1450, [mul_8999, 1280]);  view_1450 = mul_8999 = None
	        permute_155: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1454, [1, 0]);  view_1454 = None
	        addmm_76: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_15_self_attn_v_proj_bias, view_1455, permute_155);  model_audio_tower_layers_15_self_attn_v_proj_bias = view_1455 = permute_155 = None
	        view_1456: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_76, [sym_size_int, 1500, 1280]);  addmm_76 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1457: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1456, [sym_size_int, -1, 20, 64]);  view_1456 = None
	        permute_156: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1457, [0, 2, 1, 3]);  view_1457 = None
	        clone_124: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_156, memory_format = torch.contiguous_format);  permute_156 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_15 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_122, clone_123, clone_124, None, False, scale = 1.0);  clone_122 = clone_123 = clone_124 = None
	        getitem_122: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_15[0];  _scaled_dot_product_efficient_attention_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_157: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_122, [0, 2, 1, 3]);  getitem_122 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_157, [sym_size_int, 1500, -1]);  permute_157 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1459: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1458, [sym_size_int, 1500, 1280])
	        amin_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1459, [2])
	        amax_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1459, [2]);  view_1459 = None
	        full_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_93, full_186);  amin_93 = full_186 = None
	        full_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_93, full_187);  amax_93 = full_187 = None
	        sub_4265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_93, minimum_93);  maximum_93 = None
	        div_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4265, 255.0);  sub_4265 = None
	        clamp_min_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_186, 1.1920928955078125e-07);  div_186 = None
	        div_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_93, clamp_min_279);  minimum_93 = None
	        round_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_187);  div_187 = None
	        sub_4271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_187);  round_187 = None
	        clamp_min_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4271, -128);  sub_4271 = None
	        clamp_max_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_280, 127);  clamp_min_280 = None
	        _assert_tensor_metadata_839 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_279, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_839 = None
	        _assert_tensor_metadata_840 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_186, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_840 = None
	        convert_element_type_558: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_186, torch.int8);  clamp_max_186 = None
	        view_1460: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1458, [sym_size_int, 1500, 1280]);  view_1458 = None
	        view_1461: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_279, [sym_size_int, 1500, 1])
	        view_1462: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_558, [sym_size_int, 1500, 1])
	        reciprocal_93: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1461);  view_1461 = None
	        mul_9069: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_93, 1.0);  reciprocal_93 = None
	        mul_9072: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1460, mul_9069);  view_1460 = mul_9069 = None
	        round_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9072);  mul_9072 = None
	        add_14353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_188, view_1462);  round_188 = view_1462 = None
	        clamp_min_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14353, -128);  add_14353 = None
	        clamp_max_187: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_281, 127);  clamp_min_281 = None
	        view_1463: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_187, [sym_size_int, 1500, 1280]);  clamp_max_187 = None
	        _assert_tensor_metadata_841 = torch.ops.aten._assert_tensor_metadata.default(view_1463, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_841 = None
	        convert_element_type_559: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1463, torch.int8);  view_1463 = None
	        view_1464: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_559, [sym_size_int, 1500, 1280]);  convert_element_type_559 = None
	        view_1465: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_279, [sym_size_int, 1500, 1]);  clamp_min_279 = None
	        view_1466: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_558, [sym_size_int, 1500, 1]);  convert_element_type_558 = None
	        _assert_tensor_metadata_842 = torch.ops.aten._assert_tensor_metadata.default(view_1464, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_842 = None
	        convert_element_type_560: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1464, torch.float32);  view_1464 = None
	        _assert_tensor_metadata_843 = torch.ops.aten._assert_tensor_metadata.default(view_1466, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_843 = None
	        convert_element_type_561: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1466, torch.float32);  view_1466 = None
	        sub_4291: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_560, convert_element_type_561);  convert_element_type_560 = convert_element_type_561 = None
	        mul_9094: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4291, view_1465);  sub_4291 = view_1465 = None
	        view_1467: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9094, [sym_size_int, 1500, 1280]);  mul_9094 = None
	        _assert_tensor_metadata_844 = torch.ops.aten._assert_tensor_metadata.default(view_1467, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_844 = None
	        view_1468: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1469: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1470: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_845 = torch.ops.aten._assert_tensor_metadata.default(view_1468, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_845 = None
	        convert_element_type_562: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1468, torch.float32);  view_1468 = None
	        _assert_tensor_metadata_846 = torch.ops.aten._assert_tensor_metadata.default(view_1470, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_846 = None
	        convert_element_type_563: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1470, torch.float32);  view_1470 = None
	        sub_4295: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_562, convert_element_type_563);  convert_element_type_562 = convert_element_type_563 = None
	        mul_9099: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4295, view_1469);  sub_4295 = view_1469 = None
	        view_1471: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9099, [1280, 1280]);  mul_9099 = None
	        _assert_tensor_metadata_847 = torch.ops.aten._assert_tensor_metadata.default(view_1471, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_847 = None
	        mul_9104: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1472: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1467, [mul_9104, 1280]);  view_1467 = mul_9104 = None
	        permute_158: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1471, [1, 0]);  view_1471 = None
	        addmm_77: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_15_self_attn_out_proj_bias, view_1472, permute_158);  model_audio_tower_layers_15_self_attn_out_proj_bias = view_1472 = permute_158 = None
	        view_1473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_77, [sym_size_int, 1500, 1280]);  addmm_77 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_125: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1473);  view_1473 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_14416: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_13796, clone_125);  add_13796 = clone_125 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_126: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_14416, memory_format = torch.contiguous_format)
	        var_mean_31 = torch.ops.aten.var_mean.correction(clone_126, [2], correction = 0, keepdim = True)
	        getitem_126: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_31[0]
	        getitem_127: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_31[1];  var_mean_31 = None
	        add_14421: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_126, 1e-05);  getitem_126 = None
	        rsqrt_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_14421);  add_14421 = None
	        sub_4301: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_126, getitem_127);  clone_126 = getitem_127 = None
	        mul_9115: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4301, rsqrt_31);  sub_4301 = rsqrt_31 = None
	        mul_9116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9115, model_audio_tower_layers_15_final_layer_norm_weight);  mul_9115 = model_audio_tower_layers_15_final_layer_norm_weight = None
	        add_14422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9116, model_audio_tower_layers_15_final_layer_norm_bias);  mul_9116 = model_audio_tower_layers_15_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_14422, [sym_size_int, 1500, 1280])
	        amin_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1474, [2])
	        amax_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1474, [2]);  view_1474 = None
	        full_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_94, full_188);  amin_94 = full_188 = None
	        full_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_94, full_189);  amax_94 = full_189 = None
	        sub_4312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_94, minimum_94);  maximum_94 = None
	        div_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4312, 255.0);  sub_4312 = None
	        clamp_min_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_188, 1.1920928955078125e-07);  div_188 = None
	        div_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_94, clamp_min_282);  minimum_94 = None
	        round_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_189);  div_189 = None
	        sub_4318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_189);  round_189 = None
	        clamp_min_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4318, -128);  sub_4318 = None
	        clamp_max_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_283, 127);  clamp_min_283 = None
	        _assert_tensor_metadata_848 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_282, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_848 = None
	        _assert_tensor_metadata_849 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_188, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_849 = None
	        convert_element_type_564: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_188, torch.int8);  clamp_max_188 = None
	        view_1475: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_14422, [sym_size_int, 1500, 1280]);  add_14422 = None
	        view_1476: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_282, [sym_size_int, 1500, 1])
	        view_1477: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_564, [sym_size_int, 1500, 1])
	        reciprocal_94: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1476);  view_1476 = None
	        mul_9164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_94, 1.0);  reciprocal_94 = None
	        mul_9167: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1475, mul_9164);  view_1475 = mul_9164 = None
	        round_190: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9167);  mul_9167 = None
	        add_14509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_190, view_1477);  round_190 = view_1477 = None
	        clamp_min_284: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14509, -128);  add_14509 = None
	        clamp_max_189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_284, 127);  clamp_min_284 = None
	        view_1478: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_189, [sym_size_int, 1500, 1280]);  clamp_max_189 = None
	        _assert_tensor_metadata_850 = torch.ops.aten._assert_tensor_metadata.default(view_1478, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_850 = None
	        convert_element_type_565: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1478, torch.int8);  view_1478 = None
	        view_1479: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_565, [sym_size_int, 1500, 1280]);  convert_element_type_565 = None
	        view_1480: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_282, [sym_size_int, 1500, 1]);  clamp_min_282 = None
	        view_1481: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_564, [sym_size_int, 1500, 1]);  convert_element_type_564 = None
	        _assert_tensor_metadata_851 = torch.ops.aten._assert_tensor_metadata.default(view_1479, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_851 = None
	        convert_element_type_566: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1479, torch.float32);  view_1479 = None
	        _assert_tensor_metadata_852 = torch.ops.aten._assert_tensor_metadata.default(view_1481, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_852 = None
	        convert_element_type_567: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1481, torch.float32);  view_1481 = None
	        sub_4338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_566, convert_element_type_567);  convert_element_type_566 = convert_element_type_567 = None
	        mul_9189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4338, view_1480);  sub_4338 = view_1480 = None
	        view_1482: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9189, [sym_size_int, 1500, 1280]);  mul_9189 = None
	        _assert_tensor_metadata_853 = torch.ops.aten._assert_tensor_metadata.default(view_1482, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_853 = None
	        view_1483: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_15_fc1_parametrizations_weight_original0 = None
	        view_1484: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_15_fc1_parametrizations_weight_original1 = None
	        view_1485: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_15_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_854 = torch.ops.aten._assert_tensor_metadata.default(view_1483, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_854 = None
	        convert_element_type_568: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1483, torch.float32);  view_1483 = None
	        _assert_tensor_metadata_855 = torch.ops.aten._assert_tensor_metadata.default(view_1485, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_855 = None
	        convert_element_type_569: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1485, torch.float32);  view_1485 = None
	        sub_4342: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_568, convert_element_type_569);  convert_element_type_568 = convert_element_type_569 = None
	        mul_9194: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4342, view_1484);  sub_4342 = view_1484 = None
	        view_1486: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9194, [5120, 1280]);  mul_9194 = None
	        _assert_tensor_metadata_856 = torch.ops.aten._assert_tensor_metadata.default(view_1486, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_856 = None
	        mul_9199: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1487: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1482, [mul_9199, 1280]);  view_1482 = mul_9199 = None
	        permute_159: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1486, [1, 0]);  view_1486 = None
	        addmm_78: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_15_fc1_bias, view_1487, permute_159);  model_audio_tower_layers_15_fc1_bias = view_1487 = permute_159 = None
	        view_1488: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_78, [sym_size_int, 1500, 5120]);  addmm_78 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_9206: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1488, 0.5)
	        mul_9207: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1488, 0.7071067811865476);  view_1488 = None
	        erf_17: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_9207);  mul_9207 = None
	        add_14568: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_17, 1);  erf_17 = None
	        mul_9208: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9206, add_14568);  mul_9206 = add_14568 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_127: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_9208);  mul_9208 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1489: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_127, [sym_size_int, 1500, 5120])
	        amin_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1489, [2])
	        amax_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1489, [2]);  view_1489 = None
	        full_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_95, full_190);  amin_95 = full_190 = None
	        full_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_95, full_191);  amax_95 = full_191 = None
	        sub_4355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_95, minimum_95);  maximum_95 = None
	        div_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4355, 255.0);  sub_4355 = None
	        clamp_min_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_190, 1.1920928955078125e-07);  div_190 = None
	        div_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_95, clamp_min_285);  minimum_95 = None
	        round_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_191);  div_191 = None
	        sub_4361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_191);  round_191 = None
	        clamp_min_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4361, -128);  sub_4361 = None
	        clamp_max_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_286, 127);  clamp_min_286 = None
	        _assert_tensor_metadata_857 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_285, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_857 = None
	        _assert_tensor_metadata_858 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_190, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_858 = None
	        convert_element_type_570: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_190, torch.int8);  clamp_max_190 = None
	        view_1490: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_127, [sym_size_int, 1500, 5120]);  clone_127 = None
	        view_1491: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_285, [sym_size_int, 1500, 1])
	        view_1492: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_570, [sym_size_int, 1500, 1])
	        reciprocal_95: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1491);  view_1491 = None
	        mul_9254: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_95, 1.0);  reciprocal_95 = None
	        mul_9257: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1490, mul_9254);  view_1490 = mul_9254 = None
	        round_192: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_9257);  mul_9257 = None
	        add_14651: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_192, view_1492);  round_192 = view_1492 = None
	        clamp_min_287: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14651, -128);  add_14651 = None
	        clamp_max_191: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_287, 127);  clamp_min_287 = None
	        view_1493: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_191, [sym_size_int, 1500, 5120]);  clamp_max_191 = None
	        _assert_tensor_metadata_859 = torch.ops.aten._assert_tensor_metadata.default(view_1493, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_859 = None
	        convert_element_type_571: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1493, torch.int8);  view_1493 = None
	        view_1494: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_571, [sym_size_int, 1500, 5120]);  convert_element_type_571 = None
	        view_1495: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_285, [sym_size_int, 1500, 1]);  clamp_min_285 = None
	        view_1496: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_570, [sym_size_int, 1500, 1]);  convert_element_type_570 = None
	        _assert_tensor_metadata_860 = torch.ops.aten._assert_tensor_metadata.default(view_1494, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_860 = None
	        convert_element_type_572: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1494, torch.float32);  view_1494 = None
	        _assert_tensor_metadata_861 = torch.ops.aten._assert_tensor_metadata.default(view_1496, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_861 = None
	        convert_element_type_573: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1496, torch.float32);  view_1496 = None
	        sub_4381: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_572, convert_element_type_573);  convert_element_type_572 = convert_element_type_573 = None
	        mul_9279: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4381, view_1495);  sub_4381 = view_1495 = None
	        view_1497: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_9279, [sym_size_int, 1500, 5120]);  mul_9279 = None
	        _assert_tensor_metadata_862 = torch.ops.aten._assert_tensor_metadata.default(view_1497, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_862 = None
	        view_1498: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_15_fc2_parametrizations_weight_original0 = None
	        view_1499: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_15_fc2_parametrizations_weight_original1 = None
	        view_1500: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_15_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_863 = torch.ops.aten._assert_tensor_metadata.default(view_1498, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_863 = None
	        convert_element_type_574: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1498, torch.float32);  view_1498 = None
	        _assert_tensor_metadata_864 = torch.ops.aten._assert_tensor_metadata.default(view_1500, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_864 = None
	        convert_element_type_575: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1500, torch.float32);  view_1500 = None
	        sub_4385: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_574, convert_element_type_575);  convert_element_type_574 = convert_element_type_575 = None
	        mul_9284: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4385, view_1499);  sub_4385 = view_1499 = None
	        view_1501: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_9284, [1280, 5120]);  mul_9284 = None
	        _assert_tensor_metadata_865 = torch.ops.aten._assert_tensor_metadata.default(view_1501, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_865 = None
	        mul_9289: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1502: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1497, [mul_9289, 5120]);  view_1497 = mul_9289 = None
	        permute_160: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1501, [1, 0]);  view_1501 = None
	        addmm_79: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_15_fc2_bias, view_1502, permute_160);  model_audio_tower_layers_15_fc2_bias = view_1502 = permute_160 = None
	        view_1503: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_79, [sym_size_int, 1500, 1280]);  addmm_79 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1503);  view_1503 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_14714: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_14416, clone_128);  add_14416 = clone_128 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_129: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_14714, memory_format = torch.contiguous_format)
	        var_mean_32 = torch.ops.aten.var_mean.correction(clone_129, [2], correction = 0, keepdim = True)
	        getitem_128: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_32[0]
	        getitem_129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_32[1];  var_mean_32 = None
	        add_14719: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_128, 1e-05);  getitem_128 = None
	        rsqrt_32: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_14719);  add_14719 = None
	        sub_4391: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_129, getitem_129);  clone_129 = getitem_129 = None
	        mul_9300: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4391, rsqrt_32);  sub_4391 = rsqrt_32 = None
	        mul_9301: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9300, model_audio_tower_layers_16_self_attn_layer_norm_weight);  mul_9300 = model_audio_tower_layers_16_self_attn_layer_norm_weight = None
	        add_14720: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9301, model_audio_tower_layers_16_self_attn_layer_norm_bias);  mul_9301 = model_audio_tower_layers_16_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1504: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_14720, [sym_size_int, 1500, 1280])
	        amin_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1504, [2])
	        amax_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1504, [2]);  view_1504 = None
	        full_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_96, full_192);  amin_96 = full_192 = None
	        full_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_96, full_193);  amax_96 = full_193 = None
	        sub_4402: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_96, minimum_96);  maximum_96 = None
	        div_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4402, 255.0);  sub_4402 = None
	        clamp_min_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_192, 1.1920928955078125e-07);  div_192 = None
	        div_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_96, clamp_min_288);  minimum_96 = None
	        round_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_193);  div_193 = None
	        sub_4408: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_193);  round_193 = None
	        clamp_min_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4408, -128);  sub_4408 = None
	        clamp_max_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_289, 127);  clamp_min_289 = None
	        _assert_tensor_metadata_866 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_288, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_866 = None
	        _assert_tensor_metadata_867 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_192, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_867 = None
	        convert_element_type_576: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_192, torch.int8);  clamp_max_192 = None
	        view_1505: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_14720, [sym_size_int, 1500, 1280])
	        view_1506: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_288, [sym_size_int, 1500, 1])
	        view_1507: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_576, [sym_size_int, 1500, 1])
	        reciprocal_96: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1506);  view_1506 = None
	        mul_9349: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_96, 1.0);  reciprocal_96 = None
	        mul_9352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1505, mul_9349);  view_1505 = mul_9349 = None
	        round_194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9352);  mul_9352 = None
	        add_14807: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_194, view_1507);  round_194 = view_1507 = None
	        clamp_min_290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14807, -128);  add_14807 = None
	        clamp_max_193: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_290, 127);  clamp_min_290 = None
	        view_1508: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_193, [sym_size_int, 1500, 1280]);  clamp_max_193 = None
	        _assert_tensor_metadata_868 = torch.ops.aten._assert_tensor_metadata.default(view_1508, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_868 = None
	        convert_element_type_577: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1508, torch.int8);  view_1508 = None
	        view_1509: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_577, [sym_size_int, 1500, 1280]);  convert_element_type_577 = None
	        view_1510: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_288, [sym_size_int, 1500, 1]);  clamp_min_288 = None
	        view_1511: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_576, [sym_size_int, 1500, 1]);  convert_element_type_576 = None
	        _assert_tensor_metadata_869 = torch.ops.aten._assert_tensor_metadata.default(view_1509, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_869 = None
	        convert_element_type_578: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1509, torch.float32);  view_1509 = None
	        _assert_tensor_metadata_870 = torch.ops.aten._assert_tensor_metadata.default(view_1511, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_870 = None
	        convert_element_type_579: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1511, torch.float32);  view_1511 = None
	        sub_4428: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_578, convert_element_type_579);  convert_element_type_578 = convert_element_type_579 = None
	        mul_9374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4428, view_1510);  sub_4428 = view_1510 = None
	        view_1512: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9374, [sym_size_int, 1500, 1280]);  mul_9374 = None
	        _assert_tensor_metadata_871 = torch.ops.aten._assert_tensor_metadata.default(view_1512, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_871 = None
	        view_1513: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1514: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1515: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_872 = torch.ops.aten._assert_tensor_metadata.default(view_1513, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_872 = None
	        convert_element_type_580: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1513, torch.float32);  view_1513 = None
	        _assert_tensor_metadata_873 = torch.ops.aten._assert_tensor_metadata.default(view_1515, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_873 = None
	        convert_element_type_581: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1515, torch.float32);  view_1515 = None
	        sub_4432: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_580, convert_element_type_581);  convert_element_type_580 = convert_element_type_581 = None
	        mul_9379: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4432, view_1514);  sub_4432 = view_1514 = None
	        view_1516: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9379, [1280, 1280]);  mul_9379 = None
	        _assert_tensor_metadata_874 = torch.ops.aten._assert_tensor_metadata.default(view_1516, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_874 = None
	        mul_9384: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1517: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1512, [mul_9384, 1280]);  view_1512 = mul_9384 = None
	        permute_161: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1516, [1, 0]);  view_1516 = None
	        addmm_80: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_16_self_attn_q_proj_bias, view_1517, permute_161);  model_audio_tower_layers_16_self_attn_q_proj_bias = view_1517 = permute_161 = None
	        view_1518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_80, [sym_size_int, 1500, 1280]);  addmm_80 = None
	        mul_9391: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1518, 0.125);  view_1518 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1519: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_9391, [sym_size_int, 1500, 20, 64]);  mul_9391 = None
	        permute_162: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1519, [0, 2, 1, 3]);  view_1519 = None
	        clone_130: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_162, memory_format = torch.contiguous_format);  permute_162 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1520: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_14720, [sym_size_int, 1500, 1280])
	        amin_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1520, [2])
	        amax_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1520, [2]);  view_1520 = None
	        full_194: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_97, full_194);  amin_97 = full_194 = None
	        full_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_97, full_195);  amax_97 = full_195 = None
	        sub_4447: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_97, minimum_97);  maximum_97 = None
	        div_194: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4447, 255.0);  sub_4447 = None
	        clamp_min_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_194, 1.1920928955078125e-07);  div_194 = None
	        div_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_97, clamp_min_291);  minimum_97 = None
	        round_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_195);  div_195 = None
	        sub_4453: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_195);  round_195 = None
	        clamp_min_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4453, -128);  sub_4453 = None
	        clamp_max_194: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_292, 127);  clamp_min_292 = None
	        _assert_tensor_metadata_875 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_291, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_875 = None
	        _assert_tensor_metadata_876 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_194, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_876 = None
	        convert_element_type_582: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_194, torch.int8);  clamp_max_194 = None
	        view_1521: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_14720, [sym_size_int, 1500, 1280])
	        view_1522: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_291, [sym_size_int, 1500, 1])
	        view_1523: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_582, [sym_size_int, 1500, 1])
	        reciprocal_97: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1522);  view_1522 = None
	        mul_9445: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_97, 1.0);  reciprocal_97 = None
	        mul_9448: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1521, mul_9445);  view_1521 = mul_9445 = None
	        round_196: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9448);  mul_9448 = None
	        add_14959: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_196, view_1523);  round_196 = view_1523 = None
	        clamp_min_293: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14959, -128);  add_14959 = None
	        clamp_max_195: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_293, 127);  clamp_min_293 = None
	        view_1524: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_195, [sym_size_int, 1500, 1280]);  clamp_max_195 = None
	        _assert_tensor_metadata_877 = torch.ops.aten._assert_tensor_metadata.default(view_1524, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_877 = None
	        convert_element_type_583: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1524, torch.int8);  view_1524 = None
	        view_1525: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_583, [sym_size_int, 1500, 1280]);  convert_element_type_583 = None
	        view_1526: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_291, [sym_size_int, 1500, 1]);  clamp_min_291 = None
	        view_1527: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_582, [sym_size_int, 1500, 1]);  convert_element_type_582 = None
	        _assert_tensor_metadata_878 = torch.ops.aten._assert_tensor_metadata.default(view_1525, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_878 = None
	        convert_element_type_584: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1525, torch.float32);  view_1525 = None
	        _assert_tensor_metadata_879 = torch.ops.aten._assert_tensor_metadata.default(view_1527, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_879 = None
	        convert_element_type_585: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1527, torch.float32);  view_1527 = None
	        sub_4473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_584, convert_element_type_585);  convert_element_type_584 = convert_element_type_585 = None
	        mul_9470: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4473, view_1526);  sub_4473 = view_1526 = None
	        view_1528: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9470, [sym_size_int, 1500, 1280]);  mul_9470 = None
	        _assert_tensor_metadata_880 = torch.ops.aten._assert_tensor_metadata.default(view_1528, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_880 = None
	        view_1529: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1530: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1531: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_881 = torch.ops.aten._assert_tensor_metadata.default(view_1529, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_881 = None
	        convert_element_type_586: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1529, torch.float32);  view_1529 = None
	        _assert_tensor_metadata_882 = torch.ops.aten._assert_tensor_metadata.default(view_1531, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_882 = None
	        convert_element_type_587: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1531, torch.float32);  view_1531 = None
	        sub_4477: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_586, convert_element_type_587);  convert_element_type_586 = convert_element_type_587 = None
	        mul_9475: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4477, view_1530);  sub_4477 = view_1530 = None
	        view_1532: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9475, [1280, 1280]);  mul_9475 = None
	        _assert_tensor_metadata_883 = torch.ops.aten._assert_tensor_metadata.default(view_1532, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_883 = None
	        permute_163: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1532, [1, 0]);  view_1532 = None
	        mul_9478: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1533: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1528, [mul_9478, 1280]);  view_1528 = mul_9478 = None
	        mm_16: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1533, permute_163);  view_1533 = permute_163 = None
	        view_1534: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_16, [sym_size_int, 1500, 1280]);  mm_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1535: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1534, [sym_size_int, -1, 20, 64]);  view_1534 = None
	        permute_164: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1535, [0, 2, 1, 3]);  view_1535 = None
	        clone_131: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_164, memory_format = torch.contiguous_format);  permute_164 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1536: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_14720, [sym_size_int, 1500, 1280])
	        amin_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1536, [2])
	        amax_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1536, [2]);  view_1536 = None
	        full_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_98, full_196);  amin_98 = full_196 = None
	        full_197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_98, full_197);  amax_98 = full_197 = None
	        sub_4491: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_98, minimum_98);  maximum_98 = None
	        div_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4491, 255.0);  sub_4491 = None
	        clamp_min_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_196, 1.1920928955078125e-07);  div_196 = None
	        div_197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_98, clamp_min_294);  minimum_98 = None
	        round_197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_197);  div_197 = None
	        sub_4497: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_197);  round_197 = None
	        clamp_min_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4497, -128);  sub_4497 = None
	        clamp_max_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_295, 127);  clamp_min_295 = None
	        _assert_tensor_metadata_884 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_884 = None
	        _assert_tensor_metadata_885 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_196, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_885 = None
	        convert_element_type_588: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_196, torch.int8);  clamp_max_196 = None
	        view_1537: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_14720, [sym_size_int, 1500, 1280]);  add_14720 = None
	        view_1538: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_294, [sym_size_int, 1500, 1])
	        view_1539: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_588, [sym_size_int, 1500, 1])
	        reciprocal_98: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1538);  view_1538 = None
	        mul_9544: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_98, 1.0);  reciprocal_98 = None
	        mul_9547: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1537, mul_9544);  view_1537 = mul_9544 = None
	        round_198: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9547);  mul_9547 = None
	        add_15107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_198, view_1539);  round_198 = view_1539 = None
	        clamp_min_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15107, -128);  add_15107 = None
	        clamp_max_197: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_296, 127);  clamp_min_296 = None
	        view_1540: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_197, [sym_size_int, 1500, 1280]);  clamp_max_197 = None
	        _assert_tensor_metadata_886 = torch.ops.aten._assert_tensor_metadata.default(view_1540, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_886 = None
	        convert_element_type_589: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1540, torch.int8);  view_1540 = None
	        view_1541: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_589, [sym_size_int, 1500, 1280]);  convert_element_type_589 = None
	        view_1542: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_294, [sym_size_int, 1500, 1]);  clamp_min_294 = None
	        view_1543: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_588, [sym_size_int, 1500, 1]);  convert_element_type_588 = None
	        _assert_tensor_metadata_887 = torch.ops.aten._assert_tensor_metadata.default(view_1541, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_887 = None
	        convert_element_type_590: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1541, torch.float32);  view_1541 = None
	        _assert_tensor_metadata_888 = torch.ops.aten._assert_tensor_metadata.default(view_1543, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_888 = None
	        convert_element_type_591: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1543, torch.float32);  view_1543 = None
	        sub_4517: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_590, convert_element_type_591);  convert_element_type_590 = convert_element_type_591 = None
	        mul_9569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4517, view_1542);  sub_4517 = view_1542 = None
	        view_1544: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9569, [sym_size_int, 1500, 1280]);  mul_9569 = None
	        _assert_tensor_metadata_889 = torch.ops.aten._assert_tensor_metadata.default(view_1544, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_889 = None
	        view_1545: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1546: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1547: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_890 = torch.ops.aten._assert_tensor_metadata.default(view_1545, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_890 = None
	        convert_element_type_592: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1545, torch.float32);  view_1545 = None
	        _assert_tensor_metadata_891 = torch.ops.aten._assert_tensor_metadata.default(view_1547, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_891 = None
	        convert_element_type_593: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1547, torch.float32);  view_1547 = None
	        sub_4521: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_592, convert_element_type_593);  convert_element_type_592 = convert_element_type_593 = None
	        mul_9574: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4521, view_1546);  sub_4521 = view_1546 = None
	        view_1548: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9574, [1280, 1280]);  mul_9574 = None
	        _assert_tensor_metadata_892 = torch.ops.aten._assert_tensor_metadata.default(view_1548, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_892 = None
	        mul_9579: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1549: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1544, [mul_9579, 1280]);  view_1544 = mul_9579 = None
	        permute_165: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1548, [1, 0]);  view_1548 = None
	        addmm_81: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_16_self_attn_v_proj_bias, view_1549, permute_165);  model_audio_tower_layers_16_self_attn_v_proj_bias = view_1549 = permute_165 = None
	        view_1550: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_81, [sym_size_int, 1500, 1280]);  addmm_81 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1551: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1550, [sym_size_int, -1, 20, 64]);  view_1550 = None
	        permute_166: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1551, [0, 2, 1, 3]);  view_1551 = None
	        clone_132: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_166, memory_format = torch.contiguous_format);  permute_166 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_16 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_130, clone_131, clone_132, None, False, scale = 1.0);  clone_130 = clone_131 = clone_132 = None
	        getitem_130: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_16[0];  _scaled_dot_product_efficient_attention_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_167: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_130, [0, 2, 1, 3]);  getitem_130 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_167, [sym_size_int, 1500, -1]);  permute_167 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1553: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1552, [sym_size_int, 1500, 1280])
	        amin_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1553, [2])
	        amax_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1553, [2]);  view_1553 = None
	        full_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_99, full_198);  amin_99 = full_198 = None
	        full_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_99, full_199);  amax_99 = full_199 = None
	        sub_4539: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_99, minimum_99);  maximum_99 = None
	        div_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4539, 255.0);  sub_4539 = None
	        clamp_min_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_198, 1.1920928955078125e-07);  div_198 = None
	        div_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_99, clamp_min_297);  minimum_99 = None
	        round_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_199);  div_199 = None
	        sub_4545: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_199);  round_199 = None
	        clamp_min_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4545, -128);  sub_4545 = None
	        clamp_max_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_298, 127);  clamp_min_298 = None
	        _assert_tensor_metadata_893 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_297, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_893 = None
	        _assert_tensor_metadata_894 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_198, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_894 = None
	        convert_element_type_594: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_198, torch.int8);  clamp_max_198 = None
	        view_1554: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1552, [sym_size_int, 1500, 1280]);  view_1552 = None
	        view_1555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_297, [sym_size_int, 1500, 1])
	        view_1556: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_594, [sym_size_int, 1500, 1])
	        reciprocal_99: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1555);  view_1555 = None
	        mul_9649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_99, 1.0);  reciprocal_99 = None
	        mul_9652: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1554, mul_9649);  view_1554 = mul_9649 = None
	        round_200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9652);  mul_9652 = None
	        add_15271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_200, view_1556);  round_200 = view_1556 = None
	        clamp_min_299: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15271, -128);  add_15271 = None
	        clamp_max_199: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_299, 127);  clamp_min_299 = None
	        view_1557: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_199, [sym_size_int, 1500, 1280]);  clamp_max_199 = None
	        _assert_tensor_metadata_895 = torch.ops.aten._assert_tensor_metadata.default(view_1557, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_895 = None
	        convert_element_type_595: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1557, torch.int8);  view_1557 = None
	        view_1558: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_595, [sym_size_int, 1500, 1280]);  convert_element_type_595 = None
	        view_1559: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_297, [sym_size_int, 1500, 1]);  clamp_min_297 = None
	        view_1560: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_594, [sym_size_int, 1500, 1]);  convert_element_type_594 = None
	        _assert_tensor_metadata_896 = torch.ops.aten._assert_tensor_metadata.default(view_1558, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_896 = None
	        convert_element_type_596: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1558, torch.float32);  view_1558 = None
	        _assert_tensor_metadata_897 = torch.ops.aten._assert_tensor_metadata.default(view_1560, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_897 = None
	        convert_element_type_597: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1560, torch.float32);  view_1560 = None
	        sub_4565: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_596, convert_element_type_597);  convert_element_type_596 = convert_element_type_597 = None
	        mul_9674: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4565, view_1559);  sub_4565 = view_1559 = None
	        view_1561: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9674, [sym_size_int, 1500, 1280]);  mul_9674 = None
	        _assert_tensor_metadata_898 = torch.ops.aten._assert_tensor_metadata.default(view_1561, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_898 = None
	        view_1562: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1563: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1564: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_899 = torch.ops.aten._assert_tensor_metadata.default(view_1562, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_899 = None
	        convert_element_type_598: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1562, torch.float32);  view_1562 = None
	        _assert_tensor_metadata_900 = torch.ops.aten._assert_tensor_metadata.default(view_1564, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_900 = None
	        convert_element_type_599: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1564, torch.float32);  view_1564 = None
	        sub_4569: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_598, convert_element_type_599);  convert_element_type_598 = convert_element_type_599 = None
	        mul_9679: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4569, view_1563);  sub_4569 = view_1563 = None
	        view_1565: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9679, [1280, 1280]);  mul_9679 = None
	        _assert_tensor_metadata_901 = torch.ops.aten._assert_tensor_metadata.default(view_1565, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_901 = None
	        mul_9684: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1566: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1561, [mul_9684, 1280]);  view_1561 = mul_9684 = None
	        permute_168: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1565, [1, 0]);  view_1565 = None
	        addmm_82: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_16_self_attn_out_proj_bias, view_1566, permute_168);  model_audio_tower_layers_16_self_attn_out_proj_bias = view_1566 = permute_168 = None
	        view_1567: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_82, [sym_size_int, 1500, 1280]);  addmm_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_133: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1567);  view_1567 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_15334: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_14714, clone_133);  add_14714 = clone_133 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_15334, memory_format = torch.contiguous_format)
	        var_mean_33 = torch.ops.aten.var_mean.correction(clone_134, [2], correction = 0, keepdim = True)
	        getitem_134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_33[0]
	        getitem_135: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_33[1];  var_mean_33 = None
	        add_15339: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_134, 1e-05);  getitem_134 = None
	        rsqrt_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_15339);  add_15339 = None
	        sub_4575: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_134, getitem_135);  clone_134 = getitem_135 = None
	        mul_9695: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4575, rsqrt_33);  sub_4575 = rsqrt_33 = None
	        mul_9696: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9695, model_audio_tower_layers_16_final_layer_norm_weight);  mul_9695 = model_audio_tower_layers_16_final_layer_norm_weight = None
	        add_15340: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9696, model_audio_tower_layers_16_final_layer_norm_bias);  mul_9696 = model_audio_tower_layers_16_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1568: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_15340, [sym_size_int, 1500, 1280])
	        amin_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1568, [2])
	        amax_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1568, [2]);  view_1568 = None
	        full_200: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_100, full_200);  amin_100 = full_200 = None
	        full_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_100, full_201);  amax_100 = full_201 = None
	        sub_4586: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_100, minimum_100);  maximum_100 = None
	        div_200: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4586, 255.0);  sub_4586 = None
	        clamp_min_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_200, 1.1920928955078125e-07);  div_200 = None
	        div_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_100, clamp_min_300);  minimum_100 = None
	        round_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_201);  div_201 = None
	        sub_4592: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_201);  round_201 = None
	        clamp_min_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4592, -128);  sub_4592 = None
	        clamp_max_200: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_301, 127);  clamp_min_301 = None
	        _assert_tensor_metadata_902 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_300, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_902 = None
	        _assert_tensor_metadata_903 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_200, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_903 = None
	        convert_element_type_600: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_200, torch.int8);  clamp_max_200 = None
	        view_1569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_15340, [sym_size_int, 1500, 1280]);  add_15340 = None
	        view_1570: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_300, [sym_size_int, 1500, 1])
	        view_1571: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_600, [sym_size_int, 1500, 1])
	        reciprocal_100: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1570);  view_1570 = None
	        mul_9744: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_100, 1.0);  reciprocal_100 = None
	        mul_9747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1569, mul_9744);  view_1569 = mul_9744 = None
	        round_202: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9747);  mul_9747 = None
	        add_15427: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_202, view_1571);  round_202 = view_1571 = None
	        clamp_min_302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15427, -128);  add_15427 = None
	        clamp_max_201: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_302, 127);  clamp_min_302 = None
	        view_1572: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_201, [sym_size_int, 1500, 1280]);  clamp_max_201 = None
	        _assert_tensor_metadata_904 = torch.ops.aten._assert_tensor_metadata.default(view_1572, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_904 = None
	        convert_element_type_601: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1572, torch.int8);  view_1572 = None
	        view_1573: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_601, [sym_size_int, 1500, 1280]);  convert_element_type_601 = None
	        view_1574: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_300, [sym_size_int, 1500, 1]);  clamp_min_300 = None
	        view_1575: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_600, [sym_size_int, 1500, 1]);  convert_element_type_600 = None
	        _assert_tensor_metadata_905 = torch.ops.aten._assert_tensor_metadata.default(view_1573, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_905 = None
	        convert_element_type_602: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1573, torch.float32);  view_1573 = None
	        _assert_tensor_metadata_906 = torch.ops.aten._assert_tensor_metadata.default(view_1575, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_906 = None
	        convert_element_type_603: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1575, torch.float32);  view_1575 = None
	        sub_4612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_602, convert_element_type_603);  convert_element_type_602 = convert_element_type_603 = None
	        mul_9769: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4612, view_1574);  sub_4612 = view_1574 = None
	        view_1576: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9769, [sym_size_int, 1500, 1280]);  mul_9769 = None
	        _assert_tensor_metadata_907 = torch.ops.aten._assert_tensor_metadata.default(view_1576, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_907 = None
	        view_1577: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_16_fc1_parametrizations_weight_original0 = None
	        view_1578: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_16_fc1_parametrizations_weight_original1 = None
	        view_1579: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_16_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_908 = torch.ops.aten._assert_tensor_metadata.default(view_1577, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_908 = None
	        convert_element_type_604: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1577, torch.float32);  view_1577 = None
	        _assert_tensor_metadata_909 = torch.ops.aten._assert_tensor_metadata.default(view_1579, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_909 = None
	        convert_element_type_605: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1579, torch.float32);  view_1579 = None
	        sub_4616: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_604, convert_element_type_605);  convert_element_type_604 = convert_element_type_605 = None
	        mul_9774: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4616, view_1578);  sub_4616 = view_1578 = None
	        view_1580: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9774, [5120, 1280]);  mul_9774 = None
	        _assert_tensor_metadata_910 = torch.ops.aten._assert_tensor_metadata.default(view_1580, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_910 = None
	        mul_9779: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1581: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1576, [mul_9779, 1280]);  view_1576 = mul_9779 = None
	        permute_169: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1580, [1, 0]);  view_1580 = None
	        addmm_83: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_16_fc1_bias, view_1581, permute_169);  model_audio_tower_layers_16_fc1_bias = view_1581 = permute_169 = None
	        view_1582: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_83, [sym_size_int, 1500, 5120]);  addmm_83 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_9786: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1582, 0.5)
	        mul_9787: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1582, 0.7071067811865476);  view_1582 = None
	        erf_18: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_9787);  mul_9787 = None
	        add_15486: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_18, 1);  erf_18 = None
	        mul_9788: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9786, add_15486);  mul_9786 = add_15486 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_135: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_9788);  mul_9788 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1583: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_135, [sym_size_int, 1500, 5120])
	        amin_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1583, [2])
	        amax_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1583, [2]);  view_1583 = None
	        full_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_101, full_202);  amin_101 = full_202 = None
	        full_203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_101, full_203);  amax_101 = full_203 = None
	        sub_4629: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_101, minimum_101);  maximum_101 = None
	        div_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4629, 255.0);  sub_4629 = None
	        clamp_min_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_202, 1.1920928955078125e-07);  div_202 = None
	        div_203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_101, clamp_min_303);  minimum_101 = None
	        round_203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_203);  div_203 = None
	        sub_4635: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_203);  round_203 = None
	        clamp_min_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4635, -128);  sub_4635 = None
	        clamp_max_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_304, 127);  clamp_min_304 = None
	        _assert_tensor_metadata_911 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_303, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_911 = None
	        _assert_tensor_metadata_912 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_202, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_912 = None
	        convert_element_type_606: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_202, torch.int8);  clamp_max_202 = None
	        view_1584: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_135, [sym_size_int, 1500, 5120]);  clone_135 = None
	        view_1585: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_303, [sym_size_int, 1500, 1])
	        view_1586: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_606, [sym_size_int, 1500, 1])
	        reciprocal_101: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1585);  view_1585 = None
	        mul_9834: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_101, 1.0);  reciprocal_101 = None
	        mul_9837: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1584, mul_9834);  view_1584 = mul_9834 = None
	        round_204: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_9837);  mul_9837 = None
	        add_15569: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_204, view_1586);  round_204 = view_1586 = None
	        clamp_min_305: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15569, -128);  add_15569 = None
	        clamp_max_203: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_305, 127);  clamp_min_305 = None
	        view_1587: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_203, [sym_size_int, 1500, 5120]);  clamp_max_203 = None
	        _assert_tensor_metadata_913 = torch.ops.aten._assert_tensor_metadata.default(view_1587, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_913 = None
	        convert_element_type_607: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1587, torch.int8);  view_1587 = None
	        view_1588: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_607, [sym_size_int, 1500, 5120]);  convert_element_type_607 = None
	        view_1589: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_303, [sym_size_int, 1500, 1]);  clamp_min_303 = None
	        view_1590: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_606, [sym_size_int, 1500, 1]);  convert_element_type_606 = None
	        _assert_tensor_metadata_914 = torch.ops.aten._assert_tensor_metadata.default(view_1588, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_914 = None
	        convert_element_type_608: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1588, torch.float32);  view_1588 = None
	        _assert_tensor_metadata_915 = torch.ops.aten._assert_tensor_metadata.default(view_1590, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_915 = None
	        convert_element_type_609: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1590, torch.float32);  view_1590 = None
	        sub_4655: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_608, convert_element_type_609);  convert_element_type_608 = convert_element_type_609 = None
	        mul_9859: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4655, view_1589);  sub_4655 = view_1589 = None
	        view_1591: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_9859, [sym_size_int, 1500, 5120]);  mul_9859 = None
	        _assert_tensor_metadata_916 = torch.ops.aten._assert_tensor_metadata.default(view_1591, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_916 = None
	        view_1592: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_16_fc2_parametrizations_weight_original0 = None
	        view_1593: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_16_fc2_parametrizations_weight_original1 = None
	        view_1594: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_16_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_917 = torch.ops.aten._assert_tensor_metadata.default(view_1592, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_917 = None
	        convert_element_type_610: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1592, torch.float32);  view_1592 = None
	        _assert_tensor_metadata_918 = torch.ops.aten._assert_tensor_metadata.default(view_1594, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_918 = None
	        convert_element_type_611: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1594, torch.float32);  view_1594 = None
	        sub_4659: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_610, convert_element_type_611);  convert_element_type_610 = convert_element_type_611 = None
	        mul_9864: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4659, view_1593);  sub_4659 = view_1593 = None
	        view_1595: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_9864, [1280, 5120]);  mul_9864 = None
	        _assert_tensor_metadata_919 = torch.ops.aten._assert_tensor_metadata.default(view_1595, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_919 = None
	        mul_9869: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1596: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1591, [mul_9869, 5120]);  view_1591 = mul_9869 = None
	        permute_170: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1595, [1, 0]);  view_1595 = None
	        addmm_84: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_16_fc2_bias, view_1596, permute_170);  model_audio_tower_layers_16_fc2_bias = view_1596 = permute_170 = None
	        view_1597: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_84, [sym_size_int, 1500, 1280]);  addmm_84 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_136: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1597);  view_1597 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_15632: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_15334, clone_136);  add_15334 = clone_136 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_15632, memory_format = torch.contiguous_format)
	        var_mean_34 = torch.ops.aten.var_mean.correction(clone_137, [2], correction = 0, keepdim = True)
	        getitem_136: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_34[0]
	        getitem_137: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_34[1];  var_mean_34 = None
	        add_15637: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_136, 1e-05);  getitem_136 = None
	        rsqrt_34: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_15637);  add_15637 = None
	        sub_4665: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_137, getitem_137);  clone_137 = getitem_137 = None
	        mul_9880: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4665, rsqrt_34);  sub_4665 = rsqrt_34 = None
	        mul_9881: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9880, model_audio_tower_layers_17_self_attn_layer_norm_weight);  mul_9880 = model_audio_tower_layers_17_self_attn_layer_norm_weight = None
	        add_15638: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9881, model_audio_tower_layers_17_self_attn_layer_norm_bias);  mul_9881 = model_audio_tower_layers_17_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1598: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_15638, [sym_size_int, 1500, 1280])
	        amin_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1598, [2])
	        amax_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1598, [2]);  view_1598 = None
	        full_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_102, full_204);  amin_102 = full_204 = None
	        full_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_102, full_205);  amax_102 = full_205 = None
	        sub_4676: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_102, minimum_102);  maximum_102 = None
	        div_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4676, 255.0);  sub_4676 = None
	        clamp_min_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_204, 1.1920928955078125e-07);  div_204 = None
	        div_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_102, clamp_min_306);  minimum_102 = None
	        round_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_205);  div_205 = None
	        sub_4682: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_205);  round_205 = None
	        clamp_min_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4682, -128);  sub_4682 = None
	        clamp_max_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_307, 127);  clamp_min_307 = None
	        _assert_tensor_metadata_920 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_306, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_920 = None
	        _assert_tensor_metadata_921 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_204, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_921 = None
	        convert_element_type_612: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_204, torch.int8);  clamp_max_204 = None
	        view_1599: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_15638, [sym_size_int, 1500, 1280])
	        view_1600: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_306, [sym_size_int, 1500, 1])
	        view_1601: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_612, [sym_size_int, 1500, 1])
	        reciprocal_102: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1600);  view_1600 = None
	        mul_9929: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_102, 1.0);  reciprocal_102 = None
	        mul_9932: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1599, mul_9929);  view_1599 = mul_9929 = None
	        round_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9932);  mul_9932 = None
	        add_15725: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_206, view_1601);  round_206 = view_1601 = None
	        clamp_min_308: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15725, -128);  add_15725 = None
	        clamp_max_205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_308, 127);  clamp_min_308 = None
	        view_1602: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_205, [sym_size_int, 1500, 1280]);  clamp_max_205 = None
	        _assert_tensor_metadata_922 = torch.ops.aten._assert_tensor_metadata.default(view_1602, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_922 = None
	        convert_element_type_613: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1602, torch.int8);  view_1602 = None
	        view_1603: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_613, [sym_size_int, 1500, 1280]);  convert_element_type_613 = None
	        view_1604: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_306, [sym_size_int, 1500, 1]);  clamp_min_306 = None
	        view_1605: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_612, [sym_size_int, 1500, 1]);  convert_element_type_612 = None
	        _assert_tensor_metadata_923 = torch.ops.aten._assert_tensor_metadata.default(view_1603, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_923 = None
	        convert_element_type_614: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1603, torch.float32);  view_1603 = None
	        _assert_tensor_metadata_924 = torch.ops.aten._assert_tensor_metadata.default(view_1605, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_924 = None
	        convert_element_type_615: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1605, torch.float32);  view_1605 = None
	        sub_4702: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_614, convert_element_type_615);  convert_element_type_614 = convert_element_type_615 = None
	        mul_9954: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4702, view_1604);  sub_4702 = view_1604 = None
	        view_1606: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9954, [sym_size_int, 1500, 1280]);  mul_9954 = None
	        _assert_tensor_metadata_925 = torch.ops.aten._assert_tensor_metadata.default(view_1606, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_925 = None
	        view_1607: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1608: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1609: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_926 = torch.ops.aten._assert_tensor_metadata.default(view_1607, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_926 = None
	        convert_element_type_616: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1607, torch.float32);  view_1607 = None
	        _assert_tensor_metadata_927 = torch.ops.aten._assert_tensor_metadata.default(view_1609, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_927 = None
	        convert_element_type_617: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1609, torch.float32);  view_1609 = None
	        sub_4706: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_616, convert_element_type_617);  convert_element_type_616 = convert_element_type_617 = None
	        mul_9959: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4706, view_1608);  sub_4706 = view_1608 = None
	        view_1610: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9959, [1280, 1280]);  mul_9959 = None
	        _assert_tensor_metadata_928 = torch.ops.aten._assert_tensor_metadata.default(view_1610, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_928 = None
	        mul_9964: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1611: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1606, [mul_9964, 1280]);  view_1606 = mul_9964 = None
	        permute_171: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1610, [1, 0]);  view_1610 = None
	        addmm_85: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_17_self_attn_q_proj_bias, view_1611, permute_171);  model_audio_tower_layers_17_self_attn_q_proj_bias = view_1611 = permute_171 = None
	        view_1612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_85, [sym_size_int, 1500, 1280]);  addmm_85 = None
	        mul_9971: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1612, 0.125);  view_1612 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1613: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_9971, [sym_size_int, 1500, 20, 64]);  mul_9971 = None
	        permute_172: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1613, [0, 2, 1, 3]);  view_1613 = None
	        clone_138: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_172, memory_format = torch.contiguous_format);  permute_172 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1614: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_15638, [sym_size_int, 1500, 1280])
	        amin_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1614, [2])
	        amax_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1614, [2]);  view_1614 = None
	        full_206: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_103, full_206);  amin_103 = full_206 = None
	        full_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_103, full_207);  amax_103 = full_207 = None
	        sub_4721: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_103, minimum_103);  maximum_103 = None
	        div_206: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4721, 255.0);  sub_4721 = None
	        clamp_min_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_206, 1.1920928955078125e-07);  div_206 = None
	        div_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_103, clamp_min_309);  minimum_103 = None
	        round_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_207);  div_207 = None
	        sub_4727: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_207);  round_207 = None
	        clamp_min_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4727, -128);  sub_4727 = None
	        clamp_max_206: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_310, 127);  clamp_min_310 = None
	        _assert_tensor_metadata_929 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_309, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_929 = None
	        _assert_tensor_metadata_930 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_206, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_930 = None
	        convert_element_type_618: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_206, torch.int8);  clamp_max_206 = None
	        view_1615: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_15638, [sym_size_int, 1500, 1280])
	        view_1616: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_309, [sym_size_int, 1500, 1])
	        view_1617: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_618, [sym_size_int, 1500, 1])
	        reciprocal_103: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1616);  view_1616 = None
	        mul_10025: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_103, 1.0);  reciprocal_103 = None
	        mul_10028: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1615, mul_10025);  view_1615 = mul_10025 = None
	        round_208: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10028);  mul_10028 = None
	        add_15877: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_208, view_1617);  round_208 = view_1617 = None
	        clamp_min_311: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15877, -128);  add_15877 = None
	        clamp_max_207: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_311, 127);  clamp_min_311 = None
	        view_1618: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_207, [sym_size_int, 1500, 1280]);  clamp_max_207 = None
	        _assert_tensor_metadata_931 = torch.ops.aten._assert_tensor_metadata.default(view_1618, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_931 = None
	        convert_element_type_619: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1618, torch.int8);  view_1618 = None
	        view_1619: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_619, [sym_size_int, 1500, 1280]);  convert_element_type_619 = None
	        view_1620: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_309, [sym_size_int, 1500, 1]);  clamp_min_309 = None
	        view_1621: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_618, [sym_size_int, 1500, 1]);  convert_element_type_618 = None
	        _assert_tensor_metadata_932 = torch.ops.aten._assert_tensor_metadata.default(view_1619, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_932 = None
	        convert_element_type_620: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1619, torch.float32);  view_1619 = None
	        _assert_tensor_metadata_933 = torch.ops.aten._assert_tensor_metadata.default(view_1621, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_933 = None
	        convert_element_type_621: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1621, torch.float32);  view_1621 = None
	        sub_4747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_620, convert_element_type_621);  convert_element_type_620 = convert_element_type_621 = None
	        mul_10050: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4747, view_1620);  sub_4747 = view_1620 = None
	        view_1622: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10050, [sym_size_int, 1500, 1280]);  mul_10050 = None
	        _assert_tensor_metadata_934 = torch.ops.aten._assert_tensor_metadata.default(view_1622, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_934 = None
	        view_1623: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1624: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1625: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_935 = torch.ops.aten._assert_tensor_metadata.default(view_1623, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_935 = None
	        convert_element_type_622: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1623, torch.float32);  view_1623 = None
	        _assert_tensor_metadata_936 = torch.ops.aten._assert_tensor_metadata.default(view_1625, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_936 = None
	        convert_element_type_623: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1625, torch.float32);  view_1625 = None
	        sub_4751: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_622, convert_element_type_623);  convert_element_type_622 = convert_element_type_623 = None
	        mul_10055: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4751, view_1624);  sub_4751 = view_1624 = None
	        view_1626: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10055, [1280, 1280]);  mul_10055 = None
	        _assert_tensor_metadata_937 = torch.ops.aten._assert_tensor_metadata.default(view_1626, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_937 = None
	        permute_173: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1626, [1, 0]);  view_1626 = None
	        mul_10058: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1627: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1622, [mul_10058, 1280]);  view_1622 = mul_10058 = None
	        mm_17: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1627, permute_173);  view_1627 = permute_173 = None
	        view_1628: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_17, [sym_size_int, 1500, 1280]);  mm_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1629: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1628, [sym_size_int, -1, 20, 64]);  view_1628 = None
	        permute_174: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1629, [0, 2, 1, 3]);  view_1629 = None
	        clone_139: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_174, memory_format = torch.contiguous_format);  permute_174 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1630: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_15638, [sym_size_int, 1500, 1280])
	        amin_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1630, [2])
	        amax_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1630, [2]);  view_1630 = None
	        full_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_104, full_208);  amin_104 = full_208 = None
	        full_209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_104, full_209);  amax_104 = full_209 = None
	        sub_4765: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_104, minimum_104);  maximum_104 = None
	        div_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4765, 255.0);  sub_4765 = None
	        clamp_min_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_208, 1.1920928955078125e-07);  div_208 = None
	        div_209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_104, clamp_min_312);  minimum_104 = None
	        round_209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_209);  div_209 = None
	        sub_4771: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_209);  round_209 = None
	        clamp_min_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4771, -128);  sub_4771 = None
	        clamp_max_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_313, 127);  clamp_min_313 = None
	        _assert_tensor_metadata_938 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_312, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_938 = None
	        _assert_tensor_metadata_939 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_208, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_939 = None
	        convert_element_type_624: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_208, torch.int8);  clamp_max_208 = None
	        view_1631: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_15638, [sym_size_int, 1500, 1280]);  add_15638 = None
	        view_1632: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_312, [sym_size_int, 1500, 1])
	        view_1633: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_624, [sym_size_int, 1500, 1])
	        reciprocal_104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1632);  view_1632 = None
	        mul_10124: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_104, 1.0);  reciprocal_104 = None
	        mul_10127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1631, mul_10124);  view_1631 = mul_10124 = None
	        round_210: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10127);  mul_10127 = None
	        add_16025: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_210, view_1633);  round_210 = view_1633 = None
	        clamp_min_314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16025, -128);  add_16025 = None
	        clamp_max_209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_314, 127);  clamp_min_314 = None
	        view_1634: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_209, [sym_size_int, 1500, 1280]);  clamp_max_209 = None
	        _assert_tensor_metadata_940 = torch.ops.aten._assert_tensor_metadata.default(view_1634, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_940 = None
	        convert_element_type_625: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1634, torch.int8);  view_1634 = None
	        view_1635: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_625, [sym_size_int, 1500, 1280]);  convert_element_type_625 = None
	        view_1636: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_312, [sym_size_int, 1500, 1]);  clamp_min_312 = None
	        view_1637: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_624, [sym_size_int, 1500, 1]);  convert_element_type_624 = None
	        _assert_tensor_metadata_941 = torch.ops.aten._assert_tensor_metadata.default(view_1635, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_941 = None
	        convert_element_type_626: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1635, torch.float32);  view_1635 = None
	        _assert_tensor_metadata_942 = torch.ops.aten._assert_tensor_metadata.default(view_1637, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_942 = None
	        convert_element_type_627: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1637, torch.float32);  view_1637 = None
	        sub_4791: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_626, convert_element_type_627);  convert_element_type_626 = convert_element_type_627 = None
	        mul_10149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4791, view_1636);  sub_4791 = view_1636 = None
	        view_1638: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10149, [sym_size_int, 1500, 1280]);  mul_10149 = None
	        _assert_tensor_metadata_943 = torch.ops.aten._assert_tensor_metadata.default(view_1638, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_943 = None
	        view_1639: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1640: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1641: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_944 = torch.ops.aten._assert_tensor_metadata.default(view_1639, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_944 = None
	        convert_element_type_628: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1639, torch.float32);  view_1639 = None
	        _assert_tensor_metadata_945 = torch.ops.aten._assert_tensor_metadata.default(view_1641, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_945 = None
	        convert_element_type_629: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1641, torch.float32);  view_1641 = None
	        sub_4795: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_628, convert_element_type_629);  convert_element_type_628 = convert_element_type_629 = None
	        mul_10154: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4795, view_1640);  sub_4795 = view_1640 = None
	        view_1642: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10154, [1280, 1280]);  mul_10154 = None
	        _assert_tensor_metadata_946 = torch.ops.aten._assert_tensor_metadata.default(view_1642, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_946 = None
	        mul_10159: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1643: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1638, [mul_10159, 1280]);  view_1638 = mul_10159 = None
	        permute_175: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1642, [1, 0]);  view_1642 = None
	        addmm_86: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_17_self_attn_v_proj_bias, view_1643, permute_175);  model_audio_tower_layers_17_self_attn_v_proj_bias = view_1643 = permute_175 = None
	        view_1644: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_86, [sym_size_int, 1500, 1280]);  addmm_86 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1645: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1644, [sym_size_int, -1, 20, 64]);  view_1644 = None
	        permute_176: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1645, [0, 2, 1, 3]);  view_1645 = None
	        clone_140: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_176, memory_format = torch.contiguous_format);  permute_176 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_17 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_138, clone_139, clone_140, None, False, scale = 1.0);  clone_138 = clone_139 = clone_140 = None
	        getitem_138: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_17[0];  _scaled_dot_product_efficient_attention_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_177: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_138, [0, 2, 1, 3]);  getitem_138 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1646: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_177, [sym_size_int, 1500, -1]);  permute_177 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1647: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1646, [sym_size_int, 1500, 1280])
	        amin_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1647, [2])
	        amax_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1647, [2]);  view_1647 = None
	        full_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_105, full_210);  amin_105 = full_210 = None
	        full_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_105, full_211);  amax_105 = full_211 = None
	        sub_4813: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_105, minimum_105);  maximum_105 = None
	        div_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4813, 255.0);  sub_4813 = None
	        clamp_min_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_210, 1.1920928955078125e-07);  div_210 = None
	        div_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_105, clamp_min_315);  minimum_105 = None
	        round_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_211);  div_211 = None
	        sub_4819: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_211);  round_211 = None
	        clamp_min_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4819, -128);  sub_4819 = None
	        clamp_max_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_316, 127);  clamp_min_316 = None
	        _assert_tensor_metadata_947 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_315, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_947 = None
	        _assert_tensor_metadata_948 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_210, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_948 = None
	        convert_element_type_630: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_210, torch.int8);  clamp_max_210 = None
	        view_1648: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1646, [sym_size_int, 1500, 1280]);  view_1646 = None
	        view_1649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_315, [sym_size_int, 1500, 1])
	        view_1650: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_630, [sym_size_int, 1500, 1])
	        reciprocal_105: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1649);  view_1649 = None
	        mul_10229: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_105, 1.0);  reciprocal_105 = None
	        mul_10232: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1648, mul_10229);  view_1648 = mul_10229 = None
	        round_212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10232);  mul_10232 = None
	        add_16189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_212, view_1650);  round_212 = view_1650 = None
	        clamp_min_317: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16189, -128);  add_16189 = None
	        clamp_max_211: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_317, 127);  clamp_min_317 = None
	        view_1651: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_211, [sym_size_int, 1500, 1280]);  clamp_max_211 = None
	        _assert_tensor_metadata_949 = torch.ops.aten._assert_tensor_metadata.default(view_1651, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_949 = None
	        convert_element_type_631: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1651, torch.int8);  view_1651 = None
	        view_1652: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_631, [sym_size_int, 1500, 1280]);  convert_element_type_631 = None
	        view_1653: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_315, [sym_size_int, 1500, 1]);  clamp_min_315 = None
	        view_1654: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_630, [sym_size_int, 1500, 1]);  convert_element_type_630 = None
	        _assert_tensor_metadata_950 = torch.ops.aten._assert_tensor_metadata.default(view_1652, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_950 = None
	        convert_element_type_632: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1652, torch.float32);  view_1652 = None
	        _assert_tensor_metadata_951 = torch.ops.aten._assert_tensor_metadata.default(view_1654, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_951 = None
	        convert_element_type_633: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1654, torch.float32);  view_1654 = None
	        sub_4839: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_632, convert_element_type_633);  convert_element_type_632 = convert_element_type_633 = None
	        mul_10254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4839, view_1653);  sub_4839 = view_1653 = None
	        view_1655: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10254, [sym_size_int, 1500, 1280]);  mul_10254 = None
	        _assert_tensor_metadata_952 = torch.ops.aten._assert_tensor_metadata.default(view_1655, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_952 = None
	        view_1656: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1657: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1658: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_953 = torch.ops.aten._assert_tensor_metadata.default(view_1656, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_953 = None
	        convert_element_type_634: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1656, torch.float32);  view_1656 = None
	        _assert_tensor_metadata_954 = torch.ops.aten._assert_tensor_metadata.default(view_1658, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_954 = None
	        convert_element_type_635: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1658, torch.float32);  view_1658 = None
	        sub_4843: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_634, convert_element_type_635);  convert_element_type_634 = convert_element_type_635 = None
	        mul_10259: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4843, view_1657);  sub_4843 = view_1657 = None
	        view_1659: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10259, [1280, 1280]);  mul_10259 = None
	        _assert_tensor_metadata_955 = torch.ops.aten._assert_tensor_metadata.default(view_1659, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_955 = None
	        mul_10264: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1660: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1655, [mul_10264, 1280]);  view_1655 = mul_10264 = None
	        permute_178: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1659, [1, 0]);  view_1659 = None
	        addmm_87: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_17_self_attn_out_proj_bias, view_1660, permute_178);  model_audio_tower_layers_17_self_attn_out_proj_bias = view_1660 = permute_178 = None
	        view_1661: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_87, [sym_size_int, 1500, 1280]);  addmm_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_141: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1661);  view_1661 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_16252: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_15632, clone_141);  add_15632 = clone_141 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_16252, memory_format = torch.contiguous_format)
	        var_mean_35 = torch.ops.aten.var_mean.correction(clone_142, [2], correction = 0, keepdim = True)
	        getitem_142: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_35[0]
	        getitem_143: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_35[1];  var_mean_35 = None
	        add_16257: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_142, 1e-05);  getitem_142 = None
	        rsqrt_35: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_16257);  add_16257 = None
	        sub_4849: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_142, getitem_143);  clone_142 = getitem_143 = None
	        mul_10275: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4849, rsqrt_35);  sub_4849 = rsqrt_35 = None
	        mul_10276: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10275, model_audio_tower_layers_17_final_layer_norm_weight);  mul_10275 = model_audio_tower_layers_17_final_layer_norm_weight = None
	        add_16258: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_10276, model_audio_tower_layers_17_final_layer_norm_bias);  mul_10276 = model_audio_tower_layers_17_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1662: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_16258, [sym_size_int, 1500, 1280])
	        amin_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1662, [2])
	        amax_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1662, [2]);  view_1662 = None
	        full_212: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_106, full_212);  amin_106 = full_212 = None
	        full_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_106, full_213);  amax_106 = full_213 = None
	        sub_4860: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_106, minimum_106);  maximum_106 = None
	        div_212: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4860, 255.0);  sub_4860 = None
	        clamp_min_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_212, 1.1920928955078125e-07);  div_212 = None
	        div_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_106, clamp_min_318);  minimum_106 = None
	        round_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_213);  div_213 = None
	        sub_4866: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_213);  round_213 = None
	        clamp_min_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4866, -128);  sub_4866 = None
	        clamp_max_212: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_319, 127);  clamp_min_319 = None
	        _assert_tensor_metadata_956 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_318, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_956 = None
	        _assert_tensor_metadata_957 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_212, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_957 = None
	        convert_element_type_636: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_212, torch.int8);  clamp_max_212 = None
	        view_1663: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_16258, [sym_size_int, 1500, 1280]);  add_16258 = None
	        view_1664: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_318, [sym_size_int, 1500, 1])
	        view_1665: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_636, [sym_size_int, 1500, 1])
	        reciprocal_106: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1664);  view_1664 = None
	        mul_10324: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_106, 1.0);  reciprocal_106 = None
	        mul_10327: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1663, mul_10324);  view_1663 = mul_10324 = None
	        round_214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10327);  mul_10327 = None
	        add_16345: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_214, view_1665);  round_214 = view_1665 = None
	        clamp_min_320: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16345, -128);  add_16345 = None
	        clamp_max_213: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_320, 127);  clamp_min_320 = None
	        view_1666: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_213, [sym_size_int, 1500, 1280]);  clamp_max_213 = None
	        _assert_tensor_metadata_958 = torch.ops.aten._assert_tensor_metadata.default(view_1666, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_958 = None
	        convert_element_type_637: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1666, torch.int8);  view_1666 = None
	        view_1667: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_637, [sym_size_int, 1500, 1280]);  convert_element_type_637 = None
	        view_1668: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_318, [sym_size_int, 1500, 1]);  clamp_min_318 = None
	        view_1669: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_636, [sym_size_int, 1500, 1]);  convert_element_type_636 = None
	        _assert_tensor_metadata_959 = torch.ops.aten._assert_tensor_metadata.default(view_1667, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_959 = None
	        convert_element_type_638: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1667, torch.float32);  view_1667 = None
	        _assert_tensor_metadata_960 = torch.ops.aten._assert_tensor_metadata.default(view_1669, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_960 = None
	        convert_element_type_639: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1669, torch.float32);  view_1669 = None
	        sub_4886: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_638, convert_element_type_639);  convert_element_type_638 = convert_element_type_639 = None
	        mul_10349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4886, view_1668);  sub_4886 = view_1668 = None
	        view_1670: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10349, [sym_size_int, 1500, 1280]);  mul_10349 = None
	        _assert_tensor_metadata_961 = torch.ops.aten._assert_tensor_metadata.default(view_1670, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_961 = None
	        view_1671: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_17_fc1_parametrizations_weight_original0 = None
	        view_1672: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_17_fc1_parametrizations_weight_original1 = None
	        view_1673: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_17_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_962 = torch.ops.aten._assert_tensor_metadata.default(view_1671, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_962 = None
	        convert_element_type_640: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1671, torch.float32);  view_1671 = None
	        _assert_tensor_metadata_963 = torch.ops.aten._assert_tensor_metadata.default(view_1673, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_963 = None
	        convert_element_type_641: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1673, torch.float32);  view_1673 = None
	        sub_4890: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_640, convert_element_type_641);  convert_element_type_640 = convert_element_type_641 = None
	        mul_10354: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4890, view_1672);  sub_4890 = view_1672 = None
	        view_1674: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10354, [5120, 1280]);  mul_10354 = None
	        _assert_tensor_metadata_964 = torch.ops.aten._assert_tensor_metadata.default(view_1674, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_964 = None
	        mul_10359: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1675: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1670, [mul_10359, 1280]);  view_1670 = mul_10359 = None
	        permute_179: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1674, [1, 0]);  view_1674 = None
	        addmm_88: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_17_fc1_bias, view_1675, permute_179);  model_audio_tower_layers_17_fc1_bias = view_1675 = permute_179 = None
	        view_1676: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_88, [sym_size_int, 1500, 5120]);  addmm_88 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_10366: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1676, 0.5)
	        mul_10367: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1676, 0.7071067811865476);  view_1676 = None
	        erf_19: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_10367);  mul_10367 = None
	        add_16404: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_19, 1);  erf_19 = None
	        mul_10368: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10366, add_16404);  mul_10366 = add_16404 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_143: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_10368);  mul_10368 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1677: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_143, [sym_size_int, 1500, 5120])
	        amin_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1677, [2])
	        amax_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1677, [2]);  view_1677 = None
	        full_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_107, full_214);  amin_107 = full_214 = None
	        full_215: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_107, full_215);  amax_107 = full_215 = None
	        sub_4903: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_107, minimum_107);  maximum_107 = None
	        div_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4903, 255.0);  sub_4903 = None
	        clamp_min_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_214, 1.1920928955078125e-07);  div_214 = None
	        div_215: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_107, clamp_min_321);  minimum_107 = None
	        round_215: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_215);  div_215 = None
	        sub_4909: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_215);  round_215 = None
	        clamp_min_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4909, -128);  sub_4909 = None
	        clamp_max_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_322, 127);  clamp_min_322 = None
	        _assert_tensor_metadata_965 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_321, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_965 = None
	        _assert_tensor_metadata_966 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_214, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_966 = None
	        convert_element_type_642: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_214, torch.int8);  clamp_max_214 = None
	        view_1678: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_143, [sym_size_int, 1500, 5120]);  clone_143 = None
	        view_1679: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_321, [sym_size_int, 1500, 1])
	        view_1680: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_642, [sym_size_int, 1500, 1])
	        reciprocal_107: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1679);  view_1679 = None
	        mul_10414: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_107, 1.0);  reciprocal_107 = None
	        mul_10417: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1678, mul_10414);  view_1678 = mul_10414 = None
	        round_216: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_10417);  mul_10417 = None
	        add_16487: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_216, view_1680);  round_216 = view_1680 = None
	        clamp_min_323: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16487, -128);  add_16487 = None
	        clamp_max_215: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_323, 127);  clamp_min_323 = None
	        view_1681: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_215, [sym_size_int, 1500, 5120]);  clamp_max_215 = None
	        _assert_tensor_metadata_967 = torch.ops.aten._assert_tensor_metadata.default(view_1681, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_967 = None
	        convert_element_type_643: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1681, torch.int8);  view_1681 = None
	        view_1682: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_643, [sym_size_int, 1500, 5120]);  convert_element_type_643 = None
	        view_1683: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_321, [sym_size_int, 1500, 1]);  clamp_min_321 = None
	        view_1684: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_642, [sym_size_int, 1500, 1]);  convert_element_type_642 = None
	        _assert_tensor_metadata_968 = torch.ops.aten._assert_tensor_metadata.default(view_1682, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_968 = None
	        convert_element_type_644: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1682, torch.float32);  view_1682 = None
	        _assert_tensor_metadata_969 = torch.ops.aten._assert_tensor_metadata.default(view_1684, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_969 = None
	        convert_element_type_645: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1684, torch.float32);  view_1684 = None
	        sub_4929: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_644, convert_element_type_645);  convert_element_type_644 = convert_element_type_645 = None
	        mul_10439: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4929, view_1683);  sub_4929 = view_1683 = None
	        view_1685: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_10439, [sym_size_int, 1500, 5120]);  mul_10439 = None
	        _assert_tensor_metadata_970 = torch.ops.aten._assert_tensor_metadata.default(view_1685, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_970 = None
	        view_1686: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_17_fc2_parametrizations_weight_original0 = None
	        view_1687: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_17_fc2_parametrizations_weight_original1 = None
	        view_1688: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_17_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_971 = torch.ops.aten._assert_tensor_metadata.default(view_1686, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_971 = None
	        convert_element_type_646: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1686, torch.float32);  view_1686 = None
	        _assert_tensor_metadata_972 = torch.ops.aten._assert_tensor_metadata.default(view_1688, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_972 = None
	        convert_element_type_647: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1688, torch.float32);  view_1688 = None
	        sub_4933: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_646, convert_element_type_647);  convert_element_type_646 = convert_element_type_647 = None
	        mul_10444: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4933, view_1687);  sub_4933 = view_1687 = None
	        view_1689: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_10444, [1280, 5120]);  mul_10444 = None
	        _assert_tensor_metadata_973 = torch.ops.aten._assert_tensor_metadata.default(view_1689, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_973 = None
	        mul_10449: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1690: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1685, [mul_10449, 5120]);  view_1685 = mul_10449 = None
	        permute_180: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1689, [1, 0]);  view_1689 = None
	        addmm_89: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_17_fc2_bias, view_1690, permute_180);  model_audio_tower_layers_17_fc2_bias = view_1690 = permute_180 = None
	        view_1691: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_89, [sym_size_int, 1500, 1280]);  addmm_89 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_144: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1691);  view_1691 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_16550: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_16252, clone_144);  add_16252 = clone_144 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_145: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_16550, memory_format = torch.contiguous_format)
	        var_mean_36 = torch.ops.aten.var_mean.correction(clone_145, [2], correction = 0, keepdim = True)
	        getitem_144: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_36[0]
	        getitem_145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_36[1];  var_mean_36 = None
	        add_16555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_144, 1e-05);  getitem_144 = None
	        rsqrt_36: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_16555);  add_16555 = None
	        sub_4939: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_145, getitem_145);  clone_145 = getitem_145 = None
	        mul_10460: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4939, rsqrt_36);  sub_4939 = rsqrt_36 = None
	        mul_10461: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10460, model_audio_tower_layers_18_self_attn_layer_norm_weight);  mul_10460 = model_audio_tower_layers_18_self_attn_layer_norm_weight = None
	        add_16556: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_10461, model_audio_tower_layers_18_self_attn_layer_norm_bias);  mul_10461 = model_audio_tower_layers_18_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1692: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_16556, [sym_size_int, 1500, 1280])
	        amin_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1692, [2])
	        amax_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1692, [2]);  view_1692 = None
	        full_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_108, full_216);  amin_108 = full_216 = None
	        full_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_108, full_217);  amax_108 = full_217 = None
	        sub_4950: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_108, minimum_108);  maximum_108 = None
	        div_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4950, 255.0);  sub_4950 = None
	        clamp_min_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_216, 1.1920928955078125e-07);  div_216 = None
	        div_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_108, clamp_min_324);  minimum_108 = None
	        round_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_217);  div_217 = None
	        sub_4956: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_217);  round_217 = None
	        clamp_min_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4956, -128);  sub_4956 = None
	        clamp_max_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_325, 127);  clamp_min_325 = None
	        _assert_tensor_metadata_974 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_324, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_974 = None
	        _assert_tensor_metadata_975 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_216, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_975 = None
	        convert_element_type_648: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_216, torch.int8);  clamp_max_216 = None
	        view_1693: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_16556, [sym_size_int, 1500, 1280])
	        view_1694: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_324, [sym_size_int, 1500, 1])
	        view_1695: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_648, [sym_size_int, 1500, 1])
	        reciprocal_108: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1694);  view_1694 = None
	        mul_10509: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_108, 1.0);  reciprocal_108 = None
	        mul_10512: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1693, mul_10509);  view_1693 = mul_10509 = None
	        round_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10512);  mul_10512 = None
	        add_16643: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_218, view_1695);  round_218 = view_1695 = None
	        clamp_min_326: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16643, -128);  add_16643 = None
	        clamp_max_217: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_326, 127);  clamp_min_326 = None
	        view_1696: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_217, [sym_size_int, 1500, 1280]);  clamp_max_217 = None
	        _assert_tensor_metadata_976 = torch.ops.aten._assert_tensor_metadata.default(view_1696, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_976 = None
	        convert_element_type_649: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1696, torch.int8);  view_1696 = None
	        view_1697: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_649, [sym_size_int, 1500, 1280]);  convert_element_type_649 = None
	        view_1698: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_324, [sym_size_int, 1500, 1]);  clamp_min_324 = None
	        view_1699: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_648, [sym_size_int, 1500, 1]);  convert_element_type_648 = None
	        _assert_tensor_metadata_977 = torch.ops.aten._assert_tensor_metadata.default(view_1697, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_977 = None
	        convert_element_type_650: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1697, torch.float32);  view_1697 = None
	        _assert_tensor_metadata_978 = torch.ops.aten._assert_tensor_metadata.default(view_1699, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_978 = None
	        convert_element_type_651: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1699, torch.float32);  view_1699 = None
	        sub_4976: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_650, convert_element_type_651);  convert_element_type_650 = convert_element_type_651 = None
	        mul_10534: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4976, view_1698);  sub_4976 = view_1698 = None
	        view_1700: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10534, [sym_size_int, 1500, 1280]);  mul_10534 = None
	        _assert_tensor_metadata_979 = torch.ops.aten._assert_tensor_metadata.default(view_1700, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_979 = None
	        view_1701: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1702: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1703: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_980 = torch.ops.aten._assert_tensor_metadata.default(view_1701, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_980 = None
	        convert_element_type_652: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1701, torch.float32);  view_1701 = None
	        _assert_tensor_metadata_981 = torch.ops.aten._assert_tensor_metadata.default(view_1703, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_981 = None
	        convert_element_type_653: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1703, torch.float32);  view_1703 = None
	        sub_4980: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_652, convert_element_type_653);  convert_element_type_652 = convert_element_type_653 = None
	        mul_10539: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4980, view_1702);  sub_4980 = view_1702 = None
	        view_1704: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10539, [1280, 1280]);  mul_10539 = None
	        _assert_tensor_metadata_982 = torch.ops.aten._assert_tensor_metadata.default(view_1704, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_982 = None
	        mul_10544: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1705: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1700, [mul_10544, 1280]);  view_1700 = mul_10544 = None
	        permute_181: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1704, [1, 0]);  view_1704 = None
	        addmm_90: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_18_self_attn_q_proj_bias, view_1705, permute_181);  model_audio_tower_layers_18_self_attn_q_proj_bias = view_1705 = permute_181 = None
	        view_1706: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_90, [sym_size_int, 1500, 1280]);  addmm_90 = None
	        mul_10551: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1706, 0.125);  view_1706 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1707: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_10551, [sym_size_int, 1500, 20, 64]);  mul_10551 = None
	        permute_182: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1707, [0, 2, 1, 3]);  view_1707 = None
	        clone_146: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_182, memory_format = torch.contiguous_format);  permute_182 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1708: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_16556, [sym_size_int, 1500, 1280])
	        amin_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1708, [2])
	        amax_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1708, [2]);  view_1708 = None
	        full_218: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_109, full_218);  amin_109 = full_218 = None
	        full_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_109, full_219);  amax_109 = full_219 = None
	        sub_4995: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_109, minimum_109);  maximum_109 = None
	        div_218: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4995, 255.0);  sub_4995 = None
	        clamp_min_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_218, 1.1920928955078125e-07);  div_218 = None
	        div_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_109, clamp_min_327);  minimum_109 = None
	        round_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_219);  div_219 = None
	        sub_5001: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_219);  round_219 = None
	        clamp_min_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5001, -128);  sub_5001 = None
	        clamp_max_218: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_328, 127);  clamp_min_328 = None
	        _assert_tensor_metadata_983 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_327, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_983 = None
	        _assert_tensor_metadata_984 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_218, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_984 = None
	        convert_element_type_654: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_218, torch.int8);  clamp_max_218 = None
	        view_1709: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_16556, [sym_size_int, 1500, 1280])
	        view_1710: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_327, [sym_size_int, 1500, 1])
	        view_1711: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_654, [sym_size_int, 1500, 1])
	        reciprocal_109: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1710);  view_1710 = None
	        mul_10605: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_109, 1.0);  reciprocal_109 = None
	        mul_10608: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1709, mul_10605);  view_1709 = mul_10605 = None
	        round_220: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10608);  mul_10608 = None
	        add_16795: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_220, view_1711);  round_220 = view_1711 = None
	        clamp_min_329: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16795, -128);  add_16795 = None
	        clamp_max_219: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_329, 127);  clamp_min_329 = None
	        view_1712: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_219, [sym_size_int, 1500, 1280]);  clamp_max_219 = None
	        _assert_tensor_metadata_985 = torch.ops.aten._assert_tensor_metadata.default(view_1712, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_985 = None
	        convert_element_type_655: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1712, torch.int8);  view_1712 = None
	        view_1713: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_655, [sym_size_int, 1500, 1280]);  convert_element_type_655 = None
	        view_1714: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_327, [sym_size_int, 1500, 1]);  clamp_min_327 = None
	        view_1715: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_654, [sym_size_int, 1500, 1]);  convert_element_type_654 = None
	        _assert_tensor_metadata_986 = torch.ops.aten._assert_tensor_metadata.default(view_1713, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_986 = None
	        convert_element_type_656: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1713, torch.float32);  view_1713 = None
	        _assert_tensor_metadata_987 = torch.ops.aten._assert_tensor_metadata.default(view_1715, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_987 = None
	        convert_element_type_657: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1715, torch.float32);  view_1715 = None
	        sub_5021: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_656, convert_element_type_657);  convert_element_type_656 = convert_element_type_657 = None
	        mul_10630: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5021, view_1714);  sub_5021 = view_1714 = None
	        view_1716: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10630, [sym_size_int, 1500, 1280]);  mul_10630 = None
	        _assert_tensor_metadata_988 = torch.ops.aten._assert_tensor_metadata.default(view_1716, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_988 = None
	        view_1717: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1718: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1719: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_989 = torch.ops.aten._assert_tensor_metadata.default(view_1717, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_989 = None
	        convert_element_type_658: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1717, torch.float32);  view_1717 = None
	        _assert_tensor_metadata_990 = torch.ops.aten._assert_tensor_metadata.default(view_1719, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_990 = None
	        convert_element_type_659: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1719, torch.float32);  view_1719 = None
	        sub_5025: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_658, convert_element_type_659);  convert_element_type_658 = convert_element_type_659 = None
	        mul_10635: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5025, view_1718);  sub_5025 = view_1718 = None
	        view_1720: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10635, [1280, 1280]);  mul_10635 = None
	        _assert_tensor_metadata_991 = torch.ops.aten._assert_tensor_metadata.default(view_1720, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_991 = None
	        permute_183: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1720, [1, 0]);  view_1720 = None
	        mul_10638: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1721: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1716, [mul_10638, 1280]);  view_1716 = mul_10638 = None
	        mm_18: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1721, permute_183);  view_1721 = permute_183 = None
	        view_1722: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_18, [sym_size_int, 1500, 1280]);  mm_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1723: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1722, [sym_size_int, -1, 20, 64]);  view_1722 = None
	        permute_184: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1723, [0, 2, 1, 3]);  view_1723 = None
	        clone_147: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_184, memory_format = torch.contiguous_format);  permute_184 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1724: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_16556, [sym_size_int, 1500, 1280])
	        amin_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1724, [2])
	        amax_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1724, [2]);  view_1724 = None
	        full_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_110, full_220);  amin_110 = full_220 = None
	        full_221: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_110, full_221);  amax_110 = full_221 = None
	        sub_5039: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_110, minimum_110);  maximum_110 = None
	        div_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5039, 255.0);  sub_5039 = None
	        clamp_min_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_220, 1.1920928955078125e-07);  div_220 = None
	        div_221: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_110, clamp_min_330);  minimum_110 = None
	        round_221: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_221);  div_221 = None
	        sub_5045: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_221);  round_221 = None
	        clamp_min_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5045, -128);  sub_5045 = None
	        clamp_max_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_331, 127);  clamp_min_331 = None
	        _assert_tensor_metadata_992 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_330, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_992 = None
	        _assert_tensor_metadata_993 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_220, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_993 = None
	        convert_element_type_660: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_220, torch.int8);  clamp_max_220 = None
	        view_1725: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_16556, [sym_size_int, 1500, 1280]);  add_16556 = None
	        view_1726: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_330, [sym_size_int, 1500, 1])
	        view_1727: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_660, [sym_size_int, 1500, 1])
	        reciprocal_110: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1726);  view_1726 = None
	        mul_10704: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_110, 1.0);  reciprocal_110 = None
	        mul_10707: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1725, mul_10704);  view_1725 = mul_10704 = None
	        round_222: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10707);  mul_10707 = None
	        add_16943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_222, view_1727);  round_222 = view_1727 = None
	        clamp_min_332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16943, -128);  add_16943 = None
	        clamp_max_221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_332, 127);  clamp_min_332 = None
	        view_1728: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_221, [sym_size_int, 1500, 1280]);  clamp_max_221 = None
	        _assert_tensor_metadata_994 = torch.ops.aten._assert_tensor_metadata.default(view_1728, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_994 = None
	        convert_element_type_661: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1728, torch.int8);  view_1728 = None
	        view_1729: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_661, [sym_size_int, 1500, 1280]);  convert_element_type_661 = None
	        view_1730: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_330, [sym_size_int, 1500, 1]);  clamp_min_330 = None
	        view_1731: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_660, [sym_size_int, 1500, 1]);  convert_element_type_660 = None
	        _assert_tensor_metadata_995 = torch.ops.aten._assert_tensor_metadata.default(view_1729, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_995 = None
	        convert_element_type_662: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1729, torch.float32);  view_1729 = None
	        _assert_tensor_metadata_996 = torch.ops.aten._assert_tensor_metadata.default(view_1731, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_996 = None
	        convert_element_type_663: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1731, torch.float32);  view_1731 = None
	        sub_5065: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_662, convert_element_type_663);  convert_element_type_662 = convert_element_type_663 = None
	        mul_10729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5065, view_1730);  sub_5065 = view_1730 = None
	        view_1732: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10729, [sym_size_int, 1500, 1280]);  mul_10729 = None
	        _assert_tensor_metadata_997 = torch.ops.aten._assert_tensor_metadata.default(view_1732, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_997 = None
	        view_1733: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1734: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1735: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_998 = torch.ops.aten._assert_tensor_metadata.default(view_1733, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_998 = None
	        convert_element_type_664: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1733, torch.float32);  view_1733 = None
	        _assert_tensor_metadata_999 = torch.ops.aten._assert_tensor_metadata.default(view_1735, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_999 = None
	        convert_element_type_665: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1735, torch.float32);  view_1735 = None
	        sub_5069: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_664, convert_element_type_665);  convert_element_type_664 = convert_element_type_665 = None
	        mul_10734: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5069, view_1734);  sub_5069 = view_1734 = None
	        view_1736: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10734, [1280, 1280]);  mul_10734 = None
	        _assert_tensor_metadata_1000 = torch.ops.aten._assert_tensor_metadata.default(view_1736, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1000 = None
	        mul_10739: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1737: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1732, [mul_10739, 1280]);  view_1732 = mul_10739 = None
	        permute_185: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1736, [1, 0]);  view_1736 = None
	        addmm_91: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_18_self_attn_v_proj_bias, view_1737, permute_185);  model_audio_tower_layers_18_self_attn_v_proj_bias = view_1737 = permute_185 = None
	        view_1738: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_91, [sym_size_int, 1500, 1280]);  addmm_91 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1739: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1738, [sym_size_int, -1, 20, 64]);  view_1738 = None
	        permute_186: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1739, [0, 2, 1, 3]);  view_1739 = None
	        clone_148: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_186, memory_format = torch.contiguous_format);  permute_186 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_18 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_146, clone_147, clone_148, None, False, scale = 1.0);  clone_146 = clone_147 = clone_148 = None
	        getitem_146: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_18[0];  _scaled_dot_product_efficient_attention_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_187: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_146, [0, 2, 1, 3]);  getitem_146 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1740: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_187, [sym_size_int, 1500, -1]);  permute_187 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1741: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1740, [sym_size_int, 1500, 1280])
	        amin_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1741, [2])
	        amax_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1741, [2]);  view_1741 = None
	        full_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_111, full_222);  amin_111 = full_222 = None
	        full_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_111, full_223);  amax_111 = full_223 = None
	        sub_5087: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_111, minimum_111);  maximum_111 = None
	        div_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5087, 255.0);  sub_5087 = None
	        clamp_min_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_222, 1.1920928955078125e-07);  div_222 = None
	        div_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_111, clamp_min_333);  minimum_111 = None
	        round_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_223);  div_223 = None
	        sub_5093: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_223);  round_223 = None
	        clamp_min_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5093, -128);  sub_5093 = None
	        clamp_max_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_334, 127);  clamp_min_334 = None
	        _assert_tensor_metadata_1001 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_333, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1001 = None
	        _assert_tensor_metadata_1002 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_222, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1002 = None
	        convert_element_type_666: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_222, torch.int8);  clamp_max_222 = None
	        view_1742: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1740, [sym_size_int, 1500, 1280]);  view_1740 = None
	        view_1743: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_333, [sym_size_int, 1500, 1])
	        view_1744: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_666, [sym_size_int, 1500, 1])
	        reciprocal_111: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1743);  view_1743 = None
	        mul_10809: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_111, 1.0);  reciprocal_111 = None
	        mul_10812: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1742, mul_10809);  view_1742 = mul_10809 = None
	        round_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10812);  mul_10812 = None
	        add_17107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_224, view_1744);  round_224 = view_1744 = None
	        clamp_min_335: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17107, -128);  add_17107 = None
	        clamp_max_223: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_335, 127);  clamp_min_335 = None
	        view_1745: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_223, [sym_size_int, 1500, 1280]);  clamp_max_223 = None
	        _assert_tensor_metadata_1003 = torch.ops.aten._assert_tensor_metadata.default(view_1745, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1003 = None
	        convert_element_type_667: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1745, torch.int8);  view_1745 = None
	        view_1746: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_667, [sym_size_int, 1500, 1280]);  convert_element_type_667 = None
	        view_1747: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_333, [sym_size_int, 1500, 1]);  clamp_min_333 = None
	        view_1748: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_666, [sym_size_int, 1500, 1]);  convert_element_type_666 = None
	        _assert_tensor_metadata_1004 = torch.ops.aten._assert_tensor_metadata.default(view_1746, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1004 = None
	        convert_element_type_668: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1746, torch.float32);  view_1746 = None
	        _assert_tensor_metadata_1005 = torch.ops.aten._assert_tensor_metadata.default(view_1748, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1005 = None
	        convert_element_type_669: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1748, torch.float32);  view_1748 = None
	        sub_5113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_668, convert_element_type_669);  convert_element_type_668 = convert_element_type_669 = None
	        mul_10834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5113, view_1747);  sub_5113 = view_1747 = None
	        view_1749: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10834, [sym_size_int, 1500, 1280]);  mul_10834 = None
	        _assert_tensor_metadata_1006 = torch.ops.aten._assert_tensor_metadata.default(view_1749, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1006 = None
	        view_1750: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1751: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1752: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1007 = torch.ops.aten._assert_tensor_metadata.default(view_1750, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1007 = None
	        convert_element_type_670: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1750, torch.float32);  view_1750 = None
	        _assert_tensor_metadata_1008 = torch.ops.aten._assert_tensor_metadata.default(view_1752, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1008 = None
	        convert_element_type_671: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1752, torch.float32);  view_1752 = None
	        sub_5117: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_670, convert_element_type_671);  convert_element_type_670 = convert_element_type_671 = None
	        mul_10839: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5117, view_1751);  sub_5117 = view_1751 = None
	        view_1753: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10839, [1280, 1280]);  mul_10839 = None
	        _assert_tensor_metadata_1009 = torch.ops.aten._assert_tensor_metadata.default(view_1753, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1009 = None
	        mul_10844: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1754: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1749, [mul_10844, 1280]);  view_1749 = mul_10844 = None
	        permute_188: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1753, [1, 0]);  view_1753 = None
	        addmm_92: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_18_self_attn_out_proj_bias, view_1754, permute_188);  model_audio_tower_layers_18_self_attn_out_proj_bias = view_1754 = permute_188 = None
	        view_1755: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_92, [sym_size_int, 1500, 1280]);  addmm_92 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1755);  view_1755 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_17170: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_16550, clone_149);  add_16550 = clone_149 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_150: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_17170, memory_format = torch.contiguous_format)
	        var_mean_37 = torch.ops.aten.var_mean.correction(clone_150, [2], correction = 0, keepdim = True)
	        getitem_150: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_37[0]
	        getitem_151: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_37[1];  var_mean_37 = None
	        add_17175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_150, 1e-05);  getitem_150 = None
	        rsqrt_37: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_17175);  add_17175 = None
	        sub_5123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_150, getitem_151);  clone_150 = getitem_151 = None
	        mul_10855: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5123, rsqrt_37);  sub_5123 = rsqrt_37 = None
	        mul_10856: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10855, model_audio_tower_layers_18_final_layer_norm_weight);  mul_10855 = model_audio_tower_layers_18_final_layer_norm_weight = None
	        add_17176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_10856, model_audio_tower_layers_18_final_layer_norm_bias);  mul_10856 = model_audio_tower_layers_18_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1756: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_17176, [sym_size_int, 1500, 1280])
	        amin_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1756, [2])
	        amax_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1756, [2]);  view_1756 = None
	        full_224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_112, full_224);  amin_112 = full_224 = None
	        full_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_112, full_225);  amax_112 = full_225 = None
	        sub_5134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_112, minimum_112);  maximum_112 = None
	        div_224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5134, 255.0);  sub_5134 = None
	        clamp_min_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_224, 1.1920928955078125e-07);  div_224 = None
	        div_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_112, clamp_min_336);  minimum_112 = None
	        round_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_225);  div_225 = None
	        sub_5140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_225);  round_225 = None
	        clamp_min_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5140, -128);  sub_5140 = None
	        clamp_max_224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_337, 127);  clamp_min_337 = None
	        _assert_tensor_metadata_1010 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_336, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1010 = None
	        _assert_tensor_metadata_1011 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_224, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1011 = None
	        convert_element_type_672: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_224, torch.int8);  clamp_max_224 = None
	        view_1757: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_17176, [sym_size_int, 1500, 1280]);  add_17176 = None
	        view_1758: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_336, [sym_size_int, 1500, 1])
	        view_1759: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_672, [sym_size_int, 1500, 1])
	        reciprocal_112: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1758);  view_1758 = None
	        mul_10904: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_112, 1.0);  reciprocal_112 = None
	        mul_10907: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1757, mul_10904);  view_1757 = mul_10904 = None
	        round_226: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10907);  mul_10907 = None
	        add_17263: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_226, view_1759);  round_226 = view_1759 = None
	        clamp_min_338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17263, -128);  add_17263 = None
	        clamp_max_225: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_338, 127);  clamp_min_338 = None
	        view_1760: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_225, [sym_size_int, 1500, 1280]);  clamp_max_225 = None
	        _assert_tensor_metadata_1012 = torch.ops.aten._assert_tensor_metadata.default(view_1760, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1012 = None
	        convert_element_type_673: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1760, torch.int8);  view_1760 = None
	        view_1761: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_673, [sym_size_int, 1500, 1280]);  convert_element_type_673 = None
	        view_1762: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_336, [sym_size_int, 1500, 1]);  clamp_min_336 = None
	        view_1763: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_672, [sym_size_int, 1500, 1]);  convert_element_type_672 = None
	        _assert_tensor_metadata_1013 = torch.ops.aten._assert_tensor_metadata.default(view_1761, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1013 = None
	        convert_element_type_674: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1761, torch.float32);  view_1761 = None
	        _assert_tensor_metadata_1014 = torch.ops.aten._assert_tensor_metadata.default(view_1763, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1014 = None
	        convert_element_type_675: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1763, torch.float32);  view_1763 = None
	        sub_5160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_674, convert_element_type_675);  convert_element_type_674 = convert_element_type_675 = None
	        mul_10929: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5160, view_1762);  sub_5160 = view_1762 = None
	        view_1764: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10929, [sym_size_int, 1500, 1280]);  mul_10929 = None
	        _assert_tensor_metadata_1015 = torch.ops.aten._assert_tensor_metadata.default(view_1764, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1015 = None
	        view_1765: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_18_fc1_parametrizations_weight_original0 = None
	        view_1766: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_18_fc1_parametrizations_weight_original1 = None
	        view_1767: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_18_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1016 = torch.ops.aten._assert_tensor_metadata.default(view_1765, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1016 = None
	        convert_element_type_676: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1765, torch.float32);  view_1765 = None
	        _assert_tensor_metadata_1017 = torch.ops.aten._assert_tensor_metadata.default(view_1767, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1017 = None
	        convert_element_type_677: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1767, torch.float32);  view_1767 = None
	        sub_5164: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_676, convert_element_type_677);  convert_element_type_676 = convert_element_type_677 = None
	        mul_10934: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5164, view_1766);  sub_5164 = view_1766 = None
	        view_1768: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10934, [5120, 1280]);  mul_10934 = None
	        _assert_tensor_metadata_1018 = torch.ops.aten._assert_tensor_metadata.default(view_1768, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1018 = None
	        mul_10939: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1769: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1764, [mul_10939, 1280]);  view_1764 = mul_10939 = None
	        permute_189: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1768, [1, 0]);  view_1768 = None
	        addmm_93: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_18_fc1_bias, view_1769, permute_189);  model_audio_tower_layers_18_fc1_bias = view_1769 = permute_189 = None
	        view_1770: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_93, [sym_size_int, 1500, 5120]);  addmm_93 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_10946: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1770, 0.5)
	        mul_10947: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1770, 0.7071067811865476);  view_1770 = None
	        erf_20: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_10947);  mul_10947 = None
	        add_17322: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_20, 1);  erf_20 = None
	        mul_10948: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10946, add_17322);  mul_10946 = add_17322 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_151: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_10948);  mul_10948 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1771: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_151, [sym_size_int, 1500, 5120])
	        amin_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1771, [2])
	        amax_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1771, [2]);  view_1771 = None
	        full_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_113, full_226);  amin_113 = full_226 = None
	        full_227: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_113, full_227);  amax_113 = full_227 = None
	        sub_5177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_113, minimum_113);  maximum_113 = None
	        div_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5177, 255.0);  sub_5177 = None
	        clamp_min_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_226, 1.1920928955078125e-07);  div_226 = None
	        div_227: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_113, clamp_min_339);  minimum_113 = None
	        round_227: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_227);  div_227 = None
	        sub_5183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_227);  round_227 = None
	        clamp_min_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5183, -128);  sub_5183 = None
	        clamp_max_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_340, 127);  clamp_min_340 = None
	        _assert_tensor_metadata_1019 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_339, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1019 = None
	        _assert_tensor_metadata_1020 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_226, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1020 = None
	        convert_element_type_678: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_226, torch.int8);  clamp_max_226 = None
	        view_1772: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_151, [sym_size_int, 1500, 5120]);  clone_151 = None
	        view_1773: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_339, [sym_size_int, 1500, 1])
	        view_1774: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_678, [sym_size_int, 1500, 1])
	        reciprocal_113: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1773);  view_1773 = None
	        mul_10994: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_113, 1.0);  reciprocal_113 = None
	        mul_10997: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1772, mul_10994);  view_1772 = mul_10994 = None
	        round_228: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_10997);  mul_10997 = None
	        add_17405: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_228, view_1774);  round_228 = view_1774 = None
	        clamp_min_341: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17405, -128);  add_17405 = None
	        clamp_max_227: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_341, 127);  clamp_min_341 = None
	        view_1775: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_227, [sym_size_int, 1500, 5120]);  clamp_max_227 = None
	        _assert_tensor_metadata_1021 = torch.ops.aten._assert_tensor_metadata.default(view_1775, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1021 = None
	        convert_element_type_679: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1775, torch.int8);  view_1775 = None
	        view_1776: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_679, [sym_size_int, 1500, 5120]);  convert_element_type_679 = None
	        view_1777: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_339, [sym_size_int, 1500, 1]);  clamp_min_339 = None
	        view_1778: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_678, [sym_size_int, 1500, 1]);  convert_element_type_678 = None
	        _assert_tensor_metadata_1022 = torch.ops.aten._assert_tensor_metadata.default(view_1776, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1022 = None
	        convert_element_type_680: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1776, torch.float32);  view_1776 = None
	        _assert_tensor_metadata_1023 = torch.ops.aten._assert_tensor_metadata.default(view_1778, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1023 = None
	        convert_element_type_681: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1778, torch.float32);  view_1778 = None
	        sub_5203: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_680, convert_element_type_681);  convert_element_type_680 = convert_element_type_681 = None
	        mul_11019: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5203, view_1777);  sub_5203 = view_1777 = None
	        view_1779: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_11019, [sym_size_int, 1500, 5120]);  mul_11019 = None
	        _assert_tensor_metadata_1024 = torch.ops.aten._assert_tensor_metadata.default(view_1779, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1024 = None
	        view_1780: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_18_fc2_parametrizations_weight_original0 = None
	        view_1781: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_18_fc2_parametrizations_weight_original1 = None
	        view_1782: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_18_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1025 = torch.ops.aten._assert_tensor_metadata.default(view_1780, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1025 = None
	        convert_element_type_682: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1780, torch.float32);  view_1780 = None
	        _assert_tensor_metadata_1026 = torch.ops.aten._assert_tensor_metadata.default(view_1782, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1026 = None
	        convert_element_type_683: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1782, torch.float32);  view_1782 = None
	        sub_5207: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_682, convert_element_type_683);  convert_element_type_682 = convert_element_type_683 = None
	        mul_11024: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5207, view_1781);  sub_5207 = view_1781 = None
	        view_1783: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_11024, [1280, 5120]);  mul_11024 = None
	        _assert_tensor_metadata_1027 = torch.ops.aten._assert_tensor_metadata.default(view_1783, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1027 = None
	        mul_11029: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1784: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1779, [mul_11029, 5120]);  view_1779 = mul_11029 = None
	        permute_190: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1783, [1, 0]);  view_1783 = None
	        addmm_94: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_18_fc2_bias, view_1784, permute_190);  model_audio_tower_layers_18_fc2_bias = view_1784 = permute_190 = None
	        view_1785: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_94, [sym_size_int, 1500, 1280]);  addmm_94 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1785);  view_1785 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_17468: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_17170, clone_152);  add_17170 = clone_152 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_153: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_17468, memory_format = torch.contiguous_format)
	        var_mean_38 = torch.ops.aten.var_mean.correction(clone_153, [2], correction = 0, keepdim = True)
	        getitem_152: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_38[0]
	        getitem_153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_38[1];  var_mean_38 = None
	        add_17473: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_152, 1e-05);  getitem_152 = None
	        rsqrt_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_17473);  add_17473 = None
	        sub_5213: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_153, getitem_153);  clone_153 = getitem_153 = None
	        mul_11040: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5213, rsqrt_38);  sub_5213 = rsqrt_38 = None
	        mul_11041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11040, model_audio_tower_layers_19_self_attn_layer_norm_weight);  mul_11040 = model_audio_tower_layers_19_self_attn_layer_norm_weight = None
	        add_17474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_11041, model_audio_tower_layers_19_self_attn_layer_norm_bias);  mul_11041 = model_audio_tower_layers_19_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1786: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_17474, [sym_size_int, 1500, 1280])
	        amin_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1786, [2])
	        amax_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1786, [2]);  view_1786 = None
	        full_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_114, full_228);  amin_114 = full_228 = None
	        full_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_114, full_229);  amax_114 = full_229 = None
	        sub_5224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_114, minimum_114);  maximum_114 = None
	        div_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5224, 255.0);  sub_5224 = None
	        clamp_min_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_228, 1.1920928955078125e-07);  div_228 = None
	        div_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_114, clamp_min_342);  minimum_114 = None
	        round_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_229);  div_229 = None
	        sub_5230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_229);  round_229 = None
	        clamp_min_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5230, -128);  sub_5230 = None
	        clamp_max_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_343, 127);  clamp_min_343 = None
	        _assert_tensor_metadata_1028 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_342, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1028 = None
	        _assert_tensor_metadata_1029 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_228, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1029 = None
	        convert_element_type_684: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_228, torch.int8);  clamp_max_228 = None
	        view_1787: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_17474, [sym_size_int, 1500, 1280])
	        view_1788: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_342, [sym_size_int, 1500, 1])
	        view_1789: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_684, [sym_size_int, 1500, 1])
	        reciprocal_114: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1788);  view_1788 = None
	        mul_11089: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_114, 1.0);  reciprocal_114 = None
	        mul_11092: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1787, mul_11089);  view_1787 = mul_11089 = None
	        round_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11092);  mul_11092 = None
	        add_17561: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_230, view_1789);  round_230 = view_1789 = None
	        clamp_min_344: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17561, -128);  add_17561 = None
	        clamp_max_229: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_344, 127);  clamp_min_344 = None
	        view_1790: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_229, [sym_size_int, 1500, 1280]);  clamp_max_229 = None
	        _assert_tensor_metadata_1030 = torch.ops.aten._assert_tensor_metadata.default(view_1790, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1030 = None
	        convert_element_type_685: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1790, torch.int8);  view_1790 = None
	        view_1791: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_685, [sym_size_int, 1500, 1280]);  convert_element_type_685 = None
	        view_1792: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_342, [sym_size_int, 1500, 1]);  clamp_min_342 = None
	        view_1793: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_684, [sym_size_int, 1500, 1]);  convert_element_type_684 = None
	        _assert_tensor_metadata_1031 = torch.ops.aten._assert_tensor_metadata.default(view_1791, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1031 = None
	        convert_element_type_686: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1791, torch.float32);  view_1791 = None
	        _assert_tensor_metadata_1032 = torch.ops.aten._assert_tensor_metadata.default(view_1793, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1032 = None
	        convert_element_type_687: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1793, torch.float32);  view_1793 = None
	        sub_5250: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_686, convert_element_type_687);  convert_element_type_686 = convert_element_type_687 = None
	        mul_11114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5250, view_1792);  sub_5250 = view_1792 = None
	        view_1794: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11114, [sym_size_int, 1500, 1280]);  mul_11114 = None
	        _assert_tensor_metadata_1033 = torch.ops.aten._assert_tensor_metadata.default(view_1794, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1033 = None
	        view_1795: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1796: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1797: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1034 = torch.ops.aten._assert_tensor_metadata.default(view_1795, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1034 = None
	        convert_element_type_688: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1795, torch.float32);  view_1795 = None
	        _assert_tensor_metadata_1035 = torch.ops.aten._assert_tensor_metadata.default(view_1797, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1035 = None
	        convert_element_type_689: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1797, torch.float32);  view_1797 = None
	        sub_5254: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_688, convert_element_type_689);  convert_element_type_688 = convert_element_type_689 = None
	        mul_11119: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5254, view_1796);  sub_5254 = view_1796 = None
	        view_1798: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11119, [1280, 1280]);  mul_11119 = None
	        _assert_tensor_metadata_1036 = torch.ops.aten._assert_tensor_metadata.default(view_1798, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1036 = None
	        mul_11124: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1799: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1794, [mul_11124, 1280]);  view_1794 = mul_11124 = None
	        permute_191: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1798, [1, 0]);  view_1798 = None
	        addmm_95: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_19_self_attn_q_proj_bias, view_1799, permute_191);  model_audio_tower_layers_19_self_attn_q_proj_bias = view_1799 = permute_191 = None
	        view_1800: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_95, [sym_size_int, 1500, 1280]);  addmm_95 = None
	        mul_11131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1800, 0.125);  view_1800 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1801: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_11131, [sym_size_int, 1500, 20, 64]);  mul_11131 = None
	        permute_192: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1801, [0, 2, 1, 3]);  view_1801 = None
	        clone_154: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_192, memory_format = torch.contiguous_format);  permute_192 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1802: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_17474, [sym_size_int, 1500, 1280])
	        amin_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1802, [2])
	        amax_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1802, [2]);  view_1802 = None
	        full_230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_115, full_230);  amin_115 = full_230 = None
	        full_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_115, full_231);  amax_115 = full_231 = None
	        sub_5269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_115, minimum_115);  maximum_115 = None
	        div_230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5269, 255.0);  sub_5269 = None
	        clamp_min_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_230, 1.1920928955078125e-07);  div_230 = None
	        div_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_115, clamp_min_345);  minimum_115 = None
	        round_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_231);  div_231 = None
	        sub_5275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_231);  round_231 = None
	        clamp_min_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5275, -128);  sub_5275 = None
	        clamp_max_230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_346, 127);  clamp_min_346 = None
	        _assert_tensor_metadata_1037 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_345, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1037 = None
	        _assert_tensor_metadata_1038 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_230, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1038 = None
	        convert_element_type_690: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_230, torch.int8);  clamp_max_230 = None
	        view_1803: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_17474, [sym_size_int, 1500, 1280])
	        view_1804: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_345, [sym_size_int, 1500, 1])
	        view_1805: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_690, [sym_size_int, 1500, 1])
	        reciprocal_115: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1804);  view_1804 = None
	        mul_11185: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_115, 1.0);  reciprocal_115 = None
	        mul_11188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1803, mul_11185);  view_1803 = mul_11185 = None
	        round_232: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11188);  mul_11188 = None
	        add_17713: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_232, view_1805);  round_232 = view_1805 = None
	        clamp_min_347: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17713, -128);  add_17713 = None
	        clamp_max_231: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_347, 127);  clamp_min_347 = None
	        view_1806: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_231, [sym_size_int, 1500, 1280]);  clamp_max_231 = None
	        _assert_tensor_metadata_1039 = torch.ops.aten._assert_tensor_metadata.default(view_1806, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1039 = None
	        convert_element_type_691: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1806, torch.int8);  view_1806 = None
	        view_1807: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_691, [sym_size_int, 1500, 1280]);  convert_element_type_691 = None
	        view_1808: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_345, [sym_size_int, 1500, 1]);  clamp_min_345 = None
	        view_1809: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_690, [sym_size_int, 1500, 1]);  convert_element_type_690 = None
	        _assert_tensor_metadata_1040 = torch.ops.aten._assert_tensor_metadata.default(view_1807, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1040 = None
	        convert_element_type_692: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1807, torch.float32);  view_1807 = None
	        _assert_tensor_metadata_1041 = torch.ops.aten._assert_tensor_metadata.default(view_1809, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1041 = None
	        convert_element_type_693: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1809, torch.float32);  view_1809 = None
	        sub_5295: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_692, convert_element_type_693);  convert_element_type_692 = convert_element_type_693 = None
	        mul_11210: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5295, view_1808);  sub_5295 = view_1808 = None
	        view_1810: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11210, [sym_size_int, 1500, 1280]);  mul_11210 = None
	        _assert_tensor_metadata_1042 = torch.ops.aten._assert_tensor_metadata.default(view_1810, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1042 = None
	        view_1811: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1812: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1813: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1043 = torch.ops.aten._assert_tensor_metadata.default(view_1811, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1043 = None
	        convert_element_type_694: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1811, torch.float32);  view_1811 = None
	        _assert_tensor_metadata_1044 = torch.ops.aten._assert_tensor_metadata.default(view_1813, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1044 = None
	        convert_element_type_695: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1813, torch.float32);  view_1813 = None
	        sub_5299: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_694, convert_element_type_695);  convert_element_type_694 = convert_element_type_695 = None
	        mul_11215: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5299, view_1812);  sub_5299 = view_1812 = None
	        view_1814: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11215, [1280, 1280]);  mul_11215 = None
	        _assert_tensor_metadata_1045 = torch.ops.aten._assert_tensor_metadata.default(view_1814, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1045 = None
	        permute_193: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1814, [1, 0]);  view_1814 = None
	        mul_11218: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1815: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1810, [mul_11218, 1280]);  view_1810 = mul_11218 = None
	        mm_19: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1815, permute_193);  view_1815 = permute_193 = None
	        view_1816: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_19, [sym_size_int, 1500, 1280]);  mm_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1817: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1816, [sym_size_int, -1, 20, 64]);  view_1816 = None
	        permute_194: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1817, [0, 2, 1, 3]);  view_1817 = None
	        clone_155: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_194, memory_format = torch.contiguous_format);  permute_194 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1818: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_17474, [sym_size_int, 1500, 1280])
	        amin_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1818, [2])
	        amax_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1818, [2]);  view_1818 = None
	        full_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_116, full_232);  amin_116 = full_232 = None
	        full_233: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_116, full_233);  amax_116 = full_233 = None
	        sub_5313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_116, minimum_116);  maximum_116 = None
	        div_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5313, 255.0);  sub_5313 = None
	        clamp_min_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_232, 1.1920928955078125e-07);  div_232 = None
	        div_233: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_116, clamp_min_348);  minimum_116 = None
	        round_233: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_233);  div_233 = None
	        sub_5319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_233);  round_233 = None
	        clamp_min_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5319, -128);  sub_5319 = None
	        clamp_max_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_349, 127);  clamp_min_349 = None
	        _assert_tensor_metadata_1046 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_348, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1046 = None
	        _assert_tensor_metadata_1047 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_232, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1047 = None
	        convert_element_type_696: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_232, torch.int8);  clamp_max_232 = None
	        view_1819: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_17474, [sym_size_int, 1500, 1280]);  add_17474 = None
	        view_1820: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_348, [sym_size_int, 1500, 1])
	        view_1821: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_696, [sym_size_int, 1500, 1])
	        reciprocal_116: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1820);  view_1820 = None
	        mul_11284: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_116, 1.0);  reciprocal_116 = None
	        mul_11287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1819, mul_11284);  view_1819 = mul_11284 = None
	        round_234: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11287);  mul_11287 = None
	        add_17861: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_234, view_1821);  round_234 = view_1821 = None
	        clamp_min_350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17861, -128);  add_17861 = None
	        clamp_max_233: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_350, 127);  clamp_min_350 = None
	        view_1822: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_233, [sym_size_int, 1500, 1280]);  clamp_max_233 = None
	        _assert_tensor_metadata_1048 = torch.ops.aten._assert_tensor_metadata.default(view_1822, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1048 = None
	        convert_element_type_697: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1822, torch.int8);  view_1822 = None
	        view_1823: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_697, [sym_size_int, 1500, 1280]);  convert_element_type_697 = None
	        view_1824: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_348, [sym_size_int, 1500, 1]);  clamp_min_348 = None
	        view_1825: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_696, [sym_size_int, 1500, 1]);  convert_element_type_696 = None
	        _assert_tensor_metadata_1049 = torch.ops.aten._assert_tensor_metadata.default(view_1823, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1049 = None
	        convert_element_type_698: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1823, torch.float32);  view_1823 = None
	        _assert_tensor_metadata_1050 = torch.ops.aten._assert_tensor_metadata.default(view_1825, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1050 = None
	        convert_element_type_699: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1825, torch.float32);  view_1825 = None
	        sub_5339: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_698, convert_element_type_699);  convert_element_type_698 = convert_element_type_699 = None
	        mul_11309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5339, view_1824);  sub_5339 = view_1824 = None
	        view_1826: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11309, [sym_size_int, 1500, 1280]);  mul_11309 = None
	        _assert_tensor_metadata_1051 = torch.ops.aten._assert_tensor_metadata.default(view_1826, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1051 = None
	        view_1827: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1828: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1829: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1052 = torch.ops.aten._assert_tensor_metadata.default(view_1827, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1052 = None
	        convert_element_type_700: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1827, torch.float32);  view_1827 = None
	        _assert_tensor_metadata_1053 = torch.ops.aten._assert_tensor_metadata.default(view_1829, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1053 = None
	        convert_element_type_701: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1829, torch.float32);  view_1829 = None
	        sub_5343: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_700, convert_element_type_701);  convert_element_type_700 = convert_element_type_701 = None
	        mul_11314: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5343, view_1828);  sub_5343 = view_1828 = None
	        view_1830: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11314, [1280, 1280]);  mul_11314 = None
	        _assert_tensor_metadata_1054 = torch.ops.aten._assert_tensor_metadata.default(view_1830, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1054 = None
	        mul_11319: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1831: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1826, [mul_11319, 1280]);  view_1826 = mul_11319 = None
	        permute_195: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1830, [1, 0]);  view_1830 = None
	        addmm_96: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_19_self_attn_v_proj_bias, view_1831, permute_195);  model_audio_tower_layers_19_self_attn_v_proj_bias = view_1831 = permute_195 = None
	        view_1832: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_96, [sym_size_int, 1500, 1280]);  addmm_96 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1833: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1832, [sym_size_int, -1, 20, 64]);  view_1832 = None
	        permute_196: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1833, [0, 2, 1, 3]);  view_1833 = None
	        clone_156: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_196, memory_format = torch.contiguous_format);  permute_196 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_19 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_154, clone_155, clone_156, None, False, scale = 1.0);  clone_154 = clone_155 = clone_156 = None
	        getitem_154: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_19[0];  _scaled_dot_product_efficient_attention_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_197: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_154, [0, 2, 1, 3]);  getitem_154 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_197, [sym_size_int, 1500, -1]);  permute_197 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1835: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1834, [sym_size_int, 1500, 1280])
	        amin_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1835, [2])
	        amax_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1835, [2]);  view_1835 = None
	        full_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_117, full_234);  amin_117 = full_234 = None
	        full_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_117, full_235);  amax_117 = full_235 = None
	        sub_5361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_117, minimum_117);  maximum_117 = None
	        div_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5361, 255.0);  sub_5361 = None
	        clamp_min_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_234, 1.1920928955078125e-07);  div_234 = None
	        div_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_117, clamp_min_351);  minimum_117 = None
	        round_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_235);  div_235 = None
	        sub_5367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_235);  round_235 = None
	        clamp_min_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5367, -128);  sub_5367 = None
	        clamp_max_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_352, 127);  clamp_min_352 = None
	        _assert_tensor_metadata_1055 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_351, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1055 = None
	        _assert_tensor_metadata_1056 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_234, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1056 = None
	        convert_element_type_702: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_234, torch.int8);  clamp_max_234 = None
	        view_1836: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1834, [sym_size_int, 1500, 1280]);  view_1834 = None
	        view_1837: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_351, [sym_size_int, 1500, 1])
	        view_1838: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_702, [sym_size_int, 1500, 1])
	        reciprocal_117: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1837);  view_1837 = None
	        mul_11389: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_117, 1.0);  reciprocal_117 = None
	        mul_11392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1836, mul_11389);  view_1836 = mul_11389 = None
	        round_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11392);  mul_11392 = None
	        add_18025: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_236, view_1838);  round_236 = view_1838 = None
	        clamp_min_353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18025, -128);  add_18025 = None
	        clamp_max_235: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_353, 127);  clamp_min_353 = None
	        view_1839: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_235, [sym_size_int, 1500, 1280]);  clamp_max_235 = None
	        _assert_tensor_metadata_1057 = torch.ops.aten._assert_tensor_metadata.default(view_1839, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1057 = None
	        convert_element_type_703: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1839, torch.int8);  view_1839 = None
	        view_1840: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_703, [sym_size_int, 1500, 1280]);  convert_element_type_703 = None
	        view_1841: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_351, [sym_size_int, 1500, 1]);  clamp_min_351 = None
	        view_1842: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_702, [sym_size_int, 1500, 1]);  convert_element_type_702 = None
	        _assert_tensor_metadata_1058 = torch.ops.aten._assert_tensor_metadata.default(view_1840, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1058 = None
	        convert_element_type_704: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1840, torch.float32);  view_1840 = None
	        _assert_tensor_metadata_1059 = torch.ops.aten._assert_tensor_metadata.default(view_1842, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1059 = None
	        convert_element_type_705: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1842, torch.float32);  view_1842 = None
	        sub_5387: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_704, convert_element_type_705);  convert_element_type_704 = convert_element_type_705 = None
	        mul_11414: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5387, view_1841);  sub_5387 = view_1841 = None
	        view_1843: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11414, [sym_size_int, 1500, 1280]);  mul_11414 = None
	        _assert_tensor_metadata_1060 = torch.ops.aten._assert_tensor_metadata.default(view_1843, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1060 = None
	        view_1844: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1845: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1846: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1061 = torch.ops.aten._assert_tensor_metadata.default(view_1844, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1061 = None
	        convert_element_type_706: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1844, torch.float32);  view_1844 = None
	        _assert_tensor_metadata_1062 = torch.ops.aten._assert_tensor_metadata.default(view_1846, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1062 = None
	        convert_element_type_707: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1846, torch.float32);  view_1846 = None
	        sub_5391: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_706, convert_element_type_707);  convert_element_type_706 = convert_element_type_707 = None
	        mul_11419: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5391, view_1845);  sub_5391 = view_1845 = None
	        view_1847: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11419, [1280, 1280]);  mul_11419 = None
	        _assert_tensor_metadata_1063 = torch.ops.aten._assert_tensor_metadata.default(view_1847, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1063 = None
	        mul_11424: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1848: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1843, [mul_11424, 1280]);  view_1843 = mul_11424 = None
	        permute_198: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1847, [1, 0]);  view_1847 = None
	        addmm_97: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_19_self_attn_out_proj_bias, view_1848, permute_198);  model_audio_tower_layers_19_self_attn_out_proj_bias = view_1848 = permute_198 = None
	        view_1849: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_97, [sym_size_int, 1500, 1280]);  addmm_97 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_157: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1849);  view_1849 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_18088: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_17468, clone_157);  add_17468 = clone_157 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_18088, memory_format = torch.contiguous_format)
	        var_mean_39 = torch.ops.aten.var_mean.correction(clone_158, [2], correction = 0, keepdim = True)
	        getitem_158: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_39[0]
	        getitem_159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_39[1];  var_mean_39 = None
	        add_18093: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_158, 1e-05);  getitem_158 = None
	        rsqrt_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_18093);  add_18093 = None
	        sub_5397: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_158, getitem_159);  clone_158 = getitem_159 = None
	        mul_11435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5397, rsqrt_39);  sub_5397 = rsqrt_39 = None
	        mul_11436: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11435, model_audio_tower_layers_19_final_layer_norm_weight);  mul_11435 = model_audio_tower_layers_19_final_layer_norm_weight = None
	        add_18094: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_11436, model_audio_tower_layers_19_final_layer_norm_bias);  mul_11436 = model_audio_tower_layers_19_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1850: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_18094, [sym_size_int, 1500, 1280])
	        amin_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1850, [2])
	        amax_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1850, [2]);  view_1850 = None
	        full_236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_118, full_236);  amin_118 = full_236 = None
	        full_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_118, full_237);  amax_118 = full_237 = None
	        sub_5408: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_118, minimum_118);  maximum_118 = None
	        div_236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5408, 255.0);  sub_5408 = None
	        clamp_min_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_236, 1.1920928955078125e-07);  div_236 = None
	        div_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_118, clamp_min_354);  minimum_118 = None
	        round_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_237);  div_237 = None
	        sub_5414: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_237);  round_237 = None
	        clamp_min_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5414, -128);  sub_5414 = None
	        clamp_max_236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_355, 127);  clamp_min_355 = None
	        _assert_tensor_metadata_1064 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_354, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1064 = None
	        _assert_tensor_metadata_1065 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_236, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1065 = None
	        convert_element_type_708: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_236, torch.int8);  clamp_max_236 = None
	        view_1851: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_18094, [sym_size_int, 1500, 1280]);  add_18094 = None
	        view_1852: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_354, [sym_size_int, 1500, 1])
	        view_1853: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_708, [sym_size_int, 1500, 1])
	        reciprocal_118: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1852);  view_1852 = None
	        mul_11484: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_118, 1.0);  reciprocal_118 = None
	        mul_11487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1851, mul_11484);  view_1851 = mul_11484 = None
	        round_238: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11487);  mul_11487 = None
	        add_18181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_238, view_1853);  round_238 = view_1853 = None
	        clamp_min_356: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18181, -128);  add_18181 = None
	        clamp_max_237: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_356, 127);  clamp_min_356 = None
	        view_1854: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_237, [sym_size_int, 1500, 1280]);  clamp_max_237 = None
	        _assert_tensor_metadata_1066 = torch.ops.aten._assert_tensor_metadata.default(view_1854, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1066 = None
	        convert_element_type_709: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1854, torch.int8);  view_1854 = None
	        view_1855: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_709, [sym_size_int, 1500, 1280]);  convert_element_type_709 = None
	        view_1856: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_354, [sym_size_int, 1500, 1]);  clamp_min_354 = None
	        view_1857: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_708, [sym_size_int, 1500, 1]);  convert_element_type_708 = None
	        _assert_tensor_metadata_1067 = torch.ops.aten._assert_tensor_metadata.default(view_1855, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1067 = None
	        convert_element_type_710: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1855, torch.float32);  view_1855 = None
	        _assert_tensor_metadata_1068 = torch.ops.aten._assert_tensor_metadata.default(view_1857, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1068 = None
	        convert_element_type_711: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1857, torch.float32);  view_1857 = None
	        sub_5434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_710, convert_element_type_711);  convert_element_type_710 = convert_element_type_711 = None
	        mul_11509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5434, view_1856);  sub_5434 = view_1856 = None
	        view_1858: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11509, [sym_size_int, 1500, 1280]);  mul_11509 = None
	        _assert_tensor_metadata_1069 = torch.ops.aten._assert_tensor_metadata.default(view_1858, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1069 = None
	        view_1859: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_19_fc1_parametrizations_weight_original0 = None
	        view_1860: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_19_fc1_parametrizations_weight_original1 = None
	        view_1861: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_19_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1070 = torch.ops.aten._assert_tensor_metadata.default(view_1859, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1070 = None
	        convert_element_type_712: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1859, torch.float32);  view_1859 = None
	        _assert_tensor_metadata_1071 = torch.ops.aten._assert_tensor_metadata.default(view_1861, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1071 = None
	        convert_element_type_713: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1861, torch.float32);  view_1861 = None
	        sub_5438: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_712, convert_element_type_713);  convert_element_type_712 = convert_element_type_713 = None
	        mul_11514: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5438, view_1860);  sub_5438 = view_1860 = None
	        view_1862: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11514, [5120, 1280]);  mul_11514 = None
	        _assert_tensor_metadata_1072 = torch.ops.aten._assert_tensor_metadata.default(view_1862, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1072 = None
	        mul_11519: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1863: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1858, [mul_11519, 1280]);  view_1858 = mul_11519 = None
	        permute_199: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1862, [1, 0]);  view_1862 = None
	        addmm_98: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_19_fc1_bias, view_1863, permute_199);  model_audio_tower_layers_19_fc1_bias = view_1863 = permute_199 = None
	        view_1864: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_98, [sym_size_int, 1500, 5120]);  addmm_98 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_11526: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1864, 0.5)
	        mul_11527: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1864, 0.7071067811865476);  view_1864 = None
	        erf_21: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_11527);  mul_11527 = None
	        add_18240: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_21, 1);  erf_21 = None
	        mul_11528: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11526, add_18240);  mul_11526 = add_18240 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_159: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_11528);  mul_11528 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1865: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_159, [sym_size_int, 1500, 5120])
	        amin_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1865, [2])
	        amax_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1865, [2]);  view_1865 = None
	        full_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_119, full_238);  amin_119 = full_238 = None
	        full_239: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_119, full_239);  amax_119 = full_239 = None
	        sub_5451: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_119, minimum_119);  maximum_119 = None
	        div_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5451, 255.0);  sub_5451 = None
	        clamp_min_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_238, 1.1920928955078125e-07);  div_238 = None
	        div_239: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_119, clamp_min_357);  minimum_119 = None
	        round_239: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_239);  div_239 = None
	        sub_5457: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_239);  round_239 = None
	        clamp_min_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5457, -128);  sub_5457 = None
	        clamp_max_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_358, 127);  clamp_min_358 = None
	        _assert_tensor_metadata_1073 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_357, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1073 = None
	        _assert_tensor_metadata_1074 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_238, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1074 = None
	        convert_element_type_714: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_238, torch.int8);  clamp_max_238 = None
	        view_1866: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_159, [sym_size_int, 1500, 5120]);  clone_159 = None
	        view_1867: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_357, [sym_size_int, 1500, 1])
	        view_1868: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_714, [sym_size_int, 1500, 1])
	        reciprocal_119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1867);  view_1867 = None
	        mul_11574: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_119, 1.0);  reciprocal_119 = None
	        mul_11577: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1866, mul_11574);  view_1866 = mul_11574 = None
	        round_240: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_11577);  mul_11577 = None
	        add_18323: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_240, view_1868);  round_240 = view_1868 = None
	        clamp_min_359: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18323, -128);  add_18323 = None
	        clamp_max_239: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_359, 127);  clamp_min_359 = None
	        view_1869: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_239, [sym_size_int, 1500, 5120]);  clamp_max_239 = None
	        _assert_tensor_metadata_1075 = torch.ops.aten._assert_tensor_metadata.default(view_1869, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1075 = None
	        convert_element_type_715: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1869, torch.int8);  view_1869 = None
	        view_1870: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_715, [sym_size_int, 1500, 5120]);  convert_element_type_715 = None
	        view_1871: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_357, [sym_size_int, 1500, 1]);  clamp_min_357 = None
	        view_1872: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_714, [sym_size_int, 1500, 1]);  convert_element_type_714 = None
	        _assert_tensor_metadata_1076 = torch.ops.aten._assert_tensor_metadata.default(view_1870, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1076 = None
	        convert_element_type_716: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1870, torch.float32);  view_1870 = None
	        _assert_tensor_metadata_1077 = torch.ops.aten._assert_tensor_metadata.default(view_1872, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1077 = None
	        convert_element_type_717: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1872, torch.float32);  view_1872 = None
	        sub_5477: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_716, convert_element_type_717);  convert_element_type_716 = convert_element_type_717 = None
	        mul_11599: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5477, view_1871);  sub_5477 = view_1871 = None
	        view_1873: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_11599, [sym_size_int, 1500, 5120]);  mul_11599 = None
	        _assert_tensor_metadata_1078 = torch.ops.aten._assert_tensor_metadata.default(view_1873, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1078 = None
	        view_1874: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_19_fc2_parametrizations_weight_original0 = None
	        view_1875: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_19_fc2_parametrizations_weight_original1 = None
	        view_1876: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_19_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1079 = torch.ops.aten._assert_tensor_metadata.default(view_1874, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1079 = None
	        convert_element_type_718: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1874, torch.float32);  view_1874 = None
	        _assert_tensor_metadata_1080 = torch.ops.aten._assert_tensor_metadata.default(view_1876, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1080 = None
	        convert_element_type_719: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1876, torch.float32);  view_1876 = None
	        sub_5481: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_718, convert_element_type_719);  convert_element_type_718 = convert_element_type_719 = None
	        mul_11604: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5481, view_1875);  sub_5481 = view_1875 = None
	        view_1877: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_11604, [1280, 5120]);  mul_11604 = None
	        _assert_tensor_metadata_1081 = torch.ops.aten._assert_tensor_metadata.default(view_1877, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1081 = None
	        mul_11609: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1878: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1873, [mul_11609, 5120]);  view_1873 = mul_11609 = None
	        permute_200: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1877, [1, 0]);  view_1877 = None
	        addmm_99: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_19_fc2_bias, view_1878, permute_200);  model_audio_tower_layers_19_fc2_bias = view_1878 = permute_200 = None
	        view_1879: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_99, [sym_size_int, 1500, 1280]);  addmm_99 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1879);  view_1879 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_18386: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_18088, clone_160);  add_18088 = clone_160 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_18386, memory_format = torch.contiguous_format)
	        var_mean_40 = torch.ops.aten.var_mean.correction(clone_161, [2], correction = 0, keepdim = True)
	        getitem_160: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_40[0]
	        getitem_161: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_40[1];  var_mean_40 = None
	        add_18391: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_160, 1e-05);  getitem_160 = None
	        rsqrt_40: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_18391);  add_18391 = None
	        sub_5487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_161, getitem_161);  clone_161 = getitem_161 = None
	        mul_11620: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5487, rsqrt_40);  sub_5487 = rsqrt_40 = None
	        mul_11621: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11620, model_audio_tower_layers_20_self_attn_layer_norm_weight);  mul_11620 = model_audio_tower_layers_20_self_attn_layer_norm_weight = None
	        add_18392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_11621, model_audio_tower_layers_20_self_attn_layer_norm_bias);  mul_11621 = model_audio_tower_layers_20_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1880: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_18392, [sym_size_int, 1500, 1280])
	        amin_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1880, [2])
	        amax_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1880, [2]);  view_1880 = None
	        full_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_120, full_240);  amin_120 = full_240 = None
	        full_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_120, full_241);  amax_120 = full_241 = None
	        sub_5498: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_120, minimum_120);  maximum_120 = None
	        div_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5498, 255.0);  sub_5498 = None
	        clamp_min_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_240, 1.1920928955078125e-07);  div_240 = None
	        div_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_120, clamp_min_360);  minimum_120 = None
	        round_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_241);  div_241 = None
	        sub_5504: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_241);  round_241 = None
	        clamp_min_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5504, -128);  sub_5504 = None
	        clamp_max_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_361, 127);  clamp_min_361 = None
	        _assert_tensor_metadata_1082 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_360, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1082 = None
	        _assert_tensor_metadata_1083 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_240, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1083 = None
	        convert_element_type_720: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_240, torch.int8);  clamp_max_240 = None
	        view_1881: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_18392, [sym_size_int, 1500, 1280])
	        view_1882: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_360, [sym_size_int, 1500, 1])
	        view_1883: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_720, [sym_size_int, 1500, 1])
	        reciprocal_120: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1882);  view_1882 = None
	        mul_11669: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_120, 1.0);  reciprocal_120 = None
	        mul_11672: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1881, mul_11669);  view_1881 = mul_11669 = None
	        round_242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11672);  mul_11672 = None
	        add_18479: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_242, view_1883);  round_242 = view_1883 = None
	        clamp_min_362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18479, -128);  add_18479 = None
	        clamp_max_241: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_362, 127);  clamp_min_362 = None
	        view_1884: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_241, [sym_size_int, 1500, 1280]);  clamp_max_241 = None
	        _assert_tensor_metadata_1084 = torch.ops.aten._assert_tensor_metadata.default(view_1884, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1084 = None
	        convert_element_type_721: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1884, torch.int8);  view_1884 = None
	        view_1885: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_721, [sym_size_int, 1500, 1280]);  convert_element_type_721 = None
	        view_1886: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_360, [sym_size_int, 1500, 1]);  clamp_min_360 = None
	        view_1887: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_720, [sym_size_int, 1500, 1]);  convert_element_type_720 = None
	        _assert_tensor_metadata_1085 = torch.ops.aten._assert_tensor_metadata.default(view_1885, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1085 = None
	        convert_element_type_722: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1885, torch.float32);  view_1885 = None
	        _assert_tensor_metadata_1086 = torch.ops.aten._assert_tensor_metadata.default(view_1887, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1086 = None
	        convert_element_type_723: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1887, torch.float32);  view_1887 = None
	        sub_5524: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_722, convert_element_type_723);  convert_element_type_722 = convert_element_type_723 = None
	        mul_11694: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5524, view_1886);  sub_5524 = view_1886 = None
	        view_1888: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11694, [sym_size_int, 1500, 1280]);  mul_11694 = None
	        _assert_tensor_metadata_1087 = torch.ops.aten._assert_tensor_metadata.default(view_1888, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1087 = None
	        view_1889: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1890: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1891: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1088 = torch.ops.aten._assert_tensor_metadata.default(view_1889, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1088 = None
	        convert_element_type_724: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1889, torch.float32);  view_1889 = None
	        _assert_tensor_metadata_1089 = torch.ops.aten._assert_tensor_metadata.default(view_1891, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1089 = None
	        convert_element_type_725: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1891, torch.float32);  view_1891 = None
	        sub_5528: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_724, convert_element_type_725);  convert_element_type_724 = convert_element_type_725 = None
	        mul_11699: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5528, view_1890);  sub_5528 = view_1890 = None
	        view_1892: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11699, [1280, 1280]);  mul_11699 = None
	        _assert_tensor_metadata_1090 = torch.ops.aten._assert_tensor_metadata.default(view_1892, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1090 = None
	        mul_11704: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1893: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1888, [mul_11704, 1280]);  view_1888 = mul_11704 = None
	        permute_201: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1892, [1, 0]);  view_1892 = None
	        addmm_100: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_20_self_attn_q_proj_bias, view_1893, permute_201);  model_audio_tower_layers_20_self_attn_q_proj_bias = view_1893 = permute_201 = None
	        view_1894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_100, [sym_size_int, 1500, 1280]);  addmm_100 = None
	        mul_11711: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1894, 0.125);  view_1894 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1895: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_11711, [sym_size_int, 1500, 20, 64]);  mul_11711 = None
	        permute_202: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1895, [0, 2, 1, 3]);  view_1895 = None
	        clone_162: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_202, memory_format = torch.contiguous_format);  permute_202 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1896: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_18392, [sym_size_int, 1500, 1280])
	        amin_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1896, [2])
	        amax_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1896, [2]);  view_1896 = None
	        full_242: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_121, full_242);  amin_121 = full_242 = None
	        full_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_121, full_243);  amax_121 = full_243 = None
	        sub_5543: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_121, minimum_121);  maximum_121 = None
	        div_242: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5543, 255.0);  sub_5543 = None
	        clamp_min_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_242, 1.1920928955078125e-07);  div_242 = None
	        div_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_121, clamp_min_363);  minimum_121 = None
	        round_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_243);  div_243 = None
	        sub_5549: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_243);  round_243 = None
	        clamp_min_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5549, -128);  sub_5549 = None
	        clamp_max_242: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_364, 127);  clamp_min_364 = None
	        _assert_tensor_metadata_1091 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_363, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1091 = None
	        _assert_tensor_metadata_1092 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_242, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1092 = None
	        convert_element_type_726: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_242, torch.int8);  clamp_max_242 = None
	        view_1897: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_18392, [sym_size_int, 1500, 1280])
	        view_1898: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_363, [sym_size_int, 1500, 1])
	        view_1899: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_726, [sym_size_int, 1500, 1])
	        reciprocal_121: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1898);  view_1898 = None
	        mul_11765: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_121, 1.0);  reciprocal_121 = None
	        mul_11768: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1897, mul_11765);  view_1897 = mul_11765 = None
	        round_244: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11768);  mul_11768 = None
	        add_18631: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_244, view_1899);  round_244 = view_1899 = None
	        clamp_min_365: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18631, -128);  add_18631 = None
	        clamp_max_243: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_365, 127);  clamp_min_365 = None
	        view_1900: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_243, [sym_size_int, 1500, 1280]);  clamp_max_243 = None
	        _assert_tensor_metadata_1093 = torch.ops.aten._assert_tensor_metadata.default(view_1900, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1093 = None
	        convert_element_type_727: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1900, torch.int8);  view_1900 = None
	        view_1901: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_727, [sym_size_int, 1500, 1280]);  convert_element_type_727 = None
	        view_1902: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_363, [sym_size_int, 1500, 1]);  clamp_min_363 = None
	        view_1903: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_726, [sym_size_int, 1500, 1]);  convert_element_type_726 = None
	        _assert_tensor_metadata_1094 = torch.ops.aten._assert_tensor_metadata.default(view_1901, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1094 = None
	        convert_element_type_728: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1901, torch.float32);  view_1901 = None
	        _assert_tensor_metadata_1095 = torch.ops.aten._assert_tensor_metadata.default(view_1903, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1095 = None
	        convert_element_type_729: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1903, torch.float32);  view_1903 = None
	        sub_5569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_728, convert_element_type_729);  convert_element_type_728 = convert_element_type_729 = None
	        mul_11790: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5569, view_1902);  sub_5569 = view_1902 = None
	        view_1904: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11790, [sym_size_int, 1500, 1280]);  mul_11790 = None
	        _assert_tensor_metadata_1096 = torch.ops.aten._assert_tensor_metadata.default(view_1904, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1096 = None
	        view_1905: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1906: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1907: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1097 = torch.ops.aten._assert_tensor_metadata.default(view_1905, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1097 = None
	        convert_element_type_730: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1905, torch.float32);  view_1905 = None
	        _assert_tensor_metadata_1098 = torch.ops.aten._assert_tensor_metadata.default(view_1907, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1098 = None
	        convert_element_type_731: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1907, torch.float32);  view_1907 = None
	        sub_5573: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_730, convert_element_type_731);  convert_element_type_730 = convert_element_type_731 = None
	        mul_11795: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5573, view_1906);  sub_5573 = view_1906 = None
	        view_1908: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11795, [1280, 1280]);  mul_11795 = None
	        _assert_tensor_metadata_1099 = torch.ops.aten._assert_tensor_metadata.default(view_1908, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1099 = None
	        permute_203: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1908, [1, 0]);  view_1908 = None
	        mul_11798: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1909: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1904, [mul_11798, 1280]);  view_1904 = mul_11798 = None
	        mm_20: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1909, permute_203);  view_1909 = permute_203 = None
	        view_1910: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_20, [sym_size_int, 1500, 1280]);  mm_20 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1911: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1910, [sym_size_int, -1, 20, 64]);  view_1910 = None
	        permute_204: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1911, [0, 2, 1, 3]);  view_1911 = None
	        clone_163: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_204, memory_format = torch.contiguous_format);  permute_204 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_1912: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_18392, [sym_size_int, 1500, 1280])
	        amin_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1912, [2])
	        amax_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1912, [2]);  view_1912 = None
	        full_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_122, full_244);  amin_122 = full_244 = None
	        full_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_122, full_245);  amax_122 = full_245 = None
	        sub_5587: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_122, minimum_122);  maximum_122 = None
	        div_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5587, 255.0);  sub_5587 = None
	        clamp_min_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_244, 1.1920928955078125e-07);  div_244 = None
	        div_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_122, clamp_min_366);  minimum_122 = None
	        round_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_245);  div_245 = None
	        sub_5593: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_245);  round_245 = None
	        clamp_min_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5593, -128);  sub_5593 = None
	        clamp_max_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_367, 127);  clamp_min_367 = None
	        _assert_tensor_metadata_1100 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_366, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1100 = None
	        _assert_tensor_metadata_1101 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_244, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1101 = None
	        convert_element_type_732: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_244, torch.int8);  clamp_max_244 = None
	        view_1913: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_18392, [sym_size_int, 1500, 1280]);  add_18392 = None
	        view_1914: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_366, [sym_size_int, 1500, 1])
	        view_1915: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_732, [sym_size_int, 1500, 1])
	        reciprocal_122: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1914);  view_1914 = None
	        mul_11864: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_122, 1.0);  reciprocal_122 = None
	        mul_11867: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1913, mul_11864);  view_1913 = mul_11864 = None
	        round_246: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11867);  mul_11867 = None
	        add_18779: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_246, view_1915);  round_246 = view_1915 = None
	        clamp_min_368: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18779, -128);  add_18779 = None
	        clamp_max_245: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_368, 127);  clamp_min_368 = None
	        view_1916: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_245, [sym_size_int, 1500, 1280]);  clamp_max_245 = None
	        _assert_tensor_metadata_1102 = torch.ops.aten._assert_tensor_metadata.default(view_1916, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1102 = None
	        convert_element_type_733: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1916, torch.int8);  view_1916 = None
	        view_1917: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_733, [sym_size_int, 1500, 1280]);  convert_element_type_733 = None
	        view_1918: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_366, [sym_size_int, 1500, 1]);  clamp_min_366 = None
	        view_1919: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_732, [sym_size_int, 1500, 1]);  convert_element_type_732 = None
	        _assert_tensor_metadata_1103 = torch.ops.aten._assert_tensor_metadata.default(view_1917, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1103 = None
	        convert_element_type_734: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1917, torch.float32);  view_1917 = None
	        _assert_tensor_metadata_1104 = torch.ops.aten._assert_tensor_metadata.default(view_1919, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1104 = None
	        convert_element_type_735: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1919, torch.float32);  view_1919 = None
	        sub_5613: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_734, convert_element_type_735);  convert_element_type_734 = convert_element_type_735 = None
	        mul_11889: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5613, view_1918);  sub_5613 = view_1918 = None
	        view_1920: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11889, [sym_size_int, 1500, 1280]);  mul_11889 = None
	        _assert_tensor_metadata_1105 = torch.ops.aten._assert_tensor_metadata.default(view_1920, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1105 = None
	        view_1921: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1922: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1923: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1106 = torch.ops.aten._assert_tensor_metadata.default(view_1921, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1106 = None
	        convert_element_type_736: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1921, torch.float32);  view_1921 = None
	        _assert_tensor_metadata_1107 = torch.ops.aten._assert_tensor_metadata.default(view_1923, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1107 = None
	        convert_element_type_737: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1923, torch.float32);  view_1923 = None
	        sub_5617: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_736, convert_element_type_737);  convert_element_type_736 = convert_element_type_737 = None
	        mul_11894: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5617, view_1922);  sub_5617 = view_1922 = None
	        view_1924: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11894, [1280, 1280]);  mul_11894 = None
	        _assert_tensor_metadata_1108 = torch.ops.aten._assert_tensor_metadata.default(view_1924, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1108 = None
	        mul_11899: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1925: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1920, [mul_11899, 1280]);  view_1920 = mul_11899 = None
	        permute_205: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1924, [1, 0]);  view_1924 = None
	        addmm_101: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_20_self_attn_v_proj_bias, view_1925, permute_205);  model_audio_tower_layers_20_self_attn_v_proj_bias = view_1925 = permute_205 = None
	        view_1926: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_101, [sym_size_int, 1500, 1280]);  addmm_101 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1927: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1926, [sym_size_int, -1, 20, 64]);  view_1926 = None
	        permute_206: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1927, [0, 2, 1, 3]);  view_1927 = None
	        clone_164: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_206, memory_format = torch.contiguous_format);  permute_206 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_20 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_162, clone_163, clone_164, None, False, scale = 1.0);  clone_162 = clone_163 = clone_164 = None
	        getitem_162: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_20[0];  _scaled_dot_product_efficient_attention_20 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_207: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_162, [0, 2, 1, 3]);  getitem_162 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1928: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_207, [sym_size_int, 1500, -1]);  permute_207 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_1929: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1928, [sym_size_int, 1500, 1280])
	        amin_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1929, [2])
	        amax_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1929, [2]);  view_1929 = None
	        full_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_123, full_246);  amin_123 = full_246 = None
	        full_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_123, full_247);  amax_123 = full_247 = None
	        sub_5635: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_123, minimum_123);  maximum_123 = None
	        div_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5635, 255.0);  sub_5635 = None
	        clamp_min_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_246, 1.1920928955078125e-07);  div_246 = None
	        div_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_123, clamp_min_369);  minimum_123 = None
	        round_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_247);  div_247 = None
	        sub_5641: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_247);  round_247 = None
	        clamp_min_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5641, -128);  sub_5641 = None
	        clamp_max_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_370, 127);  clamp_min_370 = None
	        _assert_tensor_metadata_1109 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_369, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1109 = None
	        _assert_tensor_metadata_1110 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_246, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1110 = None
	        convert_element_type_738: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_246, torch.int8);  clamp_max_246 = None
	        view_1930: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_1928, [sym_size_int, 1500, 1280]);  view_1928 = None
	        view_1931: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_369, [sym_size_int, 1500, 1])
	        view_1932: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_738, [sym_size_int, 1500, 1])
	        reciprocal_123: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1931);  view_1931 = None
	        mul_11969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_123, 1.0);  reciprocal_123 = None
	        mul_11972: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1930, mul_11969);  view_1930 = mul_11969 = None
	        round_248: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11972);  mul_11972 = None
	        add_18943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_248, view_1932);  round_248 = view_1932 = None
	        clamp_min_371: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18943, -128);  add_18943 = None
	        clamp_max_247: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_371, 127);  clamp_min_371 = None
	        view_1933: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_247, [sym_size_int, 1500, 1280]);  clamp_max_247 = None
	        _assert_tensor_metadata_1111 = torch.ops.aten._assert_tensor_metadata.default(view_1933, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1111 = None
	        convert_element_type_739: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1933, torch.int8);  view_1933 = None
	        view_1934: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_739, [sym_size_int, 1500, 1280]);  convert_element_type_739 = None
	        view_1935: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_369, [sym_size_int, 1500, 1]);  clamp_min_369 = None
	        view_1936: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_738, [sym_size_int, 1500, 1]);  convert_element_type_738 = None
	        _assert_tensor_metadata_1112 = torch.ops.aten._assert_tensor_metadata.default(view_1934, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1112 = None
	        convert_element_type_740: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1934, torch.float32);  view_1934 = None
	        _assert_tensor_metadata_1113 = torch.ops.aten._assert_tensor_metadata.default(view_1936, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1113 = None
	        convert_element_type_741: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1936, torch.float32);  view_1936 = None
	        sub_5661: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_740, convert_element_type_741);  convert_element_type_740 = convert_element_type_741 = None
	        mul_11994: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5661, view_1935);  sub_5661 = view_1935 = None
	        view_1937: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11994, [sym_size_int, 1500, 1280]);  mul_11994 = None
	        _assert_tensor_metadata_1114 = torch.ops.aten._assert_tensor_metadata.default(view_1937, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1114 = None
	        view_1938: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1939: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1940: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1115 = torch.ops.aten._assert_tensor_metadata.default(view_1938, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1115 = None
	        convert_element_type_742: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1938, torch.float32);  view_1938 = None
	        _assert_tensor_metadata_1116 = torch.ops.aten._assert_tensor_metadata.default(view_1940, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1116 = None
	        convert_element_type_743: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1940, torch.float32);  view_1940 = None
	        sub_5665: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_742, convert_element_type_743);  convert_element_type_742 = convert_element_type_743 = None
	        mul_11999: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5665, view_1939);  sub_5665 = view_1939 = None
	        view_1941: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11999, [1280, 1280]);  mul_11999 = None
	        _assert_tensor_metadata_1117 = torch.ops.aten._assert_tensor_metadata.default(view_1941, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1117 = None
	        mul_12004: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1942: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1937, [mul_12004, 1280]);  view_1937 = mul_12004 = None
	        permute_208: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1941, [1, 0]);  view_1941 = None
	        addmm_102: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_20_self_attn_out_proj_bias, view_1942, permute_208);  model_audio_tower_layers_20_self_attn_out_proj_bias = view_1942 = permute_208 = None
	        view_1943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_102, [sym_size_int, 1500, 1280]);  addmm_102 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_165: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1943);  view_1943 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_19006: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_18386, clone_165);  add_18386 = clone_165 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_166: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_19006, memory_format = torch.contiguous_format)
	        var_mean_41 = torch.ops.aten.var_mean.correction(clone_166, [2], correction = 0, keepdim = True)
	        getitem_166: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_41[0]
	        getitem_167: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_41[1];  var_mean_41 = None
	        add_19011: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_166, 1e-05);  getitem_166 = None
	        rsqrt_41: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_19011);  add_19011 = None
	        sub_5671: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_166, getitem_167);  clone_166 = getitem_167 = None
	        mul_12015: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5671, rsqrt_41);  sub_5671 = rsqrt_41 = None
	        mul_12016: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12015, model_audio_tower_layers_20_final_layer_norm_weight);  mul_12015 = model_audio_tower_layers_20_final_layer_norm_weight = None
	        add_19012: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12016, model_audio_tower_layers_20_final_layer_norm_bias);  mul_12016 = model_audio_tower_layers_20_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_1944: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19012, [sym_size_int, 1500, 1280])
	        amin_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1944, [2])
	        amax_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1944, [2]);  view_1944 = None
	        full_248: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_124, full_248);  amin_124 = full_248 = None
	        full_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_124, full_249);  amax_124 = full_249 = None
	        sub_5682: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_124, minimum_124);  maximum_124 = None
	        div_248: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5682, 255.0);  sub_5682 = None
	        clamp_min_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_248, 1.1920928955078125e-07);  div_248 = None
	        div_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_124, clamp_min_372);  minimum_124 = None
	        round_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_249);  div_249 = None
	        sub_5688: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_249);  round_249 = None
	        clamp_min_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5688, -128);  sub_5688 = None
	        clamp_max_248: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_373, 127);  clamp_min_373 = None
	        _assert_tensor_metadata_1118 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_372, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1118 = None
	        _assert_tensor_metadata_1119 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_248, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1119 = None
	        convert_element_type_744: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_248, torch.int8);  clamp_max_248 = None
	        view_1945: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19012, [sym_size_int, 1500, 1280]);  add_19012 = None
	        view_1946: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_372, [sym_size_int, 1500, 1])
	        view_1947: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_744, [sym_size_int, 1500, 1])
	        reciprocal_124: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1946);  view_1946 = None
	        mul_12064: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_124, 1.0);  reciprocal_124 = None
	        mul_12067: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1945, mul_12064);  view_1945 = mul_12064 = None
	        round_250: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12067);  mul_12067 = None
	        add_19099: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_250, view_1947);  round_250 = view_1947 = None
	        clamp_min_374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19099, -128);  add_19099 = None
	        clamp_max_249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_374, 127);  clamp_min_374 = None
	        view_1948: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_249, [sym_size_int, 1500, 1280]);  clamp_max_249 = None
	        _assert_tensor_metadata_1120 = torch.ops.aten._assert_tensor_metadata.default(view_1948, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1120 = None
	        convert_element_type_745: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1948, torch.int8);  view_1948 = None
	        view_1949: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_745, [sym_size_int, 1500, 1280]);  convert_element_type_745 = None
	        view_1950: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_372, [sym_size_int, 1500, 1]);  clamp_min_372 = None
	        view_1951: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_744, [sym_size_int, 1500, 1]);  convert_element_type_744 = None
	        _assert_tensor_metadata_1121 = torch.ops.aten._assert_tensor_metadata.default(view_1949, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1121 = None
	        convert_element_type_746: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1949, torch.float32);  view_1949 = None
	        _assert_tensor_metadata_1122 = torch.ops.aten._assert_tensor_metadata.default(view_1951, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1122 = None
	        convert_element_type_747: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1951, torch.float32);  view_1951 = None
	        sub_5708: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_746, convert_element_type_747);  convert_element_type_746 = convert_element_type_747 = None
	        mul_12089: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5708, view_1950);  sub_5708 = view_1950 = None
	        view_1952: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12089, [sym_size_int, 1500, 1280]);  mul_12089 = None
	        _assert_tensor_metadata_1123 = torch.ops.aten._assert_tensor_metadata.default(view_1952, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1123 = None
	        view_1953: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_20_fc1_parametrizations_weight_original0 = None
	        view_1954: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_20_fc1_parametrizations_weight_original1 = None
	        view_1955: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_20_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1124 = torch.ops.aten._assert_tensor_metadata.default(view_1953, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1124 = None
	        convert_element_type_748: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1953, torch.float32);  view_1953 = None
	        _assert_tensor_metadata_1125 = torch.ops.aten._assert_tensor_metadata.default(view_1955, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1125 = None
	        convert_element_type_749: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1955, torch.float32);  view_1955 = None
	        sub_5712: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_748, convert_element_type_749);  convert_element_type_748 = convert_element_type_749 = None
	        mul_12094: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5712, view_1954);  sub_5712 = view_1954 = None
	        view_1956: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12094, [5120, 1280]);  mul_12094 = None
	        _assert_tensor_metadata_1126 = torch.ops.aten._assert_tensor_metadata.default(view_1956, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1126 = None
	        mul_12099: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1957: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1952, [mul_12099, 1280]);  view_1952 = mul_12099 = None
	        permute_209: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1956, [1, 0]);  view_1956 = None
	        addmm_103: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_20_fc1_bias, view_1957, permute_209);  model_audio_tower_layers_20_fc1_bias = view_1957 = permute_209 = None
	        view_1958: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_103, [sym_size_int, 1500, 5120]);  addmm_103 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_12106: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1958, 0.5)
	        mul_12107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1958, 0.7071067811865476);  view_1958 = None
	        erf_22: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_12107);  mul_12107 = None
	        add_19158: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_22, 1);  erf_22 = None
	        mul_12108: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12106, add_19158);  mul_12106 = add_19158 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_167: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_12108);  mul_12108 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_1959: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_167, [sym_size_int, 1500, 5120])
	        amin_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1959, [2])
	        amax_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1959, [2]);  view_1959 = None
	        full_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_125, full_250);  amin_125 = full_250 = None
	        full_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_125, full_251);  amax_125 = full_251 = None
	        sub_5725: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_125, minimum_125);  maximum_125 = None
	        div_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5725, 255.0);  sub_5725 = None
	        clamp_min_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_250, 1.1920928955078125e-07);  div_250 = None
	        div_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_125, clamp_min_375);  minimum_125 = None
	        round_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_251);  div_251 = None
	        sub_5731: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_251);  round_251 = None
	        clamp_min_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5731, -128);  sub_5731 = None
	        clamp_max_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_376, 127);  clamp_min_376 = None
	        _assert_tensor_metadata_1127 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_375, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1127 = None
	        _assert_tensor_metadata_1128 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_250, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1128 = None
	        convert_element_type_750: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_250, torch.int8);  clamp_max_250 = None
	        view_1960: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_167, [sym_size_int, 1500, 5120]);  clone_167 = None
	        view_1961: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_375, [sym_size_int, 1500, 1])
	        view_1962: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_750, [sym_size_int, 1500, 1])
	        reciprocal_125: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1961);  view_1961 = None
	        mul_12154: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_125, 1.0);  reciprocal_125 = None
	        mul_12157: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1960, mul_12154);  view_1960 = mul_12154 = None
	        round_252: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_12157);  mul_12157 = None
	        add_19241: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_252, view_1962);  round_252 = view_1962 = None
	        clamp_min_377: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19241, -128);  add_19241 = None
	        clamp_max_251: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_377, 127);  clamp_min_377 = None
	        view_1963: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_251, [sym_size_int, 1500, 5120]);  clamp_max_251 = None
	        _assert_tensor_metadata_1129 = torch.ops.aten._assert_tensor_metadata.default(view_1963, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1129 = None
	        convert_element_type_751: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1963, torch.int8);  view_1963 = None
	        view_1964: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_751, [sym_size_int, 1500, 5120]);  convert_element_type_751 = None
	        view_1965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_375, [sym_size_int, 1500, 1]);  clamp_min_375 = None
	        view_1966: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_750, [sym_size_int, 1500, 1]);  convert_element_type_750 = None
	        _assert_tensor_metadata_1130 = torch.ops.aten._assert_tensor_metadata.default(view_1964, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1130 = None
	        convert_element_type_752: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1964, torch.float32);  view_1964 = None
	        _assert_tensor_metadata_1131 = torch.ops.aten._assert_tensor_metadata.default(view_1966, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1131 = None
	        convert_element_type_753: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1966, torch.float32);  view_1966 = None
	        sub_5751: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_752, convert_element_type_753);  convert_element_type_752 = convert_element_type_753 = None
	        mul_12179: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5751, view_1965);  sub_5751 = view_1965 = None
	        view_1967: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_12179, [sym_size_int, 1500, 5120]);  mul_12179 = None
	        _assert_tensor_metadata_1132 = torch.ops.aten._assert_tensor_metadata.default(view_1967, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1132 = None
	        view_1968: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_20_fc2_parametrizations_weight_original0 = None
	        view_1969: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_20_fc2_parametrizations_weight_original1 = None
	        view_1970: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_20_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1133 = torch.ops.aten._assert_tensor_metadata.default(view_1968, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1133 = None
	        convert_element_type_754: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1968, torch.float32);  view_1968 = None
	        _assert_tensor_metadata_1134 = torch.ops.aten._assert_tensor_metadata.default(view_1970, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1134 = None
	        convert_element_type_755: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1970, torch.float32);  view_1970 = None
	        sub_5755: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_754, convert_element_type_755);  convert_element_type_754 = convert_element_type_755 = None
	        mul_12184: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5755, view_1969);  sub_5755 = view_1969 = None
	        view_1971: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_12184, [1280, 5120]);  mul_12184 = None
	        _assert_tensor_metadata_1135 = torch.ops.aten._assert_tensor_metadata.default(view_1971, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1135 = None
	        mul_12189: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1972: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_1967, [mul_12189, 5120]);  view_1967 = mul_12189 = None
	        permute_210: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1971, [1, 0]);  view_1971 = None
	        addmm_104: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_20_fc2_bias, view_1972, permute_210);  model_audio_tower_layers_20_fc2_bias = view_1972 = permute_210 = None
	        view_1973: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_104, [sym_size_int, 1500, 1280]);  addmm_104 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_168: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_1973);  view_1973 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_19304: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_19006, clone_168);  add_19006 = clone_168 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_169: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_19304, memory_format = torch.contiguous_format)
	        var_mean_42 = torch.ops.aten.var_mean.correction(clone_169, [2], correction = 0, keepdim = True)
	        getitem_168: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_42[0]
	        getitem_169: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_42[1];  var_mean_42 = None
	        add_19309: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_168, 1e-05);  getitem_168 = None
	        rsqrt_42: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_19309);  add_19309 = None
	        sub_5761: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_169, getitem_169);  clone_169 = getitem_169 = None
	        mul_12200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5761, rsqrt_42);  sub_5761 = rsqrt_42 = None
	        mul_12201: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12200, model_audio_tower_layers_21_self_attn_layer_norm_weight);  mul_12200 = model_audio_tower_layers_21_self_attn_layer_norm_weight = None
	        add_19310: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12201, model_audio_tower_layers_21_self_attn_layer_norm_bias);  mul_12201 = model_audio_tower_layers_21_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_1974: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19310, [sym_size_int, 1500, 1280])
	        amin_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1974, [2])
	        amax_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1974, [2]);  view_1974 = None
	        full_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_126, full_252);  amin_126 = full_252 = None
	        full_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_126, full_253);  amax_126 = full_253 = None
	        sub_5772: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_126, minimum_126);  maximum_126 = None
	        div_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5772, 255.0);  sub_5772 = None
	        clamp_min_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_252, 1.1920928955078125e-07);  div_252 = None
	        div_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_126, clamp_min_378);  minimum_126 = None
	        round_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_253);  div_253 = None
	        sub_5778: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_253);  round_253 = None
	        clamp_min_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5778, -128);  sub_5778 = None
	        clamp_max_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_379, 127);  clamp_min_379 = None
	        _assert_tensor_metadata_1136 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_378, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1136 = None
	        _assert_tensor_metadata_1137 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_252, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1137 = None
	        convert_element_type_756: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_252, torch.int8);  clamp_max_252 = None
	        view_1975: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19310, [sym_size_int, 1500, 1280])
	        view_1976: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_378, [sym_size_int, 1500, 1])
	        view_1977: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_756, [sym_size_int, 1500, 1])
	        reciprocal_126: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1976);  view_1976 = None
	        mul_12249: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_126, 1.0);  reciprocal_126 = None
	        mul_12252: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1975, mul_12249);  view_1975 = mul_12249 = None
	        round_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12252);  mul_12252 = None
	        add_19397: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_254, view_1977);  round_254 = view_1977 = None
	        clamp_min_380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19397, -128);  add_19397 = None
	        clamp_max_253: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_380, 127);  clamp_min_380 = None
	        view_1978: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_253, [sym_size_int, 1500, 1280]);  clamp_max_253 = None
	        _assert_tensor_metadata_1138 = torch.ops.aten._assert_tensor_metadata.default(view_1978, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1138 = None
	        convert_element_type_757: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1978, torch.int8);  view_1978 = None
	        view_1979: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_757, [sym_size_int, 1500, 1280]);  convert_element_type_757 = None
	        view_1980: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_378, [sym_size_int, 1500, 1]);  clamp_min_378 = None
	        view_1981: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_756, [sym_size_int, 1500, 1]);  convert_element_type_756 = None
	        _assert_tensor_metadata_1139 = torch.ops.aten._assert_tensor_metadata.default(view_1979, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1139 = None
	        convert_element_type_758: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1979, torch.float32);  view_1979 = None
	        _assert_tensor_metadata_1140 = torch.ops.aten._assert_tensor_metadata.default(view_1981, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1140 = None
	        convert_element_type_759: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1981, torch.float32);  view_1981 = None
	        sub_5798: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_758, convert_element_type_759);  convert_element_type_758 = convert_element_type_759 = None
	        mul_12274: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5798, view_1980);  sub_5798 = view_1980 = None
	        view_1982: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12274, [sym_size_int, 1500, 1280]);  mul_12274 = None
	        _assert_tensor_metadata_1141 = torch.ops.aten._assert_tensor_metadata.default(view_1982, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1141 = None
	        view_1983: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1984: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1985: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1142 = torch.ops.aten._assert_tensor_metadata.default(view_1983, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1142 = None
	        convert_element_type_760: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1983, torch.float32);  view_1983 = None
	        _assert_tensor_metadata_1143 = torch.ops.aten._assert_tensor_metadata.default(view_1985, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1143 = None
	        convert_element_type_761: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1985, torch.float32);  view_1985 = None
	        sub_5802: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_760, convert_element_type_761);  convert_element_type_760 = convert_element_type_761 = None
	        mul_12279: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5802, view_1984);  sub_5802 = view_1984 = None
	        view_1986: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12279, [1280, 1280]);  mul_12279 = None
	        _assert_tensor_metadata_1144 = torch.ops.aten._assert_tensor_metadata.default(view_1986, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1144 = None
	        mul_12284: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1987: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1982, [mul_12284, 1280]);  view_1982 = mul_12284 = None
	        permute_211: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1986, [1, 0]);  view_1986 = None
	        addmm_105: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_21_self_attn_q_proj_bias, view_1987, permute_211);  model_audio_tower_layers_21_self_attn_q_proj_bias = view_1987 = permute_211 = None
	        view_1988: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_105, [sym_size_int, 1500, 1280]);  addmm_105 = None
	        mul_12291: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1988, 0.125);  view_1988 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1989: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_12291, [sym_size_int, 1500, 20, 64]);  mul_12291 = None
	        permute_212: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1989, [0, 2, 1, 3]);  view_1989 = None
	        clone_170: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_212, memory_format = torch.contiguous_format);  permute_212 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_1990: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19310, [sym_size_int, 1500, 1280])
	        amin_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1990, [2])
	        amax_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1990, [2]);  view_1990 = None
	        full_254: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_127, full_254);  amin_127 = full_254 = None
	        full_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_127, full_255);  amax_127 = full_255 = None
	        sub_5817: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_127, minimum_127);  maximum_127 = None
	        div_254: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5817, 255.0);  sub_5817 = None
	        clamp_min_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_254, 1.1920928955078125e-07);  div_254 = None
	        div_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_127, clamp_min_381);  minimum_127 = None
	        round_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_255);  div_255 = None
	        sub_5823: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_255);  round_255 = None
	        clamp_min_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5823, -128);  sub_5823 = None
	        clamp_max_254: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_382, 127);  clamp_min_382 = None
	        _assert_tensor_metadata_1145 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_381, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1145 = None
	        _assert_tensor_metadata_1146 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_254, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1146 = None
	        convert_element_type_762: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_254, torch.int8);  clamp_max_254 = None
	        view_1991: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19310, [sym_size_int, 1500, 1280])
	        view_1992: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_381, [sym_size_int, 1500, 1])
	        view_1993: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_762, [sym_size_int, 1500, 1])
	        reciprocal_127: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1992);  view_1992 = None
	        mul_12345: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_127, 1.0);  reciprocal_127 = None
	        mul_12348: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1991, mul_12345);  view_1991 = mul_12345 = None
	        round_256: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12348);  mul_12348 = None
	        add_19549: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_256, view_1993);  round_256 = view_1993 = None
	        clamp_min_383: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19549, -128);  add_19549 = None
	        clamp_max_255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_383, 127);  clamp_min_383 = None
	        view_1994: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_255, [sym_size_int, 1500, 1280]);  clamp_max_255 = None
	        _assert_tensor_metadata_1147 = torch.ops.aten._assert_tensor_metadata.default(view_1994, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1147 = None
	        convert_element_type_763: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1994, torch.int8);  view_1994 = None
	        view_1995: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_763, [sym_size_int, 1500, 1280]);  convert_element_type_763 = None
	        view_1996: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_381, [sym_size_int, 1500, 1]);  clamp_min_381 = None
	        view_1997: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_762, [sym_size_int, 1500, 1]);  convert_element_type_762 = None
	        _assert_tensor_metadata_1148 = torch.ops.aten._assert_tensor_metadata.default(view_1995, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1148 = None
	        convert_element_type_764: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1995, torch.float32);  view_1995 = None
	        _assert_tensor_metadata_1149 = torch.ops.aten._assert_tensor_metadata.default(view_1997, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1149 = None
	        convert_element_type_765: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1997, torch.float32);  view_1997 = None
	        sub_5843: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_764, convert_element_type_765);  convert_element_type_764 = convert_element_type_765 = None
	        mul_12370: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5843, view_1996);  sub_5843 = view_1996 = None
	        view_1998: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12370, [sym_size_int, 1500, 1280]);  mul_12370 = None
	        _assert_tensor_metadata_1150 = torch.ops.aten._assert_tensor_metadata.default(view_1998, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1150 = None
	        view_1999: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2000: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2001: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1151 = torch.ops.aten._assert_tensor_metadata.default(view_1999, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1151 = None
	        convert_element_type_766: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1999, torch.float32);  view_1999 = None
	        _assert_tensor_metadata_1152 = torch.ops.aten._assert_tensor_metadata.default(view_2001, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1152 = None
	        convert_element_type_767: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2001, torch.float32);  view_2001 = None
	        sub_5847: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_766, convert_element_type_767);  convert_element_type_766 = convert_element_type_767 = None
	        mul_12375: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5847, view_2000);  sub_5847 = view_2000 = None
	        view_2002: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12375, [1280, 1280]);  mul_12375 = None
	        _assert_tensor_metadata_1153 = torch.ops.aten._assert_tensor_metadata.default(view_2002, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1153 = None
	        permute_213: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2002, [1, 0]);  view_2002 = None
	        mul_12378: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2003: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_1998, [mul_12378, 1280]);  view_1998 = mul_12378 = None
	        mm_21: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2003, permute_213);  view_2003 = permute_213 = None
	        view_2004: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_21, [sym_size_int, 1500, 1280]);  mm_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2005: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2004, [sym_size_int, -1, 20, 64]);  view_2004 = None
	        permute_214: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2005, [0, 2, 1, 3]);  view_2005 = None
	        clone_171: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_214, memory_format = torch.contiguous_format);  permute_214 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2006: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19310, [sym_size_int, 1500, 1280])
	        amin_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2006, [2])
	        amax_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2006, [2]);  view_2006 = None
	        full_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_128, full_256);  amin_128 = full_256 = None
	        full_257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_128, full_257);  amax_128 = full_257 = None
	        sub_5861: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_128, minimum_128);  maximum_128 = None
	        div_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5861, 255.0);  sub_5861 = None
	        clamp_min_384: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_256, 1.1920928955078125e-07);  div_256 = None
	        div_257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_128, clamp_min_384);  minimum_128 = None
	        round_257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_257);  div_257 = None
	        sub_5867: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_257);  round_257 = None
	        clamp_min_385: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5867, -128);  sub_5867 = None
	        clamp_max_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_385, 127);  clamp_min_385 = None
	        _assert_tensor_metadata_1154 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_384, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1154 = None
	        _assert_tensor_metadata_1155 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_256, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1155 = None
	        convert_element_type_768: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_256, torch.int8);  clamp_max_256 = None
	        view_2007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19310, [sym_size_int, 1500, 1280]);  add_19310 = None
	        view_2008: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_384, [sym_size_int, 1500, 1])
	        view_2009: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_768, [sym_size_int, 1500, 1])
	        reciprocal_128: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2008);  view_2008 = None
	        mul_12444: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_128, 1.0);  reciprocal_128 = None
	        mul_12447: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2007, mul_12444);  view_2007 = mul_12444 = None
	        round_258: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12447);  mul_12447 = None
	        add_19697: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_258, view_2009);  round_258 = view_2009 = None
	        clamp_min_386: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19697, -128);  add_19697 = None
	        clamp_max_257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_386, 127);  clamp_min_386 = None
	        view_2010: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_257, [sym_size_int, 1500, 1280]);  clamp_max_257 = None
	        _assert_tensor_metadata_1156 = torch.ops.aten._assert_tensor_metadata.default(view_2010, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1156 = None
	        convert_element_type_769: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2010, torch.int8);  view_2010 = None
	        view_2011: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_769, [sym_size_int, 1500, 1280]);  convert_element_type_769 = None
	        view_2012: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_384, [sym_size_int, 1500, 1]);  clamp_min_384 = None
	        view_2013: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_768, [sym_size_int, 1500, 1]);  convert_element_type_768 = None
	        _assert_tensor_metadata_1157 = torch.ops.aten._assert_tensor_metadata.default(view_2011, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1157 = None
	        convert_element_type_770: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2011, torch.float32);  view_2011 = None
	        _assert_tensor_metadata_1158 = torch.ops.aten._assert_tensor_metadata.default(view_2013, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1158 = None
	        convert_element_type_771: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2013, torch.float32);  view_2013 = None
	        sub_5887: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_770, convert_element_type_771);  convert_element_type_770 = convert_element_type_771 = None
	        mul_12469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5887, view_2012);  sub_5887 = view_2012 = None
	        view_2014: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12469, [sym_size_int, 1500, 1280]);  mul_12469 = None
	        _assert_tensor_metadata_1159 = torch.ops.aten._assert_tensor_metadata.default(view_2014, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1159 = None
	        view_2015: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2016: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2017: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1160 = torch.ops.aten._assert_tensor_metadata.default(view_2015, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1160 = None
	        convert_element_type_772: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2015, torch.float32);  view_2015 = None
	        _assert_tensor_metadata_1161 = torch.ops.aten._assert_tensor_metadata.default(view_2017, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1161 = None
	        convert_element_type_773: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2017, torch.float32);  view_2017 = None
	        sub_5891: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_772, convert_element_type_773);  convert_element_type_772 = convert_element_type_773 = None
	        mul_12474: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5891, view_2016);  sub_5891 = view_2016 = None
	        view_2018: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12474, [1280, 1280]);  mul_12474 = None
	        _assert_tensor_metadata_1162 = torch.ops.aten._assert_tensor_metadata.default(view_2018, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1162 = None
	        mul_12479: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2019: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2014, [mul_12479, 1280]);  view_2014 = mul_12479 = None
	        permute_215: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2018, [1, 0]);  view_2018 = None
	        addmm_106: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_21_self_attn_v_proj_bias, view_2019, permute_215);  model_audio_tower_layers_21_self_attn_v_proj_bias = view_2019 = permute_215 = None
	        view_2020: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_106, [sym_size_int, 1500, 1280]);  addmm_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2021: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2020, [sym_size_int, -1, 20, 64]);  view_2020 = None
	        permute_216: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2021, [0, 2, 1, 3]);  view_2021 = None
	        clone_172: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_216, memory_format = torch.contiguous_format);  permute_216 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_21 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_170, clone_171, clone_172, None, False, scale = 1.0);  clone_170 = clone_171 = clone_172 = None
	        getitem_170: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_21[0];  _scaled_dot_product_efficient_attention_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_217: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_170, [0, 2, 1, 3]);  getitem_170 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2022: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_217, [sym_size_int, 1500, -1]);  permute_217 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2023: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2022, [sym_size_int, 1500, 1280])
	        amin_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2023, [2])
	        amax_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2023, [2]);  view_2023 = None
	        full_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_129, full_258);  amin_129 = full_258 = None
	        full_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_129, full_259);  amax_129 = full_259 = None
	        sub_5909: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_129, minimum_129);  maximum_129 = None
	        div_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5909, 255.0);  sub_5909 = None
	        clamp_min_387: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_258, 1.1920928955078125e-07);  div_258 = None
	        div_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_129, clamp_min_387);  minimum_129 = None
	        round_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_259);  div_259 = None
	        sub_5915: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_259);  round_259 = None
	        clamp_min_388: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5915, -128);  sub_5915 = None
	        clamp_max_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_388, 127);  clamp_min_388 = None
	        _assert_tensor_metadata_1163 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_387, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1163 = None
	        _assert_tensor_metadata_1164 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_258, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1164 = None
	        convert_element_type_774: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_258, torch.int8);  clamp_max_258 = None
	        view_2024: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2022, [sym_size_int, 1500, 1280]);  view_2022 = None
	        view_2025: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_387, [sym_size_int, 1500, 1])
	        view_2026: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_774, [sym_size_int, 1500, 1])
	        reciprocal_129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2025);  view_2025 = None
	        mul_12549: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_129, 1.0);  reciprocal_129 = None
	        mul_12552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2024, mul_12549);  view_2024 = mul_12549 = None
	        round_260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12552);  mul_12552 = None
	        add_19861: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_260, view_2026);  round_260 = view_2026 = None
	        clamp_min_389: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19861, -128);  add_19861 = None
	        clamp_max_259: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_389, 127);  clamp_min_389 = None
	        view_2027: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_259, [sym_size_int, 1500, 1280]);  clamp_max_259 = None
	        _assert_tensor_metadata_1165 = torch.ops.aten._assert_tensor_metadata.default(view_2027, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1165 = None
	        convert_element_type_775: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2027, torch.int8);  view_2027 = None
	        view_2028: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_775, [sym_size_int, 1500, 1280]);  convert_element_type_775 = None
	        view_2029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_387, [sym_size_int, 1500, 1]);  clamp_min_387 = None
	        view_2030: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_774, [sym_size_int, 1500, 1]);  convert_element_type_774 = None
	        _assert_tensor_metadata_1166 = torch.ops.aten._assert_tensor_metadata.default(view_2028, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1166 = None
	        convert_element_type_776: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2028, torch.float32);  view_2028 = None
	        _assert_tensor_metadata_1167 = torch.ops.aten._assert_tensor_metadata.default(view_2030, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1167 = None
	        convert_element_type_777: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2030, torch.float32);  view_2030 = None
	        sub_5935: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_776, convert_element_type_777);  convert_element_type_776 = convert_element_type_777 = None
	        mul_12574: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5935, view_2029);  sub_5935 = view_2029 = None
	        view_2031: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12574, [sym_size_int, 1500, 1280]);  mul_12574 = None
	        _assert_tensor_metadata_1168 = torch.ops.aten._assert_tensor_metadata.default(view_2031, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1168 = None
	        view_2032: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2033: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2034: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1169 = torch.ops.aten._assert_tensor_metadata.default(view_2032, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1169 = None
	        convert_element_type_778: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2032, torch.float32);  view_2032 = None
	        _assert_tensor_metadata_1170 = torch.ops.aten._assert_tensor_metadata.default(view_2034, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1170 = None
	        convert_element_type_779: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2034, torch.float32);  view_2034 = None
	        sub_5939: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_778, convert_element_type_779);  convert_element_type_778 = convert_element_type_779 = None
	        mul_12579: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5939, view_2033);  sub_5939 = view_2033 = None
	        view_2035: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12579, [1280, 1280]);  mul_12579 = None
	        _assert_tensor_metadata_1171 = torch.ops.aten._assert_tensor_metadata.default(view_2035, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1171 = None
	        mul_12584: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2036: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2031, [mul_12584, 1280]);  view_2031 = mul_12584 = None
	        permute_218: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2035, [1, 0]);  view_2035 = None
	        addmm_107: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_21_self_attn_out_proj_bias, view_2036, permute_218);  model_audio_tower_layers_21_self_attn_out_proj_bias = view_2036 = permute_218 = None
	        view_2037: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_107, [sym_size_int, 1500, 1280]);  addmm_107 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2037);  view_2037 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_19924: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_19304, clone_173);  add_19304 = clone_173 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_19924, memory_format = torch.contiguous_format)
	        var_mean_43 = torch.ops.aten.var_mean.correction(clone_174, [2], correction = 0, keepdim = True)
	        getitem_174: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_43[0]
	        getitem_175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_43[1];  var_mean_43 = None
	        add_19929: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_174, 1e-05);  getitem_174 = None
	        rsqrt_43: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_19929);  add_19929 = None
	        sub_5945: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_174, getitem_175);  clone_174 = getitem_175 = None
	        mul_12595: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5945, rsqrt_43);  sub_5945 = rsqrt_43 = None
	        mul_12596: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12595, model_audio_tower_layers_21_final_layer_norm_weight);  mul_12595 = model_audio_tower_layers_21_final_layer_norm_weight = None
	        add_19930: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12596, model_audio_tower_layers_21_final_layer_norm_bias);  mul_12596 = model_audio_tower_layers_21_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2038: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19930, [sym_size_int, 1500, 1280])
	        amin_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2038, [2])
	        amax_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2038, [2]);  view_2038 = None
	        full_260: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_130, full_260);  amin_130 = full_260 = None
	        full_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_130, full_261);  amax_130 = full_261 = None
	        sub_5956: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_130, minimum_130);  maximum_130 = None
	        div_260: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5956, 255.0);  sub_5956 = None
	        clamp_min_390: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_260, 1.1920928955078125e-07);  div_260 = None
	        div_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_130, clamp_min_390);  minimum_130 = None
	        round_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_261);  div_261 = None
	        sub_5962: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_261);  round_261 = None
	        clamp_min_391: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5962, -128);  sub_5962 = None
	        clamp_max_260: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_391, 127);  clamp_min_391 = None
	        _assert_tensor_metadata_1172 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_390, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1172 = None
	        _assert_tensor_metadata_1173 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_260, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1173 = None
	        convert_element_type_780: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_260, torch.int8);  clamp_max_260 = None
	        view_2039: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_19930, [sym_size_int, 1500, 1280]);  add_19930 = None
	        view_2040: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_390, [sym_size_int, 1500, 1])
	        view_2041: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_780, [sym_size_int, 1500, 1])
	        reciprocal_130: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2040);  view_2040 = None
	        mul_12644: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_130, 1.0);  reciprocal_130 = None
	        mul_12647: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2039, mul_12644);  view_2039 = mul_12644 = None
	        round_262: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12647);  mul_12647 = None
	        add_20017: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_262, view_2041);  round_262 = view_2041 = None
	        clamp_min_392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20017, -128);  add_20017 = None
	        clamp_max_261: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_392, 127);  clamp_min_392 = None
	        view_2042: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_261, [sym_size_int, 1500, 1280]);  clamp_max_261 = None
	        _assert_tensor_metadata_1174 = torch.ops.aten._assert_tensor_metadata.default(view_2042, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1174 = None
	        convert_element_type_781: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2042, torch.int8);  view_2042 = None
	        view_2043: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_781, [sym_size_int, 1500, 1280]);  convert_element_type_781 = None
	        view_2044: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_390, [sym_size_int, 1500, 1]);  clamp_min_390 = None
	        view_2045: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_780, [sym_size_int, 1500, 1]);  convert_element_type_780 = None
	        _assert_tensor_metadata_1175 = torch.ops.aten._assert_tensor_metadata.default(view_2043, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1175 = None
	        convert_element_type_782: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2043, torch.float32);  view_2043 = None
	        _assert_tensor_metadata_1176 = torch.ops.aten._assert_tensor_metadata.default(view_2045, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1176 = None
	        convert_element_type_783: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2045, torch.float32);  view_2045 = None
	        sub_5982: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_782, convert_element_type_783);  convert_element_type_782 = convert_element_type_783 = None
	        mul_12669: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5982, view_2044);  sub_5982 = view_2044 = None
	        view_2046: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12669, [sym_size_int, 1500, 1280]);  mul_12669 = None
	        _assert_tensor_metadata_1177 = torch.ops.aten._assert_tensor_metadata.default(view_2046, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1177 = None
	        view_2047: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_21_fc1_parametrizations_weight_original0 = None
	        view_2048: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_21_fc1_parametrizations_weight_original1 = None
	        view_2049: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_21_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1178 = torch.ops.aten._assert_tensor_metadata.default(view_2047, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1178 = None
	        convert_element_type_784: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2047, torch.float32);  view_2047 = None
	        _assert_tensor_metadata_1179 = torch.ops.aten._assert_tensor_metadata.default(view_2049, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1179 = None
	        convert_element_type_785: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2049, torch.float32);  view_2049 = None
	        sub_5986: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_784, convert_element_type_785);  convert_element_type_784 = convert_element_type_785 = None
	        mul_12674: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5986, view_2048);  sub_5986 = view_2048 = None
	        view_2050: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12674, [5120, 1280]);  mul_12674 = None
	        _assert_tensor_metadata_1180 = torch.ops.aten._assert_tensor_metadata.default(view_2050, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1180 = None
	        mul_12679: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2051: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2046, [mul_12679, 1280]);  view_2046 = mul_12679 = None
	        permute_219: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2050, [1, 0]);  view_2050 = None
	        addmm_108: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_21_fc1_bias, view_2051, permute_219);  model_audio_tower_layers_21_fc1_bias = view_2051 = permute_219 = None
	        view_2052: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_108, [sym_size_int, 1500, 5120]);  addmm_108 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_12686: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2052, 0.5)
	        mul_12687: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2052, 0.7071067811865476);  view_2052 = None
	        erf_23: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_12687);  mul_12687 = None
	        add_20076: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_23, 1);  erf_23 = None
	        mul_12688: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12686, add_20076);  mul_12686 = add_20076 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_175: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_12688);  mul_12688 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2053: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_175, [sym_size_int, 1500, 5120])
	        amin_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2053, [2])
	        amax_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2053, [2]);  view_2053 = None
	        full_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_131, full_262);  amin_131 = full_262 = None
	        full_263: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_131, full_263);  amax_131 = full_263 = None
	        sub_5999: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_131, minimum_131);  maximum_131 = None
	        div_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5999, 255.0);  sub_5999 = None
	        clamp_min_393: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_262, 1.1920928955078125e-07);  div_262 = None
	        div_263: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_131, clamp_min_393);  minimum_131 = None
	        round_263: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_263);  div_263 = None
	        sub_6005: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_263);  round_263 = None
	        clamp_min_394: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6005, -128);  sub_6005 = None
	        clamp_max_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_394, 127);  clamp_min_394 = None
	        _assert_tensor_metadata_1181 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_393, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1181 = None
	        _assert_tensor_metadata_1182 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_262, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1182 = None
	        convert_element_type_786: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_262, torch.int8);  clamp_max_262 = None
	        view_2054: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_175, [sym_size_int, 1500, 5120]);  clone_175 = None
	        view_2055: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_393, [sym_size_int, 1500, 1])
	        view_2056: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_786, [sym_size_int, 1500, 1])
	        reciprocal_131: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2055);  view_2055 = None
	        mul_12734: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_131, 1.0);  reciprocal_131 = None
	        mul_12737: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2054, mul_12734);  view_2054 = mul_12734 = None
	        round_264: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_12737);  mul_12737 = None
	        add_20159: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_264, view_2056);  round_264 = view_2056 = None
	        clamp_min_395: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20159, -128);  add_20159 = None
	        clamp_max_263: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_395, 127);  clamp_min_395 = None
	        view_2057: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_263, [sym_size_int, 1500, 5120]);  clamp_max_263 = None
	        _assert_tensor_metadata_1183 = torch.ops.aten._assert_tensor_metadata.default(view_2057, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1183 = None
	        convert_element_type_787: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2057, torch.int8);  view_2057 = None
	        view_2058: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_787, [sym_size_int, 1500, 5120]);  convert_element_type_787 = None
	        view_2059: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_393, [sym_size_int, 1500, 1]);  clamp_min_393 = None
	        view_2060: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_786, [sym_size_int, 1500, 1]);  convert_element_type_786 = None
	        _assert_tensor_metadata_1184 = torch.ops.aten._assert_tensor_metadata.default(view_2058, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1184 = None
	        convert_element_type_788: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2058, torch.float32);  view_2058 = None
	        _assert_tensor_metadata_1185 = torch.ops.aten._assert_tensor_metadata.default(view_2060, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1185 = None
	        convert_element_type_789: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2060, torch.float32);  view_2060 = None
	        sub_6025: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_788, convert_element_type_789);  convert_element_type_788 = convert_element_type_789 = None
	        mul_12759: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6025, view_2059);  sub_6025 = view_2059 = None
	        view_2061: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_12759, [sym_size_int, 1500, 5120]);  mul_12759 = None
	        _assert_tensor_metadata_1186 = torch.ops.aten._assert_tensor_metadata.default(view_2061, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1186 = None
	        view_2062: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_21_fc2_parametrizations_weight_original0 = None
	        view_2063: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_21_fc2_parametrizations_weight_original1 = None
	        view_2064: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_21_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1187 = torch.ops.aten._assert_tensor_metadata.default(view_2062, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1187 = None
	        convert_element_type_790: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2062, torch.float32);  view_2062 = None
	        _assert_tensor_metadata_1188 = torch.ops.aten._assert_tensor_metadata.default(view_2064, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1188 = None
	        convert_element_type_791: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2064, torch.float32);  view_2064 = None
	        sub_6029: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_790, convert_element_type_791);  convert_element_type_790 = convert_element_type_791 = None
	        mul_12764: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6029, view_2063);  sub_6029 = view_2063 = None
	        view_2065: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_12764, [1280, 5120]);  mul_12764 = None
	        _assert_tensor_metadata_1189 = torch.ops.aten._assert_tensor_metadata.default(view_2065, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1189 = None
	        mul_12769: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2066: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2061, [mul_12769, 5120]);  view_2061 = mul_12769 = None
	        permute_220: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2065, [1, 0]);  view_2065 = None
	        addmm_109: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_21_fc2_bias, view_2066, permute_220);  model_audio_tower_layers_21_fc2_bias = view_2066 = permute_220 = None
	        view_2067: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_109, [sym_size_int, 1500, 1280]);  addmm_109 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2067);  view_2067 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_20222: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_19924, clone_176);  add_19924 = clone_176 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_177: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_20222, memory_format = torch.contiguous_format)
	        var_mean_44 = torch.ops.aten.var_mean.correction(clone_177, [2], correction = 0, keepdim = True)
	        getitem_176: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_44[0]
	        getitem_177: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_44[1];  var_mean_44 = None
	        add_20227: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_176, 1e-05);  getitem_176 = None
	        rsqrt_44: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_20227);  add_20227 = None
	        sub_6035: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_177, getitem_177);  clone_177 = getitem_177 = None
	        mul_12780: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6035, rsqrt_44);  sub_6035 = rsqrt_44 = None
	        mul_12781: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12780, model_audio_tower_layers_22_self_attn_layer_norm_weight);  mul_12780 = model_audio_tower_layers_22_self_attn_layer_norm_weight = None
	        add_20228: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12781, model_audio_tower_layers_22_self_attn_layer_norm_bias);  mul_12781 = model_audio_tower_layers_22_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2068: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_20228, [sym_size_int, 1500, 1280])
	        amin_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2068, [2])
	        amax_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2068, [2]);  view_2068 = None
	        full_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_132, full_264);  amin_132 = full_264 = None
	        full_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_132, full_265);  amax_132 = full_265 = None
	        sub_6046: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_132, minimum_132);  maximum_132 = None
	        div_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6046, 255.0);  sub_6046 = None
	        clamp_min_396: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_264, 1.1920928955078125e-07);  div_264 = None
	        div_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_132, clamp_min_396);  minimum_132 = None
	        round_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_265);  div_265 = None
	        sub_6052: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_265);  round_265 = None
	        clamp_min_397: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6052, -128);  sub_6052 = None
	        clamp_max_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_397, 127);  clamp_min_397 = None
	        _assert_tensor_metadata_1190 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_396, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1190 = None
	        _assert_tensor_metadata_1191 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_264, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1191 = None
	        convert_element_type_792: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_264, torch.int8);  clamp_max_264 = None
	        view_2069: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_20228, [sym_size_int, 1500, 1280])
	        view_2070: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_396, [sym_size_int, 1500, 1])
	        view_2071: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_792, [sym_size_int, 1500, 1])
	        reciprocal_132: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2070);  view_2070 = None
	        mul_12829: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_132, 1.0);  reciprocal_132 = None
	        mul_12832: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2069, mul_12829);  view_2069 = mul_12829 = None
	        round_266: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12832);  mul_12832 = None
	        add_20315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_266, view_2071);  round_266 = view_2071 = None
	        clamp_min_398: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20315, -128);  add_20315 = None
	        clamp_max_265: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_398, 127);  clamp_min_398 = None
	        view_2072: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_265, [sym_size_int, 1500, 1280]);  clamp_max_265 = None
	        _assert_tensor_metadata_1192 = torch.ops.aten._assert_tensor_metadata.default(view_2072, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1192 = None
	        convert_element_type_793: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2072, torch.int8);  view_2072 = None
	        view_2073: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_793, [sym_size_int, 1500, 1280]);  convert_element_type_793 = None
	        view_2074: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_396, [sym_size_int, 1500, 1]);  clamp_min_396 = None
	        view_2075: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_792, [sym_size_int, 1500, 1]);  convert_element_type_792 = None
	        _assert_tensor_metadata_1193 = torch.ops.aten._assert_tensor_metadata.default(view_2073, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1193 = None
	        convert_element_type_794: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2073, torch.float32);  view_2073 = None
	        _assert_tensor_metadata_1194 = torch.ops.aten._assert_tensor_metadata.default(view_2075, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1194 = None
	        convert_element_type_795: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2075, torch.float32);  view_2075 = None
	        sub_6072: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_794, convert_element_type_795);  convert_element_type_794 = convert_element_type_795 = None
	        mul_12854: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6072, view_2074);  sub_6072 = view_2074 = None
	        view_2076: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12854, [sym_size_int, 1500, 1280]);  mul_12854 = None
	        _assert_tensor_metadata_1195 = torch.ops.aten._assert_tensor_metadata.default(view_2076, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1195 = None
	        view_2077: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2078: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2079: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1196 = torch.ops.aten._assert_tensor_metadata.default(view_2077, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1196 = None
	        convert_element_type_796: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2077, torch.float32);  view_2077 = None
	        _assert_tensor_metadata_1197 = torch.ops.aten._assert_tensor_metadata.default(view_2079, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1197 = None
	        convert_element_type_797: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2079, torch.float32);  view_2079 = None
	        sub_6076: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_796, convert_element_type_797);  convert_element_type_796 = convert_element_type_797 = None
	        mul_12859: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6076, view_2078);  sub_6076 = view_2078 = None
	        view_2080: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12859, [1280, 1280]);  mul_12859 = None
	        _assert_tensor_metadata_1198 = torch.ops.aten._assert_tensor_metadata.default(view_2080, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1198 = None
	        mul_12864: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2081: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2076, [mul_12864, 1280]);  view_2076 = mul_12864 = None
	        permute_221: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2080, [1, 0]);  view_2080 = None
	        addmm_110: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_22_self_attn_q_proj_bias, view_2081, permute_221);  model_audio_tower_layers_22_self_attn_q_proj_bias = view_2081 = permute_221 = None
	        view_2082: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_110, [sym_size_int, 1500, 1280]);  addmm_110 = None
	        mul_12871: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2082, 0.125);  view_2082 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2083: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_12871, [sym_size_int, 1500, 20, 64]);  mul_12871 = None
	        permute_222: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2083, [0, 2, 1, 3]);  view_2083 = None
	        clone_178: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_222, memory_format = torch.contiguous_format);  permute_222 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2084: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_20228, [sym_size_int, 1500, 1280])
	        amin_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2084, [2])
	        amax_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2084, [2]);  view_2084 = None
	        full_266: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_133, full_266);  amin_133 = full_266 = None
	        full_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_133, full_267);  amax_133 = full_267 = None
	        sub_6091: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_133, minimum_133);  maximum_133 = None
	        div_266: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6091, 255.0);  sub_6091 = None
	        clamp_min_399: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_266, 1.1920928955078125e-07);  div_266 = None
	        div_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_133, clamp_min_399);  minimum_133 = None
	        round_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_267);  div_267 = None
	        sub_6097: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_267);  round_267 = None
	        clamp_min_400: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6097, -128);  sub_6097 = None
	        clamp_max_266: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_400, 127);  clamp_min_400 = None
	        _assert_tensor_metadata_1199 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_399, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1199 = None
	        _assert_tensor_metadata_1200 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_266, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1200 = None
	        convert_element_type_798: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_266, torch.int8);  clamp_max_266 = None
	        view_2085: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_20228, [sym_size_int, 1500, 1280])
	        view_2086: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_399, [sym_size_int, 1500, 1])
	        view_2087: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_798, [sym_size_int, 1500, 1])
	        reciprocal_133: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2086);  view_2086 = None
	        mul_12925: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_133, 1.0);  reciprocal_133 = None
	        mul_12928: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2085, mul_12925);  view_2085 = mul_12925 = None
	        round_268: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12928);  mul_12928 = None
	        add_20467: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_268, view_2087);  round_268 = view_2087 = None
	        clamp_min_401: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20467, -128);  add_20467 = None
	        clamp_max_267: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_401, 127);  clamp_min_401 = None
	        view_2088: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_267, [sym_size_int, 1500, 1280]);  clamp_max_267 = None
	        _assert_tensor_metadata_1201 = torch.ops.aten._assert_tensor_metadata.default(view_2088, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1201 = None
	        convert_element_type_799: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2088, torch.int8);  view_2088 = None
	        view_2089: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_799, [sym_size_int, 1500, 1280]);  convert_element_type_799 = None
	        view_2090: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_399, [sym_size_int, 1500, 1]);  clamp_min_399 = None
	        view_2091: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_798, [sym_size_int, 1500, 1]);  convert_element_type_798 = None
	        _assert_tensor_metadata_1202 = torch.ops.aten._assert_tensor_metadata.default(view_2089, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1202 = None
	        convert_element_type_800: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2089, torch.float32);  view_2089 = None
	        _assert_tensor_metadata_1203 = torch.ops.aten._assert_tensor_metadata.default(view_2091, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1203 = None
	        convert_element_type_801: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2091, torch.float32);  view_2091 = None
	        sub_6117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_800, convert_element_type_801);  convert_element_type_800 = convert_element_type_801 = None
	        mul_12950: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6117, view_2090);  sub_6117 = view_2090 = None
	        view_2092: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12950, [sym_size_int, 1500, 1280]);  mul_12950 = None
	        _assert_tensor_metadata_1204 = torch.ops.aten._assert_tensor_metadata.default(view_2092, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1204 = None
	        view_2093: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2094: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2095: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1205 = torch.ops.aten._assert_tensor_metadata.default(view_2093, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1205 = None
	        convert_element_type_802: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2093, torch.float32);  view_2093 = None
	        _assert_tensor_metadata_1206 = torch.ops.aten._assert_tensor_metadata.default(view_2095, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1206 = None
	        convert_element_type_803: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2095, torch.float32);  view_2095 = None
	        sub_6121: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_802, convert_element_type_803);  convert_element_type_802 = convert_element_type_803 = None
	        mul_12955: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6121, view_2094);  sub_6121 = view_2094 = None
	        view_2096: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12955, [1280, 1280]);  mul_12955 = None
	        _assert_tensor_metadata_1207 = torch.ops.aten._assert_tensor_metadata.default(view_2096, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1207 = None
	        permute_223: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2096, [1, 0]);  view_2096 = None
	        mul_12958: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2097: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2092, [mul_12958, 1280]);  view_2092 = mul_12958 = None
	        mm_22: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2097, permute_223);  view_2097 = permute_223 = None
	        view_2098: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_22, [sym_size_int, 1500, 1280]);  mm_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2099: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2098, [sym_size_int, -1, 20, 64]);  view_2098 = None
	        permute_224: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2099, [0, 2, 1, 3]);  view_2099 = None
	        clone_179: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_224, memory_format = torch.contiguous_format);  permute_224 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2100: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_20228, [sym_size_int, 1500, 1280])
	        amin_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2100, [2])
	        amax_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2100, [2]);  view_2100 = None
	        full_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_134, full_268);  amin_134 = full_268 = None
	        full_269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_134, full_269);  amax_134 = full_269 = None
	        sub_6135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_134, minimum_134);  maximum_134 = None
	        div_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6135, 255.0);  sub_6135 = None
	        clamp_min_402: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_268, 1.1920928955078125e-07);  div_268 = None
	        div_269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_134, clamp_min_402);  minimum_134 = None
	        round_269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_269);  div_269 = None
	        sub_6141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_269);  round_269 = None
	        clamp_min_403: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6141, -128);  sub_6141 = None
	        clamp_max_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_403, 127);  clamp_min_403 = None
	        _assert_tensor_metadata_1208 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_402, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1208 = None
	        _assert_tensor_metadata_1209 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_268, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1209 = None
	        convert_element_type_804: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_268, torch.int8);  clamp_max_268 = None
	        view_2101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_20228, [sym_size_int, 1500, 1280]);  add_20228 = None
	        view_2102: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_402, [sym_size_int, 1500, 1])
	        view_2103: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_804, [sym_size_int, 1500, 1])
	        reciprocal_134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2102);  view_2102 = None
	        mul_13024: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_134, 1.0);  reciprocal_134 = None
	        mul_13027: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2101, mul_13024);  view_2101 = mul_13024 = None
	        round_270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13027);  mul_13027 = None
	        add_20615: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_270, view_2103);  round_270 = view_2103 = None
	        clamp_min_404: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20615, -128);  add_20615 = None
	        clamp_max_269: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_404, 127);  clamp_min_404 = None
	        view_2104: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_269, [sym_size_int, 1500, 1280]);  clamp_max_269 = None
	        _assert_tensor_metadata_1210 = torch.ops.aten._assert_tensor_metadata.default(view_2104, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1210 = None
	        convert_element_type_805: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2104, torch.int8);  view_2104 = None
	        view_2105: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_805, [sym_size_int, 1500, 1280]);  convert_element_type_805 = None
	        view_2106: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_402, [sym_size_int, 1500, 1]);  clamp_min_402 = None
	        view_2107: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_804, [sym_size_int, 1500, 1]);  convert_element_type_804 = None
	        _assert_tensor_metadata_1211 = torch.ops.aten._assert_tensor_metadata.default(view_2105, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1211 = None
	        convert_element_type_806: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2105, torch.float32);  view_2105 = None
	        _assert_tensor_metadata_1212 = torch.ops.aten._assert_tensor_metadata.default(view_2107, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1212 = None
	        convert_element_type_807: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2107, torch.float32);  view_2107 = None
	        sub_6161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_806, convert_element_type_807);  convert_element_type_806 = convert_element_type_807 = None
	        mul_13049: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6161, view_2106);  sub_6161 = view_2106 = None
	        view_2108: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13049, [sym_size_int, 1500, 1280]);  mul_13049 = None
	        _assert_tensor_metadata_1213 = torch.ops.aten._assert_tensor_metadata.default(view_2108, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1213 = None
	        view_2109: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2110: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2111: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1214 = torch.ops.aten._assert_tensor_metadata.default(view_2109, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1214 = None
	        convert_element_type_808: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2109, torch.float32);  view_2109 = None
	        _assert_tensor_metadata_1215 = torch.ops.aten._assert_tensor_metadata.default(view_2111, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1215 = None
	        convert_element_type_809: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2111, torch.float32);  view_2111 = None
	        sub_6165: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_808, convert_element_type_809);  convert_element_type_808 = convert_element_type_809 = None
	        mul_13054: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6165, view_2110);  sub_6165 = view_2110 = None
	        view_2112: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13054, [1280, 1280]);  mul_13054 = None
	        _assert_tensor_metadata_1216 = torch.ops.aten._assert_tensor_metadata.default(view_2112, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1216 = None
	        mul_13059: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2113: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2108, [mul_13059, 1280]);  view_2108 = mul_13059 = None
	        permute_225: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2112, [1, 0]);  view_2112 = None
	        addmm_111: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_22_self_attn_v_proj_bias, view_2113, permute_225);  model_audio_tower_layers_22_self_attn_v_proj_bias = view_2113 = permute_225 = None
	        view_2114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_111, [sym_size_int, 1500, 1280]);  addmm_111 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2115: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2114, [sym_size_int, -1, 20, 64]);  view_2114 = None
	        permute_226: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2115, [0, 2, 1, 3]);  view_2115 = None
	        clone_180: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_226, memory_format = torch.contiguous_format);  permute_226 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_22 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_178, clone_179, clone_180, None, False, scale = 1.0);  clone_178 = clone_179 = clone_180 = None
	        getitem_178: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_22[0];  _scaled_dot_product_efficient_attention_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_227: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_178, [0, 2, 1, 3]);  getitem_178 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_227, [sym_size_int, 1500, -1]);  permute_227 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2116, [sym_size_int, 1500, 1280])
	        amin_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2117, [2])
	        amax_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2117, [2]);  view_2117 = None
	        full_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_135, full_270);  amin_135 = full_270 = None
	        full_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_135, full_271);  amax_135 = full_271 = None
	        sub_6183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_135, minimum_135);  maximum_135 = None
	        div_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6183, 255.0);  sub_6183 = None
	        clamp_min_405: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_270, 1.1920928955078125e-07);  div_270 = None
	        div_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_135, clamp_min_405);  minimum_135 = None
	        round_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_271);  div_271 = None
	        sub_6189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_271);  round_271 = None
	        clamp_min_406: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6189, -128);  sub_6189 = None
	        clamp_max_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_406, 127);  clamp_min_406 = None
	        _assert_tensor_metadata_1217 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_405, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1217 = None
	        _assert_tensor_metadata_1218 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_270, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1218 = None
	        convert_element_type_810: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_270, torch.int8);  clamp_max_270 = None
	        view_2118: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2116, [sym_size_int, 1500, 1280]);  view_2116 = None
	        view_2119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_405, [sym_size_int, 1500, 1])
	        view_2120: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_810, [sym_size_int, 1500, 1])
	        reciprocal_135: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2119);  view_2119 = None
	        mul_13129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_135, 1.0);  reciprocal_135 = None
	        mul_13132: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2118, mul_13129);  view_2118 = mul_13129 = None
	        round_272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13132);  mul_13132 = None
	        add_20779: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_272, view_2120);  round_272 = view_2120 = None
	        clamp_min_407: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20779, -128);  add_20779 = None
	        clamp_max_271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_407, 127);  clamp_min_407 = None
	        view_2121: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_271, [sym_size_int, 1500, 1280]);  clamp_max_271 = None
	        _assert_tensor_metadata_1219 = torch.ops.aten._assert_tensor_metadata.default(view_2121, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1219 = None
	        convert_element_type_811: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2121, torch.int8);  view_2121 = None
	        view_2122: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_811, [sym_size_int, 1500, 1280]);  convert_element_type_811 = None
	        view_2123: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_405, [sym_size_int, 1500, 1]);  clamp_min_405 = None
	        view_2124: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_810, [sym_size_int, 1500, 1]);  convert_element_type_810 = None
	        _assert_tensor_metadata_1220 = torch.ops.aten._assert_tensor_metadata.default(view_2122, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1220 = None
	        convert_element_type_812: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2122, torch.float32);  view_2122 = None
	        _assert_tensor_metadata_1221 = torch.ops.aten._assert_tensor_metadata.default(view_2124, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1221 = None
	        convert_element_type_813: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2124, torch.float32);  view_2124 = None
	        sub_6209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_812, convert_element_type_813);  convert_element_type_812 = convert_element_type_813 = None
	        mul_13154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6209, view_2123);  sub_6209 = view_2123 = None
	        view_2125: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13154, [sym_size_int, 1500, 1280]);  mul_13154 = None
	        _assert_tensor_metadata_1222 = torch.ops.aten._assert_tensor_metadata.default(view_2125, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1222 = None
	        view_2126: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2127: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2128: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1223 = torch.ops.aten._assert_tensor_metadata.default(view_2126, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1223 = None
	        convert_element_type_814: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2126, torch.float32);  view_2126 = None
	        _assert_tensor_metadata_1224 = torch.ops.aten._assert_tensor_metadata.default(view_2128, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1224 = None
	        convert_element_type_815: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2128, torch.float32);  view_2128 = None
	        sub_6213: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_814, convert_element_type_815);  convert_element_type_814 = convert_element_type_815 = None
	        mul_13159: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6213, view_2127);  sub_6213 = view_2127 = None
	        view_2129: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13159, [1280, 1280]);  mul_13159 = None
	        _assert_tensor_metadata_1225 = torch.ops.aten._assert_tensor_metadata.default(view_2129, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1225 = None
	        mul_13164: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2130: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2125, [mul_13164, 1280]);  view_2125 = mul_13164 = None
	        permute_228: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2129, [1, 0]);  view_2129 = None
	        addmm_112: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_22_self_attn_out_proj_bias, view_2130, permute_228);  model_audio_tower_layers_22_self_attn_out_proj_bias = view_2130 = permute_228 = None
	        view_2131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_112, [sym_size_int, 1500, 1280]);  addmm_112 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2131);  view_2131 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_20842: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_20222, clone_181);  add_20222 = clone_181 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_20842, memory_format = torch.contiguous_format)
	        var_mean_45 = torch.ops.aten.var_mean.correction(clone_182, [2], correction = 0, keepdim = True)
	        getitem_182: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_45[0]
	        getitem_183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_45[1];  var_mean_45 = None
	        add_20847: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_182, 1e-05);  getitem_182 = None
	        rsqrt_45: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_20847);  add_20847 = None
	        sub_6219: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_182, getitem_183);  clone_182 = getitem_183 = None
	        mul_13175: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6219, rsqrt_45);  sub_6219 = rsqrt_45 = None
	        mul_13176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13175, model_audio_tower_layers_22_final_layer_norm_weight);  mul_13175 = model_audio_tower_layers_22_final_layer_norm_weight = None
	        add_20848: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13176, model_audio_tower_layers_22_final_layer_norm_bias);  mul_13176 = model_audio_tower_layers_22_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2132: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_20848, [sym_size_int, 1500, 1280])
	        amin_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2132, [2])
	        amax_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2132, [2]);  view_2132 = None
	        full_272: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_136, full_272);  amin_136 = full_272 = None
	        full_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_136, full_273);  amax_136 = full_273 = None
	        sub_6230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_136, minimum_136);  maximum_136 = None
	        div_272: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6230, 255.0);  sub_6230 = None
	        clamp_min_408: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_272, 1.1920928955078125e-07);  div_272 = None
	        div_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_136, clamp_min_408);  minimum_136 = None
	        round_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_273);  div_273 = None
	        sub_6236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_273);  round_273 = None
	        clamp_min_409: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6236, -128);  sub_6236 = None
	        clamp_max_272: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_409, 127);  clamp_min_409 = None
	        _assert_tensor_metadata_1226 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_408, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1226 = None
	        _assert_tensor_metadata_1227 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_272, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1227 = None
	        convert_element_type_816: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_272, torch.int8);  clamp_max_272 = None
	        view_2133: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_20848, [sym_size_int, 1500, 1280]);  add_20848 = None
	        view_2134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_408, [sym_size_int, 1500, 1])
	        view_2135: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_816, [sym_size_int, 1500, 1])
	        reciprocal_136: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2134);  view_2134 = None
	        mul_13224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_136, 1.0);  reciprocal_136 = None
	        mul_13227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2133, mul_13224);  view_2133 = mul_13224 = None
	        round_274: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13227);  mul_13227 = None
	        add_20935: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_274, view_2135);  round_274 = view_2135 = None
	        clamp_min_410: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20935, -128);  add_20935 = None
	        clamp_max_273: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_410, 127);  clamp_min_410 = None
	        view_2136: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_273, [sym_size_int, 1500, 1280]);  clamp_max_273 = None
	        _assert_tensor_metadata_1228 = torch.ops.aten._assert_tensor_metadata.default(view_2136, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1228 = None
	        convert_element_type_817: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2136, torch.int8);  view_2136 = None
	        view_2137: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_817, [sym_size_int, 1500, 1280]);  convert_element_type_817 = None
	        view_2138: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_408, [sym_size_int, 1500, 1]);  clamp_min_408 = None
	        view_2139: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_816, [sym_size_int, 1500, 1]);  convert_element_type_816 = None
	        _assert_tensor_metadata_1229 = torch.ops.aten._assert_tensor_metadata.default(view_2137, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1229 = None
	        convert_element_type_818: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2137, torch.float32);  view_2137 = None
	        _assert_tensor_metadata_1230 = torch.ops.aten._assert_tensor_metadata.default(view_2139, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1230 = None
	        convert_element_type_819: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2139, torch.float32);  view_2139 = None
	        sub_6256: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_818, convert_element_type_819);  convert_element_type_818 = convert_element_type_819 = None
	        mul_13249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6256, view_2138);  sub_6256 = view_2138 = None
	        view_2140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13249, [sym_size_int, 1500, 1280]);  mul_13249 = None
	        _assert_tensor_metadata_1231 = torch.ops.aten._assert_tensor_metadata.default(view_2140, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1231 = None
	        view_2141: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_22_fc1_parametrizations_weight_original0 = None
	        view_2142: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_22_fc1_parametrizations_weight_original1 = None
	        view_2143: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_22_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1232 = torch.ops.aten._assert_tensor_metadata.default(view_2141, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1232 = None
	        convert_element_type_820: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2141, torch.float32);  view_2141 = None
	        _assert_tensor_metadata_1233 = torch.ops.aten._assert_tensor_metadata.default(view_2143, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1233 = None
	        convert_element_type_821: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2143, torch.float32);  view_2143 = None
	        sub_6260: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_820, convert_element_type_821);  convert_element_type_820 = convert_element_type_821 = None
	        mul_13254: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6260, view_2142);  sub_6260 = view_2142 = None
	        view_2144: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13254, [5120, 1280]);  mul_13254 = None
	        _assert_tensor_metadata_1234 = torch.ops.aten._assert_tensor_metadata.default(view_2144, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1234 = None
	        mul_13259: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2145: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2140, [mul_13259, 1280]);  view_2140 = mul_13259 = None
	        permute_229: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2144, [1, 0]);  view_2144 = None
	        addmm_113: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_22_fc1_bias, view_2145, permute_229);  model_audio_tower_layers_22_fc1_bias = view_2145 = permute_229 = None
	        view_2146: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_113, [sym_size_int, 1500, 5120]);  addmm_113 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_13266: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2146, 0.5)
	        mul_13267: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2146, 0.7071067811865476);  view_2146 = None
	        erf_24: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_13267);  mul_13267 = None
	        add_20994: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_24, 1);  erf_24 = None
	        mul_13268: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13266, add_20994);  mul_13266 = add_20994 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_183: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_13268);  mul_13268 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2147: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_183, [sym_size_int, 1500, 5120])
	        amin_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2147, [2])
	        amax_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2147, [2]);  view_2147 = None
	        full_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_137, full_274);  amin_137 = full_274 = None
	        full_275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_137, full_275);  amax_137 = full_275 = None
	        sub_6273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_137, minimum_137);  maximum_137 = None
	        div_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6273, 255.0);  sub_6273 = None
	        clamp_min_411: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_274, 1.1920928955078125e-07);  div_274 = None
	        div_275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_137, clamp_min_411);  minimum_137 = None
	        round_275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_275);  div_275 = None
	        sub_6279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_275);  round_275 = None
	        clamp_min_412: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6279, -128);  sub_6279 = None
	        clamp_max_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_412, 127);  clamp_min_412 = None
	        _assert_tensor_metadata_1235 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_411, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1235 = None
	        _assert_tensor_metadata_1236 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_274, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1236 = None
	        convert_element_type_822: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_274, torch.int8);  clamp_max_274 = None
	        view_2148: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_183, [sym_size_int, 1500, 5120]);  clone_183 = None
	        view_2149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_411, [sym_size_int, 1500, 1])
	        view_2150: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_822, [sym_size_int, 1500, 1])
	        reciprocal_137: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2149);  view_2149 = None
	        mul_13314: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_137, 1.0);  reciprocal_137 = None
	        mul_13317: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2148, mul_13314);  view_2148 = mul_13314 = None
	        round_276: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_13317);  mul_13317 = None
	        add_21077: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_276, view_2150);  round_276 = view_2150 = None
	        clamp_min_413: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21077, -128);  add_21077 = None
	        clamp_max_275: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_413, 127);  clamp_min_413 = None
	        view_2151: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_275, [sym_size_int, 1500, 5120]);  clamp_max_275 = None
	        _assert_tensor_metadata_1237 = torch.ops.aten._assert_tensor_metadata.default(view_2151, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1237 = None
	        convert_element_type_823: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2151, torch.int8);  view_2151 = None
	        view_2152: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_823, [sym_size_int, 1500, 5120]);  convert_element_type_823 = None
	        view_2153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_411, [sym_size_int, 1500, 1]);  clamp_min_411 = None
	        view_2154: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_822, [sym_size_int, 1500, 1]);  convert_element_type_822 = None
	        _assert_tensor_metadata_1238 = torch.ops.aten._assert_tensor_metadata.default(view_2152, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1238 = None
	        convert_element_type_824: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2152, torch.float32);  view_2152 = None
	        _assert_tensor_metadata_1239 = torch.ops.aten._assert_tensor_metadata.default(view_2154, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1239 = None
	        convert_element_type_825: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2154, torch.float32);  view_2154 = None
	        sub_6299: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_824, convert_element_type_825);  convert_element_type_824 = convert_element_type_825 = None
	        mul_13339: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6299, view_2153);  sub_6299 = view_2153 = None
	        view_2155: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_13339, [sym_size_int, 1500, 5120]);  mul_13339 = None
	        _assert_tensor_metadata_1240 = torch.ops.aten._assert_tensor_metadata.default(view_2155, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1240 = None
	        view_2156: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_22_fc2_parametrizations_weight_original0 = None
	        view_2157: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_22_fc2_parametrizations_weight_original1 = None
	        view_2158: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_22_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1241 = torch.ops.aten._assert_tensor_metadata.default(view_2156, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1241 = None
	        convert_element_type_826: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2156, torch.float32);  view_2156 = None
	        _assert_tensor_metadata_1242 = torch.ops.aten._assert_tensor_metadata.default(view_2158, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1242 = None
	        convert_element_type_827: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2158, torch.float32);  view_2158 = None
	        sub_6303: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_826, convert_element_type_827);  convert_element_type_826 = convert_element_type_827 = None
	        mul_13344: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6303, view_2157);  sub_6303 = view_2157 = None
	        view_2159: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_13344, [1280, 5120]);  mul_13344 = None
	        _assert_tensor_metadata_1243 = torch.ops.aten._assert_tensor_metadata.default(view_2159, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1243 = None
	        mul_13349: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2160: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2155, [mul_13349, 5120]);  view_2155 = mul_13349 = None
	        permute_230: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2159, [1, 0]);  view_2159 = None
	        addmm_114: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_22_fc2_bias, view_2160, permute_230);  model_audio_tower_layers_22_fc2_bias = view_2160 = permute_230 = None
	        view_2161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_114, [sym_size_int, 1500, 1280]);  addmm_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_184: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2161);  view_2161 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_21140: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_20842, clone_184);  add_20842 = clone_184 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_21140, memory_format = torch.contiguous_format)
	        var_mean_46 = torch.ops.aten.var_mean.correction(clone_185, [2], correction = 0, keepdim = True)
	        getitem_184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_46[0]
	        getitem_185: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_46[1];  var_mean_46 = None
	        add_21145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_184, 1e-05);  getitem_184 = None
	        rsqrt_46: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_21145);  add_21145 = None
	        sub_6309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_185, getitem_185);  clone_185 = getitem_185 = None
	        mul_13360: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6309, rsqrt_46);  sub_6309 = rsqrt_46 = None
	        mul_13361: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13360, model_audio_tower_layers_23_self_attn_layer_norm_weight);  mul_13360 = model_audio_tower_layers_23_self_attn_layer_norm_weight = None
	        add_21146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13361, model_audio_tower_layers_23_self_attn_layer_norm_bias);  mul_13361 = model_audio_tower_layers_23_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2162: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_21146, [sym_size_int, 1500, 1280])
	        amin_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2162, [2])
	        amax_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2162, [2]);  view_2162 = None
	        full_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_138, full_276);  amin_138 = full_276 = None
	        full_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_138, full_277);  amax_138 = full_277 = None
	        sub_6320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_138, minimum_138);  maximum_138 = None
	        div_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6320, 255.0);  sub_6320 = None
	        clamp_min_414: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_276, 1.1920928955078125e-07);  div_276 = None
	        div_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_138, clamp_min_414);  minimum_138 = None
	        round_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_277);  div_277 = None
	        sub_6326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_277);  round_277 = None
	        clamp_min_415: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6326, -128);  sub_6326 = None
	        clamp_max_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_415, 127);  clamp_min_415 = None
	        _assert_tensor_metadata_1244 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_414, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1244 = None
	        _assert_tensor_metadata_1245 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_276, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1245 = None
	        convert_element_type_828: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_276, torch.int8);  clamp_max_276 = None
	        view_2163: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_21146, [sym_size_int, 1500, 1280])
	        view_2164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_414, [sym_size_int, 1500, 1])
	        view_2165: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_828, [sym_size_int, 1500, 1])
	        reciprocal_138: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2164);  view_2164 = None
	        mul_13409: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_138, 1.0);  reciprocal_138 = None
	        mul_13412: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2163, mul_13409);  view_2163 = mul_13409 = None
	        round_278: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13412);  mul_13412 = None
	        add_21233: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_278, view_2165);  round_278 = view_2165 = None
	        clamp_min_416: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21233, -128);  add_21233 = None
	        clamp_max_277: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_416, 127);  clamp_min_416 = None
	        view_2166: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_277, [sym_size_int, 1500, 1280]);  clamp_max_277 = None
	        _assert_tensor_metadata_1246 = torch.ops.aten._assert_tensor_metadata.default(view_2166, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1246 = None
	        convert_element_type_829: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2166, torch.int8);  view_2166 = None
	        view_2167: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_829, [sym_size_int, 1500, 1280]);  convert_element_type_829 = None
	        view_2168: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_414, [sym_size_int, 1500, 1]);  clamp_min_414 = None
	        view_2169: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_828, [sym_size_int, 1500, 1]);  convert_element_type_828 = None
	        _assert_tensor_metadata_1247 = torch.ops.aten._assert_tensor_metadata.default(view_2167, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1247 = None
	        convert_element_type_830: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2167, torch.float32);  view_2167 = None
	        _assert_tensor_metadata_1248 = torch.ops.aten._assert_tensor_metadata.default(view_2169, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1248 = None
	        convert_element_type_831: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2169, torch.float32);  view_2169 = None
	        sub_6346: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_830, convert_element_type_831);  convert_element_type_830 = convert_element_type_831 = None
	        mul_13434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6346, view_2168);  sub_6346 = view_2168 = None
	        view_2170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13434, [sym_size_int, 1500, 1280]);  mul_13434 = None
	        _assert_tensor_metadata_1249 = torch.ops.aten._assert_tensor_metadata.default(view_2170, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1249 = None
	        view_2171: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2172: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2173: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1250 = torch.ops.aten._assert_tensor_metadata.default(view_2171, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1250 = None
	        convert_element_type_832: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2171, torch.float32);  view_2171 = None
	        _assert_tensor_metadata_1251 = torch.ops.aten._assert_tensor_metadata.default(view_2173, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1251 = None
	        convert_element_type_833: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2173, torch.float32);  view_2173 = None
	        sub_6350: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_832, convert_element_type_833);  convert_element_type_832 = convert_element_type_833 = None
	        mul_13439: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6350, view_2172);  sub_6350 = view_2172 = None
	        view_2174: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13439, [1280, 1280]);  mul_13439 = None
	        _assert_tensor_metadata_1252 = torch.ops.aten._assert_tensor_metadata.default(view_2174, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1252 = None
	        mul_13444: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2175: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2170, [mul_13444, 1280]);  view_2170 = mul_13444 = None
	        permute_231: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2174, [1, 0]);  view_2174 = None
	        addmm_115: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_23_self_attn_q_proj_bias, view_2175, permute_231);  model_audio_tower_layers_23_self_attn_q_proj_bias = view_2175 = permute_231 = None
	        view_2176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_115, [sym_size_int, 1500, 1280]);  addmm_115 = None
	        mul_13451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2176, 0.125);  view_2176 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2177: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_13451, [sym_size_int, 1500, 20, 64]);  mul_13451 = None
	        permute_232: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2177, [0, 2, 1, 3]);  view_2177 = None
	        clone_186: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_232, memory_format = torch.contiguous_format);  permute_232 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2178: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_21146, [sym_size_int, 1500, 1280])
	        amin_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2178, [2])
	        amax_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2178, [2]);  view_2178 = None
	        full_278: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_139, full_278);  amin_139 = full_278 = None
	        full_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_139, full_279);  amax_139 = full_279 = None
	        sub_6365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_139, minimum_139);  maximum_139 = None
	        div_278: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6365, 255.0);  sub_6365 = None
	        clamp_min_417: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_278, 1.1920928955078125e-07);  div_278 = None
	        div_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_139, clamp_min_417);  minimum_139 = None
	        round_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_279);  div_279 = None
	        sub_6371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_279);  round_279 = None
	        clamp_min_418: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6371, -128);  sub_6371 = None
	        clamp_max_278: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_418, 127);  clamp_min_418 = None
	        _assert_tensor_metadata_1253 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_417, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1253 = None
	        _assert_tensor_metadata_1254 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_278, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1254 = None
	        convert_element_type_834: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_278, torch.int8);  clamp_max_278 = None
	        view_2179: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_21146, [sym_size_int, 1500, 1280])
	        view_2180: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_417, [sym_size_int, 1500, 1])
	        view_2181: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_834, [sym_size_int, 1500, 1])
	        reciprocal_139: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2180);  view_2180 = None
	        mul_13505: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_139, 1.0);  reciprocal_139 = None
	        mul_13508: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2179, mul_13505);  view_2179 = mul_13505 = None
	        round_280: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13508);  mul_13508 = None
	        add_21385: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_280, view_2181);  round_280 = view_2181 = None
	        clamp_min_419: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21385, -128);  add_21385 = None
	        clamp_max_279: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_419, 127);  clamp_min_419 = None
	        view_2182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_279, [sym_size_int, 1500, 1280]);  clamp_max_279 = None
	        _assert_tensor_metadata_1255 = torch.ops.aten._assert_tensor_metadata.default(view_2182, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1255 = None
	        convert_element_type_835: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2182, torch.int8);  view_2182 = None
	        view_2183: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_835, [sym_size_int, 1500, 1280]);  convert_element_type_835 = None
	        view_2184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_417, [sym_size_int, 1500, 1]);  clamp_min_417 = None
	        view_2185: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_834, [sym_size_int, 1500, 1]);  convert_element_type_834 = None
	        _assert_tensor_metadata_1256 = torch.ops.aten._assert_tensor_metadata.default(view_2183, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1256 = None
	        convert_element_type_836: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2183, torch.float32);  view_2183 = None
	        _assert_tensor_metadata_1257 = torch.ops.aten._assert_tensor_metadata.default(view_2185, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1257 = None
	        convert_element_type_837: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2185, torch.float32);  view_2185 = None
	        sub_6391: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_836, convert_element_type_837);  convert_element_type_836 = convert_element_type_837 = None
	        mul_13530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6391, view_2184);  sub_6391 = view_2184 = None
	        view_2186: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13530, [sym_size_int, 1500, 1280]);  mul_13530 = None
	        _assert_tensor_metadata_1258 = torch.ops.aten._assert_tensor_metadata.default(view_2186, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1258 = None
	        view_2187: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2188: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2189: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1259 = torch.ops.aten._assert_tensor_metadata.default(view_2187, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1259 = None
	        convert_element_type_838: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2187, torch.float32);  view_2187 = None
	        _assert_tensor_metadata_1260 = torch.ops.aten._assert_tensor_metadata.default(view_2189, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1260 = None
	        convert_element_type_839: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2189, torch.float32);  view_2189 = None
	        sub_6395: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_838, convert_element_type_839);  convert_element_type_838 = convert_element_type_839 = None
	        mul_13535: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6395, view_2188);  sub_6395 = view_2188 = None
	        view_2190: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13535, [1280, 1280]);  mul_13535 = None
	        _assert_tensor_metadata_1261 = torch.ops.aten._assert_tensor_metadata.default(view_2190, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1261 = None
	        permute_233: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2190, [1, 0]);  view_2190 = None
	        mul_13538: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2191: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2186, [mul_13538, 1280]);  view_2186 = mul_13538 = None
	        mm_23: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2191, permute_233);  view_2191 = permute_233 = None
	        view_2192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_23, [sym_size_int, 1500, 1280]);  mm_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2193: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2192, [sym_size_int, -1, 20, 64]);  view_2192 = None
	        permute_234: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2193, [0, 2, 1, 3]);  view_2193 = None
	        clone_187: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_234, memory_format = torch.contiguous_format);  permute_234 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_21146, [sym_size_int, 1500, 1280])
	        amin_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2194, [2])
	        amax_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2194, [2]);  view_2194 = None
	        full_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_140, full_280);  amin_140 = full_280 = None
	        full_281: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_140, full_281);  amax_140 = full_281 = None
	        sub_6409: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_140, minimum_140);  maximum_140 = None
	        div_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6409, 255.0);  sub_6409 = None
	        clamp_min_420: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_280, 1.1920928955078125e-07);  div_280 = None
	        div_281: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_140, clamp_min_420);  minimum_140 = None
	        round_281: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_281);  div_281 = None
	        sub_6415: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_281);  round_281 = None
	        clamp_min_421: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6415, -128);  sub_6415 = None
	        clamp_max_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_421, 127);  clamp_min_421 = None
	        _assert_tensor_metadata_1262 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_420, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1262 = None
	        _assert_tensor_metadata_1263 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_280, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1263 = None
	        convert_element_type_840: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_280, torch.int8);  clamp_max_280 = None
	        view_2195: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_21146, [sym_size_int, 1500, 1280]);  add_21146 = None
	        view_2196: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_420, [sym_size_int, 1500, 1])
	        view_2197: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_840, [sym_size_int, 1500, 1])
	        reciprocal_140: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2196);  view_2196 = None
	        mul_13604: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_140, 1.0);  reciprocal_140 = None
	        mul_13607: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2195, mul_13604);  view_2195 = mul_13604 = None
	        round_282: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13607);  mul_13607 = None
	        add_21533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_282, view_2197);  round_282 = view_2197 = None
	        clamp_min_422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21533, -128);  add_21533 = None
	        clamp_max_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_422, 127);  clamp_min_422 = None
	        view_2198: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_281, [sym_size_int, 1500, 1280]);  clamp_max_281 = None
	        _assert_tensor_metadata_1264 = torch.ops.aten._assert_tensor_metadata.default(view_2198, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1264 = None
	        convert_element_type_841: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2198, torch.int8);  view_2198 = None
	        view_2199: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_841, [sym_size_int, 1500, 1280]);  convert_element_type_841 = None
	        view_2200: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_420, [sym_size_int, 1500, 1]);  clamp_min_420 = None
	        view_2201: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_840, [sym_size_int, 1500, 1]);  convert_element_type_840 = None
	        _assert_tensor_metadata_1265 = torch.ops.aten._assert_tensor_metadata.default(view_2199, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1265 = None
	        convert_element_type_842: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2199, torch.float32);  view_2199 = None
	        _assert_tensor_metadata_1266 = torch.ops.aten._assert_tensor_metadata.default(view_2201, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1266 = None
	        convert_element_type_843: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2201, torch.float32);  view_2201 = None
	        sub_6435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_842, convert_element_type_843);  convert_element_type_842 = convert_element_type_843 = None
	        mul_13629: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6435, view_2200);  sub_6435 = view_2200 = None
	        view_2202: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13629, [sym_size_int, 1500, 1280]);  mul_13629 = None
	        _assert_tensor_metadata_1267 = torch.ops.aten._assert_tensor_metadata.default(view_2202, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1267 = None
	        view_2203: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2204: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2205: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1268 = torch.ops.aten._assert_tensor_metadata.default(view_2203, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1268 = None
	        convert_element_type_844: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2203, torch.float32);  view_2203 = None
	        _assert_tensor_metadata_1269 = torch.ops.aten._assert_tensor_metadata.default(view_2205, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1269 = None
	        convert_element_type_845: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2205, torch.float32);  view_2205 = None
	        sub_6439: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_844, convert_element_type_845);  convert_element_type_844 = convert_element_type_845 = None
	        mul_13634: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6439, view_2204);  sub_6439 = view_2204 = None
	        view_2206: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13634, [1280, 1280]);  mul_13634 = None
	        _assert_tensor_metadata_1270 = torch.ops.aten._assert_tensor_metadata.default(view_2206, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1270 = None
	        mul_13639: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2207: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2202, [mul_13639, 1280]);  view_2202 = mul_13639 = None
	        permute_235: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2206, [1, 0]);  view_2206 = None
	        addmm_116: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_23_self_attn_v_proj_bias, view_2207, permute_235);  model_audio_tower_layers_23_self_attn_v_proj_bias = view_2207 = permute_235 = None
	        view_2208: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_116, [sym_size_int, 1500, 1280]);  addmm_116 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2209: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2208, [sym_size_int, -1, 20, 64]);  view_2208 = None
	        permute_236: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2209, [0, 2, 1, 3]);  view_2209 = None
	        clone_188: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_236, memory_format = torch.contiguous_format);  permute_236 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_23 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_186, clone_187, clone_188, None, False, scale = 1.0);  clone_186 = clone_187 = clone_188 = None
	        getitem_186: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_23[0];  _scaled_dot_product_efficient_attention_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_237: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_186, [0, 2, 1, 3]);  getitem_186 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2210: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_237, [sym_size_int, 1500, -1]);  permute_237 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2211: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2210, [sym_size_int, 1500, 1280])
	        amin_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2211, [2])
	        amax_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2211, [2]);  view_2211 = None
	        full_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_141, full_282);  amin_141 = full_282 = None
	        full_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_141, full_283);  amax_141 = full_283 = None
	        sub_6457: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_141, minimum_141);  maximum_141 = None
	        div_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6457, 255.0);  sub_6457 = None
	        clamp_min_423: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_282, 1.1920928955078125e-07);  div_282 = None
	        div_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_141, clamp_min_423);  minimum_141 = None
	        round_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_283);  div_283 = None
	        sub_6463: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_283);  round_283 = None
	        clamp_min_424: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6463, -128);  sub_6463 = None
	        clamp_max_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_424, 127);  clamp_min_424 = None
	        _assert_tensor_metadata_1271 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_423, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1271 = None
	        _assert_tensor_metadata_1272 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_282, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1272 = None
	        convert_element_type_846: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_282, torch.int8);  clamp_max_282 = None
	        view_2212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2210, [sym_size_int, 1500, 1280]);  view_2210 = None
	        view_2213: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_423, [sym_size_int, 1500, 1])
	        view_2214: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_846, [sym_size_int, 1500, 1])
	        reciprocal_141: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2213);  view_2213 = None
	        mul_13709: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_141, 1.0);  reciprocal_141 = None
	        mul_13712: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2212, mul_13709);  view_2212 = mul_13709 = None
	        round_284: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13712);  mul_13712 = None
	        add_21697: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_284, view_2214);  round_284 = view_2214 = None
	        clamp_min_425: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21697, -128);  add_21697 = None
	        clamp_max_283: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_425, 127);  clamp_min_425 = None
	        view_2215: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_283, [sym_size_int, 1500, 1280]);  clamp_max_283 = None
	        _assert_tensor_metadata_1273 = torch.ops.aten._assert_tensor_metadata.default(view_2215, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1273 = None
	        convert_element_type_847: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2215, torch.int8);  view_2215 = None
	        view_2216: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_847, [sym_size_int, 1500, 1280]);  convert_element_type_847 = None
	        view_2217: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_423, [sym_size_int, 1500, 1]);  clamp_min_423 = None
	        view_2218: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_846, [sym_size_int, 1500, 1]);  convert_element_type_846 = None
	        _assert_tensor_metadata_1274 = torch.ops.aten._assert_tensor_metadata.default(view_2216, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1274 = None
	        convert_element_type_848: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2216, torch.float32);  view_2216 = None
	        _assert_tensor_metadata_1275 = torch.ops.aten._assert_tensor_metadata.default(view_2218, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1275 = None
	        convert_element_type_849: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2218, torch.float32);  view_2218 = None
	        sub_6483: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_848, convert_element_type_849);  convert_element_type_848 = convert_element_type_849 = None
	        mul_13734: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6483, view_2217);  sub_6483 = view_2217 = None
	        view_2219: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13734, [sym_size_int, 1500, 1280]);  mul_13734 = None
	        _assert_tensor_metadata_1276 = torch.ops.aten._assert_tensor_metadata.default(view_2219, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1276 = None
	        view_2220: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2221: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2222: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1277 = torch.ops.aten._assert_tensor_metadata.default(view_2220, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1277 = None
	        convert_element_type_850: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2220, torch.float32);  view_2220 = None
	        _assert_tensor_metadata_1278 = torch.ops.aten._assert_tensor_metadata.default(view_2222, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1278 = None
	        convert_element_type_851: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2222, torch.float32);  view_2222 = None
	        sub_6487: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_850, convert_element_type_851);  convert_element_type_850 = convert_element_type_851 = None
	        mul_13739: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6487, view_2221);  sub_6487 = view_2221 = None
	        view_2223: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13739, [1280, 1280]);  mul_13739 = None
	        _assert_tensor_metadata_1279 = torch.ops.aten._assert_tensor_metadata.default(view_2223, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1279 = None
	        mul_13744: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2224: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2219, [mul_13744, 1280]);  view_2219 = mul_13744 = None
	        permute_238: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2223, [1, 0]);  view_2223 = None
	        addmm_117: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_23_self_attn_out_proj_bias, view_2224, permute_238);  model_audio_tower_layers_23_self_attn_out_proj_bias = view_2224 = permute_238 = None
	        view_2225: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_117, [sym_size_int, 1500, 1280]);  addmm_117 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2225);  view_2225 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_21760: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_21140, clone_189);  add_21140 = clone_189 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_190: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_21760, memory_format = torch.contiguous_format)
	        var_mean_47 = torch.ops.aten.var_mean.correction(clone_190, [2], correction = 0, keepdim = True)
	        getitem_190: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_47[0]
	        getitem_191: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_47[1];  var_mean_47 = None
	        add_21765: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_190, 1e-05);  getitem_190 = None
	        rsqrt_47: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_21765);  add_21765 = None
	        sub_6493: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_190, getitem_191);  clone_190 = getitem_191 = None
	        mul_13755: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6493, rsqrt_47);  sub_6493 = rsqrt_47 = None
	        mul_13756: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13755, model_audio_tower_layers_23_final_layer_norm_weight);  mul_13755 = model_audio_tower_layers_23_final_layer_norm_weight = None
	        add_21766: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13756, model_audio_tower_layers_23_final_layer_norm_bias);  mul_13756 = model_audio_tower_layers_23_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2226: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_21766, [sym_size_int, 1500, 1280])
	        amin_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2226, [2])
	        amax_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2226, [2]);  view_2226 = None
	        full_284: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_142, full_284);  amin_142 = full_284 = None
	        full_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_142, full_285);  amax_142 = full_285 = None
	        sub_6504: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_142, minimum_142);  maximum_142 = None
	        div_284: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6504, 255.0);  sub_6504 = None
	        clamp_min_426: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_284, 1.1920928955078125e-07);  div_284 = None
	        div_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_142, clamp_min_426);  minimum_142 = None
	        round_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_285);  div_285 = None
	        sub_6510: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_285);  round_285 = None
	        clamp_min_427: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6510, -128);  sub_6510 = None
	        clamp_max_284: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_427, 127);  clamp_min_427 = None
	        _assert_tensor_metadata_1280 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_426, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1280 = None
	        _assert_tensor_metadata_1281 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_284, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1281 = None
	        convert_element_type_852: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_284, torch.int8);  clamp_max_284 = None
	        view_2227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_21766, [sym_size_int, 1500, 1280]);  add_21766 = None
	        view_2228: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_426, [sym_size_int, 1500, 1])
	        view_2229: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_852, [sym_size_int, 1500, 1])
	        reciprocal_142: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2228);  view_2228 = None
	        mul_13804: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_142, 1.0);  reciprocal_142 = None
	        mul_13807: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2227, mul_13804);  view_2227 = mul_13804 = None
	        round_286: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13807);  mul_13807 = None
	        add_21853: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_286, view_2229);  round_286 = view_2229 = None
	        clamp_min_428: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21853, -128);  add_21853 = None
	        clamp_max_285: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_428, 127);  clamp_min_428 = None
	        view_2230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_285, [sym_size_int, 1500, 1280]);  clamp_max_285 = None
	        _assert_tensor_metadata_1282 = torch.ops.aten._assert_tensor_metadata.default(view_2230, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1282 = None
	        convert_element_type_853: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2230, torch.int8);  view_2230 = None
	        view_2231: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_853, [sym_size_int, 1500, 1280]);  convert_element_type_853 = None
	        view_2232: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_426, [sym_size_int, 1500, 1]);  clamp_min_426 = None
	        view_2233: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_852, [sym_size_int, 1500, 1]);  convert_element_type_852 = None
	        _assert_tensor_metadata_1283 = torch.ops.aten._assert_tensor_metadata.default(view_2231, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1283 = None
	        convert_element_type_854: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2231, torch.float32);  view_2231 = None
	        _assert_tensor_metadata_1284 = torch.ops.aten._assert_tensor_metadata.default(view_2233, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1284 = None
	        convert_element_type_855: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2233, torch.float32);  view_2233 = None
	        sub_6530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_854, convert_element_type_855);  convert_element_type_854 = convert_element_type_855 = None
	        mul_13829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6530, view_2232);  sub_6530 = view_2232 = None
	        view_2234: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13829, [sym_size_int, 1500, 1280]);  mul_13829 = None
	        _assert_tensor_metadata_1285 = torch.ops.aten._assert_tensor_metadata.default(view_2234, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1285 = None
	        view_2235: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_23_fc1_parametrizations_weight_original0 = None
	        view_2236: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_23_fc1_parametrizations_weight_original1 = None
	        view_2237: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_23_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1286 = torch.ops.aten._assert_tensor_metadata.default(view_2235, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1286 = None
	        convert_element_type_856: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2235, torch.float32);  view_2235 = None
	        _assert_tensor_metadata_1287 = torch.ops.aten._assert_tensor_metadata.default(view_2237, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1287 = None
	        convert_element_type_857: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2237, torch.float32);  view_2237 = None
	        sub_6534: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_856, convert_element_type_857);  convert_element_type_856 = convert_element_type_857 = None
	        mul_13834: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6534, view_2236);  sub_6534 = view_2236 = None
	        view_2238: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13834, [5120, 1280]);  mul_13834 = None
	        _assert_tensor_metadata_1288 = torch.ops.aten._assert_tensor_metadata.default(view_2238, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1288 = None
	        mul_13839: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2239: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2234, [mul_13839, 1280]);  view_2234 = mul_13839 = None
	        permute_239: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2238, [1, 0]);  view_2238 = None
	        addmm_118: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_23_fc1_bias, view_2239, permute_239);  model_audio_tower_layers_23_fc1_bias = view_2239 = permute_239 = None
	        view_2240: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_118, [sym_size_int, 1500, 5120]);  addmm_118 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_13846: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2240, 0.5)
	        mul_13847: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2240, 0.7071067811865476);  view_2240 = None
	        erf_25: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_13847);  mul_13847 = None
	        add_21912: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_25, 1);  erf_25 = None
	        mul_13848: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13846, add_21912);  mul_13846 = add_21912 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_191: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_13848);  mul_13848 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2241: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_191, [sym_size_int, 1500, 5120])
	        amin_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2241, [2])
	        amax_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2241, [2]);  view_2241 = None
	        full_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_143, full_286);  amin_143 = full_286 = None
	        full_287: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_143, full_287);  amax_143 = full_287 = None
	        sub_6547: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_143, minimum_143);  maximum_143 = None
	        div_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6547, 255.0);  sub_6547 = None
	        clamp_min_429: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_286, 1.1920928955078125e-07);  div_286 = None
	        div_287: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_143, clamp_min_429);  minimum_143 = None
	        round_287: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_287);  div_287 = None
	        sub_6553: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_287);  round_287 = None
	        clamp_min_430: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6553, -128);  sub_6553 = None
	        clamp_max_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_430, 127);  clamp_min_430 = None
	        _assert_tensor_metadata_1289 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_429, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1289 = None
	        _assert_tensor_metadata_1290 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_286, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1290 = None
	        convert_element_type_858: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_286, torch.int8);  clamp_max_286 = None
	        view_2242: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_191, [sym_size_int, 1500, 5120]);  clone_191 = None
	        view_2243: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_429, [sym_size_int, 1500, 1])
	        view_2244: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_858, [sym_size_int, 1500, 1])
	        reciprocal_143: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2243);  view_2243 = None
	        mul_13894: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_143, 1.0);  reciprocal_143 = None
	        mul_13897: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2242, mul_13894);  view_2242 = mul_13894 = None
	        round_288: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_13897);  mul_13897 = None
	        add_21995: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_288, view_2244);  round_288 = view_2244 = None
	        clamp_min_431: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21995, -128);  add_21995 = None
	        clamp_max_287: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_431, 127);  clamp_min_431 = None
	        view_2245: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_287, [sym_size_int, 1500, 5120]);  clamp_max_287 = None
	        _assert_tensor_metadata_1291 = torch.ops.aten._assert_tensor_metadata.default(view_2245, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1291 = None
	        convert_element_type_859: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2245, torch.int8);  view_2245 = None
	        view_2246: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_859, [sym_size_int, 1500, 5120]);  convert_element_type_859 = None
	        view_2247: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_429, [sym_size_int, 1500, 1]);  clamp_min_429 = None
	        view_2248: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_858, [sym_size_int, 1500, 1]);  convert_element_type_858 = None
	        _assert_tensor_metadata_1292 = torch.ops.aten._assert_tensor_metadata.default(view_2246, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1292 = None
	        convert_element_type_860: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2246, torch.float32);  view_2246 = None
	        _assert_tensor_metadata_1293 = torch.ops.aten._assert_tensor_metadata.default(view_2248, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1293 = None
	        convert_element_type_861: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2248, torch.float32);  view_2248 = None
	        sub_6573: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_860, convert_element_type_861);  convert_element_type_860 = convert_element_type_861 = None
	        mul_13919: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6573, view_2247);  sub_6573 = view_2247 = None
	        view_2249: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_13919, [sym_size_int, 1500, 5120]);  mul_13919 = None
	        _assert_tensor_metadata_1294 = torch.ops.aten._assert_tensor_metadata.default(view_2249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1294 = None
	        view_2250: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_23_fc2_parametrizations_weight_original0 = None
	        view_2251: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_23_fc2_parametrizations_weight_original1 = None
	        view_2252: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_23_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1295 = torch.ops.aten._assert_tensor_metadata.default(view_2250, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1295 = None
	        convert_element_type_862: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2250, torch.float32);  view_2250 = None
	        _assert_tensor_metadata_1296 = torch.ops.aten._assert_tensor_metadata.default(view_2252, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1296 = None
	        convert_element_type_863: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2252, torch.float32);  view_2252 = None
	        sub_6577: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_862, convert_element_type_863);  convert_element_type_862 = convert_element_type_863 = None
	        mul_13924: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6577, view_2251);  sub_6577 = view_2251 = None
	        view_2253: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_13924, [1280, 5120]);  mul_13924 = None
	        _assert_tensor_metadata_1297 = torch.ops.aten._assert_tensor_metadata.default(view_2253, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1297 = None
	        mul_13929: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2254: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2249, [mul_13929, 5120]);  view_2249 = mul_13929 = None
	        permute_240: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2253, [1, 0]);  view_2253 = None
	        addmm_119: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_23_fc2_bias, view_2254, permute_240);  model_audio_tower_layers_23_fc2_bias = view_2254 = permute_240 = None
	        view_2255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_119, [sym_size_int, 1500, 1280]);  addmm_119 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2255);  view_2255 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_22058: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_21760, clone_192);  add_21760 = clone_192 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_193: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_22058, memory_format = torch.contiguous_format)
	        var_mean_48 = torch.ops.aten.var_mean.correction(clone_193, [2], correction = 0, keepdim = True)
	        getitem_192: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_48[0]
	        getitem_193: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_48[1];  var_mean_48 = None
	        add_22063: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_192, 1e-05);  getitem_192 = None
	        rsqrt_48: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_22063);  add_22063 = None
	        sub_6583: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_193, getitem_193);  clone_193 = getitem_193 = None
	        mul_13940: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6583, rsqrt_48);  sub_6583 = rsqrt_48 = None
	        mul_13941: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13940, model_audio_tower_layers_24_self_attn_layer_norm_weight);  mul_13940 = model_audio_tower_layers_24_self_attn_layer_norm_weight = None
	        add_22064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13941, model_audio_tower_layers_24_self_attn_layer_norm_bias);  mul_13941 = model_audio_tower_layers_24_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2256: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22064, [sym_size_int, 1500, 1280])
	        amin_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2256, [2])
	        amax_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2256, [2]);  view_2256 = None
	        full_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_144, full_288);  amin_144 = full_288 = None
	        full_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_144, full_289);  amax_144 = full_289 = None
	        sub_6594: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_144, minimum_144);  maximum_144 = None
	        div_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6594, 255.0);  sub_6594 = None
	        clamp_min_432: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_288, 1.1920928955078125e-07);  div_288 = None
	        div_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_144, clamp_min_432);  minimum_144 = None
	        round_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_289);  div_289 = None
	        sub_6600: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_289);  round_289 = None
	        clamp_min_433: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6600, -128);  sub_6600 = None
	        clamp_max_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_433, 127);  clamp_min_433 = None
	        _assert_tensor_metadata_1298 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_432, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1298 = None
	        _assert_tensor_metadata_1299 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_288, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1299 = None
	        convert_element_type_864: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_288, torch.int8);  clamp_max_288 = None
	        view_2257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22064, [sym_size_int, 1500, 1280])
	        view_2258: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_432, [sym_size_int, 1500, 1])
	        view_2259: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_864, [sym_size_int, 1500, 1])
	        reciprocal_144: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2258);  view_2258 = None
	        mul_13989: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_144, 1.0);  reciprocal_144 = None
	        mul_13992: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2257, mul_13989);  view_2257 = mul_13989 = None
	        round_290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13992);  mul_13992 = None
	        add_22151: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_290, view_2259);  round_290 = view_2259 = None
	        clamp_min_434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22151, -128);  add_22151 = None
	        clamp_max_289: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_434, 127);  clamp_min_434 = None
	        view_2260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_289, [sym_size_int, 1500, 1280]);  clamp_max_289 = None
	        _assert_tensor_metadata_1300 = torch.ops.aten._assert_tensor_metadata.default(view_2260, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1300 = None
	        convert_element_type_865: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2260, torch.int8);  view_2260 = None
	        view_2261: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_865, [sym_size_int, 1500, 1280]);  convert_element_type_865 = None
	        view_2262: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_432, [sym_size_int, 1500, 1]);  clamp_min_432 = None
	        view_2263: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_864, [sym_size_int, 1500, 1]);  convert_element_type_864 = None
	        _assert_tensor_metadata_1301 = torch.ops.aten._assert_tensor_metadata.default(view_2261, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1301 = None
	        convert_element_type_866: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2261, torch.float32);  view_2261 = None
	        _assert_tensor_metadata_1302 = torch.ops.aten._assert_tensor_metadata.default(view_2263, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1302 = None
	        convert_element_type_867: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2263, torch.float32);  view_2263 = None
	        sub_6620: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_866, convert_element_type_867);  convert_element_type_866 = convert_element_type_867 = None
	        mul_14014: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6620, view_2262);  sub_6620 = view_2262 = None
	        view_2264: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14014, [sym_size_int, 1500, 1280]);  mul_14014 = None
	        _assert_tensor_metadata_1303 = torch.ops.aten._assert_tensor_metadata.default(view_2264, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1303 = None
	        view_2265: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2266: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2267: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1304 = torch.ops.aten._assert_tensor_metadata.default(view_2265, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1304 = None
	        convert_element_type_868: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2265, torch.float32);  view_2265 = None
	        _assert_tensor_metadata_1305 = torch.ops.aten._assert_tensor_metadata.default(view_2267, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1305 = None
	        convert_element_type_869: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2267, torch.float32);  view_2267 = None
	        sub_6624: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_868, convert_element_type_869);  convert_element_type_868 = convert_element_type_869 = None
	        mul_14019: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6624, view_2266);  sub_6624 = view_2266 = None
	        view_2268: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14019, [1280, 1280]);  mul_14019 = None
	        _assert_tensor_metadata_1306 = torch.ops.aten._assert_tensor_metadata.default(view_2268, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1306 = None
	        mul_14024: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2269: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2264, [mul_14024, 1280]);  view_2264 = mul_14024 = None
	        permute_241: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2268, [1, 0]);  view_2268 = None
	        addmm_120: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_24_self_attn_q_proj_bias, view_2269, permute_241);  model_audio_tower_layers_24_self_attn_q_proj_bias = view_2269 = permute_241 = None
	        view_2270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_120, [sym_size_int, 1500, 1280]);  addmm_120 = None
	        mul_14031: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2270, 0.125);  view_2270 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2271: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_14031, [sym_size_int, 1500, 20, 64]);  mul_14031 = None
	        permute_242: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2271, [0, 2, 1, 3]);  view_2271 = None
	        clone_194: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_242, memory_format = torch.contiguous_format);  permute_242 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22064, [sym_size_int, 1500, 1280])
	        amin_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2272, [2])
	        amax_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2272, [2]);  view_2272 = None
	        full_290: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_145, full_290);  amin_145 = full_290 = None
	        full_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_145, full_291);  amax_145 = full_291 = None
	        sub_6639: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_145, minimum_145);  maximum_145 = None
	        div_290: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6639, 255.0);  sub_6639 = None
	        clamp_min_435: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_290, 1.1920928955078125e-07);  div_290 = None
	        div_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_145, clamp_min_435);  minimum_145 = None
	        round_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_291);  div_291 = None
	        sub_6645: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_291);  round_291 = None
	        clamp_min_436: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6645, -128);  sub_6645 = None
	        clamp_max_290: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_436, 127);  clamp_min_436 = None
	        _assert_tensor_metadata_1307 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_435, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1307 = None
	        _assert_tensor_metadata_1308 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_290, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1308 = None
	        convert_element_type_870: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_290, torch.int8);  clamp_max_290 = None
	        view_2273: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22064, [sym_size_int, 1500, 1280])
	        view_2274: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_435, [sym_size_int, 1500, 1])
	        view_2275: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_870, [sym_size_int, 1500, 1])
	        reciprocal_145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2274);  view_2274 = None
	        mul_14085: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_145, 1.0);  reciprocal_145 = None
	        mul_14088: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2273, mul_14085);  view_2273 = mul_14085 = None
	        round_292: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14088);  mul_14088 = None
	        add_22303: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_292, view_2275);  round_292 = view_2275 = None
	        clamp_min_437: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22303, -128);  add_22303 = None
	        clamp_max_291: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_437, 127);  clamp_min_437 = None
	        view_2276: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_291, [sym_size_int, 1500, 1280]);  clamp_max_291 = None
	        _assert_tensor_metadata_1309 = torch.ops.aten._assert_tensor_metadata.default(view_2276, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1309 = None
	        convert_element_type_871: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2276, torch.int8);  view_2276 = None
	        view_2277: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_871, [sym_size_int, 1500, 1280]);  convert_element_type_871 = None
	        view_2278: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_435, [sym_size_int, 1500, 1]);  clamp_min_435 = None
	        view_2279: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_870, [sym_size_int, 1500, 1]);  convert_element_type_870 = None
	        _assert_tensor_metadata_1310 = torch.ops.aten._assert_tensor_metadata.default(view_2277, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1310 = None
	        convert_element_type_872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2277, torch.float32);  view_2277 = None
	        _assert_tensor_metadata_1311 = torch.ops.aten._assert_tensor_metadata.default(view_2279, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1311 = None
	        convert_element_type_873: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2279, torch.float32);  view_2279 = None
	        sub_6665: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_872, convert_element_type_873);  convert_element_type_872 = convert_element_type_873 = None
	        mul_14110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6665, view_2278);  sub_6665 = view_2278 = None
	        view_2280: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14110, [sym_size_int, 1500, 1280]);  mul_14110 = None
	        _assert_tensor_metadata_1312 = torch.ops.aten._assert_tensor_metadata.default(view_2280, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1312 = None
	        view_2281: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2282: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2283: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1313 = torch.ops.aten._assert_tensor_metadata.default(view_2281, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1313 = None
	        convert_element_type_874: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2281, torch.float32);  view_2281 = None
	        _assert_tensor_metadata_1314 = torch.ops.aten._assert_tensor_metadata.default(view_2283, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1314 = None
	        convert_element_type_875: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2283, torch.float32);  view_2283 = None
	        sub_6669: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_874, convert_element_type_875);  convert_element_type_874 = convert_element_type_875 = None
	        mul_14115: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6669, view_2282);  sub_6669 = view_2282 = None
	        view_2284: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14115, [1280, 1280]);  mul_14115 = None
	        _assert_tensor_metadata_1315 = torch.ops.aten._assert_tensor_metadata.default(view_2284, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1315 = None
	        permute_243: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2284, [1, 0]);  view_2284 = None
	        mul_14118: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2285: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2280, [mul_14118, 1280]);  view_2280 = mul_14118 = None
	        mm_24: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2285, permute_243);  view_2285 = permute_243 = None
	        view_2286: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_24, [sym_size_int, 1500, 1280]);  mm_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2287: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2286, [sym_size_int, -1, 20, 64]);  view_2286 = None
	        permute_244: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2287, [0, 2, 1, 3]);  view_2287 = None
	        clone_195: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_244, memory_format = torch.contiguous_format);  permute_244 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2288: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22064, [sym_size_int, 1500, 1280])
	        amin_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2288, [2])
	        amax_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2288, [2]);  view_2288 = None
	        full_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_146, full_292);  amin_146 = full_292 = None
	        full_293: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_146, full_293);  amax_146 = full_293 = None
	        sub_6683: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_146, minimum_146);  maximum_146 = None
	        div_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6683, 255.0);  sub_6683 = None
	        clamp_min_438: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_292, 1.1920928955078125e-07);  div_292 = None
	        div_293: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_146, clamp_min_438);  minimum_146 = None
	        round_293: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_293);  div_293 = None
	        sub_6689: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_293);  round_293 = None
	        clamp_min_439: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6689, -128);  sub_6689 = None
	        clamp_max_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_439, 127);  clamp_min_439 = None
	        _assert_tensor_metadata_1316 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_438, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1316 = None
	        _assert_tensor_metadata_1317 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_292, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1317 = None
	        convert_element_type_876: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_292, torch.int8);  clamp_max_292 = None
	        view_2289: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22064, [sym_size_int, 1500, 1280]);  add_22064 = None
	        view_2290: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_438, [sym_size_int, 1500, 1])
	        view_2291: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_876, [sym_size_int, 1500, 1])
	        reciprocal_146: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2290);  view_2290 = None
	        mul_14184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_146, 1.0);  reciprocal_146 = None
	        mul_14187: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2289, mul_14184);  view_2289 = mul_14184 = None
	        round_294: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14187);  mul_14187 = None
	        add_22451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_294, view_2291);  round_294 = view_2291 = None
	        clamp_min_440: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22451, -128);  add_22451 = None
	        clamp_max_293: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_440, 127);  clamp_min_440 = None
	        view_2292: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_293, [sym_size_int, 1500, 1280]);  clamp_max_293 = None
	        _assert_tensor_metadata_1318 = torch.ops.aten._assert_tensor_metadata.default(view_2292, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1318 = None
	        convert_element_type_877: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2292, torch.int8);  view_2292 = None
	        view_2293: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_877, [sym_size_int, 1500, 1280]);  convert_element_type_877 = None
	        view_2294: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_438, [sym_size_int, 1500, 1]);  clamp_min_438 = None
	        view_2295: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_876, [sym_size_int, 1500, 1]);  convert_element_type_876 = None
	        _assert_tensor_metadata_1319 = torch.ops.aten._assert_tensor_metadata.default(view_2293, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1319 = None
	        convert_element_type_878: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2293, torch.float32);  view_2293 = None
	        _assert_tensor_metadata_1320 = torch.ops.aten._assert_tensor_metadata.default(view_2295, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1320 = None
	        convert_element_type_879: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2295, torch.float32);  view_2295 = None
	        sub_6709: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_878, convert_element_type_879);  convert_element_type_878 = convert_element_type_879 = None
	        mul_14209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6709, view_2294);  sub_6709 = view_2294 = None
	        view_2296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14209, [sym_size_int, 1500, 1280]);  mul_14209 = None
	        _assert_tensor_metadata_1321 = torch.ops.aten._assert_tensor_metadata.default(view_2296, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1321 = None
	        view_2297: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2298: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2299: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1322 = torch.ops.aten._assert_tensor_metadata.default(view_2297, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1322 = None
	        convert_element_type_880: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2297, torch.float32);  view_2297 = None
	        _assert_tensor_metadata_1323 = torch.ops.aten._assert_tensor_metadata.default(view_2299, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1323 = None
	        convert_element_type_881: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2299, torch.float32);  view_2299 = None
	        sub_6713: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_880, convert_element_type_881);  convert_element_type_880 = convert_element_type_881 = None
	        mul_14214: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6713, view_2298);  sub_6713 = view_2298 = None
	        view_2300: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14214, [1280, 1280]);  mul_14214 = None
	        _assert_tensor_metadata_1324 = torch.ops.aten._assert_tensor_metadata.default(view_2300, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1324 = None
	        mul_14219: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2301: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2296, [mul_14219, 1280]);  view_2296 = mul_14219 = None
	        permute_245: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2300, [1, 0]);  view_2300 = None
	        addmm_121: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_24_self_attn_v_proj_bias, view_2301, permute_245);  model_audio_tower_layers_24_self_attn_v_proj_bias = view_2301 = permute_245 = None
	        view_2302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_121, [sym_size_int, 1500, 1280]);  addmm_121 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2303: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2302, [sym_size_int, -1, 20, 64]);  view_2302 = None
	        permute_246: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2303, [0, 2, 1, 3]);  view_2303 = None
	        clone_196: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_246, memory_format = torch.contiguous_format);  permute_246 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_24 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_194, clone_195, clone_196, None, False, scale = 1.0);  clone_194 = clone_195 = clone_196 = None
	        getitem_194: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_24[0];  _scaled_dot_product_efficient_attention_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_247: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_194, [0, 2, 1, 3]);  getitem_194 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2304: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_247, [sym_size_int, 1500, -1]);  permute_247 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2305: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2304, [sym_size_int, 1500, 1280])
	        amin_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2305, [2])
	        amax_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2305, [2]);  view_2305 = None
	        full_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_147, full_294);  amin_147 = full_294 = None
	        full_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_147, full_295);  amax_147 = full_295 = None
	        sub_6731: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_147, minimum_147);  maximum_147 = None
	        div_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6731, 255.0);  sub_6731 = None
	        clamp_min_441: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_294, 1.1920928955078125e-07);  div_294 = None
	        div_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_147, clamp_min_441);  minimum_147 = None
	        round_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_295);  div_295 = None
	        sub_6737: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_295);  round_295 = None
	        clamp_min_442: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6737, -128);  sub_6737 = None
	        clamp_max_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_442, 127);  clamp_min_442 = None
	        _assert_tensor_metadata_1325 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_441, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1325 = None
	        _assert_tensor_metadata_1326 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1326 = None
	        convert_element_type_882: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_294, torch.int8);  clamp_max_294 = None
	        view_2306: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2304, [sym_size_int, 1500, 1280]);  view_2304 = None
	        view_2307: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_441, [sym_size_int, 1500, 1])
	        view_2308: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_882, [sym_size_int, 1500, 1])
	        reciprocal_147: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2307);  view_2307 = None
	        mul_14289: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_147, 1.0);  reciprocal_147 = None
	        mul_14292: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2306, mul_14289);  view_2306 = mul_14289 = None
	        round_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14292);  mul_14292 = None
	        add_22615: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_296, view_2308);  round_296 = view_2308 = None
	        clamp_min_443: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22615, -128);  add_22615 = None
	        clamp_max_295: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_443, 127);  clamp_min_443 = None
	        view_2309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_295, [sym_size_int, 1500, 1280]);  clamp_max_295 = None
	        _assert_tensor_metadata_1327 = torch.ops.aten._assert_tensor_metadata.default(view_2309, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1327 = None
	        convert_element_type_883: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2309, torch.int8);  view_2309 = None
	        view_2310: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_883, [sym_size_int, 1500, 1280]);  convert_element_type_883 = None
	        view_2311: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_441, [sym_size_int, 1500, 1]);  clamp_min_441 = None
	        view_2312: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_882, [sym_size_int, 1500, 1]);  convert_element_type_882 = None
	        _assert_tensor_metadata_1328 = torch.ops.aten._assert_tensor_metadata.default(view_2310, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1328 = None
	        convert_element_type_884: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2310, torch.float32);  view_2310 = None
	        _assert_tensor_metadata_1329 = torch.ops.aten._assert_tensor_metadata.default(view_2312, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1329 = None
	        convert_element_type_885: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2312, torch.float32);  view_2312 = None
	        sub_6757: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_884, convert_element_type_885);  convert_element_type_884 = convert_element_type_885 = None
	        mul_14314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6757, view_2311);  sub_6757 = view_2311 = None
	        view_2313: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14314, [sym_size_int, 1500, 1280]);  mul_14314 = None
	        _assert_tensor_metadata_1330 = torch.ops.aten._assert_tensor_metadata.default(view_2313, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1330 = None
	        view_2314: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2315: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2316: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1331 = torch.ops.aten._assert_tensor_metadata.default(view_2314, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1331 = None
	        convert_element_type_886: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2314, torch.float32);  view_2314 = None
	        _assert_tensor_metadata_1332 = torch.ops.aten._assert_tensor_metadata.default(view_2316, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1332 = None
	        convert_element_type_887: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2316, torch.float32);  view_2316 = None
	        sub_6761: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_886, convert_element_type_887);  convert_element_type_886 = convert_element_type_887 = None
	        mul_14319: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6761, view_2315);  sub_6761 = view_2315 = None
	        view_2317: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14319, [1280, 1280]);  mul_14319 = None
	        _assert_tensor_metadata_1333 = torch.ops.aten._assert_tensor_metadata.default(view_2317, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1333 = None
	        mul_14324: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2318: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2313, [mul_14324, 1280]);  view_2313 = mul_14324 = None
	        permute_248: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2317, [1, 0]);  view_2317 = None
	        addmm_122: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_24_self_attn_out_proj_bias, view_2318, permute_248);  model_audio_tower_layers_24_self_attn_out_proj_bias = view_2318 = permute_248 = None
	        view_2319: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_122, [sym_size_int, 1500, 1280]);  addmm_122 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_197: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2319);  view_2319 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_22678: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_22058, clone_197);  add_22058 = clone_197 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_198: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_22678, memory_format = torch.contiguous_format)
	        var_mean_49 = torch.ops.aten.var_mean.correction(clone_198, [2], correction = 0, keepdim = True)
	        getitem_198: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_49[0]
	        getitem_199: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_49[1];  var_mean_49 = None
	        add_22683: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_198, 1e-05);  getitem_198 = None
	        rsqrt_49: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_22683);  add_22683 = None
	        sub_6767: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_198, getitem_199);  clone_198 = getitem_199 = None
	        mul_14335: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6767, rsqrt_49);  sub_6767 = rsqrt_49 = None
	        mul_14336: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14335, model_audio_tower_layers_24_final_layer_norm_weight);  mul_14335 = model_audio_tower_layers_24_final_layer_norm_weight = None
	        add_22684: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_14336, model_audio_tower_layers_24_final_layer_norm_bias);  mul_14336 = model_audio_tower_layers_24_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2320: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22684, [sym_size_int, 1500, 1280])
	        amin_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2320, [2])
	        amax_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2320, [2]);  view_2320 = None
	        full_296: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_148, full_296);  amin_148 = full_296 = None
	        full_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_148, full_297);  amax_148 = full_297 = None
	        sub_6778: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_148, minimum_148);  maximum_148 = None
	        div_296: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6778, 255.0);  sub_6778 = None
	        clamp_min_444: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_296, 1.1920928955078125e-07);  div_296 = None
	        div_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_148, clamp_min_444);  minimum_148 = None
	        round_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_297);  div_297 = None
	        sub_6784: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_297);  round_297 = None
	        clamp_min_445: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6784, -128);  sub_6784 = None
	        clamp_max_296: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_445, 127);  clamp_min_445 = None
	        _assert_tensor_metadata_1334 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_444, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1334 = None
	        _assert_tensor_metadata_1335 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_296, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1335 = None
	        convert_element_type_888: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_296, torch.int8);  clamp_max_296 = None
	        view_2321: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22684, [sym_size_int, 1500, 1280]);  add_22684 = None
	        view_2322: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_444, [sym_size_int, 1500, 1])
	        view_2323: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_888, [sym_size_int, 1500, 1])
	        reciprocal_148: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2322);  view_2322 = None
	        mul_14384: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_148, 1.0);  reciprocal_148 = None
	        mul_14387: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2321, mul_14384);  view_2321 = mul_14384 = None
	        round_298: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14387);  mul_14387 = None
	        add_22771: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_298, view_2323);  round_298 = view_2323 = None
	        clamp_min_446: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22771, -128);  add_22771 = None
	        clamp_max_297: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_446, 127);  clamp_min_446 = None
	        view_2324: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_297, [sym_size_int, 1500, 1280]);  clamp_max_297 = None
	        _assert_tensor_metadata_1336 = torch.ops.aten._assert_tensor_metadata.default(view_2324, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1336 = None
	        convert_element_type_889: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2324, torch.int8);  view_2324 = None
	        view_2325: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_889, [sym_size_int, 1500, 1280]);  convert_element_type_889 = None
	        view_2326: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_444, [sym_size_int, 1500, 1]);  clamp_min_444 = None
	        view_2327: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_888, [sym_size_int, 1500, 1]);  convert_element_type_888 = None
	        _assert_tensor_metadata_1337 = torch.ops.aten._assert_tensor_metadata.default(view_2325, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1337 = None
	        convert_element_type_890: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2325, torch.float32);  view_2325 = None
	        _assert_tensor_metadata_1338 = torch.ops.aten._assert_tensor_metadata.default(view_2327, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1338 = None
	        convert_element_type_891: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2327, torch.float32);  view_2327 = None
	        sub_6804: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_890, convert_element_type_891);  convert_element_type_890 = convert_element_type_891 = None
	        mul_14409: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6804, view_2326);  sub_6804 = view_2326 = None
	        view_2328: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14409, [sym_size_int, 1500, 1280]);  mul_14409 = None
	        _assert_tensor_metadata_1339 = torch.ops.aten._assert_tensor_metadata.default(view_2328, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1339 = None
	        view_2329: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_24_fc1_parametrizations_weight_original0 = None
	        view_2330: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_24_fc1_parametrizations_weight_original1 = None
	        view_2331: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_24_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1340 = torch.ops.aten._assert_tensor_metadata.default(view_2329, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1340 = None
	        convert_element_type_892: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2329, torch.float32);  view_2329 = None
	        _assert_tensor_metadata_1341 = torch.ops.aten._assert_tensor_metadata.default(view_2331, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1341 = None
	        convert_element_type_893: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2331, torch.float32);  view_2331 = None
	        sub_6808: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_892, convert_element_type_893);  convert_element_type_892 = convert_element_type_893 = None
	        mul_14414: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6808, view_2330);  sub_6808 = view_2330 = None
	        view_2332: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14414, [5120, 1280]);  mul_14414 = None
	        _assert_tensor_metadata_1342 = torch.ops.aten._assert_tensor_metadata.default(view_2332, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1342 = None
	        mul_14419: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2333: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2328, [mul_14419, 1280]);  view_2328 = mul_14419 = None
	        permute_249: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2332, [1, 0]);  view_2332 = None
	        addmm_123: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_24_fc1_bias, view_2333, permute_249);  model_audio_tower_layers_24_fc1_bias = view_2333 = permute_249 = None
	        view_2334: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_123, [sym_size_int, 1500, 5120]);  addmm_123 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_14426: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2334, 0.5)
	        mul_14427: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2334, 0.7071067811865476);  view_2334 = None
	        erf_26: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_14427);  mul_14427 = None
	        add_22830: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_26, 1);  erf_26 = None
	        mul_14428: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14426, add_22830);  mul_14426 = add_22830 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_199: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_14428);  mul_14428 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2335: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_199, [sym_size_int, 1500, 5120])
	        amin_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2335, [2])
	        amax_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2335, [2]);  view_2335 = None
	        full_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_149, full_298);  amin_149 = full_298 = None
	        full_299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_149, full_299);  amax_149 = full_299 = None
	        sub_6821: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_149, minimum_149);  maximum_149 = None
	        div_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6821, 255.0);  sub_6821 = None
	        clamp_min_447: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_298, 1.1920928955078125e-07);  div_298 = None
	        div_299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_149, clamp_min_447);  minimum_149 = None
	        round_299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_299);  div_299 = None
	        sub_6827: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_299);  round_299 = None
	        clamp_min_448: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6827, -128);  sub_6827 = None
	        clamp_max_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_448, 127);  clamp_min_448 = None
	        _assert_tensor_metadata_1343 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_447, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1343 = None
	        _assert_tensor_metadata_1344 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_298, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1344 = None
	        convert_element_type_894: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_298, torch.int8);  clamp_max_298 = None
	        view_2336: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_199, [sym_size_int, 1500, 5120]);  clone_199 = None
	        view_2337: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_447, [sym_size_int, 1500, 1])
	        view_2338: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_894, [sym_size_int, 1500, 1])
	        reciprocal_149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2337);  view_2337 = None
	        mul_14474: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_149, 1.0);  reciprocal_149 = None
	        mul_14477: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2336, mul_14474);  view_2336 = mul_14474 = None
	        round_300: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_14477);  mul_14477 = None
	        add_22913: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_300, view_2338);  round_300 = view_2338 = None
	        clamp_min_449: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22913, -128);  add_22913 = None
	        clamp_max_299: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_449, 127);  clamp_min_449 = None
	        view_2339: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_299, [sym_size_int, 1500, 5120]);  clamp_max_299 = None
	        _assert_tensor_metadata_1345 = torch.ops.aten._assert_tensor_metadata.default(view_2339, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1345 = None
	        convert_element_type_895: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2339, torch.int8);  view_2339 = None
	        view_2340: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_895, [sym_size_int, 1500, 5120]);  convert_element_type_895 = None
	        view_2341: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_447, [sym_size_int, 1500, 1]);  clamp_min_447 = None
	        view_2342: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_894, [sym_size_int, 1500, 1]);  convert_element_type_894 = None
	        _assert_tensor_metadata_1346 = torch.ops.aten._assert_tensor_metadata.default(view_2340, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1346 = None
	        convert_element_type_896: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2340, torch.float32);  view_2340 = None
	        _assert_tensor_metadata_1347 = torch.ops.aten._assert_tensor_metadata.default(view_2342, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1347 = None
	        convert_element_type_897: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2342, torch.float32);  view_2342 = None
	        sub_6847: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_896, convert_element_type_897);  convert_element_type_896 = convert_element_type_897 = None
	        mul_14499: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6847, view_2341);  sub_6847 = view_2341 = None
	        view_2343: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_14499, [sym_size_int, 1500, 5120]);  mul_14499 = None
	        _assert_tensor_metadata_1348 = torch.ops.aten._assert_tensor_metadata.default(view_2343, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1348 = None
	        view_2344: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_24_fc2_parametrizations_weight_original0 = None
	        view_2345: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_24_fc2_parametrizations_weight_original1 = None
	        view_2346: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_24_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1349 = torch.ops.aten._assert_tensor_metadata.default(view_2344, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1349 = None
	        convert_element_type_898: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2344, torch.float32);  view_2344 = None
	        _assert_tensor_metadata_1350 = torch.ops.aten._assert_tensor_metadata.default(view_2346, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1350 = None
	        convert_element_type_899: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2346, torch.float32);  view_2346 = None
	        sub_6851: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_898, convert_element_type_899);  convert_element_type_898 = convert_element_type_899 = None
	        mul_14504: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6851, view_2345);  sub_6851 = view_2345 = None
	        view_2347: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_14504, [1280, 5120]);  mul_14504 = None
	        _assert_tensor_metadata_1351 = torch.ops.aten._assert_tensor_metadata.default(view_2347, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1351 = None
	        mul_14509: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2348: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2343, [mul_14509, 5120]);  view_2343 = mul_14509 = None
	        permute_250: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2347, [1, 0]);  view_2347 = None
	        addmm_124: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_24_fc2_bias, view_2348, permute_250);  model_audio_tower_layers_24_fc2_bias = view_2348 = permute_250 = None
	        view_2349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_124, [sym_size_int, 1500, 1280]);  addmm_124 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2349);  view_2349 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_22976: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_22678, clone_200);  add_22678 = clone_200 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_201: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_22976, memory_format = torch.contiguous_format)
	        var_mean_50 = torch.ops.aten.var_mean.correction(clone_201, [2], correction = 0, keepdim = True)
	        getitem_200: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_50[0]
	        getitem_201: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_50[1];  var_mean_50 = None
	        add_22981: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_200, 1e-05);  getitem_200 = None
	        rsqrt_50: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_22981);  add_22981 = None
	        sub_6857: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_201, getitem_201);  clone_201 = getitem_201 = None
	        mul_14520: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6857, rsqrt_50);  sub_6857 = rsqrt_50 = None
	        mul_14521: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14520, model_audio_tower_layers_25_self_attn_layer_norm_weight);  mul_14520 = model_audio_tower_layers_25_self_attn_layer_norm_weight = None
	        add_22982: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_14521, model_audio_tower_layers_25_self_attn_layer_norm_bias);  mul_14521 = model_audio_tower_layers_25_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22982, [sym_size_int, 1500, 1280])
	        amin_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2350, [2])
	        amax_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2350, [2]);  view_2350 = None
	        full_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_150, full_300);  amin_150 = full_300 = None
	        full_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_150, full_301);  amax_150 = full_301 = None
	        sub_6868: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_150, minimum_150);  maximum_150 = None
	        div_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6868, 255.0);  sub_6868 = None
	        clamp_min_450: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_300, 1.1920928955078125e-07);  div_300 = None
	        div_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_150, clamp_min_450);  minimum_150 = None
	        round_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_301);  div_301 = None
	        sub_6874: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_301);  round_301 = None
	        clamp_min_451: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6874, -128);  sub_6874 = None
	        clamp_max_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_451, 127);  clamp_min_451 = None
	        _assert_tensor_metadata_1352 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_450, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1352 = None
	        _assert_tensor_metadata_1353 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_300, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1353 = None
	        convert_element_type_900: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_300, torch.int8);  clamp_max_300 = None
	        view_2351: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22982, [sym_size_int, 1500, 1280])
	        view_2352: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_450, [sym_size_int, 1500, 1])
	        view_2353: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_900, [sym_size_int, 1500, 1])
	        reciprocal_150: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2352);  view_2352 = None
	        mul_14569: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_150, 1.0);  reciprocal_150 = None
	        mul_14572: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2351, mul_14569);  view_2351 = mul_14569 = None
	        round_302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14572);  mul_14572 = None
	        add_23069: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_302, view_2353);  round_302 = view_2353 = None
	        clamp_min_452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23069, -128);  add_23069 = None
	        clamp_max_301: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_452, 127);  clamp_min_452 = None
	        view_2354: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_301, [sym_size_int, 1500, 1280]);  clamp_max_301 = None
	        _assert_tensor_metadata_1354 = torch.ops.aten._assert_tensor_metadata.default(view_2354, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1354 = None
	        convert_element_type_901: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2354, torch.int8);  view_2354 = None
	        view_2355: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_901, [sym_size_int, 1500, 1280]);  convert_element_type_901 = None
	        view_2356: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_450, [sym_size_int, 1500, 1]);  clamp_min_450 = None
	        view_2357: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_900, [sym_size_int, 1500, 1]);  convert_element_type_900 = None
	        _assert_tensor_metadata_1355 = torch.ops.aten._assert_tensor_metadata.default(view_2355, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1355 = None
	        convert_element_type_902: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2355, torch.float32);  view_2355 = None
	        _assert_tensor_metadata_1356 = torch.ops.aten._assert_tensor_metadata.default(view_2357, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1356 = None
	        convert_element_type_903: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2357, torch.float32);  view_2357 = None
	        sub_6894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_902, convert_element_type_903);  convert_element_type_902 = convert_element_type_903 = None
	        mul_14594: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6894, view_2356);  sub_6894 = view_2356 = None
	        view_2358: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14594, [sym_size_int, 1500, 1280]);  mul_14594 = None
	        _assert_tensor_metadata_1357 = torch.ops.aten._assert_tensor_metadata.default(view_2358, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1357 = None
	        view_2359: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2360: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2361: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1358 = torch.ops.aten._assert_tensor_metadata.default(view_2359, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1358 = None
	        convert_element_type_904: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2359, torch.float32);  view_2359 = None
	        _assert_tensor_metadata_1359 = torch.ops.aten._assert_tensor_metadata.default(view_2361, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1359 = None
	        convert_element_type_905: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2361, torch.float32);  view_2361 = None
	        sub_6898: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_904, convert_element_type_905);  convert_element_type_904 = convert_element_type_905 = None
	        mul_14599: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6898, view_2360);  sub_6898 = view_2360 = None
	        view_2362: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14599, [1280, 1280]);  mul_14599 = None
	        _assert_tensor_metadata_1360 = torch.ops.aten._assert_tensor_metadata.default(view_2362, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1360 = None
	        mul_14604: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2363: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2358, [mul_14604, 1280]);  view_2358 = mul_14604 = None
	        permute_251: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2362, [1, 0]);  view_2362 = None
	        addmm_125: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_25_self_attn_q_proj_bias, view_2363, permute_251);  model_audio_tower_layers_25_self_attn_q_proj_bias = view_2363 = permute_251 = None
	        view_2364: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_125, [sym_size_int, 1500, 1280]);  addmm_125 = None
	        mul_14611: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2364, 0.125);  view_2364 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2365: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_14611, [sym_size_int, 1500, 20, 64]);  mul_14611 = None
	        permute_252: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2365, [0, 2, 1, 3]);  view_2365 = None
	        clone_202: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_252, memory_format = torch.contiguous_format);  permute_252 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2366: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22982, [sym_size_int, 1500, 1280])
	        amin_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2366, [2])
	        amax_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2366, [2]);  view_2366 = None
	        full_302: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_151, full_302);  amin_151 = full_302 = None
	        full_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_151, full_303);  amax_151 = full_303 = None
	        sub_6913: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_151, minimum_151);  maximum_151 = None
	        div_302: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6913, 255.0);  sub_6913 = None
	        clamp_min_453: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_302, 1.1920928955078125e-07);  div_302 = None
	        div_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_151, clamp_min_453);  minimum_151 = None
	        round_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_303);  div_303 = None
	        sub_6919: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_303);  round_303 = None
	        clamp_min_454: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6919, -128);  sub_6919 = None
	        clamp_max_302: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_454, 127);  clamp_min_454 = None
	        _assert_tensor_metadata_1361 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_453, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1361 = None
	        _assert_tensor_metadata_1362 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_302, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1362 = None
	        convert_element_type_906: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_302, torch.int8);  clamp_max_302 = None
	        view_2367: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22982, [sym_size_int, 1500, 1280])
	        view_2368: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_453, [sym_size_int, 1500, 1])
	        view_2369: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_906, [sym_size_int, 1500, 1])
	        reciprocal_151: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2368);  view_2368 = None
	        mul_14665: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_151, 1.0);  reciprocal_151 = None
	        mul_14668: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2367, mul_14665);  view_2367 = mul_14665 = None
	        round_304: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14668);  mul_14668 = None
	        add_23221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_304, view_2369);  round_304 = view_2369 = None
	        clamp_min_455: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23221, -128);  add_23221 = None
	        clamp_max_303: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_455, 127);  clamp_min_455 = None
	        view_2370: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_303, [sym_size_int, 1500, 1280]);  clamp_max_303 = None
	        _assert_tensor_metadata_1363 = torch.ops.aten._assert_tensor_metadata.default(view_2370, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1363 = None
	        convert_element_type_907: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2370, torch.int8);  view_2370 = None
	        view_2371: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_907, [sym_size_int, 1500, 1280]);  convert_element_type_907 = None
	        view_2372: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_453, [sym_size_int, 1500, 1]);  clamp_min_453 = None
	        view_2373: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_906, [sym_size_int, 1500, 1]);  convert_element_type_906 = None
	        _assert_tensor_metadata_1364 = torch.ops.aten._assert_tensor_metadata.default(view_2371, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1364 = None
	        convert_element_type_908: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2371, torch.float32);  view_2371 = None
	        _assert_tensor_metadata_1365 = torch.ops.aten._assert_tensor_metadata.default(view_2373, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1365 = None
	        convert_element_type_909: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2373, torch.float32);  view_2373 = None
	        sub_6939: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_908, convert_element_type_909);  convert_element_type_908 = convert_element_type_909 = None
	        mul_14690: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6939, view_2372);  sub_6939 = view_2372 = None
	        view_2374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14690, [sym_size_int, 1500, 1280]);  mul_14690 = None
	        _assert_tensor_metadata_1366 = torch.ops.aten._assert_tensor_metadata.default(view_2374, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1366 = None
	        view_2375: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2376: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2377: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1367 = torch.ops.aten._assert_tensor_metadata.default(view_2375, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1367 = None
	        convert_element_type_910: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2375, torch.float32);  view_2375 = None
	        _assert_tensor_metadata_1368 = torch.ops.aten._assert_tensor_metadata.default(view_2377, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1368 = None
	        convert_element_type_911: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2377, torch.float32);  view_2377 = None
	        sub_6943: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_910, convert_element_type_911);  convert_element_type_910 = convert_element_type_911 = None
	        mul_14695: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6943, view_2376);  sub_6943 = view_2376 = None
	        view_2378: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14695, [1280, 1280]);  mul_14695 = None
	        _assert_tensor_metadata_1369 = torch.ops.aten._assert_tensor_metadata.default(view_2378, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1369 = None
	        permute_253: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2378, [1, 0]);  view_2378 = None
	        mul_14698: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2379: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2374, [mul_14698, 1280]);  view_2374 = mul_14698 = None
	        mm_25: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2379, permute_253);  view_2379 = permute_253 = None
	        view_2380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_25, [sym_size_int, 1500, 1280]);  mm_25 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2381: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2380, [sym_size_int, -1, 20, 64]);  view_2380 = None
	        permute_254: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2381, [0, 2, 1, 3]);  view_2381 = None
	        clone_203: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_254, memory_format = torch.contiguous_format);  permute_254 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2382: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22982, [sym_size_int, 1500, 1280])
	        amin_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2382, [2])
	        amax_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2382, [2]);  view_2382 = None
	        full_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_152, full_304);  amin_152 = full_304 = None
	        full_305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_152, full_305);  amax_152 = full_305 = None
	        sub_6957: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_152, minimum_152);  maximum_152 = None
	        div_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6957, 255.0);  sub_6957 = None
	        clamp_min_456: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_304, 1.1920928955078125e-07);  div_304 = None
	        div_305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_152, clamp_min_456);  minimum_152 = None
	        round_305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_305);  div_305 = None
	        sub_6963: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_305);  round_305 = None
	        clamp_min_457: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6963, -128);  sub_6963 = None
	        clamp_max_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_457, 127);  clamp_min_457 = None
	        _assert_tensor_metadata_1370 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_456, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1370 = None
	        _assert_tensor_metadata_1371 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_304, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1371 = None
	        convert_element_type_912: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_304, torch.int8);  clamp_max_304 = None
	        view_2383: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_22982, [sym_size_int, 1500, 1280]);  add_22982 = None
	        view_2384: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_456, [sym_size_int, 1500, 1])
	        view_2385: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_912, [sym_size_int, 1500, 1])
	        reciprocal_152: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2384);  view_2384 = None
	        mul_14764: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_152, 1.0);  reciprocal_152 = None
	        mul_14767: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2383, mul_14764);  view_2383 = mul_14764 = None
	        round_306: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14767);  mul_14767 = None
	        add_23369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_306, view_2385);  round_306 = view_2385 = None
	        clamp_min_458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23369, -128);  add_23369 = None
	        clamp_max_305: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_458, 127);  clamp_min_458 = None
	        view_2386: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_305, [sym_size_int, 1500, 1280]);  clamp_max_305 = None
	        _assert_tensor_metadata_1372 = torch.ops.aten._assert_tensor_metadata.default(view_2386, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1372 = None
	        convert_element_type_913: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2386, torch.int8);  view_2386 = None
	        view_2387: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_913, [sym_size_int, 1500, 1280]);  convert_element_type_913 = None
	        view_2388: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_456, [sym_size_int, 1500, 1]);  clamp_min_456 = None
	        view_2389: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_912, [sym_size_int, 1500, 1]);  convert_element_type_912 = None
	        _assert_tensor_metadata_1373 = torch.ops.aten._assert_tensor_metadata.default(view_2387, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1373 = None
	        convert_element_type_914: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2387, torch.float32);  view_2387 = None
	        _assert_tensor_metadata_1374 = torch.ops.aten._assert_tensor_metadata.default(view_2389, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1374 = None
	        convert_element_type_915: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2389, torch.float32);  view_2389 = None
	        sub_6983: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_914, convert_element_type_915);  convert_element_type_914 = convert_element_type_915 = None
	        mul_14789: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6983, view_2388);  sub_6983 = view_2388 = None
	        view_2390: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14789, [sym_size_int, 1500, 1280]);  mul_14789 = None
	        _assert_tensor_metadata_1375 = torch.ops.aten._assert_tensor_metadata.default(view_2390, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1375 = None
	        view_2391: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2392: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2393: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1376 = torch.ops.aten._assert_tensor_metadata.default(view_2391, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1376 = None
	        convert_element_type_916: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2391, torch.float32);  view_2391 = None
	        _assert_tensor_metadata_1377 = torch.ops.aten._assert_tensor_metadata.default(view_2393, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1377 = None
	        convert_element_type_917: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2393, torch.float32);  view_2393 = None
	        sub_6987: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_916, convert_element_type_917);  convert_element_type_916 = convert_element_type_917 = None
	        mul_14794: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6987, view_2392);  sub_6987 = view_2392 = None
	        view_2394: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14794, [1280, 1280]);  mul_14794 = None
	        _assert_tensor_metadata_1378 = torch.ops.aten._assert_tensor_metadata.default(view_2394, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1378 = None
	        mul_14799: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2395: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2390, [mul_14799, 1280]);  view_2390 = mul_14799 = None
	        permute_255: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2394, [1, 0]);  view_2394 = None
	        addmm_126: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_25_self_attn_v_proj_bias, view_2395, permute_255);  model_audio_tower_layers_25_self_attn_v_proj_bias = view_2395 = permute_255 = None
	        view_2396: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_126, [sym_size_int, 1500, 1280]);  addmm_126 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2397: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2396, [sym_size_int, -1, 20, 64]);  view_2396 = None
	        permute_256: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2397, [0, 2, 1, 3]);  view_2397 = None
	        clone_204: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_256, memory_format = torch.contiguous_format);  permute_256 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_25 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_202, clone_203, clone_204, None, False, scale = 1.0);  clone_202 = clone_203 = clone_204 = None
	        getitem_202: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_25[0];  _scaled_dot_product_efficient_attention_25 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_257: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_202, [0, 2, 1, 3]);  getitem_202 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2398: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_257, [sym_size_int, 1500, -1]);  permute_257 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2399: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2398, [sym_size_int, 1500, 1280])
	        amin_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2399, [2])
	        amax_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2399, [2]);  view_2399 = None
	        full_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_153, full_306);  amin_153 = full_306 = None
	        full_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_153, full_307);  amax_153 = full_307 = None
	        sub_7005: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_153, minimum_153);  maximum_153 = None
	        div_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7005, 255.0);  sub_7005 = None
	        clamp_min_459: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_306, 1.1920928955078125e-07);  div_306 = None
	        div_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_153, clamp_min_459);  minimum_153 = None
	        round_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_307);  div_307 = None
	        sub_7011: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_307);  round_307 = None
	        clamp_min_460: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7011, -128);  sub_7011 = None
	        clamp_max_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_460, 127);  clamp_min_460 = None
	        _assert_tensor_metadata_1379 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_459, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1379 = None
	        _assert_tensor_metadata_1380 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_306, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1380 = None
	        convert_element_type_918: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_306, torch.int8);  clamp_max_306 = None
	        view_2400: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2398, [sym_size_int, 1500, 1280]);  view_2398 = None
	        view_2401: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_459, [sym_size_int, 1500, 1])
	        view_2402: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_918, [sym_size_int, 1500, 1])
	        reciprocal_153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2401);  view_2401 = None
	        mul_14869: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_153, 1.0);  reciprocal_153 = None
	        mul_14872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2400, mul_14869);  view_2400 = mul_14869 = None
	        round_308: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14872);  mul_14872 = None
	        add_23533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_308, view_2402);  round_308 = view_2402 = None
	        clamp_min_461: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23533, -128);  add_23533 = None
	        clamp_max_307: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_461, 127);  clamp_min_461 = None
	        view_2403: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_307, [sym_size_int, 1500, 1280]);  clamp_max_307 = None
	        _assert_tensor_metadata_1381 = torch.ops.aten._assert_tensor_metadata.default(view_2403, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1381 = None
	        convert_element_type_919: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2403, torch.int8);  view_2403 = None
	        view_2404: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_919, [sym_size_int, 1500, 1280]);  convert_element_type_919 = None
	        view_2405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_459, [sym_size_int, 1500, 1]);  clamp_min_459 = None
	        view_2406: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_918, [sym_size_int, 1500, 1]);  convert_element_type_918 = None
	        _assert_tensor_metadata_1382 = torch.ops.aten._assert_tensor_metadata.default(view_2404, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1382 = None
	        convert_element_type_920: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2404, torch.float32);  view_2404 = None
	        _assert_tensor_metadata_1383 = torch.ops.aten._assert_tensor_metadata.default(view_2406, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1383 = None
	        convert_element_type_921: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2406, torch.float32);  view_2406 = None
	        sub_7031: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_920, convert_element_type_921);  convert_element_type_920 = convert_element_type_921 = None
	        mul_14894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7031, view_2405);  sub_7031 = view_2405 = None
	        view_2407: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14894, [sym_size_int, 1500, 1280]);  mul_14894 = None
	        _assert_tensor_metadata_1384 = torch.ops.aten._assert_tensor_metadata.default(view_2407, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1384 = None
	        view_2408: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2409: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2410: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1385 = torch.ops.aten._assert_tensor_metadata.default(view_2408, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1385 = None
	        convert_element_type_922: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2408, torch.float32);  view_2408 = None
	        _assert_tensor_metadata_1386 = torch.ops.aten._assert_tensor_metadata.default(view_2410, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1386 = None
	        convert_element_type_923: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2410, torch.float32);  view_2410 = None
	        sub_7035: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_922, convert_element_type_923);  convert_element_type_922 = convert_element_type_923 = None
	        mul_14899: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7035, view_2409);  sub_7035 = view_2409 = None
	        view_2411: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14899, [1280, 1280]);  mul_14899 = None
	        _assert_tensor_metadata_1387 = torch.ops.aten._assert_tensor_metadata.default(view_2411, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1387 = None
	        mul_14904: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2412: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2407, [mul_14904, 1280]);  view_2407 = mul_14904 = None
	        permute_258: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2411, [1, 0]);  view_2411 = None
	        addmm_127: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_25_self_attn_out_proj_bias, view_2412, permute_258);  model_audio_tower_layers_25_self_attn_out_proj_bias = view_2412 = permute_258 = None
	        view_2413: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_127, [sym_size_int, 1500, 1280]);  addmm_127 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2413);  view_2413 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_23596: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_22976, clone_205);  add_22976 = clone_205 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_23596, memory_format = torch.contiguous_format)
	        var_mean_51 = torch.ops.aten.var_mean.correction(clone_206, [2], correction = 0, keepdim = True)
	        getitem_206: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_51[0]
	        getitem_207: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_51[1];  var_mean_51 = None
	        add_23601: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_206, 1e-05);  getitem_206 = None
	        rsqrt_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_23601);  add_23601 = None
	        sub_7041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_206, getitem_207);  clone_206 = getitem_207 = None
	        mul_14915: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7041, rsqrt_51);  sub_7041 = rsqrt_51 = None
	        mul_14916: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14915, model_audio_tower_layers_25_final_layer_norm_weight);  mul_14915 = model_audio_tower_layers_25_final_layer_norm_weight = None
	        add_23602: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_14916, model_audio_tower_layers_25_final_layer_norm_bias);  mul_14916 = model_audio_tower_layers_25_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2414: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_23602, [sym_size_int, 1500, 1280])
	        amin_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2414, [2])
	        amax_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2414, [2]);  view_2414 = None
	        full_308: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_154, full_308);  amin_154 = full_308 = None
	        full_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_154, full_309);  amax_154 = full_309 = None
	        sub_7052: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_154, minimum_154);  maximum_154 = None
	        div_308: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7052, 255.0);  sub_7052 = None
	        clamp_min_462: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_308, 1.1920928955078125e-07);  div_308 = None
	        div_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_154, clamp_min_462);  minimum_154 = None
	        round_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_309);  div_309 = None
	        sub_7058: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_309);  round_309 = None
	        clamp_min_463: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7058, -128);  sub_7058 = None
	        clamp_max_308: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_463, 127);  clamp_min_463 = None
	        _assert_tensor_metadata_1388 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_462, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1388 = None
	        _assert_tensor_metadata_1389 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_308, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1389 = None
	        convert_element_type_924: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_308, torch.int8);  clamp_max_308 = None
	        view_2415: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_23602, [sym_size_int, 1500, 1280]);  add_23602 = None
	        view_2416: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_462, [sym_size_int, 1500, 1])
	        view_2417: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_924, [sym_size_int, 1500, 1])
	        reciprocal_154: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2416);  view_2416 = None
	        mul_14964: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_154, 1.0);  reciprocal_154 = None
	        mul_14967: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2415, mul_14964);  view_2415 = mul_14964 = None
	        round_310: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14967);  mul_14967 = None
	        add_23689: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_310, view_2417);  round_310 = view_2417 = None
	        clamp_min_464: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23689, -128);  add_23689 = None
	        clamp_max_309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_464, 127);  clamp_min_464 = None
	        view_2418: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_309, [sym_size_int, 1500, 1280]);  clamp_max_309 = None
	        _assert_tensor_metadata_1390 = torch.ops.aten._assert_tensor_metadata.default(view_2418, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1390 = None
	        convert_element_type_925: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2418, torch.int8);  view_2418 = None
	        view_2419: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_925, [sym_size_int, 1500, 1280]);  convert_element_type_925 = None
	        view_2420: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_462, [sym_size_int, 1500, 1]);  clamp_min_462 = None
	        view_2421: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_924, [sym_size_int, 1500, 1]);  convert_element_type_924 = None
	        _assert_tensor_metadata_1391 = torch.ops.aten._assert_tensor_metadata.default(view_2419, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1391 = None
	        convert_element_type_926: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2419, torch.float32);  view_2419 = None
	        _assert_tensor_metadata_1392 = torch.ops.aten._assert_tensor_metadata.default(view_2421, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1392 = None
	        convert_element_type_927: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2421, torch.float32);  view_2421 = None
	        sub_7078: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_926, convert_element_type_927);  convert_element_type_926 = convert_element_type_927 = None
	        mul_14989: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7078, view_2420);  sub_7078 = view_2420 = None
	        view_2422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14989, [sym_size_int, 1500, 1280]);  mul_14989 = None
	        _assert_tensor_metadata_1393 = torch.ops.aten._assert_tensor_metadata.default(view_2422, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1393 = None
	        view_2423: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_25_fc1_parametrizations_weight_original0 = None
	        view_2424: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_25_fc1_parametrizations_weight_original1 = None
	        view_2425: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_25_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1394 = torch.ops.aten._assert_tensor_metadata.default(view_2423, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1394 = None
	        convert_element_type_928: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2423, torch.float32);  view_2423 = None
	        _assert_tensor_metadata_1395 = torch.ops.aten._assert_tensor_metadata.default(view_2425, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1395 = None
	        convert_element_type_929: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2425, torch.float32);  view_2425 = None
	        sub_7082: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_928, convert_element_type_929);  convert_element_type_928 = convert_element_type_929 = None
	        mul_14994: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7082, view_2424);  sub_7082 = view_2424 = None
	        view_2426: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14994, [5120, 1280]);  mul_14994 = None
	        _assert_tensor_metadata_1396 = torch.ops.aten._assert_tensor_metadata.default(view_2426, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1396 = None
	        mul_14999: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2427: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2422, [mul_14999, 1280]);  view_2422 = mul_14999 = None
	        permute_259: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2426, [1, 0]);  view_2426 = None
	        addmm_128: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_25_fc1_bias, view_2427, permute_259);  model_audio_tower_layers_25_fc1_bias = view_2427 = permute_259 = None
	        view_2428: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_128, [sym_size_int, 1500, 5120]);  addmm_128 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_15006: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2428, 0.5)
	        mul_15007: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2428, 0.7071067811865476);  view_2428 = None
	        erf_27: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_15007);  mul_15007 = None
	        add_23748: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_27, 1);  erf_27 = None
	        mul_15008: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15006, add_23748);  mul_15006 = add_23748 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_207: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_15008);  mul_15008 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2429: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_207, [sym_size_int, 1500, 5120])
	        amin_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2429, [2])
	        amax_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2429, [2]);  view_2429 = None
	        full_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_155, full_310);  amin_155 = full_310 = None
	        full_311: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_155, full_311);  amax_155 = full_311 = None
	        sub_7095: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_155, minimum_155);  maximum_155 = None
	        div_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7095, 255.0);  sub_7095 = None
	        clamp_min_465: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_310, 1.1920928955078125e-07);  div_310 = None
	        div_311: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_155, clamp_min_465);  minimum_155 = None
	        round_311: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_311);  div_311 = None
	        sub_7101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_311);  round_311 = None
	        clamp_min_466: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7101, -128);  sub_7101 = None
	        clamp_max_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_466, 127);  clamp_min_466 = None
	        _assert_tensor_metadata_1397 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_465, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1397 = None
	        _assert_tensor_metadata_1398 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_310, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1398 = None
	        convert_element_type_930: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_310, torch.int8);  clamp_max_310 = None
	        view_2430: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_207, [sym_size_int, 1500, 5120]);  clone_207 = None
	        view_2431: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_465, [sym_size_int, 1500, 1])
	        view_2432: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_930, [sym_size_int, 1500, 1])
	        reciprocal_155: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2431);  view_2431 = None
	        mul_15054: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_155, 1.0);  reciprocal_155 = None
	        mul_15057: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2430, mul_15054);  view_2430 = mul_15054 = None
	        round_312: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_15057);  mul_15057 = None
	        add_23831: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_312, view_2432);  round_312 = view_2432 = None
	        clamp_min_467: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23831, -128);  add_23831 = None
	        clamp_max_311: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_467, 127);  clamp_min_467 = None
	        view_2433: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_311, [sym_size_int, 1500, 5120]);  clamp_max_311 = None
	        _assert_tensor_metadata_1399 = torch.ops.aten._assert_tensor_metadata.default(view_2433, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1399 = None
	        convert_element_type_931: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2433, torch.int8);  view_2433 = None
	        view_2434: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_931, [sym_size_int, 1500, 5120]);  convert_element_type_931 = None
	        view_2435: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_465, [sym_size_int, 1500, 1]);  clamp_min_465 = None
	        view_2436: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_930, [sym_size_int, 1500, 1]);  convert_element_type_930 = None
	        _assert_tensor_metadata_1400 = torch.ops.aten._assert_tensor_metadata.default(view_2434, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1400 = None
	        convert_element_type_932: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2434, torch.float32);  view_2434 = None
	        _assert_tensor_metadata_1401 = torch.ops.aten._assert_tensor_metadata.default(view_2436, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1401 = None
	        convert_element_type_933: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2436, torch.float32);  view_2436 = None
	        sub_7121: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_932, convert_element_type_933);  convert_element_type_932 = convert_element_type_933 = None
	        mul_15079: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7121, view_2435);  sub_7121 = view_2435 = None
	        view_2437: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_15079, [sym_size_int, 1500, 5120]);  mul_15079 = None
	        _assert_tensor_metadata_1402 = torch.ops.aten._assert_tensor_metadata.default(view_2437, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1402 = None
	        view_2438: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_25_fc2_parametrizations_weight_original0 = None
	        view_2439: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_25_fc2_parametrizations_weight_original1 = None
	        view_2440: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_25_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1403 = torch.ops.aten._assert_tensor_metadata.default(view_2438, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1403 = None
	        convert_element_type_934: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2438, torch.float32);  view_2438 = None
	        _assert_tensor_metadata_1404 = torch.ops.aten._assert_tensor_metadata.default(view_2440, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1404 = None
	        convert_element_type_935: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2440, torch.float32);  view_2440 = None
	        sub_7125: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_934, convert_element_type_935);  convert_element_type_934 = convert_element_type_935 = None
	        mul_15084: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7125, view_2439);  sub_7125 = view_2439 = None
	        view_2441: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_15084, [1280, 5120]);  mul_15084 = None
	        _assert_tensor_metadata_1405 = torch.ops.aten._assert_tensor_metadata.default(view_2441, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1405 = None
	        mul_15089: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2442: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2437, [mul_15089, 5120]);  view_2437 = mul_15089 = None
	        permute_260: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2441, [1, 0]);  view_2441 = None
	        addmm_129: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_25_fc2_bias, view_2442, permute_260);  model_audio_tower_layers_25_fc2_bias = view_2442 = permute_260 = None
	        view_2443: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_129, [sym_size_int, 1500, 1280]);  addmm_129 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_208: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2443);  view_2443 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_23894: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_23596, clone_208);  add_23596 = clone_208 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_23894, memory_format = torch.contiguous_format)
	        var_mean_52 = torch.ops.aten.var_mean.correction(clone_209, [2], correction = 0, keepdim = True)
	        getitem_208: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_52[0]
	        getitem_209: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_52[1];  var_mean_52 = None
	        add_23899: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_208, 1e-05);  getitem_208 = None
	        rsqrt_52: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_23899);  add_23899 = None
	        sub_7131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_209, getitem_209);  clone_209 = getitem_209 = None
	        mul_15100: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7131, rsqrt_52);  sub_7131 = rsqrt_52 = None
	        mul_15101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15100, model_audio_tower_layers_26_self_attn_layer_norm_weight);  mul_15100 = model_audio_tower_layers_26_self_attn_layer_norm_weight = None
	        add_23900: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_15101, model_audio_tower_layers_26_self_attn_layer_norm_bias);  mul_15101 = model_audio_tower_layers_26_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2444: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_23900, [sym_size_int, 1500, 1280])
	        amin_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2444, [2])
	        amax_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2444, [2]);  view_2444 = None
	        full_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_156, full_312);  amin_156 = full_312 = None
	        full_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_156, full_313);  amax_156 = full_313 = None
	        sub_7142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_156, minimum_156);  maximum_156 = None
	        div_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7142, 255.0);  sub_7142 = None
	        clamp_min_468: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_312, 1.1920928955078125e-07);  div_312 = None
	        div_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_156, clamp_min_468);  minimum_156 = None
	        round_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_313);  div_313 = None
	        sub_7148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_313);  round_313 = None
	        clamp_min_469: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7148, -128);  sub_7148 = None
	        clamp_max_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_469, 127);  clamp_min_469 = None
	        _assert_tensor_metadata_1406 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_468, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1406 = None
	        _assert_tensor_metadata_1407 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_312, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1407 = None
	        convert_element_type_936: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_312, torch.int8);  clamp_max_312 = None
	        view_2445: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_23900, [sym_size_int, 1500, 1280])
	        view_2446: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_468, [sym_size_int, 1500, 1])
	        view_2447: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_936, [sym_size_int, 1500, 1])
	        reciprocal_156: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2446);  view_2446 = None
	        mul_15149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_156, 1.0);  reciprocal_156 = None
	        mul_15152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2445, mul_15149);  view_2445 = mul_15149 = None
	        round_314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15152);  mul_15152 = None
	        add_23987: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_314, view_2447);  round_314 = view_2447 = None
	        clamp_min_470: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23987, -128);  add_23987 = None
	        clamp_max_313: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_470, 127);  clamp_min_470 = None
	        view_2448: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_313, [sym_size_int, 1500, 1280]);  clamp_max_313 = None
	        _assert_tensor_metadata_1408 = torch.ops.aten._assert_tensor_metadata.default(view_2448, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1408 = None
	        convert_element_type_937: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2448, torch.int8);  view_2448 = None
	        view_2449: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_937, [sym_size_int, 1500, 1280]);  convert_element_type_937 = None
	        view_2450: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_468, [sym_size_int, 1500, 1]);  clamp_min_468 = None
	        view_2451: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_936, [sym_size_int, 1500, 1]);  convert_element_type_936 = None
	        _assert_tensor_metadata_1409 = torch.ops.aten._assert_tensor_metadata.default(view_2449, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1409 = None
	        convert_element_type_938: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2449, torch.float32);  view_2449 = None
	        _assert_tensor_metadata_1410 = torch.ops.aten._assert_tensor_metadata.default(view_2451, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1410 = None
	        convert_element_type_939: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2451, torch.float32);  view_2451 = None
	        sub_7168: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_938, convert_element_type_939);  convert_element_type_938 = convert_element_type_939 = None
	        mul_15174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7168, view_2450);  sub_7168 = view_2450 = None
	        view_2452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15174, [sym_size_int, 1500, 1280]);  mul_15174 = None
	        _assert_tensor_metadata_1411 = torch.ops.aten._assert_tensor_metadata.default(view_2452, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1411 = None
	        view_2453: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2454: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2455: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1412 = torch.ops.aten._assert_tensor_metadata.default(view_2453, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1412 = None
	        convert_element_type_940: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2453, torch.float32);  view_2453 = None
	        _assert_tensor_metadata_1413 = torch.ops.aten._assert_tensor_metadata.default(view_2455, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1413 = None
	        convert_element_type_941: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2455, torch.float32);  view_2455 = None
	        sub_7172: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_940, convert_element_type_941);  convert_element_type_940 = convert_element_type_941 = None
	        mul_15179: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7172, view_2454);  sub_7172 = view_2454 = None
	        view_2456: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15179, [1280, 1280]);  mul_15179 = None
	        _assert_tensor_metadata_1414 = torch.ops.aten._assert_tensor_metadata.default(view_2456, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1414 = None
	        mul_15184: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2457: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2452, [mul_15184, 1280]);  view_2452 = mul_15184 = None
	        permute_261: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2456, [1, 0]);  view_2456 = None
	        addmm_130: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_26_self_attn_q_proj_bias, view_2457, permute_261);  model_audio_tower_layers_26_self_attn_q_proj_bias = view_2457 = permute_261 = None
	        view_2458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_130, [sym_size_int, 1500, 1280]);  addmm_130 = None
	        mul_15191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2458, 0.125);  view_2458 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2459: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_15191, [sym_size_int, 1500, 20, 64]);  mul_15191 = None
	        permute_262: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2459, [0, 2, 1, 3]);  view_2459 = None
	        clone_210: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_262, memory_format = torch.contiguous_format);  permute_262 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2460: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_23900, [sym_size_int, 1500, 1280])
	        amin_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2460, [2])
	        amax_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2460, [2]);  view_2460 = None
	        full_314: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_157, full_314);  amin_157 = full_314 = None
	        full_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_157, full_315);  amax_157 = full_315 = None
	        sub_7187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_157, minimum_157);  maximum_157 = None
	        div_314: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7187, 255.0);  sub_7187 = None
	        clamp_min_471: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_314, 1.1920928955078125e-07);  div_314 = None
	        div_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_157, clamp_min_471);  minimum_157 = None
	        round_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_315);  div_315 = None
	        sub_7193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_315);  round_315 = None
	        clamp_min_472: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7193, -128);  sub_7193 = None
	        clamp_max_314: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_472, 127);  clamp_min_472 = None
	        _assert_tensor_metadata_1415 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_471, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1415 = None
	        _assert_tensor_metadata_1416 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_314, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1416 = None
	        convert_element_type_942: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_314, torch.int8);  clamp_max_314 = None
	        view_2461: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_23900, [sym_size_int, 1500, 1280])
	        view_2462: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_471, [sym_size_int, 1500, 1])
	        view_2463: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_942, [sym_size_int, 1500, 1])
	        reciprocal_157: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2462);  view_2462 = None
	        mul_15245: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_157, 1.0);  reciprocal_157 = None
	        mul_15248: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2461, mul_15245);  view_2461 = mul_15245 = None
	        round_316: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15248);  mul_15248 = None
	        add_24139: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_316, view_2463);  round_316 = view_2463 = None
	        clamp_min_473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24139, -128);  add_24139 = None
	        clamp_max_315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_473, 127);  clamp_min_473 = None
	        view_2464: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_315, [sym_size_int, 1500, 1280]);  clamp_max_315 = None
	        _assert_tensor_metadata_1417 = torch.ops.aten._assert_tensor_metadata.default(view_2464, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1417 = None
	        convert_element_type_943: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2464, torch.int8);  view_2464 = None
	        view_2465: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_943, [sym_size_int, 1500, 1280]);  convert_element_type_943 = None
	        view_2466: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_471, [sym_size_int, 1500, 1]);  clamp_min_471 = None
	        view_2467: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_942, [sym_size_int, 1500, 1]);  convert_element_type_942 = None
	        _assert_tensor_metadata_1418 = torch.ops.aten._assert_tensor_metadata.default(view_2465, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1418 = None
	        convert_element_type_944: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2465, torch.float32);  view_2465 = None
	        _assert_tensor_metadata_1419 = torch.ops.aten._assert_tensor_metadata.default(view_2467, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1419 = None
	        convert_element_type_945: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2467, torch.float32);  view_2467 = None
	        sub_7213: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_944, convert_element_type_945);  convert_element_type_944 = convert_element_type_945 = None
	        mul_15270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7213, view_2466);  sub_7213 = view_2466 = None
	        view_2468: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15270, [sym_size_int, 1500, 1280]);  mul_15270 = None
	        _assert_tensor_metadata_1420 = torch.ops.aten._assert_tensor_metadata.default(view_2468, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1420 = None
	        view_2469: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2470: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2471: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1421 = torch.ops.aten._assert_tensor_metadata.default(view_2469, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1421 = None
	        convert_element_type_946: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2469, torch.float32);  view_2469 = None
	        _assert_tensor_metadata_1422 = torch.ops.aten._assert_tensor_metadata.default(view_2471, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1422 = None
	        convert_element_type_947: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2471, torch.float32);  view_2471 = None
	        sub_7217: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_946, convert_element_type_947);  convert_element_type_946 = convert_element_type_947 = None
	        mul_15275: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7217, view_2470);  sub_7217 = view_2470 = None
	        view_2472: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15275, [1280, 1280]);  mul_15275 = None
	        _assert_tensor_metadata_1423 = torch.ops.aten._assert_tensor_metadata.default(view_2472, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1423 = None
	        permute_263: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2472, [1, 0]);  view_2472 = None
	        mul_15278: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2473: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2468, [mul_15278, 1280]);  view_2468 = mul_15278 = None
	        mm_26: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2473, permute_263);  view_2473 = permute_263 = None
	        view_2474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_26, [sym_size_int, 1500, 1280]);  mm_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2475: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2474, [sym_size_int, -1, 20, 64]);  view_2474 = None
	        permute_264: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2475, [0, 2, 1, 3]);  view_2475 = None
	        clone_211: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_264, memory_format = torch.contiguous_format);  permute_264 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2476: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_23900, [sym_size_int, 1500, 1280])
	        amin_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2476, [2])
	        amax_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2476, [2]);  view_2476 = None
	        full_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_158, full_316);  amin_158 = full_316 = None
	        full_317: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_158, full_317);  amax_158 = full_317 = None
	        sub_7231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_158, minimum_158);  maximum_158 = None
	        div_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7231, 255.0);  sub_7231 = None
	        clamp_min_474: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_316, 1.1920928955078125e-07);  div_316 = None
	        div_317: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_158, clamp_min_474);  minimum_158 = None
	        round_317: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_317);  div_317 = None
	        sub_7237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_317);  round_317 = None
	        clamp_min_475: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7237, -128);  sub_7237 = None
	        clamp_max_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_475, 127);  clamp_min_475 = None
	        _assert_tensor_metadata_1424 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_474, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1424 = None
	        _assert_tensor_metadata_1425 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_316, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1425 = None
	        convert_element_type_948: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_316, torch.int8);  clamp_max_316 = None
	        view_2477: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_23900, [sym_size_int, 1500, 1280]);  add_23900 = None
	        view_2478: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_474, [sym_size_int, 1500, 1])
	        view_2479: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_948, [sym_size_int, 1500, 1])
	        reciprocal_158: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2478);  view_2478 = None
	        mul_15344: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_158, 1.0);  reciprocal_158 = None
	        mul_15347: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2477, mul_15344);  view_2477 = mul_15344 = None
	        round_318: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15347);  mul_15347 = None
	        add_24287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_318, view_2479);  round_318 = view_2479 = None
	        clamp_min_476: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24287, -128);  add_24287 = None
	        clamp_max_317: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_476, 127);  clamp_min_476 = None
	        view_2480: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_317, [sym_size_int, 1500, 1280]);  clamp_max_317 = None
	        _assert_tensor_metadata_1426 = torch.ops.aten._assert_tensor_metadata.default(view_2480, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1426 = None
	        convert_element_type_949: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2480, torch.int8);  view_2480 = None
	        view_2481: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_949, [sym_size_int, 1500, 1280]);  convert_element_type_949 = None
	        view_2482: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_474, [sym_size_int, 1500, 1]);  clamp_min_474 = None
	        view_2483: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_948, [sym_size_int, 1500, 1]);  convert_element_type_948 = None
	        _assert_tensor_metadata_1427 = torch.ops.aten._assert_tensor_metadata.default(view_2481, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1427 = None
	        convert_element_type_950: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2481, torch.float32);  view_2481 = None
	        _assert_tensor_metadata_1428 = torch.ops.aten._assert_tensor_metadata.default(view_2483, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1428 = None
	        convert_element_type_951: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2483, torch.float32);  view_2483 = None
	        sub_7257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_950, convert_element_type_951);  convert_element_type_950 = convert_element_type_951 = None
	        mul_15369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7257, view_2482);  sub_7257 = view_2482 = None
	        view_2484: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15369, [sym_size_int, 1500, 1280]);  mul_15369 = None
	        _assert_tensor_metadata_1429 = torch.ops.aten._assert_tensor_metadata.default(view_2484, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1429 = None
	        view_2485: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2486: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2487: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1430 = torch.ops.aten._assert_tensor_metadata.default(view_2485, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1430 = None
	        convert_element_type_952: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2485, torch.float32);  view_2485 = None
	        _assert_tensor_metadata_1431 = torch.ops.aten._assert_tensor_metadata.default(view_2487, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1431 = None
	        convert_element_type_953: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2487, torch.float32);  view_2487 = None
	        sub_7261: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_952, convert_element_type_953);  convert_element_type_952 = convert_element_type_953 = None
	        mul_15374: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7261, view_2486);  sub_7261 = view_2486 = None
	        view_2488: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15374, [1280, 1280]);  mul_15374 = None
	        _assert_tensor_metadata_1432 = torch.ops.aten._assert_tensor_metadata.default(view_2488, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1432 = None
	        mul_15379: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2489: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2484, [mul_15379, 1280]);  view_2484 = mul_15379 = None
	        permute_265: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2488, [1, 0]);  view_2488 = None
	        addmm_131: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_26_self_attn_v_proj_bias, view_2489, permute_265);  model_audio_tower_layers_26_self_attn_v_proj_bias = view_2489 = permute_265 = None
	        view_2490: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_131, [sym_size_int, 1500, 1280]);  addmm_131 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2491: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2490, [sym_size_int, -1, 20, 64]);  view_2490 = None
	        permute_266: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2491, [0, 2, 1, 3]);  view_2491 = None
	        clone_212: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_266, memory_format = torch.contiguous_format);  permute_266 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_26 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_210, clone_211, clone_212, None, False, scale = 1.0);  clone_210 = clone_211 = clone_212 = None
	        getitem_210: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_26[0];  _scaled_dot_product_efficient_attention_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_267: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_210, [0, 2, 1, 3]);  getitem_210 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2492: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_267, [sym_size_int, 1500, -1]);  permute_267 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2493: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2492, [sym_size_int, 1500, 1280])
	        amin_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2493, [2])
	        amax_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2493, [2]);  view_2493 = None
	        full_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_159, full_318);  amin_159 = full_318 = None
	        full_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_159, full_319);  amax_159 = full_319 = None
	        sub_7279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_159, minimum_159);  maximum_159 = None
	        div_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7279, 255.0);  sub_7279 = None
	        clamp_min_477: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_318, 1.1920928955078125e-07);  div_318 = None
	        div_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_159, clamp_min_477);  minimum_159 = None
	        round_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_319);  div_319 = None
	        sub_7285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_319);  round_319 = None
	        clamp_min_478: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7285, -128);  sub_7285 = None
	        clamp_max_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_478, 127);  clamp_min_478 = None
	        _assert_tensor_metadata_1433 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_477, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1433 = None
	        _assert_tensor_metadata_1434 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_318, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1434 = None
	        convert_element_type_954: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_318, torch.int8);  clamp_max_318 = None
	        view_2494: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2492, [sym_size_int, 1500, 1280]);  view_2492 = None
	        view_2495: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_477, [sym_size_int, 1500, 1])
	        view_2496: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_954, [sym_size_int, 1500, 1])
	        reciprocal_159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2495);  view_2495 = None
	        mul_15449: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_159, 1.0);  reciprocal_159 = None
	        mul_15452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2494, mul_15449);  view_2494 = mul_15449 = None
	        round_320: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15452);  mul_15452 = None
	        add_24451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_320, view_2496);  round_320 = view_2496 = None
	        clamp_min_479: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24451, -128);  add_24451 = None
	        clamp_max_319: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_479, 127);  clamp_min_479 = None
	        view_2497: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_319, [sym_size_int, 1500, 1280]);  clamp_max_319 = None
	        _assert_tensor_metadata_1435 = torch.ops.aten._assert_tensor_metadata.default(view_2497, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1435 = None
	        convert_element_type_955: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2497, torch.int8);  view_2497 = None
	        view_2498: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_955, [sym_size_int, 1500, 1280]);  convert_element_type_955 = None
	        view_2499: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_477, [sym_size_int, 1500, 1]);  clamp_min_477 = None
	        view_2500: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_954, [sym_size_int, 1500, 1]);  convert_element_type_954 = None
	        _assert_tensor_metadata_1436 = torch.ops.aten._assert_tensor_metadata.default(view_2498, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1436 = None
	        convert_element_type_956: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2498, torch.float32);  view_2498 = None
	        _assert_tensor_metadata_1437 = torch.ops.aten._assert_tensor_metadata.default(view_2500, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1437 = None
	        convert_element_type_957: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2500, torch.float32);  view_2500 = None
	        sub_7305: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_956, convert_element_type_957);  convert_element_type_956 = convert_element_type_957 = None
	        mul_15474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7305, view_2499);  sub_7305 = view_2499 = None
	        view_2501: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15474, [sym_size_int, 1500, 1280]);  mul_15474 = None
	        _assert_tensor_metadata_1438 = torch.ops.aten._assert_tensor_metadata.default(view_2501, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1438 = None
	        view_2502: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2503: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2504: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1439 = torch.ops.aten._assert_tensor_metadata.default(view_2502, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1439 = None
	        convert_element_type_958: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2502, torch.float32);  view_2502 = None
	        _assert_tensor_metadata_1440 = torch.ops.aten._assert_tensor_metadata.default(view_2504, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1440 = None
	        convert_element_type_959: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2504, torch.float32);  view_2504 = None
	        sub_7309: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_958, convert_element_type_959);  convert_element_type_958 = convert_element_type_959 = None
	        mul_15479: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7309, view_2503);  sub_7309 = view_2503 = None
	        view_2505: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15479, [1280, 1280]);  mul_15479 = None
	        _assert_tensor_metadata_1441 = torch.ops.aten._assert_tensor_metadata.default(view_2505, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1441 = None
	        mul_15484: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2506: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2501, [mul_15484, 1280]);  view_2501 = mul_15484 = None
	        permute_268: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2505, [1, 0]);  view_2505 = None
	        addmm_132: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_26_self_attn_out_proj_bias, view_2506, permute_268);  model_audio_tower_layers_26_self_attn_out_proj_bias = view_2506 = permute_268 = None
	        view_2507: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_132, [sym_size_int, 1500, 1280]);  addmm_132 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_213: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2507);  view_2507 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_24514: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_23894, clone_213);  add_23894 = clone_213 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_24514, memory_format = torch.contiguous_format)
	        var_mean_53 = torch.ops.aten.var_mean.correction(clone_214, [2], correction = 0, keepdim = True)
	        getitem_214: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_53[0]
	        getitem_215: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_53[1];  var_mean_53 = None
	        add_24519: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_214, 1e-05);  getitem_214 = None
	        rsqrt_53: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_24519);  add_24519 = None
	        sub_7315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_214, getitem_215);  clone_214 = getitem_215 = None
	        mul_15495: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7315, rsqrt_53);  sub_7315 = rsqrt_53 = None
	        mul_15496: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15495, model_audio_tower_layers_26_final_layer_norm_weight);  mul_15495 = model_audio_tower_layers_26_final_layer_norm_weight = None
	        add_24520: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_15496, model_audio_tower_layers_26_final_layer_norm_bias);  mul_15496 = model_audio_tower_layers_26_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2508: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_24520, [sym_size_int, 1500, 1280])
	        amin_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2508, [2])
	        amax_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2508, [2]);  view_2508 = None
	        full_320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_160, full_320);  amin_160 = full_320 = None
	        full_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_160, full_321);  amax_160 = full_321 = None
	        sub_7326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_160, minimum_160);  maximum_160 = None
	        div_320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7326, 255.0);  sub_7326 = None
	        clamp_min_480: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_320, 1.1920928955078125e-07);  div_320 = None
	        div_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_160, clamp_min_480);  minimum_160 = None
	        round_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_321);  div_321 = None
	        sub_7332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_321);  round_321 = None
	        clamp_min_481: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7332, -128);  sub_7332 = None
	        clamp_max_320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_481, 127);  clamp_min_481 = None
	        _assert_tensor_metadata_1442 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_480, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1442 = None
	        _assert_tensor_metadata_1443 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_320, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1443 = None
	        convert_element_type_960: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_320, torch.int8);  clamp_max_320 = None
	        view_2509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_24520, [sym_size_int, 1500, 1280]);  add_24520 = None
	        view_2510: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_480, [sym_size_int, 1500, 1])
	        view_2511: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_960, [sym_size_int, 1500, 1])
	        reciprocal_160: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2510);  view_2510 = None
	        mul_15544: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_160, 1.0);  reciprocal_160 = None
	        mul_15547: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2509, mul_15544);  view_2509 = mul_15544 = None
	        round_322: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15547);  mul_15547 = None
	        add_24607: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_322, view_2511);  round_322 = view_2511 = None
	        clamp_min_482: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24607, -128);  add_24607 = None
	        clamp_max_321: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_482, 127);  clamp_min_482 = None
	        view_2512: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_321, [sym_size_int, 1500, 1280]);  clamp_max_321 = None
	        _assert_tensor_metadata_1444 = torch.ops.aten._assert_tensor_metadata.default(view_2512, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1444 = None
	        convert_element_type_961: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2512, torch.int8);  view_2512 = None
	        view_2513: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_961, [sym_size_int, 1500, 1280]);  convert_element_type_961 = None
	        view_2514: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_480, [sym_size_int, 1500, 1]);  clamp_min_480 = None
	        view_2515: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_960, [sym_size_int, 1500, 1]);  convert_element_type_960 = None
	        _assert_tensor_metadata_1445 = torch.ops.aten._assert_tensor_metadata.default(view_2513, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1445 = None
	        convert_element_type_962: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2513, torch.float32);  view_2513 = None
	        _assert_tensor_metadata_1446 = torch.ops.aten._assert_tensor_metadata.default(view_2515, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1446 = None
	        convert_element_type_963: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2515, torch.float32);  view_2515 = None
	        sub_7352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_962, convert_element_type_963);  convert_element_type_962 = convert_element_type_963 = None
	        mul_15569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7352, view_2514);  sub_7352 = view_2514 = None
	        view_2516: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15569, [sym_size_int, 1500, 1280]);  mul_15569 = None
	        _assert_tensor_metadata_1447 = torch.ops.aten._assert_tensor_metadata.default(view_2516, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1447 = None
	        view_2517: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_26_fc1_parametrizations_weight_original0 = None
	        view_2518: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_26_fc1_parametrizations_weight_original1 = None
	        view_2519: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_26_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1448 = torch.ops.aten._assert_tensor_metadata.default(view_2517, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1448 = None
	        convert_element_type_964: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2517, torch.float32);  view_2517 = None
	        _assert_tensor_metadata_1449 = torch.ops.aten._assert_tensor_metadata.default(view_2519, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1449 = None
	        convert_element_type_965: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2519, torch.float32);  view_2519 = None
	        sub_7356: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_964, convert_element_type_965);  convert_element_type_964 = convert_element_type_965 = None
	        mul_15574: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7356, view_2518);  sub_7356 = view_2518 = None
	        view_2520: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15574, [5120, 1280]);  mul_15574 = None
	        _assert_tensor_metadata_1450 = torch.ops.aten._assert_tensor_metadata.default(view_2520, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1450 = None
	        mul_15579: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2521: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2516, [mul_15579, 1280]);  view_2516 = mul_15579 = None
	        permute_269: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2520, [1, 0]);  view_2520 = None
	        addmm_133: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_26_fc1_bias, view_2521, permute_269);  model_audio_tower_layers_26_fc1_bias = view_2521 = permute_269 = None
	        view_2522: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_133, [sym_size_int, 1500, 5120]);  addmm_133 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_15586: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2522, 0.5)
	        mul_15587: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2522, 0.7071067811865476);  view_2522 = None
	        erf_28: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_15587);  mul_15587 = None
	        add_24666: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_28, 1);  erf_28 = None
	        mul_15588: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15586, add_24666);  mul_15586 = add_24666 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_215: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_15588);  mul_15588 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2523: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_215, [sym_size_int, 1500, 5120])
	        amin_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2523, [2])
	        amax_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2523, [2]);  view_2523 = None
	        full_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_161, full_322);  amin_161 = full_322 = None
	        full_323: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_161, full_323);  amax_161 = full_323 = None
	        sub_7369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_161, minimum_161);  maximum_161 = None
	        div_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7369, 255.0);  sub_7369 = None
	        clamp_min_483: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_322, 1.1920928955078125e-07);  div_322 = None
	        div_323: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_161, clamp_min_483);  minimum_161 = None
	        round_323: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_323);  div_323 = None
	        sub_7375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_323);  round_323 = None
	        clamp_min_484: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7375, -128);  sub_7375 = None
	        clamp_max_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_484, 127);  clamp_min_484 = None
	        _assert_tensor_metadata_1451 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_483, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1451 = None
	        _assert_tensor_metadata_1452 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_322, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1452 = None
	        convert_element_type_966: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_322, torch.int8);  clamp_max_322 = None
	        view_2524: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_215, [sym_size_int, 1500, 5120]);  clone_215 = None
	        view_2525: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_483, [sym_size_int, 1500, 1])
	        view_2526: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_966, [sym_size_int, 1500, 1])
	        reciprocal_161: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2525);  view_2525 = None
	        mul_15634: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_161, 1.0);  reciprocal_161 = None
	        mul_15637: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2524, mul_15634);  view_2524 = mul_15634 = None
	        round_324: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_15637);  mul_15637 = None
	        add_24749: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_324, view_2526);  round_324 = view_2526 = None
	        clamp_min_485: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24749, -128);  add_24749 = None
	        clamp_max_323: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_485, 127);  clamp_min_485 = None
	        view_2527: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_323, [sym_size_int, 1500, 5120]);  clamp_max_323 = None
	        _assert_tensor_metadata_1453 = torch.ops.aten._assert_tensor_metadata.default(view_2527, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1453 = None
	        convert_element_type_967: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2527, torch.int8);  view_2527 = None
	        view_2528: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_967, [sym_size_int, 1500, 5120]);  convert_element_type_967 = None
	        view_2529: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_483, [sym_size_int, 1500, 1]);  clamp_min_483 = None
	        view_2530: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_966, [sym_size_int, 1500, 1]);  convert_element_type_966 = None
	        _assert_tensor_metadata_1454 = torch.ops.aten._assert_tensor_metadata.default(view_2528, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1454 = None
	        convert_element_type_968: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2528, torch.float32);  view_2528 = None
	        _assert_tensor_metadata_1455 = torch.ops.aten._assert_tensor_metadata.default(view_2530, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1455 = None
	        convert_element_type_969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2530, torch.float32);  view_2530 = None
	        sub_7395: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_968, convert_element_type_969);  convert_element_type_968 = convert_element_type_969 = None
	        mul_15659: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7395, view_2529);  sub_7395 = view_2529 = None
	        view_2531: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_15659, [sym_size_int, 1500, 5120]);  mul_15659 = None
	        _assert_tensor_metadata_1456 = torch.ops.aten._assert_tensor_metadata.default(view_2531, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1456 = None
	        view_2532: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_26_fc2_parametrizations_weight_original0 = None
	        view_2533: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_26_fc2_parametrizations_weight_original1 = None
	        view_2534: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_26_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1457 = torch.ops.aten._assert_tensor_metadata.default(view_2532, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1457 = None
	        convert_element_type_970: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2532, torch.float32);  view_2532 = None
	        _assert_tensor_metadata_1458 = torch.ops.aten._assert_tensor_metadata.default(view_2534, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1458 = None
	        convert_element_type_971: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2534, torch.float32);  view_2534 = None
	        sub_7399: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_970, convert_element_type_971);  convert_element_type_970 = convert_element_type_971 = None
	        mul_15664: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7399, view_2533);  sub_7399 = view_2533 = None
	        view_2535: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_15664, [1280, 5120]);  mul_15664 = None
	        _assert_tensor_metadata_1459 = torch.ops.aten._assert_tensor_metadata.default(view_2535, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1459 = None
	        mul_15669: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2536: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2531, [mul_15669, 5120]);  view_2531 = mul_15669 = None
	        permute_270: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2535, [1, 0]);  view_2535 = None
	        addmm_134: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_26_fc2_bias, view_2536, permute_270);  model_audio_tower_layers_26_fc2_bias = view_2536 = permute_270 = None
	        view_2537: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_134, [sym_size_int, 1500, 1280]);  addmm_134 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_216: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2537);  view_2537 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_24812: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_24514, clone_216);  add_24514 = clone_216 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_217: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_24812, memory_format = torch.contiguous_format)
	        var_mean_54 = torch.ops.aten.var_mean.correction(clone_217, [2], correction = 0, keepdim = True)
	        getitem_216: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_54[0]
	        getitem_217: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_54[1];  var_mean_54 = None
	        add_24817: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_216, 1e-05);  getitem_216 = None
	        rsqrt_54: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_24817);  add_24817 = None
	        sub_7405: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_217, getitem_217);  clone_217 = getitem_217 = None
	        mul_15680: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7405, rsqrt_54);  sub_7405 = rsqrt_54 = None
	        mul_15681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15680, model_audio_tower_layers_27_self_attn_layer_norm_weight);  mul_15680 = model_audio_tower_layers_27_self_attn_layer_norm_weight = None
	        add_24818: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_15681, model_audio_tower_layers_27_self_attn_layer_norm_bias);  mul_15681 = model_audio_tower_layers_27_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2538: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_24818, [sym_size_int, 1500, 1280])
	        amin_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2538, [2])
	        amax_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2538, [2]);  view_2538 = None
	        full_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_162, full_324);  amin_162 = full_324 = None
	        full_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_162, full_325);  amax_162 = full_325 = None
	        sub_7416: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_162, minimum_162);  maximum_162 = None
	        div_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7416, 255.0);  sub_7416 = None
	        clamp_min_486: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_324, 1.1920928955078125e-07);  div_324 = None
	        div_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_162, clamp_min_486);  minimum_162 = None
	        round_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_325);  div_325 = None
	        sub_7422: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_325);  round_325 = None
	        clamp_min_487: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7422, -128);  sub_7422 = None
	        clamp_max_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_487, 127);  clamp_min_487 = None
	        _assert_tensor_metadata_1460 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_486, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1460 = None
	        _assert_tensor_metadata_1461 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_324, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1461 = None
	        convert_element_type_972: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_324, torch.int8);  clamp_max_324 = None
	        view_2539: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_24818, [sym_size_int, 1500, 1280])
	        view_2540: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_486, [sym_size_int, 1500, 1])
	        view_2541: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_972, [sym_size_int, 1500, 1])
	        reciprocal_162: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2540);  view_2540 = None
	        mul_15729: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_162, 1.0);  reciprocal_162 = None
	        mul_15732: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2539, mul_15729);  view_2539 = mul_15729 = None
	        round_326: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15732);  mul_15732 = None
	        add_24905: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_326, view_2541);  round_326 = view_2541 = None
	        clamp_min_488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24905, -128);  add_24905 = None
	        clamp_max_325: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_488, 127);  clamp_min_488 = None
	        view_2542: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_325, [sym_size_int, 1500, 1280]);  clamp_max_325 = None
	        _assert_tensor_metadata_1462 = torch.ops.aten._assert_tensor_metadata.default(view_2542, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1462 = None
	        convert_element_type_973: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2542, torch.int8);  view_2542 = None
	        view_2543: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_973, [sym_size_int, 1500, 1280]);  convert_element_type_973 = None
	        view_2544: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_486, [sym_size_int, 1500, 1]);  clamp_min_486 = None
	        view_2545: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_972, [sym_size_int, 1500, 1]);  convert_element_type_972 = None
	        _assert_tensor_metadata_1463 = torch.ops.aten._assert_tensor_metadata.default(view_2543, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1463 = None
	        convert_element_type_974: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2543, torch.float32);  view_2543 = None
	        _assert_tensor_metadata_1464 = torch.ops.aten._assert_tensor_metadata.default(view_2545, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1464 = None
	        convert_element_type_975: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2545, torch.float32);  view_2545 = None
	        sub_7442: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_974, convert_element_type_975);  convert_element_type_974 = convert_element_type_975 = None
	        mul_15754: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7442, view_2544);  sub_7442 = view_2544 = None
	        view_2546: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15754, [sym_size_int, 1500, 1280]);  mul_15754 = None
	        _assert_tensor_metadata_1465 = torch.ops.aten._assert_tensor_metadata.default(view_2546, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1465 = None
	        view_2547: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2548: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2549: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1466 = torch.ops.aten._assert_tensor_metadata.default(view_2547, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1466 = None
	        convert_element_type_976: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2547, torch.float32);  view_2547 = None
	        _assert_tensor_metadata_1467 = torch.ops.aten._assert_tensor_metadata.default(view_2549, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1467 = None
	        convert_element_type_977: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2549, torch.float32);  view_2549 = None
	        sub_7446: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_976, convert_element_type_977);  convert_element_type_976 = convert_element_type_977 = None
	        mul_15759: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7446, view_2548);  sub_7446 = view_2548 = None
	        view_2550: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15759, [1280, 1280]);  mul_15759 = None
	        _assert_tensor_metadata_1468 = torch.ops.aten._assert_tensor_metadata.default(view_2550, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1468 = None
	        mul_15764: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2551: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2546, [mul_15764, 1280]);  view_2546 = mul_15764 = None
	        permute_271: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2550, [1, 0]);  view_2550 = None
	        addmm_135: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_27_self_attn_q_proj_bias, view_2551, permute_271);  model_audio_tower_layers_27_self_attn_q_proj_bias = view_2551 = permute_271 = None
	        view_2552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_135, [sym_size_int, 1500, 1280]);  addmm_135 = None
	        mul_15771: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2552, 0.125);  view_2552 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2553: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_15771, [sym_size_int, 1500, 20, 64]);  mul_15771 = None
	        permute_272: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2553, [0, 2, 1, 3]);  view_2553 = None
	        clone_218: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_272, memory_format = torch.contiguous_format);  permute_272 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2554: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_24818, [sym_size_int, 1500, 1280])
	        amin_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2554, [2])
	        amax_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2554, [2]);  view_2554 = None
	        full_326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_163, full_326);  amin_163 = full_326 = None
	        full_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_163, full_327);  amax_163 = full_327 = None
	        sub_7461: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_163, minimum_163);  maximum_163 = None
	        div_326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7461, 255.0);  sub_7461 = None
	        clamp_min_489: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_326, 1.1920928955078125e-07);  div_326 = None
	        div_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_163, clamp_min_489);  minimum_163 = None
	        round_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_327);  div_327 = None
	        sub_7467: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_327);  round_327 = None
	        clamp_min_490: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7467, -128);  sub_7467 = None
	        clamp_max_326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_490, 127);  clamp_min_490 = None
	        _assert_tensor_metadata_1469 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_489, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1469 = None
	        _assert_tensor_metadata_1470 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_326, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1470 = None
	        convert_element_type_978: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_326, torch.int8);  clamp_max_326 = None
	        view_2555: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_24818, [sym_size_int, 1500, 1280])
	        view_2556: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_489, [sym_size_int, 1500, 1])
	        view_2557: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_978, [sym_size_int, 1500, 1])
	        reciprocal_163: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2556);  view_2556 = None
	        mul_15825: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_163, 1.0);  reciprocal_163 = None
	        mul_15828: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2555, mul_15825);  view_2555 = mul_15825 = None
	        round_328: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15828);  mul_15828 = None
	        add_25057: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_328, view_2557);  round_328 = view_2557 = None
	        clamp_min_491: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25057, -128);  add_25057 = None
	        clamp_max_327: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_491, 127);  clamp_min_491 = None
	        view_2558: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_327, [sym_size_int, 1500, 1280]);  clamp_max_327 = None
	        _assert_tensor_metadata_1471 = torch.ops.aten._assert_tensor_metadata.default(view_2558, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1471 = None
	        convert_element_type_979: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2558, torch.int8);  view_2558 = None
	        view_2559: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_979, [sym_size_int, 1500, 1280]);  convert_element_type_979 = None
	        view_2560: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_489, [sym_size_int, 1500, 1]);  clamp_min_489 = None
	        view_2561: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_978, [sym_size_int, 1500, 1]);  convert_element_type_978 = None
	        _assert_tensor_metadata_1472 = torch.ops.aten._assert_tensor_metadata.default(view_2559, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1472 = None
	        convert_element_type_980: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2559, torch.float32);  view_2559 = None
	        _assert_tensor_metadata_1473 = torch.ops.aten._assert_tensor_metadata.default(view_2561, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1473 = None
	        convert_element_type_981: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2561, torch.float32);  view_2561 = None
	        sub_7487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_980, convert_element_type_981);  convert_element_type_980 = convert_element_type_981 = None
	        mul_15850: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7487, view_2560);  sub_7487 = view_2560 = None
	        view_2562: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15850, [sym_size_int, 1500, 1280]);  mul_15850 = None
	        _assert_tensor_metadata_1474 = torch.ops.aten._assert_tensor_metadata.default(view_2562, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1474 = None
	        view_2563: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2564: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2565: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1475 = torch.ops.aten._assert_tensor_metadata.default(view_2563, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1475 = None
	        convert_element_type_982: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2563, torch.float32);  view_2563 = None
	        _assert_tensor_metadata_1476 = torch.ops.aten._assert_tensor_metadata.default(view_2565, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1476 = None
	        convert_element_type_983: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2565, torch.float32);  view_2565 = None
	        sub_7491: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_982, convert_element_type_983);  convert_element_type_982 = convert_element_type_983 = None
	        mul_15855: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7491, view_2564);  sub_7491 = view_2564 = None
	        view_2566: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15855, [1280, 1280]);  mul_15855 = None
	        _assert_tensor_metadata_1477 = torch.ops.aten._assert_tensor_metadata.default(view_2566, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1477 = None
	        permute_273: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2566, [1, 0]);  view_2566 = None
	        mul_15858: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2567: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2562, [mul_15858, 1280]);  view_2562 = mul_15858 = None
	        mm_27: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2567, permute_273);  view_2567 = permute_273 = None
	        view_2568: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_27, [sym_size_int, 1500, 1280]);  mm_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2569: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2568, [sym_size_int, -1, 20, 64]);  view_2568 = None
	        permute_274: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2569, [0, 2, 1, 3]);  view_2569 = None
	        clone_219: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_274, memory_format = torch.contiguous_format);  permute_274 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2570: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_24818, [sym_size_int, 1500, 1280])
	        amin_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2570, [2])
	        amax_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2570, [2]);  view_2570 = None
	        full_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_164, full_328);  amin_164 = full_328 = None
	        full_329: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_164, full_329);  amax_164 = full_329 = None
	        sub_7505: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_164, minimum_164);  maximum_164 = None
	        div_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7505, 255.0);  sub_7505 = None
	        clamp_min_492: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_328, 1.1920928955078125e-07);  div_328 = None
	        div_329: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_164, clamp_min_492);  minimum_164 = None
	        round_329: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_329);  div_329 = None
	        sub_7511: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_329);  round_329 = None
	        clamp_min_493: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7511, -128);  sub_7511 = None
	        clamp_max_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_493, 127);  clamp_min_493 = None
	        _assert_tensor_metadata_1478 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_492, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1478 = None
	        _assert_tensor_metadata_1479 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_328, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1479 = None
	        convert_element_type_984: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_328, torch.int8);  clamp_max_328 = None
	        view_2571: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_24818, [sym_size_int, 1500, 1280]);  add_24818 = None
	        view_2572: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_492, [sym_size_int, 1500, 1])
	        view_2573: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_984, [sym_size_int, 1500, 1])
	        reciprocal_164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2572);  view_2572 = None
	        mul_15924: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_164, 1.0);  reciprocal_164 = None
	        mul_15927: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2571, mul_15924);  view_2571 = mul_15924 = None
	        round_330: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15927);  mul_15927 = None
	        add_25205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_330, view_2573);  round_330 = view_2573 = None
	        clamp_min_494: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25205, -128);  add_25205 = None
	        clamp_max_329: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_494, 127);  clamp_min_494 = None
	        view_2574: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_329, [sym_size_int, 1500, 1280]);  clamp_max_329 = None
	        _assert_tensor_metadata_1480 = torch.ops.aten._assert_tensor_metadata.default(view_2574, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1480 = None
	        convert_element_type_985: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2574, torch.int8);  view_2574 = None
	        view_2575: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_985, [sym_size_int, 1500, 1280]);  convert_element_type_985 = None
	        view_2576: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_492, [sym_size_int, 1500, 1]);  clamp_min_492 = None
	        view_2577: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_984, [sym_size_int, 1500, 1]);  convert_element_type_984 = None
	        _assert_tensor_metadata_1481 = torch.ops.aten._assert_tensor_metadata.default(view_2575, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1481 = None
	        convert_element_type_986: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2575, torch.float32);  view_2575 = None
	        _assert_tensor_metadata_1482 = torch.ops.aten._assert_tensor_metadata.default(view_2577, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1482 = None
	        convert_element_type_987: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2577, torch.float32);  view_2577 = None
	        sub_7531: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_986, convert_element_type_987);  convert_element_type_986 = convert_element_type_987 = None
	        mul_15949: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7531, view_2576);  sub_7531 = view_2576 = None
	        view_2578: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15949, [sym_size_int, 1500, 1280]);  mul_15949 = None
	        _assert_tensor_metadata_1483 = torch.ops.aten._assert_tensor_metadata.default(view_2578, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1483 = None
	        view_2579: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2580: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2581: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1484 = torch.ops.aten._assert_tensor_metadata.default(view_2579, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1484 = None
	        convert_element_type_988: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2579, torch.float32);  view_2579 = None
	        _assert_tensor_metadata_1485 = torch.ops.aten._assert_tensor_metadata.default(view_2581, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1485 = None
	        convert_element_type_989: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2581, torch.float32);  view_2581 = None
	        sub_7535: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_988, convert_element_type_989);  convert_element_type_988 = convert_element_type_989 = None
	        mul_15954: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7535, view_2580);  sub_7535 = view_2580 = None
	        view_2582: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15954, [1280, 1280]);  mul_15954 = None
	        _assert_tensor_metadata_1486 = torch.ops.aten._assert_tensor_metadata.default(view_2582, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1486 = None
	        mul_15959: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2583: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2578, [mul_15959, 1280]);  view_2578 = mul_15959 = None
	        permute_275: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2582, [1, 0]);  view_2582 = None
	        addmm_136: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_27_self_attn_v_proj_bias, view_2583, permute_275);  model_audio_tower_layers_27_self_attn_v_proj_bias = view_2583 = permute_275 = None
	        view_2584: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_136, [sym_size_int, 1500, 1280]);  addmm_136 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2585: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2584, [sym_size_int, -1, 20, 64]);  view_2584 = None
	        permute_276: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2585, [0, 2, 1, 3]);  view_2585 = None
	        clone_220: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_276, memory_format = torch.contiguous_format);  permute_276 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_27 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_218, clone_219, clone_220, None, False, scale = 1.0);  clone_218 = clone_219 = clone_220 = None
	        getitem_218: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_27[0];  _scaled_dot_product_efficient_attention_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_277: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_218, [0, 2, 1, 3]);  getitem_218 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2586: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_277, [sym_size_int, 1500, -1]);  permute_277 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2587: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2586, [sym_size_int, 1500, 1280])
	        amin_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2587, [2])
	        amax_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2587, [2]);  view_2587 = None
	        full_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_165, full_330);  amin_165 = full_330 = None
	        full_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_165, full_331);  amax_165 = full_331 = None
	        sub_7553: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_165, minimum_165);  maximum_165 = None
	        div_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7553, 255.0);  sub_7553 = None
	        clamp_min_495: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_330, 1.1920928955078125e-07);  div_330 = None
	        div_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_165, clamp_min_495);  minimum_165 = None
	        round_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_331);  div_331 = None
	        sub_7559: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_331);  round_331 = None
	        clamp_min_496: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7559, -128);  sub_7559 = None
	        clamp_max_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_496, 127);  clamp_min_496 = None
	        _assert_tensor_metadata_1487 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_495, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1487 = None
	        _assert_tensor_metadata_1488 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_330, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1488 = None
	        convert_element_type_990: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_330, torch.int8);  clamp_max_330 = None
	        view_2588: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2586, [sym_size_int, 1500, 1280]);  view_2586 = None
	        view_2589: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_495, [sym_size_int, 1500, 1])
	        view_2590: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_990, [sym_size_int, 1500, 1])
	        reciprocal_165: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2589);  view_2589 = None
	        mul_16029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_165, 1.0);  reciprocal_165 = None
	        mul_16032: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2588, mul_16029);  view_2588 = mul_16029 = None
	        round_332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16032);  mul_16032 = None
	        add_25369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_332, view_2590);  round_332 = view_2590 = None
	        clamp_min_497: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25369, -128);  add_25369 = None
	        clamp_max_331: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_497, 127);  clamp_min_497 = None
	        view_2591: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_331, [sym_size_int, 1500, 1280]);  clamp_max_331 = None
	        _assert_tensor_metadata_1489 = torch.ops.aten._assert_tensor_metadata.default(view_2591, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1489 = None
	        convert_element_type_991: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2591, torch.int8);  view_2591 = None
	        view_2592: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_991, [sym_size_int, 1500, 1280]);  convert_element_type_991 = None
	        view_2593: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_495, [sym_size_int, 1500, 1]);  clamp_min_495 = None
	        view_2594: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_990, [sym_size_int, 1500, 1]);  convert_element_type_990 = None
	        _assert_tensor_metadata_1490 = torch.ops.aten._assert_tensor_metadata.default(view_2592, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1490 = None
	        convert_element_type_992: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2592, torch.float32);  view_2592 = None
	        _assert_tensor_metadata_1491 = torch.ops.aten._assert_tensor_metadata.default(view_2594, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1491 = None
	        convert_element_type_993: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2594, torch.float32);  view_2594 = None
	        sub_7579: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_992, convert_element_type_993);  convert_element_type_992 = convert_element_type_993 = None
	        mul_16054: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7579, view_2593);  sub_7579 = view_2593 = None
	        view_2595: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16054, [sym_size_int, 1500, 1280]);  mul_16054 = None
	        _assert_tensor_metadata_1492 = torch.ops.aten._assert_tensor_metadata.default(view_2595, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1492 = None
	        view_2596: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2597: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2598: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1493 = torch.ops.aten._assert_tensor_metadata.default(view_2596, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1493 = None
	        convert_element_type_994: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2596, torch.float32);  view_2596 = None
	        _assert_tensor_metadata_1494 = torch.ops.aten._assert_tensor_metadata.default(view_2598, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1494 = None
	        convert_element_type_995: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2598, torch.float32);  view_2598 = None
	        sub_7583: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_994, convert_element_type_995);  convert_element_type_994 = convert_element_type_995 = None
	        mul_16059: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7583, view_2597);  sub_7583 = view_2597 = None
	        view_2599: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16059, [1280, 1280]);  mul_16059 = None
	        _assert_tensor_metadata_1495 = torch.ops.aten._assert_tensor_metadata.default(view_2599, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1495 = None
	        mul_16064: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2600: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2595, [mul_16064, 1280]);  view_2595 = mul_16064 = None
	        permute_278: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2599, [1, 0]);  view_2599 = None
	        addmm_137: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_27_self_attn_out_proj_bias, view_2600, permute_278);  model_audio_tower_layers_27_self_attn_out_proj_bias = view_2600 = permute_278 = None
	        view_2601: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_137, [sym_size_int, 1500, 1280]);  addmm_137 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2601);  view_2601 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_25432: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_24812, clone_221);  add_24812 = clone_221 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_222: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_25432, memory_format = torch.contiguous_format)
	        var_mean_55 = torch.ops.aten.var_mean.correction(clone_222, [2], correction = 0, keepdim = True)
	        getitem_222: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_55[0]
	        getitem_223: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_55[1];  var_mean_55 = None
	        add_25437: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_222, 1e-05);  getitem_222 = None
	        rsqrt_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_25437);  add_25437 = None
	        sub_7589: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_222, getitem_223);  clone_222 = getitem_223 = None
	        mul_16075: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7589, rsqrt_55);  sub_7589 = rsqrt_55 = None
	        mul_16076: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16075, model_audio_tower_layers_27_final_layer_norm_weight);  mul_16075 = model_audio_tower_layers_27_final_layer_norm_weight = None
	        add_25438: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16076, model_audio_tower_layers_27_final_layer_norm_bias);  mul_16076 = model_audio_tower_layers_27_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2602: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_25438, [sym_size_int, 1500, 1280])
	        amin_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2602, [2])
	        amax_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2602, [2]);  view_2602 = None
	        full_332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_166, full_332);  amin_166 = full_332 = None
	        full_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_166, full_333);  amax_166 = full_333 = None
	        sub_7600: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_166, minimum_166);  maximum_166 = None
	        div_332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7600, 255.0);  sub_7600 = None
	        clamp_min_498: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_332, 1.1920928955078125e-07);  div_332 = None
	        div_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_166, clamp_min_498);  minimum_166 = None
	        round_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_333);  div_333 = None
	        sub_7606: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_333);  round_333 = None
	        clamp_min_499: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7606, -128);  sub_7606 = None
	        clamp_max_332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_499, 127);  clamp_min_499 = None
	        _assert_tensor_metadata_1496 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_498, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1496 = None
	        _assert_tensor_metadata_1497 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_332, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1497 = None
	        convert_element_type_996: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_332, torch.int8);  clamp_max_332 = None
	        view_2603: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_25438, [sym_size_int, 1500, 1280]);  add_25438 = None
	        view_2604: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_498, [sym_size_int, 1500, 1])
	        view_2605: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_996, [sym_size_int, 1500, 1])
	        reciprocal_166: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2604);  view_2604 = None
	        mul_16124: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_166, 1.0);  reciprocal_166 = None
	        mul_16127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2603, mul_16124);  view_2603 = mul_16124 = None
	        round_334: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16127);  mul_16127 = None
	        add_25525: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_334, view_2605);  round_334 = view_2605 = None
	        clamp_min_500: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25525, -128);  add_25525 = None
	        clamp_max_333: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_500, 127);  clamp_min_500 = None
	        view_2606: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_333, [sym_size_int, 1500, 1280]);  clamp_max_333 = None
	        _assert_tensor_metadata_1498 = torch.ops.aten._assert_tensor_metadata.default(view_2606, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1498 = None
	        convert_element_type_997: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2606, torch.int8);  view_2606 = None
	        view_2607: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_997, [sym_size_int, 1500, 1280]);  convert_element_type_997 = None
	        view_2608: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_498, [sym_size_int, 1500, 1]);  clamp_min_498 = None
	        view_2609: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_996, [sym_size_int, 1500, 1]);  convert_element_type_996 = None
	        _assert_tensor_metadata_1499 = torch.ops.aten._assert_tensor_metadata.default(view_2607, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1499 = None
	        convert_element_type_998: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2607, torch.float32);  view_2607 = None
	        _assert_tensor_metadata_1500 = torch.ops.aten._assert_tensor_metadata.default(view_2609, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1500 = None
	        convert_element_type_999: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2609, torch.float32);  view_2609 = None
	        sub_7626: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_998, convert_element_type_999);  convert_element_type_998 = convert_element_type_999 = None
	        mul_16149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7626, view_2608);  sub_7626 = view_2608 = None
	        view_2610: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16149, [sym_size_int, 1500, 1280]);  mul_16149 = None
	        _assert_tensor_metadata_1501 = torch.ops.aten._assert_tensor_metadata.default(view_2610, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1501 = None
	        view_2611: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_27_fc1_parametrizations_weight_original0 = None
	        view_2612: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_27_fc1_parametrizations_weight_original1 = None
	        view_2613: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_27_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1502 = torch.ops.aten._assert_tensor_metadata.default(view_2611, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1502 = None
	        convert_element_type_1000: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2611, torch.float32);  view_2611 = None
	        _assert_tensor_metadata_1503 = torch.ops.aten._assert_tensor_metadata.default(view_2613, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1503 = None
	        convert_element_type_1001: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2613, torch.float32);  view_2613 = None
	        sub_7630: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1000, convert_element_type_1001);  convert_element_type_1000 = convert_element_type_1001 = None
	        mul_16154: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7630, view_2612);  sub_7630 = view_2612 = None
	        view_2614: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16154, [5120, 1280]);  mul_16154 = None
	        _assert_tensor_metadata_1504 = torch.ops.aten._assert_tensor_metadata.default(view_2614, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1504 = None
	        mul_16159: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2615: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2610, [mul_16159, 1280]);  view_2610 = mul_16159 = None
	        permute_279: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2614, [1, 0]);  view_2614 = None
	        addmm_138: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_27_fc1_bias, view_2615, permute_279);  model_audio_tower_layers_27_fc1_bias = view_2615 = permute_279 = None
	        view_2616: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_138, [sym_size_int, 1500, 5120]);  addmm_138 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_16166: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2616, 0.5)
	        mul_16167: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2616, 0.7071067811865476);  view_2616 = None
	        erf_29: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_16167);  mul_16167 = None
	        add_25584: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_29, 1);  erf_29 = None
	        mul_16168: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16166, add_25584);  mul_16166 = add_25584 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_223: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_16168);  mul_16168 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2617: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_223, [sym_size_int, 1500, 5120])
	        amin_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2617, [2])
	        amax_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2617, [2]);  view_2617 = None
	        full_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_167, full_334);  amin_167 = full_334 = None
	        full_335: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_167, full_335);  amax_167 = full_335 = None
	        sub_7643: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_167, minimum_167);  maximum_167 = None
	        div_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7643, 255.0);  sub_7643 = None
	        clamp_min_501: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_334, 1.1920928955078125e-07);  div_334 = None
	        div_335: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_167, clamp_min_501);  minimum_167 = None
	        round_335: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_335);  div_335 = None
	        sub_7649: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_335);  round_335 = None
	        clamp_min_502: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7649, -128);  sub_7649 = None
	        clamp_max_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_502, 127);  clamp_min_502 = None
	        _assert_tensor_metadata_1505 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_501, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1505 = None
	        _assert_tensor_metadata_1506 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_334, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1506 = None
	        convert_element_type_1002: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_334, torch.int8);  clamp_max_334 = None
	        view_2618: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_223, [sym_size_int, 1500, 5120]);  clone_223 = None
	        view_2619: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_501, [sym_size_int, 1500, 1])
	        view_2620: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1002, [sym_size_int, 1500, 1])
	        reciprocal_167: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2619);  view_2619 = None
	        mul_16214: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_167, 1.0);  reciprocal_167 = None
	        mul_16217: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2618, mul_16214);  view_2618 = mul_16214 = None
	        round_336: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_16217);  mul_16217 = None
	        add_25667: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_336, view_2620);  round_336 = view_2620 = None
	        clamp_min_503: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25667, -128);  add_25667 = None
	        clamp_max_335: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_503, 127);  clamp_min_503 = None
	        view_2621: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_335, [sym_size_int, 1500, 5120]);  clamp_max_335 = None
	        _assert_tensor_metadata_1507 = torch.ops.aten._assert_tensor_metadata.default(view_2621, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1507 = None
	        convert_element_type_1003: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2621, torch.int8);  view_2621 = None
	        view_2622: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1003, [sym_size_int, 1500, 5120]);  convert_element_type_1003 = None
	        view_2623: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_501, [sym_size_int, 1500, 1]);  clamp_min_501 = None
	        view_2624: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1002, [sym_size_int, 1500, 1]);  convert_element_type_1002 = None
	        _assert_tensor_metadata_1508 = torch.ops.aten._assert_tensor_metadata.default(view_2622, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1508 = None
	        convert_element_type_1004: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2622, torch.float32);  view_2622 = None
	        _assert_tensor_metadata_1509 = torch.ops.aten._assert_tensor_metadata.default(view_2624, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1509 = None
	        convert_element_type_1005: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2624, torch.float32);  view_2624 = None
	        sub_7669: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1004, convert_element_type_1005);  convert_element_type_1004 = convert_element_type_1005 = None
	        mul_16239: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7669, view_2623);  sub_7669 = view_2623 = None
	        view_2625: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_16239, [sym_size_int, 1500, 5120]);  mul_16239 = None
	        _assert_tensor_metadata_1510 = torch.ops.aten._assert_tensor_metadata.default(view_2625, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1510 = None
	        view_2626: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_27_fc2_parametrizations_weight_original0 = None
	        view_2627: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_27_fc2_parametrizations_weight_original1 = None
	        view_2628: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_27_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1511 = torch.ops.aten._assert_tensor_metadata.default(view_2626, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1511 = None
	        convert_element_type_1006: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2626, torch.float32);  view_2626 = None
	        _assert_tensor_metadata_1512 = torch.ops.aten._assert_tensor_metadata.default(view_2628, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1512 = None
	        convert_element_type_1007: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2628, torch.float32);  view_2628 = None
	        sub_7673: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1006, convert_element_type_1007);  convert_element_type_1006 = convert_element_type_1007 = None
	        mul_16244: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7673, view_2627);  sub_7673 = view_2627 = None
	        view_2629: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_16244, [1280, 5120]);  mul_16244 = None
	        _assert_tensor_metadata_1513 = torch.ops.aten._assert_tensor_metadata.default(view_2629, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1513 = None
	        mul_16249: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2630: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2625, [mul_16249, 5120]);  view_2625 = mul_16249 = None
	        permute_280: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2629, [1, 0]);  view_2629 = None
	        addmm_139: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_27_fc2_bias, view_2630, permute_280);  model_audio_tower_layers_27_fc2_bias = view_2630 = permute_280 = None
	        view_2631: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_139, [sym_size_int, 1500, 1280]);  addmm_139 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2631);  view_2631 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_25730: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_25432, clone_224);  add_25432 = clone_224 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_225: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_25730, memory_format = torch.contiguous_format)
	        var_mean_56 = torch.ops.aten.var_mean.correction(clone_225, [2], correction = 0, keepdim = True)
	        getitem_224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_56[0]
	        getitem_225: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_56[1];  var_mean_56 = None
	        add_25735: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_224, 1e-05);  getitem_224 = None
	        rsqrt_56: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_25735);  add_25735 = None
	        sub_7679: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_225, getitem_225);  clone_225 = getitem_225 = None
	        mul_16260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7679, rsqrt_56);  sub_7679 = rsqrt_56 = None
	        mul_16261: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16260, model_audio_tower_layers_28_self_attn_layer_norm_weight);  mul_16260 = model_audio_tower_layers_28_self_attn_layer_norm_weight = None
	        add_25736: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16261, model_audio_tower_layers_28_self_attn_layer_norm_bias);  mul_16261 = model_audio_tower_layers_28_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2632: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_25736, [sym_size_int, 1500, 1280])
	        amin_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2632, [2])
	        amax_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2632, [2]);  view_2632 = None
	        full_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_168, full_336);  amin_168 = full_336 = None
	        full_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_168, full_337);  amax_168 = full_337 = None
	        sub_7690: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_168, minimum_168);  maximum_168 = None
	        div_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7690, 255.0);  sub_7690 = None
	        clamp_min_504: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_336, 1.1920928955078125e-07);  div_336 = None
	        div_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_168, clamp_min_504);  minimum_168 = None
	        round_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_337);  div_337 = None
	        sub_7696: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_337);  round_337 = None
	        clamp_min_505: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7696, -128);  sub_7696 = None
	        clamp_max_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_505, 127);  clamp_min_505 = None
	        _assert_tensor_metadata_1514 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_504, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1514 = None
	        _assert_tensor_metadata_1515 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_336, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1515 = None
	        convert_element_type_1008: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_336, torch.int8);  clamp_max_336 = None
	        view_2633: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_25736, [sym_size_int, 1500, 1280])
	        view_2634: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_504, [sym_size_int, 1500, 1])
	        view_2635: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1008, [sym_size_int, 1500, 1])
	        reciprocal_168: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2634);  view_2634 = None
	        mul_16309: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_168, 1.0);  reciprocal_168 = None
	        mul_16312: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2633, mul_16309);  view_2633 = mul_16309 = None
	        round_338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16312);  mul_16312 = None
	        add_25823: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_338, view_2635);  round_338 = view_2635 = None
	        clamp_min_506: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25823, -128);  add_25823 = None
	        clamp_max_337: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_506, 127);  clamp_min_506 = None
	        view_2636: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_337, [sym_size_int, 1500, 1280]);  clamp_max_337 = None
	        _assert_tensor_metadata_1516 = torch.ops.aten._assert_tensor_metadata.default(view_2636, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1516 = None
	        convert_element_type_1009: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2636, torch.int8);  view_2636 = None
	        view_2637: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1009, [sym_size_int, 1500, 1280]);  convert_element_type_1009 = None
	        view_2638: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_504, [sym_size_int, 1500, 1]);  clamp_min_504 = None
	        view_2639: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1008, [sym_size_int, 1500, 1]);  convert_element_type_1008 = None
	        _assert_tensor_metadata_1517 = torch.ops.aten._assert_tensor_metadata.default(view_2637, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1517 = None
	        convert_element_type_1010: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2637, torch.float32);  view_2637 = None
	        _assert_tensor_metadata_1518 = torch.ops.aten._assert_tensor_metadata.default(view_2639, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1518 = None
	        convert_element_type_1011: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2639, torch.float32);  view_2639 = None
	        sub_7716: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1010, convert_element_type_1011);  convert_element_type_1010 = convert_element_type_1011 = None
	        mul_16334: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7716, view_2638);  sub_7716 = view_2638 = None
	        view_2640: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16334, [sym_size_int, 1500, 1280]);  mul_16334 = None
	        _assert_tensor_metadata_1519 = torch.ops.aten._assert_tensor_metadata.default(view_2640, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1519 = None
	        view_2641: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2642: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2643: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1520 = torch.ops.aten._assert_tensor_metadata.default(view_2641, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1520 = None
	        convert_element_type_1012: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2641, torch.float32);  view_2641 = None
	        _assert_tensor_metadata_1521 = torch.ops.aten._assert_tensor_metadata.default(view_2643, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1521 = None
	        convert_element_type_1013: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2643, torch.float32);  view_2643 = None
	        sub_7720: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1012, convert_element_type_1013);  convert_element_type_1012 = convert_element_type_1013 = None
	        mul_16339: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7720, view_2642);  sub_7720 = view_2642 = None
	        view_2644: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16339, [1280, 1280]);  mul_16339 = None
	        _assert_tensor_metadata_1522 = torch.ops.aten._assert_tensor_metadata.default(view_2644, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1522 = None
	        mul_16344: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2645: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2640, [mul_16344, 1280]);  view_2640 = mul_16344 = None
	        permute_281: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2644, [1, 0]);  view_2644 = None
	        addmm_140: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_28_self_attn_q_proj_bias, view_2645, permute_281);  model_audio_tower_layers_28_self_attn_q_proj_bias = view_2645 = permute_281 = None
	        view_2646: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_140, [sym_size_int, 1500, 1280]);  addmm_140 = None
	        mul_16351: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2646, 0.125);  view_2646 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2647: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_16351, [sym_size_int, 1500, 20, 64]);  mul_16351 = None
	        permute_282: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2647, [0, 2, 1, 3]);  view_2647 = None
	        clone_226: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_282, memory_format = torch.contiguous_format);  permute_282 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2648: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_25736, [sym_size_int, 1500, 1280])
	        amin_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2648, [2])
	        amax_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2648, [2]);  view_2648 = None
	        full_338: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_169, full_338);  amin_169 = full_338 = None
	        full_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_169, full_339);  amax_169 = full_339 = None
	        sub_7735: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_169, minimum_169);  maximum_169 = None
	        div_338: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7735, 255.0);  sub_7735 = None
	        clamp_min_507: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_338, 1.1920928955078125e-07);  div_338 = None
	        div_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_169, clamp_min_507);  minimum_169 = None
	        round_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_339);  div_339 = None
	        sub_7741: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_339);  round_339 = None
	        clamp_min_508: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7741, -128);  sub_7741 = None
	        clamp_max_338: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_508, 127);  clamp_min_508 = None
	        _assert_tensor_metadata_1523 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_507, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1523 = None
	        _assert_tensor_metadata_1524 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_338, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1524 = None
	        convert_element_type_1014: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_338, torch.int8);  clamp_max_338 = None
	        view_2649: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_25736, [sym_size_int, 1500, 1280])
	        view_2650: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_507, [sym_size_int, 1500, 1])
	        view_2651: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1014, [sym_size_int, 1500, 1])
	        reciprocal_169: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2650);  view_2650 = None
	        mul_16405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_169, 1.0);  reciprocal_169 = None
	        mul_16408: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2649, mul_16405);  view_2649 = mul_16405 = None
	        round_340: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16408);  mul_16408 = None
	        add_25975: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_340, view_2651);  round_340 = view_2651 = None
	        clamp_min_509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25975, -128);  add_25975 = None
	        clamp_max_339: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_509, 127);  clamp_min_509 = None
	        view_2652: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_339, [sym_size_int, 1500, 1280]);  clamp_max_339 = None
	        _assert_tensor_metadata_1525 = torch.ops.aten._assert_tensor_metadata.default(view_2652, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1525 = None
	        convert_element_type_1015: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2652, torch.int8);  view_2652 = None
	        view_2653: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1015, [sym_size_int, 1500, 1280]);  convert_element_type_1015 = None
	        view_2654: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_507, [sym_size_int, 1500, 1]);  clamp_min_507 = None
	        view_2655: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1014, [sym_size_int, 1500, 1]);  convert_element_type_1014 = None
	        _assert_tensor_metadata_1526 = torch.ops.aten._assert_tensor_metadata.default(view_2653, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1526 = None
	        convert_element_type_1016: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2653, torch.float32);  view_2653 = None
	        _assert_tensor_metadata_1527 = torch.ops.aten._assert_tensor_metadata.default(view_2655, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1527 = None
	        convert_element_type_1017: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2655, torch.float32);  view_2655 = None
	        sub_7761: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1016, convert_element_type_1017);  convert_element_type_1016 = convert_element_type_1017 = None
	        mul_16430: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7761, view_2654);  sub_7761 = view_2654 = None
	        view_2656: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16430, [sym_size_int, 1500, 1280]);  mul_16430 = None
	        _assert_tensor_metadata_1528 = torch.ops.aten._assert_tensor_metadata.default(view_2656, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1528 = None
	        view_2657: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2658: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2659: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1529 = torch.ops.aten._assert_tensor_metadata.default(view_2657, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1529 = None
	        convert_element_type_1018: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2657, torch.float32);  view_2657 = None
	        _assert_tensor_metadata_1530 = torch.ops.aten._assert_tensor_metadata.default(view_2659, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1530 = None
	        convert_element_type_1019: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2659, torch.float32);  view_2659 = None
	        sub_7765: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1018, convert_element_type_1019);  convert_element_type_1018 = convert_element_type_1019 = None
	        mul_16435: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7765, view_2658);  sub_7765 = view_2658 = None
	        view_2660: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16435, [1280, 1280]);  mul_16435 = None
	        _assert_tensor_metadata_1531 = torch.ops.aten._assert_tensor_metadata.default(view_2660, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1531 = None
	        permute_283: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2660, [1, 0]);  view_2660 = None
	        mul_16438: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2661: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2656, [mul_16438, 1280]);  view_2656 = mul_16438 = None
	        mm_28: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2661, permute_283);  view_2661 = permute_283 = None
	        view_2662: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_28, [sym_size_int, 1500, 1280]);  mm_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2663: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2662, [sym_size_int, -1, 20, 64]);  view_2662 = None
	        permute_284: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2663, [0, 2, 1, 3]);  view_2663 = None
	        clone_227: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_284, memory_format = torch.contiguous_format);  permute_284 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2664: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_25736, [sym_size_int, 1500, 1280])
	        amin_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2664, [2])
	        amax_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2664, [2]);  view_2664 = None
	        full_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_170, full_340);  amin_170 = full_340 = None
	        full_341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_170, full_341);  amax_170 = full_341 = None
	        sub_7779: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_170, minimum_170);  maximum_170 = None
	        div_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7779, 255.0);  sub_7779 = None
	        clamp_min_510: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_340, 1.1920928955078125e-07);  div_340 = None
	        div_341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_170, clamp_min_510);  minimum_170 = None
	        round_341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_341);  div_341 = None
	        sub_7785: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_341);  round_341 = None
	        clamp_min_511: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7785, -128);  sub_7785 = None
	        clamp_max_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_511, 127);  clamp_min_511 = None
	        _assert_tensor_metadata_1532 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_510, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1532 = None
	        _assert_tensor_metadata_1533 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_340, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1533 = None
	        convert_element_type_1020: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_340, torch.int8);  clamp_max_340 = None
	        view_2665: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_25736, [sym_size_int, 1500, 1280]);  add_25736 = None
	        view_2666: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_510, [sym_size_int, 1500, 1])
	        view_2667: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1020, [sym_size_int, 1500, 1])
	        reciprocal_170: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2666);  view_2666 = None
	        mul_16504: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_170, 1.0);  reciprocal_170 = None
	        mul_16507: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2665, mul_16504);  view_2665 = mul_16504 = None
	        round_342: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16507);  mul_16507 = None
	        add_26123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_342, view_2667);  round_342 = view_2667 = None
	        clamp_min_512: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26123, -128);  add_26123 = None
	        clamp_max_341: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_512, 127);  clamp_min_512 = None
	        view_2668: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_341, [sym_size_int, 1500, 1280]);  clamp_max_341 = None
	        _assert_tensor_metadata_1534 = torch.ops.aten._assert_tensor_metadata.default(view_2668, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1534 = None
	        convert_element_type_1021: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2668, torch.int8);  view_2668 = None
	        view_2669: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1021, [sym_size_int, 1500, 1280]);  convert_element_type_1021 = None
	        view_2670: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_510, [sym_size_int, 1500, 1]);  clamp_min_510 = None
	        view_2671: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1020, [sym_size_int, 1500, 1]);  convert_element_type_1020 = None
	        _assert_tensor_metadata_1535 = torch.ops.aten._assert_tensor_metadata.default(view_2669, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1535 = None
	        convert_element_type_1022: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2669, torch.float32);  view_2669 = None
	        _assert_tensor_metadata_1536 = torch.ops.aten._assert_tensor_metadata.default(view_2671, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1536 = None
	        convert_element_type_1023: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2671, torch.float32);  view_2671 = None
	        sub_7805: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1022, convert_element_type_1023);  convert_element_type_1022 = convert_element_type_1023 = None
	        mul_16529: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7805, view_2670);  sub_7805 = view_2670 = None
	        view_2672: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16529, [sym_size_int, 1500, 1280]);  mul_16529 = None
	        _assert_tensor_metadata_1537 = torch.ops.aten._assert_tensor_metadata.default(view_2672, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1537 = None
	        view_2673: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2674: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2675: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1538 = torch.ops.aten._assert_tensor_metadata.default(view_2673, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1538 = None
	        convert_element_type_1024: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2673, torch.float32);  view_2673 = None
	        _assert_tensor_metadata_1539 = torch.ops.aten._assert_tensor_metadata.default(view_2675, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1539 = None
	        convert_element_type_1025: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2675, torch.float32);  view_2675 = None
	        sub_7809: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1024, convert_element_type_1025);  convert_element_type_1024 = convert_element_type_1025 = None
	        mul_16534: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7809, view_2674);  sub_7809 = view_2674 = None
	        view_2676: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16534, [1280, 1280]);  mul_16534 = None
	        _assert_tensor_metadata_1540 = torch.ops.aten._assert_tensor_metadata.default(view_2676, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1540 = None
	        mul_16539: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2677: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2672, [mul_16539, 1280]);  view_2672 = mul_16539 = None
	        permute_285: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2676, [1, 0]);  view_2676 = None
	        addmm_141: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_28_self_attn_v_proj_bias, view_2677, permute_285);  model_audio_tower_layers_28_self_attn_v_proj_bias = view_2677 = permute_285 = None
	        view_2678: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_141, [sym_size_int, 1500, 1280]);  addmm_141 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2679: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2678, [sym_size_int, -1, 20, 64]);  view_2678 = None
	        permute_286: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2679, [0, 2, 1, 3]);  view_2679 = None
	        clone_228: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_286, memory_format = torch.contiguous_format);  permute_286 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_28 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_226, clone_227, clone_228, None, False, scale = 1.0);  clone_226 = clone_227 = clone_228 = None
	        getitem_226: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_28[0];  _scaled_dot_product_efficient_attention_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_287: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_226, [0, 2, 1, 3]);  getitem_226 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2680: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_287, [sym_size_int, 1500, -1]);  permute_287 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2680, [sym_size_int, 1500, 1280])
	        amin_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2681, [2])
	        amax_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2681, [2]);  view_2681 = None
	        full_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_171, full_342);  amin_171 = full_342 = None
	        full_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_171, full_343);  amax_171 = full_343 = None
	        sub_7827: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_171, minimum_171);  maximum_171 = None
	        div_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7827, 255.0);  sub_7827 = None
	        clamp_min_513: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_342, 1.1920928955078125e-07);  div_342 = None
	        div_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_171, clamp_min_513);  minimum_171 = None
	        round_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_343);  div_343 = None
	        sub_7833: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_343);  round_343 = None
	        clamp_min_514: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7833, -128);  sub_7833 = None
	        clamp_max_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_514, 127);  clamp_min_514 = None
	        _assert_tensor_metadata_1541 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_513, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1541 = None
	        _assert_tensor_metadata_1542 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_342, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1542 = None
	        convert_element_type_1026: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_342, torch.int8);  clamp_max_342 = None
	        view_2682: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2680, [sym_size_int, 1500, 1280]);  view_2680 = None
	        view_2683: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_513, [sym_size_int, 1500, 1])
	        view_2684: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1026, [sym_size_int, 1500, 1])
	        reciprocal_171: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2683);  view_2683 = None
	        mul_16609: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_171, 1.0);  reciprocal_171 = None
	        mul_16612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2682, mul_16609);  view_2682 = mul_16609 = None
	        round_344: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16612);  mul_16612 = None
	        add_26287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_344, view_2684);  round_344 = view_2684 = None
	        clamp_min_515: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26287, -128);  add_26287 = None
	        clamp_max_343: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_515, 127);  clamp_min_515 = None
	        view_2685: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_343, [sym_size_int, 1500, 1280]);  clamp_max_343 = None
	        _assert_tensor_metadata_1543 = torch.ops.aten._assert_tensor_metadata.default(view_2685, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1543 = None
	        convert_element_type_1027: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2685, torch.int8);  view_2685 = None
	        view_2686: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1027, [sym_size_int, 1500, 1280]);  convert_element_type_1027 = None
	        view_2687: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_513, [sym_size_int, 1500, 1]);  clamp_min_513 = None
	        view_2688: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1026, [sym_size_int, 1500, 1]);  convert_element_type_1026 = None
	        _assert_tensor_metadata_1544 = torch.ops.aten._assert_tensor_metadata.default(view_2686, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1544 = None
	        convert_element_type_1028: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2686, torch.float32);  view_2686 = None
	        _assert_tensor_metadata_1545 = torch.ops.aten._assert_tensor_metadata.default(view_2688, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1545 = None
	        convert_element_type_1029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2688, torch.float32);  view_2688 = None
	        sub_7853: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1028, convert_element_type_1029);  convert_element_type_1028 = convert_element_type_1029 = None
	        mul_16634: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7853, view_2687);  sub_7853 = view_2687 = None
	        view_2689: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16634, [sym_size_int, 1500, 1280]);  mul_16634 = None
	        _assert_tensor_metadata_1546 = torch.ops.aten._assert_tensor_metadata.default(view_2689, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1546 = None
	        view_2690: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2691: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2692: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1547 = torch.ops.aten._assert_tensor_metadata.default(view_2690, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1547 = None
	        convert_element_type_1030: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2690, torch.float32);  view_2690 = None
	        _assert_tensor_metadata_1548 = torch.ops.aten._assert_tensor_metadata.default(view_2692, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1548 = None
	        convert_element_type_1031: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2692, torch.float32);  view_2692 = None
	        sub_7857: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1030, convert_element_type_1031);  convert_element_type_1030 = convert_element_type_1031 = None
	        mul_16639: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7857, view_2691);  sub_7857 = view_2691 = None
	        view_2693: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16639, [1280, 1280]);  mul_16639 = None
	        _assert_tensor_metadata_1549 = torch.ops.aten._assert_tensor_metadata.default(view_2693, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1549 = None
	        mul_16644: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2694: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2689, [mul_16644, 1280]);  view_2689 = mul_16644 = None
	        permute_288: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2693, [1, 0]);  view_2693 = None
	        addmm_142: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_28_self_attn_out_proj_bias, view_2694, permute_288);  model_audio_tower_layers_28_self_attn_out_proj_bias = view_2694 = permute_288 = None
	        view_2695: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_142, [sym_size_int, 1500, 1280]);  addmm_142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_229: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2695);  view_2695 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_26350: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_25730, clone_229);  add_25730 = clone_229 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_26350, memory_format = torch.contiguous_format)
	        var_mean_57 = torch.ops.aten.var_mean.correction(clone_230, [2], correction = 0, keepdim = True)
	        getitem_230: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_57[0]
	        getitem_231: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_57[1];  var_mean_57 = None
	        add_26355: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_230, 1e-05);  getitem_230 = None
	        rsqrt_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_26355);  add_26355 = None
	        sub_7863: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_230, getitem_231);  clone_230 = getitem_231 = None
	        mul_16655: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7863, rsqrt_57);  sub_7863 = rsqrt_57 = None
	        mul_16656: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16655, model_audio_tower_layers_28_final_layer_norm_weight);  mul_16655 = model_audio_tower_layers_28_final_layer_norm_weight = None
	        add_26356: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16656, model_audio_tower_layers_28_final_layer_norm_bias);  mul_16656 = model_audio_tower_layers_28_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2696: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_26356, [sym_size_int, 1500, 1280])
	        amin_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2696, [2])
	        amax_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2696, [2]);  view_2696 = None
	        full_344: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_172, full_344);  amin_172 = full_344 = None
	        full_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_172, full_345);  amax_172 = full_345 = None
	        sub_7874: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_172, minimum_172);  maximum_172 = None
	        div_344: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7874, 255.0);  sub_7874 = None
	        clamp_min_516: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_344, 1.1920928955078125e-07);  div_344 = None
	        div_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_172, clamp_min_516);  minimum_172 = None
	        round_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_345);  div_345 = None
	        sub_7880: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_345);  round_345 = None
	        clamp_min_517: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7880, -128);  sub_7880 = None
	        clamp_max_344: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_517, 127);  clamp_min_517 = None
	        _assert_tensor_metadata_1550 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_516, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1550 = None
	        _assert_tensor_metadata_1551 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_344, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1551 = None
	        convert_element_type_1032: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_344, torch.int8);  clamp_max_344 = None
	        view_2697: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_26356, [sym_size_int, 1500, 1280]);  add_26356 = None
	        view_2698: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_516, [sym_size_int, 1500, 1])
	        view_2699: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1032, [sym_size_int, 1500, 1])
	        reciprocal_172: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2698);  view_2698 = None
	        mul_16704: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_172, 1.0);  reciprocal_172 = None
	        mul_16707: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2697, mul_16704);  view_2697 = mul_16704 = None
	        round_346: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16707);  mul_16707 = None
	        add_26443: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_346, view_2699);  round_346 = view_2699 = None
	        clamp_min_518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26443, -128);  add_26443 = None
	        clamp_max_345: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_518, 127);  clamp_min_518 = None
	        view_2700: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_345, [sym_size_int, 1500, 1280]);  clamp_max_345 = None
	        _assert_tensor_metadata_1552 = torch.ops.aten._assert_tensor_metadata.default(view_2700, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1552 = None
	        convert_element_type_1033: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2700, torch.int8);  view_2700 = None
	        view_2701: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1033, [sym_size_int, 1500, 1280]);  convert_element_type_1033 = None
	        view_2702: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_516, [sym_size_int, 1500, 1]);  clamp_min_516 = None
	        view_2703: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1032, [sym_size_int, 1500, 1]);  convert_element_type_1032 = None
	        _assert_tensor_metadata_1553 = torch.ops.aten._assert_tensor_metadata.default(view_2701, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1553 = None
	        convert_element_type_1034: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2701, torch.float32);  view_2701 = None
	        _assert_tensor_metadata_1554 = torch.ops.aten._assert_tensor_metadata.default(view_2703, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1554 = None
	        convert_element_type_1035: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2703, torch.float32);  view_2703 = None
	        sub_7900: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1034, convert_element_type_1035);  convert_element_type_1034 = convert_element_type_1035 = None
	        mul_16729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7900, view_2702);  sub_7900 = view_2702 = None
	        view_2704: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16729, [sym_size_int, 1500, 1280]);  mul_16729 = None
	        _assert_tensor_metadata_1555 = torch.ops.aten._assert_tensor_metadata.default(view_2704, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1555 = None
	        view_2705: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_28_fc1_parametrizations_weight_original0 = None
	        view_2706: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_28_fc1_parametrizations_weight_original1 = None
	        view_2707: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_28_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1556 = torch.ops.aten._assert_tensor_metadata.default(view_2705, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1556 = None
	        convert_element_type_1036: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2705, torch.float32);  view_2705 = None
	        _assert_tensor_metadata_1557 = torch.ops.aten._assert_tensor_metadata.default(view_2707, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1557 = None
	        convert_element_type_1037: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2707, torch.float32);  view_2707 = None
	        sub_7904: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1036, convert_element_type_1037);  convert_element_type_1036 = convert_element_type_1037 = None
	        mul_16734: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7904, view_2706);  sub_7904 = view_2706 = None
	        view_2708: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16734, [5120, 1280]);  mul_16734 = None
	        _assert_tensor_metadata_1558 = torch.ops.aten._assert_tensor_metadata.default(view_2708, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1558 = None
	        mul_16739: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2709: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2704, [mul_16739, 1280]);  view_2704 = mul_16739 = None
	        permute_289: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2708, [1, 0]);  view_2708 = None
	        addmm_143: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_28_fc1_bias, view_2709, permute_289);  model_audio_tower_layers_28_fc1_bias = view_2709 = permute_289 = None
	        view_2710: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_143, [sym_size_int, 1500, 5120]);  addmm_143 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_16746: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2710, 0.5)
	        mul_16747: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2710, 0.7071067811865476);  view_2710 = None
	        erf_30: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_16747);  mul_16747 = None
	        add_26502: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_30, 1);  erf_30 = None
	        mul_16748: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16746, add_26502);  mul_16746 = add_26502 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_231: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_16748);  mul_16748 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2711: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_231, [sym_size_int, 1500, 5120])
	        amin_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2711, [2])
	        amax_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2711, [2]);  view_2711 = None
	        full_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_173, full_346);  amin_173 = full_346 = None
	        full_347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_173, full_347);  amax_173 = full_347 = None
	        sub_7917: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_173, minimum_173);  maximum_173 = None
	        div_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7917, 255.0);  sub_7917 = None
	        clamp_min_519: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_346, 1.1920928955078125e-07);  div_346 = None
	        div_347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_173, clamp_min_519);  minimum_173 = None
	        round_347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_347);  div_347 = None
	        sub_7923: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_347);  round_347 = None
	        clamp_min_520: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7923, -128);  sub_7923 = None
	        clamp_max_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_520, 127);  clamp_min_520 = None
	        _assert_tensor_metadata_1559 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_519, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1559 = None
	        _assert_tensor_metadata_1560 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_346, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1560 = None
	        convert_element_type_1038: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_346, torch.int8);  clamp_max_346 = None
	        view_2712: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_231, [sym_size_int, 1500, 5120]);  clone_231 = None
	        view_2713: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_519, [sym_size_int, 1500, 1])
	        view_2714: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1038, [sym_size_int, 1500, 1])
	        reciprocal_173: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2713);  view_2713 = None
	        mul_16794: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_173, 1.0);  reciprocal_173 = None
	        mul_16797: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2712, mul_16794);  view_2712 = mul_16794 = None
	        round_348: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_16797);  mul_16797 = None
	        add_26585: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_348, view_2714);  round_348 = view_2714 = None
	        clamp_min_521: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26585, -128);  add_26585 = None
	        clamp_max_347: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_521, 127);  clamp_min_521 = None
	        view_2715: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_347, [sym_size_int, 1500, 5120]);  clamp_max_347 = None
	        _assert_tensor_metadata_1561 = torch.ops.aten._assert_tensor_metadata.default(view_2715, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1561 = None
	        convert_element_type_1039: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2715, torch.int8);  view_2715 = None
	        view_2716: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1039, [sym_size_int, 1500, 5120]);  convert_element_type_1039 = None
	        view_2717: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_519, [sym_size_int, 1500, 1]);  clamp_min_519 = None
	        view_2718: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1038, [sym_size_int, 1500, 1]);  convert_element_type_1038 = None
	        _assert_tensor_metadata_1562 = torch.ops.aten._assert_tensor_metadata.default(view_2716, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1562 = None
	        convert_element_type_1040: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2716, torch.float32);  view_2716 = None
	        _assert_tensor_metadata_1563 = torch.ops.aten._assert_tensor_metadata.default(view_2718, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1563 = None
	        convert_element_type_1041: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2718, torch.float32);  view_2718 = None
	        sub_7943: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1040, convert_element_type_1041);  convert_element_type_1040 = convert_element_type_1041 = None
	        mul_16819: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7943, view_2717);  sub_7943 = view_2717 = None
	        view_2719: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_16819, [sym_size_int, 1500, 5120]);  mul_16819 = None
	        _assert_tensor_metadata_1564 = torch.ops.aten._assert_tensor_metadata.default(view_2719, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1564 = None
	        view_2720: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_28_fc2_parametrizations_weight_original0 = None
	        view_2721: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_28_fc2_parametrizations_weight_original1 = None
	        view_2722: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_28_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1565 = torch.ops.aten._assert_tensor_metadata.default(view_2720, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1565 = None
	        convert_element_type_1042: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2720, torch.float32);  view_2720 = None
	        _assert_tensor_metadata_1566 = torch.ops.aten._assert_tensor_metadata.default(view_2722, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1566 = None
	        convert_element_type_1043: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2722, torch.float32);  view_2722 = None
	        sub_7947: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1042, convert_element_type_1043);  convert_element_type_1042 = convert_element_type_1043 = None
	        mul_16824: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7947, view_2721);  sub_7947 = view_2721 = None
	        view_2723: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_16824, [1280, 5120]);  mul_16824 = None
	        _assert_tensor_metadata_1567 = torch.ops.aten._assert_tensor_metadata.default(view_2723, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1567 = None
	        mul_16829: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2724: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2719, [mul_16829, 5120]);  view_2719 = mul_16829 = None
	        permute_290: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2723, [1, 0]);  view_2723 = None
	        addmm_144: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_28_fc2_bias, view_2724, permute_290);  model_audio_tower_layers_28_fc2_bias = view_2724 = permute_290 = None
	        view_2725: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_144, [sym_size_int, 1500, 1280]);  addmm_144 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_232: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2725);  view_2725 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_26648: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_26350, clone_232);  add_26350 = clone_232 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_233: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_26648, memory_format = torch.contiguous_format)
	        var_mean_58 = torch.ops.aten.var_mean.correction(clone_233, [2], correction = 0, keepdim = True)
	        getitem_232: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_58[0]
	        getitem_233: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_58[1];  var_mean_58 = None
	        add_26653: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_232, 1e-05);  getitem_232 = None
	        rsqrt_58: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_26653);  add_26653 = None
	        sub_7953: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_233, getitem_233);  clone_233 = getitem_233 = None
	        mul_16840: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7953, rsqrt_58);  sub_7953 = rsqrt_58 = None
	        mul_16841: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16840, model_audio_tower_layers_29_self_attn_layer_norm_weight);  mul_16840 = model_audio_tower_layers_29_self_attn_layer_norm_weight = None
	        add_26654: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16841, model_audio_tower_layers_29_self_attn_layer_norm_bias);  mul_16841 = model_audio_tower_layers_29_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2726: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_26654, [sym_size_int, 1500, 1280])
	        amin_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2726, [2])
	        amax_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2726, [2]);  view_2726 = None
	        full_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_174, full_348);  amin_174 = full_348 = None
	        full_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_174, full_349);  amax_174 = full_349 = None
	        sub_7964: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_174, minimum_174);  maximum_174 = None
	        div_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7964, 255.0);  sub_7964 = None
	        clamp_min_522: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_348, 1.1920928955078125e-07);  div_348 = None
	        div_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_174, clamp_min_522);  minimum_174 = None
	        round_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_349);  div_349 = None
	        sub_7970: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_349);  round_349 = None
	        clamp_min_523: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7970, -128);  sub_7970 = None
	        clamp_max_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_523, 127);  clamp_min_523 = None
	        _assert_tensor_metadata_1568 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_522, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1568 = None
	        _assert_tensor_metadata_1569 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_348, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1569 = None
	        convert_element_type_1044: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_348, torch.int8);  clamp_max_348 = None
	        view_2727: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_26654, [sym_size_int, 1500, 1280])
	        view_2728: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_522, [sym_size_int, 1500, 1])
	        view_2729: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1044, [sym_size_int, 1500, 1])
	        reciprocal_174: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2728);  view_2728 = None
	        mul_16889: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_174, 1.0);  reciprocal_174 = None
	        mul_16892: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2727, mul_16889);  view_2727 = mul_16889 = None
	        round_350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16892);  mul_16892 = None
	        add_26741: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_350, view_2729);  round_350 = view_2729 = None
	        clamp_min_524: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26741, -128);  add_26741 = None
	        clamp_max_349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_524, 127);  clamp_min_524 = None
	        view_2730: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_349, [sym_size_int, 1500, 1280]);  clamp_max_349 = None
	        _assert_tensor_metadata_1570 = torch.ops.aten._assert_tensor_metadata.default(view_2730, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1570 = None
	        convert_element_type_1045: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2730, torch.int8);  view_2730 = None
	        view_2731: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1045, [sym_size_int, 1500, 1280]);  convert_element_type_1045 = None
	        view_2732: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_522, [sym_size_int, 1500, 1]);  clamp_min_522 = None
	        view_2733: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1044, [sym_size_int, 1500, 1]);  convert_element_type_1044 = None
	        _assert_tensor_metadata_1571 = torch.ops.aten._assert_tensor_metadata.default(view_2731, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1571 = None
	        convert_element_type_1046: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2731, torch.float32);  view_2731 = None
	        _assert_tensor_metadata_1572 = torch.ops.aten._assert_tensor_metadata.default(view_2733, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1572 = None
	        convert_element_type_1047: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2733, torch.float32);  view_2733 = None
	        sub_7990: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1046, convert_element_type_1047);  convert_element_type_1046 = convert_element_type_1047 = None
	        mul_16914: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7990, view_2732);  sub_7990 = view_2732 = None
	        view_2734: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16914, [sym_size_int, 1500, 1280]);  mul_16914 = None
	        _assert_tensor_metadata_1573 = torch.ops.aten._assert_tensor_metadata.default(view_2734, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1573 = None
	        view_2735: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2736: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2737: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1574 = torch.ops.aten._assert_tensor_metadata.default(view_2735, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1574 = None
	        convert_element_type_1048: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2735, torch.float32);  view_2735 = None
	        _assert_tensor_metadata_1575 = torch.ops.aten._assert_tensor_metadata.default(view_2737, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1575 = None
	        convert_element_type_1049: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2737, torch.float32);  view_2737 = None
	        sub_7994: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1048, convert_element_type_1049);  convert_element_type_1048 = convert_element_type_1049 = None
	        mul_16919: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7994, view_2736);  sub_7994 = view_2736 = None
	        view_2738: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16919, [1280, 1280]);  mul_16919 = None
	        _assert_tensor_metadata_1576 = torch.ops.aten._assert_tensor_metadata.default(view_2738, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1576 = None
	        mul_16924: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2739: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2734, [mul_16924, 1280]);  view_2734 = mul_16924 = None
	        permute_291: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2738, [1, 0]);  view_2738 = None
	        addmm_145: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_29_self_attn_q_proj_bias, view_2739, permute_291);  model_audio_tower_layers_29_self_attn_q_proj_bias = view_2739 = permute_291 = None
	        view_2740: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_145, [sym_size_int, 1500, 1280]);  addmm_145 = None
	        mul_16931: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2740, 0.125);  view_2740 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2741: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_16931, [sym_size_int, 1500, 20, 64]);  mul_16931 = None
	        permute_292: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2741, [0, 2, 1, 3]);  view_2741 = None
	        clone_234: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_292, memory_format = torch.contiguous_format);  permute_292 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2742: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_26654, [sym_size_int, 1500, 1280])
	        amin_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2742, [2])
	        amax_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2742, [2]);  view_2742 = None
	        full_350: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_175, full_350);  amin_175 = full_350 = None
	        full_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_175, full_351);  amax_175 = full_351 = None
	        sub_8009: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_175, minimum_175);  maximum_175 = None
	        div_350: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8009, 255.0);  sub_8009 = None
	        clamp_min_525: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_350, 1.1920928955078125e-07);  div_350 = None
	        div_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_175, clamp_min_525);  minimum_175 = None
	        round_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_351);  div_351 = None
	        sub_8015: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_351);  round_351 = None
	        clamp_min_526: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8015, -128);  sub_8015 = None
	        clamp_max_350: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_526, 127);  clamp_min_526 = None
	        _assert_tensor_metadata_1577 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_525, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1577 = None
	        _assert_tensor_metadata_1578 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_350, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1578 = None
	        convert_element_type_1050: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_350, torch.int8);  clamp_max_350 = None
	        view_2743: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_26654, [sym_size_int, 1500, 1280])
	        view_2744: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_525, [sym_size_int, 1500, 1])
	        view_2745: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1050, [sym_size_int, 1500, 1])
	        reciprocal_175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2744);  view_2744 = None
	        mul_16985: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_175, 1.0);  reciprocal_175 = None
	        mul_16988: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2743, mul_16985);  view_2743 = mul_16985 = None
	        round_352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16988);  mul_16988 = None
	        add_26893: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_352, view_2745);  round_352 = view_2745 = None
	        clamp_min_527: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26893, -128);  add_26893 = None
	        clamp_max_351: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_527, 127);  clamp_min_527 = None
	        view_2746: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_351, [sym_size_int, 1500, 1280]);  clamp_max_351 = None
	        _assert_tensor_metadata_1579 = torch.ops.aten._assert_tensor_metadata.default(view_2746, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1579 = None
	        convert_element_type_1051: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2746, torch.int8);  view_2746 = None
	        view_2747: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1051, [sym_size_int, 1500, 1280]);  convert_element_type_1051 = None
	        view_2748: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_525, [sym_size_int, 1500, 1]);  clamp_min_525 = None
	        view_2749: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1050, [sym_size_int, 1500, 1]);  convert_element_type_1050 = None
	        _assert_tensor_metadata_1580 = torch.ops.aten._assert_tensor_metadata.default(view_2747, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1580 = None
	        convert_element_type_1052: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2747, torch.float32);  view_2747 = None
	        _assert_tensor_metadata_1581 = torch.ops.aten._assert_tensor_metadata.default(view_2749, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1581 = None
	        convert_element_type_1053: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2749, torch.float32);  view_2749 = None
	        sub_8035: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1052, convert_element_type_1053);  convert_element_type_1052 = convert_element_type_1053 = None
	        mul_17010: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8035, view_2748);  sub_8035 = view_2748 = None
	        view_2750: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17010, [sym_size_int, 1500, 1280]);  mul_17010 = None
	        _assert_tensor_metadata_1582 = torch.ops.aten._assert_tensor_metadata.default(view_2750, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1582 = None
	        view_2751: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2752: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2753: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1583 = torch.ops.aten._assert_tensor_metadata.default(view_2751, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1583 = None
	        convert_element_type_1054: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2751, torch.float32);  view_2751 = None
	        _assert_tensor_metadata_1584 = torch.ops.aten._assert_tensor_metadata.default(view_2753, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1584 = None
	        convert_element_type_1055: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2753, torch.float32);  view_2753 = None
	        sub_8039: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1054, convert_element_type_1055);  convert_element_type_1054 = convert_element_type_1055 = None
	        mul_17015: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8039, view_2752);  sub_8039 = view_2752 = None
	        view_2754: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17015, [1280, 1280]);  mul_17015 = None
	        _assert_tensor_metadata_1585 = torch.ops.aten._assert_tensor_metadata.default(view_2754, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1585 = None
	        permute_293: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2754, [1, 0]);  view_2754 = None
	        mul_17018: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2755: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2750, [mul_17018, 1280]);  view_2750 = mul_17018 = None
	        mm_29: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2755, permute_293);  view_2755 = permute_293 = None
	        view_2756: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_29, [sym_size_int, 1500, 1280]);  mm_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2757: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2756, [sym_size_int, -1, 20, 64]);  view_2756 = None
	        permute_294: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2757, [0, 2, 1, 3]);  view_2757 = None
	        clone_235: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_294, memory_format = torch.contiguous_format);  permute_294 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2758: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_26654, [sym_size_int, 1500, 1280])
	        amin_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2758, [2])
	        amax_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2758, [2]);  view_2758 = None
	        full_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_176, full_352);  amin_176 = full_352 = None
	        full_353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_176, full_353);  amax_176 = full_353 = None
	        sub_8053: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_176, minimum_176);  maximum_176 = None
	        div_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8053, 255.0);  sub_8053 = None
	        clamp_min_528: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_352, 1.1920928955078125e-07);  div_352 = None
	        div_353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_176, clamp_min_528);  minimum_176 = None
	        round_353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_353);  div_353 = None
	        sub_8059: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_353);  round_353 = None
	        clamp_min_529: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8059, -128);  sub_8059 = None
	        clamp_max_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_529, 127);  clamp_min_529 = None
	        _assert_tensor_metadata_1586 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_528, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1586 = None
	        _assert_tensor_metadata_1587 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_352, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1587 = None
	        convert_element_type_1056: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_352, torch.int8);  clamp_max_352 = None
	        view_2759: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_26654, [sym_size_int, 1500, 1280]);  add_26654 = None
	        view_2760: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_528, [sym_size_int, 1500, 1])
	        view_2761: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1056, [sym_size_int, 1500, 1])
	        reciprocal_176: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2760);  view_2760 = None
	        mul_17084: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_176, 1.0);  reciprocal_176 = None
	        mul_17087: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2759, mul_17084);  view_2759 = mul_17084 = None
	        round_354: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17087);  mul_17087 = None
	        add_27041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_354, view_2761);  round_354 = view_2761 = None
	        clamp_min_530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27041, -128);  add_27041 = None
	        clamp_max_353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_530, 127);  clamp_min_530 = None
	        view_2762: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_353, [sym_size_int, 1500, 1280]);  clamp_max_353 = None
	        _assert_tensor_metadata_1588 = torch.ops.aten._assert_tensor_metadata.default(view_2762, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1588 = None
	        convert_element_type_1057: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2762, torch.int8);  view_2762 = None
	        view_2763: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1057, [sym_size_int, 1500, 1280]);  convert_element_type_1057 = None
	        view_2764: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_528, [sym_size_int, 1500, 1]);  clamp_min_528 = None
	        view_2765: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1056, [sym_size_int, 1500, 1]);  convert_element_type_1056 = None
	        _assert_tensor_metadata_1589 = torch.ops.aten._assert_tensor_metadata.default(view_2763, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1589 = None
	        convert_element_type_1058: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2763, torch.float32);  view_2763 = None
	        _assert_tensor_metadata_1590 = torch.ops.aten._assert_tensor_metadata.default(view_2765, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1590 = None
	        convert_element_type_1059: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2765, torch.float32);  view_2765 = None
	        sub_8079: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1058, convert_element_type_1059);  convert_element_type_1058 = convert_element_type_1059 = None
	        mul_17109: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8079, view_2764);  sub_8079 = view_2764 = None
	        view_2766: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17109, [sym_size_int, 1500, 1280]);  mul_17109 = None
	        _assert_tensor_metadata_1591 = torch.ops.aten._assert_tensor_metadata.default(view_2766, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1591 = None
	        view_2767: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2768: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2769: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1592 = torch.ops.aten._assert_tensor_metadata.default(view_2767, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1592 = None
	        convert_element_type_1060: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2767, torch.float32);  view_2767 = None
	        _assert_tensor_metadata_1593 = torch.ops.aten._assert_tensor_metadata.default(view_2769, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1593 = None
	        convert_element_type_1061: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2769, torch.float32);  view_2769 = None
	        sub_8083: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1060, convert_element_type_1061);  convert_element_type_1060 = convert_element_type_1061 = None
	        mul_17114: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8083, view_2768);  sub_8083 = view_2768 = None
	        view_2770: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17114, [1280, 1280]);  mul_17114 = None
	        _assert_tensor_metadata_1594 = torch.ops.aten._assert_tensor_metadata.default(view_2770, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1594 = None
	        mul_17119: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2771: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2766, [mul_17119, 1280]);  view_2766 = mul_17119 = None
	        permute_295: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2770, [1, 0]);  view_2770 = None
	        addmm_146: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_29_self_attn_v_proj_bias, view_2771, permute_295);  model_audio_tower_layers_29_self_attn_v_proj_bias = view_2771 = permute_295 = None
	        view_2772: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_146, [sym_size_int, 1500, 1280]);  addmm_146 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2773: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2772, [sym_size_int, -1, 20, 64]);  view_2772 = None
	        permute_296: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2773, [0, 2, 1, 3]);  view_2773 = None
	        clone_236: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_296, memory_format = torch.contiguous_format);  permute_296 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_29 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_234, clone_235, clone_236, None, False, scale = 1.0);  clone_234 = clone_235 = clone_236 = None
	        getitem_234: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_29[0];  _scaled_dot_product_efficient_attention_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_297: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_234, [0, 2, 1, 3]);  getitem_234 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2774: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_297, [sym_size_int, 1500, -1]);  permute_297 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2775: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2774, [sym_size_int, 1500, 1280])
	        amin_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2775, [2])
	        amax_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2775, [2]);  view_2775 = None
	        full_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_177, full_354);  amin_177 = full_354 = None
	        full_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_177, full_355);  amax_177 = full_355 = None
	        sub_8101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_177, minimum_177);  maximum_177 = None
	        div_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8101, 255.0);  sub_8101 = None
	        clamp_min_531: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_354, 1.1920928955078125e-07);  div_354 = None
	        div_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_177, clamp_min_531);  minimum_177 = None
	        round_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_355);  div_355 = None
	        sub_8107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_355);  round_355 = None
	        clamp_min_532: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8107, -128);  sub_8107 = None
	        clamp_max_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_532, 127);  clamp_min_532 = None
	        _assert_tensor_metadata_1595 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_531, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1595 = None
	        _assert_tensor_metadata_1596 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_354, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1596 = None
	        convert_element_type_1062: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_354, torch.int8);  clamp_max_354 = None
	        view_2776: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2774, [sym_size_int, 1500, 1280]);  view_2774 = None
	        view_2777: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_531, [sym_size_int, 1500, 1])
	        view_2778: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1062, [sym_size_int, 1500, 1])
	        reciprocal_177: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2777);  view_2777 = None
	        mul_17189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_177, 1.0);  reciprocal_177 = None
	        mul_17192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2776, mul_17189);  view_2776 = mul_17189 = None
	        round_356: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17192);  mul_17192 = None
	        add_27205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_356, view_2778);  round_356 = view_2778 = None
	        clamp_min_533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27205, -128);  add_27205 = None
	        clamp_max_355: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_533, 127);  clamp_min_533 = None
	        view_2779: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_355, [sym_size_int, 1500, 1280]);  clamp_max_355 = None
	        _assert_tensor_metadata_1597 = torch.ops.aten._assert_tensor_metadata.default(view_2779, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1597 = None
	        convert_element_type_1063: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2779, torch.int8);  view_2779 = None
	        view_2780: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1063, [sym_size_int, 1500, 1280]);  convert_element_type_1063 = None
	        view_2781: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_531, [sym_size_int, 1500, 1]);  clamp_min_531 = None
	        view_2782: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1062, [sym_size_int, 1500, 1]);  convert_element_type_1062 = None
	        _assert_tensor_metadata_1598 = torch.ops.aten._assert_tensor_metadata.default(view_2780, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1598 = None
	        convert_element_type_1064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2780, torch.float32);  view_2780 = None
	        _assert_tensor_metadata_1599 = torch.ops.aten._assert_tensor_metadata.default(view_2782, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1599 = None
	        convert_element_type_1065: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2782, torch.float32);  view_2782 = None
	        sub_8127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1064, convert_element_type_1065);  convert_element_type_1064 = convert_element_type_1065 = None
	        mul_17214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8127, view_2781);  sub_8127 = view_2781 = None
	        view_2783: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17214, [sym_size_int, 1500, 1280]);  mul_17214 = None
	        _assert_tensor_metadata_1600 = torch.ops.aten._assert_tensor_metadata.default(view_2783, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1600 = None
	        view_2784: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2785: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2786: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1601 = torch.ops.aten._assert_tensor_metadata.default(view_2784, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1601 = None
	        convert_element_type_1066: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2784, torch.float32);  view_2784 = None
	        _assert_tensor_metadata_1602 = torch.ops.aten._assert_tensor_metadata.default(view_2786, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1602 = None
	        convert_element_type_1067: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2786, torch.float32);  view_2786 = None
	        sub_8131: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1066, convert_element_type_1067);  convert_element_type_1066 = convert_element_type_1067 = None
	        mul_17219: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8131, view_2785);  sub_8131 = view_2785 = None
	        view_2787: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17219, [1280, 1280]);  mul_17219 = None
	        _assert_tensor_metadata_1603 = torch.ops.aten._assert_tensor_metadata.default(view_2787, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1603 = None
	        mul_17224: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2788: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2783, [mul_17224, 1280]);  view_2783 = mul_17224 = None
	        permute_298: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2787, [1, 0]);  view_2787 = None
	        addmm_147: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_29_self_attn_out_proj_bias, view_2788, permute_298);  model_audio_tower_layers_29_self_attn_out_proj_bias = view_2788 = permute_298 = None
	        view_2789: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_147, [sym_size_int, 1500, 1280]);  addmm_147 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_237: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2789);  view_2789 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_27268: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_26648, clone_237);  add_26648 = clone_237 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_238: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_27268, memory_format = torch.contiguous_format)
	        var_mean_59 = torch.ops.aten.var_mean.correction(clone_238, [2], correction = 0, keepdim = True)
	        getitem_238: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_59[0]
	        getitem_239: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_59[1];  var_mean_59 = None
	        add_27273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_238, 1e-05);  getitem_238 = None
	        rsqrt_59: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_27273);  add_27273 = None
	        sub_8137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_238, getitem_239);  clone_238 = getitem_239 = None
	        mul_17235: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8137, rsqrt_59);  sub_8137 = rsqrt_59 = None
	        mul_17236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17235, model_audio_tower_layers_29_final_layer_norm_weight);  mul_17235 = model_audio_tower_layers_29_final_layer_norm_weight = None
	        add_27274: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_17236, model_audio_tower_layers_29_final_layer_norm_bias);  mul_17236 = model_audio_tower_layers_29_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2790: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_27274, [sym_size_int, 1500, 1280])
	        amin_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2790, [2])
	        amax_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2790, [2]);  view_2790 = None
	        full_356: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_178, full_356);  amin_178 = full_356 = None
	        full_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_178, full_357);  amax_178 = full_357 = None
	        sub_8148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_178, minimum_178);  maximum_178 = None
	        div_356: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8148, 255.0);  sub_8148 = None
	        clamp_min_534: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_356, 1.1920928955078125e-07);  div_356 = None
	        div_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_178, clamp_min_534);  minimum_178 = None
	        round_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_357);  div_357 = None
	        sub_8154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_357);  round_357 = None
	        clamp_min_535: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8154, -128);  sub_8154 = None
	        clamp_max_356: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_535, 127);  clamp_min_535 = None
	        _assert_tensor_metadata_1604 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_534, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1604 = None
	        _assert_tensor_metadata_1605 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_356, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1605 = None
	        convert_element_type_1068: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_356, torch.int8);  clamp_max_356 = None
	        view_2791: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_27274, [sym_size_int, 1500, 1280]);  add_27274 = None
	        view_2792: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_534, [sym_size_int, 1500, 1])
	        view_2793: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1068, [sym_size_int, 1500, 1])
	        reciprocal_178: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2792);  view_2792 = None
	        mul_17284: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_178, 1.0);  reciprocal_178 = None
	        mul_17287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2791, mul_17284);  view_2791 = mul_17284 = None
	        round_358: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17287);  mul_17287 = None
	        add_27361: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_358, view_2793);  round_358 = view_2793 = None
	        clamp_min_536: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27361, -128);  add_27361 = None
	        clamp_max_357: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_536, 127);  clamp_min_536 = None
	        view_2794: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_357, [sym_size_int, 1500, 1280]);  clamp_max_357 = None
	        _assert_tensor_metadata_1606 = torch.ops.aten._assert_tensor_metadata.default(view_2794, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1606 = None
	        convert_element_type_1069: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2794, torch.int8);  view_2794 = None
	        view_2795: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1069, [sym_size_int, 1500, 1280]);  convert_element_type_1069 = None
	        view_2796: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_534, [sym_size_int, 1500, 1]);  clamp_min_534 = None
	        view_2797: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1068, [sym_size_int, 1500, 1]);  convert_element_type_1068 = None
	        _assert_tensor_metadata_1607 = torch.ops.aten._assert_tensor_metadata.default(view_2795, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1607 = None
	        convert_element_type_1070: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2795, torch.float32);  view_2795 = None
	        _assert_tensor_metadata_1608 = torch.ops.aten._assert_tensor_metadata.default(view_2797, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1608 = None
	        convert_element_type_1071: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2797, torch.float32);  view_2797 = None
	        sub_8174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1070, convert_element_type_1071);  convert_element_type_1070 = convert_element_type_1071 = None
	        mul_17309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8174, view_2796);  sub_8174 = view_2796 = None
	        view_2798: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17309, [sym_size_int, 1500, 1280]);  mul_17309 = None
	        _assert_tensor_metadata_1609 = torch.ops.aten._assert_tensor_metadata.default(view_2798, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1609 = None
	        view_2799: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_29_fc1_parametrizations_weight_original0 = None
	        view_2800: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_29_fc1_parametrizations_weight_original1 = None
	        view_2801: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_29_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1610 = torch.ops.aten._assert_tensor_metadata.default(view_2799, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1610 = None
	        convert_element_type_1072: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2799, torch.float32);  view_2799 = None
	        _assert_tensor_metadata_1611 = torch.ops.aten._assert_tensor_metadata.default(view_2801, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1611 = None
	        convert_element_type_1073: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2801, torch.float32);  view_2801 = None
	        sub_8178: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1072, convert_element_type_1073);  convert_element_type_1072 = convert_element_type_1073 = None
	        mul_17314: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8178, view_2800);  sub_8178 = view_2800 = None
	        view_2802: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17314, [5120, 1280]);  mul_17314 = None
	        _assert_tensor_metadata_1612 = torch.ops.aten._assert_tensor_metadata.default(view_2802, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1612 = None
	        mul_17319: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2803: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2798, [mul_17319, 1280]);  view_2798 = mul_17319 = None
	        permute_299: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2802, [1, 0]);  view_2802 = None
	        addmm_148: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_29_fc1_bias, view_2803, permute_299);  model_audio_tower_layers_29_fc1_bias = view_2803 = permute_299 = None
	        view_2804: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_148, [sym_size_int, 1500, 5120]);  addmm_148 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_17326: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2804, 0.5)
	        mul_17327: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2804, 0.7071067811865476);  view_2804 = None
	        erf_31: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_17327);  mul_17327 = None
	        add_27420: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_31, 1);  erf_31 = None
	        mul_17328: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17326, add_27420);  mul_17326 = add_27420 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_239: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_17328);  mul_17328 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2805: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_239, [sym_size_int, 1500, 5120])
	        amin_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2805, [2])
	        amax_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2805, [2]);  view_2805 = None
	        full_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_179, full_358);  amin_179 = full_358 = None
	        full_359: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_179, full_359);  amax_179 = full_359 = None
	        sub_8191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_179, minimum_179);  maximum_179 = None
	        div_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8191, 255.0);  sub_8191 = None
	        clamp_min_537: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_358, 1.1920928955078125e-07);  div_358 = None
	        div_359: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_179, clamp_min_537);  minimum_179 = None
	        round_359: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_359);  div_359 = None
	        sub_8197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_359);  round_359 = None
	        clamp_min_538: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8197, -128);  sub_8197 = None
	        clamp_max_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_538, 127);  clamp_min_538 = None
	        _assert_tensor_metadata_1613 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_537, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1613 = None
	        _assert_tensor_metadata_1614 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_358, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1614 = None
	        convert_element_type_1074: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_358, torch.int8);  clamp_max_358 = None
	        view_2806: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_239, [sym_size_int, 1500, 5120]);  clone_239 = None
	        view_2807: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_537, [sym_size_int, 1500, 1])
	        view_2808: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1074, [sym_size_int, 1500, 1])
	        reciprocal_179: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2807);  view_2807 = None
	        mul_17374: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_179, 1.0);  reciprocal_179 = None
	        mul_17377: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2806, mul_17374);  view_2806 = mul_17374 = None
	        round_360: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_17377);  mul_17377 = None
	        add_27503: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_360, view_2808);  round_360 = view_2808 = None
	        clamp_min_539: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27503, -128);  add_27503 = None
	        clamp_max_359: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_539, 127);  clamp_min_539 = None
	        view_2809: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_359, [sym_size_int, 1500, 5120]);  clamp_max_359 = None
	        _assert_tensor_metadata_1615 = torch.ops.aten._assert_tensor_metadata.default(view_2809, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1615 = None
	        convert_element_type_1075: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2809, torch.int8);  view_2809 = None
	        view_2810: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1075, [sym_size_int, 1500, 5120]);  convert_element_type_1075 = None
	        view_2811: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_537, [sym_size_int, 1500, 1]);  clamp_min_537 = None
	        view_2812: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1074, [sym_size_int, 1500, 1]);  convert_element_type_1074 = None
	        _assert_tensor_metadata_1616 = torch.ops.aten._assert_tensor_metadata.default(view_2810, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1616 = None
	        convert_element_type_1076: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2810, torch.float32);  view_2810 = None
	        _assert_tensor_metadata_1617 = torch.ops.aten._assert_tensor_metadata.default(view_2812, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1617 = None
	        convert_element_type_1077: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2812, torch.float32);  view_2812 = None
	        sub_8217: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1076, convert_element_type_1077);  convert_element_type_1076 = convert_element_type_1077 = None
	        mul_17399: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8217, view_2811);  sub_8217 = view_2811 = None
	        view_2813: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_17399, [sym_size_int, 1500, 5120]);  mul_17399 = None
	        _assert_tensor_metadata_1618 = torch.ops.aten._assert_tensor_metadata.default(view_2813, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1618 = None
	        view_2814: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_29_fc2_parametrizations_weight_original0 = None
	        view_2815: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_29_fc2_parametrizations_weight_original1 = None
	        view_2816: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_29_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1619 = torch.ops.aten._assert_tensor_metadata.default(view_2814, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1619 = None
	        convert_element_type_1078: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2814, torch.float32);  view_2814 = None
	        _assert_tensor_metadata_1620 = torch.ops.aten._assert_tensor_metadata.default(view_2816, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1620 = None
	        convert_element_type_1079: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2816, torch.float32);  view_2816 = None
	        sub_8221: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1078, convert_element_type_1079);  convert_element_type_1078 = convert_element_type_1079 = None
	        mul_17404: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8221, view_2815);  sub_8221 = view_2815 = None
	        view_2817: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_17404, [1280, 5120]);  mul_17404 = None
	        _assert_tensor_metadata_1621 = torch.ops.aten._assert_tensor_metadata.default(view_2817, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1621 = None
	        mul_17409: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2818: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2813, [mul_17409, 5120]);  view_2813 = mul_17409 = None
	        permute_300: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2817, [1, 0]);  view_2817 = None
	        addmm_149: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_29_fc2_bias, view_2818, permute_300);  model_audio_tower_layers_29_fc2_bias = view_2818 = permute_300 = None
	        view_2819: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_149, [sym_size_int, 1500, 1280]);  addmm_149 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_240: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2819);  view_2819 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_27566: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_27268, clone_240);  add_27268 = clone_240 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_241: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_27566, memory_format = torch.contiguous_format)
	        var_mean_60 = torch.ops.aten.var_mean.correction(clone_241, [2], correction = 0, keepdim = True)
	        getitem_240: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_60[0]
	        getitem_241: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_60[1];  var_mean_60 = None
	        add_27571: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_240, 1e-05);  getitem_240 = None
	        rsqrt_60: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_27571);  add_27571 = None
	        sub_8227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_241, getitem_241);  clone_241 = getitem_241 = None
	        mul_17420: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8227, rsqrt_60);  sub_8227 = rsqrt_60 = None
	        mul_17421: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17420, model_audio_tower_layers_30_self_attn_layer_norm_weight);  mul_17420 = model_audio_tower_layers_30_self_attn_layer_norm_weight = None
	        add_27572: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_17421, model_audio_tower_layers_30_self_attn_layer_norm_bias);  mul_17421 = model_audio_tower_layers_30_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2820: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_27572, [sym_size_int, 1500, 1280])
	        amin_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2820, [2])
	        amax_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2820, [2]);  view_2820 = None
	        full_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_180, full_360);  amin_180 = full_360 = None
	        full_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_180, full_361);  amax_180 = full_361 = None
	        sub_8238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_180, minimum_180);  maximum_180 = None
	        div_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8238, 255.0);  sub_8238 = None
	        clamp_min_540: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_360, 1.1920928955078125e-07);  div_360 = None
	        div_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_180, clamp_min_540);  minimum_180 = None
	        round_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_361);  div_361 = None
	        sub_8244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_361);  round_361 = None
	        clamp_min_541: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8244, -128);  sub_8244 = None
	        clamp_max_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_541, 127);  clamp_min_541 = None
	        _assert_tensor_metadata_1622 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_540, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1622 = None
	        _assert_tensor_metadata_1623 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_360, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1623 = None
	        convert_element_type_1080: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_360, torch.int8);  clamp_max_360 = None
	        view_2821: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_27572, [sym_size_int, 1500, 1280])
	        view_2822: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_540, [sym_size_int, 1500, 1])
	        view_2823: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1080, [sym_size_int, 1500, 1])
	        reciprocal_180: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2822);  view_2822 = None
	        mul_17469: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_180, 1.0);  reciprocal_180 = None
	        mul_17472: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2821, mul_17469);  view_2821 = mul_17469 = None
	        round_362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17472);  mul_17472 = None
	        add_27659: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_362, view_2823);  round_362 = view_2823 = None
	        clamp_min_542: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27659, -128);  add_27659 = None
	        clamp_max_361: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_542, 127);  clamp_min_542 = None
	        view_2824: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_361, [sym_size_int, 1500, 1280]);  clamp_max_361 = None
	        _assert_tensor_metadata_1624 = torch.ops.aten._assert_tensor_metadata.default(view_2824, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1624 = None
	        convert_element_type_1081: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2824, torch.int8);  view_2824 = None
	        view_2825: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1081, [sym_size_int, 1500, 1280]);  convert_element_type_1081 = None
	        view_2826: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_540, [sym_size_int, 1500, 1]);  clamp_min_540 = None
	        view_2827: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1080, [sym_size_int, 1500, 1]);  convert_element_type_1080 = None
	        _assert_tensor_metadata_1625 = torch.ops.aten._assert_tensor_metadata.default(view_2825, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1625 = None
	        convert_element_type_1082: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2825, torch.float32);  view_2825 = None
	        _assert_tensor_metadata_1626 = torch.ops.aten._assert_tensor_metadata.default(view_2827, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1626 = None
	        convert_element_type_1083: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2827, torch.float32);  view_2827 = None
	        sub_8264: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1082, convert_element_type_1083);  convert_element_type_1082 = convert_element_type_1083 = None
	        mul_17494: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8264, view_2826);  sub_8264 = view_2826 = None
	        view_2828: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17494, [sym_size_int, 1500, 1280]);  mul_17494 = None
	        _assert_tensor_metadata_1627 = torch.ops.aten._assert_tensor_metadata.default(view_2828, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1627 = None
	        view_2829: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2830: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2831: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1628 = torch.ops.aten._assert_tensor_metadata.default(view_2829, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1628 = None
	        convert_element_type_1084: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2829, torch.float32);  view_2829 = None
	        _assert_tensor_metadata_1629 = torch.ops.aten._assert_tensor_metadata.default(view_2831, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1629 = None
	        convert_element_type_1085: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2831, torch.float32);  view_2831 = None
	        sub_8268: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1084, convert_element_type_1085);  convert_element_type_1084 = convert_element_type_1085 = None
	        mul_17499: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8268, view_2830);  sub_8268 = view_2830 = None
	        view_2832: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17499, [1280, 1280]);  mul_17499 = None
	        _assert_tensor_metadata_1630 = torch.ops.aten._assert_tensor_metadata.default(view_2832, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1630 = None
	        mul_17504: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2833: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2828, [mul_17504, 1280]);  view_2828 = mul_17504 = None
	        permute_301: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2832, [1, 0]);  view_2832 = None
	        addmm_150: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_30_self_attn_q_proj_bias, view_2833, permute_301);  model_audio_tower_layers_30_self_attn_q_proj_bias = view_2833 = permute_301 = None
	        view_2834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_150, [sym_size_int, 1500, 1280]);  addmm_150 = None
	        mul_17511: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2834, 0.125);  view_2834 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2835: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_17511, [sym_size_int, 1500, 20, 64]);  mul_17511 = None
	        permute_302: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2835, [0, 2, 1, 3]);  view_2835 = None
	        clone_242: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_302, memory_format = torch.contiguous_format);  permute_302 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2836: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_27572, [sym_size_int, 1500, 1280])
	        amin_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2836, [2])
	        amax_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2836, [2]);  view_2836 = None
	        full_362: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_181, full_362);  amin_181 = full_362 = None
	        full_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_181, full_363);  amax_181 = full_363 = None
	        sub_8283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_181, minimum_181);  maximum_181 = None
	        div_362: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8283, 255.0);  sub_8283 = None
	        clamp_min_543: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_362, 1.1920928955078125e-07);  div_362 = None
	        div_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_181, clamp_min_543);  minimum_181 = None
	        round_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_363);  div_363 = None
	        sub_8289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_363);  round_363 = None
	        clamp_min_544: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8289, -128);  sub_8289 = None
	        clamp_max_362: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_544, 127);  clamp_min_544 = None
	        _assert_tensor_metadata_1631 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_543, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1631 = None
	        _assert_tensor_metadata_1632 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_362, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1632 = None
	        convert_element_type_1086: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_362, torch.int8);  clamp_max_362 = None
	        view_2837: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_27572, [sym_size_int, 1500, 1280])
	        view_2838: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_543, [sym_size_int, 1500, 1])
	        view_2839: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1086, [sym_size_int, 1500, 1])
	        reciprocal_181: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2838);  view_2838 = None
	        mul_17565: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_181, 1.0);  reciprocal_181 = None
	        mul_17568: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2837, mul_17565);  view_2837 = mul_17565 = None
	        round_364: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17568);  mul_17568 = None
	        add_27811: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_364, view_2839);  round_364 = view_2839 = None
	        clamp_min_545: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27811, -128);  add_27811 = None
	        clamp_max_363: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_545, 127);  clamp_min_545 = None
	        view_2840: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_363, [sym_size_int, 1500, 1280]);  clamp_max_363 = None
	        _assert_tensor_metadata_1633 = torch.ops.aten._assert_tensor_metadata.default(view_2840, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1633 = None
	        convert_element_type_1087: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2840, torch.int8);  view_2840 = None
	        view_2841: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1087, [sym_size_int, 1500, 1280]);  convert_element_type_1087 = None
	        view_2842: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_543, [sym_size_int, 1500, 1]);  clamp_min_543 = None
	        view_2843: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1086, [sym_size_int, 1500, 1]);  convert_element_type_1086 = None
	        _assert_tensor_metadata_1634 = torch.ops.aten._assert_tensor_metadata.default(view_2841, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1634 = None
	        convert_element_type_1088: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2841, torch.float32);  view_2841 = None
	        _assert_tensor_metadata_1635 = torch.ops.aten._assert_tensor_metadata.default(view_2843, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1635 = None
	        convert_element_type_1089: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2843, torch.float32);  view_2843 = None
	        sub_8309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1088, convert_element_type_1089);  convert_element_type_1088 = convert_element_type_1089 = None
	        mul_17590: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8309, view_2842);  sub_8309 = view_2842 = None
	        view_2844: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17590, [sym_size_int, 1500, 1280]);  mul_17590 = None
	        _assert_tensor_metadata_1636 = torch.ops.aten._assert_tensor_metadata.default(view_2844, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1636 = None
	        view_2845: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2846: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2847: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1637 = torch.ops.aten._assert_tensor_metadata.default(view_2845, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1637 = None
	        convert_element_type_1090: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2845, torch.float32);  view_2845 = None
	        _assert_tensor_metadata_1638 = torch.ops.aten._assert_tensor_metadata.default(view_2847, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1638 = None
	        convert_element_type_1091: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2847, torch.float32);  view_2847 = None
	        sub_8313: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1090, convert_element_type_1091);  convert_element_type_1090 = convert_element_type_1091 = None
	        mul_17595: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8313, view_2846);  sub_8313 = view_2846 = None
	        view_2848: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17595, [1280, 1280]);  mul_17595 = None
	        _assert_tensor_metadata_1639 = torch.ops.aten._assert_tensor_metadata.default(view_2848, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1639 = None
	        permute_303: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2848, [1, 0]);  view_2848 = None
	        mul_17598: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2849: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2844, [mul_17598, 1280]);  view_2844 = mul_17598 = None
	        mm_30: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2849, permute_303);  view_2849 = permute_303 = None
	        view_2850: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_30, [sym_size_int, 1500, 1280]);  mm_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2851: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2850, [sym_size_int, -1, 20, 64]);  view_2850 = None
	        permute_304: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2851, [0, 2, 1, 3]);  view_2851 = None
	        clone_243: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_304, memory_format = torch.contiguous_format);  permute_304 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2852: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_27572, [sym_size_int, 1500, 1280])
	        amin_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2852, [2])
	        amax_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2852, [2]);  view_2852 = None
	        full_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_182, full_364);  amin_182 = full_364 = None
	        full_365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_182, full_365);  amax_182 = full_365 = None
	        sub_8327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_182, minimum_182);  maximum_182 = None
	        div_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8327, 255.0);  sub_8327 = None
	        clamp_min_546: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_364, 1.1920928955078125e-07);  div_364 = None
	        div_365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_182, clamp_min_546);  minimum_182 = None
	        round_365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_365);  div_365 = None
	        sub_8333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_365);  round_365 = None
	        clamp_min_547: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8333, -128);  sub_8333 = None
	        clamp_max_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_547, 127);  clamp_min_547 = None
	        _assert_tensor_metadata_1640 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_546, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1640 = None
	        _assert_tensor_metadata_1641 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_364, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1641 = None
	        convert_element_type_1092: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_364, torch.int8);  clamp_max_364 = None
	        view_2853: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_27572, [sym_size_int, 1500, 1280]);  add_27572 = None
	        view_2854: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_546, [sym_size_int, 1500, 1])
	        view_2855: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1092, [sym_size_int, 1500, 1])
	        reciprocal_182: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2854);  view_2854 = None
	        mul_17664: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_182, 1.0);  reciprocal_182 = None
	        mul_17667: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2853, mul_17664);  view_2853 = mul_17664 = None
	        round_366: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17667);  mul_17667 = None
	        add_27959: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_366, view_2855);  round_366 = view_2855 = None
	        clamp_min_548: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27959, -128);  add_27959 = None
	        clamp_max_365: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_548, 127);  clamp_min_548 = None
	        view_2856: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_365, [sym_size_int, 1500, 1280]);  clamp_max_365 = None
	        _assert_tensor_metadata_1642 = torch.ops.aten._assert_tensor_metadata.default(view_2856, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1642 = None
	        convert_element_type_1093: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2856, torch.int8);  view_2856 = None
	        view_2857: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1093, [sym_size_int, 1500, 1280]);  convert_element_type_1093 = None
	        view_2858: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_546, [sym_size_int, 1500, 1]);  clamp_min_546 = None
	        view_2859: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1092, [sym_size_int, 1500, 1]);  convert_element_type_1092 = None
	        _assert_tensor_metadata_1643 = torch.ops.aten._assert_tensor_metadata.default(view_2857, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1643 = None
	        convert_element_type_1094: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2857, torch.float32);  view_2857 = None
	        _assert_tensor_metadata_1644 = torch.ops.aten._assert_tensor_metadata.default(view_2859, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1644 = None
	        convert_element_type_1095: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2859, torch.float32);  view_2859 = None
	        sub_8353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1094, convert_element_type_1095);  convert_element_type_1094 = convert_element_type_1095 = None
	        mul_17689: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8353, view_2858);  sub_8353 = view_2858 = None
	        view_2860: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17689, [sym_size_int, 1500, 1280]);  mul_17689 = None
	        _assert_tensor_metadata_1645 = torch.ops.aten._assert_tensor_metadata.default(view_2860, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1645 = None
	        view_2861: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2862: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2863: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1646 = torch.ops.aten._assert_tensor_metadata.default(view_2861, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1646 = None
	        convert_element_type_1096: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2861, torch.float32);  view_2861 = None
	        _assert_tensor_metadata_1647 = torch.ops.aten._assert_tensor_metadata.default(view_2863, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1647 = None
	        convert_element_type_1097: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2863, torch.float32);  view_2863 = None
	        sub_8357: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1096, convert_element_type_1097);  convert_element_type_1096 = convert_element_type_1097 = None
	        mul_17694: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8357, view_2862);  sub_8357 = view_2862 = None
	        view_2864: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17694, [1280, 1280]);  mul_17694 = None
	        _assert_tensor_metadata_1648 = torch.ops.aten._assert_tensor_metadata.default(view_2864, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1648 = None
	        mul_17699: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2865: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2860, [mul_17699, 1280]);  view_2860 = mul_17699 = None
	        permute_305: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2864, [1, 0]);  view_2864 = None
	        addmm_151: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_30_self_attn_v_proj_bias, view_2865, permute_305);  model_audio_tower_layers_30_self_attn_v_proj_bias = view_2865 = permute_305 = None
	        view_2866: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_151, [sym_size_int, 1500, 1280]);  addmm_151 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2867: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2866, [sym_size_int, -1, 20, 64]);  view_2866 = None
	        permute_306: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2867, [0, 2, 1, 3]);  view_2867 = None
	        clone_244: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_306, memory_format = torch.contiguous_format);  permute_306 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_30 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_242, clone_243, clone_244, None, False, scale = 1.0);  clone_242 = clone_243 = clone_244 = None
	        getitem_242: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_30[0];  _scaled_dot_product_efficient_attention_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_307: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_242, [0, 2, 1, 3]);  getitem_242 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2868: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_307, [sym_size_int, 1500, -1]);  permute_307 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2869: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2868, [sym_size_int, 1500, 1280])
	        amin_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2869, [2])
	        amax_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2869, [2]);  view_2869 = None
	        full_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_183, full_366);  amin_183 = full_366 = None
	        full_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_183, full_367);  amax_183 = full_367 = None
	        sub_8375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_183, minimum_183);  maximum_183 = None
	        div_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8375, 255.0);  sub_8375 = None
	        clamp_min_549: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_366, 1.1920928955078125e-07);  div_366 = None
	        div_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_183, clamp_min_549);  minimum_183 = None
	        round_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_367);  div_367 = None
	        sub_8381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_367);  round_367 = None
	        clamp_min_550: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8381, -128);  sub_8381 = None
	        clamp_max_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_550, 127);  clamp_min_550 = None
	        _assert_tensor_metadata_1649 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_549, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1649 = None
	        _assert_tensor_metadata_1650 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_366, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1650 = None
	        convert_element_type_1098: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_366, torch.int8);  clamp_max_366 = None
	        view_2870: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2868, [sym_size_int, 1500, 1280]);  view_2868 = None
	        view_2871: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_549, [sym_size_int, 1500, 1])
	        view_2872: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1098, [sym_size_int, 1500, 1])
	        reciprocal_183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2871);  view_2871 = None
	        mul_17769: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_183, 1.0);  reciprocal_183 = None
	        mul_17772: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2870, mul_17769);  view_2870 = mul_17769 = None
	        round_368: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17772);  mul_17772 = None
	        add_28123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_368, view_2872);  round_368 = view_2872 = None
	        clamp_min_551: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28123, -128);  add_28123 = None
	        clamp_max_367: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_551, 127);  clamp_min_551 = None
	        view_2873: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_367, [sym_size_int, 1500, 1280]);  clamp_max_367 = None
	        _assert_tensor_metadata_1651 = torch.ops.aten._assert_tensor_metadata.default(view_2873, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1651 = None
	        convert_element_type_1099: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2873, torch.int8);  view_2873 = None
	        view_2874: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1099, [sym_size_int, 1500, 1280]);  convert_element_type_1099 = None
	        view_2875: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_549, [sym_size_int, 1500, 1]);  clamp_min_549 = None
	        view_2876: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1098, [sym_size_int, 1500, 1]);  convert_element_type_1098 = None
	        _assert_tensor_metadata_1652 = torch.ops.aten._assert_tensor_metadata.default(view_2874, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1652 = None
	        convert_element_type_1100: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2874, torch.float32);  view_2874 = None
	        _assert_tensor_metadata_1653 = torch.ops.aten._assert_tensor_metadata.default(view_2876, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1653 = None
	        convert_element_type_1101: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2876, torch.float32);  view_2876 = None
	        sub_8401: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1100, convert_element_type_1101);  convert_element_type_1100 = convert_element_type_1101 = None
	        mul_17794: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8401, view_2875);  sub_8401 = view_2875 = None
	        view_2877: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17794, [sym_size_int, 1500, 1280]);  mul_17794 = None
	        _assert_tensor_metadata_1654 = torch.ops.aten._assert_tensor_metadata.default(view_2877, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1654 = None
	        view_2878: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2879: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2880: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1655 = torch.ops.aten._assert_tensor_metadata.default(view_2878, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1655 = None
	        convert_element_type_1102: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2878, torch.float32);  view_2878 = None
	        _assert_tensor_metadata_1656 = torch.ops.aten._assert_tensor_metadata.default(view_2880, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1656 = None
	        convert_element_type_1103: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2880, torch.float32);  view_2880 = None
	        sub_8405: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1102, convert_element_type_1103);  convert_element_type_1102 = convert_element_type_1103 = None
	        mul_17799: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8405, view_2879);  sub_8405 = view_2879 = None
	        view_2881: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17799, [1280, 1280]);  mul_17799 = None
	        _assert_tensor_metadata_1657 = torch.ops.aten._assert_tensor_metadata.default(view_2881, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1657 = None
	        mul_17804: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2882: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2877, [mul_17804, 1280]);  view_2877 = mul_17804 = None
	        permute_308: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2881, [1, 0]);  view_2881 = None
	        addmm_152: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_30_self_attn_out_proj_bias, view_2882, permute_308);  model_audio_tower_layers_30_self_attn_out_proj_bias = view_2882 = permute_308 = None
	        view_2883: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_152, [sym_size_int, 1500, 1280]);  addmm_152 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_245: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2883);  view_2883 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_28186: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_27566, clone_245);  add_27566 = clone_245 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_246: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_28186, memory_format = torch.contiguous_format)
	        var_mean_61 = torch.ops.aten.var_mean.correction(clone_246, [2], correction = 0, keepdim = True)
	        getitem_246: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_61[0]
	        getitem_247: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_61[1];  var_mean_61 = None
	        add_28191: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_246, 1e-05);  getitem_246 = None
	        rsqrt_61: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_28191);  add_28191 = None
	        sub_8411: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_246, getitem_247);  clone_246 = getitem_247 = None
	        mul_17815: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8411, rsqrt_61);  sub_8411 = rsqrt_61 = None
	        mul_17816: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17815, model_audio_tower_layers_30_final_layer_norm_weight);  mul_17815 = model_audio_tower_layers_30_final_layer_norm_weight = None
	        add_28192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_17816, model_audio_tower_layers_30_final_layer_norm_bias);  mul_17816 = model_audio_tower_layers_30_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2884: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_28192, [sym_size_int, 1500, 1280])
	        amin_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2884, [2])
	        amax_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2884, [2]);  view_2884 = None
	        full_368: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_184, full_368);  amin_184 = full_368 = None
	        full_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_184, full_369);  amax_184 = full_369 = None
	        sub_8422: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_184, minimum_184);  maximum_184 = None
	        div_368: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8422, 255.0);  sub_8422 = None
	        clamp_min_552: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_368, 1.1920928955078125e-07);  div_368 = None
	        div_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_184, clamp_min_552);  minimum_184 = None
	        round_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_369);  div_369 = None
	        sub_8428: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_369);  round_369 = None
	        clamp_min_553: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8428, -128);  sub_8428 = None
	        clamp_max_368: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_553, 127);  clamp_min_553 = None
	        _assert_tensor_metadata_1658 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_552, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1658 = None
	        _assert_tensor_metadata_1659 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_368, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1659 = None
	        convert_element_type_1104: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_368, torch.int8);  clamp_max_368 = None
	        view_2885: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_28192, [sym_size_int, 1500, 1280]);  add_28192 = None
	        view_2886: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_552, [sym_size_int, 1500, 1])
	        view_2887: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1104, [sym_size_int, 1500, 1])
	        reciprocal_184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2886);  view_2886 = None
	        mul_17864: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_184, 1.0);  reciprocal_184 = None
	        mul_17867: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2885, mul_17864);  view_2885 = mul_17864 = None
	        round_370: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17867);  mul_17867 = None
	        add_28279: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_370, view_2887);  round_370 = view_2887 = None
	        clamp_min_554: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28279, -128);  add_28279 = None
	        clamp_max_369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_554, 127);  clamp_min_554 = None
	        view_2888: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_369, [sym_size_int, 1500, 1280]);  clamp_max_369 = None
	        _assert_tensor_metadata_1660 = torch.ops.aten._assert_tensor_metadata.default(view_2888, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1660 = None
	        convert_element_type_1105: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2888, torch.int8);  view_2888 = None
	        view_2889: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1105, [sym_size_int, 1500, 1280]);  convert_element_type_1105 = None
	        view_2890: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_552, [sym_size_int, 1500, 1]);  clamp_min_552 = None
	        view_2891: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1104, [sym_size_int, 1500, 1]);  convert_element_type_1104 = None
	        _assert_tensor_metadata_1661 = torch.ops.aten._assert_tensor_metadata.default(view_2889, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1661 = None
	        convert_element_type_1106: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2889, torch.float32);  view_2889 = None
	        _assert_tensor_metadata_1662 = torch.ops.aten._assert_tensor_metadata.default(view_2891, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1662 = None
	        convert_element_type_1107: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2891, torch.float32);  view_2891 = None
	        sub_8448: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1106, convert_element_type_1107);  convert_element_type_1106 = convert_element_type_1107 = None
	        mul_17889: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8448, view_2890);  sub_8448 = view_2890 = None
	        view_2892: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17889, [sym_size_int, 1500, 1280]);  mul_17889 = None
	        _assert_tensor_metadata_1663 = torch.ops.aten._assert_tensor_metadata.default(view_2892, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1663 = None
	        view_2893: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_30_fc1_parametrizations_weight_original0 = None
	        view_2894: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_30_fc1_parametrizations_weight_original1 = None
	        view_2895: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_30_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1664 = torch.ops.aten._assert_tensor_metadata.default(view_2893, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1664 = None
	        convert_element_type_1108: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2893, torch.float32);  view_2893 = None
	        _assert_tensor_metadata_1665 = torch.ops.aten._assert_tensor_metadata.default(view_2895, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1665 = None
	        convert_element_type_1109: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2895, torch.float32);  view_2895 = None
	        sub_8452: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1108, convert_element_type_1109);  convert_element_type_1108 = convert_element_type_1109 = None
	        mul_17894: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8452, view_2894);  sub_8452 = view_2894 = None
	        view_2896: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17894, [5120, 1280]);  mul_17894 = None
	        _assert_tensor_metadata_1666 = torch.ops.aten._assert_tensor_metadata.default(view_2896, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1666 = None
	        mul_17899: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2897: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2892, [mul_17899, 1280]);  view_2892 = mul_17899 = None
	        permute_309: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2896, [1, 0]);  view_2896 = None
	        addmm_153: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_30_fc1_bias, view_2897, permute_309);  model_audio_tower_layers_30_fc1_bias = view_2897 = permute_309 = None
	        view_2898: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_153, [sym_size_int, 1500, 5120]);  addmm_153 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_17906: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2898, 0.5)
	        mul_17907: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2898, 0.7071067811865476);  view_2898 = None
	        erf_32: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_17907);  mul_17907 = None
	        add_28338: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_32, 1);  erf_32 = None
	        mul_17908: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17906, add_28338);  mul_17906 = add_28338 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_247: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_17908);  mul_17908 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2899: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_247, [sym_size_int, 1500, 5120])
	        amin_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2899, [2])
	        amax_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2899, [2]);  view_2899 = None
	        full_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_185, full_370);  amin_185 = full_370 = None
	        full_371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_185, full_371);  amax_185 = full_371 = None
	        sub_8465: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_185, minimum_185);  maximum_185 = None
	        div_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8465, 255.0);  sub_8465 = None
	        clamp_min_555: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_370, 1.1920928955078125e-07);  div_370 = None
	        div_371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_185, clamp_min_555);  minimum_185 = None
	        round_371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_371);  div_371 = None
	        sub_8471: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_371);  round_371 = None
	        clamp_min_556: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8471, -128);  sub_8471 = None
	        clamp_max_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_556, 127);  clamp_min_556 = None
	        _assert_tensor_metadata_1667 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_555, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1667 = None
	        _assert_tensor_metadata_1668 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_370, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1668 = None
	        convert_element_type_1110: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_370, torch.int8);  clamp_max_370 = None
	        view_2900: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_247, [sym_size_int, 1500, 5120]);  clone_247 = None
	        view_2901: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_555, [sym_size_int, 1500, 1])
	        view_2902: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1110, [sym_size_int, 1500, 1])
	        reciprocal_185: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2901);  view_2901 = None
	        mul_17954: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_185, 1.0);  reciprocal_185 = None
	        mul_17957: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2900, mul_17954);  view_2900 = mul_17954 = None
	        round_372: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_17957);  mul_17957 = None
	        add_28421: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_372, view_2902);  round_372 = view_2902 = None
	        clamp_min_557: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28421, -128);  add_28421 = None
	        clamp_max_371: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_557, 127);  clamp_min_557 = None
	        view_2903: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_371, [sym_size_int, 1500, 5120]);  clamp_max_371 = None
	        _assert_tensor_metadata_1669 = torch.ops.aten._assert_tensor_metadata.default(view_2903, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1669 = None
	        convert_element_type_1111: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2903, torch.int8);  view_2903 = None
	        view_2904: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1111, [sym_size_int, 1500, 5120]);  convert_element_type_1111 = None
	        view_2905: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_555, [sym_size_int, 1500, 1]);  clamp_min_555 = None
	        view_2906: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1110, [sym_size_int, 1500, 1]);  convert_element_type_1110 = None
	        _assert_tensor_metadata_1670 = torch.ops.aten._assert_tensor_metadata.default(view_2904, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1670 = None
	        convert_element_type_1112: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2904, torch.float32);  view_2904 = None
	        _assert_tensor_metadata_1671 = torch.ops.aten._assert_tensor_metadata.default(view_2906, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1671 = None
	        convert_element_type_1113: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2906, torch.float32);  view_2906 = None
	        sub_8491: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1112, convert_element_type_1113);  convert_element_type_1112 = convert_element_type_1113 = None
	        mul_17979: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8491, view_2905);  sub_8491 = view_2905 = None
	        view_2907: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_17979, [sym_size_int, 1500, 5120]);  mul_17979 = None
	        _assert_tensor_metadata_1672 = torch.ops.aten._assert_tensor_metadata.default(view_2907, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1672 = None
	        view_2908: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_30_fc2_parametrizations_weight_original0 = None
	        view_2909: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_30_fc2_parametrizations_weight_original1 = None
	        view_2910: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_30_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1673 = torch.ops.aten._assert_tensor_metadata.default(view_2908, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1673 = None
	        convert_element_type_1114: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2908, torch.float32);  view_2908 = None
	        _assert_tensor_metadata_1674 = torch.ops.aten._assert_tensor_metadata.default(view_2910, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1674 = None
	        convert_element_type_1115: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2910, torch.float32);  view_2910 = None
	        sub_8495: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1114, convert_element_type_1115);  convert_element_type_1114 = convert_element_type_1115 = None
	        mul_17984: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8495, view_2909);  sub_8495 = view_2909 = None
	        view_2911: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_17984, [1280, 5120]);  mul_17984 = None
	        _assert_tensor_metadata_1675 = torch.ops.aten._assert_tensor_metadata.default(view_2911, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1675 = None
	        mul_17989: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2912: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_2907, [mul_17989, 5120]);  view_2907 = mul_17989 = None
	        permute_310: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2911, [1, 0]);  view_2911 = None
	        addmm_154: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_30_fc2_bias, view_2912, permute_310);  model_audio_tower_layers_30_fc2_bias = view_2912 = permute_310 = None
	        view_2913: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_154, [sym_size_int, 1500, 1280]);  addmm_154 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_248: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2913);  view_2913 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_28484: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_28186, clone_248);  add_28186 = clone_248 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_28484, memory_format = torch.contiguous_format)
	        var_mean_62 = torch.ops.aten.var_mean.correction(clone_249, [2], correction = 0, keepdim = True)
	        getitem_248: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_62[0]
	        getitem_249: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_62[1];  var_mean_62 = None
	        add_28489: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_248, 1e-05);  getitem_248 = None
	        rsqrt_62: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_28489);  add_28489 = None
	        sub_8501: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_249, getitem_249);  clone_249 = getitem_249 = None
	        mul_18000: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8501, rsqrt_62);  sub_8501 = rsqrt_62 = None
	        mul_18001: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18000, model_audio_tower_layers_31_self_attn_layer_norm_weight);  mul_18000 = model_audio_tower_layers_31_self_attn_layer_norm_weight = None
	        add_28490: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_18001, model_audio_tower_layers_31_self_attn_layer_norm_bias);  mul_18001 = model_audio_tower_layers_31_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        view_2914: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_28490, [sym_size_int, 1500, 1280])
	        amin_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2914, [2])
	        amax_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2914, [2]);  view_2914 = None
	        full_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_186, full_372);  amin_186 = full_372 = None
	        full_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_186, full_373);  amax_186 = full_373 = None
	        sub_8512: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_186, minimum_186);  maximum_186 = None
	        div_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8512, 255.0);  sub_8512 = None
	        clamp_min_558: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_372, 1.1920928955078125e-07);  div_372 = None
	        div_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_186, clamp_min_558);  minimum_186 = None
	        round_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_373);  div_373 = None
	        sub_8518: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_373);  round_373 = None
	        clamp_min_559: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8518, -128);  sub_8518 = None
	        clamp_max_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_559, 127);  clamp_min_559 = None
	        _assert_tensor_metadata_1676 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_558, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1676 = None
	        _assert_tensor_metadata_1677 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_372, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1677 = None
	        convert_element_type_1116: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_372, torch.int8);  clamp_max_372 = None
	        view_2915: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_28490, [sym_size_int, 1500, 1280])
	        view_2916: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_558, [sym_size_int, 1500, 1])
	        view_2917: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1116, [sym_size_int, 1500, 1])
	        reciprocal_186: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2916);  view_2916 = None
	        mul_18049: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_186, 1.0);  reciprocal_186 = None
	        mul_18052: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2915, mul_18049);  view_2915 = mul_18049 = None
	        round_374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18052);  mul_18052 = None
	        add_28577: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_374, view_2917);  round_374 = view_2917 = None
	        clamp_min_560: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28577, -128);  add_28577 = None
	        clamp_max_373: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_560, 127);  clamp_min_560 = None
	        view_2918: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_373, [sym_size_int, 1500, 1280]);  clamp_max_373 = None
	        _assert_tensor_metadata_1678 = torch.ops.aten._assert_tensor_metadata.default(view_2918, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1678 = None
	        convert_element_type_1117: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2918, torch.int8);  view_2918 = None
	        view_2919: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1117, [sym_size_int, 1500, 1280]);  convert_element_type_1117 = None
	        view_2920: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_558, [sym_size_int, 1500, 1]);  clamp_min_558 = None
	        view_2921: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1116, [sym_size_int, 1500, 1]);  convert_element_type_1116 = None
	        _assert_tensor_metadata_1679 = torch.ops.aten._assert_tensor_metadata.default(view_2919, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1679 = None
	        convert_element_type_1118: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2919, torch.float32);  view_2919 = None
	        _assert_tensor_metadata_1680 = torch.ops.aten._assert_tensor_metadata.default(view_2921, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1680 = None
	        convert_element_type_1119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2921, torch.float32);  view_2921 = None
	        sub_8538: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1118, convert_element_type_1119);  convert_element_type_1118 = convert_element_type_1119 = None
	        mul_18074: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8538, view_2920);  sub_8538 = view_2920 = None
	        view_2922: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18074, [sym_size_int, 1500, 1280]);  mul_18074 = None
	        _assert_tensor_metadata_1681 = torch.ops.aten._assert_tensor_metadata.default(view_2922, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1681 = None
	        view_2923: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2924: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2925: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1682 = torch.ops.aten._assert_tensor_metadata.default(view_2923, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1682 = None
	        convert_element_type_1120: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2923, torch.float32);  view_2923 = None
	        _assert_tensor_metadata_1683 = torch.ops.aten._assert_tensor_metadata.default(view_2925, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1683 = None
	        convert_element_type_1121: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2925, torch.float32);  view_2925 = None
	        sub_8542: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1120, convert_element_type_1121);  convert_element_type_1120 = convert_element_type_1121 = None
	        mul_18079: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8542, view_2924);  sub_8542 = view_2924 = None
	        view_2926: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18079, [1280, 1280]);  mul_18079 = None
	        _assert_tensor_metadata_1684 = torch.ops.aten._assert_tensor_metadata.default(view_2926, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1684 = None
	        mul_18084: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2927: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2922, [mul_18084, 1280]);  view_2922 = mul_18084 = None
	        permute_311: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2926, [1, 0]);  view_2926 = None
	        addmm_155: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_31_self_attn_q_proj_bias, view_2927, permute_311);  model_audio_tower_layers_31_self_attn_q_proj_bias = view_2927 = permute_311 = None
	        view_2928: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_155, [sym_size_int, 1500, 1280]);  addmm_155 = None
	        mul_18091: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2928, 0.125);  view_2928 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2929: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_18091, [sym_size_int, 1500, 20, 64]);  mul_18091 = None
	        permute_312: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2929, [0, 2, 1, 3]);  view_2929 = None
	        clone_250: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_312, memory_format = torch.contiguous_format);  permute_312 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        view_2930: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_28490, [sym_size_int, 1500, 1280])
	        amin_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2930, [2])
	        amax_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2930, [2]);  view_2930 = None
	        full_374: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_187, full_374);  amin_187 = full_374 = None
	        full_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_187, full_375);  amax_187 = full_375 = None
	        sub_8557: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_187, minimum_187);  maximum_187 = None
	        div_374: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8557, 255.0);  sub_8557 = None
	        clamp_min_561: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_374, 1.1920928955078125e-07);  div_374 = None
	        div_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_187, clamp_min_561);  minimum_187 = None
	        round_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_375);  div_375 = None
	        sub_8563: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_375);  round_375 = None
	        clamp_min_562: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8563, -128);  sub_8563 = None
	        clamp_max_374: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_562, 127);  clamp_min_562 = None
	        _assert_tensor_metadata_1685 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_561, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1685 = None
	        _assert_tensor_metadata_1686 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_374, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1686 = None
	        convert_element_type_1122: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_374, torch.int8);  clamp_max_374 = None
	        view_2931: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_28490, [sym_size_int, 1500, 1280])
	        view_2932: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_561, [sym_size_int, 1500, 1])
	        view_2933: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1122, [sym_size_int, 1500, 1])
	        reciprocal_187: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2932);  view_2932 = None
	        mul_18145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_187, 1.0);  reciprocal_187 = None
	        mul_18148: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2931, mul_18145);  view_2931 = mul_18145 = None
	        round_376: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18148);  mul_18148 = None
	        add_28729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_376, view_2933);  round_376 = view_2933 = None
	        clamp_min_563: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28729, -128);  add_28729 = None
	        clamp_max_375: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_563, 127);  clamp_min_563 = None
	        view_2934: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_375, [sym_size_int, 1500, 1280]);  clamp_max_375 = None
	        _assert_tensor_metadata_1687 = torch.ops.aten._assert_tensor_metadata.default(view_2934, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1687 = None
	        convert_element_type_1123: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2934, torch.int8);  view_2934 = None
	        view_2935: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1123, [sym_size_int, 1500, 1280]);  convert_element_type_1123 = None
	        view_2936: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_561, [sym_size_int, 1500, 1]);  clamp_min_561 = None
	        view_2937: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1122, [sym_size_int, 1500, 1]);  convert_element_type_1122 = None
	        _assert_tensor_metadata_1688 = torch.ops.aten._assert_tensor_metadata.default(view_2935, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1688 = None
	        convert_element_type_1124: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2935, torch.float32);  view_2935 = None
	        _assert_tensor_metadata_1689 = torch.ops.aten._assert_tensor_metadata.default(view_2937, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1689 = None
	        convert_element_type_1125: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2937, torch.float32);  view_2937 = None
	        sub_8583: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1124, convert_element_type_1125);  convert_element_type_1124 = convert_element_type_1125 = None
	        mul_18170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8583, view_2936);  sub_8583 = view_2936 = None
	        view_2938: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18170, [sym_size_int, 1500, 1280]);  mul_18170 = None
	        _assert_tensor_metadata_1690 = torch.ops.aten._assert_tensor_metadata.default(view_2938, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1690 = None
	        view_2939: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2940: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2941: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1691 = torch.ops.aten._assert_tensor_metadata.default(view_2939, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1691 = None
	        convert_element_type_1126: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2939, torch.float32);  view_2939 = None
	        _assert_tensor_metadata_1692 = torch.ops.aten._assert_tensor_metadata.default(view_2941, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1692 = None
	        convert_element_type_1127: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2941, torch.float32);  view_2941 = None
	        sub_8587: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1126, convert_element_type_1127);  convert_element_type_1126 = convert_element_type_1127 = None
	        mul_18175: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8587, view_2940);  sub_8587 = view_2940 = None
	        view_2942: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18175, [1280, 1280]);  mul_18175 = None
	        _assert_tensor_metadata_1693 = torch.ops.aten._assert_tensor_metadata.default(view_2942, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1693 = None
	        permute_313: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2942, [1, 0]);  view_2942 = None
	        mul_18178: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2943: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2938, [mul_18178, 1280]);  view_2938 = mul_18178 = None
	        mm_31: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2943, permute_313);  view_2943 = permute_313 = None
	        view_2944: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_31, [sym_size_int, 1500, 1280]);  mm_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2945: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2944, [sym_size_int, -1, 20, 64]);  view_2944 = None
	        permute_314: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2945, [0, 2, 1, 3]);  view_2945 = None
	        clone_251: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_314, memory_format = torch.contiguous_format);  permute_314 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        view_2946: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_28490, [sym_size_int, 1500, 1280])
	        amin_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2946, [2])
	        amax_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2946, [2]);  view_2946 = None
	        full_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_188, full_376);  amin_188 = full_376 = None
	        full_377: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_188, full_377);  amax_188 = full_377 = None
	        sub_8601: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_188, minimum_188);  maximum_188 = None
	        div_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8601, 255.0);  sub_8601 = None
	        clamp_min_564: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_376, 1.1920928955078125e-07);  div_376 = None
	        div_377: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_188, clamp_min_564);  minimum_188 = None
	        round_377: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_377);  div_377 = None
	        sub_8607: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_377);  round_377 = None
	        clamp_min_565: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8607, -128);  sub_8607 = None
	        clamp_max_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_565, 127);  clamp_min_565 = None
	        _assert_tensor_metadata_1694 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_564, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1694 = None
	        _assert_tensor_metadata_1695 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_376, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1695 = None
	        convert_element_type_1128: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_376, torch.int8);  clamp_max_376 = None
	        view_2947: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_28490, [sym_size_int, 1500, 1280]);  add_28490 = None
	        view_2948: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_564, [sym_size_int, 1500, 1])
	        view_2949: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1128, [sym_size_int, 1500, 1])
	        reciprocal_188: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2948);  view_2948 = None
	        mul_18244: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_188, 1.0);  reciprocal_188 = None
	        mul_18247: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2947, mul_18244);  view_2947 = mul_18244 = None
	        round_378: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18247);  mul_18247 = None
	        add_28877: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_378, view_2949);  round_378 = view_2949 = None
	        clamp_min_566: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28877, -128);  add_28877 = None
	        clamp_max_377: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_566, 127);  clamp_min_566 = None
	        view_2950: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_377, [sym_size_int, 1500, 1280]);  clamp_max_377 = None
	        _assert_tensor_metadata_1696 = torch.ops.aten._assert_tensor_metadata.default(view_2950, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1696 = None
	        convert_element_type_1129: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2950, torch.int8);  view_2950 = None
	        view_2951: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1129, [sym_size_int, 1500, 1280]);  convert_element_type_1129 = None
	        view_2952: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_564, [sym_size_int, 1500, 1]);  clamp_min_564 = None
	        view_2953: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1128, [sym_size_int, 1500, 1]);  convert_element_type_1128 = None
	        _assert_tensor_metadata_1697 = torch.ops.aten._assert_tensor_metadata.default(view_2951, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1697 = None
	        convert_element_type_1130: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2951, torch.float32);  view_2951 = None
	        _assert_tensor_metadata_1698 = torch.ops.aten._assert_tensor_metadata.default(view_2953, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1698 = None
	        convert_element_type_1131: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2953, torch.float32);  view_2953 = None
	        sub_8627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1130, convert_element_type_1131);  convert_element_type_1130 = convert_element_type_1131 = None
	        mul_18269: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8627, view_2952);  sub_8627 = view_2952 = None
	        view_2954: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18269, [sym_size_int, 1500, 1280]);  mul_18269 = None
	        _assert_tensor_metadata_1699 = torch.ops.aten._assert_tensor_metadata.default(view_2954, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1699 = None
	        view_2955: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2956: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2957: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1700 = torch.ops.aten._assert_tensor_metadata.default(view_2955, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1700 = None
	        convert_element_type_1132: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2955, torch.float32);  view_2955 = None
	        _assert_tensor_metadata_1701 = torch.ops.aten._assert_tensor_metadata.default(view_2957, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1701 = None
	        convert_element_type_1133: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2957, torch.float32);  view_2957 = None
	        sub_8631: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1132, convert_element_type_1133);  convert_element_type_1132 = convert_element_type_1133 = None
	        mul_18274: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8631, view_2956);  sub_8631 = view_2956 = None
	        view_2958: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18274, [1280, 1280]);  mul_18274 = None
	        _assert_tensor_metadata_1702 = torch.ops.aten._assert_tensor_metadata.default(view_2958, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1702 = None
	        mul_18279: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2959: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2954, [mul_18279, 1280]);  view_2954 = mul_18279 = None
	        permute_315: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2958, [1, 0]);  view_2958 = None
	        addmm_156: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_31_self_attn_v_proj_bias, view_2959, permute_315);  model_audio_tower_layers_31_self_attn_v_proj_bias = view_2959 = permute_315 = None
	        view_2960: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_156, [sym_size_int, 1500, 1280]);  addmm_156 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2961: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2960, [sym_size_int, -1, 20, 64]);  view_2960 = None
	        permute_316: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2961, [0, 2, 1, 3]);  view_2961 = None
	        clone_252: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_316, memory_format = torch.contiguous_format);  permute_316 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_31 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_250, clone_251, clone_252, None, False, scale = 1.0);  clone_250 = clone_251 = clone_252 = None
	        getitem_250: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_31[0];  _scaled_dot_product_efficient_attention_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_317: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_250, [0, 2, 1, 3]);  getitem_250 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2962: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_317, [sym_size_int, 1500, -1]);  permute_317 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        view_2963: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2962, [sym_size_int, 1500, 1280])
	        amin_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2963, [2])
	        amax_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2963, [2]);  view_2963 = None
	        full_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_189, full_378);  amin_189 = full_378 = None
	        full_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_189, full_379);  amax_189 = full_379 = None
	        sub_8649: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_189, minimum_189);  maximum_189 = None
	        div_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8649, 255.0);  sub_8649 = None
	        clamp_min_567: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_378, 1.1920928955078125e-07);  div_378 = None
	        div_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_189, clamp_min_567);  minimum_189 = None
	        round_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_379);  div_379 = None
	        sub_8655: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_379);  round_379 = None
	        clamp_min_568: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8655, -128);  sub_8655 = None
	        clamp_max_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_568, 127);  clamp_min_568 = None
	        _assert_tensor_metadata_1703 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_567, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1703 = None
	        _assert_tensor_metadata_1704 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_378, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1704 = None
	        convert_element_type_1134: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_378, torch.int8);  clamp_max_378 = None
	        view_2964: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(view_2962, [sym_size_int, 1500, 1280]);  view_2962 = None
	        view_2965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_567, [sym_size_int, 1500, 1])
	        view_2966: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1134, [sym_size_int, 1500, 1])
	        reciprocal_189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2965);  view_2965 = None
	        mul_18349: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_189, 1.0);  reciprocal_189 = None
	        mul_18352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2964, mul_18349);  view_2964 = mul_18349 = None
	        round_380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18352);  mul_18352 = None
	        add_29041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_380, view_2966);  round_380 = view_2966 = None
	        clamp_min_569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29041, -128);  add_29041 = None
	        clamp_max_379: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_569, 127);  clamp_min_569 = None
	        view_2967: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_379, [sym_size_int, 1500, 1280]);  clamp_max_379 = None
	        _assert_tensor_metadata_1705 = torch.ops.aten._assert_tensor_metadata.default(view_2967, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1705 = None
	        convert_element_type_1135: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2967, torch.int8);  view_2967 = None
	        view_2968: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1135, [sym_size_int, 1500, 1280]);  convert_element_type_1135 = None
	        view_2969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_567, [sym_size_int, 1500, 1]);  clamp_min_567 = None
	        view_2970: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1134, [sym_size_int, 1500, 1]);  convert_element_type_1134 = None
	        _assert_tensor_metadata_1706 = torch.ops.aten._assert_tensor_metadata.default(view_2968, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1706 = None
	        convert_element_type_1136: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2968, torch.float32);  view_2968 = None
	        _assert_tensor_metadata_1707 = torch.ops.aten._assert_tensor_metadata.default(view_2970, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1707 = None
	        convert_element_type_1137: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2970, torch.float32);  view_2970 = None
	        sub_8675: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1136, convert_element_type_1137);  convert_element_type_1136 = convert_element_type_1137 = None
	        mul_18374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8675, view_2969);  sub_8675 = view_2969 = None
	        view_2971: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18374, [sym_size_int, 1500, 1280]);  mul_18374 = None
	        _assert_tensor_metadata_1708 = torch.ops.aten._assert_tensor_metadata.default(view_2971, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1708 = None
	        view_2972: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2973: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2974: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1709 = torch.ops.aten._assert_tensor_metadata.default(view_2972, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1709 = None
	        convert_element_type_1138: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2972, torch.float32);  view_2972 = None
	        _assert_tensor_metadata_1710 = torch.ops.aten._assert_tensor_metadata.default(view_2974, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1710 = None
	        convert_element_type_1139: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2974, torch.float32);  view_2974 = None
	        sub_8679: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1138, convert_element_type_1139);  convert_element_type_1138 = convert_element_type_1139 = None
	        mul_18379: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8679, view_2973);  sub_8679 = view_2973 = None
	        view_2975: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18379, [1280, 1280]);  mul_18379 = None
	        _assert_tensor_metadata_1711 = torch.ops.aten._assert_tensor_metadata.default(view_2975, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1711 = None
	        mul_18384: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2976: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2971, [mul_18384, 1280]);  view_2971 = mul_18384 = None
	        permute_318: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2975, [1, 0]);  view_2975 = None
	        addmm_157: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_31_self_attn_out_proj_bias, view_2976, permute_318);  model_audio_tower_layers_31_self_attn_out_proj_bias = view_2976 = permute_318 = None
	        view_2977: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_157, [sym_size_int, 1500, 1280]);  addmm_157 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:209 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_253: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_2977);  view_2977 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_29104: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_28484, clone_253);  add_28484 = clone_253 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_29104, memory_format = torch.contiguous_format)
	        var_mean_63 = torch.ops.aten.var_mean.correction(clone_254, [2], correction = 0, keepdim = True)
	        getitem_254: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_63[0]
	        getitem_255: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_63[1];  var_mean_63 = None
	        add_29109: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_254, 1e-05);  getitem_254 = None
	        rsqrt_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_29109);  add_29109 = None
	        sub_8685: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_254, getitem_255);  clone_254 = getitem_255 = None
	        mul_18395: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8685, rsqrt_63);  sub_8685 = rsqrt_63 = None
	        mul_18396: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18395, model_audio_tower_layers_31_final_layer_norm_weight);  mul_18395 = model_audio_tower_layers_31_final_layer_norm_weight = None
	        add_29110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_18396, model_audio_tower_layers_31_final_layer_norm_bias);  mul_18396 = model_audio_tower_layers_31_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        view_2978: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_29110, [sym_size_int, 1500, 1280])
	        amin_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2978, [2])
	        amax_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2978, [2]);  view_2978 = None
	        full_380: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_190, full_380);  amin_190 = full_380 = None
	        full_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_190, full_381);  amax_190 = full_381 = None
	        sub_8696: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_190, minimum_190);  maximum_190 = None
	        div_380: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8696, 255.0);  sub_8696 = None
	        clamp_min_570: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_380, 1.1920928955078125e-07);  div_380 = None
	        div_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_190, clamp_min_570);  minimum_190 = None
	        round_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_381);  div_381 = None
	        sub_8702: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_381);  round_381 = None
	        clamp_min_571: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8702, -128);  sub_8702 = None
	        clamp_max_380: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_571, 127);  clamp_min_571 = None
	        _assert_tensor_metadata_1712 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_570, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1712 = None
	        _assert_tensor_metadata_1713 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_380, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1713 = None
	        convert_element_type_1140: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_380, torch.int8);  clamp_max_380 = None
	        view_2979: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(add_29110, [sym_size_int, 1500, 1280]);  add_29110 = None
	        view_2980: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_570, [sym_size_int, 1500, 1])
	        view_2981: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1140, [sym_size_int, 1500, 1])
	        reciprocal_190: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2980);  view_2980 = None
	        mul_18444: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_190, 1.0);  reciprocal_190 = None
	        mul_18447: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2979, mul_18444);  view_2979 = mul_18444 = None
	        round_382: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18447);  mul_18447 = None
	        add_29197: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_382, view_2981);  round_382 = view_2981 = None
	        clamp_min_572: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29197, -128);  add_29197 = None
	        clamp_max_381: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_572, 127);  clamp_min_572 = None
	        view_2982: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_381, [sym_size_int, 1500, 1280]);  clamp_max_381 = None
	        _assert_tensor_metadata_1714 = torch.ops.aten._assert_tensor_metadata.default(view_2982, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1714 = None
	        convert_element_type_1141: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2982, torch.int8);  view_2982 = None
	        view_2983: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1141, [sym_size_int, 1500, 1280]);  convert_element_type_1141 = None
	        view_2984: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_570, [sym_size_int, 1500, 1]);  clamp_min_570 = None
	        view_2985: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1140, [sym_size_int, 1500, 1]);  convert_element_type_1140 = None
	        _assert_tensor_metadata_1715 = torch.ops.aten._assert_tensor_metadata.default(view_2983, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1715 = None
	        convert_element_type_1142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2983, torch.float32);  view_2983 = None
	        _assert_tensor_metadata_1716 = torch.ops.aten._assert_tensor_metadata.default(view_2985, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1716 = None
	        convert_element_type_1143: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2985, torch.float32);  view_2985 = None
	        sub_8722: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1142, convert_element_type_1143);  convert_element_type_1142 = convert_element_type_1143 = None
	        mul_18469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8722, view_2984);  sub_8722 = view_2984 = None
	        view_2986: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18469, [sym_size_int, 1500, 1280]);  mul_18469 = None
	        _assert_tensor_metadata_1717 = torch.ops.aten._assert_tensor_metadata.default(view_2986, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1717 = None
	        view_2987: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_31_fc1_parametrizations_weight_original0 = None
	        view_2988: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_31_fc1_parametrizations_weight_original1 = None
	        view_2989: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_31_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1718 = torch.ops.aten._assert_tensor_metadata.default(view_2987, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1718 = None
	        convert_element_type_1144: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2987, torch.float32);  view_2987 = None
	        _assert_tensor_metadata_1719 = torch.ops.aten._assert_tensor_metadata.default(view_2989, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1719 = None
	        convert_element_type_1145: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2989, torch.float32);  view_2989 = None
	        sub_8726: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1144, convert_element_type_1145);  convert_element_type_1144 = convert_element_type_1145 = None
	        mul_18474: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8726, view_2988);  sub_8726 = view_2988 = None
	        view_2990: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18474, [5120, 1280]);  mul_18474 = None
	        _assert_tensor_metadata_1720 = torch.ops.aten._assert_tensor_metadata.default(view_2990, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1720 = None
	        mul_18479: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2991: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(view_2986, [mul_18479, 1280]);  view_2986 = mul_18479 = None
	        permute_319: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2990, [1, 0]);  view_2990 = None
	        addmm_158: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_31_fc1_bias, view_2991, permute_319);  model_audio_tower_layers_31_fc1_bias = view_2991 = permute_319 = None
	        view_2992: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_158, [sym_size_int, 1500, 5120]);  addmm_158 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_18486: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2992, 0.5)
	        mul_18487: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2992, 0.7071067811865476);  view_2992 = None
	        erf_33: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_18487);  mul_18487 = None
	        add_29256: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_33, 1);  erf_33 = None
	        mul_18488: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18486, add_29256);  mul_18486 = add_29256 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:215 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
	        clone_255: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clone.default(mul_18488);  mul_18488 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        view_2993: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_255, [sym_size_int, 1500, 5120])
	        amin_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2993, [2])
	        amax_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2993, [2]);  view_2993 = None
	        full_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_191, full_382);  amin_191 = full_382 = None
	        full_383: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_191, full_383);  amax_191 = full_383 = None
	        sub_8739: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_191, minimum_191);  maximum_191 = None
	        div_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8739, 255.0);  sub_8739 = None
	        clamp_min_573: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_382, 1.1920928955078125e-07);  div_382 = None
	        div_383: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_191, clamp_min_573);  minimum_191 = None
	        round_383: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_383);  div_383 = None
	        sub_8745: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_383);  round_383 = None
	        clamp_min_574: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8745, -128);  sub_8745 = None
	        clamp_max_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_574, 127);  clamp_min_574 = None
	        _assert_tensor_metadata_1721 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_573, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1721 = None
	        _assert_tensor_metadata_1722 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_382, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1722 = None
	        convert_element_type_1146: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_382, torch.int8);  clamp_max_382 = None
	        view_2994: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clone_255, [sym_size_int, 1500, 5120]);  clone_255 = None
	        view_2995: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_573, [sym_size_int, 1500, 1])
	        view_2996: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1146, [sym_size_int, 1500, 1])
	        reciprocal_191: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2995);  view_2995 = None
	        mul_18534: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_191, 1.0);  reciprocal_191 = None
	        mul_18537: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2994, mul_18534);  view_2994 = mul_18534 = None
	        round_384: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_18537);  mul_18537 = None
	        add_29339: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_384, view_2996);  round_384 = view_2996 = None
	        clamp_min_575: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29339, -128);  add_29339 = None
	        clamp_max_383: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_575, 127);  clamp_min_575 = None
	        view_2997: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_383, [sym_size_int, 1500, 5120]);  clamp_max_383 = None
	        _assert_tensor_metadata_1723 = torch.ops.aten._assert_tensor_metadata.default(view_2997, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1723 = None
	        convert_element_type_1147: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2997, torch.int8);  view_2997 = None
	        view_2998: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1147, [sym_size_int, 1500, 5120]);  convert_element_type_1147 = None
	        view_2999: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_573, [sym_size_int, 1500, 1]);  clamp_min_573 = None
	        view_3000: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1146, [sym_size_int, 1500, 1]);  convert_element_type_1146 = None
	        _assert_tensor_metadata_1724 = torch.ops.aten._assert_tensor_metadata.default(view_2998, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1724 = None
	        convert_element_type_1148: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2998, torch.float32);  view_2998 = None
	        _assert_tensor_metadata_1725 = torch.ops.aten._assert_tensor_metadata.default(view_3000, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1725 = None
	        convert_element_type_1149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3000, torch.float32);  view_3000 = None
	        sub_8765: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1148, convert_element_type_1149);  convert_element_type_1148 = convert_element_type_1149 = None
	        mul_18559: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8765, view_2999);  sub_8765 = view_2999 = None
	        view_3001: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(mul_18559, [sym_size_int, 1500, 5120]);  mul_18559 = None
	        _assert_tensor_metadata_1726 = torch.ops.aten._assert_tensor_metadata.default(view_3001, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1726 = None
	        view_3002: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_31_fc2_parametrizations_weight_original0 = None
	        view_3003: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_31_fc2_parametrizations_weight_original1 = None
	        view_3004: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_31_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1727 = torch.ops.aten._assert_tensor_metadata.default(view_3002, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1727 = None
	        convert_element_type_1150: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3002, torch.float32);  view_3002 = None
	        _assert_tensor_metadata_1728 = torch.ops.aten._assert_tensor_metadata.default(view_3004, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1728 = None
	        convert_element_type_1151: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3004, torch.float32);  view_3004 = None
	        sub_8769: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1150, convert_element_type_1151);  convert_element_type_1150 = convert_element_type_1151 = None
	        mul_18564: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8769, view_3003);  sub_8769 = view_3003 = None
	        view_3005: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_18564, [1280, 5120]);  mul_18564 = None
	        _assert_tensor_metadata_1729 = torch.ops.aten._assert_tensor_metadata.default(view_3005, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1729 = None
	        mul_18569: "Sym(1500*s6)" = sym_size_int * 1500
	        view_3006: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_3001, [mul_18569, 5120]);  view_3001 = mul_18569 = None
	        permute_320: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_3005, [1, 0]);  view_3005 = None
	        addmm_159: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_31_fc2_bias, view_3006, permute_320);  model_audio_tower_layers_31_fc2_bias = view_3006 = permute_320 = None
	        view_3007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_159, [sym_size_int, 1500, 1280]);  addmm_159 = sym_size_int = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:217 in forward, code: hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
	        clone_256: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(view_3007);  view_3007 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_29402: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_29104, clone_256);  add_29104 = clone_256 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:365 in forward, code: hidden_states = self.layer_norm(hidden_states)
	        clone_257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_29402, memory_format = torch.contiguous_format);  add_29402 = None
	        var_mean_64 = torch.ops.aten.var_mean.correction(clone_257, [2], correction = 0, keepdim = True)
	        getitem_256: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_64[0]
	        getitem_257: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_64[1];  var_mean_64 = None
	        add_29407: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_256, 1e-05);  getitem_256 = None
	        rsqrt_64: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_29407);  add_29407 = None
	        sub_8775: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_257, getitem_257);  clone_257 = getitem_257 = None
	        mul_18580: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8775, rsqrt_64);  sub_8775 = rsqrt_64 = None
	        mul_18581: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18580, model_audio_tower_layer_norm_weight);  mul_18580 = model_audio_tower_layer_norm_weight = None
	        add_29408: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_18581, model_audio_tower_layer_norm_bias);  mul_18581 = model_audio_tower_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:451 in get_audio_embeds, code: audio_hidden_states = audio_hidden_states.reshape(-1, self.config.audio_config.intermediate_size)
	        view_3008: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(add_29408, [-1, 5120]);  add_29408 = None
	        sym_size_int_193: "Sym(375*s6)" = torch.ops.aten.sym_size.int(view_3008, 0)
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:389 in forward, code: hidden_states = self.linear_1(audio_features)
	        view_3009: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_3008, [sym_size_int_193, 5120])
	        amin_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.amin.default(view_3009, [1])
	        amax_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.amax.default(view_3009, [1]);  view_3009 = None
	        full_384: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.minimum.default(amin_192, full_384);  amin_192 = full_384 = None
	        full_385: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.maximum.default(amax_192, full_385);  amax_192 = full_385 = None
	        sub_8787: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_192, minimum_192);  maximum_192 = None
	        div_384: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(sub_8787, 255.0);  sub_8787 = None
	        clamp_min_576: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(div_384, 1.1920928955078125e-07);  div_384 = None
	        div_385: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(minimum_192, clamp_min_576);  minimum_192 = None
	        round_385: "f32[375*s6][1]cuda:0" = torch.ops.aten.round.default(div_385);  div_385 = None
	        sub_8793: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_385);  round_385 = None
	        clamp_min_577: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8793, -128);  sub_8793 = None
	        clamp_max_384: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_577, 127);  clamp_min_577 = None
	        _assert_tensor_metadata_1730 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_576, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1730 = None
	        _assert_tensor_metadata_1731 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_384, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1731 = None
	        convert_element_type_1152: "i8[375*s6][1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_384, torch.int8);  clamp_max_384 = None
	        view_3010: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(view_3008, [sym_size_int_193, 5120]);  view_3008 = None
	        view_3011: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_576, [sym_size_int_193, 1])
	        view_3012: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1152, [sym_size_int_193, 1])
	        reciprocal_192: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_3011);  view_3011 = None
	        mul_18613: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_192, 1.0);  reciprocal_192 = None
	        mul_18615: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_3010, mul_18613);  view_3010 = mul_18613 = None
	        round_386: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.round.default(mul_18615);  mul_18615 = None
	        add_29476: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_386, view_3012);  round_386 = view_3012 = None
	        clamp_min_578: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29476, -128);  add_29476 = None
	        clamp_max_385: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_578, 127);  clamp_min_578 = None
	        view_3013: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_385, [sym_size_int_193, 5120]);  clamp_max_385 = None
	        _assert_tensor_metadata_1732 = torch.ops.aten._assert_tensor_metadata.default(view_3013, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1732 = None
	        convert_element_type_1153: "i8[375*s6, 5120][5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3013, torch.int8);  view_3013 = None
	        view_3014: "i8[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1153, [sym_size_int_193, 5120]);  convert_element_type_1153 = None
	        view_3015: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_576, [sym_size_int_193, 1]);  clamp_min_576 = None
	        view_3016: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1152, [sym_size_int_193, 1]);  convert_element_type_1152 = None
	        _assert_tensor_metadata_1733 = torch.ops.aten._assert_tensor_metadata.default(view_3014, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1733 = None
	        convert_element_type_1154: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3014, torch.float32);  view_3014 = None
	        _assert_tensor_metadata_1734 = torch.ops.aten._assert_tensor_metadata.default(view_3016, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1734 = None
	        convert_element_type_1155: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3016, torch.float32);  view_3016 = None
	        sub_8813: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1154, convert_element_type_1155);  convert_element_type_1154 = convert_element_type_1155 = None
	        mul_18634: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8813, view_3015);  sub_8813 = view_3015 = None
	        view_3017: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_18634, [sym_size_int_193, 5120]);  mul_18634 = None
	        _assert_tensor_metadata_1735 = torch.ops.aten._assert_tensor_metadata.default(view_3017, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1735 = None
	        view_3018: "i8[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_multi_modal_projector_linear_1_parametrizations_weight_original0, [3072, 160, 32]);  model_multi_modal_projector_linear_1_parametrizations_weight_original0 = None
	        view_3019: "f32[3072, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_multi_modal_projector_linear_1_parametrizations_weight_original1, [3072, 160, 1]);  model_multi_modal_projector_linear_1_parametrizations_weight_original1 = None
	        view_3020: "i8[3072, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_multi_modal_projector_linear_1_parametrizations_weight_original2, [3072, 160, 1]);  model_multi_modal_projector_linear_1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1736 = torch.ops.aten._assert_tensor_metadata.default(view_3018, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1736 = None
	        convert_element_type_1156: "f32[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3018, torch.float32);  view_3018 = None
	        _assert_tensor_metadata_1737 = torch.ops.aten._assert_tensor_metadata.default(view_3020, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1737 = None
	        convert_element_type_1157: "f32[3072, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3020, torch.float32);  view_3020 = None
	        sub_8817: "f32[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1156, convert_element_type_1157);  convert_element_type_1156 = convert_element_type_1157 = None
	        mul_18639: "f32[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8817, view_3019);  sub_8817 = view_3019 = None
	        view_3021: "f32[3072, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_18639, [3072, 5120]);  mul_18639 = None
	        _assert_tensor_metadata_1738 = torch.ops.aten._assert_tensor_metadata.default(view_3021, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1738 = None
	        permute_321: "f32[5120, 3072][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_3021, [1, 0]);  view_3021 = None
	        mm_32: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mm.default(view_3017, permute_321);  view_3017 = permute_321 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_18642: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(mm_32, 0.5)
	        mul_18643: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(mm_32, 0.7071067811865476);  mm_32 = None
	        erf_34: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.erf.default(mul_18643);  mul_18643 = None
	        add_29516: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_34, 1);  erf_34 = None
	        mul_18644: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18642, add_29516);  mul_18642 = add_29516 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:391 in forward, code: hidden_states = self.linear_2(hidden_states)
	        view_3022: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.view.default(mul_18644, [sym_size_int_193, 3072])
	        amin_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.amin.default(view_3022, [1])
	        amax_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.amax.default(view_3022, [1]);  view_3022 = None
	        full_386: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.minimum.default(amin_193, full_386);  amin_193 = full_386 = None
	        full_387: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.maximum.default(amax_193, full_387);  amax_193 = full_387 = None
	        sub_8827: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_193, minimum_193);  maximum_193 = None
	        div_386: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(sub_8827, 255.0);  sub_8827 = None
	        clamp_min_579: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(div_386, 1.1920928955078125e-07);  div_386 = None
	        div_387: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(minimum_193, clamp_min_579);  minimum_193 = None
	        round_387: "f32[375*s6][1]cuda:0" = torch.ops.aten.round.default(div_387);  div_387 = None
	        sub_8833: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_387);  round_387 = None
	        clamp_min_580: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8833, -128);  sub_8833 = None
	        clamp_max_386: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_580, 127);  clamp_min_580 = None
	        _assert_tensor_metadata_1739 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_579, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1739 = None
	        _assert_tensor_metadata_1740 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_386, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1740 = None
	        convert_element_type_1158: "i8[375*s6][1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_386, torch.int8);  clamp_max_386 = None
	        view_3023: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.view.default(mul_18644, [sym_size_int_193, 3072]);  mul_18644 = None
	        view_3024: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_579, [sym_size_int_193, 1])
	        view_3025: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1158, [sym_size_int_193, 1])
	        reciprocal_193: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_3024);  view_3024 = None
	        mul_18666: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_193, 1.0);  reciprocal_193 = None
	        mul_18668: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_3023, mul_18666);  view_3023 = mul_18666 = None
	        round_388: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.round.default(mul_18668);  mul_18668 = None
	        add_29572: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.add.Tensor(round_388, view_3025);  round_388 = view_3025 = None
	        clamp_min_581: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29572, -128);  add_29572 = None
	        clamp_max_387: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_581, 127);  clamp_min_581 = None
	        view_3026: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.view.default(clamp_max_387, [sym_size_int_193, 3072]);  clamp_max_387 = None
	        _assert_tensor_metadata_1741 = torch.ops.aten._assert_tensor_metadata.default(view_3026, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1741 = None
	        convert_element_type_1159: "i8[375*s6, 3072][3072, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3026, torch.int8);  view_3026 = None
	        view_3027: "i8[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1159, [sym_size_int_193, 3072]);  convert_element_type_1159 = None
	        view_3028: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_579, [sym_size_int_193, 1]);  clamp_min_579 = None
	        view_3029: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1158, [sym_size_int_193, 1]);  convert_element_type_1158 = None
	        _assert_tensor_metadata_1742 = torch.ops.aten._assert_tensor_metadata.default(view_3027, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1742 = None
	        convert_element_type_1160: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3027, torch.float32);  view_3027 = None
	        _assert_tensor_metadata_1743 = torch.ops.aten._assert_tensor_metadata.default(view_3029, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1743 = None
	        convert_element_type_1161: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3029, torch.float32);  view_3029 = None
	        sub_8853: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1160, convert_element_type_1161);  convert_element_type_1160 = convert_element_type_1161 = None
	        mul_18687: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8853, view_3028);  sub_8853 = view_3028 = None
	        view_3030: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.view.default(mul_18687, [sym_size_int_193, 3072]);  mul_18687 = sym_size_int_193 = None
	        _assert_tensor_metadata_1744 = torch.ops.aten._assert_tensor_metadata.default(view_3030, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1744 = None
	        view_3031: "i8[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.aten.view.default(model_multi_modal_projector_linear_2_parametrizations_weight_original0, [3072, 96, 32]);  model_multi_modal_projector_linear_2_parametrizations_weight_original0 = None
	        view_3032: "f32[3072, 96, 1][96, 1, 1]cuda:0" = torch.ops.aten.view.default(model_multi_modal_projector_linear_2_parametrizations_weight_original1, [3072, 96, 1]);  model_multi_modal_projector_linear_2_parametrizations_weight_original1 = None
	        view_3033: "i8[3072, 96, 1][96, 1, 1]cuda:0" = torch.ops.aten.view.default(model_multi_modal_projector_linear_2_parametrizations_weight_original2, [3072, 96, 1]);  model_multi_modal_projector_linear_2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1745 = torch.ops.aten._assert_tensor_metadata.default(view_3031, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1745 = None
	        convert_element_type_1162: "f32[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3031, torch.float32);  view_3031 = None
	        _assert_tensor_metadata_1746 = torch.ops.aten._assert_tensor_metadata.default(view_3033, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1746 = None
	        convert_element_type_1163: "f32[3072, 96, 1][96, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3033, torch.float32);  view_3033 = None
	        sub_8857: "f32[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1162, convert_element_type_1163);  convert_element_type_1162 = convert_element_type_1163 = None
	        mul_18692: "f32[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8857, view_3032);  sub_8857 = view_3032 = None
	        view_3034: "f32[3072, 3072][3072, 1]cuda:0" = torch.ops.aten.view.default(mul_18692, [3072, 3072]);  mul_18692 = None
	        _assert_tensor_metadata_1747 = torch.ops.aten._assert_tensor_metadata.default(view_3034, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1747 = None
	        permute_322: "f32[3072, 3072][1, 3072]cuda:0" = torch.ops.aten.permute.default(view_3034, [1, 0]);  view_3034 = None
	        mm_33: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mm.default(view_3030, permute_322);  view_3030 = permute_322 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py:83 in forward, code: return audio_embeds.unsqueeze(0)
	        unsqueeze: "f32[1, 375*s6, 3072][1152000*s6, 3072, 1]cuda:0" = torch.ops.aten.unsqueeze.default(mm_33, 0);  mm_33 = None
	        return (unsqueeze,)
	        
V0910 09:42:39.079000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2ed437d203944b9f88ebd4a0a71b94a2"}
	{
	"name": "_recursive_joint_graph_passes",
	"ts": 1757522559079854.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.794000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "624a78aae2362e866463f3008083f7ee"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561794148.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.796000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "748ae189cccf0ac4acfbe3bfdba1ef00"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561796167.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.797000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2151f12a143455e6afd9c70f5249b9f3"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561797768.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.799000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "69972df6917d51f1fa840dced8c0f321"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561799122.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.809000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d73326f877ab201e004148e3f000dc11"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561809384.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.810000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "64646944c96aedddbba921a80781ae91"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561810776.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.812000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3e4cb99cac9e84ea8ccefde0a0578b6a"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561812252.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.813000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "761f3ecb15775ce0737a4e606985ef08"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561813565.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.826000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8369a550aa29a2bda308e271a8ba7b5c"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561826638.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.828000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "55b4997f7c04f63f4c16de921b0485c3"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561828048.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.830000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "06ec7be2482a4e1b272551170367f3d4"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561830533.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.831000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d07c591f685db4b720d9d25c1f7dd9c2"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561831933.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.845000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b9a1c767dc81f15ea27da01195312191"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561845092.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.846000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "26263cdf28d5550459a07e8b41822fd1"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561846571.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.848000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0bf09779fec24ed3224b88b47f3bc861"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561848092.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.849000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "dad6200793b2697c90a41f64019292f0"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561849460.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.862000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0ddd3b264b1456ae87c3a73b49dda559"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561862153.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.863000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5e90ba544feed1c456ac3e805dbbc1d6"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561863589.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.865000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2d0f39f9f86b4c143e44b3cfe69894ae"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561865106.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.866000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b759a37460cf9c6de8f0258112a1d5fc"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561866464.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.880000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4cbed7e1cbd4fa1aa11cfc1763b920eb"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561880671.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.882000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5f2f5e19a820ef115e4131d55741ada6"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561882170.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.883000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e8b3e98b173f5aacf3344d7306bc3722"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561883700.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.885000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "34caa860f54a4c18a8f03a1a9aeeb206"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561885062.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.897000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9f0cf85e3eaa577148a3b52983724a48"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561897367.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.898000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "87ceff58dbc892b0bfa431b4fdf54b0e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561898760.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.900000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ea0703d12680421b535a6a295830dacc"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561900208.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.901000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c59522acc3c8572b0eb57aa44bc8fd82"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561901804.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.914000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d6d4007adcede23937cb6a4713d05afb"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561914632.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.916000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e0a06219def057aeeb0cf35578b85f29"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561916021.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.917000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b51da0dbc6e3e2f9063a37eab75946e9"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561917520.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.919000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "67c0ddde36695901b7aa653bf8a9c6f0"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561918957.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.930000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "44683169a8ca8d34c533c2a1d0cc12e0"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561930601.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.932000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ab8263ac44a81714df1793c0019a9b7c"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561932005.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.933000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0b757491b2d355ecfaabd9ebc4eeb537"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561933463.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.934000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "32722be3e6921191decf0d159a0db32e"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561934879.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.946000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5d48862ad0d3a0aff0551d17f805619d"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561946578.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.948000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a8e84aa662672585cb59ef98c6630b54"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561947958.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.949000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f4f6f31c369d4c829cc56aaec2230f8b"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561949408.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.950000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2e633f61ac17fa4cea0f3a265c75546d"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561950714.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.962000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c5f5e33aa2ce8fde49b544bdbbae26ff"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561962279.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.963000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0f03764d48193805804c5c7294191b0e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561963656.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.965000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "96bad7c2851450e2cfdf0de3a2108299"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561965151.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.966000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e0992691c74d1d8dd08c5e9668256b66"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561966572.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.980000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2f743e952a9785678c323b7e9259fe99"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561980352.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.981000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7cadbc3ed45d81d928d2ce967adc5770"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561981929.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.983000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f0167c12b673d051ba78394de60b56c5"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561983423.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.984000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "bc080f9ce1745b198b7019724f90129b"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561984754.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.997000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f8902ea57271ddd39f73f756641df381"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522561996974.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:41.998000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f30fbab2d524f2712542e2999ff168ed"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561998391.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.000000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4b29696aff65cb4de57ee9e420058059"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522561999973.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.001000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e0ca1811515710559d237de91d6dd3e9"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562001335.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.014000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "36500ddc3bfb33535f402fbf45884e13"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562014845.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.016000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "36cc75ff3b986c4f1a1d19e114eec74b"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562016335.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.017000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "365d7b4b114bb5306335f63d67d8317e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562017888.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.019000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3b63fa637800e31a80d4fc24fd1a81b0"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562019300.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.031000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c98f3102e8e3dc4726eae0129fae5b7d"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562031919.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.033000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ba9f49695e5c9fb04c42933171eb4f3e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562033392.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.035000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e42e70ec8a2e989d76db7ffa3406e584"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562034950.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.036000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1e45decc317b29dbc6cb6e183c2b2ebd"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562036344.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.049000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1d6e3282f0f3a7b0df8e8552dc36b504"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562049056.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.050000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5be7d9d86729bf3fcc2e14fdc8592e4e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562050532.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.052000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "599c3ffe7051cdae1e61ae674d90e9f8"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562052221.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.053000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9b40f2005f73402a04452e2c3a7535e5"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562053522.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.065000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "867f6164d55ce9226922905acd2bc77d"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562065328.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.066000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d0cd6f3fbc5e1166c94157a4d1d026d2"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562066760.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.068000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9bd0ce1b2d11c7efde1b570616d81aa0"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562068280.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.069000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "152e32f7c622a974b2ff7a96d870b2f9"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562069582.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.083000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d9e243f3bf7cc343cafdf53b578126a3"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562083563.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.085000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2c0e1e62b57848ae70cd573afde1f672"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562084957.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.086000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "71825c0fb5be903f22ed62d446bae779"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562086489.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.087000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "597201387530adf74d23a9dcac6e54ac"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562087887.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.100000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "060271876a30c4a01f4dc1bca0d78f39"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562100452.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.101000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8ea1eab5208fea2300fb8ea109613c51"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562101885.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.103000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "483c3d27130383227d4048c741c10a06"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562103420.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.104000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6e1232c682bcbb838553ffa503ebb564"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562104783.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.117000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d56cdd63b4dbf9d51739c8077fddb728"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562117756.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.119000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3de64908e0b3b3abe314430787d03398"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562119193.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.120000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a42914994fa9094c49bd8b520b9a148c"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562120684.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.122000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "070d73182ed29b99df7f011c7ac8d34d"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562121972.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.134000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "30c97c7f905d6be0a862ac7e63e0166a"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562134099.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.135000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "612f8968290f5f501ab0c9bf304e2c67"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562135762.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.137000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b9d8bc487b29febe1bdbcd89f428fd56"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562137258.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.138000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "178431053050f16f00bc5f0e0ed0d017"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562138597.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.150000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c187a4cb8f63885e1325f9261d8e0ea1"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562150857.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.152000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7cb69795219679450554385681f09c53"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562152282.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.153000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2d3dc5287e8397ff8f07b3e982367274"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562153808.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.155000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "02d1c7d6a24e66087955ed12530ab66b"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562155152.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.167000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "58a591f9a7e4303aa29592e5456382f9"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562167156.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.168000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "33f337a34db949519b1953c974ee375a"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562168606.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.170000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3aa4cec6559f02ee35fce37dd59b8246"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562170105.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.171000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "604d92085e9e4d5664781ccb67c22d80"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562171441.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.185000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3495a3041ee2d6c216a8c400dc264c54"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562185640.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.187000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9ee2df28126587b2f080b3e14d4b74a8"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562187142.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.188000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "fb012a3f1be76e42ce110891814c39bd"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562188653.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.190000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "dd89c7214f8cfaac48b0388957618572"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562189995.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.202000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3a60f54af56079e5ca3b78c40ba4aab6"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562202703.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.204000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f9dc3ffe333d6fbfa607dce96d8b01d6"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562204166.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.207000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "290cf13b1694486673d2bdb11ca8e766"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562207215.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.208000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "dad36cb58c63ecef814d5800588a6d8b"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562208572.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.221000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "35be38275a727c3142850c9eba0229f3"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562221603.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.223000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2e2da76d1fbad1ff9f4f9c3edab52396"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562223090.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.224000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0f2c7bebf877efefaa1a9f1d639c97e7"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562224601.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.226000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3a01abb1c6e4300614b13c8156cad998"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562225962.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.238000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "112c73c38f345b6fca2816cb21b249d7"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562238231.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.239000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "31c8a74aa3f97aae483cb7bcedfeebc0"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562239709.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.241000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4a3846215cf507adbe83c67d57741753"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562241199.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.242000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4611678a62dde92e2f87a153939aa004"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562242529.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.254000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d0c81d1d840db56c04910f6de0fc9688"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562254752.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.256000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4dcbb8209ead7887aae14265726a92e1"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562256214.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.257000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "74a64d57e123535f551938784c6dce95"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562257751.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.259000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "375d7b02a1a64c5178df79df68205ec3"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562259128.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.271000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a82a7ca020830f461bd5e4fe7a996d3a"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562271212.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.272000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "bbf1c498f63044b28b326c8508b09d10"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562272642.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.274000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e0a04dca395228148af7ec7ff72d653d"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562274168.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.275000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "168d103ee468b7b5d8708e39eb426b21"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562275540.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.289000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5e276b525ad13864b2457b45e15d76f0"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562289789.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.291000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d393754262c98204661136d1960042a5"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562291398.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.292000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "60c52b449293da0089a2bcb148d52036"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562292913.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.294000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2ae38abdf0640ec2ebd870946458310b"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562294274.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.307000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3b37d897dfe6bcec3b459c6af4ff040a"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562307070.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.308000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "baba6fe47ac3956e9b3f663938b909f3"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562308544.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.310000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "987b41d2c10b42437f08f29a25a34062"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562310135.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.311000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9f5374b3169f790bf909882d4a10d540"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562311479.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.324000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f7140d0e342a20c9ef1b004ea2b34f7a"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562324657.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.326000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9ab9c0cd76960dd6e72349ff7f15d987"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562326077.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.327000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d46e02abf8c94fe8260d5c1218974a68"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562327573.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.328000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ede8fb4ecd027c06f1aaf9b500ce8373"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562328925.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.341000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b1921588f2ea955fcc40381ac5ebb09b"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562340939.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.342000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "df3456f812838481c1f5a9d6622dca7d"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562342400.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.344000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ab0179924f315a01be09b544e367bddd"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562343974.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.345000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "22a0dae708c8930772b325df2e4d2844"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562345331.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.357000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "32192d6dc5b6532291127eb6b0f1aede"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562357530.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.359000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "35f9b01ade4fd69e9e4773f824b27830"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562359032.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.360000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5ea82f200c11b97bf839ad6325c9181e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562360716.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.362000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "27a0c47102937ec229f7a98374085458"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562362080.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.374000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "96a2850028010f12835b3ee5f079da73"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562374600.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.376000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "59b38db8e54fe30b8ef2d80019e48274"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562376119.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.377000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "cca6390e22bbb3fae2c885e526997051"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562377693.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.379000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "86f155cb82fd0a976f262f38395fb009"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562379095.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.393000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f31f02a6f3151dfd742ebf3a15cf0f01"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562393536.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.395000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9e2ef288e496d7150a56809a9c34f618"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562395045.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.396000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6fa652942f388af14d765afc99fb2310"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562396556.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.398000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9682601597883faa2a60dd8ebeb88f82"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562398001.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.411000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0900c6a9906d9d364efcc63eac1cd38b"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562411073.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.412000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1639f90ba8e3d704b7e888bbb2c32888"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562412574.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.414000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1d32ed1a3bd20e9fc2825848b4d7ab6e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562414134.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.415000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "029da597539cab96e25fc9f3424827c1"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562415530.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.428000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "be12e6c43e1e960bad86844d5258462c"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562428477.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.430000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3a667637dc1a8d4ec8013403e4ead1d1"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562429963.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.431000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c79645a430d83e78fdde86945923ff2b"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562431480.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.432000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "24e4d71cd494dc1168afa103b4d534e3"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562432810.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.444000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b8e1b959b0947a64e6af365b59deb030"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562444585.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.446000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c5a0de33683c37769939d0b306ae85d0"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562446225.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.447000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "adc2fb3c34e684d512fcda302674c1e2"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562447690.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.449000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ef81bd69e5d7389e04fafd519312a82e"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562449041.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.460000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7f2796c39596c6d0e05a5cd7cbcee246"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562460907.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.462000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e4cf406815baebc8163afc5d2068d8ee"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562462426.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.463000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "643d44adfa02c142578be89855de2424"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562463897.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.465000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7abf0b9ee2d6e123c424c09ae03ef94f"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562465209.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.477000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7091ae39ef78709957506242ff5afa4e"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562477348.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.479000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8e7ee4df7c858261f5d66219f3bb6baa"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562478937.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.480000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4e57dda9919859a35608a644b1c2c3a7"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562480516.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.481000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "dcf48d0a6db30ec9fa54ed204e273994"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562481860.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.495000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "861d2b709f60eb27abdd5d2f6f15a1af"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562495765.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.497000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "dc502e46e6228ba51aa36bd64629f981"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562497203.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.498000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2b5ae14156f4874974da8887d28db891"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562498713.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.500000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9eee9a58fe71dcbcd39e8527f8a58cd5"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562500083.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.513000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ccfb24278d4beaa63fc365e49a412c73"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562513285.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.514000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6a7de8336b09808de86cd135d7d395e2"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562514795.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.516000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b6634bb9c48655683696519263759f60"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562516370.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.518000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "bab2c974d6a7817c15f4b6ba6abd8334"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562517987.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.531000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "14b300138bb3efbb7c36836783a18042"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562531171.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.532000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f31162d3429c8064180d3432ed5b4e68"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562532601.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.534000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e6dc59d87be0b075afb1aac3b2cea644"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562534072.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.535000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9a4c5d7a1f5c01292dfb2ceda70cbc95"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562535423.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.547000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ac2aaae81240dd7a129bcb7236537ae2"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562547414.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.548000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "30ac963654376b453880cc7ad0cb5ba4"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562548814.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.550000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "fb0c1c13fd65e7addcf2e13addaf6d76"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562550258.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.551000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b085d634fade302d9b39d5e181b4cf4c"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562551563.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.563000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "fba86e67c1378f2f142391a519b106c8"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562563702.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.565000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "30012440b445620e31fd49812e09b3ea"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562565258.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.566000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6c56fbc58509bd51fe794986c8382d82"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562566814.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.568000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "335a27e38b812a2503261bfe163438a2"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562568207.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.581000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f3c0efd43a1fc3c925d1225a0216ffb4"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562581063.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.582000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9e41d26aa8221a9d551f10afa9e83111"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562582853.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.584000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "cbc7e4d2be22a8605fe54681b5d5ce7c"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562584492.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.585000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b0970417a33b317ca908078cceb61f9c"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562585885.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.600000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "85201a1f9cff3ca088d88d75c11dd88a"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562600218.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.602000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d183a1f06f1e7ea7da36686244c247c4"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562602216.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.603000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "dbcf318b0c4faaed320bc1cc0bbe6744"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562603771.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.605000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4427ca425c9bf8c0f42147d6e5763d4a"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562605143.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.617000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7229ac8a8cb8881158eb0d726c22051b"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562617839.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.619000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6e58d5c1ec3956e5c59c09dac14c6b26"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562619442.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.621000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "01a2379d9089b9fbddec6863a0d41785"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562620985.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.622000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "fbd9499f1c26cf5913c719ac12cdcb0a"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562622328.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.635000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0b659837f9dc948114d3bb7f6fd2a592"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562635914.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.637000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d0a89e42ae244edb9e66548862dcef84"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562637600.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.639000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "39b02e5f5a3e0a8aeb3ddce636055c65"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562639182.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.640000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3c21450849799cc0b0c4e69429d5e53d"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562640580.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.653000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6d7dd9afea13242a43735a621a9cce58"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562653262.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.654000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "eaed4364b1c4549aced121e2351086f1"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562654770.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.656000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7a933f682dd5113cfbefe1fc4bfc6c3d"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562656330.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.657000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "92da30eca59fd87fa47f38772b7c572f"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562657773.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.670000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8bfe8981df6748337a5d3586990329fc"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562670279.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.671000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6abb0f399281da64fd34ce4e91dfab7f"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562671775.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.673000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5d509e7939c64fa14acafe3c4bf2dbcd"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562673536.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.675000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5d0e839a5fe01aa91bbc83736d359a16"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562674965.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.687000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "aa33d5eb81c77effdca9cbc578f81eea"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562687253.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.688000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d82836b2fefd8fa9b1eaa697ba5bd857"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562688683.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.690000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "fcfdb0b32e0895db1cc25c1b1f76e01f"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562690252.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.691000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "067f53e17f729217bb493aafa7d61b44"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562691637.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.705000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "cb1c62734f3f50b4724a0202fd2324fd"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562705508.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.706000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a82a9f5c73d2ea8cdb991b67549dfc0a"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562706920.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.708000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2774c0e2fe48d1c7e57d35c1c10b9ea0"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562708383.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.709000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c563dd6e7ea14cad87460068f21a4c61"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562709715.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.722000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b20113605e1f178616d10ffc6f908f19"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562722083.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.723000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "20c6042dfe4e79ce2a3c3c186c29f005"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562723482.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.725000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b584cf10c2ab832f3b2d7be9b3313da9"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562724957.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.726000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "798c034f0d896de01ba3dfbf09a00247"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562726348.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.739000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "77d90b30ca12f85447fd745cd2b04d79"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562739047.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.740000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "94d26ee83d19e1933e37be95b3039999"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562740419.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.741000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ae11218ed16b52692db22fedc665434f"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562741869.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.743000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ddb247ecb67f4d0453520426e0446003"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562743173.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.754000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b9cbd87950dd5fe599dbdd914ca11f88"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562754908.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.756000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3db76fb9649da3d642ed60f9a1119a43"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562756555.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.758000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7ae71e60700e6ae30878850b34432276"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562758177.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.759000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c2c37bf5c45f36939954bd3564d02198"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562759508.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.771000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f6f8d92ebb76b0f9741dce8f54d7a981"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562771454.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.772000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "bb1164382dd8260cfb07ca494fe03382"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562772858.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.774000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ecc8f3b6488db8d9b13687487392e58c"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562774336.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.775000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9eb576f282ba8e4623f5294d9b6ea3d8"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562775648.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.787000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b696c82bae287567cbb9f61ab7a97557"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562787931.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.789000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ec021e3293a76989c2dea0a2af26211a"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562789313.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.790000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "cde0b46cc84e93461ebdf30b08902e48"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562790777.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.792000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0daa9e501ba1ae91576192d98651c722"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562792088.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.805000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "86dedc8d2130aed3e91ce137bed8f4e5"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562805818.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.807000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b713442c8ae2528a3b9a8566ffebcc6c"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562807212.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.808000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c176e7f5f16f7e576d0c9e80317be44d"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562808676.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.810000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "870350289c1636ea6ba6fd7b3d344fc2"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562810016.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.822000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4a90d0d485fc6043e6c7f509e6c06ed5"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562822222.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.823000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "62afcfad9b71ff3368aa6e0f2964411e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562823599.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.825000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "53837c92b9417fd50970fda5ed03fd00"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562825074.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.826000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "570a34ecdcab21d4a247dae61dcbc649"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562826561.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.839000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4587c4253988f7c2559b8cdd7754ba5c"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562839203.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.840000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d2fe341579f90b1c3950cde02a469a0b"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562840620.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.842000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "78c0b5c10190443de0736cf4601d8eeb"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562842088.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.843000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "232366330bff85c27154031820c5d067"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562843392.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.855000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "64b907877a7e391dc84eb82b62288212"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562855189.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.856000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5d23222b2ece5098e03819d19b7ecf8e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562856558.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.858000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "98df2ee3adb93b17cb25195dd1441be1"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562858024.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.859000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "dbc4d24b5a9811332298741a18b2d9f7"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562859325.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.871000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2e551a5e4844620e7f428c4a585da6d5"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562871129.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.872000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "fcc5648a2df7c79639cb36d3532d29e8"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562872538.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.874000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ac8e72a4a006c3560e792ec5f92b314c"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562873994.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.875000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "36fa4e3dfd2466b3a4d3497ae18b37b6"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562875338.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.887000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "02fc7bd8268b94cb9bb10a4efae8e4cf"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562887515.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.888000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "540362a364d438b0adb33ab6aac15293"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562888901.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.890000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "099d78d7d1fd11f59b5c6bded7ff8782"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562890361.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.891000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ee0cfe52fdc7aaca724f88ba5daef53c"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562891676.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.905000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9d26c4bdcc53ef2515f92f63b0c3ff36"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562905506.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.907000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f7c4d5669210f1c15018dae7ca9db0dd"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562907079.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.908000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0497e2387868dfeb53a484d9bdd84598"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562908550.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.910000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8c0930707ff4f4f19cdeeff91cfa6702"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562909956.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.922000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "36cd24fa5168dbfb4d8c99132f082235"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562922743.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.924000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "96d23a2c3f3000db3e59cd4dcde892d2"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562924135.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.925000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b6cc2a7280d251069dee500e3d8bdbbb"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562925652.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.927000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "caf7cd09a7d5561c4754e0c12adcf5f4"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562927001.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.940000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3b3e293ec1d204b6e10eada5dfe418e6"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562940024.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.941000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7bc70d85de4c41ff7d729b6a5a35bd30"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562941460.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.943000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "269fa32973ec16b8e9470b758ca03b94"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562942978.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.944000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "bb92199b586ce0a1e07516d4462d2c36"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562944319.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.956000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2273b8190b6de065c9f24dabd00f143b"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562956434.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.957000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1950f9e24c497ff73fb35c3b398534be"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562957889.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.959000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4ed5b554c492eaf89cbe22e5682cf05a"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562959380.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.960000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "09349fc6f2a7ea74d80f9495a46c375e"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562960730.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.972000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a8fe299e9591ced536a8d544b2513af5"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562972910.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.974000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "86e9abf4293db2be2c5bc0a6180a5537"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562974390.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.976000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8e4b8c9cff3223bd3c12e6b0d3872723"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562976135.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.977000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "712934af7590022352f01246b8a3e18d"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562977504.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.989000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5d37ddedbcff7fea3b8535eec1bffddf"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562989852.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.991000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7e55ef1e2b2c450a22a7ed2b1437409c"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562991276.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.992000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c815402fcb3e6933448fb2fc4f948ace"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522562992768.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:42.994000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9f435ef08f1ea41782ef89fcb4374f45"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522562994115.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.008000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c35b00f7e85ef0ac65b0431b91f0e207"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563008423.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.009000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "def9569269f9b54295b5ce877be6a705"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563009922.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.011000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d11b7a451e57646ce7827880f199403c"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563011478.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.012000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a15e39d6e2d5100ceee52b0d8cfe0a81"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563012894.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.026000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ef2e8cb5f570dcb0ff2ef92e5524e2d6"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563026133.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.027000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c99a3ca9b877d3947f0c6aab2db049f6"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563027568.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.029000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b295cfed2d80540188447c0224c2ad40"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563029102.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.030000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "eb740b752f69180be9540dd8e3c7d313"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563030456.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.043000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f4ca93011e36637df25858472c2a711c"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563043672.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.045000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "db28fe28c23c03c1b80e1e5d13019460"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563045132.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.046000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "76d349a693c36ccdbe65c8dd51d95164"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563046661.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.048000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "378837da8c810442cff828f13246a500"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563048044.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.060000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9c418a57aa7606c6055d9f25a93d9fb4"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563060358.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.062000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "cb2366d838628ae503a7ca6f3587ae66"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563061995.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.063000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "622d37e9b7a55cead46d77967e14806d"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563063567.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.065000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e6a9bc535d0e8908f3ad9ab44e5ab724"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563065000.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.077000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c223f327d75e8c477edb65d9b3003886"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563077348.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.078000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d0c0b5570ea41c2dabddd426415e5da4"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563078842.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.080000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8ace8d6f0629f382177ead45f50d0410"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563080404.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.081000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6af3a8fd6e7a5fe61bed9fb13df5d6e1"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563081791.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.094000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "54c7a06a57ce7f0838a116a55515e0c8"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563094140.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.095000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6d529976ed17b9e49be2a348b8035e56"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563095564.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.097000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8322ee3e532a16b1023503c00a8e99fe"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563097138.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.098000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "33461725f41d3322e32529a2f69da032"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563098483.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.112000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5df877dcdeaed16487d051353821310b"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563112848.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.114000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "dd4c9960d47aa021a4048ea24d0192bf"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563114341.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.115000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d3e2fe0d18879666e810ecea8cc2bcb7"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563115888.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.117000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "01dea1ae21fef3a7d6ed91ebcc526b42"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563117286.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.130000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "639e7adfc625d2fb40444fac62443cdf"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563130243.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.131000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8dc9c038da4429fa3e7537a5db961def"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563131687.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.133000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6cb7777d47281d1fc14fe1930ab02a57"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563133363.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.134000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "279740edb184e1ecc8157381765037ed"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563134789.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.148000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1019e6178b7909d57280f3968c87b380"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563148158.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.149000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "deeb2aaf73546085345d23d77e892190"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563149655.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.151000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "27cc7571a96415e3825b8e70a85bd439"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563151193.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.152000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4e92ef7afeefc09dbcbf32bd0a6f7366"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563152565.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.165000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c9ff171bf1e7bc30d92a138c7df3e141"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563165014.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.166000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "81870fe79080afe000d13867ae04308a"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563166478.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.168000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c16756decff4eb7e69b02cce43f5a384"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563168047.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.169000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1b38326d2a4289f0c419f07169a516dc"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563169456.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.182000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "edf58e199243ad80e560fc9ad946708d"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563182037.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.183000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "95992b74342e85af89fd7209a781ba3c"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563183538.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.185000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c533cffb7791aa208bf02da82020a226"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563185117.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.186000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "bc5ed6bad4288276432ec080e35dd2bf"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563186522.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.199000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6f838201fc98c16eaa35738b6f6bb000"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563199074.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.200000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a4076f71de3c7d0a856597284e447280"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563200535.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.202000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "02ac0e4350ede532b064334b64ed0d87"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563202089.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.203000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a7fc65a61fc9d36c13987fc1d587b132"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563203473.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.218000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1e9d62d3532c635b691700e374e0482b"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563218096.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.219000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "24ac5ca0eee545d5db2c3d2b962f421a"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563219805.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.221000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4b2772493f120c976915ba8a35b83650"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563221370.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.222000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a0c60975e59aa64942b667cb38106742"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563222792.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.236000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a9470b40e43582e8d6a18128c96f4924"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563235965.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.237000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "561cafff4a84ebe4e630a4529bf491ce"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563237417.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.239000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "413f5710705876da9d8cb4f352da2bf0"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563239032.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.240000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1b560fb69766c7bc489c152a8edd3d54"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563240387.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.253000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "051431c30087425187fd8c092fb5a36e"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563253526.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.255000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a0da401127ea2b887c89fc59edadde13"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563254970.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.256000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1bada83e35b7b98d9f7fc438b01b50d3"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563256499.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.257000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c794542a68ff6d4b7479d5a0567e2f77"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563257813.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.270000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "23311777f9904bc1c2e67da4896e0622"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563269942.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.271000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5d0703eb96fe13aa70414f5cda6a99e1"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563271432.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.273000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "250b72cd8f0c610446195e56c28465ec"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563272966.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.274000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "cc6929638b2fb5c2c99745a43cd87360"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563274481.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.286000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9cf3681c5b44bdc6b362eb09a9b8fa12"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563286612.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.288000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d45c4404e8de6f7e76b5049544fb0ec5"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563288118.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.289000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1d5ffc54258afff4b44c8768c4fd4099"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563289776.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.291000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "af1095765fe45031e283d435d99842c1"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563291093.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.303000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "dff71fe00c0843c275218f6ffd230ec5"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563302991.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.304000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a1507a865d82be3ecef7624471e6f5c9"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563304421.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.306000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0a31f597f338ea448ab4195830c73ae1"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563305920.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.307000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "207757c770796b4741ba70565dbc5d39"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563307287.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.321000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "570d2b4d891f6a1ddcc908b37c36e6a6"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563321821.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.323000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ff8fe92e3d9eac4758b79e654e3ef789"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563323251.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.324000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "01944214efa1a3c1f0724526302548bb"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563324758.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.326000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5afb8c8bceca84ed581e5d3077804af2"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563326111.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.338000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f145fe86046f83fd753c6a189554829a"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563338842.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.340000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "29f078d317d9dedaccb7b06e6cdf3449"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563340276.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.341000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8abc3e3fbf94f653cd796a6f86a97674"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563341848.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.343000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "487ea3120000cd7aaa0ff2d97f89a25c"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563343219.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.356000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6a8b1a6b76659919a0ab53b2d98150c1"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563355978.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.357000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5221c63e829ffba2a3b3095d5f3dd57f"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563357385.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.358000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1aa9623b5a41ddb5fc97c01109ec219b"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563358911.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.360000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f36fab6366bdee8b0579c0e186530bbe"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563360292.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.372000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "135d5ae061985e45eccbe62ea431124d"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563371952.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.373000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "78b99ff274350595c1be4821fab4c3b4"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563373549.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.375000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5ddb87246bc4e291d36e3a345cea4f6c"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563375031.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.376000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2ccdba4c4670ea373721d4b503a1927c"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563376353.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.388000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5ddc92e3f719f67995a3394884f3bab2"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563388401.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.389000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "33a6c1962dd4ccaa143c540afc369e31"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563389922.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.391000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "89ae8e4a395ad24ddf3d33867697ea25"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563391511.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.392000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f7345c496694758d5877c96b1f5f95d2"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563392852.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.404000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ccda779220db9f5a6cff192908ee9ba2"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563404820.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.406000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ed2d114154bdbfd416947c6446bc5b27"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563406242.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.407000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b4e7cdfb78f11b7a020dfbcb788d3dc2"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563407859.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.409000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c530c3bb69e4325bfb675062602f7e15"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563409292.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.423000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f397d4fac614c98d1a8d21ff6b8446c8"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563423463.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.425000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "618084abd309eb34b4f1d41f5e640922"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563424942.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.426000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a86fa7a3448f30efd3ee3e2e683ce93b"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563426465.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.427000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4cef197466eae99547a999eadac51763"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563427894.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.440000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0ec7ca12693ca1a0e29c3eea85bc3c45"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563440923.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.442000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "430ab0b5a62d4e36573e047dfc308ad2"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563442363.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.444000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "50c4f5bac32f03354e95733ca9ccc717"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563443882.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.445000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "306eacb92cd69fb20a439949c31d4587"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563445370.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.458000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "29889c61a4963771a4bcccfeaa63782a"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563458151.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.459000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "07487941b1a5ad11c70effd860efb7f9"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563459561.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.461000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9c03b9b654ea77fd57bab3a34d3d7ae4"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563461024.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.462000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "34a69c2398f3dcd31d5e58afcaf372b8"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563462321.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.473000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ea8d238860ed0beba681f80629711b5e"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563473910.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.475000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a2bf56177eaafa55dd9ee2c80177bace"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563475292.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.476000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9f4d1a3fb8fbb4823aad204065c4f476"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563476758.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.478000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "94a51d3b1075179c9af2c33504a6ac59"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563478055.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.489000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4569088f73cb43330a49c7dd27b822f4"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563489780.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.491000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e82f14aa1757b009d4e768b82fb4de48"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563491174.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.492000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1085ae97778cc7f9a49251b86e7f1069"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563492645.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.494000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "54e89a60fb8d8ade11129ec9c0aae540"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563493982.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.505000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "abfd2811566bdee0e97bfd2d7a53366e"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563505789.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.507000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9b7d2f302e64761b98a30c4bb754bd82"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563507176.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.508000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "168804c1ff12aac3b8bb1e0981d0d1c3"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563508623.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.510000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7339468be14e46adc3320da4bdba7c46"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563509941.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.523000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6b49c0b2889419c6ef6f92981ef10882"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563523447.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.525000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "bf05d05d7b87b8f8a7e3420a85976015"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563525022.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.526000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "859b255346895e39e543edfe4a5ea5ea"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563526522.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.527000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b658a158bb33f139c8bf6a007e6bb7ca"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563527937.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.540000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c4f3142b711a784e225cdc8bd22afcbf"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563540363.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.541000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "12311bb0484669338cd10c8a47154b49"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563541843.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.543000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a7cd554899af10fd748567cefcce977a"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563543395.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.544000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "99a8761e1ac198a598765791ba714f10"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563544717.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.557000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "07168e4ae7b8fac1a4f05af7706d7c95"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563557463.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.558000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8895725a740aa7a135a0a29d571e63e7"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563558849.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.560000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "40ad6fbac9988d94226e20aa4a9cb9f9"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563560301.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.561000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "13a280bee367fb1094a67a73658d9327"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563561600.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.573000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a8bd1fa188bbfc729d725a6102c5e5ee"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563573208.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.574000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2523938c553688a99ea02f1591373a61"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563574579.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.576000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a694ae9a076aadb9aadd54d4607e0d89"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563576036.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.577000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0c4b04e67825708cb3f1038627288118"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563577347.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.589000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b11290348d9d33ce0615871439e85588"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563589021.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.590000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "cc581f874213346692010fdffdd20ffb"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563590543.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.592000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ee37925685d12c1d9b07b6aca2b7e8ae"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563592220.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.593000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8ef6605723edc583eedc959738b79c95"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563593539.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.605000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "bdec707ffbd87fa85144c41b03f28303"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563605469.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.606000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "58a91b541869928e792a91b2eeb0b2ef"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563606847.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.608000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4595dcabcc01299d21535d3cfed6d524"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563608299.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.609000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "aa278d424b74323fbe370c1e5c0a2278"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563609641.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.623000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "90d5a9f97d430e4e6a9b5587fd0b9236"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563623285.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.624000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b83a369be0cc735a812ad2057f0c3f3e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563624686.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.626000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f10c5735b204268dd94df131fea3dec8"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563626137.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.627000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0c77e5f3e54b7d52d4b4bf3854a6437b"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563627437.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.640000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f1dc4b52f9fc4f346208f74d3d7794af"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563640048.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.641000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3c1387ae4d5511a750f0b5cbaa6d4d2b"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563641432.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.642000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0b804d6008210340272578bd9b6a1dee"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563642904.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.644000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "cef2c3c5c9799534ffcade90faa89700"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563644217.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.656000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "725a91845896f5d0e3e8ff0f9eefb3ff"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563656788.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.658000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7138a5252f3a12c897a3ef1cd81318d1"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563658164.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.659000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "bfb29b1747e3c2888bcb7963422ba622"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563659616.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.660000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f04d09c63f9e885364ddf48beed1f1d7"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563660920.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.672000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8f43a4c6e22bd2cabfb4e6ed06365a40"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563672503.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.674000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "00b3018cb6cef696457b3f65aa69ef01"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563674093.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.675000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "75452625a537b4d71ec6d5bdc6c4c897"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563675549.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.676000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3f37c04d9a76c33855a89526e7316464"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563676860.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.688000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "63abec4ea58b25c8db7fb082ff386649"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563688685.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.690000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5e5195f872852d01760d9ec26afc0fee"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563690090.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.691000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "bd9c0f95073795f66056e9cb82399ea0"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563691709.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.693000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0013deecb7e0f29c83d3ddd4ad6d6c8d"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563693057.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.704000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b36bf1bd5d50f358c7a16bb1e204622c"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563704885.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.706000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b80a8f18c7cbfe312a4059e94b3dbcd1"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563706270.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.707000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4f85de92fb8afb6c99d5437bd4768868"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563707726.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.709000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "16068d603e00d1566c9373c019f8068b"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563709030.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.722000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1e049ce0a1a2753b088aa8e0ec2b4567"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563722619.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.724000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f0b26fe4d7efe797b0d8a4e8bf549f91"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563724020.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.725000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a6352f19881f2fbecd9586caa6bf7fa4"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563725466.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.726000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1afc0af5ae130a1ad1bf9e5cb663c103"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563726774.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.739000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9a421661ffabcb6297dec7fd0f0a7097"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563739125.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.740000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f0a6a1f4bdfcf493dc3ca6e70ed7af3e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563740490.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.742000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b3e8677013650e75586c7dff39a41da7"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563741954.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.743000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "508532ce35f65c37ba942757bcf05158"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563743438.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.756000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "db3f7c91d110dab69475856d4a60005f"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563756092.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.757000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "70d97095ddbfe95cc018e45d6680c3bd"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563757484.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.759000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5c5744927b213ec02825e8fafa15fa1d"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563758945.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.760000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0d0f54d398bc134c1e355f4c513589bc"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563760263.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.772000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3b04eb2fa9d6284764190542b8e2afd6"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563772355.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.773000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "71f6699240ba0fc7a95a3a9a7352b967"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563773774.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.775000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ebca509e6f3745596cb147845c5c6d3e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563775276.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.776000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c1c0a2cbf5188def329d30117731d07a"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563776571.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.788000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ed5532539856c6de31ce51ba894f4ff5"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563788551.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.790000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6671699758a5bf07b72386a309fff12e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563790008.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.791000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a98175ab768a31555efb4f01e91ca621"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563791564.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.793000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "eab86f05798c7b0067a406ea63967561"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563793011.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.804000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "121ae192cbba562fbf6d5ecc538f04a7"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563804802.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.806000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "27cf39eb2657cf0366e62f0b5b97c1c5"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563806271.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.807000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "63150fdc5d2264355a4714dba5c671eb"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563807871.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.809000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4f40976890a93de569d2df0cab66507c"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563809290.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.824000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8f722c6dbe9433ca33fe403a339d4d75"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563823936.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.825000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "95124d19409190f61f5f73d14a68cfc7"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563825617.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.827000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d9369db4704c5b34db075aedf8e366e9"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563827177.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.828000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e4f67be95b412d86939e52aa5d3ab3e9"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563828572.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.841000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "59e0a3ef7728ec737514d0f9a8451d5f"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563841758.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.843000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9d3bad3a1459feb9e568b9575efb48dd"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563843251.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.844000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "07bd467342087031551759eb84dae26e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563844839.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.846000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "54a71eea8de6696029fac6b27aada58d"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563846264.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.859000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e03cb228e2f341d14e2ac4194b40df5f"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563859719.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.861000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6017cabb5782b9cfa14c17cd8fb06215"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563861208.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.862000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0364e03f98ced0884e85dc04f6d122ef"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563862765.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.864000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5b9d271c8dce700153a66c335aa97acc"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563864155.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.876000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "52daca77e9a6966cabf07ccacffae7df"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563876607.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.878000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "cc89789e9376d101e12c3b20cacf2e62"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563878086.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.879000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3bdf75dddeb865f49225089b1bbef0ea"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563879715.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.881000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3cea66640895e07b8a3652b1e12ec402"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563881121.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.893000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d5b276eb8404872f6203180e35586277"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563893617.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.895000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e8d1f76310802d478f70ebe0b9d01878"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563895116.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.896000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "fc5dce6dd6ecaa82f7f059ff00671540"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563896861.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.898000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "aa12baed2008a8a82a0a882032e0c5be"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563898255.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.910000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9a5da179b1c472de4e5c0a1e7c73ee66"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563910749.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.912000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ae26ab93c590ae6dc4a7073059a9e872"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563912237.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.913000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "694ac338d325681bd0dd02ed0d75766c"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563913800.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.915000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "51e63dea18600b3cae624df6105778b6"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563915179.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.929000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b25b205918ef0a0583f654a6d9e2580a"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563929590.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.931000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b915d2da2257515a2ad97ab547e9974a"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563931080.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.932000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d589c5e31376f7db2343f6a5669de67c"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563932651.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.934000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9a4f4d3e3e4e4a9f897c2b3058417889"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563934047.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.947000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a9573fca7dc9b7f2528527bf7089cd20"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563947210.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.948000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e0b7818c892fe74587b95d7da9245f79"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563948716.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.950000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4112e874175d579d7a73607577137558"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563950274.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.951000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "660dee70f5342afec75037e73b39c4e0"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563951677.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.965000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "01de561f242da2e3b958a22df8ef4c81"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563965229.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.966000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5f4c6e6bb247d18fbf70d4e64cd32897"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563966721.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.968000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "70141044a829c06c71d9e50c7e3ec95b"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563968263.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.969000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f97112e329a22efa20becc5dc9ba8e8b"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563969659.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.982000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "65a8329281cc0a02dcd3553157e4cc02"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563982127.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.983000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "93aec420fd52024661a00fb9361a98ae"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563983817.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.985000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e9db216ab749414842a566ab680c8f42"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522563985375.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.986000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "89449fe7f41f4dbae8bda2e75ce18052"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563986779.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:43.999000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "76752235c7e6c034cb2e624f210da68d"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522563999218.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.000000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a4e81fad19c9517341ce5eccd07c10d2"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564000739.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.002000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b24eeac67b6457a00ee54f42e0cc5ada"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564002282.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.003000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "eccb0fc77f052d3bef7c4be9205adca5"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564003702.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.016000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "21602c295288bedf33e1c1f3cc07a77d"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564016289.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.017000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "74daf88e18c1d1802cf780beeb16ab4b"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564017775.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.019000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c04514cd87a49351211c37ec42b3a4d4"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564019342.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.020000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "adeeb627d69078462619a66d7488d663"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564020760.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.035000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "380650800cbb2567441e8b4f64a8b611"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564035257.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.036000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "16a9c71d241ab0b436ba95cf35b7a6a6"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564036739.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.038000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "81c56bfca448a68ce850e9c21eec938d"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564038262.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.039000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b16609b2ce2aa3531f69a0f99b6b5dee"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564039685.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.052000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "361edf10cefb50f05d8fac33490a04c2"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564052418.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.053000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a14984b4cc77a6210e7106dd252d876e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564053850.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.055000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e47869c118085811e0b886a5773d4c32"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564055528.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.057000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "aae20bc263a4c2d36343c45967e002fd"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564057028.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.070000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b79d371097165b51c0acf20f2d5d627c"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564070053.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.071000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "28476823ebe2d727d0ecdcf6c519615f"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564071596.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.073000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b70925c73165cf91709207bb816c5aa3"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564073056.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.074000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ebfa6c73cafb51e2c11550869ce8d3b0"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564074368.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.086000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5e6f0c4d84e64dd1d725c4fbd5ad7003"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564086071.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.087000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "eadda137b8692a24f87f8c765906a157"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564087475.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.088000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7d3e467d5f388408ed6d3f93efbd04f1"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564088935.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.090000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3cce30f4bc98a36f232a8c2c1c9c95c6"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564090250.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.102000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6bd8fabde08ba41af0e5a50be4f81a36"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564102364.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.103000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "13a616999580b020325c2d98083c4fc2"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564103778.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.105000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "71cd61281af5b38bee95e6527949b29f"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564105296.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.106000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "bae91a099069fee1b2fa5baaaef02352"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564106611.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.118000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d41826e462623181d85730cbe15c0eec"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564118413.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.119000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6d2ad542d582e5f1f3bd657c2f368640"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564119887.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.121000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3214c197db42058f0aa7beea130d1169"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564121337.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.122000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "efff165b08025edac6f62b7ca1fdb160"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564122656.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.136000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "cf449e88a32379b05ac76f6732c2c907"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564136175.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.137000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7a3d98a12a0c09b97659bd952665a14b"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564137764.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.139000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c1a8be3a32d9005be571b28cd957d369"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564139214.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.140000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c0f33792808e425233af37b3588bedeb"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564140530.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.152000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e8d6e0dedb944a6e15427c27ce5a96ed"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564152686.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.154000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b48394358e70d8a7e63ddbb4cae75a13"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564154079.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.155000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "bce0b2a24303f00280257da3c0b257a8"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564155542.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.156000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8156e4d8b53b206d4445cb608974d73f"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564156850.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.169000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "304f65e49403fa7ad47fe102153a9a12"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564169433.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.170000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6398a8f9a922ad3b631c7d4875567b9c"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564170844.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.172000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "75b400f484eefb8c0245bce6a8888283"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564172337.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.173000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "716e0964f5d5889a9ad1cf486e53a043"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564173726.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.186000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "60cbddeb7122aef2f9abf8b04d56f2c6"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564185978.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.187000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "491edabc5bf8db7fdde4a0ebd5667521"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564187436.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.189000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "02b4e66807597fd1381d99f4a912350b"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564188991.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.190000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6dd5cd30dc9db122802719a861597a1c"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564190388.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.202000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "67e54a4dc719e587915680827d544a56"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564202579.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.204000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d555fb181b01f9ae308a4ce4bce3ddca"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564204040.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.205000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "44d3347fd00f4657fcfed0b85e4d845e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564205685.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.207000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "90851cf76a879753983da388536c16e8"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564207098.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.219000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "065772dfe6fb21400c2ddb90cf4d16cb"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564219035.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.220000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "47af7ddb153c7977d164e0fee6c1777a"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564220420.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.222000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "04181097fb5e50a75177a7e022d109bb"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564221946.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.223000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4d64fde4a99b7e45166ffd6467278c1f"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564223359.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.237000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "98682e77bc7b67be4e8daf065c218d40"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564236991.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.238000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c052ca550cabbfa4141a758863eabc2f"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564238388.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.239000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8090934e19121db916290816e287cf0b"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564239851.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.241000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "38ac0be2910e2893fc70854724e48fdc"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564241160.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.253000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "75bcfbb0a8146f0defd6f2eb51a838e6"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564253461.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.254000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ab66f8bebc1b659be42103da44aa363b"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564254895.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.256000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8d6260c2af6b726a58f308e49e95eb07"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564256348.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.257000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "43eadfd49cc97eff0735211826628f02"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564257658.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.270000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "fc4c7023f620b6ef56a1714a7df38ded"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564270665.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.272000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ee12689741b1c9d2ee4c809d5a222fff"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564272093.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.273000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "373e3e84fa65940d1e19e18e82226c04"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564273541.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.274000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "42413f490047f0e5a5679436d8dd4e08"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564274856.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.286000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "06a67597c31124a59151db1f519186b1"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564286552.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.288000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6e85b40a20b6b1b41d16c73f917f3c98"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564288145.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.289000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5ccf8403d00e7aa39dac1e146e0fd554"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564289595.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.290000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "043cb848786d29efc31d424c7f1546e6"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564290919.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.302000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "fd3156f61116a3c82e3a2eacfe06ccac"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564302761.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.304000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e93f5a92f04f3fa2b4a33b4870912215"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564304168.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.305000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "45b9e7b74e0fe82f5184ad3492f914cd"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564305633.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.306000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "61f5912758affe8ced10bc454eaea4b6"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564306930.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.318000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d93030e8f480336a00a1877ae361e950"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564318535.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.319000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "849b9684b1fbfde257c787426821b6af"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564319925.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.321000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "78dc8d2b20c61a8010ccf6fb1c38f3ba"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564321395.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.322000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b0934fb00a077b277e3c6b315ca45f4d"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564322708.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.336000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9f86e38713e020942089c7d032f987e6"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564336348.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.337000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "76e0a6bd9a6816d3ce0ff03ae7479fba"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564337731.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.339000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7a7466239c5813923994c6a1ea4d6e8d"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564339175.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.340000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "25d610446a8341eda1534383e617cd96"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564340475.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.352000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6bf042db2f4a3e4812f08c8e7f46676a"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564352651.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.354000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "881b91419e7dc9eb11b0166cd47a7141"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564354034.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.355000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4d26381ca10e6dac4f585fb73be6a594"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564355487.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.357000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6e283eabf8358a81559fd9f1287d1603"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564356969.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.369000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2aabde89f135798b33e56d1014007db2"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564369714.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.371000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e522494e3fe486aab999ca2c5cd0af6e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564371120.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.372000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5a01018e4afffd10d8876716e5709103"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564372582.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.373000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2128f44a3a8c5c50eccc03ec91a38811"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564373923.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.385000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "605b2d94ea58b0e5c512e35cfebc052e"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564385719.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.387000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4ed624158d9a2b0fb9fe3a0af505afd8"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564387158.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.388000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "dd4138cd33b4c79311bca2212bd31e43"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564388611.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.390000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3401c61ae431bf27bb624d3e9465e0ab"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564389946.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.402000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3bf74e2519f9c868562027e2999650ee"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564402389.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.403000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a03d4ec12cd36bfce7880fde27458aff"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564403889.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.405000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9750d780b25dddd0967e657934da8d7a"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564405355.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.406000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "fd20408a3774184f518c18be4897fa6d"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564406703.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.418000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4caaecf7eed257ddc27d2835f63da6c6"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564418615.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.420000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f92225371112353333df94f7f9af658d"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564420002.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.421000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b614d46c4abc5c60ca9b0e609da7ec2f"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564421501.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.422000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1f6ac982f25cb8bec46e394b8ef3774d"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564422861.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.437000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ca37635af4eeeb2773c5f7b98a7661c7"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564437141.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.438000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "cb9e8bc7ed178c764d1fa4171d56e54d"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564438838.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.440000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "23cc9e30f4d6b416a2a7fd58377da3f9"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564440352.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.441000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "76ad102466eb8961fcc64027f3f2cdee"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564441693.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.454000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "56bc8e2917440615571873b374c95332"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564454385.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.455000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0780bbe7e0fe7262f38b193fbd18210e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564455854.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.457000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5a08deaf8f6825f992fbf16d0b9bf500"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564457338.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.458000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2e3b73bb2dd1774a2401af87651a9e4f"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564458641.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.471000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b88e25f952b9ee2b59393ada10dc7b59"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564471343.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.472000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c363c126e248aced9cb222c423f0df48"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564472744.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.474000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9dfa5a7f6d9db0bbff0e8c590f5ccb46"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564474309.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.475000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "bacc1891ec7957294bfdb5f55cf9f691"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564475718.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.488000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f2b961cf8d3214df47befb8cdb6848b4"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564488364.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.489000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "dbfa3c3f01dc09efb8b97e4c53c4b5d1"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564489841.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.491000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "bd0bf3c17f90f5256dc8938aea4ee950"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564491370.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.492000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5d81b3e6d760ea447d38eacaaecea9b5"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564492693.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.504000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a3a847cc6991b4d8ef1f42222e728499"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564504637.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.506000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "48d49016a79d2a2cc67041694c928836"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564506017.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.507000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d85209ca6313dd79a68038655c3fb796"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564507667.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.509000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c71b54f28b5bc8331232bb94970be25a"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564508970.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.521000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e7bb1f5f3b3a19fc7d7e70e3b27fe79f"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564521191.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.522000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "646e0f3473750e023ca29b22e6428752"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564522642.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.524000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "763e4bf1351deca6c950997b008b6b58"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564524117.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.525000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5f2b96e977a56a0222ccf594766ea6b4"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564525439.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.539000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "206b924646ed4c44919806721ff9b2c4"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564539332.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.540000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "64400d257fd6a5dd21095f02459679ae"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564540727.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.542000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b6a4c576f41778f5ff8d6e940c58b2aa"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564542222.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.543000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a172b150393b59eaa63e8e8a94f02b53"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564543589.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.556000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d458a4cf9f3673984374480d4036a7b5"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564556139.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.557000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1186cace5b37c420aad20cdfa9fb1779"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564557676.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.559000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "72b1ea0e01314a0c94f002428b346971"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564559192.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.560000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6e050450ed1f43bc886811255e9c9de0"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564560513.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.573000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0bc5a994252eaec86d95c6cb009d61d5"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564573380.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.574000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e9d62de9e66feb5dc85d007fae2d67b6"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564574826.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.576000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "808c33a8ac6a7faafd839e24db763a8b"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564576315.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.577000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "00a67be93fbcdca6385d0296f60a9ab1"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564577669.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.589000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ef6a032d5e5515441f527d4ae7af986c"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564589834.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.591000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "46a5bb56e6ec85b5a660d6cdbb32363a"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564591444.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.593000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d72e76cbabbf71970ea7bab307dbd8d3"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564592988.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.594000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "fce49c445b7e80ca54620f87673d2de3"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564594380.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.606000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "788880f63a4bfdfddec8566322c35de5"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564606699.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.608000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9db781f3fe50932c42bfa84079f979b3"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564608154.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.609000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d4e4b9a8aaee05e7bf88d76b9db1061b"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564609683.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.611000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "66e5499dda499b148a6860f2f9737c89"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564611311.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.623000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "01e5ce455478990c00a5fc092f19c186"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564623552.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.625000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c1e961e9420a5cdc5c7e03c81aa92779"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564625000.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.626000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "912fabadc6d239ea14043a8057182092"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564626499.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.627000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c9321a40503cac40ab9c52a74e041059"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564627866.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.642000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a0d54989b63f82aad436db093eee614e"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564642416.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.643000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "47d065d27c38adae2822bd955da93744"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564643927.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.645000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3b96d0e5065fcf2e2739be1ceb11c531"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564645478.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.646000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ab2ec6459ee3514bdb0c590a28e98d72"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564646889.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.660000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2b1caa8ba241b05498caf487af2130dd"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564659996.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.661000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "601d7ea4f6a6dd56903a6fad614c2f8c"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564661457.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.663000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "55fd75d739d2554cbdba71657f464659"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564663046.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.664000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "61ce5d4a98aea4156c7fa831556e1cb2"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564664636.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.678000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4c100aea8657fab8103b4b55d81f05d6"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564678201.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.679000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6fd841ac792c0a21da10c7044aa1b685"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564679697.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.681000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9a0a24c29442e9a168525475da7f6feb"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564681232.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.682000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "817b6937ed37ed5e7850d158917dc193"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564682638.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.695000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3dc2bb32c941754467feada702e1f07b"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564695119.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.696000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8c9d3e9bb1ed5d20ae536c3d7898c6ab"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564696568.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.698000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7fb158e37e617ec31dcd742f76866342"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564698147.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.699000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1e9ac57560ceb0df0bb172cc13428a02"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564699535.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.712000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3a07179183a761c26f68734b0bc20aa3"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564712495.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.714000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "442c3e2cd0184f348eef6febfa48addd"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564713964.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.715000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "89835b6b8927c6c2936d818a1f317ff0"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564715529.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.717000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d1963696d79a8ddde4380b38c315b9a3"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564716951.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.729000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "21afa7b69a231119398d1abaecdf866b"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564729487.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.731000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "17471d29acf9348ad2566617040d9dd7"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564730949.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.732000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0c2982219cf1ae55cb3d0a5b51f07549"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564732476.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.733000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5f494f394268606aeb5340c0fb184618"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564733890.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.748000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9ef6e25c079ab7f236552547ebb447f3"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564748533.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.750000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c3f59126fcdff292dbb4c04ce2fe83eb"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564750206.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.751000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "297f55115cfbf1b61c4d0bd1692f6098"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564751768.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.753000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ca41639e4d1a8d2859369b9c6749a6bc"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564753153.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.766000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ad38e994c4418b1677ae5b8924af2092"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564766261.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.767000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "82d4692d27fb04ee29bb5fec7a64aa5b"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564767764.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.769000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f68d32bc4d29e97a5d1b801146fd9d43"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564769332.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.770000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9efe57623fb1ac78ab5b6bd9f710e5cf"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564770744.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.784000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9f939f39a1a5d1cdc2613a063e21b1a1"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564784297.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.785000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "387a7200cd0ec4001529831d6188a30a"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564785784.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.787000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "25e41ec467e8f5c73c2fbc3ac948b8b6"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564787341.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.788000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4597ac5b77b65a534b7ea7de7b7fc2b2"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564788749.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.801000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b50a2d9dee13503274cc44ebf20e9fa2"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564801178.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.802000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "046410ddd2f7677aa2877039f57ea36b"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564802706.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.804000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3dbf6bd6a02fdb43f1e579d0aabc4466"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564804576.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.806000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "84e016faee37d2638f01b820391fc7ee"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564806573.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.819000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2189c09721b0100b0b53fe1a07ea42f5"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564819614.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.821000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6a9cd03dfd2a0f01d561b2be35f5c6f9"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564821184.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.823000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "95124ee5085b5be1cd18046e8e756747"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564822978.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.824000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a62dc1de7b1c2c5a43c9e3253623165e"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564824398.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.837000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2f65baea43df8939b4444013e7d5f4ad"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564837003.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.838000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "88a7fe779bc4fff5b56dc7a474fc0868"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564838530.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.840000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f863df255b1a29e6ad9c68d8e45400e4"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564840103.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.841000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b9091fa84e32161803f26daefeb16c15"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564841500.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.856000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "abadad8327a7b5e0f54cf14dbfefb621"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564856154.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.857000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5c5e2c10425ba32ffa36330503153e6c"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564857644.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.859000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ebc87c66d6002364e3ba4dd2eef65289"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564859208.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.860000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f3cadb85137ff43b3b20bb5edee69a96"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564860612.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.873000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "bb4430fd10bd6e4bdb5a56901e36d152"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564873857.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.875000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8b0a5efdd3b1ecf5c166e9e208c03eed"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564875322.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.876000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "05b5ff48dd4b437dd697d620c33be88f"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564876904.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.878000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "116673d8788e768fce657998ce8487d2"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564878324.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.891000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f66a061752e5cae84e93686d9848af8a"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564891197.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.892000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4af56077758f5a51b263c8779bcea46e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564892579.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.894000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "056d617384d9de9e3770a325d1c08e52"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564894038.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.895000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7e25ec19d22b9b38c91b773fbf3e85d0"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564895323.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.907000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3f0053c224200edb8c9703ed55337be4"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564907285.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.908000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "bffcacc1f18823b1676c25b4eea9aec4"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564908890.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.910000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "13ff9dad03016b8772da3295cee82d9c"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564910411.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.911000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "259083e0b7769218fb549c796e1cf9d1"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564911754.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.923000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "120d01b207f3b2d9754e28c37dd7a66d"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564923663.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.925000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "600566b3d667b57897ea6701c736c27b"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564925058.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.926000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a16b38c07fdb6df0b85affc19c94ad97"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564926533.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.927000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8ecc73f0819bb1fc2247a4506499f9a3"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564927851.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.939000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7e8e01f3592864c26c4a1b84422611bf"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564939494.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.941000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a16c0ae2b9f1ec1b48b03f6b7eddb072"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564940941.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.942000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "503e8cde084782a24b5467afb46cbf42"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564942487.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.943000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3b451b64f18916a0a8cf3a1d34c9aaf8"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564943875.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.958000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5dc88eb307a812b06f48850f2623f04f"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564958045.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.959000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3f8e48e4026ad9285b77ce73dece251c"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564959534.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.961000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a2baa7815fdae28308edf1e850fdbb87"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564961073.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.962000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6ad2fd43be920d88113cf0865bb3d425"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564962378.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.975000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d0bc096405e774ffe6edfebe8b2ca83e"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564975190.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.976000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0f5930594d8e8c59efd3fdf73c609aab"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564976582.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.978000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "94f5f98d1a41b09320e7532af342e2f0"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564978253.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.979000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "86198268d03a2837200307b5027e9a93"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564979568.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.992000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2a9914a2cc11bf73e4b31e88fdee6786"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564992432.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.993000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "011aa77400e85a2cb1f156b6030d6ef2"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564993851.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.995000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8a39490837a7f2cbf8b0b5ef9dffdf7e"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522564995318.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:44.996000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "70cc495be02d88579ad6cf18de6c1120"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522564996613.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.008000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "dd7c34efa38ea2365847c84ab40c1df9"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522565008817.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.010000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5d3cb71cec8af8eadfff98e3df7924d5"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522565010334.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.011000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "dbe096a0d846b51134cb0e6fb5de4726"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522565011888.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.013000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "90d046c84f53d9c540ed807c2f28723e"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522565013267.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.025000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "33f90dee10d58655fea2193b8882e779"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522565025755.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.027000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "76b7644863b88727dd27860b05244fb9"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522565027168.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.028000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "436f2a1fe74278522782c6c4e510e012"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522565028794.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.030000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ba5e96e2ee974c9a7444439bd9419d8d"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522565030234.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.042000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "68a486b962eae30aeab707cba921269f"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522565042805.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.044000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5230dc6969a89ce3e16c114ff700b265"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522565044252.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.045000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "90a70de0a43b04fb96b07476c7923e02"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522565045724.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.047000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f29450ed4912b256478203d8e47af32b"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522565047144.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.061000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "02cea16aef4e9ee48e9921b4d4552e86"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522565061073.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.062000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3578916bd269067a2ef85bd62daedf09"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522565062685.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.064000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "74a86dc8619dac6f551c6b3d5bb1f115"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522565064214.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.065000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1ddf030ee8719c2fcd6db4f5f813917f"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522565065606.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.078000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9bf77919dc0701f78caf3025da3b727d"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522565078169.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.079000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "73b3c64c7e2beff8d92ef2ccb8e28aa6"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522565079555.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.081000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "22622c255f6c78a0b11e32a6e7b0e84d"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522565081041.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.082000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ef3c819d41b381c35743d8f32ef913dc"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522565082346.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.095000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c536926bf216598d32123c492d7bb693"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522565095090.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.096000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8fff92cc887339fa254713d1ce141ab5"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522565096471.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.098000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6d3d4ae11154fdc3837a42708b59f280"}
	{
	"name": "pad_mm_benchmark_get_do_bench",
	"ts": 1757522565097947.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.099000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ef5578c9f3d3a26fa326b2725485de10"}
	{
	"name": "pad_mm_benchmark",
	"ts": 1757522565099252.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.312000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "049b934503a2dbf2b64f26b6bedea821"}
	{
	"name": "_recursive_joint_graph_passes",
	"ts": 1757522565312179.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:45.907000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_inductor/compile_fx.py:2198] {"artifact": {"name": "after_joint_graph", "encoding": "string"}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2657, "name": "compile_fx", "filename": 19, "loc": "return inference_compiler(unlifted_gm, example_inputs_)"}, {"line": 1267, "name": "__call__", "filename": 33, "loc": "return self.compiler_fn(gm, example_inputs)"}, {"line": 2543, "name": "fw_compiler_base", "filename": 19, "loc": "return compile_fx_forward("}, {"line": 2198, "name": "compile_fx_forward", "filename": 19, "loc": "trace_structured("}], "has_payload": "4d99dc73eb9e1eed68053a1f239eff0d"}
	class <lambda>(torch.nn.Module):
	    def forward(self):
	        arg877_1: "f32[s6, 128, 3000][384000, 3000, 1]cuda:0"; 
	    
	        arg877_1, = fx_pytree.tree_flatten_spec([], self._in_spec)
	        # No stacktrace found for following nodes
	        model_audio_tower_embed_positions_weight: "f32[1500, 1280][1280, 1]cuda:0" = self.model.audio_tower.embed_positions.weight
	        model_audio_tower_conv1_weight: "f32[1280, 128, 3][384, 3, 1]cuda:0" = self.model.audio_tower.conv1.weight
	        model_audio_tower_conv1_bias: "f32[1280][1]cuda:0" = self.model.audio_tower.conv1.bias
	        model_audio_tower_conv2_weight: "f32[1280, 1280, 3][3840, 3, 1]cuda:0" = self.model.audio_tower.conv2.weight
	        model_audio_tower_conv2_bias: "f32[1280][1]cuda:0" = self.model.audio_tower.conv2.bias
	        model_audio_tower_layers_0_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn_layer_norm.weight
	        model_audio_tower_layers_0_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn_layer_norm.bias
	        model_audio_tower_layers_0_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.bias
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.bias
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.bias
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").final_layer_norm.weight
	        model_audio_tower_layers_0_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").final_layer_norm.bias
	        model_audio_tower_layers_0_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.bias
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_0_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.bias
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn_layer_norm.weight
	        model_audio_tower_layers_1_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn_layer_norm.bias
	        model_audio_tower_layers_1_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.bias
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.bias
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.bias
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").final_layer_norm.weight
	        model_audio_tower_layers_1_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").final_layer_norm.bias
	        model_audio_tower_layers_1_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.bias
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_1_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.bias
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn_layer_norm.weight
	        model_audio_tower_layers_2_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn_layer_norm.bias
	        model_audio_tower_layers_2_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.bias
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.bias
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.bias
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").final_layer_norm.weight
	        model_audio_tower_layers_2_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").final_layer_norm.bias
	        model_audio_tower_layers_2_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.bias
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_2_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.bias
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn_layer_norm.weight
	        model_audio_tower_layers_3_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn_layer_norm.bias
	        model_audio_tower_layers_3_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.bias
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.bias
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.bias
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").final_layer_norm.weight
	        model_audio_tower_layers_3_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").final_layer_norm.bias
	        model_audio_tower_layers_3_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.bias
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_3_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.bias
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn_layer_norm.weight
	        model_audio_tower_layers_4_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn_layer_norm.bias
	        model_audio_tower_layers_4_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.bias
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.bias
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.bias
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").final_layer_norm.weight
	        model_audio_tower_layers_4_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").final_layer_norm.bias
	        model_audio_tower_layers_4_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.bias
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_4_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.bias
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn_layer_norm.weight
	        model_audio_tower_layers_5_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn_layer_norm.bias
	        model_audio_tower_layers_5_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.bias
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.bias
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.bias
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").final_layer_norm.weight
	        model_audio_tower_layers_5_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").final_layer_norm.bias
	        model_audio_tower_layers_5_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.bias
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_5_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.bias
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn_layer_norm.weight
	        model_audio_tower_layers_6_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn_layer_norm.bias
	        model_audio_tower_layers_6_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.bias
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.bias
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.bias
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").final_layer_norm.weight
	        model_audio_tower_layers_6_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").final_layer_norm.bias
	        model_audio_tower_layers_6_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.bias
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_6_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.bias
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn_layer_norm.weight
	        model_audio_tower_layers_7_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn_layer_norm.bias
	        model_audio_tower_layers_7_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.bias
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.bias
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.bias
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").final_layer_norm.weight
	        model_audio_tower_layers_7_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").final_layer_norm.bias
	        model_audio_tower_layers_7_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.bias
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_7_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.bias
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn_layer_norm.weight
	        model_audio_tower_layers_8_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn_layer_norm.bias
	        model_audio_tower_layers_8_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.bias
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.bias
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.bias
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").final_layer_norm.weight
	        model_audio_tower_layers_8_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").final_layer_norm.bias
	        model_audio_tower_layers_8_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.bias
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_8_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.bias
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn_layer_norm.weight
	        model_audio_tower_layers_9_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn_layer_norm.bias
	        model_audio_tower_layers_9_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.bias
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.bias
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.bias
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").final_layer_norm.weight
	        model_audio_tower_layers_9_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").final_layer_norm.bias
	        model_audio_tower_layers_9_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.bias
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_9_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.bias
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn_layer_norm.weight
	        model_audio_tower_layers_10_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn_layer_norm.bias
	        model_audio_tower_layers_10_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.bias
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.bias
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.bias
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").final_layer_norm.weight
	        model_audio_tower_layers_10_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").final_layer_norm.bias
	        model_audio_tower_layers_10_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.bias
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_10_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.bias
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn_layer_norm.weight
	        model_audio_tower_layers_11_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn_layer_norm.bias
	        model_audio_tower_layers_11_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.bias
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.bias
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.bias
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").final_layer_norm.weight
	        model_audio_tower_layers_11_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").final_layer_norm.bias
	        model_audio_tower_layers_11_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.bias
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_11_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.bias
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn_layer_norm.weight
	        model_audio_tower_layers_12_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn_layer_norm.bias
	        model_audio_tower_layers_12_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.bias
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.bias
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.bias
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").final_layer_norm.weight
	        model_audio_tower_layers_12_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").final_layer_norm.bias
	        model_audio_tower_layers_12_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.bias
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_12_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.bias
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn_layer_norm.weight
	        model_audio_tower_layers_13_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn_layer_norm.bias
	        model_audio_tower_layers_13_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.bias
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.bias
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.bias
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").final_layer_norm.weight
	        model_audio_tower_layers_13_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").final_layer_norm.bias
	        model_audio_tower_layers_13_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.bias
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_13_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.bias
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn_layer_norm.weight
	        model_audio_tower_layers_14_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn_layer_norm.bias
	        model_audio_tower_layers_14_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.bias
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.bias
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.bias
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").final_layer_norm.weight
	        model_audio_tower_layers_14_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").final_layer_norm.bias
	        model_audio_tower_layers_14_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.bias
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_14_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.bias
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn_layer_norm.weight
	        model_audio_tower_layers_15_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn_layer_norm.bias
	        model_audio_tower_layers_15_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.bias
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.bias
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.bias
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").final_layer_norm.weight
	        model_audio_tower_layers_15_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").final_layer_norm.bias
	        model_audio_tower_layers_15_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.bias
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_15_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.bias
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn_layer_norm.weight
	        model_audio_tower_layers_16_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn_layer_norm.bias
	        model_audio_tower_layers_16_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.bias
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.bias
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.bias
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").final_layer_norm.weight
	        model_audio_tower_layers_16_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").final_layer_norm.bias
	        model_audio_tower_layers_16_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.bias
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_16_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.bias
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn_layer_norm.weight
	        model_audio_tower_layers_17_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn_layer_norm.bias
	        model_audio_tower_layers_17_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.bias
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.bias
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.bias
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").final_layer_norm.weight
	        model_audio_tower_layers_17_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").final_layer_norm.bias
	        model_audio_tower_layers_17_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.bias
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_17_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.bias
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn_layer_norm.weight
	        model_audio_tower_layers_18_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn_layer_norm.bias
	        model_audio_tower_layers_18_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.bias
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.bias
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.bias
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").final_layer_norm.weight
	        model_audio_tower_layers_18_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").final_layer_norm.bias
	        model_audio_tower_layers_18_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.bias
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_18_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.bias
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn_layer_norm.weight
	        model_audio_tower_layers_19_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn_layer_norm.bias
	        model_audio_tower_layers_19_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.bias
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.bias
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.bias
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").final_layer_norm.weight
	        model_audio_tower_layers_19_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").final_layer_norm.bias
	        model_audio_tower_layers_19_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.bias
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_19_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.bias
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn_layer_norm.weight
	        model_audio_tower_layers_20_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn_layer_norm.bias
	        model_audio_tower_layers_20_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.bias
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.bias
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.bias
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").final_layer_norm.weight
	        model_audio_tower_layers_20_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").final_layer_norm.bias
	        model_audio_tower_layers_20_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.bias
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_20_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.bias
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn_layer_norm.weight
	        model_audio_tower_layers_21_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn_layer_norm.bias
	        model_audio_tower_layers_21_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.bias
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.bias
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.bias
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").final_layer_norm.weight
	        model_audio_tower_layers_21_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").final_layer_norm.bias
	        model_audio_tower_layers_21_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.bias
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_21_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.bias
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn_layer_norm.weight
	        model_audio_tower_layers_22_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn_layer_norm.bias
	        model_audio_tower_layers_22_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.bias
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.bias
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.bias
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").final_layer_norm.weight
	        model_audio_tower_layers_22_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").final_layer_norm.bias
	        model_audio_tower_layers_22_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.bias
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_22_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.bias
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn_layer_norm.weight
	        model_audio_tower_layers_23_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn_layer_norm.bias
	        model_audio_tower_layers_23_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.bias
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.bias
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.bias
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").final_layer_norm.weight
	        model_audio_tower_layers_23_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").final_layer_norm.bias
	        model_audio_tower_layers_23_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.bias
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_23_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.bias
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn_layer_norm.weight
	        model_audio_tower_layers_24_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn_layer_norm.bias
	        model_audio_tower_layers_24_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.bias
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.bias
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.bias
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").final_layer_norm.weight
	        model_audio_tower_layers_24_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").final_layer_norm.bias
	        model_audio_tower_layers_24_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.bias
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_24_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.bias
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn_layer_norm.weight
	        model_audio_tower_layers_25_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn_layer_norm.bias
	        model_audio_tower_layers_25_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.bias
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.bias
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.bias
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").final_layer_norm.weight
	        model_audio_tower_layers_25_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").final_layer_norm.bias
	        model_audio_tower_layers_25_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.bias
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_25_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.bias
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn_layer_norm.weight
	        model_audio_tower_layers_26_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn_layer_norm.bias
	        model_audio_tower_layers_26_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.bias
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.bias
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.bias
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").final_layer_norm.weight
	        model_audio_tower_layers_26_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").final_layer_norm.bias
	        model_audio_tower_layers_26_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.bias
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_26_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.bias
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn_layer_norm.weight
	        model_audio_tower_layers_27_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn_layer_norm.bias
	        model_audio_tower_layers_27_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.bias
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.bias
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.bias
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").final_layer_norm.weight
	        model_audio_tower_layers_27_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").final_layer_norm.bias
	        model_audio_tower_layers_27_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.bias
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_27_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.bias
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn_layer_norm.weight
	        model_audio_tower_layers_28_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn_layer_norm.bias
	        model_audio_tower_layers_28_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.bias
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.bias
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.bias
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").final_layer_norm.weight
	        model_audio_tower_layers_28_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").final_layer_norm.bias
	        model_audio_tower_layers_28_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.bias
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_28_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.bias
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn_layer_norm.weight
	        model_audio_tower_layers_29_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn_layer_norm.bias
	        model_audio_tower_layers_29_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.bias
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.bias
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.bias
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").final_layer_norm.weight
	        model_audio_tower_layers_29_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").final_layer_norm.bias
	        model_audio_tower_layers_29_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.bias
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_29_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.bias
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn_layer_norm.weight
	        model_audio_tower_layers_30_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn_layer_norm.bias
	        model_audio_tower_layers_30_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.bias
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.bias
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.bias
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").final_layer_norm.weight
	        model_audio_tower_layers_30_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").final_layer_norm.bias
	        model_audio_tower_layers_30_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.bias
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_30_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.bias
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn_layer_norm.weight
	        model_audio_tower_layers_31_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn_layer_norm.bias
	        model_audio_tower_layers_31_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.bias
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.bias
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.bias
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").final_layer_norm.weight
	        model_audio_tower_layers_31_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").final_layer_norm.bias
	        model_audio_tower_layers_31_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.bias
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_31_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.bias
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original2
	        model_audio_tower_layer_norm_weight: "f32[1280][1]cuda:0" = self.model.audio_tower.layer_norm.weight
	        model_audio_tower_layer_norm_bias: "f32[1280][1]cuda:0" = self.model.audio_tower.layer_norm.bias
	        model_multi_modal_projector_linear_1_parametrizations_weight_original0: "i8[3072, 5120][5120, 1]cuda:0" = self.model.multi_modal_projector.linear_1.parametrizations.weight.original0
	        model_multi_modal_projector_linear_1_parametrizations_weight_original1: "f32[3072, 160][160, 1]cuda:0" = self.model.multi_modal_projector.linear_1.parametrizations.weight.original1
	        model_multi_modal_projector_linear_1_parametrizations_weight_original2: "i8[3072, 160][160, 1]cuda:0" = self.model.multi_modal_projector.linear_1.parametrizations.weight.original2
	        model_multi_modal_projector_linear_2_parametrizations_weight_original0: "i8[3072, 3072][3072, 1]cuda:0" = self.model.multi_modal_projector.linear_2.parametrizations.weight.original0
	        model_multi_modal_projector_linear_2_parametrizations_weight_original1: "f32[3072, 96][96, 1]cuda:0" = self.model.multi_modal_projector.linear_2.parametrizations.weight.original1
	        model_multi_modal_projector_linear_2_parametrizations_weight_original2: "i8[3072, 96][96, 1]cuda:0" = self.model.multi_modal_projector.linear_2.parametrizations.weight.original2
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:348 in forward, code: input_features = input_features.to(dtype=self.conv1.weight.dtype, device=self.conv1.weight.device)
	        _assert_tensor_metadata = torch.ops.aten._assert_tensor_metadata.default(arg877_1, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:349 in forward, code: inputs_embeds = nn.functional.gelu(self.conv1(input_features))
	        convolution: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.convolution.default(arg877_1, model_audio_tower_conv1_weight, model_audio_tower_conv1_bias, [1], [1], [1], False, [0], 1);  model_audio_tower_conv1_weight = model_audio_tower_conv1_bias = None
	        sym_size_int: "Sym(s6)" = torch.ops.aten.sym_size.int(arg877_1, 0);  arg877_1 = None
	        mul_2: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution, 0.5)
	        mul_3: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution, 0.7071067811865476);  convolution = None
	        erf: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.erf.default(mul_3);  mul_3 = None
	        add_4: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.add.Tensor(erf, 1);  erf = None
	        mul_4: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2, add_4);  mul_2 = add_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:350 in forward, code: inputs_embeds = nn.functional.gelu(self.conv2(inputs_embeds))
	        convolution_1: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.convolution.default(mul_4, model_audio_tower_conv2_weight, model_audio_tower_conv2_bias, [2], [1], [1], False, [0], 1);  mul_4 = model_audio_tower_conv2_weight = model_audio_tower_conv2_bias = None
	        mul_9: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution_1, 0.5)
	        mul_10: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution_1, 0.7071067811865476);  convolution_1 = None
	        erf_1: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.erf.default(mul_10);  mul_10 = None
	        add_13: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_1, 1);  erf_1 = None
	        mul_11: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9, add_13);  mul_9 = add_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:351 in forward, code: inputs_embeds = inputs_embeds.permute(0, 2, 1)
	        permute: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.permute.default(mul_11, [0, 2, 1]);  mul_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:354 in forward, code: hidden_states = (inputs_embeds + embed_pos).to(inputs_embeds.dtype)
	        add_22: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(permute, model_audio_tower_embed_positions_weight);  permute = model_audio_tower_embed_positions_weight = None
	        _assert_tensor_metadata_1 = torch.ops.aten._assert_tensor_metadata.default(add_22, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_22, memory_format = torch.contiguous_format)
	        var_mean = torch.ops.aten.var_mean.correction(clone_1, [2], correction = 0, keepdim = True)
	        getitem: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean[0]
	        getitem_1: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean[1];  var_mean = None
	        add_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem, 1e-05);  getitem = None
	        rsqrt: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_31);  add_31 = None
	        sub_7: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_1, getitem_1);  clone_1 = getitem_1 = None
	        mul_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7, rsqrt);  sub_7 = rsqrt = None
	        mul_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_20, model_audio_tower_layers_0_self_attn_layer_norm_weight);  mul_20 = model_audio_tower_layers_0_self_attn_layer_norm_weight = None
	        add_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_21, model_audio_tower_layers_0_self_attn_layer_norm_bias);  mul_21 = model_audio_tower_layers_0_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_32, [2])
	        amax: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_32, [2])
	        full: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin, full);  amin = full = None
	        full_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax, full_1);  amax = full_1 = None
	        sub_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum, minimum);  maximum = None
	        div: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_18, 255.0);  sub_18 = None
	        clamp_min: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div, 1.1920928955078125e-07);  div = None
	        div_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum, clamp_min);  minimum = None
	        round_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_1);  div_1 = None
	        sub_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_1);  round_1 = None
	        clamp_min_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_24, -128);  sub_24 = None
	        clamp_max: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_1, 127);  clamp_min_1 = None
	        _assert_tensor_metadata_2 = torch.ops.aten._assert_tensor_metadata.default(clamp_min, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_2 = None
	        _assert_tensor_metadata_3 = torch.ops.aten._assert_tensor_metadata.default(clamp_max, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_3 = None
	        convert_element_type: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max, torch.int8);  clamp_max = None
	        view_2: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min, [sym_size_int, 1500, 1])
	        view_3: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type, [sym_size_int, 1500, 1])
	        reciprocal: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2);  view_2 = None
	        mul_69: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal, 1.0);  reciprocal = None
	        mul_72: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_32, mul_69);  mul_69 = None
	        round_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_72);  mul_72 = None
	        add_119: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_2, view_3);  round_2 = view_3 = None
	        clamp_min_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_119, -128);  add_119 = None
	        clamp_max_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_2, 127);  clamp_min_2 = None
	        _assert_tensor_metadata_4 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_1, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_4 = None
	        convert_element_type_1: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_1, torch.int8);  clamp_max_1 = None
	        view_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min, [sym_size_int, 1500, 1]);  clamp_min = None
	        view_7: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type, [sym_size_int, 1500, 1]);  convert_element_type = None
	        _assert_tensor_metadata_5 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_5 = None
	        convert_element_type_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1, torch.float32);  convert_element_type_1 = None
	        _assert_tensor_metadata_6 = torch.ops.aten._assert_tensor_metadata.default(view_7, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_6 = None
	        convert_element_type_3: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_7, torch.float32);  view_7 = None
	        sub_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_2, convert_element_type_3);  convert_element_type_2 = convert_element_type_3 = None
	        mul_94: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_44, view_6);  sub_44 = view_6 = None
	        _assert_tensor_metadata_7 = torch.ops.aten._assert_tensor_metadata.default(mul_94, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_7 = None
	        view_9: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_10: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_11: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_8 = torch.ops.aten._assert_tensor_metadata.default(view_9, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_8 = None
	        convert_element_type_4: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_9, torch.float32);  view_9 = None
	        _assert_tensor_metadata_9 = torch.ops.aten._assert_tensor_metadata.default(view_11, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_9 = None
	        convert_element_type_5: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_11, torch.float32);  view_11 = None
	        sub_48: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_4, convert_element_type_5);  convert_element_type_4 = convert_element_type_5 = None
	        mul_99: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_48, view_10);  sub_48 = view_10 = None
	        view_12: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_99, [1280, 1280]);  mul_99 = None
	        _assert_tensor_metadata_10 = torch.ops.aten._assert_tensor_metadata.default(view_12, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_10 = None
	        mul_104: "Sym(1500*s6)" = sym_size_int * 1500
	        view_13: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_94, [mul_104, 1280]);  mul_94 = mul_104 = None
	        permute_1: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_12, [1, 0]);  view_12 = None
	        addmm: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_0_self_attn_q_proj_bias, view_13, permute_1);  model_audio_tower_layers_0_self_attn_q_proj_bias = view_13 = permute_1 = None
	        view_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm, [sym_size_int, 1500, 1280]);  addmm = None
	        mul_111: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_14, 0.125);  view_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_15: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_111, [sym_size_int, 1500, 20, 64]);  mul_111 = None
	        permute_2: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_15, [0, 2, 1, 3]);  view_15 = None
	        clone_2: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_2, memory_format = torch.contiguous_format);  permute_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_32, [2])
	        amax_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_32, [2])
	        full_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_1, full_2);  amin_1 = full_2 = None
	        full_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_1, full_3);  amax_1 = full_3 = None
	        sub_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_1, minimum_1);  maximum_1 = None
	        div_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_63, 255.0);  sub_63 = None
	        clamp_min_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_2, 1.1920928955078125e-07);  div_2 = None
	        div_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_1, clamp_min_3);  minimum_1 = None
	        round_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_3);  div_3 = None
	        sub_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_3);  round_3 = None
	        clamp_min_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_69, -128);  sub_69 = None
	        clamp_max_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_4, 127);  clamp_min_4 = None
	        _assert_tensor_metadata_11 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_3, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_11 = None
	        _assert_tensor_metadata_12 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_2, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_12 = None
	        convert_element_type_6: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_2, torch.int8);  clamp_max_2 = None
	        view_18: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_3, [sym_size_int, 1500, 1])
	        view_19: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_6, [sym_size_int, 1500, 1])
	        reciprocal_1: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_18);  view_18 = None
	        mul_165: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_1, 1.0);  reciprocal_1 = None
	        mul_168: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_32, mul_165);  mul_165 = None
	        round_4: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_168);  mul_168 = None
	        add_271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_4, view_19);  round_4 = view_19 = None
	        clamp_min_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_271, -128);  add_271 = None
	        clamp_max_3: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_5, 127);  clamp_min_5 = None
	        _assert_tensor_metadata_13 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_3, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_13 = None
	        convert_element_type_7: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_3, torch.int8);  clamp_max_3 = None
	        view_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_3, [sym_size_int, 1500, 1]);  clamp_min_3 = None
	        view_23: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_6, [sym_size_int, 1500, 1]);  convert_element_type_6 = None
	        _assert_tensor_metadata_14 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_7, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_14 = None
	        convert_element_type_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_7, torch.float32);  convert_element_type_7 = None
	        _assert_tensor_metadata_15 = torch.ops.aten._assert_tensor_metadata.default(view_23, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_15 = None
	        convert_element_type_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_23, torch.float32);  view_23 = None
	        sub_89: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_8, convert_element_type_9);  convert_element_type_8 = convert_element_type_9 = None
	        mul_190: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_89, view_22);  sub_89 = view_22 = None
	        _assert_tensor_metadata_16 = torch.ops.aten._assert_tensor_metadata.default(mul_190, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_16 = None
	        view_25: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_26: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_27: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_17 = torch.ops.aten._assert_tensor_metadata.default(view_25, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_17 = None
	        convert_element_type_10: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_25, torch.float32);  view_25 = None
	        _assert_tensor_metadata_18 = torch.ops.aten._assert_tensor_metadata.default(view_27, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_18 = None
	        convert_element_type_11: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_27, torch.float32);  view_27 = None
	        sub_93: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_10, convert_element_type_11);  convert_element_type_10 = convert_element_type_11 = None
	        mul_195: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_93, view_26);  sub_93 = view_26 = None
	        view_28: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_195, [1280, 1280]);  mul_195 = None
	        _assert_tensor_metadata_19 = torch.ops.aten._assert_tensor_metadata.default(view_28, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_19 = None
	        permute_3: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_28, [1, 0]);  view_28 = None
	        mul_198: "Sym(1500*s6)" = sym_size_int * 1500
	        view_29: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_190, [mul_198, 1280]);  mul_190 = mul_198 = None
	        mm: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_29, permute_3);  view_29 = permute_3 = None
	        view_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm, [sym_size_int, 1500, 1280]);  mm = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_31: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_30, [sym_size_int, -1, 20, 64]);  view_30 = None
	        permute_4: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_31, [0, 2, 1, 3]);  view_31 = None
	        clone_3: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_4, memory_format = torch.contiguous_format);  permute_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_32, [2])
	        amax_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_32, [2])
	        full_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_2, full_4);  amin_2 = full_4 = None
	        full_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_2, full_5);  amax_2 = full_5 = None
	        sub_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_2, minimum_2);  maximum_2 = None
	        div_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_107, 255.0);  sub_107 = None
	        clamp_min_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_4, 1.1920928955078125e-07);  div_4 = None
	        div_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_2, clamp_min_6);  minimum_2 = None
	        round_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_5);  div_5 = None
	        sub_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_5);  round_5 = None
	        clamp_min_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_113, -128);  sub_113 = None
	        clamp_max_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_7, 127);  clamp_min_7 = None
	        _assert_tensor_metadata_20 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_6, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_20 = None
	        _assert_tensor_metadata_21 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_4, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_21 = None
	        convert_element_type_12: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_4, torch.int8);  clamp_max_4 = None
	        view_34: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_6, [sym_size_int, 1500, 1])
	        view_35: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_12, [sym_size_int, 1500, 1])
	        reciprocal_2: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_34);  view_34 = None
	        mul_264: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_2, 1.0);  reciprocal_2 = None
	        mul_267: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_32, mul_264);  add_32 = mul_264 = None
	        round_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_267);  mul_267 = None
	        add_419: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_6, view_35);  round_6 = view_35 = None
	        clamp_min_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_419, -128);  add_419 = None
	        clamp_max_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_8, 127);  clamp_min_8 = None
	        _assert_tensor_metadata_22 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_5, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_22 = None
	        convert_element_type_13: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_5, torch.int8);  clamp_max_5 = None
	        view_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_6, [sym_size_int, 1500, 1]);  clamp_min_6 = None
	        view_39: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_12, [sym_size_int, 1500, 1]);  convert_element_type_12 = None
	        _assert_tensor_metadata_23 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_13, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_23 = None
	        convert_element_type_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_13, torch.float32);  convert_element_type_13 = None
	        _assert_tensor_metadata_24 = torch.ops.aten._assert_tensor_metadata.default(view_39, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_24 = None
	        convert_element_type_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_39, torch.float32);  view_39 = None
	        sub_133: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_14, convert_element_type_15);  convert_element_type_14 = convert_element_type_15 = None
	        mul_289: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_133, view_38);  sub_133 = view_38 = None
	        _assert_tensor_metadata_25 = torch.ops.aten._assert_tensor_metadata.default(mul_289, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_25 = None
	        view_41: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_42: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_43: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_26 = torch.ops.aten._assert_tensor_metadata.default(view_41, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_26 = None
	        convert_element_type_16: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_41, torch.float32);  view_41 = None
	        _assert_tensor_metadata_27 = torch.ops.aten._assert_tensor_metadata.default(view_43, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_27 = None
	        convert_element_type_17: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_43, torch.float32);  view_43 = None
	        sub_137: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_16, convert_element_type_17);  convert_element_type_16 = convert_element_type_17 = None
	        mul_294: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_137, view_42);  sub_137 = view_42 = None
	        view_44: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_294, [1280, 1280]);  mul_294 = None
	        _assert_tensor_metadata_28 = torch.ops.aten._assert_tensor_metadata.default(view_44, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_28 = None
	        mul_299: "Sym(1500*s6)" = sym_size_int * 1500
	        view_45: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_289, [mul_299, 1280]);  mul_289 = mul_299 = None
	        permute_5: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_44, [1, 0]);  view_44 = None
	        addmm_1: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_0_self_attn_v_proj_bias, view_45, permute_5);  model_audio_tower_layers_0_self_attn_v_proj_bias = view_45 = permute_5 = None
	        view_46: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_1, [sym_size_int, 1500, 1280]);  addmm_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_47: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_46, [sym_size_int, -1, 20, 64]);  view_46 = None
	        permute_6: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_47, [0, 2, 1, 3]);  view_47 = None
	        clone_4: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_6, memory_format = torch.contiguous_format);  permute_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_2, clone_3, clone_4, None, False, scale = 1.0);  clone_2 = clone_3 = clone_4 = None
	        getitem_2: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention[0];  _scaled_dot_product_efficient_attention = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_7: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_2, [0, 2, 1, 3]);  getitem_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_48: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_7, [sym_size_int, 1500, -1]);  permute_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_48, [2])
	        amax_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_48, [2])
	        full_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_3, full_6);  amin_3 = full_6 = None
	        full_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_3, full_7);  amax_3 = full_7 = None
	        sub_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_3, minimum_3);  maximum_3 = None
	        div_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_155, 255.0);  sub_155 = None
	        clamp_min_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_6, 1.1920928955078125e-07);  div_6 = None
	        div_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_3, clamp_min_9);  minimum_3 = None
	        round_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_7);  div_7 = None
	        sub_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_7);  round_7 = None
	        clamp_min_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_161, -128);  sub_161 = None
	        clamp_max_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_10, 127);  clamp_min_10 = None
	        _assert_tensor_metadata_29 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_9, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_29 = None
	        _assert_tensor_metadata_30 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_6, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_30 = None
	        convert_element_type_18: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_6, torch.int8);  clamp_max_6 = None
	        view_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_9, [sym_size_int, 1500, 1])
	        view_52: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_18, [sym_size_int, 1500, 1])
	        reciprocal_3: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_51);  view_51 = None
	        mul_369: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_3, 1.0);  reciprocal_3 = None
	        mul_372: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_48, mul_369);  view_48 = mul_369 = None
	        round_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_372);  mul_372 = None
	        add_583: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_8, view_52);  round_8 = view_52 = None
	        clamp_min_11: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_583, -128);  add_583 = None
	        clamp_max_7: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_11, 127);  clamp_min_11 = None
	        _assert_tensor_metadata_31 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_7, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_31 = None
	        convert_element_type_19: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_7, torch.int8);  clamp_max_7 = None
	        view_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_9, [sym_size_int, 1500, 1]);  clamp_min_9 = None
	        view_56: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_18, [sym_size_int, 1500, 1]);  convert_element_type_18 = None
	        _assert_tensor_metadata_32 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_19, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_32 = None
	        convert_element_type_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_19, torch.float32);  convert_element_type_19 = None
	        _assert_tensor_metadata_33 = torch.ops.aten._assert_tensor_metadata.default(view_56, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_33 = None
	        convert_element_type_21: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_56, torch.float32);  view_56 = None
	        sub_181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_20, convert_element_type_21);  convert_element_type_20 = convert_element_type_21 = None
	        mul_394: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_181, view_55);  sub_181 = view_55 = None
	        _assert_tensor_metadata_34 = torch.ops.aten._assert_tensor_metadata.default(mul_394, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_34 = None
	        view_58: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_59: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_60: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_35 = torch.ops.aten._assert_tensor_metadata.default(view_58, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_35 = None
	        convert_element_type_22: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_58, torch.float32);  view_58 = None
	        _assert_tensor_metadata_36 = torch.ops.aten._assert_tensor_metadata.default(view_60, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_36 = None
	        convert_element_type_23: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_60, torch.float32);  view_60 = None
	        sub_185: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_22, convert_element_type_23);  convert_element_type_22 = convert_element_type_23 = None
	        mul_399: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_185, view_59);  sub_185 = view_59 = None
	        view_61: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_399, [1280, 1280]);  mul_399 = None
	        _assert_tensor_metadata_37 = torch.ops.aten._assert_tensor_metadata.default(view_61, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_37 = None
	        mul_404: "Sym(1500*s6)" = sym_size_int * 1500
	        view_62: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_394, [mul_404, 1280]);  mul_394 = mul_404 = None
	        permute_8: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_61, [1, 0]);  view_61 = None
	        addmm_2: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_0_self_attn_out_proj_bias, view_62, permute_8);  model_audio_tower_layers_0_self_attn_out_proj_bias = view_62 = permute_8 = None
	        view_63: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_2, [sym_size_int, 1500, 1280]);  addmm_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_646: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_22, view_63);  add_22 = view_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_646, memory_format = torch.contiguous_format)
	        var_mean_1 = torch.ops.aten.var_mean.correction(clone_6, [2], correction = 0, keepdim = True)
	        getitem_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_1[0]
	        getitem_7: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_1[1];  var_mean_1 = None
	        add_651: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_6, 1e-05);  getitem_6 = None
	        rsqrt_1: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_651);  add_651 = None
	        sub_191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_6, getitem_7);  clone_6 = getitem_7 = None
	        mul_415: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_191, rsqrt_1);  sub_191 = rsqrt_1 = None
	        mul_416: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_415, model_audio_tower_layers_0_final_layer_norm_weight);  mul_415 = model_audio_tower_layers_0_final_layer_norm_weight = None
	        add_652: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_416, model_audio_tower_layers_0_final_layer_norm_bias);  mul_416 = model_audio_tower_layers_0_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_652, [2])
	        amax_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_652, [2])
	        full_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_4, full_8);  amin_4 = full_8 = None
	        full_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_4, full_9);  amax_4 = full_9 = None
	        sub_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_4, minimum_4);  maximum_4 = None
	        div_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_202, 255.0);  sub_202 = None
	        clamp_min_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_8, 1.1920928955078125e-07);  div_8 = None
	        div_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_4, clamp_min_12);  minimum_4 = None
	        round_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_9);  div_9 = None
	        sub_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_9);  round_9 = None
	        clamp_min_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_208, -128);  sub_208 = None
	        clamp_max_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_13, 127);  clamp_min_13 = None
	        _assert_tensor_metadata_38 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_12, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_38 = None
	        _assert_tensor_metadata_39 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_8, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_39 = None
	        convert_element_type_24: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_8, torch.int8);  clamp_max_8 = None
	        view_66: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_12, [sym_size_int, 1500, 1])
	        view_67: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_24, [sym_size_int, 1500, 1])
	        reciprocal_4: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_66);  view_66 = None
	        mul_464: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_4, 1.0);  reciprocal_4 = None
	        mul_467: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_652, mul_464);  add_652 = mul_464 = None
	        round_10: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_467);  mul_467 = None
	        add_739: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_10, view_67);  round_10 = view_67 = None
	        clamp_min_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_739, -128);  add_739 = None
	        clamp_max_9: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_14, 127);  clamp_min_14 = None
	        _assert_tensor_metadata_40 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_9, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_40 = None
	        convert_element_type_25: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_9, torch.int8);  clamp_max_9 = None
	        view_70: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_12, [sym_size_int, 1500, 1]);  clamp_min_12 = None
	        view_71: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_24, [sym_size_int, 1500, 1]);  convert_element_type_24 = None
	        _assert_tensor_metadata_41 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_25, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_41 = None
	        convert_element_type_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_25, torch.float32);  convert_element_type_25 = None
	        _assert_tensor_metadata_42 = torch.ops.aten._assert_tensor_metadata.default(view_71, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_42 = None
	        convert_element_type_27: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_71, torch.float32);  view_71 = None
	        sub_228: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_26, convert_element_type_27);  convert_element_type_26 = convert_element_type_27 = None
	        mul_489: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_228, view_70);  sub_228 = view_70 = None
	        _assert_tensor_metadata_43 = torch.ops.aten._assert_tensor_metadata.default(mul_489, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_43 = None
	        view_73: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_0_fc1_parametrizations_weight_original0 = None
	        view_74: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_0_fc1_parametrizations_weight_original1 = None
	        view_75: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_0_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_44 = torch.ops.aten._assert_tensor_metadata.default(view_73, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_44 = None
	        convert_element_type_28: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_73, torch.float32);  view_73 = None
	        _assert_tensor_metadata_45 = torch.ops.aten._assert_tensor_metadata.default(view_75, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_45 = None
	        convert_element_type_29: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_75, torch.float32);  view_75 = None
	        sub_232: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_28, convert_element_type_29);  convert_element_type_28 = convert_element_type_29 = None
	        mul_494: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_232, view_74);  sub_232 = view_74 = None
	        view_76: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_494, [5120, 1280]);  mul_494 = None
	        _assert_tensor_metadata_46 = torch.ops.aten._assert_tensor_metadata.default(view_76, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_46 = None
	        mul_499: "Sym(1500*s6)" = sym_size_int * 1500
	        view_77: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_489, [mul_499, 1280]);  mul_489 = mul_499 = None
	        permute_9: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_76, [1, 0]);  view_76 = None
	        addmm_3: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_0_fc1_bias, view_77, permute_9);  model_audio_tower_layers_0_fc1_bias = view_77 = permute_9 = None
	        view_78: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_3, [sym_size_int, 1500, 5120]);  addmm_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_506: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_78, 0.5)
	        mul_507: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_78, 0.7071067811865476);  view_78 = None
	        erf_2: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_507);  mul_507 = None
	        add_798: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_2, 1);  erf_2 = None
	        mul_508: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_506, add_798);  mul_506 = add_798 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_508, [2])
	        amax_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_508, [2])
	        full_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_5, full_10);  amin_5 = full_10 = None
	        full_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_5, full_11);  amax_5 = full_11 = None
	        sub_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_5, minimum_5);  maximum_5 = None
	        div_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_245, 255.0);  sub_245 = None
	        clamp_min_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_10, 1.1920928955078125e-07);  div_10 = None
	        div_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_5, clamp_min_15);  minimum_5 = None
	        round_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_11);  div_11 = None
	        sub_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_11);  round_11 = None
	        clamp_min_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_251, -128);  sub_251 = None
	        clamp_max_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_16, 127);  clamp_min_16 = None
	        _assert_tensor_metadata_47 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_15, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_47 = None
	        _assert_tensor_metadata_48 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_10, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_48 = None
	        convert_element_type_30: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_10, torch.int8);  clamp_max_10 = None
	        view_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_15, [sym_size_int, 1500, 1])
	        view_82: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_30, [sym_size_int, 1500, 1])
	        reciprocal_5: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_81);  view_81 = None
	        mul_554: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_5, 1.0);  reciprocal_5 = None
	        mul_557: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_508, mul_554);  mul_508 = mul_554 = None
	        round_12: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_557);  mul_557 = None
	        add_881: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_12, view_82);  round_12 = view_82 = None
	        clamp_min_17: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_881, -128);  add_881 = None
	        clamp_max_11: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_17, 127);  clamp_min_17 = None
	        _assert_tensor_metadata_49 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_11, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_49 = None
	        convert_element_type_31: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_11, torch.int8);  clamp_max_11 = None
	        view_85: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_15, [sym_size_int, 1500, 1]);  clamp_min_15 = None
	        view_86: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_30, [sym_size_int, 1500, 1]);  convert_element_type_30 = None
	        _assert_tensor_metadata_50 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_31, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_50 = None
	        convert_element_type_32: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_31, torch.float32);  convert_element_type_31 = None
	        _assert_tensor_metadata_51 = torch.ops.aten._assert_tensor_metadata.default(view_86, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_51 = None
	        convert_element_type_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_86, torch.float32);  view_86 = None
	        sub_271: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_32, convert_element_type_33);  convert_element_type_32 = convert_element_type_33 = None
	        mul_579: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_271, view_85);  sub_271 = view_85 = None
	        _assert_tensor_metadata_52 = torch.ops.aten._assert_tensor_metadata.default(mul_579, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_52 = None
	        view_88: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_0_fc2_parametrizations_weight_original0 = None
	        view_89: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_0_fc2_parametrizations_weight_original1 = None
	        view_90: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_0_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_0_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_53 = torch.ops.aten._assert_tensor_metadata.default(view_88, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_53 = None
	        convert_element_type_34: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_88, torch.float32);  view_88 = None
	        _assert_tensor_metadata_54 = torch.ops.aten._assert_tensor_metadata.default(view_90, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_54 = None
	        convert_element_type_35: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_90, torch.float32);  view_90 = None
	        sub_275: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_34, convert_element_type_35);  convert_element_type_34 = convert_element_type_35 = None
	        mul_584: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_275, view_89);  sub_275 = view_89 = None
	        view_91: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_584, [1280, 5120]);  mul_584 = None
	        _assert_tensor_metadata_55 = torch.ops.aten._assert_tensor_metadata.default(view_91, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_55 = None
	        mul_589: "Sym(1500*s6)" = sym_size_int * 1500
	        view_92: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_579, [mul_589, 5120]);  mul_579 = mul_589 = None
	        permute_10: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_91, [1, 0]);  view_91 = None
	        addmm_4: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_0_fc2_bias, view_92, permute_10);  model_audio_tower_layers_0_fc2_bias = view_92 = permute_10 = None
	        view_93: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_4, [sym_size_int, 1500, 1280]);  addmm_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_944: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_646, view_93);  add_646 = view_93 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_9: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_944, memory_format = torch.contiguous_format)
	        var_mean_2 = torch.ops.aten.var_mean.correction(clone_9, [2], correction = 0, keepdim = True)
	        getitem_8: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_2[0]
	        getitem_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_2[1];  var_mean_2 = None
	        add_949: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_8, 1e-05);  getitem_8 = None
	        rsqrt_2: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_949);  add_949 = None
	        sub_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_9, getitem_9);  clone_9 = getitem_9 = None
	        mul_600: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_281, rsqrt_2);  sub_281 = rsqrt_2 = None
	        mul_601: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_600, model_audio_tower_layers_1_self_attn_layer_norm_weight);  mul_600 = model_audio_tower_layers_1_self_attn_layer_norm_weight = None
	        add_950: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_601, model_audio_tower_layers_1_self_attn_layer_norm_bias);  mul_601 = model_audio_tower_layers_1_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_950, [2])
	        amax_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_950, [2])
	        full_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_6, full_12);  amin_6 = full_12 = None
	        full_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_6, full_13);  amax_6 = full_13 = None
	        sub_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_6, minimum_6);  maximum_6 = None
	        div_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_292, 255.0);  sub_292 = None
	        clamp_min_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_12, 1.1920928955078125e-07);  div_12 = None
	        div_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_6, clamp_min_18);  minimum_6 = None
	        round_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_13);  div_13 = None
	        sub_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_13);  round_13 = None
	        clamp_min_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_298, -128);  sub_298 = None
	        clamp_max_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_19, 127);  clamp_min_19 = None
	        _assert_tensor_metadata_56 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_18, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_56 = None
	        _assert_tensor_metadata_57 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_12, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_57 = None
	        convert_element_type_36: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_12, torch.int8);  clamp_max_12 = None
	        view_96: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_18, [sym_size_int, 1500, 1])
	        view_97: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_36, [sym_size_int, 1500, 1])
	        reciprocal_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_96);  view_96 = None
	        mul_649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_6, 1.0);  reciprocal_6 = None
	        mul_652: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_950, mul_649);  mul_649 = None
	        round_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_652);  mul_652 = None
	        add_1037: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_14, view_97);  round_14 = view_97 = None
	        clamp_min_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1037, -128);  add_1037 = None
	        clamp_max_13: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_20, 127);  clamp_min_20 = None
	        _assert_tensor_metadata_58 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_13, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_58 = None
	        convert_element_type_37: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_13, torch.int8);  clamp_max_13 = None
	        view_100: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_18, [sym_size_int, 1500, 1]);  clamp_min_18 = None
	        view_101: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_36, [sym_size_int, 1500, 1]);  convert_element_type_36 = None
	        _assert_tensor_metadata_59 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_37, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_59 = None
	        convert_element_type_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_37, torch.float32);  convert_element_type_37 = None
	        _assert_tensor_metadata_60 = torch.ops.aten._assert_tensor_metadata.default(view_101, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_60 = None
	        convert_element_type_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_101, torch.float32);  view_101 = None
	        sub_318: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_38, convert_element_type_39);  convert_element_type_38 = convert_element_type_39 = None
	        mul_674: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_318, view_100);  sub_318 = view_100 = None
	        _assert_tensor_metadata_61 = torch.ops.aten._assert_tensor_metadata.default(mul_674, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_61 = None
	        view_103: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_104: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_105: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_62 = torch.ops.aten._assert_tensor_metadata.default(view_103, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_62 = None
	        convert_element_type_40: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_103, torch.float32);  view_103 = None
	        _assert_tensor_metadata_63 = torch.ops.aten._assert_tensor_metadata.default(view_105, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_63 = None
	        convert_element_type_41: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_105, torch.float32);  view_105 = None
	        sub_322: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_40, convert_element_type_41);  convert_element_type_40 = convert_element_type_41 = None
	        mul_679: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_322, view_104);  sub_322 = view_104 = None
	        view_106: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_679, [1280, 1280]);  mul_679 = None
	        _assert_tensor_metadata_64 = torch.ops.aten._assert_tensor_metadata.default(view_106, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_64 = None
	        mul_684: "Sym(1500*s6)" = sym_size_int * 1500
	        view_107: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_674, [mul_684, 1280]);  mul_674 = mul_684 = None
	        permute_11: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_106, [1, 0]);  view_106 = None
	        addmm_5: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_1_self_attn_q_proj_bias, view_107, permute_11);  model_audio_tower_layers_1_self_attn_q_proj_bias = view_107 = permute_11 = None
	        view_108: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_5, [sym_size_int, 1500, 1280]);  addmm_5 = None
	        mul_691: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_108, 0.125);  view_108 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_109: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_691, [sym_size_int, 1500, 20, 64]);  mul_691 = None
	        permute_12: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_109, [0, 2, 1, 3]);  view_109 = None
	        clone_10: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_12, memory_format = torch.contiguous_format);  permute_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_950, [2])
	        amax_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_950, [2])
	        full_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_7, full_14);  amin_7 = full_14 = None
	        full_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_7, full_15);  amax_7 = full_15 = None
	        sub_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_7, minimum_7);  maximum_7 = None
	        div_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_337, 255.0);  sub_337 = None
	        clamp_min_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_14, 1.1920928955078125e-07);  div_14 = None
	        div_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_7, clamp_min_21);  minimum_7 = None
	        round_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_15);  div_15 = None
	        sub_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_15);  round_15 = None
	        clamp_min_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_343, -128);  sub_343 = None
	        clamp_max_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_22, 127);  clamp_min_22 = None
	        _assert_tensor_metadata_65 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_21, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_65 = None
	        _assert_tensor_metadata_66 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_14, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_66 = None
	        convert_element_type_42: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_14, torch.int8);  clamp_max_14 = None
	        view_112: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_21, [sym_size_int, 1500, 1])
	        view_113: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_42, [sym_size_int, 1500, 1])
	        reciprocal_7: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_112);  view_112 = None
	        mul_745: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_7, 1.0);  reciprocal_7 = None
	        mul_748: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_950, mul_745);  mul_745 = None
	        round_16: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_748);  mul_748 = None
	        add_1189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_16, view_113);  round_16 = view_113 = None
	        clamp_min_23: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1189, -128);  add_1189 = None
	        clamp_max_15: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_23, 127);  clamp_min_23 = None
	        _assert_tensor_metadata_67 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_15, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_67 = None
	        convert_element_type_43: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_15, torch.int8);  clamp_max_15 = None
	        view_116: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_21, [sym_size_int, 1500, 1]);  clamp_min_21 = None
	        view_117: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_42, [sym_size_int, 1500, 1]);  convert_element_type_42 = None
	        _assert_tensor_metadata_68 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_43, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_68 = None
	        convert_element_type_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_43, torch.float32);  convert_element_type_43 = None
	        _assert_tensor_metadata_69 = torch.ops.aten._assert_tensor_metadata.default(view_117, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_69 = None
	        convert_element_type_45: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_117, torch.float32);  view_117 = None
	        sub_363: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_44, convert_element_type_45);  convert_element_type_44 = convert_element_type_45 = None
	        mul_770: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_363, view_116);  sub_363 = view_116 = None
	        _assert_tensor_metadata_70 = torch.ops.aten._assert_tensor_metadata.default(mul_770, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_70 = None
	        view_119: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_120: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_121: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_71 = torch.ops.aten._assert_tensor_metadata.default(view_119, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_71 = None
	        convert_element_type_46: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_119, torch.float32);  view_119 = None
	        _assert_tensor_metadata_72 = torch.ops.aten._assert_tensor_metadata.default(view_121, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_72 = None
	        convert_element_type_47: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_121, torch.float32);  view_121 = None
	        sub_367: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_46, convert_element_type_47);  convert_element_type_46 = convert_element_type_47 = None
	        mul_775: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_367, view_120);  sub_367 = view_120 = None
	        view_122: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_775, [1280, 1280]);  mul_775 = None
	        _assert_tensor_metadata_73 = torch.ops.aten._assert_tensor_metadata.default(view_122, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_73 = None
	        permute_13: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_122, [1, 0]);  view_122 = None
	        mul_778: "Sym(1500*s6)" = sym_size_int * 1500
	        view_123: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_770, [mul_778, 1280]);  mul_770 = mul_778 = None
	        mm_1: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_123, permute_13);  view_123 = permute_13 = None
	        view_124: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_1, [sym_size_int, 1500, 1280]);  mm_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_125: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_124, [sym_size_int, -1, 20, 64]);  view_124 = None
	        permute_14: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_125, [0, 2, 1, 3]);  view_125 = None
	        clone_11: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_14, memory_format = torch.contiguous_format);  permute_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_950, [2])
	        amax_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_950, [2])
	        full_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_8, full_16);  amin_8 = full_16 = None
	        full_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_8, full_17);  amax_8 = full_17 = None
	        sub_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_8, minimum_8);  maximum_8 = None
	        div_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_381, 255.0);  sub_381 = None
	        clamp_min_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_16, 1.1920928955078125e-07);  div_16 = None
	        div_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_8, clamp_min_24);  minimum_8 = None
	        round_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_17);  div_17 = None
	        sub_387: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_17);  round_17 = None
	        clamp_min_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_387, -128);  sub_387 = None
	        clamp_max_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_25, 127);  clamp_min_25 = None
	        _assert_tensor_metadata_74 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_24, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_74 = None
	        _assert_tensor_metadata_75 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_16, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_75 = None
	        convert_element_type_48: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_16, torch.int8);  clamp_max_16 = None
	        view_128: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_24, [sym_size_int, 1500, 1])
	        view_129: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_48, [sym_size_int, 1500, 1])
	        reciprocal_8: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_128);  view_128 = None
	        mul_844: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_8, 1.0);  reciprocal_8 = None
	        mul_847: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_950, mul_844);  add_950 = mul_844 = None
	        round_18: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_847);  mul_847 = None
	        add_1337: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_18, view_129);  round_18 = view_129 = None
	        clamp_min_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1337, -128);  add_1337 = None
	        clamp_max_17: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_26, 127);  clamp_min_26 = None
	        _assert_tensor_metadata_76 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_17, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_76 = None
	        convert_element_type_49: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_17, torch.int8);  clamp_max_17 = None
	        view_132: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_24, [sym_size_int, 1500, 1]);  clamp_min_24 = None
	        view_133: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_48, [sym_size_int, 1500, 1]);  convert_element_type_48 = None
	        _assert_tensor_metadata_77 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_49, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_77 = None
	        convert_element_type_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_49, torch.float32);  convert_element_type_49 = None
	        _assert_tensor_metadata_78 = torch.ops.aten._assert_tensor_metadata.default(view_133, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_78 = None
	        convert_element_type_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_133, torch.float32);  view_133 = None
	        sub_407: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_50, convert_element_type_51);  convert_element_type_50 = convert_element_type_51 = None
	        mul_869: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_407, view_132);  sub_407 = view_132 = None
	        _assert_tensor_metadata_79 = torch.ops.aten._assert_tensor_metadata.default(mul_869, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_79 = None
	        view_135: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_136: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_137: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_80 = torch.ops.aten._assert_tensor_metadata.default(view_135, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_80 = None
	        convert_element_type_52: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_135, torch.float32);  view_135 = None
	        _assert_tensor_metadata_81 = torch.ops.aten._assert_tensor_metadata.default(view_137, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_81 = None
	        convert_element_type_53: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_137, torch.float32);  view_137 = None
	        sub_411: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_52, convert_element_type_53);  convert_element_type_52 = convert_element_type_53 = None
	        mul_874: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_411, view_136);  sub_411 = view_136 = None
	        view_138: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_874, [1280, 1280]);  mul_874 = None
	        _assert_tensor_metadata_82 = torch.ops.aten._assert_tensor_metadata.default(view_138, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_82 = None
	        mul_879: "Sym(1500*s6)" = sym_size_int * 1500
	        view_139: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_869, [mul_879, 1280]);  mul_869 = mul_879 = None
	        permute_15: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_138, [1, 0]);  view_138 = None
	        addmm_6: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_1_self_attn_v_proj_bias, view_139, permute_15);  model_audio_tower_layers_1_self_attn_v_proj_bias = view_139 = permute_15 = None
	        view_140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_6, [sym_size_int, 1500, 1280]);  addmm_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_141: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_140, [sym_size_int, -1, 20, 64]);  view_140 = None
	        permute_16: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_141, [0, 2, 1, 3]);  view_141 = None
	        clone_12: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_16, memory_format = torch.contiguous_format);  permute_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_1 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_10, clone_11, clone_12, None, False, scale = 1.0);  clone_10 = clone_11 = clone_12 = None
	        getitem_10: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_1[0];  _scaled_dot_product_efficient_attention_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_17: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_10, [0, 2, 1, 3]);  getitem_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_17, [sym_size_int, 1500, -1]);  permute_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_142, [2])
	        amax_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_142, [2])
	        full_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_9, full_18);  amin_9 = full_18 = None
	        full_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_9, full_19);  amax_9 = full_19 = None
	        sub_429: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_9, minimum_9);  maximum_9 = None
	        div_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_429, 255.0);  sub_429 = None
	        clamp_min_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_18, 1.1920928955078125e-07);  div_18 = None
	        div_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_9, clamp_min_27);  minimum_9 = None
	        round_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_19);  div_19 = None
	        sub_435: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_19);  round_19 = None
	        clamp_min_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_435, -128);  sub_435 = None
	        clamp_max_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_28, 127);  clamp_min_28 = None
	        _assert_tensor_metadata_83 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_27, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_83 = None
	        _assert_tensor_metadata_84 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_18, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_84 = None
	        convert_element_type_54: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_18, torch.int8);  clamp_max_18 = None
	        view_145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_27, [sym_size_int, 1500, 1])
	        view_146: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_54, [sym_size_int, 1500, 1])
	        reciprocal_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_145);  view_145 = None
	        mul_949: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_9, 1.0);  reciprocal_9 = None
	        mul_952: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_142, mul_949);  view_142 = mul_949 = None
	        round_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_952);  mul_952 = None
	        add_1501: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_20, view_146);  round_20 = view_146 = None
	        clamp_min_29: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1501, -128);  add_1501 = None
	        clamp_max_19: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_29, 127);  clamp_min_29 = None
	        _assert_tensor_metadata_85 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_19, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_85 = None
	        convert_element_type_55: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_19, torch.int8);  clamp_max_19 = None
	        view_149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_27, [sym_size_int, 1500, 1]);  clamp_min_27 = None
	        view_150: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_54, [sym_size_int, 1500, 1]);  convert_element_type_54 = None
	        _assert_tensor_metadata_86 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_55, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_86 = None
	        convert_element_type_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_55, torch.float32);  convert_element_type_55 = None
	        _assert_tensor_metadata_87 = torch.ops.aten._assert_tensor_metadata.default(view_150, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_87 = None
	        convert_element_type_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_150, torch.float32);  view_150 = None
	        sub_455: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_56, convert_element_type_57);  convert_element_type_56 = convert_element_type_57 = None
	        mul_974: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_455, view_149);  sub_455 = view_149 = None
	        _assert_tensor_metadata_88 = torch.ops.aten._assert_tensor_metadata.default(mul_974, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_88 = None
	        view_152: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_153: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_154: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_89 = torch.ops.aten._assert_tensor_metadata.default(view_152, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_89 = None
	        convert_element_type_58: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_152, torch.float32);  view_152 = None
	        _assert_tensor_metadata_90 = torch.ops.aten._assert_tensor_metadata.default(view_154, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_90 = None
	        convert_element_type_59: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_154, torch.float32);  view_154 = None
	        sub_459: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_58, convert_element_type_59);  convert_element_type_58 = convert_element_type_59 = None
	        mul_979: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_459, view_153);  sub_459 = view_153 = None
	        view_155: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_979, [1280, 1280]);  mul_979 = None
	        _assert_tensor_metadata_91 = torch.ops.aten._assert_tensor_metadata.default(view_155, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_91 = None
	        mul_984: "Sym(1500*s6)" = sym_size_int * 1500
	        view_156: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_974, [mul_984, 1280]);  mul_974 = mul_984 = None
	        permute_18: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_155, [1, 0]);  view_155 = None
	        addmm_7: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_1_self_attn_out_proj_bias, view_156, permute_18);  model_audio_tower_layers_1_self_attn_out_proj_bias = view_156 = permute_18 = None
	        view_157: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_7, [sym_size_int, 1500, 1280]);  addmm_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_1564: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_944, view_157);  add_944 = view_157 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_1564, memory_format = torch.contiguous_format)
	        var_mean_3 = torch.ops.aten.var_mean.correction(clone_14, [2], correction = 0, keepdim = True)
	        getitem_14: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_3[0]
	        getitem_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_3[1];  var_mean_3 = None
	        add_1569: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_14, 1e-05);  getitem_14 = None
	        rsqrt_3: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_1569);  add_1569 = None
	        sub_465: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_14, getitem_15);  clone_14 = getitem_15 = None
	        mul_995: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_465, rsqrt_3);  sub_465 = rsqrt_3 = None
	        mul_996: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_995, model_audio_tower_layers_1_final_layer_norm_weight);  mul_995 = model_audio_tower_layers_1_final_layer_norm_weight = None
	        add_1570: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_996, model_audio_tower_layers_1_final_layer_norm_bias);  mul_996 = model_audio_tower_layers_1_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_1570, [2])
	        amax_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_1570, [2])
	        full_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_10, full_20);  amin_10 = full_20 = None
	        full_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_10, full_21);  amax_10 = full_21 = None
	        sub_476: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_10, minimum_10);  maximum_10 = None
	        div_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_476, 255.0);  sub_476 = None
	        clamp_min_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_20, 1.1920928955078125e-07);  div_20 = None
	        div_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_10, clamp_min_30);  minimum_10 = None
	        round_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_21);  div_21 = None
	        sub_482: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_21);  round_21 = None
	        clamp_min_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_482, -128);  sub_482 = None
	        clamp_max_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_31, 127);  clamp_min_31 = None
	        _assert_tensor_metadata_92 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_30, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_92 = None
	        _assert_tensor_metadata_93 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_20, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_93 = None
	        convert_element_type_60: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_20, torch.int8);  clamp_max_20 = None
	        view_160: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_30, [sym_size_int, 1500, 1])
	        view_161: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_60, [sym_size_int, 1500, 1])
	        reciprocal_10: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_160);  view_160 = None
	        mul_1044: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_10, 1.0);  reciprocal_10 = None
	        mul_1047: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_1570, mul_1044);  add_1570 = mul_1044 = None
	        round_22: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1047);  mul_1047 = None
	        add_1657: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_22, view_161);  round_22 = view_161 = None
	        clamp_min_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1657, -128);  add_1657 = None
	        clamp_max_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_32, 127);  clamp_min_32 = None
	        _assert_tensor_metadata_94 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_21, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_94 = None
	        convert_element_type_61: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_21, torch.int8);  clamp_max_21 = None
	        view_164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_30, [sym_size_int, 1500, 1]);  clamp_min_30 = None
	        view_165: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_60, [sym_size_int, 1500, 1]);  convert_element_type_60 = None
	        _assert_tensor_metadata_95 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_61, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_95 = None
	        convert_element_type_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_61, torch.float32);  convert_element_type_61 = None
	        _assert_tensor_metadata_96 = torch.ops.aten._assert_tensor_metadata.default(view_165, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_96 = None
	        convert_element_type_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_165, torch.float32);  view_165 = None
	        sub_502: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_62, convert_element_type_63);  convert_element_type_62 = convert_element_type_63 = None
	        mul_1069: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_502, view_164);  sub_502 = view_164 = None
	        _assert_tensor_metadata_97 = torch.ops.aten._assert_tensor_metadata.default(mul_1069, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_97 = None
	        view_167: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_1_fc1_parametrizations_weight_original0 = None
	        view_168: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_1_fc1_parametrizations_weight_original1 = None
	        view_169: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_1_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_98 = torch.ops.aten._assert_tensor_metadata.default(view_167, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_98 = None
	        convert_element_type_64: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_167, torch.float32);  view_167 = None
	        _assert_tensor_metadata_99 = torch.ops.aten._assert_tensor_metadata.default(view_169, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_99 = None
	        convert_element_type_65: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_169, torch.float32);  view_169 = None
	        sub_506: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_64, convert_element_type_65);  convert_element_type_64 = convert_element_type_65 = None
	        mul_1074: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_506, view_168);  sub_506 = view_168 = None
	        view_170: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1074, [5120, 1280]);  mul_1074 = None
	        _assert_tensor_metadata_100 = torch.ops.aten._assert_tensor_metadata.default(view_170, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_100 = None
	        mul_1079: "Sym(1500*s6)" = sym_size_int * 1500
	        view_171: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1069, [mul_1079, 1280]);  mul_1069 = mul_1079 = None
	        permute_19: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_170, [1, 0]);  view_170 = None
	        addmm_8: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_1_fc1_bias, view_171, permute_19);  model_audio_tower_layers_1_fc1_bias = view_171 = permute_19 = None
	        view_172: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_8, [sym_size_int, 1500, 5120]);  addmm_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_1086: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_172, 0.5)
	        mul_1087: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_172, 0.7071067811865476);  view_172 = None
	        erf_3: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_1087);  mul_1087 = None
	        add_1716: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_3, 1);  erf_3 = None
	        mul_1088: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1086, add_1716);  mul_1086 = add_1716 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_1088, [2])
	        amax_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_1088, [2])
	        full_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_11, full_22);  amin_11 = full_22 = None
	        full_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_11, full_23);  amax_11 = full_23 = None
	        sub_519: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_11, minimum_11);  maximum_11 = None
	        div_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_519, 255.0);  sub_519 = None
	        clamp_min_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_22, 1.1920928955078125e-07);  div_22 = None
	        div_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_11, clamp_min_33);  minimum_11 = None
	        round_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_23);  div_23 = None
	        sub_525: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_23);  round_23 = None
	        clamp_min_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_525, -128);  sub_525 = None
	        clamp_max_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_34, 127);  clamp_min_34 = None
	        _assert_tensor_metadata_101 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_33, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_101 = None
	        _assert_tensor_metadata_102 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_22, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_102 = None
	        convert_element_type_66: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_22, torch.int8);  clamp_max_22 = None
	        view_175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_33, [sym_size_int, 1500, 1])
	        view_176: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_66, [sym_size_int, 1500, 1])
	        reciprocal_11: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_175);  view_175 = None
	        mul_1134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_11, 1.0);  reciprocal_11 = None
	        mul_1137: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1088, mul_1134);  mul_1088 = mul_1134 = None
	        round_24: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_1137);  mul_1137 = None
	        add_1799: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_24, view_176);  round_24 = view_176 = None
	        clamp_min_35: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1799, -128);  add_1799 = None
	        clamp_max_23: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_35, 127);  clamp_min_35 = None
	        _assert_tensor_metadata_103 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_23, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_103 = None
	        convert_element_type_67: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_23, torch.int8);  clamp_max_23 = None
	        view_179: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_33, [sym_size_int, 1500, 1]);  clamp_min_33 = None
	        view_180: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_66, [sym_size_int, 1500, 1]);  convert_element_type_66 = None
	        _assert_tensor_metadata_104 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_67, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_104 = None
	        convert_element_type_68: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_67, torch.float32);  convert_element_type_67 = None
	        _assert_tensor_metadata_105 = torch.ops.aten._assert_tensor_metadata.default(view_180, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_105 = None
	        convert_element_type_69: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_180, torch.float32);  view_180 = None
	        sub_545: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_68, convert_element_type_69);  convert_element_type_68 = convert_element_type_69 = None
	        mul_1159: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_545, view_179);  sub_545 = view_179 = None
	        _assert_tensor_metadata_106 = torch.ops.aten._assert_tensor_metadata.default(mul_1159, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_106 = None
	        view_182: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_1_fc2_parametrizations_weight_original0 = None
	        view_183: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_1_fc2_parametrizations_weight_original1 = None
	        view_184: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_1_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_1_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_107 = torch.ops.aten._assert_tensor_metadata.default(view_182, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_107 = None
	        convert_element_type_70: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_182, torch.float32);  view_182 = None
	        _assert_tensor_metadata_108 = torch.ops.aten._assert_tensor_metadata.default(view_184, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_108 = None
	        convert_element_type_71: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_184, torch.float32);  view_184 = None
	        sub_549: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_70, convert_element_type_71);  convert_element_type_70 = convert_element_type_71 = None
	        mul_1164: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_549, view_183);  sub_549 = view_183 = None
	        view_185: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_1164, [1280, 5120]);  mul_1164 = None
	        _assert_tensor_metadata_109 = torch.ops.aten._assert_tensor_metadata.default(view_185, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_109 = None
	        mul_1169: "Sym(1500*s6)" = sym_size_int * 1500
	        view_186: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_1159, [mul_1169, 5120]);  mul_1159 = mul_1169 = None
	        permute_20: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_185, [1, 0]);  view_185 = None
	        addmm_9: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_1_fc2_bias, view_186, permute_20);  model_audio_tower_layers_1_fc2_bias = view_186 = permute_20 = None
	        view_187: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_9, [sym_size_int, 1500, 1280]);  addmm_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_1862: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_1564, view_187);  add_1564 = view_187 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_17: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_1862, memory_format = torch.contiguous_format)
	        var_mean_4 = torch.ops.aten.var_mean.correction(clone_17, [2], correction = 0, keepdim = True)
	        getitem_16: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_4[0]
	        getitem_17: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_4[1];  var_mean_4 = None
	        add_1867: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_16, 1e-05);  getitem_16 = None
	        rsqrt_4: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_1867);  add_1867 = None
	        sub_555: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_17, getitem_17);  clone_17 = getitem_17 = None
	        mul_1180: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_555, rsqrt_4);  sub_555 = rsqrt_4 = None
	        mul_1181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1180, model_audio_tower_layers_2_self_attn_layer_norm_weight);  mul_1180 = model_audio_tower_layers_2_self_attn_layer_norm_weight = None
	        add_1868: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_1181, model_audio_tower_layers_2_self_attn_layer_norm_bias);  mul_1181 = model_audio_tower_layers_2_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_1868, [2])
	        amax_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_1868, [2])
	        full_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_12, full_24);  amin_12 = full_24 = None
	        full_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_12, full_25);  amax_12 = full_25 = None
	        sub_566: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_12, minimum_12);  maximum_12 = None
	        div_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_566, 255.0);  sub_566 = None
	        clamp_min_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_24, 1.1920928955078125e-07);  div_24 = None
	        div_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_12, clamp_min_36);  minimum_12 = None
	        round_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_25);  div_25 = None
	        sub_572: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_25);  round_25 = None
	        clamp_min_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_572, -128);  sub_572 = None
	        clamp_max_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_37, 127);  clamp_min_37 = None
	        _assert_tensor_metadata_110 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_36, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_110 = None
	        _assert_tensor_metadata_111 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_24, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_111 = None
	        convert_element_type_72: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_24, torch.int8);  clamp_max_24 = None
	        view_190: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_36, [sym_size_int, 1500, 1])
	        view_191: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_72, [sym_size_int, 1500, 1])
	        reciprocal_12: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_190);  view_190 = None
	        mul_1229: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_12, 1.0);  reciprocal_12 = None
	        mul_1232: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_1868, mul_1229);  mul_1229 = None
	        round_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1232);  mul_1232 = None
	        add_1955: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_26, view_191);  round_26 = view_191 = None
	        clamp_min_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1955, -128);  add_1955 = None
	        clamp_max_25: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_38, 127);  clamp_min_38 = None
	        _assert_tensor_metadata_112 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_25, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_112 = None
	        convert_element_type_73: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_25, torch.int8);  clamp_max_25 = None
	        view_194: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_36, [sym_size_int, 1500, 1]);  clamp_min_36 = None
	        view_195: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_72, [sym_size_int, 1500, 1]);  convert_element_type_72 = None
	        _assert_tensor_metadata_113 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_73, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_113 = None
	        convert_element_type_74: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_73, torch.float32);  convert_element_type_73 = None
	        _assert_tensor_metadata_114 = torch.ops.aten._assert_tensor_metadata.default(view_195, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_114 = None
	        convert_element_type_75: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_195, torch.float32);  view_195 = None
	        sub_592: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_74, convert_element_type_75);  convert_element_type_74 = convert_element_type_75 = None
	        mul_1254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_592, view_194);  sub_592 = view_194 = None
	        _assert_tensor_metadata_115 = torch.ops.aten._assert_tensor_metadata.default(mul_1254, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_115 = None
	        view_197: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_198: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_199: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_116 = torch.ops.aten._assert_tensor_metadata.default(view_197, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_116 = None
	        convert_element_type_76: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_197, torch.float32);  view_197 = None
	        _assert_tensor_metadata_117 = torch.ops.aten._assert_tensor_metadata.default(view_199, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_117 = None
	        convert_element_type_77: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_199, torch.float32);  view_199 = None
	        sub_596: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_76, convert_element_type_77);  convert_element_type_76 = convert_element_type_77 = None
	        mul_1259: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_596, view_198);  sub_596 = view_198 = None
	        view_200: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1259, [1280, 1280]);  mul_1259 = None
	        _assert_tensor_metadata_118 = torch.ops.aten._assert_tensor_metadata.default(view_200, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_118 = None
	        mul_1264: "Sym(1500*s6)" = sym_size_int * 1500
	        view_201: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1254, [mul_1264, 1280]);  mul_1254 = mul_1264 = None
	        permute_21: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_200, [1, 0]);  view_200 = None
	        addmm_10: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_2_self_attn_q_proj_bias, view_201, permute_21);  model_audio_tower_layers_2_self_attn_q_proj_bias = view_201 = permute_21 = None
	        view_202: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_10, [sym_size_int, 1500, 1280]);  addmm_10 = None
	        mul_1271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_202, 0.125);  view_202 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_203: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_1271, [sym_size_int, 1500, 20, 64]);  mul_1271 = None
	        permute_22: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_203, [0, 2, 1, 3]);  view_203 = None
	        clone_18: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_22, memory_format = torch.contiguous_format);  permute_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_1868, [2])
	        amax_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_1868, [2])
	        full_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_13, full_26);  amin_13 = full_26 = None
	        full_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_13, full_27);  amax_13 = full_27 = None
	        sub_611: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_13, minimum_13);  maximum_13 = None
	        div_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_611, 255.0);  sub_611 = None
	        clamp_min_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_26, 1.1920928955078125e-07);  div_26 = None
	        div_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_13, clamp_min_39);  minimum_13 = None
	        round_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_27);  div_27 = None
	        sub_617: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_27);  round_27 = None
	        clamp_min_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_617, -128);  sub_617 = None
	        clamp_max_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_40, 127);  clamp_min_40 = None
	        _assert_tensor_metadata_119 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_39, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_119 = None
	        _assert_tensor_metadata_120 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_26, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_120 = None
	        convert_element_type_78: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_26, torch.int8);  clamp_max_26 = None
	        view_206: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_39, [sym_size_int, 1500, 1])
	        view_207: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_78, [sym_size_int, 1500, 1])
	        reciprocal_13: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_206);  view_206 = None
	        mul_1325: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_13, 1.0);  reciprocal_13 = None
	        mul_1328: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_1868, mul_1325);  mul_1325 = None
	        round_28: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1328);  mul_1328 = None
	        add_2107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_28, view_207);  round_28 = view_207 = None
	        clamp_min_41: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2107, -128);  add_2107 = None
	        clamp_max_27: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_41, 127);  clamp_min_41 = None
	        _assert_tensor_metadata_121 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_27, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_121 = None
	        convert_element_type_79: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_27, torch.int8);  clamp_max_27 = None
	        view_210: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_39, [sym_size_int, 1500, 1]);  clamp_min_39 = None
	        view_211: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_78, [sym_size_int, 1500, 1]);  convert_element_type_78 = None
	        _assert_tensor_metadata_122 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_79, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_122 = None
	        convert_element_type_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_79, torch.float32);  convert_element_type_79 = None
	        _assert_tensor_metadata_123 = torch.ops.aten._assert_tensor_metadata.default(view_211, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_123 = None
	        convert_element_type_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_211, torch.float32);  view_211 = None
	        sub_637: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_80, convert_element_type_81);  convert_element_type_80 = convert_element_type_81 = None
	        mul_1350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_637, view_210);  sub_637 = view_210 = None
	        _assert_tensor_metadata_124 = torch.ops.aten._assert_tensor_metadata.default(mul_1350, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_124 = None
	        view_213: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_214: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_215: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_125 = torch.ops.aten._assert_tensor_metadata.default(view_213, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_125 = None
	        convert_element_type_82: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_213, torch.float32);  view_213 = None
	        _assert_tensor_metadata_126 = torch.ops.aten._assert_tensor_metadata.default(view_215, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_126 = None
	        convert_element_type_83: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_215, torch.float32);  view_215 = None
	        sub_641: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_82, convert_element_type_83);  convert_element_type_82 = convert_element_type_83 = None
	        mul_1355: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_641, view_214);  sub_641 = view_214 = None
	        view_216: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1355, [1280, 1280]);  mul_1355 = None
	        _assert_tensor_metadata_127 = torch.ops.aten._assert_tensor_metadata.default(view_216, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_127 = None
	        permute_23: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_216, [1, 0]);  view_216 = None
	        mul_1358: "Sym(1500*s6)" = sym_size_int * 1500
	        view_217: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1350, [mul_1358, 1280]);  mul_1350 = mul_1358 = None
	        mm_2: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_217, permute_23);  view_217 = permute_23 = None
	        view_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_2, [sym_size_int, 1500, 1280]);  mm_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_219: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_218, [sym_size_int, -1, 20, 64]);  view_218 = None
	        permute_24: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_219, [0, 2, 1, 3]);  view_219 = None
	        clone_19: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_24, memory_format = torch.contiguous_format);  permute_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_1868, [2])
	        amax_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_1868, [2])
	        full_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_14, full_28);  amin_14 = full_28 = None
	        full_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_14, full_29);  amax_14 = full_29 = None
	        sub_655: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_14, minimum_14);  maximum_14 = None
	        div_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_655, 255.0);  sub_655 = None
	        clamp_min_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_28, 1.1920928955078125e-07);  div_28 = None
	        div_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_14, clamp_min_42);  minimum_14 = None
	        round_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_29);  div_29 = None
	        sub_661: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_29);  round_29 = None
	        clamp_min_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_661, -128);  sub_661 = None
	        clamp_max_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_43, 127);  clamp_min_43 = None
	        _assert_tensor_metadata_128 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_42, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_128 = None
	        _assert_tensor_metadata_129 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_28, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_129 = None
	        convert_element_type_84: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_28, torch.int8);  clamp_max_28 = None
	        view_222: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_42, [sym_size_int, 1500, 1])
	        view_223: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_84, [sym_size_int, 1500, 1])
	        reciprocal_14: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_222);  view_222 = None
	        mul_1424: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_14, 1.0);  reciprocal_14 = None
	        mul_1427: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_1868, mul_1424);  add_1868 = mul_1424 = None
	        round_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1427);  mul_1427 = None
	        add_2255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_30, view_223);  round_30 = view_223 = None
	        clamp_min_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2255, -128);  add_2255 = None
	        clamp_max_29: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_44, 127);  clamp_min_44 = None
	        _assert_tensor_metadata_130 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_29, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_130 = None
	        convert_element_type_85: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_29, torch.int8);  clamp_max_29 = None
	        view_226: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_42, [sym_size_int, 1500, 1]);  clamp_min_42 = None
	        view_227: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_84, [sym_size_int, 1500, 1]);  convert_element_type_84 = None
	        _assert_tensor_metadata_131 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_85, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_131 = None
	        convert_element_type_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_85, torch.float32);  convert_element_type_85 = None
	        _assert_tensor_metadata_132 = torch.ops.aten._assert_tensor_metadata.default(view_227, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_132 = None
	        convert_element_type_87: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_227, torch.float32);  view_227 = None
	        sub_681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_86, convert_element_type_87);  convert_element_type_86 = convert_element_type_87 = None
	        mul_1449: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_681, view_226);  sub_681 = view_226 = None
	        _assert_tensor_metadata_133 = torch.ops.aten._assert_tensor_metadata.default(mul_1449, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_133 = None
	        view_229: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_230: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_231: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_134 = torch.ops.aten._assert_tensor_metadata.default(view_229, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_134 = None
	        convert_element_type_88: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_229, torch.float32);  view_229 = None
	        _assert_tensor_metadata_135 = torch.ops.aten._assert_tensor_metadata.default(view_231, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_135 = None
	        convert_element_type_89: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_231, torch.float32);  view_231 = None
	        sub_685: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_88, convert_element_type_89);  convert_element_type_88 = convert_element_type_89 = None
	        mul_1454: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_685, view_230);  sub_685 = view_230 = None
	        view_232: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1454, [1280, 1280]);  mul_1454 = None
	        _assert_tensor_metadata_136 = torch.ops.aten._assert_tensor_metadata.default(view_232, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_136 = None
	        mul_1459: "Sym(1500*s6)" = sym_size_int * 1500
	        view_233: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1449, [mul_1459, 1280]);  mul_1449 = mul_1459 = None
	        permute_25: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_232, [1, 0]);  view_232 = None
	        addmm_11: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_2_self_attn_v_proj_bias, view_233, permute_25);  model_audio_tower_layers_2_self_attn_v_proj_bias = view_233 = permute_25 = None
	        view_234: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_11, [sym_size_int, 1500, 1280]);  addmm_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_235: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_234, [sym_size_int, -1, 20, 64]);  view_234 = None
	        permute_26: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_235, [0, 2, 1, 3]);  view_235 = None
	        clone_20: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_26, memory_format = torch.contiguous_format);  permute_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_2 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_18, clone_19, clone_20, None, False, scale = 1.0);  clone_18 = clone_19 = clone_20 = None
	        getitem_18: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_2[0];  _scaled_dot_product_efficient_attention_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_27: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_18, [0, 2, 1, 3]);  getitem_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_27, [sym_size_int, 1500, -1]);  permute_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_236, [2])
	        amax_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_236, [2])
	        full_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_15, full_30);  amin_15 = full_30 = None
	        full_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_15, full_31);  amax_15 = full_31 = None
	        sub_703: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_15, minimum_15);  maximum_15 = None
	        div_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_703, 255.0);  sub_703 = None
	        clamp_min_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_30, 1.1920928955078125e-07);  div_30 = None
	        div_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_15, clamp_min_45);  minimum_15 = None
	        round_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_31);  div_31 = None
	        sub_709: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_31);  round_31 = None
	        clamp_min_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_709, -128);  sub_709 = None
	        clamp_max_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_46, 127);  clamp_min_46 = None
	        _assert_tensor_metadata_137 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_45, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_137 = None
	        _assert_tensor_metadata_138 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_30, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_138 = None
	        convert_element_type_90: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_30, torch.int8);  clamp_max_30 = None
	        view_239: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_45, [sym_size_int, 1500, 1])
	        view_240: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_90, [sym_size_int, 1500, 1])
	        reciprocal_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_239);  view_239 = None
	        mul_1529: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_15, 1.0);  reciprocal_15 = None
	        mul_1532: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_236, mul_1529);  view_236 = mul_1529 = None
	        round_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1532);  mul_1532 = None
	        add_2419: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_32, view_240);  round_32 = view_240 = None
	        clamp_min_47: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2419, -128);  add_2419 = None
	        clamp_max_31: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_47, 127);  clamp_min_47 = None
	        _assert_tensor_metadata_139 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_31, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_139 = None
	        convert_element_type_91: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_31, torch.int8);  clamp_max_31 = None
	        view_243: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_45, [sym_size_int, 1500, 1]);  clamp_min_45 = None
	        view_244: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_90, [sym_size_int, 1500, 1]);  convert_element_type_90 = None
	        _assert_tensor_metadata_140 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_91, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_140 = None
	        convert_element_type_92: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_91, torch.float32);  convert_element_type_91 = None
	        _assert_tensor_metadata_141 = torch.ops.aten._assert_tensor_metadata.default(view_244, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_141 = None
	        convert_element_type_93: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_244, torch.float32);  view_244 = None
	        sub_729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_92, convert_element_type_93);  convert_element_type_92 = convert_element_type_93 = None
	        mul_1554: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_729, view_243);  sub_729 = view_243 = None
	        _assert_tensor_metadata_142 = torch.ops.aten._assert_tensor_metadata.default(mul_1554, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_142 = None
	        view_246: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_247: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_248: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_143 = torch.ops.aten._assert_tensor_metadata.default(view_246, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_143 = None
	        convert_element_type_94: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_246, torch.float32);  view_246 = None
	        _assert_tensor_metadata_144 = torch.ops.aten._assert_tensor_metadata.default(view_248, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_144 = None
	        convert_element_type_95: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_248, torch.float32);  view_248 = None
	        sub_733: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_94, convert_element_type_95);  convert_element_type_94 = convert_element_type_95 = None
	        mul_1559: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_733, view_247);  sub_733 = view_247 = None
	        view_249: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1559, [1280, 1280]);  mul_1559 = None
	        _assert_tensor_metadata_145 = torch.ops.aten._assert_tensor_metadata.default(view_249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_145 = None
	        mul_1564: "Sym(1500*s6)" = sym_size_int * 1500
	        view_250: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1554, [mul_1564, 1280]);  mul_1554 = mul_1564 = None
	        permute_28: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_249, [1, 0]);  view_249 = None
	        addmm_12: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_2_self_attn_out_proj_bias, view_250, permute_28);  model_audio_tower_layers_2_self_attn_out_proj_bias = view_250 = permute_28 = None
	        view_251: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_12, [sym_size_int, 1500, 1280]);  addmm_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_2482: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_1862, view_251);  add_1862 = view_251 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_22: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_2482, memory_format = torch.contiguous_format)
	        var_mean_5 = torch.ops.aten.var_mean.correction(clone_22, [2], correction = 0, keepdim = True)
	        getitem_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_5[0]
	        getitem_23: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_5[1];  var_mean_5 = None
	        add_2487: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_22, 1e-05);  getitem_22 = None
	        rsqrt_5: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_2487);  add_2487 = None
	        sub_739: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_22, getitem_23);  clone_22 = getitem_23 = None
	        mul_1575: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_739, rsqrt_5);  sub_739 = rsqrt_5 = None
	        mul_1576: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1575, model_audio_tower_layers_2_final_layer_norm_weight);  mul_1575 = model_audio_tower_layers_2_final_layer_norm_weight = None
	        add_2488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_1576, model_audio_tower_layers_2_final_layer_norm_bias);  mul_1576 = model_audio_tower_layers_2_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_2488, [2])
	        amax_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_2488, [2])
	        full_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_16, full_32);  amin_16 = full_32 = None
	        full_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_16, full_33);  amax_16 = full_33 = None
	        sub_750: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_16, minimum_16);  maximum_16 = None
	        div_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_750, 255.0);  sub_750 = None
	        clamp_min_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_32, 1.1920928955078125e-07);  div_32 = None
	        div_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_16, clamp_min_48);  minimum_16 = None
	        round_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_33);  div_33 = None
	        sub_756: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_33);  round_33 = None
	        clamp_min_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_756, -128);  sub_756 = None
	        clamp_max_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_49, 127);  clamp_min_49 = None
	        _assert_tensor_metadata_146 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_48, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_146 = None
	        _assert_tensor_metadata_147 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_32, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_147 = None
	        convert_element_type_96: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_32, torch.int8);  clamp_max_32 = None
	        view_254: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_48, [sym_size_int, 1500, 1])
	        view_255: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_96, [sym_size_int, 1500, 1])
	        reciprocal_16: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_254);  view_254 = None
	        mul_1624: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_16, 1.0);  reciprocal_16 = None
	        mul_1627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_2488, mul_1624);  add_2488 = mul_1624 = None
	        round_34: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1627);  mul_1627 = None
	        add_2575: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_34, view_255);  round_34 = view_255 = None
	        clamp_min_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2575, -128);  add_2575 = None
	        clamp_max_33: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_50, 127);  clamp_min_50 = None
	        _assert_tensor_metadata_148 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_33, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_148 = None
	        convert_element_type_97: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_33, torch.int8);  clamp_max_33 = None
	        view_258: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_48, [sym_size_int, 1500, 1]);  clamp_min_48 = None
	        view_259: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_96, [sym_size_int, 1500, 1]);  convert_element_type_96 = None
	        _assert_tensor_metadata_149 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_97, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_149 = None
	        convert_element_type_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_97, torch.float32);  convert_element_type_97 = None
	        _assert_tensor_metadata_150 = torch.ops.aten._assert_tensor_metadata.default(view_259, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_150 = None
	        convert_element_type_99: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_259, torch.float32);  view_259 = None
	        sub_776: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_98, convert_element_type_99);  convert_element_type_98 = convert_element_type_99 = None
	        mul_1649: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_776, view_258);  sub_776 = view_258 = None
	        _assert_tensor_metadata_151 = torch.ops.aten._assert_tensor_metadata.default(mul_1649, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_151 = None
	        view_261: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_2_fc1_parametrizations_weight_original0 = None
	        view_262: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_2_fc1_parametrizations_weight_original1 = None
	        view_263: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_2_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_152 = torch.ops.aten._assert_tensor_metadata.default(view_261, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_152 = None
	        convert_element_type_100: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_261, torch.float32);  view_261 = None
	        _assert_tensor_metadata_153 = torch.ops.aten._assert_tensor_metadata.default(view_263, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_153 = None
	        convert_element_type_101: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_263, torch.float32);  view_263 = None
	        sub_780: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_100, convert_element_type_101);  convert_element_type_100 = convert_element_type_101 = None
	        mul_1654: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_780, view_262);  sub_780 = view_262 = None
	        view_264: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1654, [5120, 1280]);  mul_1654 = None
	        _assert_tensor_metadata_154 = torch.ops.aten._assert_tensor_metadata.default(view_264, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_154 = None
	        mul_1659: "Sym(1500*s6)" = sym_size_int * 1500
	        view_265: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1649, [mul_1659, 1280]);  mul_1649 = mul_1659 = None
	        permute_29: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_264, [1, 0]);  view_264 = None
	        addmm_13: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_2_fc1_bias, view_265, permute_29);  model_audio_tower_layers_2_fc1_bias = view_265 = permute_29 = None
	        view_266: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_13, [sym_size_int, 1500, 5120]);  addmm_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_1666: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_266, 0.5)
	        mul_1667: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_266, 0.7071067811865476);  view_266 = None
	        erf_4: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_1667);  mul_1667 = None
	        add_2634: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_4, 1);  erf_4 = None
	        mul_1668: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1666, add_2634);  mul_1666 = add_2634 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_1668, [2])
	        amax_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_1668, [2])
	        full_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_17, full_34);  amin_17 = full_34 = None
	        full_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_17, full_35);  amax_17 = full_35 = None
	        sub_793: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_17, minimum_17);  maximum_17 = None
	        div_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_793, 255.0);  sub_793 = None
	        clamp_min_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_34, 1.1920928955078125e-07);  div_34 = None
	        div_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_17, clamp_min_51);  minimum_17 = None
	        round_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_35);  div_35 = None
	        sub_799: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_35);  round_35 = None
	        clamp_min_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_799, -128);  sub_799 = None
	        clamp_max_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_52, 127);  clamp_min_52 = None
	        _assert_tensor_metadata_155 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_51, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_155 = None
	        _assert_tensor_metadata_156 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_34, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_156 = None
	        convert_element_type_102: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_34, torch.int8);  clamp_max_34 = None
	        view_269: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_51, [sym_size_int, 1500, 1])
	        view_270: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_102, [sym_size_int, 1500, 1])
	        reciprocal_17: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_269);  view_269 = None
	        mul_1714: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_17, 1.0);  reciprocal_17 = None
	        mul_1717: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1668, mul_1714);  mul_1668 = mul_1714 = None
	        round_36: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_1717);  mul_1717 = None
	        add_2717: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_36, view_270);  round_36 = view_270 = None
	        clamp_min_53: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2717, -128);  add_2717 = None
	        clamp_max_35: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_53, 127);  clamp_min_53 = None
	        _assert_tensor_metadata_157 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_35, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_157 = None
	        convert_element_type_103: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_35, torch.int8);  clamp_max_35 = None
	        view_273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_51, [sym_size_int, 1500, 1]);  clamp_min_51 = None
	        view_274: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_102, [sym_size_int, 1500, 1]);  convert_element_type_102 = None
	        _assert_tensor_metadata_158 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_103, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_158 = None
	        convert_element_type_104: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_103, torch.float32);  convert_element_type_103 = None
	        _assert_tensor_metadata_159 = torch.ops.aten._assert_tensor_metadata.default(view_274, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_159 = None
	        convert_element_type_105: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_274, torch.float32);  view_274 = None
	        sub_819: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_104, convert_element_type_105);  convert_element_type_104 = convert_element_type_105 = None
	        mul_1739: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_819, view_273);  sub_819 = view_273 = None
	        _assert_tensor_metadata_160 = torch.ops.aten._assert_tensor_metadata.default(mul_1739, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_160 = None
	        view_276: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_2_fc2_parametrizations_weight_original0 = None
	        view_277: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_2_fc2_parametrizations_weight_original1 = None
	        view_278: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_2_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_2_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_161 = torch.ops.aten._assert_tensor_metadata.default(view_276, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_161 = None
	        convert_element_type_106: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_276, torch.float32);  view_276 = None
	        _assert_tensor_metadata_162 = torch.ops.aten._assert_tensor_metadata.default(view_278, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_162 = None
	        convert_element_type_107: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_278, torch.float32);  view_278 = None
	        sub_823: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_106, convert_element_type_107);  convert_element_type_106 = convert_element_type_107 = None
	        mul_1744: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_823, view_277);  sub_823 = view_277 = None
	        view_279: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_1744, [1280, 5120]);  mul_1744 = None
	        _assert_tensor_metadata_163 = torch.ops.aten._assert_tensor_metadata.default(view_279, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_163 = None
	        mul_1749: "Sym(1500*s6)" = sym_size_int * 1500
	        view_280: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_1739, [mul_1749, 5120]);  mul_1739 = mul_1749 = None
	        permute_30: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_279, [1, 0]);  view_279 = None
	        addmm_14: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_2_fc2_bias, view_280, permute_30);  model_audio_tower_layers_2_fc2_bias = view_280 = permute_30 = None
	        view_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_14, [sym_size_int, 1500, 1280]);  addmm_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_2780: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_2482, view_281);  add_2482 = view_281 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_25: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_2780, memory_format = torch.contiguous_format)
	        var_mean_6 = torch.ops.aten.var_mean.correction(clone_25, [2], correction = 0, keepdim = True)
	        getitem_24: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_6[0]
	        getitem_25: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_6[1];  var_mean_6 = None
	        add_2785: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_24, 1e-05);  getitem_24 = None
	        rsqrt_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_2785);  add_2785 = None
	        sub_829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_25, getitem_25);  clone_25 = getitem_25 = None
	        mul_1760: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_829, rsqrt_6);  sub_829 = rsqrt_6 = None
	        mul_1761: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1760, model_audio_tower_layers_3_self_attn_layer_norm_weight);  mul_1760 = model_audio_tower_layers_3_self_attn_layer_norm_weight = None
	        add_2786: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_1761, model_audio_tower_layers_3_self_attn_layer_norm_bias);  mul_1761 = model_audio_tower_layers_3_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_2786, [2])
	        amax_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_2786, [2])
	        full_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_18, full_36);  amin_18 = full_36 = None
	        full_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_18, full_37);  amax_18 = full_37 = None
	        sub_840: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_18, minimum_18);  maximum_18 = None
	        div_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_840, 255.0);  sub_840 = None
	        clamp_min_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_36, 1.1920928955078125e-07);  div_36 = None
	        div_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_18, clamp_min_54);  minimum_18 = None
	        round_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_37);  div_37 = None
	        sub_846: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_37);  round_37 = None
	        clamp_min_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_846, -128);  sub_846 = None
	        clamp_max_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_55, 127);  clamp_min_55 = None
	        _assert_tensor_metadata_164 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_54, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_164 = None
	        _assert_tensor_metadata_165 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_36, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_165 = None
	        convert_element_type_108: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_36, torch.int8);  clamp_max_36 = None
	        view_284: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_54, [sym_size_int, 1500, 1])
	        view_285: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_108, [sym_size_int, 1500, 1])
	        reciprocal_18: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_284);  view_284 = None
	        mul_1809: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_18, 1.0);  reciprocal_18 = None
	        mul_1812: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_2786, mul_1809);  mul_1809 = None
	        round_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1812);  mul_1812 = None
	        add_2873: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_38, view_285);  round_38 = view_285 = None
	        clamp_min_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2873, -128);  add_2873 = None
	        clamp_max_37: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_56, 127);  clamp_min_56 = None
	        _assert_tensor_metadata_166 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_37, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_166 = None
	        convert_element_type_109: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_37, torch.int8);  clamp_max_37 = None
	        view_288: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_54, [sym_size_int, 1500, 1]);  clamp_min_54 = None
	        view_289: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_108, [sym_size_int, 1500, 1]);  convert_element_type_108 = None
	        _assert_tensor_metadata_167 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_109, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_167 = None
	        convert_element_type_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_109, torch.float32);  convert_element_type_109 = None
	        _assert_tensor_metadata_168 = torch.ops.aten._assert_tensor_metadata.default(view_289, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_168 = None
	        convert_element_type_111: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_289, torch.float32);  view_289 = None
	        sub_866: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_110, convert_element_type_111);  convert_element_type_110 = convert_element_type_111 = None
	        mul_1834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_866, view_288);  sub_866 = view_288 = None
	        _assert_tensor_metadata_169 = torch.ops.aten._assert_tensor_metadata.default(mul_1834, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_169 = None
	        view_291: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_292: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_293: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_170 = torch.ops.aten._assert_tensor_metadata.default(view_291, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_170 = None
	        convert_element_type_112: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_291, torch.float32);  view_291 = None
	        _assert_tensor_metadata_171 = torch.ops.aten._assert_tensor_metadata.default(view_293, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_171 = None
	        convert_element_type_113: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_293, torch.float32);  view_293 = None
	        sub_870: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_112, convert_element_type_113);  convert_element_type_112 = convert_element_type_113 = None
	        mul_1839: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_870, view_292);  sub_870 = view_292 = None
	        view_294: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1839, [1280, 1280]);  mul_1839 = None
	        _assert_tensor_metadata_172 = torch.ops.aten._assert_tensor_metadata.default(view_294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_172 = None
	        mul_1844: "Sym(1500*s6)" = sym_size_int * 1500
	        view_295: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1834, [mul_1844, 1280]);  mul_1834 = mul_1844 = None
	        permute_31: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_294, [1, 0]);  view_294 = None
	        addmm_15: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_3_self_attn_q_proj_bias, view_295, permute_31);  model_audio_tower_layers_3_self_attn_q_proj_bias = view_295 = permute_31 = None
	        view_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_15, [sym_size_int, 1500, 1280]);  addmm_15 = None
	        mul_1851: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_296, 0.125);  view_296 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_297: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_1851, [sym_size_int, 1500, 20, 64]);  mul_1851 = None
	        permute_32: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_297, [0, 2, 1, 3]);  view_297 = None
	        clone_26: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_32, memory_format = torch.contiguous_format);  permute_32 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_2786, [2])
	        amax_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_2786, [2])
	        full_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_19, full_38);  amin_19 = full_38 = None
	        full_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_19, full_39);  amax_19 = full_39 = None
	        sub_885: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_19, minimum_19);  maximum_19 = None
	        div_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_885, 255.0);  sub_885 = None
	        clamp_min_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_38, 1.1920928955078125e-07);  div_38 = None
	        div_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_19, clamp_min_57);  minimum_19 = None
	        round_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_39);  div_39 = None
	        sub_891: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_39);  round_39 = None
	        clamp_min_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_891, -128);  sub_891 = None
	        clamp_max_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_58, 127);  clamp_min_58 = None
	        _assert_tensor_metadata_173 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_57, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_173 = None
	        _assert_tensor_metadata_174 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_38, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_174 = None
	        convert_element_type_114: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_38, torch.int8);  clamp_max_38 = None
	        view_300: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_57, [sym_size_int, 1500, 1])
	        view_301: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_114, [sym_size_int, 1500, 1])
	        reciprocal_19: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_300);  view_300 = None
	        mul_1905: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_19, 1.0);  reciprocal_19 = None
	        mul_1908: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_2786, mul_1905);  mul_1905 = None
	        round_40: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1908);  mul_1908 = None
	        add_3025: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_40, view_301);  round_40 = view_301 = None
	        clamp_min_59: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3025, -128);  add_3025 = None
	        clamp_max_39: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_59, 127);  clamp_min_59 = None
	        _assert_tensor_metadata_175 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_39, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_175 = None
	        convert_element_type_115: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_39, torch.int8);  clamp_max_39 = None
	        view_304: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_57, [sym_size_int, 1500, 1]);  clamp_min_57 = None
	        view_305: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_114, [sym_size_int, 1500, 1]);  convert_element_type_114 = None
	        _assert_tensor_metadata_176 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_115, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_176 = None
	        convert_element_type_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_115, torch.float32);  convert_element_type_115 = None
	        _assert_tensor_metadata_177 = torch.ops.aten._assert_tensor_metadata.default(view_305, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_177 = None
	        convert_element_type_117: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_305, torch.float32);  view_305 = None
	        sub_911: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_116, convert_element_type_117);  convert_element_type_116 = convert_element_type_117 = None
	        mul_1930: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_911, view_304);  sub_911 = view_304 = None
	        _assert_tensor_metadata_178 = torch.ops.aten._assert_tensor_metadata.default(mul_1930, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_178 = None
	        view_307: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_308: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_309: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_179 = torch.ops.aten._assert_tensor_metadata.default(view_307, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_179 = None
	        convert_element_type_118: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_307, torch.float32);  view_307 = None
	        _assert_tensor_metadata_180 = torch.ops.aten._assert_tensor_metadata.default(view_309, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_180 = None
	        convert_element_type_119: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_309, torch.float32);  view_309 = None
	        sub_915: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_118, convert_element_type_119);  convert_element_type_118 = convert_element_type_119 = None
	        mul_1935: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_915, view_308);  sub_915 = view_308 = None
	        view_310: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1935, [1280, 1280]);  mul_1935 = None
	        _assert_tensor_metadata_181 = torch.ops.aten._assert_tensor_metadata.default(view_310, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_181 = None
	        permute_33: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_310, [1, 0]);  view_310 = None
	        mul_1938: "Sym(1500*s6)" = sym_size_int * 1500
	        view_311: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_1930, [mul_1938, 1280]);  mul_1930 = mul_1938 = None
	        mm_3: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_311, permute_33);  view_311 = permute_33 = None
	        view_312: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_3, [sym_size_int, 1500, 1280]);  mm_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_313: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_312, [sym_size_int, -1, 20, 64]);  view_312 = None
	        permute_34: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_313, [0, 2, 1, 3]);  view_313 = None
	        clone_27: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_34, memory_format = torch.contiguous_format);  permute_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_2786, [2])
	        amax_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_2786, [2])
	        full_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_20, full_40);  amin_20 = full_40 = None
	        full_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_20, full_41);  amax_20 = full_41 = None
	        sub_929: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_20, minimum_20);  maximum_20 = None
	        div_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_929, 255.0);  sub_929 = None
	        clamp_min_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_40, 1.1920928955078125e-07);  div_40 = None
	        div_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_20, clamp_min_60);  minimum_20 = None
	        round_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_41);  div_41 = None
	        sub_935: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_41);  round_41 = None
	        clamp_min_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_935, -128);  sub_935 = None
	        clamp_max_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_61, 127);  clamp_min_61 = None
	        _assert_tensor_metadata_182 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_60, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_182 = None
	        _assert_tensor_metadata_183 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_40, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_183 = None
	        convert_element_type_120: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_40, torch.int8);  clamp_max_40 = None
	        view_316: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_60, [sym_size_int, 1500, 1])
	        view_317: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_120, [sym_size_int, 1500, 1])
	        reciprocal_20: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_316);  view_316 = None
	        mul_2004: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_20, 1.0);  reciprocal_20 = None
	        mul_2007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_2786, mul_2004);  add_2786 = mul_2004 = None
	        round_42: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2007);  mul_2007 = None
	        add_3173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_42, view_317);  round_42 = view_317 = None
	        clamp_min_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3173, -128);  add_3173 = None
	        clamp_max_41: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_62, 127);  clamp_min_62 = None
	        _assert_tensor_metadata_184 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_41, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_184 = None
	        convert_element_type_121: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_41, torch.int8);  clamp_max_41 = None
	        view_320: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_60, [sym_size_int, 1500, 1]);  clamp_min_60 = None
	        view_321: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_120, [sym_size_int, 1500, 1]);  convert_element_type_120 = None
	        _assert_tensor_metadata_185 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_121, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_185 = None
	        convert_element_type_122: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_121, torch.float32);  convert_element_type_121 = None
	        _assert_tensor_metadata_186 = torch.ops.aten._assert_tensor_metadata.default(view_321, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_186 = None
	        convert_element_type_123: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_321, torch.float32);  view_321 = None
	        sub_955: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_122, convert_element_type_123);  convert_element_type_122 = convert_element_type_123 = None
	        mul_2029: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_955, view_320);  sub_955 = view_320 = None
	        _assert_tensor_metadata_187 = torch.ops.aten._assert_tensor_metadata.default(mul_2029, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_187 = None
	        view_323: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_324: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_325: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_188 = torch.ops.aten._assert_tensor_metadata.default(view_323, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_188 = None
	        convert_element_type_124: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_323, torch.float32);  view_323 = None
	        _assert_tensor_metadata_189 = torch.ops.aten._assert_tensor_metadata.default(view_325, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_189 = None
	        convert_element_type_125: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_325, torch.float32);  view_325 = None
	        sub_959: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_124, convert_element_type_125);  convert_element_type_124 = convert_element_type_125 = None
	        mul_2034: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_959, view_324);  sub_959 = view_324 = None
	        view_326: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2034, [1280, 1280]);  mul_2034 = None
	        _assert_tensor_metadata_190 = torch.ops.aten._assert_tensor_metadata.default(view_326, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_190 = None
	        mul_2039: "Sym(1500*s6)" = sym_size_int * 1500
	        view_327: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2029, [mul_2039, 1280]);  mul_2029 = mul_2039 = None
	        permute_35: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_326, [1, 0]);  view_326 = None
	        addmm_16: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_3_self_attn_v_proj_bias, view_327, permute_35);  model_audio_tower_layers_3_self_attn_v_proj_bias = view_327 = permute_35 = None
	        view_328: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_16, [sym_size_int, 1500, 1280]);  addmm_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_329: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_328, [sym_size_int, -1, 20, 64]);  view_328 = None
	        permute_36: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_329, [0, 2, 1, 3]);  view_329 = None
	        clone_28: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_36, memory_format = torch.contiguous_format);  permute_36 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_3 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_26, clone_27, clone_28, None, False, scale = 1.0);  clone_26 = clone_27 = clone_28 = None
	        getitem_26: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_3[0];  _scaled_dot_product_efficient_attention_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_37: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_26, [0, 2, 1, 3]);  getitem_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_330: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_37, [sym_size_int, 1500, -1]);  permute_37 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_330, [2])
	        amax_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_330, [2])
	        full_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_21, full_42);  amin_21 = full_42 = None
	        full_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_21, full_43);  amax_21 = full_43 = None
	        sub_977: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_21, minimum_21);  maximum_21 = None
	        div_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_977, 255.0);  sub_977 = None
	        clamp_min_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_42, 1.1920928955078125e-07);  div_42 = None
	        div_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_21, clamp_min_63);  minimum_21 = None
	        round_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_43);  div_43 = None
	        sub_983: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_43);  round_43 = None
	        clamp_min_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_983, -128);  sub_983 = None
	        clamp_max_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_64, 127);  clamp_min_64 = None
	        _assert_tensor_metadata_191 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_63, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_191 = None
	        _assert_tensor_metadata_192 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_42, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_192 = None
	        convert_element_type_126: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_42, torch.int8);  clamp_max_42 = None
	        view_333: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_63, [sym_size_int, 1500, 1])
	        view_334: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_126, [sym_size_int, 1500, 1])
	        reciprocal_21: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_333);  view_333 = None
	        mul_2109: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_21, 1.0);  reciprocal_21 = None
	        mul_2112: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_330, mul_2109);  view_330 = mul_2109 = None
	        round_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2112);  mul_2112 = None
	        add_3337: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_44, view_334);  round_44 = view_334 = None
	        clamp_min_65: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3337, -128);  add_3337 = None
	        clamp_max_43: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_65, 127);  clamp_min_65 = None
	        _assert_tensor_metadata_193 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_43, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_193 = None
	        convert_element_type_127: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_43, torch.int8);  clamp_max_43 = None
	        view_337: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_63, [sym_size_int, 1500, 1]);  clamp_min_63 = None
	        view_338: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_126, [sym_size_int, 1500, 1]);  convert_element_type_126 = None
	        _assert_tensor_metadata_194 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_127, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_194 = None
	        convert_element_type_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_127, torch.float32);  convert_element_type_127 = None
	        _assert_tensor_metadata_195 = torch.ops.aten._assert_tensor_metadata.default(view_338, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_195 = None
	        convert_element_type_129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_338, torch.float32);  view_338 = None
	        sub_1003: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_128, convert_element_type_129);  convert_element_type_128 = convert_element_type_129 = None
	        mul_2134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1003, view_337);  sub_1003 = view_337 = None
	        _assert_tensor_metadata_196 = torch.ops.aten._assert_tensor_metadata.default(mul_2134, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_196 = None
	        view_340: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_341: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_342: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_197 = torch.ops.aten._assert_tensor_metadata.default(view_340, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_197 = None
	        convert_element_type_130: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_340, torch.float32);  view_340 = None
	        _assert_tensor_metadata_198 = torch.ops.aten._assert_tensor_metadata.default(view_342, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_198 = None
	        convert_element_type_131: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_342, torch.float32);  view_342 = None
	        sub_1007: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_130, convert_element_type_131);  convert_element_type_130 = convert_element_type_131 = None
	        mul_2139: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1007, view_341);  sub_1007 = view_341 = None
	        view_343: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2139, [1280, 1280]);  mul_2139 = None
	        _assert_tensor_metadata_199 = torch.ops.aten._assert_tensor_metadata.default(view_343, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_199 = None
	        mul_2144: "Sym(1500*s6)" = sym_size_int * 1500
	        view_344: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2134, [mul_2144, 1280]);  mul_2134 = mul_2144 = None
	        permute_38: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_343, [1, 0]);  view_343 = None
	        addmm_17: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_3_self_attn_out_proj_bias, view_344, permute_38);  model_audio_tower_layers_3_self_attn_out_proj_bias = view_344 = permute_38 = None
	        view_345: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_17, [sym_size_int, 1500, 1280]);  addmm_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_3400: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_2780, view_345);  add_2780 = view_345 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_3400, memory_format = torch.contiguous_format)
	        var_mean_7 = torch.ops.aten.var_mean.correction(clone_30, [2], correction = 0, keepdim = True)
	        getitem_30: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_7[0]
	        getitem_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_7[1];  var_mean_7 = None
	        add_3405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_30, 1e-05);  getitem_30 = None
	        rsqrt_7: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_3405);  add_3405 = None
	        sub_1013: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_30, getitem_31);  clone_30 = getitem_31 = None
	        mul_2155: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1013, rsqrt_7);  sub_1013 = rsqrt_7 = None
	        mul_2156: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2155, model_audio_tower_layers_3_final_layer_norm_weight);  mul_2155 = model_audio_tower_layers_3_final_layer_norm_weight = None
	        add_3406: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2156, model_audio_tower_layers_3_final_layer_norm_bias);  mul_2156 = model_audio_tower_layers_3_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_3406, [2])
	        amax_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_3406, [2])
	        full_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_22, full_44);  amin_22 = full_44 = None
	        full_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_22, full_45);  amax_22 = full_45 = None
	        sub_1024: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_22, minimum_22);  maximum_22 = None
	        div_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1024, 255.0);  sub_1024 = None
	        clamp_min_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_44, 1.1920928955078125e-07);  div_44 = None
	        div_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_22, clamp_min_66);  minimum_22 = None
	        round_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_45);  div_45 = None
	        sub_1030: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_45);  round_45 = None
	        clamp_min_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1030, -128);  sub_1030 = None
	        clamp_max_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_67, 127);  clamp_min_67 = None
	        _assert_tensor_metadata_200 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_66, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_200 = None
	        _assert_tensor_metadata_201 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_44, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_201 = None
	        convert_element_type_132: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_44, torch.int8);  clamp_max_44 = None
	        view_348: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_66, [sym_size_int, 1500, 1])
	        view_349: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_132, [sym_size_int, 1500, 1])
	        reciprocal_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_348);  view_348 = None
	        mul_2204: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_22, 1.0);  reciprocal_22 = None
	        mul_2207: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_3406, mul_2204);  add_3406 = mul_2204 = None
	        round_46: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2207);  mul_2207 = None
	        add_3493: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_46, view_349);  round_46 = view_349 = None
	        clamp_min_68: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3493, -128);  add_3493 = None
	        clamp_max_45: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_68, 127);  clamp_min_68 = None
	        _assert_tensor_metadata_202 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_45, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_202 = None
	        convert_element_type_133: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_45, torch.int8);  clamp_max_45 = None
	        view_352: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_66, [sym_size_int, 1500, 1]);  clamp_min_66 = None
	        view_353: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_132, [sym_size_int, 1500, 1]);  convert_element_type_132 = None
	        _assert_tensor_metadata_203 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_133, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_203 = None
	        convert_element_type_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_133, torch.float32);  convert_element_type_133 = None
	        _assert_tensor_metadata_204 = torch.ops.aten._assert_tensor_metadata.default(view_353, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_204 = None
	        convert_element_type_135: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_353, torch.float32);  view_353 = None
	        sub_1050: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_134, convert_element_type_135);  convert_element_type_134 = convert_element_type_135 = None
	        mul_2229: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1050, view_352);  sub_1050 = view_352 = None
	        _assert_tensor_metadata_205 = torch.ops.aten._assert_tensor_metadata.default(mul_2229, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_205 = None
	        view_355: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_3_fc1_parametrizations_weight_original0 = None
	        view_356: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_3_fc1_parametrizations_weight_original1 = None
	        view_357: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_3_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_206 = torch.ops.aten._assert_tensor_metadata.default(view_355, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_206 = None
	        convert_element_type_136: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_355, torch.float32);  view_355 = None
	        _assert_tensor_metadata_207 = torch.ops.aten._assert_tensor_metadata.default(view_357, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_207 = None
	        convert_element_type_137: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_357, torch.float32);  view_357 = None
	        sub_1054: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_136, convert_element_type_137);  convert_element_type_136 = convert_element_type_137 = None
	        mul_2234: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1054, view_356);  sub_1054 = view_356 = None
	        view_358: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2234, [5120, 1280]);  mul_2234 = None
	        _assert_tensor_metadata_208 = torch.ops.aten._assert_tensor_metadata.default(view_358, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_208 = None
	        mul_2239: "Sym(1500*s6)" = sym_size_int * 1500
	        view_359: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2229, [mul_2239, 1280]);  mul_2229 = mul_2239 = None
	        permute_39: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_358, [1, 0]);  view_358 = None
	        addmm_18: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_3_fc1_bias, view_359, permute_39);  model_audio_tower_layers_3_fc1_bias = view_359 = permute_39 = None
	        view_360: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_18, [sym_size_int, 1500, 5120]);  addmm_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_2246: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_360, 0.5)
	        mul_2247: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_360, 0.7071067811865476);  view_360 = None
	        erf_5: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_2247);  mul_2247 = None
	        add_3552: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_5, 1);  erf_5 = None
	        mul_2248: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2246, add_3552);  mul_2246 = add_3552 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_2248, [2])
	        amax_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_2248, [2])
	        full_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_23, full_46);  amin_23 = full_46 = None
	        full_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_23, full_47);  amax_23 = full_47 = None
	        sub_1067: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_23, minimum_23);  maximum_23 = None
	        div_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1067, 255.0);  sub_1067 = None
	        clamp_min_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_46, 1.1920928955078125e-07);  div_46 = None
	        div_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_23, clamp_min_69);  minimum_23 = None
	        round_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_47);  div_47 = None
	        sub_1073: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_47);  round_47 = None
	        clamp_min_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1073, -128);  sub_1073 = None
	        clamp_max_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_70, 127);  clamp_min_70 = None
	        _assert_tensor_metadata_209 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_69, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_209 = None
	        _assert_tensor_metadata_210 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_46, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_210 = None
	        convert_element_type_138: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_46, torch.int8);  clamp_max_46 = None
	        view_363: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_69, [sym_size_int, 1500, 1])
	        view_364: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_138, [sym_size_int, 1500, 1])
	        reciprocal_23: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_363);  view_363 = None
	        mul_2294: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_23, 1.0);  reciprocal_23 = None
	        mul_2297: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2248, mul_2294);  mul_2248 = mul_2294 = None
	        round_48: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_2297);  mul_2297 = None
	        add_3635: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_48, view_364);  round_48 = view_364 = None
	        clamp_min_71: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3635, -128);  add_3635 = None
	        clamp_max_47: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_71, 127);  clamp_min_71 = None
	        _assert_tensor_metadata_211 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_47, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_211 = None
	        convert_element_type_139: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_47, torch.int8);  clamp_max_47 = None
	        view_367: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_69, [sym_size_int, 1500, 1]);  clamp_min_69 = None
	        view_368: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_138, [sym_size_int, 1500, 1]);  convert_element_type_138 = None
	        _assert_tensor_metadata_212 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_139, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_212 = None
	        convert_element_type_140: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_139, torch.float32);  convert_element_type_139 = None
	        _assert_tensor_metadata_213 = torch.ops.aten._assert_tensor_metadata.default(view_368, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_213 = None
	        convert_element_type_141: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_368, torch.float32);  view_368 = None
	        sub_1093: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_140, convert_element_type_141);  convert_element_type_140 = convert_element_type_141 = None
	        mul_2319: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1093, view_367);  sub_1093 = view_367 = None
	        _assert_tensor_metadata_214 = torch.ops.aten._assert_tensor_metadata.default(mul_2319, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_214 = None
	        view_370: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_3_fc2_parametrizations_weight_original0 = None
	        view_371: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_3_fc2_parametrizations_weight_original1 = None
	        view_372: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_3_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_3_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_215 = torch.ops.aten._assert_tensor_metadata.default(view_370, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_215 = None
	        convert_element_type_142: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_370, torch.float32);  view_370 = None
	        _assert_tensor_metadata_216 = torch.ops.aten._assert_tensor_metadata.default(view_372, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_216 = None
	        convert_element_type_143: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_372, torch.float32);  view_372 = None
	        sub_1097: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_142, convert_element_type_143);  convert_element_type_142 = convert_element_type_143 = None
	        mul_2324: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1097, view_371);  sub_1097 = view_371 = None
	        view_373: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_2324, [1280, 5120]);  mul_2324 = None
	        _assert_tensor_metadata_217 = torch.ops.aten._assert_tensor_metadata.default(view_373, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_217 = None
	        mul_2329: "Sym(1500*s6)" = sym_size_int * 1500
	        view_374: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_2319, [mul_2329, 5120]);  mul_2319 = mul_2329 = None
	        permute_40: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_373, [1, 0]);  view_373 = None
	        addmm_19: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_3_fc2_bias, view_374, permute_40);  model_audio_tower_layers_3_fc2_bias = view_374 = permute_40 = None
	        view_375: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_19, [sym_size_int, 1500, 1280]);  addmm_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_3698: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_3400, view_375);  add_3400 = view_375 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_33: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_3698, memory_format = torch.contiguous_format)
	        var_mean_8 = torch.ops.aten.var_mean.correction(clone_33, [2], correction = 0, keepdim = True)
	        getitem_32: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_8[0]
	        getitem_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_8[1];  var_mean_8 = None
	        add_3703: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_32, 1e-05);  getitem_32 = None
	        rsqrt_8: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_3703);  add_3703 = None
	        sub_1103: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_33, getitem_33);  clone_33 = getitem_33 = None
	        mul_2340: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1103, rsqrt_8);  sub_1103 = rsqrt_8 = None
	        mul_2341: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2340, model_audio_tower_layers_4_self_attn_layer_norm_weight);  mul_2340 = model_audio_tower_layers_4_self_attn_layer_norm_weight = None
	        add_3704: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2341, model_audio_tower_layers_4_self_attn_layer_norm_bias);  mul_2341 = model_audio_tower_layers_4_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_3704, [2])
	        amax_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_3704, [2])
	        full_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_24, full_48);  amin_24 = full_48 = None
	        full_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_24, full_49);  amax_24 = full_49 = None
	        sub_1114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_24, minimum_24);  maximum_24 = None
	        div_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1114, 255.0);  sub_1114 = None
	        clamp_min_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_48, 1.1920928955078125e-07);  div_48 = None
	        div_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_24, clamp_min_72);  minimum_24 = None
	        round_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_49);  div_49 = None
	        sub_1120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_49);  round_49 = None
	        clamp_min_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1120, -128);  sub_1120 = None
	        clamp_max_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_73, 127);  clamp_min_73 = None
	        _assert_tensor_metadata_218 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_72, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_218 = None
	        _assert_tensor_metadata_219 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_48, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_219 = None
	        convert_element_type_144: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_48, torch.int8);  clamp_max_48 = None
	        view_378: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_72, [sym_size_int, 1500, 1])
	        view_379: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_144, [sym_size_int, 1500, 1])
	        reciprocal_24: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_378);  view_378 = None
	        mul_2389: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_24, 1.0);  reciprocal_24 = None
	        mul_2392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_3704, mul_2389);  mul_2389 = None
	        round_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2392);  mul_2392 = None
	        add_3791: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_50, view_379);  round_50 = view_379 = None
	        clamp_min_74: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3791, -128);  add_3791 = None
	        clamp_max_49: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_74, 127);  clamp_min_74 = None
	        _assert_tensor_metadata_220 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_49, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_220 = None
	        convert_element_type_145: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_49, torch.int8);  clamp_max_49 = None
	        view_382: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_72, [sym_size_int, 1500, 1]);  clamp_min_72 = None
	        view_383: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_144, [sym_size_int, 1500, 1]);  convert_element_type_144 = None
	        _assert_tensor_metadata_221 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_145, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_221 = None
	        convert_element_type_146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_145, torch.float32);  convert_element_type_145 = None
	        _assert_tensor_metadata_222 = torch.ops.aten._assert_tensor_metadata.default(view_383, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_222 = None
	        convert_element_type_147: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_383, torch.float32);  view_383 = None
	        sub_1140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_146, convert_element_type_147);  convert_element_type_146 = convert_element_type_147 = None
	        mul_2414: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1140, view_382);  sub_1140 = view_382 = None
	        _assert_tensor_metadata_223 = torch.ops.aten._assert_tensor_metadata.default(mul_2414, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_223 = None
	        view_385: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_386: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_387: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_224 = torch.ops.aten._assert_tensor_metadata.default(view_385, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_224 = None
	        convert_element_type_148: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_385, torch.float32);  view_385 = None
	        _assert_tensor_metadata_225 = torch.ops.aten._assert_tensor_metadata.default(view_387, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_225 = None
	        convert_element_type_149: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_387, torch.float32);  view_387 = None
	        sub_1144: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_148, convert_element_type_149);  convert_element_type_148 = convert_element_type_149 = None
	        mul_2419: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1144, view_386);  sub_1144 = view_386 = None
	        view_388: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2419, [1280, 1280]);  mul_2419 = None
	        _assert_tensor_metadata_226 = torch.ops.aten._assert_tensor_metadata.default(view_388, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_226 = None
	        mul_2424: "Sym(1500*s6)" = sym_size_int * 1500
	        view_389: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2414, [mul_2424, 1280]);  mul_2414 = mul_2424 = None
	        permute_41: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_388, [1, 0]);  view_388 = None
	        addmm_20: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_4_self_attn_q_proj_bias, view_389, permute_41);  model_audio_tower_layers_4_self_attn_q_proj_bias = view_389 = permute_41 = None
	        view_390: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_20, [sym_size_int, 1500, 1280]);  addmm_20 = None
	        mul_2431: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_390, 0.125);  view_390 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_391: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_2431, [sym_size_int, 1500, 20, 64]);  mul_2431 = None
	        permute_42: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_391, [0, 2, 1, 3]);  view_391 = None
	        clone_34: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_42, memory_format = torch.contiguous_format);  permute_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_3704, [2])
	        amax_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_3704, [2])
	        full_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_25, full_50);  amin_25 = full_50 = None
	        full_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_25, full_51);  amax_25 = full_51 = None
	        sub_1159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_25, minimum_25);  maximum_25 = None
	        div_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1159, 255.0);  sub_1159 = None
	        clamp_min_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_50, 1.1920928955078125e-07);  div_50 = None
	        div_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_25, clamp_min_75);  minimum_25 = None
	        round_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_51);  div_51 = None
	        sub_1165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_51);  round_51 = None
	        clamp_min_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1165, -128);  sub_1165 = None
	        clamp_max_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_76, 127);  clamp_min_76 = None
	        _assert_tensor_metadata_227 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_75, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_227 = None
	        _assert_tensor_metadata_228 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_50, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_228 = None
	        convert_element_type_150: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_50, torch.int8);  clamp_max_50 = None
	        view_394: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_75, [sym_size_int, 1500, 1])
	        view_395: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_150, [sym_size_int, 1500, 1])
	        reciprocal_25: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_394);  view_394 = None
	        mul_2485: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_25, 1.0);  reciprocal_25 = None
	        mul_2488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_3704, mul_2485);  mul_2485 = None
	        round_52: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2488);  mul_2488 = None
	        add_3943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_52, view_395);  round_52 = view_395 = None
	        clamp_min_77: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3943, -128);  add_3943 = None
	        clamp_max_51: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_77, 127);  clamp_min_77 = None
	        _assert_tensor_metadata_229 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_51, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_229 = None
	        convert_element_type_151: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_51, torch.int8);  clamp_max_51 = None
	        view_398: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_75, [sym_size_int, 1500, 1]);  clamp_min_75 = None
	        view_399: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_150, [sym_size_int, 1500, 1]);  convert_element_type_150 = None
	        _assert_tensor_metadata_230 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_151, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_230 = None
	        convert_element_type_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_151, torch.float32);  convert_element_type_151 = None
	        _assert_tensor_metadata_231 = torch.ops.aten._assert_tensor_metadata.default(view_399, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_231 = None
	        convert_element_type_153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_399, torch.float32);  view_399 = None
	        sub_1185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_152, convert_element_type_153);  convert_element_type_152 = convert_element_type_153 = None
	        mul_2510: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1185, view_398);  sub_1185 = view_398 = None
	        _assert_tensor_metadata_232 = torch.ops.aten._assert_tensor_metadata.default(mul_2510, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_232 = None
	        view_401: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_402: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_403: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_233 = torch.ops.aten._assert_tensor_metadata.default(view_401, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_233 = None
	        convert_element_type_154: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_401, torch.float32);  view_401 = None
	        _assert_tensor_metadata_234 = torch.ops.aten._assert_tensor_metadata.default(view_403, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_234 = None
	        convert_element_type_155: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_403, torch.float32);  view_403 = None
	        sub_1189: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_154, convert_element_type_155);  convert_element_type_154 = convert_element_type_155 = None
	        mul_2515: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1189, view_402);  sub_1189 = view_402 = None
	        view_404: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2515, [1280, 1280]);  mul_2515 = None
	        _assert_tensor_metadata_235 = torch.ops.aten._assert_tensor_metadata.default(view_404, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_235 = None
	        permute_43: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_404, [1, 0]);  view_404 = None
	        mul_2518: "Sym(1500*s6)" = sym_size_int * 1500
	        view_405: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2510, [mul_2518, 1280]);  mul_2510 = mul_2518 = None
	        mm_4: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_405, permute_43);  view_405 = permute_43 = None
	        view_406: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_4, [sym_size_int, 1500, 1280]);  mm_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_407: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_406, [sym_size_int, -1, 20, 64]);  view_406 = None
	        permute_44: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_407, [0, 2, 1, 3]);  view_407 = None
	        clone_35: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_44, memory_format = torch.contiguous_format);  permute_44 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_3704, [2])
	        amax_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_3704, [2])
	        full_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_26, full_52);  amin_26 = full_52 = None
	        full_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_26, full_53);  amax_26 = full_53 = None
	        sub_1203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_26, minimum_26);  maximum_26 = None
	        div_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1203, 255.0);  sub_1203 = None
	        clamp_min_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_52, 1.1920928955078125e-07);  div_52 = None
	        div_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_26, clamp_min_78);  minimum_26 = None
	        round_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_53);  div_53 = None
	        sub_1209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_53);  round_53 = None
	        clamp_min_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1209, -128);  sub_1209 = None
	        clamp_max_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_79, 127);  clamp_min_79 = None
	        _assert_tensor_metadata_236 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_78, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_236 = None
	        _assert_tensor_metadata_237 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_52, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_237 = None
	        convert_element_type_156: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_52, torch.int8);  clamp_max_52 = None
	        view_410: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_78, [sym_size_int, 1500, 1])
	        view_411: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_156, [sym_size_int, 1500, 1])
	        reciprocal_26: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_410);  view_410 = None
	        mul_2584: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_26, 1.0);  reciprocal_26 = None
	        mul_2587: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_3704, mul_2584);  add_3704 = mul_2584 = None
	        round_54: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2587);  mul_2587 = None
	        add_4091: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_54, view_411);  round_54 = view_411 = None
	        clamp_min_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4091, -128);  add_4091 = None
	        clamp_max_53: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_80, 127);  clamp_min_80 = None
	        _assert_tensor_metadata_238 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_53, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_238 = None
	        convert_element_type_157: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_53, torch.int8);  clamp_max_53 = None
	        view_414: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_78, [sym_size_int, 1500, 1]);  clamp_min_78 = None
	        view_415: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_156, [sym_size_int, 1500, 1]);  convert_element_type_156 = None
	        _assert_tensor_metadata_239 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_157, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_239 = None
	        convert_element_type_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_157, torch.float32);  convert_element_type_157 = None
	        _assert_tensor_metadata_240 = torch.ops.aten._assert_tensor_metadata.default(view_415, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_240 = None
	        convert_element_type_159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_415, torch.float32);  view_415 = None
	        sub_1229: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_158, convert_element_type_159);  convert_element_type_158 = convert_element_type_159 = None
	        mul_2609: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1229, view_414);  sub_1229 = view_414 = None
	        _assert_tensor_metadata_241 = torch.ops.aten._assert_tensor_metadata.default(mul_2609, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_241 = None
	        view_417: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_418: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_419: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_242 = torch.ops.aten._assert_tensor_metadata.default(view_417, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_242 = None
	        convert_element_type_160: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_417, torch.float32);  view_417 = None
	        _assert_tensor_metadata_243 = torch.ops.aten._assert_tensor_metadata.default(view_419, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_243 = None
	        convert_element_type_161: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_419, torch.float32);  view_419 = None
	        sub_1233: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_160, convert_element_type_161);  convert_element_type_160 = convert_element_type_161 = None
	        mul_2614: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1233, view_418);  sub_1233 = view_418 = None
	        view_420: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2614, [1280, 1280]);  mul_2614 = None
	        _assert_tensor_metadata_244 = torch.ops.aten._assert_tensor_metadata.default(view_420, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_244 = None
	        mul_2619: "Sym(1500*s6)" = sym_size_int * 1500
	        view_421: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2609, [mul_2619, 1280]);  mul_2609 = mul_2619 = None
	        permute_45: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_420, [1, 0]);  view_420 = None
	        addmm_21: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_4_self_attn_v_proj_bias, view_421, permute_45);  model_audio_tower_layers_4_self_attn_v_proj_bias = view_421 = permute_45 = None
	        view_422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_21, [sym_size_int, 1500, 1280]);  addmm_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_423: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_422, [sym_size_int, -1, 20, 64]);  view_422 = None
	        permute_46: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_423, [0, 2, 1, 3]);  view_423 = None
	        clone_36: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_46, memory_format = torch.contiguous_format);  permute_46 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_4 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_34, clone_35, clone_36, None, False, scale = 1.0);  clone_34 = clone_35 = clone_36 = None
	        getitem_34: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_4[0];  _scaled_dot_product_efficient_attention_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_47: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_34, [0, 2, 1, 3]);  getitem_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_424: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_47, [sym_size_int, 1500, -1]);  permute_47 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_424, [2])
	        amax_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_424, [2])
	        full_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_27, full_54);  amin_27 = full_54 = None
	        full_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_27, full_55);  amax_27 = full_55 = None
	        sub_1251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_27, minimum_27);  maximum_27 = None
	        div_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1251, 255.0);  sub_1251 = None
	        clamp_min_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_54, 1.1920928955078125e-07);  div_54 = None
	        div_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_27, clamp_min_81);  minimum_27 = None
	        round_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_55);  div_55 = None
	        sub_1257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_55);  round_55 = None
	        clamp_min_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1257, -128);  sub_1257 = None
	        clamp_max_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_82, 127);  clamp_min_82 = None
	        _assert_tensor_metadata_245 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_81, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_245 = None
	        _assert_tensor_metadata_246 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_54, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_246 = None
	        convert_element_type_162: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_54, torch.int8);  clamp_max_54 = None
	        view_427: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_81, [sym_size_int, 1500, 1])
	        view_428: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_162, [sym_size_int, 1500, 1])
	        reciprocal_27: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_427);  view_427 = None
	        mul_2689: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_27, 1.0);  reciprocal_27 = None
	        mul_2692: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_424, mul_2689);  view_424 = mul_2689 = None
	        round_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2692);  mul_2692 = None
	        add_4255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_56, view_428);  round_56 = view_428 = None
	        clamp_min_83: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4255, -128);  add_4255 = None
	        clamp_max_55: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_83, 127);  clamp_min_83 = None
	        _assert_tensor_metadata_247 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_55, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_247 = None
	        convert_element_type_163: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_55, torch.int8);  clamp_max_55 = None
	        view_431: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_81, [sym_size_int, 1500, 1]);  clamp_min_81 = None
	        view_432: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_162, [sym_size_int, 1500, 1]);  convert_element_type_162 = None
	        _assert_tensor_metadata_248 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_163, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_248 = None
	        convert_element_type_164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_163, torch.float32);  convert_element_type_163 = None
	        _assert_tensor_metadata_249 = torch.ops.aten._assert_tensor_metadata.default(view_432, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_249 = None
	        convert_element_type_165: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_432, torch.float32);  view_432 = None
	        sub_1277: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_164, convert_element_type_165);  convert_element_type_164 = convert_element_type_165 = None
	        mul_2714: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1277, view_431);  sub_1277 = view_431 = None
	        _assert_tensor_metadata_250 = torch.ops.aten._assert_tensor_metadata.default(mul_2714, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_250 = None
	        view_434: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_435: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_436: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_251 = torch.ops.aten._assert_tensor_metadata.default(view_434, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_251 = None
	        convert_element_type_166: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_434, torch.float32);  view_434 = None
	        _assert_tensor_metadata_252 = torch.ops.aten._assert_tensor_metadata.default(view_436, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_252 = None
	        convert_element_type_167: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_436, torch.float32);  view_436 = None
	        sub_1281: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_166, convert_element_type_167);  convert_element_type_166 = convert_element_type_167 = None
	        mul_2719: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1281, view_435);  sub_1281 = view_435 = None
	        view_437: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2719, [1280, 1280]);  mul_2719 = None
	        _assert_tensor_metadata_253 = torch.ops.aten._assert_tensor_metadata.default(view_437, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_253 = None
	        mul_2724: "Sym(1500*s6)" = sym_size_int * 1500
	        view_438: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2714, [mul_2724, 1280]);  mul_2714 = mul_2724 = None
	        permute_48: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_437, [1, 0]);  view_437 = None
	        addmm_22: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_4_self_attn_out_proj_bias, view_438, permute_48);  model_audio_tower_layers_4_self_attn_out_proj_bias = view_438 = permute_48 = None
	        view_439: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_22, [sym_size_int, 1500, 1280]);  addmm_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_4318: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_3698, view_439);  add_3698 = view_439 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_4318, memory_format = torch.contiguous_format)
	        var_mean_9 = torch.ops.aten.var_mean.correction(clone_38, [2], correction = 0, keepdim = True)
	        getitem_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_9[0]
	        getitem_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_9[1];  var_mean_9 = None
	        add_4323: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_38, 1e-05);  getitem_38 = None
	        rsqrt_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_4323);  add_4323 = None
	        sub_1287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_38, getitem_39);  clone_38 = getitem_39 = None
	        mul_2735: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1287, rsqrt_9);  sub_1287 = rsqrt_9 = None
	        mul_2736: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2735, model_audio_tower_layers_4_final_layer_norm_weight);  mul_2735 = model_audio_tower_layers_4_final_layer_norm_weight = None
	        add_4324: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2736, model_audio_tower_layers_4_final_layer_norm_bias);  mul_2736 = model_audio_tower_layers_4_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_4324, [2])
	        amax_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_4324, [2])
	        full_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_28, full_56);  amin_28 = full_56 = None
	        full_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_28, full_57);  amax_28 = full_57 = None
	        sub_1298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_28, minimum_28);  maximum_28 = None
	        div_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1298, 255.0);  sub_1298 = None
	        clamp_min_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_56, 1.1920928955078125e-07);  div_56 = None
	        div_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_28, clamp_min_84);  minimum_28 = None
	        round_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_57);  div_57 = None
	        sub_1304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_57);  round_57 = None
	        clamp_min_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1304, -128);  sub_1304 = None
	        clamp_max_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_85, 127);  clamp_min_85 = None
	        _assert_tensor_metadata_254 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_84, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_254 = None
	        _assert_tensor_metadata_255 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_56, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_255 = None
	        convert_element_type_168: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_56, torch.int8);  clamp_max_56 = None
	        view_442: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_84, [sym_size_int, 1500, 1])
	        view_443: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_168, [sym_size_int, 1500, 1])
	        reciprocal_28: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_442);  view_442 = None
	        mul_2784: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_28, 1.0);  reciprocal_28 = None
	        mul_2787: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_4324, mul_2784);  add_4324 = mul_2784 = None
	        round_58: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2787);  mul_2787 = None
	        add_4411: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_58, view_443);  round_58 = view_443 = None
	        clamp_min_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4411, -128);  add_4411 = None
	        clamp_max_57: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_86, 127);  clamp_min_86 = None
	        _assert_tensor_metadata_256 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_57, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_256 = None
	        convert_element_type_169: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_57, torch.int8);  clamp_max_57 = None
	        view_446: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_84, [sym_size_int, 1500, 1]);  clamp_min_84 = None
	        view_447: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_168, [sym_size_int, 1500, 1]);  convert_element_type_168 = None
	        _assert_tensor_metadata_257 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_169, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_257 = None
	        convert_element_type_170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_169, torch.float32);  convert_element_type_169 = None
	        _assert_tensor_metadata_258 = torch.ops.aten._assert_tensor_metadata.default(view_447, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_258 = None
	        convert_element_type_171: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_447, torch.float32);  view_447 = None
	        sub_1324: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_170, convert_element_type_171);  convert_element_type_170 = convert_element_type_171 = None
	        mul_2809: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1324, view_446);  sub_1324 = view_446 = None
	        _assert_tensor_metadata_259 = torch.ops.aten._assert_tensor_metadata.default(mul_2809, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_259 = None
	        view_449: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_4_fc1_parametrizations_weight_original0 = None
	        view_450: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_4_fc1_parametrizations_weight_original1 = None
	        view_451: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_4_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_260 = torch.ops.aten._assert_tensor_metadata.default(view_449, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_260 = None
	        convert_element_type_172: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_449, torch.float32);  view_449 = None
	        _assert_tensor_metadata_261 = torch.ops.aten._assert_tensor_metadata.default(view_451, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_261 = None
	        convert_element_type_173: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_451, torch.float32);  view_451 = None
	        sub_1328: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_172, convert_element_type_173);  convert_element_type_172 = convert_element_type_173 = None
	        mul_2814: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1328, view_450);  sub_1328 = view_450 = None
	        view_452: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2814, [5120, 1280]);  mul_2814 = None
	        _assert_tensor_metadata_262 = torch.ops.aten._assert_tensor_metadata.default(view_452, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_262 = None
	        mul_2819: "Sym(1500*s6)" = sym_size_int * 1500
	        view_453: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2809, [mul_2819, 1280]);  mul_2809 = mul_2819 = None
	        permute_49: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_452, [1, 0]);  view_452 = None
	        addmm_23: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_4_fc1_bias, view_453, permute_49);  model_audio_tower_layers_4_fc1_bias = view_453 = permute_49 = None
	        view_454: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_23, [sym_size_int, 1500, 5120]);  addmm_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_2826: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_454, 0.5)
	        mul_2827: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_454, 0.7071067811865476);  view_454 = None
	        erf_6: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_2827);  mul_2827 = None
	        add_4470: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_6, 1);  erf_6 = None
	        mul_2828: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2826, add_4470);  mul_2826 = add_4470 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_2828, [2])
	        amax_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_2828, [2])
	        full_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_29, full_58);  amin_29 = full_58 = None
	        full_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_29, full_59);  amax_29 = full_59 = None
	        sub_1341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_29, minimum_29);  maximum_29 = None
	        div_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1341, 255.0);  sub_1341 = None
	        clamp_min_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_58, 1.1920928955078125e-07);  div_58 = None
	        div_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_29, clamp_min_87);  minimum_29 = None
	        round_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_59);  div_59 = None
	        sub_1347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_59);  round_59 = None
	        clamp_min_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1347, -128);  sub_1347 = None
	        clamp_max_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_88, 127);  clamp_min_88 = None
	        _assert_tensor_metadata_263 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_87, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_263 = None
	        _assert_tensor_metadata_264 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_58, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_264 = None
	        convert_element_type_174: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_58, torch.int8);  clamp_max_58 = None
	        view_457: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_87, [sym_size_int, 1500, 1])
	        view_458: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_174, [sym_size_int, 1500, 1])
	        reciprocal_29: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_457);  view_457 = None
	        mul_2874: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_29, 1.0);  reciprocal_29 = None
	        mul_2877: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2828, mul_2874);  mul_2828 = mul_2874 = None
	        round_60: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_2877);  mul_2877 = None
	        add_4553: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_60, view_458);  round_60 = view_458 = None
	        clamp_min_89: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4553, -128);  add_4553 = None
	        clamp_max_59: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_89, 127);  clamp_min_89 = None
	        _assert_tensor_metadata_265 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_59, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_265 = None
	        convert_element_type_175: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_59, torch.int8);  clamp_max_59 = None
	        view_461: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_87, [sym_size_int, 1500, 1]);  clamp_min_87 = None
	        view_462: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_174, [sym_size_int, 1500, 1]);  convert_element_type_174 = None
	        _assert_tensor_metadata_266 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_175, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_266 = None
	        convert_element_type_176: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_175, torch.float32);  convert_element_type_175 = None
	        _assert_tensor_metadata_267 = torch.ops.aten._assert_tensor_metadata.default(view_462, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_267 = None
	        convert_element_type_177: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_462, torch.float32);  view_462 = None
	        sub_1367: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_176, convert_element_type_177);  convert_element_type_176 = convert_element_type_177 = None
	        mul_2899: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1367, view_461);  sub_1367 = view_461 = None
	        _assert_tensor_metadata_268 = torch.ops.aten._assert_tensor_metadata.default(mul_2899, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_268 = None
	        view_464: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_4_fc2_parametrizations_weight_original0 = None
	        view_465: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_4_fc2_parametrizations_weight_original1 = None
	        view_466: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_4_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_4_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_269 = torch.ops.aten._assert_tensor_metadata.default(view_464, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_269 = None
	        convert_element_type_178: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_464, torch.float32);  view_464 = None
	        _assert_tensor_metadata_270 = torch.ops.aten._assert_tensor_metadata.default(view_466, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_270 = None
	        convert_element_type_179: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_466, torch.float32);  view_466 = None
	        sub_1371: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_178, convert_element_type_179);  convert_element_type_178 = convert_element_type_179 = None
	        mul_2904: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1371, view_465);  sub_1371 = view_465 = None
	        view_467: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_2904, [1280, 5120]);  mul_2904 = None
	        _assert_tensor_metadata_271 = torch.ops.aten._assert_tensor_metadata.default(view_467, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_271 = None
	        mul_2909: "Sym(1500*s6)" = sym_size_int * 1500
	        view_468: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_2899, [mul_2909, 5120]);  mul_2899 = mul_2909 = None
	        permute_50: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_467, [1, 0]);  view_467 = None
	        addmm_24: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_4_fc2_bias, view_468, permute_50);  model_audio_tower_layers_4_fc2_bias = view_468 = permute_50 = None
	        view_469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_24, [sym_size_int, 1500, 1280]);  addmm_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_4616: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_4318, view_469);  add_4318 = view_469 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_41: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_4616, memory_format = torch.contiguous_format)
	        var_mean_10 = torch.ops.aten.var_mean.correction(clone_41, [2], correction = 0, keepdim = True)
	        getitem_40: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_10[0]
	        getitem_41: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_10[1];  var_mean_10 = None
	        add_4621: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_40, 1e-05);  getitem_40 = None
	        rsqrt_10: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_4621);  add_4621 = None
	        sub_1377: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_41, getitem_41);  clone_41 = getitem_41 = None
	        mul_2920: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1377, rsqrt_10);  sub_1377 = rsqrt_10 = None
	        mul_2921: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2920, model_audio_tower_layers_5_self_attn_layer_norm_weight);  mul_2920 = model_audio_tower_layers_5_self_attn_layer_norm_weight = None
	        add_4622: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2921, model_audio_tower_layers_5_self_attn_layer_norm_bias);  mul_2921 = model_audio_tower_layers_5_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_4622, [2])
	        amax_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_4622, [2])
	        full_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_30, full_60);  amin_30 = full_60 = None
	        full_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_30, full_61);  amax_30 = full_61 = None
	        sub_1388: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_30, minimum_30);  maximum_30 = None
	        div_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1388, 255.0);  sub_1388 = None
	        clamp_min_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_60, 1.1920928955078125e-07);  div_60 = None
	        div_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_30, clamp_min_90);  minimum_30 = None
	        round_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_61);  div_61 = None
	        sub_1394: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_61);  round_61 = None
	        clamp_min_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1394, -128);  sub_1394 = None
	        clamp_max_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_91, 127);  clamp_min_91 = None
	        _assert_tensor_metadata_272 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_90, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_272 = None
	        _assert_tensor_metadata_273 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_60, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_273 = None
	        convert_element_type_180: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_60, torch.int8);  clamp_max_60 = None
	        view_472: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_90, [sym_size_int, 1500, 1])
	        view_473: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_180, [sym_size_int, 1500, 1])
	        reciprocal_30: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_472);  view_472 = None
	        mul_2969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_30, 1.0);  reciprocal_30 = None
	        mul_2972: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_4622, mul_2969);  mul_2969 = None
	        round_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2972);  mul_2972 = None
	        add_4709: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_62, view_473);  round_62 = view_473 = None
	        clamp_min_92: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4709, -128);  add_4709 = None
	        clamp_max_61: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_92, 127);  clamp_min_92 = None
	        _assert_tensor_metadata_274 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_61, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_274 = None
	        convert_element_type_181: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_61, torch.int8);  clamp_max_61 = None
	        view_476: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_90, [sym_size_int, 1500, 1]);  clamp_min_90 = None
	        view_477: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_180, [sym_size_int, 1500, 1]);  convert_element_type_180 = None
	        _assert_tensor_metadata_275 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_181, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_275 = None
	        convert_element_type_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_181, torch.float32);  convert_element_type_181 = None
	        _assert_tensor_metadata_276 = torch.ops.aten._assert_tensor_metadata.default(view_477, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_276 = None
	        convert_element_type_183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_477, torch.float32);  view_477 = None
	        sub_1414: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_182, convert_element_type_183);  convert_element_type_182 = convert_element_type_183 = None
	        mul_2994: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1414, view_476);  sub_1414 = view_476 = None
	        _assert_tensor_metadata_277 = torch.ops.aten._assert_tensor_metadata.default(mul_2994, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_277 = None
	        view_479: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_480: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_481: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_278 = torch.ops.aten._assert_tensor_metadata.default(view_479, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_278 = None
	        convert_element_type_184: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_479, torch.float32);  view_479 = None
	        _assert_tensor_metadata_279 = torch.ops.aten._assert_tensor_metadata.default(view_481, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_279 = None
	        convert_element_type_185: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_481, torch.float32);  view_481 = None
	        sub_1418: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_184, convert_element_type_185);  convert_element_type_184 = convert_element_type_185 = None
	        mul_2999: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1418, view_480);  sub_1418 = view_480 = None
	        view_482: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2999, [1280, 1280]);  mul_2999 = None
	        _assert_tensor_metadata_280 = torch.ops.aten._assert_tensor_metadata.default(view_482, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_280 = None
	        mul_3004: "Sym(1500*s6)" = sym_size_int * 1500
	        view_483: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_2994, [mul_3004, 1280]);  mul_2994 = mul_3004 = None
	        permute_51: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_482, [1, 0]);  view_482 = None
	        addmm_25: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_5_self_attn_q_proj_bias, view_483, permute_51);  model_audio_tower_layers_5_self_attn_q_proj_bias = view_483 = permute_51 = None
	        view_484: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_25, [sym_size_int, 1500, 1280]);  addmm_25 = None
	        mul_3011: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_484, 0.125);  view_484 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_485: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_3011, [sym_size_int, 1500, 20, 64]);  mul_3011 = None
	        permute_52: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_485, [0, 2, 1, 3]);  view_485 = None
	        clone_42: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_52, memory_format = torch.contiguous_format);  permute_52 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_4622, [2])
	        amax_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_4622, [2])
	        full_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_31, full_62);  amin_31 = full_62 = None
	        full_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_31, full_63);  amax_31 = full_63 = None
	        sub_1433: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_31, minimum_31);  maximum_31 = None
	        div_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1433, 255.0);  sub_1433 = None
	        clamp_min_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_62, 1.1920928955078125e-07);  div_62 = None
	        div_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_31, clamp_min_93);  minimum_31 = None
	        round_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_63);  div_63 = None
	        sub_1439: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_63);  round_63 = None
	        clamp_min_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1439, -128);  sub_1439 = None
	        clamp_max_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_94, 127);  clamp_min_94 = None
	        _assert_tensor_metadata_281 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_93, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_281 = None
	        _assert_tensor_metadata_282 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_62, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_282 = None
	        convert_element_type_186: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_62, torch.int8);  clamp_max_62 = None
	        view_488: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_93, [sym_size_int, 1500, 1])
	        view_489: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_186, [sym_size_int, 1500, 1])
	        reciprocal_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_488);  view_488 = None
	        mul_3065: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_31, 1.0);  reciprocal_31 = None
	        mul_3068: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_4622, mul_3065);  mul_3065 = None
	        round_64: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3068);  mul_3068 = None
	        add_4861: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_64, view_489);  round_64 = view_489 = None
	        clamp_min_95: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4861, -128);  add_4861 = None
	        clamp_max_63: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_95, 127);  clamp_min_95 = None
	        _assert_tensor_metadata_283 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_63, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_283 = None
	        convert_element_type_187: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_63, torch.int8);  clamp_max_63 = None
	        view_492: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_93, [sym_size_int, 1500, 1]);  clamp_min_93 = None
	        view_493: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_186, [sym_size_int, 1500, 1]);  convert_element_type_186 = None
	        _assert_tensor_metadata_284 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_187, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_284 = None
	        convert_element_type_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_187, torch.float32);  convert_element_type_187 = None
	        _assert_tensor_metadata_285 = torch.ops.aten._assert_tensor_metadata.default(view_493, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_285 = None
	        convert_element_type_189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_493, torch.float32);  view_493 = None
	        sub_1459: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_188, convert_element_type_189);  convert_element_type_188 = convert_element_type_189 = None
	        mul_3090: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1459, view_492);  sub_1459 = view_492 = None
	        _assert_tensor_metadata_286 = torch.ops.aten._assert_tensor_metadata.default(mul_3090, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_286 = None
	        view_495: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_496: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_497: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_287 = torch.ops.aten._assert_tensor_metadata.default(view_495, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_287 = None
	        convert_element_type_190: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_495, torch.float32);  view_495 = None
	        _assert_tensor_metadata_288 = torch.ops.aten._assert_tensor_metadata.default(view_497, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_288 = None
	        convert_element_type_191: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_497, torch.float32);  view_497 = None
	        sub_1463: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_190, convert_element_type_191);  convert_element_type_190 = convert_element_type_191 = None
	        mul_3095: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1463, view_496);  sub_1463 = view_496 = None
	        view_498: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3095, [1280, 1280]);  mul_3095 = None
	        _assert_tensor_metadata_289 = torch.ops.aten._assert_tensor_metadata.default(view_498, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_289 = None
	        permute_53: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_498, [1, 0]);  view_498 = None
	        mul_3098: "Sym(1500*s6)" = sym_size_int * 1500
	        view_499: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3090, [mul_3098, 1280]);  mul_3090 = mul_3098 = None
	        mm_5: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_499, permute_53);  view_499 = permute_53 = None
	        view_500: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_5, [sym_size_int, 1500, 1280]);  mm_5 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_501: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_500, [sym_size_int, -1, 20, 64]);  view_500 = None
	        permute_54: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_501, [0, 2, 1, 3]);  view_501 = None
	        clone_43: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_54, memory_format = torch.contiguous_format);  permute_54 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_4622, [2])
	        amax_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_4622, [2])
	        full_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_32, full_64);  amin_32 = full_64 = None
	        full_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_32, full_65);  amax_32 = full_65 = None
	        sub_1477: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_32, minimum_32);  maximum_32 = None
	        div_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1477, 255.0);  sub_1477 = None
	        clamp_min_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_64, 1.1920928955078125e-07);  div_64 = None
	        div_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_32, clamp_min_96);  minimum_32 = None
	        round_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_65);  div_65 = None
	        sub_1483: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_65);  round_65 = None
	        clamp_min_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1483, -128);  sub_1483 = None
	        clamp_max_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_97, 127);  clamp_min_97 = None
	        _assert_tensor_metadata_290 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_96, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_290 = None
	        _assert_tensor_metadata_291 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_64, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_291 = None
	        convert_element_type_192: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_64, torch.int8);  clamp_max_64 = None
	        view_504: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_96, [sym_size_int, 1500, 1])
	        view_505: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_192, [sym_size_int, 1500, 1])
	        reciprocal_32: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_504);  view_504 = None
	        mul_3164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_32, 1.0);  reciprocal_32 = None
	        mul_3167: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_4622, mul_3164);  add_4622 = mul_3164 = None
	        round_66: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3167);  mul_3167 = None
	        add_5009: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_66, view_505);  round_66 = view_505 = None
	        clamp_min_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5009, -128);  add_5009 = None
	        clamp_max_65: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_98, 127);  clamp_min_98 = None
	        _assert_tensor_metadata_292 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_65, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_292 = None
	        convert_element_type_193: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_65, torch.int8);  clamp_max_65 = None
	        view_508: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_96, [sym_size_int, 1500, 1]);  clamp_min_96 = None
	        view_509: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_192, [sym_size_int, 1500, 1]);  convert_element_type_192 = None
	        _assert_tensor_metadata_293 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_193, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_293 = None
	        convert_element_type_194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_193, torch.float32);  convert_element_type_193 = None
	        _assert_tensor_metadata_294 = torch.ops.aten._assert_tensor_metadata.default(view_509, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_294 = None
	        convert_element_type_195: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_509, torch.float32);  view_509 = None
	        sub_1503: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_194, convert_element_type_195);  convert_element_type_194 = convert_element_type_195 = None
	        mul_3189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1503, view_508);  sub_1503 = view_508 = None
	        _assert_tensor_metadata_295 = torch.ops.aten._assert_tensor_metadata.default(mul_3189, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_295 = None
	        view_511: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_512: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_513: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_296 = torch.ops.aten._assert_tensor_metadata.default(view_511, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_296 = None
	        convert_element_type_196: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_511, torch.float32);  view_511 = None
	        _assert_tensor_metadata_297 = torch.ops.aten._assert_tensor_metadata.default(view_513, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_297 = None
	        convert_element_type_197: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_513, torch.float32);  view_513 = None
	        sub_1507: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_196, convert_element_type_197);  convert_element_type_196 = convert_element_type_197 = None
	        mul_3194: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1507, view_512);  sub_1507 = view_512 = None
	        view_514: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3194, [1280, 1280]);  mul_3194 = None
	        _assert_tensor_metadata_298 = torch.ops.aten._assert_tensor_metadata.default(view_514, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_298 = None
	        mul_3199: "Sym(1500*s6)" = sym_size_int * 1500
	        view_515: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3189, [mul_3199, 1280]);  mul_3189 = mul_3199 = None
	        permute_55: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_514, [1, 0]);  view_514 = None
	        addmm_26: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_5_self_attn_v_proj_bias, view_515, permute_55);  model_audio_tower_layers_5_self_attn_v_proj_bias = view_515 = permute_55 = None
	        view_516: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_26, [sym_size_int, 1500, 1280]);  addmm_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_517: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_516, [sym_size_int, -1, 20, 64]);  view_516 = None
	        permute_56: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_517, [0, 2, 1, 3]);  view_517 = None
	        clone_44: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_56, memory_format = torch.contiguous_format);  permute_56 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_5 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_42, clone_43, clone_44, None, False, scale = 1.0);  clone_42 = clone_43 = clone_44 = None
	        getitem_42: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_5[0];  _scaled_dot_product_efficient_attention_5 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_57: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_42, [0, 2, 1, 3]);  getitem_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_57, [sym_size_int, 1500, -1]);  permute_57 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_518, [2])
	        amax_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_518, [2])
	        full_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_33, full_66);  amin_33 = full_66 = None
	        full_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_33, full_67);  amax_33 = full_67 = None
	        sub_1525: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_33, minimum_33);  maximum_33 = None
	        div_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1525, 255.0);  sub_1525 = None
	        clamp_min_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_66, 1.1920928955078125e-07);  div_66 = None
	        div_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_33, clamp_min_99);  minimum_33 = None
	        round_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_67);  div_67 = None
	        sub_1531: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_67);  round_67 = None
	        clamp_min_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1531, -128);  sub_1531 = None
	        clamp_max_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_100, 127);  clamp_min_100 = None
	        _assert_tensor_metadata_299 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_99, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_299 = None
	        _assert_tensor_metadata_300 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_66, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_300 = None
	        convert_element_type_198: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_66, torch.int8);  clamp_max_66 = None
	        view_521: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_99, [sym_size_int, 1500, 1])
	        view_522: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_198, [sym_size_int, 1500, 1])
	        reciprocal_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_521);  view_521 = None
	        mul_3269: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_33, 1.0);  reciprocal_33 = None
	        mul_3272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_518, mul_3269);  view_518 = mul_3269 = None
	        round_68: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3272);  mul_3272 = None
	        add_5173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_68, view_522);  round_68 = view_522 = None
	        clamp_min_101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5173, -128);  add_5173 = None
	        clamp_max_67: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_101, 127);  clamp_min_101 = None
	        _assert_tensor_metadata_301 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_67, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_301 = None
	        convert_element_type_199: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_67, torch.int8);  clamp_max_67 = None
	        view_525: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_99, [sym_size_int, 1500, 1]);  clamp_min_99 = None
	        view_526: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_198, [sym_size_int, 1500, 1]);  convert_element_type_198 = None
	        _assert_tensor_metadata_302 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_199, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_302 = None
	        convert_element_type_200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_199, torch.float32);  convert_element_type_199 = None
	        _assert_tensor_metadata_303 = torch.ops.aten._assert_tensor_metadata.default(view_526, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_303 = None
	        convert_element_type_201: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_526, torch.float32);  view_526 = None
	        sub_1551: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_200, convert_element_type_201);  convert_element_type_200 = convert_element_type_201 = None
	        mul_3294: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1551, view_525);  sub_1551 = view_525 = None
	        _assert_tensor_metadata_304 = torch.ops.aten._assert_tensor_metadata.default(mul_3294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_304 = None
	        view_528: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_529: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_530: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_305 = torch.ops.aten._assert_tensor_metadata.default(view_528, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_305 = None
	        convert_element_type_202: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_528, torch.float32);  view_528 = None
	        _assert_tensor_metadata_306 = torch.ops.aten._assert_tensor_metadata.default(view_530, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_306 = None
	        convert_element_type_203: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_530, torch.float32);  view_530 = None
	        sub_1555: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_202, convert_element_type_203);  convert_element_type_202 = convert_element_type_203 = None
	        mul_3299: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1555, view_529);  sub_1555 = view_529 = None
	        view_531: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3299, [1280, 1280]);  mul_3299 = None
	        _assert_tensor_metadata_307 = torch.ops.aten._assert_tensor_metadata.default(view_531, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_307 = None
	        mul_3304: "Sym(1500*s6)" = sym_size_int * 1500
	        view_532: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3294, [mul_3304, 1280]);  mul_3294 = mul_3304 = None
	        permute_58: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_531, [1, 0]);  view_531 = None
	        addmm_27: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_5_self_attn_out_proj_bias, view_532, permute_58);  model_audio_tower_layers_5_self_attn_out_proj_bias = view_532 = permute_58 = None
	        view_533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_27, [sym_size_int, 1500, 1280]);  addmm_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_5236: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_4616, view_533);  add_4616 = view_533 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_46: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_5236, memory_format = torch.contiguous_format)
	        var_mean_11 = torch.ops.aten.var_mean.correction(clone_46, [2], correction = 0, keepdim = True)
	        getitem_46: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_11[0]
	        getitem_47: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_11[1];  var_mean_11 = None
	        add_5241: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_46, 1e-05);  getitem_46 = None
	        rsqrt_11: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_5241);  add_5241 = None
	        sub_1561: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_46, getitem_47);  clone_46 = getitem_47 = None
	        mul_3315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1561, rsqrt_11);  sub_1561 = rsqrt_11 = None
	        mul_3316: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3315, model_audio_tower_layers_5_final_layer_norm_weight);  mul_3315 = model_audio_tower_layers_5_final_layer_norm_weight = None
	        add_5242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_3316, model_audio_tower_layers_5_final_layer_norm_bias);  mul_3316 = model_audio_tower_layers_5_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_5242, [2])
	        amax_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_5242, [2])
	        full_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_34, full_68);  amin_34 = full_68 = None
	        full_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_34, full_69);  amax_34 = full_69 = None
	        sub_1572: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_34, minimum_34);  maximum_34 = None
	        div_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1572, 255.0);  sub_1572 = None
	        clamp_min_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_68, 1.1920928955078125e-07);  div_68 = None
	        div_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_34, clamp_min_102);  minimum_34 = None
	        round_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_69);  div_69 = None
	        sub_1578: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_69);  round_69 = None
	        clamp_min_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1578, -128);  sub_1578 = None
	        clamp_max_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_103, 127);  clamp_min_103 = None
	        _assert_tensor_metadata_308 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_102, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_308 = None
	        _assert_tensor_metadata_309 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_68, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_309 = None
	        convert_element_type_204: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_68, torch.int8);  clamp_max_68 = None
	        view_536: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_102, [sym_size_int, 1500, 1])
	        view_537: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_204, [sym_size_int, 1500, 1])
	        reciprocal_34: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_536);  view_536 = None
	        mul_3364: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_34, 1.0);  reciprocal_34 = None
	        mul_3367: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_5242, mul_3364);  add_5242 = mul_3364 = None
	        round_70: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3367);  mul_3367 = None
	        add_5329: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_70, view_537);  round_70 = view_537 = None
	        clamp_min_104: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5329, -128);  add_5329 = None
	        clamp_max_69: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_104, 127);  clamp_min_104 = None
	        _assert_tensor_metadata_310 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_69, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_310 = None
	        convert_element_type_205: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_69, torch.int8);  clamp_max_69 = None
	        view_540: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_102, [sym_size_int, 1500, 1]);  clamp_min_102 = None
	        view_541: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_204, [sym_size_int, 1500, 1]);  convert_element_type_204 = None
	        _assert_tensor_metadata_311 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_205, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_311 = None
	        convert_element_type_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_205, torch.float32);  convert_element_type_205 = None
	        _assert_tensor_metadata_312 = torch.ops.aten._assert_tensor_metadata.default(view_541, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_312 = None
	        convert_element_type_207: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_541, torch.float32);  view_541 = None
	        sub_1598: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_206, convert_element_type_207);  convert_element_type_206 = convert_element_type_207 = None
	        mul_3389: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1598, view_540);  sub_1598 = view_540 = None
	        _assert_tensor_metadata_313 = torch.ops.aten._assert_tensor_metadata.default(mul_3389, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_313 = None
	        view_543: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_5_fc1_parametrizations_weight_original0 = None
	        view_544: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_5_fc1_parametrizations_weight_original1 = None
	        view_545: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_5_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_314 = torch.ops.aten._assert_tensor_metadata.default(view_543, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_314 = None
	        convert_element_type_208: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_543, torch.float32);  view_543 = None
	        _assert_tensor_metadata_315 = torch.ops.aten._assert_tensor_metadata.default(view_545, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_315 = None
	        convert_element_type_209: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_545, torch.float32);  view_545 = None
	        sub_1602: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_208, convert_element_type_209);  convert_element_type_208 = convert_element_type_209 = None
	        mul_3394: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1602, view_544);  sub_1602 = view_544 = None
	        view_546: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3394, [5120, 1280]);  mul_3394 = None
	        _assert_tensor_metadata_316 = torch.ops.aten._assert_tensor_metadata.default(view_546, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_316 = None
	        mul_3399: "Sym(1500*s6)" = sym_size_int * 1500
	        view_547: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3389, [mul_3399, 1280]);  mul_3389 = mul_3399 = None
	        permute_59: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_546, [1, 0]);  view_546 = None
	        addmm_28: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_5_fc1_bias, view_547, permute_59);  model_audio_tower_layers_5_fc1_bias = view_547 = permute_59 = None
	        view_548: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_28, [sym_size_int, 1500, 5120]);  addmm_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_3406: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_548, 0.5)
	        mul_3407: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_548, 0.7071067811865476);  view_548 = None
	        erf_7: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_3407);  mul_3407 = None
	        add_5388: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_7, 1);  erf_7 = None
	        mul_3408: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3406, add_5388);  mul_3406 = add_5388 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_3408, [2])
	        amax_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_3408, [2])
	        full_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_35, full_70);  amin_35 = full_70 = None
	        full_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_35, full_71);  amax_35 = full_71 = None
	        sub_1615: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_35, minimum_35);  maximum_35 = None
	        div_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1615, 255.0);  sub_1615 = None
	        clamp_min_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_70, 1.1920928955078125e-07);  div_70 = None
	        div_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_35, clamp_min_105);  minimum_35 = None
	        round_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_71);  div_71 = None
	        sub_1621: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_71);  round_71 = None
	        clamp_min_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1621, -128);  sub_1621 = None
	        clamp_max_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_106, 127);  clamp_min_106 = None
	        _assert_tensor_metadata_317 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_105, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_317 = None
	        _assert_tensor_metadata_318 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_70, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_318 = None
	        convert_element_type_210: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_70, torch.int8);  clamp_max_70 = None
	        view_551: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_105, [sym_size_int, 1500, 1])
	        view_552: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_210, [sym_size_int, 1500, 1])
	        reciprocal_35: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_551);  view_551 = None
	        mul_3454: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_35, 1.0);  reciprocal_35 = None
	        mul_3457: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3408, mul_3454);  mul_3408 = mul_3454 = None
	        round_72: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_3457);  mul_3457 = None
	        add_5471: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_72, view_552);  round_72 = view_552 = None
	        clamp_min_107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5471, -128);  add_5471 = None
	        clamp_max_71: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_107, 127);  clamp_min_107 = None
	        _assert_tensor_metadata_319 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_71, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_319 = None
	        convert_element_type_211: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_71, torch.int8);  clamp_max_71 = None
	        view_555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_105, [sym_size_int, 1500, 1]);  clamp_min_105 = None
	        view_556: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_210, [sym_size_int, 1500, 1]);  convert_element_type_210 = None
	        _assert_tensor_metadata_320 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_211, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_320 = None
	        convert_element_type_212: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_211, torch.float32);  convert_element_type_211 = None
	        _assert_tensor_metadata_321 = torch.ops.aten._assert_tensor_metadata.default(view_556, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_321 = None
	        convert_element_type_213: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_556, torch.float32);  view_556 = None
	        sub_1641: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_212, convert_element_type_213);  convert_element_type_212 = convert_element_type_213 = None
	        mul_3479: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1641, view_555);  sub_1641 = view_555 = None
	        _assert_tensor_metadata_322 = torch.ops.aten._assert_tensor_metadata.default(mul_3479, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_322 = None
	        view_558: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_5_fc2_parametrizations_weight_original0 = None
	        view_559: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_5_fc2_parametrizations_weight_original1 = None
	        view_560: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_5_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_5_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_323 = torch.ops.aten._assert_tensor_metadata.default(view_558, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_323 = None
	        convert_element_type_214: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_558, torch.float32);  view_558 = None
	        _assert_tensor_metadata_324 = torch.ops.aten._assert_tensor_metadata.default(view_560, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_324 = None
	        convert_element_type_215: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_560, torch.float32);  view_560 = None
	        sub_1645: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_214, convert_element_type_215);  convert_element_type_214 = convert_element_type_215 = None
	        mul_3484: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1645, view_559);  sub_1645 = view_559 = None
	        view_561: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_3484, [1280, 5120]);  mul_3484 = None
	        _assert_tensor_metadata_325 = torch.ops.aten._assert_tensor_metadata.default(view_561, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_325 = None
	        mul_3489: "Sym(1500*s6)" = sym_size_int * 1500
	        view_562: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_3479, [mul_3489, 5120]);  mul_3479 = mul_3489 = None
	        permute_60: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_561, [1, 0]);  view_561 = None
	        addmm_29: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_5_fc2_bias, view_562, permute_60);  model_audio_tower_layers_5_fc2_bias = view_562 = permute_60 = None
	        view_563: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_29, [sym_size_int, 1500, 1280]);  addmm_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_5534: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_5236, view_563);  add_5236 = view_563 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_49: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_5534, memory_format = torch.contiguous_format)
	        var_mean_12 = torch.ops.aten.var_mean.correction(clone_49, [2], correction = 0, keepdim = True)
	        getitem_48: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_12[0]
	        getitem_49: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_12[1];  var_mean_12 = None
	        add_5539: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_48, 1e-05);  getitem_48 = None
	        rsqrt_12: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_5539);  add_5539 = None
	        sub_1651: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_49, getitem_49);  clone_49 = getitem_49 = None
	        mul_3500: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1651, rsqrt_12);  sub_1651 = rsqrt_12 = None
	        mul_3501: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3500, model_audio_tower_layers_6_self_attn_layer_norm_weight);  mul_3500 = model_audio_tower_layers_6_self_attn_layer_norm_weight = None
	        add_5540: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_3501, model_audio_tower_layers_6_self_attn_layer_norm_bias);  mul_3501 = model_audio_tower_layers_6_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_5540, [2])
	        amax_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_5540, [2])
	        full_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_36, full_72);  amin_36 = full_72 = None
	        full_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_36, full_73);  amax_36 = full_73 = None
	        sub_1662: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_36, minimum_36);  maximum_36 = None
	        div_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1662, 255.0);  sub_1662 = None
	        clamp_min_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_72, 1.1920928955078125e-07);  div_72 = None
	        div_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_36, clamp_min_108);  minimum_36 = None
	        round_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_73);  div_73 = None
	        sub_1668: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_73);  round_73 = None
	        clamp_min_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1668, -128);  sub_1668 = None
	        clamp_max_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_109, 127);  clamp_min_109 = None
	        _assert_tensor_metadata_326 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_108, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_326 = None
	        _assert_tensor_metadata_327 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_72, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_327 = None
	        convert_element_type_216: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_72, torch.int8);  clamp_max_72 = None
	        view_566: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_108, [sym_size_int, 1500, 1])
	        view_567: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_216, [sym_size_int, 1500, 1])
	        reciprocal_36: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_566);  view_566 = None
	        mul_3549: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_36, 1.0);  reciprocal_36 = None
	        mul_3552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_5540, mul_3549);  mul_3549 = None
	        round_74: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3552);  mul_3552 = None
	        add_5627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_74, view_567);  round_74 = view_567 = None
	        clamp_min_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5627, -128);  add_5627 = None
	        clamp_max_73: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_110, 127);  clamp_min_110 = None
	        _assert_tensor_metadata_328 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_73, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_328 = None
	        convert_element_type_217: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_73, torch.int8);  clamp_max_73 = None
	        view_570: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_108, [sym_size_int, 1500, 1]);  clamp_min_108 = None
	        view_571: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_216, [sym_size_int, 1500, 1]);  convert_element_type_216 = None
	        _assert_tensor_metadata_329 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_217, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_329 = None
	        convert_element_type_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_217, torch.float32);  convert_element_type_217 = None
	        _assert_tensor_metadata_330 = torch.ops.aten._assert_tensor_metadata.default(view_571, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_330 = None
	        convert_element_type_219: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_571, torch.float32);  view_571 = None
	        sub_1688: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_218, convert_element_type_219);  convert_element_type_218 = convert_element_type_219 = None
	        mul_3574: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1688, view_570);  sub_1688 = view_570 = None
	        _assert_tensor_metadata_331 = torch.ops.aten._assert_tensor_metadata.default(mul_3574, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_331 = None
	        view_573: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_574: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_575: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_332 = torch.ops.aten._assert_tensor_metadata.default(view_573, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_332 = None
	        convert_element_type_220: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_573, torch.float32);  view_573 = None
	        _assert_tensor_metadata_333 = torch.ops.aten._assert_tensor_metadata.default(view_575, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_333 = None
	        convert_element_type_221: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_575, torch.float32);  view_575 = None
	        sub_1692: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_220, convert_element_type_221);  convert_element_type_220 = convert_element_type_221 = None
	        mul_3579: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1692, view_574);  sub_1692 = view_574 = None
	        view_576: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3579, [1280, 1280]);  mul_3579 = None
	        _assert_tensor_metadata_334 = torch.ops.aten._assert_tensor_metadata.default(view_576, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_334 = None
	        mul_3584: "Sym(1500*s6)" = sym_size_int * 1500
	        view_577: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3574, [mul_3584, 1280]);  mul_3574 = mul_3584 = None
	        permute_61: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_576, [1, 0]);  view_576 = None
	        addmm_30: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_6_self_attn_q_proj_bias, view_577, permute_61);  model_audio_tower_layers_6_self_attn_q_proj_bias = view_577 = permute_61 = None
	        view_578: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_30, [sym_size_int, 1500, 1280]);  addmm_30 = None
	        mul_3591: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_578, 0.125);  view_578 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_579: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_3591, [sym_size_int, 1500, 20, 64]);  mul_3591 = None
	        permute_62: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_579, [0, 2, 1, 3]);  view_579 = None
	        clone_50: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_62, memory_format = torch.contiguous_format);  permute_62 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_5540, [2])
	        amax_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_5540, [2])
	        full_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_37, full_74);  amin_37 = full_74 = None
	        full_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_37, full_75);  amax_37 = full_75 = None
	        sub_1707: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_37, minimum_37);  maximum_37 = None
	        div_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1707, 255.0);  sub_1707 = None
	        clamp_min_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_74, 1.1920928955078125e-07);  div_74 = None
	        div_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_37, clamp_min_111);  minimum_37 = None
	        round_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_75);  div_75 = None
	        sub_1713: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_75);  round_75 = None
	        clamp_min_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1713, -128);  sub_1713 = None
	        clamp_max_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_112, 127);  clamp_min_112 = None
	        _assert_tensor_metadata_335 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_111, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_335 = None
	        _assert_tensor_metadata_336 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_74, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_336 = None
	        convert_element_type_222: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_74, torch.int8);  clamp_max_74 = None
	        view_582: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_111, [sym_size_int, 1500, 1])
	        view_583: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_222, [sym_size_int, 1500, 1])
	        reciprocal_37: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_582);  view_582 = None
	        mul_3645: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_37, 1.0);  reciprocal_37 = None
	        mul_3648: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_5540, mul_3645);  mul_3645 = None
	        round_76: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3648);  mul_3648 = None
	        add_5779: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_76, view_583);  round_76 = view_583 = None
	        clamp_min_113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5779, -128);  add_5779 = None
	        clamp_max_75: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_113, 127);  clamp_min_113 = None
	        _assert_tensor_metadata_337 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_75, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_337 = None
	        convert_element_type_223: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_75, torch.int8);  clamp_max_75 = None
	        view_586: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_111, [sym_size_int, 1500, 1]);  clamp_min_111 = None
	        view_587: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_222, [sym_size_int, 1500, 1]);  convert_element_type_222 = None
	        _assert_tensor_metadata_338 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_223, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_338 = None
	        convert_element_type_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_223, torch.float32);  convert_element_type_223 = None
	        _assert_tensor_metadata_339 = torch.ops.aten._assert_tensor_metadata.default(view_587, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_339 = None
	        convert_element_type_225: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_587, torch.float32);  view_587 = None
	        sub_1733: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_224, convert_element_type_225);  convert_element_type_224 = convert_element_type_225 = None
	        mul_3670: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1733, view_586);  sub_1733 = view_586 = None
	        _assert_tensor_metadata_340 = torch.ops.aten._assert_tensor_metadata.default(mul_3670, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_340 = None
	        view_589: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_590: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_591: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_341 = torch.ops.aten._assert_tensor_metadata.default(view_589, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_341 = None
	        convert_element_type_226: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_589, torch.float32);  view_589 = None
	        _assert_tensor_metadata_342 = torch.ops.aten._assert_tensor_metadata.default(view_591, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_342 = None
	        convert_element_type_227: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_591, torch.float32);  view_591 = None
	        sub_1737: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_226, convert_element_type_227);  convert_element_type_226 = convert_element_type_227 = None
	        mul_3675: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1737, view_590);  sub_1737 = view_590 = None
	        view_592: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3675, [1280, 1280]);  mul_3675 = None
	        _assert_tensor_metadata_343 = torch.ops.aten._assert_tensor_metadata.default(view_592, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_343 = None
	        permute_63: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_592, [1, 0]);  view_592 = None
	        mul_3678: "Sym(1500*s6)" = sym_size_int * 1500
	        view_593: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3670, [mul_3678, 1280]);  mul_3670 = mul_3678 = None
	        mm_6: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_593, permute_63);  view_593 = permute_63 = None
	        view_594: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_6, [sym_size_int, 1500, 1280]);  mm_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_595: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_594, [sym_size_int, -1, 20, 64]);  view_594 = None
	        permute_64: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_595, [0, 2, 1, 3]);  view_595 = None
	        clone_51: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_64, memory_format = torch.contiguous_format);  permute_64 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_5540, [2])
	        amax_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_5540, [2])
	        full_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_38, full_76);  amin_38 = full_76 = None
	        full_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_38, full_77);  amax_38 = full_77 = None
	        sub_1751: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_38, minimum_38);  maximum_38 = None
	        div_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1751, 255.0);  sub_1751 = None
	        clamp_min_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_76, 1.1920928955078125e-07);  div_76 = None
	        div_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_38, clamp_min_114);  minimum_38 = None
	        round_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_77);  div_77 = None
	        sub_1757: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_77);  round_77 = None
	        clamp_min_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1757, -128);  sub_1757 = None
	        clamp_max_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_115, 127);  clamp_min_115 = None
	        _assert_tensor_metadata_344 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_114, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_344 = None
	        _assert_tensor_metadata_345 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_76, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_345 = None
	        convert_element_type_228: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_76, torch.int8);  clamp_max_76 = None
	        view_598: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_114, [sym_size_int, 1500, 1])
	        view_599: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_228, [sym_size_int, 1500, 1])
	        reciprocal_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_598);  view_598 = None
	        mul_3744: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_38, 1.0);  reciprocal_38 = None
	        mul_3747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_5540, mul_3744);  add_5540 = mul_3744 = None
	        round_78: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3747);  mul_3747 = None
	        add_5927: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_78, view_599);  round_78 = view_599 = None
	        clamp_min_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5927, -128);  add_5927 = None
	        clamp_max_77: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_116, 127);  clamp_min_116 = None
	        _assert_tensor_metadata_346 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_77, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_346 = None
	        convert_element_type_229: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_77, torch.int8);  clamp_max_77 = None
	        view_602: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_114, [sym_size_int, 1500, 1]);  clamp_min_114 = None
	        view_603: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_228, [sym_size_int, 1500, 1]);  convert_element_type_228 = None
	        _assert_tensor_metadata_347 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_229, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_347 = None
	        convert_element_type_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_229, torch.float32);  convert_element_type_229 = None
	        _assert_tensor_metadata_348 = torch.ops.aten._assert_tensor_metadata.default(view_603, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_348 = None
	        convert_element_type_231: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_603, torch.float32);  view_603 = None
	        sub_1777: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_230, convert_element_type_231);  convert_element_type_230 = convert_element_type_231 = None
	        mul_3769: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1777, view_602);  sub_1777 = view_602 = None
	        _assert_tensor_metadata_349 = torch.ops.aten._assert_tensor_metadata.default(mul_3769, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_349 = None
	        view_605: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_606: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_607: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_350 = torch.ops.aten._assert_tensor_metadata.default(view_605, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_350 = None
	        convert_element_type_232: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_605, torch.float32);  view_605 = None
	        _assert_tensor_metadata_351 = torch.ops.aten._assert_tensor_metadata.default(view_607, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_351 = None
	        convert_element_type_233: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_607, torch.float32);  view_607 = None
	        sub_1781: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_232, convert_element_type_233);  convert_element_type_232 = convert_element_type_233 = None
	        mul_3774: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1781, view_606);  sub_1781 = view_606 = None
	        view_608: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3774, [1280, 1280]);  mul_3774 = None
	        _assert_tensor_metadata_352 = torch.ops.aten._assert_tensor_metadata.default(view_608, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_352 = None
	        mul_3779: "Sym(1500*s6)" = sym_size_int * 1500
	        view_609: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3769, [mul_3779, 1280]);  mul_3769 = mul_3779 = None
	        permute_65: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_608, [1, 0]);  view_608 = None
	        addmm_31: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_6_self_attn_v_proj_bias, view_609, permute_65);  model_audio_tower_layers_6_self_attn_v_proj_bias = view_609 = permute_65 = None
	        view_610: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_31, [sym_size_int, 1500, 1280]);  addmm_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_611: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_610, [sym_size_int, -1, 20, 64]);  view_610 = None
	        permute_66: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_611, [0, 2, 1, 3]);  view_611 = None
	        clone_52: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_66, memory_format = torch.contiguous_format);  permute_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_6 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_50, clone_51, clone_52, None, False, scale = 1.0);  clone_50 = clone_51 = clone_52 = None
	        getitem_50: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_6[0];  _scaled_dot_product_efficient_attention_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_67: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_50, [0, 2, 1, 3]);  getitem_50 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_67, [sym_size_int, 1500, -1]);  permute_67 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_612, [2])
	        amax_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_612, [2])
	        full_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_39, full_78);  amin_39 = full_78 = None
	        full_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_39, full_79);  amax_39 = full_79 = None
	        sub_1799: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_39, minimum_39);  maximum_39 = None
	        div_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1799, 255.0);  sub_1799 = None
	        clamp_min_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_78, 1.1920928955078125e-07);  div_78 = None
	        div_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_39, clamp_min_117);  minimum_39 = None
	        round_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_79);  div_79 = None
	        sub_1805: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_79);  round_79 = None
	        clamp_min_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1805, -128);  sub_1805 = None
	        clamp_max_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_118, 127);  clamp_min_118 = None
	        _assert_tensor_metadata_353 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_117, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_353 = None
	        _assert_tensor_metadata_354 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_78, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_354 = None
	        convert_element_type_234: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_78, torch.int8);  clamp_max_78 = None
	        view_615: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_117, [sym_size_int, 1500, 1])
	        view_616: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_234, [sym_size_int, 1500, 1])
	        reciprocal_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_615);  view_615 = None
	        mul_3849: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_39, 1.0);  reciprocal_39 = None
	        mul_3852: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_612, mul_3849);  view_612 = mul_3849 = None
	        round_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3852);  mul_3852 = None
	        add_6091: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_80, view_616);  round_80 = view_616 = None
	        clamp_min_119: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6091, -128);  add_6091 = None
	        clamp_max_79: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_119, 127);  clamp_min_119 = None
	        _assert_tensor_metadata_355 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_79, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_355 = None
	        convert_element_type_235: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_79, torch.int8);  clamp_max_79 = None
	        view_619: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_117, [sym_size_int, 1500, 1]);  clamp_min_117 = None
	        view_620: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_234, [sym_size_int, 1500, 1]);  convert_element_type_234 = None
	        _assert_tensor_metadata_356 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_235, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_356 = None
	        convert_element_type_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_235, torch.float32);  convert_element_type_235 = None
	        _assert_tensor_metadata_357 = torch.ops.aten._assert_tensor_metadata.default(view_620, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_357 = None
	        convert_element_type_237: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_620, torch.float32);  view_620 = None
	        sub_1825: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_236, convert_element_type_237);  convert_element_type_236 = convert_element_type_237 = None
	        mul_3874: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1825, view_619);  sub_1825 = view_619 = None
	        _assert_tensor_metadata_358 = torch.ops.aten._assert_tensor_metadata.default(mul_3874, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_358 = None
	        view_622: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_623: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_624: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_359 = torch.ops.aten._assert_tensor_metadata.default(view_622, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_359 = None
	        convert_element_type_238: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_622, torch.float32);  view_622 = None
	        _assert_tensor_metadata_360 = torch.ops.aten._assert_tensor_metadata.default(view_624, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_360 = None
	        convert_element_type_239: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_624, torch.float32);  view_624 = None
	        sub_1829: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_238, convert_element_type_239);  convert_element_type_238 = convert_element_type_239 = None
	        mul_3879: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1829, view_623);  sub_1829 = view_623 = None
	        view_625: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3879, [1280, 1280]);  mul_3879 = None
	        _assert_tensor_metadata_361 = torch.ops.aten._assert_tensor_metadata.default(view_625, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_361 = None
	        mul_3884: "Sym(1500*s6)" = sym_size_int * 1500
	        view_626: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3874, [mul_3884, 1280]);  mul_3874 = mul_3884 = None
	        permute_68: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_625, [1, 0]);  view_625 = None
	        addmm_32: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_6_self_attn_out_proj_bias, view_626, permute_68);  model_audio_tower_layers_6_self_attn_out_proj_bias = view_626 = permute_68 = None
	        view_627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_32, [sym_size_int, 1500, 1280]);  addmm_32 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_6154: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_5534, view_627);  add_5534 = view_627 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_54: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_6154, memory_format = torch.contiguous_format)
	        var_mean_13 = torch.ops.aten.var_mean.correction(clone_54, [2], correction = 0, keepdim = True)
	        getitem_54: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_13[0]
	        getitem_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_13[1];  var_mean_13 = None
	        add_6159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_54, 1e-05);  getitem_54 = None
	        rsqrt_13: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_6159);  add_6159 = None
	        sub_1835: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_54, getitem_55);  clone_54 = getitem_55 = None
	        mul_3895: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1835, rsqrt_13);  sub_1835 = rsqrt_13 = None
	        mul_3896: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3895, model_audio_tower_layers_6_final_layer_norm_weight);  mul_3895 = model_audio_tower_layers_6_final_layer_norm_weight = None
	        add_6160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_3896, model_audio_tower_layers_6_final_layer_norm_bias);  mul_3896 = model_audio_tower_layers_6_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_6160, [2])
	        amax_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_6160, [2])
	        full_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_40, full_80);  amin_40 = full_80 = None
	        full_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_40, full_81);  amax_40 = full_81 = None
	        sub_1846: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_40, minimum_40);  maximum_40 = None
	        div_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1846, 255.0);  sub_1846 = None
	        clamp_min_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_80, 1.1920928955078125e-07);  div_80 = None
	        div_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_40, clamp_min_120);  minimum_40 = None
	        round_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_81);  div_81 = None
	        sub_1852: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_81);  round_81 = None
	        clamp_min_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1852, -128);  sub_1852 = None
	        clamp_max_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_121, 127);  clamp_min_121 = None
	        _assert_tensor_metadata_362 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_120, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_362 = None
	        _assert_tensor_metadata_363 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_80, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_363 = None
	        convert_element_type_240: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_80, torch.int8);  clamp_max_80 = None
	        view_630: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_120, [sym_size_int, 1500, 1])
	        view_631: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_240, [sym_size_int, 1500, 1])
	        reciprocal_40: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_630);  view_630 = None
	        mul_3944: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_40, 1.0);  reciprocal_40 = None
	        mul_3947: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_6160, mul_3944);  add_6160 = mul_3944 = None
	        round_82: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3947);  mul_3947 = None
	        add_6247: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_82, view_631);  round_82 = view_631 = None
	        clamp_min_122: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6247, -128);  add_6247 = None
	        clamp_max_81: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_122, 127);  clamp_min_122 = None
	        _assert_tensor_metadata_364 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_81, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_364 = None
	        convert_element_type_241: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_81, torch.int8);  clamp_max_81 = None
	        view_634: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_120, [sym_size_int, 1500, 1]);  clamp_min_120 = None
	        view_635: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_240, [sym_size_int, 1500, 1]);  convert_element_type_240 = None
	        _assert_tensor_metadata_365 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_241, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_365 = None
	        convert_element_type_242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_241, torch.float32);  convert_element_type_241 = None
	        _assert_tensor_metadata_366 = torch.ops.aten._assert_tensor_metadata.default(view_635, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_366 = None
	        convert_element_type_243: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_635, torch.float32);  view_635 = None
	        sub_1872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_242, convert_element_type_243);  convert_element_type_242 = convert_element_type_243 = None
	        mul_3969: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1872, view_634);  sub_1872 = view_634 = None
	        _assert_tensor_metadata_367 = torch.ops.aten._assert_tensor_metadata.default(mul_3969, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_367 = None
	        view_637: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_6_fc1_parametrizations_weight_original0 = None
	        view_638: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_6_fc1_parametrizations_weight_original1 = None
	        view_639: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_6_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_368 = torch.ops.aten._assert_tensor_metadata.default(view_637, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_368 = None
	        convert_element_type_244: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_637, torch.float32);  view_637 = None
	        _assert_tensor_metadata_369 = torch.ops.aten._assert_tensor_metadata.default(view_639, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_369 = None
	        convert_element_type_245: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_639, torch.float32);  view_639 = None
	        sub_1876: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_244, convert_element_type_245);  convert_element_type_244 = convert_element_type_245 = None
	        mul_3974: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1876, view_638);  sub_1876 = view_638 = None
	        view_640: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3974, [5120, 1280]);  mul_3974 = None
	        _assert_tensor_metadata_370 = torch.ops.aten._assert_tensor_metadata.default(view_640, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_370 = None
	        mul_3979: "Sym(1500*s6)" = sym_size_int * 1500
	        view_641: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_3969, [mul_3979, 1280]);  mul_3969 = mul_3979 = None
	        permute_69: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_640, [1, 0]);  view_640 = None
	        addmm_33: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_6_fc1_bias, view_641, permute_69);  model_audio_tower_layers_6_fc1_bias = view_641 = permute_69 = None
	        view_642: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_33, [sym_size_int, 1500, 5120]);  addmm_33 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_3986: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_642, 0.5)
	        mul_3987: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_642, 0.7071067811865476);  view_642 = None
	        erf_8: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_3987);  mul_3987 = None
	        add_6306: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_8, 1);  erf_8 = None
	        mul_3988: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3986, add_6306);  mul_3986 = add_6306 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_3988, [2])
	        amax_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_3988, [2])
	        full_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_41, full_82);  amin_41 = full_82 = None
	        full_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_41, full_83);  amax_41 = full_83 = None
	        sub_1889: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_41, minimum_41);  maximum_41 = None
	        div_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1889, 255.0);  sub_1889 = None
	        clamp_min_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_82, 1.1920928955078125e-07);  div_82 = None
	        div_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_41, clamp_min_123);  minimum_41 = None
	        round_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_83);  div_83 = None
	        sub_1895: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_83);  round_83 = None
	        clamp_min_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1895, -128);  sub_1895 = None
	        clamp_max_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_124, 127);  clamp_min_124 = None
	        _assert_tensor_metadata_371 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_123, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_371 = None
	        _assert_tensor_metadata_372 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_82, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_372 = None
	        convert_element_type_246: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_82, torch.int8);  clamp_max_82 = None
	        view_645: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_123, [sym_size_int, 1500, 1])
	        view_646: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_246, [sym_size_int, 1500, 1])
	        reciprocal_41: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_645);  view_645 = None
	        mul_4034: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_41, 1.0);  reciprocal_41 = None
	        mul_4037: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3988, mul_4034);  mul_3988 = mul_4034 = None
	        round_84: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_4037);  mul_4037 = None
	        add_6389: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_84, view_646);  round_84 = view_646 = None
	        clamp_min_125: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6389, -128);  add_6389 = None
	        clamp_max_83: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_125, 127);  clamp_min_125 = None
	        _assert_tensor_metadata_373 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_83, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_373 = None
	        convert_element_type_247: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_83, torch.int8);  clamp_max_83 = None
	        view_649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_123, [sym_size_int, 1500, 1]);  clamp_min_123 = None
	        view_650: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_246, [sym_size_int, 1500, 1]);  convert_element_type_246 = None
	        _assert_tensor_metadata_374 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_247, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_374 = None
	        convert_element_type_248: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_247, torch.float32);  convert_element_type_247 = None
	        _assert_tensor_metadata_375 = torch.ops.aten._assert_tensor_metadata.default(view_650, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_375 = None
	        convert_element_type_249: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_650, torch.float32);  view_650 = None
	        sub_1915: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_248, convert_element_type_249);  convert_element_type_248 = convert_element_type_249 = None
	        mul_4059: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1915, view_649);  sub_1915 = view_649 = None
	        _assert_tensor_metadata_376 = torch.ops.aten._assert_tensor_metadata.default(mul_4059, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_376 = None
	        view_652: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_6_fc2_parametrizations_weight_original0 = None
	        view_653: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_6_fc2_parametrizations_weight_original1 = None
	        view_654: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_6_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_6_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_377 = torch.ops.aten._assert_tensor_metadata.default(view_652, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_377 = None
	        convert_element_type_250: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_652, torch.float32);  view_652 = None
	        _assert_tensor_metadata_378 = torch.ops.aten._assert_tensor_metadata.default(view_654, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_378 = None
	        convert_element_type_251: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_654, torch.float32);  view_654 = None
	        sub_1919: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_250, convert_element_type_251);  convert_element_type_250 = convert_element_type_251 = None
	        mul_4064: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1919, view_653);  sub_1919 = view_653 = None
	        view_655: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_4064, [1280, 5120]);  mul_4064 = None
	        _assert_tensor_metadata_379 = torch.ops.aten._assert_tensor_metadata.default(view_655, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_379 = None
	        mul_4069: "Sym(1500*s6)" = sym_size_int * 1500
	        view_656: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_4059, [mul_4069, 5120]);  mul_4059 = mul_4069 = None
	        permute_70: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_655, [1, 0]);  view_655 = None
	        addmm_34: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_6_fc2_bias, view_656, permute_70);  model_audio_tower_layers_6_fc2_bias = view_656 = permute_70 = None
	        view_657: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_34, [sym_size_int, 1500, 1280]);  addmm_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_6452: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_6154, view_657);  add_6154 = view_657 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_57: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_6452, memory_format = torch.contiguous_format)
	        var_mean_14 = torch.ops.aten.var_mean.correction(clone_57, [2], correction = 0, keepdim = True)
	        getitem_56: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_14[0]
	        getitem_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_14[1];  var_mean_14 = None
	        add_6457: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_56, 1e-05);  getitem_56 = None
	        rsqrt_14: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_6457);  add_6457 = None
	        sub_1925: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_57, getitem_57);  clone_57 = getitem_57 = None
	        mul_4080: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1925, rsqrt_14);  sub_1925 = rsqrt_14 = None
	        mul_4081: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4080, model_audio_tower_layers_7_self_attn_layer_norm_weight);  mul_4080 = model_audio_tower_layers_7_self_attn_layer_norm_weight = None
	        add_6458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_4081, model_audio_tower_layers_7_self_attn_layer_norm_bias);  mul_4081 = model_audio_tower_layers_7_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_6458, [2])
	        amax_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_6458, [2])
	        full_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_42, full_84);  amin_42 = full_84 = None
	        full_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_42, full_85);  amax_42 = full_85 = None
	        sub_1936: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_42, minimum_42);  maximum_42 = None
	        div_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1936, 255.0);  sub_1936 = None
	        clamp_min_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_84, 1.1920928955078125e-07);  div_84 = None
	        div_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_42, clamp_min_126);  minimum_42 = None
	        round_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_85);  div_85 = None
	        sub_1942: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_85);  round_85 = None
	        clamp_min_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1942, -128);  sub_1942 = None
	        clamp_max_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_127, 127);  clamp_min_127 = None
	        _assert_tensor_metadata_380 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_126, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_380 = None
	        _assert_tensor_metadata_381 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_84, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_381 = None
	        convert_element_type_252: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_84, torch.int8);  clamp_max_84 = None
	        view_660: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_126, [sym_size_int, 1500, 1])
	        view_661: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_252, [sym_size_int, 1500, 1])
	        reciprocal_42: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_660);  view_660 = None
	        mul_4129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_42, 1.0);  reciprocal_42 = None
	        mul_4132: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_6458, mul_4129);  mul_4129 = None
	        round_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4132);  mul_4132 = None
	        add_6545: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_86, view_661);  round_86 = view_661 = None
	        clamp_min_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6545, -128);  add_6545 = None
	        clamp_max_85: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_128, 127);  clamp_min_128 = None
	        _assert_tensor_metadata_382 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_85, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_382 = None
	        convert_element_type_253: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_85, torch.int8);  clamp_max_85 = None
	        view_664: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_126, [sym_size_int, 1500, 1]);  clamp_min_126 = None
	        view_665: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_252, [sym_size_int, 1500, 1]);  convert_element_type_252 = None
	        _assert_tensor_metadata_383 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_253, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_383 = None
	        convert_element_type_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_253, torch.float32);  convert_element_type_253 = None
	        _assert_tensor_metadata_384 = torch.ops.aten._assert_tensor_metadata.default(view_665, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_384 = None
	        convert_element_type_255: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_665, torch.float32);  view_665 = None
	        sub_1962: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_254, convert_element_type_255);  convert_element_type_254 = convert_element_type_255 = None
	        mul_4154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1962, view_664);  sub_1962 = view_664 = None
	        _assert_tensor_metadata_385 = torch.ops.aten._assert_tensor_metadata.default(mul_4154, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_385 = None
	        view_667: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_668: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_669: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_386 = torch.ops.aten._assert_tensor_metadata.default(view_667, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_386 = None
	        convert_element_type_256: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_667, torch.float32);  view_667 = None
	        _assert_tensor_metadata_387 = torch.ops.aten._assert_tensor_metadata.default(view_669, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_387 = None
	        convert_element_type_257: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_669, torch.float32);  view_669 = None
	        sub_1966: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_256, convert_element_type_257);  convert_element_type_256 = convert_element_type_257 = None
	        mul_4159: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1966, view_668);  sub_1966 = view_668 = None
	        view_670: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4159, [1280, 1280]);  mul_4159 = None
	        _assert_tensor_metadata_388 = torch.ops.aten._assert_tensor_metadata.default(view_670, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_388 = None
	        mul_4164: "Sym(1500*s6)" = sym_size_int * 1500
	        view_671: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4154, [mul_4164, 1280]);  mul_4154 = mul_4164 = None
	        permute_71: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_670, [1, 0]);  view_670 = None
	        addmm_35: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_7_self_attn_q_proj_bias, view_671, permute_71);  model_audio_tower_layers_7_self_attn_q_proj_bias = view_671 = permute_71 = None
	        view_672: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_35, [sym_size_int, 1500, 1280]);  addmm_35 = None
	        mul_4171: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_672, 0.125);  view_672 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_673: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_4171, [sym_size_int, 1500, 20, 64]);  mul_4171 = None
	        permute_72: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_673, [0, 2, 1, 3]);  view_673 = None
	        clone_58: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_72, memory_format = torch.contiguous_format);  permute_72 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_6458, [2])
	        amax_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_6458, [2])
	        full_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_43, full_86);  amin_43 = full_86 = None
	        full_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_43, full_87);  amax_43 = full_87 = None
	        sub_1981: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_43, minimum_43);  maximum_43 = None
	        div_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1981, 255.0);  sub_1981 = None
	        clamp_min_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_86, 1.1920928955078125e-07);  div_86 = None
	        div_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_43, clamp_min_129);  minimum_43 = None
	        round_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_87);  div_87 = None
	        sub_1987: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_87);  round_87 = None
	        clamp_min_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1987, -128);  sub_1987 = None
	        clamp_max_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_130, 127);  clamp_min_130 = None
	        _assert_tensor_metadata_389 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_129, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_389 = None
	        _assert_tensor_metadata_390 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_86, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_390 = None
	        convert_element_type_258: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_86, torch.int8);  clamp_max_86 = None
	        view_676: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_129, [sym_size_int, 1500, 1])
	        view_677: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_258, [sym_size_int, 1500, 1])
	        reciprocal_43: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_676);  view_676 = None
	        mul_4225: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_43, 1.0);  reciprocal_43 = None
	        mul_4228: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_6458, mul_4225);  mul_4225 = None
	        round_88: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4228);  mul_4228 = None
	        add_6697: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_88, view_677);  round_88 = view_677 = None
	        clamp_min_131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6697, -128);  add_6697 = None
	        clamp_max_87: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_131, 127);  clamp_min_131 = None
	        _assert_tensor_metadata_391 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_87, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_391 = None
	        convert_element_type_259: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_87, torch.int8);  clamp_max_87 = None
	        view_680: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_129, [sym_size_int, 1500, 1]);  clamp_min_129 = None
	        view_681: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_258, [sym_size_int, 1500, 1]);  convert_element_type_258 = None
	        _assert_tensor_metadata_392 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_259, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_392 = None
	        convert_element_type_260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_259, torch.float32);  convert_element_type_259 = None
	        _assert_tensor_metadata_393 = torch.ops.aten._assert_tensor_metadata.default(view_681, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_393 = None
	        convert_element_type_261: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_681, torch.float32);  view_681 = None
	        sub_2007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_260, convert_element_type_261);  convert_element_type_260 = convert_element_type_261 = None
	        mul_4250: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2007, view_680);  sub_2007 = view_680 = None
	        _assert_tensor_metadata_394 = torch.ops.aten._assert_tensor_metadata.default(mul_4250, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_394 = None
	        view_683: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_684: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_685: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_395 = torch.ops.aten._assert_tensor_metadata.default(view_683, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_395 = None
	        convert_element_type_262: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_683, torch.float32);  view_683 = None
	        _assert_tensor_metadata_396 = torch.ops.aten._assert_tensor_metadata.default(view_685, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_396 = None
	        convert_element_type_263: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_685, torch.float32);  view_685 = None
	        sub_2011: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_262, convert_element_type_263);  convert_element_type_262 = convert_element_type_263 = None
	        mul_4255: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2011, view_684);  sub_2011 = view_684 = None
	        view_686: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4255, [1280, 1280]);  mul_4255 = None
	        _assert_tensor_metadata_397 = torch.ops.aten._assert_tensor_metadata.default(view_686, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_397 = None
	        permute_73: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_686, [1, 0]);  view_686 = None
	        mul_4258: "Sym(1500*s6)" = sym_size_int * 1500
	        view_687: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4250, [mul_4258, 1280]);  mul_4250 = mul_4258 = None
	        mm_7: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_687, permute_73);  view_687 = permute_73 = None
	        view_688: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_7, [sym_size_int, 1500, 1280]);  mm_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_689: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_688, [sym_size_int, -1, 20, 64]);  view_688 = None
	        permute_74: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_689, [0, 2, 1, 3]);  view_689 = None
	        clone_59: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_74, memory_format = torch.contiguous_format);  permute_74 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_6458, [2])
	        amax_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_6458, [2])
	        full_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_44, full_88);  amin_44 = full_88 = None
	        full_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_44, full_89);  amax_44 = full_89 = None
	        sub_2025: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_44, minimum_44);  maximum_44 = None
	        div_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2025, 255.0);  sub_2025 = None
	        clamp_min_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_88, 1.1920928955078125e-07);  div_88 = None
	        div_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_44, clamp_min_132);  minimum_44 = None
	        round_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_89);  div_89 = None
	        sub_2031: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_89);  round_89 = None
	        clamp_min_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2031, -128);  sub_2031 = None
	        clamp_max_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_133, 127);  clamp_min_133 = None
	        _assert_tensor_metadata_398 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_132, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_398 = None
	        _assert_tensor_metadata_399 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_88, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_399 = None
	        convert_element_type_264: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_88, torch.int8);  clamp_max_88 = None
	        view_692: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_132, [sym_size_int, 1500, 1])
	        view_693: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_264, [sym_size_int, 1500, 1])
	        reciprocal_44: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_692);  view_692 = None
	        mul_4324: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_44, 1.0);  reciprocal_44 = None
	        mul_4327: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_6458, mul_4324);  add_6458 = mul_4324 = None
	        round_90: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4327);  mul_4327 = None
	        add_6845: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_90, view_693);  round_90 = view_693 = None
	        clamp_min_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6845, -128);  add_6845 = None
	        clamp_max_89: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_134, 127);  clamp_min_134 = None
	        _assert_tensor_metadata_400 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_89, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_400 = None
	        convert_element_type_265: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_89, torch.int8);  clamp_max_89 = None
	        view_696: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_132, [sym_size_int, 1500, 1]);  clamp_min_132 = None
	        view_697: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_264, [sym_size_int, 1500, 1]);  convert_element_type_264 = None
	        _assert_tensor_metadata_401 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_265, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_401 = None
	        convert_element_type_266: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_265, torch.float32);  convert_element_type_265 = None
	        _assert_tensor_metadata_402 = torch.ops.aten._assert_tensor_metadata.default(view_697, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_402 = None
	        convert_element_type_267: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_697, torch.float32);  view_697 = None
	        sub_2051: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_266, convert_element_type_267);  convert_element_type_266 = convert_element_type_267 = None
	        mul_4349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2051, view_696);  sub_2051 = view_696 = None
	        _assert_tensor_metadata_403 = torch.ops.aten._assert_tensor_metadata.default(mul_4349, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_403 = None
	        view_699: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_700: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_701: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_404 = torch.ops.aten._assert_tensor_metadata.default(view_699, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_404 = None
	        convert_element_type_268: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_699, torch.float32);  view_699 = None
	        _assert_tensor_metadata_405 = torch.ops.aten._assert_tensor_metadata.default(view_701, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_405 = None
	        convert_element_type_269: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_701, torch.float32);  view_701 = None
	        sub_2055: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_268, convert_element_type_269);  convert_element_type_268 = convert_element_type_269 = None
	        mul_4354: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2055, view_700);  sub_2055 = view_700 = None
	        view_702: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4354, [1280, 1280]);  mul_4354 = None
	        _assert_tensor_metadata_406 = torch.ops.aten._assert_tensor_metadata.default(view_702, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_406 = None
	        mul_4359: "Sym(1500*s6)" = sym_size_int * 1500
	        view_703: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4349, [mul_4359, 1280]);  mul_4349 = mul_4359 = None
	        permute_75: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_702, [1, 0]);  view_702 = None
	        addmm_36: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_7_self_attn_v_proj_bias, view_703, permute_75);  model_audio_tower_layers_7_self_attn_v_proj_bias = view_703 = permute_75 = None
	        view_704: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_36, [sym_size_int, 1500, 1280]);  addmm_36 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_705: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_704, [sym_size_int, -1, 20, 64]);  view_704 = None
	        permute_76: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_705, [0, 2, 1, 3]);  view_705 = None
	        clone_60: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_76, memory_format = torch.contiguous_format);  permute_76 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_7 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_58, clone_59, clone_60, None, False, scale = 1.0);  clone_58 = clone_59 = clone_60 = None
	        getitem_58: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_7[0];  _scaled_dot_product_efficient_attention_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_77: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_58, [0, 2, 1, 3]);  getitem_58 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_706: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_77, [sym_size_int, 1500, -1]);  permute_77 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_706, [2])
	        amax_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_706, [2])
	        full_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_45, full_90);  amin_45 = full_90 = None
	        full_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_45, full_91);  amax_45 = full_91 = None
	        sub_2073: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_45, minimum_45);  maximum_45 = None
	        div_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2073, 255.0);  sub_2073 = None
	        clamp_min_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_90, 1.1920928955078125e-07);  div_90 = None
	        div_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_45, clamp_min_135);  minimum_45 = None
	        round_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_91);  div_91 = None
	        sub_2079: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_91);  round_91 = None
	        clamp_min_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2079, -128);  sub_2079 = None
	        clamp_max_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_136, 127);  clamp_min_136 = None
	        _assert_tensor_metadata_407 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_135, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_407 = None
	        _assert_tensor_metadata_408 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_90, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_408 = None
	        convert_element_type_270: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_90, torch.int8);  clamp_max_90 = None
	        view_709: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_135, [sym_size_int, 1500, 1])
	        view_710: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_270, [sym_size_int, 1500, 1])
	        reciprocal_45: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_709);  view_709 = None
	        mul_4429: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_45, 1.0);  reciprocal_45 = None
	        mul_4432: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_706, mul_4429);  view_706 = mul_4429 = None
	        round_92: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4432);  mul_4432 = None
	        add_7009: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_92, view_710);  round_92 = view_710 = None
	        clamp_min_137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7009, -128);  add_7009 = None
	        clamp_max_91: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_137, 127);  clamp_min_137 = None
	        _assert_tensor_metadata_409 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_91, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_409 = None
	        convert_element_type_271: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_91, torch.int8);  clamp_max_91 = None
	        view_713: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_135, [sym_size_int, 1500, 1]);  clamp_min_135 = None
	        view_714: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_270, [sym_size_int, 1500, 1]);  convert_element_type_270 = None
	        _assert_tensor_metadata_410 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_271, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_410 = None
	        convert_element_type_272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_271, torch.float32);  convert_element_type_271 = None
	        _assert_tensor_metadata_411 = torch.ops.aten._assert_tensor_metadata.default(view_714, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_411 = None
	        convert_element_type_273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_714, torch.float32);  view_714 = None
	        sub_2099: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_272, convert_element_type_273);  convert_element_type_272 = convert_element_type_273 = None
	        mul_4454: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2099, view_713);  sub_2099 = view_713 = None
	        _assert_tensor_metadata_412 = torch.ops.aten._assert_tensor_metadata.default(mul_4454, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_412 = None
	        view_716: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_717: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_718: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_413 = torch.ops.aten._assert_tensor_metadata.default(view_716, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_413 = None
	        convert_element_type_274: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_716, torch.float32);  view_716 = None
	        _assert_tensor_metadata_414 = torch.ops.aten._assert_tensor_metadata.default(view_718, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_414 = None
	        convert_element_type_275: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_718, torch.float32);  view_718 = None
	        sub_2103: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_274, convert_element_type_275);  convert_element_type_274 = convert_element_type_275 = None
	        mul_4459: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2103, view_717);  sub_2103 = view_717 = None
	        view_719: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4459, [1280, 1280]);  mul_4459 = None
	        _assert_tensor_metadata_415 = torch.ops.aten._assert_tensor_metadata.default(view_719, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_415 = None
	        mul_4464: "Sym(1500*s6)" = sym_size_int * 1500
	        view_720: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4454, [mul_4464, 1280]);  mul_4454 = mul_4464 = None
	        permute_78: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_719, [1, 0]);  view_719 = None
	        addmm_37: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_7_self_attn_out_proj_bias, view_720, permute_78);  model_audio_tower_layers_7_self_attn_out_proj_bias = view_720 = permute_78 = None
	        view_721: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_37, [sym_size_int, 1500, 1280]);  addmm_37 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_7072: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_6452, view_721);  add_6452 = view_721 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_7072, memory_format = torch.contiguous_format)
	        var_mean_15 = torch.ops.aten.var_mean.correction(clone_62, [2], correction = 0, keepdim = True)
	        getitem_62: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_15[0]
	        getitem_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_15[1];  var_mean_15 = None
	        add_7077: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_62, 1e-05);  getitem_62 = None
	        rsqrt_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_7077);  add_7077 = None
	        sub_2109: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_62, getitem_63);  clone_62 = getitem_63 = None
	        mul_4475: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2109, rsqrt_15);  sub_2109 = rsqrt_15 = None
	        mul_4476: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4475, model_audio_tower_layers_7_final_layer_norm_weight);  mul_4475 = model_audio_tower_layers_7_final_layer_norm_weight = None
	        add_7078: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_4476, model_audio_tower_layers_7_final_layer_norm_bias);  mul_4476 = model_audio_tower_layers_7_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_7078, [2])
	        amax_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_7078, [2])
	        full_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_46, full_92);  amin_46 = full_92 = None
	        full_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_46, full_93);  amax_46 = full_93 = None
	        sub_2120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_46, minimum_46);  maximum_46 = None
	        div_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2120, 255.0);  sub_2120 = None
	        clamp_min_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_92, 1.1920928955078125e-07);  div_92 = None
	        div_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_46, clamp_min_138);  minimum_46 = None
	        round_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_93);  div_93 = None
	        sub_2126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_93);  round_93 = None
	        clamp_min_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2126, -128);  sub_2126 = None
	        clamp_max_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_139, 127);  clamp_min_139 = None
	        _assert_tensor_metadata_416 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_138, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_416 = None
	        _assert_tensor_metadata_417 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_92, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_417 = None
	        convert_element_type_276: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_92, torch.int8);  clamp_max_92 = None
	        view_724: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_138, [sym_size_int, 1500, 1])
	        view_725: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_276, [sym_size_int, 1500, 1])
	        reciprocal_46: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_724);  view_724 = None
	        mul_4524: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_46, 1.0);  reciprocal_46 = None
	        mul_4527: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_7078, mul_4524);  add_7078 = mul_4524 = None
	        round_94: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4527);  mul_4527 = None
	        add_7165: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_94, view_725);  round_94 = view_725 = None
	        clamp_min_140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7165, -128);  add_7165 = None
	        clamp_max_93: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_140, 127);  clamp_min_140 = None
	        _assert_tensor_metadata_418 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_93, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_418 = None
	        convert_element_type_277: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_93, torch.int8);  clamp_max_93 = None
	        view_728: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_138, [sym_size_int, 1500, 1]);  clamp_min_138 = None
	        view_729: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_276, [sym_size_int, 1500, 1]);  convert_element_type_276 = None
	        _assert_tensor_metadata_419 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_277, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_419 = None
	        convert_element_type_278: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_277, torch.float32);  convert_element_type_277 = None
	        _assert_tensor_metadata_420 = torch.ops.aten._assert_tensor_metadata.default(view_729, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_420 = None
	        convert_element_type_279: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_729, torch.float32);  view_729 = None
	        sub_2146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_278, convert_element_type_279);  convert_element_type_278 = convert_element_type_279 = None
	        mul_4549: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2146, view_728);  sub_2146 = view_728 = None
	        _assert_tensor_metadata_421 = torch.ops.aten._assert_tensor_metadata.default(mul_4549, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_421 = None
	        view_731: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_7_fc1_parametrizations_weight_original0 = None
	        view_732: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_7_fc1_parametrizations_weight_original1 = None
	        view_733: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_7_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_422 = torch.ops.aten._assert_tensor_metadata.default(view_731, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_422 = None
	        convert_element_type_280: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_731, torch.float32);  view_731 = None
	        _assert_tensor_metadata_423 = torch.ops.aten._assert_tensor_metadata.default(view_733, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_423 = None
	        convert_element_type_281: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_733, torch.float32);  view_733 = None
	        sub_2150: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_280, convert_element_type_281);  convert_element_type_280 = convert_element_type_281 = None
	        mul_4554: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2150, view_732);  sub_2150 = view_732 = None
	        view_734: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4554, [5120, 1280]);  mul_4554 = None
	        _assert_tensor_metadata_424 = torch.ops.aten._assert_tensor_metadata.default(view_734, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_424 = None
	        mul_4559: "Sym(1500*s6)" = sym_size_int * 1500
	        view_735: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4549, [mul_4559, 1280]);  mul_4549 = mul_4559 = None
	        permute_79: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_734, [1, 0]);  view_734 = None
	        addmm_38: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_7_fc1_bias, view_735, permute_79);  model_audio_tower_layers_7_fc1_bias = view_735 = permute_79 = None
	        view_736: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_38, [sym_size_int, 1500, 5120]);  addmm_38 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_4566: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_736, 0.5)
	        mul_4567: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_736, 0.7071067811865476);  view_736 = None
	        erf_9: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_4567);  mul_4567 = None
	        add_7224: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_9, 1);  erf_9 = None
	        mul_4568: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4566, add_7224);  mul_4566 = add_7224 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_4568, [2])
	        amax_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_4568, [2])
	        full_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_47, full_94);  amin_47 = full_94 = None
	        full_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_47, full_95);  amax_47 = full_95 = None
	        sub_2163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_47, minimum_47);  maximum_47 = None
	        div_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2163, 255.0);  sub_2163 = None
	        clamp_min_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_94, 1.1920928955078125e-07);  div_94 = None
	        div_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_47, clamp_min_141);  minimum_47 = None
	        round_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_95);  div_95 = None
	        sub_2169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_95);  round_95 = None
	        clamp_min_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2169, -128);  sub_2169 = None
	        clamp_max_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_142, 127);  clamp_min_142 = None
	        _assert_tensor_metadata_425 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_141, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_425 = None
	        _assert_tensor_metadata_426 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_94, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_426 = None
	        convert_element_type_282: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_94, torch.int8);  clamp_max_94 = None
	        view_739: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_141, [sym_size_int, 1500, 1])
	        view_740: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_282, [sym_size_int, 1500, 1])
	        reciprocal_47: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_739);  view_739 = None
	        mul_4614: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_47, 1.0);  reciprocal_47 = None
	        mul_4617: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4568, mul_4614);  mul_4568 = mul_4614 = None
	        round_96: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_4617);  mul_4617 = None
	        add_7307: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_96, view_740);  round_96 = view_740 = None
	        clamp_min_143: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7307, -128);  add_7307 = None
	        clamp_max_95: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_143, 127);  clamp_min_143 = None
	        _assert_tensor_metadata_427 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_95, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_427 = None
	        convert_element_type_283: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_95, torch.int8);  clamp_max_95 = None
	        view_743: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_141, [sym_size_int, 1500, 1]);  clamp_min_141 = None
	        view_744: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_282, [sym_size_int, 1500, 1]);  convert_element_type_282 = None
	        _assert_tensor_metadata_428 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_283, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_428 = None
	        convert_element_type_284: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_283, torch.float32);  convert_element_type_283 = None
	        _assert_tensor_metadata_429 = torch.ops.aten._assert_tensor_metadata.default(view_744, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_429 = None
	        convert_element_type_285: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_744, torch.float32);  view_744 = None
	        sub_2189: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_284, convert_element_type_285);  convert_element_type_284 = convert_element_type_285 = None
	        mul_4639: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2189, view_743);  sub_2189 = view_743 = None
	        _assert_tensor_metadata_430 = torch.ops.aten._assert_tensor_metadata.default(mul_4639, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_430 = None
	        view_746: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_7_fc2_parametrizations_weight_original0 = None
	        view_747: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_7_fc2_parametrizations_weight_original1 = None
	        view_748: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_7_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_7_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_431 = torch.ops.aten._assert_tensor_metadata.default(view_746, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_431 = None
	        convert_element_type_286: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_746, torch.float32);  view_746 = None
	        _assert_tensor_metadata_432 = torch.ops.aten._assert_tensor_metadata.default(view_748, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_432 = None
	        convert_element_type_287: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_748, torch.float32);  view_748 = None
	        sub_2193: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_286, convert_element_type_287);  convert_element_type_286 = convert_element_type_287 = None
	        mul_4644: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2193, view_747);  sub_2193 = view_747 = None
	        view_749: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_4644, [1280, 5120]);  mul_4644 = None
	        _assert_tensor_metadata_433 = torch.ops.aten._assert_tensor_metadata.default(view_749, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_433 = None
	        mul_4649: "Sym(1500*s6)" = sym_size_int * 1500
	        view_750: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_4639, [mul_4649, 5120]);  mul_4639 = mul_4649 = None
	        permute_80: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_749, [1, 0]);  view_749 = None
	        addmm_39: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_7_fc2_bias, view_750, permute_80);  model_audio_tower_layers_7_fc2_bias = view_750 = permute_80 = None
	        view_751: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_39, [sym_size_int, 1500, 1280]);  addmm_39 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_7370: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_7072, view_751);  add_7072 = view_751 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_65: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_7370, memory_format = torch.contiguous_format)
	        var_mean_16 = torch.ops.aten.var_mean.correction(clone_65, [2], correction = 0, keepdim = True)
	        getitem_64: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_16[0]
	        getitem_65: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_16[1];  var_mean_16 = None
	        add_7375: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_64, 1e-05);  getitem_64 = None
	        rsqrt_16: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_7375);  add_7375 = None
	        sub_2199: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_65, getitem_65);  clone_65 = getitem_65 = None
	        mul_4660: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2199, rsqrt_16);  sub_2199 = rsqrt_16 = None
	        mul_4661: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4660, model_audio_tower_layers_8_self_attn_layer_norm_weight);  mul_4660 = model_audio_tower_layers_8_self_attn_layer_norm_weight = None
	        add_7376: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_4661, model_audio_tower_layers_8_self_attn_layer_norm_bias);  mul_4661 = model_audio_tower_layers_8_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_7376, [2])
	        amax_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_7376, [2])
	        full_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_48, full_96);  amin_48 = full_96 = None
	        full_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_48, full_97);  amax_48 = full_97 = None
	        sub_2210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_48, minimum_48);  maximum_48 = None
	        div_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2210, 255.0);  sub_2210 = None
	        clamp_min_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_96, 1.1920928955078125e-07);  div_96 = None
	        div_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_48, clamp_min_144);  minimum_48 = None
	        round_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_97);  div_97 = None
	        sub_2216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_97);  round_97 = None
	        clamp_min_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2216, -128);  sub_2216 = None
	        clamp_max_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_145, 127);  clamp_min_145 = None
	        _assert_tensor_metadata_434 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_144, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_434 = None
	        _assert_tensor_metadata_435 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_96, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_435 = None
	        convert_element_type_288: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_96, torch.int8);  clamp_max_96 = None
	        view_754: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_144, [sym_size_int, 1500, 1])
	        view_755: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_288, [sym_size_int, 1500, 1])
	        reciprocal_48: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_754);  view_754 = None
	        mul_4709: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_48, 1.0);  reciprocal_48 = None
	        mul_4712: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_7376, mul_4709);  mul_4709 = None
	        round_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4712);  mul_4712 = None
	        add_7463: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_98, view_755);  round_98 = view_755 = None
	        clamp_min_146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7463, -128);  add_7463 = None
	        clamp_max_97: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_146, 127);  clamp_min_146 = None
	        _assert_tensor_metadata_436 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_97, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_436 = None
	        convert_element_type_289: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_97, torch.int8);  clamp_max_97 = None
	        view_758: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_144, [sym_size_int, 1500, 1]);  clamp_min_144 = None
	        view_759: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_288, [sym_size_int, 1500, 1]);  convert_element_type_288 = None
	        _assert_tensor_metadata_437 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_289, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_437 = None
	        convert_element_type_290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_289, torch.float32);  convert_element_type_289 = None
	        _assert_tensor_metadata_438 = torch.ops.aten._assert_tensor_metadata.default(view_759, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_438 = None
	        convert_element_type_291: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_759, torch.float32);  view_759 = None
	        sub_2236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_290, convert_element_type_291);  convert_element_type_290 = convert_element_type_291 = None
	        mul_4734: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2236, view_758);  sub_2236 = view_758 = None
	        _assert_tensor_metadata_439 = torch.ops.aten._assert_tensor_metadata.default(mul_4734, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_439 = None
	        view_761: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_762: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_763: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_440 = torch.ops.aten._assert_tensor_metadata.default(view_761, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_440 = None
	        convert_element_type_292: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_761, torch.float32);  view_761 = None
	        _assert_tensor_metadata_441 = torch.ops.aten._assert_tensor_metadata.default(view_763, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_441 = None
	        convert_element_type_293: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_763, torch.float32);  view_763 = None
	        sub_2240: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_292, convert_element_type_293);  convert_element_type_292 = convert_element_type_293 = None
	        mul_4739: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2240, view_762);  sub_2240 = view_762 = None
	        view_764: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4739, [1280, 1280]);  mul_4739 = None
	        _assert_tensor_metadata_442 = torch.ops.aten._assert_tensor_metadata.default(view_764, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_442 = None
	        mul_4744: "Sym(1500*s6)" = sym_size_int * 1500
	        view_765: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4734, [mul_4744, 1280]);  mul_4734 = mul_4744 = None
	        permute_81: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_764, [1, 0]);  view_764 = None
	        addmm_40: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_8_self_attn_q_proj_bias, view_765, permute_81);  model_audio_tower_layers_8_self_attn_q_proj_bias = view_765 = permute_81 = None
	        view_766: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_40, [sym_size_int, 1500, 1280]);  addmm_40 = None
	        mul_4751: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_766, 0.125);  view_766 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_767: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_4751, [sym_size_int, 1500, 20, 64]);  mul_4751 = None
	        permute_82: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_767, [0, 2, 1, 3]);  view_767 = None
	        clone_66: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_82, memory_format = torch.contiguous_format);  permute_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_7376, [2])
	        amax_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_7376, [2])
	        full_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_49, full_98);  amin_49 = full_98 = None
	        full_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_49, full_99);  amax_49 = full_99 = None
	        sub_2255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_49, minimum_49);  maximum_49 = None
	        div_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2255, 255.0);  sub_2255 = None
	        clamp_min_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_98, 1.1920928955078125e-07);  div_98 = None
	        div_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_49, clamp_min_147);  minimum_49 = None
	        round_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_99);  div_99 = None
	        sub_2261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_99);  round_99 = None
	        clamp_min_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2261, -128);  sub_2261 = None
	        clamp_max_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_148, 127);  clamp_min_148 = None
	        _assert_tensor_metadata_443 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_147, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_443 = None
	        _assert_tensor_metadata_444 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_98, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_444 = None
	        convert_element_type_294: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_98, torch.int8);  clamp_max_98 = None
	        view_770: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_147, [sym_size_int, 1500, 1])
	        view_771: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_294, [sym_size_int, 1500, 1])
	        reciprocal_49: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_770);  view_770 = None
	        mul_4805: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_49, 1.0);  reciprocal_49 = None
	        mul_4808: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_7376, mul_4805);  mul_4805 = None
	        round_100: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4808);  mul_4808 = None
	        add_7615: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_100, view_771);  round_100 = view_771 = None
	        clamp_min_149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7615, -128);  add_7615 = None
	        clamp_max_99: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_149, 127);  clamp_min_149 = None
	        _assert_tensor_metadata_445 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_99, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_445 = None
	        convert_element_type_295: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_99, torch.int8);  clamp_max_99 = None
	        view_774: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_147, [sym_size_int, 1500, 1]);  clamp_min_147 = None
	        view_775: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_294, [sym_size_int, 1500, 1]);  convert_element_type_294 = None
	        _assert_tensor_metadata_446 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_295, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_446 = None
	        convert_element_type_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_295, torch.float32);  convert_element_type_295 = None
	        _assert_tensor_metadata_447 = torch.ops.aten._assert_tensor_metadata.default(view_775, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_447 = None
	        convert_element_type_297: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_775, torch.float32);  view_775 = None
	        sub_2281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_296, convert_element_type_297);  convert_element_type_296 = convert_element_type_297 = None
	        mul_4830: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2281, view_774);  sub_2281 = view_774 = None
	        _assert_tensor_metadata_448 = torch.ops.aten._assert_tensor_metadata.default(mul_4830, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_448 = None
	        view_777: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_778: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_779: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_449 = torch.ops.aten._assert_tensor_metadata.default(view_777, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_449 = None
	        convert_element_type_298: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_777, torch.float32);  view_777 = None
	        _assert_tensor_metadata_450 = torch.ops.aten._assert_tensor_metadata.default(view_779, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_450 = None
	        convert_element_type_299: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_779, torch.float32);  view_779 = None
	        sub_2285: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_298, convert_element_type_299);  convert_element_type_298 = convert_element_type_299 = None
	        mul_4835: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2285, view_778);  sub_2285 = view_778 = None
	        view_780: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4835, [1280, 1280]);  mul_4835 = None
	        _assert_tensor_metadata_451 = torch.ops.aten._assert_tensor_metadata.default(view_780, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_451 = None
	        permute_83: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_780, [1, 0]);  view_780 = None
	        mul_4838: "Sym(1500*s6)" = sym_size_int * 1500
	        view_781: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4830, [mul_4838, 1280]);  mul_4830 = mul_4838 = None
	        mm_8: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_781, permute_83);  view_781 = permute_83 = None
	        view_782: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_8, [sym_size_int, 1500, 1280]);  mm_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_783: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_782, [sym_size_int, -1, 20, 64]);  view_782 = None
	        permute_84: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_783, [0, 2, 1, 3]);  view_783 = None
	        clone_67: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_84, memory_format = torch.contiguous_format);  permute_84 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_7376, [2])
	        amax_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_7376, [2])
	        full_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_50, full_100);  amin_50 = full_100 = None
	        full_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_50, full_101);  amax_50 = full_101 = None
	        sub_2299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_50, minimum_50);  maximum_50 = None
	        div_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2299, 255.0);  sub_2299 = None
	        clamp_min_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_100, 1.1920928955078125e-07);  div_100 = None
	        div_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_50, clamp_min_150);  minimum_50 = None
	        round_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_101);  div_101 = None
	        sub_2305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_101);  round_101 = None
	        clamp_min_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2305, -128);  sub_2305 = None
	        clamp_max_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_151, 127);  clamp_min_151 = None
	        _assert_tensor_metadata_452 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_150, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_452 = None
	        _assert_tensor_metadata_453 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_100, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_453 = None
	        convert_element_type_300: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_100, torch.int8);  clamp_max_100 = None
	        view_786: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_150, [sym_size_int, 1500, 1])
	        view_787: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_300, [sym_size_int, 1500, 1])
	        reciprocal_50: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_786);  view_786 = None
	        mul_4904: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_50, 1.0);  reciprocal_50 = None
	        mul_4907: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_7376, mul_4904);  add_7376 = mul_4904 = None
	        round_102: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4907);  mul_4907 = None
	        add_7763: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_102, view_787);  round_102 = view_787 = None
	        clamp_min_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7763, -128);  add_7763 = None
	        clamp_max_101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_152, 127);  clamp_min_152 = None
	        _assert_tensor_metadata_454 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_101, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_454 = None
	        convert_element_type_301: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_101, torch.int8);  clamp_max_101 = None
	        view_790: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_150, [sym_size_int, 1500, 1]);  clamp_min_150 = None
	        view_791: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_300, [sym_size_int, 1500, 1]);  convert_element_type_300 = None
	        _assert_tensor_metadata_455 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_301, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_455 = None
	        convert_element_type_302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_301, torch.float32);  convert_element_type_301 = None
	        _assert_tensor_metadata_456 = torch.ops.aten._assert_tensor_metadata.default(view_791, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_456 = None
	        convert_element_type_303: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_791, torch.float32);  view_791 = None
	        sub_2325: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_302, convert_element_type_303);  convert_element_type_302 = convert_element_type_303 = None
	        mul_4929: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2325, view_790);  sub_2325 = view_790 = None
	        _assert_tensor_metadata_457 = torch.ops.aten._assert_tensor_metadata.default(mul_4929, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_457 = None
	        view_793: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_794: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_795: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_458 = torch.ops.aten._assert_tensor_metadata.default(view_793, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_458 = None
	        convert_element_type_304: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_793, torch.float32);  view_793 = None
	        _assert_tensor_metadata_459 = torch.ops.aten._assert_tensor_metadata.default(view_795, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_459 = None
	        convert_element_type_305: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_795, torch.float32);  view_795 = None
	        sub_2329: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_304, convert_element_type_305);  convert_element_type_304 = convert_element_type_305 = None
	        mul_4934: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2329, view_794);  sub_2329 = view_794 = None
	        view_796: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4934, [1280, 1280]);  mul_4934 = None
	        _assert_tensor_metadata_460 = torch.ops.aten._assert_tensor_metadata.default(view_796, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_460 = None
	        mul_4939: "Sym(1500*s6)" = sym_size_int * 1500
	        view_797: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_4929, [mul_4939, 1280]);  mul_4929 = mul_4939 = None
	        permute_85: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_796, [1, 0]);  view_796 = None
	        addmm_41: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_8_self_attn_v_proj_bias, view_797, permute_85);  model_audio_tower_layers_8_self_attn_v_proj_bias = view_797 = permute_85 = None
	        view_798: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_41, [sym_size_int, 1500, 1280]);  addmm_41 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_799: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_798, [sym_size_int, -1, 20, 64]);  view_798 = None
	        permute_86: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_799, [0, 2, 1, 3]);  view_799 = None
	        clone_68: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_86, memory_format = torch.contiguous_format);  permute_86 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_8 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_66, clone_67, clone_68, None, False, scale = 1.0);  clone_66 = clone_67 = clone_68 = None
	        getitem_66: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_8[0];  _scaled_dot_product_efficient_attention_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_87: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_66, [0, 2, 1, 3]);  getitem_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_800: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_87, [sym_size_int, 1500, -1]);  permute_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_800, [2])
	        amax_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_800, [2])
	        full_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_51, full_102);  amin_51 = full_102 = None
	        full_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_51, full_103);  amax_51 = full_103 = None
	        sub_2347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_51, minimum_51);  maximum_51 = None
	        div_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2347, 255.0);  sub_2347 = None
	        clamp_min_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_102, 1.1920928955078125e-07);  div_102 = None
	        div_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_51, clamp_min_153);  minimum_51 = None
	        round_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_103);  div_103 = None
	        sub_2353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_103);  round_103 = None
	        clamp_min_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2353, -128);  sub_2353 = None
	        clamp_max_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_154, 127);  clamp_min_154 = None
	        _assert_tensor_metadata_461 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_153, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_461 = None
	        _assert_tensor_metadata_462 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_102, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_462 = None
	        convert_element_type_306: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_102, torch.int8);  clamp_max_102 = None
	        view_803: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_153, [sym_size_int, 1500, 1])
	        view_804: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_306, [sym_size_int, 1500, 1])
	        reciprocal_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_803);  view_803 = None
	        mul_5009: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_51, 1.0);  reciprocal_51 = None
	        mul_5012: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_800, mul_5009);  view_800 = mul_5009 = None
	        round_104: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5012);  mul_5012 = None
	        add_7927: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_104, view_804);  round_104 = view_804 = None
	        clamp_min_155: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7927, -128);  add_7927 = None
	        clamp_max_103: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_155, 127);  clamp_min_155 = None
	        _assert_tensor_metadata_463 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_103, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_463 = None
	        convert_element_type_307: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_103, torch.int8);  clamp_max_103 = None
	        view_807: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_153, [sym_size_int, 1500, 1]);  clamp_min_153 = None
	        view_808: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_306, [sym_size_int, 1500, 1]);  convert_element_type_306 = None
	        _assert_tensor_metadata_464 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_307, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_464 = None
	        convert_element_type_308: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_307, torch.float32);  convert_element_type_307 = None
	        _assert_tensor_metadata_465 = torch.ops.aten._assert_tensor_metadata.default(view_808, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_465 = None
	        convert_element_type_309: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_808, torch.float32);  view_808 = None
	        sub_2373: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_308, convert_element_type_309);  convert_element_type_308 = convert_element_type_309 = None
	        mul_5034: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2373, view_807);  sub_2373 = view_807 = None
	        _assert_tensor_metadata_466 = torch.ops.aten._assert_tensor_metadata.default(mul_5034, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_466 = None
	        view_810: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_811: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_812: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_467 = torch.ops.aten._assert_tensor_metadata.default(view_810, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_467 = None
	        convert_element_type_310: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_810, torch.float32);  view_810 = None
	        _assert_tensor_metadata_468 = torch.ops.aten._assert_tensor_metadata.default(view_812, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_468 = None
	        convert_element_type_311: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_812, torch.float32);  view_812 = None
	        sub_2377: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_310, convert_element_type_311);  convert_element_type_310 = convert_element_type_311 = None
	        mul_5039: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2377, view_811);  sub_2377 = view_811 = None
	        view_813: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5039, [1280, 1280]);  mul_5039 = None
	        _assert_tensor_metadata_469 = torch.ops.aten._assert_tensor_metadata.default(view_813, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_469 = None
	        mul_5044: "Sym(1500*s6)" = sym_size_int * 1500
	        view_814: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5034, [mul_5044, 1280]);  mul_5034 = mul_5044 = None
	        permute_88: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_813, [1, 0]);  view_813 = None
	        addmm_42: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_8_self_attn_out_proj_bias, view_814, permute_88);  model_audio_tower_layers_8_self_attn_out_proj_bias = view_814 = permute_88 = None
	        view_815: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_42, [sym_size_int, 1500, 1280]);  addmm_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_7990: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_7370, view_815);  add_7370 = view_815 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_70: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_7990, memory_format = torch.contiguous_format)
	        var_mean_17 = torch.ops.aten.var_mean.correction(clone_70, [2], correction = 0, keepdim = True)
	        getitem_70: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_17[0]
	        getitem_71: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_17[1];  var_mean_17 = None
	        add_7995: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_70, 1e-05);  getitem_70 = None
	        rsqrt_17: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_7995);  add_7995 = None
	        sub_2383: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_70, getitem_71);  clone_70 = getitem_71 = None
	        mul_5055: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2383, rsqrt_17);  sub_2383 = rsqrt_17 = None
	        mul_5056: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5055, model_audio_tower_layers_8_final_layer_norm_weight);  mul_5055 = model_audio_tower_layers_8_final_layer_norm_weight = None
	        add_7996: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5056, model_audio_tower_layers_8_final_layer_norm_bias);  mul_5056 = model_audio_tower_layers_8_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_7996, [2])
	        amax_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_7996, [2])
	        full_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_52, full_104);  amin_52 = full_104 = None
	        full_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_52, full_105);  amax_52 = full_105 = None
	        sub_2394: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_52, minimum_52);  maximum_52 = None
	        div_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2394, 255.0);  sub_2394 = None
	        clamp_min_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_104, 1.1920928955078125e-07);  div_104 = None
	        div_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_52, clamp_min_156);  minimum_52 = None
	        round_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_105);  div_105 = None
	        sub_2400: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_105);  round_105 = None
	        clamp_min_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2400, -128);  sub_2400 = None
	        clamp_max_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_157, 127);  clamp_min_157 = None
	        _assert_tensor_metadata_470 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_156, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_470 = None
	        _assert_tensor_metadata_471 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_104, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_471 = None
	        convert_element_type_312: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_104, torch.int8);  clamp_max_104 = None
	        view_818: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_156, [sym_size_int, 1500, 1])
	        view_819: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_312, [sym_size_int, 1500, 1])
	        reciprocal_52: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_818);  view_818 = None
	        mul_5104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_52, 1.0);  reciprocal_52 = None
	        mul_5107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_7996, mul_5104);  add_7996 = mul_5104 = None
	        round_106: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5107);  mul_5107 = None
	        add_8083: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_106, view_819);  round_106 = view_819 = None
	        clamp_min_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8083, -128);  add_8083 = None
	        clamp_max_105: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_158, 127);  clamp_min_158 = None
	        _assert_tensor_metadata_472 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_105, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_472 = None
	        convert_element_type_313: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_105, torch.int8);  clamp_max_105 = None
	        view_822: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_156, [sym_size_int, 1500, 1]);  clamp_min_156 = None
	        view_823: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_312, [sym_size_int, 1500, 1]);  convert_element_type_312 = None
	        _assert_tensor_metadata_473 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_313, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_473 = None
	        convert_element_type_314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_313, torch.float32);  convert_element_type_313 = None
	        _assert_tensor_metadata_474 = torch.ops.aten._assert_tensor_metadata.default(view_823, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_474 = None
	        convert_element_type_315: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_823, torch.float32);  view_823 = None
	        sub_2420: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_314, convert_element_type_315);  convert_element_type_314 = convert_element_type_315 = None
	        mul_5129: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2420, view_822);  sub_2420 = view_822 = None
	        _assert_tensor_metadata_475 = torch.ops.aten._assert_tensor_metadata.default(mul_5129, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_475 = None
	        view_825: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_8_fc1_parametrizations_weight_original0 = None
	        view_826: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_8_fc1_parametrizations_weight_original1 = None
	        view_827: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_8_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_476 = torch.ops.aten._assert_tensor_metadata.default(view_825, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_476 = None
	        convert_element_type_316: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_825, torch.float32);  view_825 = None
	        _assert_tensor_metadata_477 = torch.ops.aten._assert_tensor_metadata.default(view_827, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_477 = None
	        convert_element_type_317: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_827, torch.float32);  view_827 = None
	        sub_2424: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_316, convert_element_type_317);  convert_element_type_316 = convert_element_type_317 = None
	        mul_5134: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2424, view_826);  sub_2424 = view_826 = None
	        view_828: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5134, [5120, 1280]);  mul_5134 = None
	        _assert_tensor_metadata_478 = torch.ops.aten._assert_tensor_metadata.default(view_828, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_478 = None
	        mul_5139: "Sym(1500*s6)" = sym_size_int * 1500
	        view_829: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5129, [mul_5139, 1280]);  mul_5129 = mul_5139 = None
	        permute_89: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_828, [1, 0]);  view_828 = None
	        addmm_43: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_8_fc1_bias, view_829, permute_89);  model_audio_tower_layers_8_fc1_bias = view_829 = permute_89 = None
	        view_830: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_43, [sym_size_int, 1500, 5120]);  addmm_43 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_5146: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_830, 0.5)
	        mul_5147: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_830, 0.7071067811865476);  view_830 = None
	        erf_10: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_5147);  mul_5147 = None
	        add_8142: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_10, 1);  erf_10 = None
	        mul_5148: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5146, add_8142);  mul_5146 = add_8142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_5148, [2])
	        amax_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_5148, [2])
	        full_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_53, full_106);  amin_53 = full_106 = None
	        full_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_53, full_107);  amax_53 = full_107 = None
	        sub_2437: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_53, minimum_53);  maximum_53 = None
	        div_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2437, 255.0);  sub_2437 = None
	        clamp_min_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_106, 1.1920928955078125e-07);  div_106 = None
	        div_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_53, clamp_min_159);  minimum_53 = None
	        round_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_107);  div_107 = None
	        sub_2443: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_107);  round_107 = None
	        clamp_min_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2443, -128);  sub_2443 = None
	        clamp_max_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_160, 127);  clamp_min_160 = None
	        _assert_tensor_metadata_479 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_159, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_479 = None
	        _assert_tensor_metadata_480 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_106, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_480 = None
	        convert_element_type_318: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_106, torch.int8);  clamp_max_106 = None
	        view_833: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_159, [sym_size_int, 1500, 1])
	        view_834: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_318, [sym_size_int, 1500, 1])
	        reciprocal_53: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_833);  view_833 = None
	        mul_5194: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_53, 1.0);  reciprocal_53 = None
	        mul_5197: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5148, mul_5194);  mul_5148 = mul_5194 = None
	        round_108: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_5197);  mul_5197 = None
	        add_8225: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_108, view_834);  round_108 = view_834 = None
	        clamp_min_161: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8225, -128);  add_8225 = None
	        clamp_max_107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_161, 127);  clamp_min_161 = None
	        _assert_tensor_metadata_481 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_107, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_481 = None
	        convert_element_type_319: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_107, torch.int8);  clamp_max_107 = None
	        view_837: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_159, [sym_size_int, 1500, 1]);  clamp_min_159 = None
	        view_838: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_318, [sym_size_int, 1500, 1]);  convert_element_type_318 = None
	        _assert_tensor_metadata_482 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_319, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_482 = None
	        convert_element_type_320: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_319, torch.float32);  convert_element_type_319 = None
	        _assert_tensor_metadata_483 = torch.ops.aten._assert_tensor_metadata.default(view_838, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_483 = None
	        convert_element_type_321: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_838, torch.float32);  view_838 = None
	        sub_2463: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_320, convert_element_type_321);  convert_element_type_320 = convert_element_type_321 = None
	        mul_5219: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2463, view_837);  sub_2463 = view_837 = None
	        _assert_tensor_metadata_484 = torch.ops.aten._assert_tensor_metadata.default(mul_5219, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_484 = None
	        view_840: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_8_fc2_parametrizations_weight_original0 = None
	        view_841: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_8_fc2_parametrizations_weight_original1 = None
	        view_842: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_8_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_8_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_485 = torch.ops.aten._assert_tensor_metadata.default(view_840, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_485 = None
	        convert_element_type_322: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_840, torch.float32);  view_840 = None
	        _assert_tensor_metadata_486 = torch.ops.aten._assert_tensor_metadata.default(view_842, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_486 = None
	        convert_element_type_323: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_842, torch.float32);  view_842 = None
	        sub_2467: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_322, convert_element_type_323);  convert_element_type_322 = convert_element_type_323 = None
	        mul_5224: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2467, view_841);  sub_2467 = view_841 = None
	        view_843: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_5224, [1280, 5120]);  mul_5224 = None
	        _assert_tensor_metadata_487 = torch.ops.aten._assert_tensor_metadata.default(view_843, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_487 = None
	        mul_5229: "Sym(1500*s6)" = sym_size_int * 1500
	        view_844: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_5219, [mul_5229, 5120]);  mul_5219 = mul_5229 = None
	        permute_90: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_843, [1, 0]);  view_843 = None
	        addmm_44: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_8_fc2_bias, view_844, permute_90);  model_audio_tower_layers_8_fc2_bias = view_844 = permute_90 = None
	        view_845: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_44, [sym_size_int, 1500, 1280]);  addmm_44 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_8288: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_7990, view_845);  add_7990 = view_845 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_73: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_8288, memory_format = torch.contiguous_format)
	        var_mean_18 = torch.ops.aten.var_mean.correction(clone_73, [2], correction = 0, keepdim = True)
	        getitem_72: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_18[0]
	        getitem_73: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_18[1];  var_mean_18 = None
	        add_8293: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_72, 1e-05);  getitem_72 = None
	        rsqrt_18: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_8293);  add_8293 = None
	        sub_2473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_73, getitem_73);  clone_73 = getitem_73 = None
	        mul_5240: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2473, rsqrt_18);  sub_2473 = rsqrt_18 = None
	        mul_5241: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5240, model_audio_tower_layers_9_self_attn_layer_norm_weight);  mul_5240 = model_audio_tower_layers_9_self_attn_layer_norm_weight = None
	        add_8294: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5241, model_audio_tower_layers_9_self_attn_layer_norm_bias);  mul_5241 = model_audio_tower_layers_9_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_8294, [2])
	        amax_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_8294, [2])
	        full_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_54, full_108);  amin_54 = full_108 = None
	        full_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_54, full_109);  amax_54 = full_109 = None
	        sub_2484: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_54, minimum_54);  maximum_54 = None
	        div_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2484, 255.0);  sub_2484 = None
	        clamp_min_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_108, 1.1920928955078125e-07);  div_108 = None
	        div_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_54, clamp_min_162);  minimum_54 = None
	        round_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_109);  div_109 = None
	        sub_2490: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_109);  round_109 = None
	        clamp_min_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2490, -128);  sub_2490 = None
	        clamp_max_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_163, 127);  clamp_min_163 = None
	        _assert_tensor_metadata_488 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_162, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_488 = None
	        _assert_tensor_metadata_489 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_108, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_489 = None
	        convert_element_type_324: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_108, torch.int8);  clamp_max_108 = None
	        view_848: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_162, [sym_size_int, 1500, 1])
	        view_849: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_324, [sym_size_int, 1500, 1])
	        reciprocal_54: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_848);  view_848 = None
	        mul_5289: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_54, 1.0);  reciprocal_54 = None
	        mul_5292: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_8294, mul_5289);  mul_5289 = None
	        round_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5292);  mul_5292 = None
	        add_8381: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_110, view_849);  round_110 = view_849 = None
	        clamp_min_164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8381, -128);  add_8381 = None
	        clamp_max_109: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_164, 127);  clamp_min_164 = None
	        _assert_tensor_metadata_490 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_109, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_490 = None
	        convert_element_type_325: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_109, torch.int8);  clamp_max_109 = None
	        view_852: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_162, [sym_size_int, 1500, 1]);  clamp_min_162 = None
	        view_853: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_324, [sym_size_int, 1500, 1]);  convert_element_type_324 = None
	        _assert_tensor_metadata_491 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_325, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_491 = None
	        convert_element_type_326: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_325, torch.float32);  convert_element_type_325 = None
	        _assert_tensor_metadata_492 = torch.ops.aten._assert_tensor_metadata.default(view_853, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_492 = None
	        convert_element_type_327: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_853, torch.float32);  view_853 = None
	        sub_2510: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_326, convert_element_type_327);  convert_element_type_326 = convert_element_type_327 = None
	        mul_5314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2510, view_852);  sub_2510 = view_852 = None
	        _assert_tensor_metadata_493 = torch.ops.aten._assert_tensor_metadata.default(mul_5314, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_493 = None
	        view_855: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_856: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_857: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_494 = torch.ops.aten._assert_tensor_metadata.default(view_855, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_494 = None
	        convert_element_type_328: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_855, torch.float32);  view_855 = None
	        _assert_tensor_metadata_495 = torch.ops.aten._assert_tensor_metadata.default(view_857, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_495 = None
	        convert_element_type_329: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_857, torch.float32);  view_857 = None
	        sub_2514: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_328, convert_element_type_329);  convert_element_type_328 = convert_element_type_329 = None
	        mul_5319: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2514, view_856);  sub_2514 = view_856 = None
	        view_858: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5319, [1280, 1280]);  mul_5319 = None
	        _assert_tensor_metadata_496 = torch.ops.aten._assert_tensor_metadata.default(view_858, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_496 = None
	        mul_5324: "Sym(1500*s6)" = sym_size_int * 1500
	        view_859: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5314, [mul_5324, 1280]);  mul_5314 = mul_5324 = None
	        permute_91: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_858, [1, 0]);  view_858 = None
	        addmm_45: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_9_self_attn_q_proj_bias, view_859, permute_91);  model_audio_tower_layers_9_self_attn_q_proj_bias = view_859 = permute_91 = None
	        view_860: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_45, [sym_size_int, 1500, 1280]);  addmm_45 = None
	        mul_5331: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_860, 0.125);  view_860 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_861: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_5331, [sym_size_int, 1500, 20, 64]);  mul_5331 = None
	        permute_92: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_861, [0, 2, 1, 3]);  view_861 = None
	        clone_74: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_92, memory_format = torch.contiguous_format);  permute_92 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_8294, [2])
	        amax_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_8294, [2])
	        full_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_55, full_110);  amin_55 = full_110 = None
	        full_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_55, full_111);  amax_55 = full_111 = None
	        sub_2529: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_55, minimum_55);  maximum_55 = None
	        div_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2529, 255.0);  sub_2529 = None
	        clamp_min_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_110, 1.1920928955078125e-07);  div_110 = None
	        div_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_55, clamp_min_165);  minimum_55 = None
	        round_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_111);  div_111 = None
	        sub_2535: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_111);  round_111 = None
	        clamp_min_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2535, -128);  sub_2535 = None
	        clamp_max_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_166, 127);  clamp_min_166 = None
	        _assert_tensor_metadata_497 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_165, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_497 = None
	        _assert_tensor_metadata_498 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_110, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_498 = None
	        convert_element_type_330: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_110, torch.int8);  clamp_max_110 = None
	        view_864: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_165, [sym_size_int, 1500, 1])
	        view_865: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_330, [sym_size_int, 1500, 1])
	        reciprocal_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_864);  view_864 = None
	        mul_5385: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_55, 1.0);  reciprocal_55 = None
	        mul_5388: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_8294, mul_5385);  mul_5385 = None
	        round_112: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5388);  mul_5388 = None
	        add_8533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_112, view_865);  round_112 = view_865 = None
	        clamp_min_167: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8533, -128);  add_8533 = None
	        clamp_max_111: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_167, 127);  clamp_min_167 = None
	        _assert_tensor_metadata_499 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_111, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_499 = None
	        convert_element_type_331: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_111, torch.int8);  clamp_max_111 = None
	        view_868: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_165, [sym_size_int, 1500, 1]);  clamp_min_165 = None
	        view_869: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_330, [sym_size_int, 1500, 1]);  convert_element_type_330 = None
	        _assert_tensor_metadata_500 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_331, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_500 = None
	        convert_element_type_332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_331, torch.float32);  convert_element_type_331 = None
	        _assert_tensor_metadata_501 = torch.ops.aten._assert_tensor_metadata.default(view_869, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_501 = None
	        convert_element_type_333: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_869, torch.float32);  view_869 = None
	        sub_2555: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_332, convert_element_type_333);  convert_element_type_332 = convert_element_type_333 = None
	        mul_5410: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2555, view_868);  sub_2555 = view_868 = None
	        _assert_tensor_metadata_502 = torch.ops.aten._assert_tensor_metadata.default(mul_5410, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_502 = None
	        view_871: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_872: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_873: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_503 = torch.ops.aten._assert_tensor_metadata.default(view_871, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_503 = None
	        convert_element_type_334: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_871, torch.float32);  view_871 = None
	        _assert_tensor_metadata_504 = torch.ops.aten._assert_tensor_metadata.default(view_873, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_504 = None
	        convert_element_type_335: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_873, torch.float32);  view_873 = None
	        sub_2559: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_334, convert_element_type_335);  convert_element_type_334 = convert_element_type_335 = None
	        mul_5415: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2559, view_872);  sub_2559 = view_872 = None
	        view_874: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5415, [1280, 1280]);  mul_5415 = None
	        _assert_tensor_metadata_505 = torch.ops.aten._assert_tensor_metadata.default(view_874, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_505 = None
	        permute_93: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_874, [1, 0]);  view_874 = None
	        mul_5418: "Sym(1500*s6)" = sym_size_int * 1500
	        view_875: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5410, [mul_5418, 1280]);  mul_5410 = mul_5418 = None
	        mm_9: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_875, permute_93);  view_875 = permute_93 = None
	        view_876: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_9, [sym_size_int, 1500, 1280]);  mm_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_877: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_876, [sym_size_int, -1, 20, 64]);  view_876 = None
	        permute_94: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_877, [0, 2, 1, 3]);  view_877 = None
	        clone_75: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_94, memory_format = torch.contiguous_format);  permute_94 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_8294, [2])
	        amax_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_8294, [2])
	        full_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_56, full_112);  amin_56 = full_112 = None
	        full_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_56, full_113);  amax_56 = full_113 = None
	        sub_2573: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_56, minimum_56);  maximum_56 = None
	        div_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2573, 255.0);  sub_2573 = None
	        clamp_min_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_112, 1.1920928955078125e-07);  div_112 = None
	        div_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_56, clamp_min_168);  minimum_56 = None
	        round_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_113);  div_113 = None
	        sub_2579: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_113);  round_113 = None
	        clamp_min_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2579, -128);  sub_2579 = None
	        clamp_max_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_169, 127);  clamp_min_169 = None
	        _assert_tensor_metadata_506 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_168, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_506 = None
	        _assert_tensor_metadata_507 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_112, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_507 = None
	        convert_element_type_336: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_112, torch.int8);  clamp_max_112 = None
	        view_880: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_168, [sym_size_int, 1500, 1])
	        view_881: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_336, [sym_size_int, 1500, 1])
	        reciprocal_56: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_880);  view_880 = None
	        mul_5484: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_56, 1.0);  reciprocal_56 = None
	        mul_5487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_8294, mul_5484);  add_8294 = mul_5484 = None
	        round_114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5487);  mul_5487 = None
	        add_8681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_114, view_881);  round_114 = view_881 = None
	        clamp_min_170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8681, -128);  add_8681 = None
	        clamp_max_113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_170, 127);  clamp_min_170 = None
	        _assert_tensor_metadata_508 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_113, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_508 = None
	        convert_element_type_337: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_113, torch.int8);  clamp_max_113 = None
	        view_884: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_168, [sym_size_int, 1500, 1]);  clamp_min_168 = None
	        view_885: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_336, [sym_size_int, 1500, 1]);  convert_element_type_336 = None
	        _assert_tensor_metadata_509 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_337, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_509 = None
	        convert_element_type_338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_337, torch.float32);  convert_element_type_337 = None
	        _assert_tensor_metadata_510 = torch.ops.aten._assert_tensor_metadata.default(view_885, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_510 = None
	        convert_element_type_339: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_885, torch.float32);  view_885 = None
	        sub_2599: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_338, convert_element_type_339);  convert_element_type_338 = convert_element_type_339 = None
	        mul_5509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2599, view_884);  sub_2599 = view_884 = None
	        _assert_tensor_metadata_511 = torch.ops.aten._assert_tensor_metadata.default(mul_5509, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_511 = None
	        view_887: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_888: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_889: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_512 = torch.ops.aten._assert_tensor_metadata.default(view_887, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_512 = None
	        convert_element_type_340: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_887, torch.float32);  view_887 = None
	        _assert_tensor_metadata_513 = torch.ops.aten._assert_tensor_metadata.default(view_889, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_513 = None
	        convert_element_type_341: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_889, torch.float32);  view_889 = None
	        sub_2603: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_340, convert_element_type_341);  convert_element_type_340 = convert_element_type_341 = None
	        mul_5514: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2603, view_888);  sub_2603 = view_888 = None
	        view_890: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5514, [1280, 1280]);  mul_5514 = None
	        _assert_tensor_metadata_514 = torch.ops.aten._assert_tensor_metadata.default(view_890, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_514 = None
	        mul_5519: "Sym(1500*s6)" = sym_size_int * 1500
	        view_891: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5509, [mul_5519, 1280]);  mul_5509 = mul_5519 = None
	        permute_95: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_890, [1, 0]);  view_890 = None
	        addmm_46: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_9_self_attn_v_proj_bias, view_891, permute_95);  model_audio_tower_layers_9_self_attn_v_proj_bias = view_891 = permute_95 = None
	        view_892: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_46, [sym_size_int, 1500, 1280]);  addmm_46 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_893: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_892, [sym_size_int, -1, 20, 64]);  view_892 = None
	        permute_96: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_893, [0, 2, 1, 3]);  view_893 = None
	        clone_76: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_96, memory_format = torch.contiguous_format);  permute_96 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_9 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_74, clone_75, clone_76, None, False, scale = 1.0);  clone_74 = clone_75 = clone_76 = None
	        getitem_74: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_9[0];  _scaled_dot_product_efficient_attention_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_97: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_74, [0, 2, 1, 3]);  getitem_74 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_97, [sym_size_int, 1500, -1]);  permute_97 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_894, [2])
	        amax_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_894, [2])
	        full_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_57, full_114);  amin_57 = full_114 = None
	        full_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_57, full_115);  amax_57 = full_115 = None
	        sub_2621: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_57, minimum_57);  maximum_57 = None
	        div_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2621, 255.0);  sub_2621 = None
	        clamp_min_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_114, 1.1920928955078125e-07);  div_114 = None
	        div_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_57, clamp_min_171);  minimum_57 = None
	        round_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_115);  div_115 = None
	        sub_2627: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_115);  round_115 = None
	        clamp_min_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2627, -128);  sub_2627 = None
	        clamp_max_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_172, 127);  clamp_min_172 = None
	        _assert_tensor_metadata_515 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_171, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_515 = None
	        _assert_tensor_metadata_516 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_114, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_516 = None
	        convert_element_type_342: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_114, torch.int8);  clamp_max_114 = None
	        view_897: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_171, [sym_size_int, 1500, 1])
	        view_898: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_342, [sym_size_int, 1500, 1])
	        reciprocal_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_897);  view_897 = None
	        mul_5589: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_57, 1.0);  reciprocal_57 = None
	        mul_5592: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_894, mul_5589);  view_894 = mul_5589 = None
	        round_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5592);  mul_5592 = None
	        add_8845: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_116, view_898);  round_116 = view_898 = None
	        clamp_min_173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8845, -128);  add_8845 = None
	        clamp_max_115: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_173, 127);  clamp_min_173 = None
	        _assert_tensor_metadata_517 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_115, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_517 = None
	        convert_element_type_343: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_115, torch.int8);  clamp_max_115 = None
	        view_901: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_171, [sym_size_int, 1500, 1]);  clamp_min_171 = None
	        view_902: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_342, [sym_size_int, 1500, 1]);  convert_element_type_342 = None
	        _assert_tensor_metadata_518 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_343, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_518 = None
	        convert_element_type_344: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_343, torch.float32);  convert_element_type_343 = None
	        _assert_tensor_metadata_519 = torch.ops.aten._assert_tensor_metadata.default(view_902, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_519 = None
	        convert_element_type_345: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_902, torch.float32);  view_902 = None
	        sub_2647: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_344, convert_element_type_345);  convert_element_type_344 = convert_element_type_345 = None
	        mul_5614: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2647, view_901);  sub_2647 = view_901 = None
	        _assert_tensor_metadata_520 = torch.ops.aten._assert_tensor_metadata.default(mul_5614, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_520 = None
	        view_904: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_905: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_906: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_521 = torch.ops.aten._assert_tensor_metadata.default(view_904, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_521 = None
	        convert_element_type_346: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_904, torch.float32);  view_904 = None
	        _assert_tensor_metadata_522 = torch.ops.aten._assert_tensor_metadata.default(view_906, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_522 = None
	        convert_element_type_347: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_906, torch.float32);  view_906 = None
	        sub_2651: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_346, convert_element_type_347);  convert_element_type_346 = convert_element_type_347 = None
	        mul_5619: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2651, view_905);  sub_2651 = view_905 = None
	        view_907: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5619, [1280, 1280]);  mul_5619 = None
	        _assert_tensor_metadata_523 = torch.ops.aten._assert_tensor_metadata.default(view_907, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_523 = None
	        mul_5624: "Sym(1500*s6)" = sym_size_int * 1500
	        view_908: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5614, [mul_5624, 1280]);  mul_5614 = mul_5624 = None
	        permute_98: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_907, [1, 0]);  view_907 = None
	        addmm_47: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_9_self_attn_out_proj_bias, view_908, permute_98);  model_audio_tower_layers_9_self_attn_out_proj_bias = view_908 = permute_98 = None
	        view_909: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_47, [sym_size_int, 1500, 1280]);  addmm_47 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_8908: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_8288, view_909);  add_8288 = view_909 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_78: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_8908, memory_format = torch.contiguous_format)
	        var_mean_19 = torch.ops.aten.var_mean.correction(clone_78, [2], correction = 0, keepdim = True)
	        getitem_78: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_19[0]
	        getitem_79: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_19[1];  var_mean_19 = None
	        add_8913: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_78, 1e-05);  getitem_78 = None
	        rsqrt_19: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_8913);  add_8913 = None
	        sub_2657: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_78, getitem_79);  clone_78 = getitem_79 = None
	        mul_5635: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2657, rsqrt_19);  sub_2657 = rsqrt_19 = None
	        mul_5636: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5635, model_audio_tower_layers_9_final_layer_norm_weight);  mul_5635 = model_audio_tower_layers_9_final_layer_norm_weight = None
	        add_8914: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5636, model_audio_tower_layers_9_final_layer_norm_bias);  mul_5636 = model_audio_tower_layers_9_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_8914, [2])
	        amax_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_8914, [2])
	        full_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_58, full_116);  amin_58 = full_116 = None
	        full_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_58, full_117);  amax_58 = full_117 = None
	        sub_2668: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_58, minimum_58);  maximum_58 = None
	        div_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2668, 255.0);  sub_2668 = None
	        clamp_min_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_116, 1.1920928955078125e-07);  div_116 = None
	        div_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_58, clamp_min_174);  minimum_58 = None
	        round_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_117);  div_117 = None
	        sub_2674: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_117);  round_117 = None
	        clamp_min_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2674, -128);  sub_2674 = None
	        clamp_max_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_175, 127);  clamp_min_175 = None
	        _assert_tensor_metadata_524 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_174, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_524 = None
	        _assert_tensor_metadata_525 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_116, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_525 = None
	        convert_element_type_348: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_116, torch.int8);  clamp_max_116 = None
	        view_912: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_174, [sym_size_int, 1500, 1])
	        view_913: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_348, [sym_size_int, 1500, 1])
	        reciprocal_58: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_912);  view_912 = None
	        mul_5684: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_58, 1.0);  reciprocal_58 = None
	        mul_5687: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_8914, mul_5684);  add_8914 = mul_5684 = None
	        round_118: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5687);  mul_5687 = None
	        add_9001: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_118, view_913);  round_118 = view_913 = None
	        clamp_min_176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9001, -128);  add_9001 = None
	        clamp_max_117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_176, 127);  clamp_min_176 = None
	        _assert_tensor_metadata_526 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_117, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_526 = None
	        convert_element_type_349: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_117, torch.int8);  clamp_max_117 = None
	        view_916: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_174, [sym_size_int, 1500, 1]);  clamp_min_174 = None
	        view_917: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_348, [sym_size_int, 1500, 1]);  convert_element_type_348 = None
	        _assert_tensor_metadata_527 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_349, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_527 = None
	        convert_element_type_350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_349, torch.float32);  convert_element_type_349 = None
	        _assert_tensor_metadata_528 = torch.ops.aten._assert_tensor_metadata.default(view_917, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_528 = None
	        convert_element_type_351: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_917, torch.float32);  view_917 = None
	        sub_2694: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_350, convert_element_type_351);  convert_element_type_350 = convert_element_type_351 = None
	        mul_5709: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2694, view_916);  sub_2694 = view_916 = None
	        _assert_tensor_metadata_529 = torch.ops.aten._assert_tensor_metadata.default(mul_5709, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_529 = None
	        view_919: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_9_fc1_parametrizations_weight_original0 = None
	        view_920: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_9_fc1_parametrizations_weight_original1 = None
	        view_921: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_9_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_530 = torch.ops.aten._assert_tensor_metadata.default(view_919, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_530 = None
	        convert_element_type_352: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_919, torch.float32);  view_919 = None
	        _assert_tensor_metadata_531 = torch.ops.aten._assert_tensor_metadata.default(view_921, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_531 = None
	        convert_element_type_353: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_921, torch.float32);  view_921 = None
	        sub_2698: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_352, convert_element_type_353);  convert_element_type_352 = convert_element_type_353 = None
	        mul_5714: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2698, view_920);  sub_2698 = view_920 = None
	        view_922: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5714, [5120, 1280]);  mul_5714 = None
	        _assert_tensor_metadata_532 = torch.ops.aten._assert_tensor_metadata.default(view_922, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_532 = None
	        mul_5719: "Sym(1500*s6)" = sym_size_int * 1500
	        view_923: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5709, [mul_5719, 1280]);  mul_5709 = mul_5719 = None
	        permute_99: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_922, [1, 0]);  view_922 = None
	        addmm_48: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_9_fc1_bias, view_923, permute_99);  model_audio_tower_layers_9_fc1_bias = view_923 = permute_99 = None
	        view_924: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_48, [sym_size_int, 1500, 5120]);  addmm_48 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_5726: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_924, 0.5)
	        mul_5727: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_924, 0.7071067811865476);  view_924 = None
	        erf_11: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_5727);  mul_5727 = None
	        add_9060: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_11, 1);  erf_11 = None
	        mul_5728: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5726, add_9060);  mul_5726 = add_9060 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_5728, [2])
	        amax_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_5728, [2])
	        full_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_59, full_118);  amin_59 = full_118 = None
	        full_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_59, full_119);  amax_59 = full_119 = None
	        sub_2711: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_59, minimum_59);  maximum_59 = None
	        div_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2711, 255.0);  sub_2711 = None
	        clamp_min_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_118, 1.1920928955078125e-07);  div_118 = None
	        div_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_59, clamp_min_177);  minimum_59 = None
	        round_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_119);  div_119 = None
	        sub_2717: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_119);  round_119 = None
	        clamp_min_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2717, -128);  sub_2717 = None
	        clamp_max_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_178, 127);  clamp_min_178 = None
	        _assert_tensor_metadata_533 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_177, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_533 = None
	        _assert_tensor_metadata_534 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_118, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_534 = None
	        convert_element_type_354: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_118, torch.int8);  clamp_max_118 = None
	        view_927: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_177, [sym_size_int, 1500, 1])
	        view_928: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_354, [sym_size_int, 1500, 1])
	        reciprocal_59: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_927);  view_927 = None
	        mul_5774: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_59, 1.0);  reciprocal_59 = None
	        mul_5777: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5728, mul_5774);  mul_5728 = mul_5774 = None
	        round_120: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_5777);  mul_5777 = None
	        add_9143: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_120, view_928);  round_120 = view_928 = None
	        clamp_min_179: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9143, -128);  add_9143 = None
	        clamp_max_119: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_179, 127);  clamp_min_179 = None
	        _assert_tensor_metadata_535 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_119, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_535 = None
	        convert_element_type_355: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_119, torch.int8);  clamp_max_119 = None
	        view_931: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_177, [sym_size_int, 1500, 1]);  clamp_min_177 = None
	        view_932: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_354, [sym_size_int, 1500, 1]);  convert_element_type_354 = None
	        _assert_tensor_metadata_536 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_355, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_536 = None
	        convert_element_type_356: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_355, torch.float32);  convert_element_type_355 = None
	        _assert_tensor_metadata_537 = torch.ops.aten._assert_tensor_metadata.default(view_932, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_537 = None
	        convert_element_type_357: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_932, torch.float32);  view_932 = None
	        sub_2737: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_356, convert_element_type_357);  convert_element_type_356 = convert_element_type_357 = None
	        mul_5799: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2737, view_931);  sub_2737 = view_931 = None
	        _assert_tensor_metadata_538 = torch.ops.aten._assert_tensor_metadata.default(mul_5799, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_538 = None
	        view_934: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_9_fc2_parametrizations_weight_original0 = None
	        view_935: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_9_fc2_parametrizations_weight_original1 = None
	        view_936: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_9_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_9_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_539 = torch.ops.aten._assert_tensor_metadata.default(view_934, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_539 = None
	        convert_element_type_358: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_934, torch.float32);  view_934 = None
	        _assert_tensor_metadata_540 = torch.ops.aten._assert_tensor_metadata.default(view_936, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_540 = None
	        convert_element_type_359: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_936, torch.float32);  view_936 = None
	        sub_2741: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_358, convert_element_type_359);  convert_element_type_358 = convert_element_type_359 = None
	        mul_5804: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2741, view_935);  sub_2741 = view_935 = None
	        view_937: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_5804, [1280, 5120]);  mul_5804 = None
	        _assert_tensor_metadata_541 = torch.ops.aten._assert_tensor_metadata.default(view_937, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_541 = None
	        mul_5809: "Sym(1500*s6)" = sym_size_int * 1500
	        view_938: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_5799, [mul_5809, 5120]);  mul_5799 = mul_5809 = None
	        permute_100: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_937, [1, 0]);  view_937 = None
	        addmm_49: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_9_fc2_bias, view_938, permute_100);  model_audio_tower_layers_9_fc2_bias = view_938 = permute_100 = None
	        view_939: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_49, [sym_size_int, 1500, 1280]);  addmm_49 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_9206: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_8908, view_939);  add_8908 = view_939 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_81: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_9206, memory_format = torch.contiguous_format)
	        var_mean_20 = torch.ops.aten.var_mean.correction(clone_81, [2], correction = 0, keepdim = True)
	        getitem_80: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_20[0]
	        getitem_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_20[1];  var_mean_20 = None
	        add_9211: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_80, 1e-05);  getitem_80 = None
	        rsqrt_20: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_9211);  add_9211 = None
	        sub_2747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_81, getitem_81);  clone_81 = getitem_81 = None
	        mul_5820: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2747, rsqrt_20);  sub_2747 = rsqrt_20 = None
	        mul_5821: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5820, model_audio_tower_layers_10_self_attn_layer_norm_weight);  mul_5820 = model_audio_tower_layers_10_self_attn_layer_norm_weight = None
	        add_9212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5821, model_audio_tower_layers_10_self_attn_layer_norm_bias);  mul_5821 = model_audio_tower_layers_10_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_9212, [2])
	        amax_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_9212, [2])
	        full_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_60, full_120);  amin_60 = full_120 = None
	        full_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_60, full_121);  amax_60 = full_121 = None
	        sub_2758: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_60, minimum_60);  maximum_60 = None
	        div_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2758, 255.0);  sub_2758 = None
	        clamp_min_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_120, 1.1920928955078125e-07);  div_120 = None
	        div_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_60, clamp_min_180);  minimum_60 = None
	        round_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_121);  div_121 = None
	        sub_2764: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_121);  round_121 = None
	        clamp_min_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2764, -128);  sub_2764 = None
	        clamp_max_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_181, 127);  clamp_min_181 = None
	        _assert_tensor_metadata_542 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_180, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_542 = None
	        _assert_tensor_metadata_543 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_120, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_543 = None
	        convert_element_type_360: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_120, torch.int8);  clamp_max_120 = None
	        view_942: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_180, [sym_size_int, 1500, 1])
	        view_943: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_360, [sym_size_int, 1500, 1])
	        reciprocal_60: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_942);  view_942 = None
	        mul_5869: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_60, 1.0);  reciprocal_60 = None
	        mul_5872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_9212, mul_5869);  mul_5869 = None
	        round_122: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5872);  mul_5872 = None
	        add_9299: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_122, view_943);  round_122 = view_943 = None
	        clamp_min_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9299, -128);  add_9299 = None
	        clamp_max_121: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_182, 127);  clamp_min_182 = None
	        _assert_tensor_metadata_544 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_121, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_544 = None
	        convert_element_type_361: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_121, torch.int8);  clamp_max_121 = None
	        view_946: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_180, [sym_size_int, 1500, 1]);  clamp_min_180 = None
	        view_947: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_360, [sym_size_int, 1500, 1]);  convert_element_type_360 = None
	        _assert_tensor_metadata_545 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_361, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_545 = None
	        convert_element_type_362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_361, torch.float32);  convert_element_type_361 = None
	        _assert_tensor_metadata_546 = torch.ops.aten._assert_tensor_metadata.default(view_947, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_546 = None
	        convert_element_type_363: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_947, torch.float32);  view_947 = None
	        sub_2784: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_362, convert_element_type_363);  convert_element_type_362 = convert_element_type_363 = None
	        mul_5894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2784, view_946);  sub_2784 = view_946 = None
	        _assert_tensor_metadata_547 = torch.ops.aten._assert_tensor_metadata.default(mul_5894, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_547 = None
	        view_949: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_950: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_951: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_548 = torch.ops.aten._assert_tensor_metadata.default(view_949, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_548 = None
	        convert_element_type_364: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_949, torch.float32);  view_949 = None
	        _assert_tensor_metadata_549 = torch.ops.aten._assert_tensor_metadata.default(view_951, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_549 = None
	        convert_element_type_365: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_951, torch.float32);  view_951 = None
	        sub_2788: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_364, convert_element_type_365);  convert_element_type_364 = convert_element_type_365 = None
	        mul_5899: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2788, view_950);  sub_2788 = view_950 = None
	        view_952: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5899, [1280, 1280]);  mul_5899 = None
	        _assert_tensor_metadata_550 = torch.ops.aten._assert_tensor_metadata.default(view_952, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_550 = None
	        mul_5904: "Sym(1500*s6)" = sym_size_int * 1500
	        view_953: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5894, [mul_5904, 1280]);  mul_5894 = mul_5904 = None
	        permute_101: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_952, [1, 0]);  view_952 = None
	        addmm_50: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_10_self_attn_q_proj_bias, view_953, permute_101);  model_audio_tower_layers_10_self_attn_q_proj_bias = view_953 = permute_101 = None
	        view_954: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_50, [sym_size_int, 1500, 1280]);  addmm_50 = None
	        mul_5911: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_954, 0.125);  view_954 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_955: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_5911, [sym_size_int, 1500, 20, 64]);  mul_5911 = None
	        permute_102: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_955, [0, 2, 1, 3]);  view_955 = None
	        clone_82: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_102, memory_format = torch.contiguous_format);  permute_102 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_9212, [2])
	        amax_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_9212, [2])
	        full_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_61, full_122);  amin_61 = full_122 = None
	        full_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_61, full_123);  amax_61 = full_123 = None
	        sub_2803: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_61, minimum_61);  maximum_61 = None
	        div_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2803, 255.0);  sub_2803 = None
	        clamp_min_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_122, 1.1920928955078125e-07);  div_122 = None
	        div_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_61, clamp_min_183);  minimum_61 = None
	        round_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_123);  div_123 = None
	        sub_2809: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_123);  round_123 = None
	        clamp_min_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2809, -128);  sub_2809 = None
	        clamp_max_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_184, 127);  clamp_min_184 = None
	        _assert_tensor_metadata_551 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_183, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_551 = None
	        _assert_tensor_metadata_552 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_122, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_552 = None
	        convert_element_type_366: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_122, torch.int8);  clamp_max_122 = None
	        view_958: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_183, [sym_size_int, 1500, 1])
	        view_959: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_366, [sym_size_int, 1500, 1])
	        reciprocal_61: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_958);  view_958 = None
	        mul_5965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_61, 1.0);  reciprocal_61 = None
	        mul_5968: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_9212, mul_5965);  mul_5965 = None
	        round_124: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5968);  mul_5968 = None
	        add_9451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_124, view_959);  round_124 = view_959 = None
	        clamp_min_185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9451, -128);  add_9451 = None
	        clamp_max_123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_185, 127);  clamp_min_185 = None
	        _assert_tensor_metadata_553 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_123, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_553 = None
	        convert_element_type_367: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_123, torch.int8);  clamp_max_123 = None
	        view_962: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_183, [sym_size_int, 1500, 1]);  clamp_min_183 = None
	        view_963: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_366, [sym_size_int, 1500, 1]);  convert_element_type_366 = None
	        _assert_tensor_metadata_554 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_367, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_554 = None
	        convert_element_type_368: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_367, torch.float32);  convert_element_type_367 = None
	        _assert_tensor_metadata_555 = torch.ops.aten._assert_tensor_metadata.default(view_963, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_555 = None
	        convert_element_type_369: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_963, torch.float32);  view_963 = None
	        sub_2829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_368, convert_element_type_369);  convert_element_type_368 = convert_element_type_369 = None
	        mul_5990: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2829, view_962);  sub_2829 = view_962 = None
	        _assert_tensor_metadata_556 = torch.ops.aten._assert_tensor_metadata.default(mul_5990, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_556 = None
	        view_965: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_966: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_967: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_557 = torch.ops.aten._assert_tensor_metadata.default(view_965, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_557 = None
	        convert_element_type_370: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_965, torch.float32);  view_965 = None
	        _assert_tensor_metadata_558 = torch.ops.aten._assert_tensor_metadata.default(view_967, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_558 = None
	        convert_element_type_371: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_967, torch.float32);  view_967 = None
	        sub_2833: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_370, convert_element_type_371);  convert_element_type_370 = convert_element_type_371 = None
	        mul_5995: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2833, view_966);  sub_2833 = view_966 = None
	        view_968: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5995, [1280, 1280]);  mul_5995 = None
	        _assert_tensor_metadata_559 = torch.ops.aten._assert_tensor_metadata.default(view_968, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_559 = None
	        permute_103: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_968, [1, 0]);  view_968 = None
	        mul_5998: "Sym(1500*s6)" = sym_size_int * 1500
	        view_969: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_5990, [mul_5998, 1280]);  mul_5990 = mul_5998 = None
	        mm_10: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_969, permute_103);  view_969 = permute_103 = None
	        view_970: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_10, [sym_size_int, 1500, 1280]);  mm_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_971: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_970, [sym_size_int, -1, 20, 64]);  view_970 = None
	        permute_104: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_971, [0, 2, 1, 3]);  view_971 = None
	        clone_83: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_104, memory_format = torch.contiguous_format);  permute_104 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_9212, [2])
	        amax_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_9212, [2])
	        full_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_62, full_124);  amin_62 = full_124 = None
	        full_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_62, full_125);  amax_62 = full_125 = None
	        sub_2847: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_62, minimum_62);  maximum_62 = None
	        div_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2847, 255.0);  sub_2847 = None
	        clamp_min_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_124, 1.1920928955078125e-07);  div_124 = None
	        div_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_62, clamp_min_186);  minimum_62 = None
	        round_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_125);  div_125 = None
	        sub_2853: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_125);  round_125 = None
	        clamp_min_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2853, -128);  sub_2853 = None
	        clamp_max_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_187, 127);  clamp_min_187 = None
	        _assert_tensor_metadata_560 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_186, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_560 = None
	        _assert_tensor_metadata_561 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_124, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_561 = None
	        convert_element_type_372: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_124, torch.int8);  clamp_max_124 = None
	        view_974: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_186, [sym_size_int, 1500, 1])
	        view_975: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_372, [sym_size_int, 1500, 1])
	        reciprocal_62: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_974);  view_974 = None
	        mul_6064: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_62, 1.0);  reciprocal_62 = None
	        mul_6067: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_9212, mul_6064);  add_9212 = mul_6064 = None
	        round_126: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6067);  mul_6067 = None
	        add_9599: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_126, view_975);  round_126 = view_975 = None
	        clamp_min_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9599, -128);  add_9599 = None
	        clamp_max_125: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_188, 127);  clamp_min_188 = None
	        _assert_tensor_metadata_562 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_125, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_562 = None
	        convert_element_type_373: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_125, torch.int8);  clamp_max_125 = None
	        view_978: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_186, [sym_size_int, 1500, 1]);  clamp_min_186 = None
	        view_979: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_372, [sym_size_int, 1500, 1]);  convert_element_type_372 = None
	        _assert_tensor_metadata_563 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_373, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_563 = None
	        convert_element_type_374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_373, torch.float32);  convert_element_type_373 = None
	        _assert_tensor_metadata_564 = torch.ops.aten._assert_tensor_metadata.default(view_979, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_564 = None
	        convert_element_type_375: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_979, torch.float32);  view_979 = None
	        sub_2873: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_374, convert_element_type_375);  convert_element_type_374 = convert_element_type_375 = None
	        mul_6089: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2873, view_978);  sub_2873 = view_978 = None
	        _assert_tensor_metadata_565 = torch.ops.aten._assert_tensor_metadata.default(mul_6089, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_565 = None
	        view_981: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_982: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_983: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_566 = torch.ops.aten._assert_tensor_metadata.default(view_981, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_566 = None
	        convert_element_type_376: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_981, torch.float32);  view_981 = None
	        _assert_tensor_metadata_567 = torch.ops.aten._assert_tensor_metadata.default(view_983, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_567 = None
	        convert_element_type_377: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_983, torch.float32);  view_983 = None
	        sub_2877: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_376, convert_element_type_377);  convert_element_type_376 = convert_element_type_377 = None
	        mul_6094: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2877, view_982);  sub_2877 = view_982 = None
	        view_984: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6094, [1280, 1280]);  mul_6094 = None
	        _assert_tensor_metadata_568 = torch.ops.aten._assert_tensor_metadata.default(view_984, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_568 = None
	        mul_6099: "Sym(1500*s6)" = sym_size_int * 1500
	        view_985: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6089, [mul_6099, 1280]);  mul_6089 = mul_6099 = None
	        permute_105: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_984, [1, 0]);  view_984 = None
	        addmm_51: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_10_self_attn_v_proj_bias, view_985, permute_105);  model_audio_tower_layers_10_self_attn_v_proj_bias = view_985 = permute_105 = None
	        view_986: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_51, [sym_size_int, 1500, 1280]);  addmm_51 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_987: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_986, [sym_size_int, -1, 20, 64]);  view_986 = None
	        permute_106: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_987, [0, 2, 1, 3]);  view_987 = None
	        clone_84: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_106, memory_format = torch.contiguous_format);  permute_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_10 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_82, clone_83, clone_84, None, False, scale = 1.0);  clone_82 = clone_83 = clone_84 = None
	        getitem_82: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_10[0];  _scaled_dot_product_efficient_attention_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_107: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_82, [0, 2, 1, 3]);  getitem_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_988: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_107, [sym_size_int, 1500, -1]);  permute_107 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_988, [2])
	        amax_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_988, [2])
	        full_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_63, full_126);  amin_63 = full_126 = None
	        full_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_63, full_127);  amax_63 = full_127 = None
	        sub_2895: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_63, minimum_63);  maximum_63 = None
	        div_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2895, 255.0);  sub_2895 = None
	        clamp_min_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_126, 1.1920928955078125e-07);  div_126 = None
	        div_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_63, clamp_min_189);  minimum_63 = None
	        round_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_127);  div_127 = None
	        sub_2901: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_127);  round_127 = None
	        clamp_min_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2901, -128);  sub_2901 = None
	        clamp_max_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_190, 127);  clamp_min_190 = None
	        _assert_tensor_metadata_569 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_189, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_569 = None
	        _assert_tensor_metadata_570 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_126, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_570 = None
	        convert_element_type_378: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_126, torch.int8);  clamp_max_126 = None
	        view_991: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_189, [sym_size_int, 1500, 1])
	        view_992: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_378, [sym_size_int, 1500, 1])
	        reciprocal_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_991);  view_991 = None
	        mul_6169: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_63, 1.0);  reciprocal_63 = None
	        mul_6172: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_988, mul_6169);  view_988 = mul_6169 = None
	        round_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6172);  mul_6172 = None
	        add_9763: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_128, view_992);  round_128 = view_992 = None
	        clamp_min_191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9763, -128);  add_9763 = None
	        clamp_max_127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_191, 127);  clamp_min_191 = None
	        _assert_tensor_metadata_571 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_127, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_571 = None
	        convert_element_type_379: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_127, torch.int8);  clamp_max_127 = None
	        view_995: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_189, [sym_size_int, 1500, 1]);  clamp_min_189 = None
	        view_996: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_378, [sym_size_int, 1500, 1]);  convert_element_type_378 = None
	        _assert_tensor_metadata_572 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_379, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_572 = None
	        convert_element_type_380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_379, torch.float32);  convert_element_type_379 = None
	        _assert_tensor_metadata_573 = torch.ops.aten._assert_tensor_metadata.default(view_996, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_573 = None
	        convert_element_type_381: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_996, torch.float32);  view_996 = None
	        sub_2921: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_380, convert_element_type_381);  convert_element_type_380 = convert_element_type_381 = None
	        mul_6194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2921, view_995);  sub_2921 = view_995 = None
	        _assert_tensor_metadata_574 = torch.ops.aten._assert_tensor_metadata.default(mul_6194, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_574 = None
	        view_998: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_999: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1000: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_575 = torch.ops.aten._assert_tensor_metadata.default(view_998, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_575 = None
	        convert_element_type_382: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_998, torch.float32);  view_998 = None
	        _assert_tensor_metadata_576 = torch.ops.aten._assert_tensor_metadata.default(view_1000, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_576 = None
	        convert_element_type_383: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1000, torch.float32);  view_1000 = None
	        sub_2925: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_382, convert_element_type_383);  convert_element_type_382 = convert_element_type_383 = None
	        mul_6199: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2925, view_999);  sub_2925 = view_999 = None
	        view_1001: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6199, [1280, 1280]);  mul_6199 = None
	        _assert_tensor_metadata_577 = torch.ops.aten._assert_tensor_metadata.default(view_1001, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_577 = None
	        mul_6204: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1002: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6194, [mul_6204, 1280]);  mul_6194 = mul_6204 = None
	        permute_108: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1001, [1, 0]);  view_1001 = None
	        addmm_52: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_10_self_attn_out_proj_bias, view_1002, permute_108);  model_audio_tower_layers_10_self_attn_out_proj_bias = view_1002 = permute_108 = None
	        view_1003: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_52, [sym_size_int, 1500, 1280]);  addmm_52 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_9826: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_9206, view_1003);  add_9206 = view_1003 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_9826, memory_format = torch.contiguous_format)
	        var_mean_21 = torch.ops.aten.var_mean.correction(clone_86, [2], correction = 0, keepdim = True)
	        getitem_86: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_21[0]
	        getitem_87: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_21[1];  var_mean_21 = None
	        add_9831: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_86, 1e-05);  getitem_86 = None
	        rsqrt_21: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_9831);  add_9831 = None
	        sub_2931: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_86, getitem_87);  clone_86 = getitem_87 = None
	        mul_6215: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2931, rsqrt_21);  sub_2931 = rsqrt_21 = None
	        mul_6216: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6215, model_audio_tower_layers_10_final_layer_norm_weight);  mul_6215 = model_audio_tower_layers_10_final_layer_norm_weight = None
	        add_9832: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6216, model_audio_tower_layers_10_final_layer_norm_bias);  mul_6216 = model_audio_tower_layers_10_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_9832, [2])
	        amax_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_9832, [2])
	        full_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_64, full_128);  amin_64 = full_128 = None
	        full_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_64, full_129);  amax_64 = full_129 = None
	        sub_2942: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_64, minimum_64);  maximum_64 = None
	        div_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2942, 255.0);  sub_2942 = None
	        clamp_min_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_128, 1.1920928955078125e-07);  div_128 = None
	        div_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_64, clamp_min_192);  minimum_64 = None
	        round_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_129);  div_129 = None
	        sub_2948: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_129);  round_129 = None
	        clamp_min_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2948, -128);  sub_2948 = None
	        clamp_max_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_193, 127);  clamp_min_193 = None
	        _assert_tensor_metadata_578 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_192, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_578 = None
	        _assert_tensor_metadata_579 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_128, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_579 = None
	        convert_element_type_384: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_128, torch.int8);  clamp_max_128 = None
	        view_1006: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_192, [sym_size_int, 1500, 1])
	        view_1007: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_384, [sym_size_int, 1500, 1])
	        reciprocal_64: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1006);  view_1006 = None
	        mul_6264: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_64, 1.0);  reciprocal_64 = None
	        mul_6267: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_9832, mul_6264);  add_9832 = mul_6264 = None
	        round_130: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6267);  mul_6267 = None
	        add_9919: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_130, view_1007);  round_130 = view_1007 = None
	        clamp_min_194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9919, -128);  add_9919 = None
	        clamp_max_129: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_194, 127);  clamp_min_194 = None
	        _assert_tensor_metadata_580 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_129, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_580 = None
	        convert_element_type_385: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_129, torch.int8);  clamp_max_129 = None
	        view_1010: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_192, [sym_size_int, 1500, 1]);  clamp_min_192 = None
	        view_1011: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_384, [sym_size_int, 1500, 1]);  convert_element_type_384 = None
	        _assert_tensor_metadata_581 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_385, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_581 = None
	        convert_element_type_386: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_385, torch.float32);  convert_element_type_385 = None
	        _assert_tensor_metadata_582 = torch.ops.aten._assert_tensor_metadata.default(view_1011, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_582 = None
	        convert_element_type_387: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1011, torch.float32);  view_1011 = None
	        sub_2968: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_386, convert_element_type_387);  convert_element_type_386 = convert_element_type_387 = None
	        mul_6289: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2968, view_1010);  sub_2968 = view_1010 = None
	        _assert_tensor_metadata_583 = torch.ops.aten._assert_tensor_metadata.default(mul_6289, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_583 = None
	        view_1013: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_10_fc1_parametrizations_weight_original0 = None
	        view_1014: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_10_fc1_parametrizations_weight_original1 = None
	        view_1015: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_10_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_584 = torch.ops.aten._assert_tensor_metadata.default(view_1013, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_584 = None
	        convert_element_type_388: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1013, torch.float32);  view_1013 = None
	        _assert_tensor_metadata_585 = torch.ops.aten._assert_tensor_metadata.default(view_1015, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_585 = None
	        convert_element_type_389: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1015, torch.float32);  view_1015 = None
	        sub_2972: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_388, convert_element_type_389);  convert_element_type_388 = convert_element_type_389 = None
	        mul_6294: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2972, view_1014);  sub_2972 = view_1014 = None
	        view_1016: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6294, [5120, 1280]);  mul_6294 = None
	        _assert_tensor_metadata_586 = torch.ops.aten._assert_tensor_metadata.default(view_1016, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_586 = None
	        mul_6299: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1017: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6289, [mul_6299, 1280]);  mul_6289 = mul_6299 = None
	        permute_109: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1016, [1, 0]);  view_1016 = None
	        addmm_53: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_10_fc1_bias, view_1017, permute_109);  model_audio_tower_layers_10_fc1_bias = view_1017 = permute_109 = None
	        view_1018: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_53, [sym_size_int, 1500, 5120]);  addmm_53 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_6306: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1018, 0.5)
	        mul_6307: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1018, 0.7071067811865476);  view_1018 = None
	        erf_12: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_6307);  mul_6307 = None
	        add_9978: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_12, 1);  erf_12 = None
	        mul_6308: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6306, add_9978);  mul_6306 = add_9978 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_6308, [2])
	        amax_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_6308, [2])
	        full_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_65, full_130);  amin_65 = full_130 = None
	        full_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_65, full_131);  amax_65 = full_131 = None
	        sub_2985: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_65, minimum_65);  maximum_65 = None
	        div_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2985, 255.0);  sub_2985 = None
	        clamp_min_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_130, 1.1920928955078125e-07);  div_130 = None
	        div_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_65, clamp_min_195);  minimum_65 = None
	        round_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_131);  div_131 = None
	        sub_2991: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_131);  round_131 = None
	        clamp_min_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2991, -128);  sub_2991 = None
	        clamp_max_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_196, 127);  clamp_min_196 = None
	        _assert_tensor_metadata_587 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_195, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_587 = None
	        _assert_tensor_metadata_588 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_130, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_588 = None
	        convert_element_type_390: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_130, torch.int8);  clamp_max_130 = None
	        view_1021: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_195, [sym_size_int, 1500, 1])
	        view_1022: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_390, [sym_size_int, 1500, 1])
	        reciprocal_65: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1021);  view_1021 = None
	        mul_6354: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_65, 1.0);  reciprocal_65 = None
	        mul_6357: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6308, mul_6354);  mul_6308 = mul_6354 = None
	        round_132: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_6357);  mul_6357 = None
	        add_10061: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_132, view_1022);  round_132 = view_1022 = None
	        clamp_min_197: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10061, -128);  add_10061 = None
	        clamp_max_131: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_197, 127);  clamp_min_197 = None
	        _assert_tensor_metadata_589 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_131, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_589 = None
	        convert_element_type_391: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_131, torch.int8);  clamp_max_131 = None
	        view_1025: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_195, [sym_size_int, 1500, 1]);  clamp_min_195 = None
	        view_1026: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_390, [sym_size_int, 1500, 1]);  convert_element_type_390 = None
	        _assert_tensor_metadata_590 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_391, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_590 = None
	        convert_element_type_392: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_391, torch.float32);  convert_element_type_391 = None
	        _assert_tensor_metadata_591 = torch.ops.aten._assert_tensor_metadata.default(view_1026, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_591 = None
	        convert_element_type_393: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1026, torch.float32);  view_1026 = None
	        sub_3011: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_392, convert_element_type_393);  convert_element_type_392 = convert_element_type_393 = None
	        mul_6379: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3011, view_1025);  sub_3011 = view_1025 = None
	        _assert_tensor_metadata_592 = torch.ops.aten._assert_tensor_metadata.default(mul_6379, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_592 = None
	        view_1028: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_10_fc2_parametrizations_weight_original0 = None
	        view_1029: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_10_fc2_parametrizations_weight_original1 = None
	        view_1030: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_10_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_10_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_593 = torch.ops.aten._assert_tensor_metadata.default(view_1028, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_593 = None
	        convert_element_type_394: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1028, torch.float32);  view_1028 = None
	        _assert_tensor_metadata_594 = torch.ops.aten._assert_tensor_metadata.default(view_1030, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_594 = None
	        convert_element_type_395: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1030, torch.float32);  view_1030 = None
	        sub_3015: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_394, convert_element_type_395);  convert_element_type_394 = convert_element_type_395 = None
	        mul_6384: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3015, view_1029);  sub_3015 = view_1029 = None
	        view_1031: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_6384, [1280, 5120]);  mul_6384 = None
	        _assert_tensor_metadata_595 = torch.ops.aten._assert_tensor_metadata.default(view_1031, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_595 = None
	        mul_6389: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1032: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_6379, [mul_6389, 5120]);  mul_6379 = mul_6389 = None
	        permute_110: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1031, [1, 0]);  view_1031 = None
	        addmm_54: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_10_fc2_bias, view_1032, permute_110);  model_audio_tower_layers_10_fc2_bias = view_1032 = permute_110 = None
	        view_1033: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_54, [sym_size_int, 1500, 1280]);  addmm_54 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_10124: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_9826, view_1033);  add_9826 = view_1033 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_89: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_10124, memory_format = torch.contiguous_format)
	        var_mean_22 = torch.ops.aten.var_mean.correction(clone_89, [2], correction = 0, keepdim = True)
	        getitem_88: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_22[0]
	        getitem_89: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_22[1];  var_mean_22 = None
	        add_10129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_88, 1e-05);  getitem_88 = None
	        rsqrt_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_10129);  add_10129 = None
	        sub_3021: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_89, getitem_89);  clone_89 = getitem_89 = None
	        mul_6400: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3021, rsqrt_22);  sub_3021 = rsqrt_22 = None
	        mul_6401: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6400, model_audio_tower_layers_11_self_attn_layer_norm_weight);  mul_6400 = model_audio_tower_layers_11_self_attn_layer_norm_weight = None
	        add_10130: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6401, model_audio_tower_layers_11_self_attn_layer_norm_bias);  mul_6401 = model_audio_tower_layers_11_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_10130, [2])
	        amax_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_10130, [2])
	        full_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_66, full_132);  amin_66 = full_132 = None
	        full_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_66, full_133);  amax_66 = full_133 = None
	        sub_3032: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_66, minimum_66);  maximum_66 = None
	        div_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3032, 255.0);  sub_3032 = None
	        clamp_min_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_132, 1.1920928955078125e-07);  div_132 = None
	        div_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_66, clamp_min_198);  minimum_66 = None
	        round_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_133);  div_133 = None
	        sub_3038: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_133);  round_133 = None
	        clamp_min_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3038, -128);  sub_3038 = None
	        clamp_max_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_199, 127);  clamp_min_199 = None
	        _assert_tensor_metadata_596 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_198, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_596 = None
	        _assert_tensor_metadata_597 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_132, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_597 = None
	        convert_element_type_396: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_132, torch.int8);  clamp_max_132 = None
	        view_1036: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_198, [sym_size_int, 1500, 1])
	        view_1037: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_396, [sym_size_int, 1500, 1])
	        reciprocal_66: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1036);  view_1036 = None
	        mul_6449: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_66, 1.0);  reciprocal_66 = None
	        mul_6452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_10130, mul_6449);  mul_6449 = None
	        round_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6452);  mul_6452 = None
	        add_10217: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_134, view_1037);  round_134 = view_1037 = None
	        clamp_min_200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10217, -128);  add_10217 = None
	        clamp_max_133: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_200, 127);  clamp_min_200 = None
	        _assert_tensor_metadata_598 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_133, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_598 = None
	        convert_element_type_397: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_133, torch.int8);  clamp_max_133 = None
	        view_1040: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_198, [sym_size_int, 1500, 1]);  clamp_min_198 = None
	        view_1041: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_396, [sym_size_int, 1500, 1]);  convert_element_type_396 = None
	        _assert_tensor_metadata_599 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_397, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_599 = None
	        convert_element_type_398: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_397, torch.float32);  convert_element_type_397 = None
	        _assert_tensor_metadata_600 = torch.ops.aten._assert_tensor_metadata.default(view_1041, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_600 = None
	        convert_element_type_399: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1041, torch.float32);  view_1041 = None
	        sub_3058: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_398, convert_element_type_399);  convert_element_type_398 = convert_element_type_399 = None
	        mul_6474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3058, view_1040);  sub_3058 = view_1040 = None
	        _assert_tensor_metadata_601 = torch.ops.aten._assert_tensor_metadata.default(mul_6474, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_601 = None
	        view_1043: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1044: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1045: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_602 = torch.ops.aten._assert_tensor_metadata.default(view_1043, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_602 = None
	        convert_element_type_400: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1043, torch.float32);  view_1043 = None
	        _assert_tensor_metadata_603 = torch.ops.aten._assert_tensor_metadata.default(view_1045, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_603 = None
	        convert_element_type_401: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1045, torch.float32);  view_1045 = None
	        sub_3062: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_400, convert_element_type_401);  convert_element_type_400 = convert_element_type_401 = None
	        mul_6479: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3062, view_1044);  sub_3062 = view_1044 = None
	        view_1046: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6479, [1280, 1280]);  mul_6479 = None
	        _assert_tensor_metadata_604 = torch.ops.aten._assert_tensor_metadata.default(view_1046, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_604 = None
	        mul_6484: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1047: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6474, [mul_6484, 1280]);  mul_6474 = mul_6484 = None
	        permute_111: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1046, [1, 0]);  view_1046 = None
	        addmm_55: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_11_self_attn_q_proj_bias, view_1047, permute_111);  model_audio_tower_layers_11_self_attn_q_proj_bias = view_1047 = permute_111 = None
	        view_1048: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_55, [sym_size_int, 1500, 1280]);  addmm_55 = None
	        mul_6491: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1048, 0.125);  view_1048 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1049: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_6491, [sym_size_int, 1500, 20, 64]);  mul_6491 = None
	        permute_112: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1049, [0, 2, 1, 3]);  view_1049 = None
	        clone_90: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_112, memory_format = torch.contiguous_format);  permute_112 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_10130, [2])
	        amax_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_10130, [2])
	        full_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_67, full_134);  amin_67 = full_134 = None
	        full_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_67, full_135);  amax_67 = full_135 = None
	        sub_3077: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_67, minimum_67);  maximum_67 = None
	        div_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3077, 255.0);  sub_3077 = None
	        clamp_min_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_134, 1.1920928955078125e-07);  div_134 = None
	        div_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_67, clamp_min_201);  minimum_67 = None
	        round_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_135);  div_135 = None
	        sub_3083: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_135);  round_135 = None
	        clamp_min_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3083, -128);  sub_3083 = None
	        clamp_max_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_202, 127);  clamp_min_202 = None
	        _assert_tensor_metadata_605 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_201, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_605 = None
	        _assert_tensor_metadata_606 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_134, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_606 = None
	        convert_element_type_402: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_134, torch.int8);  clamp_max_134 = None
	        view_1052: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_201, [sym_size_int, 1500, 1])
	        view_1053: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_402, [sym_size_int, 1500, 1])
	        reciprocal_67: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1052);  view_1052 = None
	        mul_6545: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_67, 1.0);  reciprocal_67 = None
	        mul_6548: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_10130, mul_6545);  mul_6545 = None
	        round_136: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6548);  mul_6548 = None
	        add_10369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_136, view_1053);  round_136 = view_1053 = None
	        clamp_min_203: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10369, -128);  add_10369 = None
	        clamp_max_135: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_203, 127);  clamp_min_203 = None
	        _assert_tensor_metadata_607 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_135, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_607 = None
	        convert_element_type_403: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_135, torch.int8);  clamp_max_135 = None
	        view_1056: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_201, [sym_size_int, 1500, 1]);  clamp_min_201 = None
	        view_1057: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_402, [sym_size_int, 1500, 1]);  convert_element_type_402 = None
	        _assert_tensor_metadata_608 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_403, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_608 = None
	        convert_element_type_404: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_403, torch.float32);  convert_element_type_403 = None
	        _assert_tensor_metadata_609 = torch.ops.aten._assert_tensor_metadata.default(view_1057, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_609 = None
	        convert_element_type_405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1057, torch.float32);  view_1057 = None
	        sub_3103: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_404, convert_element_type_405);  convert_element_type_404 = convert_element_type_405 = None
	        mul_6570: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3103, view_1056);  sub_3103 = view_1056 = None
	        _assert_tensor_metadata_610 = torch.ops.aten._assert_tensor_metadata.default(mul_6570, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_610 = None
	        view_1059: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1060: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1061: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_611 = torch.ops.aten._assert_tensor_metadata.default(view_1059, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_611 = None
	        convert_element_type_406: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1059, torch.float32);  view_1059 = None
	        _assert_tensor_metadata_612 = torch.ops.aten._assert_tensor_metadata.default(view_1061, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_612 = None
	        convert_element_type_407: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1061, torch.float32);  view_1061 = None
	        sub_3107: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_406, convert_element_type_407);  convert_element_type_406 = convert_element_type_407 = None
	        mul_6575: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3107, view_1060);  sub_3107 = view_1060 = None
	        view_1062: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6575, [1280, 1280]);  mul_6575 = None
	        _assert_tensor_metadata_613 = torch.ops.aten._assert_tensor_metadata.default(view_1062, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_613 = None
	        permute_113: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1062, [1, 0]);  view_1062 = None
	        mul_6578: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1063: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6570, [mul_6578, 1280]);  mul_6570 = mul_6578 = None
	        mm_11: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1063, permute_113);  view_1063 = permute_113 = None
	        view_1064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_11, [sym_size_int, 1500, 1280]);  mm_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1065: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1064, [sym_size_int, -1, 20, 64]);  view_1064 = None
	        permute_114: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1065, [0, 2, 1, 3]);  view_1065 = None
	        clone_91: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_114, memory_format = torch.contiguous_format);  permute_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_10130, [2])
	        amax_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_10130, [2])
	        full_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_68, full_136);  amin_68 = full_136 = None
	        full_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_68, full_137);  amax_68 = full_137 = None
	        sub_3121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_68, minimum_68);  maximum_68 = None
	        div_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3121, 255.0);  sub_3121 = None
	        clamp_min_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_136, 1.1920928955078125e-07);  div_136 = None
	        div_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_68, clamp_min_204);  minimum_68 = None
	        round_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_137);  div_137 = None
	        sub_3127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_137);  round_137 = None
	        clamp_min_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3127, -128);  sub_3127 = None
	        clamp_max_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_205, 127);  clamp_min_205 = None
	        _assert_tensor_metadata_614 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_204, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_614 = None
	        _assert_tensor_metadata_615 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_136, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_615 = None
	        convert_element_type_408: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_136, torch.int8);  clamp_max_136 = None
	        view_1068: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_204, [sym_size_int, 1500, 1])
	        view_1069: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_408, [sym_size_int, 1500, 1])
	        reciprocal_68: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1068);  view_1068 = None
	        mul_6644: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_68, 1.0);  reciprocal_68 = None
	        mul_6647: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_10130, mul_6644);  add_10130 = mul_6644 = None
	        round_138: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6647);  mul_6647 = None
	        add_10517: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_138, view_1069);  round_138 = view_1069 = None
	        clamp_min_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10517, -128);  add_10517 = None
	        clamp_max_137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_206, 127);  clamp_min_206 = None
	        _assert_tensor_metadata_616 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_137, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_616 = None
	        convert_element_type_409: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_137, torch.int8);  clamp_max_137 = None
	        view_1072: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_204, [sym_size_int, 1500, 1]);  clamp_min_204 = None
	        view_1073: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_408, [sym_size_int, 1500, 1]);  convert_element_type_408 = None
	        _assert_tensor_metadata_617 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_409, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_617 = None
	        convert_element_type_410: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_409, torch.float32);  convert_element_type_409 = None
	        _assert_tensor_metadata_618 = torch.ops.aten._assert_tensor_metadata.default(view_1073, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_618 = None
	        convert_element_type_411: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1073, torch.float32);  view_1073 = None
	        sub_3147: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_410, convert_element_type_411);  convert_element_type_410 = convert_element_type_411 = None
	        mul_6669: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3147, view_1072);  sub_3147 = view_1072 = None
	        _assert_tensor_metadata_619 = torch.ops.aten._assert_tensor_metadata.default(mul_6669, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_619 = None
	        view_1075: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1076: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1077: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_620 = torch.ops.aten._assert_tensor_metadata.default(view_1075, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_620 = None
	        convert_element_type_412: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1075, torch.float32);  view_1075 = None
	        _assert_tensor_metadata_621 = torch.ops.aten._assert_tensor_metadata.default(view_1077, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_621 = None
	        convert_element_type_413: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1077, torch.float32);  view_1077 = None
	        sub_3151: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_412, convert_element_type_413);  convert_element_type_412 = convert_element_type_413 = None
	        mul_6674: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3151, view_1076);  sub_3151 = view_1076 = None
	        view_1078: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6674, [1280, 1280]);  mul_6674 = None
	        _assert_tensor_metadata_622 = torch.ops.aten._assert_tensor_metadata.default(view_1078, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_622 = None
	        mul_6679: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1079: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6669, [mul_6679, 1280]);  mul_6669 = mul_6679 = None
	        permute_115: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1078, [1, 0]);  view_1078 = None
	        addmm_56: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_11_self_attn_v_proj_bias, view_1079, permute_115);  model_audio_tower_layers_11_self_attn_v_proj_bias = view_1079 = permute_115 = None
	        view_1080: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_56, [sym_size_int, 1500, 1280]);  addmm_56 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1081: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1080, [sym_size_int, -1, 20, 64]);  view_1080 = None
	        permute_116: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1081, [0, 2, 1, 3]);  view_1081 = None
	        clone_92: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_116, memory_format = torch.contiguous_format);  permute_116 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_11 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_90, clone_91, clone_92, None, False, scale = 1.0);  clone_90 = clone_91 = clone_92 = None
	        getitem_90: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_11[0];  _scaled_dot_product_efficient_attention_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_117: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_90, [0, 2, 1, 3]);  getitem_90 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1082: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_117, [sym_size_int, 1500, -1]);  permute_117 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1082, [2])
	        amax_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1082, [2])
	        full_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_69, full_138);  amin_69 = full_138 = None
	        full_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_69, full_139);  amax_69 = full_139 = None
	        sub_3169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_69, minimum_69);  maximum_69 = None
	        div_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3169, 255.0);  sub_3169 = None
	        clamp_min_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_138, 1.1920928955078125e-07);  div_138 = None
	        div_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_69, clamp_min_207);  minimum_69 = None
	        round_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_139);  div_139 = None
	        sub_3175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_139);  round_139 = None
	        clamp_min_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3175, -128);  sub_3175 = None
	        clamp_max_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_208, 127);  clamp_min_208 = None
	        _assert_tensor_metadata_623 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_207, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_623 = None
	        _assert_tensor_metadata_624 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_138, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_624 = None
	        convert_element_type_414: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_138, torch.int8);  clamp_max_138 = None
	        view_1085: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_207, [sym_size_int, 1500, 1])
	        view_1086: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_414, [sym_size_int, 1500, 1])
	        reciprocal_69: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1085);  view_1085 = None
	        mul_6749: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_69, 1.0);  reciprocal_69 = None
	        mul_6752: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1082, mul_6749);  view_1082 = mul_6749 = None
	        round_140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6752);  mul_6752 = None
	        add_10681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_140, view_1086);  round_140 = view_1086 = None
	        clamp_min_209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10681, -128);  add_10681 = None
	        clamp_max_139: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_209, 127);  clamp_min_209 = None
	        _assert_tensor_metadata_625 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_139, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_625 = None
	        convert_element_type_415: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_139, torch.int8);  clamp_max_139 = None
	        view_1089: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_207, [sym_size_int, 1500, 1]);  clamp_min_207 = None
	        view_1090: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_414, [sym_size_int, 1500, 1]);  convert_element_type_414 = None
	        _assert_tensor_metadata_626 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_415, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_626 = None
	        convert_element_type_416: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_415, torch.float32);  convert_element_type_415 = None
	        _assert_tensor_metadata_627 = torch.ops.aten._assert_tensor_metadata.default(view_1090, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_627 = None
	        convert_element_type_417: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1090, torch.float32);  view_1090 = None
	        sub_3195: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_416, convert_element_type_417);  convert_element_type_416 = convert_element_type_417 = None
	        mul_6774: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3195, view_1089);  sub_3195 = view_1089 = None
	        _assert_tensor_metadata_628 = torch.ops.aten._assert_tensor_metadata.default(mul_6774, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_628 = None
	        view_1092: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1093: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1094: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_629 = torch.ops.aten._assert_tensor_metadata.default(view_1092, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_629 = None
	        convert_element_type_418: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1092, torch.float32);  view_1092 = None
	        _assert_tensor_metadata_630 = torch.ops.aten._assert_tensor_metadata.default(view_1094, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_630 = None
	        convert_element_type_419: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1094, torch.float32);  view_1094 = None
	        sub_3199: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_418, convert_element_type_419);  convert_element_type_418 = convert_element_type_419 = None
	        mul_6779: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3199, view_1093);  sub_3199 = view_1093 = None
	        view_1095: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6779, [1280, 1280]);  mul_6779 = None
	        _assert_tensor_metadata_631 = torch.ops.aten._assert_tensor_metadata.default(view_1095, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_631 = None
	        mul_6784: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1096: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6774, [mul_6784, 1280]);  mul_6774 = mul_6784 = None
	        permute_118: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1095, [1, 0]);  view_1095 = None
	        addmm_57: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_11_self_attn_out_proj_bias, view_1096, permute_118);  model_audio_tower_layers_11_self_attn_out_proj_bias = view_1096 = permute_118 = None
	        view_1097: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_57, [sym_size_int, 1500, 1280]);  addmm_57 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_10744: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_10124, view_1097);  add_10124 = view_1097 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_94: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_10744, memory_format = torch.contiguous_format)
	        var_mean_23 = torch.ops.aten.var_mean.correction(clone_94, [2], correction = 0, keepdim = True)
	        getitem_94: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_23[0]
	        getitem_95: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_23[1];  var_mean_23 = None
	        add_10749: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_94, 1e-05);  getitem_94 = None
	        rsqrt_23: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_10749);  add_10749 = None
	        sub_3205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_94, getitem_95);  clone_94 = getitem_95 = None
	        mul_6795: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3205, rsqrt_23);  sub_3205 = rsqrt_23 = None
	        mul_6796: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6795, model_audio_tower_layers_11_final_layer_norm_weight);  mul_6795 = model_audio_tower_layers_11_final_layer_norm_weight = None
	        add_10750: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6796, model_audio_tower_layers_11_final_layer_norm_bias);  mul_6796 = model_audio_tower_layers_11_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_10750, [2])
	        amax_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_10750, [2])
	        full_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_70, full_140);  amin_70 = full_140 = None
	        full_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_70, full_141);  amax_70 = full_141 = None
	        sub_3216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_70, minimum_70);  maximum_70 = None
	        div_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3216, 255.0);  sub_3216 = None
	        clamp_min_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_140, 1.1920928955078125e-07);  div_140 = None
	        div_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_70, clamp_min_210);  minimum_70 = None
	        round_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_141);  div_141 = None
	        sub_3222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_141);  round_141 = None
	        clamp_min_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3222, -128);  sub_3222 = None
	        clamp_max_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_211, 127);  clamp_min_211 = None
	        _assert_tensor_metadata_632 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_210, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_632 = None
	        _assert_tensor_metadata_633 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_140, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_633 = None
	        convert_element_type_420: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_140, torch.int8);  clamp_max_140 = None
	        view_1100: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_210, [sym_size_int, 1500, 1])
	        view_1101: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_420, [sym_size_int, 1500, 1])
	        reciprocal_70: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1100);  view_1100 = None
	        mul_6844: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_70, 1.0);  reciprocal_70 = None
	        mul_6847: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_10750, mul_6844);  add_10750 = mul_6844 = None
	        round_142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6847);  mul_6847 = None
	        add_10837: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_142, view_1101);  round_142 = view_1101 = None
	        clamp_min_212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10837, -128);  add_10837 = None
	        clamp_max_141: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_212, 127);  clamp_min_212 = None
	        _assert_tensor_metadata_634 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_141, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_634 = None
	        convert_element_type_421: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_141, torch.int8);  clamp_max_141 = None
	        view_1104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_210, [sym_size_int, 1500, 1]);  clamp_min_210 = None
	        view_1105: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_420, [sym_size_int, 1500, 1]);  convert_element_type_420 = None
	        _assert_tensor_metadata_635 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_421, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_635 = None
	        convert_element_type_422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_421, torch.float32);  convert_element_type_421 = None
	        _assert_tensor_metadata_636 = torch.ops.aten._assert_tensor_metadata.default(view_1105, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_636 = None
	        convert_element_type_423: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1105, torch.float32);  view_1105 = None
	        sub_3242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_422, convert_element_type_423);  convert_element_type_422 = convert_element_type_423 = None
	        mul_6869: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3242, view_1104);  sub_3242 = view_1104 = None
	        _assert_tensor_metadata_637 = torch.ops.aten._assert_tensor_metadata.default(mul_6869, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_637 = None
	        view_1107: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_11_fc1_parametrizations_weight_original0 = None
	        view_1108: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_11_fc1_parametrizations_weight_original1 = None
	        view_1109: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_11_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_638 = torch.ops.aten._assert_tensor_metadata.default(view_1107, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_638 = None
	        convert_element_type_424: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1107, torch.float32);  view_1107 = None
	        _assert_tensor_metadata_639 = torch.ops.aten._assert_tensor_metadata.default(view_1109, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_639 = None
	        convert_element_type_425: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1109, torch.float32);  view_1109 = None
	        sub_3246: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_424, convert_element_type_425);  convert_element_type_424 = convert_element_type_425 = None
	        mul_6874: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3246, view_1108);  sub_3246 = view_1108 = None
	        view_1110: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6874, [5120, 1280]);  mul_6874 = None
	        _assert_tensor_metadata_640 = torch.ops.aten._assert_tensor_metadata.default(view_1110, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_640 = None
	        mul_6879: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1111: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_6869, [mul_6879, 1280]);  mul_6869 = mul_6879 = None
	        permute_119: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1110, [1, 0]);  view_1110 = None
	        addmm_58: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_11_fc1_bias, view_1111, permute_119);  model_audio_tower_layers_11_fc1_bias = view_1111 = permute_119 = None
	        view_1112: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_58, [sym_size_int, 1500, 5120]);  addmm_58 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_6886: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1112, 0.5)
	        mul_6887: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1112, 0.7071067811865476);  view_1112 = None
	        erf_13: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_6887);  mul_6887 = None
	        add_10896: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_13, 1);  erf_13 = None
	        mul_6888: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6886, add_10896);  mul_6886 = add_10896 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_6888, [2])
	        amax_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_6888, [2])
	        full_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_71, full_142);  amin_71 = full_142 = None
	        full_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_71, full_143);  amax_71 = full_143 = None
	        sub_3259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_71, minimum_71);  maximum_71 = None
	        div_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3259, 255.0);  sub_3259 = None
	        clamp_min_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_142, 1.1920928955078125e-07);  div_142 = None
	        div_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_71, clamp_min_213);  minimum_71 = None
	        round_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_143);  div_143 = None
	        sub_3265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_143);  round_143 = None
	        clamp_min_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3265, -128);  sub_3265 = None
	        clamp_max_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_214, 127);  clamp_min_214 = None
	        _assert_tensor_metadata_641 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_213, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_641 = None
	        _assert_tensor_metadata_642 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_142, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_642 = None
	        convert_element_type_426: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_142, torch.int8);  clamp_max_142 = None
	        view_1115: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_213, [sym_size_int, 1500, 1])
	        view_1116: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_426, [sym_size_int, 1500, 1])
	        reciprocal_71: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1115);  view_1115 = None
	        mul_6934: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_71, 1.0);  reciprocal_71 = None
	        mul_6937: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6888, mul_6934);  mul_6888 = mul_6934 = None
	        round_144: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_6937);  mul_6937 = None
	        add_10979: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_144, view_1116);  round_144 = view_1116 = None
	        clamp_min_215: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10979, -128);  add_10979 = None
	        clamp_max_143: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_215, 127);  clamp_min_215 = None
	        _assert_tensor_metadata_643 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_143, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_643 = None
	        convert_element_type_427: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_143, torch.int8);  clamp_max_143 = None
	        view_1119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_213, [sym_size_int, 1500, 1]);  clamp_min_213 = None
	        view_1120: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_426, [sym_size_int, 1500, 1]);  convert_element_type_426 = None
	        _assert_tensor_metadata_644 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_427, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_644 = None
	        convert_element_type_428: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_427, torch.float32);  convert_element_type_427 = None
	        _assert_tensor_metadata_645 = torch.ops.aten._assert_tensor_metadata.default(view_1120, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_645 = None
	        convert_element_type_429: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1120, torch.float32);  view_1120 = None
	        sub_3285: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_428, convert_element_type_429);  convert_element_type_428 = convert_element_type_429 = None
	        mul_6959: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3285, view_1119);  sub_3285 = view_1119 = None
	        _assert_tensor_metadata_646 = torch.ops.aten._assert_tensor_metadata.default(mul_6959, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_646 = None
	        view_1122: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_11_fc2_parametrizations_weight_original0 = None
	        view_1123: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_11_fc2_parametrizations_weight_original1 = None
	        view_1124: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_11_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_11_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_647 = torch.ops.aten._assert_tensor_metadata.default(view_1122, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_647 = None
	        convert_element_type_430: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1122, torch.float32);  view_1122 = None
	        _assert_tensor_metadata_648 = torch.ops.aten._assert_tensor_metadata.default(view_1124, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_648 = None
	        convert_element_type_431: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1124, torch.float32);  view_1124 = None
	        sub_3289: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_430, convert_element_type_431);  convert_element_type_430 = convert_element_type_431 = None
	        mul_6964: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3289, view_1123);  sub_3289 = view_1123 = None
	        view_1125: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_6964, [1280, 5120]);  mul_6964 = None
	        _assert_tensor_metadata_649 = torch.ops.aten._assert_tensor_metadata.default(view_1125, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_649 = None
	        mul_6969: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1126: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_6959, [mul_6969, 5120]);  mul_6959 = mul_6969 = None
	        permute_120: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1125, [1, 0]);  view_1125 = None
	        addmm_59: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_11_fc2_bias, view_1126, permute_120);  model_audio_tower_layers_11_fc2_bias = view_1126 = permute_120 = None
	        view_1127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_59, [sym_size_int, 1500, 1280]);  addmm_59 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_11042: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_10744, view_1127);  add_10744 = view_1127 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_97: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_11042, memory_format = torch.contiguous_format)
	        var_mean_24 = torch.ops.aten.var_mean.correction(clone_97, [2], correction = 0, keepdim = True)
	        getitem_96: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_24[0]
	        getitem_97: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_24[1];  var_mean_24 = None
	        add_11047: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_96, 1e-05);  getitem_96 = None
	        rsqrt_24: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_11047);  add_11047 = None
	        sub_3295: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_97, getitem_97);  clone_97 = getitem_97 = None
	        mul_6980: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3295, rsqrt_24);  sub_3295 = rsqrt_24 = None
	        mul_6981: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6980, model_audio_tower_layers_12_self_attn_layer_norm_weight);  mul_6980 = model_audio_tower_layers_12_self_attn_layer_norm_weight = None
	        add_11048: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6981, model_audio_tower_layers_12_self_attn_layer_norm_bias);  mul_6981 = model_audio_tower_layers_12_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11048, [2])
	        amax_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11048, [2])
	        full_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_72, full_144);  amin_72 = full_144 = None
	        full_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_72, full_145);  amax_72 = full_145 = None
	        sub_3306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_72, minimum_72);  maximum_72 = None
	        div_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3306, 255.0);  sub_3306 = None
	        clamp_min_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_144, 1.1920928955078125e-07);  div_144 = None
	        div_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_72, clamp_min_216);  minimum_72 = None
	        round_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_145);  div_145 = None
	        sub_3312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_145);  round_145 = None
	        clamp_min_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3312, -128);  sub_3312 = None
	        clamp_max_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_217, 127);  clamp_min_217 = None
	        _assert_tensor_metadata_650 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_216, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_650 = None
	        _assert_tensor_metadata_651 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_144, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_651 = None
	        convert_element_type_432: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_144, torch.int8);  clamp_max_144 = None
	        view_1130: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_216, [sym_size_int, 1500, 1])
	        view_1131: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_432, [sym_size_int, 1500, 1])
	        reciprocal_72: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1130);  view_1130 = None
	        mul_7029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_72, 1.0);  reciprocal_72 = None
	        mul_7032: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11048, mul_7029);  mul_7029 = None
	        round_146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7032);  mul_7032 = None
	        add_11135: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_146, view_1131);  round_146 = view_1131 = None
	        clamp_min_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11135, -128);  add_11135 = None
	        clamp_max_145: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_218, 127);  clamp_min_218 = None
	        _assert_tensor_metadata_652 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_145, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_652 = None
	        convert_element_type_433: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_145, torch.int8);  clamp_max_145 = None
	        view_1134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_216, [sym_size_int, 1500, 1]);  clamp_min_216 = None
	        view_1135: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_432, [sym_size_int, 1500, 1]);  convert_element_type_432 = None
	        _assert_tensor_metadata_653 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_433, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_653 = None
	        convert_element_type_434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_433, torch.float32);  convert_element_type_433 = None
	        _assert_tensor_metadata_654 = torch.ops.aten._assert_tensor_metadata.default(view_1135, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_654 = None
	        convert_element_type_435: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1135, torch.float32);  view_1135 = None
	        sub_3332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_434, convert_element_type_435);  convert_element_type_434 = convert_element_type_435 = None
	        mul_7054: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3332, view_1134);  sub_3332 = view_1134 = None
	        _assert_tensor_metadata_655 = torch.ops.aten._assert_tensor_metadata.default(mul_7054, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_655 = None
	        view_1137: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1138: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1139: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_656 = torch.ops.aten._assert_tensor_metadata.default(view_1137, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_656 = None
	        convert_element_type_436: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1137, torch.float32);  view_1137 = None
	        _assert_tensor_metadata_657 = torch.ops.aten._assert_tensor_metadata.default(view_1139, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_657 = None
	        convert_element_type_437: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1139, torch.float32);  view_1139 = None
	        sub_3336: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_436, convert_element_type_437);  convert_element_type_436 = convert_element_type_437 = None
	        mul_7059: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3336, view_1138);  sub_3336 = view_1138 = None
	        view_1140: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7059, [1280, 1280]);  mul_7059 = None
	        _assert_tensor_metadata_658 = torch.ops.aten._assert_tensor_metadata.default(view_1140, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_658 = None
	        mul_7064: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1141: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7054, [mul_7064, 1280]);  mul_7054 = mul_7064 = None
	        permute_121: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1140, [1, 0]);  view_1140 = None
	        addmm_60: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_12_self_attn_q_proj_bias, view_1141, permute_121);  model_audio_tower_layers_12_self_attn_q_proj_bias = view_1141 = permute_121 = None
	        view_1142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_60, [sym_size_int, 1500, 1280]);  addmm_60 = None
	        mul_7071: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1142, 0.125);  view_1142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1143: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_7071, [sym_size_int, 1500, 20, 64]);  mul_7071 = None
	        permute_122: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1143, [0, 2, 1, 3]);  view_1143 = None
	        clone_98: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_122, memory_format = torch.contiguous_format);  permute_122 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11048, [2])
	        amax_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11048, [2])
	        full_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_73, full_146);  amin_73 = full_146 = None
	        full_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_73, full_147);  amax_73 = full_147 = None
	        sub_3351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_73, minimum_73);  maximum_73 = None
	        div_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3351, 255.0);  sub_3351 = None
	        clamp_min_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_146, 1.1920928955078125e-07);  div_146 = None
	        div_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_73, clamp_min_219);  minimum_73 = None
	        round_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_147);  div_147 = None
	        sub_3357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_147);  round_147 = None
	        clamp_min_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3357, -128);  sub_3357 = None
	        clamp_max_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_220, 127);  clamp_min_220 = None
	        _assert_tensor_metadata_659 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_219, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_659 = None
	        _assert_tensor_metadata_660 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_146, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_660 = None
	        convert_element_type_438: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_146, torch.int8);  clamp_max_146 = None
	        view_1146: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_219, [sym_size_int, 1500, 1])
	        view_1147: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_438, [sym_size_int, 1500, 1])
	        reciprocal_73: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1146);  view_1146 = None
	        mul_7125: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_73, 1.0);  reciprocal_73 = None
	        mul_7128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11048, mul_7125);  mul_7125 = None
	        round_148: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7128);  mul_7128 = None
	        add_11287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_148, view_1147);  round_148 = view_1147 = None
	        clamp_min_221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11287, -128);  add_11287 = None
	        clamp_max_147: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_221, 127);  clamp_min_221 = None
	        _assert_tensor_metadata_661 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_147, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_661 = None
	        convert_element_type_439: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_147, torch.int8);  clamp_max_147 = None
	        view_1150: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_219, [sym_size_int, 1500, 1]);  clamp_min_219 = None
	        view_1151: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_438, [sym_size_int, 1500, 1]);  convert_element_type_438 = None
	        _assert_tensor_metadata_662 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_439, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_662 = None
	        convert_element_type_440: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_439, torch.float32);  convert_element_type_439 = None
	        _assert_tensor_metadata_663 = torch.ops.aten._assert_tensor_metadata.default(view_1151, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_663 = None
	        convert_element_type_441: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1151, torch.float32);  view_1151 = None
	        sub_3377: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_440, convert_element_type_441);  convert_element_type_440 = convert_element_type_441 = None
	        mul_7150: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3377, view_1150);  sub_3377 = view_1150 = None
	        _assert_tensor_metadata_664 = torch.ops.aten._assert_tensor_metadata.default(mul_7150, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_664 = None
	        view_1153: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1154: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1155: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_665 = torch.ops.aten._assert_tensor_metadata.default(view_1153, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_665 = None
	        convert_element_type_442: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1153, torch.float32);  view_1153 = None
	        _assert_tensor_metadata_666 = torch.ops.aten._assert_tensor_metadata.default(view_1155, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_666 = None
	        convert_element_type_443: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1155, torch.float32);  view_1155 = None
	        sub_3381: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_442, convert_element_type_443);  convert_element_type_442 = convert_element_type_443 = None
	        mul_7155: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3381, view_1154);  sub_3381 = view_1154 = None
	        view_1156: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7155, [1280, 1280]);  mul_7155 = None
	        _assert_tensor_metadata_667 = torch.ops.aten._assert_tensor_metadata.default(view_1156, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_667 = None
	        permute_123: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1156, [1, 0]);  view_1156 = None
	        mul_7158: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1157: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7150, [mul_7158, 1280]);  mul_7150 = mul_7158 = None
	        mm_12: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1157, permute_123);  view_1157 = permute_123 = None
	        view_1158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_12, [sym_size_int, 1500, 1280]);  mm_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1159: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1158, [sym_size_int, -1, 20, 64]);  view_1158 = None
	        permute_124: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1159, [0, 2, 1, 3]);  view_1159 = None
	        clone_99: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_124, memory_format = torch.contiguous_format);  permute_124 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11048, [2])
	        amax_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11048, [2])
	        full_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_74, full_148);  amin_74 = full_148 = None
	        full_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_74, full_149);  amax_74 = full_149 = None
	        sub_3395: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_74, minimum_74);  maximum_74 = None
	        div_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3395, 255.0);  sub_3395 = None
	        clamp_min_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_148, 1.1920928955078125e-07);  div_148 = None
	        div_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_74, clamp_min_222);  minimum_74 = None
	        round_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_149);  div_149 = None
	        sub_3401: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_149);  round_149 = None
	        clamp_min_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3401, -128);  sub_3401 = None
	        clamp_max_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_223, 127);  clamp_min_223 = None
	        _assert_tensor_metadata_668 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_222, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_668 = None
	        _assert_tensor_metadata_669 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_148, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_669 = None
	        convert_element_type_444: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_148, torch.int8);  clamp_max_148 = None
	        view_1162: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_222, [sym_size_int, 1500, 1])
	        view_1163: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_444, [sym_size_int, 1500, 1])
	        reciprocal_74: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1162);  view_1162 = None
	        mul_7224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_74, 1.0);  reciprocal_74 = None
	        mul_7227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11048, mul_7224);  add_11048 = mul_7224 = None
	        round_150: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7227);  mul_7227 = None
	        add_11435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_150, view_1163);  round_150 = view_1163 = None
	        clamp_min_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11435, -128);  add_11435 = None
	        clamp_max_149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_224, 127);  clamp_min_224 = None
	        _assert_tensor_metadata_670 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_149, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_670 = None
	        convert_element_type_445: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_149, torch.int8);  clamp_max_149 = None
	        view_1166: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_222, [sym_size_int, 1500, 1]);  clamp_min_222 = None
	        view_1167: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_444, [sym_size_int, 1500, 1]);  convert_element_type_444 = None
	        _assert_tensor_metadata_671 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_445, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_671 = None
	        convert_element_type_446: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_445, torch.float32);  convert_element_type_445 = None
	        _assert_tensor_metadata_672 = torch.ops.aten._assert_tensor_metadata.default(view_1167, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_672 = None
	        convert_element_type_447: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1167, torch.float32);  view_1167 = None
	        sub_3421: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_446, convert_element_type_447);  convert_element_type_446 = convert_element_type_447 = None
	        mul_7249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3421, view_1166);  sub_3421 = view_1166 = None
	        _assert_tensor_metadata_673 = torch.ops.aten._assert_tensor_metadata.default(mul_7249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_673 = None
	        view_1169: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1170: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1171: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_674 = torch.ops.aten._assert_tensor_metadata.default(view_1169, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_674 = None
	        convert_element_type_448: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1169, torch.float32);  view_1169 = None
	        _assert_tensor_metadata_675 = torch.ops.aten._assert_tensor_metadata.default(view_1171, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_675 = None
	        convert_element_type_449: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1171, torch.float32);  view_1171 = None
	        sub_3425: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_448, convert_element_type_449);  convert_element_type_448 = convert_element_type_449 = None
	        mul_7254: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3425, view_1170);  sub_3425 = view_1170 = None
	        view_1172: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7254, [1280, 1280]);  mul_7254 = None
	        _assert_tensor_metadata_676 = torch.ops.aten._assert_tensor_metadata.default(view_1172, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_676 = None
	        mul_7259: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1173: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7249, [mul_7259, 1280]);  mul_7249 = mul_7259 = None
	        permute_125: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1172, [1, 0]);  view_1172 = None
	        addmm_61: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_12_self_attn_v_proj_bias, view_1173, permute_125);  model_audio_tower_layers_12_self_attn_v_proj_bias = view_1173 = permute_125 = None
	        view_1174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_61, [sym_size_int, 1500, 1280]);  addmm_61 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1175: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1174, [sym_size_int, -1, 20, 64]);  view_1174 = None
	        permute_126: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1175, [0, 2, 1, 3]);  view_1175 = None
	        clone_100: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_126, memory_format = torch.contiguous_format);  permute_126 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_12 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_98, clone_99, clone_100, None, False, scale = 1.0);  clone_98 = clone_99 = clone_100 = None
	        getitem_98: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_12[0];  _scaled_dot_product_efficient_attention_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_127: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_98, [0, 2, 1, 3]);  getitem_98 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_127, [sym_size_int, 1500, -1]);  permute_127 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1176, [2])
	        amax_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1176, [2])
	        full_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_75, full_150);  amin_75 = full_150 = None
	        full_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_75, full_151);  amax_75 = full_151 = None
	        sub_3443: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_75, minimum_75);  maximum_75 = None
	        div_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3443, 255.0);  sub_3443 = None
	        clamp_min_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_150, 1.1920928955078125e-07);  div_150 = None
	        div_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_75, clamp_min_225);  minimum_75 = None
	        round_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_151);  div_151 = None
	        sub_3449: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_151);  round_151 = None
	        clamp_min_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3449, -128);  sub_3449 = None
	        clamp_max_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_226, 127);  clamp_min_226 = None
	        _assert_tensor_metadata_677 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_225, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_677 = None
	        _assert_tensor_metadata_678 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_150, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_678 = None
	        convert_element_type_450: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_150, torch.int8);  clamp_max_150 = None
	        view_1179: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_225, [sym_size_int, 1500, 1])
	        view_1180: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_450, [sym_size_int, 1500, 1])
	        reciprocal_75: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1179);  view_1179 = None
	        mul_7329: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_75, 1.0);  reciprocal_75 = None
	        mul_7332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1176, mul_7329);  view_1176 = mul_7329 = None
	        round_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7332);  mul_7332 = None
	        add_11599: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_152, view_1180);  round_152 = view_1180 = None
	        clamp_min_227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11599, -128);  add_11599 = None
	        clamp_max_151: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_227, 127);  clamp_min_227 = None
	        _assert_tensor_metadata_679 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_151, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_679 = None
	        convert_element_type_451: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_151, torch.int8);  clamp_max_151 = None
	        view_1183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_225, [sym_size_int, 1500, 1]);  clamp_min_225 = None
	        view_1184: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_450, [sym_size_int, 1500, 1]);  convert_element_type_450 = None
	        _assert_tensor_metadata_680 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_451, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_680 = None
	        convert_element_type_452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_451, torch.float32);  convert_element_type_451 = None
	        _assert_tensor_metadata_681 = torch.ops.aten._assert_tensor_metadata.default(view_1184, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_681 = None
	        convert_element_type_453: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1184, torch.float32);  view_1184 = None
	        sub_3469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_452, convert_element_type_453);  convert_element_type_452 = convert_element_type_453 = None
	        mul_7354: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3469, view_1183);  sub_3469 = view_1183 = None
	        _assert_tensor_metadata_682 = torch.ops.aten._assert_tensor_metadata.default(mul_7354, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_682 = None
	        view_1186: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1187: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1188: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_683 = torch.ops.aten._assert_tensor_metadata.default(view_1186, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_683 = None
	        convert_element_type_454: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1186, torch.float32);  view_1186 = None
	        _assert_tensor_metadata_684 = torch.ops.aten._assert_tensor_metadata.default(view_1188, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_684 = None
	        convert_element_type_455: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1188, torch.float32);  view_1188 = None
	        sub_3473: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_454, convert_element_type_455);  convert_element_type_454 = convert_element_type_455 = None
	        mul_7359: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3473, view_1187);  sub_3473 = view_1187 = None
	        view_1189: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7359, [1280, 1280]);  mul_7359 = None
	        _assert_tensor_metadata_685 = torch.ops.aten._assert_tensor_metadata.default(view_1189, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_685 = None
	        mul_7364: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1190: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7354, [mul_7364, 1280]);  mul_7354 = mul_7364 = None
	        permute_128: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1189, [1, 0]);  view_1189 = None
	        addmm_62: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_12_self_attn_out_proj_bias, view_1190, permute_128);  model_audio_tower_layers_12_self_attn_out_proj_bias = view_1190 = permute_128 = None
	        view_1191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_62, [sym_size_int, 1500, 1280]);  addmm_62 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_11662: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_11042, view_1191);  add_11042 = view_1191 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_102: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_11662, memory_format = torch.contiguous_format)
	        var_mean_25 = torch.ops.aten.var_mean.correction(clone_102, [2], correction = 0, keepdim = True)
	        getitem_102: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_25[0]
	        getitem_103: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_25[1];  var_mean_25 = None
	        add_11667: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_102, 1e-05);  getitem_102 = None
	        rsqrt_25: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_11667);  add_11667 = None
	        sub_3479: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_102, getitem_103);  clone_102 = getitem_103 = None
	        mul_7375: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3479, rsqrt_25);  sub_3479 = rsqrt_25 = None
	        mul_7376: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7375, model_audio_tower_layers_12_final_layer_norm_weight);  mul_7375 = model_audio_tower_layers_12_final_layer_norm_weight = None
	        add_11668: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_7376, model_audio_tower_layers_12_final_layer_norm_bias);  mul_7376 = model_audio_tower_layers_12_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11668, [2])
	        amax_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11668, [2])
	        full_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_76, full_152);  amin_76 = full_152 = None
	        full_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_76, full_153);  amax_76 = full_153 = None
	        sub_3490: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_76, minimum_76);  maximum_76 = None
	        div_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3490, 255.0);  sub_3490 = None
	        clamp_min_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_152, 1.1920928955078125e-07);  div_152 = None
	        div_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_76, clamp_min_228);  minimum_76 = None
	        round_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_153);  div_153 = None
	        sub_3496: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_153);  round_153 = None
	        clamp_min_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3496, -128);  sub_3496 = None
	        clamp_max_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_229, 127);  clamp_min_229 = None
	        _assert_tensor_metadata_686 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_228, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_686 = None
	        _assert_tensor_metadata_687 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_152, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_687 = None
	        convert_element_type_456: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_152, torch.int8);  clamp_max_152 = None
	        view_1194: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_228, [sym_size_int, 1500, 1])
	        view_1195: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_456, [sym_size_int, 1500, 1])
	        reciprocal_76: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1194);  view_1194 = None
	        mul_7424: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_76, 1.0);  reciprocal_76 = None
	        mul_7427: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11668, mul_7424);  add_11668 = mul_7424 = None
	        round_154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7427);  mul_7427 = None
	        add_11755: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_154, view_1195);  round_154 = view_1195 = None
	        clamp_min_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11755, -128);  add_11755 = None
	        clamp_max_153: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_230, 127);  clamp_min_230 = None
	        _assert_tensor_metadata_688 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_153, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_688 = None
	        convert_element_type_457: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_153, torch.int8);  clamp_max_153 = None
	        view_1198: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_228, [sym_size_int, 1500, 1]);  clamp_min_228 = None
	        view_1199: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_456, [sym_size_int, 1500, 1]);  convert_element_type_456 = None
	        _assert_tensor_metadata_689 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_457, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_689 = None
	        convert_element_type_458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_457, torch.float32);  convert_element_type_457 = None
	        _assert_tensor_metadata_690 = torch.ops.aten._assert_tensor_metadata.default(view_1199, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_690 = None
	        convert_element_type_459: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1199, torch.float32);  view_1199 = None
	        sub_3516: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_458, convert_element_type_459);  convert_element_type_458 = convert_element_type_459 = None
	        mul_7449: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3516, view_1198);  sub_3516 = view_1198 = None
	        _assert_tensor_metadata_691 = torch.ops.aten._assert_tensor_metadata.default(mul_7449, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_691 = None
	        view_1201: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_12_fc1_parametrizations_weight_original0 = None
	        view_1202: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_12_fc1_parametrizations_weight_original1 = None
	        view_1203: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_12_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_692 = torch.ops.aten._assert_tensor_metadata.default(view_1201, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_692 = None
	        convert_element_type_460: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1201, torch.float32);  view_1201 = None
	        _assert_tensor_metadata_693 = torch.ops.aten._assert_tensor_metadata.default(view_1203, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_693 = None
	        convert_element_type_461: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1203, torch.float32);  view_1203 = None
	        sub_3520: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_460, convert_element_type_461);  convert_element_type_460 = convert_element_type_461 = None
	        mul_7454: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3520, view_1202);  sub_3520 = view_1202 = None
	        view_1204: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7454, [5120, 1280]);  mul_7454 = None
	        _assert_tensor_metadata_694 = torch.ops.aten._assert_tensor_metadata.default(view_1204, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_694 = None
	        mul_7459: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1205: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7449, [mul_7459, 1280]);  mul_7449 = mul_7459 = None
	        permute_129: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1204, [1, 0]);  view_1204 = None
	        addmm_63: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_12_fc1_bias, view_1205, permute_129);  model_audio_tower_layers_12_fc1_bias = view_1205 = permute_129 = None
	        view_1206: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_63, [sym_size_int, 1500, 5120]);  addmm_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_7466: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1206, 0.5)
	        mul_7467: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1206, 0.7071067811865476);  view_1206 = None
	        erf_14: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_7467);  mul_7467 = None
	        add_11814: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_14, 1);  erf_14 = None
	        mul_7468: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7466, add_11814);  mul_7466 = add_11814 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_7468, [2])
	        amax_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_7468, [2])
	        full_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_77, full_154);  amin_77 = full_154 = None
	        full_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_77, full_155);  amax_77 = full_155 = None
	        sub_3533: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_77, minimum_77);  maximum_77 = None
	        div_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3533, 255.0);  sub_3533 = None
	        clamp_min_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_154, 1.1920928955078125e-07);  div_154 = None
	        div_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_77, clamp_min_231);  minimum_77 = None
	        round_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_155);  div_155 = None
	        sub_3539: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_155);  round_155 = None
	        clamp_min_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3539, -128);  sub_3539 = None
	        clamp_max_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_232, 127);  clamp_min_232 = None
	        _assert_tensor_metadata_695 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_231, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_695 = None
	        _assert_tensor_metadata_696 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_154, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_696 = None
	        convert_element_type_462: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_154, torch.int8);  clamp_max_154 = None
	        view_1209: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_231, [sym_size_int, 1500, 1])
	        view_1210: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_462, [sym_size_int, 1500, 1])
	        reciprocal_77: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1209);  view_1209 = None
	        mul_7514: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_77, 1.0);  reciprocal_77 = None
	        mul_7517: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7468, mul_7514);  mul_7468 = mul_7514 = None
	        round_156: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_7517);  mul_7517 = None
	        add_11897: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_156, view_1210);  round_156 = view_1210 = None
	        clamp_min_233: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11897, -128);  add_11897 = None
	        clamp_max_155: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_233, 127);  clamp_min_233 = None
	        _assert_tensor_metadata_697 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_155, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_697 = None
	        convert_element_type_463: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_155, torch.int8);  clamp_max_155 = None
	        view_1213: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_231, [sym_size_int, 1500, 1]);  clamp_min_231 = None
	        view_1214: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_462, [sym_size_int, 1500, 1]);  convert_element_type_462 = None
	        _assert_tensor_metadata_698 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_463, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_698 = None
	        convert_element_type_464: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_463, torch.float32);  convert_element_type_463 = None
	        _assert_tensor_metadata_699 = torch.ops.aten._assert_tensor_metadata.default(view_1214, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_699 = None
	        convert_element_type_465: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1214, torch.float32);  view_1214 = None
	        sub_3559: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_464, convert_element_type_465);  convert_element_type_464 = convert_element_type_465 = None
	        mul_7539: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3559, view_1213);  sub_3559 = view_1213 = None
	        _assert_tensor_metadata_700 = torch.ops.aten._assert_tensor_metadata.default(mul_7539, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_700 = None
	        view_1216: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_12_fc2_parametrizations_weight_original0 = None
	        view_1217: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_12_fc2_parametrizations_weight_original1 = None
	        view_1218: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_12_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_12_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_701 = torch.ops.aten._assert_tensor_metadata.default(view_1216, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_701 = None
	        convert_element_type_466: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1216, torch.float32);  view_1216 = None
	        _assert_tensor_metadata_702 = torch.ops.aten._assert_tensor_metadata.default(view_1218, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_702 = None
	        convert_element_type_467: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1218, torch.float32);  view_1218 = None
	        sub_3563: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_466, convert_element_type_467);  convert_element_type_466 = convert_element_type_467 = None
	        mul_7544: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3563, view_1217);  sub_3563 = view_1217 = None
	        view_1219: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_7544, [1280, 5120]);  mul_7544 = None
	        _assert_tensor_metadata_703 = torch.ops.aten._assert_tensor_metadata.default(view_1219, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_703 = None
	        mul_7549: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1220: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_7539, [mul_7549, 5120]);  mul_7539 = mul_7549 = None
	        permute_130: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1219, [1, 0]);  view_1219 = None
	        addmm_64: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_12_fc2_bias, view_1220, permute_130);  model_audio_tower_layers_12_fc2_bias = view_1220 = permute_130 = None
	        view_1221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_64, [sym_size_int, 1500, 1280]);  addmm_64 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_11960: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_11662, view_1221);  add_11662 = view_1221 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_105: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_11960, memory_format = torch.contiguous_format)
	        var_mean_26 = torch.ops.aten.var_mean.correction(clone_105, [2], correction = 0, keepdim = True)
	        getitem_104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_26[0]
	        getitem_105: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_26[1];  var_mean_26 = None
	        add_11965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_104, 1e-05);  getitem_104 = None
	        rsqrt_26: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_11965);  add_11965 = None
	        sub_3569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_105, getitem_105);  clone_105 = getitem_105 = None
	        mul_7560: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3569, rsqrt_26);  sub_3569 = rsqrt_26 = None
	        mul_7561: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7560, model_audio_tower_layers_13_self_attn_layer_norm_weight);  mul_7560 = model_audio_tower_layers_13_self_attn_layer_norm_weight = None
	        add_11966: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_7561, model_audio_tower_layers_13_self_attn_layer_norm_bias);  mul_7561 = model_audio_tower_layers_13_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11966, [2])
	        amax_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11966, [2])
	        full_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_78, full_156);  amin_78 = full_156 = None
	        full_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_78, full_157);  amax_78 = full_157 = None
	        sub_3580: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_78, minimum_78);  maximum_78 = None
	        div_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3580, 255.0);  sub_3580 = None
	        clamp_min_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_156, 1.1920928955078125e-07);  div_156 = None
	        div_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_78, clamp_min_234);  minimum_78 = None
	        round_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_157);  div_157 = None
	        sub_3586: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_157);  round_157 = None
	        clamp_min_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3586, -128);  sub_3586 = None
	        clamp_max_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_235, 127);  clamp_min_235 = None
	        _assert_tensor_metadata_704 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_234, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_704 = None
	        _assert_tensor_metadata_705 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_156, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_705 = None
	        convert_element_type_468: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_156, torch.int8);  clamp_max_156 = None
	        view_1224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_234, [sym_size_int, 1500, 1])
	        view_1225: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_468, [sym_size_int, 1500, 1])
	        reciprocal_78: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1224);  view_1224 = None
	        mul_7609: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_78, 1.0);  reciprocal_78 = None
	        mul_7612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11966, mul_7609);  mul_7609 = None
	        round_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7612);  mul_7612 = None
	        add_12053: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_158, view_1225);  round_158 = view_1225 = None
	        clamp_min_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12053, -128);  add_12053 = None
	        clamp_max_157: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_236, 127);  clamp_min_236 = None
	        _assert_tensor_metadata_706 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_157, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_706 = None
	        convert_element_type_469: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_157, torch.int8);  clamp_max_157 = None
	        view_1228: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_234, [sym_size_int, 1500, 1]);  clamp_min_234 = None
	        view_1229: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_468, [sym_size_int, 1500, 1]);  convert_element_type_468 = None
	        _assert_tensor_metadata_707 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_469, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_707 = None
	        convert_element_type_470: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_469, torch.float32);  convert_element_type_469 = None
	        _assert_tensor_metadata_708 = torch.ops.aten._assert_tensor_metadata.default(view_1229, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_708 = None
	        convert_element_type_471: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1229, torch.float32);  view_1229 = None
	        sub_3606: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_470, convert_element_type_471);  convert_element_type_470 = convert_element_type_471 = None
	        mul_7634: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3606, view_1228);  sub_3606 = view_1228 = None
	        _assert_tensor_metadata_709 = torch.ops.aten._assert_tensor_metadata.default(mul_7634, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_709 = None
	        view_1231: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1232: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1233: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_710 = torch.ops.aten._assert_tensor_metadata.default(view_1231, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_710 = None
	        convert_element_type_472: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1231, torch.float32);  view_1231 = None
	        _assert_tensor_metadata_711 = torch.ops.aten._assert_tensor_metadata.default(view_1233, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_711 = None
	        convert_element_type_473: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1233, torch.float32);  view_1233 = None
	        sub_3610: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_472, convert_element_type_473);  convert_element_type_472 = convert_element_type_473 = None
	        mul_7639: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3610, view_1232);  sub_3610 = view_1232 = None
	        view_1234: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7639, [1280, 1280]);  mul_7639 = None
	        _assert_tensor_metadata_712 = torch.ops.aten._assert_tensor_metadata.default(view_1234, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_712 = None
	        mul_7644: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1235: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7634, [mul_7644, 1280]);  mul_7634 = mul_7644 = None
	        permute_131: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1234, [1, 0]);  view_1234 = None
	        addmm_65: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_13_self_attn_q_proj_bias, view_1235, permute_131);  model_audio_tower_layers_13_self_attn_q_proj_bias = view_1235 = permute_131 = None
	        view_1236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_65, [sym_size_int, 1500, 1280]);  addmm_65 = None
	        mul_7651: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1236, 0.125);  view_1236 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1237: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_7651, [sym_size_int, 1500, 20, 64]);  mul_7651 = None
	        permute_132: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1237, [0, 2, 1, 3]);  view_1237 = None
	        clone_106: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_132, memory_format = torch.contiguous_format);  permute_132 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11966, [2])
	        amax_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11966, [2])
	        full_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_79, full_158);  amin_79 = full_158 = None
	        full_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_79, full_159);  amax_79 = full_159 = None
	        sub_3625: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_79, minimum_79);  maximum_79 = None
	        div_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3625, 255.0);  sub_3625 = None
	        clamp_min_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_158, 1.1920928955078125e-07);  div_158 = None
	        div_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_79, clamp_min_237);  minimum_79 = None
	        round_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_159);  div_159 = None
	        sub_3631: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_159);  round_159 = None
	        clamp_min_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3631, -128);  sub_3631 = None
	        clamp_max_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_238, 127);  clamp_min_238 = None
	        _assert_tensor_metadata_713 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_237, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_713 = None
	        _assert_tensor_metadata_714 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_158, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_714 = None
	        convert_element_type_474: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_158, torch.int8);  clamp_max_158 = None
	        view_1240: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_237, [sym_size_int, 1500, 1])
	        view_1241: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_474, [sym_size_int, 1500, 1])
	        reciprocal_79: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1240);  view_1240 = None
	        mul_7705: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_79, 1.0);  reciprocal_79 = None
	        mul_7708: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11966, mul_7705);  mul_7705 = None
	        round_160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7708);  mul_7708 = None
	        add_12205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_160, view_1241);  round_160 = view_1241 = None
	        clamp_min_239: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12205, -128);  add_12205 = None
	        clamp_max_159: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_239, 127);  clamp_min_239 = None
	        _assert_tensor_metadata_715 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_159, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_715 = None
	        convert_element_type_475: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_159, torch.int8);  clamp_max_159 = None
	        view_1244: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_237, [sym_size_int, 1500, 1]);  clamp_min_237 = None
	        view_1245: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_474, [sym_size_int, 1500, 1]);  convert_element_type_474 = None
	        _assert_tensor_metadata_716 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_475, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_716 = None
	        convert_element_type_476: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_475, torch.float32);  convert_element_type_475 = None
	        _assert_tensor_metadata_717 = torch.ops.aten._assert_tensor_metadata.default(view_1245, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_717 = None
	        convert_element_type_477: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1245, torch.float32);  view_1245 = None
	        sub_3651: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_476, convert_element_type_477);  convert_element_type_476 = convert_element_type_477 = None
	        mul_7730: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3651, view_1244);  sub_3651 = view_1244 = None
	        _assert_tensor_metadata_718 = torch.ops.aten._assert_tensor_metadata.default(mul_7730, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_718 = None
	        view_1247: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1248: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1249: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_719 = torch.ops.aten._assert_tensor_metadata.default(view_1247, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_719 = None
	        convert_element_type_478: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1247, torch.float32);  view_1247 = None
	        _assert_tensor_metadata_720 = torch.ops.aten._assert_tensor_metadata.default(view_1249, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_720 = None
	        convert_element_type_479: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1249, torch.float32);  view_1249 = None
	        sub_3655: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_478, convert_element_type_479);  convert_element_type_478 = convert_element_type_479 = None
	        mul_7735: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3655, view_1248);  sub_3655 = view_1248 = None
	        view_1250: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7735, [1280, 1280]);  mul_7735 = None
	        _assert_tensor_metadata_721 = torch.ops.aten._assert_tensor_metadata.default(view_1250, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_721 = None
	        permute_133: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1250, [1, 0]);  view_1250 = None
	        mul_7738: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1251: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7730, [mul_7738, 1280]);  mul_7730 = mul_7738 = None
	        mm_13: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1251, permute_133);  view_1251 = permute_133 = None
	        view_1252: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_13, [sym_size_int, 1500, 1280]);  mm_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1253: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1252, [sym_size_int, -1, 20, 64]);  view_1252 = None
	        permute_134: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1253, [0, 2, 1, 3]);  view_1253 = None
	        clone_107: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_134, memory_format = torch.contiguous_format);  permute_134 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11966, [2])
	        amax_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11966, [2])
	        full_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_80, full_160);  amin_80 = full_160 = None
	        full_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_80, full_161);  amax_80 = full_161 = None
	        sub_3669: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_80, minimum_80);  maximum_80 = None
	        div_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3669, 255.0);  sub_3669 = None
	        clamp_min_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_160, 1.1920928955078125e-07);  div_160 = None
	        div_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_80, clamp_min_240);  minimum_80 = None
	        round_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_161);  div_161 = None
	        sub_3675: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_161);  round_161 = None
	        clamp_min_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3675, -128);  sub_3675 = None
	        clamp_max_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_241, 127);  clamp_min_241 = None
	        _assert_tensor_metadata_722 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_240, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_722 = None
	        _assert_tensor_metadata_723 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_160, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_723 = None
	        convert_element_type_480: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_160, torch.int8);  clamp_max_160 = None
	        view_1256: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_240, [sym_size_int, 1500, 1])
	        view_1257: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_480, [sym_size_int, 1500, 1])
	        reciprocal_80: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1256);  view_1256 = None
	        mul_7804: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_80, 1.0);  reciprocal_80 = None
	        mul_7807: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11966, mul_7804);  add_11966 = mul_7804 = None
	        round_162: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7807);  mul_7807 = None
	        add_12353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_162, view_1257);  round_162 = view_1257 = None
	        clamp_min_242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12353, -128);  add_12353 = None
	        clamp_max_161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_242, 127);  clamp_min_242 = None
	        _assert_tensor_metadata_724 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_161, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_724 = None
	        convert_element_type_481: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_161, torch.int8);  clamp_max_161 = None
	        view_1260: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_240, [sym_size_int, 1500, 1]);  clamp_min_240 = None
	        view_1261: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_480, [sym_size_int, 1500, 1]);  convert_element_type_480 = None
	        _assert_tensor_metadata_725 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_481, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_725 = None
	        convert_element_type_482: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_481, torch.float32);  convert_element_type_481 = None
	        _assert_tensor_metadata_726 = torch.ops.aten._assert_tensor_metadata.default(view_1261, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_726 = None
	        convert_element_type_483: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1261, torch.float32);  view_1261 = None
	        sub_3695: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_482, convert_element_type_483);  convert_element_type_482 = convert_element_type_483 = None
	        mul_7829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3695, view_1260);  sub_3695 = view_1260 = None
	        _assert_tensor_metadata_727 = torch.ops.aten._assert_tensor_metadata.default(mul_7829, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_727 = None
	        view_1263: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1264: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1265: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_728 = torch.ops.aten._assert_tensor_metadata.default(view_1263, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_728 = None
	        convert_element_type_484: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1263, torch.float32);  view_1263 = None
	        _assert_tensor_metadata_729 = torch.ops.aten._assert_tensor_metadata.default(view_1265, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_729 = None
	        convert_element_type_485: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1265, torch.float32);  view_1265 = None
	        sub_3699: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_484, convert_element_type_485);  convert_element_type_484 = convert_element_type_485 = None
	        mul_7834: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3699, view_1264);  sub_3699 = view_1264 = None
	        view_1266: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7834, [1280, 1280]);  mul_7834 = None
	        _assert_tensor_metadata_730 = torch.ops.aten._assert_tensor_metadata.default(view_1266, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_730 = None
	        mul_7839: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1267: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7829, [mul_7839, 1280]);  mul_7829 = mul_7839 = None
	        permute_135: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1266, [1, 0]);  view_1266 = None
	        addmm_66: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_13_self_attn_v_proj_bias, view_1267, permute_135);  model_audio_tower_layers_13_self_attn_v_proj_bias = view_1267 = permute_135 = None
	        view_1268: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_66, [sym_size_int, 1500, 1280]);  addmm_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1269: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1268, [sym_size_int, -1, 20, 64]);  view_1268 = None
	        permute_136: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1269, [0, 2, 1, 3]);  view_1269 = None
	        clone_108: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_136, memory_format = torch.contiguous_format);  permute_136 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_13 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_106, clone_107, clone_108, None, False, scale = 1.0);  clone_106 = clone_107 = clone_108 = None
	        getitem_106: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_13[0];  _scaled_dot_product_efficient_attention_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_137: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_106, [0, 2, 1, 3]);  getitem_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_137, [sym_size_int, 1500, -1]);  permute_137 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1270, [2])
	        amax_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1270, [2])
	        full_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_81, full_162);  amin_81 = full_162 = None
	        full_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_81, full_163);  amax_81 = full_163 = None
	        sub_3717: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_81, minimum_81);  maximum_81 = None
	        div_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3717, 255.0);  sub_3717 = None
	        clamp_min_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_162, 1.1920928955078125e-07);  div_162 = None
	        div_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_81, clamp_min_243);  minimum_81 = None
	        round_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_163);  div_163 = None
	        sub_3723: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_163);  round_163 = None
	        clamp_min_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3723, -128);  sub_3723 = None
	        clamp_max_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_244, 127);  clamp_min_244 = None
	        _assert_tensor_metadata_731 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_243, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_731 = None
	        _assert_tensor_metadata_732 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_162, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_732 = None
	        convert_element_type_486: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_162, torch.int8);  clamp_max_162 = None
	        view_1273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_243, [sym_size_int, 1500, 1])
	        view_1274: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_486, [sym_size_int, 1500, 1])
	        reciprocal_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1273);  view_1273 = None
	        mul_7909: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_81, 1.0);  reciprocal_81 = None
	        mul_7912: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1270, mul_7909);  view_1270 = mul_7909 = None
	        round_164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7912);  mul_7912 = None
	        add_12517: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_164, view_1274);  round_164 = view_1274 = None
	        clamp_min_245: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12517, -128);  add_12517 = None
	        clamp_max_163: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_245, 127);  clamp_min_245 = None
	        _assert_tensor_metadata_733 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_163, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_733 = None
	        convert_element_type_487: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_163, torch.int8);  clamp_max_163 = None
	        view_1277: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_243, [sym_size_int, 1500, 1]);  clamp_min_243 = None
	        view_1278: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_486, [sym_size_int, 1500, 1]);  convert_element_type_486 = None
	        _assert_tensor_metadata_734 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_487, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_734 = None
	        convert_element_type_488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_487, torch.float32);  convert_element_type_487 = None
	        _assert_tensor_metadata_735 = torch.ops.aten._assert_tensor_metadata.default(view_1278, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_735 = None
	        convert_element_type_489: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1278, torch.float32);  view_1278 = None
	        sub_3743: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_488, convert_element_type_489);  convert_element_type_488 = convert_element_type_489 = None
	        mul_7934: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3743, view_1277);  sub_3743 = view_1277 = None
	        _assert_tensor_metadata_736 = torch.ops.aten._assert_tensor_metadata.default(mul_7934, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_736 = None
	        view_1280: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1281: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1282: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_737 = torch.ops.aten._assert_tensor_metadata.default(view_1280, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_737 = None
	        convert_element_type_490: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1280, torch.float32);  view_1280 = None
	        _assert_tensor_metadata_738 = torch.ops.aten._assert_tensor_metadata.default(view_1282, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_738 = None
	        convert_element_type_491: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1282, torch.float32);  view_1282 = None
	        sub_3747: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_490, convert_element_type_491);  convert_element_type_490 = convert_element_type_491 = None
	        mul_7939: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3747, view_1281);  sub_3747 = view_1281 = None
	        view_1283: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7939, [1280, 1280]);  mul_7939 = None
	        _assert_tensor_metadata_739 = torch.ops.aten._assert_tensor_metadata.default(view_1283, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_739 = None
	        mul_7944: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1284: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_7934, [mul_7944, 1280]);  mul_7934 = mul_7944 = None
	        permute_138: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1283, [1, 0]);  view_1283 = None
	        addmm_67: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_13_self_attn_out_proj_bias, view_1284, permute_138);  model_audio_tower_layers_13_self_attn_out_proj_bias = view_1284 = permute_138 = None
	        view_1285: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_67, [sym_size_int, 1500, 1280]);  addmm_67 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_12580: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_11960, view_1285);  add_11960 = view_1285 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_12580, memory_format = torch.contiguous_format)
	        var_mean_27 = torch.ops.aten.var_mean.correction(clone_110, [2], correction = 0, keepdim = True)
	        getitem_110: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_27[0]
	        getitem_111: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_27[1];  var_mean_27 = None
	        add_12585: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_110, 1e-05);  getitem_110 = None
	        rsqrt_27: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_12585);  add_12585 = None
	        sub_3753: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_110, getitem_111);  clone_110 = getitem_111 = None
	        mul_7955: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3753, rsqrt_27);  sub_3753 = rsqrt_27 = None
	        mul_7956: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7955, model_audio_tower_layers_13_final_layer_norm_weight);  mul_7955 = model_audio_tower_layers_13_final_layer_norm_weight = None
	        add_12586: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_7956, model_audio_tower_layers_13_final_layer_norm_bias);  mul_7956 = model_audio_tower_layers_13_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_12586, [2])
	        amax_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_12586, [2])
	        full_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_82, full_164);  amin_82 = full_164 = None
	        full_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_82, full_165);  amax_82 = full_165 = None
	        sub_3764: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_82, minimum_82);  maximum_82 = None
	        div_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3764, 255.0);  sub_3764 = None
	        clamp_min_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_164, 1.1920928955078125e-07);  div_164 = None
	        div_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_82, clamp_min_246);  minimum_82 = None
	        round_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_165);  div_165 = None
	        sub_3770: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_165);  round_165 = None
	        clamp_min_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3770, -128);  sub_3770 = None
	        clamp_max_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_247, 127);  clamp_min_247 = None
	        _assert_tensor_metadata_740 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_246, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_740 = None
	        _assert_tensor_metadata_741 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_164, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_741 = None
	        convert_element_type_492: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_164, torch.int8);  clamp_max_164 = None
	        view_1288: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_246, [sym_size_int, 1500, 1])
	        view_1289: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_492, [sym_size_int, 1500, 1])
	        reciprocal_82: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1288);  view_1288 = None
	        mul_8004: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_82, 1.0);  reciprocal_82 = None
	        mul_8007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_12586, mul_8004);  add_12586 = mul_8004 = None
	        round_166: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8007);  mul_8007 = None
	        add_12673: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_166, view_1289);  round_166 = view_1289 = None
	        clamp_min_248: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12673, -128);  add_12673 = None
	        clamp_max_165: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_248, 127);  clamp_min_248 = None
	        _assert_tensor_metadata_742 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_165, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_742 = None
	        convert_element_type_493: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_165, torch.int8);  clamp_max_165 = None
	        view_1292: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_246, [sym_size_int, 1500, 1]);  clamp_min_246 = None
	        view_1293: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_492, [sym_size_int, 1500, 1]);  convert_element_type_492 = None
	        _assert_tensor_metadata_743 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_493, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_743 = None
	        convert_element_type_494: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_493, torch.float32);  convert_element_type_493 = None
	        _assert_tensor_metadata_744 = torch.ops.aten._assert_tensor_metadata.default(view_1293, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_744 = None
	        convert_element_type_495: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1293, torch.float32);  view_1293 = None
	        sub_3790: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_494, convert_element_type_495);  convert_element_type_494 = convert_element_type_495 = None
	        mul_8029: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3790, view_1292);  sub_3790 = view_1292 = None
	        _assert_tensor_metadata_745 = torch.ops.aten._assert_tensor_metadata.default(mul_8029, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_745 = None
	        view_1295: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_13_fc1_parametrizations_weight_original0 = None
	        view_1296: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_13_fc1_parametrizations_weight_original1 = None
	        view_1297: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_13_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_746 = torch.ops.aten._assert_tensor_metadata.default(view_1295, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_746 = None
	        convert_element_type_496: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1295, torch.float32);  view_1295 = None
	        _assert_tensor_metadata_747 = torch.ops.aten._assert_tensor_metadata.default(view_1297, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_747 = None
	        convert_element_type_497: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1297, torch.float32);  view_1297 = None
	        sub_3794: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_496, convert_element_type_497);  convert_element_type_496 = convert_element_type_497 = None
	        mul_8034: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3794, view_1296);  sub_3794 = view_1296 = None
	        view_1298: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8034, [5120, 1280]);  mul_8034 = None
	        _assert_tensor_metadata_748 = torch.ops.aten._assert_tensor_metadata.default(view_1298, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_748 = None
	        mul_8039: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1299: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8029, [mul_8039, 1280]);  mul_8029 = mul_8039 = None
	        permute_139: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1298, [1, 0]);  view_1298 = None
	        addmm_68: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_13_fc1_bias, view_1299, permute_139);  model_audio_tower_layers_13_fc1_bias = view_1299 = permute_139 = None
	        view_1300: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_68, [sym_size_int, 1500, 5120]);  addmm_68 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_8046: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1300, 0.5)
	        mul_8047: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1300, 0.7071067811865476);  view_1300 = None
	        erf_15: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_8047);  mul_8047 = None
	        add_12732: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_15, 1);  erf_15 = None
	        mul_8048: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8046, add_12732);  mul_8046 = add_12732 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_8048, [2])
	        amax_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_8048, [2])
	        full_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_83, full_166);  amin_83 = full_166 = None
	        full_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_83, full_167);  amax_83 = full_167 = None
	        sub_3807: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_83, minimum_83);  maximum_83 = None
	        div_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3807, 255.0);  sub_3807 = None
	        clamp_min_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_166, 1.1920928955078125e-07);  div_166 = None
	        div_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_83, clamp_min_249);  minimum_83 = None
	        round_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_167);  div_167 = None
	        sub_3813: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_167);  round_167 = None
	        clamp_min_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3813, -128);  sub_3813 = None
	        clamp_max_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_250, 127);  clamp_min_250 = None
	        _assert_tensor_metadata_749 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_749 = None
	        _assert_tensor_metadata_750 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_166, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_750 = None
	        convert_element_type_498: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_166, torch.int8);  clamp_max_166 = None
	        view_1303: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_249, [sym_size_int, 1500, 1])
	        view_1304: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_498, [sym_size_int, 1500, 1])
	        reciprocal_83: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1303);  view_1303 = None
	        mul_8094: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_83, 1.0);  reciprocal_83 = None
	        mul_8097: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8048, mul_8094);  mul_8048 = mul_8094 = None
	        round_168: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_8097);  mul_8097 = None
	        add_12815: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_168, view_1304);  round_168 = view_1304 = None
	        clamp_min_251: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12815, -128);  add_12815 = None
	        clamp_max_167: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_251, 127);  clamp_min_251 = None
	        _assert_tensor_metadata_751 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_167, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_751 = None
	        convert_element_type_499: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_167, torch.int8);  clamp_max_167 = None
	        view_1307: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_249, [sym_size_int, 1500, 1]);  clamp_min_249 = None
	        view_1308: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_498, [sym_size_int, 1500, 1]);  convert_element_type_498 = None
	        _assert_tensor_metadata_752 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_499, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_752 = None
	        convert_element_type_500: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_499, torch.float32);  convert_element_type_499 = None
	        _assert_tensor_metadata_753 = torch.ops.aten._assert_tensor_metadata.default(view_1308, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_753 = None
	        convert_element_type_501: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1308, torch.float32);  view_1308 = None
	        sub_3833: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_500, convert_element_type_501);  convert_element_type_500 = convert_element_type_501 = None
	        mul_8119: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3833, view_1307);  sub_3833 = view_1307 = None
	        _assert_tensor_metadata_754 = torch.ops.aten._assert_tensor_metadata.default(mul_8119, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_754 = None
	        view_1310: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_13_fc2_parametrizations_weight_original0 = None
	        view_1311: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_13_fc2_parametrizations_weight_original1 = None
	        view_1312: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_13_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_13_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_755 = torch.ops.aten._assert_tensor_metadata.default(view_1310, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_755 = None
	        convert_element_type_502: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1310, torch.float32);  view_1310 = None
	        _assert_tensor_metadata_756 = torch.ops.aten._assert_tensor_metadata.default(view_1312, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_756 = None
	        convert_element_type_503: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1312, torch.float32);  view_1312 = None
	        sub_3837: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_502, convert_element_type_503);  convert_element_type_502 = convert_element_type_503 = None
	        mul_8124: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3837, view_1311);  sub_3837 = view_1311 = None
	        view_1313: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_8124, [1280, 5120]);  mul_8124 = None
	        _assert_tensor_metadata_757 = torch.ops.aten._assert_tensor_metadata.default(view_1313, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_757 = None
	        mul_8129: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1314: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_8119, [mul_8129, 5120]);  mul_8119 = mul_8129 = None
	        permute_140: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1313, [1, 0]);  view_1313 = None
	        addmm_69: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_13_fc2_bias, view_1314, permute_140);  model_audio_tower_layers_13_fc2_bias = view_1314 = permute_140 = None
	        view_1315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_69, [sym_size_int, 1500, 1280]);  addmm_69 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_12878: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_12580, view_1315);  add_12580 = view_1315 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_12878, memory_format = torch.contiguous_format)
	        var_mean_28 = torch.ops.aten.var_mean.correction(clone_113, [2], correction = 0, keepdim = True)
	        getitem_112: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_28[0]
	        getitem_113: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_28[1];  var_mean_28 = None
	        add_12883: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_112, 1e-05);  getitem_112 = None
	        rsqrt_28: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_12883);  add_12883 = None
	        sub_3843: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_113, getitem_113);  clone_113 = getitem_113 = None
	        mul_8140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3843, rsqrt_28);  sub_3843 = rsqrt_28 = None
	        mul_8141: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8140, model_audio_tower_layers_14_self_attn_layer_norm_weight);  mul_8140 = model_audio_tower_layers_14_self_attn_layer_norm_weight = None
	        add_12884: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_8141, model_audio_tower_layers_14_self_attn_layer_norm_bias);  mul_8141 = model_audio_tower_layers_14_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_12884, [2])
	        amax_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_12884, [2])
	        full_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_84, full_168);  amin_84 = full_168 = None
	        full_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_84, full_169);  amax_84 = full_169 = None
	        sub_3854: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_84, minimum_84);  maximum_84 = None
	        div_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3854, 255.0);  sub_3854 = None
	        clamp_min_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_168, 1.1920928955078125e-07);  div_168 = None
	        div_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_84, clamp_min_252);  minimum_84 = None
	        round_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_169);  div_169 = None
	        sub_3860: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_169);  round_169 = None
	        clamp_min_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3860, -128);  sub_3860 = None
	        clamp_max_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_253, 127);  clamp_min_253 = None
	        _assert_tensor_metadata_758 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_252, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_758 = None
	        _assert_tensor_metadata_759 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_168, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_759 = None
	        convert_element_type_504: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_168, torch.int8);  clamp_max_168 = None
	        view_1318: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_252, [sym_size_int, 1500, 1])
	        view_1319: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_504, [sym_size_int, 1500, 1])
	        reciprocal_84: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1318);  view_1318 = None
	        mul_8189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_84, 1.0);  reciprocal_84 = None
	        mul_8192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_12884, mul_8189);  mul_8189 = None
	        round_170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8192);  mul_8192 = None
	        add_12971: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_170, view_1319);  round_170 = view_1319 = None
	        clamp_min_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12971, -128);  add_12971 = None
	        clamp_max_169: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_254, 127);  clamp_min_254 = None
	        _assert_tensor_metadata_760 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_169, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_760 = None
	        convert_element_type_505: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_169, torch.int8);  clamp_max_169 = None
	        view_1322: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_252, [sym_size_int, 1500, 1]);  clamp_min_252 = None
	        view_1323: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_504, [sym_size_int, 1500, 1]);  convert_element_type_504 = None
	        _assert_tensor_metadata_761 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_505, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_761 = None
	        convert_element_type_506: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_505, torch.float32);  convert_element_type_505 = None
	        _assert_tensor_metadata_762 = torch.ops.aten._assert_tensor_metadata.default(view_1323, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_762 = None
	        convert_element_type_507: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1323, torch.float32);  view_1323 = None
	        sub_3880: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_506, convert_element_type_507);  convert_element_type_506 = convert_element_type_507 = None
	        mul_8214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3880, view_1322);  sub_3880 = view_1322 = None
	        _assert_tensor_metadata_763 = torch.ops.aten._assert_tensor_metadata.default(mul_8214, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_763 = None
	        view_1325: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1326: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1327: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_764 = torch.ops.aten._assert_tensor_metadata.default(view_1325, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_764 = None
	        convert_element_type_508: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1325, torch.float32);  view_1325 = None
	        _assert_tensor_metadata_765 = torch.ops.aten._assert_tensor_metadata.default(view_1327, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_765 = None
	        convert_element_type_509: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1327, torch.float32);  view_1327 = None
	        sub_3884: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_508, convert_element_type_509);  convert_element_type_508 = convert_element_type_509 = None
	        mul_8219: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3884, view_1326);  sub_3884 = view_1326 = None
	        view_1328: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8219, [1280, 1280]);  mul_8219 = None
	        _assert_tensor_metadata_766 = torch.ops.aten._assert_tensor_metadata.default(view_1328, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_766 = None
	        mul_8224: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1329: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8214, [mul_8224, 1280]);  mul_8214 = mul_8224 = None
	        permute_141: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1328, [1, 0]);  view_1328 = None
	        addmm_70: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_14_self_attn_q_proj_bias, view_1329, permute_141);  model_audio_tower_layers_14_self_attn_q_proj_bias = view_1329 = permute_141 = None
	        view_1330: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_70, [sym_size_int, 1500, 1280]);  addmm_70 = None
	        mul_8231: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1330, 0.125);  view_1330 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1331: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_8231, [sym_size_int, 1500, 20, 64]);  mul_8231 = None
	        permute_142: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1331, [0, 2, 1, 3]);  view_1331 = None
	        clone_114: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_142, memory_format = torch.contiguous_format);  permute_142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_12884, [2])
	        amax_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_12884, [2])
	        full_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_85, full_170);  amin_85 = full_170 = None
	        full_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_85, full_171);  amax_85 = full_171 = None
	        sub_3899: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_85, minimum_85);  maximum_85 = None
	        div_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3899, 255.0);  sub_3899 = None
	        clamp_min_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_170, 1.1920928955078125e-07);  div_170 = None
	        div_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_85, clamp_min_255);  minimum_85 = None
	        round_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_171);  div_171 = None
	        sub_3905: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_171);  round_171 = None
	        clamp_min_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3905, -128);  sub_3905 = None
	        clamp_max_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_256, 127);  clamp_min_256 = None
	        _assert_tensor_metadata_767 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_255, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_767 = None
	        _assert_tensor_metadata_768 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_170, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_768 = None
	        convert_element_type_510: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_170, torch.int8);  clamp_max_170 = None
	        view_1334: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_255, [sym_size_int, 1500, 1])
	        view_1335: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_510, [sym_size_int, 1500, 1])
	        reciprocal_85: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1334);  view_1334 = None
	        mul_8285: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_85, 1.0);  reciprocal_85 = None
	        mul_8288: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_12884, mul_8285);  mul_8285 = None
	        round_172: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8288);  mul_8288 = None
	        add_13123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_172, view_1335);  round_172 = view_1335 = None
	        clamp_min_257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13123, -128);  add_13123 = None
	        clamp_max_171: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_257, 127);  clamp_min_257 = None
	        _assert_tensor_metadata_769 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_171, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_769 = None
	        convert_element_type_511: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_171, torch.int8);  clamp_max_171 = None
	        view_1338: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_255, [sym_size_int, 1500, 1]);  clamp_min_255 = None
	        view_1339: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_510, [sym_size_int, 1500, 1]);  convert_element_type_510 = None
	        _assert_tensor_metadata_770 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_511, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_770 = None
	        convert_element_type_512: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_511, torch.float32);  convert_element_type_511 = None
	        _assert_tensor_metadata_771 = torch.ops.aten._assert_tensor_metadata.default(view_1339, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_771 = None
	        convert_element_type_513: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1339, torch.float32);  view_1339 = None
	        sub_3925: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_512, convert_element_type_513);  convert_element_type_512 = convert_element_type_513 = None
	        mul_8310: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3925, view_1338);  sub_3925 = view_1338 = None
	        _assert_tensor_metadata_772 = torch.ops.aten._assert_tensor_metadata.default(mul_8310, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_772 = None
	        view_1341: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1342: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1343: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_773 = torch.ops.aten._assert_tensor_metadata.default(view_1341, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_773 = None
	        convert_element_type_514: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1341, torch.float32);  view_1341 = None
	        _assert_tensor_metadata_774 = torch.ops.aten._assert_tensor_metadata.default(view_1343, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_774 = None
	        convert_element_type_515: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1343, torch.float32);  view_1343 = None
	        sub_3929: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_514, convert_element_type_515);  convert_element_type_514 = convert_element_type_515 = None
	        mul_8315: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3929, view_1342);  sub_3929 = view_1342 = None
	        view_1344: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8315, [1280, 1280]);  mul_8315 = None
	        _assert_tensor_metadata_775 = torch.ops.aten._assert_tensor_metadata.default(view_1344, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_775 = None
	        permute_143: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1344, [1, 0]);  view_1344 = None
	        mul_8318: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1345: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8310, [mul_8318, 1280]);  mul_8310 = mul_8318 = None
	        mm_14: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1345, permute_143);  view_1345 = permute_143 = None
	        view_1346: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_14, [sym_size_int, 1500, 1280]);  mm_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1347: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1346, [sym_size_int, -1, 20, 64]);  view_1346 = None
	        permute_144: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1347, [0, 2, 1, 3]);  view_1347 = None
	        clone_115: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_144, memory_format = torch.contiguous_format);  permute_144 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_12884, [2])
	        amax_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_12884, [2])
	        full_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_86, full_172);  amin_86 = full_172 = None
	        full_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_86, full_173);  amax_86 = full_173 = None
	        sub_3943: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_86, minimum_86);  maximum_86 = None
	        div_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3943, 255.0);  sub_3943 = None
	        clamp_min_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_172, 1.1920928955078125e-07);  div_172 = None
	        div_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_86, clamp_min_258);  minimum_86 = None
	        round_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_173);  div_173 = None
	        sub_3949: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_173);  round_173 = None
	        clamp_min_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3949, -128);  sub_3949 = None
	        clamp_max_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_259, 127);  clamp_min_259 = None
	        _assert_tensor_metadata_776 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_258, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_776 = None
	        _assert_tensor_metadata_777 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_172, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_777 = None
	        convert_element_type_516: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_172, torch.int8);  clamp_max_172 = None
	        view_1350: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_258, [sym_size_int, 1500, 1])
	        view_1351: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_516, [sym_size_int, 1500, 1])
	        reciprocal_86: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1350);  view_1350 = None
	        mul_8384: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_86, 1.0);  reciprocal_86 = None
	        mul_8387: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_12884, mul_8384);  add_12884 = mul_8384 = None
	        round_174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8387);  mul_8387 = None
	        add_13271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_174, view_1351);  round_174 = view_1351 = None
	        clamp_min_260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13271, -128);  add_13271 = None
	        clamp_max_173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_260, 127);  clamp_min_260 = None
	        _assert_tensor_metadata_778 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_173, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_778 = None
	        convert_element_type_517: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_173, torch.int8);  clamp_max_173 = None
	        view_1354: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_258, [sym_size_int, 1500, 1]);  clamp_min_258 = None
	        view_1355: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_516, [sym_size_int, 1500, 1]);  convert_element_type_516 = None
	        _assert_tensor_metadata_779 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_517, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_779 = None
	        convert_element_type_518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_517, torch.float32);  convert_element_type_517 = None
	        _assert_tensor_metadata_780 = torch.ops.aten._assert_tensor_metadata.default(view_1355, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_780 = None
	        convert_element_type_519: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1355, torch.float32);  view_1355 = None
	        sub_3969: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_518, convert_element_type_519);  convert_element_type_518 = convert_element_type_519 = None
	        mul_8409: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3969, view_1354);  sub_3969 = view_1354 = None
	        _assert_tensor_metadata_781 = torch.ops.aten._assert_tensor_metadata.default(mul_8409, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_781 = None
	        view_1357: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1358: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1359: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_782 = torch.ops.aten._assert_tensor_metadata.default(view_1357, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_782 = None
	        convert_element_type_520: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1357, torch.float32);  view_1357 = None
	        _assert_tensor_metadata_783 = torch.ops.aten._assert_tensor_metadata.default(view_1359, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_783 = None
	        convert_element_type_521: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1359, torch.float32);  view_1359 = None
	        sub_3973: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_520, convert_element_type_521);  convert_element_type_520 = convert_element_type_521 = None
	        mul_8414: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3973, view_1358);  sub_3973 = view_1358 = None
	        view_1360: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8414, [1280, 1280]);  mul_8414 = None
	        _assert_tensor_metadata_784 = torch.ops.aten._assert_tensor_metadata.default(view_1360, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_784 = None
	        mul_8419: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1361: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8409, [mul_8419, 1280]);  mul_8409 = mul_8419 = None
	        permute_145: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1360, [1, 0]);  view_1360 = None
	        addmm_71: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_14_self_attn_v_proj_bias, view_1361, permute_145);  model_audio_tower_layers_14_self_attn_v_proj_bias = view_1361 = permute_145 = None
	        view_1362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_71, [sym_size_int, 1500, 1280]);  addmm_71 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1363: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1362, [sym_size_int, -1, 20, 64]);  view_1362 = None
	        permute_146: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1363, [0, 2, 1, 3]);  view_1363 = None
	        clone_116: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_146, memory_format = torch.contiguous_format);  permute_146 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_14 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_114, clone_115, clone_116, None, False, scale = 1.0);  clone_114 = clone_115 = clone_116 = None
	        getitem_114: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_14[0];  _scaled_dot_product_efficient_attention_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_147: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_114, [0, 2, 1, 3]);  getitem_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1364: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_147, [sym_size_int, 1500, -1]);  permute_147 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1364, [2])
	        amax_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1364, [2])
	        full_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_87, full_174);  amin_87 = full_174 = None
	        full_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_87, full_175);  amax_87 = full_175 = None
	        sub_3991: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_87, minimum_87);  maximum_87 = None
	        div_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3991, 255.0);  sub_3991 = None
	        clamp_min_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_174, 1.1920928955078125e-07);  div_174 = None
	        div_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_87, clamp_min_261);  minimum_87 = None
	        round_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_175);  div_175 = None
	        sub_3997: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_175);  round_175 = None
	        clamp_min_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3997, -128);  sub_3997 = None
	        clamp_max_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_262, 127);  clamp_min_262 = None
	        _assert_tensor_metadata_785 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_261, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_785 = None
	        _assert_tensor_metadata_786 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_174, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_786 = None
	        convert_element_type_522: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_174, torch.int8);  clamp_max_174 = None
	        view_1367: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_261, [sym_size_int, 1500, 1])
	        view_1368: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_522, [sym_size_int, 1500, 1])
	        reciprocal_87: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1367);  view_1367 = None
	        mul_8489: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_87, 1.0);  reciprocal_87 = None
	        mul_8492: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1364, mul_8489);  view_1364 = mul_8489 = None
	        round_176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8492);  mul_8492 = None
	        add_13435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_176, view_1368);  round_176 = view_1368 = None
	        clamp_min_263: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13435, -128);  add_13435 = None
	        clamp_max_175: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_263, 127);  clamp_min_263 = None
	        _assert_tensor_metadata_787 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_175, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_787 = None
	        convert_element_type_523: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_175, torch.int8);  clamp_max_175 = None
	        view_1371: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_261, [sym_size_int, 1500, 1]);  clamp_min_261 = None
	        view_1372: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_522, [sym_size_int, 1500, 1]);  convert_element_type_522 = None
	        _assert_tensor_metadata_788 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_523, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_788 = None
	        convert_element_type_524: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_523, torch.float32);  convert_element_type_523 = None
	        _assert_tensor_metadata_789 = torch.ops.aten._assert_tensor_metadata.default(view_1372, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_789 = None
	        convert_element_type_525: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1372, torch.float32);  view_1372 = None
	        sub_4017: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_524, convert_element_type_525);  convert_element_type_524 = convert_element_type_525 = None
	        mul_8514: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4017, view_1371);  sub_4017 = view_1371 = None
	        _assert_tensor_metadata_790 = torch.ops.aten._assert_tensor_metadata.default(mul_8514, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_790 = None
	        view_1374: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1375: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1376: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_791 = torch.ops.aten._assert_tensor_metadata.default(view_1374, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_791 = None
	        convert_element_type_526: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1374, torch.float32);  view_1374 = None
	        _assert_tensor_metadata_792 = torch.ops.aten._assert_tensor_metadata.default(view_1376, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_792 = None
	        convert_element_type_527: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1376, torch.float32);  view_1376 = None
	        sub_4021: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_526, convert_element_type_527);  convert_element_type_526 = convert_element_type_527 = None
	        mul_8519: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4021, view_1375);  sub_4021 = view_1375 = None
	        view_1377: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8519, [1280, 1280]);  mul_8519 = None
	        _assert_tensor_metadata_793 = torch.ops.aten._assert_tensor_metadata.default(view_1377, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_793 = None
	        mul_8524: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1378: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8514, [mul_8524, 1280]);  mul_8514 = mul_8524 = None
	        permute_148: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1377, [1, 0]);  view_1377 = None
	        addmm_72: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_14_self_attn_out_proj_bias, view_1378, permute_148);  model_audio_tower_layers_14_self_attn_out_proj_bias = view_1378 = permute_148 = None
	        view_1379: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_72, [sym_size_int, 1500, 1280]);  addmm_72 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_13498: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_12878, view_1379);  add_12878 = view_1379 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_118: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_13498, memory_format = torch.contiguous_format)
	        var_mean_29 = torch.ops.aten.var_mean.correction(clone_118, [2], correction = 0, keepdim = True)
	        getitem_118: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_29[0]
	        getitem_119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_29[1];  var_mean_29 = None
	        add_13503: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_118, 1e-05);  getitem_118 = None
	        rsqrt_29: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_13503);  add_13503 = None
	        sub_4027: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_118, getitem_119);  clone_118 = getitem_119 = None
	        mul_8535: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4027, rsqrt_29);  sub_4027 = rsqrt_29 = None
	        mul_8536: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8535, model_audio_tower_layers_14_final_layer_norm_weight);  mul_8535 = model_audio_tower_layers_14_final_layer_norm_weight = None
	        add_13504: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_8536, model_audio_tower_layers_14_final_layer_norm_bias);  mul_8536 = model_audio_tower_layers_14_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_13504, [2])
	        amax_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_13504, [2])
	        full_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_88, full_176);  amin_88 = full_176 = None
	        full_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_88, full_177);  amax_88 = full_177 = None
	        sub_4038: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_88, minimum_88);  maximum_88 = None
	        div_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4038, 255.0);  sub_4038 = None
	        clamp_min_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_176, 1.1920928955078125e-07);  div_176 = None
	        div_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_88, clamp_min_264);  minimum_88 = None
	        round_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_177);  div_177 = None
	        sub_4044: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_177);  round_177 = None
	        clamp_min_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4044, -128);  sub_4044 = None
	        clamp_max_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_265, 127);  clamp_min_265 = None
	        _assert_tensor_metadata_794 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_264, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_794 = None
	        _assert_tensor_metadata_795 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_176, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_795 = None
	        convert_element_type_528: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_176, torch.int8);  clamp_max_176 = None
	        view_1382: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_264, [sym_size_int, 1500, 1])
	        view_1383: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_528, [sym_size_int, 1500, 1])
	        reciprocal_88: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1382);  view_1382 = None
	        mul_8584: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_88, 1.0);  reciprocal_88 = None
	        mul_8587: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_13504, mul_8584);  add_13504 = mul_8584 = None
	        round_178: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8587);  mul_8587 = None
	        add_13591: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_178, view_1383);  round_178 = view_1383 = None
	        clamp_min_266: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13591, -128);  add_13591 = None
	        clamp_max_177: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_266, 127);  clamp_min_266 = None
	        _assert_tensor_metadata_796 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_177, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_796 = None
	        convert_element_type_529: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_177, torch.int8);  clamp_max_177 = None
	        view_1386: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_264, [sym_size_int, 1500, 1]);  clamp_min_264 = None
	        view_1387: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_528, [sym_size_int, 1500, 1]);  convert_element_type_528 = None
	        _assert_tensor_metadata_797 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_529, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_797 = None
	        convert_element_type_530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_529, torch.float32);  convert_element_type_529 = None
	        _assert_tensor_metadata_798 = torch.ops.aten._assert_tensor_metadata.default(view_1387, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_798 = None
	        convert_element_type_531: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1387, torch.float32);  view_1387 = None
	        sub_4064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_530, convert_element_type_531);  convert_element_type_530 = convert_element_type_531 = None
	        mul_8609: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4064, view_1386);  sub_4064 = view_1386 = None
	        _assert_tensor_metadata_799 = torch.ops.aten._assert_tensor_metadata.default(mul_8609, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_799 = None
	        view_1389: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_14_fc1_parametrizations_weight_original0 = None
	        view_1390: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_14_fc1_parametrizations_weight_original1 = None
	        view_1391: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_14_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_800 = torch.ops.aten._assert_tensor_metadata.default(view_1389, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_800 = None
	        convert_element_type_532: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1389, torch.float32);  view_1389 = None
	        _assert_tensor_metadata_801 = torch.ops.aten._assert_tensor_metadata.default(view_1391, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_801 = None
	        convert_element_type_533: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1391, torch.float32);  view_1391 = None
	        sub_4068: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_532, convert_element_type_533);  convert_element_type_532 = convert_element_type_533 = None
	        mul_8614: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4068, view_1390);  sub_4068 = view_1390 = None
	        view_1392: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8614, [5120, 1280]);  mul_8614 = None
	        _assert_tensor_metadata_802 = torch.ops.aten._assert_tensor_metadata.default(view_1392, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_802 = None
	        mul_8619: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1393: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8609, [mul_8619, 1280]);  mul_8609 = mul_8619 = None
	        permute_149: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1392, [1, 0]);  view_1392 = None
	        addmm_73: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_14_fc1_bias, view_1393, permute_149);  model_audio_tower_layers_14_fc1_bias = view_1393 = permute_149 = None
	        view_1394: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_73, [sym_size_int, 1500, 5120]);  addmm_73 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_8626: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1394, 0.5)
	        mul_8627: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1394, 0.7071067811865476);  view_1394 = None
	        erf_16: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_8627);  mul_8627 = None
	        add_13650: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_16, 1);  erf_16 = None
	        mul_8628: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8626, add_13650);  mul_8626 = add_13650 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_8628, [2])
	        amax_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_8628, [2])
	        full_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_89, full_178);  amin_89 = full_178 = None
	        full_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_89, full_179);  amax_89 = full_179 = None
	        sub_4081: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_89, minimum_89);  maximum_89 = None
	        div_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4081, 255.0);  sub_4081 = None
	        clamp_min_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_178, 1.1920928955078125e-07);  div_178 = None
	        div_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_89, clamp_min_267);  minimum_89 = None
	        round_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_179);  div_179 = None
	        sub_4087: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_179);  round_179 = None
	        clamp_min_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4087, -128);  sub_4087 = None
	        clamp_max_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_268, 127);  clamp_min_268 = None
	        _assert_tensor_metadata_803 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_267, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_803 = None
	        _assert_tensor_metadata_804 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_178, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_804 = None
	        convert_element_type_534: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_178, torch.int8);  clamp_max_178 = None
	        view_1397: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_267, [sym_size_int, 1500, 1])
	        view_1398: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_534, [sym_size_int, 1500, 1])
	        reciprocal_89: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1397);  view_1397 = None
	        mul_8674: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_89, 1.0);  reciprocal_89 = None
	        mul_8677: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8628, mul_8674);  mul_8628 = mul_8674 = None
	        round_180: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_8677);  mul_8677 = None
	        add_13733: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_180, view_1398);  round_180 = view_1398 = None
	        clamp_min_269: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13733, -128);  add_13733 = None
	        clamp_max_179: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_269, 127);  clamp_min_269 = None
	        _assert_tensor_metadata_805 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_179, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_805 = None
	        convert_element_type_535: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_179, torch.int8);  clamp_max_179 = None
	        view_1401: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_267, [sym_size_int, 1500, 1]);  clamp_min_267 = None
	        view_1402: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_534, [sym_size_int, 1500, 1]);  convert_element_type_534 = None
	        _assert_tensor_metadata_806 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_535, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_806 = None
	        convert_element_type_536: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_535, torch.float32);  convert_element_type_535 = None
	        _assert_tensor_metadata_807 = torch.ops.aten._assert_tensor_metadata.default(view_1402, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_807 = None
	        convert_element_type_537: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1402, torch.float32);  view_1402 = None
	        sub_4107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_536, convert_element_type_537);  convert_element_type_536 = convert_element_type_537 = None
	        mul_8699: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4107, view_1401);  sub_4107 = view_1401 = None
	        _assert_tensor_metadata_808 = torch.ops.aten._assert_tensor_metadata.default(mul_8699, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_808 = None
	        view_1404: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_14_fc2_parametrizations_weight_original0 = None
	        view_1405: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_14_fc2_parametrizations_weight_original1 = None
	        view_1406: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_14_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_14_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_809 = torch.ops.aten._assert_tensor_metadata.default(view_1404, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_809 = None
	        convert_element_type_538: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1404, torch.float32);  view_1404 = None
	        _assert_tensor_metadata_810 = torch.ops.aten._assert_tensor_metadata.default(view_1406, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_810 = None
	        convert_element_type_539: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1406, torch.float32);  view_1406 = None
	        sub_4111: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_538, convert_element_type_539);  convert_element_type_538 = convert_element_type_539 = None
	        mul_8704: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4111, view_1405);  sub_4111 = view_1405 = None
	        view_1407: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_8704, [1280, 5120]);  mul_8704 = None
	        _assert_tensor_metadata_811 = torch.ops.aten._assert_tensor_metadata.default(view_1407, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_811 = None
	        mul_8709: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1408: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_8699, [mul_8709, 5120]);  mul_8699 = mul_8709 = None
	        permute_150: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1407, [1, 0]);  view_1407 = None
	        addmm_74: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_14_fc2_bias, view_1408, permute_150);  model_audio_tower_layers_14_fc2_bias = view_1408 = permute_150 = None
	        view_1409: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_74, [sym_size_int, 1500, 1280]);  addmm_74 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_13796: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_13498, view_1409);  add_13498 = view_1409 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_121: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_13796, memory_format = torch.contiguous_format)
	        var_mean_30 = torch.ops.aten.var_mean.correction(clone_121, [2], correction = 0, keepdim = True)
	        getitem_120: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_30[0]
	        getitem_121: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_30[1];  var_mean_30 = None
	        add_13801: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_120, 1e-05);  getitem_120 = None
	        rsqrt_30: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_13801);  add_13801 = None
	        sub_4117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_121, getitem_121);  clone_121 = getitem_121 = None
	        mul_8720: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4117, rsqrt_30);  sub_4117 = rsqrt_30 = None
	        mul_8721: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8720, model_audio_tower_layers_15_self_attn_layer_norm_weight);  mul_8720 = model_audio_tower_layers_15_self_attn_layer_norm_weight = None
	        add_13802: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_8721, model_audio_tower_layers_15_self_attn_layer_norm_bias);  mul_8721 = model_audio_tower_layers_15_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_13802, [2])
	        amax_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_13802, [2])
	        full_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_90, full_180);  amin_90 = full_180 = None
	        full_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_90, full_181);  amax_90 = full_181 = None
	        sub_4128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_90, minimum_90);  maximum_90 = None
	        div_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4128, 255.0);  sub_4128 = None
	        clamp_min_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_180, 1.1920928955078125e-07);  div_180 = None
	        div_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_90, clamp_min_270);  minimum_90 = None
	        round_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_181);  div_181 = None
	        sub_4134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_181);  round_181 = None
	        clamp_min_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4134, -128);  sub_4134 = None
	        clamp_max_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_271, 127);  clamp_min_271 = None
	        _assert_tensor_metadata_812 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_270, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_812 = None
	        _assert_tensor_metadata_813 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_180, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_813 = None
	        convert_element_type_540: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_180, torch.int8);  clamp_max_180 = None
	        view_1412: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_270, [sym_size_int, 1500, 1])
	        view_1413: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_540, [sym_size_int, 1500, 1])
	        reciprocal_90: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1412);  view_1412 = None
	        mul_8769: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_90, 1.0);  reciprocal_90 = None
	        mul_8772: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_13802, mul_8769);  mul_8769 = None
	        round_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8772);  mul_8772 = None
	        add_13889: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_182, view_1413);  round_182 = view_1413 = None
	        clamp_min_272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13889, -128);  add_13889 = None
	        clamp_max_181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_272, 127);  clamp_min_272 = None
	        _assert_tensor_metadata_814 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_181, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_814 = None
	        convert_element_type_541: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_181, torch.int8);  clamp_max_181 = None
	        view_1416: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_270, [sym_size_int, 1500, 1]);  clamp_min_270 = None
	        view_1417: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_540, [sym_size_int, 1500, 1]);  convert_element_type_540 = None
	        _assert_tensor_metadata_815 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_541, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_815 = None
	        convert_element_type_542: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_541, torch.float32);  convert_element_type_541 = None
	        _assert_tensor_metadata_816 = torch.ops.aten._assert_tensor_metadata.default(view_1417, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_816 = None
	        convert_element_type_543: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1417, torch.float32);  view_1417 = None
	        sub_4154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_542, convert_element_type_543);  convert_element_type_542 = convert_element_type_543 = None
	        mul_8794: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4154, view_1416);  sub_4154 = view_1416 = None
	        _assert_tensor_metadata_817 = torch.ops.aten._assert_tensor_metadata.default(mul_8794, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_817 = None
	        view_1419: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1420: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1421: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_818 = torch.ops.aten._assert_tensor_metadata.default(view_1419, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_818 = None
	        convert_element_type_544: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1419, torch.float32);  view_1419 = None
	        _assert_tensor_metadata_819 = torch.ops.aten._assert_tensor_metadata.default(view_1421, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_819 = None
	        convert_element_type_545: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1421, torch.float32);  view_1421 = None
	        sub_4158: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_544, convert_element_type_545);  convert_element_type_544 = convert_element_type_545 = None
	        mul_8799: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4158, view_1420);  sub_4158 = view_1420 = None
	        view_1422: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8799, [1280, 1280]);  mul_8799 = None
	        _assert_tensor_metadata_820 = torch.ops.aten._assert_tensor_metadata.default(view_1422, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_820 = None
	        mul_8804: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1423: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8794, [mul_8804, 1280]);  mul_8794 = mul_8804 = None
	        permute_151: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1422, [1, 0]);  view_1422 = None
	        addmm_75: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_15_self_attn_q_proj_bias, view_1423, permute_151);  model_audio_tower_layers_15_self_attn_q_proj_bias = view_1423 = permute_151 = None
	        view_1424: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_75, [sym_size_int, 1500, 1280]);  addmm_75 = None
	        mul_8811: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1424, 0.125);  view_1424 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1425: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_8811, [sym_size_int, 1500, 20, 64]);  mul_8811 = None
	        permute_152: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1425, [0, 2, 1, 3]);  view_1425 = None
	        clone_122: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_152, memory_format = torch.contiguous_format);  permute_152 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_13802, [2])
	        amax_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_13802, [2])
	        full_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_91, full_182);  amin_91 = full_182 = None
	        full_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_91, full_183);  amax_91 = full_183 = None
	        sub_4173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_91, minimum_91);  maximum_91 = None
	        div_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4173, 255.0);  sub_4173 = None
	        clamp_min_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_182, 1.1920928955078125e-07);  div_182 = None
	        div_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_91, clamp_min_273);  minimum_91 = None
	        round_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_183);  div_183 = None
	        sub_4179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_183);  round_183 = None
	        clamp_min_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4179, -128);  sub_4179 = None
	        clamp_max_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_274, 127);  clamp_min_274 = None
	        _assert_tensor_metadata_821 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_273, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_821 = None
	        _assert_tensor_metadata_822 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_182, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_822 = None
	        convert_element_type_546: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_182, torch.int8);  clamp_max_182 = None
	        view_1428: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_273, [sym_size_int, 1500, 1])
	        view_1429: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_546, [sym_size_int, 1500, 1])
	        reciprocal_91: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1428);  view_1428 = None
	        mul_8865: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_91, 1.0);  reciprocal_91 = None
	        mul_8868: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_13802, mul_8865);  mul_8865 = None
	        round_184: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8868);  mul_8868 = None
	        add_14041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_184, view_1429);  round_184 = view_1429 = None
	        clamp_min_275: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14041, -128);  add_14041 = None
	        clamp_max_183: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_275, 127);  clamp_min_275 = None
	        _assert_tensor_metadata_823 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_183, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_823 = None
	        convert_element_type_547: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_183, torch.int8);  clamp_max_183 = None
	        view_1432: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_273, [sym_size_int, 1500, 1]);  clamp_min_273 = None
	        view_1433: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_546, [sym_size_int, 1500, 1]);  convert_element_type_546 = None
	        _assert_tensor_metadata_824 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_547, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_824 = None
	        convert_element_type_548: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_547, torch.float32);  convert_element_type_547 = None
	        _assert_tensor_metadata_825 = torch.ops.aten._assert_tensor_metadata.default(view_1433, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_825 = None
	        convert_element_type_549: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1433, torch.float32);  view_1433 = None
	        sub_4199: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_548, convert_element_type_549);  convert_element_type_548 = convert_element_type_549 = None
	        mul_8890: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4199, view_1432);  sub_4199 = view_1432 = None
	        _assert_tensor_metadata_826 = torch.ops.aten._assert_tensor_metadata.default(mul_8890, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_826 = None
	        view_1435: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1436: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1437: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_827 = torch.ops.aten._assert_tensor_metadata.default(view_1435, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_827 = None
	        convert_element_type_550: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1435, torch.float32);  view_1435 = None
	        _assert_tensor_metadata_828 = torch.ops.aten._assert_tensor_metadata.default(view_1437, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_828 = None
	        convert_element_type_551: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1437, torch.float32);  view_1437 = None
	        sub_4203: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_550, convert_element_type_551);  convert_element_type_550 = convert_element_type_551 = None
	        mul_8895: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4203, view_1436);  sub_4203 = view_1436 = None
	        view_1438: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8895, [1280, 1280]);  mul_8895 = None
	        _assert_tensor_metadata_829 = torch.ops.aten._assert_tensor_metadata.default(view_1438, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_829 = None
	        permute_153: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1438, [1, 0]);  view_1438 = None
	        mul_8898: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1439: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8890, [mul_8898, 1280]);  mul_8890 = mul_8898 = None
	        mm_15: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1439, permute_153);  view_1439 = permute_153 = None
	        view_1440: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_15, [sym_size_int, 1500, 1280]);  mm_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1441: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1440, [sym_size_int, -1, 20, 64]);  view_1440 = None
	        permute_154: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1441, [0, 2, 1, 3]);  view_1441 = None
	        clone_123: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_154, memory_format = torch.contiguous_format);  permute_154 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_13802, [2])
	        amax_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_13802, [2])
	        full_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_92, full_184);  amin_92 = full_184 = None
	        full_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_92, full_185);  amax_92 = full_185 = None
	        sub_4217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_92, minimum_92);  maximum_92 = None
	        div_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4217, 255.0);  sub_4217 = None
	        clamp_min_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_184, 1.1920928955078125e-07);  div_184 = None
	        div_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_92, clamp_min_276);  minimum_92 = None
	        round_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_185);  div_185 = None
	        sub_4223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_185);  round_185 = None
	        clamp_min_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4223, -128);  sub_4223 = None
	        clamp_max_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_277, 127);  clamp_min_277 = None
	        _assert_tensor_metadata_830 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_276, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_830 = None
	        _assert_tensor_metadata_831 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_184, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_831 = None
	        convert_element_type_552: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_184, torch.int8);  clamp_max_184 = None
	        view_1444: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_276, [sym_size_int, 1500, 1])
	        view_1445: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_552, [sym_size_int, 1500, 1])
	        reciprocal_92: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1444);  view_1444 = None
	        mul_8964: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_92, 1.0);  reciprocal_92 = None
	        mul_8967: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_13802, mul_8964);  add_13802 = mul_8964 = None
	        round_186: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8967);  mul_8967 = None
	        add_14189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_186, view_1445);  round_186 = view_1445 = None
	        clamp_min_278: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14189, -128);  add_14189 = None
	        clamp_max_185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_278, 127);  clamp_min_278 = None
	        _assert_tensor_metadata_832 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_185, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_832 = None
	        convert_element_type_553: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_185, torch.int8);  clamp_max_185 = None
	        view_1448: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_276, [sym_size_int, 1500, 1]);  clamp_min_276 = None
	        view_1449: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_552, [sym_size_int, 1500, 1]);  convert_element_type_552 = None
	        _assert_tensor_metadata_833 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_553, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_833 = None
	        convert_element_type_554: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_553, torch.float32);  convert_element_type_553 = None
	        _assert_tensor_metadata_834 = torch.ops.aten._assert_tensor_metadata.default(view_1449, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_834 = None
	        convert_element_type_555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1449, torch.float32);  view_1449 = None
	        sub_4243: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_554, convert_element_type_555);  convert_element_type_554 = convert_element_type_555 = None
	        mul_8989: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4243, view_1448);  sub_4243 = view_1448 = None
	        _assert_tensor_metadata_835 = torch.ops.aten._assert_tensor_metadata.default(mul_8989, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_835 = None
	        view_1451: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1452: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1453: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_836 = torch.ops.aten._assert_tensor_metadata.default(view_1451, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_836 = None
	        convert_element_type_556: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1451, torch.float32);  view_1451 = None
	        _assert_tensor_metadata_837 = torch.ops.aten._assert_tensor_metadata.default(view_1453, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_837 = None
	        convert_element_type_557: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1453, torch.float32);  view_1453 = None
	        sub_4247: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_556, convert_element_type_557);  convert_element_type_556 = convert_element_type_557 = None
	        mul_8994: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4247, view_1452);  sub_4247 = view_1452 = None
	        view_1454: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8994, [1280, 1280]);  mul_8994 = None
	        _assert_tensor_metadata_838 = torch.ops.aten._assert_tensor_metadata.default(view_1454, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_838 = None
	        mul_8999: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1455: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_8989, [mul_8999, 1280]);  mul_8989 = mul_8999 = None
	        permute_155: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1454, [1, 0]);  view_1454 = None
	        addmm_76: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_15_self_attn_v_proj_bias, view_1455, permute_155);  model_audio_tower_layers_15_self_attn_v_proj_bias = view_1455 = permute_155 = None
	        view_1456: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_76, [sym_size_int, 1500, 1280]);  addmm_76 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1457: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1456, [sym_size_int, -1, 20, 64]);  view_1456 = None
	        permute_156: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1457, [0, 2, 1, 3]);  view_1457 = None
	        clone_124: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_156, memory_format = torch.contiguous_format);  permute_156 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_15 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_122, clone_123, clone_124, None, False, scale = 1.0);  clone_122 = clone_123 = clone_124 = None
	        getitem_122: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_15[0];  _scaled_dot_product_efficient_attention_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_157: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_122, [0, 2, 1, 3]);  getitem_122 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_157, [sym_size_int, 1500, -1]);  permute_157 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1458, [2])
	        amax_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1458, [2])
	        full_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_93, full_186);  amin_93 = full_186 = None
	        full_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_93, full_187);  amax_93 = full_187 = None
	        sub_4265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_93, minimum_93);  maximum_93 = None
	        div_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4265, 255.0);  sub_4265 = None
	        clamp_min_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_186, 1.1920928955078125e-07);  div_186 = None
	        div_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_93, clamp_min_279);  minimum_93 = None
	        round_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_187);  div_187 = None
	        sub_4271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_187);  round_187 = None
	        clamp_min_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4271, -128);  sub_4271 = None
	        clamp_max_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_280, 127);  clamp_min_280 = None
	        _assert_tensor_metadata_839 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_279, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_839 = None
	        _assert_tensor_metadata_840 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_186, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_840 = None
	        convert_element_type_558: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_186, torch.int8);  clamp_max_186 = None
	        view_1461: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_279, [sym_size_int, 1500, 1])
	        view_1462: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_558, [sym_size_int, 1500, 1])
	        reciprocal_93: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1461);  view_1461 = None
	        mul_9069: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_93, 1.0);  reciprocal_93 = None
	        mul_9072: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1458, mul_9069);  view_1458 = mul_9069 = None
	        round_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9072);  mul_9072 = None
	        add_14353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_188, view_1462);  round_188 = view_1462 = None
	        clamp_min_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14353, -128);  add_14353 = None
	        clamp_max_187: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_281, 127);  clamp_min_281 = None
	        _assert_tensor_metadata_841 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_187, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_841 = None
	        convert_element_type_559: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_187, torch.int8);  clamp_max_187 = None
	        view_1465: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_279, [sym_size_int, 1500, 1]);  clamp_min_279 = None
	        view_1466: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_558, [sym_size_int, 1500, 1]);  convert_element_type_558 = None
	        _assert_tensor_metadata_842 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_559, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_842 = None
	        convert_element_type_560: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_559, torch.float32);  convert_element_type_559 = None
	        _assert_tensor_metadata_843 = torch.ops.aten._assert_tensor_metadata.default(view_1466, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_843 = None
	        convert_element_type_561: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1466, torch.float32);  view_1466 = None
	        sub_4291: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_560, convert_element_type_561);  convert_element_type_560 = convert_element_type_561 = None
	        mul_9094: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4291, view_1465);  sub_4291 = view_1465 = None
	        _assert_tensor_metadata_844 = torch.ops.aten._assert_tensor_metadata.default(mul_9094, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_844 = None
	        view_1468: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1469: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1470: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_845 = torch.ops.aten._assert_tensor_metadata.default(view_1468, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_845 = None
	        convert_element_type_562: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1468, torch.float32);  view_1468 = None
	        _assert_tensor_metadata_846 = torch.ops.aten._assert_tensor_metadata.default(view_1470, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_846 = None
	        convert_element_type_563: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1470, torch.float32);  view_1470 = None
	        sub_4295: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_562, convert_element_type_563);  convert_element_type_562 = convert_element_type_563 = None
	        mul_9099: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4295, view_1469);  sub_4295 = view_1469 = None
	        view_1471: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9099, [1280, 1280]);  mul_9099 = None
	        _assert_tensor_metadata_847 = torch.ops.aten._assert_tensor_metadata.default(view_1471, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_847 = None
	        mul_9104: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1472: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9094, [mul_9104, 1280]);  mul_9094 = mul_9104 = None
	        permute_158: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1471, [1, 0]);  view_1471 = None
	        addmm_77: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_15_self_attn_out_proj_bias, view_1472, permute_158);  model_audio_tower_layers_15_self_attn_out_proj_bias = view_1472 = permute_158 = None
	        view_1473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_77, [sym_size_int, 1500, 1280]);  addmm_77 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_14416: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_13796, view_1473);  add_13796 = view_1473 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_126: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_14416, memory_format = torch.contiguous_format)
	        var_mean_31 = torch.ops.aten.var_mean.correction(clone_126, [2], correction = 0, keepdim = True)
	        getitem_126: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_31[0]
	        getitem_127: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_31[1];  var_mean_31 = None
	        add_14421: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_126, 1e-05);  getitem_126 = None
	        rsqrt_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_14421);  add_14421 = None
	        sub_4301: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_126, getitem_127);  clone_126 = getitem_127 = None
	        mul_9115: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4301, rsqrt_31);  sub_4301 = rsqrt_31 = None
	        mul_9116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9115, model_audio_tower_layers_15_final_layer_norm_weight);  mul_9115 = model_audio_tower_layers_15_final_layer_norm_weight = None
	        add_14422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9116, model_audio_tower_layers_15_final_layer_norm_bias);  mul_9116 = model_audio_tower_layers_15_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_14422, [2])
	        amax_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_14422, [2])
	        full_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_94, full_188);  amin_94 = full_188 = None
	        full_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_94, full_189);  amax_94 = full_189 = None
	        sub_4312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_94, minimum_94);  maximum_94 = None
	        div_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4312, 255.0);  sub_4312 = None
	        clamp_min_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_188, 1.1920928955078125e-07);  div_188 = None
	        div_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_94, clamp_min_282);  minimum_94 = None
	        round_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_189);  div_189 = None
	        sub_4318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_189);  round_189 = None
	        clamp_min_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4318, -128);  sub_4318 = None
	        clamp_max_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_283, 127);  clamp_min_283 = None
	        _assert_tensor_metadata_848 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_282, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_848 = None
	        _assert_tensor_metadata_849 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_188, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_849 = None
	        convert_element_type_564: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_188, torch.int8);  clamp_max_188 = None
	        view_1476: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_282, [sym_size_int, 1500, 1])
	        view_1477: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_564, [sym_size_int, 1500, 1])
	        reciprocal_94: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1476);  view_1476 = None
	        mul_9164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_94, 1.0);  reciprocal_94 = None
	        mul_9167: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_14422, mul_9164);  add_14422 = mul_9164 = None
	        round_190: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9167);  mul_9167 = None
	        add_14509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_190, view_1477);  round_190 = view_1477 = None
	        clamp_min_284: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14509, -128);  add_14509 = None
	        clamp_max_189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_284, 127);  clamp_min_284 = None
	        _assert_tensor_metadata_850 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_189, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_850 = None
	        convert_element_type_565: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_189, torch.int8);  clamp_max_189 = None
	        view_1480: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_282, [sym_size_int, 1500, 1]);  clamp_min_282 = None
	        view_1481: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_564, [sym_size_int, 1500, 1]);  convert_element_type_564 = None
	        _assert_tensor_metadata_851 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_565, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_851 = None
	        convert_element_type_566: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_565, torch.float32);  convert_element_type_565 = None
	        _assert_tensor_metadata_852 = torch.ops.aten._assert_tensor_metadata.default(view_1481, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_852 = None
	        convert_element_type_567: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1481, torch.float32);  view_1481 = None
	        sub_4338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_566, convert_element_type_567);  convert_element_type_566 = convert_element_type_567 = None
	        mul_9189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4338, view_1480);  sub_4338 = view_1480 = None
	        _assert_tensor_metadata_853 = torch.ops.aten._assert_tensor_metadata.default(mul_9189, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_853 = None
	        view_1483: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_15_fc1_parametrizations_weight_original0 = None
	        view_1484: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_15_fc1_parametrizations_weight_original1 = None
	        view_1485: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_15_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_854 = torch.ops.aten._assert_tensor_metadata.default(view_1483, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_854 = None
	        convert_element_type_568: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1483, torch.float32);  view_1483 = None
	        _assert_tensor_metadata_855 = torch.ops.aten._assert_tensor_metadata.default(view_1485, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_855 = None
	        convert_element_type_569: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1485, torch.float32);  view_1485 = None
	        sub_4342: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_568, convert_element_type_569);  convert_element_type_568 = convert_element_type_569 = None
	        mul_9194: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4342, view_1484);  sub_4342 = view_1484 = None
	        view_1486: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9194, [5120, 1280]);  mul_9194 = None
	        _assert_tensor_metadata_856 = torch.ops.aten._assert_tensor_metadata.default(view_1486, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_856 = None
	        mul_9199: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1487: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9189, [mul_9199, 1280]);  mul_9189 = mul_9199 = None
	        permute_159: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1486, [1, 0]);  view_1486 = None
	        addmm_78: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_15_fc1_bias, view_1487, permute_159);  model_audio_tower_layers_15_fc1_bias = view_1487 = permute_159 = None
	        view_1488: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_78, [sym_size_int, 1500, 5120]);  addmm_78 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_9206: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1488, 0.5)
	        mul_9207: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1488, 0.7071067811865476);  view_1488 = None
	        erf_17: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_9207);  mul_9207 = None
	        add_14568: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_17, 1);  erf_17 = None
	        mul_9208: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9206, add_14568);  mul_9206 = add_14568 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_9208, [2])
	        amax_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_9208, [2])
	        full_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_95, full_190);  amin_95 = full_190 = None
	        full_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_95, full_191);  amax_95 = full_191 = None
	        sub_4355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_95, minimum_95);  maximum_95 = None
	        div_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4355, 255.0);  sub_4355 = None
	        clamp_min_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_190, 1.1920928955078125e-07);  div_190 = None
	        div_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_95, clamp_min_285);  minimum_95 = None
	        round_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_191);  div_191 = None
	        sub_4361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_191);  round_191 = None
	        clamp_min_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4361, -128);  sub_4361 = None
	        clamp_max_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_286, 127);  clamp_min_286 = None
	        _assert_tensor_metadata_857 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_285, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_857 = None
	        _assert_tensor_metadata_858 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_190, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_858 = None
	        convert_element_type_570: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_190, torch.int8);  clamp_max_190 = None
	        view_1491: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_285, [sym_size_int, 1500, 1])
	        view_1492: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_570, [sym_size_int, 1500, 1])
	        reciprocal_95: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1491);  view_1491 = None
	        mul_9254: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_95, 1.0);  reciprocal_95 = None
	        mul_9257: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9208, mul_9254);  mul_9208 = mul_9254 = None
	        round_192: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_9257);  mul_9257 = None
	        add_14651: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_192, view_1492);  round_192 = view_1492 = None
	        clamp_min_287: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14651, -128);  add_14651 = None
	        clamp_max_191: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_287, 127);  clamp_min_287 = None
	        _assert_tensor_metadata_859 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_191, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_859 = None
	        convert_element_type_571: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_191, torch.int8);  clamp_max_191 = None
	        view_1495: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_285, [sym_size_int, 1500, 1]);  clamp_min_285 = None
	        view_1496: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_570, [sym_size_int, 1500, 1]);  convert_element_type_570 = None
	        _assert_tensor_metadata_860 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_571, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_860 = None
	        convert_element_type_572: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_571, torch.float32);  convert_element_type_571 = None
	        _assert_tensor_metadata_861 = torch.ops.aten._assert_tensor_metadata.default(view_1496, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_861 = None
	        convert_element_type_573: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1496, torch.float32);  view_1496 = None
	        sub_4381: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_572, convert_element_type_573);  convert_element_type_572 = convert_element_type_573 = None
	        mul_9279: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4381, view_1495);  sub_4381 = view_1495 = None
	        _assert_tensor_metadata_862 = torch.ops.aten._assert_tensor_metadata.default(mul_9279, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_862 = None
	        view_1498: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_15_fc2_parametrizations_weight_original0 = None
	        view_1499: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_15_fc2_parametrizations_weight_original1 = None
	        view_1500: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_15_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_15_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_863 = torch.ops.aten._assert_tensor_metadata.default(view_1498, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_863 = None
	        convert_element_type_574: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1498, torch.float32);  view_1498 = None
	        _assert_tensor_metadata_864 = torch.ops.aten._assert_tensor_metadata.default(view_1500, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_864 = None
	        convert_element_type_575: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1500, torch.float32);  view_1500 = None
	        sub_4385: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_574, convert_element_type_575);  convert_element_type_574 = convert_element_type_575 = None
	        mul_9284: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4385, view_1499);  sub_4385 = view_1499 = None
	        view_1501: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_9284, [1280, 5120]);  mul_9284 = None
	        _assert_tensor_metadata_865 = torch.ops.aten._assert_tensor_metadata.default(view_1501, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_865 = None
	        mul_9289: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1502: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_9279, [mul_9289, 5120]);  mul_9279 = mul_9289 = None
	        permute_160: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1501, [1, 0]);  view_1501 = None
	        addmm_79: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_15_fc2_bias, view_1502, permute_160);  model_audio_tower_layers_15_fc2_bias = view_1502 = permute_160 = None
	        view_1503: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_79, [sym_size_int, 1500, 1280]);  addmm_79 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_14714: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_14416, view_1503);  add_14416 = view_1503 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_129: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_14714, memory_format = torch.contiguous_format)
	        var_mean_32 = torch.ops.aten.var_mean.correction(clone_129, [2], correction = 0, keepdim = True)
	        getitem_128: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_32[0]
	        getitem_129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_32[1];  var_mean_32 = None
	        add_14719: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_128, 1e-05);  getitem_128 = None
	        rsqrt_32: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_14719);  add_14719 = None
	        sub_4391: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_129, getitem_129);  clone_129 = getitem_129 = None
	        mul_9300: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4391, rsqrt_32);  sub_4391 = rsqrt_32 = None
	        mul_9301: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9300, model_audio_tower_layers_16_self_attn_layer_norm_weight);  mul_9300 = model_audio_tower_layers_16_self_attn_layer_norm_weight = None
	        add_14720: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9301, model_audio_tower_layers_16_self_attn_layer_norm_bias);  mul_9301 = model_audio_tower_layers_16_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_14720, [2])
	        amax_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_14720, [2])
	        full_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_96, full_192);  amin_96 = full_192 = None
	        full_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_96, full_193);  amax_96 = full_193 = None
	        sub_4402: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_96, minimum_96);  maximum_96 = None
	        div_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4402, 255.0);  sub_4402 = None
	        clamp_min_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_192, 1.1920928955078125e-07);  div_192 = None
	        div_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_96, clamp_min_288);  minimum_96 = None
	        round_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_193);  div_193 = None
	        sub_4408: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_193);  round_193 = None
	        clamp_min_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4408, -128);  sub_4408 = None
	        clamp_max_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_289, 127);  clamp_min_289 = None
	        _assert_tensor_metadata_866 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_288, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_866 = None
	        _assert_tensor_metadata_867 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_192, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_867 = None
	        convert_element_type_576: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_192, torch.int8);  clamp_max_192 = None
	        view_1506: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_288, [sym_size_int, 1500, 1])
	        view_1507: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_576, [sym_size_int, 1500, 1])
	        reciprocal_96: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1506);  view_1506 = None
	        mul_9349: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_96, 1.0);  reciprocal_96 = None
	        mul_9352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_14720, mul_9349);  mul_9349 = None
	        round_194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9352);  mul_9352 = None
	        add_14807: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_194, view_1507);  round_194 = view_1507 = None
	        clamp_min_290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14807, -128);  add_14807 = None
	        clamp_max_193: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_290, 127);  clamp_min_290 = None
	        _assert_tensor_metadata_868 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_193, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_868 = None
	        convert_element_type_577: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_193, torch.int8);  clamp_max_193 = None
	        view_1510: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_288, [sym_size_int, 1500, 1]);  clamp_min_288 = None
	        view_1511: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_576, [sym_size_int, 1500, 1]);  convert_element_type_576 = None
	        _assert_tensor_metadata_869 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_577, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_869 = None
	        convert_element_type_578: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_577, torch.float32);  convert_element_type_577 = None
	        _assert_tensor_metadata_870 = torch.ops.aten._assert_tensor_metadata.default(view_1511, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_870 = None
	        convert_element_type_579: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1511, torch.float32);  view_1511 = None
	        sub_4428: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_578, convert_element_type_579);  convert_element_type_578 = convert_element_type_579 = None
	        mul_9374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4428, view_1510);  sub_4428 = view_1510 = None
	        _assert_tensor_metadata_871 = torch.ops.aten._assert_tensor_metadata.default(mul_9374, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_871 = None
	        view_1513: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1514: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1515: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_872 = torch.ops.aten._assert_tensor_metadata.default(view_1513, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_872 = None
	        convert_element_type_580: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1513, torch.float32);  view_1513 = None
	        _assert_tensor_metadata_873 = torch.ops.aten._assert_tensor_metadata.default(view_1515, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_873 = None
	        convert_element_type_581: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1515, torch.float32);  view_1515 = None
	        sub_4432: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_580, convert_element_type_581);  convert_element_type_580 = convert_element_type_581 = None
	        mul_9379: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4432, view_1514);  sub_4432 = view_1514 = None
	        view_1516: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9379, [1280, 1280]);  mul_9379 = None
	        _assert_tensor_metadata_874 = torch.ops.aten._assert_tensor_metadata.default(view_1516, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_874 = None
	        mul_9384: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1517: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9374, [mul_9384, 1280]);  mul_9374 = mul_9384 = None
	        permute_161: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1516, [1, 0]);  view_1516 = None
	        addmm_80: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_16_self_attn_q_proj_bias, view_1517, permute_161);  model_audio_tower_layers_16_self_attn_q_proj_bias = view_1517 = permute_161 = None
	        view_1518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_80, [sym_size_int, 1500, 1280]);  addmm_80 = None
	        mul_9391: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1518, 0.125);  view_1518 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1519: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_9391, [sym_size_int, 1500, 20, 64]);  mul_9391 = None
	        permute_162: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1519, [0, 2, 1, 3]);  view_1519 = None
	        clone_130: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_162, memory_format = torch.contiguous_format);  permute_162 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_14720, [2])
	        amax_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_14720, [2])
	        full_194: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_97, full_194);  amin_97 = full_194 = None
	        full_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_97, full_195);  amax_97 = full_195 = None
	        sub_4447: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_97, minimum_97);  maximum_97 = None
	        div_194: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4447, 255.0);  sub_4447 = None
	        clamp_min_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_194, 1.1920928955078125e-07);  div_194 = None
	        div_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_97, clamp_min_291);  minimum_97 = None
	        round_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_195);  div_195 = None
	        sub_4453: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_195);  round_195 = None
	        clamp_min_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4453, -128);  sub_4453 = None
	        clamp_max_194: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_292, 127);  clamp_min_292 = None
	        _assert_tensor_metadata_875 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_291, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_875 = None
	        _assert_tensor_metadata_876 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_194, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_876 = None
	        convert_element_type_582: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_194, torch.int8);  clamp_max_194 = None
	        view_1522: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_291, [sym_size_int, 1500, 1])
	        view_1523: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_582, [sym_size_int, 1500, 1])
	        reciprocal_97: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1522);  view_1522 = None
	        mul_9445: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_97, 1.0);  reciprocal_97 = None
	        mul_9448: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_14720, mul_9445);  mul_9445 = None
	        round_196: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9448);  mul_9448 = None
	        add_14959: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_196, view_1523);  round_196 = view_1523 = None
	        clamp_min_293: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14959, -128);  add_14959 = None
	        clamp_max_195: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_293, 127);  clamp_min_293 = None
	        _assert_tensor_metadata_877 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_195, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_877 = None
	        convert_element_type_583: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_195, torch.int8);  clamp_max_195 = None
	        view_1526: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_291, [sym_size_int, 1500, 1]);  clamp_min_291 = None
	        view_1527: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_582, [sym_size_int, 1500, 1]);  convert_element_type_582 = None
	        _assert_tensor_metadata_878 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_583, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_878 = None
	        convert_element_type_584: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_583, torch.float32);  convert_element_type_583 = None
	        _assert_tensor_metadata_879 = torch.ops.aten._assert_tensor_metadata.default(view_1527, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_879 = None
	        convert_element_type_585: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1527, torch.float32);  view_1527 = None
	        sub_4473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_584, convert_element_type_585);  convert_element_type_584 = convert_element_type_585 = None
	        mul_9470: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4473, view_1526);  sub_4473 = view_1526 = None
	        _assert_tensor_metadata_880 = torch.ops.aten._assert_tensor_metadata.default(mul_9470, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_880 = None
	        view_1529: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1530: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1531: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_881 = torch.ops.aten._assert_tensor_metadata.default(view_1529, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_881 = None
	        convert_element_type_586: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1529, torch.float32);  view_1529 = None
	        _assert_tensor_metadata_882 = torch.ops.aten._assert_tensor_metadata.default(view_1531, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_882 = None
	        convert_element_type_587: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1531, torch.float32);  view_1531 = None
	        sub_4477: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_586, convert_element_type_587);  convert_element_type_586 = convert_element_type_587 = None
	        mul_9475: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4477, view_1530);  sub_4477 = view_1530 = None
	        view_1532: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9475, [1280, 1280]);  mul_9475 = None
	        _assert_tensor_metadata_883 = torch.ops.aten._assert_tensor_metadata.default(view_1532, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_883 = None
	        permute_163: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1532, [1, 0]);  view_1532 = None
	        mul_9478: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1533: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9470, [mul_9478, 1280]);  mul_9470 = mul_9478 = None
	        mm_16: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1533, permute_163);  view_1533 = permute_163 = None
	        view_1534: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_16, [sym_size_int, 1500, 1280]);  mm_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1535: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1534, [sym_size_int, -1, 20, 64]);  view_1534 = None
	        permute_164: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1535, [0, 2, 1, 3]);  view_1535 = None
	        clone_131: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_164, memory_format = torch.contiguous_format);  permute_164 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_14720, [2])
	        amax_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_14720, [2])
	        full_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_98, full_196);  amin_98 = full_196 = None
	        full_197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_98, full_197);  amax_98 = full_197 = None
	        sub_4491: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_98, minimum_98);  maximum_98 = None
	        div_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4491, 255.0);  sub_4491 = None
	        clamp_min_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_196, 1.1920928955078125e-07);  div_196 = None
	        div_197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_98, clamp_min_294);  minimum_98 = None
	        round_197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_197);  div_197 = None
	        sub_4497: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_197);  round_197 = None
	        clamp_min_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4497, -128);  sub_4497 = None
	        clamp_max_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_295, 127);  clamp_min_295 = None
	        _assert_tensor_metadata_884 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_884 = None
	        _assert_tensor_metadata_885 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_196, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_885 = None
	        convert_element_type_588: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_196, torch.int8);  clamp_max_196 = None
	        view_1538: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_294, [sym_size_int, 1500, 1])
	        view_1539: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_588, [sym_size_int, 1500, 1])
	        reciprocal_98: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1538);  view_1538 = None
	        mul_9544: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_98, 1.0);  reciprocal_98 = None
	        mul_9547: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_14720, mul_9544);  add_14720 = mul_9544 = None
	        round_198: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9547);  mul_9547 = None
	        add_15107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_198, view_1539);  round_198 = view_1539 = None
	        clamp_min_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15107, -128);  add_15107 = None
	        clamp_max_197: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_296, 127);  clamp_min_296 = None
	        _assert_tensor_metadata_886 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_197, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_886 = None
	        convert_element_type_589: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_197, torch.int8);  clamp_max_197 = None
	        view_1542: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_294, [sym_size_int, 1500, 1]);  clamp_min_294 = None
	        view_1543: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_588, [sym_size_int, 1500, 1]);  convert_element_type_588 = None
	        _assert_tensor_metadata_887 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_589, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_887 = None
	        convert_element_type_590: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_589, torch.float32);  convert_element_type_589 = None
	        _assert_tensor_metadata_888 = torch.ops.aten._assert_tensor_metadata.default(view_1543, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_888 = None
	        convert_element_type_591: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1543, torch.float32);  view_1543 = None
	        sub_4517: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_590, convert_element_type_591);  convert_element_type_590 = convert_element_type_591 = None
	        mul_9569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4517, view_1542);  sub_4517 = view_1542 = None
	        _assert_tensor_metadata_889 = torch.ops.aten._assert_tensor_metadata.default(mul_9569, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_889 = None
	        view_1545: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1546: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1547: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_890 = torch.ops.aten._assert_tensor_metadata.default(view_1545, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_890 = None
	        convert_element_type_592: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1545, torch.float32);  view_1545 = None
	        _assert_tensor_metadata_891 = torch.ops.aten._assert_tensor_metadata.default(view_1547, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_891 = None
	        convert_element_type_593: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1547, torch.float32);  view_1547 = None
	        sub_4521: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_592, convert_element_type_593);  convert_element_type_592 = convert_element_type_593 = None
	        mul_9574: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4521, view_1546);  sub_4521 = view_1546 = None
	        view_1548: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9574, [1280, 1280]);  mul_9574 = None
	        _assert_tensor_metadata_892 = torch.ops.aten._assert_tensor_metadata.default(view_1548, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_892 = None
	        mul_9579: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1549: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9569, [mul_9579, 1280]);  mul_9569 = mul_9579 = None
	        permute_165: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1548, [1, 0]);  view_1548 = None
	        addmm_81: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_16_self_attn_v_proj_bias, view_1549, permute_165);  model_audio_tower_layers_16_self_attn_v_proj_bias = view_1549 = permute_165 = None
	        view_1550: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_81, [sym_size_int, 1500, 1280]);  addmm_81 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1551: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1550, [sym_size_int, -1, 20, 64]);  view_1550 = None
	        permute_166: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1551, [0, 2, 1, 3]);  view_1551 = None
	        clone_132: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_166, memory_format = torch.contiguous_format);  permute_166 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_16 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_130, clone_131, clone_132, None, False, scale = 1.0);  clone_130 = clone_131 = clone_132 = None
	        getitem_130: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_16[0];  _scaled_dot_product_efficient_attention_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_167: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_130, [0, 2, 1, 3]);  getitem_130 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_167, [sym_size_int, 1500, -1]);  permute_167 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1552, [2])
	        amax_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1552, [2])
	        full_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_99, full_198);  amin_99 = full_198 = None
	        full_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_99, full_199);  amax_99 = full_199 = None
	        sub_4539: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_99, minimum_99);  maximum_99 = None
	        div_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4539, 255.0);  sub_4539 = None
	        clamp_min_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_198, 1.1920928955078125e-07);  div_198 = None
	        div_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_99, clamp_min_297);  minimum_99 = None
	        round_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_199);  div_199 = None
	        sub_4545: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_199);  round_199 = None
	        clamp_min_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4545, -128);  sub_4545 = None
	        clamp_max_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_298, 127);  clamp_min_298 = None
	        _assert_tensor_metadata_893 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_297, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_893 = None
	        _assert_tensor_metadata_894 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_198, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_894 = None
	        convert_element_type_594: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_198, torch.int8);  clamp_max_198 = None
	        view_1555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_297, [sym_size_int, 1500, 1])
	        view_1556: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_594, [sym_size_int, 1500, 1])
	        reciprocal_99: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1555);  view_1555 = None
	        mul_9649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_99, 1.0);  reciprocal_99 = None
	        mul_9652: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1552, mul_9649);  view_1552 = mul_9649 = None
	        round_200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9652);  mul_9652 = None
	        add_15271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_200, view_1556);  round_200 = view_1556 = None
	        clamp_min_299: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15271, -128);  add_15271 = None
	        clamp_max_199: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_299, 127);  clamp_min_299 = None
	        _assert_tensor_metadata_895 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_199, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_895 = None
	        convert_element_type_595: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_199, torch.int8);  clamp_max_199 = None
	        view_1559: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_297, [sym_size_int, 1500, 1]);  clamp_min_297 = None
	        view_1560: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_594, [sym_size_int, 1500, 1]);  convert_element_type_594 = None
	        _assert_tensor_metadata_896 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_595, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_896 = None
	        convert_element_type_596: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_595, torch.float32);  convert_element_type_595 = None
	        _assert_tensor_metadata_897 = torch.ops.aten._assert_tensor_metadata.default(view_1560, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_897 = None
	        convert_element_type_597: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1560, torch.float32);  view_1560 = None
	        sub_4565: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_596, convert_element_type_597);  convert_element_type_596 = convert_element_type_597 = None
	        mul_9674: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4565, view_1559);  sub_4565 = view_1559 = None
	        _assert_tensor_metadata_898 = torch.ops.aten._assert_tensor_metadata.default(mul_9674, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_898 = None
	        view_1562: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1563: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1564: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_899 = torch.ops.aten._assert_tensor_metadata.default(view_1562, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_899 = None
	        convert_element_type_598: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1562, torch.float32);  view_1562 = None
	        _assert_tensor_metadata_900 = torch.ops.aten._assert_tensor_metadata.default(view_1564, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_900 = None
	        convert_element_type_599: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1564, torch.float32);  view_1564 = None
	        sub_4569: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_598, convert_element_type_599);  convert_element_type_598 = convert_element_type_599 = None
	        mul_9679: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4569, view_1563);  sub_4569 = view_1563 = None
	        view_1565: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9679, [1280, 1280]);  mul_9679 = None
	        _assert_tensor_metadata_901 = torch.ops.aten._assert_tensor_metadata.default(view_1565, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_901 = None
	        mul_9684: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1566: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9674, [mul_9684, 1280]);  mul_9674 = mul_9684 = None
	        permute_168: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1565, [1, 0]);  view_1565 = None
	        addmm_82: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_16_self_attn_out_proj_bias, view_1566, permute_168);  model_audio_tower_layers_16_self_attn_out_proj_bias = view_1566 = permute_168 = None
	        view_1567: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_82, [sym_size_int, 1500, 1280]);  addmm_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_15334: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_14714, view_1567);  add_14714 = view_1567 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_15334, memory_format = torch.contiguous_format)
	        var_mean_33 = torch.ops.aten.var_mean.correction(clone_134, [2], correction = 0, keepdim = True)
	        getitem_134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_33[0]
	        getitem_135: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_33[1];  var_mean_33 = None
	        add_15339: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_134, 1e-05);  getitem_134 = None
	        rsqrt_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_15339);  add_15339 = None
	        sub_4575: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_134, getitem_135);  clone_134 = getitem_135 = None
	        mul_9695: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4575, rsqrt_33);  sub_4575 = rsqrt_33 = None
	        mul_9696: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9695, model_audio_tower_layers_16_final_layer_norm_weight);  mul_9695 = model_audio_tower_layers_16_final_layer_norm_weight = None
	        add_15340: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9696, model_audio_tower_layers_16_final_layer_norm_bias);  mul_9696 = model_audio_tower_layers_16_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_15340, [2])
	        amax_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_15340, [2])
	        full_200: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_100, full_200);  amin_100 = full_200 = None
	        full_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_100, full_201);  amax_100 = full_201 = None
	        sub_4586: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_100, minimum_100);  maximum_100 = None
	        div_200: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4586, 255.0);  sub_4586 = None
	        clamp_min_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_200, 1.1920928955078125e-07);  div_200 = None
	        div_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_100, clamp_min_300);  minimum_100 = None
	        round_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_201);  div_201 = None
	        sub_4592: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_201);  round_201 = None
	        clamp_min_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4592, -128);  sub_4592 = None
	        clamp_max_200: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_301, 127);  clamp_min_301 = None
	        _assert_tensor_metadata_902 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_300, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_902 = None
	        _assert_tensor_metadata_903 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_200, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_903 = None
	        convert_element_type_600: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_200, torch.int8);  clamp_max_200 = None
	        view_1570: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_300, [sym_size_int, 1500, 1])
	        view_1571: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_600, [sym_size_int, 1500, 1])
	        reciprocal_100: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1570);  view_1570 = None
	        mul_9744: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_100, 1.0);  reciprocal_100 = None
	        mul_9747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_15340, mul_9744);  add_15340 = mul_9744 = None
	        round_202: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9747);  mul_9747 = None
	        add_15427: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_202, view_1571);  round_202 = view_1571 = None
	        clamp_min_302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15427, -128);  add_15427 = None
	        clamp_max_201: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_302, 127);  clamp_min_302 = None
	        _assert_tensor_metadata_904 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_201, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_904 = None
	        convert_element_type_601: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_201, torch.int8);  clamp_max_201 = None
	        view_1574: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_300, [sym_size_int, 1500, 1]);  clamp_min_300 = None
	        view_1575: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_600, [sym_size_int, 1500, 1]);  convert_element_type_600 = None
	        _assert_tensor_metadata_905 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_601, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_905 = None
	        convert_element_type_602: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_601, torch.float32);  convert_element_type_601 = None
	        _assert_tensor_metadata_906 = torch.ops.aten._assert_tensor_metadata.default(view_1575, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_906 = None
	        convert_element_type_603: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1575, torch.float32);  view_1575 = None
	        sub_4612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_602, convert_element_type_603);  convert_element_type_602 = convert_element_type_603 = None
	        mul_9769: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4612, view_1574);  sub_4612 = view_1574 = None
	        _assert_tensor_metadata_907 = torch.ops.aten._assert_tensor_metadata.default(mul_9769, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_907 = None
	        view_1577: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_16_fc1_parametrizations_weight_original0 = None
	        view_1578: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_16_fc1_parametrizations_weight_original1 = None
	        view_1579: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_16_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_908 = torch.ops.aten._assert_tensor_metadata.default(view_1577, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_908 = None
	        convert_element_type_604: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1577, torch.float32);  view_1577 = None
	        _assert_tensor_metadata_909 = torch.ops.aten._assert_tensor_metadata.default(view_1579, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_909 = None
	        convert_element_type_605: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1579, torch.float32);  view_1579 = None
	        sub_4616: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_604, convert_element_type_605);  convert_element_type_604 = convert_element_type_605 = None
	        mul_9774: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4616, view_1578);  sub_4616 = view_1578 = None
	        view_1580: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9774, [5120, 1280]);  mul_9774 = None
	        _assert_tensor_metadata_910 = torch.ops.aten._assert_tensor_metadata.default(view_1580, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_910 = None
	        mul_9779: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1581: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9769, [mul_9779, 1280]);  mul_9769 = mul_9779 = None
	        permute_169: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1580, [1, 0]);  view_1580 = None
	        addmm_83: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_16_fc1_bias, view_1581, permute_169);  model_audio_tower_layers_16_fc1_bias = view_1581 = permute_169 = None
	        view_1582: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_83, [sym_size_int, 1500, 5120]);  addmm_83 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_9786: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1582, 0.5)
	        mul_9787: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1582, 0.7071067811865476);  view_1582 = None
	        erf_18: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_9787);  mul_9787 = None
	        add_15486: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_18, 1);  erf_18 = None
	        mul_9788: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9786, add_15486);  mul_9786 = add_15486 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_9788, [2])
	        amax_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_9788, [2])
	        full_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_101, full_202);  amin_101 = full_202 = None
	        full_203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_101, full_203);  amax_101 = full_203 = None
	        sub_4629: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_101, minimum_101);  maximum_101 = None
	        div_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4629, 255.0);  sub_4629 = None
	        clamp_min_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_202, 1.1920928955078125e-07);  div_202 = None
	        div_203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_101, clamp_min_303);  minimum_101 = None
	        round_203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_203);  div_203 = None
	        sub_4635: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_203);  round_203 = None
	        clamp_min_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4635, -128);  sub_4635 = None
	        clamp_max_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_304, 127);  clamp_min_304 = None
	        _assert_tensor_metadata_911 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_303, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_911 = None
	        _assert_tensor_metadata_912 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_202, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_912 = None
	        convert_element_type_606: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_202, torch.int8);  clamp_max_202 = None
	        view_1585: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_303, [sym_size_int, 1500, 1])
	        view_1586: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_606, [sym_size_int, 1500, 1])
	        reciprocal_101: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1585);  view_1585 = None
	        mul_9834: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_101, 1.0);  reciprocal_101 = None
	        mul_9837: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9788, mul_9834);  mul_9788 = mul_9834 = None
	        round_204: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_9837);  mul_9837 = None
	        add_15569: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_204, view_1586);  round_204 = view_1586 = None
	        clamp_min_305: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15569, -128);  add_15569 = None
	        clamp_max_203: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_305, 127);  clamp_min_305 = None
	        _assert_tensor_metadata_913 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_203, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_913 = None
	        convert_element_type_607: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_203, torch.int8);  clamp_max_203 = None
	        view_1589: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_303, [sym_size_int, 1500, 1]);  clamp_min_303 = None
	        view_1590: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_606, [sym_size_int, 1500, 1]);  convert_element_type_606 = None
	        _assert_tensor_metadata_914 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_607, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_914 = None
	        convert_element_type_608: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_607, torch.float32);  convert_element_type_607 = None
	        _assert_tensor_metadata_915 = torch.ops.aten._assert_tensor_metadata.default(view_1590, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_915 = None
	        convert_element_type_609: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1590, torch.float32);  view_1590 = None
	        sub_4655: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_608, convert_element_type_609);  convert_element_type_608 = convert_element_type_609 = None
	        mul_9859: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4655, view_1589);  sub_4655 = view_1589 = None
	        _assert_tensor_metadata_916 = torch.ops.aten._assert_tensor_metadata.default(mul_9859, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_916 = None
	        view_1592: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_16_fc2_parametrizations_weight_original0 = None
	        view_1593: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_16_fc2_parametrizations_weight_original1 = None
	        view_1594: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_16_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_16_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_917 = torch.ops.aten._assert_tensor_metadata.default(view_1592, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_917 = None
	        convert_element_type_610: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1592, torch.float32);  view_1592 = None
	        _assert_tensor_metadata_918 = torch.ops.aten._assert_tensor_metadata.default(view_1594, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_918 = None
	        convert_element_type_611: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1594, torch.float32);  view_1594 = None
	        sub_4659: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_610, convert_element_type_611);  convert_element_type_610 = convert_element_type_611 = None
	        mul_9864: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4659, view_1593);  sub_4659 = view_1593 = None
	        view_1595: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_9864, [1280, 5120]);  mul_9864 = None
	        _assert_tensor_metadata_919 = torch.ops.aten._assert_tensor_metadata.default(view_1595, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_919 = None
	        mul_9869: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1596: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_9859, [mul_9869, 5120]);  mul_9859 = mul_9869 = None
	        permute_170: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1595, [1, 0]);  view_1595 = None
	        addmm_84: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_16_fc2_bias, view_1596, permute_170);  model_audio_tower_layers_16_fc2_bias = view_1596 = permute_170 = None
	        view_1597: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_84, [sym_size_int, 1500, 1280]);  addmm_84 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_15632: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_15334, view_1597);  add_15334 = view_1597 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_15632, memory_format = torch.contiguous_format)
	        var_mean_34 = torch.ops.aten.var_mean.correction(clone_137, [2], correction = 0, keepdim = True)
	        getitem_136: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_34[0]
	        getitem_137: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_34[1];  var_mean_34 = None
	        add_15637: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_136, 1e-05);  getitem_136 = None
	        rsqrt_34: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_15637);  add_15637 = None
	        sub_4665: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_137, getitem_137);  clone_137 = getitem_137 = None
	        mul_9880: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4665, rsqrt_34);  sub_4665 = rsqrt_34 = None
	        mul_9881: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9880, model_audio_tower_layers_17_self_attn_layer_norm_weight);  mul_9880 = model_audio_tower_layers_17_self_attn_layer_norm_weight = None
	        add_15638: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9881, model_audio_tower_layers_17_self_attn_layer_norm_bias);  mul_9881 = model_audio_tower_layers_17_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_15638, [2])
	        amax_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_15638, [2])
	        full_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_102, full_204);  amin_102 = full_204 = None
	        full_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_102, full_205);  amax_102 = full_205 = None
	        sub_4676: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_102, minimum_102);  maximum_102 = None
	        div_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4676, 255.0);  sub_4676 = None
	        clamp_min_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_204, 1.1920928955078125e-07);  div_204 = None
	        div_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_102, clamp_min_306);  minimum_102 = None
	        round_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_205);  div_205 = None
	        sub_4682: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_205);  round_205 = None
	        clamp_min_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4682, -128);  sub_4682 = None
	        clamp_max_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_307, 127);  clamp_min_307 = None
	        _assert_tensor_metadata_920 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_306, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_920 = None
	        _assert_tensor_metadata_921 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_204, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_921 = None
	        convert_element_type_612: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_204, torch.int8);  clamp_max_204 = None
	        view_1600: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_306, [sym_size_int, 1500, 1])
	        view_1601: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_612, [sym_size_int, 1500, 1])
	        reciprocal_102: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1600);  view_1600 = None
	        mul_9929: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_102, 1.0);  reciprocal_102 = None
	        mul_9932: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_15638, mul_9929);  mul_9929 = None
	        round_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9932);  mul_9932 = None
	        add_15725: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_206, view_1601);  round_206 = view_1601 = None
	        clamp_min_308: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15725, -128);  add_15725 = None
	        clamp_max_205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_308, 127);  clamp_min_308 = None
	        _assert_tensor_metadata_922 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_205, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_922 = None
	        convert_element_type_613: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_205, torch.int8);  clamp_max_205 = None
	        view_1604: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_306, [sym_size_int, 1500, 1]);  clamp_min_306 = None
	        view_1605: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_612, [sym_size_int, 1500, 1]);  convert_element_type_612 = None
	        _assert_tensor_metadata_923 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_613, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_923 = None
	        convert_element_type_614: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_613, torch.float32);  convert_element_type_613 = None
	        _assert_tensor_metadata_924 = torch.ops.aten._assert_tensor_metadata.default(view_1605, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_924 = None
	        convert_element_type_615: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1605, torch.float32);  view_1605 = None
	        sub_4702: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_614, convert_element_type_615);  convert_element_type_614 = convert_element_type_615 = None
	        mul_9954: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4702, view_1604);  sub_4702 = view_1604 = None
	        _assert_tensor_metadata_925 = torch.ops.aten._assert_tensor_metadata.default(mul_9954, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_925 = None
	        view_1607: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1608: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1609: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_926 = torch.ops.aten._assert_tensor_metadata.default(view_1607, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_926 = None
	        convert_element_type_616: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1607, torch.float32);  view_1607 = None
	        _assert_tensor_metadata_927 = torch.ops.aten._assert_tensor_metadata.default(view_1609, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_927 = None
	        convert_element_type_617: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1609, torch.float32);  view_1609 = None
	        sub_4706: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_616, convert_element_type_617);  convert_element_type_616 = convert_element_type_617 = None
	        mul_9959: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4706, view_1608);  sub_4706 = view_1608 = None
	        view_1610: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9959, [1280, 1280]);  mul_9959 = None
	        _assert_tensor_metadata_928 = torch.ops.aten._assert_tensor_metadata.default(view_1610, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_928 = None
	        mul_9964: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1611: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_9954, [mul_9964, 1280]);  mul_9954 = mul_9964 = None
	        permute_171: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1610, [1, 0]);  view_1610 = None
	        addmm_85: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_17_self_attn_q_proj_bias, view_1611, permute_171);  model_audio_tower_layers_17_self_attn_q_proj_bias = view_1611 = permute_171 = None
	        view_1612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_85, [sym_size_int, 1500, 1280]);  addmm_85 = None
	        mul_9971: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1612, 0.125);  view_1612 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1613: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_9971, [sym_size_int, 1500, 20, 64]);  mul_9971 = None
	        permute_172: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1613, [0, 2, 1, 3]);  view_1613 = None
	        clone_138: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_172, memory_format = torch.contiguous_format);  permute_172 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_15638, [2])
	        amax_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_15638, [2])
	        full_206: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_103, full_206);  amin_103 = full_206 = None
	        full_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_103, full_207);  amax_103 = full_207 = None
	        sub_4721: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_103, minimum_103);  maximum_103 = None
	        div_206: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4721, 255.0);  sub_4721 = None
	        clamp_min_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_206, 1.1920928955078125e-07);  div_206 = None
	        div_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_103, clamp_min_309);  minimum_103 = None
	        round_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_207);  div_207 = None
	        sub_4727: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_207);  round_207 = None
	        clamp_min_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4727, -128);  sub_4727 = None
	        clamp_max_206: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_310, 127);  clamp_min_310 = None
	        _assert_tensor_metadata_929 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_309, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_929 = None
	        _assert_tensor_metadata_930 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_206, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_930 = None
	        convert_element_type_618: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_206, torch.int8);  clamp_max_206 = None
	        view_1616: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_309, [sym_size_int, 1500, 1])
	        view_1617: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_618, [sym_size_int, 1500, 1])
	        reciprocal_103: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1616);  view_1616 = None
	        mul_10025: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_103, 1.0);  reciprocal_103 = None
	        mul_10028: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_15638, mul_10025);  mul_10025 = None
	        round_208: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10028);  mul_10028 = None
	        add_15877: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_208, view_1617);  round_208 = view_1617 = None
	        clamp_min_311: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15877, -128);  add_15877 = None
	        clamp_max_207: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_311, 127);  clamp_min_311 = None
	        _assert_tensor_metadata_931 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_207, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_931 = None
	        convert_element_type_619: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_207, torch.int8);  clamp_max_207 = None
	        view_1620: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_309, [sym_size_int, 1500, 1]);  clamp_min_309 = None
	        view_1621: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_618, [sym_size_int, 1500, 1]);  convert_element_type_618 = None
	        _assert_tensor_metadata_932 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_619, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_932 = None
	        convert_element_type_620: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_619, torch.float32);  convert_element_type_619 = None
	        _assert_tensor_metadata_933 = torch.ops.aten._assert_tensor_metadata.default(view_1621, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_933 = None
	        convert_element_type_621: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1621, torch.float32);  view_1621 = None
	        sub_4747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_620, convert_element_type_621);  convert_element_type_620 = convert_element_type_621 = None
	        mul_10050: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4747, view_1620);  sub_4747 = view_1620 = None
	        _assert_tensor_metadata_934 = torch.ops.aten._assert_tensor_metadata.default(mul_10050, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_934 = None
	        view_1623: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1624: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1625: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_935 = torch.ops.aten._assert_tensor_metadata.default(view_1623, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_935 = None
	        convert_element_type_622: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1623, torch.float32);  view_1623 = None
	        _assert_tensor_metadata_936 = torch.ops.aten._assert_tensor_metadata.default(view_1625, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_936 = None
	        convert_element_type_623: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1625, torch.float32);  view_1625 = None
	        sub_4751: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_622, convert_element_type_623);  convert_element_type_622 = convert_element_type_623 = None
	        mul_10055: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4751, view_1624);  sub_4751 = view_1624 = None
	        view_1626: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10055, [1280, 1280]);  mul_10055 = None
	        _assert_tensor_metadata_937 = torch.ops.aten._assert_tensor_metadata.default(view_1626, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_937 = None
	        permute_173: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1626, [1, 0]);  view_1626 = None
	        mul_10058: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1627: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10050, [mul_10058, 1280]);  mul_10050 = mul_10058 = None
	        mm_17: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1627, permute_173);  view_1627 = permute_173 = None
	        view_1628: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_17, [sym_size_int, 1500, 1280]);  mm_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1629: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1628, [sym_size_int, -1, 20, 64]);  view_1628 = None
	        permute_174: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1629, [0, 2, 1, 3]);  view_1629 = None
	        clone_139: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_174, memory_format = torch.contiguous_format);  permute_174 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_15638, [2])
	        amax_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_15638, [2])
	        full_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_104, full_208);  amin_104 = full_208 = None
	        full_209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_104, full_209);  amax_104 = full_209 = None
	        sub_4765: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_104, minimum_104);  maximum_104 = None
	        div_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4765, 255.0);  sub_4765 = None
	        clamp_min_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_208, 1.1920928955078125e-07);  div_208 = None
	        div_209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_104, clamp_min_312);  minimum_104 = None
	        round_209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_209);  div_209 = None
	        sub_4771: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_209);  round_209 = None
	        clamp_min_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4771, -128);  sub_4771 = None
	        clamp_max_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_313, 127);  clamp_min_313 = None
	        _assert_tensor_metadata_938 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_312, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_938 = None
	        _assert_tensor_metadata_939 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_208, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_939 = None
	        convert_element_type_624: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_208, torch.int8);  clamp_max_208 = None
	        view_1632: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_312, [sym_size_int, 1500, 1])
	        view_1633: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_624, [sym_size_int, 1500, 1])
	        reciprocal_104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1632);  view_1632 = None
	        mul_10124: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_104, 1.0);  reciprocal_104 = None
	        mul_10127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_15638, mul_10124);  add_15638 = mul_10124 = None
	        round_210: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10127);  mul_10127 = None
	        add_16025: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_210, view_1633);  round_210 = view_1633 = None
	        clamp_min_314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16025, -128);  add_16025 = None
	        clamp_max_209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_314, 127);  clamp_min_314 = None
	        _assert_tensor_metadata_940 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_209, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_940 = None
	        convert_element_type_625: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_209, torch.int8);  clamp_max_209 = None
	        view_1636: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_312, [sym_size_int, 1500, 1]);  clamp_min_312 = None
	        view_1637: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_624, [sym_size_int, 1500, 1]);  convert_element_type_624 = None
	        _assert_tensor_metadata_941 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_625, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_941 = None
	        convert_element_type_626: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_625, torch.float32);  convert_element_type_625 = None
	        _assert_tensor_metadata_942 = torch.ops.aten._assert_tensor_metadata.default(view_1637, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_942 = None
	        convert_element_type_627: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1637, torch.float32);  view_1637 = None
	        sub_4791: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_626, convert_element_type_627);  convert_element_type_626 = convert_element_type_627 = None
	        mul_10149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4791, view_1636);  sub_4791 = view_1636 = None
	        _assert_tensor_metadata_943 = torch.ops.aten._assert_tensor_metadata.default(mul_10149, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_943 = None
	        view_1639: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1640: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1641: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_944 = torch.ops.aten._assert_tensor_metadata.default(view_1639, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_944 = None
	        convert_element_type_628: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1639, torch.float32);  view_1639 = None
	        _assert_tensor_metadata_945 = torch.ops.aten._assert_tensor_metadata.default(view_1641, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_945 = None
	        convert_element_type_629: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1641, torch.float32);  view_1641 = None
	        sub_4795: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_628, convert_element_type_629);  convert_element_type_628 = convert_element_type_629 = None
	        mul_10154: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4795, view_1640);  sub_4795 = view_1640 = None
	        view_1642: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10154, [1280, 1280]);  mul_10154 = None
	        _assert_tensor_metadata_946 = torch.ops.aten._assert_tensor_metadata.default(view_1642, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_946 = None
	        mul_10159: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1643: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10149, [mul_10159, 1280]);  mul_10149 = mul_10159 = None
	        permute_175: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1642, [1, 0]);  view_1642 = None
	        addmm_86: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_17_self_attn_v_proj_bias, view_1643, permute_175);  model_audio_tower_layers_17_self_attn_v_proj_bias = view_1643 = permute_175 = None
	        view_1644: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_86, [sym_size_int, 1500, 1280]);  addmm_86 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1645: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1644, [sym_size_int, -1, 20, 64]);  view_1644 = None
	        permute_176: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1645, [0, 2, 1, 3]);  view_1645 = None
	        clone_140: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_176, memory_format = torch.contiguous_format);  permute_176 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_17 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_138, clone_139, clone_140, None, False, scale = 1.0);  clone_138 = clone_139 = clone_140 = None
	        getitem_138: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_17[0];  _scaled_dot_product_efficient_attention_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_177: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_138, [0, 2, 1, 3]);  getitem_138 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1646: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_177, [sym_size_int, 1500, -1]);  permute_177 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1646, [2])
	        amax_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1646, [2])
	        full_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_105, full_210);  amin_105 = full_210 = None
	        full_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_105, full_211);  amax_105 = full_211 = None
	        sub_4813: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_105, minimum_105);  maximum_105 = None
	        div_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4813, 255.0);  sub_4813 = None
	        clamp_min_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_210, 1.1920928955078125e-07);  div_210 = None
	        div_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_105, clamp_min_315);  minimum_105 = None
	        round_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_211);  div_211 = None
	        sub_4819: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_211);  round_211 = None
	        clamp_min_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4819, -128);  sub_4819 = None
	        clamp_max_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_316, 127);  clamp_min_316 = None
	        _assert_tensor_metadata_947 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_315, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_947 = None
	        _assert_tensor_metadata_948 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_210, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_948 = None
	        convert_element_type_630: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_210, torch.int8);  clamp_max_210 = None
	        view_1649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_315, [sym_size_int, 1500, 1])
	        view_1650: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_630, [sym_size_int, 1500, 1])
	        reciprocal_105: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1649);  view_1649 = None
	        mul_10229: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_105, 1.0);  reciprocal_105 = None
	        mul_10232: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1646, mul_10229);  view_1646 = mul_10229 = None
	        round_212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10232);  mul_10232 = None
	        add_16189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_212, view_1650);  round_212 = view_1650 = None
	        clamp_min_317: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16189, -128);  add_16189 = None
	        clamp_max_211: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_317, 127);  clamp_min_317 = None
	        _assert_tensor_metadata_949 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_211, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_949 = None
	        convert_element_type_631: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_211, torch.int8);  clamp_max_211 = None
	        view_1653: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_315, [sym_size_int, 1500, 1]);  clamp_min_315 = None
	        view_1654: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_630, [sym_size_int, 1500, 1]);  convert_element_type_630 = None
	        _assert_tensor_metadata_950 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_631, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_950 = None
	        convert_element_type_632: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_631, torch.float32);  convert_element_type_631 = None
	        _assert_tensor_metadata_951 = torch.ops.aten._assert_tensor_metadata.default(view_1654, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_951 = None
	        convert_element_type_633: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1654, torch.float32);  view_1654 = None
	        sub_4839: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_632, convert_element_type_633);  convert_element_type_632 = convert_element_type_633 = None
	        mul_10254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4839, view_1653);  sub_4839 = view_1653 = None
	        _assert_tensor_metadata_952 = torch.ops.aten._assert_tensor_metadata.default(mul_10254, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_952 = None
	        view_1656: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1657: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1658: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_953 = torch.ops.aten._assert_tensor_metadata.default(view_1656, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_953 = None
	        convert_element_type_634: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1656, torch.float32);  view_1656 = None
	        _assert_tensor_metadata_954 = torch.ops.aten._assert_tensor_metadata.default(view_1658, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_954 = None
	        convert_element_type_635: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1658, torch.float32);  view_1658 = None
	        sub_4843: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_634, convert_element_type_635);  convert_element_type_634 = convert_element_type_635 = None
	        mul_10259: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4843, view_1657);  sub_4843 = view_1657 = None
	        view_1659: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10259, [1280, 1280]);  mul_10259 = None
	        _assert_tensor_metadata_955 = torch.ops.aten._assert_tensor_metadata.default(view_1659, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_955 = None
	        mul_10264: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1660: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10254, [mul_10264, 1280]);  mul_10254 = mul_10264 = None
	        permute_178: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1659, [1, 0]);  view_1659 = None
	        addmm_87: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_17_self_attn_out_proj_bias, view_1660, permute_178);  model_audio_tower_layers_17_self_attn_out_proj_bias = view_1660 = permute_178 = None
	        view_1661: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_87, [sym_size_int, 1500, 1280]);  addmm_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_16252: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_15632, view_1661);  add_15632 = view_1661 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_16252, memory_format = torch.contiguous_format)
	        var_mean_35 = torch.ops.aten.var_mean.correction(clone_142, [2], correction = 0, keepdim = True)
	        getitem_142: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_35[0]
	        getitem_143: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_35[1];  var_mean_35 = None
	        add_16257: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_142, 1e-05);  getitem_142 = None
	        rsqrt_35: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_16257);  add_16257 = None
	        sub_4849: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_142, getitem_143);  clone_142 = getitem_143 = None
	        mul_10275: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4849, rsqrt_35);  sub_4849 = rsqrt_35 = None
	        mul_10276: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10275, model_audio_tower_layers_17_final_layer_norm_weight);  mul_10275 = model_audio_tower_layers_17_final_layer_norm_weight = None
	        add_16258: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_10276, model_audio_tower_layers_17_final_layer_norm_bias);  mul_10276 = model_audio_tower_layers_17_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_16258, [2])
	        amax_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_16258, [2])
	        full_212: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_106, full_212);  amin_106 = full_212 = None
	        full_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_106, full_213);  amax_106 = full_213 = None
	        sub_4860: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_106, minimum_106);  maximum_106 = None
	        div_212: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4860, 255.0);  sub_4860 = None
	        clamp_min_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_212, 1.1920928955078125e-07);  div_212 = None
	        div_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_106, clamp_min_318);  minimum_106 = None
	        round_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_213);  div_213 = None
	        sub_4866: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_213);  round_213 = None
	        clamp_min_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4866, -128);  sub_4866 = None
	        clamp_max_212: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_319, 127);  clamp_min_319 = None
	        _assert_tensor_metadata_956 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_318, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_956 = None
	        _assert_tensor_metadata_957 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_212, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_957 = None
	        convert_element_type_636: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_212, torch.int8);  clamp_max_212 = None
	        view_1664: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_318, [sym_size_int, 1500, 1])
	        view_1665: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_636, [sym_size_int, 1500, 1])
	        reciprocal_106: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1664);  view_1664 = None
	        mul_10324: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_106, 1.0);  reciprocal_106 = None
	        mul_10327: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_16258, mul_10324);  add_16258 = mul_10324 = None
	        round_214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10327);  mul_10327 = None
	        add_16345: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_214, view_1665);  round_214 = view_1665 = None
	        clamp_min_320: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16345, -128);  add_16345 = None
	        clamp_max_213: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_320, 127);  clamp_min_320 = None
	        _assert_tensor_metadata_958 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_213, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_958 = None
	        convert_element_type_637: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_213, torch.int8);  clamp_max_213 = None
	        view_1668: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_318, [sym_size_int, 1500, 1]);  clamp_min_318 = None
	        view_1669: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_636, [sym_size_int, 1500, 1]);  convert_element_type_636 = None
	        _assert_tensor_metadata_959 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_637, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_959 = None
	        convert_element_type_638: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_637, torch.float32);  convert_element_type_637 = None
	        _assert_tensor_metadata_960 = torch.ops.aten._assert_tensor_metadata.default(view_1669, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_960 = None
	        convert_element_type_639: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1669, torch.float32);  view_1669 = None
	        sub_4886: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_638, convert_element_type_639);  convert_element_type_638 = convert_element_type_639 = None
	        mul_10349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4886, view_1668);  sub_4886 = view_1668 = None
	        _assert_tensor_metadata_961 = torch.ops.aten._assert_tensor_metadata.default(mul_10349, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_961 = None
	        view_1671: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_17_fc1_parametrizations_weight_original0 = None
	        view_1672: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_17_fc1_parametrizations_weight_original1 = None
	        view_1673: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_17_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_962 = torch.ops.aten._assert_tensor_metadata.default(view_1671, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_962 = None
	        convert_element_type_640: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1671, torch.float32);  view_1671 = None
	        _assert_tensor_metadata_963 = torch.ops.aten._assert_tensor_metadata.default(view_1673, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_963 = None
	        convert_element_type_641: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1673, torch.float32);  view_1673 = None
	        sub_4890: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_640, convert_element_type_641);  convert_element_type_640 = convert_element_type_641 = None
	        mul_10354: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4890, view_1672);  sub_4890 = view_1672 = None
	        view_1674: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10354, [5120, 1280]);  mul_10354 = None
	        _assert_tensor_metadata_964 = torch.ops.aten._assert_tensor_metadata.default(view_1674, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_964 = None
	        mul_10359: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1675: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10349, [mul_10359, 1280]);  mul_10349 = mul_10359 = None
	        permute_179: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1674, [1, 0]);  view_1674 = None
	        addmm_88: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_17_fc1_bias, view_1675, permute_179);  model_audio_tower_layers_17_fc1_bias = view_1675 = permute_179 = None
	        view_1676: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_88, [sym_size_int, 1500, 5120]);  addmm_88 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_10366: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1676, 0.5)
	        mul_10367: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1676, 0.7071067811865476);  view_1676 = None
	        erf_19: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_10367);  mul_10367 = None
	        add_16404: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_19, 1);  erf_19 = None
	        mul_10368: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10366, add_16404);  mul_10366 = add_16404 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_10368, [2])
	        amax_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_10368, [2])
	        full_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_107, full_214);  amin_107 = full_214 = None
	        full_215: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_107, full_215);  amax_107 = full_215 = None
	        sub_4903: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_107, minimum_107);  maximum_107 = None
	        div_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4903, 255.0);  sub_4903 = None
	        clamp_min_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_214, 1.1920928955078125e-07);  div_214 = None
	        div_215: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_107, clamp_min_321);  minimum_107 = None
	        round_215: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_215);  div_215 = None
	        sub_4909: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_215);  round_215 = None
	        clamp_min_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4909, -128);  sub_4909 = None
	        clamp_max_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_322, 127);  clamp_min_322 = None
	        _assert_tensor_metadata_965 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_321, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_965 = None
	        _assert_tensor_metadata_966 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_214, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_966 = None
	        convert_element_type_642: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_214, torch.int8);  clamp_max_214 = None
	        view_1679: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_321, [sym_size_int, 1500, 1])
	        view_1680: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_642, [sym_size_int, 1500, 1])
	        reciprocal_107: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1679);  view_1679 = None
	        mul_10414: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_107, 1.0);  reciprocal_107 = None
	        mul_10417: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10368, mul_10414);  mul_10368 = mul_10414 = None
	        round_216: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_10417);  mul_10417 = None
	        add_16487: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_216, view_1680);  round_216 = view_1680 = None
	        clamp_min_323: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16487, -128);  add_16487 = None
	        clamp_max_215: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_323, 127);  clamp_min_323 = None
	        _assert_tensor_metadata_967 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_215, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_967 = None
	        convert_element_type_643: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_215, torch.int8);  clamp_max_215 = None
	        view_1683: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_321, [sym_size_int, 1500, 1]);  clamp_min_321 = None
	        view_1684: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_642, [sym_size_int, 1500, 1]);  convert_element_type_642 = None
	        _assert_tensor_metadata_968 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_643, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_968 = None
	        convert_element_type_644: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_643, torch.float32);  convert_element_type_643 = None
	        _assert_tensor_metadata_969 = torch.ops.aten._assert_tensor_metadata.default(view_1684, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_969 = None
	        convert_element_type_645: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1684, torch.float32);  view_1684 = None
	        sub_4929: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_644, convert_element_type_645);  convert_element_type_644 = convert_element_type_645 = None
	        mul_10439: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4929, view_1683);  sub_4929 = view_1683 = None
	        _assert_tensor_metadata_970 = torch.ops.aten._assert_tensor_metadata.default(mul_10439, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_970 = None
	        view_1686: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_17_fc2_parametrizations_weight_original0 = None
	        view_1687: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_17_fc2_parametrizations_weight_original1 = None
	        view_1688: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_17_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_17_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_971 = torch.ops.aten._assert_tensor_metadata.default(view_1686, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_971 = None
	        convert_element_type_646: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1686, torch.float32);  view_1686 = None
	        _assert_tensor_metadata_972 = torch.ops.aten._assert_tensor_metadata.default(view_1688, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_972 = None
	        convert_element_type_647: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1688, torch.float32);  view_1688 = None
	        sub_4933: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_646, convert_element_type_647);  convert_element_type_646 = convert_element_type_647 = None
	        mul_10444: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4933, view_1687);  sub_4933 = view_1687 = None
	        view_1689: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_10444, [1280, 5120]);  mul_10444 = None
	        _assert_tensor_metadata_973 = torch.ops.aten._assert_tensor_metadata.default(view_1689, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_973 = None
	        mul_10449: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1690: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_10439, [mul_10449, 5120]);  mul_10439 = mul_10449 = None
	        permute_180: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1689, [1, 0]);  view_1689 = None
	        addmm_89: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_17_fc2_bias, view_1690, permute_180);  model_audio_tower_layers_17_fc2_bias = view_1690 = permute_180 = None
	        view_1691: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_89, [sym_size_int, 1500, 1280]);  addmm_89 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_16550: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_16252, view_1691);  add_16252 = view_1691 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_145: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_16550, memory_format = torch.contiguous_format)
	        var_mean_36 = torch.ops.aten.var_mean.correction(clone_145, [2], correction = 0, keepdim = True)
	        getitem_144: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_36[0]
	        getitem_145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_36[1];  var_mean_36 = None
	        add_16555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_144, 1e-05);  getitem_144 = None
	        rsqrt_36: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_16555);  add_16555 = None
	        sub_4939: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_145, getitem_145);  clone_145 = getitem_145 = None
	        mul_10460: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4939, rsqrt_36);  sub_4939 = rsqrt_36 = None
	        mul_10461: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10460, model_audio_tower_layers_18_self_attn_layer_norm_weight);  mul_10460 = model_audio_tower_layers_18_self_attn_layer_norm_weight = None
	        add_16556: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_10461, model_audio_tower_layers_18_self_attn_layer_norm_bias);  mul_10461 = model_audio_tower_layers_18_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_16556, [2])
	        amax_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_16556, [2])
	        full_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_108, full_216);  amin_108 = full_216 = None
	        full_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_108, full_217);  amax_108 = full_217 = None
	        sub_4950: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_108, minimum_108);  maximum_108 = None
	        div_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4950, 255.0);  sub_4950 = None
	        clamp_min_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_216, 1.1920928955078125e-07);  div_216 = None
	        div_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_108, clamp_min_324);  minimum_108 = None
	        round_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_217);  div_217 = None
	        sub_4956: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_217);  round_217 = None
	        clamp_min_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4956, -128);  sub_4956 = None
	        clamp_max_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_325, 127);  clamp_min_325 = None
	        _assert_tensor_metadata_974 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_324, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_974 = None
	        _assert_tensor_metadata_975 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_216, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_975 = None
	        convert_element_type_648: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_216, torch.int8);  clamp_max_216 = None
	        view_1694: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_324, [sym_size_int, 1500, 1])
	        view_1695: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_648, [sym_size_int, 1500, 1])
	        reciprocal_108: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1694);  view_1694 = None
	        mul_10509: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_108, 1.0);  reciprocal_108 = None
	        mul_10512: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_16556, mul_10509);  mul_10509 = None
	        round_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10512);  mul_10512 = None
	        add_16643: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_218, view_1695);  round_218 = view_1695 = None
	        clamp_min_326: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16643, -128);  add_16643 = None
	        clamp_max_217: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_326, 127);  clamp_min_326 = None
	        _assert_tensor_metadata_976 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_217, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_976 = None
	        convert_element_type_649: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_217, torch.int8);  clamp_max_217 = None
	        view_1698: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_324, [sym_size_int, 1500, 1]);  clamp_min_324 = None
	        view_1699: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_648, [sym_size_int, 1500, 1]);  convert_element_type_648 = None
	        _assert_tensor_metadata_977 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_649, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_977 = None
	        convert_element_type_650: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_649, torch.float32);  convert_element_type_649 = None
	        _assert_tensor_metadata_978 = torch.ops.aten._assert_tensor_metadata.default(view_1699, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_978 = None
	        convert_element_type_651: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1699, torch.float32);  view_1699 = None
	        sub_4976: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_650, convert_element_type_651);  convert_element_type_650 = convert_element_type_651 = None
	        mul_10534: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4976, view_1698);  sub_4976 = view_1698 = None
	        _assert_tensor_metadata_979 = torch.ops.aten._assert_tensor_metadata.default(mul_10534, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_979 = None
	        view_1701: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1702: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1703: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_980 = torch.ops.aten._assert_tensor_metadata.default(view_1701, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_980 = None
	        convert_element_type_652: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1701, torch.float32);  view_1701 = None
	        _assert_tensor_metadata_981 = torch.ops.aten._assert_tensor_metadata.default(view_1703, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_981 = None
	        convert_element_type_653: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1703, torch.float32);  view_1703 = None
	        sub_4980: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_652, convert_element_type_653);  convert_element_type_652 = convert_element_type_653 = None
	        mul_10539: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4980, view_1702);  sub_4980 = view_1702 = None
	        view_1704: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10539, [1280, 1280]);  mul_10539 = None
	        _assert_tensor_metadata_982 = torch.ops.aten._assert_tensor_metadata.default(view_1704, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_982 = None
	        mul_10544: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1705: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10534, [mul_10544, 1280]);  mul_10534 = mul_10544 = None
	        permute_181: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1704, [1, 0]);  view_1704 = None
	        addmm_90: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_18_self_attn_q_proj_bias, view_1705, permute_181);  model_audio_tower_layers_18_self_attn_q_proj_bias = view_1705 = permute_181 = None
	        view_1706: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_90, [sym_size_int, 1500, 1280]);  addmm_90 = None
	        mul_10551: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1706, 0.125);  view_1706 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1707: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_10551, [sym_size_int, 1500, 20, 64]);  mul_10551 = None
	        permute_182: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1707, [0, 2, 1, 3]);  view_1707 = None
	        clone_146: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_182, memory_format = torch.contiguous_format);  permute_182 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_16556, [2])
	        amax_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_16556, [2])
	        full_218: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_109, full_218);  amin_109 = full_218 = None
	        full_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_109, full_219);  amax_109 = full_219 = None
	        sub_4995: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_109, minimum_109);  maximum_109 = None
	        div_218: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4995, 255.0);  sub_4995 = None
	        clamp_min_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_218, 1.1920928955078125e-07);  div_218 = None
	        div_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_109, clamp_min_327);  minimum_109 = None
	        round_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_219);  div_219 = None
	        sub_5001: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_219);  round_219 = None
	        clamp_min_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5001, -128);  sub_5001 = None
	        clamp_max_218: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_328, 127);  clamp_min_328 = None
	        _assert_tensor_metadata_983 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_327, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_983 = None
	        _assert_tensor_metadata_984 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_218, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_984 = None
	        convert_element_type_654: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_218, torch.int8);  clamp_max_218 = None
	        view_1710: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_327, [sym_size_int, 1500, 1])
	        view_1711: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_654, [sym_size_int, 1500, 1])
	        reciprocal_109: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1710);  view_1710 = None
	        mul_10605: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_109, 1.0);  reciprocal_109 = None
	        mul_10608: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_16556, mul_10605);  mul_10605 = None
	        round_220: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10608);  mul_10608 = None
	        add_16795: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_220, view_1711);  round_220 = view_1711 = None
	        clamp_min_329: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16795, -128);  add_16795 = None
	        clamp_max_219: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_329, 127);  clamp_min_329 = None
	        _assert_tensor_metadata_985 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_219, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_985 = None
	        convert_element_type_655: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_219, torch.int8);  clamp_max_219 = None
	        view_1714: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_327, [sym_size_int, 1500, 1]);  clamp_min_327 = None
	        view_1715: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_654, [sym_size_int, 1500, 1]);  convert_element_type_654 = None
	        _assert_tensor_metadata_986 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_655, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_986 = None
	        convert_element_type_656: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_655, torch.float32);  convert_element_type_655 = None
	        _assert_tensor_metadata_987 = torch.ops.aten._assert_tensor_metadata.default(view_1715, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_987 = None
	        convert_element_type_657: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1715, torch.float32);  view_1715 = None
	        sub_5021: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_656, convert_element_type_657);  convert_element_type_656 = convert_element_type_657 = None
	        mul_10630: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5021, view_1714);  sub_5021 = view_1714 = None
	        _assert_tensor_metadata_988 = torch.ops.aten._assert_tensor_metadata.default(mul_10630, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_988 = None
	        view_1717: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1718: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1719: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_989 = torch.ops.aten._assert_tensor_metadata.default(view_1717, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_989 = None
	        convert_element_type_658: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1717, torch.float32);  view_1717 = None
	        _assert_tensor_metadata_990 = torch.ops.aten._assert_tensor_metadata.default(view_1719, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_990 = None
	        convert_element_type_659: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1719, torch.float32);  view_1719 = None
	        sub_5025: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_658, convert_element_type_659);  convert_element_type_658 = convert_element_type_659 = None
	        mul_10635: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5025, view_1718);  sub_5025 = view_1718 = None
	        view_1720: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10635, [1280, 1280]);  mul_10635 = None
	        _assert_tensor_metadata_991 = torch.ops.aten._assert_tensor_metadata.default(view_1720, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_991 = None
	        permute_183: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1720, [1, 0]);  view_1720 = None
	        mul_10638: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1721: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10630, [mul_10638, 1280]);  mul_10630 = mul_10638 = None
	        mm_18: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1721, permute_183);  view_1721 = permute_183 = None
	        view_1722: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_18, [sym_size_int, 1500, 1280]);  mm_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1723: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1722, [sym_size_int, -1, 20, 64]);  view_1722 = None
	        permute_184: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1723, [0, 2, 1, 3]);  view_1723 = None
	        clone_147: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_184, memory_format = torch.contiguous_format);  permute_184 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_16556, [2])
	        amax_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_16556, [2])
	        full_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_110, full_220);  amin_110 = full_220 = None
	        full_221: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_110, full_221);  amax_110 = full_221 = None
	        sub_5039: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_110, minimum_110);  maximum_110 = None
	        div_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5039, 255.0);  sub_5039 = None
	        clamp_min_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_220, 1.1920928955078125e-07);  div_220 = None
	        div_221: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_110, clamp_min_330);  minimum_110 = None
	        round_221: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_221);  div_221 = None
	        sub_5045: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_221);  round_221 = None
	        clamp_min_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5045, -128);  sub_5045 = None
	        clamp_max_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_331, 127);  clamp_min_331 = None
	        _assert_tensor_metadata_992 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_330, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_992 = None
	        _assert_tensor_metadata_993 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_220, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_993 = None
	        convert_element_type_660: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_220, torch.int8);  clamp_max_220 = None
	        view_1726: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_330, [sym_size_int, 1500, 1])
	        view_1727: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_660, [sym_size_int, 1500, 1])
	        reciprocal_110: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1726);  view_1726 = None
	        mul_10704: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_110, 1.0);  reciprocal_110 = None
	        mul_10707: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_16556, mul_10704);  add_16556 = mul_10704 = None
	        round_222: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10707);  mul_10707 = None
	        add_16943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_222, view_1727);  round_222 = view_1727 = None
	        clamp_min_332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16943, -128);  add_16943 = None
	        clamp_max_221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_332, 127);  clamp_min_332 = None
	        _assert_tensor_metadata_994 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_221, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_994 = None
	        convert_element_type_661: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_221, torch.int8);  clamp_max_221 = None
	        view_1730: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_330, [sym_size_int, 1500, 1]);  clamp_min_330 = None
	        view_1731: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_660, [sym_size_int, 1500, 1]);  convert_element_type_660 = None
	        _assert_tensor_metadata_995 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_661, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_995 = None
	        convert_element_type_662: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_661, torch.float32);  convert_element_type_661 = None
	        _assert_tensor_metadata_996 = torch.ops.aten._assert_tensor_metadata.default(view_1731, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_996 = None
	        convert_element_type_663: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1731, torch.float32);  view_1731 = None
	        sub_5065: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_662, convert_element_type_663);  convert_element_type_662 = convert_element_type_663 = None
	        mul_10729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5065, view_1730);  sub_5065 = view_1730 = None
	        _assert_tensor_metadata_997 = torch.ops.aten._assert_tensor_metadata.default(mul_10729, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_997 = None
	        view_1733: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1734: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1735: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_998 = torch.ops.aten._assert_tensor_metadata.default(view_1733, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_998 = None
	        convert_element_type_664: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1733, torch.float32);  view_1733 = None
	        _assert_tensor_metadata_999 = torch.ops.aten._assert_tensor_metadata.default(view_1735, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_999 = None
	        convert_element_type_665: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1735, torch.float32);  view_1735 = None
	        sub_5069: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_664, convert_element_type_665);  convert_element_type_664 = convert_element_type_665 = None
	        mul_10734: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5069, view_1734);  sub_5069 = view_1734 = None
	        view_1736: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10734, [1280, 1280]);  mul_10734 = None
	        _assert_tensor_metadata_1000 = torch.ops.aten._assert_tensor_metadata.default(view_1736, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1000 = None
	        mul_10739: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1737: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10729, [mul_10739, 1280]);  mul_10729 = mul_10739 = None
	        permute_185: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1736, [1, 0]);  view_1736 = None
	        addmm_91: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_18_self_attn_v_proj_bias, view_1737, permute_185);  model_audio_tower_layers_18_self_attn_v_proj_bias = view_1737 = permute_185 = None
	        view_1738: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_91, [sym_size_int, 1500, 1280]);  addmm_91 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1739: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1738, [sym_size_int, -1, 20, 64]);  view_1738 = None
	        permute_186: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1739, [0, 2, 1, 3]);  view_1739 = None
	        clone_148: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_186, memory_format = torch.contiguous_format);  permute_186 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_18 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_146, clone_147, clone_148, None, False, scale = 1.0);  clone_146 = clone_147 = clone_148 = None
	        getitem_146: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_18[0];  _scaled_dot_product_efficient_attention_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_187: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_146, [0, 2, 1, 3]);  getitem_146 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1740: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_187, [sym_size_int, 1500, -1]);  permute_187 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1740, [2])
	        amax_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1740, [2])
	        full_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_111, full_222);  amin_111 = full_222 = None
	        full_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_111, full_223);  amax_111 = full_223 = None
	        sub_5087: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_111, minimum_111);  maximum_111 = None
	        div_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5087, 255.0);  sub_5087 = None
	        clamp_min_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_222, 1.1920928955078125e-07);  div_222 = None
	        div_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_111, clamp_min_333);  minimum_111 = None
	        round_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_223);  div_223 = None
	        sub_5093: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_223);  round_223 = None
	        clamp_min_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5093, -128);  sub_5093 = None
	        clamp_max_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_334, 127);  clamp_min_334 = None
	        _assert_tensor_metadata_1001 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_333, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1001 = None
	        _assert_tensor_metadata_1002 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_222, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1002 = None
	        convert_element_type_666: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_222, torch.int8);  clamp_max_222 = None
	        view_1743: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_333, [sym_size_int, 1500, 1])
	        view_1744: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_666, [sym_size_int, 1500, 1])
	        reciprocal_111: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1743);  view_1743 = None
	        mul_10809: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_111, 1.0);  reciprocal_111 = None
	        mul_10812: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1740, mul_10809);  view_1740 = mul_10809 = None
	        round_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10812);  mul_10812 = None
	        add_17107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_224, view_1744);  round_224 = view_1744 = None
	        clamp_min_335: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17107, -128);  add_17107 = None
	        clamp_max_223: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_335, 127);  clamp_min_335 = None
	        _assert_tensor_metadata_1003 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_223, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1003 = None
	        convert_element_type_667: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_223, torch.int8);  clamp_max_223 = None
	        view_1747: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_333, [sym_size_int, 1500, 1]);  clamp_min_333 = None
	        view_1748: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_666, [sym_size_int, 1500, 1]);  convert_element_type_666 = None
	        _assert_tensor_metadata_1004 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_667, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1004 = None
	        convert_element_type_668: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_667, torch.float32);  convert_element_type_667 = None
	        _assert_tensor_metadata_1005 = torch.ops.aten._assert_tensor_metadata.default(view_1748, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1005 = None
	        convert_element_type_669: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1748, torch.float32);  view_1748 = None
	        sub_5113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_668, convert_element_type_669);  convert_element_type_668 = convert_element_type_669 = None
	        mul_10834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5113, view_1747);  sub_5113 = view_1747 = None
	        _assert_tensor_metadata_1006 = torch.ops.aten._assert_tensor_metadata.default(mul_10834, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1006 = None
	        view_1750: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1751: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1752: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1007 = torch.ops.aten._assert_tensor_metadata.default(view_1750, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1007 = None
	        convert_element_type_670: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1750, torch.float32);  view_1750 = None
	        _assert_tensor_metadata_1008 = torch.ops.aten._assert_tensor_metadata.default(view_1752, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1008 = None
	        convert_element_type_671: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1752, torch.float32);  view_1752 = None
	        sub_5117: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_670, convert_element_type_671);  convert_element_type_670 = convert_element_type_671 = None
	        mul_10839: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5117, view_1751);  sub_5117 = view_1751 = None
	        view_1753: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10839, [1280, 1280]);  mul_10839 = None
	        _assert_tensor_metadata_1009 = torch.ops.aten._assert_tensor_metadata.default(view_1753, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1009 = None
	        mul_10844: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1754: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10834, [mul_10844, 1280]);  mul_10834 = mul_10844 = None
	        permute_188: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1753, [1, 0]);  view_1753 = None
	        addmm_92: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_18_self_attn_out_proj_bias, view_1754, permute_188);  model_audio_tower_layers_18_self_attn_out_proj_bias = view_1754 = permute_188 = None
	        view_1755: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_92, [sym_size_int, 1500, 1280]);  addmm_92 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_17170: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_16550, view_1755);  add_16550 = view_1755 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_150: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_17170, memory_format = torch.contiguous_format)
	        var_mean_37 = torch.ops.aten.var_mean.correction(clone_150, [2], correction = 0, keepdim = True)
	        getitem_150: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_37[0]
	        getitem_151: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_37[1];  var_mean_37 = None
	        add_17175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_150, 1e-05);  getitem_150 = None
	        rsqrt_37: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_17175);  add_17175 = None
	        sub_5123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_150, getitem_151);  clone_150 = getitem_151 = None
	        mul_10855: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5123, rsqrt_37);  sub_5123 = rsqrt_37 = None
	        mul_10856: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10855, model_audio_tower_layers_18_final_layer_norm_weight);  mul_10855 = model_audio_tower_layers_18_final_layer_norm_weight = None
	        add_17176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_10856, model_audio_tower_layers_18_final_layer_norm_bias);  mul_10856 = model_audio_tower_layers_18_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_17176, [2])
	        amax_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_17176, [2])
	        full_224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_112, full_224);  amin_112 = full_224 = None
	        full_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_112, full_225);  amax_112 = full_225 = None
	        sub_5134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_112, minimum_112);  maximum_112 = None
	        div_224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5134, 255.0);  sub_5134 = None
	        clamp_min_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_224, 1.1920928955078125e-07);  div_224 = None
	        div_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_112, clamp_min_336);  minimum_112 = None
	        round_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_225);  div_225 = None
	        sub_5140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_225);  round_225 = None
	        clamp_min_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5140, -128);  sub_5140 = None
	        clamp_max_224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_337, 127);  clamp_min_337 = None
	        _assert_tensor_metadata_1010 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_336, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1010 = None
	        _assert_tensor_metadata_1011 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_224, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1011 = None
	        convert_element_type_672: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_224, torch.int8);  clamp_max_224 = None
	        view_1758: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_336, [sym_size_int, 1500, 1])
	        view_1759: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_672, [sym_size_int, 1500, 1])
	        reciprocal_112: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1758);  view_1758 = None
	        mul_10904: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_112, 1.0);  reciprocal_112 = None
	        mul_10907: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_17176, mul_10904);  add_17176 = mul_10904 = None
	        round_226: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10907);  mul_10907 = None
	        add_17263: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_226, view_1759);  round_226 = view_1759 = None
	        clamp_min_338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17263, -128);  add_17263 = None
	        clamp_max_225: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_338, 127);  clamp_min_338 = None
	        _assert_tensor_metadata_1012 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_225, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1012 = None
	        convert_element_type_673: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_225, torch.int8);  clamp_max_225 = None
	        view_1762: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_336, [sym_size_int, 1500, 1]);  clamp_min_336 = None
	        view_1763: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_672, [sym_size_int, 1500, 1]);  convert_element_type_672 = None
	        _assert_tensor_metadata_1013 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_673, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1013 = None
	        convert_element_type_674: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_673, torch.float32);  convert_element_type_673 = None
	        _assert_tensor_metadata_1014 = torch.ops.aten._assert_tensor_metadata.default(view_1763, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1014 = None
	        convert_element_type_675: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1763, torch.float32);  view_1763 = None
	        sub_5160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_674, convert_element_type_675);  convert_element_type_674 = convert_element_type_675 = None
	        mul_10929: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5160, view_1762);  sub_5160 = view_1762 = None
	        _assert_tensor_metadata_1015 = torch.ops.aten._assert_tensor_metadata.default(mul_10929, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1015 = None
	        view_1765: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_18_fc1_parametrizations_weight_original0 = None
	        view_1766: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_18_fc1_parametrizations_weight_original1 = None
	        view_1767: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_18_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1016 = torch.ops.aten._assert_tensor_metadata.default(view_1765, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1016 = None
	        convert_element_type_676: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1765, torch.float32);  view_1765 = None
	        _assert_tensor_metadata_1017 = torch.ops.aten._assert_tensor_metadata.default(view_1767, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1017 = None
	        convert_element_type_677: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1767, torch.float32);  view_1767 = None
	        sub_5164: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_676, convert_element_type_677);  convert_element_type_676 = convert_element_type_677 = None
	        mul_10934: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5164, view_1766);  sub_5164 = view_1766 = None
	        view_1768: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10934, [5120, 1280]);  mul_10934 = None
	        _assert_tensor_metadata_1018 = torch.ops.aten._assert_tensor_metadata.default(view_1768, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1018 = None
	        mul_10939: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1769: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_10929, [mul_10939, 1280]);  mul_10929 = mul_10939 = None
	        permute_189: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1768, [1, 0]);  view_1768 = None
	        addmm_93: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_18_fc1_bias, view_1769, permute_189);  model_audio_tower_layers_18_fc1_bias = view_1769 = permute_189 = None
	        view_1770: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_93, [sym_size_int, 1500, 5120]);  addmm_93 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_10946: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1770, 0.5)
	        mul_10947: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1770, 0.7071067811865476);  view_1770 = None
	        erf_20: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_10947);  mul_10947 = None
	        add_17322: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_20, 1);  erf_20 = None
	        mul_10948: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10946, add_17322);  mul_10946 = add_17322 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_10948, [2])
	        amax_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_10948, [2])
	        full_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_113, full_226);  amin_113 = full_226 = None
	        full_227: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_113, full_227);  amax_113 = full_227 = None
	        sub_5177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_113, minimum_113);  maximum_113 = None
	        div_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5177, 255.0);  sub_5177 = None
	        clamp_min_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_226, 1.1920928955078125e-07);  div_226 = None
	        div_227: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_113, clamp_min_339);  minimum_113 = None
	        round_227: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_227);  div_227 = None
	        sub_5183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_227);  round_227 = None
	        clamp_min_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5183, -128);  sub_5183 = None
	        clamp_max_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_340, 127);  clamp_min_340 = None
	        _assert_tensor_metadata_1019 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_339, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1019 = None
	        _assert_tensor_metadata_1020 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_226, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1020 = None
	        convert_element_type_678: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_226, torch.int8);  clamp_max_226 = None
	        view_1773: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_339, [sym_size_int, 1500, 1])
	        view_1774: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_678, [sym_size_int, 1500, 1])
	        reciprocal_113: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1773);  view_1773 = None
	        mul_10994: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_113, 1.0);  reciprocal_113 = None
	        mul_10997: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10948, mul_10994);  mul_10948 = mul_10994 = None
	        round_228: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_10997);  mul_10997 = None
	        add_17405: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_228, view_1774);  round_228 = view_1774 = None
	        clamp_min_341: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17405, -128);  add_17405 = None
	        clamp_max_227: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_341, 127);  clamp_min_341 = None
	        _assert_tensor_metadata_1021 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_227, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1021 = None
	        convert_element_type_679: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_227, torch.int8);  clamp_max_227 = None
	        view_1777: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_339, [sym_size_int, 1500, 1]);  clamp_min_339 = None
	        view_1778: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_678, [sym_size_int, 1500, 1]);  convert_element_type_678 = None
	        _assert_tensor_metadata_1022 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_679, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1022 = None
	        convert_element_type_680: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_679, torch.float32);  convert_element_type_679 = None
	        _assert_tensor_metadata_1023 = torch.ops.aten._assert_tensor_metadata.default(view_1778, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1023 = None
	        convert_element_type_681: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1778, torch.float32);  view_1778 = None
	        sub_5203: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_680, convert_element_type_681);  convert_element_type_680 = convert_element_type_681 = None
	        mul_11019: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5203, view_1777);  sub_5203 = view_1777 = None
	        _assert_tensor_metadata_1024 = torch.ops.aten._assert_tensor_metadata.default(mul_11019, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1024 = None
	        view_1780: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_18_fc2_parametrizations_weight_original0 = None
	        view_1781: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_18_fc2_parametrizations_weight_original1 = None
	        view_1782: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_18_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_18_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1025 = torch.ops.aten._assert_tensor_metadata.default(view_1780, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1025 = None
	        convert_element_type_682: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1780, torch.float32);  view_1780 = None
	        _assert_tensor_metadata_1026 = torch.ops.aten._assert_tensor_metadata.default(view_1782, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1026 = None
	        convert_element_type_683: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1782, torch.float32);  view_1782 = None
	        sub_5207: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_682, convert_element_type_683);  convert_element_type_682 = convert_element_type_683 = None
	        mul_11024: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5207, view_1781);  sub_5207 = view_1781 = None
	        view_1783: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_11024, [1280, 5120]);  mul_11024 = None
	        _assert_tensor_metadata_1027 = torch.ops.aten._assert_tensor_metadata.default(view_1783, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1027 = None
	        mul_11029: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1784: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_11019, [mul_11029, 5120]);  mul_11019 = mul_11029 = None
	        permute_190: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1783, [1, 0]);  view_1783 = None
	        addmm_94: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_18_fc2_bias, view_1784, permute_190);  model_audio_tower_layers_18_fc2_bias = view_1784 = permute_190 = None
	        view_1785: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_94, [sym_size_int, 1500, 1280]);  addmm_94 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_17468: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_17170, view_1785);  add_17170 = view_1785 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_153: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_17468, memory_format = torch.contiguous_format)
	        var_mean_38 = torch.ops.aten.var_mean.correction(clone_153, [2], correction = 0, keepdim = True)
	        getitem_152: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_38[0]
	        getitem_153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_38[1];  var_mean_38 = None
	        add_17473: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_152, 1e-05);  getitem_152 = None
	        rsqrt_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_17473);  add_17473 = None
	        sub_5213: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_153, getitem_153);  clone_153 = getitem_153 = None
	        mul_11040: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5213, rsqrt_38);  sub_5213 = rsqrt_38 = None
	        mul_11041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11040, model_audio_tower_layers_19_self_attn_layer_norm_weight);  mul_11040 = model_audio_tower_layers_19_self_attn_layer_norm_weight = None
	        add_17474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_11041, model_audio_tower_layers_19_self_attn_layer_norm_bias);  mul_11041 = model_audio_tower_layers_19_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_17474, [2])
	        amax_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_17474, [2])
	        full_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_114, full_228);  amin_114 = full_228 = None
	        full_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_114, full_229);  amax_114 = full_229 = None
	        sub_5224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_114, minimum_114);  maximum_114 = None
	        div_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5224, 255.0);  sub_5224 = None
	        clamp_min_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_228, 1.1920928955078125e-07);  div_228 = None
	        div_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_114, clamp_min_342);  minimum_114 = None
	        round_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_229);  div_229 = None
	        sub_5230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_229);  round_229 = None
	        clamp_min_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5230, -128);  sub_5230 = None
	        clamp_max_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_343, 127);  clamp_min_343 = None
	        _assert_tensor_metadata_1028 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_342, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1028 = None
	        _assert_tensor_metadata_1029 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_228, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1029 = None
	        convert_element_type_684: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_228, torch.int8);  clamp_max_228 = None
	        view_1788: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_342, [sym_size_int, 1500, 1])
	        view_1789: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_684, [sym_size_int, 1500, 1])
	        reciprocal_114: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1788);  view_1788 = None
	        mul_11089: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_114, 1.0);  reciprocal_114 = None
	        mul_11092: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_17474, mul_11089);  mul_11089 = None
	        round_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11092);  mul_11092 = None
	        add_17561: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_230, view_1789);  round_230 = view_1789 = None
	        clamp_min_344: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17561, -128);  add_17561 = None
	        clamp_max_229: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_344, 127);  clamp_min_344 = None
	        _assert_tensor_metadata_1030 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_229, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1030 = None
	        convert_element_type_685: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_229, torch.int8);  clamp_max_229 = None
	        view_1792: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_342, [sym_size_int, 1500, 1]);  clamp_min_342 = None
	        view_1793: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_684, [sym_size_int, 1500, 1]);  convert_element_type_684 = None
	        _assert_tensor_metadata_1031 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_685, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1031 = None
	        convert_element_type_686: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_685, torch.float32);  convert_element_type_685 = None
	        _assert_tensor_metadata_1032 = torch.ops.aten._assert_tensor_metadata.default(view_1793, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1032 = None
	        convert_element_type_687: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1793, torch.float32);  view_1793 = None
	        sub_5250: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_686, convert_element_type_687);  convert_element_type_686 = convert_element_type_687 = None
	        mul_11114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5250, view_1792);  sub_5250 = view_1792 = None
	        _assert_tensor_metadata_1033 = torch.ops.aten._assert_tensor_metadata.default(mul_11114, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1033 = None
	        view_1795: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1796: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1797: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1034 = torch.ops.aten._assert_tensor_metadata.default(view_1795, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1034 = None
	        convert_element_type_688: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1795, torch.float32);  view_1795 = None
	        _assert_tensor_metadata_1035 = torch.ops.aten._assert_tensor_metadata.default(view_1797, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1035 = None
	        convert_element_type_689: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1797, torch.float32);  view_1797 = None
	        sub_5254: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_688, convert_element_type_689);  convert_element_type_688 = convert_element_type_689 = None
	        mul_11119: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5254, view_1796);  sub_5254 = view_1796 = None
	        view_1798: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11119, [1280, 1280]);  mul_11119 = None
	        _assert_tensor_metadata_1036 = torch.ops.aten._assert_tensor_metadata.default(view_1798, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1036 = None
	        mul_11124: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1799: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11114, [mul_11124, 1280]);  mul_11114 = mul_11124 = None
	        permute_191: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1798, [1, 0]);  view_1798 = None
	        addmm_95: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_19_self_attn_q_proj_bias, view_1799, permute_191);  model_audio_tower_layers_19_self_attn_q_proj_bias = view_1799 = permute_191 = None
	        view_1800: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_95, [sym_size_int, 1500, 1280]);  addmm_95 = None
	        mul_11131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1800, 0.125);  view_1800 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1801: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_11131, [sym_size_int, 1500, 20, 64]);  mul_11131 = None
	        permute_192: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1801, [0, 2, 1, 3]);  view_1801 = None
	        clone_154: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_192, memory_format = torch.contiguous_format);  permute_192 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_17474, [2])
	        amax_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_17474, [2])
	        full_230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_115, full_230);  amin_115 = full_230 = None
	        full_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_115, full_231);  amax_115 = full_231 = None
	        sub_5269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_115, minimum_115);  maximum_115 = None
	        div_230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5269, 255.0);  sub_5269 = None
	        clamp_min_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_230, 1.1920928955078125e-07);  div_230 = None
	        div_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_115, clamp_min_345);  minimum_115 = None
	        round_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_231);  div_231 = None
	        sub_5275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_231);  round_231 = None
	        clamp_min_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5275, -128);  sub_5275 = None
	        clamp_max_230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_346, 127);  clamp_min_346 = None
	        _assert_tensor_metadata_1037 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_345, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1037 = None
	        _assert_tensor_metadata_1038 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_230, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1038 = None
	        convert_element_type_690: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_230, torch.int8);  clamp_max_230 = None
	        view_1804: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_345, [sym_size_int, 1500, 1])
	        view_1805: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_690, [sym_size_int, 1500, 1])
	        reciprocal_115: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1804);  view_1804 = None
	        mul_11185: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_115, 1.0);  reciprocal_115 = None
	        mul_11188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_17474, mul_11185);  mul_11185 = None
	        round_232: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11188);  mul_11188 = None
	        add_17713: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_232, view_1805);  round_232 = view_1805 = None
	        clamp_min_347: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17713, -128);  add_17713 = None
	        clamp_max_231: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_347, 127);  clamp_min_347 = None
	        _assert_tensor_metadata_1039 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_231, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1039 = None
	        convert_element_type_691: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_231, torch.int8);  clamp_max_231 = None
	        view_1808: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_345, [sym_size_int, 1500, 1]);  clamp_min_345 = None
	        view_1809: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_690, [sym_size_int, 1500, 1]);  convert_element_type_690 = None
	        _assert_tensor_metadata_1040 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_691, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1040 = None
	        convert_element_type_692: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_691, torch.float32);  convert_element_type_691 = None
	        _assert_tensor_metadata_1041 = torch.ops.aten._assert_tensor_metadata.default(view_1809, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1041 = None
	        convert_element_type_693: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1809, torch.float32);  view_1809 = None
	        sub_5295: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_692, convert_element_type_693);  convert_element_type_692 = convert_element_type_693 = None
	        mul_11210: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5295, view_1808);  sub_5295 = view_1808 = None
	        _assert_tensor_metadata_1042 = torch.ops.aten._assert_tensor_metadata.default(mul_11210, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1042 = None
	        view_1811: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1812: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1813: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1043 = torch.ops.aten._assert_tensor_metadata.default(view_1811, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1043 = None
	        convert_element_type_694: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1811, torch.float32);  view_1811 = None
	        _assert_tensor_metadata_1044 = torch.ops.aten._assert_tensor_metadata.default(view_1813, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1044 = None
	        convert_element_type_695: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1813, torch.float32);  view_1813 = None
	        sub_5299: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_694, convert_element_type_695);  convert_element_type_694 = convert_element_type_695 = None
	        mul_11215: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5299, view_1812);  sub_5299 = view_1812 = None
	        view_1814: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11215, [1280, 1280]);  mul_11215 = None
	        _assert_tensor_metadata_1045 = torch.ops.aten._assert_tensor_metadata.default(view_1814, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1045 = None
	        permute_193: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1814, [1, 0]);  view_1814 = None
	        mul_11218: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1815: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11210, [mul_11218, 1280]);  mul_11210 = mul_11218 = None
	        mm_19: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1815, permute_193);  view_1815 = permute_193 = None
	        view_1816: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_19, [sym_size_int, 1500, 1280]);  mm_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1817: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1816, [sym_size_int, -1, 20, 64]);  view_1816 = None
	        permute_194: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1817, [0, 2, 1, 3]);  view_1817 = None
	        clone_155: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_194, memory_format = torch.contiguous_format);  permute_194 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_17474, [2])
	        amax_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_17474, [2])
	        full_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_116, full_232);  amin_116 = full_232 = None
	        full_233: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_116, full_233);  amax_116 = full_233 = None
	        sub_5313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_116, minimum_116);  maximum_116 = None
	        div_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5313, 255.0);  sub_5313 = None
	        clamp_min_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_232, 1.1920928955078125e-07);  div_232 = None
	        div_233: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_116, clamp_min_348);  minimum_116 = None
	        round_233: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_233);  div_233 = None
	        sub_5319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_233);  round_233 = None
	        clamp_min_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5319, -128);  sub_5319 = None
	        clamp_max_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_349, 127);  clamp_min_349 = None
	        _assert_tensor_metadata_1046 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_348, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1046 = None
	        _assert_tensor_metadata_1047 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_232, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1047 = None
	        convert_element_type_696: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_232, torch.int8);  clamp_max_232 = None
	        view_1820: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_348, [sym_size_int, 1500, 1])
	        view_1821: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_696, [sym_size_int, 1500, 1])
	        reciprocal_116: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1820);  view_1820 = None
	        mul_11284: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_116, 1.0);  reciprocal_116 = None
	        mul_11287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_17474, mul_11284);  add_17474 = mul_11284 = None
	        round_234: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11287);  mul_11287 = None
	        add_17861: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_234, view_1821);  round_234 = view_1821 = None
	        clamp_min_350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17861, -128);  add_17861 = None
	        clamp_max_233: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_350, 127);  clamp_min_350 = None
	        _assert_tensor_metadata_1048 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_233, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1048 = None
	        convert_element_type_697: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_233, torch.int8);  clamp_max_233 = None
	        view_1824: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_348, [sym_size_int, 1500, 1]);  clamp_min_348 = None
	        view_1825: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_696, [sym_size_int, 1500, 1]);  convert_element_type_696 = None
	        _assert_tensor_metadata_1049 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_697, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1049 = None
	        convert_element_type_698: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_697, torch.float32);  convert_element_type_697 = None
	        _assert_tensor_metadata_1050 = torch.ops.aten._assert_tensor_metadata.default(view_1825, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1050 = None
	        convert_element_type_699: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1825, torch.float32);  view_1825 = None
	        sub_5339: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_698, convert_element_type_699);  convert_element_type_698 = convert_element_type_699 = None
	        mul_11309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5339, view_1824);  sub_5339 = view_1824 = None
	        _assert_tensor_metadata_1051 = torch.ops.aten._assert_tensor_metadata.default(mul_11309, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1051 = None
	        view_1827: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1828: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1829: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1052 = torch.ops.aten._assert_tensor_metadata.default(view_1827, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1052 = None
	        convert_element_type_700: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1827, torch.float32);  view_1827 = None
	        _assert_tensor_metadata_1053 = torch.ops.aten._assert_tensor_metadata.default(view_1829, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1053 = None
	        convert_element_type_701: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1829, torch.float32);  view_1829 = None
	        sub_5343: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_700, convert_element_type_701);  convert_element_type_700 = convert_element_type_701 = None
	        mul_11314: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5343, view_1828);  sub_5343 = view_1828 = None
	        view_1830: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11314, [1280, 1280]);  mul_11314 = None
	        _assert_tensor_metadata_1054 = torch.ops.aten._assert_tensor_metadata.default(view_1830, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1054 = None
	        mul_11319: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1831: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11309, [mul_11319, 1280]);  mul_11309 = mul_11319 = None
	        permute_195: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1830, [1, 0]);  view_1830 = None
	        addmm_96: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_19_self_attn_v_proj_bias, view_1831, permute_195);  model_audio_tower_layers_19_self_attn_v_proj_bias = view_1831 = permute_195 = None
	        view_1832: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_96, [sym_size_int, 1500, 1280]);  addmm_96 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1833: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1832, [sym_size_int, -1, 20, 64]);  view_1832 = None
	        permute_196: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1833, [0, 2, 1, 3]);  view_1833 = None
	        clone_156: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_196, memory_format = torch.contiguous_format);  permute_196 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_19 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_154, clone_155, clone_156, None, False, scale = 1.0);  clone_154 = clone_155 = clone_156 = None
	        getitem_154: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_19[0];  _scaled_dot_product_efficient_attention_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_197: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_154, [0, 2, 1, 3]);  getitem_154 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_197, [sym_size_int, 1500, -1]);  permute_197 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1834, [2])
	        amax_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1834, [2])
	        full_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_117, full_234);  amin_117 = full_234 = None
	        full_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_117, full_235);  amax_117 = full_235 = None
	        sub_5361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_117, minimum_117);  maximum_117 = None
	        div_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5361, 255.0);  sub_5361 = None
	        clamp_min_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_234, 1.1920928955078125e-07);  div_234 = None
	        div_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_117, clamp_min_351);  minimum_117 = None
	        round_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_235);  div_235 = None
	        sub_5367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_235);  round_235 = None
	        clamp_min_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5367, -128);  sub_5367 = None
	        clamp_max_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_352, 127);  clamp_min_352 = None
	        _assert_tensor_metadata_1055 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_351, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1055 = None
	        _assert_tensor_metadata_1056 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_234, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1056 = None
	        convert_element_type_702: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_234, torch.int8);  clamp_max_234 = None
	        view_1837: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_351, [sym_size_int, 1500, 1])
	        view_1838: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_702, [sym_size_int, 1500, 1])
	        reciprocal_117: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1837);  view_1837 = None
	        mul_11389: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_117, 1.0);  reciprocal_117 = None
	        mul_11392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1834, mul_11389);  view_1834 = mul_11389 = None
	        round_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11392);  mul_11392 = None
	        add_18025: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_236, view_1838);  round_236 = view_1838 = None
	        clamp_min_353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18025, -128);  add_18025 = None
	        clamp_max_235: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_353, 127);  clamp_min_353 = None
	        _assert_tensor_metadata_1057 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_235, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1057 = None
	        convert_element_type_703: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_235, torch.int8);  clamp_max_235 = None
	        view_1841: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_351, [sym_size_int, 1500, 1]);  clamp_min_351 = None
	        view_1842: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_702, [sym_size_int, 1500, 1]);  convert_element_type_702 = None
	        _assert_tensor_metadata_1058 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_703, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1058 = None
	        convert_element_type_704: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_703, torch.float32);  convert_element_type_703 = None
	        _assert_tensor_metadata_1059 = torch.ops.aten._assert_tensor_metadata.default(view_1842, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1059 = None
	        convert_element_type_705: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1842, torch.float32);  view_1842 = None
	        sub_5387: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_704, convert_element_type_705);  convert_element_type_704 = convert_element_type_705 = None
	        mul_11414: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5387, view_1841);  sub_5387 = view_1841 = None
	        _assert_tensor_metadata_1060 = torch.ops.aten._assert_tensor_metadata.default(mul_11414, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1060 = None
	        view_1844: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1845: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1846: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1061 = torch.ops.aten._assert_tensor_metadata.default(view_1844, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1061 = None
	        convert_element_type_706: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1844, torch.float32);  view_1844 = None
	        _assert_tensor_metadata_1062 = torch.ops.aten._assert_tensor_metadata.default(view_1846, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1062 = None
	        convert_element_type_707: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1846, torch.float32);  view_1846 = None
	        sub_5391: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_706, convert_element_type_707);  convert_element_type_706 = convert_element_type_707 = None
	        mul_11419: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5391, view_1845);  sub_5391 = view_1845 = None
	        view_1847: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11419, [1280, 1280]);  mul_11419 = None
	        _assert_tensor_metadata_1063 = torch.ops.aten._assert_tensor_metadata.default(view_1847, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1063 = None
	        mul_11424: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1848: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11414, [mul_11424, 1280]);  mul_11414 = mul_11424 = None
	        permute_198: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1847, [1, 0]);  view_1847 = None
	        addmm_97: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_19_self_attn_out_proj_bias, view_1848, permute_198);  model_audio_tower_layers_19_self_attn_out_proj_bias = view_1848 = permute_198 = None
	        view_1849: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_97, [sym_size_int, 1500, 1280]);  addmm_97 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_18088: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_17468, view_1849);  add_17468 = view_1849 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_18088, memory_format = torch.contiguous_format)
	        var_mean_39 = torch.ops.aten.var_mean.correction(clone_158, [2], correction = 0, keepdim = True)
	        getitem_158: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_39[0]
	        getitem_159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_39[1];  var_mean_39 = None
	        add_18093: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_158, 1e-05);  getitem_158 = None
	        rsqrt_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_18093);  add_18093 = None
	        sub_5397: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_158, getitem_159);  clone_158 = getitem_159 = None
	        mul_11435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5397, rsqrt_39);  sub_5397 = rsqrt_39 = None
	        mul_11436: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11435, model_audio_tower_layers_19_final_layer_norm_weight);  mul_11435 = model_audio_tower_layers_19_final_layer_norm_weight = None
	        add_18094: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_11436, model_audio_tower_layers_19_final_layer_norm_bias);  mul_11436 = model_audio_tower_layers_19_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_18094, [2])
	        amax_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_18094, [2])
	        full_236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_118, full_236);  amin_118 = full_236 = None
	        full_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_118, full_237);  amax_118 = full_237 = None
	        sub_5408: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_118, minimum_118);  maximum_118 = None
	        div_236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5408, 255.0);  sub_5408 = None
	        clamp_min_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_236, 1.1920928955078125e-07);  div_236 = None
	        div_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_118, clamp_min_354);  minimum_118 = None
	        round_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_237);  div_237 = None
	        sub_5414: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_237);  round_237 = None
	        clamp_min_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5414, -128);  sub_5414 = None
	        clamp_max_236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_355, 127);  clamp_min_355 = None
	        _assert_tensor_metadata_1064 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_354, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1064 = None
	        _assert_tensor_metadata_1065 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_236, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1065 = None
	        convert_element_type_708: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_236, torch.int8);  clamp_max_236 = None
	        view_1852: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_354, [sym_size_int, 1500, 1])
	        view_1853: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_708, [sym_size_int, 1500, 1])
	        reciprocal_118: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1852);  view_1852 = None
	        mul_11484: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_118, 1.0);  reciprocal_118 = None
	        mul_11487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_18094, mul_11484);  add_18094 = mul_11484 = None
	        round_238: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11487);  mul_11487 = None
	        add_18181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_238, view_1853);  round_238 = view_1853 = None
	        clamp_min_356: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18181, -128);  add_18181 = None
	        clamp_max_237: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_356, 127);  clamp_min_356 = None
	        _assert_tensor_metadata_1066 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_237, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1066 = None
	        convert_element_type_709: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_237, torch.int8);  clamp_max_237 = None
	        view_1856: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_354, [sym_size_int, 1500, 1]);  clamp_min_354 = None
	        view_1857: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_708, [sym_size_int, 1500, 1]);  convert_element_type_708 = None
	        _assert_tensor_metadata_1067 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_709, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1067 = None
	        convert_element_type_710: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_709, torch.float32);  convert_element_type_709 = None
	        _assert_tensor_metadata_1068 = torch.ops.aten._assert_tensor_metadata.default(view_1857, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1068 = None
	        convert_element_type_711: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1857, torch.float32);  view_1857 = None
	        sub_5434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_710, convert_element_type_711);  convert_element_type_710 = convert_element_type_711 = None
	        mul_11509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5434, view_1856);  sub_5434 = view_1856 = None
	        _assert_tensor_metadata_1069 = torch.ops.aten._assert_tensor_metadata.default(mul_11509, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1069 = None
	        view_1859: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_19_fc1_parametrizations_weight_original0 = None
	        view_1860: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_19_fc1_parametrizations_weight_original1 = None
	        view_1861: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_19_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1070 = torch.ops.aten._assert_tensor_metadata.default(view_1859, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1070 = None
	        convert_element_type_712: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1859, torch.float32);  view_1859 = None
	        _assert_tensor_metadata_1071 = torch.ops.aten._assert_tensor_metadata.default(view_1861, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1071 = None
	        convert_element_type_713: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1861, torch.float32);  view_1861 = None
	        sub_5438: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_712, convert_element_type_713);  convert_element_type_712 = convert_element_type_713 = None
	        mul_11514: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5438, view_1860);  sub_5438 = view_1860 = None
	        view_1862: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11514, [5120, 1280]);  mul_11514 = None
	        _assert_tensor_metadata_1072 = torch.ops.aten._assert_tensor_metadata.default(view_1862, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1072 = None
	        mul_11519: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1863: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11509, [mul_11519, 1280]);  mul_11509 = mul_11519 = None
	        permute_199: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1862, [1, 0]);  view_1862 = None
	        addmm_98: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_19_fc1_bias, view_1863, permute_199);  model_audio_tower_layers_19_fc1_bias = view_1863 = permute_199 = None
	        view_1864: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_98, [sym_size_int, 1500, 5120]);  addmm_98 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_11526: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1864, 0.5)
	        mul_11527: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1864, 0.7071067811865476);  view_1864 = None
	        erf_21: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_11527);  mul_11527 = None
	        add_18240: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_21, 1);  erf_21 = None
	        mul_11528: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11526, add_18240);  mul_11526 = add_18240 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_11528, [2])
	        amax_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_11528, [2])
	        full_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_119, full_238);  amin_119 = full_238 = None
	        full_239: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_119, full_239);  amax_119 = full_239 = None
	        sub_5451: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_119, minimum_119);  maximum_119 = None
	        div_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5451, 255.0);  sub_5451 = None
	        clamp_min_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_238, 1.1920928955078125e-07);  div_238 = None
	        div_239: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_119, clamp_min_357);  minimum_119 = None
	        round_239: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_239);  div_239 = None
	        sub_5457: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_239);  round_239 = None
	        clamp_min_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5457, -128);  sub_5457 = None
	        clamp_max_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_358, 127);  clamp_min_358 = None
	        _assert_tensor_metadata_1073 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_357, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1073 = None
	        _assert_tensor_metadata_1074 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_238, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1074 = None
	        convert_element_type_714: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_238, torch.int8);  clamp_max_238 = None
	        view_1867: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_357, [sym_size_int, 1500, 1])
	        view_1868: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_714, [sym_size_int, 1500, 1])
	        reciprocal_119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1867);  view_1867 = None
	        mul_11574: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_119, 1.0);  reciprocal_119 = None
	        mul_11577: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11528, mul_11574);  mul_11528 = mul_11574 = None
	        round_240: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_11577);  mul_11577 = None
	        add_18323: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_240, view_1868);  round_240 = view_1868 = None
	        clamp_min_359: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18323, -128);  add_18323 = None
	        clamp_max_239: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_359, 127);  clamp_min_359 = None
	        _assert_tensor_metadata_1075 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_239, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1075 = None
	        convert_element_type_715: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_239, torch.int8);  clamp_max_239 = None
	        view_1871: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_357, [sym_size_int, 1500, 1]);  clamp_min_357 = None
	        view_1872: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_714, [sym_size_int, 1500, 1]);  convert_element_type_714 = None
	        _assert_tensor_metadata_1076 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_715, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1076 = None
	        convert_element_type_716: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_715, torch.float32);  convert_element_type_715 = None
	        _assert_tensor_metadata_1077 = torch.ops.aten._assert_tensor_metadata.default(view_1872, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1077 = None
	        convert_element_type_717: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1872, torch.float32);  view_1872 = None
	        sub_5477: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_716, convert_element_type_717);  convert_element_type_716 = convert_element_type_717 = None
	        mul_11599: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5477, view_1871);  sub_5477 = view_1871 = None
	        _assert_tensor_metadata_1078 = torch.ops.aten._assert_tensor_metadata.default(mul_11599, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1078 = None
	        view_1874: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_19_fc2_parametrizations_weight_original0 = None
	        view_1875: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_19_fc2_parametrizations_weight_original1 = None
	        view_1876: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_19_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_19_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1079 = torch.ops.aten._assert_tensor_metadata.default(view_1874, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1079 = None
	        convert_element_type_718: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1874, torch.float32);  view_1874 = None
	        _assert_tensor_metadata_1080 = torch.ops.aten._assert_tensor_metadata.default(view_1876, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1080 = None
	        convert_element_type_719: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1876, torch.float32);  view_1876 = None
	        sub_5481: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_718, convert_element_type_719);  convert_element_type_718 = convert_element_type_719 = None
	        mul_11604: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5481, view_1875);  sub_5481 = view_1875 = None
	        view_1877: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_11604, [1280, 5120]);  mul_11604 = None
	        _assert_tensor_metadata_1081 = torch.ops.aten._assert_tensor_metadata.default(view_1877, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1081 = None
	        mul_11609: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1878: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_11599, [mul_11609, 5120]);  mul_11599 = mul_11609 = None
	        permute_200: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1877, [1, 0]);  view_1877 = None
	        addmm_99: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_19_fc2_bias, view_1878, permute_200);  model_audio_tower_layers_19_fc2_bias = view_1878 = permute_200 = None
	        view_1879: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_99, [sym_size_int, 1500, 1280]);  addmm_99 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_18386: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_18088, view_1879);  add_18088 = view_1879 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_18386, memory_format = torch.contiguous_format)
	        var_mean_40 = torch.ops.aten.var_mean.correction(clone_161, [2], correction = 0, keepdim = True)
	        getitem_160: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_40[0]
	        getitem_161: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_40[1];  var_mean_40 = None
	        add_18391: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_160, 1e-05);  getitem_160 = None
	        rsqrt_40: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_18391);  add_18391 = None
	        sub_5487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_161, getitem_161);  clone_161 = getitem_161 = None
	        mul_11620: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5487, rsqrt_40);  sub_5487 = rsqrt_40 = None
	        mul_11621: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11620, model_audio_tower_layers_20_self_attn_layer_norm_weight);  mul_11620 = model_audio_tower_layers_20_self_attn_layer_norm_weight = None
	        add_18392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_11621, model_audio_tower_layers_20_self_attn_layer_norm_bias);  mul_11621 = model_audio_tower_layers_20_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_18392, [2])
	        amax_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_18392, [2])
	        full_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_120, full_240);  amin_120 = full_240 = None
	        full_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_120, full_241);  amax_120 = full_241 = None
	        sub_5498: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_120, minimum_120);  maximum_120 = None
	        div_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5498, 255.0);  sub_5498 = None
	        clamp_min_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_240, 1.1920928955078125e-07);  div_240 = None
	        div_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_120, clamp_min_360);  minimum_120 = None
	        round_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_241);  div_241 = None
	        sub_5504: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_241);  round_241 = None
	        clamp_min_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5504, -128);  sub_5504 = None
	        clamp_max_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_361, 127);  clamp_min_361 = None
	        _assert_tensor_metadata_1082 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_360, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1082 = None
	        _assert_tensor_metadata_1083 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_240, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1083 = None
	        convert_element_type_720: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_240, torch.int8);  clamp_max_240 = None
	        view_1882: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_360, [sym_size_int, 1500, 1])
	        view_1883: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_720, [sym_size_int, 1500, 1])
	        reciprocal_120: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1882);  view_1882 = None
	        mul_11669: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_120, 1.0);  reciprocal_120 = None
	        mul_11672: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_18392, mul_11669);  mul_11669 = None
	        round_242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11672);  mul_11672 = None
	        add_18479: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_242, view_1883);  round_242 = view_1883 = None
	        clamp_min_362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18479, -128);  add_18479 = None
	        clamp_max_241: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_362, 127);  clamp_min_362 = None
	        _assert_tensor_metadata_1084 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_241, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1084 = None
	        convert_element_type_721: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_241, torch.int8);  clamp_max_241 = None
	        view_1886: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_360, [sym_size_int, 1500, 1]);  clamp_min_360 = None
	        view_1887: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_720, [sym_size_int, 1500, 1]);  convert_element_type_720 = None
	        _assert_tensor_metadata_1085 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_721, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1085 = None
	        convert_element_type_722: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_721, torch.float32);  convert_element_type_721 = None
	        _assert_tensor_metadata_1086 = torch.ops.aten._assert_tensor_metadata.default(view_1887, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1086 = None
	        convert_element_type_723: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1887, torch.float32);  view_1887 = None
	        sub_5524: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_722, convert_element_type_723);  convert_element_type_722 = convert_element_type_723 = None
	        mul_11694: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5524, view_1886);  sub_5524 = view_1886 = None
	        _assert_tensor_metadata_1087 = torch.ops.aten._assert_tensor_metadata.default(mul_11694, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1087 = None
	        view_1889: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1890: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1891: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1088 = torch.ops.aten._assert_tensor_metadata.default(view_1889, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1088 = None
	        convert_element_type_724: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1889, torch.float32);  view_1889 = None
	        _assert_tensor_metadata_1089 = torch.ops.aten._assert_tensor_metadata.default(view_1891, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1089 = None
	        convert_element_type_725: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1891, torch.float32);  view_1891 = None
	        sub_5528: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_724, convert_element_type_725);  convert_element_type_724 = convert_element_type_725 = None
	        mul_11699: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5528, view_1890);  sub_5528 = view_1890 = None
	        view_1892: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11699, [1280, 1280]);  mul_11699 = None
	        _assert_tensor_metadata_1090 = torch.ops.aten._assert_tensor_metadata.default(view_1892, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1090 = None
	        mul_11704: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1893: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11694, [mul_11704, 1280]);  mul_11694 = mul_11704 = None
	        permute_201: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1892, [1, 0]);  view_1892 = None
	        addmm_100: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_20_self_attn_q_proj_bias, view_1893, permute_201);  model_audio_tower_layers_20_self_attn_q_proj_bias = view_1893 = permute_201 = None
	        view_1894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_100, [sym_size_int, 1500, 1280]);  addmm_100 = None
	        mul_11711: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1894, 0.125);  view_1894 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1895: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_11711, [sym_size_int, 1500, 20, 64]);  mul_11711 = None
	        permute_202: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1895, [0, 2, 1, 3]);  view_1895 = None
	        clone_162: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_202, memory_format = torch.contiguous_format);  permute_202 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_18392, [2])
	        amax_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_18392, [2])
	        full_242: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_121, full_242);  amin_121 = full_242 = None
	        full_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_121, full_243);  amax_121 = full_243 = None
	        sub_5543: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_121, minimum_121);  maximum_121 = None
	        div_242: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5543, 255.0);  sub_5543 = None
	        clamp_min_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_242, 1.1920928955078125e-07);  div_242 = None
	        div_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_121, clamp_min_363);  minimum_121 = None
	        round_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_243);  div_243 = None
	        sub_5549: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_243);  round_243 = None
	        clamp_min_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5549, -128);  sub_5549 = None
	        clamp_max_242: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_364, 127);  clamp_min_364 = None
	        _assert_tensor_metadata_1091 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_363, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1091 = None
	        _assert_tensor_metadata_1092 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_242, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1092 = None
	        convert_element_type_726: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_242, torch.int8);  clamp_max_242 = None
	        view_1898: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_363, [sym_size_int, 1500, 1])
	        view_1899: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_726, [sym_size_int, 1500, 1])
	        reciprocal_121: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1898);  view_1898 = None
	        mul_11765: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_121, 1.0);  reciprocal_121 = None
	        mul_11768: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_18392, mul_11765);  mul_11765 = None
	        round_244: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11768);  mul_11768 = None
	        add_18631: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_244, view_1899);  round_244 = view_1899 = None
	        clamp_min_365: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18631, -128);  add_18631 = None
	        clamp_max_243: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_365, 127);  clamp_min_365 = None
	        _assert_tensor_metadata_1093 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_243, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1093 = None
	        convert_element_type_727: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_243, torch.int8);  clamp_max_243 = None
	        view_1902: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_363, [sym_size_int, 1500, 1]);  clamp_min_363 = None
	        view_1903: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_726, [sym_size_int, 1500, 1]);  convert_element_type_726 = None
	        _assert_tensor_metadata_1094 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_727, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1094 = None
	        convert_element_type_728: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_727, torch.float32);  convert_element_type_727 = None
	        _assert_tensor_metadata_1095 = torch.ops.aten._assert_tensor_metadata.default(view_1903, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1095 = None
	        convert_element_type_729: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1903, torch.float32);  view_1903 = None
	        sub_5569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_728, convert_element_type_729);  convert_element_type_728 = convert_element_type_729 = None
	        mul_11790: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5569, view_1902);  sub_5569 = view_1902 = None
	        _assert_tensor_metadata_1096 = torch.ops.aten._assert_tensor_metadata.default(mul_11790, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1096 = None
	        view_1905: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1906: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1907: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1097 = torch.ops.aten._assert_tensor_metadata.default(view_1905, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1097 = None
	        convert_element_type_730: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1905, torch.float32);  view_1905 = None
	        _assert_tensor_metadata_1098 = torch.ops.aten._assert_tensor_metadata.default(view_1907, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1098 = None
	        convert_element_type_731: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1907, torch.float32);  view_1907 = None
	        sub_5573: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_730, convert_element_type_731);  convert_element_type_730 = convert_element_type_731 = None
	        mul_11795: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5573, view_1906);  sub_5573 = view_1906 = None
	        view_1908: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11795, [1280, 1280]);  mul_11795 = None
	        _assert_tensor_metadata_1099 = torch.ops.aten._assert_tensor_metadata.default(view_1908, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1099 = None
	        permute_203: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1908, [1, 0]);  view_1908 = None
	        mul_11798: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1909: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11790, [mul_11798, 1280]);  mul_11790 = mul_11798 = None
	        mm_20: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1909, permute_203);  view_1909 = permute_203 = None
	        view_1910: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_20, [sym_size_int, 1500, 1280]);  mm_20 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1911: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1910, [sym_size_int, -1, 20, 64]);  view_1910 = None
	        permute_204: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1911, [0, 2, 1, 3]);  view_1911 = None
	        clone_163: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_204, memory_format = torch.contiguous_format);  permute_204 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_18392, [2])
	        amax_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_18392, [2])
	        full_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_122, full_244);  amin_122 = full_244 = None
	        full_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_122, full_245);  amax_122 = full_245 = None
	        sub_5587: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_122, minimum_122);  maximum_122 = None
	        div_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5587, 255.0);  sub_5587 = None
	        clamp_min_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_244, 1.1920928955078125e-07);  div_244 = None
	        div_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_122, clamp_min_366);  minimum_122 = None
	        round_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_245);  div_245 = None
	        sub_5593: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_245);  round_245 = None
	        clamp_min_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5593, -128);  sub_5593 = None
	        clamp_max_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_367, 127);  clamp_min_367 = None
	        _assert_tensor_metadata_1100 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_366, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1100 = None
	        _assert_tensor_metadata_1101 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_244, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1101 = None
	        convert_element_type_732: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_244, torch.int8);  clamp_max_244 = None
	        view_1914: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_366, [sym_size_int, 1500, 1])
	        view_1915: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_732, [sym_size_int, 1500, 1])
	        reciprocal_122: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1914);  view_1914 = None
	        mul_11864: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_122, 1.0);  reciprocal_122 = None
	        mul_11867: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_18392, mul_11864);  add_18392 = mul_11864 = None
	        round_246: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11867);  mul_11867 = None
	        add_18779: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_246, view_1915);  round_246 = view_1915 = None
	        clamp_min_368: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18779, -128);  add_18779 = None
	        clamp_max_245: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_368, 127);  clamp_min_368 = None
	        _assert_tensor_metadata_1102 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_245, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1102 = None
	        convert_element_type_733: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_245, torch.int8);  clamp_max_245 = None
	        view_1918: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_366, [sym_size_int, 1500, 1]);  clamp_min_366 = None
	        view_1919: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_732, [sym_size_int, 1500, 1]);  convert_element_type_732 = None
	        _assert_tensor_metadata_1103 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_733, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1103 = None
	        convert_element_type_734: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_733, torch.float32);  convert_element_type_733 = None
	        _assert_tensor_metadata_1104 = torch.ops.aten._assert_tensor_metadata.default(view_1919, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1104 = None
	        convert_element_type_735: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1919, torch.float32);  view_1919 = None
	        sub_5613: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_734, convert_element_type_735);  convert_element_type_734 = convert_element_type_735 = None
	        mul_11889: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5613, view_1918);  sub_5613 = view_1918 = None
	        _assert_tensor_metadata_1105 = torch.ops.aten._assert_tensor_metadata.default(mul_11889, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1105 = None
	        view_1921: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1922: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1923: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1106 = torch.ops.aten._assert_tensor_metadata.default(view_1921, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1106 = None
	        convert_element_type_736: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1921, torch.float32);  view_1921 = None
	        _assert_tensor_metadata_1107 = torch.ops.aten._assert_tensor_metadata.default(view_1923, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1107 = None
	        convert_element_type_737: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1923, torch.float32);  view_1923 = None
	        sub_5617: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_736, convert_element_type_737);  convert_element_type_736 = convert_element_type_737 = None
	        mul_11894: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5617, view_1922);  sub_5617 = view_1922 = None
	        view_1924: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11894, [1280, 1280]);  mul_11894 = None
	        _assert_tensor_metadata_1108 = torch.ops.aten._assert_tensor_metadata.default(view_1924, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1108 = None
	        mul_11899: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1925: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11889, [mul_11899, 1280]);  mul_11889 = mul_11899 = None
	        permute_205: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1924, [1, 0]);  view_1924 = None
	        addmm_101: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_20_self_attn_v_proj_bias, view_1925, permute_205);  model_audio_tower_layers_20_self_attn_v_proj_bias = view_1925 = permute_205 = None
	        view_1926: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_101, [sym_size_int, 1500, 1280]);  addmm_101 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1927: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_1926, [sym_size_int, -1, 20, 64]);  view_1926 = None
	        permute_206: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1927, [0, 2, 1, 3]);  view_1927 = None
	        clone_164: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_206, memory_format = torch.contiguous_format);  permute_206 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_20 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_162, clone_163, clone_164, None, False, scale = 1.0);  clone_162 = clone_163 = clone_164 = None
	        getitem_162: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_20[0];  _scaled_dot_product_efficient_attention_20 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_207: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_162, [0, 2, 1, 3]);  getitem_162 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1928: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_207, [sym_size_int, 1500, -1]);  permute_207 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1928, [2])
	        amax_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1928, [2])
	        full_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_123, full_246);  amin_123 = full_246 = None
	        full_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_123, full_247);  amax_123 = full_247 = None
	        sub_5635: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_123, minimum_123);  maximum_123 = None
	        div_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5635, 255.0);  sub_5635 = None
	        clamp_min_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_246, 1.1920928955078125e-07);  div_246 = None
	        div_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_123, clamp_min_369);  minimum_123 = None
	        round_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_247);  div_247 = None
	        sub_5641: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_247);  round_247 = None
	        clamp_min_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5641, -128);  sub_5641 = None
	        clamp_max_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_370, 127);  clamp_min_370 = None
	        _assert_tensor_metadata_1109 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_369, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1109 = None
	        _assert_tensor_metadata_1110 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_246, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1110 = None
	        convert_element_type_738: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_246, torch.int8);  clamp_max_246 = None
	        view_1931: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_369, [sym_size_int, 1500, 1])
	        view_1932: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_738, [sym_size_int, 1500, 1])
	        reciprocal_123: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1931);  view_1931 = None
	        mul_11969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_123, 1.0);  reciprocal_123 = None
	        mul_11972: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1928, mul_11969);  view_1928 = mul_11969 = None
	        round_248: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11972);  mul_11972 = None
	        add_18943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_248, view_1932);  round_248 = view_1932 = None
	        clamp_min_371: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18943, -128);  add_18943 = None
	        clamp_max_247: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_371, 127);  clamp_min_371 = None
	        _assert_tensor_metadata_1111 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_247, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1111 = None
	        convert_element_type_739: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_247, torch.int8);  clamp_max_247 = None
	        view_1935: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_369, [sym_size_int, 1500, 1]);  clamp_min_369 = None
	        view_1936: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_738, [sym_size_int, 1500, 1]);  convert_element_type_738 = None
	        _assert_tensor_metadata_1112 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_739, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1112 = None
	        convert_element_type_740: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_739, torch.float32);  convert_element_type_739 = None
	        _assert_tensor_metadata_1113 = torch.ops.aten._assert_tensor_metadata.default(view_1936, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1113 = None
	        convert_element_type_741: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1936, torch.float32);  view_1936 = None
	        sub_5661: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_740, convert_element_type_741);  convert_element_type_740 = convert_element_type_741 = None
	        mul_11994: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5661, view_1935);  sub_5661 = view_1935 = None
	        _assert_tensor_metadata_1114 = torch.ops.aten._assert_tensor_metadata.default(mul_11994, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1114 = None
	        view_1938: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1939: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1940: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1115 = torch.ops.aten._assert_tensor_metadata.default(view_1938, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1115 = None
	        convert_element_type_742: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1938, torch.float32);  view_1938 = None
	        _assert_tensor_metadata_1116 = torch.ops.aten._assert_tensor_metadata.default(view_1940, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1116 = None
	        convert_element_type_743: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1940, torch.float32);  view_1940 = None
	        sub_5665: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_742, convert_element_type_743);  convert_element_type_742 = convert_element_type_743 = None
	        mul_11999: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5665, view_1939);  sub_5665 = view_1939 = None
	        view_1941: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11999, [1280, 1280]);  mul_11999 = None
	        _assert_tensor_metadata_1117 = torch.ops.aten._assert_tensor_metadata.default(view_1941, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1117 = None
	        mul_12004: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1942: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_11994, [mul_12004, 1280]);  mul_11994 = mul_12004 = None
	        permute_208: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1941, [1, 0]);  view_1941 = None
	        addmm_102: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_20_self_attn_out_proj_bias, view_1942, permute_208);  model_audio_tower_layers_20_self_attn_out_proj_bias = view_1942 = permute_208 = None
	        view_1943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_102, [sym_size_int, 1500, 1280]);  addmm_102 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_19006: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_18386, view_1943);  add_18386 = view_1943 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_166: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_19006, memory_format = torch.contiguous_format)
	        var_mean_41 = torch.ops.aten.var_mean.correction(clone_166, [2], correction = 0, keepdim = True)
	        getitem_166: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_41[0]
	        getitem_167: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_41[1];  var_mean_41 = None
	        add_19011: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_166, 1e-05);  getitem_166 = None
	        rsqrt_41: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_19011);  add_19011 = None
	        sub_5671: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_166, getitem_167);  clone_166 = getitem_167 = None
	        mul_12015: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5671, rsqrt_41);  sub_5671 = rsqrt_41 = None
	        mul_12016: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12015, model_audio_tower_layers_20_final_layer_norm_weight);  mul_12015 = model_audio_tower_layers_20_final_layer_norm_weight = None
	        add_19012: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12016, model_audio_tower_layers_20_final_layer_norm_bias);  mul_12016 = model_audio_tower_layers_20_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_19012, [2])
	        amax_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_19012, [2])
	        full_248: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_124, full_248);  amin_124 = full_248 = None
	        full_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_124, full_249);  amax_124 = full_249 = None
	        sub_5682: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_124, minimum_124);  maximum_124 = None
	        div_248: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5682, 255.0);  sub_5682 = None
	        clamp_min_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_248, 1.1920928955078125e-07);  div_248 = None
	        div_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_124, clamp_min_372);  minimum_124 = None
	        round_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_249);  div_249 = None
	        sub_5688: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_249);  round_249 = None
	        clamp_min_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5688, -128);  sub_5688 = None
	        clamp_max_248: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_373, 127);  clamp_min_373 = None
	        _assert_tensor_metadata_1118 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_372, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1118 = None
	        _assert_tensor_metadata_1119 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_248, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1119 = None
	        convert_element_type_744: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_248, torch.int8);  clamp_max_248 = None
	        view_1946: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_372, [sym_size_int, 1500, 1])
	        view_1947: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_744, [sym_size_int, 1500, 1])
	        reciprocal_124: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1946);  view_1946 = None
	        mul_12064: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_124, 1.0);  reciprocal_124 = None
	        mul_12067: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_19012, mul_12064);  add_19012 = mul_12064 = None
	        round_250: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12067);  mul_12067 = None
	        add_19099: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_250, view_1947);  round_250 = view_1947 = None
	        clamp_min_374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19099, -128);  add_19099 = None
	        clamp_max_249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_374, 127);  clamp_min_374 = None
	        _assert_tensor_metadata_1120 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1120 = None
	        convert_element_type_745: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_249, torch.int8);  clamp_max_249 = None
	        view_1950: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_372, [sym_size_int, 1500, 1]);  clamp_min_372 = None
	        view_1951: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_744, [sym_size_int, 1500, 1]);  convert_element_type_744 = None
	        _assert_tensor_metadata_1121 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_745, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1121 = None
	        convert_element_type_746: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_745, torch.float32);  convert_element_type_745 = None
	        _assert_tensor_metadata_1122 = torch.ops.aten._assert_tensor_metadata.default(view_1951, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1122 = None
	        convert_element_type_747: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1951, torch.float32);  view_1951 = None
	        sub_5708: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_746, convert_element_type_747);  convert_element_type_746 = convert_element_type_747 = None
	        mul_12089: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5708, view_1950);  sub_5708 = view_1950 = None
	        _assert_tensor_metadata_1123 = torch.ops.aten._assert_tensor_metadata.default(mul_12089, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1123 = None
	        view_1953: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_20_fc1_parametrizations_weight_original0 = None
	        view_1954: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_20_fc1_parametrizations_weight_original1 = None
	        view_1955: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_20_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1124 = torch.ops.aten._assert_tensor_metadata.default(view_1953, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1124 = None
	        convert_element_type_748: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1953, torch.float32);  view_1953 = None
	        _assert_tensor_metadata_1125 = torch.ops.aten._assert_tensor_metadata.default(view_1955, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1125 = None
	        convert_element_type_749: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1955, torch.float32);  view_1955 = None
	        sub_5712: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_748, convert_element_type_749);  convert_element_type_748 = convert_element_type_749 = None
	        mul_12094: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5712, view_1954);  sub_5712 = view_1954 = None
	        view_1956: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12094, [5120, 1280]);  mul_12094 = None
	        _assert_tensor_metadata_1126 = torch.ops.aten._assert_tensor_metadata.default(view_1956, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1126 = None
	        mul_12099: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1957: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12089, [mul_12099, 1280]);  mul_12089 = mul_12099 = None
	        permute_209: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1956, [1, 0]);  view_1956 = None
	        addmm_103: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_20_fc1_bias, view_1957, permute_209);  model_audio_tower_layers_20_fc1_bias = view_1957 = permute_209 = None
	        view_1958: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_103, [sym_size_int, 1500, 5120]);  addmm_103 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_12106: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1958, 0.5)
	        mul_12107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1958, 0.7071067811865476);  view_1958 = None
	        erf_22: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_12107);  mul_12107 = None
	        add_19158: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_22, 1);  erf_22 = None
	        mul_12108: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12106, add_19158);  mul_12106 = add_19158 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_12108, [2])
	        amax_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_12108, [2])
	        full_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_125, full_250);  amin_125 = full_250 = None
	        full_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_125, full_251);  amax_125 = full_251 = None
	        sub_5725: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_125, minimum_125);  maximum_125 = None
	        div_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5725, 255.0);  sub_5725 = None
	        clamp_min_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_250, 1.1920928955078125e-07);  div_250 = None
	        div_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_125, clamp_min_375);  minimum_125 = None
	        round_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_251);  div_251 = None
	        sub_5731: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_251);  round_251 = None
	        clamp_min_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5731, -128);  sub_5731 = None
	        clamp_max_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_376, 127);  clamp_min_376 = None
	        _assert_tensor_metadata_1127 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_375, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1127 = None
	        _assert_tensor_metadata_1128 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_250, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1128 = None
	        convert_element_type_750: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_250, torch.int8);  clamp_max_250 = None
	        view_1961: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_375, [sym_size_int, 1500, 1])
	        view_1962: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_750, [sym_size_int, 1500, 1])
	        reciprocal_125: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1961);  view_1961 = None
	        mul_12154: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_125, 1.0);  reciprocal_125 = None
	        mul_12157: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12108, mul_12154);  mul_12108 = mul_12154 = None
	        round_252: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_12157);  mul_12157 = None
	        add_19241: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_252, view_1962);  round_252 = view_1962 = None
	        clamp_min_377: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19241, -128);  add_19241 = None
	        clamp_max_251: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_377, 127);  clamp_min_377 = None
	        _assert_tensor_metadata_1129 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_251, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1129 = None
	        convert_element_type_751: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_251, torch.int8);  clamp_max_251 = None
	        view_1965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_375, [sym_size_int, 1500, 1]);  clamp_min_375 = None
	        view_1966: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_750, [sym_size_int, 1500, 1]);  convert_element_type_750 = None
	        _assert_tensor_metadata_1130 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_751, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1130 = None
	        convert_element_type_752: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_751, torch.float32);  convert_element_type_751 = None
	        _assert_tensor_metadata_1131 = torch.ops.aten._assert_tensor_metadata.default(view_1966, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1131 = None
	        convert_element_type_753: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1966, torch.float32);  view_1966 = None
	        sub_5751: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_752, convert_element_type_753);  convert_element_type_752 = convert_element_type_753 = None
	        mul_12179: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5751, view_1965);  sub_5751 = view_1965 = None
	        _assert_tensor_metadata_1132 = torch.ops.aten._assert_tensor_metadata.default(mul_12179, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1132 = None
	        view_1968: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_20_fc2_parametrizations_weight_original0 = None
	        view_1969: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_20_fc2_parametrizations_weight_original1 = None
	        view_1970: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_20_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_20_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1133 = torch.ops.aten._assert_tensor_metadata.default(view_1968, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1133 = None
	        convert_element_type_754: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1968, torch.float32);  view_1968 = None
	        _assert_tensor_metadata_1134 = torch.ops.aten._assert_tensor_metadata.default(view_1970, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1134 = None
	        convert_element_type_755: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1970, torch.float32);  view_1970 = None
	        sub_5755: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_754, convert_element_type_755);  convert_element_type_754 = convert_element_type_755 = None
	        mul_12184: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5755, view_1969);  sub_5755 = view_1969 = None
	        view_1971: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_12184, [1280, 5120]);  mul_12184 = None
	        _assert_tensor_metadata_1135 = torch.ops.aten._assert_tensor_metadata.default(view_1971, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1135 = None
	        mul_12189: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1972: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_12179, [mul_12189, 5120]);  mul_12179 = mul_12189 = None
	        permute_210: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1971, [1, 0]);  view_1971 = None
	        addmm_104: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_20_fc2_bias, view_1972, permute_210);  model_audio_tower_layers_20_fc2_bias = view_1972 = permute_210 = None
	        view_1973: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_104, [sym_size_int, 1500, 1280]);  addmm_104 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_19304: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_19006, view_1973);  add_19006 = view_1973 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_169: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_19304, memory_format = torch.contiguous_format)
	        var_mean_42 = torch.ops.aten.var_mean.correction(clone_169, [2], correction = 0, keepdim = True)
	        getitem_168: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_42[0]
	        getitem_169: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_42[1];  var_mean_42 = None
	        add_19309: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_168, 1e-05);  getitem_168 = None
	        rsqrt_42: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_19309);  add_19309 = None
	        sub_5761: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_169, getitem_169);  clone_169 = getitem_169 = None
	        mul_12200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5761, rsqrt_42);  sub_5761 = rsqrt_42 = None
	        mul_12201: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12200, model_audio_tower_layers_21_self_attn_layer_norm_weight);  mul_12200 = model_audio_tower_layers_21_self_attn_layer_norm_weight = None
	        add_19310: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12201, model_audio_tower_layers_21_self_attn_layer_norm_bias);  mul_12201 = model_audio_tower_layers_21_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_19310, [2])
	        amax_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_19310, [2])
	        full_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_126, full_252);  amin_126 = full_252 = None
	        full_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_126, full_253);  amax_126 = full_253 = None
	        sub_5772: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_126, minimum_126);  maximum_126 = None
	        div_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5772, 255.0);  sub_5772 = None
	        clamp_min_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_252, 1.1920928955078125e-07);  div_252 = None
	        div_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_126, clamp_min_378);  minimum_126 = None
	        round_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_253);  div_253 = None
	        sub_5778: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_253);  round_253 = None
	        clamp_min_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5778, -128);  sub_5778 = None
	        clamp_max_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_379, 127);  clamp_min_379 = None
	        _assert_tensor_metadata_1136 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_378, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1136 = None
	        _assert_tensor_metadata_1137 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_252, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1137 = None
	        convert_element_type_756: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_252, torch.int8);  clamp_max_252 = None
	        view_1976: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_378, [sym_size_int, 1500, 1])
	        view_1977: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_756, [sym_size_int, 1500, 1])
	        reciprocal_126: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1976);  view_1976 = None
	        mul_12249: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_126, 1.0);  reciprocal_126 = None
	        mul_12252: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_19310, mul_12249);  mul_12249 = None
	        round_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12252);  mul_12252 = None
	        add_19397: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_254, view_1977);  round_254 = view_1977 = None
	        clamp_min_380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19397, -128);  add_19397 = None
	        clamp_max_253: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_380, 127);  clamp_min_380 = None
	        _assert_tensor_metadata_1138 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_253, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1138 = None
	        convert_element_type_757: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_253, torch.int8);  clamp_max_253 = None
	        view_1980: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_378, [sym_size_int, 1500, 1]);  clamp_min_378 = None
	        view_1981: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_756, [sym_size_int, 1500, 1]);  convert_element_type_756 = None
	        _assert_tensor_metadata_1139 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_757, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1139 = None
	        convert_element_type_758: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_757, torch.float32);  convert_element_type_757 = None
	        _assert_tensor_metadata_1140 = torch.ops.aten._assert_tensor_metadata.default(view_1981, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1140 = None
	        convert_element_type_759: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1981, torch.float32);  view_1981 = None
	        sub_5798: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_758, convert_element_type_759);  convert_element_type_758 = convert_element_type_759 = None
	        mul_12274: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5798, view_1980);  sub_5798 = view_1980 = None
	        _assert_tensor_metadata_1141 = torch.ops.aten._assert_tensor_metadata.default(mul_12274, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1141 = None
	        view_1983: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1984: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1985: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1142 = torch.ops.aten._assert_tensor_metadata.default(view_1983, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1142 = None
	        convert_element_type_760: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1983, torch.float32);  view_1983 = None
	        _assert_tensor_metadata_1143 = torch.ops.aten._assert_tensor_metadata.default(view_1985, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1143 = None
	        convert_element_type_761: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1985, torch.float32);  view_1985 = None
	        sub_5802: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_760, convert_element_type_761);  convert_element_type_760 = convert_element_type_761 = None
	        mul_12279: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5802, view_1984);  sub_5802 = view_1984 = None
	        view_1986: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12279, [1280, 1280]);  mul_12279 = None
	        _assert_tensor_metadata_1144 = torch.ops.aten._assert_tensor_metadata.default(view_1986, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1144 = None
	        mul_12284: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1987: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12274, [mul_12284, 1280]);  mul_12274 = mul_12284 = None
	        permute_211: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1986, [1, 0]);  view_1986 = None
	        addmm_105: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_21_self_attn_q_proj_bias, view_1987, permute_211);  model_audio_tower_layers_21_self_attn_q_proj_bias = view_1987 = permute_211 = None
	        view_1988: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_105, [sym_size_int, 1500, 1280]);  addmm_105 = None
	        mul_12291: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1988, 0.125);  view_1988 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1989: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_12291, [sym_size_int, 1500, 20, 64]);  mul_12291 = None
	        permute_212: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1989, [0, 2, 1, 3]);  view_1989 = None
	        clone_170: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_212, memory_format = torch.contiguous_format);  permute_212 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_19310, [2])
	        amax_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_19310, [2])
	        full_254: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_127, full_254);  amin_127 = full_254 = None
	        full_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_127, full_255);  amax_127 = full_255 = None
	        sub_5817: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_127, minimum_127);  maximum_127 = None
	        div_254: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5817, 255.0);  sub_5817 = None
	        clamp_min_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_254, 1.1920928955078125e-07);  div_254 = None
	        div_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_127, clamp_min_381);  minimum_127 = None
	        round_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_255);  div_255 = None
	        sub_5823: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_255);  round_255 = None
	        clamp_min_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5823, -128);  sub_5823 = None
	        clamp_max_254: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_382, 127);  clamp_min_382 = None
	        _assert_tensor_metadata_1145 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_381, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1145 = None
	        _assert_tensor_metadata_1146 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_254, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1146 = None
	        convert_element_type_762: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_254, torch.int8);  clamp_max_254 = None
	        view_1992: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_381, [sym_size_int, 1500, 1])
	        view_1993: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_762, [sym_size_int, 1500, 1])
	        reciprocal_127: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1992);  view_1992 = None
	        mul_12345: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_127, 1.0);  reciprocal_127 = None
	        mul_12348: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_19310, mul_12345);  mul_12345 = None
	        round_256: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12348);  mul_12348 = None
	        add_19549: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_256, view_1993);  round_256 = view_1993 = None
	        clamp_min_383: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19549, -128);  add_19549 = None
	        clamp_max_255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_383, 127);  clamp_min_383 = None
	        _assert_tensor_metadata_1147 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_255, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1147 = None
	        convert_element_type_763: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_255, torch.int8);  clamp_max_255 = None
	        view_1996: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_381, [sym_size_int, 1500, 1]);  clamp_min_381 = None
	        view_1997: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_762, [sym_size_int, 1500, 1]);  convert_element_type_762 = None
	        _assert_tensor_metadata_1148 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_763, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1148 = None
	        convert_element_type_764: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_763, torch.float32);  convert_element_type_763 = None
	        _assert_tensor_metadata_1149 = torch.ops.aten._assert_tensor_metadata.default(view_1997, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1149 = None
	        convert_element_type_765: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1997, torch.float32);  view_1997 = None
	        sub_5843: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_764, convert_element_type_765);  convert_element_type_764 = convert_element_type_765 = None
	        mul_12370: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5843, view_1996);  sub_5843 = view_1996 = None
	        _assert_tensor_metadata_1150 = torch.ops.aten._assert_tensor_metadata.default(mul_12370, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1150 = None
	        view_1999: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2000: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2001: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1151 = torch.ops.aten._assert_tensor_metadata.default(view_1999, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1151 = None
	        convert_element_type_766: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1999, torch.float32);  view_1999 = None
	        _assert_tensor_metadata_1152 = torch.ops.aten._assert_tensor_metadata.default(view_2001, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1152 = None
	        convert_element_type_767: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2001, torch.float32);  view_2001 = None
	        sub_5847: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_766, convert_element_type_767);  convert_element_type_766 = convert_element_type_767 = None
	        mul_12375: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5847, view_2000);  sub_5847 = view_2000 = None
	        view_2002: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12375, [1280, 1280]);  mul_12375 = None
	        _assert_tensor_metadata_1153 = torch.ops.aten._assert_tensor_metadata.default(view_2002, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1153 = None
	        permute_213: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2002, [1, 0]);  view_2002 = None
	        mul_12378: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2003: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12370, [mul_12378, 1280]);  mul_12370 = mul_12378 = None
	        mm_21: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2003, permute_213);  view_2003 = permute_213 = None
	        view_2004: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_21, [sym_size_int, 1500, 1280]);  mm_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2005: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2004, [sym_size_int, -1, 20, 64]);  view_2004 = None
	        permute_214: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2005, [0, 2, 1, 3]);  view_2005 = None
	        clone_171: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_214, memory_format = torch.contiguous_format);  permute_214 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_19310, [2])
	        amax_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_19310, [2])
	        full_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_128, full_256);  amin_128 = full_256 = None
	        full_257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_128, full_257);  amax_128 = full_257 = None
	        sub_5861: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_128, minimum_128);  maximum_128 = None
	        div_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5861, 255.0);  sub_5861 = None
	        clamp_min_384: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_256, 1.1920928955078125e-07);  div_256 = None
	        div_257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_128, clamp_min_384);  minimum_128 = None
	        round_257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_257);  div_257 = None
	        sub_5867: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_257);  round_257 = None
	        clamp_min_385: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5867, -128);  sub_5867 = None
	        clamp_max_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_385, 127);  clamp_min_385 = None
	        _assert_tensor_metadata_1154 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_384, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1154 = None
	        _assert_tensor_metadata_1155 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_256, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1155 = None
	        convert_element_type_768: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_256, torch.int8);  clamp_max_256 = None
	        view_2008: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_384, [sym_size_int, 1500, 1])
	        view_2009: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_768, [sym_size_int, 1500, 1])
	        reciprocal_128: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2008);  view_2008 = None
	        mul_12444: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_128, 1.0);  reciprocal_128 = None
	        mul_12447: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_19310, mul_12444);  add_19310 = mul_12444 = None
	        round_258: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12447);  mul_12447 = None
	        add_19697: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_258, view_2009);  round_258 = view_2009 = None
	        clamp_min_386: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19697, -128);  add_19697 = None
	        clamp_max_257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_386, 127);  clamp_min_386 = None
	        _assert_tensor_metadata_1156 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_257, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1156 = None
	        convert_element_type_769: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_257, torch.int8);  clamp_max_257 = None
	        view_2012: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_384, [sym_size_int, 1500, 1]);  clamp_min_384 = None
	        view_2013: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_768, [sym_size_int, 1500, 1]);  convert_element_type_768 = None
	        _assert_tensor_metadata_1157 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_769, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1157 = None
	        convert_element_type_770: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_769, torch.float32);  convert_element_type_769 = None
	        _assert_tensor_metadata_1158 = torch.ops.aten._assert_tensor_metadata.default(view_2013, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1158 = None
	        convert_element_type_771: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2013, torch.float32);  view_2013 = None
	        sub_5887: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_770, convert_element_type_771);  convert_element_type_770 = convert_element_type_771 = None
	        mul_12469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5887, view_2012);  sub_5887 = view_2012 = None
	        _assert_tensor_metadata_1159 = torch.ops.aten._assert_tensor_metadata.default(mul_12469, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1159 = None
	        view_2015: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2016: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2017: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1160 = torch.ops.aten._assert_tensor_metadata.default(view_2015, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1160 = None
	        convert_element_type_772: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2015, torch.float32);  view_2015 = None
	        _assert_tensor_metadata_1161 = torch.ops.aten._assert_tensor_metadata.default(view_2017, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1161 = None
	        convert_element_type_773: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2017, torch.float32);  view_2017 = None
	        sub_5891: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_772, convert_element_type_773);  convert_element_type_772 = convert_element_type_773 = None
	        mul_12474: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5891, view_2016);  sub_5891 = view_2016 = None
	        view_2018: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12474, [1280, 1280]);  mul_12474 = None
	        _assert_tensor_metadata_1162 = torch.ops.aten._assert_tensor_metadata.default(view_2018, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1162 = None
	        mul_12479: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2019: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12469, [mul_12479, 1280]);  mul_12469 = mul_12479 = None
	        permute_215: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2018, [1, 0]);  view_2018 = None
	        addmm_106: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_21_self_attn_v_proj_bias, view_2019, permute_215);  model_audio_tower_layers_21_self_attn_v_proj_bias = view_2019 = permute_215 = None
	        view_2020: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_106, [sym_size_int, 1500, 1280]);  addmm_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2021: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2020, [sym_size_int, -1, 20, 64]);  view_2020 = None
	        permute_216: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2021, [0, 2, 1, 3]);  view_2021 = None
	        clone_172: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_216, memory_format = torch.contiguous_format);  permute_216 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_21 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_170, clone_171, clone_172, None, False, scale = 1.0);  clone_170 = clone_171 = clone_172 = None
	        getitem_170: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_21[0];  _scaled_dot_product_efficient_attention_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_217: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_170, [0, 2, 1, 3]);  getitem_170 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2022: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_217, [sym_size_int, 1500, -1]);  permute_217 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2022, [2])
	        amax_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2022, [2])
	        full_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_129, full_258);  amin_129 = full_258 = None
	        full_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_129, full_259);  amax_129 = full_259 = None
	        sub_5909: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_129, minimum_129);  maximum_129 = None
	        div_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5909, 255.0);  sub_5909 = None
	        clamp_min_387: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_258, 1.1920928955078125e-07);  div_258 = None
	        div_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_129, clamp_min_387);  minimum_129 = None
	        round_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_259);  div_259 = None
	        sub_5915: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_259);  round_259 = None
	        clamp_min_388: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5915, -128);  sub_5915 = None
	        clamp_max_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_388, 127);  clamp_min_388 = None
	        _assert_tensor_metadata_1163 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_387, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1163 = None
	        _assert_tensor_metadata_1164 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_258, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1164 = None
	        convert_element_type_774: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_258, torch.int8);  clamp_max_258 = None
	        view_2025: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_387, [sym_size_int, 1500, 1])
	        view_2026: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_774, [sym_size_int, 1500, 1])
	        reciprocal_129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2025);  view_2025 = None
	        mul_12549: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_129, 1.0);  reciprocal_129 = None
	        mul_12552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2022, mul_12549);  view_2022 = mul_12549 = None
	        round_260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12552);  mul_12552 = None
	        add_19861: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_260, view_2026);  round_260 = view_2026 = None
	        clamp_min_389: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19861, -128);  add_19861 = None
	        clamp_max_259: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_389, 127);  clamp_min_389 = None
	        _assert_tensor_metadata_1165 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_259, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1165 = None
	        convert_element_type_775: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_259, torch.int8);  clamp_max_259 = None
	        view_2029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_387, [sym_size_int, 1500, 1]);  clamp_min_387 = None
	        view_2030: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_774, [sym_size_int, 1500, 1]);  convert_element_type_774 = None
	        _assert_tensor_metadata_1166 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_775, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1166 = None
	        convert_element_type_776: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_775, torch.float32);  convert_element_type_775 = None
	        _assert_tensor_metadata_1167 = torch.ops.aten._assert_tensor_metadata.default(view_2030, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1167 = None
	        convert_element_type_777: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2030, torch.float32);  view_2030 = None
	        sub_5935: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_776, convert_element_type_777);  convert_element_type_776 = convert_element_type_777 = None
	        mul_12574: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5935, view_2029);  sub_5935 = view_2029 = None
	        _assert_tensor_metadata_1168 = torch.ops.aten._assert_tensor_metadata.default(mul_12574, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1168 = None
	        view_2032: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2033: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2034: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1169 = torch.ops.aten._assert_tensor_metadata.default(view_2032, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1169 = None
	        convert_element_type_778: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2032, torch.float32);  view_2032 = None
	        _assert_tensor_metadata_1170 = torch.ops.aten._assert_tensor_metadata.default(view_2034, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1170 = None
	        convert_element_type_779: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2034, torch.float32);  view_2034 = None
	        sub_5939: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_778, convert_element_type_779);  convert_element_type_778 = convert_element_type_779 = None
	        mul_12579: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5939, view_2033);  sub_5939 = view_2033 = None
	        view_2035: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12579, [1280, 1280]);  mul_12579 = None
	        _assert_tensor_metadata_1171 = torch.ops.aten._assert_tensor_metadata.default(view_2035, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1171 = None
	        mul_12584: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2036: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12574, [mul_12584, 1280]);  mul_12574 = mul_12584 = None
	        permute_218: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2035, [1, 0]);  view_2035 = None
	        addmm_107: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_21_self_attn_out_proj_bias, view_2036, permute_218);  model_audio_tower_layers_21_self_attn_out_proj_bias = view_2036 = permute_218 = None
	        view_2037: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_107, [sym_size_int, 1500, 1280]);  addmm_107 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_19924: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_19304, view_2037);  add_19304 = view_2037 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_19924, memory_format = torch.contiguous_format)
	        var_mean_43 = torch.ops.aten.var_mean.correction(clone_174, [2], correction = 0, keepdim = True)
	        getitem_174: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_43[0]
	        getitem_175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_43[1];  var_mean_43 = None
	        add_19929: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_174, 1e-05);  getitem_174 = None
	        rsqrt_43: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_19929);  add_19929 = None
	        sub_5945: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_174, getitem_175);  clone_174 = getitem_175 = None
	        mul_12595: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5945, rsqrt_43);  sub_5945 = rsqrt_43 = None
	        mul_12596: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12595, model_audio_tower_layers_21_final_layer_norm_weight);  mul_12595 = model_audio_tower_layers_21_final_layer_norm_weight = None
	        add_19930: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12596, model_audio_tower_layers_21_final_layer_norm_bias);  mul_12596 = model_audio_tower_layers_21_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_19930, [2])
	        amax_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_19930, [2])
	        full_260: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_130, full_260);  amin_130 = full_260 = None
	        full_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_130, full_261);  amax_130 = full_261 = None
	        sub_5956: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_130, minimum_130);  maximum_130 = None
	        div_260: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5956, 255.0);  sub_5956 = None
	        clamp_min_390: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_260, 1.1920928955078125e-07);  div_260 = None
	        div_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_130, clamp_min_390);  minimum_130 = None
	        round_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_261);  div_261 = None
	        sub_5962: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_261);  round_261 = None
	        clamp_min_391: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5962, -128);  sub_5962 = None
	        clamp_max_260: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_391, 127);  clamp_min_391 = None
	        _assert_tensor_metadata_1172 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_390, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1172 = None
	        _assert_tensor_metadata_1173 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_260, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1173 = None
	        convert_element_type_780: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_260, torch.int8);  clamp_max_260 = None
	        view_2040: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_390, [sym_size_int, 1500, 1])
	        view_2041: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_780, [sym_size_int, 1500, 1])
	        reciprocal_130: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2040);  view_2040 = None
	        mul_12644: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_130, 1.0);  reciprocal_130 = None
	        mul_12647: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_19930, mul_12644);  add_19930 = mul_12644 = None
	        round_262: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12647);  mul_12647 = None
	        add_20017: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_262, view_2041);  round_262 = view_2041 = None
	        clamp_min_392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20017, -128);  add_20017 = None
	        clamp_max_261: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_392, 127);  clamp_min_392 = None
	        _assert_tensor_metadata_1174 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_261, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1174 = None
	        convert_element_type_781: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_261, torch.int8);  clamp_max_261 = None
	        view_2044: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_390, [sym_size_int, 1500, 1]);  clamp_min_390 = None
	        view_2045: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_780, [sym_size_int, 1500, 1]);  convert_element_type_780 = None
	        _assert_tensor_metadata_1175 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_781, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1175 = None
	        convert_element_type_782: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_781, torch.float32);  convert_element_type_781 = None
	        _assert_tensor_metadata_1176 = torch.ops.aten._assert_tensor_metadata.default(view_2045, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1176 = None
	        convert_element_type_783: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2045, torch.float32);  view_2045 = None
	        sub_5982: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_782, convert_element_type_783);  convert_element_type_782 = convert_element_type_783 = None
	        mul_12669: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5982, view_2044);  sub_5982 = view_2044 = None
	        _assert_tensor_metadata_1177 = torch.ops.aten._assert_tensor_metadata.default(mul_12669, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1177 = None
	        view_2047: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_21_fc1_parametrizations_weight_original0 = None
	        view_2048: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_21_fc1_parametrizations_weight_original1 = None
	        view_2049: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_21_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1178 = torch.ops.aten._assert_tensor_metadata.default(view_2047, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1178 = None
	        convert_element_type_784: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2047, torch.float32);  view_2047 = None
	        _assert_tensor_metadata_1179 = torch.ops.aten._assert_tensor_metadata.default(view_2049, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1179 = None
	        convert_element_type_785: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2049, torch.float32);  view_2049 = None
	        sub_5986: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_784, convert_element_type_785);  convert_element_type_784 = convert_element_type_785 = None
	        mul_12674: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5986, view_2048);  sub_5986 = view_2048 = None
	        view_2050: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12674, [5120, 1280]);  mul_12674 = None
	        _assert_tensor_metadata_1180 = torch.ops.aten._assert_tensor_metadata.default(view_2050, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1180 = None
	        mul_12679: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2051: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12669, [mul_12679, 1280]);  mul_12669 = mul_12679 = None
	        permute_219: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2050, [1, 0]);  view_2050 = None
	        addmm_108: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_21_fc1_bias, view_2051, permute_219);  model_audio_tower_layers_21_fc1_bias = view_2051 = permute_219 = None
	        view_2052: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_108, [sym_size_int, 1500, 5120]);  addmm_108 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_12686: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2052, 0.5)
	        mul_12687: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2052, 0.7071067811865476);  view_2052 = None
	        erf_23: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_12687);  mul_12687 = None
	        add_20076: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_23, 1);  erf_23 = None
	        mul_12688: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12686, add_20076);  mul_12686 = add_20076 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_12688, [2])
	        amax_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_12688, [2])
	        full_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_131, full_262);  amin_131 = full_262 = None
	        full_263: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_131, full_263);  amax_131 = full_263 = None
	        sub_5999: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_131, minimum_131);  maximum_131 = None
	        div_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5999, 255.0);  sub_5999 = None
	        clamp_min_393: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_262, 1.1920928955078125e-07);  div_262 = None
	        div_263: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_131, clamp_min_393);  minimum_131 = None
	        round_263: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_263);  div_263 = None
	        sub_6005: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_263);  round_263 = None
	        clamp_min_394: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6005, -128);  sub_6005 = None
	        clamp_max_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_394, 127);  clamp_min_394 = None
	        _assert_tensor_metadata_1181 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_393, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1181 = None
	        _assert_tensor_metadata_1182 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_262, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1182 = None
	        convert_element_type_786: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_262, torch.int8);  clamp_max_262 = None
	        view_2055: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_393, [sym_size_int, 1500, 1])
	        view_2056: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_786, [sym_size_int, 1500, 1])
	        reciprocal_131: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2055);  view_2055 = None
	        mul_12734: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_131, 1.0);  reciprocal_131 = None
	        mul_12737: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12688, mul_12734);  mul_12688 = mul_12734 = None
	        round_264: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_12737);  mul_12737 = None
	        add_20159: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_264, view_2056);  round_264 = view_2056 = None
	        clamp_min_395: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20159, -128);  add_20159 = None
	        clamp_max_263: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_395, 127);  clamp_min_395 = None
	        _assert_tensor_metadata_1183 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_263, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1183 = None
	        convert_element_type_787: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_263, torch.int8);  clamp_max_263 = None
	        view_2059: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_393, [sym_size_int, 1500, 1]);  clamp_min_393 = None
	        view_2060: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_786, [sym_size_int, 1500, 1]);  convert_element_type_786 = None
	        _assert_tensor_metadata_1184 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_787, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1184 = None
	        convert_element_type_788: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_787, torch.float32);  convert_element_type_787 = None
	        _assert_tensor_metadata_1185 = torch.ops.aten._assert_tensor_metadata.default(view_2060, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1185 = None
	        convert_element_type_789: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2060, torch.float32);  view_2060 = None
	        sub_6025: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_788, convert_element_type_789);  convert_element_type_788 = convert_element_type_789 = None
	        mul_12759: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6025, view_2059);  sub_6025 = view_2059 = None
	        _assert_tensor_metadata_1186 = torch.ops.aten._assert_tensor_metadata.default(mul_12759, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1186 = None
	        view_2062: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_21_fc2_parametrizations_weight_original0 = None
	        view_2063: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_21_fc2_parametrizations_weight_original1 = None
	        view_2064: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_21_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_21_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1187 = torch.ops.aten._assert_tensor_metadata.default(view_2062, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1187 = None
	        convert_element_type_790: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2062, torch.float32);  view_2062 = None
	        _assert_tensor_metadata_1188 = torch.ops.aten._assert_tensor_metadata.default(view_2064, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1188 = None
	        convert_element_type_791: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2064, torch.float32);  view_2064 = None
	        sub_6029: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_790, convert_element_type_791);  convert_element_type_790 = convert_element_type_791 = None
	        mul_12764: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6029, view_2063);  sub_6029 = view_2063 = None
	        view_2065: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_12764, [1280, 5120]);  mul_12764 = None
	        _assert_tensor_metadata_1189 = torch.ops.aten._assert_tensor_metadata.default(view_2065, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1189 = None
	        mul_12769: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2066: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_12759, [mul_12769, 5120]);  mul_12759 = mul_12769 = None
	        permute_220: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2065, [1, 0]);  view_2065 = None
	        addmm_109: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_21_fc2_bias, view_2066, permute_220);  model_audio_tower_layers_21_fc2_bias = view_2066 = permute_220 = None
	        view_2067: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_109, [sym_size_int, 1500, 1280]);  addmm_109 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_20222: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_19924, view_2067);  add_19924 = view_2067 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_177: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_20222, memory_format = torch.contiguous_format)
	        var_mean_44 = torch.ops.aten.var_mean.correction(clone_177, [2], correction = 0, keepdim = True)
	        getitem_176: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_44[0]
	        getitem_177: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_44[1];  var_mean_44 = None
	        add_20227: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_176, 1e-05);  getitem_176 = None
	        rsqrt_44: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_20227);  add_20227 = None
	        sub_6035: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_177, getitem_177);  clone_177 = getitem_177 = None
	        mul_12780: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6035, rsqrt_44);  sub_6035 = rsqrt_44 = None
	        mul_12781: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12780, model_audio_tower_layers_22_self_attn_layer_norm_weight);  mul_12780 = model_audio_tower_layers_22_self_attn_layer_norm_weight = None
	        add_20228: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12781, model_audio_tower_layers_22_self_attn_layer_norm_bias);  mul_12781 = model_audio_tower_layers_22_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_20228, [2])
	        amax_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_20228, [2])
	        full_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_132, full_264);  amin_132 = full_264 = None
	        full_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_132, full_265);  amax_132 = full_265 = None
	        sub_6046: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_132, minimum_132);  maximum_132 = None
	        div_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6046, 255.0);  sub_6046 = None
	        clamp_min_396: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_264, 1.1920928955078125e-07);  div_264 = None
	        div_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_132, clamp_min_396);  minimum_132 = None
	        round_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_265);  div_265 = None
	        sub_6052: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_265);  round_265 = None
	        clamp_min_397: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6052, -128);  sub_6052 = None
	        clamp_max_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_397, 127);  clamp_min_397 = None
	        _assert_tensor_metadata_1190 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_396, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1190 = None
	        _assert_tensor_metadata_1191 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_264, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1191 = None
	        convert_element_type_792: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_264, torch.int8);  clamp_max_264 = None
	        view_2070: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_396, [sym_size_int, 1500, 1])
	        view_2071: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_792, [sym_size_int, 1500, 1])
	        reciprocal_132: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2070);  view_2070 = None
	        mul_12829: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_132, 1.0);  reciprocal_132 = None
	        mul_12832: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_20228, mul_12829);  mul_12829 = None
	        round_266: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12832);  mul_12832 = None
	        add_20315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_266, view_2071);  round_266 = view_2071 = None
	        clamp_min_398: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20315, -128);  add_20315 = None
	        clamp_max_265: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_398, 127);  clamp_min_398 = None
	        _assert_tensor_metadata_1192 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_265, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1192 = None
	        convert_element_type_793: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_265, torch.int8);  clamp_max_265 = None
	        view_2074: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_396, [sym_size_int, 1500, 1]);  clamp_min_396 = None
	        view_2075: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_792, [sym_size_int, 1500, 1]);  convert_element_type_792 = None
	        _assert_tensor_metadata_1193 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_793, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1193 = None
	        convert_element_type_794: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_793, torch.float32);  convert_element_type_793 = None
	        _assert_tensor_metadata_1194 = torch.ops.aten._assert_tensor_metadata.default(view_2075, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1194 = None
	        convert_element_type_795: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2075, torch.float32);  view_2075 = None
	        sub_6072: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_794, convert_element_type_795);  convert_element_type_794 = convert_element_type_795 = None
	        mul_12854: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6072, view_2074);  sub_6072 = view_2074 = None
	        _assert_tensor_metadata_1195 = torch.ops.aten._assert_tensor_metadata.default(mul_12854, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1195 = None
	        view_2077: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2078: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2079: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1196 = torch.ops.aten._assert_tensor_metadata.default(view_2077, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1196 = None
	        convert_element_type_796: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2077, torch.float32);  view_2077 = None
	        _assert_tensor_metadata_1197 = torch.ops.aten._assert_tensor_metadata.default(view_2079, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1197 = None
	        convert_element_type_797: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2079, torch.float32);  view_2079 = None
	        sub_6076: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_796, convert_element_type_797);  convert_element_type_796 = convert_element_type_797 = None
	        mul_12859: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6076, view_2078);  sub_6076 = view_2078 = None
	        view_2080: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12859, [1280, 1280]);  mul_12859 = None
	        _assert_tensor_metadata_1198 = torch.ops.aten._assert_tensor_metadata.default(view_2080, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1198 = None
	        mul_12864: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2081: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12854, [mul_12864, 1280]);  mul_12854 = mul_12864 = None
	        permute_221: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2080, [1, 0]);  view_2080 = None
	        addmm_110: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_22_self_attn_q_proj_bias, view_2081, permute_221);  model_audio_tower_layers_22_self_attn_q_proj_bias = view_2081 = permute_221 = None
	        view_2082: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_110, [sym_size_int, 1500, 1280]);  addmm_110 = None
	        mul_12871: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2082, 0.125);  view_2082 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2083: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_12871, [sym_size_int, 1500, 20, 64]);  mul_12871 = None
	        permute_222: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2083, [0, 2, 1, 3]);  view_2083 = None
	        clone_178: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_222, memory_format = torch.contiguous_format);  permute_222 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_20228, [2])
	        amax_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_20228, [2])
	        full_266: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_133, full_266);  amin_133 = full_266 = None
	        full_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_133, full_267);  amax_133 = full_267 = None
	        sub_6091: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_133, minimum_133);  maximum_133 = None
	        div_266: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6091, 255.0);  sub_6091 = None
	        clamp_min_399: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_266, 1.1920928955078125e-07);  div_266 = None
	        div_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_133, clamp_min_399);  minimum_133 = None
	        round_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_267);  div_267 = None
	        sub_6097: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_267);  round_267 = None
	        clamp_min_400: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6097, -128);  sub_6097 = None
	        clamp_max_266: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_400, 127);  clamp_min_400 = None
	        _assert_tensor_metadata_1199 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_399, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1199 = None
	        _assert_tensor_metadata_1200 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_266, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1200 = None
	        convert_element_type_798: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_266, torch.int8);  clamp_max_266 = None
	        view_2086: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_399, [sym_size_int, 1500, 1])
	        view_2087: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_798, [sym_size_int, 1500, 1])
	        reciprocal_133: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2086);  view_2086 = None
	        mul_12925: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_133, 1.0);  reciprocal_133 = None
	        mul_12928: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_20228, mul_12925);  mul_12925 = None
	        round_268: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12928);  mul_12928 = None
	        add_20467: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_268, view_2087);  round_268 = view_2087 = None
	        clamp_min_401: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20467, -128);  add_20467 = None
	        clamp_max_267: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_401, 127);  clamp_min_401 = None
	        _assert_tensor_metadata_1201 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_267, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1201 = None
	        convert_element_type_799: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_267, torch.int8);  clamp_max_267 = None
	        view_2090: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_399, [sym_size_int, 1500, 1]);  clamp_min_399 = None
	        view_2091: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_798, [sym_size_int, 1500, 1]);  convert_element_type_798 = None
	        _assert_tensor_metadata_1202 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_799, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1202 = None
	        convert_element_type_800: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_799, torch.float32);  convert_element_type_799 = None
	        _assert_tensor_metadata_1203 = torch.ops.aten._assert_tensor_metadata.default(view_2091, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1203 = None
	        convert_element_type_801: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2091, torch.float32);  view_2091 = None
	        sub_6117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_800, convert_element_type_801);  convert_element_type_800 = convert_element_type_801 = None
	        mul_12950: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6117, view_2090);  sub_6117 = view_2090 = None
	        _assert_tensor_metadata_1204 = torch.ops.aten._assert_tensor_metadata.default(mul_12950, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1204 = None
	        view_2093: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2094: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2095: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1205 = torch.ops.aten._assert_tensor_metadata.default(view_2093, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1205 = None
	        convert_element_type_802: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2093, torch.float32);  view_2093 = None
	        _assert_tensor_metadata_1206 = torch.ops.aten._assert_tensor_metadata.default(view_2095, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1206 = None
	        convert_element_type_803: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2095, torch.float32);  view_2095 = None
	        sub_6121: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_802, convert_element_type_803);  convert_element_type_802 = convert_element_type_803 = None
	        mul_12955: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6121, view_2094);  sub_6121 = view_2094 = None
	        view_2096: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12955, [1280, 1280]);  mul_12955 = None
	        _assert_tensor_metadata_1207 = torch.ops.aten._assert_tensor_metadata.default(view_2096, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1207 = None
	        permute_223: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2096, [1, 0]);  view_2096 = None
	        mul_12958: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2097: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_12950, [mul_12958, 1280]);  mul_12950 = mul_12958 = None
	        mm_22: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2097, permute_223);  view_2097 = permute_223 = None
	        view_2098: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_22, [sym_size_int, 1500, 1280]);  mm_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2099: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2098, [sym_size_int, -1, 20, 64]);  view_2098 = None
	        permute_224: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2099, [0, 2, 1, 3]);  view_2099 = None
	        clone_179: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_224, memory_format = torch.contiguous_format);  permute_224 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_20228, [2])
	        amax_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_20228, [2])
	        full_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_134, full_268);  amin_134 = full_268 = None
	        full_269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_134, full_269);  amax_134 = full_269 = None
	        sub_6135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_134, minimum_134);  maximum_134 = None
	        div_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6135, 255.0);  sub_6135 = None
	        clamp_min_402: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_268, 1.1920928955078125e-07);  div_268 = None
	        div_269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_134, clamp_min_402);  minimum_134 = None
	        round_269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_269);  div_269 = None
	        sub_6141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_269);  round_269 = None
	        clamp_min_403: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6141, -128);  sub_6141 = None
	        clamp_max_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_403, 127);  clamp_min_403 = None
	        _assert_tensor_metadata_1208 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_402, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1208 = None
	        _assert_tensor_metadata_1209 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_268, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1209 = None
	        convert_element_type_804: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_268, torch.int8);  clamp_max_268 = None
	        view_2102: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_402, [sym_size_int, 1500, 1])
	        view_2103: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_804, [sym_size_int, 1500, 1])
	        reciprocal_134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2102);  view_2102 = None
	        mul_13024: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_134, 1.0);  reciprocal_134 = None
	        mul_13027: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_20228, mul_13024);  add_20228 = mul_13024 = None
	        round_270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13027);  mul_13027 = None
	        add_20615: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_270, view_2103);  round_270 = view_2103 = None
	        clamp_min_404: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20615, -128);  add_20615 = None
	        clamp_max_269: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_404, 127);  clamp_min_404 = None
	        _assert_tensor_metadata_1210 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_269, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1210 = None
	        convert_element_type_805: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_269, torch.int8);  clamp_max_269 = None
	        view_2106: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_402, [sym_size_int, 1500, 1]);  clamp_min_402 = None
	        view_2107: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_804, [sym_size_int, 1500, 1]);  convert_element_type_804 = None
	        _assert_tensor_metadata_1211 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_805, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1211 = None
	        convert_element_type_806: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_805, torch.float32);  convert_element_type_805 = None
	        _assert_tensor_metadata_1212 = torch.ops.aten._assert_tensor_metadata.default(view_2107, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1212 = None
	        convert_element_type_807: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2107, torch.float32);  view_2107 = None
	        sub_6161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_806, convert_element_type_807);  convert_element_type_806 = convert_element_type_807 = None
	        mul_13049: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6161, view_2106);  sub_6161 = view_2106 = None
	        _assert_tensor_metadata_1213 = torch.ops.aten._assert_tensor_metadata.default(mul_13049, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1213 = None
	        view_2109: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2110: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2111: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1214 = torch.ops.aten._assert_tensor_metadata.default(view_2109, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1214 = None
	        convert_element_type_808: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2109, torch.float32);  view_2109 = None
	        _assert_tensor_metadata_1215 = torch.ops.aten._assert_tensor_metadata.default(view_2111, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1215 = None
	        convert_element_type_809: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2111, torch.float32);  view_2111 = None
	        sub_6165: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_808, convert_element_type_809);  convert_element_type_808 = convert_element_type_809 = None
	        mul_13054: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6165, view_2110);  sub_6165 = view_2110 = None
	        view_2112: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13054, [1280, 1280]);  mul_13054 = None
	        _assert_tensor_metadata_1216 = torch.ops.aten._assert_tensor_metadata.default(view_2112, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1216 = None
	        mul_13059: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2113: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13049, [mul_13059, 1280]);  mul_13049 = mul_13059 = None
	        permute_225: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2112, [1, 0]);  view_2112 = None
	        addmm_111: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_22_self_attn_v_proj_bias, view_2113, permute_225);  model_audio_tower_layers_22_self_attn_v_proj_bias = view_2113 = permute_225 = None
	        view_2114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_111, [sym_size_int, 1500, 1280]);  addmm_111 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2115: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2114, [sym_size_int, -1, 20, 64]);  view_2114 = None
	        permute_226: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2115, [0, 2, 1, 3]);  view_2115 = None
	        clone_180: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_226, memory_format = torch.contiguous_format);  permute_226 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_22 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_178, clone_179, clone_180, None, False, scale = 1.0);  clone_178 = clone_179 = clone_180 = None
	        getitem_178: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_22[0];  _scaled_dot_product_efficient_attention_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_227: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_178, [0, 2, 1, 3]);  getitem_178 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_227, [sym_size_int, 1500, -1]);  permute_227 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2116, [2])
	        amax_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2116, [2])
	        full_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_135, full_270);  amin_135 = full_270 = None
	        full_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_135, full_271);  amax_135 = full_271 = None
	        sub_6183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_135, minimum_135);  maximum_135 = None
	        div_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6183, 255.0);  sub_6183 = None
	        clamp_min_405: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_270, 1.1920928955078125e-07);  div_270 = None
	        div_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_135, clamp_min_405);  minimum_135 = None
	        round_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_271);  div_271 = None
	        sub_6189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_271);  round_271 = None
	        clamp_min_406: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6189, -128);  sub_6189 = None
	        clamp_max_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_406, 127);  clamp_min_406 = None
	        _assert_tensor_metadata_1217 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_405, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1217 = None
	        _assert_tensor_metadata_1218 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_270, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1218 = None
	        convert_element_type_810: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_270, torch.int8);  clamp_max_270 = None
	        view_2119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_405, [sym_size_int, 1500, 1])
	        view_2120: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_810, [sym_size_int, 1500, 1])
	        reciprocal_135: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2119);  view_2119 = None
	        mul_13129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_135, 1.0);  reciprocal_135 = None
	        mul_13132: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2116, mul_13129);  view_2116 = mul_13129 = None
	        round_272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13132);  mul_13132 = None
	        add_20779: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_272, view_2120);  round_272 = view_2120 = None
	        clamp_min_407: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20779, -128);  add_20779 = None
	        clamp_max_271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_407, 127);  clamp_min_407 = None
	        _assert_tensor_metadata_1219 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_271, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1219 = None
	        convert_element_type_811: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_271, torch.int8);  clamp_max_271 = None
	        view_2123: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_405, [sym_size_int, 1500, 1]);  clamp_min_405 = None
	        view_2124: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_810, [sym_size_int, 1500, 1]);  convert_element_type_810 = None
	        _assert_tensor_metadata_1220 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_811, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1220 = None
	        convert_element_type_812: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_811, torch.float32);  convert_element_type_811 = None
	        _assert_tensor_metadata_1221 = torch.ops.aten._assert_tensor_metadata.default(view_2124, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1221 = None
	        convert_element_type_813: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2124, torch.float32);  view_2124 = None
	        sub_6209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_812, convert_element_type_813);  convert_element_type_812 = convert_element_type_813 = None
	        mul_13154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6209, view_2123);  sub_6209 = view_2123 = None
	        _assert_tensor_metadata_1222 = torch.ops.aten._assert_tensor_metadata.default(mul_13154, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1222 = None
	        view_2126: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2127: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2128: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1223 = torch.ops.aten._assert_tensor_metadata.default(view_2126, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1223 = None
	        convert_element_type_814: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2126, torch.float32);  view_2126 = None
	        _assert_tensor_metadata_1224 = torch.ops.aten._assert_tensor_metadata.default(view_2128, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1224 = None
	        convert_element_type_815: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2128, torch.float32);  view_2128 = None
	        sub_6213: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_814, convert_element_type_815);  convert_element_type_814 = convert_element_type_815 = None
	        mul_13159: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6213, view_2127);  sub_6213 = view_2127 = None
	        view_2129: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13159, [1280, 1280]);  mul_13159 = None
	        _assert_tensor_metadata_1225 = torch.ops.aten._assert_tensor_metadata.default(view_2129, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1225 = None
	        mul_13164: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2130: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13154, [mul_13164, 1280]);  mul_13154 = mul_13164 = None
	        permute_228: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2129, [1, 0]);  view_2129 = None
	        addmm_112: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_22_self_attn_out_proj_bias, view_2130, permute_228);  model_audio_tower_layers_22_self_attn_out_proj_bias = view_2130 = permute_228 = None
	        view_2131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_112, [sym_size_int, 1500, 1280]);  addmm_112 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_20842: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_20222, view_2131);  add_20222 = view_2131 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_20842, memory_format = torch.contiguous_format)
	        var_mean_45 = torch.ops.aten.var_mean.correction(clone_182, [2], correction = 0, keepdim = True)
	        getitem_182: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_45[0]
	        getitem_183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_45[1];  var_mean_45 = None
	        add_20847: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_182, 1e-05);  getitem_182 = None
	        rsqrt_45: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_20847);  add_20847 = None
	        sub_6219: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_182, getitem_183);  clone_182 = getitem_183 = None
	        mul_13175: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6219, rsqrt_45);  sub_6219 = rsqrt_45 = None
	        mul_13176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13175, model_audio_tower_layers_22_final_layer_norm_weight);  mul_13175 = model_audio_tower_layers_22_final_layer_norm_weight = None
	        add_20848: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13176, model_audio_tower_layers_22_final_layer_norm_bias);  mul_13176 = model_audio_tower_layers_22_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_20848, [2])
	        amax_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_20848, [2])
	        full_272: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_136, full_272);  amin_136 = full_272 = None
	        full_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_136, full_273);  amax_136 = full_273 = None
	        sub_6230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_136, minimum_136);  maximum_136 = None
	        div_272: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6230, 255.0);  sub_6230 = None
	        clamp_min_408: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_272, 1.1920928955078125e-07);  div_272 = None
	        div_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_136, clamp_min_408);  minimum_136 = None
	        round_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_273);  div_273 = None
	        sub_6236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_273);  round_273 = None
	        clamp_min_409: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6236, -128);  sub_6236 = None
	        clamp_max_272: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_409, 127);  clamp_min_409 = None
	        _assert_tensor_metadata_1226 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_408, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1226 = None
	        _assert_tensor_metadata_1227 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_272, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1227 = None
	        convert_element_type_816: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_272, torch.int8);  clamp_max_272 = None
	        view_2134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_408, [sym_size_int, 1500, 1])
	        view_2135: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_816, [sym_size_int, 1500, 1])
	        reciprocal_136: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2134);  view_2134 = None
	        mul_13224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_136, 1.0);  reciprocal_136 = None
	        mul_13227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_20848, mul_13224);  add_20848 = mul_13224 = None
	        round_274: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13227);  mul_13227 = None
	        add_20935: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_274, view_2135);  round_274 = view_2135 = None
	        clamp_min_410: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20935, -128);  add_20935 = None
	        clamp_max_273: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_410, 127);  clamp_min_410 = None
	        _assert_tensor_metadata_1228 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_273, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1228 = None
	        convert_element_type_817: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_273, torch.int8);  clamp_max_273 = None
	        view_2138: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_408, [sym_size_int, 1500, 1]);  clamp_min_408 = None
	        view_2139: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_816, [sym_size_int, 1500, 1]);  convert_element_type_816 = None
	        _assert_tensor_metadata_1229 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_817, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1229 = None
	        convert_element_type_818: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_817, torch.float32);  convert_element_type_817 = None
	        _assert_tensor_metadata_1230 = torch.ops.aten._assert_tensor_metadata.default(view_2139, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1230 = None
	        convert_element_type_819: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2139, torch.float32);  view_2139 = None
	        sub_6256: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_818, convert_element_type_819);  convert_element_type_818 = convert_element_type_819 = None
	        mul_13249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6256, view_2138);  sub_6256 = view_2138 = None
	        _assert_tensor_metadata_1231 = torch.ops.aten._assert_tensor_metadata.default(mul_13249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1231 = None
	        view_2141: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_22_fc1_parametrizations_weight_original0 = None
	        view_2142: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_22_fc1_parametrizations_weight_original1 = None
	        view_2143: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_22_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1232 = torch.ops.aten._assert_tensor_metadata.default(view_2141, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1232 = None
	        convert_element_type_820: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2141, torch.float32);  view_2141 = None
	        _assert_tensor_metadata_1233 = torch.ops.aten._assert_tensor_metadata.default(view_2143, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1233 = None
	        convert_element_type_821: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2143, torch.float32);  view_2143 = None
	        sub_6260: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_820, convert_element_type_821);  convert_element_type_820 = convert_element_type_821 = None
	        mul_13254: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6260, view_2142);  sub_6260 = view_2142 = None
	        view_2144: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13254, [5120, 1280]);  mul_13254 = None
	        _assert_tensor_metadata_1234 = torch.ops.aten._assert_tensor_metadata.default(view_2144, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1234 = None
	        mul_13259: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2145: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13249, [mul_13259, 1280]);  mul_13249 = mul_13259 = None
	        permute_229: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2144, [1, 0]);  view_2144 = None
	        addmm_113: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_22_fc1_bias, view_2145, permute_229);  model_audio_tower_layers_22_fc1_bias = view_2145 = permute_229 = None
	        view_2146: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_113, [sym_size_int, 1500, 5120]);  addmm_113 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_13266: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2146, 0.5)
	        mul_13267: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2146, 0.7071067811865476);  view_2146 = None
	        erf_24: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_13267);  mul_13267 = None
	        add_20994: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_24, 1);  erf_24 = None
	        mul_13268: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13266, add_20994);  mul_13266 = add_20994 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_13268, [2])
	        amax_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_13268, [2])
	        full_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_137, full_274);  amin_137 = full_274 = None
	        full_275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_137, full_275);  amax_137 = full_275 = None
	        sub_6273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_137, minimum_137);  maximum_137 = None
	        div_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6273, 255.0);  sub_6273 = None
	        clamp_min_411: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_274, 1.1920928955078125e-07);  div_274 = None
	        div_275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_137, clamp_min_411);  minimum_137 = None
	        round_275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_275);  div_275 = None
	        sub_6279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_275);  round_275 = None
	        clamp_min_412: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6279, -128);  sub_6279 = None
	        clamp_max_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_412, 127);  clamp_min_412 = None
	        _assert_tensor_metadata_1235 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_411, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1235 = None
	        _assert_tensor_metadata_1236 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_274, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1236 = None
	        convert_element_type_822: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_274, torch.int8);  clamp_max_274 = None
	        view_2149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_411, [sym_size_int, 1500, 1])
	        view_2150: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_822, [sym_size_int, 1500, 1])
	        reciprocal_137: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2149);  view_2149 = None
	        mul_13314: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_137, 1.0);  reciprocal_137 = None
	        mul_13317: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13268, mul_13314);  mul_13268 = mul_13314 = None
	        round_276: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_13317);  mul_13317 = None
	        add_21077: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_276, view_2150);  round_276 = view_2150 = None
	        clamp_min_413: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21077, -128);  add_21077 = None
	        clamp_max_275: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_413, 127);  clamp_min_413 = None
	        _assert_tensor_metadata_1237 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_275, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1237 = None
	        convert_element_type_823: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_275, torch.int8);  clamp_max_275 = None
	        view_2153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_411, [sym_size_int, 1500, 1]);  clamp_min_411 = None
	        view_2154: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_822, [sym_size_int, 1500, 1]);  convert_element_type_822 = None
	        _assert_tensor_metadata_1238 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_823, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1238 = None
	        convert_element_type_824: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_823, torch.float32);  convert_element_type_823 = None
	        _assert_tensor_metadata_1239 = torch.ops.aten._assert_tensor_metadata.default(view_2154, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1239 = None
	        convert_element_type_825: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2154, torch.float32);  view_2154 = None
	        sub_6299: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_824, convert_element_type_825);  convert_element_type_824 = convert_element_type_825 = None
	        mul_13339: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6299, view_2153);  sub_6299 = view_2153 = None
	        _assert_tensor_metadata_1240 = torch.ops.aten._assert_tensor_metadata.default(mul_13339, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1240 = None
	        view_2156: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_22_fc2_parametrizations_weight_original0 = None
	        view_2157: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_22_fc2_parametrizations_weight_original1 = None
	        view_2158: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_22_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_22_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1241 = torch.ops.aten._assert_tensor_metadata.default(view_2156, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1241 = None
	        convert_element_type_826: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2156, torch.float32);  view_2156 = None
	        _assert_tensor_metadata_1242 = torch.ops.aten._assert_tensor_metadata.default(view_2158, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1242 = None
	        convert_element_type_827: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2158, torch.float32);  view_2158 = None
	        sub_6303: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_826, convert_element_type_827);  convert_element_type_826 = convert_element_type_827 = None
	        mul_13344: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6303, view_2157);  sub_6303 = view_2157 = None
	        view_2159: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_13344, [1280, 5120]);  mul_13344 = None
	        _assert_tensor_metadata_1243 = torch.ops.aten._assert_tensor_metadata.default(view_2159, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1243 = None
	        mul_13349: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2160: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_13339, [mul_13349, 5120]);  mul_13339 = mul_13349 = None
	        permute_230: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2159, [1, 0]);  view_2159 = None
	        addmm_114: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_22_fc2_bias, view_2160, permute_230);  model_audio_tower_layers_22_fc2_bias = view_2160 = permute_230 = None
	        view_2161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_114, [sym_size_int, 1500, 1280]);  addmm_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_21140: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_20842, view_2161);  add_20842 = view_2161 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_21140, memory_format = torch.contiguous_format)
	        var_mean_46 = torch.ops.aten.var_mean.correction(clone_185, [2], correction = 0, keepdim = True)
	        getitem_184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_46[0]
	        getitem_185: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_46[1];  var_mean_46 = None
	        add_21145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_184, 1e-05);  getitem_184 = None
	        rsqrt_46: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_21145);  add_21145 = None
	        sub_6309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_185, getitem_185);  clone_185 = getitem_185 = None
	        mul_13360: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6309, rsqrt_46);  sub_6309 = rsqrt_46 = None
	        mul_13361: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13360, model_audio_tower_layers_23_self_attn_layer_norm_weight);  mul_13360 = model_audio_tower_layers_23_self_attn_layer_norm_weight = None
	        add_21146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13361, model_audio_tower_layers_23_self_attn_layer_norm_bias);  mul_13361 = model_audio_tower_layers_23_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_21146, [2])
	        amax_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_21146, [2])
	        full_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_138, full_276);  amin_138 = full_276 = None
	        full_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_138, full_277);  amax_138 = full_277 = None
	        sub_6320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_138, minimum_138);  maximum_138 = None
	        div_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6320, 255.0);  sub_6320 = None
	        clamp_min_414: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_276, 1.1920928955078125e-07);  div_276 = None
	        div_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_138, clamp_min_414);  minimum_138 = None
	        round_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_277);  div_277 = None
	        sub_6326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_277);  round_277 = None
	        clamp_min_415: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6326, -128);  sub_6326 = None
	        clamp_max_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_415, 127);  clamp_min_415 = None
	        _assert_tensor_metadata_1244 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_414, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1244 = None
	        _assert_tensor_metadata_1245 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_276, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1245 = None
	        convert_element_type_828: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_276, torch.int8);  clamp_max_276 = None
	        view_2164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_414, [sym_size_int, 1500, 1])
	        view_2165: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_828, [sym_size_int, 1500, 1])
	        reciprocal_138: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2164);  view_2164 = None
	        mul_13409: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_138, 1.0);  reciprocal_138 = None
	        mul_13412: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_21146, mul_13409);  mul_13409 = None
	        round_278: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13412);  mul_13412 = None
	        add_21233: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_278, view_2165);  round_278 = view_2165 = None
	        clamp_min_416: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21233, -128);  add_21233 = None
	        clamp_max_277: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_416, 127);  clamp_min_416 = None
	        _assert_tensor_metadata_1246 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_277, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1246 = None
	        convert_element_type_829: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_277, torch.int8);  clamp_max_277 = None
	        view_2168: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_414, [sym_size_int, 1500, 1]);  clamp_min_414 = None
	        view_2169: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_828, [sym_size_int, 1500, 1]);  convert_element_type_828 = None
	        _assert_tensor_metadata_1247 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_829, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1247 = None
	        convert_element_type_830: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_829, torch.float32);  convert_element_type_829 = None
	        _assert_tensor_metadata_1248 = torch.ops.aten._assert_tensor_metadata.default(view_2169, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1248 = None
	        convert_element_type_831: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2169, torch.float32);  view_2169 = None
	        sub_6346: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_830, convert_element_type_831);  convert_element_type_830 = convert_element_type_831 = None
	        mul_13434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6346, view_2168);  sub_6346 = view_2168 = None
	        _assert_tensor_metadata_1249 = torch.ops.aten._assert_tensor_metadata.default(mul_13434, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1249 = None
	        view_2171: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2172: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2173: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1250 = torch.ops.aten._assert_tensor_metadata.default(view_2171, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1250 = None
	        convert_element_type_832: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2171, torch.float32);  view_2171 = None
	        _assert_tensor_metadata_1251 = torch.ops.aten._assert_tensor_metadata.default(view_2173, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1251 = None
	        convert_element_type_833: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2173, torch.float32);  view_2173 = None
	        sub_6350: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_832, convert_element_type_833);  convert_element_type_832 = convert_element_type_833 = None
	        mul_13439: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6350, view_2172);  sub_6350 = view_2172 = None
	        view_2174: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13439, [1280, 1280]);  mul_13439 = None
	        _assert_tensor_metadata_1252 = torch.ops.aten._assert_tensor_metadata.default(view_2174, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1252 = None
	        mul_13444: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2175: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13434, [mul_13444, 1280]);  mul_13434 = mul_13444 = None
	        permute_231: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2174, [1, 0]);  view_2174 = None
	        addmm_115: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_23_self_attn_q_proj_bias, view_2175, permute_231);  model_audio_tower_layers_23_self_attn_q_proj_bias = view_2175 = permute_231 = None
	        view_2176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_115, [sym_size_int, 1500, 1280]);  addmm_115 = None
	        mul_13451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2176, 0.125);  view_2176 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2177: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_13451, [sym_size_int, 1500, 20, 64]);  mul_13451 = None
	        permute_232: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2177, [0, 2, 1, 3]);  view_2177 = None
	        clone_186: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_232, memory_format = torch.contiguous_format);  permute_232 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_21146, [2])
	        amax_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_21146, [2])
	        full_278: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_139, full_278);  amin_139 = full_278 = None
	        full_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_139, full_279);  amax_139 = full_279 = None
	        sub_6365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_139, minimum_139);  maximum_139 = None
	        div_278: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6365, 255.0);  sub_6365 = None
	        clamp_min_417: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_278, 1.1920928955078125e-07);  div_278 = None
	        div_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_139, clamp_min_417);  minimum_139 = None
	        round_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_279);  div_279 = None
	        sub_6371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_279);  round_279 = None
	        clamp_min_418: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6371, -128);  sub_6371 = None
	        clamp_max_278: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_418, 127);  clamp_min_418 = None
	        _assert_tensor_metadata_1253 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_417, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1253 = None
	        _assert_tensor_metadata_1254 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_278, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1254 = None
	        convert_element_type_834: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_278, torch.int8);  clamp_max_278 = None
	        view_2180: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_417, [sym_size_int, 1500, 1])
	        view_2181: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_834, [sym_size_int, 1500, 1])
	        reciprocal_139: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2180);  view_2180 = None
	        mul_13505: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_139, 1.0);  reciprocal_139 = None
	        mul_13508: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_21146, mul_13505);  mul_13505 = None
	        round_280: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13508);  mul_13508 = None
	        add_21385: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_280, view_2181);  round_280 = view_2181 = None
	        clamp_min_419: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21385, -128);  add_21385 = None
	        clamp_max_279: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_419, 127);  clamp_min_419 = None
	        _assert_tensor_metadata_1255 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_279, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1255 = None
	        convert_element_type_835: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_279, torch.int8);  clamp_max_279 = None
	        view_2184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_417, [sym_size_int, 1500, 1]);  clamp_min_417 = None
	        view_2185: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_834, [sym_size_int, 1500, 1]);  convert_element_type_834 = None
	        _assert_tensor_metadata_1256 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_835, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1256 = None
	        convert_element_type_836: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_835, torch.float32);  convert_element_type_835 = None
	        _assert_tensor_metadata_1257 = torch.ops.aten._assert_tensor_metadata.default(view_2185, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1257 = None
	        convert_element_type_837: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2185, torch.float32);  view_2185 = None
	        sub_6391: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_836, convert_element_type_837);  convert_element_type_836 = convert_element_type_837 = None
	        mul_13530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6391, view_2184);  sub_6391 = view_2184 = None
	        _assert_tensor_metadata_1258 = torch.ops.aten._assert_tensor_metadata.default(mul_13530, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1258 = None
	        view_2187: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2188: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2189: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1259 = torch.ops.aten._assert_tensor_metadata.default(view_2187, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1259 = None
	        convert_element_type_838: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2187, torch.float32);  view_2187 = None
	        _assert_tensor_metadata_1260 = torch.ops.aten._assert_tensor_metadata.default(view_2189, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1260 = None
	        convert_element_type_839: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2189, torch.float32);  view_2189 = None
	        sub_6395: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_838, convert_element_type_839);  convert_element_type_838 = convert_element_type_839 = None
	        mul_13535: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6395, view_2188);  sub_6395 = view_2188 = None
	        view_2190: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13535, [1280, 1280]);  mul_13535 = None
	        _assert_tensor_metadata_1261 = torch.ops.aten._assert_tensor_metadata.default(view_2190, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1261 = None
	        permute_233: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2190, [1, 0]);  view_2190 = None
	        mul_13538: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2191: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13530, [mul_13538, 1280]);  mul_13530 = mul_13538 = None
	        mm_23: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2191, permute_233);  view_2191 = permute_233 = None
	        view_2192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_23, [sym_size_int, 1500, 1280]);  mm_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2193: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2192, [sym_size_int, -1, 20, 64]);  view_2192 = None
	        permute_234: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2193, [0, 2, 1, 3]);  view_2193 = None
	        clone_187: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_234, memory_format = torch.contiguous_format);  permute_234 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_21146, [2])
	        amax_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_21146, [2])
	        full_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_140, full_280);  amin_140 = full_280 = None
	        full_281: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_140, full_281);  amax_140 = full_281 = None
	        sub_6409: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_140, minimum_140);  maximum_140 = None
	        div_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6409, 255.0);  sub_6409 = None
	        clamp_min_420: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_280, 1.1920928955078125e-07);  div_280 = None
	        div_281: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_140, clamp_min_420);  minimum_140 = None
	        round_281: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_281);  div_281 = None
	        sub_6415: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_281);  round_281 = None
	        clamp_min_421: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6415, -128);  sub_6415 = None
	        clamp_max_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_421, 127);  clamp_min_421 = None
	        _assert_tensor_metadata_1262 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_420, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1262 = None
	        _assert_tensor_metadata_1263 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_280, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1263 = None
	        convert_element_type_840: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_280, torch.int8);  clamp_max_280 = None
	        view_2196: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_420, [sym_size_int, 1500, 1])
	        view_2197: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_840, [sym_size_int, 1500, 1])
	        reciprocal_140: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2196);  view_2196 = None
	        mul_13604: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_140, 1.0);  reciprocal_140 = None
	        mul_13607: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_21146, mul_13604);  add_21146 = mul_13604 = None
	        round_282: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13607);  mul_13607 = None
	        add_21533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_282, view_2197);  round_282 = view_2197 = None
	        clamp_min_422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21533, -128);  add_21533 = None
	        clamp_max_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_422, 127);  clamp_min_422 = None
	        _assert_tensor_metadata_1264 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_281, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1264 = None
	        convert_element_type_841: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_281, torch.int8);  clamp_max_281 = None
	        view_2200: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_420, [sym_size_int, 1500, 1]);  clamp_min_420 = None
	        view_2201: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_840, [sym_size_int, 1500, 1]);  convert_element_type_840 = None
	        _assert_tensor_metadata_1265 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_841, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1265 = None
	        convert_element_type_842: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_841, torch.float32);  convert_element_type_841 = None
	        _assert_tensor_metadata_1266 = torch.ops.aten._assert_tensor_metadata.default(view_2201, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1266 = None
	        convert_element_type_843: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2201, torch.float32);  view_2201 = None
	        sub_6435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_842, convert_element_type_843);  convert_element_type_842 = convert_element_type_843 = None
	        mul_13629: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6435, view_2200);  sub_6435 = view_2200 = None
	        _assert_tensor_metadata_1267 = torch.ops.aten._assert_tensor_metadata.default(mul_13629, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1267 = None
	        view_2203: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2204: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2205: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1268 = torch.ops.aten._assert_tensor_metadata.default(view_2203, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1268 = None
	        convert_element_type_844: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2203, torch.float32);  view_2203 = None
	        _assert_tensor_metadata_1269 = torch.ops.aten._assert_tensor_metadata.default(view_2205, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1269 = None
	        convert_element_type_845: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2205, torch.float32);  view_2205 = None
	        sub_6439: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_844, convert_element_type_845);  convert_element_type_844 = convert_element_type_845 = None
	        mul_13634: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6439, view_2204);  sub_6439 = view_2204 = None
	        view_2206: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13634, [1280, 1280]);  mul_13634 = None
	        _assert_tensor_metadata_1270 = torch.ops.aten._assert_tensor_metadata.default(view_2206, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1270 = None
	        mul_13639: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2207: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13629, [mul_13639, 1280]);  mul_13629 = mul_13639 = None
	        permute_235: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2206, [1, 0]);  view_2206 = None
	        addmm_116: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_23_self_attn_v_proj_bias, view_2207, permute_235);  model_audio_tower_layers_23_self_attn_v_proj_bias = view_2207 = permute_235 = None
	        view_2208: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_116, [sym_size_int, 1500, 1280]);  addmm_116 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2209: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2208, [sym_size_int, -1, 20, 64]);  view_2208 = None
	        permute_236: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2209, [0, 2, 1, 3]);  view_2209 = None
	        clone_188: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_236, memory_format = torch.contiguous_format);  permute_236 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_23 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_186, clone_187, clone_188, None, False, scale = 1.0);  clone_186 = clone_187 = clone_188 = None
	        getitem_186: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_23[0];  _scaled_dot_product_efficient_attention_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_237: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_186, [0, 2, 1, 3]);  getitem_186 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2210: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_237, [sym_size_int, 1500, -1]);  permute_237 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2210, [2])
	        amax_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2210, [2])
	        full_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_141, full_282);  amin_141 = full_282 = None
	        full_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_141, full_283);  amax_141 = full_283 = None
	        sub_6457: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_141, minimum_141);  maximum_141 = None
	        div_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6457, 255.0);  sub_6457 = None
	        clamp_min_423: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_282, 1.1920928955078125e-07);  div_282 = None
	        div_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_141, clamp_min_423);  minimum_141 = None
	        round_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_283);  div_283 = None
	        sub_6463: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_283);  round_283 = None
	        clamp_min_424: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6463, -128);  sub_6463 = None
	        clamp_max_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_424, 127);  clamp_min_424 = None
	        _assert_tensor_metadata_1271 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_423, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1271 = None
	        _assert_tensor_metadata_1272 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_282, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1272 = None
	        convert_element_type_846: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_282, torch.int8);  clamp_max_282 = None
	        view_2213: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_423, [sym_size_int, 1500, 1])
	        view_2214: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_846, [sym_size_int, 1500, 1])
	        reciprocal_141: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2213);  view_2213 = None
	        mul_13709: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_141, 1.0);  reciprocal_141 = None
	        mul_13712: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2210, mul_13709);  view_2210 = mul_13709 = None
	        round_284: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13712);  mul_13712 = None
	        add_21697: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_284, view_2214);  round_284 = view_2214 = None
	        clamp_min_425: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21697, -128);  add_21697 = None
	        clamp_max_283: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_425, 127);  clamp_min_425 = None
	        _assert_tensor_metadata_1273 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_283, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1273 = None
	        convert_element_type_847: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_283, torch.int8);  clamp_max_283 = None
	        view_2217: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_423, [sym_size_int, 1500, 1]);  clamp_min_423 = None
	        view_2218: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_846, [sym_size_int, 1500, 1]);  convert_element_type_846 = None
	        _assert_tensor_metadata_1274 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_847, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1274 = None
	        convert_element_type_848: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_847, torch.float32);  convert_element_type_847 = None
	        _assert_tensor_metadata_1275 = torch.ops.aten._assert_tensor_metadata.default(view_2218, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1275 = None
	        convert_element_type_849: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2218, torch.float32);  view_2218 = None
	        sub_6483: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_848, convert_element_type_849);  convert_element_type_848 = convert_element_type_849 = None
	        mul_13734: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6483, view_2217);  sub_6483 = view_2217 = None
	        _assert_tensor_metadata_1276 = torch.ops.aten._assert_tensor_metadata.default(mul_13734, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1276 = None
	        view_2220: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2221: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2222: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1277 = torch.ops.aten._assert_tensor_metadata.default(view_2220, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1277 = None
	        convert_element_type_850: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2220, torch.float32);  view_2220 = None
	        _assert_tensor_metadata_1278 = torch.ops.aten._assert_tensor_metadata.default(view_2222, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1278 = None
	        convert_element_type_851: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2222, torch.float32);  view_2222 = None
	        sub_6487: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_850, convert_element_type_851);  convert_element_type_850 = convert_element_type_851 = None
	        mul_13739: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6487, view_2221);  sub_6487 = view_2221 = None
	        view_2223: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13739, [1280, 1280]);  mul_13739 = None
	        _assert_tensor_metadata_1279 = torch.ops.aten._assert_tensor_metadata.default(view_2223, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1279 = None
	        mul_13744: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2224: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13734, [mul_13744, 1280]);  mul_13734 = mul_13744 = None
	        permute_238: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2223, [1, 0]);  view_2223 = None
	        addmm_117: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_23_self_attn_out_proj_bias, view_2224, permute_238);  model_audio_tower_layers_23_self_attn_out_proj_bias = view_2224 = permute_238 = None
	        view_2225: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_117, [sym_size_int, 1500, 1280]);  addmm_117 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_21760: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_21140, view_2225);  add_21140 = view_2225 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_190: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_21760, memory_format = torch.contiguous_format)
	        var_mean_47 = torch.ops.aten.var_mean.correction(clone_190, [2], correction = 0, keepdim = True)
	        getitem_190: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_47[0]
	        getitem_191: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_47[1];  var_mean_47 = None
	        add_21765: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_190, 1e-05);  getitem_190 = None
	        rsqrt_47: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_21765);  add_21765 = None
	        sub_6493: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_190, getitem_191);  clone_190 = getitem_191 = None
	        mul_13755: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6493, rsqrt_47);  sub_6493 = rsqrt_47 = None
	        mul_13756: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13755, model_audio_tower_layers_23_final_layer_norm_weight);  mul_13755 = model_audio_tower_layers_23_final_layer_norm_weight = None
	        add_21766: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13756, model_audio_tower_layers_23_final_layer_norm_bias);  mul_13756 = model_audio_tower_layers_23_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_21766, [2])
	        amax_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_21766, [2])
	        full_284: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_142, full_284);  amin_142 = full_284 = None
	        full_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_142, full_285);  amax_142 = full_285 = None
	        sub_6504: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_142, minimum_142);  maximum_142 = None
	        div_284: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6504, 255.0);  sub_6504 = None
	        clamp_min_426: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_284, 1.1920928955078125e-07);  div_284 = None
	        div_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_142, clamp_min_426);  minimum_142 = None
	        round_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_285);  div_285 = None
	        sub_6510: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_285);  round_285 = None
	        clamp_min_427: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6510, -128);  sub_6510 = None
	        clamp_max_284: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_427, 127);  clamp_min_427 = None
	        _assert_tensor_metadata_1280 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_426, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1280 = None
	        _assert_tensor_metadata_1281 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_284, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1281 = None
	        convert_element_type_852: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_284, torch.int8);  clamp_max_284 = None
	        view_2228: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_426, [sym_size_int, 1500, 1])
	        view_2229: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_852, [sym_size_int, 1500, 1])
	        reciprocal_142: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2228);  view_2228 = None
	        mul_13804: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_142, 1.0);  reciprocal_142 = None
	        mul_13807: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_21766, mul_13804);  add_21766 = mul_13804 = None
	        round_286: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13807);  mul_13807 = None
	        add_21853: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_286, view_2229);  round_286 = view_2229 = None
	        clamp_min_428: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21853, -128);  add_21853 = None
	        clamp_max_285: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_428, 127);  clamp_min_428 = None
	        _assert_tensor_metadata_1282 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_285, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1282 = None
	        convert_element_type_853: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_285, torch.int8);  clamp_max_285 = None
	        view_2232: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_426, [sym_size_int, 1500, 1]);  clamp_min_426 = None
	        view_2233: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_852, [sym_size_int, 1500, 1]);  convert_element_type_852 = None
	        _assert_tensor_metadata_1283 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_853, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1283 = None
	        convert_element_type_854: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_853, torch.float32);  convert_element_type_853 = None
	        _assert_tensor_metadata_1284 = torch.ops.aten._assert_tensor_metadata.default(view_2233, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1284 = None
	        convert_element_type_855: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2233, torch.float32);  view_2233 = None
	        sub_6530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_854, convert_element_type_855);  convert_element_type_854 = convert_element_type_855 = None
	        mul_13829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6530, view_2232);  sub_6530 = view_2232 = None
	        _assert_tensor_metadata_1285 = torch.ops.aten._assert_tensor_metadata.default(mul_13829, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1285 = None
	        view_2235: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_23_fc1_parametrizations_weight_original0 = None
	        view_2236: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_23_fc1_parametrizations_weight_original1 = None
	        view_2237: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_23_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1286 = torch.ops.aten._assert_tensor_metadata.default(view_2235, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1286 = None
	        convert_element_type_856: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2235, torch.float32);  view_2235 = None
	        _assert_tensor_metadata_1287 = torch.ops.aten._assert_tensor_metadata.default(view_2237, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1287 = None
	        convert_element_type_857: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2237, torch.float32);  view_2237 = None
	        sub_6534: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_856, convert_element_type_857);  convert_element_type_856 = convert_element_type_857 = None
	        mul_13834: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6534, view_2236);  sub_6534 = view_2236 = None
	        view_2238: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13834, [5120, 1280]);  mul_13834 = None
	        _assert_tensor_metadata_1288 = torch.ops.aten._assert_tensor_metadata.default(view_2238, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1288 = None
	        mul_13839: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2239: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_13829, [mul_13839, 1280]);  mul_13829 = mul_13839 = None
	        permute_239: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2238, [1, 0]);  view_2238 = None
	        addmm_118: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_23_fc1_bias, view_2239, permute_239);  model_audio_tower_layers_23_fc1_bias = view_2239 = permute_239 = None
	        view_2240: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_118, [sym_size_int, 1500, 5120]);  addmm_118 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_13846: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2240, 0.5)
	        mul_13847: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2240, 0.7071067811865476);  view_2240 = None
	        erf_25: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_13847);  mul_13847 = None
	        add_21912: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_25, 1);  erf_25 = None
	        mul_13848: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13846, add_21912);  mul_13846 = add_21912 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_13848, [2])
	        amax_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_13848, [2])
	        full_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_143, full_286);  amin_143 = full_286 = None
	        full_287: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_143, full_287);  amax_143 = full_287 = None
	        sub_6547: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_143, minimum_143);  maximum_143 = None
	        div_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6547, 255.0);  sub_6547 = None
	        clamp_min_429: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_286, 1.1920928955078125e-07);  div_286 = None
	        div_287: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_143, clamp_min_429);  minimum_143 = None
	        round_287: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_287);  div_287 = None
	        sub_6553: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_287);  round_287 = None
	        clamp_min_430: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6553, -128);  sub_6553 = None
	        clamp_max_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_430, 127);  clamp_min_430 = None
	        _assert_tensor_metadata_1289 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_429, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1289 = None
	        _assert_tensor_metadata_1290 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_286, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1290 = None
	        convert_element_type_858: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_286, torch.int8);  clamp_max_286 = None
	        view_2243: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_429, [sym_size_int, 1500, 1])
	        view_2244: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_858, [sym_size_int, 1500, 1])
	        reciprocal_143: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2243);  view_2243 = None
	        mul_13894: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_143, 1.0);  reciprocal_143 = None
	        mul_13897: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13848, mul_13894);  mul_13848 = mul_13894 = None
	        round_288: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_13897);  mul_13897 = None
	        add_21995: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_288, view_2244);  round_288 = view_2244 = None
	        clamp_min_431: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21995, -128);  add_21995 = None
	        clamp_max_287: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_431, 127);  clamp_min_431 = None
	        _assert_tensor_metadata_1291 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_287, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1291 = None
	        convert_element_type_859: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_287, torch.int8);  clamp_max_287 = None
	        view_2247: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_429, [sym_size_int, 1500, 1]);  clamp_min_429 = None
	        view_2248: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_858, [sym_size_int, 1500, 1]);  convert_element_type_858 = None
	        _assert_tensor_metadata_1292 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_859, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1292 = None
	        convert_element_type_860: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_859, torch.float32);  convert_element_type_859 = None
	        _assert_tensor_metadata_1293 = torch.ops.aten._assert_tensor_metadata.default(view_2248, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1293 = None
	        convert_element_type_861: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2248, torch.float32);  view_2248 = None
	        sub_6573: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_860, convert_element_type_861);  convert_element_type_860 = convert_element_type_861 = None
	        mul_13919: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6573, view_2247);  sub_6573 = view_2247 = None
	        _assert_tensor_metadata_1294 = torch.ops.aten._assert_tensor_metadata.default(mul_13919, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1294 = None
	        view_2250: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_23_fc2_parametrizations_weight_original0 = None
	        view_2251: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_23_fc2_parametrizations_weight_original1 = None
	        view_2252: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_23_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_23_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1295 = torch.ops.aten._assert_tensor_metadata.default(view_2250, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1295 = None
	        convert_element_type_862: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2250, torch.float32);  view_2250 = None
	        _assert_tensor_metadata_1296 = torch.ops.aten._assert_tensor_metadata.default(view_2252, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1296 = None
	        convert_element_type_863: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2252, torch.float32);  view_2252 = None
	        sub_6577: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_862, convert_element_type_863);  convert_element_type_862 = convert_element_type_863 = None
	        mul_13924: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6577, view_2251);  sub_6577 = view_2251 = None
	        view_2253: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_13924, [1280, 5120]);  mul_13924 = None
	        _assert_tensor_metadata_1297 = torch.ops.aten._assert_tensor_metadata.default(view_2253, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1297 = None
	        mul_13929: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2254: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_13919, [mul_13929, 5120]);  mul_13919 = mul_13929 = None
	        permute_240: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2253, [1, 0]);  view_2253 = None
	        addmm_119: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_23_fc2_bias, view_2254, permute_240);  model_audio_tower_layers_23_fc2_bias = view_2254 = permute_240 = None
	        view_2255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_119, [sym_size_int, 1500, 1280]);  addmm_119 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_22058: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_21760, view_2255);  add_21760 = view_2255 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_193: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_22058, memory_format = torch.contiguous_format)
	        var_mean_48 = torch.ops.aten.var_mean.correction(clone_193, [2], correction = 0, keepdim = True)
	        getitem_192: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_48[0]
	        getitem_193: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_48[1];  var_mean_48 = None
	        add_22063: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_192, 1e-05);  getitem_192 = None
	        rsqrt_48: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_22063);  add_22063 = None
	        sub_6583: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_193, getitem_193);  clone_193 = getitem_193 = None
	        mul_13940: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6583, rsqrt_48);  sub_6583 = rsqrt_48 = None
	        mul_13941: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13940, model_audio_tower_layers_24_self_attn_layer_norm_weight);  mul_13940 = model_audio_tower_layers_24_self_attn_layer_norm_weight = None
	        add_22064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13941, model_audio_tower_layers_24_self_attn_layer_norm_bias);  mul_13941 = model_audio_tower_layers_24_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22064, [2])
	        amax_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22064, [2])
	        full_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_144, full_288);  amin_144 = full_288 = None
	        full_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_144, full_289);  amax_144 = full_289 = None
	        sub_6594: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_144, minimum_144);  maximum_144 = None
	        div_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6594, 255.0);  sub_6594 = None
	        clamp_min_432: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_288, 1.1920928955078125e-07);  div_288 = None
	        div_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_144, clamp_min_432);  minimum_144 = None
	        round_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_289);  div_289 = None
	        sub_6600: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_289);  round_289 = None
	        clamp_min_433: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6600, -128);  sub_6600 = None
	        clamp_max_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_433, 127);  clamp_min_433 = None
	        _assert_tensor_metadata_1298 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_432, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1298 = None
	        _assert_tensor_metadata_1299 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_288, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1299 = None
	        convert_element_type_864: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_288, torch.int8);  clamp_max_288 = None
	        view_2258: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_432, [sym_size_int, 1500, 1])
	        view_2259: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_864, [sym_size_int, 1500, 1])
	        reciprocal_144: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2258);  view_2258 = None
	        mul_13989: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_144, 1.0);  reciprocal_144 = None
	        mul_13992: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22064, mul_13989);  mul_13989 = None
	        round_290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13992);  mul_13992 = None
	        add_22151: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_290, view_2259);  round_290 = view_2259 = None
	        clamp_min_434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22151, -128);  add_22151 = None
	        clamp_max_289: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_434, 127);  clamp_min_434 = None
	        _assert_tensor_metadata_1300 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_289, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1300 = None
	        convert_element_type_865: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_289, torch.int8);  clamp_max_289 = None
	        view_2262: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_432, [sym_size_int, 1500, 1]);  clamp_min_432 = None
	        view_2263: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_864, [sym_size_int, 1500, 1]);  convert_element_type_864 = None
	        _assert_tensor_metadata_1301 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_865, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1301 = None
	        convert_element_type_866: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_865, torch.float32);  convert_element_type_865 = None
	        _assert_tensor_metadata_1302 = torch.ops.aten._assert_tensor_metadata.default(view_2263, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1302 = None
	        convert_element_type_867: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2263, torch.float32);  view_2263 = None
	        sub_6620: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_866, convert_element_type_867);  convert_element_type_866 = convert_element_type_867 = None
	        mul_14014: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6620, view_2262);  sub_6620 = view_2262 = None
	        _assert_tensor_metadata_1303 = torch.ops.aten._assert_tensor_metadata.default(mul_14014, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1303 = None
	        view_2265: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2266: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2267: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1304 = torch.ops.aten._assert_tensor_metadata.default(view_2265, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1304 = None
	        convert_element_type_868: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2265, torch.float32);  view_2265 = None
	        _assert_tensor_metadata_1305 = torch.ops.aten._assert_tensor_metadata.default(view_2267, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1305 = None
	        convert_element_type_869: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2267, torch.float32);  view_2267 = None
	        sub_6624: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_868, convert_element_type_869);  convert_element_type_868 = convert_element_type_869 = None
	        mul_14019: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6624, view_2266);  sub_6624 = view_2266 = None
	        view_2268: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14019, [1280, 1280]);  mul_14019 = None
	        _assert_tensor_metadata_1306 = torch.ops.aten._assert_tensor_metadata.default(view_2268, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1306 = None
	        mul_14024: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2269: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14014, [mul_14024, 1280]);  mul_14014 = mul_14024 = None
	        permute_241: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2268, [1, 0]);  view_2268 = None
	        addmm_120: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_24_self_attn_q_proj_bias, view_2269, permute_241);  model_audio_tower_layers_24_self_attn_q_proj_bias = view_2269 = permute_241 = None
	        view_2270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_120, [sym_size_int, 1500, 1280]);  addmm_120 = None
	        mul_14031: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2270, 0.125);  view_2270 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2271: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_14031, [sym_size_int, 1500, 20, 64]);  mul_14031 = None
	        permute_242: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2271, [0, 2, 1, 3]);  view_2271 = None
	        clone_194: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_242, memory_format = torch.contiguous_format);  permute_242 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22064, [2])
	        amax_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22064, [2])
	        full_290: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_145, full_290);  amin_145 = full_290 = None
	        full_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_145, full_291);  amax_145 = full_291 = None
	        sub_6639: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_145, minimum_145);  maximum_145 = None
	        div_290: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6639, 255.0);  sub_6639 = None
	        clamp_min_435: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_290, 1.1920928955078125e-07);  div_290 = None
	        div_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_145, clamp_min_435);  minimum_145 = None
	        round_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_291);  div_291 = None
	        sub_6645: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_291);  round_291 = None
	        clamp_min_436: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6645, -128);  sub_6645 = None
	        clamp_max_290: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_436, 127);  clamp_min_436 = None
	        _assert_tensor_metadata_1307 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_435, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1307 = None
	        _assert_tensor_metadata_1308 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_290, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1308 = None
	        convert_element_type_870: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_290, torch.int8);  clamp_max_290 = None
	        view_2274: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_435, [sym_size_int, 1500, 1])
	        view_2275: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_870, [sym_size_int, 1500, 1])
	        reciprocal_145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2274);  view_2274 = None
	        mul_14085: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_145, 1.0);  reciprocal_145 = None
	        mul_14088: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22064, mul_14085);  mul_14085 = None
	        round_292: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14088);  mul_14088 = None
	        add_22303: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_292, view_2275);  round_292 = view_2275 = None
	        clamp_min_437: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22303, -128);  add_22303 = None
	        clamp_max_291: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_437, 127);  clamp_min_437 = None
	        _assert_tensor_metadata_1309 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_291, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1309 = None
	        convert_element_type_871: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_291, torch.int8);  clamp_max_291 = None
	        view_2278: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_435, [sym_size_int, 1500, 1]);  clamp_min_435 = None
	        view_2279: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_870, [sym_size_int, 1500, 1]);  convert_element_type_870 = None
	        _assert_tensor_metadata_1310 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_871, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1310 = None
	        convert_element_type_872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_871, torch.float32);  convert_element_type_871 = None
	        _assert_tensor_metadata_1311 = torch.ops.aten._assert_tensor_metadata.default(view_2279, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1311 = None
	        convert_element_type_873: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2279, torch.float32);  view_2279 = None
	        sub_6665: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_872, convert_element_type_873);  convert_element_type_872 = convert_element_type_873 = None
	        mul_14110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6665, view_2278);  sub_6665 = view_2278 = None
	        _assert_tensor_metadata_1312 = torch.ops.aten._assert_tensor_metadata.default(mul_14110, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1312 = None
	        view_2281: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2282: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2283: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1313 = torch.ops.aten._assert_tensor_metadata.default(view_2281, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1313 = None
	        convert_element_type_874: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2281, torch.float32);  view_2281 = None
	        _assert_tensor_metadata_1314 = torch.ops.aten._assert_tensor_metadata.default(view_2283, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1314 = None
	        convert_element_type_875: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2283, torch.float32);  view_2283 = None
	        sub_6669: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_874, convert_element_type_875);  convert_element_type_874 = convert_element_type_875 = None
	        mul_14115: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6669, view_2282);  sub_6669 = view_2282 = None
	        view_2284: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14115, [1280, 1280]);  mul_14115 = None
	        _assert_tensor_metadata_1315 = torch.ops.aten._assert_tensor_metadata.default(view_2284, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1315 = None
	        permute_243: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2284, [1, 0]);  view_2284 = None
	        mul_14118: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2285: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14110, [mul_14118, 1280]);  mul_14110 = mul_14118 = None
	        mm_24: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2285, permute_243);  view_2285 = permute_243 = None
	        view_2286: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_24, [sym_size_int, 1500, 1280]);  mm_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2287: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2286, [sym_size_int, -1, 20, 64]);  view_2286 = None
	        permute_244: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2287, [0, 2, 1, 3]);  view_2287 = None
	        clone_195: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_244, memory_format = torch.contiguous_format);  permute_244 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22064, [2])
	        amax_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22064, [2])
	        full_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_146, full_292);  amin_146 = full_292 = None
	        full_293: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_146, full_293);  amax_146 = full_293 = None
	        sub_6683: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_146, minimum_146);  maximum_146 = None
	        div_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6683, 255.0);  sub_6683 = None
	        clamp_min_438: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_292, 1.1920928955078125e-07);  div_292 = None
	        div_293: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_146, clamp_min_438);  minimum_146 = None
	        round_293: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_293);  div_293 = None
	        sub_6689: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_293);  round_293 = None
	        clamp_min_439: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6689, -128);  sub_6689 = None
	        clamp_max_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_439, 127);  clamp_min_439 = None
	        _assert_tensor_metadata_1316 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_438, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1316 = None
	        _assert_tensor_metadata_1317 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_292, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1317 = None
	        convert_element_type_876: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_292, torch.int8);  clamp_max_292 = None
	        view_2290: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_438, [sym_size_int, 1500, 1])
	        view_2291: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_876, [sym_size_int, 1500, 1])
	        reciprocal_146: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2290);  view_2290 = None
	        mul_14184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_146, 1.0);  reciprocal_146 = None
	        mul_14187: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22064, mul_14184);  add_22064 = mul_14184 = None
	        round_294: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14187);  mul_14187 = None
	        add_22451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_294, view_2291);  round_294 = view_2291 = None
	        clamp_min_440: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22451, -128);  add_22451 = None
	        clamp_max_293: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_440, 127);  clamp_min_440 = None
	        _assert_tensor_metadata_1318 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_293, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1318 = None
	        convert_element_type_877: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_293, torch.int8);  clamp_max_293 = None
	        view_2294: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_438, [sym_size_int, 1500, 1]);  clamp_min_438 = None
	        view_2295: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_876, [sym_size_int, 1500, 1]);  convert_element_type_876 = None
	        _assert_tensor_metadata_1319 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_877, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1319 = None
	        convert_element_type_878: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_877, torch.float32);  convert_element_type_877 = None
	        _assert_tensor_metadata_1320 = torch.ops.aten._assert_tensor_metadata.default(view_2295, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1320 = None
	        convert_element_type_879: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2295, torch.float32);  view_2295 = None
	        sub_6709: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_878, convert_element_type_879);  convert_element_type_878 = convert_element_type_879 = None
	        mul_14209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6709, view_2294);  sub_6709 = view_2294 = None
	        _assert_tensor_metadata_1321 = torch.ops.aten._assert_tensor_metadata.default(mul_14209, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1321 = None
	        view_2297: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2298: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2299: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1322 = torch.ops.aten._assert_tensor_metadata.default(view_2297, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1322 = None
	        convert_element_type_880: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2297, torch.float32);  view_2297 = None
	        _assert_tensor_metadata_1323 = torch.ops.aten._assert_tensor_metadata.default(view_2299, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1323 = None
	        convert_element_type_881: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2299, torch.float32);  view_2299 = None
	        sub_6713: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_880, convert_element_type_881);  convert_element_type_880 = convert_element_type_881 = None
	        mul_14214: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6713, view_2298);  sub_6713 = view_2298 = None
	        view_2300: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14214, [1280, 1280]);  mul_14214 = None
	        _assert_tensor_metadata_1324 = torch.ops.aten._assert_tensor_metadata.default(view_2300, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1324 = None
	        mul_14219: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2301: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14209, [mul_14219, 1280]);  mul_14209 = mul_14219 = None
	        permute_245: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2300, [1, 0]);  view_2300 = None
	        addmm_121: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_24_self_attn_v_proj_bias, view_2301, permute_245);  model_audio_tower_layers_24_self_attn_v_proj_bias = view_2301 = permute_245 = None
	        view_2302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_121, [sym_size_int, 1500, 1280]);  addmm_121 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2303: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2302, [sym_size_int, -1, 20, 64]);  view_2302 = None
	        permute_246: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2303, [0, 2, 1, 3]);  view_2303 = None
	        clone_196: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_246, memory_format = torch.contiguous_format);  permute_246 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_24 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_194, clone_195, clone_196, None, False, scale = 1.0);  clone_194 = clone_195 = clone_196 = None
	        getitem_194: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_24[0];  _scaled_dot_product_efficient_attention_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_247: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_194, [0, 2, 1, 3]);  getitem_194 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2304: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_247, [sym_size_int, 1500, -1]);  permute_247 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2304, [2])
	        amax_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2304, [2])
	        full_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_147, full_294);  amin_147 = full_294 = None
	        full_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_147, full_295);  amax_147 = full_295 = None
	        sub_6731: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_147, minimum_147);  maximum_147 = None
	        div_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6731, 255.0);  sub_6731 = None
	        clamp_min_441: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_294, 1.1920928955078125e-07);  div_294 = None
	        div_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_147, clamp_min_441);  minimum_147 = None
	        round_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_295);  div_295 = None
	        sub_6737: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_295);  round_295 = None
	        clamp_min_442: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6737, -128);  sub_6737 = None
	        clamp_max_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_442, 127);  clamp_min_442 = None
	        _assert_tensor_metadata_1325 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_441, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1325 = None
	        _assert_tensor_metadata_1326 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1326 = None
	        convert_element_type_882: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_294, torch.int8);  clamp_max_294 = None
	        view_2307: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_441, [sym_size_int, 1500, 1])
	        view_2308: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_882, [sym_size_int, 1500, 1])
	        reciprocal_147: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2307);  view_2307 = None
	        mul_14289: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_147, 1.0);  reciprocal_147 = None
	        mul_14292: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2304, mul_14289);  view_2304 = mul_14289 = None
	        round_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14292);  mul_14292 = None
	        add_22615: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_296, view_2308);  round_296 = view_2308 = None
	        clamp_min_443: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22615, -128);  add_22615 = None
	        clamp_max_295: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_443, 127);  clamp_min_443 = None
	        _assert_tensor_metadata_1327 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_295, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1327 = None
	        convert_element_type_883: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_295, torch.int8);  clamp_max_295 = None
	        view_2311: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_441, [sym_size_int, 1500, 1]);  clamp_min_441 = None
	        view_2312: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_882, [sym_size_int, 1500, 1]);  convert_element_type_882 = None
	        _assert_tensor_metadata_1328 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_883, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1328 = None
	        convert_element_type_884: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_883, torch.float32);  convert_element_type_883 = None
	        _assert_tensor_metadata_1329 = torch.ops.aten._assert_tensor_metadata.default(view_2312, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1329 = None
	        convert_element_type_885: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2312, torch.float32);  view_2312 = None
	        sub_6757: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_884, convert_element_type_885);  convert_element_type_884 = convert_element_type_885 = None
	        mul_14314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6757, view_2311);  sub_6757 = view_2311 = None
	        _assert_tensor_metadata_1330 = torch.ops.aten._assert_tensor_metadata.default(mul_14314, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1330 = None
	        view_2314: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2315: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2316: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1331 = torch.ops.aten._assert_tensor_metadata.default(view_2314, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1331 = None
	        convert_element_type_886: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2314, torch.float32);  view_2314 = None
	        _assert_tensor_metadata_1332 = torch.ops.aten._assert_tensor_metadata.default(view_2316, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1332 = None
	        convert_element_type_887: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2316, torch.float32);  view_2316 = None
	        sub_6761: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_886, convert_element_type_887);  convert_element_type_886 = convert_element_type_887 = None
	        mul_14319: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6761, view_2315);  sub_6761 = view_2315 = None
	        view_2317: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14319, [1280, 1280]);  mul_14319 = None
	        _assert_tensor_metadata_1333 = torch.ops.aten._assert_tensor_metadata.default(view_2317, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1333 = None
	        mul_14324: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2318: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14314, [mul_14324, 1280]);  mul_14314 = mul_14324 = None
	        permute_248: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2317, [1, 0]);  view_2317 = None
	        addmm_122: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_24_self_attn_out_proj_bias, view_2318, permute_248);  model_audio_tower_layers_24_self_attn_out_proj_bias = view_2318 = permute_248 = None
	        view_2319: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_122, [sym_size_int, 1500, 1280]);  addmm_122 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_22678: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_22058, view_2319);  add_22058 = view_2319 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_198: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_22678, memory_format = torch.contiguous_format)
	        var_mean_49 = torch.ops.aten.var_mean.correction(clone_198, [2], correction = 0, keepdim = True)
	        getitem_198: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_49[0]
	        getitem_199: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_49[1];  var_mean_49 = None
	        add_22683: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_198, 1e-05);  getitem_198 = None
	        rsqrt_49: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_22683);  add_22683 = None
	        sub_6767: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_198, getitem_199);  clone_198 = getitem_199 = None
	        mul_14335: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6767, rsqrt_49);  sub_6767 = rsqrt_49 = None
	        mul_14336: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14335, model_audio_tower_layers_24_final_layer_norm_weight);  mul_14335 = model_audio_tower_layers_24_final_layer_norm_weight = None
	        add_22684: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_14336, model_audio_tower_layers_24_final_layer_norm_bias);  mul_14336 = model_audio_tower_layers_24_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22684, [2])
	        amax_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22684, [2])
	        full_296: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_148, full_296);  amin_148 = full_296 = None
	        full_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_148, full_297);  amax_148 = full_297 = None
	        sub_6778: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_148, minimum_148);  maximum_148 = None
	        div_296: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6778, 255.0);  sub_6778 = None
	        clamp_min_444: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_296, 1.1920928955078125e-07);  div_296 = None
	        div_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_148, clamp_min_444);  minimum_148 = None
	        round_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_297);  div_297 = None
	        sub_6784: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_297);  round_297 = None
	        clamp_min_445: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6784, -128);  sub_6784 = None
	        clamp_max_296: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_445, 127);  clamp_min_445 = None
	        _assert_tensor_metadata_1334 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_444, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1334 = None
	        _assert_tensor_metadata_1335 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_296, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1335 = None
	        convert_element_type_888: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_296, torch.int8);  clamp_max_296 = None
	        view_2322: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_444, [sym_size_int, 1500, 1])
	        view_2323: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_888, [sym_size_int, 1500, 1])
	        reciprocal_148: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2322);  view_2322 = None
	        mul_14384: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_148, 1.0);  reciprocal_148 = None
	        mul_14387: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22684, mul_14384);  add_22684 = mul_14384 = None
	        round_298: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14387);  mul_14387 = None
	        add_22771: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_298, view_2323);  round_298 = view_2323 = None
	        clamp_min_446: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22771, -128);  add_22771 = None
	        clamp_max_297: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_446, 127);  clamp_min_446 = None
	        _assert_tensor_metadata_1336 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_297, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1336 = None
	        convert_element_type_889: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_297, torch.int8);  clamp_max_297 = None
	        view_2326: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_444, [sym_size_int, 1500, 1]);  clamp_min_444 = None
	        view_2327: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_888, [sym_size_int, 1500, 1]);  convert_element_type_888 = None
	        _assert_tensor_metadata_1337 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_889, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1337 = None
	        convert_element_type_890: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_889, torch.float32);  convert_element_type_889 = None
	        _assert_tensor_metadata_1338 = torch.ops.aten._assert_tensor_metadata.default(view_2327, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1338 = None
	        convert_element_type_891: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2327, torch.float32);  view_2327 = None
	        sub_6804: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_890, convert_element_type_891);  convert_element_type_890 = convert_element_type_891 = None
	        mul_14409: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6804, view_2326);  sub_6804 = view_2326 = None
	        _assert_tensor_metadata_1339 = torch.ops.aten._assert_tensor_metadata.default(mul_14409, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1339 = None
	        view_2329: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_24_fc1_parametrizations_weight_original0 = None
	        view_2330: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_24_fc1_parametrizations_weight_original1 = None
	        view_2331: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_24_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1340 = torch.ops.aten._assert_tensor_metadata.default(view_2329, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1340 = None
	        convert_element_type_892: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2329, torch.float32);  view_2329 = None
	        _assert_tensor_metadata_1341 = torch.ops.aten._assert_tensor_metadata.default(view_2331, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1341 = None
	        convert_element_type_893: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2331, torch.float32);  view_2331 = None
	        sub_6808: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_892, convert_element_type_893);  convert_element_type_892 = convert_element_type_893 = None
	        mul_14414: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6808, view_2330);  sub_6808 = view_2330 = None
	        view_2332: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14414, [5120, 1280]);  mul_14414 = None
	        _assert_tensor_metadata_1342 = torch.ops.aten._assert_tensor_metadata.default(view_2332, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1342 = None
	        mul_14419: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2333: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14409, [mul_14419, 1280]);  mul_14409 = mul_14419 = None
	        permute_249: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2332, [1, 0]);  view_2332 = None
	        addmm_123: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_24_fc1_bias, view_2333, permute_249);  model_audio_tower_layers_24_fc1_bias = view_2333 = permute_249 = None
	        view_2334: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_123, [sym_size_int, 1500, 5120]);  addmm_123 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_14426: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2334, 0.5)
	        mul_14427: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2334, 0.7071067811865476);  view_2334 = None
	        erf_26: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_14427);  mul_14427 = None
	        add_22830: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_26, 1);  erf_26 = None
	        mul_14428: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14426, add_22830);  mul_14426 = add_22830 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_14428, [2])
	        amax_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_14428, [2])
	        full_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_149, full_298);  amin_149 = full_298 = None
	        full_299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_149, full_299);  amax_149 = full_299 = None
	        sub_6821: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_149, minimum_149);  maximum_149 = None
	        div_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6821, 255.0);  sub_6821 = None
	        clamp_min_447: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_298, 1.1920928955078125e-07);  div_298 = None
	        div_299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_149, clamp_min_447);  minimum_149 = None
	        round_299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_299);  div_299 = None
	        sub_6827: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_299);  round_299 = None
	        clamp_min_448: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6827, -128);  sub_6827 = None
	        clamp_max_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_448, 127);  clamp_min_448 = None
	        _assert_tensor_metadata_1343 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_447, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1343 = None
	        _assert_tensor_metadata_1344 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_298, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1344 = None
	        convert_element_type_894: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_298, torch.int8);  clamp_max_298 = None
	        view_2337: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_447, [sym_size_int, 1500, 1])
	        view_2338: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_894, [sym_size_int, 1500, 1])
	        reciprocal_149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2337);  view_2337 = None
	        mul_14474: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_149, 1.0);  reciprocal_149 = None
	        mul_14477: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14428, mul_14474);  mul_14428 = mul_14474 = None
	        round_300: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_14477);  mul_14477 = None
	        add_22913: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_300, view_2338);  round_300 = view_2338 = None
	        clamp_min_449: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22913, -128);  add_22913 = None
	        clamp_max_299: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_449, 127);  clamp_min_449 = None
	        _assert_tensor_metadata_1345 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_299, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1345 = None
	        convert_element_type_895: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_299, torch.int8);  clamp_max_299 = None
	        view_2341: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_447, [sym_size_int, 1500, 1]);  clamp_min_447 = None
	        view_2342: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_894, [sym_size_int, 1500, 1]);  convert_element_type_894 = None
	        _assert_tensor_metadata_1346 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_895, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1346 = None
	        convert_element_type_896: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_895, torch.float32);  convert_element_type_895 = None
	        _assert_tensor_metadata_1347 = torch.ops.aten._assert_tensor_metadata.default(view_2342, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1347 = None
	        convert_element_type_897: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2342, torch.float32);  view_2342 = None
	        sub_6847: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_896, convert_element_type_897);  convert_element_type_896 = convert_element_type_897 = None
	        mul_14499: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6847, view_2341);  sub_6847 = view_2341 = None
	        _assert_tensor_metadata_1348 = torch.ops.aten._assert_tensor_metadata.default(mul_14499, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1348 = None
	        view_2344: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_24_fc2_parametrizations_weight_original0 = None
	        view_2345: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_24_fc2_parametrizations_weight_original1 = None
	        view_2346: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_24_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_24_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1349 = torch.ops.aten._assert_tensor_metadata.default(view_2344, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1349 = None
	        convert_element_type_898: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2344, torch.float32);  view_2344 = None
	        _assert_tensor_metadata_1350 = torch.ops.aten._assert_tensor_metadata.default(view_2346, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1350 = None
	        convert_element_type_899: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2346, torch.float32);  view_2346 = None
	        sub_6851: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_898, convert_element_type_899);  convert_element_type_898 = convert_element_type_899 = None
	        mul_14504: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6851, view_2345);  sub_6851 = view_2345 = None
	        view_2347: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_14504, [1280, 5120]);  mul_14504 = None
	        _assert_tensor_metadata_1351 = torch.ops.aten._assert_tensor_metadata.default(view_2347, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1351 = None
	        mul_14509: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2348: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_14499, [mul_14509, 5120]);  mul_14499 = mul_14509 = None
	        permute_250: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2347, [1, 0]);  view_2347 = None
	        addmm_124: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_24_fc2_bias, view_2348, permute_250);  model_audio_tower_layers_24_fc2_bias = view_2348 = permute_250 = None
	        view_2349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_124, [sym_size_int, 1500, 1280]);  addmm_124 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_22976: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_22678, view_2349);  add_22678 = view_2349 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_201: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_22976, memory_format = torch.contiguous_format)
	        var_mean_50 = torch.ops.aten.var_mean.correction(clone_201, [2], correction = 0, keepdim = True)
	        getitem_200: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_50[0]
	        getitem_201: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_50[1];  var_mean_50 = None
	        add_22981: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_200, 1e-05);  getitem_200 = None
	        rsqrt_50: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_22981);  add_22981 = None
	        sub_6857: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_201, getitem_201);  clone_201 = getitem_201 = None
	        mul_14520: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6857, rsqrt_50);  sub_6857 = rsqrt_50 = None
	        mul_14521: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14520, model_audio_tower_layers_25_self_attn_layer_norm_weight);  mul_14520 = model_audio_tower_layers_25_self_attn_layer_norm_weight = None
	        add_22982: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_14521, model_audio_tower_layers_25_self_attn_layer_norm_bias);  mul_14521 = model_audio_tower_layers_25_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22982, [2])
	        amax_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22982, [2])
	        full_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_150, full_300);  amin_150 = full_300 = None
	        full_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_150, full_301);  amax_150 = full_301 = None
	        sub_6868: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_150, minimum_150);  maximum_150 = None
	        div_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6868, 255.0);  sub_6868 = None
	        clamp_min_450: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_300, 1.1920928955078125e-07);  div_300 = None
	        div_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_150, clamp_min_450);  minimum_150 = None
	        round_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_301);  div_301 = None
	        sub_6874: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_301);  round_301 = None
	        clamp_min_451: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6874, -128);  sub_6874 = None
	        clamp_max_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_451, 127);  clamp_min_451 = None
	        _assert_tensor_metadata_1352 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_450, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1352 = None
	        _assert_tensor_metadata_1353 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_300, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1353 = None
	        convert_element_type_900: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_300, torch.int8);  clamp_max_300 = None
	        view_2352: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_450, [sym_size_int, 1500, 1])
	        view_2353: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_900, [sym_size_int, 1500, 1])
	        reciprocal_150: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2352);  view_2352 = None
	        mul_14569: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_150, 1.0);  reciprocal_150 = None
	        mul_14572: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22982, mul_14569);  mul_14569 = None
	        round_302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14572);  mul_14572 = None
	        add_23069: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_302, view_2353);  round_302 = view_2353 = None
	        clamp_min_452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23069, -128);  add_23069 = None
	        clamp_max_301: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_452, 127);  clamp_min_452 = None
	        _assert_tensor_metadata_1354 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_301, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1354 = None
	        convert_element_type_901: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_301, torch.int8);  clamp_max_301 = None
	        view_2356: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_450, [sym_size_int, 1500, 1]);  clamp_min_450 = None
	        view_2357: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_900, [sym_size_int, 1500, 1]);  convert_element_type_900 = None
	        _assert_tensor_metadata_1355 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_901, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1355 = None
	        convert_element_type_902: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_901, torch.float32);  convert_element_type_901 = None
	        _assert_tensor_metadata_1356 = torch.ops.aten._assert_tensor_metadata.default(view_2357, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1356 = None
	        convert_element_type_903: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2357, torch.float32);  view_2357 = None
	        sub_6894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_902, convert_element_type_903);  convert_element_type_902 = convert_element_type_903 = None
	        mul_14594: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6894, view_2356);  sub_6894 = view_2356 = None
	        _assert_tensor_metadata_1357 = torch.ops.aten._assert_tensor_metadata.default(mul_14594, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1357 = None
	        view_2359: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2360: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2361: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1358 = torch.ops.aten._assert_tensor_metadata.default(view_2359, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1358 = None
	        convert_element_type_904: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2359, torch.float32);  view_2359 = None
	        _assert_tensor_metadata_1359 = torch.ops.aten._assert_tensor_metadata.default(view_2361, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1359 = None
	        convert_element_type_905: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2361, torch.float32);  view_2361 = None
	        sub_6898: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_904, convert_element_type_905);  convert_element_type_904 = convert_element_type_905 = None
	        mul_14599: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6898, view_2360);  sub_6898 = view_2360 = None
	        view_2362: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14599, [1280, 1280]);  mul_14599 = None
	        _assert_tensor_metadata_1360 = torch.ops.aten._assert_tensor_metadata.default(view_2362, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1360 = None
	        mul_14604: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2363: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14594, [mul_14604, 1280]);  mul_14594 = mul_14604 = None
	        permute_251: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2362, [1, 0]);  view_2362 = None
	        addmm_125: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_25_self_attn_q_proj_bias, view_2363, permute_251);  model_audio_tower_layers_25_self_attn_q_proj_bias = view_2363 = permute_251 = None
	        view_2364: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_125, [sym_size_int, 1500, 1280]);  addmm_125 = None
	        mul_14611: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2364, 0.125);  view_2364 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2365: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_14611, [sym_size_int, 1500, 20, 64]);  mul_14611 = None
	        permute_252: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2365, [0, 2, 1, 3]);  view_2365 = None
	        clone_202: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_252, memory_format = torch.contiguous_format);  permute_252 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22982, [2])
	        amax_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22982, [2])
	        full_302: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_151, full_302);  amin_151 = full_302 = None
	        full_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_151, full_303);  amax_151 = full_303 = None
	        sub_6913: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_151, minimum_151);  maximum_151 = None
	        div_302: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6913, 255.0);  sub_6913 = None
	        clamp_min_453: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_302, 1.1920928955078125e-07);  div_302 = None
	        div_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_151, clamp_min_453);  minimum_151 = None
	        round_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_303);  div_303 = None
	        sub_6919: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_303);  round_303 = None
	        clamp_min_454: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6919, -128);  sub_6919 = None
	        clamp_max_302: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_454, 127);  clamp_min_454 = None
	        _assert_tensor_metadata_1361 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_453, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1361 = None
	        _assert_tensor_metadata_1362 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_302, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1362 = None
	        convert_element_type_906: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_302, torch.int8);  clamp_max_302 = None
	        view_2368: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_453, [sym_size_int, 1500, 1])
	        view_2369: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_906, [sym_size_int, 1500, 1])
	        reciprocal_151: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2368);  view_2368 = None
	        mul_14665: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_151, 1.0);  reciprocal_151 = None
	        mul_14668: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22982, mul_14665);  mul_14665 = None
	        round_304: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14668);  mul_14668 = None
	        add_23221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_304, view_2369);  round_304 = view_2369 = None
	        clamp_min_455: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23221, -128);  add_23221 = None
	        clamp_max_303: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_455, 127);  clamp_min_455 = None
	        _assert_tensor_metadata_1363 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_303, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1363 = None
	        convert_element_type_907: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_303, torch.int8);  clamp_max_303 = None
	        view_2372: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_453, [sym_size_int, 1500, 1]);  clamp_min_453 = None
	        view_2373: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_906, [sym_size_int, 1500, 1]);  convert_element_type_906 = None
	        _assert_tensor_metadata_1364 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_907, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1364 = None
	        convert_element_type_908: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_907, torch.float32);  convert_element_type_907 = None
	        _assert_tensor_metadata_1365 = torch.ops.aten._assert_tensor_metadata.default(view_2373, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1365 = None
	        convert_element_type_909: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2373, torch.float32);  view_2373 = None
	        sub_6939: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_908, convert_element_type_909);  convert_element_type_908 = convert_element_type_909 = None
	        mul_14690: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6939, view_2372);  sub_6939 = view_2372 = None
	        _assert_tensor_metadata_1366 = torch.ops.aten._assert_tensor_metadata.default(mul_14690, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1366 = None
	        view_2375: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2376: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2377: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1367 = torch.ops.aten._assert_tensor_metadata.default(view_2375, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1367 = None
	        convert_element_type_910: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2375, torch.float32);  view_2375 = None
	        _assert_tensor_metadata_1368 = torch.ops.aten._assert_tensor_metadata.default(view_2377, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1368 = None
	        convert_element_type_911: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2377, torch.float32);  view_2377 = None
	        sub_6943: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_910, convert_element_type_911);  convert_element_type_910 = convert_element_type_911 = None
	        mul_14695: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6943, view_2376);  sub_6943 = view_2376 = None
	        view_2378: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14695, [1280, 1280]);  mul_14695 = None
	        _assert_tensor_metadata_1369 = torch.ops.aten._assert_tensor_metadata.default(view_2378, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1369 = None
	        permute_253: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2378, [1, 0]);  view_2378 = None
	        mul_14698: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2379: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14690, [mul_14698, 1280]);  mul_14690 = mul_14698 = None
	        mm_25: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2379, permute_253);  view_2379 = permute_253 = None
	        view_2380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_25, [sym_size_int, 1500, 1280]);  mm_25 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2381: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2380, [sym_size_int, -1, 20, 64]);  view_2380 = None
	        permute_254: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2381, [0, 2, 1, 3]);  view_2381 = None
	        clone_203: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_254, memory_format = torch.contiguous_format);  permute_254 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22982, [2])
	        amax_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22982, [2])
	        full_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_152, full_304);  amin_152 = full_304 = None
	        full_305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_152, full_305);  amax_152 = full_305 = None
	        sub_6957: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_152, minimum_152);  maximum_152 = None
	        div_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6957, 255.0);  sub_6957 = None
	        clamp_min_456: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_304, 1.1920928955078125e-07);  div_304 = None
	        div_305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_152, clamp_min_456);  minimum_152 = None
	        round_305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_305);  div_305 = None
	        sub_6963: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_305);  round_305 = None
	        clamp_min_457: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6963, -128);  sub_6963 = None
	        clamp_max_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_457, 127);  clamp_min_457 = None
	        _assert_tensor_metadata_1370 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_456, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1370 = None
	        _assert_tensor_metadata_1371 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_304, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1371 = None
	        convert_element_type_912: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_304, torch.int8);  clamp_max_304 = None
	        view_2384: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_456, [sym_size_int, 1500, 1])
	        view_2385: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_912, [sym_size_int, 1500, 1])
	        reciprocal_152: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2384);  view_2384 = None
	        mul_14764: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_152, 1.0);  reciprocal_152 = None
	        mul_14767: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22982, mul_14764);  add_22982 = mul_14764 = None
	        round_306: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14767);  mul_14767 = None
	        add_23369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_306, view_2385);  round_306 = view_2385 = None
	        clamp_min_458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23369, -128);  add_23369 = None
	        clamp_max_305: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_458, 127);  clamp_min_458 = None
	        _assert_tensor_metadata_1372 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_305, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1372 = None
	        convert_element_type_913: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_305, torch.int8);  clamp_max_305 = None
	        view_2388: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_456, [sym_size_int, 1500, 1]);  clamp_min_456 = None
	        view_2389: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_912, [sym_size_int, 1500, 1]);  convert_element_type_912 = None
	        _assert_tensor_metadata_1373 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_913, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1373 = None
	        convert_element_type_914: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_913, torch.float32);  convert_element_type_913 = None
	        _assert_tensor_metadata_1374 = torch.ops.aten._assert_tensor_metadata.default(view_2389, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1374 = None
	        convert_element_type_915: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2389, torch.float32);  view_2389 = None
	        sub_6983: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_914, convert_element_type_915);  convert_element_type_914 = convert_element_type_915 = None
	        mul_14789: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6983, view_2388);  sub_6983 = view_2388 = None
	        _assert_tensor_metadata_1375 = torch.ops.aten._assert_tensor_metadata.default(mul_14789, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1375 = None
	        view_2391: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2392: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2393: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1376 = torch.ops.aten._assert_tensor_metadata.default(view_2391, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1376 = None
	        convert_element_type_916: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2391, torch.float32);  view_2391 = None
	        _assert_tensor_metadata_1377 = torch.ops.aten._assert_tensor_metadata.default(view_2393, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1377 = None
	        convert_element_type_917: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2393, torch.float32);  view_2393 = None
	        sub_6987: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_916, convert_element_type_917);  convert_element_type_916 = convert_element_type_917 = None
	        mul_14794: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6987, view_2392);  sub_6987 = view_2392 = None
	        view_2394: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14794, [1280, 1280]);  mul_14794 = None
	        _assert_tensor_metadata_1378 = torch.ops.aten._assert_tensor_metadata.default(view_2394, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1378 = None
	        mul_14799: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2395: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14789, [mul_14799, 1280]);  mul_14789 = mul_14799 = None
	        permute_255: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2394, [1, 0]);  view_2394 = None
	        addmm_126: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_25_self_attn_v_proj_bias, view_2395, permute_255);  model_audio_tower_layers_25_self_attn_v_proj_bias = view_2395 = permute_255 = None
	        view_2396: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_126, [sym_size_int, 1500, 1280]);  addmm_126 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2397: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2396, [sym_size_int, -1, 20, 64]);  view_2396 = None
	        permute_256: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2397, [0, 2, 1, 3]);  view_2397 = None
	        clone_204: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_256, memory_format = torch.contiguous_format);  permute_256 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_25 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_202, clone_203, clone_204, None, False, scale = 1.0);  clone_202 = clone_203 = clone_204 = None
	        getitem_202: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_25[0];  _scaled_dot_product_efficient_attention_25 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_257: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_202, [0, 2, 1, 3]);  getitem_202 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2398: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_257, [sym_size_int, 1500, -1]);  permute_257 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2398, [2])
	        amax_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2398, [2])
	        full_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_153, full_306);  amin_153 = full_306 = None
	        full_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_153, full_307);  amax_153 = full_307 = None
	        sub_7005: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_153, minimum_153);  maximum_153 = None
	        div_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7005, 255.0);  sub_7005 = None
	        clamp_min_459: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_306, 1.1920928955078125e-07);  div_306 = None
	        div_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_153, clamp_min_459);  minimum_153 = None
	        round_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_307);  div_307 = None
	        sub_7011: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_307);  round_307 = None
	        clamp_min_460: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7011, -128);  sub_7011 = None
	        clamp_max_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_460, 127);  clamp_min_460 = None
	        _assert_tensor_metadata_1379 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_459, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1379 = None
	        _assert_tensor_metadata_1380 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_306, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1380 = None
	        convert_element_type_918: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_306, torch.int8);  clamp_max_306 = None
	        view_2401: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_459, [sym_size_int, 1500, 1])
	        view_2402: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_918, [sym_size_int, 1500, 1])
	        reciprocal_153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2401);  view_2401 = None
	        mul_14869: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_153, 1.0);  reciprocal_153 = None
	        mul_14872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2398, mul_14869);  view_2398 = mul_14869 = None
	        round_308: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14872);  mul_14872 = None
	        add_23533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_308, view_2402);  round_308 = view_2402 = None
	        clamp_min_461: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23533, -128);  add_23533 = None
	        clamp_max_307: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_461, 127);  clamp_min_461 = None
	        _assert_tensor_metadata_1381 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_307, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1381 = None
	        convert_element_type_919: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_307, torch.int8);  clamp_max_307 = None
	        view_2405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_459, [sym_size_int, 1500, 1]);  clamp_min_459 = None
	        view_2406: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_918, [sym_size_int, 1500, 1]);  convert_element_type_918 = None
	        _assert_tensor_metadata_1382 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_919, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1382 = None
	        convert_element_type_920: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_919, torch.float32);  convert_element_type_919 = None
	        _assert_tensor_metadata_1383 = torch.ops.aten._assert_tensor_metadata.default(view_2406, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1383 = None
	        convert_element_type_921: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2406, torch.float32);  view_2406 = None
	        sub_7031: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_920, convert_element_type_921);  convert_element_type_920 = convert_element_type_921 = None
	        mul_14894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7031, view_2405);  sub_7031 = view_2405 = None
	        _assert_tensor_metadata_1384 = torch.ops.aten._assert_tensor_metadata.default(mul_14894, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1384 = None
	        view_2408: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2409: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2410: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1385 = torch.ops.aten._assert_tensor_metadata.default(view_2408, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1385 = None
	        convert_element_type_922: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2408, torch.float32);  view_2408 = None
	        _assert_tensor_metadata_1386 = torch.ops.aten._assert_tensor_metadata.default(view_2410, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1386 = None
	        convert_element_type_923: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2410, torch.float32);  view_2410 = None
	        sub_7035: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_922, convert_element_type_923);  convert_element_type_922 = convert_element_type_923 = None
	        mul_14899: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7035, view_2409);  sub_7035 = view_2409 = None
	        view_2411: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14899, [1280, 1280]);  mul_14899 = None
	        _assert_tensor_metadata_1387 = torch.ops.aten._assert_tensor_metadata.default(view_2411, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1387 = None
	        mul_14904: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2412: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14894, [mul_14904, 1280]);  mul_14894 = mul_14904 = None
	        permute_258: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2411, [1, 0]);  view_2411 = None
	        addmm_127: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_25_self_attn_out_proj_bias, view_2412, permute_258);  model_audio_tower_layers_25_self_attn_out_proj_bias = view_2412 = permute_258 = None
	        view_2413: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_127, [sym_size_int, 1500, 1280]);  addmm_127 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_23596: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_22976, view_2413);  add_22976 = view_2413 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_23596, memory_format = torch.contiguous_format)
	        var_mean_51 = torch.ops.aten.var_mean.correction(clone_206, [2], correction = 0, keepdim = True)
	        getitem_206: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_51[0]
	        getitem_207: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_51[1];  var_mean_51 = None
	        add_23601: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_206, 1e-05);  getitem_206 = None
	        rsqrt_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_23601);  add_23601 = None
	        sub_7041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_206, getitem_207);  clone_206 = getitem_207 = None
	        mul_14915: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7041, rsqrt_51);  sub_7041 = rsqrt_51 = None
	        mul_14916: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14915, model_audio_tower_layers_25_final_layer_norm_weight);  mul_14915 = model_audio_tower_layers_25_final_layer_norm_weight = None
	        add_23602: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_14916, model_audio_tower_layers_25_final_layer_norm_bias);  mul_14916 = model_audio_tower_layers_25_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_23602, [2])
	        amax_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_23602, [2])
	        full_308: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_154, full_308);  amin_154 = full_308 = None
	        full_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_154, full_309);  amax_154 = full_309 = None
	        sub_7052: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_154, minimum_154);  maximum_154 = None
	        div_308: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7052, 255.0);  sub_7052 = None
	        clamp_min_462: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_308, 1.1920928955078125e-07);  div_308 = None
	        div_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_154, clamp_min_462);  minimum_154 = None
	        round_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_309);  div_309 = None
	        sub_7058: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_309);  round_309 = None
	        clamp_min_463: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7058, -128);  sub_7058 = None
	        clamp_max_308: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_463, 127);  clamp_min_463 = None
	        _assert_tensor_metadata_1388 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_462, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1388 = None
	        _assert_tensor_metadata_1389 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_308, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1389 = None
	        convert_element_type_924: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_308, torch.int8);  clamp_max_308 = None
	        view_2416: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_462, [sym_size_int, 1500, 1])
	        view_2417: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_924, [sym_size_int, 1500, 1])
	        reciprocal_154: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2416);  view_2416 = None
	        mul_14964: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_154, 1.0);  reciprocal_154 = None
	        mul_14967: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_23602, mul_14964);  add_23602 = mul_14964 = None
	        round_310: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14967);  mul_14967 = None
	        add_23689: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_310, view_2417);  round_310 = view_2417 = None
	        clamp_min_464: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23689, -128);  add_23689 = None
	        clamp_max_309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_464, 127);  clamp_min_464 = None
	        _assert_tensor_metadata_1390 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_309, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1390 = None
	        convert_element_type_925: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_309, torch.int8);  clamp_max_309 = None
	        view_2420: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_462, [sym_size_int, 1500, 1]);  clamp_min_462 = None
	        view_2421: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_924, [sym_size_int, 1500, 1]);  convert_element_type_924 = None
	        _assert_tensor_metadata_1391 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_925, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1391 = None
	        convert_element_type_926: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_925, torch.float32);  convert_element_type_925 = None
	        _assert_tensor_metadata_1392 = torch.ops.aten._assert_tensor_metadata.default(view_2421, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1392 = None
	        convert_element_type_927: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2421, torch.float32);  view_2421 = None
	        sub_7078: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_926, convert_element_type_927);  convert_element_type_926 = convert_element_type_927 = None
	        mul_14989: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7078, view_2420);  sub_7078 = view_2420 = None
	        _assert_tensor_metadata_1393 = torch.ops.aten._assert_tensor_metadata.default(mul_14989, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1393 = None
	        view_2423: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_25_fc1_parametrizations_weight_original0 = None
	        view_2424: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_25_fc1_parametrizations_weight_original1 = None
	        view_2425: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_25_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1394 = torch.ops.aten._assert_tensor_metadata.default(view_2423, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1394 = None
	        convert_element_type_928: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2423, torch.float32);  view_2423 = None
	        _assert_tensor_metadata_1395 = torch.ops.aten._assert_tensor_metadata.default(view_2425, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1395 = None
	        convert_element_type_929: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2425, torch.float32);  view_2425 = None
	        sub_7082: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_928, convert_element_type_929);  convert_element_type_928 = convert_element_type_929 = None
	        mul_14994: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7082, view_2424);  sub_7082 = view_2424 = None
	        view_2426: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14994, [5120, 1280]);  mul_14994 = None
	        _assert_tensor_metadata_1396 = torch.ops.aten._assert_tensor_metadata.default(view_2426, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1396 = None
	        mul_14999: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2427: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_14989, [mul_14999, 1280]);  mul_14989 = mul_14999 = None
	        permute_259: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2426, [1, 0]);  view_2426 = None
	        addmm_128: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_25_fc1_bias, view_2427, permute_259);  model_audio_tower_layers_25_fc1_bias = view_2427 = permute_259 = None
	        view_2428: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_128, [sym_size_int, 1500, 5120]);  addmm_128 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_15006: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2428, 0.5)
	        mul_15007: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2428, 0.7071067811865476);  view_2428 = None
	        erf_27: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_15007);  mul_15007 = None
	        add_23748: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_27, 1);  erf_27 = None
	        mul_15008: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15006, add_23748);  mul_15006 = add_23748 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_15008, [2])
	        amax_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_15008, [2])
	        full_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_155, full_310);  amin_155 = full_310 = None
	        full_311: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_155, full_311);  amax_155 = full_311 = None
	        sub_7095: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_155, minimum_155);  maximum_155 = None
	        div_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7095, 255.0);  sub_7095 = None
	        clamp_min_465: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_310, 1.1920928955078125e-07);  div_310 = None
	        div_311: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_155, clamp_min_465);  minimum_155 = None
	        round_311: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_311);  div_311 = None
	        sub_7101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_311);  round_311 = None
	        clamp_min_466: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7101, -128);  sub_7101 = None
	        clamp_max_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_466, 127);  clamp_min_466 = None
	        _assert_tensor_metadata_1397 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_465, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1397 = None
	        _assert_tensor_metadata_1398 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_310, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1398 = None
	        convert_element_type_930: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_310, torch.int8);  clamp_max_310 = None
	        view_2431: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_465, [sym_size_int, 1500, 1])
	        view_2432: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_930, [sym_size_int, 1500, 1])
	        reciprocal_155: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2431);  view_2431 = None
	        mul_15054: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_155, 1.0);  reciprocal_155 = None
	        mul_15057: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15008, mul_15054);  mul_15008 = mul_15054 = None
	        round_312: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_15057);  mul_15057 = None
	        add_23831: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_312, view_2432);  round_312 = view_2432 = None
	        clamp_min_467: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23831, -128);  add_23831 = None
	        clamp_max_311: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_467, 127);  clamp_min_467 = None
	        _assert_tensor_metadata_1399 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_311, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1399 = None
	        convert_element_type_931: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_311, torch.int8);  clamp_max_311 = None
	        view_2435: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_465, [sym_size_int, 1500, 1]);  clamp_min_465 = None
	        view_2436: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_930, [sym_size_int, 1500, 1]);  convert_element_type_930 = None
	        _assert_tensor_metadata_1400 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_931, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1400 = None
	        convert_element_type_932: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_931, torch.float32);  convert_element_type_931 = None
	        _assert_tensor_metadata_1401 = torch.ops.aten._assert_tensor_metadata.default(view_2436, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1401 = None
	        convert_element_type_933: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2436, torch.float32);  view_2436 = None
	        sub_7121: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_932, convert_element_type_933);  convert_element_type_932 = convert_element_type_933 = None
	        mul_15079: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7121, view_2435);  sub_7121 = view_2435 = None
	        _assert_tensor_metadata_1402 = torch.ops.aten._assert_tensor_metadata.default(mul_15079, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1402 = None
	        view_2438: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_25_fc2_parametrizations_weight_original0 = None
	        view_2439: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_25_fc2_parametrizations_weight_original1 = None
	        view_2440: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_25_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_25_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1403 = torch.ops.aten._assert_tensor_metadata.default(view_2438, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1403 = None
	        convert_element_type_934: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2438, torch.float32);  view_2438 = None
	        _assert_tensor_metadata_1404 = torch.ops.aten._assert_tensor_metadata.default(view_2440, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1404 = None
	        convert_element_type_935: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2440, torch.float32);  view_2440 = None
	        sub_7125: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_934, convert_element_type_935);  convert_element_type_934 = convert_element_type_935 = None
	        mul_15084: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7125, view_2439);  sub_7125 = view_2439 = None
	        view_2441: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_15084, [1280, 5120]);  mul_15084 = None
	        _assert_tensor_metadata_1405 = torch.ops.aten._assert_tensor_metadata.default(view_2441, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1405 = None
	        mul_15089: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2442: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_15079, [mul_15089, 5120]);  mul_15079 = mul_15089 = None
	        permute_260: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2441, [1, 0]);  view_2441 = None
	        addmm_129: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_25_fc2_bias, view_2442, permute_260);  model_audio_tower_layers_25_fc2_bias = view_2442 = permute_260 = None
	        view_2443: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_129, [sym_size_int, 1500, 1280]);  addmm_129 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_23894: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_23596, view_2443);  add_23596 = view_2443 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_23894, memory_format = torch.contiguous_format)
	        var_mean_52 = torch.ops.aten.var_mean.correction(clone_209, [2], correction = 0, keepdim = True)
	        getitem_208: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_52[0]
	        getitem_209: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_52[1];  var_mean_52 = None
	        add_23899: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_208, 1e-05);  getitem_208 = None
	        rsqrt_52: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_23899);  add_23899 = None
	        sub_7131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_209, getitem_209);  clone_209 = getitem_209 = None
	        mul_15100: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7131, rsqrt_52);  sub_7131 = rsqrt_52 = None
	        mul_15101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15100, model_audio_tower_layers_26_self_attn_layer_norm_weight);  mul_15100 = model_audio_tower_layers_26_self_attn_layer_norm_weight = None
	        add_23900: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_15101, model_audio_tower_layers_26_self_attn_layer_norm_bias);  mul_15101 = model_audio_tower_layers_26_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_23900, [2])
	        amax_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_23900, [2])
	        full_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_156, full_312);  amin_156 = full_312 = None
	        full_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_156, full_313);  amax_156 = full_313 = None
	        sub_7142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_156, minimum_156);  maximum_156 = None
	        div_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7142, 255.0);  sub_7142 = None
	        clamp_min_468: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_312, 1.1920928955078125e-07);  div_312 = None
	        div_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_156, clamp_min_468);  minimum_156 = None
	        round_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_313);  div_313 = None
	        sub_7148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_313);  round_313 = None
	        clamp_min_469: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7148, -128);  sub_7148 = None
	        clamp_max_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_469, 127);  clamp_min_469 = None
	        _assert_tensor_metadata_1406 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_468, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1406 = None
	        _assert_tensor_metadata_1407 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_312, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1407 = None
	        convert_element_type_936: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_312, torch.int8);  clamp_max_312 = None
	        view_2446: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_468, [sym_size_int, 1500, 1])
	        view_2447: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_936, [sym_size_int, 1500, 1])
	        reciprocal_156: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2446);  view_2446 = None
	        mul_15149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_156, 1.0);  reciprocal_156 = None
	        mul_15152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_23900, mul_15149);  mul_15149 = None
	        round_314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15152);  mul_15152 = None
	        add_23987: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_314, view_2447);  round_314 = view_2447 = None
	        clamp_min_470: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23987, -128);  add_23987 = None
	        clamp_max_313: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_470, 127);  clamp_min_470 = None
	        _assert_tensor_metadata_1408 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_313, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1408 = None
	        convert_element_type_937: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_313, torch.int8);  clamp_max_313 = None
	        view_2450: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_468, [sym_size_int, 1500, 1]);  clamp_min_468 = None
	        view_2451: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_936, [sym_size_int, 1500, 1]);  convert_element_type_936 = None
	        _assert_tensor_metadata_1409 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_937, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1409 = None
	        convert_element_type_938: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_937, torch.float32);  convert_element_type_937 = None
	        _assert_tensor_metadata_1410 = torch.ops.aten._assert_tensor_metadata.default(view_2451, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1410 = None
	        convert_element_type_939: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2451, torch.float32);  view_2451 = None
	        sub_7168: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_938, convert_element_type_939);  convert_element_type_938 = convert_element_type_939 = None
	        mul_15174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7168, view_2450);  sub_7168 = view_2450 = None
	        _assert_tensor_metadata_1411 = torch.ops.aten._assert_tensor_metadata.default(mul_15174, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1411 = None
	        view_2453: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2454: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2455: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1412 = torch.ops.aten._assert_tensor_metadata.default(view_2453, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1412 = None
	        convert_element_type_940: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2453, torch.float32);  view_2453 = None
	        _assert_tensor_metadata_1413 = torch.ops.aten._assert_tensor_metadata.default(view_2455, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1413 = None
	        convert_element_type_941: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2455, torch.float32);  view_2455 = None
	        sub_7172: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_940, convert_element_type_941);  convert_element_type_940 = convert_element_type_941 = None
	        mul_15179: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7172, view_2454);  sub_7172 = view_2454 = None
	        view_2456: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15179, [1280, 1280]);  mul_15179 = None
	        _assert_tensor_metadata_1414 = torch.ops.aten._assert_tensor_metadata.default(view_2456, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1414 = None
	        mul_15184: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2457: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15174, [mul_15184, 1280]);  mul_15174 = mul_15184 = None
	        permute_261: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2456, [1, 0]);  view_2456 = None
	        addmm_130: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_26_self_attn_q_proj_bias, view_2457, permute_261);  model_audio_tower_layers_26_self_attn_q_proj_bias = view_2457 = permute_261 = None
	        view_2458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_130, [sym_size_int, 1500, 1280]);  addmm_130 = None
	        mul_15191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2458, 0.125);  view_2458 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2459: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_15191, [sym_size_int, 1500, 20, 64]);  mul_15191 = None
	        permute_262: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2459, [0, 2, 1, 3]);  view_2459 = None
	        clone_210: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_262, memory_format = torch.contiguous_format);  permute_262 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_23900, [2])
	        amax_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_23900, [2])
	        full_314: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_157, full_314);  amin_157 = full_314 = None
	        full_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_157, full_315);  amax_157 = full_315 = None
	        sub_7187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_157, minimum_157);  maximum_157 = None
	        div_314: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7187, 255.0);  sub_7187 = None
	        clamp_min_471: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_314, 1.1920928955078125e-07);  div_314 = None
	        div_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_157, clamp_min_471);  minimum_157 = None
	        round_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_315);  div_315 = None
	        sub_7193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_315);  round_315 = None
	        clamp_min_472: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7193, -128);  sub_7193 = None
	        clamp_max_314: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_472, 127);  clamp_min_472 = None
	        _assert_tensor_metadata_1415 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_471, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1415 = None
	        _assert_tensor_metadata_1416 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_314, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1416 = None
	        convert_element_type_942: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_314, torch.int8);  clamp_max_314 = None
	        view_2462: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_471, [sym_size_int, 1500, 1])
	        view_2463: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_942, [sym_size_int, 1500, 1])
	        reciprocal_157: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2462);  view_2462 = None
	        mul_15245: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_157, 1.0);  reciprocal_157 = None
	        mul_15248: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_23900, mul_15245);  mul_15245 = None
	        round_316: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15248);  mul_15248 = None
	        add_24139: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_316, view_2463);  round_316 = view_2463 = None
	        clamp_min_473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24139, -128);  add_24139 = None
	        clamp_max_315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_473, 127);  clamp_min_473 = None
	        _assert_tensor_metadata_1417 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_315, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1417 = None
	        convert_element_type_943: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_315, torch.int8);  clamp_max_315 = None
	        view_2466: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_471, [sym_size_int, 1500, 1]);  clamp_min_471 = None
	        view_2467: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_942, [sym_size_int, 1500, 1]);  convert_element_type_942 = None
	        _assert_tensor_metadata_1418 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_943, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1418 = None
	        convert_element_type_944: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_943, torch.float32);  convert_element_type_943 = None
	        _assert_tensor_metadata_1419 = torch.ops.aten._assert_tensor_metadata.default(view_2467, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1419 = None
	        convert_element_type_945: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2467, torch.float32);  view_2467 = None
	        sub_7213: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_944, convert_element_type_945);  convert_element_type_944 = convert_element_type_945 = None
	        mul_15270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7213, view_2466);  sub_7213 = view_2466 = None
	        _assert_tensor_metadata_1420 = torch.ops.aten._assert_tensor_metadata.default(mul_15270, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1420 = None
	        view_2469: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2470: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2471: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1421 = torch.ops.aten._assert_tensor_metadata.default(view_2469, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1421 = None
	        convert_element_type_946: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2469, torch.float32);  view_2469 = None
	        _assert_tensor_metadata_1422 = torch.ops.aten._assert_tensor_metadata.default(view_2471, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1422 = None
	        convert_element_type_947: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2471, torch.float32);  view_2471 = None
	        sub_7217: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_946, convert_element_type_947);  convert_element_type_946 = convert_element_type_947 = None
	        mul_15275: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7217, view_2470);  sub_7217 = view_2470 = None
	        view_2472: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15275, [1280, 1280]);  mul_15275 = None
	        _assert_tensor_metadata_1423 = torch.ops.aten._assert_tensor_metadata.default(view_2472, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1423 = None
	        permute_263: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2472, [1, 0]);  view_2472 = None
	        mul_15278: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2473: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15270, [mul_15278, 1280]);  mul_15270 = mul_15278 = None
	        mm_26: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2473, permute_263);  view_2473 = permute_263 = None
	        view_2474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_26, [sym_size_int, 1500, 1280]);  mm_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2475: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2474, [sym_size_int, -1, 20, 64]);  view_2474 = None
	        permute_264: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2475, [0, 2, 1, 3]);  view_2475 = None
	        clone_211: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_264, memory_format = torch.contiguous_format);  permute_264 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_23900, [2])
	        amax_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_23900, [2])
	        full_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_158, full_316);  amin_158 = full_316 = None
	        full_317: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_158, full_317);  amax_158 = full_317 = None
	        sub_7231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_158, minimum_158);  maximum_158 = None
	        div_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7231, 255.0);  sub_7231 = None
	        clamp_min_474: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_316, 1.1920928955078125e-07);  div_316 = None
	        div_317: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_158, clamp_min_474);  minimum_158 = None
	        round_317: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_317);  div_317 = None
	        sub_7237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_317);  round_317 = None
	        clamp_min_475: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7237, -128);  sub_7237 = None
	        clamp_max_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_475, 127);  clamp_min_475 = None
	        _assert_tensor_metadata_1424 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_474, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1424 = None
	        _assert_tensor_metadata_1425 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_316, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1425 = None
	        convert_element_type_948: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_316, torch.int8);  clamp_max_316 = None
	        view_2478: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_474, [sym_size_int, 1500, 1])
	        view_2479: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_948, [sym_size_int, 1500, 1])
	        reciprocal_158: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2478);  view_2478 = None
	        mul_15344: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_158, 1.0);  reciprocal_158 = None
	        mul_15347: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_23900, mul_15344);  add_23900 = mul_15344 = None
	        round_318: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15347);  mul_15347 = None
	        add_24287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_318, view_2479);  round_318 = view_2479 = None
	        clamp_min_476: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24287, -128);  add_24287 = None
	        clamp_max_317: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_476, 127);  clamp_min_476 = None
	        _assert_tensor_metadata_1426 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_317, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1426 = None
	        convert_element_type_949: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_317, torch.int8);  clamp_max_317 = None
	        view_2482: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_474, [sym_size_int, 1500, 1]);  clamp_min_474 = None
	        view_2483: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_948, [sym_size_int, 1500, 1]);  convert_element_type_948 = None
	        _assert_tensor_metadata_1427 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_949, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1427 = None
	        convert_element_type_950: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_949, torch.float32);  convert_element_type_949 = None
	        _assert_tensor_metadata_1428 = torch.ops.aten._assert_tensor_metadata.default(view_2483, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1428 = None
	        convert_element_type_951: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2483, torch.float32);  view_2483 = None
	        sub_7257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_950, convert_element_type_951);  convert_element_type_950 = convert_element_type_951 = None
	        mul_15369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7257, view_2482);  sub_7257 = view_2482 = None
	        _assert_tensor_metadata_1429 = torch.ops.aten._assert_tensor_metadata.default(mul_15369, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1429 = None
	        view_2485: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2486: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2487: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1430 = torch.ops.aten._assert_tensor_metadata.default(view_2485, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1430 = None
	        convert_element_type_952: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2485, torch.float32);  view_2485 = None
	        _assert_tensor_metadata_1431 = torch.ops.aten._assert_tensor_metadata.default(view_2487, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1431 = None
	        convert_element_type_953: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2487, torch.float32);  view_2487 = None
	        sub_7261: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_952, convert_element_type_953);  convert_element_type_952 = convert_element_type_953 = None
	        mul_15374: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7261, view_2486);  sub_7261 = view_2486 = None
	        view_2488: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15374, [1280, 1280]);  mul_15374 = None
	        _assert_tensor_metadata_1432 = torch.ops.aten._assert_tensor_metadata.default(view_2488, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1432 = None
	        mul_15379: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2489: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15369, [mul_15379, 1280]);  mul_15369 = mul_15379 = None
	        permute_265: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2488, [1, 0]);  view_2488 = None
	        addmm_131: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_26_self_attn_v_proj_bias, view_2489, permute_265);  model_audio_tower_layers_26_self_attn_v_proj_bias = view_2489 = permute_265 = None
	        view_2490: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_131, [sym_size_int, 1500, 1280]);  addmm_131 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2491: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2490, [sym_size_int, -1, 20, 64]);  view_2490 = None
	        permute_266: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2491, [0, 2, 1, 3]);  view_2491 = None
	        clone_212: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_266, memory_format = torch.contiguous_format);  permute_266 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_26 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_210, clone_211, clone_212, None, False, scale = 1.0);  clone_210 = clone_211 = clone_212 = None
	        getitem_210: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_26[0];  _scaled_dot_product_efficient_attention_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_267: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_210, [0, 2, 1, 3]);  getitem_210 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2492: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_267, [sym_size_int, 1500, -1]);  permute_267 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2492, [2])
	        amax_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2492, [2])
	        full_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_159, full_318);  amin_159 = full_318 = None
	        full_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_159, full_319);  amax_159 = full_319 = None
	        sub_7279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_159, minimum_159);  maximum_159 = None
	        div_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7279, 255.0);  sub_7279 = None
	        clamp_min_477: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_318, 1.1920928955078125e-07);  div_318 = None
	        div_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_159, clamp_min_477);  minimum_159 = None
	        round_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_319);  div_319 = None
	        sub_7285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_319);  round_319 = None
	        clamp_min_478: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7285, -128);  sub_7285 = None
	        clamp_max_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_478, 127);  clamp_min_478 = None
	        _assert_tensor_metadata_1433 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_477, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1433 = None
	        _assert_tensor_metadata_1434 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_318, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1434 = None
	        convert_element_type_954: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_318, torch.int8);  clamp_max_318 = None
	        view_2495: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_477, [sym_size_int, 1500, 1])
	        view_2496: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_954, [sym_size_int, 1500, 1])
	        reciprocal_159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2495);  view_2495 = None
	        mul_15449: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_159, 1.0);  reciprocal_159 = None
	        mul_15452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2492, mul_15449);  view_2492 = mul_15449 = None
	        round_320: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15452);  mul_15452 = None
	        add_24451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_320, view_2496);  round_320 = view_2496 = None
	        clamp_min_479: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24451, -128);  add_24451 = None
	        clamp_max_319: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_479, 127);  clamp_min_479 = None
	        _assert_tensor_metadata_1435 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_319, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1435 = None
	        convert_element_type_955: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_319, torch.int8);  clamp_max_319 = None
	        view_2499: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_477, [sym_size_int, 1500, 1]);  clamp_min_477 = None
	        view_2500: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_954, [sym_size_int, 1500, 1]);  convert_element_type_954 = None
	        _assert_tensor_metadata_1436 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_955, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1436 = None
	        convert_element_type_956: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_955, torch.float32);  convert_element_type_955 = None
	        _assert_tensor_metadata_1437 = torch.ops.aten._assert_tensor_metadata.default(view_2500, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1437 = None
	        convert_element_type_957: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2500, torch.float32);  view_2500 = None
	        sub_7305: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_956, convert_element_type_957);  convert_element_type_956 = convert_element_type_957 = None
	        mul_15474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7305, view_2499);  sub_7305 = view_2499 = None
	        _assert_tensor_metadata_1438 = torch.ops.aten._assert_tensor_metadata.default(mul_15474, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1438 = None
	        view_2502: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2503: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2504: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1439 = torch.ops.aten._assert_tensor_metadata.default(view_2502, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1439 = None
	        convert_element_type_958: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2502, torch.float32);  view_2502 = None
	        _assert_tensor_metadata_1440 = torch.ops.aten._assert_tensor_metadata.default(view_2504, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1440 = None
	        convert_element_type_959: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2504, torch.float32);  view_2504 = None
	        sub_7309: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_958, convert_element_type_959);  convert_element_type_958 = convert_element_type_959 = None
	        mul_15479: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7309, view_2503);  sub_7309 = view_2503 = None
	        view_2505: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15479, [1280, 1280]);  mul_15479 = None
	        _assert_tensor_metadata_1441 = torch.ops.aten._assert_tensor_metadata.default(view_2505, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1441 = None
	        mul_15484: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2506: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15474, [mul_15484, 1280]);  mul_15474 = mul_15484 = None
	        permute_268: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2505, [1, 0]);  view_2505 = None
	        addmm_132: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_26_self_attn_out_proj_bias, view_2506, permute_268);  model_audio_tower_layers_26_self_attn_out_proj_bias = view_2506 = permute_268 = None
	        view_2507: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_132, [sym_size_int, 1500, 1280]);  addmm_132 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_24514: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_23894, view_2507);  add_23894 = view_2507 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_24514, memory_format = torch.contiguous_format)
	        var_mean_53 = torch.ops.aten.var_mean.correction(clone_214, [2], correction = 0, keepdim = True)
	        getitem_214: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_53[0]
	        getitem_215: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_53[1];  var_mean_53 = None
	        add_24519: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_214, 1e-05);  getitem_214 = None
	        rsqrt_53: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_24519);  add_24519 = None
	        sub_7315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_214, getitem_215);  clone_214 = getitem_215 = None
	        mul_15495: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7315, rsqrt_53);  sub_7315 = rsqrt_53 = None
	        mul_15496: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15495, model_audio_tower_layers_26_final_layer_norm_weight);  mul_15495 = model_audio_tower_layers_26_final_layer_norm_weight = None
	        add_24520: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_15496, model_audio_tower_layers_26_final_layer_norm_bias);  mul_15496 = model_audio_tower_layers_26_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_24520, [2])
	        amax_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_24520, [2])
	        full_320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_160, full_320);  amin_160 = full_320 = None
	        full_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_160, full_321);  amax_160 = full_321 = None
	        sub_7326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_160, minimum_160);  maximum_160 = None
	        div_320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7326, 255.0);  sub_7326 = None
	        clamp_min_480: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_320, 1.1920928955078125e-07);  div_320 = None
	        div_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_160, clamp_min_480);  minimum_160 = None
	        round_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_321);  div_321 = None
	        sub_7332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_321);  round_321 = None
	        clamp_min_481: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7332, -128);  sub_7332 = None
	        clamp_max_320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_481, 127);  clamp_min_481 = None
	        _assert_tensor_metadata_1442 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_480, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1442 = None
	        _assert_tensor_metadata_1443 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_320, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1443 = None
	        convert_element_type_960: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_320, torch.int8);  clamp_max_320 = None
	        view_2510: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_480, [sym_size_int, 1500, 1])
	        view_2511: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_960, [sym_size_int, 1500, 1])
	        reciprocal_160: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2510);  view_2510 = None
	        mul_15544: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_160, 1.0);  reciprocal_160 = None
	        mul_15547: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_24520, mul_15544);  add_24520 = mul_15544 = None
	        round_322: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15547);  mul_15547 = None
	        add_24607: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_322, view_2511);  round_322 = view_2511 = None
	        clamp_min_482: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24607, -128);  add_24607 = None
	        clamp_max_321: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_482, 127);  clamp_min_482 = None
	        _assert_tensor_metadata_1444 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_321, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1444 = None
	        convert_element_type_961: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_321, torch.int8);  clamp_max_321 = None
	        view_2514: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_480, [sym_size_int, 1500, 1]);  clamp_min_480 = None
	        view_2515: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_960, [sym_size_int, 1500, 1]);  convert_element_type_960 = None
	        _assert_tensor_metadata_1445 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_961, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1445 = None
	        convert_element_type_962: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_961, torch.float32);  convert_element_type_961 = None
	        _assert_tensor_metadata_1446 = torch.ops.aten._assert_tensor_metadata.default(view_2515, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1446 = None
	        convert_element_type_963: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2515, torch.float32);  view_2515 = None
	        sub_7352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_962, convert_element_type_963);  convert_element_type_962 = convert_element_type_963 = None
	        mul_15569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7352, view_2514);  sub_7352 = view_2514 = None
	        _assert_tensor_metadata_1447 = torch.ops.aten._assert_tensor_metadata.default(mul_15569, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1447 = None
	        view_2517: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_26_fc1_parametrizations_weight_original0 = None
	        view_2518: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_26_fc1_parametrizations_weight_original1 = None
	        view_2519: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_26_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1448 = torch.ops.aten._assert_tensor_metadata.default(view_2517, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1448 = None
	        convert_element_type_964: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2517, torch.float32);  view_2517 = None
	        _assert_tensor_metadata_1449 = torch.ops.aten._assert_tensor_metadata.default(view_2519, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1449 = None
	        convert_element_type_965: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2519, torch.float32);  view_2519 = None
	        sub_7356: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_964, convert_element_type_965);  convert_element_type_964 = convert_element_type_965 = None
	        mul_15574: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7356, view_2518);  sub_7356 = view_2518 = None
	        view_2520: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15574, [5120, 1280]);  mul_15574 = None
	        _assert_tensor_metadata_1450 = torch.ops.aten._assert_tensor_metadata.default(view_2520, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1450 = None
	        mul_15579: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2521: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15569, [mul_15579, 1280]);  mul_15569 = mul_15579 = None
	        permute_269: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2520, [1, 0]);  view_2520 = None
	        addmm_133: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_26_fc1_bias, view_2521, permute_269);  model_audio_tower_layers_26_fc1_bias = view_2521 = permute_269 = None
	        view_2522: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_133, [sym_size_int, 1500, 5120]);  addmm_133 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_15586: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2522, 0.5)
	        mul_15587: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2522, 0.7071067811865476);  view_2522 = None
	        erf_28: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_15587);  mul_15587 = None
	        add_24666: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_28, 1);  erf_28 = None
	        mul_15588: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15586, add_24666);  mul_15586 = add_24666 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_15588, [2])
	        amax_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_15588, [2])
	        full_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_161, full_322);  amin_161 = full_322 = None
	        full_323: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_161, full_323);  amax_161 = full_323 = None
	        sub_7369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_161, minimum_161);  maximum_161 = None
	        div_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7369, 255.0);  sub_7369 = None
	        clamp_min_483: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_322, 1.1920928955078125e-07);  div_322 = None
	        div_323: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_161, clamp_min_483);  minimum_161 = None
	        round_323: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_323);  div_323 = None
	        sub_7375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_323);  round_323 = None
	        clamp_min_484: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7375, -128);  sub_7375 = None
	        clamp_max_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_484, 127);  clamp_min_484 = None
	        _assert_tensor_metadata_1451 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_483, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1451 = None
	        _assert_tensor_metadata_1452 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_322, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1452 = None
	        convert_element_type_966: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_322, torch.int8);  clamp_max_322 = None
	        view_2525: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_483, [sym_size_int, 1500, 1])
	        view_2526: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_966, [sym_size_int, 1500, 1])
	        reciprocal_161: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2525);  view_2525 = None
	        mul_15634: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_161, 1.0);  reciprocal_161 = None
	        mul_15637: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15588, mul_15634);  mul_15588 = mul_15634 = None
	        round_324: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_15637);  mul_15637 = None
	        add_24749: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_324, view_2526);  round_324 = view_2526 = None
	        clamp_min_485: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24749, -128);  add_24749 = None
	        clamp_max_323: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_485, 127);  clamp_min_485 = None
	        _assert_tensor_metadata_1453 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_323, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1453 = None
	        convert_element_type_967: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_323, torch.int8);  clamp_max_323 = None
	        view_2529: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_483, [sym_size_int, 1500, 1]);  clamp_min_483 = None
	        view_2530: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_966, [sym_size_int, 1500, 1]);  convert_element_type_966 = None
	        _assert_tensor_metadata_1454 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_967, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1454 = None
	        convert_element_type_968: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_967, torch.float32);  convert_element_type_967 = None
	        _assert_tensor_metadata_1455 = torch.ops.aten._assert_tensor_metadata.default(view_2530, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1455 = None
	        convert_element_type_969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2530, torch.float32);  view_2530 = None
	        sub_7395: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_968, convert_element_type_969);  convert_element_type_968 = convert_element_type_969 = None
	        mul_15659: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7395, view_2529);  sub_7395 = view_2529 = None
	        _assert_tensor_metadata_1456 = torch.ops.aten._assert_tensor_metadata.default(mul_15659, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1456 = None
	        view_2532: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_26_fc2_parametrizations_weight_original0 = None
	        view_2533: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_26_fc2_parametrizations_weight_original1 = None
	        view_2534: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_26_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_26_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1457 = torch.ops.aten._assert_tensor_metadata.default(view_2532, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1457 = None
	        convert_element_type_970: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2532, torch.float32);  view_2532 = None
	        _assert_tensor_metadata_1458 = torch.ops.aten._assert_tensor_metadata.default(view_2534, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1458 = None
	        convert_element_type_971: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2534, torch.float32);  view_2534 = None
	        sub_7399: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_970, convert_element_type_971);  convert_element_type_970 = convert_element_type_971 = None
	        mul_15664: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7399, view_2533);  sub_7399 = view_2533 = None
	        view_2535: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_15664, [1280, 5120]);  mul_15664 = None
	        _assert_tensor_metadata_1459 = torch.ops.aten._assert_tensor_metadata.default(view_2535, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1459 = None
	        mul_15669: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2536: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_15659, [mul_15669, 5120]);  mul_15659 = mul_15669 = None
	        permute_270: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2535, [1, 0]);  view_2535 = None
	        addmm_134: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_26_fc2_bias, view_2536, permute_270);  model_audio_tower_layers_26_fc2_bias = view_2536 = permute_270 = None
	        view_2537: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_134, [sym_size_int, 1500, 1280]);  addmm_134 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_24812: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_24514, view_2537);  add_24514 = view_2537 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_217: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_24812, memory_format = torch.contiguous_format)
	        var_mean_54 = torch.ops.aten.var_mean.correction(clone_217, [2], correction = 0, keepdim = True)
	        getitem_216: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_54[0]
	        getitem_217: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_54[1];  var_mean_54 = None
	        add_24817: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_216, 1e-05);  getitem_216 = None
	        rsqrt_54: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_24817);  add_24817 = None
	        sub_7405: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_217, getitem_217);  clone_217 = getitem_217 = None
	        mul_15680: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7405, rsqrt_54);  sub_7405 = rsqrt_54 = None
	        mul_15681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15680, model_audio_tower_layers_27_self_attn_layer_norm_weight);  mul_15680 = model_audio_tower_layers_27_self_attn_layer_norm_weight = None
	        add_24818: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_15681, model_audio_tower_layers_27_self_attn_layer_norm_bias);  mul_15681 = model_audio_tower_layers_27_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_24818, [2])
	        amax_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_24818, [2])
	        full_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_162, full_324);  amin_162 = full_324 = None
	        full_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_162, full_325);  amax_162 = full_325 = None
	        sub_7416: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_162, minimum_162);  maximum_162 = None
	        div_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7416, 255.0);  sub_7416 = None
	        clamp_min_486: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_324, 1.1920928955078125e-07);  div_324 = None
	        div_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_162, clamp_min_486);  minimum_162 = None
	        round_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_325);  div_325 = None
	        sub_7422: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_325);  round_325 = None
	        clamp_min_487: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7422, -128);  sub_7422 = None
	        clamp_max_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_487, 127);  clamp_min_487 = None
	        _assert_tensor_metadata_1460 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_486, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1460 = None
	        _assert_tensor_metadata_1461 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_324, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1461 = None
	        convert_element_type_972: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_324, torch.int8);  clamp_max_324 = None
	        view_2540: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_486, [sym_size_int, 1500, 1])
	        view_2541: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_972, [sym_size_int, 1500, 1])
	        reciprocal_162: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2540);  view_2540 = None
	        mul_15729: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_162, 1.0);  reciprocal_162 = None
	        mul_15732: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_24818, mul_15729);  mul_15729 = None
	        round_326: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15732);  mul_15732 = None
	        add_24905: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_326, view_2541);  round_326 = view_2541 = None
	        clamp_min_488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24905, -128);  add_24905 = None
	        clamp_max_325: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_488, 127);  clamp_min_488 = None
	        _assert_tensor_metadata_1462 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_325, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1462 = None
	        convert_element_type_973: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_325, torch.int8);  clamp_max_325 = None
	        view_2544: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_486, [sym_size_int, 1500, 1]);  clamp_min_486 = None
	        view_2545: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_972, [sym_size_int, 1500, 1]);  convert_element_type_972 = None
	        _assert_tensor_metadata_1463 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_973, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1463 = None
	        convert_element_type_974: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_973, torch.float32);  convert_element_type_973 = None
	        _assert_tensor_metadata_1464 = torch.ops.aten._assert_tensor_metadata.default(view_2545, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1464 = None
	        convert_element_type_975: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2545, torch.float32);  view_2545 = None
	        sub_7442: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_974, convert_element_type_975);  convert_element_type_974 = convert_element_type_975 = None
	        mul_15754: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7442, view_2544);  sub_7442 = view_2544 = None
	        _assert_tensor_metadata_1465 = torch.ops.aten._assert_tensor_metadata.default(mul_15754, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1465 = None
	        view_2547: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2548: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2549: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1466 = torch.ops.aten._assert_tensor_metadata.default(view_2547, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1466 = None
	        convert_element_type_976: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2547, torch.float32);  view_2547 = None
	        _assert_tensor_metadata_1467 = torch.ops.aten._assert_tensor_metadata.default(view_2549, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1467 = None
	        convert_element_type_977: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2549, torch.float32);  view_2549 = None
	        sub_7446: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_976, convert_element_type_977);  convert_element_type_976 = convert_element_type_977 = None
	        mul_15759: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7446, view_2548);  sub_7446 = view_2548 = None
	        view_2550: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15759, [1280, 1280]);  mul_15759 = None
	        _assert_tensor_metadata_1468 = torch.ops.aten._assert_tensor_metadata.default(view_2550, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1468 = None
	        mul_15764: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2551: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15754, [mul_15764, 1280]);  mul_15754 = mul_15764 = None
	        permute_271: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2550, [1, 0]);  view_2550 = None
	        addmm_135: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_27_self_attn_q_proj_bias, view_2551, permute_271);  model_audio_tower_layers_27_self_attn_q_proj_bias = view_2551 = permute_271 = None
	        view_2552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_135, [sym_size_int, 1500, 1280]);  addmm_135 = None
	        mul_15771: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2552, 0.125);  view_2552 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2553: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_15771, [sym_size_int, 1500, 20, 64]);  mul_15771 = None
	        permute_272: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2553, [0, 2, 1, 3]);  view_2553 = None
	        clone_218: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_272, memory_format = torch.contiguous_format);  permute_272 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_24818, [2])
	        amax_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_24818, [2])
	        full_326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_163, full_326);  amin_163 = full_326 = None
	        full_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_163, full_327);  amax_163 = full_327 = None
	        sub_7461: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_163, minimum_163);  maximum_163 = None
	        div_326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7461, 255.0);  sub_7461 = None
	        clamp_min_489: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_326, 1.1920928955078125e-07);  div_326 = None
	        div_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_163, clamp_min_489);  minimum_163 = None
	        round_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_327);  div_327 = None
	        sub_7467: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_327);  round_327 = None
	        clamp_min_490: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7467, -128);  sub_7467 = None
	        clamp_max_326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_490, 127);  clamp_min_490 = None
	        _assert_tensor_metadata_1469 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_489, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1469 = None
	        _assert_tensor_metadata_1470 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_326, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1470 = None
	        convert_element_type_978: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_326, torch.int8);  clamp_max_326 = None
	        view_2556: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_489, [sym_size_int, 1500, 1])
	        view_2557: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_978, [sym_size_int, 1500, 1])
	        reciprocal_163: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2556);  view_2556 = None
	        mul_15825: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_163, 1.0);  reciprocal_163 = None
	        mul_15828: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_24818, mul_15825);  mul_15825 = None
	        round_328: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15828);  mul_15828 = None
	        add_25057: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_328, view_2557);  round_328 = view_2557 = None
	        clamp_min_491: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25057, -128);  add_25057 = None
	        clamp_max_327: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_491, 127);  clamp_min_491 = None
	        _assert_tensor_metadata_1471 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_327, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1471 = None
	        convert_element_type_979: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_327, torch.int8);  clamp_max_327 = None
	        view_2560: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_489, [sym_size_int, 1500, 1]);  clamp_min_489 = None
	        view_2561: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_978, [sym_size_int, 1500, 1]);  convert_element_type_978 = None
	        _assert_tensor_metadata_1472 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_979, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1472 = None
	        convert_element_type_980: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_979, torch.float32);  convert_element_type_979 = None
	        _assert_tensor_metadata_1473 = torch.ops.aten._assert_tensor_metadata.default(view_2561, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1473 = None
	        convert_element_type_981: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2561, torch.float32);  view_2561 = None
	        sub_7487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_980, convert_element_type_981);  convert_element_type_980 = convert_element_type_981 = None
	        mul_15850: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7487, view_2560);  sub_7487 = view_2560 = None
	        _assert_tensor_metadata_1474 = torch.ops.aten._assert_tensor_metadata.default(mul_15850, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1474 = None
	        view_2563: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2564: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2565: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1475 = torch.ops.aten._assert_tensor_metadata.default(view_2563, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1475 = None
	        convert_element_type_982: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2563, torch.float32);  view_2563 = None
	        _assert_tensor_metadata_1476 = torch.ops.aten._assert_tensor_metadata.default(view_2565, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1476 = None
	        convert_element_type_983: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2565, torch.float32);  view_2565 = None
	        sub_7491: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_982, convert_element_type_983);  convert_element_type_982 = convert_element_type_983 = None
	        mul_15855: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7491, view_2564);  sub_7491 = view_2564 = None
	        view_2566: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15855, [1280, 1280]);  mul_15855 = None
	        _assert_tensor_metadata_1477 = torch.ops.aten._assert_tensor_metadata.default(view_2566, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1477 = None
	        permute_273: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2566, [1, 0]);  view_2566 = None
	        mul_15858: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2567: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15850, [mul_15858, 1280]);  mul_15850 = mul_15858 = None
	        mm_27: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2567, permute_273);  view_2567 = permute_273 = None
	        view_2568: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_27, [sym_size_int, 1500, 1280]);  mm_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2569: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2568, [sym_size_int, -1, 20, 64]);  view_2568 = None
	        permute_274: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2569, [0, 2, 1, 3]);  view_2569 = None
	        clone_219: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_274, memory_format = torch.contiguous_format);  permute_274 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_24818, [2])
	        amax_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_24818, [2])
	        full_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_164, full_328);  amin_164 = full_328 = None
	        full_329: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_164, full_329);  amax_164 = full_329 = None
	        sub_7505: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_164, minimum_164);  maximum_164 = None
	        div_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7505, 255.0);  sub_7505 = None
	        clamp_min_492: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_328, 1.1920928955078125e-07);  div_328 = None
	        div_329: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_164, clamp_min_492);  minimum_164 = None
	        round_329: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_329);  div_329 = None
	        sub_7511: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_329);  round_329 = None
	        clamp_min_493: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7511, -128);  sub_7511 = None
	        clamp_max_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_493, 127);  clamp_min_493 = None
	        _assert_tensor_metadata_1478 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_492, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1478 = None
	        _assert_tensor_metadata_1479 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_328, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1479 = None
	        convert_element_type_984: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_328, torch.int8);  clamp_max_328 = None
	        view_2572: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_492, [sym_size_int, 1500, 1])
	        view_2573: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_984, [sym_size_int, 1500, 1])
	        reciprocal_164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2572);  view_2572 = None
	        mul_15924: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_164, 1.0);  reciprocal_164 = None
	        mul_15927: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_24818, mul_15924);  add_24818 = mul_15924 = None
	        round_330: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15927);  mul_15927 = None
	        add_25205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_330, view_2573);  round_330 = view_2573 = None
	        clamp_min_494: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25205, -128);  add_25205 = None
	        clamp_max_329: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_494, 127);  clamp_min_494 = None
	        _assert_tensor_metadata_1480 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_329, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1480 = None
	        convert_element_type_985: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_329, torch.int8);  clamp_max_329 = None
	        view_2576: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_492, [sym_size_int, 1500, 1]);  clamp_min_492 = None
	        view_2577: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_984, [sym_size_int, 1500, 1]);  convert_element_type_984 = None
	        _assert_tensor_metadata_1481 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_985, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1481 = None
	        convert_element_type_986: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_985, torch.float32);  convert_element_type_985 = None
	        _assert_tensor_metadata_1482 = torch.ops.aten._assert_tensor_metadata.default(view_2577, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1482 = None
	        convert_element_type_987: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2577, torch.float32);  view_2577 = None
	        sub_7531: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_986, convert_element_type_987);  convert_element_type_986 = convert_element_type_987 = None
	        mul_15949: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7531, view_2576);  sub_7531 = view_2576 = None
	        _assert_tensor_metadata_1483 = torch.ops.aten._assert_tensor_metadata.default(mul_15949, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1483 = None
	        view_2579: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2580: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2581: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1484 = torch.ops.aten._assert_tensor_metadata.default(view_2579, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1484 = None
	        convert_element_type_988: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2579, torch.float32);  view_2579 = None
	        _assert_tensor_metadata_1485 = torch.ops.aten._assert_tensor_metadata.default(view_2581, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1485 = None
	        convert_element_type_989: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2581, torch.float32);  view_2581 = None
	        sub_7535: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_988, convert_element_type_989);  convert_element_type_988 = convert_element_type_989 = None
	        mul_15954: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7535, view_2580);  sub_7535 = view_2580 = None
	        view_2582: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15954, [1280, 1280]);  mul_15954 = None
	        _assert_tensor_metadata_1486 = torch.ops.aten._assert_tensor_metadata.default(view_2582, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1486 = None
	        mul_15959: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2583: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_15949, [mul_15959, 1280]);  mul_15949 = mul_15959 = None
	        permute_275: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2582, [1, 0]);  view_2582 = None
	        addmm_136: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_27_self_attn_v_proj_bias, view_2583, permute_275);  model_audio_tower_layers_27_self_attn_v_proj_bias = view_2583 = permute_275 = None
	        view_2584: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_136, [sym_size_int, 1500, 1280]);  addmm_136 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2585: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2584, [sym_size_int, -1, 20, 64]);  view_2584 = None
	        permute_276: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2585, [0, 2, 1, 3]);  view_2585 = None
	        clone_220: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_276, memory_format = torch.contiguous_format);  permute_276 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_27 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_218, clone_219, clone_220, None, False, scale = 1.0);  clone_218 = clone_219 = clone_220 = None
	        getitem_218: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_27[0];  _scaled_dot_product_efficient_attention_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_277: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_218, [0, 2, 1, 3]);  getitem_218 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2586: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_277, [sym_size_int, 1500, -1]);  permute_277 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2586, [2])
	        amax_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2586, [2])
	        full_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_165, full_330);  amin_165 = full_330 = None
	        full_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_165, full_331);  amax_165 = full_331 = None
	        sub_7553: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_165, minimum_165);  maximum_165 = None
	        div_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7553, 255.0);  sub_7553 = None
	        clamp_min_495: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_330, 1.1920928955078125e-07);  div_330 = None
	        div_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_165, clamp_min_495);  minimum_165 = None
	        round_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_331);  div_331 = None
	        sub_7559: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_331);  round_331 = None
	        clamp_min_496: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7559, -128);  sub_7559 = None
	        clamp_max_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_496, 127);  clamp_min_496 = None
	        _assert_tensor_metadata_1487 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_495, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1487 = None
	        _assert_tensor_metadata_1488 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_330, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1488 = None
	        convert_element_type_990: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_330, torch.int8);  clamp_max_330 = None
	        view_2589: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_495, [sym_size_int, 1500, 1])
	        view_2590: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_990, [sym_size_int, 1500, 1])
	        reciprocal_165: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2589);  view_2589 = None
	        mul_16029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_165, 1.0);  reciprocal_165 = None
	        mul_16032: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2586, mul_16029);  view_2586 = mul_16029 = None
	        round_332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16032);  mul_16032 = None
	        add_25369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_332, view_2590);  round_332 = view_2590 = None
	        clamp_min_497: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25369, -128);  add_25369 = None
	        clamp_max_331: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_497, 127);  clamp_min_497 = None
	        _assert_tensor_metadata_1489 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_331, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1489 = None
	        convert_element_type_991: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_331, torch.int8);  clamp_max_331 = None
	        view_2593: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_495, [sym_size_int, 1500, 1]);  clamp_min_495 = None
	        view_2594: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_990, [sym_size_int, 1500, 1]);  convert_element_type_990 = None
	        _assert_tensor_metadata_1490 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_991, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1490 = None
	        convert_element_type_992: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_991, torch.float32);  convert_element_type_991 = None
	        _assert_tensor_metadata_1491 = torch.ops.aten._assert_tensor_metadata.default(view_2594, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1491 = None
	        convert_element_type_993: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2594, torch.float32);  view_2594 = None
	        sub_7579: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_992, convert_element_type_993);  convert_element_type_992 = convert_element_type_993 = None
	        mul_16054: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7579, view_2593);  sub_7579 = view_2593 = None
	        _assert_tensor_metadata_1492 = torch.ops.aten._assert_tensor_metadata.default(mul_16054, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1492 = None
	        view_2596: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2597: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2598: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1493 = torch.ops.aten._assert_tensor_metadata.default(view_2596, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1493 = None
	        convert_element_type_994: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2596, torch.float32);  view_2596 = None
	        _assert_tensor_metadata_1494 = torch.ops.aten._assert_tensor_metadata.default(view_2598, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1494 = None
	        convert_element_type_995: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2598, torch.float32);  view_2598 = None
	        sub_7583: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_994, convert_element_type_995);  convert_element_type_994 = convert_element_type_995 = None
	        mul_16059: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7583, view_2597);  sub_7583 = view_2597 = None
	        view_2599: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16059, [1280, 1280]);  mul_16059 = None
	        _assert_tensor_metadata_1495 = torch.ops.aten._assert_tensor_metadata.default(view_2599, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1495 = None
	        mul_16064: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2600: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16054, [mul_16064, 1280]);  mul_16054 = mul_16064 = None
	        permute_278: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2599, [1, 0]);  view_2599 = None
	        addmm_137: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_27_self_attn_out_proj_bias, view_2600, permute_278);  model_audio_tower_layers_27_self_attn_out_proj_bias = view_2600 = permute_278 = None
	        view_2601: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_137, [sym_size_int, 1500, 1280]);  addmm_137 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_25432: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_24812, view_2601);  add_24812 = view_2601 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_222: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_25432, memory_format = torch.contiguous_format)
	        var_mean_55 = torch.ops.aten.var_mean.correction(clone_222, [2], correction = 0, keepdim = True)
	        getitem_222: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_55[0]
	        getitem_223: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_55[1];  var_mean_55 = None
	        add_25437: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_222, 1e-05);  getitem_222 = None
	        rsqrt_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_25437);  add_25437 = None
	        sub_7589: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_222, getitem_223);  clone_222 = getitem_223 = None
	        mul_16075: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7589, rsqrt_55);  sub_7589 = rsqrt_55 = None
	        mul_16076: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16075, model_audio_tower_layers_27_final_layer_norm_weight);  mul_16075 = model_audio_tower_layers_27_final_layer_norm_weight = None
	        add_25438: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16076, model_audio_tower_layers_27_final_layer_norm_bias);  mul_16076 = model_audio_tower_layers_27_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_25438, [2])
	        amax_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_25438, [2])
	        full_332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_166, full_332);  amin_166 = full_332 = None
	        full_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_166, full_333);  amax_166 = full_333 = None
	        sub_7600: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_166, minimum_166);  maximum_166 = None
	        div_332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7600, 255.0);  sub_7600 = None
	        clamp_min_498: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_332, 1.1920928955078125e-07);  div_332 = None
	        div_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_166, clamp_min_498);  minimum_166 = None
	        round_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_333);  div_333 = None
	        sub_7606: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_333);  round_333 = None
	        clamp_min_499: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7606, -128);  sub_7606 = None
	        clamp_max_332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_499, 127);  clamp_min_499 = None
	        _assert_tensor_metadata_1496 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_498, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1496 = None
	        _assert_tensor_metadata_1497 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_332, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1497 = None
	        convert_element_type_996: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_332, torch.int8);  clamp_max_332 = None
	        view_2604: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_498, [sym_size_int, 1500, 1])
	        view_2605: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_996, [sym_size_int, 1500, 1])
	        reciprocal_166: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2604);  view_2604 = None
	        mul_16124: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_166, 1.0);  reciprocal_166 = None
	        mul_16127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_25438, mul_16124);  add_25438 = mul_16124 = None
	        round_334: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16127);  mul_16127 = None
	        add_25525: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_334, view_2605);  round_334 = view_2605 = None
	        clamp_min_500: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25525, -128);  add_25525 = None
	        clamp_max_333: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_500, 127);  clamp_min_500 = None
	        _assert_tensor_metadata_1498 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_333, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1498 = None
	        convert_element_type_997: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_333, torch.int8);  clamp_max_333 = None
	        view_2608: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_498, [sym_size_int, 1500, 1]);  clamp_min_498 = None
	        view_2609: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_996, [sym_size_int, 1500, 1]);  convert_element_type_996 = None
	        _assert_tensor_metadata_1499 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_997, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1499 = None
	        convert_element_type_998: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_997, torch.float32);  convert_element_type_997 = None
	        _assert_tensor_metadata_1500 = torch.ops.aten._assert_tensor_metadata.default(view_2609, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1500 = None
	        convert_element_type_999: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2609, torch.float32);  view_2609 = None
	        sub_7626: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_998, convert_element_type_999);  convert_element_type_998 = convert_element_type_999 = None
	        mul_16149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7626, view_2608);  sub_7626 = view_2608 = None
	        _assert_tensor_metadata_1501 = torch.ops.aten._assert_tensor_metadata.default(mul_16149, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1501 = None
	        view_2611: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_27_fc1_parametrizations_weight_original0 = None
	        view_2612: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_27_fc1_parametrizations_weight_original1 = None
	        view_2613: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_27_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1502 = torch.ops.aten._assert_tensor_metadata.default(view_2611, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1502 = None
	        convert_element_type_1000: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2611, torch.float32);  view_2611 = None
	        _assert_tensor_metadata_1503 = torch.ops.aten._assert_tensor_metadata.default(view_2613, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1503 = None
	        convert_element_type_1001: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2613, torch.float32);  view_2613 = None
	        sub_7630: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1000, convert_element_type_1001);  convert_element_type_1000 = convert_element_type_1001 = None
	        mul_16154: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7630, view_2612);  sub_7630 = view_2612 = None
	        view_2614: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16154, [5120, 1280]);  mul_16154 = None
	        _assert_tensor_metadata_1504 = torch.ops.aten._assert_tensor_metadata.default(view_2614, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1504 = None
	        mul_16159: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2615: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16149, [mul_16159, 1280]);  mul_16149 = mul_16159 = None
	        permute_279: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2614, [1, 0]);  view_2614 = None
	        addmm_138: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_27_fc1_bias, view_2615, permute_279);  model_audio_tower_layers_27_fc1_bias = view_2615 = permute_279 = None
	        view_2616: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_138, [sym_size_int, 1500, 5120]);  addmm_138 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_16166: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2616, 0.5)
	        mul_16167: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2616, 0.7071067811865476);  view_2616 = None
	        erf_29: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_16167);  mul_16167 = None
	        add_25584: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_29, 1);  erf_29 = None
	        mul_16168: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16166, add_25584);  mul_16166 = add_25584 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_16168, [2])
	        amax_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_16168, [2])
	        full_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_167, full_334);  amin_167 = full_334 = None
	        full_335: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_167, full_335);  amax_167 = full_335 = None
	        sub_7643: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_167, minimum_167);  maximum_167 = None
	        div_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7643, 255.0);  sub_7643 = None
	        clamp_min_501: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_334, 1.1920928955078125e-07);  div_334 = None
	        div_335: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_167, clamp_min_501);  minimum_167 = None
	        round_335: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_335);  div_335 = None
	        sub_7649: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_335);  round_335 = None
	        clamp_min_502: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7649, -128);  sub_7649 = None
	        clamp_max_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_502, 127);  clamp_min_502 = None
	        _assert_tensor_metadata_1505 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_501, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1505 = None
	        _assert_tensor_metadata_1506 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_334, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1506 = None
	        convert_element_type_1002: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_334, torch.int8);  clamp_max_334 = None
	        view_2619: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_501, [sym_size_int, 1500, 1])
	        view_2620: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1002, [sym_size_int, 1500, 1])
	        reciprocal_167: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2619);  view_2619 = None
	        mul_16214: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_167, 1.0);  reciprocal_167 = None
	        mul_16217: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16168, mul_16214);  mul_16168 = mul_16214 = None
	        round_336: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_16217);  mul_16217 = None
	        add_25667: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_336, view_2620);  round_336 = view_2620 = None
	        clamp_min_503: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25667, -128);  add_25667 = None
	        clamp_max_335: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_503, 127);  clamp_min_503 = None
	        _assert_tensor_metadata_1507 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_335, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1507 = None
	        convert_element_type_1003: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_335, torch.int8);  clamp_max_335 = None
	        view_2623: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_501, [sym_size_int, 1500, 1]);  clamp_min_501 = None
	        view_2624: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1002, [sym_size_int, 1500, 1]);  convert_element_type_1002 = None
	        _assert_tensor_metadata_1508 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1003, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1508 = None
	        convert_element_type_1004: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1003, torch.float32);  convert_element_type_1003 = None
	        _assert_tensor_metadata_1509 = torch.ops.aten._assert_tensor_metadata.default(view_2624, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1509 = None
	        convert_element_type_1005: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2624, torch.float32);  view_2624 = None
	        sub_7669: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1004, convert_element_type_1005);  convert_element_type_1004 = convert_element_type_1005 = None
	        mul_16239: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7669, view_2623);  sub_7669 = view_2623 = None
	        _assert_tensor_metadata_1510 = torch.ops.aten._assert_tensor_metadata.default(mul_16239, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1510 = None
	        view_2626: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_27_fc2_parametrizations_weight_original0 = None
	        view_2627: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_27_fc2_parametrizations_weight_original1 = None
	        view_2628: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_27_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_27_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1511 = torch.ops.aten._assert_tensor_metadata.default(view_2626, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1511 = None
	        convert_element_type_1006: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2626, torch.float32);  view_2626 = None
	        _assert_tensor_metadata_1512 = torch.ops.aten._assert_tensor_metadata.default(view_2628, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1512 = None
	        convert_element_type_1007: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2628, torch.float32);  view_2628 = None
	        sub_7673: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1006, convert_element_type_1007);  convert_element_type_1006 = convert_element_type_1007 = None
	        mul_16244: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7673, view_2627);  sub_7673 = view_2627 = None
	        view_2629: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_16244, [1280, 5120]);  mul_16244 = None
	        _assert_tensor_metadata_1513 = torch.ops.aten._assert_tensor_metadata.default(view_2629, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1513 = None
	        mul_16249: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2630: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_16239, [mul_16249, 5120]);  mul_16239 = mul_16249 = None
	        permute_280: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2629, [1, 0]);  view_2629 = None
	        addmm_139: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_27_fc2_bias, view_2630, permute_280);  model_audio_tower_layers_27_fc2_bias = view_2630 = permute_280 = None
	        view_2631: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_139, [sym_size_int, 1500, 1280]);  addmm_139 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_25730: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_25432, view_2631);  add_25432 = view_2631 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_225: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_25730, memory_format = torch.contiguous_format)
	        var_mean_56 = torch.ops.aten.var_mean.correction(clone_225, [2], correction = 0, keepdim = True)
	        getitem_224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_56[0]
	        getitem_225: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_56[1];  var_mean_56 = None
	        add_25735: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_224, 1e-05);  getitem_224 = None
	        rsqrt_56: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_25735);  add_25735 = None
	        sub_7679: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_225, getitem_225);  clone_225 = getitem_225 = None
	        mul_16260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7679, rsqrt_56);  sub_7679 = rsqrt_56 = None
	        mul_16261: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16260, model_audio_tower_layers_28_self_attn_layer_norm_weight);  mul_16260 = model_audio_tower_layers_28_self_attn_layer_norm_weight = None
	        add_25736: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16261, model_audio_tower_layers_28_self_attn_layer_norm_bias);  mul_16261 = model_audio_tower_layers_28_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_25736, [2])
	        amax_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_25736, [2])
	        full_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_168, full_336);  amin_168 = full_336 = None
	        full_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_168, full_337);  amax_168 = full_337 = None
	        sub_7690: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_168, minimum_168);  maximum_168 = None
	        div_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7690, 255.0);  sub_7690 = None
	        clamp_min_504: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_336, 1.1920928955078125e-07);  div_336 = None
	        div_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_168, clamp_min_504);  minimum_168 = None
	        round_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_337);  div_337 = None
	        sub_7696: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_337);  round_337 = None
	        clamp_min_505: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7696, -128);  sub_7696 = None
	        clamp_max_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_505, 127);  clamp_min_505 = None
	        _assert_tensor_metadata_1514 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_504, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1514 = None
	        _assert_tensor_metadata_1515 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_336, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1515 = None
	        convert_element_type_1008: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_336, torch.int8);  clamp_max_336 = None
	        view_2634: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_504, [sym_size_int, 1500, 1])
	        view_2635: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1008, [sym_size_int, 1500, 1])
	        reciprocal_168: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2634);  view_2634 = None
	        mul_16309: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_168, 1.0);  reciprocal_168 = None
	        mul_16312: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_25736, mul_16309);  mul_16309 = None
	        round_338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16312);  mul_16312 = None
	        add_25823: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_338, view_2635);  round_338 = view_2635 = None
	        clamp_min_506: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25823, -128);  add_25823 = None
	        clamp_max_337: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_506, 127);  clamp_min_506 = None
	        _assert_tensor_metadata_1516 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_337, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1516 = None
	        convert_element_type_1009: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_337, torch.int8);  clamp_max_337 = None
	        view_2638: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_504, [sym_size_int, 1500, 1]);  clamp_min_504 = None
	        view_2639: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1008, [sym_size_int, 1500, 1]);  convert_element_type_1008 = None
	        _assert_tensor_metadata_1517 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1009, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1517 = None
	        convert_element_type_1010: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1009, torch.float32);  convert_element_type_1009 = None
	        _assert_tensor_metadata_1518 = torch.ops.aten._assert_tensor_metadata.default(view_2639, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1518 = None
	        convert_element_type_1011: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2639, torch.float32);  view_2639 = None
	        sub_7716: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1010, convert_element_type_1011);  convert_element_type_1010 = convert_element_type_1011 = None
	        mul_16334: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7716, view_2638);  sub_7716 = view_2638 = None
	        _assert_tensor_metadata_1519 = torch.ops.aten._assert_tensor_metadata.default(mul_16334, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1519 = None
	        view_2641: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2642: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2643: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1520 = torch.ops.aten._assert_tensor_metadata.default(view_2641, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1520 = None
	        convert_element_type_1012: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2641, torch.float32);  view_2641 = None
	        _assert_tensor_metadata_1521 = torch.ops.aten._assert_tensor_metadata.default(view_2643, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1521 = None
	        convert_element_type_1013: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2643, torch.float32);  view_2643 = None
	        sub_7720: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1012, convert_element_type_1013);  convert_element_type_1012 = convert_element_type_1013 = None
	        mul_16339: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7720, view_2642);  sub_7720 = view_2642 = None
	        view_2644: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16339, [1280, 1280]);  mul_16339 = None
	        _assert_tensor_metadata_1522 = torch.ops.aten._assert_tensor_metadata.default(view_2644, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1522 = None
	        mul_16344: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2645: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16334, [mul_16344, 1280]);  mul_16334 = mul_16344 = None
	        permute_281: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2644, [1, 0]);  view_2644 = None
	        addmm_140: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_28_self_attn_q_proj_bias, view_2645, permute_281);  model_audio_tower_layers_28_self_attn_q_proj_bias = view_2645 = permute_281 = None
	        view_2646: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_140, [sym_size_int, 1500, 1280]);  addmm_140 = None
	        mul_16351: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2646, 0.125);  view_2646 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2647: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_16351, [sym_size_int, 1500, 20, 64]);  mul_16351 = None
	        permute_282: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2647, [0, 2, 1, 3]);  view_2647 = None
	        clone_226: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_282, memory_format = torch.contiguous_format);  permute_282 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_25736, [2])
	        amax_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_25736, [2])
	        full_338: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_169, full_338);  amin_169 = full_338 = None
	        full_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_169, full_339);  amax_169 = full_339 = None
	        sub_7735: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_169, minimum_169);  maximum_169 = None
	        div_338: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7735, 255.0);  sub_7735 = None
	        clamp_min_507: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_338, 1.1920928955078125e-07);  div_338 = None
	        div_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_169, clamp_min_507);  minimum_169 = None
	        round_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_339);  div_339 = None
	        sub_7741: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_339);  round_339 = None
	        clamp_min_508: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7741, -128);  sub_7741 = None
	        clamp_max_338: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_508, 127);  clamp_min_508 = None
	        _assert_tensor_metadata_1523 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_507, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1523 = None
	        _assert_tensor_metadata_1524 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_338, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1524 = None
	        convert_element_type_1014: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_338, torch.int8);  clamp_max_338 = None
	        view_2650: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_507, [sym_size_int, 1500, 1])
	        view_2651: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1014, [sym_size_int, 1500, 1])
	        reciprocal_169: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2650);  view_2650 = None
	        mul_16405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_169, 1.0);  reciprocal_169 = None
	        mul_16408: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_25736, mul_16405);  mul_16405 = None
	        round_340: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16408);  mul_16408 = None
	        add_25975: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_340, view_2651);  round_340 = view_2651 = None
	        clamp_min_509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25975, -128);  add_25975 = None
	        clamp_max_339: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_509, 127);  clamp_min_509 = None
	        _assert_tensor_metadata_1525 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_339, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1525 = None
	        convert_element_type_1015: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_339, torch.int8);  clamp_max_339 = None
	        view_2654: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_507, [sym_size_int, 1500, 1]);  clamp_min_507 = None
	        view_2655: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1014, [sym_size_int, 1500, 1]);  convert_element_type_1014 = None
	        _assert_tensor_metadata_1526 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1015, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1526 = None
	        convert_element_type_1016: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1015, torch.float32);  convert_element_type_1015 = None
	        _assert_tensor_metadata_1527 = torch.ops.aten._assert_tensor_metadata.default(view_2655, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1527 = None
	        convert_element_type_1017: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2655, torch.float32);  view_2655 = None
	        sub_7761: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1016, convert_element_type_1017);  convert_element_type_1016 = convert_element_type_1017 = None
	        mul_16430: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7761, view_2654);  sub_7761 = view_2654 = None
	        _assert_tensor_metadata_1528 = torch.ops.aten._assert_tensor_metadata.default(mul_16430, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1528 = None
	        view_2657: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2658: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2659: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1529 = torch.ops.aten._assert_tensor_metadata.default(view_2657, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1529 = None
	        convert_element_type_1018: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2657, torch.float32);  view_2657 = None
	        _assert_tensor_metadata_1530 = torch.ops.aten._assert_tensor_metadata.default(view_2659, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1530 = None
	        convert_element_type_1019: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2659, torch.float32);  view_2659 = None
	        sub_7765: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1018, convert_element_type_1019);  convert_element_type_1018 = convert_element_type_1019 = None
	        mul_16435: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7765, view_2658);  sub_7765 = view_2658 = None
	        view_2660: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16435, [1280, 1280]);  mul_16435 = None
	        _assert_tensor_metadata_1531 = torch.ops.aten._assert_tensor_metadata.default(view_2660, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1531 = None
	        permute_283: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2660, [1, 0]);  view_2660 = None
	        mul_16438: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2661: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16430, [mul_16438, 1280]);  mul_16430 = mul_16438 = None
	        mm_28: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2661, permute_283);  view_2661 = permute_283 = None
	        view_2662: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_28, [sym_size_int, 1500, 1280]);  mm_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2663: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2662, [sym_size_int, -1, 20, 64]);  view_2662 = None
	        permute_284: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2663, [0, 2, 1, 3]);  view_2663 = None
	        clone_227: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_284, memory_format = torch.contiguous_format);  permute_284 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_25736, [2])
	        amax_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_25736, [2])
	        full_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_170, full_340);  amin_170 = full_340 = None
	        full_341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_170, full_341);  amax_170 = full_341 = None
	        sub_7779: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_170, minimum_170);  maximum_170 = None
	        div_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7779, 255.0);  sub_7779 = None
	        clamp_min_510: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_340, 1.1920928955078125e-07);  div_340 = None
	        div_341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_170, clamp_min_510);  minimum_170 = None
	        round_341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_341);  div_341 = None
	        sub_7785: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_341);  round_341 = None
	        clamp_min_511: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7785, -128);  sub_7785 = None
	        clamp_max_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_511, 127);  clamp_min_511 = None
	        _assert_tensor_metadata_1532 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_510, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1532 = None
	        _assert_tensor_metadata_1533 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_340, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1533 = None
	        convert_element_type_1020: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_340, torch.int8);  clamp_max_340 = None
	        view_2666: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_510, [sym_size_int, 1500, 1])
	        view_2667: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1020, [sym_size_int, 1500, 1])
	        reciprocal_170: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2666);  view_2666 = None
	        mul_16504: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_170, 1.0);  reciprocal_170 = None
	        mul_16507: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_25736, mul_16504);  add_25736 = mul_16504 = None
	        round_342: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16507);  mul_16507 = None
	        add_26123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_342, view_2667);  round_342 = view_2667 = None
	        clamp_min_512: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26123, -128);  add_26123 = None
	        clamp_max_341: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_512, 127);  clamp_min_512 = None
	        _assert_tensor_metadata_1534 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_341, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1534 = None
	        convert_element_type_1021: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_341, torch.int8);  clamp_max_341 = None
	        view_2670: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_510, [sym_size_int, 1500, 1]);  clamp_min_510 = None
	        view_2671: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1020, [sym_size_int, 1500, 1]);  convert_element_type_1020 = None
	        _assert_tensor_metadata_1535 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1021, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1535 = None
	        convert_element_type_1022: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1021, torch.float32);  convert_element_type_1021 = None
	        _assert_tensor_metadata_1536 = torch.ops.aten._assert_tensor_metadata.default(view_2671, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1536 = None
	        convert_element_type_1023: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2671, torch.float32);  view_2671 = None
	        sub_7805: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1022, convert_element_type_1023);  convert_element_type_1022 = convert_element_type_1023 = None
	        mul_16529: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7805, view_2670);  sub_7805 = view_2670 = None
	        _assert_tensor_metadata_1537 = torch.ops.aten._assert_tensor_metadata.default(mul_16529, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1537 = None
	        view_2673: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2674: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2675: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1538 = torch.ops.aten._assert_tensor_metadata.default(view_2673, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1538 = None
	        convert_element_type_1024: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2673, torch.float32);  view_2673 = None
	        _assert_tensor_metadata_1539 = torch.ops.aten._assert_tensor_metadata.default(view_2675, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1539 = None
	        convert_element_type_1025: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2675, torch.float32);  view_2675 = None
	        sub_7809: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1024, convert_element_type_1025);  convert_element_type_1024 = convert_element_type_1025 = None
	        mul_16534: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7809, view_2674);  sub_7809 = view_2674 = None
	        view_2676: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16534, [1280, 1280]);  mul_16534 = None
	        _assert_tensor_metadata_1540 = torch.ops.aten._assert_tensor_metadata.default(view_2676, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1540 = None
	        mul_16539: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2677: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16529, [mul_16539, 1280]);  mul_16529 = mul_16539 = None
	        permute_285: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2676, [1, 0]);  view_2676 = None
	        addmm_141: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_28_self_attn_v_proj_bias, view_2677, permute_285);  model_audio_tower_layers_28_self_attn_v_proj_bias = view_2677 = permute_285 = None
	        view_2678: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_141, [sym_size_int, 1500, 1280]);  addmm_141 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2679: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2678, [sym_size_int, -1, 20, 64]);  view_2678 = None
	        permute_286: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2679, [0, 2, 1, 3]);  view_2679 = None
	        clone_228: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_286, memory_format = torch.contiguous_format);  permute_286 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_28 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_226, clone_227, clone_228, None, False, scale = 1.0);  clone_226 = clone_227 = clone_228 = None
	        getitem_226: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_28[0];  _scaled_dot_product_efficient_attention_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_287: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_226, [0, 2, 1, 3]);  getitem_226 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2680: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_287, [sym_size_int, 1500, -1]);  permute_287 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2680, [2])
	        amax_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2680, [2])
	        full_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_171, full_342);  amin_171 = full_342 = None
	        full_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_171, full_343);  amax_171 = full_343 = None
	        sub_7827: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_171, minimum_171);  maximum_171 = None
	        div_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7827, 255.0);  sub_7827 = None
	        clamp_min_513: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_342, 1.1920928955078125e-07);  div_342 = None
	        div_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_171, clamp_min_513);  minimum_171 = None
	        round_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_343);  div_343 = None
	        sub_7833: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_343);  round_343 = None
	        clamp_min_514: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7833, -128);  sub_7833 = None
	        clamp_max_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_514, 127);  clamp_min_514 = None
	        _assert_tensor_metadata_1541 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_513, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1541 = None
	        _assert_tensor_metadata_1542 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_342, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1542 = None
	        convert_element_type_1026: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_342, torch.int8);  clamp_max_342 = None
	        view_2683: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_513, [sym_size_int, 1500, 1])
	        view_2684: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1026, [sym_size_int, 1500, 1])
	        reciprocal_171: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2683);  view_2683 = None
	        mul_16609: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_171, 1.0);  reciprocal_171 = None
	        mul_16612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2680, mul_16609);  view_2680 = mul_16609 = None
	        round_344: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16612);  mul_16612 = None
	        add_26287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_344, view_2684);  round_344 = view_2684 = None
	        clamp_min_515: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26287, -128);  add_26287 = None
	        clamp_max_343: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_515, 127);  clamp_min_515 = None
	        _assert_tensor_metadata_1543 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_343, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1543 = None
	        convert_element_type_1027: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_343, torch.int8);  clamp_max_343 = None
	        view_2687: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_513, [sym_size_int, 1500, 1]);  clamp_min_513 = None
	        view_2688: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1026, [sym_size_int, 1500, 1]);  convert_element_type_1026 = None
	        _assert_tensor_metadata_1544 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1027, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1544 = None
	        convert_element_type_1028: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1027, torch.float32);  convert_element_type_1027 = None
	        _assert_tensor_metadata_1545 = torch.ops.aten._assert_tensor_metadata.default(view_2688, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1545 = None
	        convert_element_type_1029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2688, torch.float32);  view_2688 = None
	        sub_7853: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1028, convert_element_type_1029);  convert_element_type_1028 = convert_element_type_1029 = None
	        mul_16634: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7853, view_2687);  sub_7853 = view_2687 = None
	        _assert_tensor_metadata_1546 = torch.ops.aten._assert_tensor_metadata.default(mul_16634, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1546 = None
	        view_2690: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2691: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2692: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1547 = torch.ops.aten._assert_tensor_metadata.default(view_2690, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1547 = None
	        convert_element_type_1030: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2690, torch.float32);  view_2690 = None
	        _assert_tensor_metadata_1548 = torch.ops.aten._assert_tensor_metadata.default(view_2692, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1548 = None
	        convert_element_type_1031: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2692, torch.float32);  view_2692 = None
	        sub_7857: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1030, convert_element_type_1031);  convert_element_type_1030 = convert_element_type_1031 = None
	        mul_16639: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7857, view_2691);  sub_7857 = view_2691 = None
	        view_2693: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16639, [1280, 1280]);  mul_16639 = None
	        _assert_tensor_metadata_1549 = torch.ops.aten._assert_tensor_metadata.default(view_2693, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1549 = None
	        mul_16644: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2694: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16634, [mul_16644, 1280]);  mul_16634 = mul_16644 = None
	        permute_288: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2693, [1, 0]);  view_2693 = None
	        addmm_142: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_28_self_attn_out_proj_bias, view_2694, permute_288);  model_audio_tower_layers_28_self_attn_out_proj_bias = view_2694 = permute_288 = None
	        view_2695: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_142, [sym_size_int, 1500, 1280]);  addmm_142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_26350: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_25730, view_2695);  add_25730 = view_2695 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_26350, memory_format = torch.contiguous_format)
	        var_mean_57 = torch.ops.aten.var_mean.correction(clone_230, [2], correction = 0, keepdim = True)
	        getitem_230: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_57[0]
	        getitem_231: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_57[1];  var_mean_57 = None
	        add_26355: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_230, 1e-05);  getitem_230 = None
	        rsqrt_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_26355);  add_26355 = None
	        sub_7863: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_230, getitem_231);  clone_230 = getitem_231 = None
	        mul_16655: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7863, rsqrt_57);  sub_7863 = rsqrt_57 = None
	        mul_16656: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16655, model_audio_tower_layers_28_final_layer_norm_weight);  mul_16655 = model_audio_tower_layers_28_final_layer_norm_weight = None
	        add_26356: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16656, model_audio_tower_layers_28_final_layer_norm_bias);  mul_16656 = model_audio_tower_layers_28_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_26356, [2])
	        amax_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_26356, [2])
	        full_344: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_172, full_344);  amin_172 = full_344 = None
	        full_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_172, full_345);  amax_172 = full_345 = None
	        sub_7874: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_172, minimum_172);  maximum_172 = None
	        div_344: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7874, 255.0);  sub_7874 = None
	        clamp_min_516: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_344, 1.1920928955078125e-07);  div_344 = None
	        div_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_172, clamp_min_516);  minimum_172 = None
	        round_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_345);  div_345 = None
	        sub_7880: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_345);  round_345 = None
	        clamp_min_517: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7880, -128);  sub_7880 = None
	        clamp_max_344: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_517, 127);  clamp_min_517 = None
	        _assert_tensor_metadata_1550 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_516, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1550 = None
	        _assert_tensor_metadata_1551 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_344, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1551 = None
	        convert_element_type_1032: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_344, torch.int8);  clamp_max_344 = None
	        view_2698: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_516, [sym_size_int, 1500, 1])
	        view_2699: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1032, [sym_size_int, 1500, 1])
	        reciprocal_172: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2698);  view_2698 = None
	        mul_16704: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_172, 1.0);  reciprocal_172 = None
	        mul_16707: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_26356, mul_16704);  add_26356 = mul_16704 = None
	        round_346: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16707);  mul_16707 = None
	        add_26443: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_346, view_2699);  round_346 = view_2699 = None
	        clamp_min_518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26443, -128);  add_26443 = None
	        clamp_max_345: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_518, 127);  clamp_min_518 = None
	        _assert_tensor_metadata_1552 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_345, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1552 = None
	        convert_element_type_1033: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_345, torch.int8);  clamp_max_345 = None
	        view_2702: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_516, [sym_size_int, 1500, 1]);  clamp_min_516 = None
	        view_2703: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1032, [sym_size_int, 1500, 1]);  convert_element_type_1032 = None
	        _assert_tensor_metadata_1553 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1033, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1553 = None
	        convert_element_type_1034: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1033, torch.float32);  convert_element_type_1033 = None
	        _assert_tensor_metadata_1554 = torch.ops.aten._assert_tensor_metadata.default(view_2703, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1554 = None
	        convert_element_type_1035: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2703, torch.float32);  view_2703 = None
	        sub_7900: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1034, convert_element_type_1035);  convert_element_type_1034 = convert_element_type_1035 = None
	        mul_16729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7900, view_2702);  sub_7900 = view_2702 = None
	        _assert_tensor_metadata_1555 = torch.ops.aten._assert_tensor_metadata.default(mul_16729, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1555 = None
	        view_2705: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_28_fc1_parametrizations_weight_original0 = None
	        view_2706: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_28_fc1_parametrizations_weight_original1 = None
	        view_2707: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_28_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1556 = torch.ops.aten._assert_tensor_metadata.default(view_2705, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1556 = None
	        convert_element_type_1036: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2705, torch.float32);  view_2705 = None
	        _assert_tensor_metadata_1557 = torch.ops.aten._assert_tensor_metadata.default(view_2707, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1557 = None
	        convert_element_type_1037: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2707, torch.float32);  view_2707 = None
	        sub_7904: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1036, convert_element_type_1037);  convert_element_type_1036 = convert_element_type_1037 = None
	        mul_16734: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7904, view_2706);  sub_7904 = view_2706 = None
	        view_2708: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16734, [5120, 1280]);  mul_16734 = None
	        _assert_tensor_metadata_1558 = torch.ops.aten._assert_tensor_metadata.default(view_2708, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1558 = None
	        mul_16739: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2709: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16729, [mul_16739, 1280]);  mul_16729 = mul_16739 = None
	        permute_289: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2708, [1, 0]);  view_2708 = None
	        addmm_143: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_28_fc1_bias, view_2709, permute_289);  model_audio_tower_layers_28_fc1_bias = view_2709 = permute_289 = None
	        view_2710: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_143, [sym_size_int, 1500, 5120]);  addmm_143 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_16746: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2710, 0.5)
	        mul_16747: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2710, 0.7071067811865476);  view_2710 = None
	        erf_30: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_16747);  mul_16747 = None
	        add_26502: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_30, 1);  erf_30 = None
	        mul_16748: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16746, add_26502);  mul_16746 = add_26502 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_16748, [2])
	        amax_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_16748, [2])
	        full_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_173, full_346);  amin_173 = full_346 = None
	        full_347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_173, full_347);  amax_173 = full_347 = None
	        sub_7917: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_173, minimum_173);  maximum_173 = None
	        div_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7917, 255.0);  sub_7917 = None
	        clamp_min_519: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_346, 1.1920928955078125e-07);  div_346 = None
	        div_347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_173, clamp_min_519);  minimum_173 = None
	        round_347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_347);  div_347 = None
	        sub_7923: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_347);  round_347 = None
	        clamp_min_520: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7923, -128);  sub_7923 = None
	        clamp_max_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_520, 127);  clamp_min_520 = None
	        _assert_tensor_metadata_1559 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_519, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1559 = None
	        _assert_tensor_metadata_1560 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_346, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1560 = None
	        convert_element_type_1038: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_346, torch.int8);  clamp_max_346 = None
	        view_2713: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_519, [sym_size_int, 1500, 1])
	        view_2714: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1038, [sym_size_int, 1500, 1])
	        reciprocal_173: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2713);  view_2713 = None
	        mul_16794: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_173, 1.0);  reciprocal_173 = None
	        mul_16797: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16748, mul_16794);  mul_16748 = mul_16794 = None
	        round_348: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_16797);  mul_16797 = None
	        add_26585: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_348, view_2714);  round_348 = view_2714 = None
	        clamp_min_521: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26585, -128);  add_26585 = None
	        clamp_max_347: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_521, 127);  clamp_min_521 = None
	        _assert_tensor_metadata_1561 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_347, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1561 = None
	        convert_element_type_1039: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_347, torch.int8);  clamp_max_347 = None
	        view_2717: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_519, [sym_size_int, 1500, 1]);  clamp_min_519 = None
	        view_2718: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1038, [sym_size_int, 1500, 1]);  convert_element_type_1038 = None
	        _assert_tensor_metadata_1562 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1039, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1562 = None
	        convert_element_type_1040: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1039, torch.float32);  convert_element_type_1039 = None
	        _assert_tensor_metadata_1563 = torch.ops.aten._assert_tensor_metadata.default(view_2718, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1563 = None
	        convert_element_type_1041: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2718, torch.float32);  view_2718 = None
	        sub_7943: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1040, convert_element_type_1041);  convert_element_type_1040 = convert_element_type_1041 = None
	        mul_16819: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7943, view_2717);  sub_7943 = view_2717 = None
	        _assert_tensor_metadata_1564 = torch.ops.aten._assert_tensor_metadata.default(mul_16819, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1564 = None
	        view_2720: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_28_fc2_parametrizations_weight_original0 = None
	        view_2721: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_28_fc2_parametrizations_weight_original1 = None
	        view_2722: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_28_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_28_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1565 = torch.ops.aten._assert_tensor_metadata.default(view_2720, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1565 = None
	        convert_element_type_1042: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2720, torch.float32);  view_2720 = None
	        _assert_tensor_metadata_1566 = torch.ops.aten._assert_tensor_metadata.default(view_2722, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1566 = None
	        convert_element_type_1043: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2722, torch.float32);  view_2722 = None
	        sub_7947: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1042, convert_element_type_1043);  convert_element_type_1042 = convert_element_type_1043 = None
	        mul_16824: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7947, view_2721);  sub_7947 = view_2721 = None
	        view_2723: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_16824, [1280, 5120]);  mul_16824 = None
	        _assert_tensor_metadata_1567 = torch.ops.aten._assert_tensor_metadata.default(view_2723, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1567 = None
	        mul_16829: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2724: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_16819, [mul_16829, 5120]);  mul_16819 = mul_16829 = None
	        permute_290: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2723, [1, 0]);  view_2723 = None
	        addmm_144: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_28_fc2_bias, view_2724, permute_290);  model_audio_tower_layers_28_fc2_bias = view_2724 = permute_290 = None
	        view_2725: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_144, [sym_size_int, 1500, 1280]);  addmm_144 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_26648: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_26350, view_2725);  add_26350 = view_2725 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_233: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_26648, memory_format = torch.contiguous_format)
	        var_mean_58 = torch.ops.aten.var_mean.correction(clone_233, [2], correction = 0, keepdim = True)
	        getitem_232: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_58[0]
	        getitem_233: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_58[1];  var_mean_58 = None
	        add_26653: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_232, 1e-05);  getitem_232 = None
	        rsqrt_58: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_26653);  add_26653 = None
	        sub_7953: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_233, getitem_233);  clone_233 = getitem_233 = None
	        mul_16840: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7953, rsqrt_58);  sub_7953 = rsqrt_58 = None
	        mul_16841: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16840, model_audio_tower_layers_29_self_attn_layer_norm_weight);  mul_16840 = model_audio_tower_layers_29_self_attn_layer_norm_weight = None
	        add_26654: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16841, model_audio_tower_layers_29_self_attn_layer_norm_bias);  mul_16841 = model_audio_tower_layers_29_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_26654, [2])
	        amax_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_26654, [2])
	        full_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_174, full_348);  amin_174 = full_348 = None
	        full_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_174, full_349);  amax_174 = full_349 = None
	        sub_7964: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_174, minimum_174);  maximum_174 = None
	        div_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7964, 255.0);  sub_7964 = None
	        clamp_min_522: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_348, 1.1920928955078125e-07);  div_348 = None
	        div_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_174, clamp_min_522);  minimum_174 = None
	        round_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_349);  div_349 = None
	        sub_7970: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_349);  round_349 = None
	        clamp_min_523: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7970, -128);  sub_7970 = None
	        clamp_max_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_523, 127);  clamp_min_523 = None
	        _assert_tensor_metadata_1568 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_522, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1568 = None
	        _assert_tensor_metadata_1569 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_348, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1569 = None
	        convert_element_type_1044: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_348, torch.int8);  clamp_max_348 = None
	        view_2728: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_522, [sym_size_int, 1500, 1])
	        view_2729: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1044, [sym_size_int, 1500, 1])
	        reciprocal_174: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2728);  view_2728 = None
	        mul_16889: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_174, 1.0);  reciprocal_174 = None
	        mul_16892: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_26654, mul_16889);  mul_16889 = None
	        round_350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16892);  mul_16892 = None
	        add_26741: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_350, view_2729);  round_350 = view_2729 = None
	        clamp_min_524: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26741, -128);  add_26741 = None
	        clamp_max_349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_524, 127);  clamp_min_524 = None
	        _assert_tensor_metadata_1570 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_349, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1570 = None
	        convert_element_type_1045: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_349, torch.int8);  clamp_max_349 = None
	        view_2732: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_522, [sym_size_int, 1500, 1]);  clamp_min_522 = None
	        view_2733: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1044, [sym_size_int, 1500, 1]);  convert_element_type_1044 = None
	        _assert_tensor_metadata_1571 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1045, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1571 = None
	        convert_element_type_1046: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1045, torch.float32);  convert_element_type_1045 = None
	        _assert_tensor_metadata_1572 = torch.ops.aten._assert_tensor_metadata.default(view_2733, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1572 = None
	        convert_element_type_1047: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2733, torch.float32);  view_2733 = None
	        sub_7990: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1046, convert_element_type_1047);  convert_element_type_1046 = convert_element_type_1047 = None
	        mul_16914: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7990, view_2732);  sub_7990 = view_2732 = None
	        _assert_tensor_metadata_1573 = torch.ops.aten._assert_tensor_metadata.default(mul_16914, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1573 = None
	        view_2735: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2736: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2737: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1574 = torch.ops.aten._assert_tensor_metadata.default(view_2735, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1574 = None
	        convert_element_type_1048: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2735, torch.float32);  view_2735 = None
	        _assert_tensor_metadata_1575 = torch.ops.aten._assert_tensor_metadata.default(view_2737, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1575 = None
	        convert_element_type_1049: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2737, torch.float32);  view_2737 = None
	        sub_7994: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1048, convert_element_type_1049);  convert_element_type_1048 = convert_element_type_1049 = None
	        mul_16919: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7994, view_2736);  sub_7994 = view_2736 = None
	        view_2738: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16919, [1280, 1280]);  mul_16919 = None
	        _assert_tensor_metadata_1576 = torch.ops.aten._assert_tensor_metadata.default(view_2738, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1576 = None
	        mul_16924: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2739: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_16914, [mul_16924, 1280]);  mul_16914 = mul_16924 = None
	        permute_291: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2738, [1, 0]);  view_2738 = None
	        addmm_145: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_29_self_attn_q_proj_bias, view_2739, permute_291);  model_audio_tower_layers_29_self_attn_q_proj_bias = view_2739 = permute_291 = None
	        view_2740: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_145, [sym_size_int, 1500, 1280]);  addmm_145 = None
	        mul_16931: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2740, 0.125);  view_2740 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2741: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_16931, [sym_size_int, 1500, 20, 64]);  mul_16931 = None
	        permute_292: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2741, [0, 2, 1, 3]);  view_2741 = None
	        clone_234: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_292, memory_format = torch.contiguous_format);  permute_292 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_26654, [2])
	        amax_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_26654, [2])
	        full_350: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_175, full_350);  amin_175 = full_350 = None
	        full_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_175, full_351);  amax_175 = full_351 = None
	        sub_8009: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_175, minimum_175);  maximum_175 = None
	        div_350: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8009, 255.0);  sub_8009 = None
	        clamp_min_525: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_350, 1.1920928955078125e-07);  div_350 = None
	        div_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_175, clamp_min_525);  minimum_175 = None
	        round_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_351);  div_351 = None
	        sub_8015: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_351);  round_351 = None
	        clamp_min_526: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8015, -128);  sub_8015 = None
	        clamp_max_350: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_526, 127);  clamp_min_526 = None
	        _assert_tensor_metadata_1577 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_525, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1577 = None
	        _assert_tensor_metadata_1578 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_350, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1578 = None
	        convert_element_type_1050: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_350, torch.int8);  clamp_max_350 = None
	        view_2744: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_525, [sym_size_int, 1500, 1])
	        view_2745: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1050, [sym_size_int, 1500, 1])
	        reciprocal_175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2744);  view_2744 = None
	        mul_16985: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_175, 1.0);  reciprocal_175 = None
	        mul_16988: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_26654, mul_16985);  mul_16985 = None
	        round_352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16988);  mul_16988 = None
	        add_26893: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_352, view_2745);  round_352 = view_2745 = None
	        clamp_min_527: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26893, -128);  add_26893 = None
	        clamp_max_351: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_527, 127);  clamp_min_527 = None
	        _assert_tensor_metadata_1579 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_351, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1579 = None
	        convert_element_type_1051: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_351, torch.int8);  clamp_max_351 = None
	        view_2748: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_525, [sym_size_int, 1500, 1]);  clamp_min_525 = None
	        view_2749: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1050, [sym_size_int, 1500, 1]);  convert_element_type_1050 = None
	        _assert_tensor_metadata_1580 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1051, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1580 = None
	        convert_element_type_1052: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1051, torch.float32);  convert_element_type_1051 = None
	        _assert_tensor_metadata_1581 = torch.ops.aten._assert_tensor_metadata.default(view_2749, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1581 = None
	        convert_element_type_1053: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2749, torch.float32);  view_2749 = None
	        sub_8035: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1052, convert_element_type_1053);  convert_element_type_1052 = convert_element_type_1053 = None
	        mul_17010: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8035, view_2748);  sub_8035 = view_2748 = None
	        _assert_tensor_metadata_1582 = torch.ops.aten._assert_tensor_metadata.default(mul_17010, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1582 = None
	        view_2751: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2752: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2753: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1583 = torch.ops.aten._assert_tensor_metadata.default(view_2751, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1583 = None
	        convert_element_type_1054: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2751, torch.float32);  view_2751 = None
	        _assert_tensor_metadata_1584 = torch.ops.aten._assert_tensor_metadata.default(view_2753, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1584 = None
	        convert_element_type_1055: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2753, torch.float32);  view_2753 = None
	        sub_8039: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1054, convert_element_type_1055);  convert_element_type_1054 = convert_element_type_1055 = None
	        mul_17015: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8039, view_2752);  sub_8039 = view_2752 = None
	        view_2754: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17015, [1280, 1280]);  mul_17015 = None
	        _assert_tensor_metadata_1585 = torch.ops.aten._assert_tensor_metadata.default(view_2754, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1585 = None
	        permute_293: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2754, [1, 0]);  view_2754 = None
	        mul_17018: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2755: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17010, [mul_17018, 1280]);  mul_17010 = mul_17018 = None
	        mm_29: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2755, permute_293);  view_2755 = permute_293 = None
	        view_2756: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_29, [sym_size_int, 1500, 1280]);  mm_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2757: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2756, [sym_size_int, -1, 20, 64]);  view_2756 = None
	        permute_294: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2757, [0, 2, 1, 3]);  view_2757 = None
	        clone_235: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_294, memory_format = torch.contiguous_format);  permute_294 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_26654, [2])
	        amax_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_26654, [2])
	        full_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_176, full_352);  amin_176 = full_352 = None
	        full_353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_176, full_353);  amax_176 = full_353 = None
	        sub_8053: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_176, minimum_176);  maximum_176 = None
	        div_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8053, 255.0);  sub_8053 = None
	        clamp_min_528: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_352, 1.1920928955078125e-07);  div_352 = None
	        div_353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_176, clamp_min_528);  minimum_176 = None
	        round_353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_353);  div_353 = None
	        sub_8059: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_353);  round_353 = None
	        clamp_min_529: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8059, -128);  sub_8059 = None
	        clamp_max_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_529, 127);  clamp_min_529 = None
	        _assert_tensor_metadata_1586 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_528, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1586 = None
	        _assert_tensor_metadata_1587 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_352, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1587 = None
	        convert_element_type_1056: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_352, torch.int8);  clamp_max_352 = None
	        view_2760: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_528, [sym_size_int, 1500, 1])
	        view_2761: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1056, [sym_size_int, 1500, 1])
	        reciprocal_176: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2760);  view_2760 = None
	        mul_17084: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_176, 1.0);  reciprocal_176 = None
	        mul_17087: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_26654, mul_17084);  add_26654 = mul_17084 = None
	        round_354: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17087);  mul_17087 = None
	        add_27041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_354, view_2761);  round_354 = view_2761 = None
	        clamp_min_530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27041, -128);  add_27041 = None
	        clamp_max_353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_530, 127);  clamp_min_530 = None
	        _assert_tensor_metadata_1588 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_353, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1588 = None
	        convert_element_type_1057: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_353, torch.int8);  clamp_max_353 = None
	        view_2764: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_528, [sym_size_int, 1500, 1]);  clamp_min_528 = None
	        view_2765: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1056, [sym_size_int, 1500, 1]);  convert_element_type_1056 = None
	        _assert_tensor_metadata_1589 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1057, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1589 = None
	        convert_element_type_1058: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1057, torch.float32);  convert_element_type_1057 = None
	        _assert_tensor_metadata_1590 = torch.ops.aten._assert_tensor_metadata.default(view_2765, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1590 = None
	        convert_element_type_1059: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2765, torch.float32);  view_2765 = None
	        sub_8079: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1058, convert_element_type_1059);  convert_element_type_1058 = convert_element_type_1059 = None
	        mul_17109: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8079, view_2764);  sub_8079 = view_2764 = None
	        _assert_tensor_metadata_1591 = torch.ops.aten._assert_tensor_metadata.default(mul_17109, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1591 = None
	        view_2767: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2768: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2769: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1592 = torch.ops.aten._assert_tensor_metadata.default(view_2767, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1592 = None
	        convert_element_type_1060: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2767, torch.float32);  view_2767 = None
	        _assert_tensor_metadata_1593 = torch.ops.aten._assert_tensor_metadata.default(view_2769, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1593 = None
	        convert_element_type_1061: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2769, torch.float32);  view_2769 = None
	        sub_8083: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1060, convert_element_type_1061);  convert_element_type_1060 = convert_element_type_1061 = None
	        mul_17114: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8083, view_2768);  sub_8083 = view_2768 = None
	        view_2770: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17114, [1280, 1280]);  mul_17114 = None
	        _assert_tensor_metadata_1594 = torch.ops.aten._assert_tensor_metadata.default(view_2770, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1594 = None
	        mul_17119: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2771: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17109, [mul_17119, 1280]);  mul_17109 = mul_17119 = None
	        permute_295: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2770, [1, 0]);  view_2770 = None
	        addmm_146: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_29_self_attn_v_proj_bias, view_2771, permute_295);  model_audio_tower_layers_29_self_attn_v_proj_bias = view_2771 = permute_295 = None
	        view_2772: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_146, [sym_size_int, 1500, 1280]);  addmm_146 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2773: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2772, [sym_size_int, -1, 20, 64]);  view_2772 = None
	        permute_296: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2773, [0, 2, 1, 3]);  view_2773 = None
	        clone_236: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_296, memory_format = torch.contiguous_format);  permute_296 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_29 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_234, clone_235, clone_236, None, False, scale = 1.0);  clone_234 = clone_235 = clone_236 = None
	        getitem_234: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_29[0];  _scaled_dot_product_efficient_attention_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_297: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_234, [0, 2, 1, 3]);  getitem_234 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2774: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_297, [sym_size_int, 1500, -1]);  permute_297 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2774, [2])
	        amax_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2774, [2])
	        full_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_177, full_354);  amin_177 = full_354 = None
	        full_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_177, full_355);  amax_177 = full_355 = None
	        sub_8101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_177, minimum_177);  maximum_177 = None
	        div_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8101, 255.0);  sub_8101 = None
	        clamp_min_531: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_354, 1.1920928955078125e-07);  div_354 = None
	        div_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_177, clamp_min_531);  minimum_177 = None
	        round_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_355);  div_355 = None
	        sub_8107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_355);  round_355 = None
	        clamp_min_532: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8107, -128);  sub_8107 = None
	        clamp_max_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_532, 127);  clamp_min_532 = None
	        _assert_tensor_metadata_1595 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_531, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1595 = None
	        _assert_tensor_metadata_1596 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_354, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1596 = None
	        convert_element_type_1062: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_354, torch.int8);  clamp_max_354 = None
	        view_2777: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_531, [sym_size_int, 1500, 1])
	        view_2778: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1062, [sym_size_int, 1500, 1])
	        reciprocal_177: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2777);  view_2777 = None
	        mul_17189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_177, 1.0);  reciprocal_177 = None
	        mul_17192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2774, mul_17189);  view_2774 = mul_17189 = None
	        round_356: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17192);  mul_17192 = None
	        add_27205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_356, view_2778);  round_356 = view_2778 = None
	        clamp_min_533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27205, -128);  add_27205 = None
	        clamp_max_355: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_533, 127);  clamp_min_533 = None
	        _assert_tensor_metadata_1597 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_355, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1597 = None
	        convert_element_type_1063: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_355, torch.int8);  clamp_max_355 = None
	        view_2781: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_531, [sym_size_int, 1500, 1]);  clamp_min_531 = None
	        view_2782: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1062, [sym_size_int, 1500, 1]);  convert_element_type_1062 = None
	        _assert_tensor_metadata_1598 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1063, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1598 = None
	        convert_element_type_1064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1063, torch.float32);  convert_element_type_1063 = None
	        _assert_tensor_metadata_1599 = torch.ops.aten._assert_tensor_metadata.default(view_2782, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1599 = None
	        convert_element_type_1065: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2782, torch.float32);  view_2782 = None
	        sub_8127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1064, convert_element_type_1065);  convert_element_type_1064 = convert_element_type_1065 = None
	        mul_17214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8127, view_2781);  sub_8127 = view_2781 = None
	        _assert_tensor_metadata_1600 = torch.ops.aten._assert_tensor_metadata.default(mul_17214, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1600 = None
	        view_2784: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2785: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2786: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1601 = torch.ops.aten._assert_tensor_metadata.default(view_2784, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1601 = None
	        convert_element_type_1066: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2784, torch.float32);  view_2784 = None
	        _assert_tensor_metadata_1602 = torch.ops.aten._assert_tensor_metadata.default(view_2786, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1602 = None
	        convert_element_type_1067: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2786, torch.float32);  view_2786 = None
	        sub_8131: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1066, convert_element_type_1067);  convert_element_type_1066 = convert_element_type_1067 = None
	        mul_17219: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8131, view_2785);  sub_8131 = view_2785 = None
	        view_2787: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17219, [1280, 1280]);  mul_17219 = None
	        _assert_tensor_metadata_1603 = torch.ops.aten._assert_tensor_metadata.default(view_2787, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1603 = None
	        mul_17224: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2788: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17214, [mul_17224, 1280]);  mul_17214 = mul_17224 = None
	        permute_298: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2787, [1, 0]);  view_2787 = None
	        addmm_147: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_29_self_attn_out_proj_bias, view_2788, permute_298);  model_audio_tower_layers_29_self_attn_out_proj_bias = view_2788 = permute_298 = None
	        view_2789: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_147, [sym_size_int, 1500, 1280]);  addmm_147 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_27268: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_26648, view_2789);  add_26648 = view_2789 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_238: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_27268, memory_format = torch.contiguous_format)
	        var_mean_59 = torch.ops.aten.var_mean.correction(clone_238, [2], correction = 0, keepdim = True)
	        getitem_238: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_59[0]
	        getitem_239: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_59[1];  var_mean_59 = None
	        add_27273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_238, 1e-05);  getitem_238 = None
	        rsqrt_59: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_27273);  add_27273 = None
	        sub_8137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_238, getitem_239);  clone_238 = getitem_239 = None
	        mul_17235: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8137, rsqrt_59);  sub_8137 = rsqrt_59 = None
	        mul_17236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17235, model_audio_tower_layers_29_final_layer_norm_weight);  mul_17235 = model_audio_tower_layers_29_final_layer_norm_weight = None
	        add_27274: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_17236, model_audio_tower_layers_29_final_layer_norm_bias);  mul_17236 = model_audio_tower_layers_29_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_27274, [2])
	        amax_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_27274, [2])
	        full_356: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_178, full_356);  amin_178 = full_356 = None
	        full_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_178, full_357);  amax_178 = full_357 = None
	        sub_8148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_178, minimum_178);  maximum_178 = None
	        div_356: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8148, 255.0);  sub_8148 = None
	        clamp_min_534: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_356, 1.1920928955078125e-07);  div_356 = None
	        div_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_178, clamp_min_534);  minimum_178 = None
	        round_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_357);  div_357 = None
	        sub_8154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_357);  round_357 = None
	        clamp_min_535: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8154, -128);  sub_8154 = None
	        clamp_max_356: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_535, 127);  clamp_min_535 = None
	        _assert_tensor_metadata_1604 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_534, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1604 = None
	        _assert_tensor_metadata_1605 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_356, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1605 = None
	        convert_element_type_1068: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_356, torch.int8);  clamp_max_356 = None
	        view_2792: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_534, [sym_size_int, 1500, 1])
	        view_2793: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1068, [sym_size_int, 1500, 1])
	        reciprocal_178: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2792);  view_2792 = None
	        mul_17284: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_178, 1.0);  reciprocal_178 = None
	        mul_17287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_27274, mul_17284);  add_27274 = mul_17284 = None
	        round_358: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17287);  mul_17287 = None
	        add_27361: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_358, view_2793);  round_358 = view_2793 = None
	        clamp_min_536: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27361, -128);  add_27361 = None
	        clamp_max_357: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_536, 127);  clamp_min_536 = None
	        _assert_tensor_metadata_1606 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_357, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1606 = None
	        convert_element_type_1069: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_357, torch.int8);  clamp_max_357 = None
	        view_2796: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_534, [sym_size_int, 1500, 1]);  clamp_min_534 = None
	        view_2797: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1068, [sym_size_int, 1500, 1]);  convert_element_type_1068 = None
	        _assert_tensor_metadata_1607 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1069, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1607 = None
	        convert_element_type_1070: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1069, torch.float32);  convert_element_type_1069 = None
	        _assert_tensor_metadata_1608 = torch.ops.aten._assert_tensor_metadata.default(view_2797, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1608 = None
	        convert_element_type_1071: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2797, torch.float32);  view_2797 = None
	        sub_8174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1070, convert_element_type_1071);  convert_element_type_1070 = convert_element_type_1071 = None
	        mul_17309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8174, view_2796);  sub_8174 = view_2796 = None
	        _assert_tensor_metadata_1609 = torch.ops.aten._assert_tensor_metadata.default(mul_17309, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1609 = None
	        view_2799: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_29_fc1_parametrizations_weight_original0 = None
	        view_2800: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_29_fc1_parametrizations_weight_original1 = None
	        view_2801: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_29_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1610 = torch.ops.aten._assert_tensor_metadata.default(view_2799, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1610 = None
	        convert_element_type_1072: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2799, torch.float32);  view_2799 = None
	        _assert_tensor_metadata_1611 = torch.ops.aten._assert_tensor_metadata.default(view_2801, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1611 = None
	        convert_element_type_1073: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2801, torch.float32);  view_2801 = None
	        sub_8178: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1072, convert_element_type_1073);  convert_element_type_1072 = convert_element_type_1073 = None
	        mul_17314: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8178, view_2800);  sub_8178 = view_2800 = None
	        view_2802: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17314, [5120, 1280]);  mul_17314 = None
	        _assert_tensor_metadata_1612 = torch.ops.aten._assert_tensor_metadata.default(view_2802, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1612 = None
	        mul_17319: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2803: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17309, [mul_17319, 1280]);  mul_17309 = mul_17319 = None
	        permute_299: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2802, [1, 0]);  view_2802 = None
	        addmm_148: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_29_fc1_bias, view_2803, permute_299);  model_audio_tower_layers_29_fc1_bias = view_2803 = permute_299 = None
	        view_2804: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_148, [sym_size_int, 1500, 5120]);  addmm_148 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_17326: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2804, 0.5)
	        mul_17327: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2804, 0.7071067811865476);  view_2804 = None
	        erf_31: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_17327);  mul_17327 = None
	        add_27420: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_31, 1);  erf_31 = None
	        mul_17328: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17326, add_27420);  mul_17326 = add_27420 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_17328, [2])
	        amax_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_17328, [2])
	        full_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_179, full_358);  amin_179 = full_358 = None
	        full_359: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_179, full_359);  amax_179 = full_359 = None
	        sub_8191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_179, minimum_179);  maximum_179 = None
	        div_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8191, 255.0);  sub_8191 = None
	        clamp_min_537: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_358, 1.1920928955078125e-07);  div_358 = None
	        div_359: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_179, clamp_min_537);  minimum_179 = None
	        round_359: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_359);  div_359 = None
	        sub_8197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_359);  round_359 = None
	        clamp_min_538: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8197, -128);  sub_8197 = None
	        clamp_max_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_538, 127);  clamp_min_538 = None
	        _assert_tensor_metadata_1613 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_537, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1613 = None
	        _assert_tensor_metadata_1614 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_358, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1614 = None
	        convert_element_type_1074: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_358, torch.int8);  clamp_max_358 = None
	        view_2807: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_537, [sym_size_int, 1500, 1])
	        view_2808: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1074, [sym_size_int, 1500, 1])
	        reciprocal_179: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2807);  view_2807 = None
	        mul_17374: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_179, 1.0);  reciprocal_179 = None
	        mul_17377: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17328, mul_17374);  mul_17328 = mul_17374 = None
	        round_360: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_17377);  mul_17377 = None
	        add_27503: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_360, view_2808);  round_360 = view_2808 = None
	        clamp_min_539: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27503, -128);  add_27503 = None
	        clamp_max_359: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_539, 127);  clamp_min_539 = None
	        _assert_tensor_metadata_1615 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_359, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1615 = None
	        convert_element_type_1075: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_359, torch.int8);  clamp_max_359 = None
	        view_2811: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_537, [sym_size_int, 1500, 1]);  clamp_min_537 = None
	        view_2812: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1074, [sym_size_int, 1500, 1]);  convert_element_type_1074 = None
	        _assert_tensor_metadata_1616 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1075, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1616 = None
	        convert_element_type_1076: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1075, torch.float32);  convert_element_type_1075 = None
	        _assert_tensor_metadata_1617 = torch.ops.aten._assert_tensor_metadata.default(view_2812, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1617 = None
	        convert_element_type_1077: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2812, torch.float32);  view_2812 = None
	        sub_8217: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1076, convert_element_type_1077);  convert_element_type_1076 = convert_element_type_1077 = None
	        mul_17399: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8217, view_2811);  sub_8217 = view_2811 = None
	        _assert_tensor_metadata_1618 = torch.ops.aten._assert_tensor_metadata.default(mul_17399, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1618 = None
	        view_2814: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_29_fc2_parametrizations_weight_original0 = None
	        view_2815: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_29_fc2_parametrizations_weight_original1 = None
	        view_2816: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_29_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_29_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1619 = torch.ops.aten._assert_tensor_metadata.default(view_2814, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1619 = None
	        convert_element_type_1078: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2814, torch.float32);  view_2814 = None
	        _assert_tensor_metadata_1620 = torch.ops.aten._assert_tensor_metadata.default(view_2816, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1620 = None
	        convert_element_type_1079: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2816, torch.float32);  view_2816 = None
	        sub_8221: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1078, convert_element_type_1079);  convert_element_type_1078 = convert_element_type_1079 = None
	        mul_17404: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8221, view_2815);  sub_8221 = view_2815 = None
	        view_2817: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_17404, [1280, 5120]);  mul_17404 = None
	        _assert_tensor_metadata_1621 = torch.ops.aten._assert_tensor_metadata.default(view_2817, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1621 = None
	        mul_17409: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2818: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_17399, [mul_17409, 5120]);  mul_17399 = mul_17409 = None
	        permute_300: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2817, [1, 0]);  view_2817 = None
	        addmm_149: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_29_fc2_bias, view_2818, permute_300);  model_audio_tower_layers_29_fc2_bias = view_2818 = permute_300 = None
	        view_2819: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_149, [sym_size_int, 1500, 1280]);  addmm_149 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_27566: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_27268, view_2819);  add_27268 = view_2819 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_241: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_27566, memory_format = torch.contiguous_format)
	        var_mean_60 = torch.ops.aten.var_mean.correction(clone_241, [2], correction = 0, keepdim = True)
	        getitem_240: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_60[0]
	        getitem_241: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_60[1];  var_mean_60 = None
	        add_27571: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_240, 1e-05);  getitem_240 = None
	        rsqrt_60: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_27571);  add_27571 = None
	        sub_8227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_241, getitem_241);  clone_241 = getitem_241 = None
	        mul_17420: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8227, rsqrt_60);  sub_8227 = rsqrt_60 = None
	        mul_17421: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17420, model_audio_tower_layers_30_self_attn_layer_norm_weight);  mul_17420 = model_audio_tower_layers_30_self_attn_layer_norm_weight = None
	        add_27572: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_17421, model_audio_tower_layers_30_self_attn_layer_norm_bias);  mul_17421 = model_audio_tower_layers_30_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_27572, [2])
	        amax_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_27572, [2])
	        full_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_180, full_360);  amin_180 = full_360 = None
	        full_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_180, full_361);  amax_180 = full_361 = None
	        sub_8238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_180, minimum_180);  maximum_180 = None
	        div_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8238, 255.0);  sub_8238 = None
	        clamp_min_540: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_360, 1.1920928955078125e-07);  div_360 = None
	        div_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_180, clamp_min_540);  minimum_180 = None
	        round_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_361);  div_361 = None
	        sub_8244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_361);  round_361 = None
	        clamp_min_541: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8244, -128);  sub_8244 = None
	        clamp_max_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_541, 127);  clamp_min_541 = None
	        _assert_tensor_metadata_1622 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_540, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1622 = None
	        _assert_tensor_metadata_1623 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_360, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1623 = None
	        convert_element_type_1080: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_360, torch.int8);  clamp_max_360 = None
	        view_2822: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_540, [sym_size_int, 1500, 1])
	        view_2823: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1080, [sym_size_int, 1500, 1])
	        reciprocal_180: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2822);  view_2822 = None
	        mul_17469: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_180, 1.0);  reciprocal_180 = None
	        mul_17472: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_27572, mul_17469);  mul_17469 = None
	        round_362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17472);  mul_17472 = None
	        add_27659: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_362, view_2823);  round_362 = view_2823 = None
	        clamp_min_542: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27659, -128);  add_27659 = None
	        clamp_max_361: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_542, 127);  clamp_min_542 = None
	        _assert_tensor_metadata_1624 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_361, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1624 = None
	        convert_element_type_1081: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_361, torch.int8);  clamp_max_361 = None
	        view_2826: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_540, [sym_size_int, 1500, 1]);  clamp_min_540 = None
	        view_2827: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1080, [sym_size_int, 1500, 1]);  convert_element_type_1080 = None
	        _assert_tensor_metadata_1625 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1081, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1625 = None
	        convert_element_type_1082: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1081, torch.float32);  convert_element_type_1081 = None
	        _assert_tensor_metadata_1626 = torch.ops.aten._assert_tensor_metadata.default(view_2827, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1626 = None
	        convert_element_type_1083: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2827, torch.float32);  view_2827 = None
	        sub_8264: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1082, convert_element_type_1083);  convert_element_type_1082 = convert_element_type_1083 = None
	        mul_17494: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8264, view_2826);  sub_8264 = view_2826 = None
	        _assert_tensor_metadata_1627 = torch.ops.aten._assert_tensor_metadata.default(mul_17494, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1627 = None
	        view_2829: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2830: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2831: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1628 = torch.ops.aten._assert_tensor_metadata.default(view_2829, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1628 = None
	        convert_element_type_1084: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2829, torch.float32);  view_2829 = None
	        _assert_tensor_metadata_1629 = torch.ops.aten._assert_tensor_metadata.default(view_2831, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1629 = None
	        convert_element_type_1085: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2831, torch.float32);  view_2831 = None
	        sub_8268: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1084, convert_element_type_1085);  convert_element_type_1084 = convert_element_type_1085 = None
	        mul_17499: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8268, view_2830);  sub_8268 = view_2830 = None
	        view_2832: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17499, [1280, 1280]);  mul_17499 = None
	        _assert_tensor_metadata_1630 = torch.ops.aten._assert_tensor_metadata.default(view_2832, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1630 = None
	        mul_17504: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2833: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17494, [mul_17504, 1280]);  mul_17494 = mul_17504 = None
	        permute_301: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2832, [1, 0]);  view_2832 = None
	        addmm_150: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_30_self_attn_q_proj_bias, view_2833, permute_301);  model_audio_tower_layers_30_self_attn_q_proj_bias = view_2833 = permute_301 = None
	        view_2834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_150, [sym_size_int, 1500, 1280]);  addmm_150 = None
	        mul_17511: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2834, 0.125);  view_2834 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2835: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_17511, [sym_size_int, 1500, 20, 64]);  mul_17511 = None
	        permute_302: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2835, [0, 2, 1, 3]);  view_2835 = None
	        clone_242: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_302, memory_format = torch.contiguous_format);  permute_302 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_27572, [2])
	        amax_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_27572, [2])
	        full_362: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_181, full_362);  amin_181 = full_362 = None
	        full_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_181, full_363);  amax_181 = full_363 = None
	        sub_8283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_181, minimum_181);  maximum_181 = None
	        div_362: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8283, 255.0);  sub_8283 = None
	        clamp_min_543: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_362, 1.1920928955078125e-07);  div_362 = None
	        div_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_181, clamp_min_543);  minimum_181 = None
	        round_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_363);  div_363 = None
	        sub_8289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_363);  round_363 = None
	        clamp_min_544: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8289, -128);  sub_8289 = None
	        clamp_max_362: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_544, 127);  clamp_min_544 = None
	        _assert_tensor_metadata_1631 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_543, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1631 = None
	        _assert_tensor_metadata_1632 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_362, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1632 = None
	        convert_element_type_1086: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_362, torch.int8);  clamp_max_362 = None
	        view_2838: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_543, [sym_size_int, 1500, 1])
	        view_2839: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1086, [sym_size_int, 1500, 1])
	        reciprocal_181: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2838);  view_2838 = None
	        mul_17565: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_181, 1.0);  reciprocal_181 = None
	        mul_17568: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_27572, mul_17565);  mul_17565 = None
	        round_364: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17568);  mul_17568 = None
	        add_27811: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_364, view_2839);  round_364 = view_2839 = None
	        clamp_min_545: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27811, -128);  add_27811 = None
	        clamp_max_363: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_545, 127);  clamp_min_545 = None
	        _assert_tensor_metadata_1633 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_363, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1633 = None
	        convert_element_type_1087: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_363, torch.int8);  clamp_max_363 = None
	        view_2842: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_543, [sym_size_int, 1500, 1]);  clamp_min_543 = None
	        view_2843: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1086, [sym_size_int, 1500, 1]);  convert_element_type_1086 = None
	        _assert_tensor_metadata_1634 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1087, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1634 = None
	        convert_element_type_1088: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1087, torch.float32);  convert_element_type_1087 = None
	        _assert_tensor_metadata_1635 = torch.ops.aten._assert_tensor_metadata.default(view_2843, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1635 = None
	        convert_element_type_1089: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2843, torch.float32);  view_2843 = None
	        sub_8309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1088, convert_element_type_1089);  convert_element_type_1088 = convert_element_type_1089 = None
	        mul_17590: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8309, view_2842);  sub_8309 = view_2842 = None
	        _assert_tensor_metadata_1636 = torch.ops.aten._assert_tensor_metadata.default(mul_17590, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1636 = None
	        view_2845: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2846: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2847: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1637 = torch.ops.aten._assert_tensor_metadata.default(view_2845, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1637 = None
	        convert_element_type_1090: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2845, torch.float32);  view_2845 = None
	        _assert_tensor_metadata_1638 = torch.ops.aten._assert_tensor_metadata.default(view_2847, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1638 = None
	        convert_element_type_1091: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2847, torch.float32);  view_2847 = None
	        sub_8313: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1090, convert_element_type_1091);  convert_element_type_1090 = convert_element_type_1091 = None
	        mul_17595: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8313, view_2846);  sub_8313 = view_2846 = None
	        view_2848: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17595, [1280, 1280]);  mul_17595 = None
	        _assert_tensor_metadata_1639 = torch.ops.aten._assert_tensor_metadata.default(view_2848, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1639 = None
	        permute_303: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2848, [1, 0]);  view_2848 = None
	        mul_17598: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2849: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17590, [mul_17598, 1280]);  mul_17590 = mul_17598 = None
	        mm_30: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2849, permute_303);  view_2849 = permute_303 = None
	        view_2850: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_30, [sym_size_int, 1500, 1280]);  mm_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2851: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2850, [sym_size_int, -1, 20, 64]);  view_2850 = None
	        permute_304: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2851, [0, 2, 1, 3]);  view_2851 = None
	        clone_243: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_304, memory_format = torch.contiguous_format);  permute_304 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_27572, [2])
	        amax_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_27572, [2])
	        full_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_182, full_364);  amin_182 = full_364 = None
	        full_365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_182, full_365);  amax_182 = full_365 = None
	        sub_8327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_182, minimum_182);  maximum_182 = None
	        div_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8327, 255.0);  sub_8327 = None
	        clamp_min_546: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_364, 1.1920928955078125e-07);  div_364 = None
	        div_365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_182, clamp_min_546);  minimum_182 = None
	        round_365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_365);  div_365 = None
	        sub_8333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_365);  round_365 = None
	        clamp_min_547: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8333, -128);  sub_8333 = None
	        clamp_max_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_547, 127);  clamp_min_547 = None
	        _assert_tensor_metadata_1640 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_546, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1640 = None
	        _assert_tensor_metadata_1641 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_364, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1641 = None
	        convert_element_type_1092: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_364, torch.int8);  clamp_max_364 = None
	        view_2854: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_546, [sym_size_int, 1500, 1])
	        view_2855: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1092, [sym_size_int, 1500, 1])
	        reciprocal_182: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2854);  view_2854 = None
	        mul_17664: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_182, 1.0);  reciprocal_182 = None
	        mul_17667: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_27572, mul_17664);  add_27572 = mul_17664 = None
	        round_366: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17667);  mul_17667 = None
	        add_27959: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_366, view_2855);  round_366 = view_2855 = None
	        clamp_min_548: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27959, -128);  add_27959 = None
	        clamp_max_365: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_548, 127);  clamp_min_548 = None
	        _assert_tensor_metadata_1642 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_365, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1642 = None
	        convert_element_type_1093: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_365, torch.int8);  clamp_max_365 = None
	        view_2858: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_546, [sym_size_int, 1500, 1]);  clamp_min_546 = None
	        view_2859: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1092, [sym_size_int, 1500, 1]);  convert_element_type_1092 = None
	        _assert_tensor_metadata_1643 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1093, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1643 = None
	        convert_element_type_1094: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1093, torch.float32);  convert_element_type_1093 = None
	        _assert_tensor_metadata_1644 = torch.ops.aten._assert_tensor_metadata.default(view_2859, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1644 = None
	        convert_element_type_1095: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2859, torch.float32);  view_2859 = None
	        sub_8353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1094, convert_element_type_1095);  convert_element_type_1094 = convert_element_type_1095 = None
	        mul_17689: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8353, view_2858);  sub_8353 = view_2858 = None
	        _assert_tensor_metadata_1645 = torch.ops.aten._assert_tensor_metadata.default(mul_17689, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1645 = None
	        view_2861: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2862: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2863: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1646 = torch.ops.aten._assert_tensor_metadata.default(view_2861, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1646 = None
	        convert_element_type_1096: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2861, torch.float32);  view_2861 = None
	        _assert_tensor_metadata_1647 = torch.ops.aten._assert_tensor_metadata.default(view_2863, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1647 = None
	        convert_element_type_1097: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2863, torch.float32);  view_2863 = None
	        sub_8357: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1096, convert_element_type_1097);  convert_element_type_1096 = convert_element_type_1097 = None
	        mul_17694: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8357, view_2862);  sub_8357 = view_2862 = None
	        view_2864: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17694, [1280, 1280]);  mul_17694 = None
	        _assert_tensor_metadata_1648 = torch.ops.aten._assert_tensor_metadata.default(view_2864, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1648 = None
	        mul_17699: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2865: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17689, [mul_17699, 1280]);  mul_17689 = mul_17699 = None
	        permute_305: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2864, [1, 0]);  view_2864 = None
	        addmm_151: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_30_self_attn_v_proj_bias, view_2865, permute_305);  model_audio_tower_layers_30_self_attn_v_proj_bias = view_2865 = permute_305 = None
	        view_2866: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_151, [sym_size_int, 1500, 1280]);  addmm_151 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2867: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2866, [sym_size_int, -1, 20, 64]);  view_2866 = None
	        permute_306: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2867, [0, 2, 1, 3]);  view_2867 = None
	        clone_244: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_306, memory_format = torch.contiguous_format);  permute_306 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_30 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_242, clone_243, clone_244, None, False, scale = 1.0);  clone_242 = clone_243 = clone_244 = None
	        getitem_242: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_30[0];  _scaled_dot_product_efficient_attention_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_307: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_242, [0, 2, 1, 3]);  getitem_242 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2868: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_307, [sym_size_int, 1500, -1]);  permute_307 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2868, [2])
	        amax_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2868, [2])
	        full_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_183, full_366);  amin_183 = full_366 = None
	        full_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_183, full_367);  amax_183 = full_367 = None
	        sub_8375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_183, minimum_183);  maximum_183 = None
	        div_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8375, 255.0);  sub_8375 = None
	        clamp_min_549: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_366, 1.1920928955078125e-07);  div_366 = None
	        div_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_183, clamp_min_549);  minimum_183 = None
	        round_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_367);  div_367 = None
	        sub_8381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_367);  round_367 = None
	        clamp_min_550: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8381, -128);  sub_8381 = None
	        clamp_max_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_550, 127);  clamp_min_550 = None
	        _assert_tensor_metadata_1649 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_549, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1649 = None
	        _assert_tensor_metadata_1650 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_366, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1650 = None
	        convert_element_type_1098: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_366, torch.int8);  clamp_max_366 = None
	        view_2871: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_549, [sym_size_int, 1500, 1])
	        view_2872: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1098, [sym_size_int, 1500, 1])
	        reciprocal_183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2871);  view_2871 = None
	        mul_17769: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_183, 1.0);  reciprocal_183 = None
	        mul_17772: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2868, mul_17769);  view_2868 = mul_17769 = None
	        round_368: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17772);  mul_17772 = None
	        add_28123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_368, view_2872);  round_368 = view_2872 = None
	        clamp_min_551: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28123, -128);  add_28123 = None
	        clamp_max_367: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_551, 127);  clamp_min_551 = None
	        _assert_tensor_metadata_1651 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_367, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1651 = None
	        convert_element_type_1099: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_367, torch.int8);  clamp_max_367 = None
	        view_2875: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_549, [sym_size_int, 1500, 1]);  clamp_min_549 = None
	        view_2876: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1098, [sym_size_int, 1500, 1]);  convert_element_type_1098 = None
	        _assert_tensor_metadata_1652 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1099, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1652 = None
	        convert_element_type_1100: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1099, torch.float32);  convert_element_type_1099 = None
	        _assert_tensor_metadata_1653 = torch.ops.aten._assert_tensor_metadata.default(view_2876, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1653 = None
	        convert_element_type_1101: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2876, torch.float32);  view_2876 = None
	        sub_8401: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1100, convert_element_type_1101);  convert_element_type_1100 = convert_element_type_1101 = None
	        mul_17794: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8401, view_2875);  sub_8401 = view_2875 = None
	        _assert_tensor_metadata_1654 = torch.ops.aten._assert_tensor_metadata.default(mul_17794, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1654 = None
	        view_2878: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2879: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2880: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1655 = torch.ops.aten._assert_tensor_metadata.default(view_2878, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1655 = None
	        convert_element_type_1102: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2878, torch.float32);  view_2878 = None
	        _assert_tensor_metadata_1656 = torch.ops.aten._assert_tensor_metadata.default(view_2880, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1656 = None
	        convert_element_type_1103: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2880, torch.float32);  view_2880 = None
	        sub_8405: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1102, convert_element_type_1103);  convert_element_type_1102 = convert_element_type_1103 = None
	        mul_17799: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8405, view_2879);  sub_8405 = view_2879 = None
	        view_2881: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17799, [1280, 1280]);  mul_17799 = None
	        _assert_tensor_metadata_1657 = torch.ops.aten._assert_tensor_metadata.default(view_2881, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1657 = None
	        mul_17804: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2882: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17794, [mul_17804, 1280]);  mul_17794 = mul_17804 = None
	        permute_308: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2881, [1, 0]);  view_2881 = None
	        addmm_152: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_30_self_attn_out_proj_bias, view_2882, permute_308);  model_audio_tower_layers_30_self_attn_out_proj_bias = view_2882 = permute_308 = None
	        view_2883: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_152, [sym_size_int, 1500, 1280]);  addmm_152 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_28186: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_27566, view_2883);  add_27566 = view_2883 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_246: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_28186, memory_format = torch.contiguous_format)
	        var_mean_61 = torch.ops.aten.var_mean.correction(clone_246, [2], correction = 0, keepdim = True)
	        getitem_246: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_61[0]
	        getitem_247: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_61[1];  var_mean_61 = None
	        add_28191: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_246, 1e-05);  getitem_246 = None
	        rsqrt_61: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_28191);  add_28191 = None
	        sub_8411: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_246, getitem_247);  clone_246 = getitem_247 = None
	        mul_17815: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8411, rsqrt_61);  sub_8411 = rsqrt_61 = None
	        mul_17816: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17815, model_audio_tower_layers_30_final_layer_norm_weight);  mul_17815 = model_audio_tower_layers_30_final_layer_norm_weight = None
	        add_28192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_17816, model_audio_tower_layers_30_final_layer_norm_bias);  mul_17816 = model_audio_tower_layers_30_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_28192, [2])
	        amax_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_28192, [2])
	        full_368: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_184, full_368);  amin_184 = full_368 = None
	        full_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_184, full_369);  amax_184 = full_369 = None
	        sub_8422: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_184, minimum_184);  maximum_184 = None
	        div_368: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8422, 255.0);  sub_8422 = None
	        clamp_min_552: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_368, 1.1920928955078125e-07);  div_368 = None
	        div_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_184, clamp_min_552);  minimum_184 = None
	        round_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_369);  div_369 = None
	        sub_8428: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_369);  round_369 = None
	        clamp_min_553: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8428, -128);  sub_8428 = None
	        clamp_max_368: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_553, 127);  clamp_min_553 = None
	        _assert_tensor_metadata_1658 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_552, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1658 = None
	        _assert_tensor_metadata_1659 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_368, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1659 = None
	        convert_element_type_1104: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_368, torch.int8);  clamp_max_368 = None
	        view_2886: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_552, [sym_size_int, 1500, 1])
	        view_2887: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1104, [sym_size_int, 1500, 1])
	        reciprocal_184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2886);  view_2886 = None
	        mul_17864: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_184, 1.0);  reciprocal_184 = None
	        mul_17867: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_28192, mul_17864);  add_28192 = mul_17864 = None
	        round_370: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17867);  mul_17867 = None
	        add_28279: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_370, view_2887);  round_370 = view_2887 = None
	        clamp_min_554: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28279, -128);  add_28279 = None
	        clamp_max_369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_554, 127);  clamp_min_554 = None
	        _assert_tensor_metadata_1660 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_369, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1660 = None
	        convert_element_type_1105: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_369, torch.int8);  clamp_max_369 = None
	        view_2890: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_552, [sym_size_int, 1500, 1]);  clamp_min_552 = None
	        view_2891: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1104, [sym_size_int, 1500, 1]);  convert_element_type_1104 = None
	        _assert_tensor_metadata_1661 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1105, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1661 = None
	        convert_element_type_1106: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1105, torch.float32);  convert_element_type_1105 = None
	        _assert_tensor_metadata_1662 = torch.ops.aten._assert_tensor_metadata.default(view_2891, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1662 = None
	        convert_element_type_1107: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2891, torch.float32);  view_2891 = None
	        sub_8448: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1106, convert_element_type_1107);  convert_element_type_1106 = convert_element_type_1107 = None
	        mul_17889: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8448, view_2890);  sub_8448 = view_2890 = None
	        _assert_tensor_metadata_1663 = torch.ops.aten._assert_tensor_metadata.default(mul_17889, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1663 = None
	        view_2893: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_30_fc1_parametrizations_weight_original0 = None
	        view_2894: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_30_fc1_parametrizations_weight_original1 = None
	        view_2895: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_30_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1664 = torch.ops.aten._assert_tensor_metadata.default(view_2893, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1664 = None
	        convert_element_type_1108: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2893, torch.float32);  view_2893 = None
	        _assert_tensor_metadata_1665 = torch.ops.aten._assert_tensor_metadata.default(view_2895, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1665 = None
	        convert_element_type_1109: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2895, torch.float32);  view_2895 = None
	        sub_8452: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1108, convert_element_type_1109);  convert_element_type_1108 = convert_element_type_1109 = None
	        mul_17894: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8452, view_2894);  sub_8452 = view_2894 = None
	        view_2896: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17894, [5120, 1280]);  mul_17894 = None
	        _assert_tensor_metadata_1666 = torch.ops.aten._assert_tensor_metadata.default(view_2896, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1666 = None
	        mul_17899: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2897: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_17889, [mul_17899, 1280]);  mul_17889 = mul_17899 = None
	        permute_309: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2896, [1, 0]);  view_2896 = None
	        addmm_153: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_30_fc1_bias, view_2897, permute_309);  model_audio_tower_layers_30_fc1_bias = view_2897 = permute_309 = None
	        view_2898: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_153, [sym_size_int, 1500, 5120]);  addmm_153 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_17906: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2898, 0.5)
	        mul_17907: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2898, 0.7071067811865476);  view_2898 = None
	        erf_32: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_17907);  mul_17907 = None
	        add_28338: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_32, 1);  erf_32 = None
	        mul_17908: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17906, add_28338);  mul_17906 = add_28338 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_17908, [2])
	        amax_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_17908, [2])
	        full_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_185, full_370);  amin_185 = full_370 = None
	        full_371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_185, full_371);  amax_185 = full_371 = None
	        sub_8465: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_185, minimum_185);  maximum_185 = None
	        div_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8465, 255.0);  sub_8465 = None
	        clamp_min_555: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_370, 1.1920928955078125e-07);  div_370 = None
	        div_371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_185, clamp_min_555);  minimum_185 = None
	        round_371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_371);  div_371 = None
	        sub_8471: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_371);  round_371 = None
	        clamp_min_556: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8471, -128);  sub_8471 = None
	        clamp_max_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_556, 127);  clamp_min_556 = None
	        _assert_tensor_metadata_1667 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_555, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1667 = None
	        _assert_tensor_metadata_1668 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_370, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1668 = None
	        convert_element_type_1110: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_370, torch.int8);  clamp_max_370 = None
	        view_2901: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_555, [sym_size_int, 1500, 1])
	        view_2902: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1110, [sym_size_int, 1500, 1])
	        reciprocal_185: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2901);  view_2901 = None
	        mul_17954: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_185, 1.0);  reciprocal_185 = None
	        mul_17957: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17908, mul_17954);  mul_17908 = mul_17954 = None
	        round_372: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_17957);  mul_17957 = None
	        add_28421: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_372, view_2902);  round_372 = view_2902 = None
	        clamp_min_557: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28421, -128);  add_28421 = None
	        clamp_max_371: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_557, 127);  clamp_min_557 = None
	        _assert_tensor_metadata_1669 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_371, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1669 = None
	        convert_element_type_1111: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_371, torch.int8);  clamp_max_371 = None
	        view_2905: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_555, [sym_size_int, 1500, 1]);  clamp_min_555 = None
	        view_2906: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1110, [sym_size_int, 1500, 1]);  convert_element_type_1110 = None
	        _assert_tensor_metadata_1670 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1111, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1670 = None
	        convert_element_type_1112: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1111, torch.float32);  convert_element_type_1111 = None
	        _assert_tensor_metadata_1671 = torch.ops.aten._assert_tensor_metadata.default(view_2906, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1671 = None
	        convert_element_type_1113: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2906, torch.float32);  view_2906 = None
	        sub_8491: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1112, convert_element_type_1113);  convert_element_type_1112 = convert_element_type_1113 = None
	        mul_17979: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8491, view_2905);  sub_8491 = view_2905 = None
	        _assert_tensor_metadata_1672 = torch.ops.aten._assert_tensor_metadata.default(mul_17979, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1672 = None
	        view_2908: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_30_fc2_parametrizations_weight_original0 = None
	        view_2909: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_30_fc2_parametrizations_weight_original1 = None
	        view_2910: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_30_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_30_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1673 = torch.ops.aten._assert_tensor_metadata.default(view_2908, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1673 = None
	        convert_element_type_1114: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2908, torch.float32);  view_2908 = None
	        _assert_tensor_metadata_1674 = torch.ops.aten._assert_tensor_metadata.default(view_2910, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1674 = None
	        convert_element_type_1115: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2910, torch.float32);  view_2910 = None
	        sub_8495: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1114, convert_element_type_1115);  convert_element_type_1114 = convert_element_type_1115 = None
	        mul_17984: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8495, view_2909);  sub_8495 = view_2909 = None
	        view_2911: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_17984, [1280, 5120]);  mul_17984 = None
	        _assert_tensor_metadata_1675 = torch.ops.aten._assert_tensor_metadata.default(view_2911, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1675 = None
	        mul_17989: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2912: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_17979, [mul_17989, 5120]);  mul_17979 = mul_17989 = None
	        permute_310: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2911, [1, 0]);  view_2911 = None
	        addmm_154: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_30_fc2_bias, view_2912, permute_310);  model_audio_tower_layers_30_fc2_bias = view_2912 = permute_310 = None
	        view_2913: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_154, [sym_size_int, 1500, 1280]);  addmm_154 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_28484: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_28186, view_2913);  add_28186 = view_2913 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_28484, memory_format = torch.contiguous_format)
	        var_mean_62 = torch.ops.aten.var_mean.correction(clone_249, [2], correction = 0, keepdim = True)
	        getitem_248: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_62[0]
	        getitem_249: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_62[1];  var_mean_62 = None
	        add_28489: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_248, 1e-05);  getitem_248 = None
	        rsqrt_62: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_28489);  add_28489 = None
	        sub_8501: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_249, getitem_249);  clone_249 = getitem_249 = None
	        mul_18000: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8501, rsqrt_62);  sub_8501 = rsqrt_62 = None
	        mul_18001: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18000, model_audio_tower_layers_31_self_attn_layer_norm_weight);  mul_18000 = model_audio_tower_layers_31_self_attn_layer_norm_weight = None
	        add_28490: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_18001, model_audio_tower_layers_31_self_attn_layer_norm_bias);  mul_18001 = model_audio_tower_layers_31_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_28490, [2])
	        amax_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_28490, [2])
	        full_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_186, full_372);  amin_186 = full_372 = None
	        full_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_186, full_373);  amax_186 = full_373 = None
	        sub_8512: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_186, minimum_186);  maximum_186 = None
	        div_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8512, 255.0);  sub_8512 = None
	        clamp_min_558: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_372, 1.1920928955078125e-07);  div_372 = None
	        div_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_186, clamp_min_558);  minimum_186 = None
	        round_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_373);  div_373 = None
	        sub_8518: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_373);  round_373 = None
	        clamp_min_559: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8518, -128);  sub_8518 = None
	        clamp_max_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_559, 127);  clamp_min_559 = None
	        _assert_tensor_metadata_1676 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_558, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1676 = None
	        _assert_tensor_metadata_1677 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_372, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1677 = None
	        convert_element_type_1116: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_372, torch.int8);  clamp_max_372 = None
	        view_2916: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_558, [sym_size_int, 1500, 1])
	        view_2917: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1116, [sym_size_int, 1500, 1])
	        reciprocal_186: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2916);  view_2916 = None
	        mul_18049: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_186, 1.0);  reciprocal_186 = None
	        mul_18052: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_28490, mul_18049);  mul_18049 = None
	        round_374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18052);  mul_18052 = None
	        add_28577: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_374, view_2917);  round_374 = view_2917 = None
	        clamp_min_560: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28577, -128);  add_28577 = None
	        clamp_max_373: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_560, 127);  clamp_min_560 = None
	        _assert_tensor_metadata_1678 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_373, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1678 = None
	        convert_element_type_1117: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_373, torch.int8);  clamp_max_373 = None
	        view_2920: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_558, [sym_size_int, 1500, 1]);  clamp_min_558 = None
	        view_2921: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1116, [sym_size_int, 1500, 1]);  convert_element_type_1116 = None
	        _assert_tensor_metadata_1679 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1117, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1679 = None
	        convert_element_type_1118: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1117, torch.float32);  convert_element_type_1117 = None
	        _assert_tensor_metadata_1680 = torch.ops.aten._assert_tensor_metadata.default(view_2921, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1680 = None
	        convert_element_type_1119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2921, torch.float32);  view_2921 = None
	        sub_8538: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1118, convert_element_type_1119);  convert_element_type_1118 = convert_element_type_1119 = None
	        mul_18074: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8538, view_2920);  sub_8538 = view_2920 = None
	        _assert_tensor_metadata_1681 = torch.ops.aten._assert_tensor_metadata.default(mul_18074, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1681 = None
	        view_2923: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2924: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2925: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1682 = torch.ops.aten._assert_tensor_metadata.default(view_2923, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1682 = None
	        convert_element_type_1120: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2923, torch.float32);  view_2923 = None
	        _assert_tensor_metadata_1683 = torch.ops.aten._assert_tensor_metadata.default(view_2925, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1683 = None
	        convert_element_type_1121: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2925, torch.float32);  view_2925 = None
	        sub_8542: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1120, convert_element_type_1121);  convert_element_type_1120 = convert_element_type_1121 = None
	        mul_18079: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8542, view_2924);  sub_8542 = view_2924 = None
	        view_2926: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18079, [1280, 1280]);  mul_18079 = None
	        _assert_tensor_metadata_1684 = torch.ops.aten._assert_tensor_metadata.default(view_2926, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1684 = None
	        mul_18084: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2927: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18074, [mul_18084, 1280]);  mul_18074 = mul_18084 = None
	        permute_311: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2926, [1, 0]);  view_2926 = None
	        addmm_155: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_31_self_attn_q_proj_bias, view_2927, permute_311);  model_audio_tower_layers_31_self_attn_q_proj_bias = view_2927 = permute_311 = None
	        view_2928: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_155, [sym_size_int, 1500, 1280]);  addmm_155 = None
	        mul_18091: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2928, 0.125);  view_2928 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2929: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(mul_18091, [sym_size_int, 1500, 20, 64]);  mul_18091 = None
	        permute_312: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2929, [0, 2, 1, 3]);  view_2929 = None
	        clone_250: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_312, memory_format = torch.contiguous_format);  permute_312 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_28490, [2])
	        amax_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_28490, [2])
	        full_374: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_187, full_374);  amin_187 = full_374 = None
	        full_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_187, full_375);  amax_187 = full_375 = None
	        sub_8557: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_187, minimum_187);  maximum_187 = None
	        div_374: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8557, 255.0);  sub_8557 = None
	        clamp_min_561: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_374, 1.1920928955078125e-07);  div_374 = None
	        div_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_187, clamp_min_561);  minimum_187 = None
	        round_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_375);  div_375 = None
	        sub_8563: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_375);  round_375 = None
	        clamp_min_562: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8563, -128);  sub_8563 = None
	        clamp_max_374: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_562, 127);  clamp_min_562 = None
	        _assert_tensor_metadata_1685 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_561, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1685 = None
	        _assert_tensor_metadata_1686 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_374, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1686 = None
	        convert_element_type_1122: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_374, torch.int8);  clamp_max_374 = None
	        view_2932: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_561, [sym_size_int, 1500, 1])
	        view_2933: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1122, [sym_size_int, 1500, 1])
	        reciprocal_187: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2932);  view_2932 = None
	        mul_18145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_187, 1.0);  reciprocal_187 = None
	        mul_18148: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_28490, mul_18145);  mul_18145 = None
	        round_376: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18148);  mul_18148 = None
	        add_28729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_376, view_2933);  round_376 = view_2933 = None
	        clamp_min_563: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28729, -128);  add_28729 = None
	        clamp_max_375: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_563, 127);  clamp_min_563 = None
	        _assert_tensor_metadata_1687 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_375, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1687 = None
	        convert_element_type_1123: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_375, torch.int8);  clamp_max_375 = None
	        view_2936: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_561, [sym_size_int, 1500, 1]);  clamp_min_561 = None
	        view_2937: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1122, [sym_size_int, 1500, 1]);  convert_element_type_1122 = None
	        _assert_tensor_metadata_1688 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1123, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1688 = None
	        convert_element_type_1124: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1123, torch.float32);  convert_element_type_1123 = None
	        _assert_tensor_metadata_1689 = torch.ops.aten._assert_tensor_metadata.default(view_2937, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1689 = None
	        convert_element_type_1125: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2937, torch.float32);  view_2937 = None
	        sub_8583: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1124, convert_element_type_1125);  convert_element_type_1124 = convert_element_type_1125 = None
	        mul_18170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8583, view_2936);  sub_8583 = view_2936 = None
	        _assert_tensor_metadata_1690 = torch.ops.aten._assert_tensor_metadata.default(mul_18170, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1690 = None
	        view_2939: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2940: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2941: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1691 = torch.ops.aten._assert_tensor_metadata.default(view_2939, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1691 = None
	        convert_element_type_1126: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2939, torch.float32);  view_2939 = None
	        _assert_tensor_metadata_1692 = torch.ops.aten._assert_tensor_metadata.default(view_2941, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1692 = None
	        convert_element_type_1127: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2941, torch.float32);  view_2941 = None
	        sub_8587: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1126, convert_element_type_1127);  convert_element_type_1126 = convert_element_type_1127 = None
	        mul_18175: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8587, view_2940);  sub_8587 = view_2940 = None
	        view_2942: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18175, [1280, 1280]);  mul_18175 = None
	        _assert_tensor_metadata_1693 = torch.ops.aten._assert_tensor_metadata.default(view_2942, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1693 = None
	        permute_313: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2942, [1, 0]);  view_2942 = None
	        mul_18178: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2943: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18170, [mul_18178, 1280]);  mul_18170 = mul_18178 = None
	        mm_31: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2943, permute_313);  view_2943 = permute_313 = None
	        view_2944: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(mm_31, [sym_size_int, 1500, 1280]);  mm_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2945: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2944, [sym_size_int, -1, 20, 64]);  view_2944 = None
	        permute_314: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2945, [0, 2, 1, 3]);  view_2945 = None
	        clone_251: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_314, memory_format = torch.contiguous_format);  permute_314 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_28490, [2])
	        amax_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_28490, [2])
	        full_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_188, full_376);  amin_188 = full_376 = None
	        full_377: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_188, full_377);  amax_188 = full_377 = None
	        sub_8601: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_188, minimum_188);  maximum_188 = None
	        div_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8601, 255.0);  sub_8601 = None
	        clamp_min_564: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_376, 1.1920928955078125e-07);  div_376 = None
	        div_377: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_188, clamp_min_564);  minimum_188 = None
	        round_377: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_377);  div_377 = None
	        sub_8607: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_377);  round_377 = None
	        clamp_min_565: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8607, -128);  sub_8607 = None
	        clamp_max_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_565, 127);  clamp_min_565 = None
	        _assert_tensor_metadata_1694 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_564, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1694 = None
	        _assert_tensor_metadata_1695 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_376, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1695 = None
	        convert_element_type_1128: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_376, torch.int8);  clamp_max_376 = None
	        view_2948: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_564, [sym_size_int, 1500, 1])
	        view_2949: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1128, [sym_size_int, 1500, 1])
	        reciprocal_188: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2948);  view_2948 = None
	        mul_18244: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_188, 1.0);  reciprocal_188 = None
	        mul_18247: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_28490, mul_18244);  add_28490 = mul_18244 = None
	        round_378: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18247);  mul_18247 = None
	        add_28877: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_378, view_2949);  round_378 = view_2949 = None
	        clamp_min_566: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28877, -128);  add_28877 = None
	        clamp_max_377: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_566, 127);  clamp_min_566 = None
	        _assert_tensor_metadata_1696 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_377, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1696 = None
	        convert_element_type_1129: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_377, torch.int8);  clamp_max_377 = None
	        view_2952: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_564, [sym_size_int, 1500, 1]);  clamp_min_564 = None
	        view_2953: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1128, [sym_size_int, 1500, 1]);  convert_element_type_1128 = None
	        _assert_tensor_metadata_1697 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1129, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1697 = None
	        convert_element_type_1130: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1129, torch.float32);  convert_element_type_1129 = None
	        _assert_tensor_metadata_1698 = torch.ops.aten._assert_tensor_metadata.default(view_2953, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1698 = None
	        convert_element_type_1131: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2953, torch.float32);  view_2953 = None
	        sub_8627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1130, convert_element_type_1131);  convert_element_type_1130 = convert_element_type_1131 = None
	        mul_18269: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8627, view_2952);  sub_8627 = view_2952 = None
	        _assert_tensor_metadata_1699 = torch.ops.aten._assert_tensor_metadata.default(mul_18269, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1699 = None
	        view_2955: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2956: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2957: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1700 = torch.ops.aten._assert_tensor_metadata.default(view_2955, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1700 = None
	        convert_element_type_1132: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2955, torch.float32);  view_2955 = None
	        _assert_tensor_metadata_1701 = torch.ops.aten._assert_tensor_metadata.default(view_2957, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1701 = None
	        convert_element_type_1133: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2957, torch.float32);  view_2957 = None
	        sub_8631: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1132, convert_element_type_1133);  convert_element_type_1132 = convert_element_type_1133 = None
	        mul_18274: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8631, view_2956);  sub_8631 = view_2956 = None
	        view_2958: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18274, [1280, 1280]);  mul_18274 = None
	        _assert_tensor_metadata_1702 = torch.ops.aten._assert_tensor_metadata.default(view_2958, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1702 = None
	        mul_18279: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2959: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18269, [mul_18279, 1280]);  mul_18269 = mul_18279 = None
	        permute_315: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2958, [1, 0]);  view_2958 = None
	        addmm_156: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_31_self_attn_v_proj_bias, view_2959, permute_315);  model_audio_tower_layers_31_self_attn_v_proj_bias = view_2959 = permute_315 = None
	        view_2960: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_156, [sym_size_int, 1500, 1280]);  addmm_156 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2961: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.view.default(view_2960, [sym_size_int, -1, 20, 64]);  view_2960 = None
	        permute_316: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2961, [0, 2, 1, 3]);  view_2961 = None
	        clone_252: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_316, memory_format = torch.contiguous_format);  permute_316 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_31 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_250, clone_251, clone_252, None, False, scale = 1.0);  clone_250 = clone_251 = clone_252 = None
	        getitem_250: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_31[0];  _scaled_dot_product_efficient_attention_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_317: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_250, [0, 2, 1, 3]);  getitem_250 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2962: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(permute_317, [sym_size_int, 1500, -1]);  permute_317 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2962, [2])
	        amax_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2962, [2])
	        full_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_189, full_378);  amin_189 = full_378 = None
	        full_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_189, full_379);  amax_189 = full_379 = None
	        sub_8649: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_189, minimum_189);  maximum_189 = None
	        div_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8649, 255.0);  sub_8649 = None
	        clamp_min_567: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_378, 1.1920928955078125e-07);  div_378 = None
	        div_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_189, clamp_min_567);  minimum_189 = None
	        round_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_379);  div_379 = None
	        sub_8655: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_379);  round_379 = None
	        clamp_min_568: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8655, -128);  sub_8655 = None
	        clamp_max_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_568, 127);  clamp_min_568 = None
	        _assert_tensor_metadata_1703 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_567, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1703 = None
	        _assert_tensor_metadata_1704 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_378, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1704 = None
	        convert_element_type_1134: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_378, torch.int8);  clamp_max_378 = None
	        view_2965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_567, [sym_size_int, 1500, 1])
	        view_2966: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1134, [sym_size_int, 1500, 1])
	        reciprocal_189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2965);  view_2965 = None
	        mul_18349: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_189, 1.0);  reciprocal_189 = None
	        mul_18352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2962, mul_18349);  view_2962 = mul_18349 = None
	        round_380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18352);  mul_18352 = None
	        add_29041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_380, view_2966);  round_380 = view_2966 = None
	        clamp_min_569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29041, -128);  add_29041 = None
	        clamp_max_379: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_569, 127);  clamp_min_569 = None
	        _assert_tensor_metadata_1705 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_379, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1705 = None
	        convert_element_type_1135: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_379, torch.int8);  clamp_max_379 = None
	        view_2969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_567, [sym_size_int, 1500, 1]);  clamp_min_567 = None
	        view_2970: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1134, [sym_size_int, 1500, 1]);  convert_element_type_1134 = None
	        _assert_tensor_metadata_1706 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1135, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1706 = None
	        convert_element_type_1136: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1135, torch.float32);  convert_element_type_1135 = None
	        _assert_tensor_metadata_1707 = torch.ops.aten._assert_tensor_metadata.default(view_2970, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1707 = None
	        convert_element_type_1137: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2970, torch.float32);  view_2970 = None
	        sub_8675: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1136, convert_element_type_1137);  convert_element_type_1136 = convert_element_type_1137 = None
	        mul_18374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8675, view_2969);  sub_8675 = view_2969 = None
	        _assert_tensor_metadata_1708 = torch.ops.aten._assert_tensor_metadata.default(mul_18374, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1708 = None
	        view_2972: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2973: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2974: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1709 = torch.ops.aten._assert_tensor_metadata.default(view_2972, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1709 = None
	        convert_element_type_1138: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2972, torch.float32);  view_2972 = None
	        _assert_tensor_metadata_1710 = torch.ops.aten._assert_tensor_metadata.default(view_2974, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1710 = None
	        convert_element_type_1139: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2974, torch.float32);  view_2974 = None
	        sub_8679: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1138, convert_element_type_1139);  convert_element_type_1138 = convert_element_type_1139 = None
	        mul_18379: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8679, view_2973);  sub_8679 = view_2973 = None
	        view_2975: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18379, [1280, 1280]);  mul_18379 = None
	        _assert_tensor_metadata_1711 = torch.ops.aten._assert_tensor_metadata.default(view_2975, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1711 = None
	        mul_18384: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2976: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18374, [mul_18384, 1280]);  mul_18374 = mul_18384 = None
	        permute_318: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2975, [1, 0]);  view_2975 = None
	        addmm_157: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_31_self_attn_out_proj_bias, view_2976, permute_318);  model_audio_tower_layers_31_self_attn_out_proj_bias = view_2976 = permute_318 = None
	        view_2977: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_157, [sym_size_int, 1500, 1280]);  addmm_157 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_29104: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_28484, view_2977);  add_28484 = view_2977 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_29104, memory_format = torch.contiguous_format)
	        var_mean_63 = torch.ops.aten.var_mean.correction(clone_254, [2], correction = 0, keepdim = True)
	        getitem_254: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_63[0]
	        getitem_255: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_63[1];  var_mean_63 = None
	        add_29109: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_254, 1e-05);  getitem_254 = None
	        rsqrt_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_29109);  add_29109 = None
	        sub_8685: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_254, getitem_255);  clone_254 = getitem_255 = None
	        mul_18395: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8685, rsqrt_63);  sub_8685 = rsqrt_63 = None
	        mul_18396: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18395, model_audio_tower_layers_31_final_layer_norm_weight);  mul_18395 = model_audio_tower_layers_31_final_layer_norm_weight = None
	        add_29110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_18396, model_audio_tower_layers_31_final_layer_norm_bias);  mul_18396 = model_audio_tower_layers_31_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_29110, [2])
	        amax_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_29110, [2])
	        full_380: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_190, full_380);  amin_190 = full_380 = None
	        full_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_190, full_381);  amax_190 = full_381 = None
	        sub_8696: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_190, minimum_190);  maximum_190 = None
	        div_380: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8696, 255.0);  sub_8696 = None
	        clamp_min_570: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_380, 1.1920928955078125e-07);  div_380 = None
	        div_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_190, clamp_min_570);  minimum_190 = None
	        round_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_381);  div_381 = None
	        sub_8702: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_381);  round_381 = None
	        clamp_min_571: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8702, -128);  sub_8702 = None
	        clamp_max_380: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_571, 127);  clamp_min_571 = None
	        _assert_tensor_metadata_1712 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_570, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1712 = None
	        _assert_tensor_metadata_1713 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_380, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1713 = None
	        convert_element_type_1140: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_380, torch.int8);  clamp_max_380 = None
	        view_2980: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_570, [sym_size_int, 1500, 1])
	        view_2981: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1140, [sym_size_int, 1500, 1])
	        reciprocal_190: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2980);  view_2980 = None
	        mul_18444: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_190, 1.0);  reciprocal_190 = None
	        mul_18447: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_29110, mul_18444);  add_29110 = mul_18444 = None
	        round_382: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18447);  mul_18447 = None
	        add_29197: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_382, view_2981);  round_382 = view_2981 = None
	        clamp_min_572: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29197, -128);  add_29197 = None
	        clamp_max_381: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_572, 127);  clamp_min_572 = None
	        _assert_tensor_metadata_1714 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_381, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1714 = None
	        convert_element_type_1141: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_381, torch.int8);  clamp_max_381 = None
	        view_2984: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_570, [sym_size_int, 1500, 1]);  clamp_min_570 = None
	        view_2985: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1140, [sym_size_int, 1500, 1]);  convert_element_type_1140 = None
	        _assert_tensor_metadata_1715 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1141, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1715 = None
	        convert_element_type_1142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1141, torch.float32);  convert_element_type_1141 = None
	        _assert_tensor_metadata_1716 = torch.ops.aten._assert_tensor_metadata.default(view_2985, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1716 = None
	        convert_element_type_1143: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2985, torch.float32);  view_2985 = None
	        sub_8722: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1142, convert_element_type_1143);  convert_element_type_1142 = convert_element_type_1143 = None
	        mul_18469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8722, view_2984);  sub_8722 = view_2984 = None
	        _assert_tensor_metadata_1717 = torch.ops.aten._assert_tensor_metadata.default(mul_18469, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1717 = None
	        view_2987: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_31_fc1_parametrizations_weight_original0 = None
	        view_2988: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_31_fc1_parametrizations_weight_original1 = None
	        view_2989: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_31_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1718 = torch.ops.aten._assert_tensor_metadata.default(view_2987, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1718 = None
	        convert_element_type_1144: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2987, torch.float32);  view_2987 = None
	        _assert_tensor_metadata_1719 = torch.ops.aten._assert_tensor_metadata.default(view_2989, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1719 = None
	        convert_element_type_1145: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2989, torch.float32);  view_2989 = None
	        sub_8726: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1144, convert_element_type_1145);  convert_element_type_1144 = convert_element_type_1145 = None
	        mul_18474: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8726, view_2988);  sub_8726 = view_2988 = None
	        view_2990: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18474, [5120, 1280]);  mul_18474 = None
	        _assert_tensor_metadata_1720 = torch.ops.aten._assert_tensor_metadata.default(view_2990, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1720 = None
	        mul_18479: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2991: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.view.default(mul_18469, [mul_18479, 1280]);  mul_18469 = mul_18479 = None
	        permute_319: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2990, [1, 0]);  view_2990 = None
	        addmm_158: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_31_fc1_bias, view_2991, permute_319);  model_audio_tower_layers_31_fc1_bias = view_2991 = permute_319 = None
	        view_2992: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.view.default(addmm_158, [sym_size_int, 1500, 5120]);  addmm_158 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_18486: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2992, 0.5)
	        mul_18487: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2992, 0.7071067811865476);  view_2992 = None
	        erf_33: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_18487);  mul_18487 = None
	        add_29256: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_33, 1);  erf_33 = None
	        mul_18488: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18486, add_29256);  mul_18486 = add_29256 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_18488, [2])
	        amax_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_18488, [2])
	        full_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_191, full_382);  amin_191 = full_382 = None
	        full_383: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_191, full_383);  amax_191 = full_383 = None
	        sub_8739: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_191, minimum_191);  maximum_191 = None
	        div_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8739, 255.0);  sub_8739 = None
	        clamp_min_573: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_382, 1.1920928955078125e-07);  div_382 = None
	        div_383: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_191, clamp_min_573);  minimum_191 = None
	        round_383: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_383);  div_383 = None
	        sub_8745: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_383);  round_383 = None
	        clamp_min_574: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8745, -128);  sub_8745 = None
	        clamp_max_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_574, 127);  clamp_min_574 = None
	        _assert_tensor_metadata_1721 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_573, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1721 = None
	        _assert_tensor_metadata_1722 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_382, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1722 = None
	        convert_element_type_1146: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_382, torch.int8);  clamp_max_382 = None
	        view_2995: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_573, [sym_size_int, 1500, 1])
	        view_2996: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1146, [sym_size_int, 1500, 1])
	        reciprocal_191: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2995);  view_2995 = None
	        mul_18534: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_191, 1.0);  reciprocal_191 = None
	        mul_18537: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18488, mul_18534);  mul_18488 = mul_18534 = None
	        round_384: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_18537);  mul_18537 = None
	        add_29339: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_384, view_2996);  round_384 = view_2996 = None
	        clamp_min_575: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29339, -128);  add_29339 = None
	        clamp_max_383: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_575, 127);  clamp_min_575 = None
	        _assert_tensor_metadata_1723 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_383, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1723 = None
	        convert_element_type_1147: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_383, torch.int8);  clamp_max_383 = None
	        view_2999: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_573, [sym_size_int, 1500, 1]);  clamp_min_573 = None
	        view_3000: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1146, [sym_size_int, 1500, 1]);  convert_element_type_1146 = None
	        _assert_tensor_metadata_1724 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1147, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1724 = None
	        convert_element_type_1148: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1147, torch.float32);  convert_element_type_1147 = None
	        _assert_tensor_metadata_1725 = torch.ops.aten._assert_tensor_metadata.default(view_3000, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1725 = None
	        convert_element_type_1149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3000, torch.float32);  view_3000 = None
	        sub_8765: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1148, convert_element_type_1149);  convert_element_type_1148 = convert_element_type_1149 = None
	        mul_18559: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8765, view_2999);  sub_8765 = view_2999 = None
	        _assert_tensor_metadata_1726 = torch.ops.aten._assert_tensor_metadata.default(mul_18559, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1726 = None
	        view_3002: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_31_fc2_parametrizations_weight_original0 = None
	        view_3003: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_31_fc2_parametrizations_weight_original1 = None
	        view_3004: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_audio_tower_layers_31_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_31_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1727 = torch.ops.aten._assert_tensor_metadata.default(view_3002, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1727 = None
	        convert_element_type_1150: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3002, torch.float32);  view_3002 = None
	        _assert_tensor_metadata_1728 = torch.ops.aten._assert_tensor_metadata.default(view_3004, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1728 = None
	        convert_element_type_1151: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3004, torch.float32);  view_3004 = None
	        sub_8769: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1150, convert_element_type_1151);  convert_element_type_1150 = convert_element_type_1151 = None
	        mul_18564: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8769, view_3003);  sub_8769 = view_3003 = None
	        view_3005: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_18564, [1280, 5120]);  mul_18564 = None
	        _assert_tensor_metadata_1729 = torch.ops.aten._assert_tensor_metadata.default(view_3005, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1729 = None
	        mul_18569: "Sym(1500*s6)" = sym_size_int * 1500
	        view_3006: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_18559, [mul_18569, 5120]);  mul_18559 = mul_18569 = None
	        permute_320: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_3005, [1, 0]);  view_3005 = None
	        addmm_159: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_31_fc2_bias, view_3006, permute_320);  model_audio_tower_layers_31_fc2_bias = view_3006 = permute_320 = None
	        view_3007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.view.default(addmm_159, [sym_size_int, 1500, 1280]);  addmm_159 = sym_size_int = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_29402: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_29104, view_3007);  add_29104 = view_3007 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:365 in forward, code: hidden_states = self.layer_norm(hidden_states)
	        clone_257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_29402, memory_format = torch.contiguous_format);  add_29402 = None
	        var_mean_64 = torch.ops.aten.var_mean.correction(clone_257, [2], correction = 0, keepdim = True)
	        getitem_256: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_64[0]
	        getitem_257: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_64[1];  var_mean_64 = None
	        add_29407: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_256, 1e-05);  getitem_256 = None
	        rsqrt_64: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_29407);  add_29407 = None
	        sub_8775: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_257, getitem_257);  clone_257 = getitem_257 = None
	        mul_18580: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8775, rsqrt_64);  sub_8775 = rsqrt_64 = None
	        mul_18581: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18580, model_audio_tower_layer_norm_weight);  mul_18580 = model_audio_tower_layer_norm_weight = None
	        add_29408: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_18581, model_audio_tower_layer_norm_bias);  mul_18581 = model_audio_tower_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:451 in get_audio_embeds, code: audio_hidden_states = audio_hidden_states.reshape(-1, self.config.audio_config.intermediate_size)
	        view_3008: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(add_29408, [-1, 5120]);  add_29408 = None
	        sym_size_int_193: "Sym(375*s6)" = torch.ops.aten.sym_size.int(view_3008, 0)
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:389 in forward, code: hidden_states = self.linear_1(audio_features)
	        amin_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.amin.default(view_3008, [1])
	        amax_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.amax.default(view_3008, [1])
	        full_384: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.minimum.default(amin_192, full_384);  amin_192 = full_384 = None
	        full_385: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.maximum.default(amax_192, full_385);  amax_192 = full_385 = None
	        sub_8787: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_192, minimum_192);  maximum_192 = None
	        div_384: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(sub_8787, 255.0);  sub_8787 = None
	        clamp_min_576: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(div_384, 1.1920928955078125e-07);  div_384 = None
	        div_385: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(minimum_192, clamp_min_576);  minimum_192 = None
	        round_385: "f32[375*s6][1]cuda:0" = torch.ops.aten.round.default(div_385);  div_385 = None
	        sub_8793: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_385);  round_385 = None
	        clamp_min_577: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8793, -128);  sub_8793 = None
	        clamp_max_384: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_577, 127);  clamp_min_577 = None
	        _assert_tensor_metadata_1730 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_576, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1730 = None
	        _assert_tensor_metadata_1731 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_384, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1731 = None
	        convert_element_type_1152: "i8[375*s6][1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_384, torch.int8);  clamp_max_384 = None
	        view_3011: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_576, [sym_size_int_193, 1])
	        view_3012: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1152, [sym_size_int_193, 1])
	        reciprocal_192: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_3011);  view_3011 = None
	        mul_18613: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_192, 1.0);  reciprocal_192 = None
	        mul_18615: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_3008, mul_18613);  view_3008 = mul_18613 = None
	        round_386: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.round.default(mul_18615);  mul_18615 = None
	        add_29476: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_386, view_3012);  round_386 = view_3012 = None
	        clamp_min_578: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29476, -128);  add_29476 = None
	        clamp_max_385: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_578, 127);  clamp_min_578 = None
	        _assert_tensor_metadata_1732 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_385, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1732 = None
	        convert_element_type_1153: "i8[375*s6, 5120][5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_385, torch.int8);  clamp_max_385 = None
	        view_3015: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_576, [sym_size_int_193, 1]);  clamp_min_576 = None
	        view_3016: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1152, [sym_size_int_193, 1]);  convert_element_type_1152 = None
	        _assert_tensor_metadata_1733 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1153, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1733 = None
	        convert_element_type_1154: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1153, torch.float32);  convert_element_type_1153 = None
	        _assert_tensor_metadata_1734 = torch.ops.aten._assert_tensor_metadata.default(view_3016, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1734 = None
	        convert_element_type_1155: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3016, torch.float32);  view_3016 = None
	        sub_8813: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1154, convert_element_type_1155);  convert_element_type_1154 = convert_element_type_1155 = None
	        mul_18634: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8813, view_3015);  sub_8813 = view_3015 = None
	        _assert_tensor_metadata_1735 = torch.ops.aten._assert_tensor_metadata.default(mul_18634, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1735 = None
	        view_3018: "i8[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.view.default(model_multi_modal_projector_linear_1_parametrizations_weight_original0, [3072, 160, 32]);  model_multi_modal_projector_linear_1_parametrizations_weight_original0 = None
	        view_3019: "f32[3072, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_multi_modal_projector_linear_1_parametrizations_weight_original1, [3072, 160, 1]);  model_multi_modal_projector_linear_1_parametrizations_weight_original1 = None
	        view_3020: "i8[3072, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.view.default(model_multi_modal_projector_linear_1_parametrizations_weight_original2, [3072, 160, 1]);  model_multi_modal_projector_linear_1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1736 = torch.ops.aten._assert_tensor_metadata.default(view_3018, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1736 = None
	        convert_element_type_1156: "f32[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3018, torch.float32);  view_3018 = None
	        _assert_tensor_metadata_1737 = torch.ops.aten._assert_tensor_metadata.default(view_3020, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1737 = None
	        convert_element_type_1157: "f32[3072, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3020, torch.float32);  view_3020 = None
	        sub_8817: "f32[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1156, convert_element_type_1157);  convert_element_type_1156 = convert_element_type_1157 = None
	        mul_18639: "f32[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8817, view_3019);  sub_8817 = view_3019 = None
	        view_3021: "f32[3072, 5120][5120, 1]cuda:0" = torch.ops.aten.view.default(mul_18639, [3072, 5120]);  mul_18639 = None
	        _assert_tensor_metadata_1738 = torch.ops.aten._assert_tensor_metadata.default(view_3021, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1738 = None
	        permute_321: "f32[5120, 3072][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_3021, [1, 0]);  view_3021 = None
	        mm_32: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mm.default(mul_18634, permute_321);  mul_18634 = permute_321 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_18642: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(mm_32, 0.5)
	        mul_18643: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(mm_32, 0.7071067811865476);  mm_32 = None
	        erf_34: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.erf.default(mul_18643);  mul_18643 = None
	        add_29516: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_34, 1);  erf_34 = None
	        mul_18644: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18642, add_29516);  mul_18642 = add_29516 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:391 in forward, code: hidden_states = self.linear_2(hidden_states)
	        amin_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.amin.default(mul_18644, [1])
	        amax_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.amax.default(mul_18644, [1])
	        full_386: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.minimum.default(amin_193, full_386);  amin_193 = full_386 = None
	        full_387: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.maximum.default(amax_193, full_387);  amax_193 = full_387 = None
	        sub_8827: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_193, minimum_193);  maximum_193 = None
	        div_386: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(sub_8827, 255.0);  sub_8827 = None
	        clamp_min_579: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(div_386, 1.1920928955078125e-07);  div_386 = None
	        div_387: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(minimum_193, clamp_min_579);  minimum_193 = None
	        round_387: "f32[375*s6][1]cuda:0" = torch.ops.aten.round.default(div_387);  div_387 = None
	        sub_8833: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_387);  round_387 = None
	        clamp_min_580: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8833, -128);  sub_8833 = None
	        clamp_max_386: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_580, 127);  clamp_min_580 = None
	        _assert_tensor_metadata_1739 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_579, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1739 = None
	        _assert_tensor_metadata_1740 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_386, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1740 = None
	        convert_element_type_1158: "i8[375*s6][1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_386, torch.int8);  clamp_max_386 = None
	        view_3024: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_579, [sym_size_int_193, 1])
	        view_3025: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1158, [sym_size_int_193, 1])
	        reciprocal_193: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_3024);  view_3024 = None
	        mul_18666: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_193, 1.0);  reciprocal_193 = None
	        mul_18668: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18644, mul_18666);  mul_18644 = mul_18666 = None
	        round_388: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.round.default(mul_18668);  mul_18668 = None
	        add_29572: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.add.Tensor(round_388, view_3025);  round_388 = view_3025 = None
	        clamp_min_581: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29572, -128);  add_29572 = None
	        clamp_max_387: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_581, 127);  clamp_min_581 = None
	        _assert_tensor_metadata_1741 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_387, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1741 = None
	        convert_element_type_1159: "i8[375*s6, 3072][3072, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_387, torch.int8);  clamp_max_387 = None
	        view_3028: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(clamp_min_579, [sym_size_int_193, 1]);  clamp_min_579 = None
	        view_3029: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.view.default(convert_element_type_1158, [sym_size_int_193, 1]);  convert_element_type_1158 = sym_size_int_193 = None
	        _assert_tensor_metadata_1742 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1159, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1742 = None
	        convert_element_type_1160: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1159, torch.float32);  convert_element_type_1159 = None
	        _assert_tensor_metadata_1743 = torch.ops.aten._assert_tensor_metadata.default(view_3029, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1743 = None
	        convert_element_type_1161: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3029, torch.float32);  view_3029 = None
	        sub_8853: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1160, convert_element_type_1161);  convert_element_type_1160 = convert_element_type_1161 = None
	        mul_18687: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8853, view_3028);  sub_8853 = view_3028 = None
	        _assert_tensor_metadata_1744 = torch.ops.aten._assert_tensor_metadata.default(mul_18687, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1744 = None
	        view_3031: "i8[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.aten.view.default(model_multi_modal_projector_linear_2_parametrizations_weight_original0, [3072, 96, 32]);  model_multi_modal_projector_linear_2_parametrizations_weight_original0 = None
	        view_3032: "f32[3072, 96, 1][96, 1, 1]cuda:0" = torch.ops.aten.view.default(model_multi_modal_projector_linear_2_parametrizations_weight_original1, [3072, 96, 1]);  model_multi_modal_projector_linear_2_parametrizations_weight_original1 = None
	        view_3033: "i8[3072, 96, 1][96, 1, 1]cuda:0" = torch.ops.aten.view.default(model_multi_modal_projector_linear_2_parametrizations_weight_original2, [3072, 96, 1]);  model_multi_modal_projector_linear_2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1745 = torch.ops.aten._assert_tensor_metadata.default(view_3031, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1745 = None
	        convert_element_type_1162: "f32[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3031, torch.float32);  view_3031 = None
	        _assert_tensor_metadata_1746 = torch.ops.aten._assert_tensor_metadata.default(view_3033, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1746 = None
	        convert_element_type_1163: "f32[3072, 96, 1][96, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3033, torch.float32);  view_3033 = None
	        sub_8857: "f32[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1162, convert_element_type_1163);  convert_element_type_1162 = convert_element_type_1163 = None
	        mul_18692: "f32[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8857, view_3032);  sub_8857 = view_3032 = None
	        view_3034: "f32[3072, 3072][3072, 1]cuda:0" = torch.ops.aten.view.default(mul_18692, [3072, 3072]);  mul_18692 = None
	        _assert_tensor_metadata_1747 = torch.ops.aten._assert_tensor_metadata.default(view_3034, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1747 = None
	        permute_322: "f32[3072, 3072][1, 3072]cuda:0" = torch.ops.aten.permute.default(view_3034, [1, 0]);  view_3034 = None
	        mm_33: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mm.default(mul_18687, permute_322);  mul_18687 = permute_322 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py:83 in forward, code: return audio_embeds.unsqueeze(0)
	        unsqueeze: "f32[1, 375*s6, 3072][1152000*s6, 3072, 1]cuda:0" = torch.ops.aten.unsqueeze.default(mm_33, 0);  mm_33 = None
	        return (unsqueeze,)
	        
V0910 09:42:45.918000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3fcae5a24839459979b27e8265661386"}
	{
	"name": "inductor_compile",
	"ts": 1757522565918088.5,
	"args": {
	"fn_name": "compile_fx_inner",
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:46.214000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "166e0dc7c73c8d21323252d1149ace62"}
	{
	"name": "fx_codegen_and_compile",
	"ts": 1757522566214859.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:46.988000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/usr/local/fbcode/platform010/lib/python3.12/contextlib.py", 34]}
V0910 09:42:46.988000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_dynamo/repro/after_aot.py", 35]}
V0910 09:42:46.989000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_inductor/fb/utils.py", 36]}
V0910 09:42:46.994000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_inductor/compile_fx.py:1215] {"artifact": {"name": "fx_graph_runnable", "encoding": "string"}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2657, "name": "compile_fx", "filename": 19, "loc": "return inference_compiler(unlifted_gm, example_inputs_)"}, {"line": 1267, "name": "__call__", "filename": 33, "loc": "return self.compiler_fn(gm, example_inputs)"}, {"line": 2543, "name": "fw_compiler_base", "filename": 19, "loc": "return compile_fx_forward("}, {"line": 2260, "name": "compile_fx_forward", "filename": 19, "loc": "return inner_compile("}, {"line": 81, "name": "inner", "filename": 34, "loc": "return func(*args, **kwds)"}, {"line": 781, "name": "compile_fx_inner", "filename": 19, "loc": "return wrap_compiler_debug(_compile_fx_inner, compiler_name=\"inductor\")("}, {"line": 144, "name": "debug_wrapper", "filename": 35, "loc": "inner_compiled_fn = compiler_fn(gm, example_inputs)"}, {"line": 167, "name": "newFunction", "filename": 36, "loc": "return old_func(*args, **kwargs)"}, {"line": 962, "name": "_compile_fx_inner", "filename": 19, "loc": "mb_compiled_graph = fx_codegen_and_compile("}, {"line": 1694, "name": "fx_codegen_and_compile", "filename": 19, "loc": "return scheme.codegen_and_compile(gm, example_inputs, inputs_to_check, graph_kwargs)"}, {"line": 1215, "name": "codegen_and_compile", "filename": 19, "loc": "trace_structured("}], "has_payload": "46d3b1641c522a5a6dd1860dddc1d126"}
	s6 = 3
	
	
	import os
	os.environ['TORCH_TRACE'] = '/home/shangdiy/my_trace_log_dir'
	os.environ['INDUCTOR_PROVENANCE'] = '1'
	os.environ['PYTORCH_DDP_USE_SIDE_STREAM'] = '0'
	os.environ['TRITON_ALLOW_NON_CONSTEXPR_GLOBALS'] = '1'
	os.environ['TRITON_LIBHIP_PATH'] = '/usr/local/fbcode/platform010/lib/rocm-6.2.1/lib/libamdhip64.so'
	os.environ['TRITON_CUPTI_LIB_PATH'] = '/usr/local/fbcode/platform010/lib/libcupti.so'
	os.environ['TRITON_HOME'] = '/tmp/shangdiy'
	os.environ['TORCHINDUCTOR_CACHE_DIR'] = '/var/tmp/torchinductor_shangdiy'
	
	import torch
	from torch import tensor, device
	import torch.fx as fx
	from torch._dynamo.testing import rand_strided
	from math import inf
	import torch._inductor.inductor_prims
	
	
	
	import torch._dynamo.config
	import torch._inductor.config
	import torch._functorch.config
	import torch.fx.experimental._config
	
	torch._inductor.config.cpp_wrapper = True
	torch._inductor.config.fx_wrapper = False
	torch._inductor.config.compile_threads = 32
	torch._inductor.config.triton.cudagraphs = False
	torch._inductor.config.triton.autotune_cublasLt = False
	torch._inductor.config.triton.autotune_at_compile_time = True
	torch._inductor.config.triton.store_cubin = True
	torch._inductor.config.aot_inductor.output_path = 'cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj'
	torch._inductor.config.aot_inductor.serialized_in_spec = '[1, {"type": "builtins.tuple", "context": "null", "children_spec": [{"type": "builtins.tuple", "context": "null", "children_spec": []}, {"type": "builtins.dict", "context": "[\\"input_features\\"]", "children_spec": [{"type": null, "context": null, "children_spec": []}]}]}]'
	torch._inductor.config.aot_inductor.serialized_out_spec = '[1, {"type": null, "context": null, "children_spec": []}]'
	torch._inductor.config.aot_inductor.package = True
	torch._functorch.config.functionalize_rng_ops = False
	torch._functorch.config.unlift_effect_tokens = False
	
	
	
	isolate_fails_code_str = None
	
	torch.ops.load_library("//caffe2/torch/fb/sparsenn:sparsenn_operators_gpu")
	torch.ops.load_library("//caffe2/torch/fb/sparsenn:sparsenn_operators")
	torch.ops.load_library("//deeplearning/fbgemm/fbgemm_gpu:sparse_ops_cpu")
	torch.ops.load_library("//deeplearning/fbgemm/fbgemm_gpu:sparse_ops")
	
	"""
	To run this script in fbcode:
	- Create a directory (//scripts/{your_unixname}/repro)
	- Put this file in scripts/{your_unixname}/repro/fx_graph_runnable.py
	- Add a TARGETS file that looks like the following
	- `buck2 run //scripts/{your_unixname}/repro:repro`
	
	NOTE: you may need additional deps to actually be able to run the script.
	```
	# Contents of TARGETS file
	load("@fbcode_macros//build_defs:python_binary.bzl", "python_binary")
	
	python_binary(
	    name = "repro",
	    main_src = "fx_graph_runnable.py",
	    deps = [
	        "//caffe2:torch",
	        "//caffe2/torch/fb/sparsenn:sparsenn_operators_gpu",
	        "//caffe2/torch/fb/sparsenn:sparsenn_operators",
	        "//deeplearning/fbgemm/fbgemm_gpu:sparse_ops_cpu",
	        "//deeplearning/fbgemm/fbgemm_gpu:sparse_ops",
	    ],
	)
	```
	"""
	
	# torch version: 2.9.0a0+fb
	# torch cuda version: 12.4.0
	# CUDA Info: 
	# nvcc: NVIDIA (R) Cuda compiler driver 
	# Copyright (c) 2005-2024 NVIDIA Corporation 
	# Built on Tue_Oct_29_23:50:19_PDT_2024 
	# Cuda compilation tools, release 12.6, V12.6.85 
	# Build cuda_12.6.r12.6/compiler.35059454_0 
	
	# GPU Hardware Info: 
	# NVIDIA PG509-210 : 8 
	
	
	from torch.nn import *
	class Repro(torch.nn.Module):
	    def __init__(self) -> None:
	        super().__init__()
	        self.model = Module(
	  (audio_tower): Module(
	    (embed_positions): Module()
	    (conv1): Module()
	    (conv2): Module()
	    (layers): Module(
	      (0): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (1): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (2): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (3): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (4): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (5): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (6): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (7): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (8): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (9): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (10): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (11): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (12): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (13): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (14): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (15): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (16): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (17): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (18): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (19): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (20): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (21): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (22): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (23): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (24): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (25): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (26): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (27): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (28): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (29): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (30): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	      (31): Module(
	        (self_attn_layer_norm): Module()
	        (self_attn): Module(
	          (q_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (k_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (v_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	          (out_proj): Module(
	            (parametrizations): Module(
	              (weight): Module()
	            )
	          )
	        )
	        (final_layer_norm): Module()
	        (fc1): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	        (fc2): Module(
	          (parametrizations): Module(
	            (weight): Module()
	          )
	        )
	      )
	    )
	    (layer_norm): Module()
	  )
	  (multi_modal_projector): Module(
	    (linear_1): Module(
	      (parametrizations): Module(
	        (weight): Module()
	      )
	    )
	    (linear_2): Module(
	      (parametrizations): Module(
	        (weight): Module()
	      )
	    )
	  )
	).cuda()
	
	    
	    
	    def forward(self):
	        arg877_1, = fx_pytree.tree_flatten_spec([], self._in_spec)
	        model_audio_tower_embed_positions_weight = self.model.audio_tower.embed_positions.weight
	        model_audio_tower_conv1_weight = self.model.audio_tower.conv1.weight
	        model_audio_tower_conv1_bias = self.model.audio_tower.conv1.bias
	        model_audio_tower_conv2_weight = self.model.audio_tower.conv2.weight
	        model_audio_tower_conv2_bias = self.model.audio_tower.conv2.bias
	        model_audio_tower_layers_0_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "0").self_attn_layer_norm.weight
	        model_audio_tower_layers_0_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "0").self_attn_layer_norm.bias
	        model_audio_tower_layers_0_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.bias
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.bias
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.bias
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "0").final_layer_norm.weight
	        model_audio_tower_layers_0_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "0").final_layer_norm.bias
	        model_audio_tower_layers_0_fc1_bias = getattr(self.model.audio_tower.layers, "0").fc1.bias
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_0_fc2_bias = getattr(self.model.audio_tower.layers, "0").fc2.bias
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "1").self_attn_layer_norm.weight
	        model_audio_tower_layers_1_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "1").self_attn_layer_norm.bias
	        model_audio_tower_layers_1_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.bias
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.bias
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.bias
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "1").final_layer_norm.weight
	        model_audio_tower_layers_1_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "1").final_layer_norm.bias
	        model_audio_tower_layers_1_fc1_bias = getattr(self.model.audio_tower.layers, "1").fc1.bias
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_1_fc2_bias = getattr(self.model.audio_tower.layers, "1").fc2.bias
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "2").self_attn_layer_norm.weight
	        model_audio_tower_layers_2_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "2").self_attn_layer_norm.bias
	        model_audio_tower_layers_2_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.bias
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.bias
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.bias
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "2").final_layer_norm.weight
	        model_audio_tower_layers_2_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "2").final_layer_norm.bias
	        model_audio_tower_layers_2_fc1_bias = getattr(self.model.audio_tower.layers, "2").fc1.bias
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_2_fc2_bias = getattr(self.model.audio_tower.layers, "2").fc2.bias
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "3").self_attn_layer_norm.weight
	        model_audio_tower_layers_3_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "3").self_attn_layer_norm.bias
	        model_audio_tower_layers_3_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.bias
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.bias
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.bias
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "3").final_layer_norm.weight
	        model_audio_tower_layers_3_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "3").final_layer_norm.bias
	        model_audio_tower_layers_3_fc1_bias = getattr(self.model.audio_tower.layers, "3").fc1.bias
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_3_fc2_bias = getattr(self.model.audio_tower.layers, "3").fc2.bias
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "4").self_attn_layer_norm.weight
	        model_audio_tower_layers_4_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "4").self_attn_layer_norm.bias
	        model_audio_tower_layers_4_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.bias
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.bias
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.bias
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "4").final_layer_norm.weight
	        model_audio_tower_layers_4_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "4").final_layer_norm.bias
	        model_audio_tower_layers_4_fc1_bias = getattr(self.model.audio_tower.layers, "4").fc1.bias
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_4_fc2_bias = getattr(self.model.audio_tower.layers, "4").fc2.bias
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "5").self_attn_layer_norm.weight
	        model_audio_tower_layers_5_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "5").self_attn_layer_norm.bias
	        model_audio_tower_layers_5_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.bias
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.bias
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.bias
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "5").final_layer_norm.weight
	        model_audio_tower_layers_5_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "5").final_layer_norm.bias
	        model_audio_tower_layers_5_fc1_bias = getattr(self.model.audio_tower.layers, "5").fc1.bias
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_5_fc2_bias = getattr(self.model.audio_tower.layers, "5").fc2.bias
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "6").self_attn_layer_norm.weight
	        model_audio_tower_layers_6_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "6").self_attn_layer_norm.bias
	        model_audio_tower_layers_6_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.bias
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.bias
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.bias
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "6").final_layer_norm.weight
	        model_audio_tower_layers_6_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "6").final_layer_norm.bias
	        model_audio_tower_layers_6_fc1_bias = getattr(self.model.audio_tower.layers, "6").fc1.bias
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_6_fc2_bias = getattr(self.model.audio_tower.layers, "6").fc2.bias
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "7").self_attn_layer_norm.weight
	        model_audio_tower_layers_7_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "7").self_attn_layer_norm.bias
	        model_audio_tower_layers_7_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.bias
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.bias
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.bias
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "7").final_layer_norm.weight
	        model_audio_tower_layers_7_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "7").final_layer_norm.bias
	        model_audio_tower_layers_7_fc1_bias = getattr(self.model.audio_tower.layers, "7").fc1.bias
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_7_fc2_bias = getattr(self.model.audio_tower.layers, "7").fc2.bias
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "8").self_attn_layer_norm.weight
	        model_audio_tower_layers_8_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "8").self_attn_layer_norm.bias
	        model_audio_tower_layers_8_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.bias
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.bias
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.bias
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "8").final_layer_norm.weight
	        model_audio_tower_layers_8_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "8").final_layer_norm.bias
	        model_audio_tower_layers_8_fc1_bias = getattr(self.model.audio_tower.layers, "8").fc1.bias
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_8_fc2_bias = getattr(self.model.audio_tower.layers, "8").fc2.bias
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "9").self_attn_layer_norm.weight
	        model_audio_tower_layers_9_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "9").self_attn_layer_norm.bias
	        model_audio_tower_layers_9_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.bias
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.bias
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.bias
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "9").final_layer_norm.weight
	        model_audio_tower_layers_9_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "9").final_layer_norm.bias
	        model_audio_tower_layers_9_fc1_bias = getattr(self.model.audio_tower.layers, "9").fc1.bias
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_9_fc2_bias = getattr(self.model.audio_tower.layers, "9").fc2.bias
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "10").self_attn_layer_norm.weight
	        model_audio_tower_layers_10_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "10").self_attn_layer_norm.bias
	        model_audio_tower_layers_10_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.bias
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.bias
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.bias
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "10").final_layer_norm.weight
	        model_audio_tower_layers_10_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "10").final_layer_norm.bias
	        model_audio_tower_layers_10_fc1_bias = getattr(self.model.audio_tower.layers, "10").fc1.bias
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_10_fc2_bias = getattr(self.model.audio_tower.layers, "10").fc2.bias
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "11").self_attn_layer_norm.weight
	        model_audio_tower_layers_11_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "11").self_attn_layer_norm.bias
	        model_audio_tower_layers_11_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.bias
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.bias
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.bias
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "11").final_layer_norm.weight
	        model_audio_tower_layers_11_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "11").final_layer_norm.bias
	        model_audio_tower_layers_11_fc1_bias = getattr(self.model.audio_tower.layers, "11").fc1.bias
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_11_fc2_bias = getattr(self.model.audio_tower.layers, "11").fc2.bias
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "12").self_attn_layer_norm.weight
	        model_audio_tower_layers_12_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "12").self_attn_layer_norm.bias
	        model_audio_tower_layers_12_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.bias
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.bias
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.bias
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "12").final_layer_norm.weight
	        model_audio_tower_layers_12_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "12").final_layer_norm.bias
	        model_audio_tower_layers_12_fc1_bias = getattr(self.model.audio_tower.layers, "12").fc1.bias
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_12_fc2_bias = getattr(self.model.audio_tower.layers, "12").fc2.bias
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "13").self_attn_layer_norm.weight
	        model_audio_tower_layers_13_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "13").self_attn_layer_norm.bias
	        model_audio_tower_layers_13_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.bias
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.bias
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.bias
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "13").final_layer_norm.weight
	        model_audio_tower_layers_13_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "13").final_layer_norm.bias
	        model_audio_tower_layers_13_fc1_bias = getattr(self.model.audio_tower.layers, "13").fc1.bias
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_13_fc2_bias = getattr(self.model.audio_tower.layers, "13").fc2.bias
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "14").self_attn_layer_norm.weight
	        model_audio_tower_layers_14_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "14").self_attn_layer_norm.bias
	        model_audio_tower_layers_14_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.bias
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.bias
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.bias
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "14").final_layer_norm.weight
	        model_audio_tower_layers_14_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "14").final_layer_norm.bias
	        model_audio_tower_layers_14_fc1_bias = getattr(self.model.audio_tower.layers, "14").fc1.bias
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_14_fc2_bias = getattr(self.model.audio_tower.layers, "14").fc2.bias
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "15").self_attn_layer_norm.weight
	        model_audio_tower_layers_15_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "15").self_attn_layer_norm.bias
	        model_audio_tower_layers_15_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.bias
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.bias
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.bias
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "15").final_layer_norm.weight
	        model_audio_tower_layers_15_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "15").final_layer_norm.bias
	        model_audio_tower_layers_15_fc1_bias = getattr(self.model.audio_tower.layers, "15").fc1.bias
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_15_fc2_bias = getattr(self.model.audio_tower.layers, "15").fc2.bias
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "16").self_attn_layer_norm.weight
	        model_audio_tower_layers_16_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "16").self_attn_layer_norm.bias
	        model_audio_tower_layers_16_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.bias
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.bias
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.bias
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "16").final_layer_norm.weight
	        model_audio_tower_layers_16_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "16").final_layer_norm.bias
	        model_audio_tower_layers_16_fc1_bias = getattr(self.model.audio_tower.layers, "16").fc1.bias
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_16_fc2_bias = getattr(self.model.audio_tower.layers, "16").fc2.bias
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "17").self_attn_layer_norm.weight
	        model_audio_tower_layers_17_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "17").self_attn_layer_norm.bias
	        model_audio_tower_layers_17_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.bias
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.bias
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.bias
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "17").final_layer_norm.weight
	        model_audio_tower_layers_17_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "17").final_layer_norm.bias
	        model_audio_tower_layers_17_fc1_bias = getattr(self.model.audio_tower.layers, "17").fc1.bias
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_17_fc2_bias = getattr(self.model.audio_tower.layers, "17").fc2.bias
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "18").self_attn_layer_norm.weight
	        model_audio_tower_layers_18_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "18").self_attn_layer_norm.bias
	        model_audio_tower_layers_18_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.bias
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.bias
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.bias
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "18").final_layer_norm.weight
	        model_audio_tower_layers_18_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "18").final_layer_norm.bias
	        model_audio_tower_layers_18_fc1_bias = getattr(self.model.audio_tower.layers, "18").fc1.bias
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_18_fc2_bias = getattr(self.model.audio_tower.layers, "18").fc2.bias
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "19").self_attn_layer_norm.weight
	        model_audio_tower_layers_19_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "19").self_attn_layer_norm.bias
	        model_audio_tower_layers_19_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.bias
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.bias
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.bias
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "19").final_layer_norm.weight
	        model_audio_tower_layers_19_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "19").final_layer_norm.bias
	        model_audio_tower_layers_19_fc1_bias = getattr(self.model.audio_tower.layers, "19").fc1.bias
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_19_fc2_bias = getattr(self.model.audio_tower.layers, "19").fc2.bias
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "20").self_attn_layer_norm.weight
	        model_audio_tower_layers_20_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "20").self_attn_layer_norm.bias
	        model_audio_tower_layers_20_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.bias
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.bias
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.bias
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "20").final_layer_norm.weight
	        model_audio_tower_layers_20_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "20").final_layer_norm.bias
	        model_audio_tower_layers_20_fc1_bias = getattr(self.model.audio_tower.layers, "20").fc1.bias
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_20_fc2_bias = getattr(self.model.audio_tower.layers, "20").fc2.bias
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "21").self_attn_layer_norm.weight
	        model_audio_tower_layers_21_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "21").self_attn_layer_norm.bias
	        model_audio_tower_layers_21_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.bias
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.bias
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.bias
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "21").final_layer_norm.weight
	        model_audio_tower_layers_21_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "21").final_layer_norm.bias
	        model_audio_tower_layers_21_fc1_bias = getattr(self.model.audio_tower.layers, "21").fc1.bias
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_21_fc2_bias = getattr(self.model.audio_tower.layers, "21").fc2.bias
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "22").self_attn_layer_norm.weight
	        model_audio_tower_layers_22_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "22").self_attn_layer_norm.bias
	        model_audio_tower_layers_22_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.bias
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.bias
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.bias
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "22").final_layer_norm.weight
	        model_audio_tower_layers_22_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "22").final_layer_norm.bias
	        model_audio_tower_layers_22_fc1_bias = getattr(self.model.audio_tower.layers, "22").fc1.bias
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_22_fc2_bias = getattr(self.model.audio_tower.layers, "22").fc2.bias
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "23").self_attn_layer_norm.weight
	        model_audio_tower_layers_23_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "23").self_attn_layer_norm.bias
	        model_audio_tower_layers_23_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.bias
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.bias
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.bias
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "23").final_layer_norm.weight
	        model_audio_tower_layers_23_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "23").final_layer_norm.bias
	        model_audio_tower_layers_23_fc1_bias = getattr(self.model.audio_tower.layers, "23").fc1.bias
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_23_fc2_bias = getattr(self.model.audio_tower.layers, "23").fc2.bias
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "24").self_attn_layer_norm.weight
	        model_audio_tower_layers_24_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "24").self_attn_layer_norm.bias
	        model_audio_tower_layers_24_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.bias
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.bias
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.bias
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "24").final_layer_norm.weight
	        model_audio_tower_layers_24_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "24").final_layer_norm.bias
	        model_audio_tower_layers_24_fc1_bias = getattr(self.model.audio_tower.layers, "24").fc1.bias
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_24_fc2_bias = getattr(self.model.audio_tower.layers, "24").fc2.bias
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "25").self_attn_layer_norm.weight
	        model_audio_tower_layers_25_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "25").self_attn_layer_norm.bias
	        model_audio_tower_layers_25_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.bias
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.bias
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.bias
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "25").final_layer_norm.weight
	        model_audio_tower_layers_25_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "25").final_layer_norm.bias
	        model_audio_tower_layers_25_fc1_bias = getattr(self.model.audio_tower.layers, "25").fc1.bias
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_25_fc2_bias = getattr(self.model.audio_tower.layers, "25").fc2.bias
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "26").self_attn_layer_norm.weight
	        model_audio_tower_layers_26_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "26").self_attn_layer_norm.bias
	        model_audio_tower_layers_26_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.bias
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.bias
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.bias
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "26").final_layer_norm.weight
	        model_audio_tower_layers_26_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "26").final_layer_norm.bias
	        model_audio_tower_layers_26_fc1_bias = getattr(self.model.audio_tower.layers, "26").fc1.bias
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_26_fc2_bias = getattr(self.model.audio_tower.layers, "26").fc2.bias
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "27").self_attn_layer_norm.weight
	        model_audio_tower_layers_27_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "27").self_attn_layer_norm.bias
	        model_audio_tower_layers_27_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.bias
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.bias
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.bias
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "27").final_layer_norm.weight
	        model_audio_tower_layers_27_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "27").final_layer_norm.bias
	        model_audio_tower_layers_27_fc1_bias = getattr(self.model.audio_tower.layers, "27").fc1.bias
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_27_fc2_bias = getattr(self.model.audio_tower.layers, "27").fc2.bias
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "28").self_attn_layer_norm.weight
	        model_audio_tower_layers_28_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "28").self_attn_layer_norm.bias
	        model_audio_tower_layers_28_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.bias
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.bias
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.bias
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "28").final_layer_norm.weight
	        model_audio_tower_layers_28_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "28").final_layer_norm.bias
	        model_audio_tower_layers_28_fc1_bias = getattr(self.model.audio_tower.layers, "28").fc1.bias
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_28_fc2_bias = getattr(self.model.audio_tower.layers, "28").fc2.bias
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "29").self_attn_layer_norm.weight
	        model_audio_tower_layers_29_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "29").self_attn_layer_norm.bias
	        model_audio_tower_layers_29_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.bias
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.bias
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.bias
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "29").final_layer_norm.weight
	        model_audio_tower_layers_29_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "29").final_layer_norm.bias
	        model_audio_tower_layers_29_fc1_bias = getattr(self.model.audio_tower.layers, "29").fc1.bias
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_29_fc2_bias = getattr(self.model.audio_tower.layers, "29").fc2.bias
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "30").self_attn_layer_norm.weight
	        model_audio_tower_layers_30_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "30").self_attn_layer_norm.bias
	        model_audio_tower_layers_30_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.bias
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.bias
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.bias
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "30").final_layer_norm.weight
	        model_audio_tower_layers_30_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "30").final_layer_norm.bias
	        model_audio_tower_layers_30_fc1_bias = getattr(self.model.audio_tower.layers, "30").fc1.bias
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_30_fc2_bias = getattr(self.model.audio_tower.layers, "30").fc2.bias
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_layer_norm_weight = getattr(self.model.audio_tower.layers, "31").self_attn_layer_norm.weight
	        model_audio_tower_layers_31_self_attn_layer_norm_bias = getattr(self.model.audio_tower.layers, "31").self_attn_layer_norm.bias
	        model_audio_tower_layers_31_self_attn_q_proj_bias = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.bias
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_v_proj_bias = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.bias
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_out_proj_bias = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.bias
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_final_layer_norm_weight = getattr(self.model.audio_tower.layers, "31").final_layer_norm.weight
	        model_audio_tower_layers_31_final_layer_norm_bias = getattr(self.model.audio_tower.layers, "31").final_layer_norm.bias
	        model_audio_tower_layers_31_fc1_bias = getattr(self.model.audio_tower.layers, "31").fc1.bias
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_31_fc2_bias = getattr(self.model.audio_tower.layers, "31").fc2.bias
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original0 = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original1 = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original2 = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original2
	        model_audio_tower_layer_norm_weight = self.model.audio_tower.layer_norm.weight
	        model_audio_tower_layer_norm_bias = self.model.audio_tower.layer_norm.bias
	        model_multi_modal_projector_linear_1_parametrizations_weight_original0 = self.model.multi_modal_projector.linear_1.parametrizations.weight.original0
	        model_multi_modal_projector_linear_1_parametrizations_weight_original1 = self.model.multi_modal_projector.linear_1.parametrizations.weight.original1
	        model_multi_modal_projector_linear_1_parametrizations_weight_original2 = self.model.multi_modal_projector.linear_1.parametrizations.weight.original2
	        model_multi_modal_projector_linear_2_parametrizations_weight_original0 = self.model.multi_modal_projector.linear_2.parametrizations.weight.original0
	        model_multi_modal_projector_linear_2_parametrizations_weight_original1 = self.model.multi_modal_projector.linear_2.parametrizations.weight.original1
	        model_multi_modal_projector_linear_2_parametrizations_weight_original2 = self.model.multi_modal_projector.linear_2.parametrizations.weight.original2
	        _assert_tensor_metadata = torch.ops.aten._assert_tensor_metadata.default(arg877_1, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata = None
	        convolution = torch.ops.aten.convolution.default(arg877_1, model_audio_tower_conv1_weight, model_audio_tower_conv1_bias, [1], [1], [1], False, [0], 1);  model_audio_tower_conv1_weight = model_audio_tower_conv1_bias = None
	        sym_size_int = torch.ops.aten.sym_size.int(arg877_1, 0);  arg877_1 = None
	        mul_2 = torch.ops.aten.mul.Tensor(convolution, 0.5)
	        mul_3 = torch.ops.aten.mul.Tensor(convolution, 0.7071067811865476);  convolution = None
	        erf = torch.ops.aten.erf.default(mul_3);  mul_3 = None
	        add_4 = torch.ops.aten.add.Tensor(erf, 1);  erf = None
	        mul_4 = torch.ops.aten.mul.Tensor(mul_2, add_4);  mul_2 = add_4 = None
	        convolution_1 = torch.ops.aten.convolution.default(mul_4, model_audio_tower_conv2_weight, model_audio_tower_conv2_bias, [2], [1], [1], False, [0], 1);  mul_4 = model_audio_tower_conv2_weight = model_audio_tower_conv2_bias = None
	        mul_9 = torch.ops.aten.mul.Tensor(convolution_1, 0.5)
	        mul_10 = torch.ops.aten.mul.Tensor(convolution_1, 0.7071067811865476);  convolution_1 = None
	        erf_1 = torch.ops.aten.erf.default(mul_10);  mul_10 = None
	        add_13 = torch.ops.aten.add.Tensor(erf_1, 1);  erf_1 = None
	        mul_11 = torch.ops.aten.mul.Tensor(mul_9, add_13);  mul_9 = add_13 = None
	        permute = torch.ops.aten.permute.default(mul_11, [0, 2, 1]);  mul_11 = None
	        add_22 = torch.ops.aten.add.Tensor(permute, model_audio_tower_embed_positions_weight);  permute = model_audio_tower_embed_positions_weight = None
	        _assert_tensor_metadata_1 = torch.ops.aten._assert_tensor_metadata.default(add_22, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1 = None
	        clone_1 = torch.ops.aten.clone.default(add_22, memory_format = torch.contiguous_format)
	        var_mean = torch.ops.aten.var_mean.correction(clone_1, [2], correction = 0, keepdim = True)
	        getitem = var_mean[0]
	        getitem_1 = var_mean[1];  var_mean = None
	        add_31 = torch.ops.aten.add.Tensor(getitem, 1e-05);  getitem = None
	        rsqrt = torch.ops.aten.rsqrt.default(add_31);  add_31 = None
	        sub_7 = torch.ops.aten.sub.Tensor(clone_1, getitem_1);  clone_1 = getitem_1 = None
	        mul_20 = torch.ops.aten.mul.Tensor(sub_7, rsqrt);  sub_7 = rsqrt = None
	        mul_21 = torch.ops.aten.mul.Tensor(mul_20, model_audio_tower_layers_0_self_attn_layer_norm_weight);  mul_20 = model_audio_tower_layers_0_self_attn_layer_norm_weight = None
	        add_32 = torch.ops.aten.add.Tensor(mul_21, model_audio_tower_layers_0_self_attn_layer_norm_bias);  mul_21 = model_audio_tower_layers_0_self_attn_layer_norm_bias = None
	        amin = torch.ops.aten.amin.default(add_32, [2])
	        amax = torch.ops.aten.amax.default(add_32, [2])
	        full = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum = torch.ops.aten.minimum.default(amin, full);  amin = full = None
	        full_1 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum = torch.ops.aten.maximum.default(amax, full_1);  amax = full_1 = None
	        sub_18 = torch.ops.aten.sub.Tensor(maximum, minimum);  maximum = None
	        div = torch.ops.aten.div.Tensor(sub_18, 255.0);  sub_18 = None
	        clamp_min = torch.ops.aten.clamp_min.default(div, 1.1920928955078125e-07);  div = None
	        div_1 = torch.ops.aten.div.Tensor(minimum, clamp_min);  minimum = None
	        round_1 = torch.ops.aten.round.default(div_1);  div_1 = None
	        sub_24 = torch.ops.aten.sub.Tensor(-128, round_1);  round_1 = None
	        clamp_min_1 = torch.ops.aten.clamp_min.default(sub_24, -128);  sub_24 = None
	        clamp_max = torch.ops.aten.clamp_max.default(clamp_min_1, 127);  clamp_min_1 = None
	        _assert_tensor_metadata_2 = torch.ops.aten._assert_tensor_metadata.default(clamp_min, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_2 = None
	        _assert_tensor_metadata_3 = torch.ops.aten._assert_tensor_metadata.default(clamp_max, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_3 = None
	        convert_element_type = torch.ops.prims.convert_element_type.default(clamp_max, torch.int8);  clamp_max = None
	        view_2 = torch.ops.aten.view.default(clamp_min, [sym_size_int, 1500, 1])
	        view_3 = torch.ops.aten.view.default(convert_element_type, [sym_size_int, 1500, 1])
	        reciprocal = torch.ops.aten.reciprocal.default(view_2);  view_2 = None
	        mul_69 = torch.ops.aten.mul.Tensor(reciprocal, 1.0);  reciprocal = None
	        mul_72 = torch.ops.aten.mul.Tensor(add_32, mul_69);  mul_69 = None
	        round_2 = torch.ops.aten.round.default(mul_72);  mul_72 = None
	        add_119 = torch.ops.aten.add.Tensor(round_2, view_3);  round_2 = view_3 = None
	        clamp_min_2 = torch.ops.aten.clamp_min.default(add_119, -128);  add_119 = None
	        clamp_max_1 = torch.ops.aten.clamp_max.default(clamp_min_2, 127);  clamp_min_2 = None
	        _assert_tensor_metadata_4 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_1, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_4 = None
	        convert_element_type_1 = torch.ops.prims.convert_element_type.default(clamp_max_1, torch.int8);  clamp_max_1 = None
	        view_6 = torch.ops.aten.view.default(clamp_min, [sym_size_int, 1500, 1]);  clamp_min = None
	        view_7 = torch.ops.aten.view.default(convert_element_type, [sym_size_int, 1500, 1]);  convert_element_type = None
	        _assert_tensor_metadata_5 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_5 = None
	        convert_element_type_2 = torch.ops.prims.convert_element_type.default(convert_element_type_1, torch.float32);  convert_element_type_1 = None
	        _assert_tensor_metadata_6 = torch.ops.aten._assert_tensor_metadata.default(view_7, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_6 = None
	        convert_element_type_3 = torch.ops.prims.convert_element_type.default(view_7, torch.float32);  view_7 = None
	        sub_44 = torch.ops.aten.sub.Tensor(convert_element_type_2, convert_element_type_3);  convert_element_type_2 = convert_element_type_3 = None
	        mul_94 = torch.ops.aten.mul.Tensor(sub_44, view_6);  sub_44 = view_6 = None
	        _assert_tensor_metadata_7 = torch.ops.aten._assert_tensor_metadata.default(mul_94, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_7 = None
	        view_9 = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_10 = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_11 = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_8 = torch.ops.aten._assert_tensor_metadata.default(view_9, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_8 = None
	        convert_element_type_4 = torch.ops.prims.convert_element_type.default(view_9, torch.float32);  view_9 = None
	        _assert_tensor_metadata_9 = torch.ops.aten._assert_tensor_metadata.default(view_11, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_9 = None
	        convert_element_type_5 = torch.ops.prims.convert_element_type.default(view_11, torch.float32);  view_11 = None
	        sub_48 = torch.ops.aten.sub.Tensor(convert_element_type_4, convert_element_type_5);  convert_element_type_4 = convert_element_type_5 = None
	        mul_99 = torch.ops.aten.mul.Tensor(sub_48, view_10);  sub_48 = view_10 = None
	        view_12 = torch.ops.aten.view.default(mul_99, [1280, 1280]);  mul_99 = None
	        _assert_tensor_metadata_10 = torch.ops.aten._assert_tensor_metadata.default(view_12, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_10 = None
	        mul_104 = sym_size_int * 1500
	        view_13 = torch.ops.aten.view.default(mul_94, [mul_104, 1280]);  mul_94 = mul_104 = None
	        permute_1 = torch.ops.aten.permute.default(view_12, [1, 0]);  view_12 = None
	        addmm = torch.ops.aten.addmm.default(model_audio_tower_layers_0_self_attn_q_proj_bias, view_13, permute_1);  model_audio_tower_layers_0_self_attn_q_proj_bias = view_13 = permute_1 = None
	        view_14 = torch.ops.aten.view.default(addmm, [sym_size_int, 1500, 1280]);  addmm = None
	        mul_111 = torch.ops.aten.mul.Tensor(view_14, 0.125);  view_14 = None
	        view_15 = torch.ops.aten.view.default(mul_111, [sym_size_int, 1500, 20, 64]);  mul_111 = None
	        permute_2 = torch.ops.aten.permute.default(view_15, [0, 2, 1, 3]);  view_15 = None
	        clone_2 = torch.ops.aten.clone.default(permute_2, memory_format = torch.contiguous_format);  permute_2 = None
	        amin_1 = torch.ops.aten.amin.default(add_32, [2])
	        amax_1 = torch.ops.aten.amax.default(add_32, [2])
	        full_2 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_1 = torch.ops.aten.minimum.default(amin_1, full_2);  amin_1 = full_2 = None
	        full_3 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_1 = torch.ops.aten.maximum.default(amax_1, full_3);  amax_1 = full_3 = None
	        sub_63 = torch.ops.aten.sub.Tensor(maximum_1, minimum_1);  maximum_1 = None
	        div_2 = torch.ops.aten.div.Tensor(sub_63, 255.0);  sub_63 = None
	        clamp_min_3 = torch.ops.aten.clamp_min.default(div_2, 1.1920928955078125e-07);  div_2 = None
	        div_3 = torch.ops.aten.div.Tensor(minimum_1, clamp_min_3);  minimum_1 = None
	        round_3 = torch.ops.aten.round.default(div_3);  div_3 = None
	        sub_69 = torch.ops.aten.sub.Tensor(-128, round_3);  round_3 = None
	        clamp_min_4 = torch.ops.aten.clamp_min.default(sub_69, -128);  sub_69 = None
	        clamp_max_2 = torch.ops.aten.clamp_max.default(clamp_min_4, 127);  clamp_min_4 = None
	        _assert_tensor_metadata_11 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_3, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_11 = None
	        _assert_tensor_metadata_12 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_2, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_12 = None
	        convert_element_type_6 = torch.ops.prims.convert_element_type.default(clamp_max_2, torch.int8);  clamp_max_2 = None
	        view_18 = torch.ops.aten.view.default(clamp_min_3, [sym_size_int, 1500, 1])
	        view_19 = torch.ops.aten.view.default(convert_element_type_6, [sym_size_int, 1500, 1])
	        reciprocal_1 = torch.ops.aten.reciprocal.default(view_18);  view_18 = None
	        mul_165 = torch.ops.aten.mul.Tensor(reciprocal_1, 1.0);  reciprocal_1 = None
	        mul_168 = torch.ops.aten.mul.Tensor(add_32, mul_165);  mul_165 = None
	        round_4 = torch.ops.aten.round.default(mul_168);  mul_168 = None
	        add_271 = torch.ops.aten.add.Tensor(round_4, view_19);  round_4 = view_19 = None
	        clamp_min_5 = torch.ops.aten.clamp_min.default(add_271, -128);  add_271 = None
	        clamp_max_3 = torch.ops.aten.clamp_max.default(clamp_min_5, 127);  clamp_min_5 = None
	        _assert_tensor_metadata_13 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_3, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_13 = None
	        convert_element_type_7 = torch.ops.prims.convert_element_type.default(clamp_max_3, torch.int8);  clamp_max_3 = None
	        view_22 = torch.ops.aten.view.default(clamp_min_3, [sym_size_int, 1500, 1]);  clamp_min_3 = None
	        view_23 = torch.ops.aten.view.default(convert_element_type_6, [sym_size_int, 1500, 1]);  convert_element_type_6 = None
	        _assert_tensor_metadata_14 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_7, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_14 = None
	        convert_element_type_8 = torch.ops.prims.convert_element_type.default(convert_element_type_7, torch.float32);  convert_element_type_7 = None
	        _assert_tensor_metadata_15 = torch.ops.aten._assert_tensor_metadata.default(view_23, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_15 = None
	        convert_element_type_9 = torch.ops.prims.convert_element_type.default(view_23, torch.float32);  view_23 = None
	        sub_89 = torch.ops.aten.sub.Tensor(convert_element_type_8, convert_element_type_9);  convert_element_type_8 = convert_element_type_9 = None
	        mul_190 = torch.ops.aten.mul.Tensor(sub_89, view_22);  sub_89 = view_22 = None
	        _assert_tensor_metadata_16 = torch.ops.aten._assert_tensor_metadata.default(mul_190, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_16 = None
	        view_25 = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_26 = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_27 = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_17 = torch.ops.aten._assert_tensor_metadata.default(view_25, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_17 = None
	        convert_element_type_10 = torch.ops.prims.convert_element_type.default(view_25, torch.float32);  view_25 = None
	        _assert_tensor_metadata_18 = torch.ops.aten._assert_tensor_metadata.default(view_27, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_18 = None
	        convert_element_type_11 = torch.ops.prims.convert_element_type.default(view_27, torch.float32);  view_27 = None
	        sub_93 = torch.ops.aten.sub.Tensor(convert_element_type_10, convert_element_type_11);  convert_element_type_10 = convert_element_type_11 = None
	        mul_195 = torch.ops.aten.mul.Tensor(sub_93, view_26);  sub_93 = view_26 = None
	        view_28 = torch.ops.aten.view.default(mul_195, [1280, 1280]);  mul_195 = None
	        _assert_tensor_metadata_19 = torch.ops.aten._assert_tensor_metadata.default(view_28, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_19 = None
	        permute_3 = torch.ops.aten.permute.default(view_28, [1, 0]);  view_28 = None
	        mul_198 = sym_size_int * 1500
	        view_29 = torch.ops.aten.view.default(mul_190, [mul_198, 1280]);  mul_190 = mul_198 = None
	        mm = torch.ops.aten.mm.default(view_29, permute_3);  view_29 = permute_3 = None
	        view_30 = torch.ops.aten.view.default(mm, [sym_size_int, 1500, 1280]);  mm = None
	        view_31 = torch.ops.aten.view.default(view_30, [sym_size_int, -1, 20, 64]);  view_30 = None
	        permute_4 = torch.ops.aten.permute.default(view_31, [0, 2, 1, 3]);  view_31 = None
	        clone_3 = torch.ops.aten.clone.default(permute_4, memory_format = torch.contiguous_format);  permute_4 = None
	        amin_2 = torch.ops.aten.amin.default(add_32, [2])
	        amax_2 = torch.ops.aten.amax.default(add_32, [2])
	        full_4 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_2 = torch.ops.aten.minimum.default(amin_2, full_4);  amin_2 = full_4 = None
	        full_5 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_2 = torch.ops.aten.maximum.default(amax_2, full_5);  amax_2 = full_5 = None
	        sub_107 = torch.ops.aten.sub.Tensor(maximum_2, minimum_2);  maximum_2 = None
	        div_4 = torch.ops.aten.div.Tensor(sub_107, 255.0);  sub_107 = None
	        clamp_min_6 = torch.ops.aten.clamp_min.default(div_4, 1.1920928955078125e-07);  div_4 = None
	        div_5 = torch.ops.aten.div.Tensor(minimum_2, clamp_min_6);  minimum_2 = None
	        round_5 = torch.ops.aten.round.default(div_5);  div_5 = None
	        sub_113 = torch.ops.aten.sub.Tensor(-128, round_5);  round_5 = None
	        clamp_min_7 = torch.ops.aten.clamp_min.default(sub_113, -128);  sub_113 = None
	        clamp_max_4 = torch.ops.aten.clamp_max.default(clamp_min_7, 127);  clamp_min_7 = None
	        _assert_tensor_metadata_20 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_6, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_20 = None
	        _assert_tensor_metadata_21 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_4, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_21 = None
	        convert_element_type_12 = torch.ops.prims.convert_element_type.default(clamp_max_4, torch.int8);  clamp_max_4 = None
	        view_34 = torch.ops.aten.view.default(clamp_min_6, [sym_size_int, 1500, 1])
	        view_35 = torch.ops.aten.view.default(convert_element_type_12, [sym_size_int, 1500, 1])
	        reciprocal_2 = torch.ops.aten.reciprocal.default(view_34);  view_34 = None
	        mul_264 = torch.ops.aten.mul.Tensor(reciprocal_2, 1.0);  reciprocal_2 = None
	        mul_267 = torch.ops.aten.mul.Tensor(add_32, mul_264);  add_32 = mul_264 = None
	        round_6 = torch.ops.aten.round.default(mul_267);  mul_267 = None
	        add_419 = torch.ops.aten.add.Tensor(round_6, view_35);  round_6 = view_35 = None
	        clamp_min_8 = torch.ops.aten.clamp_min.default(add_419, -128);  add_419 = None
	        clamp_max_5 = torch.ops.aten.clamp_max.default(clamp_min_8, 127);  clamp_min_8 = None
	        _assert_tensor_metadata_22 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_5, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_22 = None
	        convert_element_type_13 = torch.ops.prims.convert_element_type.default(clamp_max_5, torch.int8);  clamp_max_5 = None
	        view_38 = torch.ops.aten.view.default(clamp_min_6, [sym_size_int, 1500, 1]);  clamp_min_6 = None
	        view_39 = torch.ops.aten.view.default(convert_element_type_12, [sym_size_int, 1500, 1]);  convert_element_type_12 = None
	        _assert_tensor_metadata_23 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_13, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_23 = None
	        convert_element_type_14 = torch.ops.prims.convert_element_type.default(convert_element_type_13, torch.float32);  convert_element_type_13 = None
	        _assert_tensor_metadata_24 = torch.ops.aten._assert_tensor_metadata.default(view_39, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_24 = None
	        convert_element_type_15 = torch.ops.prims.convert_element_type.default(view_39, torch.float32);  view_39 = None
	        sub_133 = torch.ops.aten.sub.Tensor(convert_element_type_14, convert_element_type_15);  convert_element_type_14 = convert_element_type_15 = None
	        mul_289 = torch.ops.aten.mul.Tensor(sub_133, view_38);  sub_133 = view_38 = None
	        _assert_tensor_metadata_25 = torch.ops.aten._assert_tensor_metadata.default(mul_289, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_25 = None
	        view_41 = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_42 = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_43 = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_26 = torch.ops.aten._assert_tensor_metadata.default(view_41, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_26 = None
	        convert_element_type_16 = torch.ops.prims.convert_element_type.default(view_41, torch.float32);  view_41 = None
	        _assert_tensor_metadata_27 = torch.ops.aten._assert_tensor_metadata.default(view_43, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_27 = None
	        convert_element_type_17 = torch.ops.prims.convert_element_type.default(view_43, torch.float32);  view_43 = None
	        sub_137 = torch.ops.aten.sub.Tensor(convert_element_type_16, convert_element_type_17);  convert_element_type_16 = convert_element_type_17 = None
	        mul_294 = torch.ops.aten.mul.Tensor(sub_137, view_42);  sub_137 = view_42 = None
	        view_44 = torch.ops.aten.view.default(mul_294, [1280, 1280]);  mul_294 = None
	        _assert_tensor_metadata_28 = torch.ops.aten._assert_tensor_metadata.default(view_44, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_28 = None
	        mul_299 = sym_size_int * 1500
	        view_45 = torch.ops.aten.view.default(mul_289, [mul_299, 1280]);  mul_289 = mul_299 = None
	        permute_5 = torch.ops.aten.permute.default(view_44, [1, 0]);  view_44 = None
	        addmm_1 = torch.ops.aten.addmm.default(model_audio_tower_layers_0_self_attn_v_proj_bias, view_45, permute_5);  model_audio_tower_layers_0_self_attn_v_proj_bias = view_45 = permute_5 = None
	        view_46 = torch.ops.aten.view.default(addmm_1, [sym_size_int, 1500, 1280]);  addmm_1 = None
	        view_47 = torch.ops.aten.view.default(view_46, [sym_size_int, -1, 20, 64]);  view_46 = None
	        permute_6 = torch.ops.aten.permute.default(view_47, [0, 2, 1, 3]);  view_47 = None
	        clone_4 = torch.ops.aten.clone.default(permute_6, memory_format = torch.contiguous_format);  permute_6 = None
	        _scaled_dot_product_efficient_attention = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_2, clone_3, clone_4, None, False, scale = 1.0);  clone_2 = clone_3 = clone_4 = None
	        getitem_2 = _scaled_dot_product_efficient_attention[0];  _scaled_dot_product_efficient_attention = None
	        permute_7 = torch.ops.aten.permute.default(getitem_2, [0, 2, 1, 3]);  getitem_2 = None
	        view_48 = torch.ops.aten.view.default(permute_7, [sym_size_int, 1500, -1]);  permute_7 = None
	        amin_3 = torch.ops.aten.amin.default(view_48, [2])
	        amax_3 = torch.ops.aten.amax.default(view_48, [2])
	        full_6 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_3 = torch.ops.aten.minimum.default(amin_3, full_6);  amin_3 = full_6 = None
	        full_7 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_3 = torch.ops.aten.maximum.default(amax_3, full_7);  amax_3 = full_7 = None
	        sub_155 = torch.ops.aten.sub.Tensor(maximum_3, minimum_3);  maximum_3 = None
	        div_6 = torch.ops.aten.div.Tensor(sub_155, 255.0);  sub_155 = None
	        clamp_min_9 = torch.ops.aten.clamp_min.default(div_6, 1.1920928955078125e-07);  div_6 = None
	        div_7 = torch.ops.aten.div.Tensor(minimum_3, clamp_min_9);  minimum_3 = None
	        round_7 = torch.ops.aten.round.default(div_7);  div_7 = None
	        sub_161 = torch.ops.aten.sub.Tensor(-128, round_7);  round_7 = None
	        clamp_min_10 = torch.ops.aten.clamp_min.default(sub_161, -128);  sub_161 = None
	        clamp_max_6 = torch.ops.aten.clamp_max.default(clamp_min_10, 127);  clamp_min_10 = None
	        _assert_tensor_metadata_29 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_9, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_29 = None
	        _assert_tensor_metadata_30 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_6, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_30 = None
	        convert_element_type_18 = torch.ops.prims.convert_element_type.default(clamp_max_6, torch.int8);  clamp_max_6 = None
	        view_51 = torch.ops.aten.view.default(clamp_min_9, [sym_size_int, 1500, 1])
	        view_52 = torch.ops.aten.view.default(convert_element_type_18, [sym_size_int, 1500, 1])
	        reciprocal_3 = torch.ops.aten.reciprocal.default(view_51);  view_51 = None
	        mul_369 = torch.ops.aten.mul.Tensor(reciprocal_3, 1.0);  reciprocal_3 = None
	        mul_372 = torch.ops.aten.mul.Tensor(view_48, mul_369);  view_48 = mul_369 = None
	        round_8 = torch.ops.aten.round.default(mul_372);  mul_372 = None
	        add_583 = torch.ops.aten.add.Tensor(round_8, view_52);  round_8 = view_52 = None
	        clamp_min_11 = torch.ops.aten.clamp_min.default(add_583, -128);  add_583 = None
	        clamp_max_7 = torch.ops.aten.clamp_max.default(clamp_min_11, 127);  clamp_min_11 = None
	        _assert_tensor_metadata_31 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_7, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_31 = None
	        convert_element_type_19 = torch.ops.prims.convert_element_type.default(clamp_max_7, torch.int8);  clamp_max_7 = None
	        view_55 = torch.ops.aten.view.default(clamp_min_9, [sym_size_int, 1500, 1]);  clamp_min_9 = None
	        view_56 = torch.ops.aten.view.default(convert_element_type_18, [sym_size_int, 1500, 1]);  convert_element_type_18 = None
	        _assert_tensor_metadata_32 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_19, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_32 = None
	        convert_element_type_20 = torch.ops.prims.convert_element_type.default(convert_element_type_19, torch.float32);  convert_element_type_19 = None
	        _assert_tensor_metadata_33 = torch.ops.aten._assert_tensor_metadata.default(view_56, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_33 = None
	        convert_element_type_21 = torch.ops.prims.convert_element_type.default(view_56, torch.float32);  view_56 = None
	        sub_181 = torch.ops.aten.sub.Tensor(convert_element_type_20, convert_element_type_21);  convert_element_type_20 = convert_element_type_21 = None
	        mul_394 = torch.ops.aten.mul.Tensor(sub_181, view_55);  sub_181 = view_55 = None
	        _assert_tensor_metadata_34 = torch.ops.aten._assert_tensor_metadata.default(mul_394, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_34 = None
	        view_58 = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_59 = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_60 = torch.ops.aten.view.default(model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_35 = torch.ops.aten._assert_tensor_metadata.default(view_58, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_35 = None
	        convert_element_type_22 = torch.ops.prims.convert_element_type.default(view_58, torch.float32);  view_58 = None
	        _assert_tensor_metadata_36 = torch.ops.aten._assert_tensor_metadata.default(view_60, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_36 = None
	        convert_element_type_23 = torch.ops.prims.convert_element_type.default(view_60, torch.float32);  view_60 = None
	        sub_185 = torch.ops.aten.sub.Tensor(convert_element_type_22, convert_element_type_23);  convert_element_type_22 = convert_element_type_23 = None
	        mul_399 = torch.ops.aten.mul.Tensor(sub_185, view_59);  sub_185 = view_59 = None
	        view_61 = torch.ops.aten.view.default(mul_399, [1280, 1280]);  mul_399 = None
	        _assert_tensor_metadata_37 = torch.ops.aten._assert_tensor_metadata.default(view_61, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_37 = None
	        mul_404 = sym_size_int * 1500
	        view_62 = torch.ops.aten.view.default(mul_394, [mul_404, 1280]);  mul_394 = mul_404 = None
	        permute_8 = torch.ops.aten.permute.default(view_61, [1, 0]);  view_61 = None
	        addmm_2 = torch.ops.aten.addmm.default(model_audio_tower_layers_0_self_attn_out_proj_bias, view_62, permute_8);  model_audio_tower_layers_0_self_attn_out_proj_bias = view_62 = permute_8 = None
	        view_63 = torch.ops.aten.view.default(addmm_2, [sym_size_int, 1500, 1280]);  addmm_2 = None
	        add_646 = torch.ops.aten.add.Tensor(add_22, view_63);  add_22 = view_63 = None
	        clone_6 = torch.ops.aten.clone.default(add_646, memory_format = torch.contiguous_format)
	        var_mean_1 = torch.ops.aten.var_mean.correction(clone_6, [2], correction = 0, keepdim = True)
	        getitem_6 = var_mean_1[0]
	        getitem_7 = var_mean_1[1];  var_mean_1 = None
	        add_651 = torch.ops.aten.add.Tensor(getitem_6, 1e-05);  getitem_6 = None
	        rsqrt_1 = torch.ops.aten.rsqrt.default(add_651);  add_651 = None
	        sub_191 = torch.ops.aten.sub.Tensor(clone_6, getitem_7);  clone_6 = getitem_7 = None
	        mul_415 = torch.ops.aten.mul.Tensor(sub_191, rsqrt_1);  sub_191 = rsqrt_1 = None
	        mul_416 = torch.ops.aten.mul.Tensor(mul_415, model_audio_tower_layers_0_final_layer_norm_weight);  mul_415 = model_audio_tower_layers_0_final_layer_norm_weight = None
	        add_652 = torch.ops.aten.add.Tensor(mul_416, model_audio_tower_layers_0_final_layer_norm_bias);  mul_416 = model_audio_tower_layers_0_final_layer_norm_bias = None
	        amin_4 = torch.ops.aten.amin.default(add_652, [2])
	        amax_4 = torch.ops.aten.amax.default(add_652, [2])
	        full_8 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_4 = torch.ops.aten.minimum.default(amin_4, full_8);  amin_4 = full_8 = None
	        full_9 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_4 = torch.ops.aten.maximum.default(amax_4, full_9);  amax_4 = full_9 = None
	        sub_202 = torch.ops.aten.sub.Tensor(maximum_4, minimum_4);  maximum_4 = None
	        div_8 = torch.ops.aten.div.Tensor(sub_202, 255.0);  sub_202 = None
	        clamp_min_12 = torch.ops.aten.clamp_min.default(div_8, 1.1920928955078125e-07);  div_8 = None
	        div_9 = torch.ops.aten.div.Tensor(minimum_4, clamp_min_12);  minimum_4 = None
	        round_9 = torch.ops.aten.round.default(div_9);  div_9 = None
	        sub_208 = torch.ops.aten.sub.Tensor(-128, round_9);  round_9 = None
	        clamp_min_13 = torch.ops.aten.clamp_min.default(sub_208, -128);  sub_208 = None
	        clamp_max_8 = torch.ops.aten.clamp_max.default(clamp_min_13, 127);  clamp_min_13 = None
	        _assert_tensor_metadata_38 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_12, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_38 = None
	        _assert_tensor_metadata_39 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_8, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_39 = None
	        convert_element_type_24 = torch.ops.prims.convert_element_type.default(clamp_max_8, torch.int8);  clamp_max_8 = None
	        view_66 = torch.ops.aten.view.default(clamp_min_12, [sym_size_int, 1500, 1])
	        view_67 = torch.ops.aten.view.default(convert_element_type_24, [sym_size_int, 1500, 1])
	        reciprocal_4 = torch.ops.aten.reciprocal.default(view_66);  view_66 = None
	        mul_464 = torch.ops.aten.mul.Tensor(reciprocal_4, 1.0);  reciprocal_4 = None
	        mul_467 = torch.ops.aten.mul.Tensor(add_652, mul_464);  add_652 = mul_464 = None
	        round_10 = torch.ops.aten.round.default(mul_467);  mul_467 = None
	        add_739 = torch.ops.aten.add.Tensor(round_10, view_67);  round_10 = view_67 = None
	        clamp_min_14 = torch.ops.aten.clamp_min.default(add_739, -128);  add_739 = None
	        clamp_max_9 = torch.ops.aten.clamp_max.default(clamp_min_14, 127);  clamp_min_14 = None
	        _assert_tensor_metadata_40 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_9, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_40 = None
	        convert_element_type_25 = torch.ops.prims.convert_element_type.default(clamp_max_9, torch.int8);  clamp_max_9 = None
	        view_70 = torch.ops.aten.view.default(clamp_min_12, [sym_size_int, 1500, 1]);  clamp_min_12 = None
	        view_71 = torch.ops.aten.view.default(convert_element_type_24, [sym_size_int, 1500, 1]);  convert_element_type_24 = None
	        _assert_tensor_metadata_41 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_25, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_41 = None
	        convert_element_type_26 = torch.ops.prims.convert_element_type.default(convert_element_type_25, torch.float32);  convert_element_type_25 = None
	        _assert_tensor_metadata_42 = torch.ops.aten._assert_tensor_metadata.default(view_71, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_42 = None
	        convert_element_type_27 = torch.ops.prims.convert_element_type.default(view_71, torch.float32);  view_71 = None
	        sub_228 = torch.ops.aten.sub.Tensor(convert_element_type_26, convert_element_type_27);  convert_element_type_26 = convert_element_type_27 = None
	        mul_489 = torch.ops.aten.mul.Tensor(sub_228, view_70);  sub_228 = view_70 = None
	        _assert_tensor_metadata_43 = torch.ops.aten._assert_tensor_metadata.default(mul_489, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_43 = None
	        view_73 = torch.ops.aten.view.default(model_audio_tower_layers_0_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_0_fc1_parametrizations_weight_original0 = None
	        view_74 = torch.ops.aten.view.default(model_audio_tower_layers_0_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_0_fc1_parametrizations_weight_original1 = None
	        view_75 = torch.ops.aten.view.default(model_audio_tower_layers_0_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_0_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_44 = torch.ops.aten._assert_tensor_metadata.default(view_73, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_44 = None
	        convert_element_type_28 = torch.ops.prims.convert_element_type.default(view_73, torch.float32);  view_73 = None
	        _assert_tensor_metadata_45 = torch.ops.aten._assert_tensor_metadata.default(view_75, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_45 = None
	        convert_element_type_29 = torch.ops.prims.convert_element_type.default(view_75, torch.float32);  view_75 = None
	        sub_232 = torch.ops.aten.sub.Tensor(convert_element_type_28, convert_element_type_29);  convert_element_type_28 = convert_element_type_29 = None
	        mul_494 = torch.ops.aten.mul.Tensor(sub_232, view_74);  sub_232 = view_74 = None
	        view_76 = torch.ops.aten.view.default(mul_494, [5120, 1280]);  mul_494 = None
	        _assert_tensor_metadata_46 = torch.ops.aten._assert_tensor_metadata.default(view_76, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_46 = None
	        mul_499 = sym_size_int * 1500
	        view_77 = torch.ops.aten.view.default(mul_489, [mul_499, 1280]);  mul_489 = mul_499 = None
	        permute_9 = torch.ops.aten.permute.default(view_76, [1, 0]);  view_76 = None
	        addmm_3 = torch.ops.aten.addmm.default(model_audio_tower_layers_0_fc1_bias, view_77, permute_9);  model_audio_tower_layers_0_fc1_bias = view_77 = permute_9 = None
	        view_78 = torch.ops.aten.view.default(addmm_3, [sym_size_int, 1500, 5120]);  addmm_3 = None
	        mul_506 = torch.ops.aten.mul.Tensor(view_78, 0.5)
	        mul_507 = torch.ops.aten.mul.Tensor(view_78, 0.7071067811865476);  view_78 = None
	        erf_2 = torch.ops.aten.erf.default(mul_507);  mul_507 = None
	        add_798 = torch.ops.aten.add.Tensor(erf_2, 1);  erf_2 = None
	        mul_508 = torch.ops.aten.mul.Tensor(mul_506, add_798);  mul_506 = add_798 = None
	        amin_5 = torch.ops.aten.amin.default(mul_508, [2])
	        amax_5 = torch.ops.aten.amax.default(mul_508, [2])
	        full_10 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_5 = torch.ops.aten.minimum.default(amin_5, full_10);  amin_5 = full_10 = None
	        full_11 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_5 = torch.ops.aten.maximum.default(amax_5, full_11);  amax_5 = full_11 = None
	        sub_245 = torch.ops.aten.sub.Tensor(maximum_5, minimum_5);  maximum_5 = None
	        div_10 = torch.ops.aten.div.Tensor(sub_245, 255.0);  sub_245 = None
	        clamp_min_15 = torch.ops.aten.clamp_min.default(div_10, 1.1920928955078125e-07);  div_10 = None
	        div_11 = torch.ops.aten.div.Tensor(minimum_5, clamp_min_15);  minimum_5 = None
	        round_11 = torch.ops.aten.round.default(div_11);  div_11 = None
	        sub_251 = torch.ops.aten.sub.Tensor(-128, round_11);  round_11 = None
	        clamp_min_16 = torch.ops.aten.clamp_min.default(sub_251, -128);  sub_251 = None
	        clamp_max_10 = torch.ops.aten.clamp_max.default(clamp_min_16, 127);  clamp_min_16 = None
	        _assert_tensor_metadata_47 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_15, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_47 = None
	        _assert_tensor_metadata_48 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_10, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_48 = None
	        convert_element_type_30 = torch.ops.prims.convert_element_type.default(clamp_max_10, torch.int8);  clamp_max_10 = None
	        view_81 = torch.ops.aten.view.default(clamp_min_15, [sym_size_int, 1500, 1])
	        view_82 = torch.ops.aten.view.default(convert_element_type_30, [sym_size_int, 1500, 1])
	        reciprocal_5 = torch.ops.aten.reciprocal.default(view_81);  view_81 = None
	        mul_554 = torch.ops.aten.mul.Tensor(reciprocal_5, 1.0);  reciprocal_5 = None
	        mul_557 = torch.ops.aten.mul.Tensor(mul_508, mul_554);  mul_508 = mul_554 = None
	        round_12 = torch.ops.aten.round.default(mul_557);  mul_557 = None
	        add_881 = torch.ops.aten.add.Tensor(round_12, view_82);  round_12 = view_82 = None
	        clamp_min_17 = torch.ops.aten.clamp_min.default(add_881, -128);  add_881 = None
	        clamp_max_11 = torch.ops.aten.clamp_max.default(clamp_min_17, 127);  clamp_min_17 = None
	        _assert_tensor_metadata_49 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_11, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_49 = None
	        convert_element_type_31 = torch.ops.prims.convert_element_type.default(clamp_max_11, torch.int8);  clamp_max_11 = None
	        view_85 = torch.ops.aten.view.default(clamp_min_15, [sym_size_int, 1500, 1]);  clamp_min_15 = None
	        view_86 = torch.ops.aten.view.default(convert_element_type_30, [sym_size_int, 1500, 1]);  convert_element_type_30 = None
	        _assert_tensor_metadata_50 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_31, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_50 = None
	        convert_element_type_32 = torch.ops.prims.convert_element_type.default(convert_element_type_31, torch.float32);  convert_element_type_31 = None
	        _assert_tensor_metadata_51 = torch.ops.aten._assert_tensor_metadata.default(view_86, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_51 = None
	        convert_element_type_33 = torch.ops.prims.convert_element_type.default(view_86, torch.float32);  view_86 = None
	        sub_271 = torch.ops.aten.sub.Tensor(convert_element_type_32, convert_element_type_33);  convert_element_type_32 = convert_element_type_33 = None
	        mul_579 = torch.ops.aten.mul.Tensor(sub_271, view_85);  sub_271 = view_85 = None
	        _assert_tensor_metadata_52 = torch.ops.aten._assert_tensor_metadata.default(mul_579, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_52 = None
	        view_88 = torch.ops.aten.view.default(model_audio_tower_layers_0_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_0_fc2_parametrizations_weight_original0 = None
	        view_89 = torch.ops.aten.view.default(model_audio_tower_layers_0_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_0_fc2_parametrizations_weight_original1 = None
	        view_90 = torch.ops.aten.view.default(model_audio_tower_layers_0_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_0_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_53 = torch.ops.aten._assert_tensor_metadata.default(view_88, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_53 = None
	        convert_element_type_34 = torch.ops.prims.convert_element_type.default(view_88, torch.float32);  view_88 = None
	        _assert_tensor_metadata_54 = torch.ops.aten._assert_tensor_metadata.default(view_90, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_54 = None
	        convert_element_type_35 = torch.ops.prims.convert_element_type.default(view_90, torch.float32);  view_90 = None
	        sub_275 = torch.ops.aten.sub.Tensor(convert_element_type_34, convert_element_type_35);  convert_element_type_34 = convert_element_type_35 = None
	        mul_584 = torch.ops.aten.mul.Tensor(sub_275, view_89);  sub_275 = view_89 = None
	        view_91 = torch.ops.aten.view.default(mul_584, [1280, 5120]);  mul_584 = None
	        _assert_tensor_metadata_55 = torch.ops.aten._assert_tensor_metadata.default(view_91, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_55 = None
	        mul_589 = sym_size_int * 1500
	        view_92 = torch.ops.aten.view.default(mul_579, [mul_589, 5120]);  mul_579 = mul_589 = None
	        permute_10 = torch.ops.aten.permute.default(view_91, [1, 0]);  view_91 = None
	        addmm_4 = torch.ops.aten.addmm.default(model_audio_tower_layers_0_fc2_bias, view_92, permute_10);  model_audio_tower_layers_0_fc2_bias = view_92 = permute_10 = None
	        view_93 = torch.ops.aten.view.default(addmm_4, [sym_size_int, 1500, 1280]);  addmm_4 = None
	        add_944 = torch.ops.aten.add.Tensor(add_646, view_93);  add_646 = view_93 = None
	        clone_9 = torch.ops.aten.clone.default(add_944, memory_format = torch.contiguous_format)
	        var_mean_2 = torch.ops.aten.var_mean.correction(clone_9, [2], correction = 0, keepdim = True)
	        getitem_8 = var_mean_2[0]
	        getitem_9 = var_mean_2[1];  var_mean_2 = None
	        add_949 = torch.ops.aten.add.Tensor(getitem_8, 1e-05);  getitem_8 = None
	        rsqrt_2 = torch.ops.aten.rsqrt.default(add_949);  add_949 = None
	        sub_281 = torch.ops.aten.sub.Tensor(clone_9, getitem_9);  clone_9 = getitem_9 = None
	        mul_600 = torch.ops.aten.mul.Tensor(sub_281, rsqrt_2);  sub_281 = rsqrt_2 = None
	        mul_601 = torch.ops.aten.mul.Tensor(mul_600, model_audio_tower_layers_1_self_attn_layer_norm_weight);  mul_600 = model_audio_tower_layers_1_self_attn_layer_norm_weight = None
	        add_950 = torch.ops.aten.add.Tensor(mul_601, model_audio_tower_layers_1_self_attn_layer_norm_bias);  mul_601 = model_audio_tower_layers_1_self_attn_layer_norm_bias = None
	        amin_6 = torch.ops.aten.amin.default(add_950, [2])
	        amax_6 = torch.ops.aten.amax.default(add_950, [2])
	        full_12 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_6 = torch.ops.aten.minimum.default(amin_6, full_12);  amin_6 = full_12 = None
	        full_13 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_6 = torch.ops.aten.maximum.default(amax_6, full_13);  amax_6 = full_13 = None
	        sub_292 = torch.ops.aten.sub.Tensor(maximum_6, minimum_6);  maximum_6 = None
	        div_12 = torch.ops.aten.div.Tensor(sub_292, 255.0);  sub_292 = None
	        clamp_min_18 = torch.ops.aten.clamp_min.default(div_12, 1.1920928955078125e-07);  div_12 = None
	        div_13 = torch.ops.aten.div.Tensor(minimum_6, clamp_min_18);  minimum_6 = None
	        round_13 = torch.ops.aten.round.default(div_13);  div_13 = None
	        sub_298 = torch.ops.aten.sub.Tensor(-128, round_13);  round_13 = None
	        clamp_min_19 = torch.ops.aten.clamp_min.default(sub_298, -128);  sub_298 = None
	        clamp_max_12 = torch.ops.aten.clamp_max.default(clamp_min_19, 127);  clamp_min_19 = None
	        _assert_tensor_metadata_56 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_18, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_56 = None
	        _assert_tensor_metadata_57 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_12, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_57 = None
	        convert_element_type_36 = torch.ops.prims.convert_element_type.default(clamp_max_12, torch.int8);  clamp_max_12 = None
	        view_96 = torch.ops.aten.view.default(clamp_min_18, [sym_size_int, 1500, 1])
	        view_97 = torch.ops.aten.view.default(convert_element_type_36, [sym_size_int, 1500, 1])
	        reciprocal_6 = torch.ops.aten.reciprocal.default(view_96);  view_96 = None
	        mul_649 = torch.ops.aten.mul.Tensor(reciprocal_6, 1.0);  reciprocal_6 = None
	        mul_652 = torch.ops.aten.mul.Tensor(add_950, mul_649);  mul_649 = None
	        round_14 = torch.ops.aten.round.default(mul_652);  mul_652 = None
	        add_1037 = torch.ops.aten.add.Tensor(round_14, view_97);  round_14 = view_97 = None
	        clamp_min_20 = torch.ops.aten.clamp_min.default(add_1037, -128);  add_1037 = None
	        clamp_max_13 = torch.ops.aten.clamp_max.default(clamp_min_20, 127);  clamp_min_20 = None
	        _assert_tensor_metadata_58 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_13, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_58 = None
	        convert_element_type_37 = torch.ops.prims.convert_element_type.default(clamp_max_13, torch.int8);  clamp_max_13 = None
	        view_100 = torch.ops.aten.view.default(clamp_min_18, [sym_size_int, 1500, 1]);  clamp_min_18 = None
	        view_101 = torch.ops.aten.view.default(convert_element_type_36, [sym_size_int, 1500, 1]);  convert_element_type_36 = None
	        _assert_tensor_metadata_59 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_37, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_59 = None
	        convert_element_type_38 = torch.ops.prims.convert_element_type.default(convert_element_type_37, torch.float32);  convert_element_type_37 = None
	        _assert_tensor_metadata_60 = torch.ops.aten._assert_tensor_metadata.default(view_101, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_60 = None
	        convert_element_type_39 = torch.ops.prims.convert_element_type.default(view_101, torch.float32);  view_101 = None
	        sub_318 = torch.ops.aten.sub.Tensor(convert_element_type_38, convert_element_type_39);  convert_element_type_38 = convert_element_type_39 = None
	        mul_674 = torch.ops.aten.mul.Tensor(sub_318, view_100);  sub_318 = view_100 = None
	        _assert_tensor_metadata_61 = torch.ops.aten._assert_tensor_metadata.default(mul_674, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_61 = None
	        view_103 = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_104 = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_105 = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_62 = torch.ops.aten._assert_tensor_metadata.default(view_103, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_62 = None
	        convert_element_type_40 = torch.ops.prims.convert_element_type.default(view_103, torch.float32);  view_103 = None
	        _assert_tensor_metadata_63 = torch.ops.aten._assert_tensor_metadata.default(view_105, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_63 = None
	        convert_element_type_41 = torch.ops.prims.convert_element_type.default(view_105, torch.float32);  view_105 = None
	        sub_322 = torch.ops.aten.sub.Tensor(convert_element_type_40, convert_element_type_41);  convert_element_type_40 = convert_element_type_41 = None
	        mul_679 = torch.ops.aten.mul.Tensor(sub_322, view_104);  sub_322 = view_104 = None
	        view_106 = torch.ops.aten.view.default(mul_679, [1280, 1280]);  mul_679 = None
	        _assert_tensor_metadata_64 = torch.ops.aten._assert_tensor_metadata.default(view_106, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_64 = None
	        mul_684 = sym_size_int * 1500
	        view_107 = torch.ops.aten.view.default(mul_674, [mul_684, 1280]);  mul_674 = mul_684 = None
	        permute_11 = torch.ops.aten.permute.default(view_106, [1, 0]);  view_106 = None
	        addmm_5 = torch.ops.aten.addmm.default(model_audio_tower_layers_1_self_attn_q_proj_bias, view_107, permute_11);  model_audio_tower_layers_1_self_attn_q_proj_bias = view_107 = permute_11 = None
	        view_108 = torch.ops.aten.view.default(addmm_5, [sym_size_int, 1500, 1280]);  addmm_5 = None
	        mul_691 = torch.ops.aten.mul.Tensor(view_108, 0.125);  view_108 = None
	        view_109 = torch.ops.aten.view.default(mul_691, [sym_size_int, 1500, 20, 64]);  mul_691 = None
	        permute_12 = torch.ops.aten.permute.default(view_109, [0, 2, 1, 3]);  view_109 = None
	        clone_10 = torch.ops.aten.clone.default(permute_12, memory_format = torch.contiguous_format);  permute_12 = None
	        amin_7 = torch.ops.aten.amin.default(add_950, [2])
	        amax_7 = torch.ops.aten.amax.default(add_950, [2])
	        full_14 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_7 = torch.ops.aten.minimum.default(amin_7, full_14);  amin_7 = full_14 = None
	        full_15 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_7 = torch.ops.aten.maximum.default(amax_7, full_15);  amax_7 = full_15 = None
	        sub_337 = torch.ops.aten.sub.Tensor(maximum_7, minimum_7);  maximum_7 = None
	        div_14 = torch.ops.aten.div.Tensor(sub_337, 255.0);  sub_337 = None
	        clamp_min_21 = torch.ops.aten.clamp_min.default(div_14, 1.1920928955078125e-07);  div_14 = None
	        div_15 = torch.ops.aten.div.Tensor(minimum_7, clamp_min_21);  minimum_7 = None
	        round_15 = torch.ops.aten.round.default(div_15);  div_15 = None
	        sub_343 = torch.ops.aten.sub.Tensor(-128, round_15);  round_15 = None
	        clamp_min_22 = torch.ops.aten.clamp_min.default(sub_343, -128);  sub_343 = None
	        clamp_max_14 = torch.ops.aten.clamp_max.default(clamp_min_22, 127);  clamp_min_22 = None
	        _assert_tensor_metadata_65 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_21, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_65 = None
	        _assert_tensor_metadata_66 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_14, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_66 = None
	        convert_element_type_42 = torch.ops.prims.convert_element_type.default(clamp_max_14, torch.int8);  clamp_max_14 = None
	        view_112 = torch.ops.aten.view.default(clamp_min_21, [sym_size_int, 1500, 1])
	        view_113 = torch.ops.aten.view.default(convert_element_type_42, [sym_size_int, 1500, 1])
	        reciprocal_7 = torch.ops.aten.reciprocal.default(view_112);  view_112 = None
	        mul_745 = torch.ops.aten.mul.Tensor(reciprocal_7, 1.0);  reciprocal_7 = None
	        mul_748 = torch.ops.aten.mul.Tensor(add_950, mul_745);  mul_745 = None
	        round_16 = torch.ops.aten.round.default(mul_748);  mul_748 = None
	        add_1189 = torch.ops.aten.add.Tensor(round_16, view_113);  round_16 = view_113 = None
	        clamp_min_23 = torch.ops.aten.clamp_min.default(add_1189, -128);  add_1189 = None
	        clamp_max_15 = torch.ops.aten.clamp_max.default(clamp_min_23, 127);  clamp_min_23 = None
	        _assert_tensor_metadata_67 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_15, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_67 = None
	        convert_element_type_43 = torch.ops.prims.convert_element_type.default(clamp_max_15, torch.int8);  clamp_max_15 = None
	        view_116 = torch.ops.aten.view.default(clamp_min_21, [sym_size_int, 1500, 1]);  clamp_min_21 = None
	        view_117 = torch.ops.aten.view.default(convert_element_type_42, [sym_size_int, 1500, 1]);  convert_element_type_42 = None
	        _assert_tensor_metadata_68 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_43, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_68 = None
	        convert_element_type_44 = torch.ops.prims.convert_element_type.default(convert_element_type_43, torch.float32);  convert_element_type_43 = None
	        _assert_tensor_metadata_69 = torch.ops.aten._assert_tensor_metadata.default(view_117, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_69 = None
	        convert_element_type_45 = torch.ops.prims.convert_element_type.default(view_117, torch.float32);  view_117 = None
	        sub_363 = torch.ops.aten.sub.Tensor(convert_element_type_44, convert_element_type_45);  convert_element_type_44 = convert_element_type_45 = None
	        mul_770 = torch.ops.aten.mul.Tensor(sub_363, view_116);  sub_363 = view_116 = None
	        _assert_tensor_metadata_70 = torch.ops.aten._assert_tensor_metadata.default(mul_770, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_70 = None
	        view_119 = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_120 = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_121 = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_71 = torch.ops.aten._assert_tensor_metadata.default(view_119, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_71 = None
	        convert_element_type_46 = torch.ops.prims.convert_element_type.default(view_119, torch.float32);  view_119 = None
	        _assert_tensor_metadata_72 = torch.ops.aten._assert_tensor_metadata.default(view_121, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_72 = None
	        convert_element_type_47 = torch.ops.prims.convert_element_type.default(view_121, torch.float32);  view_121 = None
	        sub_367 = torch.ops.aten.sub.Tensor(convert_element_type_46, convert_element_type_47);  convert_element_type_46 = convert_element_type_47 = None
	        mul_775 = torch.ops.aten.mul.Tensor(sub_367, view_120);  sub_367 = view_120 = None
	        view_122 = torch.ops.aten.view.default(mul_775, [1280, 1280]);  mul_775 = None
	        _assert_tensor_metadata_73 = torch.ops.aten._assert_tensor_metadata.default(view_122, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_73 = None
	        permute_13 = torch.ops.aten.permute.default(view_122, [1, 0]);  view_122 = None
	        mul_778 = sym_size_int * 1500
	        view_123 = torch.ops.aten.view.default(mul_770, [mul_778, 1280]);  mul_770 = mul_778 = None
	        mm_1 = torch.ops.aten.mm.default(view_123, permute_13);  view_123 = permute_13 = None
	        view_124 = torch.ops.aten.view.default(mm_1, [sym_size_int, 1500, 1280]);  mm_1 = None
	        view_125 = torch.ops.aten.view.default(view_124, [sym_size_int, -1, 20, 64]);  view_124 = None
	        permute_14 = torch.ops.aten.permute.default(view_125, [0, 2, 1, 3]);  view_125 = None
	        clone_11 = torch.ops.aten.clone.default(permute_14, memory_format = torch.contiguous_format);  permute_14 = None
	        amin_8 = torch.ops.aten.amin.default(add_950, [2])
	        amax_8 = torch.ops.aten.amax.default(add_950, [2])
	        full_16 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_8 = torch.ops.aten.minimum.default(amin_8, full_16);  amin_8 = full_16 = None
	        full_17 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_8 = torch.ops.aten.maximum.default(amax_8, full_17);  amax_8 = full_17 = None
	        sub_381 = torch.ops.aten.sub.Tensor(maximum_8, minimum_8);  maximum_8 = None
	        div_16 = torch.ops.aten.div.Tensor(sub_381, 255.0);  sub_381 = None
	        clamp_min_24 = torch.ops.aten.clamp_min.default(div_16, 1.1920928955078125e-07);  div_16 = None
	        div_17 = torch.ops.aten.div.Tensor(minimum_8, clamp_min_24);  minimum_8 = None
	        round_17 = torch.ops.aten.round.default(div_17);  div_17 = None
	        sub_387 = torch.ops.aten.sub.Tensor(-128, round_17);  round_17 = None
	        clamp_min_25 = torch.ops.aten.clamp_min.default(sub_387, -128);  sub_387 = None
	        clamp_max_16 = torch.ops.aten.clamp_max.default(clamp_min_25, 127);  clamp_min_25 = None
	        _assert_tensor_metadata_74 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_24, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_74 = None
	        _assert_tensor_metadata_75 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_16, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_75 = None
	        convert_element_type_48 = torch.ops.prims.convert_element_type.default(clamp_max_16, torch.int8);  clamp_max_16 = None
	        view_128 = torch.ops.aten.view.default(clamp_min_24, [sym_size_int, 1500, 1])
	        view_129 = torch.ops.aten.view.default(convert_element_type_48, [sym_size_int, 1500, 1])
	        reciprocal_8 = torch.ops.aten.reciprocal.default(view_128);  view_128 = None
	        mul_844 = torch.ops.aten.mul.Tensor(reciprocal_8, 1.0);  reciprocal_8 = None
	        mul_847 = torch.ops.aten.mul.Tensor(add_950, mul_844);  add_950 = mul_844 = None
	        round_18 = torch.ops.aten.round.default(mul_847);  mul_847 = None
	        add_1337 = torch.ops.aten.add.Tensor(round_18, view_129);  round_18 = view_129 = None
	        clamp_min_26 = torch.ops.aten.clamp_min.default(add_1337, -128);  add_1337 = None
	        clamp_max_17 = torch.ops.aten.clamp_max.default(clamp_min_26, 127);  clamp_min_26 = None
	        _assert_tensor_metadata_76 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_17, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_76 = None
	        convert_element_type_49 = torch.ops.prims.convert_element_type.default(clamp_max_17, torch.int8);  clamp_max_17 = None
	        view_132 = torch.ops.aten.view.default(clamp_min_24, [sym_size_int, 1500, 1]);  clamp_min_24 = None
	        view_133 = torch.ops.aten.view.default(convert_element_type_48, [sym_size_int, 1500, 1]);  convert_element_type_48 = None
	        _assert_tensor_metadata_77 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_49, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_77 = None
	        convert_element_type_50 = torch.ops.prims.convert_element_type.default(convert_element_type_49, torch.float32);  convert_element_type_49 = None
	        _assert_tensor_metadata_78 = torch.ops.aten._assert_tensor_metadata.default(view_133, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_78 = None
	        convert_element_type_51 = torch.ops.prims.convert_element_type.default(view_133, torch.float32);  view_133 = None
	        sub_407 = torch.ops.aten.sub.Tensor(convert_element_type_50, convert_element_type_51);  convert_element_type_50 = convert_element_type_51 = None
	        mul_869 = torch.ops.aten.mul.Tensor(sub_407, view_132);  sub_407 = view_132 = None
	        _assert_tensor_metadata_79 = torch.ops.aten._assert_tensor_metadata.default(mul_869, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_79 = None
	        view_135 = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_136 = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_137 = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_80 = torch.ops.aten._assert_tensor_metadata.default(view_135, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_80 = None
	        convert_element_type_52 = torch.ops.prims.convert_element_type.default(view_135, torch.float32);  view_135 = None
	        _assert_tensor_metadata_81 = torch.ops.aten._assert_tensor_metadata.default(view_137, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_81 = None
	        convert_element_type_53 = torch.ops.prims.convert_element_type.default(view_137, torch.float32);  view_137 = None
	        sub_411 = torch.ops.aten.sub.Tensor(convert_element_type_52, convert_element_type_53);  convert_element_type_52 = convert_element_type_53 = None
	        mul_874 = torch.ops.aten.mul.Tensor(sub_411, view_136);  sub_411 = view_136 = None
	        view_138 = torch.ops.aten.view.default(mul_874, [1280, 1280]);  mul_874 = None
	        _assert_tensor_metadata_82 = torch.ops.aten._assert_tensor_metadata.default(view_138, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_82 = None
	        mul_879 = sym_size_int * 1500
	        view_139 = torch.ops.aten.view.default(mul_869, [mul_879, 1280]);  mul_869 = mul_879 = None
	        permute_15 = torch.ops.aten.permute.default(view_138, [1, 0]);  view_138 = None
	        addmm_6 = torch.ops.aten.addmm.default(model_audio_tower_layers_1_self_attn_v_proj_bias, view_139, permute_15);  model_audio_tower_layers_1_self_attn_v_proj_bias = view_139 = permute_15 = None
	        view_140 = torch.ops.aten.view.default(addmm_6, [sym_size_int, 1500, 1280]);  addmm_6 = None
	        view_141 = torch.ops.aten.view.default(view_140, [sym_size_int, -1, 20, 64]);  view_140 = None
	        permute_16 = torch.ops.aten.permute.default(view_141, [0, 2, 1, 3]);  view_141 = None
	        clone_12 = torch.ops.aten.clone.default(permute_16, memory_format = torch.contiguous_format);  permute_16 = None
	        _scaled_dot_product_efficient_attention_1 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_10, clone_11, clone_12, None, False, scale = 1.0);  clone_10 = clone_11 = clone_12 = None
	        getitem_10 = _scaled_dot_product_efficient_attention_1[0];  _scaled_dot_product_efficient_attention_1 = None
	        permute_17 = torch.ops.aten.permute.default(getitem_10, [0, 2, 1, 3]);  getitem_10 = None
	        view_142 = torch.ops.aten.view.default(permute_17, [sym_size_int, 1500, -1]);  permute_17 = None
	        amin_9 = torch.ops.aten.amin.default(view_142, [2])
	        amax_9 = torch.ops.aten.amax.default(view_142, [2])
	        full_18 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_9 = torch.ops.aten.minimum.default(amin_9, full_18);  amin_9 = full_18 = None
	        full_19 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_9 = torch.ops.aten.maximum.default(amax_9, full_19);  amax_9 = full_19 = None
	        sub_429 = torch.ops.aten.sub.Tensor(maximum_9, minimum_9);  maximum_9 = None
	        div_18 = torch.ops.aten.div.Tensor(sub_429, 255.0);  sub_429 = None
	        clamp_min_27 = torch.ops.aten.clamp_min.default(div_18, 1.1920928955078125e-07);  div_18 = None
	        div_19 = torch.ops.aten.div.Tensor(minimum_9, clamp_min_27);  minimum_9 = None
	        round_19 = torch.ops.aten.round.default(div_19);  div_19 = None
	        sub_435 = torch.ops.aten.sub.Tensor(-128, round_19);  round_19 = None
	        clamp_min_28 = torch.ops.aten.clamp_min.default(sub_435, -128);  sub_435 = None
	        clamp_max_18 = torch.ops.aten.clamp_max.default(clamp_min_28, 127);  clamp_min_28 = None
	        _assert_tensor_metadata_83 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_27, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_83 = None
	        _assert_tensor_metadata_84 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_18, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_84 = None
	        convert_element_type_54 = torch.ops.prims.convert_element_type.default(clamp_max_18, torch.int8);  clamp_max_18 = None
	        view_145 = torch.ops.aten.view.default(clamp_min_27, [sym_size_int, 1500, 1])
	        view_146 = torch.ops.aten.view.default(convert_element_type_54, [sym_size_int, 1500, 1])
	        reciprocal_9 = torch.ops.aten.reciprocal.default(view_145);  view_145 = None
	        mul_949 = torch.ops.aten.mul.Tensor(reciprocal_9, 1.0);  reciprocal_9 = None
	        mul_952 = torch.ops.aten.mul.Tensor(view_142, mul_949);  view_142 = mul_949 = None
	        round_20 = torch.ops.aten.round.default(mul_952);  mul_952 = None
	        add_1501 = torch.ops.aten.add.Tensor(round_20, view_146);  round_20 = view_146 = None
	        clamp_min_29 = torch.ops.aten.clamp_min.default(add_1501, -128);  add_1501 = None
	        clamp_max_19 = torch.ops.aten.clamp_max.default(clamp_min_29, 127);  clamp_min_29 = None
	        _assert_tensor_metadata_85 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_19, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_85 = None
	        convert_element_type_55 = torch.ops.prims.convert_element_type.default(clamp_max_19, torch.int8);  clamp_max_19 = None
	        view_149 = torch.ops.aten.view.default(clamp_min_27, [sym_size_int, 1500, 1]);  clamp_min_27 = None
	        view_150 = torch.ops.aten.view.default(convert_element_type_54, [sym_size_int, 1500, 1]);  convert_element_type_54 = None
	        _assert_tensor_metadata_86 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_55, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_86 = None
	        convert_element_type_56 = torch.ops.prims.convert_element_type.default(convert_element_type_55, torch.float32);  convert_element_type_55 = None
	        _assert_tensor_metadata_87 = torch.ops.aten._assert_tensor_metadata.default(view_150, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_87 = None
	        convert_element_type_57 = torch.ops.prims.convert_element_type.default(view_150, torch.float32);  view_150 = None
	        sub_455 = torch.ops.aten.sub.Tensor(convert_element_type_56, convert_element_type_57);  convert_element_type_56 = convert_element_type_57 = None
	        mul_974 = torch.ops.aten.mul.Tensor(sub_455, view_149);  sub_455 = view_149 = None
	        _assert_tensor_metadata_88 = torch.ops.aten._assert_tensor_metadata.default(mul_974, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_88 = None
	        view_152 = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_153 = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_154 = torch.ops.aten.view.default(model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_89 = torch.ops.aten._assert_tensor_metadata.default(view_152, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_89 = None
	        convert_element_type_58 = torch.ops.prims.convert_element_type.default(view_152, torch.float32);  view_152 = None
	        _assert_tensor_metadata_90 = torch.ops.aten._assert_tensor_metadata.default(view_154, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_90 = None
	        convert_element_type_59 = torch.ops.prims.convert_element_type.default(view_154, torch.float32);  view_154 = None
	        sub_459 = torch.ops.aten.sub.Tensor(convert_element_type_58, convert_element_type_59);  convert_element_type_58 = convert_element_type_59 = None
	        mul_979 = torch.ops.aten.mul.Tensor(sub_459, view_153);  sub_459 = view_153 = None
	        view_155 = torch.ops.aten.view.default(mul_979, [1280, 1280]);  mul_979 = None
	        _assert_tensor_metadata_91 = torch.ops.aten._assert_tensor_metadata.default(view_155, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_91 = None
	        mul_984 = sym_size_int * 1500
	        view_156 = torch.ops.aten.view.default(mul_974, [mul_984, 1280]);  mul_974 = mul_984 = None
	        permute_18 = torch.ops.aten.permute.default(view_155, [1, 0]);  view_155 = None
	        addmm_7 = torch.ops.aten.addmm.default(model_audio_tower_layers_1_self_attn_out_proj_bias, view_156, permute_18);  model_audio_tower_layers_1_self_attn_out_proj_bias = view_156 = permute_18 = None
	        view_157 = torch.ops.aten.view.default(addmm_7, [sym_size_int, 1500, 1280]);  addmm_7 = None
	        add_1564 = torch.ops.aten.add.Tensor(add_944, view_157);  add_944 = view_157 = None
	        clone_14 = torch.ops.aten.clone.default(add_1564, memory_format = torch.contiguous_format)
	        var_mean_3 = torch.ops.aten.var_mean.correction(clone_14, [2], correction = 0, keepdim = True)
	        getitem_14 = var_mean_3[0]
	        getitem_15 = var_mean_3[1];  var_mean_3 = None
	        add_1569 = torch.ops.aten.add.Tensor(getitem_14, 1e-05);  getitem_14 = None
	        rsqrt_3 = torch.ops.aten.rsqrt.default(add_1569);  add_1569 = None
	        sub_465 = torch.ops.aten.sub.Tensor(clone_14, getitem_15);  clone_14 = getitem_15 = None
	        mul_995 = torch.ops.aten.mul.Tensor(sub_465, rsqrt_3);  sub_465 = rsqrt_3 = None
	        mul_996 = torch.ops.aten.mul.Tensor(mul_995, model_audio_tower_layers_1_final_layer_norm_weight);  mul_995 = model_audio_tower_layers_1_final_layer_norm_weight = None
	        add_1570 = torch.ops.aten.add.Tensor(mul_996, model_audio_tower_layers_1_final_layer_norm_bias);  mul_996 = model_audio_tower_layers_1_final_layer_norm_bias = None
	        amin_10 = torch.ops.aten.amin.default(add_1570, [2])
	        amax_10 = torch.ops.aten.amax.default(add_1570, [2])
	        full_20 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_10 = torch.ops.aten.minimum.default(amin_10, full_20);  amin_10 = full_20 = None
	        full_21 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_10 = torch.ops.aten.maximum.default(amax_10, full_21);  amax_10 = full_21 = None
	        sub_476 = torch.ops.aten.sub.Tensor(maximum_10, minimum_10);  maximum_10 = None
	        div_20 = torch.ops.aten.div.Tensor(sub_476, 255.0);  sub_476 = None
	        clamp_min_30 = torch.ops.aten.clamp_min.default(div_20, 1.1920928955078125e-07);  div_20 = None
	        div_21 = torch.ops.aten.div.Tensor(minimum_10, clamp_min_30);  minimum_10 = None
	        round_21 = torch.ops.aten.round.default(div_21);  div_21 = None
	        sub_482 = torch.ops.aten.sub.Tensor(-128, round_21);  round_21 = None
	        clamp_min_31 = torch.ops.aten.clamp_min.default(sub_482, -128);  sub_482 = None
	        clamp_max_20 = torch.ops.aten.clamp_max.default(clamp_min_31, 127);  clamp_min_31 = None
	        _assert_tensor_metadata_92 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_30, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_92 = None
	        _assert_tensor_metadata_93 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_20, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_93 = None
	        convert_element_type_60 = torch.ops.prims.convert_element_type.default(clamp_max_20, torch.int8);  clamp_max_20 = None
	        view_160 = torch.ops.aten.view.default(clamp_min_30, [sym_size_int, 1500, 1])
	        view_161 = torch.ops.aten.view.default(convert_element_type_60, [sym_size_int, 1500, 1])
	        reciprocal_10 = torch.ops.aten.reciprocal.default(view_160);  view_160 = None
	        mul_1044 = torch.ops.aten.mul.Tensor(reciprocal_10, 1.0);  reciprocal_10 = None
	        mul_1047 = torch.ops.aten.mul.Tensor(add_1570, mul_1044);  add_1570 = mul_1044 = None
	        round_22 = torch.ops.aten.round.default(mul_1047);  mul_1047 = None
	        add_1657 = torch.ops.aten.add.Tensor(round_22, view_161);  round_22 = view_161 = None
	        clamp_min_32 = torch.ops.aten.clamp_min.default(add_1657, -128);  add_1657 = None
	        clamp_max_21 = torch.ops.aten.clamp_max.default(clamp_min_32, 127);  clamp_min_32 = None
	        _assert_tensor_metadata_94 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_21, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_94 = None
	        convert_element_type_61 = torch.ops.prims.convert_element_type.default(clamp_max_21, torch.int8);  clamp_max_21 = None
	        view_164 = torch.ops.aten.view.default(clamp_min_30, [sym_size_int, 1500, 1]);  clamp_min_30 = None
	        view_165 = torch.ops.aten.view.default(convert_element_type_60, [sym_size_int, 1500, 1]);  convert_element_type_60 = None
	        _assert_tensor_metadata_95 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_61, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_95 = None
	        convert_element_type_62 = torch.ops.prims.convert_element_type.default(convert_element_type_61, torch.float32);  convert_element_type_61 = None
	        _assert_tensor_metadata_96 = torch.ops.aten._assert_tensor_metadata.default(view_165, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_96 = None
	        convert_element_type_63 = torch.ops.prims.convert_element_type.default(view_165, torch.float32);  view_165 = None
	        sub_502 = torch.ops.aten.sub.Tensor(convert_element_type_62, convert_element_type_63);  convert_element_type_62 = convert_element_type_63 = None
	        mul_1069 = torch.ops.aten.mul.Tensor(sub_502, view_164);  sub_502 = view_164 = None
	        _assert_tensor_metadata_97 = torch.ops.aten._assert_tensor_metadata.default(mul_1069, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_97 = None
	        view_167 = torch.ops.aten.view.default(model_audio_tower_layers_1_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_1_fc1_parametrizations_weight_original0 = None
	        view_168 = torch.ops.aten.view.default(model_audio_tower_layers_1_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_1_fc1_parametrizations_weight_original1 = None
	        view_169 = torch.ops.aten.view.default(model_audio_tower_layers_1_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_1_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_98 = torch.ops.aten._assert_tensor_metadata.default(view_167, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_98 = None
	        convert_element_type_64 = torch.ops.prims.convert_element_type.default(view_167, torch.float32);  view_167 = None
	        _assert_tensor_metadata_99 = torch.ops.aten._assert_tensor_metadata.default(view_169, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_99 = None
	        convert_element_type_65 = torch.ops.prims.convert_element_type.default(view_169, torch.float32);  view_169 = None
	        sub_506 = torch.ops.aten.sub.Tensor(convert_element_type_64, convert_element_type_65);  convert_element_type_64 = convert_element_type_65 = None
	        mul_1074 = torch.ops.aten.mul.Tensor(sub_506, view_168);  sub_506 = view_168 = None
	        view_170 = torch.ops.aten.view.default(mul_1074, [5120, 1280]);  mul_1074 = None
	        _assert_tensor_metadata_100 = torch.ops.aten._assert_tensor_metadata.default(view_170, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_100 = None
	        mul_1079 = sym_size_int * 1500
	        view_171 = torch.ops.aten.view.default(mul_1069, [mul_1079, 1280]);  mul_1069 = mul_1079 = None
	        permute_19 = torch.ops.aten.permute.default(view_170, [1, 0]);  view_170 = None
	        addmm_8 = torch.ops.aten.addmm.default(model_audio_tower_layers_1_fc1_bias, view_171, permute_19);  model_audio_tower_layers_1_fc1_bias = view_171 = permute_19 = None
	        view_172 = torch.ops.aten.view.default(addmm_8, [sym_size_int, 1500, 5120]);  addmm_8 = None
	        mul_1086 = torch.ops.aten.mul.Tensor(view_172, 0.5)
	        mul_1087 = torch.ops.aten.mul.Tensor(view_172, 0.7071067811865476);  view_172 = None
	        erf_3 = torch.ops.aten.erf.default(mul_1087);  mul_1087 = None
	        add_1716 = torch.ops.aten.add.Tensor(erf_3, 1);  erf_3 = None
	        mul_1088 = torch.ops.aten.mul.Tensor(mul_1086, add_1716);  mul_1086 = add_1716 = None
	        amin_11 = torch.ops.aten.amin.default(mul_1088, [2])
	        amax_11 = torch.ops.aten.amax.default(mul_1088, [2])
	        full_22 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_11 = torch.ops.aten.minimum.default(amin_11, full_22);  amin_11 = full_22 = None
	        full_23 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_11 = torch.ops.aten.maximum.default(amax_11, full_23);  amax_11 = full_23 = None
	        sub_519 = torch.ops.aten.sub.Tensor(maximum_11, minimum_11);  maximum_11 = None
	        div_22 = torch.ops.aten.div.Tensor(sub_519, 255.0);  sub_519 = None
	        clamp_min_33 = torch.ops.aten.clamp_min.default(div_22, 1.1920928955078125e-07);  div_22 = None
	        div_23 = torch.ops.aten.div.Tensor(minimum_11, clamp_min_33);  minimum_11 = None
	        round_23 = torch.ops.aten.round.default(div_23);  div_23 = None
	        sub_525 = torch.ops.aten.sub.Tensor(-128, round_23);  round_23 = None
	        clamp_min_34 = torch.ops.aten.clamp_min.default(sub_525, -128);  sub_525 = None
	        clamp_max_22 = torch.ops.aten.clamp_max.default(clamp_min_34, 127);  clamp_min_34 = None
	        _assert_tensor_metadata_101 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_33, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_101 = None
	        _assert_tensor_metadata_102 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_22, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_102 = None
	        convert_element_type_66 = torch.ops.prims.convert_element_type.default(clamp_max_22, torch.int8);  clamp_max_22 = None
	        view_175 = torch.ops.aten.view.default(clamp_min_33, [sym_size_int, 1500, 1])
	        view_176 = torch.ops.aten.view.default(convert_element_type_66, [sym_size_int, 1500, 1])
	        reciprocal_11 = torch.ops.aten.reciprocal.default(view_175);  view_175 = None
	        mul_1134 = torch.ops.aten.mul.Tensor(reciprocal_11, 1.0);  reciprocal_11 = None
	        mul_1137 = torch.ops.aten.mul.Tensor(mul_1088, mul_1134);  mul_1088 = mul_1134 = None
	        round_24 = torch.ops.aten.round.default(mul_1137);  mul_1137 = None
	        add_1799 = torch.ops.aten.add.Tensor(round_24, view_176);  round_24 = view_176 = None
	        clamp_min_35 = torch.ops.aten.clamp_min.default(add_1799, -128);  add_1799 = None
	        clamp_max_23 = torch.ops.aten.clamp_max.default(clamp_min_35, 127);  clamp_min_35 = None
	        _assert_tensor_metadata_103 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_23, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_103 = None
	        convert_element_type_67 = torch.ops.prims.convert_element_type.default(clamp_max_23, torch.int8);  clamp_max_23 = None
	        view_179 = torch.ops.aten.view.default(clamp_min_33, [sym_size_int, 1500, 1]);  clamp_min_33 = None
	        view_180 = torch.ops.aten.view.default(convert_element_type_66, [sym_size_int, 1500, 1]);  convert_element_type_66 = None
	        _assert_tensor_metadata_104 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_67, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_104 = None
	        convert_element_type_68 = torch.ops.prims.convert_element_type.default(convert_element_type_67, torch.float32);  convert_element_type_67 = None
	        _assert_tensor_metadata_105 = torch.ops.aten._assert_tensor_metadata.default(view_180, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_105 = None
	        convert_element_type_69 = torch.ops.prims.convert_element_type.default(view_180, torch.float32);  view_180 = None
	        sub_545 = torch.ops.aten.sub.Tensor(convert_element_type_68, convert_element_type_69);  convert_element_type_68 = convert_element_type_69 = None
	        mul_1159 = torch.ops.aten.mul.Tensor(sub_545, view_179);  sub_545 = view_179 = None
	        _assert_tensor_metadata_106 = torch.ops.aten._assert_tensor_metadata.default(mul_1159, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_106 = None
	        view_182 = torch.ops.aten.view.default(model_audio_tower_layers_1_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_1_fc2_parametrizations_weight_original0 = None
	        view_183 = torch.ops.aten.view.default(model_audio_tower_layers_1_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_1_fc2_parametrizations_weight_original1 = None
	        view_184 = torch.ops.aten.view.default(model_audio_tower_layers_1_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_1_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_107 = torch.ops.aten._assert_tensor_metadata.default(view_182, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_107 = None
	        convert_element_type_70 = torch.ops.prims.convert_element_type.default(view_182, torch.float32);  view_182 = None
	        _assert_tensor_metadata_108 = torch.ops.aten._assert_tensor_metadata.default(view_184, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_108 = None
	        convert_element_type_71 = torch.ops.prims.convert_element_type.default(view_184, torch.float32);  view_184 = None
	        sub_549 = torch.ops.aten.sub.Tensor(convert_element_type_70, convert_element_type_71);  convert_element_type_70 = convert_element_type_71 = None
	        mul_1164 = torch.ops.aten.mul.Tensor(sub_549, view_183);  sub_549 = view_183 = None
	        view_185 = torch.ops.aten.view.default(mul_1164, [1280, 5120]);  mul_1164 = None
	        _assert_tensor_metadata_109 = torch.ops.aten._assert_tensor_metadata.default(view_185, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_109 = None
	        mul_1169 = sym_size_int * 1500
	        view_186 = torch.ops.aten.view.default(mul_1159, [mul_1169, 5120]);  mul_1159 = mul_1169 = None
	        permute_20 = torch.ops.aten.permute.default(view_185, [1, 0]);  view_185 = None
	        addmm_9 = torch.ops.aten.addmm.default(model_audio_tower_layers_1_fc2_bias, view_186, permute_20);  model_audio_tower_layers_1_fc2_bias = view_186 = permute_20 = None
	        view_187 = torch.ops.aten.view.default(addmm_9, [sym_size_int, 1500, 1280]);  addmm_9 = None
	        add_1862 = torch.ops.aten.add.Tensor(add_1564, view_187);  add_1564 = view_187 = None
	        clone_17 = torch.ops.aten.clone.default(add_1862, memory_format = torch.contiguous_format)
	        var_mean_4 = torch.ops.aten.var_mean.correction(clone_17, [2], correction = 0, keepdim = True)
	        getitem_16 = var_mean_4[0]
	        getitem_17 = var_mean_4[1];  var_mean_4 = None
	        add_1867 = torch.ops.aten.add.Tensor(getitem_16, 1e-05);  getitem_16 = None
	        rsqrt_4 = torch.ops.aten.rsqrt.default(add_1867);  add_1867 = None
	        sub_555 = torch.ops.aten.sub.Tensor(clone_17, getitem_17);  clone_17 = getitem_17 = None
	        mul_1180 = torch.ops.aten.mul.Tensor(sub_555, rsqrt_4);  sub_555 = rsqrt_4 = None
	        mul_1181 = torch.ops.aten.mul.Tensor(mul_1180, model_audio_tower_layers_2_self_attn_layer_norm_weight);  mul_1180 = model_audio_tower_layers_2_self_attn_layer_norm_weight = None
	        add_1868 = torch.ops.aten.add.Tensor(mul_1181, model_audio_tower_layers_2_self_attn_layer_norm_bias);  mul_1181 = model_audio_tower_layers_2_self_attn_layer_norm_bias = None
	        amin_12 = torch.ops.aten.amin.default(add_1868, [2])
	        amax_12 = torch.ops.aten.amax.default(add_1868, [2])
	        full_24 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_12 = torch.ops.aten.minimum.default(amin_12, full_24);  amin_12 = full_24 = None
	        full_25 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_12 = torch.ops.aten.maximum.default(amax_12, full_25);  amax_12 = full_25 = None
	        sub_566 = torch.ops.aten.sub.Tensor(maximum_12, minimum_12);  maximum_12 = None
	        div_24 = torch.ops.aten.div.Tensor(sub_566, 255.0);  sub_566 = None
	        clamp_min_36 = torch.ops.aten.clamp_min.default(div_24, 1.1920928955078125e-07);  div_24 = None
	        div_25 = torch.ops.aten.div.Tensor(minimum_12, clamp_min_36);  minimum_12 = None
	        round_25 = torch.ops.aten.round.default(div_25);  div_25 = None
	        sub_572 = torch.ops.aten.sub.Tensor(-128, round_25);  round_25 = None
	        clamp_min_37 = torch.ops.aten.clamp_min.default(sub_572, -128);  sub_572 = None
	        clamp_max_24 = torch.ops.aten.clamp_max.default(clamp_min_37, 127);  clamp_min_37 = None
	        _assert_tensor_metadata_110 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_36, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_110 = None
	        _assert_tensor_metadata_111 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_24, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_111 = None
	        convert_element_type_72 = torch.ops.prims.convert_element_type.default(clamp_max_24, torch.int8);  clamp_max_24 = None
	        view_190 = torch.ops.aten.view.default(clamp_min_36, [sym_size_int, 1500, 1])
	        view_191 = torch.ops.aten.view.default(convert_element_type_72, [sym_size_int, 1500, 1])
	        reciprocal_12 = torch.ops.aten.reciprocal.default(view_190);  view_190 = None
	        mul_1229 = torch.ops.aten.mul.Tensor(reciprocal_12, 1.0);  reciprocal_12 = None
	        mul_1232 = torch.ops.aten.mul.Tensor(add_1868, mul_1229);  mul_1229 = None
	        round_26 = torch.ops.aten.round.default(mul_1232);  mul_1232 = None
	        add_1955 = torch.ops.aten.add.Tensor(round_26, view_191);  round_26 = view_191 = None
	        clamp_min_38 = torch.ops.aten.clamp_min.default(add_1955, -128);  add_1955 = None
	        clamp_max_25 = torch.ops.aten.clamp_max.default(clamp_min_38, 127);  clamp_min_38 = None
	        _assert_tensor_metadata_112 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_25, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_112 = None
	        convert_element_type_73 = torch.ops.prims.convert_element_type.default(clamp_max_25, torch.int8);  clamp_max_25 = None
	        view_194 = torch.ops.aten.view.default(clamp_min_36, [sym_size_int, 1500, 1]);  clamp_min_36 = None
	        view_195 = torch.ops.aten.view.default(convert_element_type_72, [sym_size_int, 1500, 1]);  convert_element_type_72 = None
	        _assert_tensor_metadata_113 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_73, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_113 = None
	        convert_element_type_74 = torch.ops.prims.convert_element_type.default(convert_element_type_73, torch.float32);  convert_element_type_73 = None
	        _assert_tensor_metadata_114 = torch.ops.aten._assert_tensor_metadata.default(view_195, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_114 = None
	        convert_element_type_75 = torch.ops.prims.convert_element_type.default(view_195, torch.float32);  view_195 = None
	        sub_592 = torch.ops.aten.sub.Tensor(convert_element_type_74, convert_element_type_75);  convert_element_type_74 = convert_element_type_75 = None
	        mul_1254 = torch.ops.aten.mul.Tensor(sub_592, view_194);  sub_592 = view_194 = None
	        _assert_tensor_metadata_115 = torch.ops.aten._assert_tensor_metadata.default(mul_1254, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_115 = None
	        view_197 = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_198 = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_199 = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_116 = torch.ops.aten._assert_tensor_metadata.default(view_197, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_116 = None
	        convert_element_type_76 = torch.ops.prims.convert_element_type.default(view_197, torch.float32);  view_197 = None
	        _assert_tensor_metadata_117 = torch.ops.aten._assert_tensor_metadata.default(view_199, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_117 = None
	        convert_element_type_77 = torch.ops.prims.convert_element_type.default(view_199, torch.float32);  view_199 = None
	        sub_596 = torch.ops.aten.sub.Tensor(convert_element_type_76, convert_element_type_77);  convert_element_type_76 = convert_element_type_77 = None
	        mul_1259 = torch.ops.aten.mul.Tensor(sub_596, view_198);  sub_596 = view_198 = None
	        view_200 = torch.ops.aten.view.default(mul_1259, [1280, 1280]);  mul_1259 = None
	        _assert_tensor_metadata_118 = torch.ops.aten._assert_tensor_metadata.default(view_200, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_118 = None
	        mul_1264 = sym_size_int * 1500
	        view_201 = torch.ops.aten.view.default(mul_1254, [mul_1264, 1280]);  mul_1254 = mul_1264 = None
	        permute_21 = torch.ops.aten.permute.default(view_200, [1, 0]);  view_200 = None
	        addmm_10 = torch.ops.aten.addmm.default(model_audio_tower_layers_2_self_attn_q_proj_bias, view_201, permute_21);  model_audio_tower_layers_2_self_attn_q_proj_bias = view_201 = permute_21 = None
	        view_202 = torch.ops.aten.view.default(addmm_10, [sym_size_int, 1500, 1280]);  addmm_10 = None
	        mul_1271 = torch.ops.aten.mul.Tensor(view_202, 0.125);  view_202 = None
	        view_203 = torch.ops.aten.view.default(mul_1271, [sym_size_int, 1500, 20, 64]);  mul_1271 = None
	        permute_22 = torch.ops.aten.permute.default(view_203, [0, 2, 1, 3]);  view_203 = None
	        clone_18 = torch.ops.aten.clone.default(permute_22, memory_format = torch.contiguous_format);  permute_22 = None
	        amin_13 = torch.ops.aten.amin.default(add_1868, [2])
	        amax_13 = torch.ops.aten.amax.default(add_1868, [2])
	        full_26 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_13 = torch.ops.aten.minimum.default(amin_13, full_26);  amin_13 = full_26 = None
	        full_27 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_13 = torch.ops.aten.maximum.default(amax_13, full_27);  amax_13 = full_27 = None
	        sub_611 = torch.ops.aten.sub.Tensor(maximum_13, minimum_13);  maximum_13 = None
	        div_26 = torch.ops.aten.div.Tensor(sub_611, 255.0);  sub_611 = None
	        clamp_min_39 = torch.ops.aten.clamp_min.default(div_26, 1.1920928955078125e-07);  div_26 = None
	        div_27 = torch.ops.aten.div.Tensor(minimum_13, clamp_min_39);  minimum_13 = None
	        round_27 = torch.ops.aten.round.default(div_27);  div_27 = None
	        sub_617 = torch.ops.aten.sub.Tensor(-128, round_27);  round_27 = None
	        clamp_min_40 = torch.ops.aten.clamp_min.default(sub_617, -128);  sub_617 = None
	        clamp_max_26 = torch.ops.aten.clamp_max.default(clamp_min_40, 127);  clamp_min_40 = None
	        _assert_tensor_metadata_119 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_39, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_119 = None
	        _assert_tensor_metadata_120 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_26, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_120 = None
	        convert_element_type_78 = torch.ops.prims.convert_element_type.default(clamp_max_26, torch.int8);  clamp_max_26 = None
	        view_206 = torch.ops.aten.view.default(clamp_min_39, [sym_size_int, 1500, 1])
	        view_207 = torch.ops.aten.view.default(convert_element_type_78, [sym_size_int, 1500, 1])
	        reciprocal_13 = torch.ops.aten.reciprocal.default(view_206);  view_206 = None
	        mul_1325 = torch.ops.aten.mul.Tensor(reciprocal_13, 1.0);  reciprocal_13 = None
	        mul_1328 = torch.ops.aten.mul.Tensor(add_1868, mul_1325);  mul_1325 = None
	        round_28 = torch.ops.aten.round.default(mul_1328);  mul_1328 = None
	        add_2107 = torch.ops.aten.add.Tensor(round_28, view_207);  round_28 = view_207 = None
	        clamp_min_41 = torch.ops.aten.clamp_min.default(add_2107, -128);  add_2107 = None
	        clamp_max_27 = torch.ops.aten.clamp_max.default(clamp_min_41, 127);  clamp_min_41 = None
	        _assert_tensor_metadata_121 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_27, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_121 = None
	        convert_element_type_79 = torch.ops.prims.convert_element_type.default(clamp_max_27, torch.int8);  clamp_max_27 = None
	        view_210 = torch.ops.aten.view.default(clamp_min_39, [sym_size_int, 1500, 1]);  clamp_min_39 = None
	        view_211 = torch.ops.aten.view.default(convert_element_type_78, [sym_size_int, 1500, 1]);  convert_element_type_78 = None
	        _assert_tensor_metadata_122 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_79, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_122 = None
	        convert_element_type_80 = torch.ops.prims.convert_element_type.default(convert_element_type_79, torch.float32);  convert_element_type_79 = None
	        _assert_tensor_metadata_123 = torch.ops.aten._assert_tensor_metadata.default(view_211, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_123 = None
	        convert_element_type_81 = torch.ops.prims.convert_element_type.default(view_211, torch.float32);  view_211 = None
	        sub_637 = torch.ops.aten.sub.Tensor(convert_element_type_80, convert_element_type_81);  convert_element_type_80 = convert_element_type_81 = None
	        mul_1350 = torch.ops.aten.mul.Tensor(sub_637, view_210);  sub_637 = view_210 = None
	        _assert_tensor_metadata_124 = torch.ops.aten._assert_tensor_metadata.default(mul_1350, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_124 = None
	        view_213 = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_214 = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_215 = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_125 = torch.ops.aten._assert_tensor_metadata.default(view_213, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_125 = None
	        convert_element_type_82 = torch.ops.prims.convert_element_type.default(view_213, torch.float32);  view_213 = None
	        _assert_tensor_metadata_126 = torch.ops.aten._assert_tensor_metadata.default(view_215, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_126 = None
	        convert_element_type_83 = torch.ops.prims.convert_element_type.default(view_215, torch.float32);  view_215 = None
	        sub_641 = torch.ops.aten.sub.Tensor(convert_element_type_82, convert_element_type_83);  convert_element_type_82 = convert_element_type_83 = None
	        mul_1355 = torch.ops.aten.mul.Tensor(sub_641, view_214);  sub_641 = view_214 = None
	        view_216 = torch.ops.aten.view.default(mul_1355, [1280, 1280]);  mul_1355 = None
	        _assert_tensor_metadata_127 = torch.ops.aten._assert_tensor_metadata.default(view_216, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_127 = None
	        permute_23 = torch.ops.aten.permute.default(view_216, [1, 0]);  view_216 = None
	        mul_1358 = sym_size_int * 1500
	        view_217 = torch.ops.aten.view.default(mul_1350, [mul_1358, 1280]);  mul_1350 = mul_1358 = None
	        mm_2 = torch.ops.aten.mm.default(view_217, permute_23);  view_217 = permute_23 = None
	        view_218 = torch.ops.aten.view.default(mm_2, [sym_size_int, 1500, 1280]);  mm_2 = None
	        view_219 = torch.ops.aten.view.default(view_218, [sym_size_int, -1, 20, 64]);  view_218 = None
	        permute_24 = torch.ops.aten.permute.default(view_219, [0, 2, 1, 3]);  view_219 = None
	        clone_19 = torch.ops.aten.clone.default(permute_24, memory_format = torch.contiguous_format);  permute_24 = None
	        amin_14 = torch.ops.aten.amin.default(add_1868, [2])
	        amax_14 = torch.ops.aten.amax.default(add_1868, [2])
	        full_28 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_14 = torch.ops.aten.minimum.default(amin_14, full_28);  amin_14 = full_28 = None
	        full_29 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_14 = torch.ops.aten.maximum.default(amax_14, full_29);  amax_14 = full_29 = None
	        sub_655 = torch.ops.aten.sub.Tensor(maximum_14, minimum_14);  maximum_14 = None
	        div_28 = torch.ops.aten.div.Tensor(sub_655, 255.0);  sub_655 = None
	        clamp_min_42 = torch.ops.aten.clamp_min.default(div_28, 1.1920928955078125e-07);  div_28 = None
	        div_29 = torch.ops.aten.div.Tensor(minimum_14, clamp_min_42);  minimum_14 = None
	        round_29 = torch.ops.aten.round.default(div_29);  div_29 = None
	        sub_661 = torch.ops.aten.sub.Tensor(-128, round_29);  round_29 = None
	        clamp_min_43 = torch.ops.aten.clamp_min.default(sub_661, -128);  sub_661 = None
	        clamp_max_28 = torch.ops.aten.clamp_max.default(clamp_min_43, 127);  clamp_min_43 = None
	        _assert_tensor_metadata_128 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_42, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_128 = None
	        _assert_tensor_metadata_129 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_28, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_129 = None
	        convert_element_type_84 = torch.ops.prims.convert_element_type.default(clamp_max_28, torch.int8);  clamp_max_28 = None
	        view_222 = torch.ops.aten.view.default(clamp_min_42, [sym_size_int, 1500, 1])
	        view_223 = torch.ops.aten.view.default(convert_element_type_84, [sym_size_int, 1500, 1])
	        reciprocal_14 = torch.ops.aten.reciprocal.default(view_222);  view_222 = None
	        mul_1424 = torch.ops.aten.mul.Tensor(reciprocal_14, 1.0);  reciprocal_14 = None
	        mul_1427 = torch.ops.aten.mul.Tensor(add_1868, mul_1424);  add_1868 = mul_1424 = None
	        round_30 = torch.ops.aten.round.default(mul_1427);  mul_1427 = None
	        add_2255 = torch.ops.aten.add.Tensor(round_30, view_223);  round_30 = view_223 = None
	        clamp_min_44 = torch.ops.aten.clamp_min.default(add_2255, -128);  add_2255 = None
	        clamp_max_29 = torch.ops.aten.clamp_max.default(clamp_min_44, 127);  clamp_min_44 = None
	        _assert_tensor_metadata_130 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_29, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_130 = None
	        convert_element_type_85 = torch.ops.prims.convert_element_type.default(clamp_max_29, torch.int8);  clamp_max_29 = None
	        view_226 = torch.ops.aten.view.default(clamp_min_42, [sym_size_int, 1500, 1]);  clamp_min_42 = None
	        view_227 = torch.ops.aten.view.default(convert_element_type_84, [sym_size_int, 1500, 1]);  convert_element_type_84 = None
	        _assert_tensor_metadata_131 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_85, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_131 = None
	        convert_element_type_86 = torch.ops.prims.convert_element_type.default(convert_element_type_85, torch.float32);  convert_element_type_85 = None
	        _assert_tensor_metadata_132 = torch.ops.aten._assert_tensor_metadata.default(view_227, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_132 = None
	        convert_element_type_87 = torch.ops.prims.convert_element_type.default(view_227, torch.float32);  view_227 = None
	        sub_681 = torch.ops.aten.sub.Tensor(convert_element_type_86, convert_element_type_87);  convert_element_type_86 = convert_element_type_87 = None
	        mul_1449 = torch.ops.aten.mul.Tensor(sub_681, view_226);  sub_681 = view_226 = None
	        _assert_tensor_metadata_133 = torch.ops.aten._assert_tensor_metadata.default(mul_1449, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_133 = None
	        view_229 = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_230 = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_231 = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_134 = torch.ops.aten._assert_tensor_metadata.default(view_229, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_134 = None
	        convert_element_type_88 = torch.ops.prims.convert_element_type.default(view_229, torch.float32);  view_229 = None
	        _assert_tensor_metadata_135 = torch.ops.aten._assert_tensor_metadata.default(view_231, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_135 = None
	        convert_element_type_89 = torch.ops.prims.convert_element_type.default(view_231, torch.float32);  view_231 = None
	        sub_685 = torch.ops.aten.sub.Tensor(convert_element_type_88, convert_element_type_89);  convert_element_type_88 = convert_element_type_89 = None
	        mul_1454 = torch.ops.aten.mul.Tensor(sub_685, view_230);  sub_685 = view_230 = None
	        view_232 = torch.ops.aten.view.default(mul_1454, [1280, 1280]);  mul_1454 = None
	        _assert_tensor_metadata_136 = torch.ops.aten._assert_tensor_metadata.default(view_232, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_136 = None
	        mul_1459 = sym_size_int * 1500
	        view_233 = torch.ops.aten.view.default(mul_1449, [mul_1459, 1280]);  mul_1449 = mul_1459 = None
	        permute_25 = torch.ops.aten.permute.default(view_232, [1, 0]);  view_232 = None
	        addmm_11 = torch.ops.aten.addmm.default(model_audio_tower_layers_2_self_attn_v_proj_bias, view_233, permute_25);  model_audio_tower_layers_2_self_attn_v_proj_bias = view_233 = permute_25 = None
	        view_234 = torch.ops.aten.view.default(addmm_11, [sym_size_int, 1500, 1280]);  addmm_11 = None
	        view_235 = torch.ops.aten.view.default(view_234, [sym_size_int, -1, 20, 64]);  view_234 = None
	        permute_26 = torch.ops.aten.permute.default(view_235, [0, 2, 1, 3]);  view_235 = None
	        clone_20 = torch.ops.aten.clone.default(permute_26, memory_format = torch.contiguous_format);  permute_26 = None
	        _scaled_dot_product_efficient_attention_2 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_18, clone_19, clone_20, None, False, scale = 1.0);  clone_18 = clone_19 = clone_20 = None
	        getitem_18 = _scaled_dot_product_efficient_attention_2[0];  _scaled_dot_product_efficient_attention_2 = None
	        permute_27 = torch.ops.aten.permute.default(getitem_18, [0, 2, 1, 3]);  getitem_18 = None
	        view_236 = torch.ops.aten.view.default(permute_27, [sym_size_int, 1500, -1]);  permute_27 = None
	        amin_15 = torch.ops.aten.amin.default(view_236, [2])
	        amax_15 = torch.ops.aten.amax.default(view_236, [2])
	        full_30 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_15 = torch.ops.aten.minimum.default(amin_15, full_30);  amin_15 = full_30 = None
	        full_31 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_15 = torch.ops.aten.maximum.default(amax_15, full_31);  amax_15 = full_31 = None
	        sub_703 = torch.ops.aten.sub.Tensor(maximum_15, minimum_15);  maximum_15 = None
	        div_30 = torch.ops.aten.div.Tensor(sub_703, 255.0);  sub_703 = None
	        clamp_min_45 = torch.ops.aten.clamp_min.default(div_30, 1.1920928955078125e-07);  div_30 = None
	        div_31 = torch.ops.aten.div.Tensor(minimum_15, clamp_min_45);  minimum_15 = None
	        round_31 = torch.ops.aten.round.default(div_31);  div_31 = None
	        sub_709 = torch.ops.aten.sub.Tensor(-128, round_31);  round_31 = None
	        clamp_min_46 = torch.ops.aten.clamp_min.default(sub_709, -128);  sub_709 = None
	        clamp_max_30 = torch.ops.aten.clamp_max.default(clamp_min_46, 127);  clamp_min_46 = None
	        _assert_tensor_metadata_137 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_45, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_137 = None
	        _assert_tensor_metadata_138 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_30, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_138 = None
	        convert_element_type_90 = torch.ops.prims.convert_element_type.default(clamp_max_30, torch.int8);  clamp_max_30 = None
	        view_239 = torch.ops.aten.view.default(clamp_min_45, [sym_size_int, 1500, 1])
	        view_240 = torch.ops.aten.view.default(convert_element_type_90, [sym_size_int, 1500, 1])
	        reciprocal_15 = torch.ops.aten.reciprocal.default(view_239);  view_239 = None
	        mul_1529 = torch.ops.aten.mul.Tensor(reciprocal_15, 1.0);  reciprocal_15 = None
	        mul_1532 = torch.ops.aten.mul.Tensor(view_236, mul_1529);  view_236 = mul_1529 = None
	        round_32 = torch.ops.aten.round.default(mul_1532);  mul_1532 = None
	        add_2419 = torch.ops.aten.add.Tensor(round_32, view_240);  round_32 = view_240 = None
	        clamp_min_47 = torch.ops.aten.clamp_min.default(add_2419, -128);  add_2419 = None
	        clamp_max_31 = torch.ops.aten.clamp_max.default(clamp_min_47, 127);  clamp_min_47 = None
	        _assert_tensor_metadata_139 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_31, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_139 = None
	        convert_element_type_91 = torch.ops.prims.convert_element_type.default(clamp_max_31, torch.int8);  clamp_max_31 = None
	        view_243 = torch.ops.aten.view.default(clamp_min_45, [sym_size_int, 1500, 1]);  clamp_min_45 = None
	        view_244 = torch.ops.aten.view.default(convert_element_type_90, [sym_size_int, 1500, 1]);  convert_element_type_90 = None
	        _assert_tensor_metadata_140 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_91, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_140 = None
	        convert_element_type_92 = torch.ops.prims.convert_element_type.default(convert_element_type_91, torch.float32);  convert_element_type_91 = None
	        _assert_tensor_metadata_141 = torch.ops.aten._assert_tensor_metadata.default(view_244, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_141 = None
	        convert_element_type_93 = torch.ops.prims.convert_element_type.default(view_244, torch.float32);  view_244 = None
	        sub_729 = torch.ops.aten.sub.Tensor(convert_element_type_92, convert_element_type_93);  convert_element_type_92 = convert_element_type_93 = None
	        mul_1554 = torch.ops.aten.mul.Tensor(sub_729, view_243);  sub_729 = view_243 = None
	        _assert_tensor_metadata_142 = torch.ops.aten._assert_tensor_metadata.default(mul_1554, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_142 = None
	        view_246 = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_247 = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_248 = torch.ops.aten.view.default(model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_143 = torch.ops.aten._assert_tensor_metadata.default(view_246, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_143 = None
	        convert_element_type_94 = torch.ops.prims.convert_element_type.default(view_246, torch.float32);  view_246 = None
	        _assert_tensor_metadata_144 = torch.ops.aten._assert_tensor_metadata.default(view_248, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_144 = None
	        convert_element_type_95 = torch.ops.prims.convert_element_type.default(view_248, torch.float32);  view_248 = None
	        sub_733 = torch.ops.aten.sub.Tensor(convert_element_type_94, convert_element_type_95);  convert_element_type_94 = convert_element_type_95 = None
	        mul_1559 = torch.ops.aten.mul.Tensor(sub_733, view_247);  sub_733 = view_247 = None
	        view_249 = torch.ops.aten.view.default(mul_1559, [1280, 1280]);  mul_1559 = None
	        _assert_tensor_metadata_145 = torch.ops.aten._assert_tensor_metadata.default(view_249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_145 = None
	        mul_1564 = sym_size_int * 1500
	        view_250 = torch.ops.aten.view.default(mul_1554, [mul_1564, 1280]);  mul_1554 = mul_1564 = None
	        permute_28 = torch.ops.aten.permute.default(view_249, [1, 0]);  view_249 = None
	        addmm_12 = torch.ops.aten.addmm.default(model_audio_tower_layers_2_self_attn_out_proj_bias, view_250, permute_28);  model_audio_tower_layers_2_self_attn_out_proj_bias = view_250 = permute_28 = None
	        view_251 = torch.ops.aten.view.default(addmm_12, [sym_size_int, 1500, 1280]);  addmm_12 = None
	        add_2482 = torch.ops.aten.add.Tensor(add_1862, view_251);  add_1862 = view_251 = None
	        clone_22 = torch.ops.aten.clone.default(add_2482, memory_format = torch.contiguous_format)
	        var_mean_5 = torch.ops.aten.var_mean.correction(clone_22, [2], correction = 0, keepdim = True)
	        getitem_22 = var_mean_5[0]
	        getitem_23 = var_mean_5[1];  var_mean_5 = None
	        add_2487 = torch.ops.aten.add.Tensor(getitem_22, 1e-05);  getitem_22 = None
	        rsqrt_5 = torch.ops.aten.rsqrt.default(add_2487);  add_2487 = None
	        sub_739 = torch.ops.aten.sub.Tensor(clone_22, getitem_23);  clone_22 = getitem_23 = None
	        mul_1575 = torch.ops.aten.mul.Tensor(sub_739, rsqrt_5);  sub_739 = rsqrt_5 = None
	        mul_1576 = torch.ops.aten.mul.Tensor(mul_1575, model_audio_tower_layers_2_final_layer_norm_weight);  mul_1575 = model_audio_tower_layers_2_final_layer_norm_weight = None
	        add_2488 = torch.ops.aten.add.Tensor(mul_1576, model_audio_tower_layers_2_final_layer_norm_bias);  mul_1576 = model_audio_tower_layers_2_final_layer_norm_bias = None
	        amin_16 = torch.ops.aten.amin.default(add_2488, [2])
	        amax_16 = torch.ops.aten.amax.default(add_2488, [2])
	        full_32 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_16 = torch.ops.aten.minimum.default(amin_16, full_32);  amin_16 = full_32 = None
	        full_33 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_16 = torch.ops.aten.maximum.default(amax_16, full_33);  amax_16 = full_33 = None
	        sub_750 = torch.ops.aten.sub.Tensor(maximum_16, minimum_16);  maximum_16 = None
	        div_32 = torch.ops.aten.div.Tensor(sub_750, 255.0);  sub_750 = None
	        clamp_min_48 = torch.ops.aten.clamp_min.default(div_32, 1.1920928955078125e-07);  div_32 = None
	        div_33 = torch.ops.aten.div.Tensor(minimum_16, clamp_min_48);  minimum_16 = None
	        round_33 = torch.ops.aten.round.default(div_33);  div_33 = None
	        sub_756 = torch.ops.aten.sub.Tensor(-128, round_33);  round_33 = None
	        clamp_min_49 = torch.ops.aten.clamp_min.default(sub_756, -128);  sub_756 = None
	        clamp_max_32 = torch.ops.aten.clamp_max.default(clamp_min_49, 127);  clamp_min_49 = None
	        _assert_tensor_metadata_146 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_48, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_146 = None
	        _assert_tensor_metadata_147 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_32, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_147 = None
	        convert_element_type_96 = torch.ops.prims.convert_element_type.default(clamp_max_32, torch.int8);  clamp_max_32 = None
	        view_254 = torch.ops.aten.view.default(clamp_min_48, [sym_size_int, 1500, 1])
	        view_255 = torch.ops.aten.view.default(convert_element_type_96, [sym_size_int, 1500, 1])
	        reciprocal_16 = torch.ops.aten.reciprocal.default(view_254);  view_254 = None
	        mul_1624 = torch.ops.aten.mul.Tensor(reciprocal_16, 1.0);  reciprocal_16 = None
	        mul_1627 = torch.ops.aten.mul.Tensor(add_2488, mul_1624);  add_2488 = mul_1624 = None
	        round_34 = torch.ops.aten.round.default(mul_1627);  mul_1627 = None
	        add_2575 = torch.ops.aten.add.Tensor(round_34, view_255);  round_34 = view_255 = None
	        clamp_min_50 = torch.ops.aten.clamp_min.default(add_2575, -128);  add_2575 = None
	        clamp_max_33 = torch.ops.aten.clamp_max.default(clamp_min_50, 127);  clamp_min_50 = None
	        _assert_tensor_metadata_148 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_33, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_148 = None
	        convert_element_type_97 = torch.ops.prims.convert_element_type.default(clamp_max_33, torch.int8);  clamp_max_33 = None
	        view_258 = torch.ops.aten.view.default(clamp_min_48, [sym_size_int, 1500, 1]);  clamp_min_48 = None
	        view_259 = torch.ops.aten.view.default(convert_element_type_96, [sym_size_int, 1500, 1]);  convert_element_type_96 = None
	        _assert_tensor_metadata_149 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_97, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_149 = None
	        convert_element_type_98 = torch.ops.prims.convert_element_type.default(convert_element_type_97, torch.float32);  convert_element_type_97 = None
	        _assert_tensor_metadata_150 = torch.ops.aten._assert_tensor_metadata.default(view_259, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_150 = None
	        convert_element_type_99 = torch.ops.prims.convert_element_type.default(view_259, torch.float32);  view_259 = None
	        sub_776 = torch.ops.aten.sub.Tensor(convert_element_type_98, convert_element_type_99);  convert_element_type_98 = convert_element_type_99 = None
	        mul_1649 = torch.ops.aten.mul.Tensor(sub_776, view_258);  sub_776 = view_258 = None
	        _assert_tensor_metadata_151 = torch.ops.aten._assert_tensor_metadata.default(mul_1649, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_151 = None
	        view_261 = torch.ops.aten.view.default(model_audio_tower_layers_2_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_2_fc1_parametrizations_weight_original0 = None
	        view_262 = torch.ops.aten.view.default(model_audio_tower_layers_2_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_2_fc1_parametrizations_weight_original1 = None
	        view_263 = torch.ops.aten.view.default(model_audio_tower_layers_2_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_2_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_152 = torch.ops.aten._assert_tensor_metadata.default(view_261, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_152 = None
	        convert_element_type_100 = torch.ops.prims.convert_element_type.default(view_261, torch.float32);  view_261 = None
	        _assert_tensor_metadata_153 = torch.ops.aten._assert_tensor_metadata.default(view_263, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_153 = None
	        convert_element_type_101 = torch.ops.prims.convert_element_type.default(view_263, torch.float32);  view_263 = None
	        sub_780 = torch.ops.aten.sub.Tensor(convert_element_type_100, convert_element_type_101);  convert_element_type_100 = convert_element_type_101 = None
	        mul_1654 = torch.ops.aten.mul.Tensor(sub_780, view_262);  sub_780 = view_262 = None
	        view_264 = torch.ops.aten.view.default(mul_1654, [5120, 1280]);  mul_1654 = None
	        _assert_tensor_metadata_154 = torch.ops.aten._assert_tensor_metadata.default(view_264, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_154 = None
	        mul_1659 = sym_size_int * 1500
	        view_265 = torch.ops.aten.view.default(mul_1649, [mul_1659, 1280]);  mul_1649 = mul_1659 = None
	        permute_29 = torch.ops.aten.permute.default(view_264, [1, 0]);  view_264 = None
	        addmm_13 = torch.ops.aten.addmm.default(model_audio_tower_layers_2_fc1_bias, view_265, permute_29);  model_audio_tower_layers_2_fc1_bias = view_265 = permute_29 = None
	        view_266 = torch.ops.aten.view.default(addmm_13, [sym_size_int, 1500, 5120]);  addmm_13 = None
	        mul_1666 = torch.ops.aten.mul.Tensor(view_266, 0.5)
	        mul_1667 = torch.ops.aten.mul.Tensor(view_266, 0.7071067811865476);  view_266 = None
	        erf_4 = torch.ops.aten.erf.default(mul_1667);  mul_1667 = None
	        add_2634 = torch.ops.aten.add.Tensor(erf_4, 1);  erf_4 = None
	        mul_1668 = torch.ops.aten.mul.Tensor(mul_1666, add_2634);  mul_1666 = add_2634 = None
	        amin_17 = torch.ops.aten.amin.default(mul_1668, [2])
	        amax_17 = torch.ops.aten.amax.default(mul_1668, [2])
	        full_34 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_17 = torch.ops.aten.minimum.default(amin_17, full_34);  amin_17 = full_34 = None
	        full_35 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_17 = torch.ops.aten.maximum.default(amax_17, full_35);  amax_17 = full_35 = None
	        sub_793 = torch.ops.aten.sub.Tensor(maximum_17, minimum_17);  maximum_17 = None
	        div_34 = torch.ops.aten.div.Tensor(sub_793, 255.0);  sub_793 = None
	        clamp_min_51 = torch.ops.aten.clamp_min.default(div_34, 1.1920928955078125e-07);  div_34 = None
	        div_35 = torch.ops.aten.div.Tensor(minimum_17, clamp_min_51);  minimum_17 = None
	        round_35 = torch.ops.aten.round.default(div_35);  div_35 = None
	        sub_799 = torch.ops.aten.sub.Tensor(-128, round_35);  round_35 = None
	        clamp_min_52 = torch.ops.aten.clamp_min.default(sub_799, -128);  sub_799 = None
	        clamp_max_34 = torch.ops.aten.clamp_max.default(clamp_min_52, 127);  clamp_min_52 = None
	        _assert_tensor_metadata_155 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_51, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_155 = None
	        _assert_tensor_metadata_156 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_34, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_156 = None
	        convert_element_type_102 = torch.ops.prims.convert_element_type.default(clamp_max_34, torch.int8);  clamp_max_34 = None
	        view_269 = torch.ops.aten.view.default(clamp_min_51, [sym_size_int, 1500, 1])
	        view_270 = torch.ops.aten.view.default(convert_element_type_102, [sym_size_int, 1500, 1])
	        reciprocal_17 = torch.ops.aten.reciprocal.default(view_269);  view_269 = None
	        mul_1714 = torch.ops.aten.mul.Tensor(reciprocal_17, 1.0);  reciprocal_17 = None
	        mul_1717 = torch.ops.aten.mul.Tensor(mul_1668, mul_1714);  mul_1668 = mul_1714 = None
	        round_36 = torch.ops.aten.round.default(mul_1717);  mul_1717 = None
	        add_2717 = torch.ops.aten.add.Tensor(round_36, view_270);  round_36 = view_270 = None
	        clamp_min_53 = torch.ops.aten.clamp_min.default(add_2717, -128);  add_2717 = None
	        clamp_max_35 = torch.ops.aten.clamp_max.default(clamp_min_53, 127);  clamp_min_53 = None
	        _assert_tensor_metadata_157 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_35, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_157 = None
	        convert_element_type_103 = torch.ops.prims.convert_element_type.default(clamp_max_35, torch.int8);  clamp_max_35 = None
	        view_273 = torch.ops.aten.view.default(clamp_min_51, [sym_size_int, 1500, 1]);  clamp_min_51 = None
	        view_274 = torch.ops.aten.view.default(convert_element_type_102, [sym_size_int, 1500, 1]);  convert_element_type_102 = None
	        _assert_tensor_metadata_158 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_103, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_158 = None
	        convert_element_type_104 = torch.ops.prims.convert_element_type.default(convert_element_type_103, torch.float32);  convert_element_type_103 = None
	        _assert_tensor_metadata_159 = torch.ops.aten._assert_tensor_metadata.default(view_274, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_159 = None
	        convert_element_type_105 = torch.ops.prims.convert_element_type.default(view_274, torch.float32);  view_274 = None
	        sub_819 = torch.ops.aten.sub.Tensor(convert_element_type_104, convert_element_type_105);  convert_element_type_104 = convert_element_type_105 = None
	        mul_1739 = torch.ops.aten.mul.Tensor(sub_819, view_273);  sub_819 = view_273 = None
	        _assert_tensor_metadata_160 = torch.ops.aten._assert_tensor_metadata.default(mul_1739, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_160 = None
	        view_276 = torch.ops.aten.view.default(model_audio_tower_layers_2_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_2_fc2_parametrizations_weight_original0 = None
	        view_277 = torch.ops.aten.view.default(model_audio_tower_layers_2_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_2_fc2_parametrizations_weight_original1 = None
	        view_278 = torch.ops.aten.view.default(model_audio_tower_layers_2_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_2_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_161 = torch.ops.aten._assert_tensor_metadata.default(view_276, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_161 = None
	        convert_element_type_106 = torch.ops.prims.convert_element_type.default(view_276, torch.float32);  view_276 = None
	        _assert_tensor_metadata_162 = torch.ops.aten._assert_tensor_metadata.default(view_278, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_162 = None
	        convert_element_type_107 = torch.ops.prims.convert_element_type.default(view_278, torch.float32);  view_278 = None
	        sub_823 = torch.ops.aten.sub.Tensor(convert_element_type_106, convert_element_type_107);  convert_element_type_106 = convert_element_type_107 = None
	        mul_1744 = torch.ops.aten.mul.Tensor(sub_823, view_277);  sub_823 = view_277 = None
	        view_279 = torch.ops.aten.view.default(mul_1744, [1280, 5120]);  mul_1744 = None
	        _assert_tensor_metadata_163 = torch.ops.aten._assert_tensor_metadata.default(view_279, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_163 = None
	        mul_1749 = sym_size_int * 1500
	        view_280 = torch.ops.aten.view.default(mul_1739, [mul_1749, 5120]);  mul_1739 = mul_1749 = None
	        permute_30 = torch.ops.aten.permute.default(view_279, [1, 0]);  view_279 = None
	        addmm_14 = torch.ops.aten.addmm.default(model_audio_tower_layers_2_fc2_bias, view_280, permute_30);  model_audio_tower_layers_2_fc2_bias = view_280 = permute_30 = None
	        view_281 = torch.ops.aten.view.default(addmm_14, [sym_size_int, 1500, 1280]);  addmm_14 = None
	        add_2780 = torch.ops.aten.add.Tensor(add_2482, view_281);  add_2482 = view_281 = None
	        clone_25 = torch.ops.aten.clone.default(add_2780, memory_format = torch.contiguous_format)
	        var_mean_6 = torch.ops.aten.var_mean.correction(clone_25, [2], correction = 0, keepdim = True)
	        getitem_24 = var_mean_6[0]
	        getitem_25 = var_mean_6[1];  var_mean_6 = None
	        add_2785 = torch.ops.aten.add.Tensor(getitem_24, 1e-05);  getitem_24 = None
	        rsqrt_6 = torch.ops.aten.rsqrt.default(add_2785);  add_2785 = None
	        sub_829 = torch.ops.aten.sub.Tensor(clone_25, getitem_25);  clone_25 = getitem_25 = None
	        mul_1760 = torch.ops.aten.mul.Tensor(sub_829, rsqrt_6);  sub_829 = rsqrt_6 = None
	        mul_1761 = torch.ops.aten.mul.Tensor(mul_1760, model_audio_tower_layers_3_self_attn_layer_norm_weight);  mul_1760 = model_audio_tower_layers_3_self_attn_layer_norm_weight = None
	        add_2786 = torch.ops.aten.add.Tensor(mul_1761, model_audio_tower_layers_3_self_attn_layer_norm_bias);  mul_1761 = model_audio_tower_layers_3_self_attn_layer_norm_bias = None
	        amin_18 = torch.ops.aten.amin.default(add_2786, [2])
	        amax_18 = torch.ops.aten.amax.default(add_2786, [2])
	        full_36 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_18 = torch.ops.aten.minimum.default(amin_18, full_36);  amin_18 = full_36 = None
	        full_37 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_18 = torch.ops.aten.maximum.default(amax_18, full_37);  amax_18 = full_37 = None
	        sub_840 = torch.ops.aten.sub.Tensor(maximum_18, minimum_18);  maximum_18 = None
	        div_36 = torch.ops.aten.div.Tensor(sub_840, 255.0);  sub_840 = None
	        clamp_min_54 = torch.ops.aten.clamp_min.default(div_36, 1.1920928955078125e-07);  div_36 = None
	        div_37 = torch.ops.aten.div.Tensor(minimum_18, clamp_min_54);  minimum_18 = None
	        round_37 = torch.ops.aten.round.default(div_37);  div_37 = None
	        sub_846 = torch.ops.aten.sub.Tensor(-128, round_37);  round_37 = None
	        clamp_min_55 = torch.ops.aten.clamp_min.default(sub_846, -128);  sub_846 = None
	        clamp_max_36 = torch.ops.aten.clamp_max.default(clamp_min_55, 127);  clamp_min_55 = None
	        _assert_tensor_metadata_164 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_54, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_164 = None
	        _assert_tensor_metadata_165 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_36, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_165 = None
	        convert_element_type_108 = torch.ops.prims.convert_element_type.default(clamp_max_36, torch.int8);  clamp_max_36 = None
	        view_284 = torch.ops.aten.view.default(clamp_min_54, [sym_size_int, 1500, 1])
	        view_285 = torch.ops.aten.view.default(convert_element_type_108, [sym_size_int, 1500, 1])
	        reciprocal_18 = torch.ops.aten.reciprocal.default(view_284);  view_284 = None
	        mul_1809 = torch.ops.aten.mul.Tensor(reciprocal_18, 1.0);  reciprocal_18 = None
	        mul_1812 = torch.ops.aten.mul.Tensor(add_2786, mul_1809);  mul_1809 = None
	        round_38 = torch.ops.aten.round.default(mul_1812);  mul_1812 = None
	        add_2873 = torch.ops.aten.add.Tensor(round_38, view_285);  round_38 = view_285 = None
	        clamp_min_56 = torch.ops.aten.clamp_min.default(add_2873, -128);  add_2873 = None
	        clamp_max_37 = torch.ops.aten.clamp_max.default(clamp_min_56, 127);  clamp_min_56 = None
	        _assert_tensor_metadata_166 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_37, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_166 = None
	        convert_element_type_109 = torch.ops.prims.convert_element_type.default(clamp_max_37, torch.int8);  clamp_max_37 = None
	        view_288 = torch.ops.aten.view.default(clamp_min_54, [sym_size_int, 1500, 1]);  clamp_min_54 = None
	        view_289 = torch.ops.aten.view.default(convert_element_type_108, [sym_size_int, 1500, 1]);  convert_element_type_108 = None
	        _assert_tensor_metadata_167 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_109, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_167 = None
	        convert_element_type_110 = torch.ops.prims.convert_element_type.default(convert_element_type_109, torch.float32);  convert_element_type_109 = None
	        _assert_tensor_metadata_168 = torch.ops.aten._assert_tensor_metadata.default(view_289, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_168 = None
	        convert_element_type_111 = torch.ops.prims.convert_element_type.default(view_289, torch.float32);  view_289 = None
	        sub_866 = torch.ops.aten.sub.Tensor(convert_element_type_110, convert_element_type_111);  convert_element_type_110 = convert_element_type_111 = None
	        mul_1834 = torch.ops.aten.mul.Tensor(sub_866, view_288);  sub_866 = view_288 = None
	        _assert_tensor_metadata_169 = torch.ops.aten._assert_tensor_metadata.default(mul_1834, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_169 = None
	        view_291 = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_292 = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_293 = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_170 = torch.ops.aten._assert_tensor_metadata.default(view_291, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_170 = None
	        convert_element_type_112 = torch.ops.prims.convert_element_type.default(view_291, torch.float32);  view_291 = None
	        _assert_tensor_metadata_171 = torch.ops.aten._assert_tensor_metadata.default(view_293, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_171 = None
	        convert_element_type_113 = torch.ops.prims.convert_element_type.default(view_293, torch.float32);  view_293 = None
	        sub_870 = torch.ops.aten.sub.Tensor(convert_element_type_112, convert_element_type_113);  convert_element_type_112 = convert_element_type_113 = None
	        mul_1839 = torch.ops.aten.mul.Tensor(sub_870, view_292);  sub_870 = view_292 = None
	        view_294 = torch.ops.aten.view.default(mul_1839, [1280, 1280]);  mul_1839 = None
	        _assert_tensor_metadata_172 = torch.ops.aten._assert_tensor_metadata.default(view_294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_172 = None
	        mul_1844 = sym_size_int * 1500
	        view_295 = torch.ops.aten.view.default(mul_1834, [mul_1844, 1280]);  mul_1834 = mul_1844 = None
	        permute_31 = torch.ops.aten.permute.default(view_294, [1, 0]);  view_294 = None
	        addmm_15 = torch.ops.aten.addmm.default(model_audio_tower_layers_3_self_attn_q_proj_bias, view_295, permute_31);  model_audio_tower_layers_3_self_attn_q_proj_bias = view_295 = permute_31 = None
	        view_296 = torch.ops.aten.view.default(addmm_15, [sym_size_int, 1500, 1280]);  addmm_15 = None
	        mul_1851 = torch.ops.aten.mul.Tensor(view_296, 0.125);  view_296 = None
	        view_297 = torch.ops.aten.view.default(mul_1851, [sym_size_int, 1500, 20, 64]);  mul_1851 = None
	        permute_32 = torch.ops.aten.permute.default(view_297, [0, 2, 1, 3]);  view_297 = None
	        clone_26 = torch.ops.aten.clone.default(permute_32, memory_format = torch.contiguous_format);  permute_32 = None
	        amin_19 = torch.ops.aten.amin.default(add_2786, [2])
	        amax_19 = torch.ops.aten.amax.default(add_2786, [2])
	        full_38 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_19 = torch.ops.aten.minimum.default(amin_19, full_38);  amin_19 = full_38 = None
	        full_39 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_19 = torch.ops.aten.maximum.default(amax_19, full_39);  amax_19 = full_39 = None
	        sub_885 = torch.ops.aten.sub.Tensor(maximum_19, minimum_19);  maximum_19 = None
	        div_38 = torch.ops.aten.div.Tensor(sub_885, 255.0);  sub_885 = None
	        clamp_min_57 = torch.ops.aten.clamp_min.default(div_38, 1.1920928955078125e-07);  div_38 = None
	        div_39 = torch.ops.aten.div.Tensor(minimum_19, clamp_min_57);  minimum_19 = None
	        round_39 = torch.ops.aten.round.default(div_39);  div_39 = None
	        sub_891 = torch.ops.aten.sub.Tensor(-128, round_39);  round_39 = None
	        clamp_min_58 = torch.ops.aten.clamp_min.default(sub_891, -128);  sub_891 = None
	        clamp_max_38 = torch.ops.aten.clamp_max.default(clamp_min_58, 127);  clamp_min_58 = None
	        _assert_tensor_metadata_173 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_57, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_173 = None
	        _assert_tensor_metadata_174 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_38, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_174 = None
	        convert_element_type_114 = torch.ops.prims.convert_element_type.default(clamp_max_38, torch.int8);  clamp_max_38 = None
	        view_300 = torch.ops.aten.view.default(clamp_min_57, [sym_size_int, 1500, 1])
	        view_301 = torch.ops.aten.view.default(convert_element_type_114, [sym_size_int, 1500, 1])
	        reciprocal_19 = torch.ops.aten.reciprocal.default(view_300);  view_300 = None
	        mul_1905 = torch.ops.aten.mul.Tensor(reciprocal_19, 1.0);  reciprocal_19 = None
	        mul_1908 = torch.ops.aten.mul.Tensor(add_2786, mul_1905);  mul_1905 = None
	        round_40 = torch.ops.aten.round.default(mul_1908);  mul_1908 = None
	        add_3025 = torch.ops.aten.add.Tensor(round_40, view_301);  round_40 = view_301 = None
	        clamp_min_59 = torch.ops.aten.clamp_min.default(add_3025, -128);  add_3025 = None
	        clamp_max_39 = torch.ops.aten.clamp_max.default(clamp_min_59, 127);  clamp_min_59 = None
	        _assert_tensor_metadata_175 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_39, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_175 = None
	        convert_element_type_115 = torch.ops.prims.convert_element_type.default(clamp_max_39, torch.int8);  clamp_max_39 = None
	        view_304 = torch.ops.aten.view.default(clamp_min_57, [sym_size_int, 1500, 1]);  clamp_min_57 = None
	        view_305 = torch.ops.aten.view.default(convert_element_type_114, [sym_size_int, 1500, 1]);  convert_element_type_114 = None
	        _assert_tensor_metadata_176 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_115, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_176 = None
	        convert_element_type_116 = torch.ops.prims.convert_element_type.default(convert_element_type_115, torch.float32);  convert_element_type_115 = None
	        _assert_tensor_metadata_177 = torch.ops.aten._assert_tensor_metadata.default(view_305, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_177 = None
	        convert_element_type_117 = torch.ops.prims.convert_element_type.default(view_305, torch.float32);  view_305 = None
	        sub_911 = torch.ops.aten.sub.Tensor(convert_element_type_116, convert_element_type_117);  convert_element_type_116 = convert_element_type_117 = None
	        mul_1930 = torch.ops.aten.mul.Tensor(sub_911, view_304);  sub_911 = view_304 = None
	        _assert_tensor_metadata_178 = torch.ops.aten._assert_tensor_metadata.default(mul_1930, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_178 = None
	        view_307 = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_308 = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_309 = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_179 = torch.ops.aten._assert_tensor_metadata.default(view_307, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_179 = None
	        convert_element_type_118 = torch.ops.prims.convert_element_type.default(view_307, torch.float32);  view_307 = None
	        _assert_tensor_metadata_180 = torch.ops.aten._assert_tensor_metadata.default(view_309, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_180 = None
	        convert_element_type_119 = torch.ops.prims.convert_element_type.default(view_309, torch.float32);  view_309 = None
	        sub_915 = torch.ops.aten.sub.Tensor(convert_element_type_118, convert_element_type_119);  convert_element_type_118 = convert_element_type_119 = None
	        mul_1935 = torch.ops.aten.mul.Tensor(sub_915, view_308);  sub_915 = view_308 = None
	        view_310 = torch.ops.aten.view.default(mul_1935, [1280, 1280]);  mul_1935 = None
	        _assert_tensor_metadata_181 = torch.ops.aten._assert_tensor_metadata.default(view_310, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_181 = None
	        permute_33 = torch.ops.aten.permute.default(view_310, [1, 0]);  view_310 = None
	        mul_1938 = sym_size_int * 1500
	        view_311 = torch.ops.aten.view.default(mul_1930, [mul_1938, 1280]);  mul_1930 = mul_1938 = None
	        mm_3 = torch.ops.aten.mm.default(view_311, permute_33);  view_311 = permute_33 = None
	        view_312 = torch.ops.aten.view.default(mm_3, [sym_size_int, 1500, 1280]);  mm_3 = None
	        view_313 = torch.ops.aten.view.default(view_312, [sym_size_int, -1, 20, 64]);  view_312 = None
	        permute_34 = torch.ops.aten.permute.default(view_313, [0, 2, 1, 3]);  view_313 = None
	        clone_27 = torch.ops.aten.clone.default(permute_34, memory_format = torch.contiguous_format);  permute_34 = None
	        amin_20 = torch.ops.aten.amin.default(add_2786, [2])
	        amax_20 = torch.ops.aten.amax.default(add_2786, [2])
	        full_40 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_20 = torch.ops.aten.minimum.default(amin_20, full_40);  amin_20 = full_40 = None
	        full_41 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_20 = torch.ops.aten.maximum.default(amax_20, full_41);  amax_20 = full_41 = None
	        sub_929 = torch.ops.aten.sub.Tensor(maximum_20, minimum_20);  maximum_20 = None
	        div_40 = torch.ops.aten.div.Tensor(sub_929, 255.0);  sub_929 = None
	        clamp_min_60 = torch.ops.aten.clamp_min.default(div_40, 1.1920928955078125e-07);  div_40 = None
	        div_41 = torch.ops.aten.div.Tensor(minimum_20, clamp_min_60);  minimum_20 = None
	        round_41 = torch.ops.aten.round.default(div_41);  div_41 = None
	        sub_935 = torch.ops.aten.sub.Tensor(-128, round_41);  round_41 = None
	        clamp_min_61 = torch.ops.aten.clamp_min.default(sub_935, -128);  sub_935 = None
	        clamp_max_40 = torch.ops.aten.clamp_max.default(clamp_min_61, 127);  clamp_min_61 = None
	        _assert_tensor_metadata_182 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_60, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_182 = None
	        _assert_tensor_metadata_183 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_40, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_183 = None
	        convert_element_type_120 = torch.ops.prims.convert_element_type.default(clamp_max_40, torch.int8);  clamp_max_40 = None
	        view_316 = torch.ops.aten.view.default(clamp_min_60, [sym_size_int, 1500, 1])
	        view_317 = torch.ops.aten.view.default(convert_element_type_120, [sym_size_int, 1500, 1])
	        reciprocal_20 = torch.ops.aten.reciprocal.default(view_316);  view_316 = None
	        mul_2004 = torch.ops.aten.mul.Tensor(reciprocal_20, 1.0);  reciprocal_20 = None
	        mul_2007 = torch.ops.aten.mul.Tensor(add_2786, mul_2004);  add_2786 = mul_2004 = None
	        round_42 = torch.ops.aten.round.default(mul_2007);  mul_2007 = None
	        add_3173 = torch.ops.aten.add.Tensor(round_42, view_317);  round_42 = view_317 = None
	        clamp_min_62 = torch.ops.aten.clamp_min.default(add_3173, -128);  add_3173 = None
	        clamp_max_41 = torch.ops.aten.clamp_max.default(clamp_min_62, 127);  clamp_min_62 = None
	        _assert_tensor_metadata_184 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_41, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_184 = None
	        convert_element_type_121 = torch.ops.prims.convert_element_type.default(clamp_max_41, torch.int8);  clamp_max_41 = None
	        view_320 = torch.ops.aten.view.default(clamp_min_60, [sym_size_int, 1500, 1]);  clamp_min_60 = None
	        view_321 = torch.ops.aten.view.default(convert_element_type_120, [sym_size_int, 1500, 1]);  convert_element_type_120 = None
	        _assert_tensor_metadata_185 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_121, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_185 = None
	        convert_element_type_122 = torch.ops.prims.convert_element_type.default(convert_element_type_121, torch.float32);  convert_element_type_121 = None
	        _assert_tensor_metadata_186 = torch.ops.aten._assert_tensor_metadata.default(view_321, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_186 = None
	        convert_element_type_123 = torch.ops.prims.convert_element_type.default(view_321, torch.float32);  view_321 = None
	        sub_955 = torch.ops.aten.sub.Tensor(convert_element_type_122, convert_element_type_123);  convert_element_type_122 = convert_element_type_123 = None
	        mul_2029 = torch.ops.aten.mul.Tensor(sub_955, view_320);  sub_955 = view_320 = None
	        _assert_tensor_metadata_187 = torch.ops.aten._assert_tensor_metadata.default(mul_2029, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_187 = None
	        view_323 = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_324 = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_325 = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_188 = torch.ops.aten._assert_tensor_metadata.default(view_323, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_188 = None
	        convert_element_type_124 = torch.ops.prims.convert_element_type.default(view_323, torch.float32);  view_323 = None
	        _assert_tensor_metadata_189 = torch.ops.aten._assert_tensor_metadata.default(view_325, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_189 = None
	        convert_element_type_125 = torch.ops.prims.convert_element_type.default(view_325, torch.float32);  view_325 = None
	        sub_959 = torch.ops.aten.sub.Tensor(convert_element_type_124, convert_element_type_125);  convert_element_type_124 = convert_element_type_125 = None
	        mul_2034 = torch.ops.aten.mul.Tensor(sub_959, view_324);  sub_959 = view_324 = None
	        view_326 = torch.ops.aten.view.default(mul_2034, [1280, 1280]);  mul_2034 = None
	        _assert_tensor_metadata_190 = torch.ops.aten._assert_tensor_metadata.default(view_326, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_190 = None
	        mul_2039 = sym_size_int * 1500
	        view_327 = torch.ops.aten.view.default(mul_2029, [mul_2039, 1280]);  mul_2029 = mul_2039 = None
	        permute_35 = torch.ops.aten.permute.default(view_326, [1, 0]);  view_326 = None
	        addmm_16 = torch.ops.aten.addmm.default(model_audio_tower_layers_3_self_attn_v_proj_bias, view_327, permute_35);  model_audio_tower_layers_3_self_attn_v_proj_bias = view_327 = permute_35 = None
	        view_328 = torch.ops.aten.view.default(addmm_16, [sym_size_int, 1500, 1280]);  addmm_16 = None
	        view_329 = torch.ops.aten.view.default(view_328, [sym_size_int, -1, 20, 64]);  view_328 = None
	        permute_36 = torch.ops.aten.permute.default(view_329, [0, 2, 1, 3]);  view_329 = None
	        clone_28 = torch.ops.aten.clone.default(permute_36, memory_format = torch.contiguous_format);  permute_36 = None
	        _scaled_dot_product_efficient_attention_3 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_26, clone_27, clone_28, None, False, scale = 1.0);  clone_26 = clone_27 = clone_28 = None
	        getitem_26 = _scaled_dot_product_efficient_attention_3[0];  _scaled_dot_product_efficient_attention_3 = None
	        permute_37 = torch.ops.aten.permute.default(getitem_26, [0, 2, 1, 3]);  getitem_26 = None
	        view_330 = torch.ops.aten.view.default(permute_37, [sym_size_int, 1500, -1]);  permute_37 = None
	        amin_21 = torch.ops.aten.amin.default(view_330, [2])
	        amax_21 = torch.ops.aten.amax.default(view_330, [2])
	        full_42 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_21 = torch.ops.aten.minimum.default(amin_21, full_42);  amin_21 = full_42 = None
	        full_43 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_21 = torch.ops.aten.maximum.default(amax_21, full_43);  amax_21 = full_43 = None
	        sub_977 = torch.ops.aten.sub.Tensor(maximum_21, minimum_21);  maximum_21 = None
	        div_42 = torch.ops.aten.div.Tensor(sub_977, 255.0);  sub_977 = None
	        clamp_min_63 = torch.ops.aten.clamp_min.default(div_42, 1.1920928955078125e-07);  div_42 = None
	        div_43 = torch.ops.aten.div.Tensor(minimum_21, clamp_min_63);  minimum_21 = None
	        round_43 = torch.ops.aten.round.default(div_43);  div_43 = None
	        sub_983 = torch.ops.aten.sub.Tensor(-128, round_43);  round_43 = None
	        clamp_min_64 = torch.ops.aten.clamp_min.default(sub_983, -128);  sub_983 = None
	        clamp_max_42 = torch.ops.aten.clamp_max.default(clamp_min_64, 127);  clamp_min_64 = None
	        _assert_tensor_metadata_191 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_63, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_191 = None
	        _assert_tensor_metadata_192 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_42, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_192 = None
	        convert_element_type_126 = torch.ops.prims.convert_element_type.default(clamp_max_42, torch.int8);  clamp_max_42 = None
	        view_333 = torch.ops.aten.view.default(clamp_min_63, [sym_size_int, 1500, 1])
	        view_334 = torch.ops.aten.view.default(convert_element_type_126, [sym_size_int, 1500, 1])
	        reciprocal_21 = torch.ops.aten.reciprocal.default(view_333);  view_333 = None
	        mul_2109 = torch.ops.aten.mul.Tensor(reciprocal_21, 1.0);  reciprocal_21 = None
	        mul_2112 = torch.ops.aten.mul.Tensor(view_330, mul_2109);  view_330 = mul_2109 = None
	        round_44 = torch.ops.aten.round.default(mul_2112);  mul_2112 = None
	        add_3337 = torch.ops.aten.add.Tensor(round_44, view_334);  round_44 = view_334 = None
	        clamp_min_65 = torch.ops.aten.clamp_min.default(add_3337, -128);  add_3337 = None
	        clamp_max_43 = torch.ops.aten.clamp_max.default(clamp_min_65, 127);  clamp_min_65 = None
	        _assert_tensor_metadata_193 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_43, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_193 = None
	        convert_element_type_127 = torch.ops.prims.convert_element_type.default(clamp_max_43, torch.int8);  clamp_max_43 = None
	        view_337 = torch.ops.aten.view.default(clamp_min_63, [sym_size_int, 1500, 1]);  clamp_min_63 = None
	        view_338 = torch.ops.aten.view.default(convert_element_type_126, [sym_size_int, 1500, 1]);  convert_element_type_126 = None
	        _assert_tensor_metadata_194 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_127, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_194 = None
	        convert_element_type_128 = torch.ops.prims.convert_element_type.default(convert_element_type_127, torch.float32);  convert_element_type_127 = None
	        _assert_tensor_metadata_195 = torch.ops.aten._assert_tensor_metadata.default(view_338, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_195 = None
	        convert_element_type_129 = torch.ops.prims.convert_element_type.default(view_338, torch.float32);  view_338 = None
	        sub_1003 = torch.ops.aten.sub.Tensor(convert_element_type_128, convert_element_type_129);  convert_element_type_128 = convert_element_type_129 = None
	        mul_2134 = torch.ops.aten.mul.Tensor(sub_1003, view_337);  sub_1003 = view_337 = None
	        _assert_tensor_metadata_196 = torch.ops.aten._assert_tensor_metadata.default(mul_2134, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_196 = None
	        view_340 = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_341 = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_342 = torch.ops.aten.view.default(model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_197 = torch.ops.aten._assert_tensor_metadata.default(view_340, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_197 = None
	        convert_element_type_130 = torch.ops.prims.convert_element_type.default(view_340, torch.float32);  view_340 = None
	        _assert_tensor_metadata_198 = torch.ops.aten._assert_tensor_metadata.default(view_342, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_198 = None
	        convert_element_type_131 = torch.ops.prims.convert_element_type.default(view_342, torch.float32);  view_342 = None
	        sub_1007 = torch.ops.aten.sub.Tensor(convert_element_type_130, convert_element_type_131);  convert_element_type_130 = convert_element_type_131 = None
	        mul_2139 = torch.ops.aten.mul.Tensor(sub_1007, view_341);  sub_1007 = view_341 = None
	        view_343 = torch.ops.aten.view.default(mul_2139, [1280, 1280]);  mul_2139 = None
	        _assert_tensor_metadata_199 = torch.ops.aten._assert_tensor_metadata.default(view_343, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_199 = None
	        mul_2144 = sym_size_int * 1500
	        view_344 = torch.ops.aten.view.default(mul_2134, [mul_2144, 1280]);  mul_2134 = mul_2144 = None
	        permute_38 = torch.ops.aten.permute.default(view_343, [1, 0]);  view_343 = None
	        addmm_17 = torch.ops.aten.addmm.default(model_audio_tower_layers_3_self_attn_out_proj_bias, view_344, permute_38);  model_audio_tower_layers_3_self_attn_out_proj_bias = view_344 = permute_38 = None
	        view_345 = torch.ops.aten.view.default(addmm_17, [sym_size_int, 1500, 1280]);  addmm_17 = None
	        add_3400 = torch.ops.aten.add.Tensor(add_2780, view_345);  add_2780 = view_345 = None
	        clone_30 = torch.ops.aten.clone.default(add_3400, memory_format = torch.contiguous_format)
	        var_mean_7 = torch.ops.aten.var_mean.correction(clone_30, [2], correction = 0, keepdim = True)
	        getitem_30 = var_mean_7[0]
	        getitem_31 = var_mean_7[1];  var_mean_7 = None
	        add_3405 = torch.ops.aten.add.Tensor(getitem_30, 1e-05);  getitem_30 = None
	        rsqrt_7 = torch.ops.aten.rsqrt.default(add_3405);  add_3405 = None
	        sub_1013 = torch.ops.aten.sub.Tensor(clone_30, getitem_31);  clone_30 = getitem_31 = None
	        mul_2155 = torch.ops.aten.mul.Tensor(sub_1013, rsqrt_7);  sub_1013 = rsqrt_7 = None
	        mul_2156 = torch.ops.aten.mul.Tensor(mul_2155, model_audio_tower_layers_3_final_layer_norm_weight);  mul_2155 = model_audio_tower_layers_3_final_layer_norm_weight = None
	        add_3406 = torch.ops.aten.add.Tensor(mul_2156, model_audio_tower_layers_3_final_layer_norm_bias);  mul_2156 = model_audio_tower_layers_3_final_layer_norm_bias = None
	        amin_22 = torch.ops.aten.amin.default(add_3406, [2])
	        amax_22 = torch.ops.aten.amax.default(add_3406, [2])
	        full_44 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_22 = torch.ops.aten.minimum.default(amin_22, full_44);  amin_22 = full_44 = None
	        full_45 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_22 = torch.ops.aten.maximum.default(amax_22, full_45);  amax_22 = full_45 = None
	        sub_1024 = torch.ops.aten.sub.Tensor(maximum_22, minimum_22);  maximum_22 = None
	        div_44 = torch.ops.aten.div.Tensor(sub_1024, 255.0);  sub_1024 = None
	        clamp_min_66 = torch.ops.aten.clamp_min.default(div_44, 1.1920928955078125e-07);  div_44 = None
	        div_45 = torch.ops.aten.div.Tensor(minimum_22, clamp_min_66);  minimum_22 = None
	        round_45 = torch.ops.aten.round.default(div_45);  div_45 = None
	        sub_1030 = torch.ops.aten.sub.Tensor(-128, round_45);  round_45 = None
	        clamp_min_67 = torch.ops.aten.clamp_min.default(sub_1030, -128);  sub_1030 = None
	        clamp_max_44 = torch.ops.aten.clamp_max.default(clamp_min_67, 127);  clamp_min_67 = None
	        _assert_tensor_metadata_200 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_66, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_200 = None
	        _assert_tensor_metadata_201 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_44, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_201 = None
	        convert_element_type_132 = torch.ops.prims.convert_element_type.default(clamp_max_44, torch.int8);  clamp_max_44 = None
	        view_348 = torch.ops.aten.view.default(clamp_min_66, [sym_size_int, 1500, 1])
	        view_349 = torch.ops.aten.view.default(convert_element_type_132, [sym_size_int, 1500, 1])
	        reciprocal_22 = torch.ops.aten.reciprocal.default(view_348);  view_348 = None
	        mul_2204 = torch.ops.aten.mul.Tensor(reciprocal_22, 1.0);  reciprocal_22 = None
	        mul_2207 = torch.ops.aten.mul.Tensor(add_3406, mul_2204);  add_3406 = mul_2204 = None
	        round_46 = torch.ops.aten.round.default(mul_2207);  mul_2207 = None
	        add_3493 = torch.ops.aten.add.Tensor(round_46, view_349);  round_46 = view_349 = None
	        clamp_min_68 = torch.ops.aten.clamp_min.default(add_3493, -128);  add_3493 = None
	        clamp_max_45 = torch.ops.aten.clamp_max.default(clamp_min_68, 127);  clamp_min_68 = None
	        _assert_tensor_metadata_202 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_45, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_202 = None
	        convert_element_type_133 = torch.ops.prims.convert_element_type.default(clamp_max_45, torch.int8);  clamp_max_45 = None
	        view_352 = torch.ops.aten.view.default(clamp_min_66, [sym_size_int, 1500, 1]);  clamp_min_66 = None
	        view_353 = torch.ops.aten.view.default(convert_element_type_132, [sym_size_int, 1500, 1]);  convert_element_type_132 = None
	        _assert_tensor_metadata_203 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_133, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_203 = None
	        convert_element_type_134 = torch.ops.prims.convert_element_type.default(convert_element_type_133, torch.float32);  convert_element_type_133 = None
	        _assert_tensor_metadata_204 = torch.ops.aten._assert_tensor_metadata.default(view_353, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_204 = None
	        convert_element_type_135 = torch.ops.prims.convert_element_type.default(view_353, torch.float32);  view_353 = None
	        sub_1050 = torch.ops.aten.sub.Tensor(convert_element_type_134, convert_element_type_135);  convert_element_type_134 = convert_element_type_135 = None
	        mul_2229 = torch.ops.aten.mul.Tensor(sub_1050, view_352);  sub_1050 = view_352 = None
	        _assert_tensor_metadata_205 = torch.ops.aten._assert_tensor_metadata.default(mul_2229, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_205 = None
	        view_355 = torch.ops.aten.view.default(model_audio_tower_layers_3_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_3_fc1_parametrizations_weight_original0 = None
	        view_356 = torch.ops.aten.view.default(model_audio_tower_layers_3_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_3_fc1_parametrizations_weight_original1 = None
	        view_357 = torch.ops.aten.view.default(model_audio_tower_layers_3_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_3_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_206 = torch.ops.aten._assert_tensor_metadata.default(view_355, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_206 = None
	        convert_element_type_136 = torch.ops.prims.convert_element_type.default(view_355, torch.float32);  view_355 = None
	        _assert_tensor_metadata_207 = torch.ops.aten._assert_tensor_metadata.default(view_357, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_207 = None
	        convert_element_type_137 = torch.ops.prims.convert_element_type.default(view_357, torch.float32);  view_357 = None
	        sub_1054 = torch.ops.aten.sub.Tensor(convert_element_type_136, convert_element_type_137);  convert_element_type_136 = convert_element_type_137 = None
	        mul_2234 = torch.ops.aten.mul.Tensor(sub_1054, view_356);  sub_1054 = view_356 = None
	        view_358 = torch.ops.aten.view.default(mul_2234, [5120, 1280]);  mul_2234 = None
	        _assert_tensor_metadata_208 = torch.ops.aten._assert_tensor_metadata.default(view_358, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_208 = None
	        mul_2239 = sym_size_int * 1500
	        view_359 = torch.ops.aten.view.default(mul_2229, [mul_2239, 1280]);  mul_2229 = mul_2239 = None
	        permute_39 = torch.ops.aten.permute.default(view_358, [1, 0]);  view_358 = None
	        addmm_18 = torch.ops.aten.addmm.default(model_audio_tower_layers_3_fc1_bias, view_359, permute_39);  model_audio_tower_layers_3_fc1_bias = view_359 = permute_39 = None
	        view_360 = torch.ops.aten.view.default(addmm_18, [sym_size_int, 1500, 5120]);  addmm_18 = None
	        mul_2246 = torch.ops.aten.mul.Tensor(view_360, 0.5)
	        mul_2247 = torch.ops.aten.mul.Tensor(view_360, 0.7071067811865476);  view_360 = None
	        erf_5 = torch.ops.aten.erf.default(mul_2247);  mul_2247 = None
	        add_3552 = torch.ops.aten.add.Tensor(erf_5, 1);  erf_5 = None
	        mul_2248 = torch.ops.aten.mul.Tensor(mul_2246, add_3552);  mul_2246 = add_3552 = None
	        amin_23 = torch.ops.aten.amin.default(mul_2248, [2])
	        amax_23 = torch.ops.aten.amax.default(mul_2248, [2])
	        full_46 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_23 = torch.ops.aten.minimum.default(amin_23, full_46);  amin_23 = full_46 = None
	        full_47 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_23 = torch.ops.aten.maximum.default(amax_23, full_47);  amax_23 = full_47 = None
	        sub_1067 = torch.ops.aten.sub.Tensor(maximum_23, minimum_23);  maximum_23 = None
	        div_46 = torch.ops.aten.div.Tensor(sub_1067, 255.0);  sub_1067 = None
	        clamp_min_69 = torch.ops.aten.clamp_min.default(div_46, 1.1920928955078125e-07);  div_46 = None
	        div_47 = torch.ops.aten.div.Tensor(minimum_23, clamp_min_69);  minimum_23 = None
	        round_47 = torch.ops.aten.round.default(div_47);  div_47 = None
	        sub_1073 = torch.ops.aten.sub.Tensor(-128, round_47);  round_47 = None
	        clamp_min_70 = torch.ops.aten.clamp_min.default(sub_1073, -128);  sub_1073 = None
	        clamp_max_46 = torch.ops.aten.clamp_max.default(clamp_min_70, 127);  clamp_min_70 = None
	        _assert_tensor_metadata_209 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_69, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_209 = None
	        _assert_tensor_metadata_210 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_46, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_210 = None
	        convert_element_type_138 = torch.ops.prims.convert_element_type.default(clamp_max_46, torch.int8);  clamp_max_46 = None
	        view_363 = torch.ops.aten.view.default(clamp_min_69, [sym_size_int, 1500, 1])
	        view_364 = torch.ops.aten.view.default(convert_element_type_138, [sym_size_int, 1500, 1])
	        reciprocal_23 = torch.ops.aten.reciprocal.default(view_363);  view_363 = None
	        mul_2294 = torch.ops.aten.mul.Tensor(reciprocal_23, 1.0);  reciprocal_23 = None
	        mul_2297 = torch.ops.aten.mul.Tensor(mul_2248, mul_2294);  mul_2248 = mul_2294 = None
	        round_48 = torch.ops.aten.round.default(mul_2297);  mul_2297 = None
	        add_3635 = torch.ops.aten.add.Tensor(round_48, view_364);  round_48 = view_364 = None
	        clamp_min_71 = torch.ops.aten.clamp_min.default(add_3635, -128);  add_3635 = None
	        clamp_max_47 = torch.ops.aten.clamp_max.default(clamp_min_71, 127);  clamp_min_71 = None
	        _assert_tensor_metadata_211 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_47, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_211 = None
	        convert_element_type_139 = torch.ops.prims.convert_element_type.default(clamp_max_47, torch.int8);  clamp_max_47 = None
	        view_367 = torch.ops.aten.view.default(clamp_min_69, [sym_size_int, 1500, 1]);  clamp_min_69 = None
	        view_368 = torch.ops.aten.view.default(convert_element_type_138, [sym_size_int, 1500, 1]);  convert_element_type_138 = None
	        _assert_tensor_metadata_212 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_139, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_212 = None
	        convert_element_type_140 = torch.ops.prims.convert_element_type.default(convert_element_type_139, torch.float32);  convert_element_type_139 = None
	        _assert_tensor_metadata_213 = torch.ops.aten._assert_tensor_metadata.default(view_368, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_213 = None
	        convert_element_type_141 = torch.ops.prims.convert_element_type.default(view_368, torch.float32);  view_368 = None
	        sub_1093 = torch.ops.aten.sub.Tensor(convert_element_type_140, convert_element_type_141);  convert_element_type_140 = convert_element_type_141 = None
	        mul_2319 = torch.ops.aten.mul.Tensor(sub_1093, view_367);  sub_1093 = view_367 = None
	        _assert_tensor_metadata_214 = torch.ops.aten._assert_tensor_metadata.default(mul_2319, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_214 = None
	        view_370 = torch.ops.aten.view.default(model_audio_tower_layers_3_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_3_fc2_parametrizations_weight_original0 = None
	        view_371 = torch.ops.aten.view.default(model_audio_tower_layers_3_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_3_fc2_parametrizations_weight_original1 = None
	        view_372 = torch.ops.aten.view.default(model_audio_tower_layers_3_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_3_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_215 = torch.ops.aten._assert_tensor_metadata.default(view_370, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_215 = None
	        convert_element_type_142 = torch.ops.prims.convert_element_type.default(view_370, torch.float32);  view_370 = None
	        _assert_tensor_metadata_216 = torch.ops.aten._assert_tensor_metadata.default(view_372, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_216 = None
	        convert_element_type_143 = torch.ops.prims.convert_element_type.default(view_372, torch.float32);  view_372 = None
	        sub_1097 = torch.ops.aten.sub.Tensor(convert_element_type_142, convert_element_type_143);  convert_element_type_142 = convert_element_type_143 = None
	        mul_2324 = torch.ops.aten.mul.Tensor(sub_1097, view_371);  sub_1097 = view_371 = None
	        view_373 = torch.ops.aten.view.default(mul_2324, [1280, 5120]);  mul_2324 = None
	        _assert_tensor_metadata_217 = torch.ops.aten._assert_tensor_metadata.default(view_373, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_217 = None
	        mul_2329 = sym_size_int * 1500
	        view_374 = torch.ops.aten.view.default(mul_2319, [mul_2329, 5120]);  mul_2319 = mul_2329 = None
	        permute_40 = torch.ops.aten.permute.default(view_373, [1, 0]);  view_373 = None
	        addmm_19 = torch.ops.aten.addmm.default(model_audio_tower_layers_3_fc2_bias, view_374, permute_40);  model_audio_tower_layers_3_fc2_bias = view_374 = permute_40 = None
	        view_375 = torch.ops.aten.view.default(addmm_19, [sym_size_int, 1500, 1280]);  addmm_19 = None
	        add_3698 = torch.ops.aten.add.Tensor(add_3400, view_375);  add_3400 = view_375 = None
	        clone_33 = torch.ops.aten.clone.default(add_3698, memory_format = torch.contiguous_format)
	        var_mean_8 = torch.ops.aten.var_mean.correction(clone_33, [2], correction = 0, keepdim = True)
	        getitem_32 = var_mean_8[0]
	        getitem_33 = var_mean_8[1];  var_mean_8 = None
	        add_3703 = torch.ops.aten.add.Tensor(getitem_32, 1e-05);  getitem_32 = None
	        rsqrt_8 = torch.ops.aten.rsqrt.default(add_3703);  add_3703 = None
	        sub_1103 = torch.ops.aten.sub.Tensor(clone_33, getitem_33);  clone_33 = getitem_33 = None
	        mul_2340 = torch.ops.aten.mul.Tensor(sub_1103, rsqrt_8);  sub_1103 = rsqrt_8 = None
	        mul_2341 = torch.ops.aten.mul.Tensor(mul_2340, model_audio_tower_layers_4_self_attn_layer_norm_weight);  mul_2340 = model_audio_tower_layers_4_self_attn_layer_norm_weight = None
	        add_3704 = torch.ops.aten.add.Tensor(mul_2341, model_audio_tower_layers_4_self_attn_layer_norm_bias);  mul_2341 = model_audio_tower_layers_4_self_attn_layer_norm_bias = None
	        amin_24 = torch.ops.aten.amin.default(add_3704, [2])
	        amax_24 = torch.ops.aten.amax.default(add_3704, [2])
	        full_48 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_24 = torch.ops.aten.minimum.default(amin_24, full_48);  amin_24 = full_48 = None
	        full_49 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_24 = torch.ops.aten.maximum.default(amax_24, full_49);  amax_24 = full_49 = None
	        sub_1114 = torch.ops.aten.sub.Tensor(maximum_24, minimum_24);  maximum_24 = None
	        div_48 = torch.ops.aten.div.Tensor(sub_1114, 255.0);  sub_1114 = None
	        clamp_min_72 = torch.ops.aten.clamp_min.default(div_48, 1.1920928955078125e-07);  div_48 = None
	        div_49 = torch.ops.aten.div.Tensor(minimum_24, clamp_min_72);  minimum_24 = None
	        round_49 = torch.ops.aten.round.default(div_49);  div_49 = None
	        sub_1120 = torch.ops.aten.sub.Tensor(-128, round_49);  round_49 = None
	        clamp_min_73 = torch.ops.aten.clamp_min.default(sub_1120, -128);  sub_1120 = None
	        clamp_max_48 = torch.ops.aten.clamp_max.default(clamp_min_73, 127);  clamp_min_73 = None
	        _assert_tensor_metadata_218 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_72, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_218 = None
	        _assert_tensor_metadata_219 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_48, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_219 = None
	        convert_element_type_144 = torch.ops.prims.convert_element_type.default(clamp_max_48, torch.int8);  clamp_max_48 = None
	        view_378 = torch.ops.aten.view.default(clamp_min_72, [sym_size_int, 1500, 1])
	        view_379 = torch.ops.aten.view.default(convert_element_type_144, [sym_size_int, 1500, 1])
	        reciprocal_24 = torch.ops.aten.reciprocal.default(view_378);  view_378 = None
	        mul_2389 = torch.ops.aten.mul.Tensor(reciprocal_24, 1.0);  reciprocal_24 = None
	        mul_2392 = torch.ops.aten.mul.Tensor(add_3704, mul_2389);  mul_2389 = None
	        round_50 = torch.ops.aten.round.default(mul_2392);  mul_2392 = None
	        add_3791 = torch.ops.aten.add.Tensor(round_50, view_379);  round_50 = view_379 = None
	        clamp_min_74 = torch.ops.aten.clamp_min.default(add_3791, -128);  add_3791 = None
	        clamp_max_49 = torch.ops.aten.clamp_max.default(clamp_min_74, 127);  clamp_min_74 = None
	        _assert_tensor_metadata_220 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_49, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_220 = None
	        convert_element_type_145 = torch.ops.prims.convert_element_type.default(clamp_max_49, torch.int8);  clamp_max_49 = None
	        view_382 = torch.ops.aten.view.default(clamp_min_72, [sym_size_int, 1500, 1]);  clamp_min_72 = None
	        view_383 = torch.ops.aten.view.default(convert_element_type_144, [sym_size_int, 1500, 1]);  convert_element_type_144 = None
	        _assert_tensor_metadata_221 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_145, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_221 = None
	        convert_element_type_146 = torch.ops.prims.convert_element_type.default(convert_element_type_145, torch.float32);  convert_element_type_145 = None
	        _assert_tensor_metadata_222 = torch.ops.aten._assert_tensor_metadata.default(view_383, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_222 = None
	        convert_element_type_147 = torch.ops.prims.convert_element_type.default(view_383, torch.float32);  view_383 = None
	        sub_1140 = torch.ops.aten.sub.Tensor(convert_element_type_146, convert_element_type_147);  convert_element_type_146 = convert_element_type_147 = None
	        mul_2414 = torch.ops.aten.mul.Tensor(sub_1140, view_382);  sub_1140 = view_382 = None
	        _assert_tensor_metadata_223 = torch.ops.aten._assert_tensor_metadata.default(mul_2414, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_223 = None
	        view_385 = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_386 = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_387 = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_224 = torch.ops.aten._assert_tensor_metadata.default(view_385, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_224 = None
	        convert_element_type_148 = torch.ops.prims.convert_element_type.default(view_385, torch.float32);  view_385 = None
	        _assert_tensor_metadata_225 = torch.ops.aten._assert_tensor_metadata.default(view_387, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_225 = None
	        convert_element_type_149 = torch.ops.prims.convert_element_type.default(view_387, torch.float32);  view_387 = None
	        sub_1144 = torch.ops.aten.sub.Tensor(convert_element_type_148, convert_element_type_149);  convert_element_type_148 = convert_element_type_149 = None
	        mul_2419 = torch.ops.aten.mul.Tensor(sub_1144, view_386);  sub_1144 = view_386 = None
	        view_388 = torch.ops.aten.view.default(mul_2419, [1280, 1280]);  mul_2419 = None
	        _assert_tensor_metadata_226 = torch.ops.aten._assert_tensor_metadata.default(view_388, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_226 = None
	        mul_2424 = sym_size_int * 1500
	        view_389 = torch.ops.aten.view.default(mul_2414, [mul_2424, 1280]);  mul_2414 = mul_2424 = None
	        permute_41 = torch.ops.aten.permute.default(view_388, [1, 0]);  view_388 = None
	        addmm_20 = torch.ops.aten.addmm.default(model_audio_tower_layers_4_self_attn_q_proj_bias, view_389, permute_41);  model_audio_tower_layers_4_self_attn_q_proj_bias = view_389 = permute_41 = None
	        view_390 = torch.ops.aten.view.default(addmm_20, [sym_size_int, 1500, 1280]);  addmm_20 = None
	        mul_2431 = torch.ops.aten.mul.Tensor(view_390, 0.125);  view_390 = None
	        view_391 = torch.ops.aten.view.default(mul_2431, [sym_size_int, 1500, 20, 64]);  mul_2431 = None
	        permute_42 = torch.ops.aten.permute.default(view_391, [0, 2, 1, 3]);  view_391 = None
	        clone_34 = torch.ops.aten.clone.default(permute_42, memory_format = torch.contiguous_format);  permute_42 = None
	        amin_25 = torch.ops.aten.amin.default(add_3704, [2])
	        amax_25 = torch.ops.aten.amax.default(add_3704, [2])
	        full_50 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_25 = torch.ops.aten.minimum.default(amin_25, full_50);  amin_25 = full_50 = None
	        full_51 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_25 = torch.ops.aten.maximum.default(amax_25, full_51);  amax_25 = full_51 = None
	        sub_1159 = torch.ops.aten.sub.Tensor(maximum_25, minimum_25);  maximum_25 = None
	        div_50 = torch.ops.aten.div.Tensor(sub_1159, 255.0);  sub_1159 = None
	        clamp_min_75 = torch.ops.aten.clamp_min.default(div_50, 1.1920928955078125e-07);  div_50 = None
	        div_51 = torch.ops.aten.div.Tensor(minimum_25, clamp_min_75);  minimum_25 = None
	        round_51 = torch.ops.aten.round.default(div_51);  div_51 = None
	        sub_1165 = torch.ops.aten.sub.Tensor(-128, round_51);  round_51 = None
	        clamp_min_76 = torch.ops.aten.clamp_min.default(sub_1165, -128);  sub_1165 = None
	        clamp_max_50 = torch.ops.aten.clamp_max.default(clamp_min_76, 127);  clamp_min_76 = None
	        _assert_tensor_metadata_227 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_75, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_227 = None
	        _assert_tensor_metadata_228 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_50, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_228 = None
	        convert_element_type_150 = torch.ops.prims.convert_element_type.default(clamp_max_50, torch.int8);  clamp_max_50 = None
	        view_394 = torch.ops.aten.view.default(clamp_min_75, [sym_size_int, 1500, 1])
	        view_395 = torch.ops.aten.view.default(convert_element_type_150, [sym_size_int, 1500, 1])
	        reciprocal_25 = torch.ops.aten.reciprocal.default(view_394);  view_394 = None
	        mul_2485 = torch.ops.aten.mul.Tensor(reciprocal_25, 1.0);  reciprocal_25 = None
	        mul_2488 = torch.ops.aten.mul.Tensor(add_3704, mul_2485);  mul_2485 = None
	        round_52 = torch.ops.aten.round.default(mul_2488);  mul_2488 = None
	        add_3943 = torch.ops.aten.add.Tensor(round_52, view_395);  round_52 = view_395 = None
	        clamp_min_77 = torch.ops.aten.clamp_min.default(add_3943, -128);  add_3943 = None
	        clamp_max_51 = torch.ops.aten.clamp_max.default(clamp_min_77, 127);  clamp_min_77 = None
	        _assert_tensor_metadata_229 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_51, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_229 = None
	        convert_element_type_151 = torch.ops.prims.convert_element_type.default(clamp_max_51, torch.int8);  clamp_max_51 = None
	        view_398 = torch.ops.aten.view.default(clamp_min_75, [sym_size_int, 1500, 1]);  clamp_min_75 = None
	        view_399 = torch.ops.aten.view.default(convert_element_type_150, [sym_size_int, 1500, 1]);  convert_element_type_150 = None
	        _assert_tensor_metadata_230 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_151, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_230 = None
	        convert_element_type_152 = torch.ops.prims.convert_element_type.default(convert_element_type_151, torch.float32);  convert_element_type_151 = None
	        _assert_tensor_metadata_231 = torch.ops.aten._assert_tensor_metadata.default(view_399, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_231 = None
	        convert_element_type_153 = torch.ops.prims.convert_element_type.default(view_399, torch.float32);  view_399 = None
	        sub_1185 = torch.ops.aten.sub.Tensor(convert_element_type_152, convert_element_type_153);  convert_element_type_152 = convert_element_type_153 = None
	        mul_2510 = torch.ops.aten.mul.Tensor(sub_1185, view_398);  sub_1185 = view_398 = None
	        _assert_tensor_metadata_232 = torch.ops.aten._assert_tensor_metadata.default(mul_2510, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_232 = None
	        view_401 = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_402 = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_403 = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_233 = torch.ops.aten._assert_tensor_metadata.default(view_401, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_233 = None
	        convert_element_type_154 = torch.ops.prims.convert_element_type.default(view_401, torch.float32);  view_401 = None
	        _assert_tensor_metadata_234 = torch.ops.aten._assert_tensor_metadata.default(view_403, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_234 = None
	        convert_element_type_155 = torch.ops.prims.convert_element_type.default(view_403, torch.float32);  view_403 = None
	        sub_1189 = torch.ops.aten.sub.Tensor(convert_element_type_154, convert_element_type_155);  convert_element_type_154 = convert_element_type_155 = None
	        mul_2515 = torch.ops.aten.mul.Tensor(sub_1189, view_402);  sub_1189 = view_402 = None
	        view_404 = torch.ops.aten.view.default(mul_2515, [1280, 1280]);  mul_2515 = None
	        _assert_tensor_metadata_235 = torch.ops.aten._assert_tensor_metadata.default(view_404, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_235 = None
	        permute_43 = torch.ops.aten.permute.default(view_404, [1, 0]);  view_404 = None
	        mul_2518 = sym_size_int * 1500
	        view_405 = torch.ops.aten.view.default(mul_2510, [mul_2518, 1280]);  mul_2510 = mul_2518 = None
	        mm_4 = torch.ops.aten.mm.default(view_405, permute_43);  view_405 = permute_43 = None
	        view_406 = torch.ops.aten.view.default(mm_4, [sym_size_int, 1500, 1280]);  mm_4 = None
	        view_407 = torch.ops.aten.view.default(view_406, [sym_size_int, -1, 20, 64]);  view_406 = None
	        permute_44 = torch.ops.aten.permute.default(view_407, [0, 2, 1, 3]);  view_407 = None
	        clone_35 = torch.ops.aten.clone.default(permute_44, memory_format = torch.contiguous_format);  permute_44 = None
	        amin_26 = torch.ops.aten.amin.default(add_3704, [2])
	        amax_26 = torch.ops.aten.amax.default(add_3704, [2])
	        full_52 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_26 = torch.ops.aten.minimum.default(amin_26, full_52);  amin_26 = full_52 = None
	        full_53 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_26 = torch.ops.aten.maximum.default(amax_26, full_53);  amax_26 = full_53 = None
	        sub_1203 = torch.ops.aten.sub.Tensor(maximum_26, minimum_26);  maximum_26 = None
	        div_52 = torch.ops.aten.div.Tensor(sub_1203, 255.0);  sub_1203 = None
	        clamp_min_78 = torch.ops.aten.clamp_min.default(div_52, 1.1920928955078125e-07);  div_52 = None
	        div_53 = torch.ops.aten.div.Tensor(minimum_26, clamp_min_78);  minimum_26 = None
	        round_53 = torch.ops.aten.round.default(div_53);  div_53 = None
	        sub_1209 = torch.ops.aten.sub.Tensor(-128, round_53);  round_53 = None
	        clamp_min_79 = torch.ops.aten.clamp_min.default(sub_1209, -128);  sub_1209 = None
	        clamp_max_52 = torch.ops.aten.clamp_max.default(clamp_min_79, 127);  clamp_min_79 = None
	        _assert_tensor_metadata_236 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_78, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_236 = None
	        _assert_tensor_metadata_237 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_52, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_237 = None
	        convert_element_type_156 = torch.ops.prims.convert_element_type.default(clamp_max_52, torch.int8);  clamp_max_52 = None
	        view_410 = torch.ops.aten.view.default(clamp_min_78, [sym_size_int, 1500, 1])
	        view_411 = torch.ops.aten.view.default(convert_element_type_156, [sym_size_int, 1500, 1])
	        reciprocal_26 = torch.ops.aten.reciprocal.default(view_410);  view_410 = None
	        mul_2584 = torch.ops.aten.mul.Tensor(reciprocal_26, 1.0);  reciprocal_26 = None
	        mul_2587 = torch.ops.aten.mul.Tensor(add_3704, mul_2584);  add_3704 = mul_2584 = None
	        round_54 = torch.ops.aten.round.default(mul_2587);  mul_2587 = None
	        add_4091 = torch.ops.aten.add.Tensor(round_54, view_411);  round_54 = view_411 = None
	        clamp_min_80 = torch.ops.aten.clamp_min.default(add_4091, -128);  add_4091 = None
	        clamp_max_53 = torch.ops.aten.clamp_max.default(clamp_min_80, 127);  clamp_min_80 = None
	        _assert_tensor_metadata_238 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_53, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_238 = None
	        convert_element_type_157 = torch.ops.prims.convert_element_type.default(clamp_max_53, torch.int8);  clamp_max_53 = None
	        view_414 = torch.ops.aten.view.default(clamp_min_78, [sym_size_int, 1500, 1]);  clamp_min_78 = None
	        view_415 = torch.ops.aten.view.default(convert_element_type_156, [sym_size_int, 1500, 1]);  convert_element_type_156 = None
	        _assert_tensor_metadata_239 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_157, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_239 = None
	        convert_element_type_158 = torch.ops.prims.convert_element_type.default(convert_element_type_157, torch.float32);  convert_element_type_157 = None
	        _assert_tensor_metadata_240 = torch.ops.aten._assert_tensor_metadata.default(view_415, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_240 = None
	        convert_element_type_159 = torch.ops.prims.convert_element_type.default(view_415, torch.float32);  view_415 = None
	        sub_1229 = torch.ops.aten.sub.Tensor(convert_element_type_158, convert_element_type_159);  convert_element_type_158 = convert_element_type_159 = None
	        mul_2609 = torch.ops.aten.mul.Tensor(sub_1229, view_414);  sub_1229 = view_414 = None
	        _assert_tensor_metadata_241 = torch.ops.aten._assert_tensor_metadata.default(mul_2609, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_241 = None
	        view_417 = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_418 = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_419 = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_242 = torch.ops.aten._assert_tensor_metadata.default(view_417, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_242 = None
	        convert_element_type_160 = torch.ops.prims.convert_element_type.default(view_417, torch.float32);  view_417 = None
	        _assert_tensor_metadata_243 = torch.ops.aten._assert_tensor_metadata.default(view_419, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_243 = None
	        convert_element_type_161 = torch.ops.prims.convert_element_type.default(view_419, torch.float32);  view_419 = None
	        sub_1233 = torch.ops.aten.sub.Tensor(convert_element_type_160, convert_element_type_161);  convert_element_type_160 = convert_element_type_161 = None
	        mul_2614 = torch.ops.aten.mul.Tensor(sub_1233, view_418);  sub_1233 = view_418 = None
	        view_420 = torch.ops.aten.view.default(mul_2614, [1280, 1280]);  mul_2614 = None
	        _assert_tensor_metadata_244 = torch.ops.aten._assert_tensor_metadata.default(view_420, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_244 = None
	        mul_2619 = sym_size_int * 1500
	        view_421 = torch.ops.aten.view.default(mul_2609, [mul_2619, 1280]);  mul_2609 = mul_2619 = None
	        permute_45 = torch.ops.aten.permute.default(view_420, [1, 0]);  view_420 = None
	        addmm_21 = torch.ops.aten.addmm.default(model_audio_tower_layers_4_self_attn_v_proj_bias, view_421, permute_45);  model_audio_tower_layers_4_self_attn_v_proj_bias = view_421 = permute_45 = None
	        view_422 = torch.ops.aten.view.default(addmm_21, [sym_size_int, 1500, 1280]);  addmm_21 = None
	        view_423 = torch.ops.aten.view.default(view_422, [sym_size_int, -1, 20, 64]);  view_422 = None
	        permute_46 = torch.ops.aten.permute.default(view_423, [0, 2, 1, 3]);  view_423 = None
	        clone_36 = torch.ops.aten.clone.default(permute_46, memory_format = torch.contiguous_format);  permute_46 = None
	        _scaled_dot_product_efficient_attention_4 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_34, clone_35, clone_36, None, False, scale = 1.0);  clone_34 = clone_35 = clone_36 = None
	        getitem_34 = _scaled_dot_product_efficient_attention_4[0];  _scaled_dot_product_efficient_attention_4 = None
	        permute_47 = torch.ops.aten.permute.default(getitem_34, [0, 2, 1, 3]);  getitem_34 = None
	        view_424 = torch.ops.aten.view.default(permute_47, [sym_size_int, 1500, -1]);  permute_47 = None
	        amin_27 = torch.ops.aten.amin.default(view_424, [2])
	        amax_27 = torch.ops.aten.amax.default(view_424, [2])
	        full_54 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_27 = torch.ops.aten.minimum.default(amin_27, full_54);  amin_27 = full_54 = None
	        full_55 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_27 = torch.ops.aten.maximum.default(amax_27, full_55);  amax_27 = full_55 = None
	        sub_1251 = torch.ops.aten.sub.Tensor(maximum_27, minimum_27);  maximum_27 = None
	        div_54 = torch.ops.aten.div.Tensor(sub_1251, 255.0);  sub_1251 = None
	        clamp_min_81 = torch.ops.aten.clamp_min.default(div_54, 1.1920928955078125e-07);  div_54 = None
	        div_55 = torch.ops.aten.div.Tensor(minimum_27, clamp_min_81);  minimum_27 = None
	        round_55 = torch.ops.aten.round.default(div_55);  div_55 = None
	        sub_1257 = torch.ops.aten.sub.Tensor(-128, round_55);  round_55 = None
	        clamp_min_82 = torch.ops.aten.clamp_min.default(sub_1257, -128);  sub_1257 = None
	        clamp_max_54 = torch.ops.aten.clamp_max.default(clamp_min_82, 127);  clamp_min_82 = None
	        _assert_tensor_metadata_245 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_81, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_245 = None
	        _assert_tensor_metadata_246 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_54, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_246 = None
	        convert_element_type_162 = torch.ops.prims.convert_element_type.default(clamp_max_54, torch.int8);  clamp_max_54 = None
	        view_427 = torch.ops.aten.view.default(clamp_min_81, [sym_size_int, 1500, 1])
	        view_428 = torch.ops.aten.view.default(convert_element_type_162, [sym_size_int, 1500, 1])
	        reciprocal_27 = torch.ops.aten.reciprocal.default(view_427);  view_427 = None
	        mul_2689 = torch.ops.aten.mul.Tensor(reciprocal_27, 1.0);  reciprocal_27 = None
	        mul_2692 = torch.ops.aten.mul.Tensor(view_424, mul_2689);  view_424 = mul_2689 = None
	        round_56 = torch.ops.aten.round.default(mul_2692);  mul_2692 = None
	        add_4255 = torch.ops.aten.add.Tensor(round_56, view_428);  round_56 = view_428 = None
	        clamp_min_83 = torch.ops.aten.clamp_min.default(add_4255, -128);  add_4255 = None
	        clamp_max_55 = torch.ops.aten.clamp_max.default(clamp_min_83, 127);  clamp_min_83 = None
	        _assert_tensor_metadata_247 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_55, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_247 = None
	        convert_element_type_163 = torch.ops.prims.convert_element_type.default(clamp_max_55, torch.int8);  clamp_max_55 = None
	        view_431 = torch.ops.aten.view.default(clamp_min_81, [sym_size_int, 1500, 1]);  clamp_min_81 = None
	        view_432 = torch.ops.aten.view.default(convert_element_type_162, [sym_size_int, 1500, 1]);  convert_element_type_162 = None
	        _assert_tensor_metadata_248 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_163, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_248 = None
	        convert_element_type_164 = torch.ops.prims.convert_element_type.default(convert_element_type_163, torch.float32);  convert_element_type_163 = None
	        _assert_tensor_metadata_249 = torch.ops.aten._assert_tensor_metadata.default(view_432, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_249 = None
	        convert_element_type_165 = torch.ops.prims.convert_element_type.default(view_432, torch.float32);  view_432 = None
	        sub_1277 = torch.ops.aten.sub.Tensor(convert_element_type_164, convert_element_type_165);  convert_element_type_164 = convert_element_type_165 = None
	        mul_2714 = torch.ops.aten.mul.Tensor(sub_1277, view_431);  sub_1277 = view_431 = None
	        _assert_tensor_metadata_250 = torch.ops.aten._assert_tensor_metadata.default(mul_2714, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_250 = None
	        view_434 = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_435 = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_436 = torch.ops.aten.view.default(model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_251 = torch.ops.aten._assert_tensor_metadata.default(view_434, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_251 = None
	        convert_element_type_166 = torch.ops.prims.convert_element_type.default(view_434, torch.float32);  view_434 = None
	        _assert_tensor_metadata_252 = torch.ops.aten._assert_tensor_metadata.default(view_436, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_252 = None
	        convert_element_type_167 = torch.ops.prims.convert_element_type.default(view_436, torch.float32);  view_436 = None
	        sub_1281 = torch.ops.aten.sub.Tensor(convert_element_type_166, convert_element_type_167);  convert_element_type_166 = convert_element_type_167 = None
	        mul_2719 = torch.ops.aten.mul.Tensor(sub_1281, view_435);  sub_1281 = view_435 = None
	        view_437 = torch.ops.aten.view.default(mul_2719, [1280, 1280]);  mul_2719 = None
	        _assert_tensor_metadata_253 = torch.ops.aten._assert_tensor_metadata.default(view_437, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_253 = None
	        mul_2724 = sym_size_int * 1500
	        view_438 = torch.ops.aten.view.default(mul_2714, [mul_2724, 1280]);  mul_2714 = mul_2724 = None
	        permute_48 = torch.ops.aten.permute.default(view_437, [1, 0]);  view_437 = None
	        addmm_22 = torch.ops.aten.addmm.default(model_audio_tower_layers_4_self_attn_out_proj_bias, view_438, permute_48);  model_audio_tower_layers_4_self_attn_out_proj_bias = view_438 = permute_48 = None
	        view_439 = torch.ops.aten.view.default(addmm_22, [sym_size_int, 1500, 1280]);  addmm_22 = None
	        add_4318 = torch.ops.aten.add.Tensor(add_3698, view_439);  add_3698 = view_439 = None
	        clone_38 = torch.ops.aten.clone.default(add_4318, memory_format = torch.contiguous_format)
	        var_mean_9 = torch.ops.aten.var_mean.correction(clone_38, [2], correction = 0, keepdim = True)
	        getitem_38 = var_mean_9[0]
	        getitem_39 = var_mean_9[1];  var_mean_9 = None
	        add_4323 = torch.ops.aten.add.Tensor(getitem_38, 1e-05);  getitem_38 = None
	        rsqrt_9 = torch.ops.aten.rsqrt.default(add_4323);  add_4323 = None
	        sub_1287 = torch.ops.aten.sub.Tensor(clone_38, getitem_39);  clone_38 = getitem_39 = None
	        mul_2735 = torch.ops.aten.mul.Tensor(sub_1287, rsqrt_9);  sub_1287 = rsqrt_9 = None
	        mul_2736 = torch.ops.aten.mul.Tensor(mul_2735, model_audio_tower_layers_4_final_layer_norm_weight);  mul_2735 = model_audio_tower_layers_4_final_layer_norm_weight = None
	        add_4324 = torch.ops.aten.add.Tensor(mul_2736, model_audio_tower_layers_4_final_layer_norm_bias);  mul_2736 = model_audio_tower_layers_4_final_layer_norm_bias = None
	        amin_28 = torch.ops.aten.amin.default(add_4324, [2])
	        amax_28 = torch.ops.aten.amax.default(add_4324, [2])
	        full_56 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_28 = torch.ops.aten.minimum.default(amin_28, full_56);  amin_28 = full_56 = None
	        full_57 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_28 = torch.ops.aten.maximum.default(amax_28, full_57);  amax_28 = full_57 = None
	        sub_1298 = torch.ops.aten.sub.Tensor(maximum_28, minimum_28);  maximum_28 = None
	        div_56 = torch.ops.aten.div.Tensor(sub_1298, 255.0);  sub_1298 = None
	        clamp_min_84 = torch.ops.aten.clamp_min.default(div_56, 1.1920928955078125e-07);  div_56 = None
	        div_57 = torch.ops.aten.div.Tensor(minimum_28, clamp_min_84);  minimum_28 = None
	        round_57 = torch.ops.aten.round.default(div_57);  div_57 = None
	        sub_1304 = torch.ops.aten.sub.Tensor(-128, round_57);  round_57 = None
	        clamp_min_85 = torch.ops.aten.clamp_min.default(sub_1304, -128);  sub_1304 = None
	        clamp_max_56 = torch.ops.aten.clamp_max.default(clamp_min_85, 127);  clamp_min_85 = None
	        _assert_tensor_metadata_254 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_84, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_254 = None
	        _assert_tensor_metadata_255 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_56, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_255 = None
	        convert_element_type_168 = torch.ops.prims.convert_element_type.default(clamp_max_56, torch.int8);  clamp_max_56 = None
	        view_442 = torch.ops.aten.view.default(clamp_min_84, [sym_size_int, 1500, 1])
	        view_443 = torch.ops.aten.view.default(convert_element_type_168, [sym_size_int, 1500, 1])
	        reciprocal_28 = torch.ops.aten.reciprocal.default(view_442);  view_442 = None
	        mul_2784 = torch.ops.aten.mul.Tensor(reciprocal_28, 1.0);  reciprocal_28 = None
	        mul_2787 = torch.ops.aten.mul.Tensor(add_4324, mul_2784);  add_4324 = mul_2784 = None
	        round_58 = torch.ops.aten.round.default(mul_2787);  mul_2787 = None
	        add_4411 = torch.ops.aten.add.Tensor(round_58, view_443);  round_58 = view_443 = None
	        clamp_min_86 = torch.ops.aten.clamp_min.default(add_4411, -128);  add_4411 = None
	        clamp_max_57 = torch.ops.aten.clamp_max.default(clamp_min_86, 127);  clamp_min_86 = None
	        _assert_tensor_metadata_256 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_57, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_256 = None
	        convert_element_type_169 = torch.ops.prims.convert_element_type.default(clamp_max_57, torch.int8);  clamp_max_57 = None
	        view_446 = torch.ops.aten.view.default(clamp_min_84, [sym_size_int, 1500, 1]);  clamp_min_84 = None
	        view_447 = torch.ops.aten.view.default(convert_element_type_168, [sym_size_int, 1500, 1]);  convert_element_type_168 = None
	        _assert_tensor_metadata_257 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_169, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_257 = None
	        convert_element_type_170 = torch.ops.prims.convert_element_type.default(convert_element_type_169, torch.float32);  convert_element_type_169 = None
	        _assert_tensor_metadata_258 = torch.ops.aten._assert_tensor_metadata.default(view_447, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_258 = None
	        convert_element_type_171 = torch.ops.prims.convert_element_type.default(view_447, torch.float32);  view_447 = None
	        sub_1324 = torch.ops.aten.sub.Tensor(convert_element_type_170, convert_element_type_171);  convert_element_type_170 = convert_element_type_171 = None
	        mul_2809 = torch.ops.aten.mul.Tensor(sub_1324, view_446);  sub_1324 = view_446 = None
	        _assert_tensor_metadata_259 = torch.ops.aten._assert_tensor_metadata.default(mul_2809, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_259 = None
	        view_449 = torch.ops.aten.view.default(model_audio_tower_layers_4_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_4_fc1_parametrizations_weight_original0 = None
	        view_450 = torch.ops.aten.view.default(model_audio_tower_layers_4_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_4_fc1_parametrizations_weight_original1 = None
	        view_451 = torch.ops.aten.view.default(model_audio_tower_layers_4_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_4_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_260 = torch.ops.aten._assert_tensor_metadata.default(view_449, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_260 = None
	        convert_element_type_172 = torch.ops.prims.convert_element_type.default(view_449, torch.float32);  view_449 = None
	        _assert_tensor_metadata_261 = torch.ops.aten._assert_tensor_metadata.default(view_451, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_261 = None
	        convert_element_type_173 = torch.ops.prims.convert_element_type.default(view_451, torch.float32);  view_451 = None
	        sub_1328 = torch.ops.aten.sub.Tensor(convert_element_type_172, convert_element_type_173);  convert_element_type_172 = convert_element_type_173 = None
	        mul_2814 = torch.ops.aten.mul.Tensor(sub_1328, view_450);  sub_1328 = view_450 = None
	        view_452 = torch.ops.aten.view.default(mul_2814, [5120, 1280]);  mul_2814 = None
	        _assert_tensor_metadata_262 = torch.ops.aten._assert_tensor_metadata.default(view_452, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_262 = None
	        mul_2819 = sym_size_int * 1500
	        view_453 = torch.ops.aten.view.default(mul_2809, [mul_2819, 1280]);  mul_2809 = mul_2819 = None
	        permute_49 = torch.ops.aten.permute.default(view_452, [1, 0]);  view_452 = None
	        addmm_23 = torch.ops.aten.addmm.default(model_audio_tower_layers_4_fc1_bias, view_453, permute_49);  model_audio_tower_layers_4_fc1_bias = view_453 = permute_49 = None
	        view_454 = torch.ops.aten.view.default(addmm_23, [sym_size_int, 1500, 5120]);  addmm_23 = None
	        mul_2826 = torch.ops.aten.mul.Tensor(view_454, 0.5)
	        mul_2827 = torch.ops.aten.mul.Tensor(view_454, 0.7071067811865476);  view_454 = None
	        erf_6 = torch.ops.aten.erf.default(mul_2827);  mul_2827 = None
	        add_4470 = torch.ops.aten.add.Tensor(erf_6, 1);  erf_6 = None
	        mul_2828 = torch.ops.aten.mul.Tensor(mul_2826, add_4470);  mul_2826 = add_4470 = None
	        amin_29 = torch.ops.aten.amin.default(mul_2828, [2])
	        amax_29 = torch.ops.aten.amax.default(mul_2828, [2])
	        full_58 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_29 = torch.ops.aten.minimum.default(amin_29, full_58);  amin_29 = full_58 = None
	        full_59 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_29 = torch.ops.aten.maximum.default(amax_29, full_59);  amax_29 = full_59 = None
	        sub_1341 = torch.ops.aten.sub.Tensor(maximum_29, minimum_29);  maximum_29 = None
	        div_58 = torch.ops.aten.div.Tensor(sub_1341, 255.0);  sub_1341 = None
	        clamp_min_87 = torch.ops.aten.clamp_min.default(div_58, 1.1920928955078125e-07);  div_58 = None
	        div_59 = torch.ops.aten.div.Tensor(minimum_29, clamp_min_87);  minimum_29 = None
	        round_59 = torch.ops.aten.round.default(div_59);  div_59 = None
	        sub_1347 = torch.ops.aten.sub.Tensor(-128, round_59);  round_59 = None
	        clamp_min_88 = torch.ops.aten.clamp_min.default(sub_1347, -128);  sub_1347 = None
	        clamp_max_58 = torch.ops.aten.clamp_max.default(clamp_min_88, 127);  clamp_min_88 = None
	        _assert_tensor_metadata_263 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_87, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_263 = None
	        _assert_tensor_metadata_264 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_58, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_264 = None
	        convert_element_type_174 = torch.ops.prims.convert_element_type.default(clamp_max_58, torch.int8);  clamp_max_58 = None
	        view_457 = torch.ops.aten.view.default(clamp_min_87, [sym_size_int, 1500, 1])
	        view_458 = torch.ops.aten.view.default(convert_element_type_174, [sym_size_int, 1500, 1])
	        reciprocal_29 = torch.ops.aten.reciprocal.default(view_457);  view_457 = None
	        mul_2874 = torch.ops.aten.mul.Tensor(reciprocal_29, 1.0);  reciprocal_29 = None
	        mul_2877 = torch.ops.aten.mul.Tensor(mul_2828, mul_2874);  mul_2828 = mul_2874 = None
	        round_60 = torch.ops.aten.round.default(mul_2877);  mul_2877 = None
	        add_4553 = torch.ops.aten.add.Tensor(round_60, view_458);  round_60 = view_458 = None
	        clamp_min_89 = torch.ops.aten.clamp_min.default(add_4553, -128);  add_4553 = None
	        clamp_max_59 = torch.ops.aten.clamp_max.default(clamp_min_89, 127);  clamp_min_89 = None
	        _assert_tensor_metadata_265 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_59, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_265 = None
	        convert_element_type_175 = torch.ops.prims.convert_element_type.default(clamp_max_59, torch.int8);  clamp_max_59 = None
	        view_461 = torch.ops.aten.view.default(clamp_min_87, [sym_size_int, 1500, 1]);  clamp_min_87 = None
	        view_462 = torch.ops.aten.view.default(convert_element_type_174, [sym_size_int, 1500, 1]);  convert_element_type_174 = None
	        _assert_tensor_metadata_266 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_175, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_266 = None
	        convert_element_type_176 = torch.ops.prims.convert_element_type.default(convert_element_type_175, torch.float32);  convert_element_type_175 = None
	        _assert_tensor_metadata_267 = torch.ops.aten._assert_tensor_metadata.default(view_462, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_267 = None
	        convert_element_type_177 = torch.ops.prims.convert_element_type.default(view_462, torch.float32);  view_462 = None
	        sub_1367 = torch.ops.aten.sub.Tensor(convert_element_type_176, convert_element_type_177);  convert_element_type_176 = convert_element_type_177 = None
	        mul_2899 = torch.ops.aten.mul.Tensor(sub_1367, view_461);  sub_1367 = view_461 = None
	        _assert_tensor_metadata_268 = torch.ops.aten._assert_tensor_metadata.default(mul_2899, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_268 = None
	        view_464 = torch.ops.aten.view.default(model_audio_tower_layers_4_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_4_fc2_parametrizations_weight_original0 = None
	        view_465 = torch.ops.aten.view.default(model_audio_tower_layers_4_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_4_fc2_parametrizations_weight_original1 = None
	        view_466 = torch.ops.aten.view.default(model_audio_tower_layers_4_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_4_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_269 = torch.ops.aten._assert_tensor_metadata.default(view_464, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_269 = None
	        convert_element_type_178 = torch.ops.prims.convert_element_type.default(view_464, torch.float32);  view_464 = None
	        _assert_tensor_metadata_270 = torch.ops.aten._assert_tensor_metadata.default(view_466, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_270 = None
	        convert_element_type_179 = torch.ops.prims.convert_element_type.default(view_466, torch.float32);  view_466 = None
	        sub_1371 = torch.ops.aten.sub.Tensor(convert_element_type_178, convert_element_type_179);  convert_element_type_178 = convert_element_type_179 = None
	        mul_2904 = torch.ops.aten.mul.Tensor(sub_1371, view_465);  sub_1371 = view_465 = None
	        view_467 = torch.ops.aten.view.default(mul_2904, [1280, 5120]);  mul_2904 = None
	        _assert_tensor_metadata_271 = torch.ops.aten._assert_tensor_metadata.default(view_467, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_271 = None
	        mul_2909 = sym_size_int * 1500
	        view_468 = torch.ops.aten.view.default(mul_2899, [mul_2909, 5120]);  mul_2899 = mul_2909 = None
	        permute_50 = torch.ops.aten.permute.default(view_467, [1, 0]);  view_467 = None
	        addmm_24 = torch.ops.aten.addmm.default(model_audio_tower_layers_4_fc2_bias, view_468, permute_50);  model_audio_tower_layers_4_fc2_bias = view_468 = permute_50 = None
	        view_469 = torch.ops.aten.view.default(addmm_24, [sym_size_int, 1500, 1280]);  addmm_24 = None
	        add_4616 = torch.ops.aten.add.Tensor(add_4318, view_469);  add_4318 = view_469 = None
	        clone_41 = torch.ops.aten.clone.default(add_4616, memory_format = torch.contiguous_format)
	        var_mean_10 = torch.ops.aten.var_mean.correction(clone_41, [2], correction = 0, keepdim = True)
	        getitem_40 = var_mean_10[0]
	        getitem_41 = var_mean_10[1];  var_mean_10 = None
	        add_4621 = torch.ops.aten.add.Tensor(getitem_40, 1e-05);  getitem_40 = None
	        rsqrt_10 = torch.ops.aten.rsqrt.default(add_4621);  add_4621 = None
	        sub_1377 = torch.ops.aten.sub.Tensor(clone_41, getitem_41);  clone_41 = getitem_41 = None
	        mul_2920 = torch.ops.aten.mul.Tensor(sub_1377, rsqrt_10);  sub_1377 = rsqrt_10 = None
	        mul_2921 = torch.ops.aten.mul.Tensor(mul_2920, model_audio_tower_layers_5_self_attn_layer_norm_weight);  mul_2920 = model_audio_tower_layers_5_self_attn_layer_norm_weight = None
	        add_4622 = torch.ops.aten.add.Tensor(mul_2921, model_audio_tower_layers_5_self_attn_layer_norm_bias);  mul_2921 = model_audio_tower_layers_5_self_attn_layer_norm_bias = None
	        amin_30 = torch.ops.aten.amin.default(add_4622, [2])
	        amax_30 = torch.ops.aten.amax.default(add_4622, [2])
	        full_60 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_30 = torch.ops.aten.minimum.default(amin_30, full_60);  amin_30 = full_60 = None
	        full_61 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_30 = torch.ops.aten.maximum.default(amax_30, full_61);  amax_30 = full_61 = None
	        sub_1388 = torch.ops.aten.sub.Tensor(maximum_30, minimum_30);  maximum_30 = None
	        div_60 = torch.ops.aten.div.Tensor(sub_1388, 255.0);  sub_1388 = None
	        clamp_min_90 = torch.ops.aten.clamp_min.default(div_60, 1.1920928955078125e-07);  div_60 = None
	        div_61 = torch.ops.aten.div.Tensor(minimum_30, clamp_min_90);  minimum_30 = None
	        round_61 = torch.ops.aten.round.default(div_61);  div_61 = None
	        sub_1394 = torch.ops.aten.sub.Tensor(-128, round_61);  round_61 = None
	        clamp_min_91 = torch.ops.aten.clamp_min.default(sub_1394, -128);  sub_1394 = None
	        clamp_max_60 = torch.ops.aten.clamp_max.default(clamp_min_91, 127);  clamp_min_91 = None
	        _assert_tensor_metadata_272 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_90, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_272 = None
	        _assert_tensor_metadata_273 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_60, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_273 = None
	        convert_element_type_180 = torch.ops.prims.convert_element_type.default(clamp_max_60, torch.int8);  clamp_max_60 = None
	        view_472 = torch.ops.aten.view.default(clamp_min_90, [sym_size_int, 1500, 1])
	        view_473 = torch.ops.aten.view.default(convert_element_type_180, [sym_size_int, 1500, 1])
	        reciprocal_30 = torch.ops.aten.reciprocal.default(view_472);  view_472 = None
	        mul_2969 = torch.ops.aten.mul.Tensor(reciprocal_30, 1.0);  reciprocal_30 = None
	        mul_2972 = torch.ops.aten.mul.Tensor(add_4622, mul_2969);  mul_2969 = None
	        round_62 = torch.ops.aten.round.default(mul_2972);  mul_2972 = None
	        add_4709 = torch.ops.aten.add.Tensor(round_62, view_473);  round_62 = view_473 = None
	        clamp_min_92 = torch.ops.aten.clamp_min.default(add_4709, -128);  add_4709 = None
	        clamp_max_61 = torch.ops.aten.clamp_max.default(clamp_min_92, 127);  clamp_min_92 = None
	        _assert_tensor_metadata_274 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_61, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_274 = None
	        convert_element_type_181 = torch.ops.prims.convert_element_type.default(clamp_max_61, torch.int8);  clamp_max_61 = None
	        view_476 = torch.ops.aten.view.default(clamp_min_90, [sym_size_int, 1500, 1]);  clamp_min_90 = None
	        view_477 = torch.ops.aten.view.default(convert_element_type_180, [sym_size_int, 1500, 1]);  convert_element_type_180 = None
	        _assert_tensor_metadata_275 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_181, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_275 = None
	        convert_element_type_182 = torch.ops.prims.convert_element_type.default(convert_element_type_181, torch.float32);  convert_element_type_181 = None
	        _assert_tensor_metadata_276 = torch.ops.aten._assert_tensor_metadata.default(view_477, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_276 = None
	        convert_element_type_183 = torch.ops.prims.convert_element_type.default(view_477, torch.float32);  view_477 = None
	        sub_1414 = torch.ops.aten.sub.Tensor(convert_element_type_182, convert_element_type_183);  convert_element_type_182 = convert_element_type_183 = None
	        mul_2994 = torch.ops.aten.mul.Tensor(sub_1414, view_476);  sub_1414 = view_476 = None
	        _assert_tensor_metadata_277 = torch.ops.aten._assert_tensor_metadata.default(mul_2994, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_277 = None
	        view_479 = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_480 = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_481 = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_278 = torch.ops.aten._assert_tensor_metadata.default(view_479, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_278 = None
	        convert_element_type_184 = torch.ops.prims.convert_element_type.default(view_479, torch.float32);  view_479 = None
	        _assert_tensor_metadata_279 = torch.ops.aten._assert_tensor_metadata.default(view_481, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_279 = None
	        convert_element_type_185 = torch.ops.prims.convert_element_type.default(view_481, torch.float32);  view_481 = None
	        sub_1418 = torch.ops.aten.sub.Tensor(convert_element_type_184, convert_element_type_185);  convert_element_type_184 = convert_element_type_185 = None
	        mul_2999 = torch.ops.aten.mul.Tensor(sub_1418, view_480);  sub_1418 = view_480 = None
	        view_482 = torch.ops.aten.view.default(mul_2999, [1280, 1280]);  mul_2999 = None
	        _assert_tensor_metadata_280 = torch.ops.aten._assert_tensor_metadata.default(view_482, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_280 = None
	        mul_3004 = sym_size_int * 1500
	        view_483 = torch.ops.aten.view.default(mul_2994, [mul_3004, 1280]);  mul_2994 = mul_3004 = None
	        permute_51 = torch.ops.aten.permute.default(view_482, [1, 0]);  view_482 = None
	        addmm_25 = torch.ops.aten.addmm.default(model_audio_tower_layers_5_self_attn_q_proj_bias, view_483, permute_51);  model_audio_tower_layers_5_self_attn_q_proj_bias = view_483 = permute_51 = None
	        view_484 = torch.ops.aten.view.default(addmm_25, [sym_size_int, 1500, 1280]);  addmm_25 = None
	        mul_3011 = torch.ops.aten.mul.Tensor(view_484, 0.125);  view_484 = None
	        view_485 = torch.ops.aten.view.default(mul_3011, [sym_size_int, 1500, 20, 64]);  mul_3011 = None
	        permute_52 = torch.ops.aten.permute.default(view_485, [0, 2, 1, 3]);  view_485 = None
	        clone_42 = torch.ops.aten.clone.default(permute_52, memory_format = torch.contiguous_format);  permute_52 = None
	        amin_31 = torch.ops.aten.amin.default(add_4622, [2])
	        amax_31 = torch.ops.aten.amax.default(add_4622, [2])
	        full_62 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_31 = torch.ops.aten.minimum.default(amin_31, full_62);  amin_31 = full_62 = None
	        full_63 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_31 = torch.ops.aten.maximum.default(amax_31, full_63);  amax_31 = full_63 = None
	        sub_1433 = torch.ops.aten.sub.Tensor(maximum_31, minimum_31);  maximum_31 = None
	        div_62 = torch.ops.aten.div.Tensor(sub_1433, 255.0);  sub_1433 = None
	        clamp_min_93 = torch.ops.aten.clamp_min.default(div_62, 1.1920928955078125e-07);  div_62 = None
	        div_63 = torch.ops.aten.div.Tensor(minimum_31, clamp_min_93);  minimum_31 = None
	        round_63 = torch.ops.aten.round.default(div_63);  div_63 = None
	        sub_1439 = torch.ops.aten.sub.Tensor(-128, round_63);  round_63 = None
	        clamp_min_94 = torch.ops.aten.clamp_min.default(sub_1439, -128);  sub_1439 = None
	        clamp_max_62 = torch.ops.aten.clamp_max.default(clamp_min_94, 127);  clamp_min_94 = None
	        _assert_tensor_metadata_281 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_93, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_281 = None
	        _assert_tensor_metadata_282 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_62, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_282 = None
	        convert_element_type_186 = torch.ops.prims.convert_element_type.default(clamp_max_62, torch.int8);  clamp_max_62 = None
	        view_488 = torch.ops.aten.view.default(clamp_min_93, [sym_size_int, 1500, 1])
	        view_489 = torch.ops.aten.view.default(convert_element_type_186, [sym_size_int, 1500, 1])
	        reciprocal_31 = torch.ops.aten.reciprocal.default(view_488);  view_488 = None
	        mul_3065 = torch.ops.aten.mul.Tensor(reciprocal_31, 1.0);  reciprocal_31 = None
	        mul_3068 = torch.ops.aten.mul.Tensor(add_4622, mul_3065);  mul_3065 = None
	        round_64 = torch.ops.aten.round.default(mul_3068);  mul_3068 = None
	        add_4861 = torch.ops.aten.add.Tensor(round_64, view_489);  round_64 = view_489 = None
	        clamp_min_95 = torch.ops.aten.clamp_min.default(add_4861, -128);  add_4861 = None
	        clamp_max_63 = torch.ops.aten.clamp_max.default(clamp_min_95, 127);  clamp_min_95 = None
	        _assert_tensor_metadata_283 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_63, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_283 = None
	        convert_element_type_187 = torch.ops.prims.convert_element_type.default(clamp_max_63, torch.int8);  clamp_max_63 = None
	        view_492 = torch.ops.aten.view.default(clamp_min_93, [sym_size_int, 1500, 1]);  clamp_min_93 = None
	        view_493 = torch.ops.aten.view.default(convert_element_type_186, [sym_size_int, 1500, 1]);  convert_element_type_186 = None
	        _assert_tensor_metadata_284 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_187, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_284 = None
	        convert_element_type_188 = torch.ops.prims.convert_element_type.default(convert_element_type_187, torch.float32);  convert_element_type_187 = None
	        _assert_tensor_metadata_285 = torch.ops.aten._assert_tensor_metadata.default(view_493, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_285 = None
	        convert_element_type_189 = torch.ops.prims.convert_element_type.default(view_493, torch.float32);  view_493 = None
	        sub_1459 = torch.ops.aten.sub.Tensor(convert_element_type_188, convert_element_type_189);  convert_element_type_188 = convert_element_type_189 = None
	        mul_3090 = torch.ops.aten.mul.Tensor(sub_1459, view_492);  sub_1459 = view_492 = None
	        _assert_tensor_metadata_286 = torch.ops.aten._assert_tensor_metadata.default(mul_3090, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_286 = None
	        view_495 = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_496 = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_497 = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_287 = torch.ops.aten._assert_tensor_metadata.default(view_495, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_287 = None
	        convert_element_type_190 = torch.ops.prims.convert_element_type.default(view_495, torch.float32);  view_495 = None
	        _assert_tensor_metadata_288 = torch.ops.aten._assert_tensor_metadata.default(view_497, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_288 = None
	        convert_element_type_191 = torch.ops.prims.convert_element_type.default(view_497, torch.float32);  view_497 = None
	        sub_1463 = torch.ops.aten.sub.Tensor(convert_element_type_190, convert_element_type_191);  convert_element_type_190 = convert_element_type_191 = None
	        mul_3095 = torch.ops.aten.mul.Tensor(sub_1463, view_496);  sub_1463 = view_496 = None
	        view_498 = torch.ops.aten.view.default(mul_3095, [1280, 1280]);  mul_3095 = None
	        _assert_tensor_metadata_289 = torch.ops.aten._assert_tensor_metadata.default(view_498, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_289 = None
	        permute_53 = torch.ops.aten.permute.default(view_498, [1, 0]);  view_498 = None
	        mul_3098 = sym_size_int * 1500
	        view_499 = torch.ops.aten.view.default(mul_3090, [mul_3098, 1280]);  mul_3090 = mul_3098 = None
	        mm_5 = torch.ops.aten.mm.default(view_499, permute_53);  view_499 = permute_53 = None
	        view_500 = torch.ops.aten.view.default(mm_5, [sym_size_int, 1500, 1280]);  mm_5 = None
	        view_501 = torch.ops.aten.view.default(view_500, [sym_size_int, -1, 20, 64]);  view_500 = None
	        permute_54 = torch.ops.aten.permute.default(view_501, [0, 2, 1, 3]);  view_501 = None
	        clone_43 = torch.ops.aten.clone.default(permute_54, memory_format = torch.contiguous_format);  permute_54 = None
	        amin_32 = torch.ops.aten.amin.default(add_4622, [2])
	        amax_32 = torch.ops.aten.amax.default(add_4622, [2])
	        full_64 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_32 = torch.ops.aten.minimum.default(amin_32, full_64);  amin_32 = full_64 = None
	        full_65 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_32 = torch.ops.aten.maximum.default(amax_32, full_65);  amax_32 = full_65 = None
	        sub_1477 = torch.ops.aten.sub.Tensor(maximum_32, minimum_32);  maximum_32 = None
	        div_64 = torch.ops.aten.div.Tensor(sub_1477, 255.0);  sub_1477 = None
	        clamp_min_96 = torch.ops.aten.clamp_min.default(div_64, 1.1920928955078125e-07);  div_64 = None
	        div_65 = torch.ops.aten.div.Tensor(minimum_32, clamp_min_96);  minimum_32 = None
	        round_65 = torch.ops.aten.round.default(div_65);  div_65 = None
	        sub_1483 = torch.ops.aten.sub.Tensor(-128, round_65);  round_65 = None
	        clamp_min_97 = torch.ops.aten.clamp_min.default(sub_1483, -128);  sub_1483 = None
	        clamp_max_64 = torch.ops.aten.clamp_max.default(clamp_min_97, 127);  clamp_min_97 = None
	        _assert_tensor_metadata_290 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_96, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_290 = None
	        _assert_tensor_metadata_291 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_64, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_291 = None
	        convert_element_type_192 = torch.ops.prims.convert_element_type.default(clamp_max_64, torch.int8);  clamp_max_64 = None
	        view_504 = torch.ops.aten.view.default(clamp_min_96, [sym_size_int, 1500, 1])
	        view_505 = torch.ops.aten.view.default(convert_element_type_192, [sym_size_int, 1500, 1])
	        reciprocal_32 = torch.ops.aten.reciprocal.default(view_504);  view_504 = None
	        mul_3164 = torch.ops.aten.mul.Tensor(reciprocal_32, 1.0);  reciprocal_32 = None
	        mul_3167 = torch.ops.aten.mul.Tensor(add_4622, mul_3164);  add_4622 = mul_3164 = None
	        round_66 = torch.ops.aten.round.default(mul_3167);  mul_3167 = None
	        add_5009 = torch.ops.aten.add.Tensor(round_66, view_505);  round_66 = view_505 = None
	        clamp_min_98 = torch.ops.aten.clamp_min.default(add_5009, -128);  add_5009 = None
	        clamp_max_65 = torch.ops.aten.clamp_max.default(clamp_min_98, 127);  clamp_min_98 = None
	        _assert_tensor_metadata_292 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_65, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_292 = None
	        convert_element_type_193 = torch.ops.prims.convert_element_type.default(clamp_max_65, torch.int8);  clamp_max_65 = None
	        view_508 = torch.ops.aten.view.default(clamp_min_96, [sym_size_int, 1500, 1]);  clamp_min_96 = None
	        view_509 = torch.ops.aten.view.default(convert_element_type_192, [sym_size_int, 1500, 1]);  convert_element_type_192 = None
	        _assert_tensor_metadata_293 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_193, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_293 = None
	        convert_element_type_194 = torch.ops.prims.convert_element_type.default(convert_element_type_193, torch.float32);  convert_element_type_193 = None
	        _assert_tensor_metadata_294 = torch.ops.aten._assert_tensor_metadata.default(view_509, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_294 = None
	        convert_element_type_195 = torch.ops.prims.convert_element_type.default(view_509, torch.float32);  view_509 = None
	        sub_1503 = torch.ops.aten.sub.Tensor(convert_element_type_194, convert_element_type_195);  convert_element_type_194 = convert_element_type_195 = None
	        mul_3189 = torch.ops.aten.mul.Tensor(sub_1503, view_508);  sub_1503 = view_508 = None
	        _assert_tensor_metadata_295 = torch.ops.aten._assert_tensor_metadata.default(mul_3189, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_295 = None
	        view_511 = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_512 = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_513 = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_296 = torch.ops.aten._assert_tensor_metadata.default(view_511, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_296 = None
	        convert_element_type_196 = torch.ops.prims.convert_element_type.default(view_511, torch.float32);  view_511 = None
	        _assert_tensor_metadata_297 = torch.ops.aten._assert_tensor_metadata.default(view_513, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_297 = None
	        convert_element_type_197 = torch.ops.prims.convert_element_type.default(view_513, torch.float32);  view_513 = None
	        sub_1507 = torch.ops.aten.sub.Tensor(convert_element_type_196, convert_element_type_197);  convert_element_type_196 = convert_element_type_197 = None
	        mul_3194 = torch.ops.aten.mul.Tensor(sub_1507, view_512);  sub_1507 = view_512 = None
	        view_514 = torch.ops.aten.view.default(mul_3194, [1280, 1280]);  mul_3194 = None
	        _assert_tensor_metadata_298 = torch.ops.aten._assert_tensor_metadata.default(view_514, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_298 = None
	        mul_3199 = sym_size_int * 1500
	        view_515 = torch.ops.aten.view.default(mul_3189, [mul_3199, 1280]);  mul_3189 = mul_3199 = None
	        permute_55 = torch.ops.aten.permute.default(view_514, [1, 0]);  view_514 = None
	        addmm_26 = torch.ops.aten.addmm.default(model_audio_tower_layers_5_self_attn_v_proj_bias, view_515, permute_55);  model_audio_tower_layers_5_self_attn_v_proj_bias = view_515 = permute_55 = None
	        view_516 = torch.ops.aten.view.default(addmm_26, [sym_size_int, 1500, 1280]);  addmm_26 = None
	        view_517 = torch.ops.aten.view.default(view_516, [sym_size_int, -1, 20, 64]);  view_516 = None
	        permute_56 = torch.ops.aten.permute.default(view_517, [0, 2, 1, 3]);  view_517 = None
	        clone_44 = torch.ops.aten.clone.default(permute_56, memory_format = torch.contiguous_format);  permute_56 = None
	        _scaled_dot_product_efficient_attention_5 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_42, clone_43, clone_44, None, False, scale = 1.0);  clone_42 = clone_43 = clone_44 = None
	        getitem_42 = _scaled_dot_product_efficient_attention_5[0];  _scaled_dot_product_efficient_attention_5 = None
	        permute_57 = torch.ops.aten.permute.default(getitem_42, [0, 2, 1, 3]);  getitem_42 = None
	        view_518 = torch.ops.aten.view.default(permute_57, [sym_size_int, 1500, -1]);  permute_57 = None
	        amin_33 = torch.ops.aten.amin.default(view_518, [2])
	        amax_33 = torch.ops.aten.amax.default(view_518, [2])
	        full_66 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_33 = torch.ops.aten.minimum.default(amin_33, full_66);  amin_33 = full_66 = None
	        full_67 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_33 = torch.ops.aten.maximum.default(amax_33, full_67);  amax_33 = full_67 = None
	        sub_1525 = torch.ops.aten.sub.Tensor(maximum_33, minimum_33);  maximum_33 = None
	        div_66 = torch.ops.aten.div.Tensor(sub_1525, 255.0);  sub_1525 = None
	        clamp_min_99 = torch.ops.aten.clamp_min.default(div_66, 1.1920928955078125e-07);  div_66 = None
	        div_67 = torch.ops.aten.div.Tensor(minimum_33, clamp_min_99);  minimum_33 = None
	        round_67 = torch.ops.aten.round.default(div_67);  div_67 = None
	        sub_1531 = torch.ops.aten.sub.Tensor(-128, round_67);  round_67 = None
	        clamp_min_100 = torch.ops.aten.clamp_min.default(sub_1531, -128);  sub_1531 = None
	        clamp_max_66 = torch.ops.aten.clamp_max.default(clamp_min_100, 127);  clamp_min_100 = None
	        _assert_tensor_metadata_299 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_99, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_299 = None
	        _assert_tensor_metadata_300 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_66, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_300 = None
	        convert_element_type_198 = torch.ops.prims.convert_element_type.default(clamp_max_66, torch.int8);  clamp_max_66 = None
	        view_521 = torch.ops.aten.view.default(clamp_min_99, [sym_size_int, 1500, 1])
	        view_522 = torch.ops.aten.view.default(convert_element_type_198, [sym_size_int, 1500, 1])
	        reciprocal_33 = torch.ops.aten.reciprocal.default(view_521);  view_521 = None
	        mul_3269 = torch.ops.aten.mul.Tensor(reciprocal_33, 1.0);  reciprocal_33 = None
	        mul_3272 = torch.ops.aten.mul.Tensor(view_518, mul_3269);  view_518 = mul_3269 = None
	        round_68 = torch.ops.aten.round.default(mul_3272);  mul_3272 = None
	        add_5173 = torch.ops.aten.add.Tensor(round_68, view_522);  round_68 = view_522 = None
	        clamp_min_101 = torch.ops.aten.clamp_min.default(add_5173, -128);  add_5173 = None
	        clamp_max_67 = torch.ops.aten.clamp_max.default(clamp_min_101, 127);  clamp_min_101 = None
	        _assert_tensor_metadata_301 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_67, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_301 = None
	        convert_element_type_199 = torch.ops.prims.convert_element_type.default(clamp_max_67, torch.int8);  clamp_max_67 = None
	        view_525 = torch.ops.aten.view.default(clamp_min_99, [sym_size_int, 1500, 1]);  clamp_min_99 = None
	        view_526 = torch.ops.aten.view.default(convert_element_type_198, [sym_size_int, 1500, 1]);  convert_element_type_198 = None
	        _assert_tensor_metadata_302 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_199, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_302 = None
	        convert_element_type_200 = torch.ops.prims.convert_element_type.default(convert_element_type_199, torch.float32);  convert_element_type_199 = None
	        _assert_tensor_metadata_303 = torch.ops.aten._assert_tensor_metadata.default(view_526, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_303 = None
	        convert_element_type_201 = torch.ops.prims.convert_element_type.default(view_526, torch.float32);  view_526 = None
	        sub_1551 = torch.ops.aten.sub.Tensor(convert_element_type_200, convert_element_type_201);  convert_element_type_200 = convert_element_type_201 = None
	        mul_3294 = torch.ops.aten.mul.Tensor(sub_1551, view_525);  sub_1551 = view_525 = None
	        _assert_tensor_metadata_304 = torch.ops.aten._assert_tensor_metadata.default(mul_3294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_304 = None
	        view_528 = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_529 = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_530 = torch.ops.aten.view.default(model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_305 = torch.ops.aten._assert_tensor_metadata.default(view_528, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_305 = None
	        convert_element_type_202 = torch.ops.prims.convert_element_type.default(view_528, torch.float32);  view_528 = None
	        _assert_tensor_metadata_306 = torch.ops.aten._assert_tensor_metadata.default(view_530, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_306 = None
	        convert_element_type_203 = torch.ops.prims.convert_element_type.default(view_530, torch.float32);  view_530 = None
	        sub_1555 = torch.ops.aten.sub.Tensor(convert_element_type_202, convert_element_type_203);  convert_element_type_202 = convert_element_type_203 = None
	        mul_3299 = torch.ops.aten.mul.Tensor(sub_1555, view_529);  sub_1555 = view_529 = None
	        view_531 = torch.ops.aten.view.default(mul_3299, [1280, 1280]);  mul_3299 = None
	        _assert_tensor_metadata_307 = torch.ops.aten._assert_tensor_metadata.default(view_531, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_307 = None
	        mul_3304 = sym_size_int * 1500
	        view_532 = torch.ops.aten.view.default(mul_3294, [mul_3304, 1280]);  mul_3294 = mul_3304 = None
	        permute_58 = torch.ops.aten.permute.default(view_531, [1, 0]);  view_531 = None
	        addmm_27 = torch.ops.aten.addmm.default(model_audio_tower_layers_5_self_attn_out_proj_bias, view_532, permute_58);  model_audio_tower_layers_5_self_attn_out_proj_bias = view_532 = permute_58 = None
	        view_533 = torch.ops.aten.view.default(addmm_27, [sym_size_int, 1500, 1280]);  addmm_27 = None
	        add_5236 = torch.ops.aten.add.Tensor(add_4616, view_533);  add_4616 = view_533 = None
	        clone_46 = torch.ops.aten.clone.default(add_5236, memory_format = torch.contiguous_format)
	        var_mean_11 = torch.ops.aten.var_mean.correction(clone_46, [2], correction = 0, keepdim = True)
	        getitem_46 = var_mean_11[0]
	        getitem_47 = var_mean_11[1];  var_mean_11 = None
	        add_5241 = torch.ops.aten.add.Tensor(getitem_46, 1e-05);  getitem_46 = None
	        rsqrt_11 = torch.ops.aten.rsqrt.default(add_5241);  add_5241 = None
	        sub_1561 = torch.ops.aten.sub.Tensor(clone_46, getitem_47);  clone_46 = getitem_47 = None
	        mul_3315 = torch.ops.aten.mul.Tensor(sub_1561, rsqrt_11);  sub_1561 = rsqrt_11 = None
	        mul_3316 = torch.ops.aten.mul.Tensor(mul_3315, model_audio_tower_layers_5_final_layer_norm_weight);  mul_3315 = model_audio_tower_layers_5_final_layer_norm_weight = None
	        add_5242 = torch.ops.aten.add.Tensor(mul_3316, model_audio_tower_layers_5_final_layer_norm_bias);  mul_3316 = model_audio_tower_layers_5_final_layer_norm_bias = None
	        amin_34 = torch.ops.aten.amin.default(add_5242, [2])
	        amax_34 = torch.ops.aten.amax.default(add_5242, [2])
	        full_68 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_34 = torch.ops.aten.minimum.default(amin_34, full_68);  amin_34 = full_68 = None
	        full_69 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_34 = torch.ops.aten.maximum.default(amax_34, full_69);  amax_34 = full_69 = None
	        sub_1572 = torch.ops.aten.sub.Tensor(maximum_34, minimum_34);  maximum_34 = None
	        div_68 = torch.ops.aten.div.Tensor(sub_1572, 255.0);  sub_1572 = None
	        clamp_min_102 = torch.ops.aten.clamp_min.default(div_68, 1.1920928955078125e-07);  div_68 = None
	        div_69 = torch.ops.aten.div.Tensor(minimum_34, clamp_min_102);  minimum_34 = None
	        round_69 = torch.ops.aten.round.default(div_69);  div_69 = None
	        sub_1578 = torch.ops.aten.sub.Tensor(-128, round_69);  round_69 = None
	        clamp_min_103 = torch.ops.aten.clamp_min.default(sub_1578, -128);  sub_1578 = None
	        clamp_max_68 = torch.ops.aten.clamp_max.default(clamp_min_103, 127);  clamp_min_103 = None
	        _assert_tensor_metadata_308 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_102, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_308 = None
	        _assert_tensor_metadata_309 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_68, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_309 = None
	        convert_element_type_204 = torch.ops.prims.convert_element_type.default(clamp_max_68, torch.int8);  clamp_max_68 = None
	        view_536 = torch.ops.aten.view.default(clamp_min_102, [sym_size_int, 1500, 1])
	        view_537 = torch.ops.aten.view.default(convert_element_type_204, [sym_size_int, 1500, 1])
	        reciprocal_34 = torch.ops.aten.reciprocal.default(view_536);  view_536 = None
	        mul_3364 = torch.ops.aten.mul.Tensor(reciprocal_34, 1.0);  reciprocal_34 = None
	        mul_3367 = torch.ops.aten.mul.Tensor(add_5242, mul_3364);  add_5242 = mul_3364 = None
	        round_70 = torch.ops.aten.round.default(mul_3367);  mul_3367 = None
	        add_5329 = torch.ops.aten.add.Tensor(round_70, view_537);  round_70 = view_537 = None
	        clamp_min_104 = torch.ops.aten.clamp_min.default(add_5329, -128);  add_5329 = None
	        clamp_max_69 = torch.ops.aten.clamp_max.default(clamp_min_104, 127);  clamp_min_104 = None
	        _assert_tensor_metadata_310 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_69, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_310 = None
	        convert_element_type_205 = torch.ops.prims.convert_element_type.default(clamp_max_69, torch.int8);  clamp_max_69 = None
	        view_540 = torch.ops.aten.view.default(clamp_min_102, [sym_size_int, 1500, 1]);  clamp_min_102 = None
	        view_541 = torch.ops.aten.view.default(convert_element_type_204, [sym_size_int, 1500, 1]);  convert_element_type_204 = None
	        _assert_tensor_metadata_311 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_205, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_311 = None
	        convert_element_type_206 = torch.ops.prims.convert_element_type.default(convert_element_type_205, torch.float32);  convert_element_type_205 = None
	        _assert_tensor_metadata_312 = torch.ops.aten._assert_tensor_metadata.default(view_541, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_312 = None
	        convert_element_type_207 = torch.ops.prims.convert_element_type.default(view_541, torch.float32);  view_541 = None
	        sub_1598 = torch.ops.aten.sub.Tensor(convert_element_type_206, convert_element_type_207);  convert_element_type_206 = convert_element_type_207 = None
	        mul_3389 = torch.ops.aten.mul.Tensor(sub_1598, view_540);  sub_1598 = view_540 = None
	        _assert_tensor_metadata_313 = torch.ops.aten._assert_tensor_metadata.default(mul_3389, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_313 = None
	        view_543 = torch.ops.aten.view.default(model_audio_tower_layers_5_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_5_fc1_parametrizations_weight_original0 = None
	        view_544 = torch.ops.aten.view.default(model_audio_tower_layers_5_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_5_fc1_parametrizations_weight_original1 = None
	        view_545 = torch.ops.aten.view.default(model_audio_tower_layers_5_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_5_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_314 = torch.ops.aten._assert_tensor_metadata.default(view_543, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_314 = None
	        convert_element_type_208 = torch.ops.prims.convert_element_type.default(view_543, torch.float32);  view_543 = None
	        _assert_tensor_metadata_315 = torch.ops.aten._assert_tensor_metadata.default(view_545, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_315 = None
	        convert_element_type_209 = torch.ops.prims.convert_element_type.default(view_545, torch.float32);  view_545 = None
	        sub_1602 = torch.ops.aten.sub.Tensor(convert_element_type_208, convert_element_type_209);  convert_element_type_208 = convert_element_type_209 = None
	        mul_3394 = torch.ops.aten.mul.Tensor(sub_1602, view_544);  sub_1602 = view_544 = None
	        view_546 = torch.ops.aten.view.default(mul_3394, [5120, 1280]);  mul_3394 = None
	        _assert_tensor_metadata_316 = torch.ops.aten._assert_tensor_metadata.default(view_546, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_316 = None
	        mul_3399 = sym_size_int * 1500
	        view_547 = torch.ops.aten.view.default(mul_3389, [mul_3399, 1280]);  mul_3389 = mul_3399 = None
	        permute_59 = torch.ops.aten.permute.default(view_546, [1, 0]);  view_546 = None
	        addmm_28 = torch.ops.aten.addmm.default(model_audio_tower_layers_5_fc1_bias, view_547, permute_59);  model_audio_tower_layers_5_fc1_bias = view_547 = permute_59 = None
	        view_548 = torch.ops.aten.view.default(addmm_28, [sym_size_int, 1500, 5120]);  addmm_28 = None
	        mul_3406 = torch.ops.aten.mul.Tensor(view_548, 0.5)
	        mul_3407 = torch.ops.aten.mul.Tensor(view_548, 0.7071067811865476);  view_548 = None
	        erf_7 = torch.ops.aten.erf.default(mul_3407);  mul_3407 = None
	        add_5388 = torch.ops.aten.add.Tensor(erf_7, 1);  erf_7 = None
	        mul_3408 = torch.ops.aten.mul.Tensor(mul_3406, add_5388);  mul_3406 = add_5388 = None
	        amin_35 = torch.ops.aten.amin.default(mul_3408, [2])
	        amax_35 = torch.ops.aten.amax.default(mul_3408, [2])
	        full_70 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_35 = torch.ops.aten.minimum.default(amin_35, full_70);  amin_35 = full_70 = None
	        full_71 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_35 = torch.ops.aten.maximum.default(amax_35, full_71);  amax_35 = full_71 = None
	        sub_1615 = torch.ops.aten.sub.Tensor(maximum_35, minimum_35);  maximum_35 = None
	        div_70 = torch.ops.aten.div.Tensor(sub_1615, 255.0);  sub_1615 = None
	        clamp_min_105 = torch.ops.aten.clamp_min.default(div_70, 1.1920928955078125e-07);  div_70 = None
	        div_71 = torch.ops.aten.div.Tensor(minimum_35, clamp_min_105);  minimum_35 = None
	        round_71 = torch.ops.aten.round.default(div_71);  div_71 = None
	        sub_1621 = torch.ops.aten.sub.Tensor(-128, round_71);  round_71 = None
	        clamp_min_106 = torch.ops.aten.clamp_min.default(sub_1621, -128);  sub_1621 = None
	        clamp_max_70 = torch.ops.aten.clamp_max.default(clamp_min_106, 127);  clamp_min_106 = None
	        _assert_tensor_metadata_317 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_105, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_317 = None
	        _assert_tensor_metadata_318 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_70, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_318 = None
	        convert_element_type_210 = torch.ops.prims.convert_element_type.default(clamp_max_70, torch.int8);  clamp_max_70 = None
	        view_551 = torch.ops.aten.view.default(clamp_min_105, [sym_size_int, 1500, 1])
	        view_552 = torch.ops.aten.view.default(convert_element_type_210, [sym_size_int, 1500, 1])
	        reciprocal_35 = torch.ops.aten.reciprocal.default(view_551);  view_551 = None
	        mul_3454 = torch.ops.aten.mul.Tensor(reciprocal_35, 1.0);  reciprocal_35 = None
	        mul_3457 = torch.ops.aten.mul.Tensor(mul_3408, mul_3454);  mul_3408 = mul_3454 = None
	        round_72 = torch.ops.aten.round.default(mul_3457);  mul_3457 = None
	        add_5471 = torch.ops.aten.add.Tensor(round_72, view_552);  round_72 = view_552 = None
	        clamp_min_107 = torch.ops.aten.clamp_min.default(add_5471, -128);  add_5471 = None
	        clamp_max_71 = torch.ops.aten.clamp_max.default(clamp_min_107, 127);  clamp_min_107 = None
	        _assert_tensor_metadata_319 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_71, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_319 = None
	        convert_element_type_211 = torch.ops.prims.convert_element_type.default(clamp_max_71, torch.int8);  clamp_max_71 = None
	        view_555 = torch.ops.aten.view.default(clamp_min_105, [sym_size_int, 1500, 1]);  clamp_min_105 = None
	        view_556 = torch.ops.aten.view.default(convert_element_type_210, [sym_size_int, 1500, 1]);  convert_element_type_210 = None
	        _assert_tensor_metadata_320 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_211, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_320 = None
	        convert_element_type_212 = torch.ops.prims.convert_element_type.default(convert_element_type_211, torch.float32);  convert_element_type_211 = None
	        _assert_tensor_metadata_321 = torch.ops.aten._assert_tensor_metadata.default(view_556, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_321 = None
	        convert_element_type_213 = torch.ops.prims.convert_element_type.default(view_556, torch.float32);  view_556 = None
	        sub_1641 = torch.ops.aten.sub.Tensor(convert_element_type_212, convert_element_type_213);  convert_element_type_212 = convert_element_type_213 = None
	        mul_3479 = torch.ops.aten.mul.Tensor(sub_1641, view_555);  sub_1641 = view_555 = None
	        _assert_tensor_metadata_322 = torch.ops.aten._assert_tensor_metadata.default(mul_3479, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_322 = None
	        view_558 = torch.ops.aten.view.default(model_audio_tower_layers_5_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_5_fc2_parametrizations_weight_original0 = None
	        view_559 = torch.ops.aten.view.default(model_audio_tower_layers_5_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_5_fc2_parametrizations_weight_original1 = None
	        view_560 = torch.ops.aten.view.default(model_audio_tower_layers_5_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_5_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_323 = torch.ops.aten._assert_tensor_metadata.default(view_558, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_323 = None
	        convert_element_type_214 = torch.ops.prims.convert_element_type.default(view_558, torch.float32);  view_558 = None
	        _assert_tensor_metadata_324 = torch.ops.aten._assert_tensor_metadata.default(view_560, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_324 = None
	        convert_element_type_215 = torch.ops.prims.convert_element_type.default(view_560, torch.float32);  view_560 = None
	        sub_1645 = torch.ops.aten.sub.Tensor(convert_element_type_214, convert_element_type_215);  convert_element_type_214 = convert_element_type_215 = None
	        mul_3484 = torch.ops.aten.mul.Tensor(sub_1645, view_559);  sub_1645 = view_559 = None
	        view_561 = torch.ops.aten.view.default(mul_3484, [1280, 5120]);  mul_3484 = None
	        _assert_tensor_metadata_325 = torch.ops.aten._assert_tensor_metadata.default(view_561, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_325 = None
	        mul_3489 = sym_size_int * 1500
	        view_562 = torch.ops.aten.view.default(mul_3479, [mul_3489, 5120]);  mul_3479 = mul_3489 = None
	        permute_60 = torch.ops.aten.permute.default(view_561, [1, 0]);  view_561 = None
	        addmm_29 = torch.ops.aten.addmm.default(model_audio_tower_layers_5_fc2_bias, view_562, permute_60);  model_audio_tower_layers_5_fc2_bias = view_562 = permute_60 = None
	        view_563 = torch.ops.aten.view.default(addmm_29, [sym_size_int, 1500, 1280]);  addmm_29 = None
	        add_5534 = torch.ops.aten.add.Tensor(add_5236, view_563);  add_5236 = view_563 = None
	        clone_49 = torch.ops.aten.clone.default(add_5534, memory_format = torch.contiguous_format)
	        var_mean_12 = torch.ops.aten.var_mean.correction(clone_49, [2], correction = 0, keepdim = True)
	        getitem_48 = var_mean_12[0]
	        getitem_49 = var_mean_12[1];  var_mean_12 = None
	        add_5539 = torch.ops.aten.add.Tensor(getitem_48, 1e-05);  getitem_48 = None
	        rsqrt_12 = torch.ops.aten.rsqrt.default(add_5539);  add_5539 = None
	        sub_1651 = torch.ops.aten.sub.Tensor(clone_49, getitem_49);  clone_49 = getitem_49 = None
	        mul_3500 = torch.ops.aten.mul.Tensor(sub_1651, rsqrt_12);  sub_1651 = rsqrt_12 = None
	        mul_3501 = torch.ops.aten.mul.Tensor(mul_3500, model_audio_tower_layers_6_self_attn_layer_norm_weight);  mul_3500 = model_audio_tower_layers_6_self_attn_layer_norm_weight = None
	        add_5540 = torch.ops.aten.add.Tensor(mul_3501, model_audio_tower_layers_6_self_attn_layer_norm_bias);  mul_3501 = model_audio_tower_layers_6_self_attn_layer_norm_bias = None
	        amin_36 = torch.ops.aten.amin.default(add_5540, [2])
	        amax_36 = torch.ops.aten.amax.default(add_5540, [2])
	        full_72 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_36 = torch.ops.aten.minimum.default(amin_36, full_72);  amin_36 = full_72 = None
	        full_73 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_36 = torch.ops.aten.maximum.default(amax_36, full_73);  amax_36 = full_73 = None
	        sub_1662 = torch.ops.aten.sub.Tensor(maximum_36, minimum_36);  maximum_36 = None
	        div_72 = torch.ops.aten.div.Tensor(sub_1662, 255.0);  sub_1662 = None
	        clamp_min_108 = torch.ops.aten.clamp_min.default(div_72, 1.1920928955078125e-07);  div_72 = None
	        div_73 = torch.ops.aten.div.Tensor(minimum_36, clamp_min_108);  minimum_36 = None
	        round_73 = torch.ops.aten.round.default(div_73);  div_73 = None
	        sub_1668 = torch.ops.aten.sub.Tensor(-128, round_73);  round_73 = None
	        clamp_min_109 = torch.ops.aten.clamp_min.default(sub_1668, -128);  sub_1668 = None
	        clamp_max_72 = torch.ops.aten.clamp_max.default(clamp_min_109, 127);  clamp_min_109 = None
	        _assert_tensor_metadata_326 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_108, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_326 = None
	        _assert_tensor_metadata_327 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_72, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_327 = None
	        convert_element_type_216 = torch.ops.prims.convert_element_type.default(clamp_max_72, torch.int8);  clamp_max_72 = None
	        view_566 = torch.ops.aten.view.default(clamp_min_108, [sym_size_int, 1500, 1])
	        view_567 = torch.ops.aten.view.default(convert_element_type_216, [sym_size_int, 1500, 1])
	        reciprocal_36 = torch.ops.aten.reciprocal.default(view_566);  view_566 = None
	        mul_3549 = torch.ops.aten.mul.Tensor(reciprocal_36, 1.0);  reciprocal_36 = None
	        mul_3552 = torch.ops.aten.mul.Tensor(add_5540, mul_3549);  mul_3549 = None
	        round_74 = torch.ops.aten.round.default(mul_3552);  mul_3552 = None
	        add_5627 = torch.ops.aten.add.Tensor(round_74, view_567);  round_74 = view_567 = None
	        clamp_min_110 = torch.ops.aten.clamp_min.default(add_5627, -128);  add_5627 = None
	        clamp_max_73 = torch.ops.aten.clamp_max.default(clamp_min_110, 127);  clamp_min_110 = None
	        _assert_tensor_metadata_328 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_73, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_328 = None
	        convert_element_type_217 = torch.ops.prims.convert_element_type.default(clamp_max_73, torch.int8);  clamp_max_73 = None
	        view_570 = torch.ops.aten.view.default(clamp_min_108, [sym_size_int, 1500, 1]);  clamp_min_108 = None
	        view_571 = torch.ops.aten.view.default(convert_element_type_216, [sym_size_int, 1500, 1]);  convert_element_type_216 = None
	        _assert_tensor_metadata_329 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_217, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_329 = None
	        convert_element_type_218 = torch.ops.prims.convert_element_type.default(convert_element_type_217, torch.float32);  convert_element_type_217 = None
	        _assert_tensor_metadata_330 = torch.ops.aten._assert_tensor_metadata.default(view_571, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_330 = None
	        convert_element_type_219 = torch.ops.prims.convert_element_type.default(view_571, torch.float32);  view_571 = None
	        sub_1688 = torch.ops.aten.sub.Tensor(convert_element_type_218, convert_element_type_219);  convert_element_type_218 = convert_element_type_219 = None
	        mul_3574 = torch.ops.aten.mul.Tensor(sub_1688, view_570);  sub_1688 = view_570 = None
	        _assert_tensor_metadata_331 = torch.ops.aten._assert_tensor_metadata.default(mul_3574, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_331 = None
	        view_573 = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_574 = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_575 = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_332 = torch.ops.aten._assert_tensor_metadata.default(view_573, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_332 = None
	        convert_element_type_220 = torch.ops.prims.convert_element_type.default(view_573, torch.float32);  view_573 = None
	        _assert_tensor_metadata_333 = torch.ops.aten._assert_tensor_metadata.default(view_575, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_333 = None
	        convert_element_type_221 = torch.ops.prims.convert_element_type.default(view_575, torch.float32);  view_575 = None
	        sub_1692 = torch.ops.aten.sub.Tensor(convert_element_type_220, convert_element_type_221);  convert_element_type_220 = convert_element_type_221 = None
	        mul_3579 = torch.ops.aten.mul.Tensor(sub_1692, view_574);  sub_1692 = view_574 = None
	        view_576 = torch.ops.aten.view.default(mul_3579, [1280, 1280]);  mul_3579 = None
	        _assert_tensor_metadata_334 = torch.ops.aten._assert_tensor_metadata.default(view_576, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_334 = None
	        mul_3584 = sym_size_int * 1500
	        view_577 = torch.ops.aten.view.default(mul_3574, [mul_3584, 1280]);  mul_3574 = mul_3584 = None
	        permute_61 = torch.ops.aten.permute.default(view_576, [1, 0]);  view_576 = None
	        addmm_30 = torch.ops.aten.addmm.default(model_audio_tower_layers_6_self_attn_q_proj_bias, view_577, permute_61);  model_audio_tower_layers_6_self_attn_q_proj_bias = view_577 = permute_61 = None
	        view_578 = torch.ops.aten.view.default(addmm_30, [sym_size_int, 1500, 1280]);  addmm_30 = None
	        mul_3591 = torch.ops.aten.mul.Tensor(view_578, 0.125);  view_578 = None
	        view_579 = torch.ops.aten.view.default(mul_3591, [sym_size_int, 1500, 20, 64]);  mul_3591 = None
	        permute_62 = torch.ops.aten.permute.default(view_579, [0, 2, 1, 3]);  view_579 = None
	        clone_50 = torch.ops.aten.clone.default(permute_62, memory_format = torch.contiguous_format);  permute_62 = None
	        amin_37 = torch.ops.aten.amin.default(add_5540, [2])
	        amax_37 = torch.ops.aten.amax.default(add_5540, [2])
	        full_74 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_37 = torch.ops.aten.minimum.default(amin_37, full_74);  amin_37 = full_74 = None
	        full_75 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_37 = torch.ops.aten.maximum.default(amax_37, full_75);  amax_37 = full_75 = None
	        sub_1707 = torch.ops.aten.sub.Tensor(maximum_37, minimum_37);  maximum_37 = None
	        div_74 = torch.ops.aten.div.Tensor(sub_1707, 255.0);  sub_1707 = None
	        clamp_min_111 = torch.ops.aten.clamp_min.default(div_74, 1.1920928955078125e-07);  div_74 = None
	        div_75 = torch.ops.aten.div.Tensor(minimum_37, clamp_min_111);  minimum_37 = None
	        round_75 = torch.ops.aten.round.default(div_75);  div_75 = None
	        sub_1713 = torch.ops.aten.sub.Tensor(-128, round_75);  round_75 = None
	        clamp_min_112 = torch.ops.aten.clamp_min.default(sub_1713, -128);  sub_1713 = None
	        clamp_max_74 = torch.ops.aten.clamp_max.default(clamp_min_112, 127);  clamp_min_112 = None
	        _assert_tensor_metadata_335 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_111, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_335 = None
	        _assert_tensor_metadata_336 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_74, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_336 = None
	        convert_element_type_222 = torch.ops.prims.convert_element_type.default(clamp_max_74, torch.int8);  clamp_max_74 = None
	        view_582 = torch.ops.aten.view.default(clamp_min_111, [sym_size_int, 1500, 1])
	        view_583 = torch.ops.aten.view.default(convert_element_type_222, [sym_size_int, 1500, 1])
	        reciprocal_37 = torch.ops.aten.reciprocal.default(view_582);  view_582 = None
	        mul_3645 = torch.ops.aten.mul.Tensor(reciprocal_37, 1.0);  reciprocal_37 = None
	        mul_3648 = torch.ops.aten.mul.Tensor(add_5540, mul_3645);  mul_3645 = None
	        round_76 = torch.ops.aten.round.default(mul_3648);  mul_3648 = None
	        add_5779 = torch.ops.aten.add.Tensor(round_76, view_583);  round_76 = view_583 = None
	        clamp_min_113 = torch.ops.aten.clamp_min.default(add_5779, -128);  add_5779 = None
	        clamp_max_75 = torch.ops.aten.clamp_max.default(clamp_min_113, 127);  clamp_min_113 = None
	        _assert_tensor_metadata_337 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_75, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_337 = None
	        convert_element_type_223 = torch.ops.prims.convert_element_type.default(clamp_max_75, torch.int8);  clamp_max_75 = None
	        view_586 = torch.ops.aten.view.default(clamp_min_111, [sym_size_int, 1500, 1]);  clamp_min_111 = None
	        view_587 = torch.ops.aten.view.default(convert_element_type_222, [sym_size_int, 1500, 1]);  convert_element_type_222 = None
	        _assert_tensor_metadata_338 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_223, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_338 = None
	        convert_element_type_224 = torch.ops.prims.convert_element_type.default(convert_element_type_223, torch.float32);  convert_element_type_223 = None
	        _assert_tensor_metadata_339 = torch.ops.aten._assert_tensor_metadata.default(view_587, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_339 = None
	        convert_element_type_225 = torch.ops.prims.convert_element_type.default(view_587, torch.float32);  view_587 = None
	        sub_1733 = torch.ops.aten.sub.Tensor(convert_element_type_224, convert_element_type_225);  convert_element_type_224 = convert_element_type_225 = None
	        mul_3670 = torch.ops.aten.mul.Tensor(sub_1733, view_586);  sub_1733 = view_586 = None
	        _assert_tensor_metadata_340 = torch.ops.aten._assert_tensor_metadata.default(mul_3670, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_340 = None
	        view_589 = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_590 = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_591 = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_341 = torch.ops.aten._assert_tensor_metadata.default(view_589, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_341 = None
	        convert_element_type_226 = torch.ops.prims.convert_element_type.default(view_589, torch.float32);  view_589 = None
	        _assert_tensor_metadata_342 = torch.ops.aten._assert_tensor_metadata.default(view_591, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_342 = None
	        convert_element_type_227 = torch.ops.prims.convert_element_type.default(view_591, torch.float32);  view_591 = None
	        sub_1737 = torch.ops.aten.sub.Tensor(convert_element_type_226, convert_element_type_227);  convert_element_type_226 = convert_element_type_227 = None
	        mul_3675 = torch.ops.aten.mul.Tensor(sub_1737, view_590);  sub_1737 = view_590 = None
	        view_592 = torch.ops.aten.view.default(mul_3675, [1280, 1280]);  mul_3675 = None
	        _assert_tensor_metadata_343 = torch.ops.aten._assert_tensor_metadata.default(view_592, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_343 = None
	        permute_63 = torch.ops.aten.permute.default(view_592, [1, 0]);  view_592 = None
	        mul_3678 = sym_size_int * 1500
	        view_593 = torch.ops.aten.view.default(mul_3670, [mul_3678, 1280]);  mul_3670 = mul_3678 = None
	        mm_6 = torch.ops.aten.mm.default(view_593, permute_63);  view_593 = permute_63 = None
	        view_594 = torch.ops.aten.view.default(mm_6, [sym_size_int, 1500, 1280]);  mm_6 = None
	        view_595 = torch.ops.aten.view.default(view_594, [sym_size_int, -1, 20, 64]);  view_594 = None
	        permute_64 = torch.ops.aten.permute.default(view_595, [0, 2, 1, 3]);  view_595 = None
	        clone_51 = torch.ops.aten.clone.default(permute_64, memory_format = torch.contiguous_format);  permute_64 = None
	        amin_38 = torch.ops.aten.amin.default(add_5540, [2])
	        amax_38 = torch.ops.aten.amax.default(add_5540, [2])
	        full_76 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_38 = torch.ops.aten.minimum.default(amin_38, full_76);  amin_38 = full_76 = None
	        full_77 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_38 = torch.ops.aten.maximum.default(amax_38, full_77);  amax_38 = full_77 = None
	        sub_1751 = torch.ops.aten.sub.Tensor(maximum_38, minimum_38);  maximum_38 = None
	        div_76 = torch.ops.aten.div.Tensor(sub_1751, 255.0);  sub_1751 = None
	        clamp_min_114 = torch.ops.aten.clamp_min.default(div_76, 1.1920928955078125e-07);  div_76 = None
	        div_77 = torch.ops.aten.div.Tensor(minimum_38, clamp_min_114);  minimum_38 = None
	        round_77 = torch.ops.aten.round.default(div_77);  div_77 = None
	        sub_1757 = torch.ops.aten.sub.Tensor(-128, round_77);  round_77 = None
	        clamp_min_115 = torch.ops.aten.clamp_min.default(sub_1757, -128);  sub_1757 = None
	        clamp_max_76 = torch.ops.aten.clamp_max.default(clamp_min_115, 127);  clamp_min_115 = None
	        _assert_tensor_metadata_344 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_114, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_344 = None
	        _assert_tensor_metadata_345 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_76, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_345 = None
	        convert_element_type_228 = torch.ops.prims.convert_element_type.default(clamp_max_76, torch.int8);  clamp_max_76 = None
	        view_598 = torch.ops.aten.view.default(clamp_min_114, [sym_size_int, 1500, 1])
	        view_599 = torch.ops.aten.view.default(convert_element_type_228, [sym_size_int, 1500, 1])
	        reciprocal_38 = torch.ops.aten.reciprocal.default(view_598);  view_598 = None
	        mul_3744 = torch.ops.aten.mul.Tensor(reciprocal_38, 1.0);  reciprocal_38 = None
	        mul_3747 = torch.ops.aten.mul.Tensor(add_5540, mul_3744);  add_5540 = mul_3744 = None
	        round_78 = torch.ops.aten.round.default(mul_3747);  mul_3747 = None
	        add_5927 = torch.ops.aten.add.Tensor(round_78, view_599);  round_78 = view_599 = None
	        clamp_min_116 = torch.ops.aten.clamp_min.default(add_5927, -128);  add_5927 = None
	        clamp_max_77 = torch.ops.aten.clamp_max.default(clamp_min_116, 127);  clamp_min_116 = None
	        _assert_tensor_metadata_346 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_77, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_346 = None
	        convert_element_type_229 = torch.ops.prims.convert_element_type.default(clamp_max_77, torch.int8);  clamp_max_77 = None
	        view_602 = torch.ops.aten.view.default(clamp_min_114, [sym_size_int, 1500, 1]);  clamp_min_114 = None
	        view_603 = torch.ops.aten.view.default(convert_element_type_228, [sym_size_int, 1500, 1]);  convert_element_type_228 = None
	        _assert_tensor_metadata_347 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_229, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_347 = None
	        convert_element_type_230 = torch.ops.prims.convert_element_type.default(convert_element_type_229, torch.float32);  convert_element_type_229 = None
	        _assert_tensor_metadata_348 = torch.ops.aten._assert_tensor_metadata.default(view_603, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_348 = None
	        convert_element_type_231 = torch.ops.prims.convert_element_type.default(view_603, torch.float32);  view_603 = None
	        sub_1777 = torch.ops.aten.sub.Tensor(convert_element_type_230, convert_element_type_231);  convert_element_type_230 = convert_element_type_231 = None
	        mul_3769 = torch.ops.aten.mul.Tensor(sub_1777, view_602);  sub_1777 = view_602 = None
	        _assert_tensor_metadata_349 = torch.ops.aten._assert_tensor_metadata.default(mul_3769, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_349 = None
	        view_605 = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_606 = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_607 = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_350 = torch.ops.aten._assert_tensor_metadata.default(view_605, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_350 = None
	        convert_element_type_232 = torch.ops.prims.convert_element_type.default(view_605, torch.float32);  view_605 = None
	        _assert_tensor_metadata_351 = torch.ops.aten._assert_tensor_metadata.default(view_607, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_351 = None
	        convert_element_type_233 = torch.ops.prims.convert_element_type.default(view_607, torch.float32);  view_607 = None
	        sub_1781 = torch.ops.aten.sub.Tensor(convert_element_type_232, convert_element_type_233);  convert_element_type_232 = convert_element_type_233 = None
	        mul_3774 = torch.ops.aten.mul.Tensor(sub_1781, view_606);  sub_1781 = view_606 = None
	        view_608 = torch.ops.aten.view.default(mul_3774, [1280, 1280]);  mul_3774 = None
	        _assert_tensor_metadata_352 = torch.ops.aten._assert_tensor_metadata.default(view_608, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_352 = None
	        mul_3779 = sym_size_int * 1500
	        view_609 = torch.ops.aten.view.default(mul_3769, [mul_3779, 1280]);  mul_3769 = mul_3779 = None
	        permute_65 = torch.ops.aten.permute.default(view_608, [1, 0]);  view_608 = None
	        addmm_31 = torch.ops.aten.addmm.default(model_audio_tower_layers_6_self_attn_v_proj_bias, view_609, permute_65);  model_audio_tower_layers_6_self_attn_v_proj_bias = view_609 = permute_65 = None
	        view_610 = torch.ops.aten.view.default(addmm_31, [sym_size_int, 1500, 1280]);  addmm_31 = None
	        view_611 = torch.ops.aten.view.default(view_610, [sym_size_int, -1, 20, 64]);  view_610 = None
	        permute_66 = torch.ops.aten.permute.default(view_611, [0, 2, 1, 3]);  view_611 = None
	        clone_52 = torch.ops.aten.clone.default(permute_66, memory_format = torch.contiguous_format);  permute_66 = None
	        _scaled_dot_product_efficient_attention_6 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_50, clone_51, clone_52, None, False, scale = 1.0);  clone_50 = clone_51 = clone_52 = None
	        getitem_50 = _scaled_dot_product_efficient_attention_6[0];  _scaled_dot_product_efficient_attention_6 = None
	        permute_67 = torch.ops.aten.permute.default(getitem_50, [0, 2, 1, 3]);  getitem_50 = None
	        view_612 = torch.ops.aten.view.default(permute_67, [sym_size_int, 1500, -1]);  permute_67 = None
	        amin_39 = torch.ops.aten.amin.default(view_612, [2])
	        amax_39 = torch.ops.aten.amax.default(view_612, [2])
	        full_78 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_39 = torch.ops.aten.minimum.default(amin_39, full_78);  amin_39 = full_78 = None
	        full_79 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_39 = torch.ops.aten.maximum.default(amax_39, full_79);  amax_39 = full_79 = None
	        sub_1799 = torch.ops.aten.sub.Tensor(maximum_39, minimum_39);  maximum_39 = None
	        div_78 = torch.ops.aten.div.Tensor(sub_1799, 255.0);  sub_1799 = None
	        clamp_min_117 = torch.ops.aten.clamp_min.default(div_78, 1.1920928955078125e-07);  div_78 = None
	        div_79 = torch.ops.aten.div.Tensor(minimum_39, clamp_min_117);  minimum_39 = None
	        round_79 = torch.ops.aten.round.default(div_79);  div_79 = None
	        sub_1805 = torch.ops.aten.sub.Tensor(-128, round_79);  round_79 = None
	        clamp_min_118 = torch.ops.aten.clamp_min.default(sub_1805, -128);  sub_1805 = None
	        clamp_max_78 = torch.ops.aten.clamp_max.default(clamp_min_118, 127);  clamp_min_118 = None
	        _assert_tensor_metadata_353 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_117, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_353 = None
	        _assert_tensor_metadata_354 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_78, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_354 = None
	        convert_element_type_234 = torch.ops.prims.convert_element_type.default(clamp_max_78, torch.int8);  clamp_max_78 = None
	        view_615 = torch.ops.aten.view.default(clamp_min_117, [sym_size_int, 1500, 1])
	        view_616 = torch.ops.aten.view.default(convert_element_type_234, [sym_size_int, 1500, 1])
	        reciprocal_39 = torch.ops.aten.reciprocal.default(view_615);  view_615 = None
	        mul_3849 = torch.ops.aten.mul.Tensor(reciprocal_39, 1.0);  reciprocal_39 = None
	        mul_3852 = torch.ops.aten.mul.Tensor(view_612, mul_3849);  view_612 = mul_3849 = None
	        round_80 = torch.ops.aten.round.default(mul_3852);  mul_3852 = None
	        add_6091 = torch.ops.aten.add.Tensor(round_80, view_616);  round_80 = view_616 = None
	        clamp_min_119 = torch.ops.aten.clamp_min.default(add_6091, -128);  add_6091 = None
	        clamp_max_79 = torch.ops.aten.clamp_max.default(clamp_min_119, 127);  clamp_min_119 = None
	        _assert_tensor_metadata_355 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_79, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_355 = None
	        convert_element_type_235 = torch.ops.prims.convert_element_type.default(clamp_max_79, torch.int8);  clamp_max_79 = None
	        view_619 = torch.ops.aten.view.default(clamp_min_117, [sym_size_int, 1500, 1]);  clamp_min_117 = None
	        view_620 = torch.ops.aten.view.default(convert_element_type_234, [sym_size_int, 1500, 1]);  convert_element_type_234 = None
	        _assert_tensor_metadata_356 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_235, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_356 = None
	        convert_element_type_236 = torch.ops.prims.convert_element_type.default(convert_element_type_235, torch.float32);  convert_element_type_235 = None
	        _assert_tensor_metadata_357 = torch.ops.aten._assert_tensor_metadata.default(view_620, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_357 = None
	        convert_element_type_237 = torch.ops.prims.convert_element_type.default(view_620, torch.float32);  view_620 = None
	        sub_1825 = torch.ops.aten.sub.Tensor(convert_element_type_236, convert_element_type_237);  convert_element_type_236 = convert_element_type_237 = None
	        mul_3874 = torch.ops.aten.mul.Tensor(sub_1825, view_619);  sub_1825 = view_619 = None
	        _assert_tensor_metadata_358 = torch.ops.aten._assert_tensor_metadata.default(mul_3874, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_358 = None
	        view_622 = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_623 = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_624 = torch.ops.aten.view.default(model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_359 = torch.ops.aten._assert_tensor_metadata.default(view_622, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_359 = None
	        convert_element_type_238 = torch.ops.prims.convert_element_type.default(view_622, torch.float32);  view_622 = None
	        _assert_tensor_metadata_360 = torch.ops.aten._assert_tensor_metadata.default(view_624, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_360 = None
	        convert_element_type_239 = torch.ops.prims.convert_element_type.default(view_624, torch.float32);  view_624 = None
	        sub_1829 = torch.ops.aten.sub.Tensor(convert_element_type_238, convert_element_type_239);  convert_element_type_238 = convert_element_type_239 = None
	        mul_3879 = torch.ops.aten.mul.Tensor(sub_1829, view_623);  sub_1829 = view_623 = None
	        view_625 = torch.ops.aten.view.default(mul_3879, [1280, 1280]);  mul_3879 = None
	        _assert_tensor_metadata_361 = torch.ops.aten._assert_tensor_metadata.default(view_625, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_361 = None
	        mul_3884 = sym_size_int * 1500
	        view_626 = torch.ops.aten.view.default(mul_3874, [mul_3884, 1280]);  mul_3874 = mul_3884 = None
	        permute_68 = torch.ops.aten.permute.default(view_625, [1, 0]);  view_625 = None
	        addmm_32 = torch.ops.aten.addmm.default(model_audio_tower_layers_6_self_attn_out_proj_bias, view_626, permute_68);  model_audio_tower_layers_6_self_attn_out_proj_bias = view_626 = permute_68 = None
	        view_627 = torch.ops.aten.view.default(addmm_32, [sym_size_int, 1500, 1280]);  addmm_32 = None
	        add_6154 = torch.ops.aten.add.Tensor(add_5534, view_627);  add_5534 = view_627 = None
	        clone_54 = torch.ops.aten.clone.default(add_6154, memory_format = torch.contiguous_format)
	        var_mean_13 = torch.ops.aten.var_mean.correction(clone_54, [2], correction = 0, keepdim = True)
	        getitem_54 = var_mean_13[0]
	        getitem_55 = var_mean_13[1];  var_mean_13 = None
	        add_6159 = torch.ops.aten.add.Tensor(getitem_54, 1e-05);  getitem_54 = None
	        rsqrt_13 = torch.ops.aten.rsqrt.default(add_6159);  add_6159 = None
	        sub_1835 = torch.ops.aten.sub.Tensor(clone_54, getitem_55);  clone_54 = getitem_55 = None
	        mul_3895 = torch.ops.aten.mul.Tensor(sub_1835, rsqrt_13);  sub_1835 = rsqrt_13 = None
	        mul_3896 = torch.ops.aten.mul.Tensor(mul_3895, model_audio_tower_layers_6_final_layer_norm_weight);  mul_3895 = model_audio_tower_layers_6_final_layer_norm_weight = None
	        add_6160 = torch.ops.aten.add.Tensor(mul_3896, model_audio_tower_layers_6_final_layer_norm_bias);  mul_3896 = model_audio_tower_layers_6_final_layer_norm_bias = None
	        amin_40 = torch.ops.aten.amin.default(add_6160, [2])
	        amax_40 = torch.ops.aten.amax.default(add_6160, [2])
	        full_80 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_40 = torch.ops.aten.minimum.default(amin_40, full_80);  amin_40 = full_80 = None
	        full_81 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_40 = torch.ops.aten.maximum.default(amax_40, full_81);  amax_40 = full_81 = None
	        sub_1846 = torch.ops.aten.sub.Tensor(maximum_40, minimum_40);  maximum_40 = None
	        div_80 = torch.ops.aten.div.Tensor(sub_1846, 255.0);  sub_1846 = None
	        clamp_min_120 = torch.ops.aten.clamp_min.default(div_80, 1.1920928955078125e-07);  div_80 = None
	        div_81 = torch.ops.aten.div.Tensor(minimum_40, clamp_min_120);  minimum_40 = None
	        round_81 = torch.ops.aten.round.default(div_81);  div_81 = None
	        sub_1852 = torch.ops.aten.sub.Tensor(-128, round_81);  round_81 = None
	        clamp_min_121 = torch.ops.aten.clamp_min.default(sub_1852, -128);  sub_1852 = None
	        clamp_max_80 = torch.ops.aten.clamp_max.default(clamp_min_121, 127);  clamp_min_121 = None
	        _assert_tensor_metadata_362 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_120, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_362 = None
	        _assert_tensor_metadata_363 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_80, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_363 = None
	        convert_element_type_240 = torch.ops.prims.convert_element_type.default(clamp_max_80, torch.int8);  clamp_max_80 = None
	        view_630 = torch.ops.aten.view.default(clamp_min_120, [sym_size_int, 1500, 1])
	        view_631 = torch.ops.aten.view.default(convert_element_type_240, [sym_size_int, 1500, 1])
	        reciprocal_40 = torch.ops.aten.reciprocal.default(view_630);  view_630 = None
	        mul_3944 = torch.ops.aten.mul.Tensor(reciprocal_40, 1.0);  reciprocal_40 = None
	        mul_3947 = torch.ops.aten.mul.Tensor(add_6160, mul_3944);  add_6160 = mul_3944 = None
	        round_82 = torch.ops.aten.round.default(mul_3947);  mul_3947 = None
	        add_6247 = torch.ops.aten.add.Tensor(round_82, view_631);  round_82 = view_631 = None
	        clamp_min_122 = torch.ops.aten.clamp_min.default(add_6247, -128);  add_6247 = None
	        clamp_max_81 = torch.ops.aten.clamp_max.default(clamp_min_122, 127);  clamp_min_122 = None
	        _assert_tensor_metadata_364 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_81, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_364 = None
	        convert_element_type_241 = torch.ops.prims.convert_element_type.default(clamp_max_81, torch.int8);  clamp_max_81 = None
	        view_634 = torch.ops.aten.view.default(clamp_min_120, [sym_size_int, 1500, 1]);  clamp_min_120 = None
	        view_635 = torch.ops.aten.view.default(convert_element_type_240, [sym_size_int, 1500, 1]);  convert_element_type_240 = None
	        _assert_tensor_metadata_365 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_241, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_365 = None
	        convert_element_type_242 = torch.ops.prims.convert_element_type.default(convert_element_type_241, torch.float32);  convert_element_type_241 = None
	        _assert_tensor_metadata_366 = torch.ops.aten._assert_tensor_metadata.default(view_635, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_366 = None
	        convert_element_type_243 = torch.ops.prims.convert_element_type.default(view_635, torch.float32);  view_635 = None
	        sub_1872 = torch.ops.aten.sub.Tensor(convert_element_type_242, convert_element_type_243);  convert_element_type_242 = convert_element_type_243 = None
	        mul_3969 = torch.ops.aten.mul.Tensor(sub_1872, view_634);  sub_1872 = view_634 = None
	        _assert_tensor_metadata_367 = torch.ops.aten._assert_tensor_metadata.default(mul_3969, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_367 = None
	        view_637 = torch.ops.aten.view.default(model_audio_tower_layers_6_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_6_fc1_parametrizations_weight_original0 = None
	        view_638 = torch.ops.aten.view.default(model_audio_tower_layers_6_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_6_fc1_parametrizations_weight_original1 = None
	        view_639 = torch.ops.aten.view.default(model_audio_tower_layers_6_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_6_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_368 = torch.ops.aten._assert_tensor_metadata.default(view_637, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_368 = None
	        convert_element_type_244 = torch.ops.prims.convert_element_type.default(view_637, torch.float32);  view_637 = None
	        _assert_tensor_metadata_369 = torch.ops.aten._assert_tensor_metadata.default(view_639, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_369 = None
	        convert_element_type_245 = torch.ops.prims.convert_element_type.default(view_639, torch.float32);  view_639 = None
	        sub_1876 = torch.ops.aten.sub.Tensor(convert_element_type_244, convert_element_type_245);  convert_element_type_244 = convert_element_type_245 = None
	        mul_3974 = torch.ops.aten.mul.Tensor(sub_1876, view_638);  sub_1876 = view_638 = None
	        view_640 = torch.ops.aten.view.default(mul_3974, [5120, 1280]);  mul_3974 = None
	        _assert_tensor_metadata_370 = torch.ops.aten._assert_tensor_metadata.default(view_640, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_370 = None
	        mul_3979 = sym_size_int * 1500
	        view_641 = torch.ops.aten.view.default(mul_3969, [mul_3979, 1280]);  mul_3969 = mul_3979 = None
	        permute_69 = torch.ops.aten.permute.default(view_640, [1, 0]);  view_640 = None
	        addmm_33 = torch.ops.aten.addmm.default(model_audio_tower_layers_6_fc1_bias, view_641, permute_69);  model_audio_tower_layers_6_fc1_bias = view_641 = permute_69 = None
	        view_642 = torch.ops.aten.view.default(addmm_33, [sym_size_int, 1500, 5120]);  addmm_33 = None
	        mul_3986 = torch.ops.aten.mul.Tensor(view_642, 0.5)
	        mul_3987 = torch.ops.aten.mul.Tensor(view_642, 0.7071067811865476);  view_642 = None
	        erf_8 = torch.ops.aten.erf.default(mul_3987);  mul_3987 = None
	        add_6306 = torch.ops.aten.add.Tensor(erf_8, 1);  erf_8 = None
	        mul_3988 = torch.ops.aten.mul.Tensor(mul_3986, add_6306);  mul_3986 = add_6306 = None
	        amin_41 = torch.ops.aten.amin.default(mul_3988, [2])
	        amax_41 = torch.ops.aten.amax.default(mul_3988, [2])
	        full_82 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_41 = torch.ops.aten.minimum.default(amin_41, full_82);  amin_41 = full_82 = None
	        full_83 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_41 = torch.ops.aten.maximum.default(amax_41, full_83);  amax_41 = full_83 = None
	        sub_1889 = torch.ops.aten.sub.Tensor(maximum_41, minimum_41);  maximum_41 = None
	        div_82 = torch.ops.aten.div.Tensor(sub_1889, 255.0);  sub_1889 = None
	        clamp_min_123 = torch.ops.aten.clamp_min.default(div_82, 1.1920928955078125e-07);  div_82 = None
	        div_83 = torch.ops.aten.div.Tensor(minimum_41, clamp_min_123);  minimum_41 = None
	        round_83 = torch.ops.aten.round.default(div_83);  div_83 = None
	        sub_1895 = torch.ops.aten.sub.Tensor(-128, round_83);  round_83 = None
	        clamp_min_124 = torch.ops.aten.clamp_min.default(sub_1895, -128);  sub_1895 = None
	        clamp_max_82 = torch.ops.aten.clamp_max.default(clamp_min_124, 127);  clamp_min_124 = None
	        _assert_tensor_metadata_371 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_123, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_371 = None
	        _assert_tensor_metadata_372 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_82, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_372 = None
	        convert_element_type_246 = torch.ops.prims.convert_element_type.default(clamp_max_82, torch.int8);  clamp_max_82 = None
	        view_645 = torch.ops.aten.view.default(clamp_min_123, [sym_size_int, 1500, 1])
	        view_646 = torch.ops.aten.view.default(convert_element_type_246, [sym_size_int, 1500, 1])
	        reciprocal_41 = torch.ops.aten.reciprocal.default(view_645);  view_645 = None
	        mul_4034 = torch.ops.aten.mul.Tensor(reciprocal_41, 1.0);  reciprocal_41 = None
	        mul_4037 = torch.ops.aten.mul.Tensor(mul_3988, mul_4034);  mul_3988 = mul_4034 = None
	        round_84 = torch.ops.aten.round.default(mul_4037);  mul_4037 = None
	        add_6389 = torch.ops.aten.add.Tensor(round_84, view_646);  round_84 = view_646 = None
	        clamp_min_125 = torch.ops.aten.clamp_min.default(add_6389, -128);  add_6389 = None
	        clamp_max_83 = torch.ops.aten.clamp_max.default(clamp_min_125, 127);  clamp_min_125 = None
	        _assert_tensor_metadata_373 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_83, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_373 = None
	        convert_element_type_247 = torch.ops.prims.convert_element_type.default(clamp_max_83, torch.int8);  clamp_max_83 = None
	        view_649 = torch.ops.aten.view.default(clamp_min_123, [sym_size_int, 1500, 1]);  clamp_min_123 = None
	        view_650 = torch.ops.aten.view.default(convert_element_type_246, [sym_size_int, 1500, 1]);  convert_element_type_246 = None
	        _assert_tensor_metadata_374 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_247, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_374 = None
	        convert_element_type_248 = torch.ops.prims.convert_element_type.default(convert_element_type_247, torch.float32);  convert_element_type_247 = None
	        _assert_tensor_metadata_375 = torch.ops.aten._assert_tensor_metadata.default(view_650, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_375 = None
	        convert_element_type_249 = torch.ops.prims.convert_element_type.default(view_650, torch.float32);  view_650 = None
	        sub_1915 = torch.ops.aten.sub.Tensor(convert_element_type_248, convert_element_type_249);  convert_element_type_248 = convert_element_type_249 = None
	        mul_4059 = torch.ops.aten.mul.Tensor(sub_1915, view_649);  sub_1915 = view_649 = None
	        _assert_tensor_metadata_376 = torch.ops.aten._assert_tensor_metadata.default(mul_4059, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_376 = None
	        view_652 = torch.ops.aten.view.default(model_audio_tower_layers_6_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_6_fc2_parametrizations_weight_original0 = None
	        view_653 = torch.ops.aten.view.default(model_audio_tower_layers_6_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_6_fc2_parametrizations_weight_original1 = None
	        view_654 = torch.ops.aten.view.default(model_audio_tower_layers_6_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_6_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_377 = torch.ops.aten._assert_tensor_metadata.default(view_652, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_377 = None
	        convert_element_type_250 = torch.ops.prims.convert_element_type.default(view_652, torch.float32);  view_652 = None
	        _assert_tensor_metadata_378 = torch.ops.aten._assert_tensor_metadata.default(view_654, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_378 = None
	        convert_element_type_251 = torch.ops.prims.convert_element_type.default(view_654, torch.float32);  view_654 = None
	        sub_1919 = torch.ops.aten.sub.Tensor(convert_element_type_250, convert_element_type_251);  convert_element_type_250 = convert_element_type_251 = None
	        mul_4064 = torch.ops.aten.mul.Tensor(sub_1919, view_653);  sub_1919 = view_653 = None
	        view_655 = torch.ops.aten.view.default(mul_4064, [1280, 5120]);  mul_4064 = None
	        _assert_tensor_metadata_379 = torch.ops.aten._assert_tensor_metadata.default(view_655, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_379 = None
	        mul_4069 = sym_size_int * 1500
	        view_656 = torch.ops.aten.view.default(mul_4059, [mul_4069, 5120]);  mul_4059 = mul_4069 = None
	        permute_70 = torch.ops.aten.permute.default(view_655, [1, 0]);  view_655 = None
	        addmm_34 = torch.ops.aten.addmm.default(model_audio_tower_layers_6_fc2_bias, view_656, permute_70);  model_audio_tower_layers_6_fc2_bias = view_656 = permute_70 = None
	        view_657 = torch.ops.aten.view.default(addmm_34, [sym_size_int, 1500, 1280]);  addmm_34 = None
	        add_6452 = torch.ops.aten.add.Tensor(add_6154, view_657);  add_6154 = view_657 = None
	        clone_57 = torch.ops.aten.clone.default(add_6452, memory_format = torch.contiguous_format)
	        var_mean_14 = torch.ops.aten.var_mean.correction(clone_57, [2], correction = 0, keepdim = True)
	        getitem_56 = var_mean_14[0]
	        getitem_57 = var_mean_14[1];  var_mean_14 = None
	        add_6457 = torch.ops.aten.add.Tensor(getitem_56, 1e-05);  getitem_56 = None
	        rsqrt_14 = torch.ops.aten.rsqrt.default(add_6457);  add_6457 = None
	        sub_1925 = torch.ops.aten.sub.Tensor(clone_57, getitem_57);  clone_57 = getitem_57 = None
	        mul_4080 = torch.ops.aten.mul.Tensor(sub_1925, rsqrt_14);  sub_1925 = rsqrt_14 = None
	        mul_4081 = torch.ops.aten.mul.Tensor(mul_4080, model_audio_tower_layers_7_self_attn_layer_norm_weight);  mul_4080 = model_audio_tower_layers_7_self_attn_layer_norm_weight = None
	        add_6458 = torch.ops.aten.add.Tensor(mul_4081, model_audio_tower_layers_7_self_attn_layer_norm_bias);  mul_4081 = model_audio_tower_layers_7_self_attn_layer_norm_bias = None
	        amin_42 = torch.ops.aten.amin.default(add_6458, [2])
	        amax_42 = torch.ops.aten.amax.default(add_6458, [2])
	        full_84 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_42 = torch.ops.aten.minimum.default(amin_42, full_84);  amin_42 = full_84 = None
	        full_85 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_42 = torch.ops.aten.maximum.default(amax_42, full_85);  amax_42 = full_85 = None
	        sub_1936 = torch.ops.aten.sub.Tensor(maximum_42, minimum_42);  maximum_42 = None
	        div_84 = torch.ops.aten.div.Tensor(sub_1936, 255.0);  sub_1936 = None
	        clamp_min_126 = torch.ops.aten.clamp_min.default(div_84, 1.1920928955078125e-07);  div_84 = None
	        div_85 = torch.ops.aten.div.Tensor(minimum_42, clamp_min_126);  minimum_42 = None
	        round_85 = torch.ops.aten.round.default(div_85);  div_85 = None
	        sub_1942 = torch.ops.aten.sub.Tensor(-128, round_85);  round_85 = None
	        clamp_min_127 = torch.ops.aten.clamp_min.default(sub_1942, -128);  sub_1942 = None
	        clamp_max_84 = torch.ops.aten.clamp_max.default(clamp_min_127, 127);  clamp_min_127 = None
	        _assert_tensor_metadata_380 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_126, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_380 = None
	        _assert_tensor_metadata_381 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_84, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_381 = None
	        convert_element_type_252 = torch.ops.prims.convert_element_type.default(clamp_max_84, torch.int8);  clamp_max_84 = None
	        view_660 = torch.ops.aten.view.default(clamp_min_126, [sym_size_int, 1500, 1])
	        view_661 = torch.ops.aten.view.default(convert_element_type_252, [sym_size_int, 1500, 1])
	        reciprocal_42 = torch.ops.aten.reciprocal.default(view_660);  view_660 = None
	        mul_4129 = torch.ops.aten.mul.Tensor(reciprocal_42, 1.0);  reciprocal_42 = None
	        mul_4132 = torch.ops.aten.mul.Tensor(add_6458, mul_4129);  mul_4129 = None
	        round_86 = torch.ops.aten.round.default(mul_4132);  mul_4132 = None
	        add_6545 = torch.ops.aten.add.Tensor(round_86, view_661);  round_86 = view_661 = None
	        clamp_min_128 = torch.ops.aten.clamp_min.default(add_6545, -128);  add_6545 = None
	        clamp_max_85 = torch.ops.aten.clamp_max.default(clamp_min_128, 127);  clamp_min_128 = None
	        _assert_tensor_metadata_382 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_85, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_382 = None
	        convert_element_type_253 = torch.ops.prims.convert_element_type.default(clamp_max_85, torch.int8);  clamp_max_85 = None
	        view_664 = torch.ops.aten.view.default(clamp_min_126, [sym_size_int, 1500, 1]);  clamp_min_126 = None
	        view_665 = torch.ops.aten.view.default(convert_element_type_252, [sym_size_int, 1500, 1]);  convert_element_type_252 = None
	        _assert_tensor_metadata_383 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_253, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_383 = None
	        convert_element_type_254 = torch.ops.prims.convert_element_type.default(convert_element_type_253, torch.float32);  convert_element_type_253 = None
	        _assert_tensor_metadata_384 = torch.ops.aten._assert_tensor_metadata.default(view_665, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_384 = None
	        convert_element_type_255 = torch.ops.prims.convert_element_type.default(view_665, torch.float32);  view_665 = None
	        sub_1962 = torch.ops.aten.sub.Tensor(convert_element_type_254, convert_element_type_255);  convert_element_type_254 = convert_element_type_255 = None
	        mul_4154 = torch.ops.aten.mul.Tensor(sub_1962, view_664);  sub_1962 = view_664 = None
	        _assert_tensor_metadata_385 = torch.ops.aten._assert_tensor_metadata.default(mul_4154, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_385 = None
	        view_667 = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_668 = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_669 = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_386 = torch.ops.aten._assert_tensor_metadata.default(view_667, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_386 = None
	        convert_element_type_256 = torch.ops.prims.convert_element_type.default(view_667, torch.float32);  view_667 = None
	        _assert_tensor_metadata_387 = torch.ops.aten._assert_tensor_metadata.default(view_669, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_387 = None
	        convert_element_type_257 = torch.ops.prims.convert_element_type.default(view_669, torch.float32);  view_669 = None
	        sub_1966 = torch.ops.aten.sub.Tensor(convert_element_type_256, convert_element_type_257);  convert_element_type_256 = convert_element_type_257 = None
	        mul_4159 = torch.ops.aten.mul.Tensor(sub_1966, view_668);  sub_1966 = view_668 = None
	        view_670 = torch.ops.aten.view.default(mul_4159, [1280, 1280]);  mul_4159 = None
	        _assert_tensor_metadata_388 = torch.ops.aten._assert_tensor_metadata.default(view_670, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_388 = None
	        mul_4164 = sym_size_int * 1500
	        view_671 = torch.ops.aten.view.default(mul_4154, [mul_4164, 1280]);  mul_4154 = mul_4164 = None
	        permute_71 = torch.ops.aten.permute.default(view_670, [1, 0]);  view_670 = None
	        addmm_35 = torch.ops.aten.addmm.default(model_audio_tower_layers_7_self_attn_q_proj_bias, view_671, permute_71);  model_audio_tower_layers_7_self_attn_q_proj_bias = view_671 = permute_71 = None
	        view_672 = torch.ops.aten.view.default(addmm_35, [sym_size_int, 1500, 1280]);  addmm_35 = None
	        mul_4171 = torch.ops.aten.mul.Tensor(view_672, 0.125);  view_672 = None
	        view_673 = torch.ops.aten.view.default(mul_4171, [sym_size_int, 1500, 20, 64]);  mul_4171 = None
	        permute_72 = torch.ops.aten.permute.default(view_673, [0, 2, 1, 3]);  view_673 = None
	        clone_58 = torch.ops.aten.clone.default(permute_72, memory_format = torch.contiguous_format);  permute_72 = None
	        amin_43 = torch.ops.aten.amin.default(add_6458, [2])
	        amax_43 = torch.ops.aten.amax.default(add_6458, [2])
	        full_86 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_43 = torch.ops.aten.minimum.default(amin_43, full_86);  amin_43 = full_86 = None
	        full_87 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_43 = torch.ops.aten.maximum.default(amax_43, full_87);  amax_43 = full_87 = None
	        sub_1981 = torch.ops.aten.sub.Tensor(maximum_43, minimum_43);  maximum_43 = None
	        div_86 = torch.ops.aten.div.Tensor(sub_1981, 255.0);  sub_1981 = None
	        clamp_min_129 = torch.ops.aten.clamp_min.default(div_86, 1.1920928955078125e-07);  div_86 = None
	        div_87 = torch.ops.aten.div.Tensor(minimum_43, clamp_min_129);  minimum_43 = None
	        round_87 = torch.ops.aten.round.default(div_87);  div_87 = None
	        sub_1987 = torch.ops.aten.sub.Tensor(-128, round_87);  round_87 = None
	        clamp_min_130 = torch.ops.aten.clamp_min.default(sub_1987, -128);  sub_1987 = None
	        clamp_max_86 = torch.ops.aten.clamp_max.default(clamp_min_130, 127);  clamp_min_130 = None
	        _assert_tensor_metadata_389 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_129, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_389 = None
	        _assert_tensor_metadata_390 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_86, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_390 = None
	        convert_element_type_258 = torch.ops.prims.convert_element_type.default(clamp_max_86, torch.int8);  clamp_max_86 = None
	        view_676 = torch.ops.aten.view.default(clamp_min_129, [sym_size_int, 1500, 1])
	        view_677 = torch.ops.aten.view.default(convert_element_type_258, [sym_size_int, 1500, 1])
	        reciprocal_43 = torch.ops.aten.reciprocal.default(view_676);  view_676 = None
	        mul_4225 = torch.ops.aten.mul.Tensor(reciprocal_43, 1.0);  reciprocal_43 = None
	        mul_4228 = torch.ops.aten.mul.Tensor(add_6458, mul_4225);  mul_4225 = None
	        round_88 = torch.ops.aten.round.default(mul_4228);  mul_4228 = None
	        add_6697 = torch.ops.aten.add.Tensor(round_88, view_677);  round_88 = view_677 = None
	        clamp_min_131 = torch.ops.aten.clamp_min.default(add_6697, -128);  add_6697 = None
	        clamp_max_87 = torch.ops.aten.clamp_max.default(clamp_min_131, 127);  clamp_min_131 = None
	        _assert_tensor_metadata_391 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_87, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_391 = None
	        convert_element_type_259 = torch.ops.prims.convert_element_type.default(clamp_max_87, torch.int8);  clamp_max_87 = None
	        view_680 = torch.ops.aten.view.default(clamp_min_129, [sym_size_int, 1500, 1]);  clamp_min_129 = None
	        view_681 = torch.ops.aten.view.default(convert_element_type_258, [sym_size_int, 1500, 1]);  convert_element_type_258 = None
	        _assert_tensor_metadata_392 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_259, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_392 = None
	        convert_element_type_260 = torch.ops.prims.convert_element_type.default(convert_element_type_259, torch.float32);  convert_element_type_259 = None
	        _assert_tensor_metadata_393 = torch.ops.aten._assert_tensor_metadata.default(view_681, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_393 = None
	        convert_element_type_261 = torch.ops.prims.convert_element_type.default(view_681, torch.float32);  view_681 = None
	        sub_2007 = torch.ops.aten.sub.Tensor(convert_element_type_260, convert_element_type_261);  convert_element_type_260 = convert_element_type_261 = None
	        mul_4250 = torch.ops.aten.mul.Tensor(sub_2007, view_680);  sub_2007 = view_680 = None
	        _assert_tensor_metadata_394 = torch.ops.aten._assert_tensor_metadata.default(mul_4250, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_394 = None
	        view_683 = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_684 = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_685 = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_395 = torch.ops.aten._assert_tensor_metadata.default(view_683, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_395 = None
	        convert_element_type_262 = torch.ops.prims.convert_element_type.default(view_683, torch.float32);  view_683 = None
	        _assert_tensor_metadata_396 = torch.ops.aten._assert_tensor_metadata.default(view_685, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_396 = None
	        convert_element_type_263 = torch.ops.prims.convert_element_type.default(view_685, torch.float32);  view_685 = None
	        sub_2011 = torch.ops.aten.sub.Tensor(convert_element_type_262, convert_element_type_263);  convert_element_type_262 = convert_element_type_263 = None
	        mul_4255 = torch.ops.aten.mul.Tensor(sub_2011, view_684);  sub_2011 = view_684 = None
	        view_686 = torch.ops.aten.view.default(mul_4255, [1280, 1280]);  mul_4255 = None
	        _assert_tensor_metadata_397 = torch.ops.aten._assert_tensor_metadata.default(view_686, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_397 = None
	        permute_73 = torch.ops.aten.permute.default(view_686, [1, 0]);  view_686 = None
	        mul_4258 = sym_size_int * 1500
	        view_687 = torch.ops.aten.view.default(mul_4250, [mul_4258, 1280]);  mul_4250 = mul_4258 = None
	        mm_7 = torch.ops.aten.mm.default(view_687, permute_73);  view_687 = permute_73 = None
	        view_688 = torch.ops.aten.view.default(mm_7, [sym_size_int, 1500, 1280]);  mm_7 = None
	        view_689 = torch.ops.aten.view.default(view_688, [sym_size_int, -1, 20, 64]);  view_688 = None
	        permute_74 = torch.ops.aten.permute.default(view_689, [0, 2, 1, 3]);  view_689 = None
	        clone_59 = torch.ops.aten.clone.default(permute_74, memory_format = torch.contiguous_format);  permute_74 = None
	        amin_44 = torch.ops.aten.amin.default(add_6458, [2])
	        amax_44 = torch.ops.aten.amax.default(add_6458, [2])
	        full_88 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_44 = torch.ops.aten.minimum.default(amin_44, full_88);  amin_44 = full_88 = None
	        full_89 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_44 = torch.ops.aten.maximum.default(amax_44, full_89);  amax_44 = full_89 = None
	        sub_2025 = torch.ops.aten.sub.Tensor(maximum_44, minimum_44);  maximum_44 = None
	        div_88 = torch.ops.aten.div.Tensor(sub_2025, 255.0);  sub_2025 = None
	        clamp_min_132 = torch.ops.aten.clamp_min.default(div_88, 1.1920928955078125e-07);  div_88 = None
	        div_89 = torch.ops.aten.div.Tensor(minimum_44, clamp_min_132);  minimum_44 = None
	        round_89 = torch.ops.aten.round.default(div_89);  div_89 = None
	        sub_2031 = torch.ops.aten.sub.Tensor(-128, round_89);  round_89 = None
	        clamp_min_133 = torch.ops.aten.clamp_min.default(sub_2031, -128);  sub_2031 = None
	        clamp_max_88 = torch.ops.aten.clamp_max.default(clamp_min_133, 127);  clamp_min_133 = None
	        _assert_tensor_metadata_398 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_132, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_398 = None
	        _assert_tensor_metadata_399 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_88, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_399 = None
	        convert_element_type_264 = torch.ops.prims.convert_element_type.default(clamp_max_88, torch.int8);  clamp_max_88 = None
	        view_692 = torch.ops.aten.view.default(clamp_min_132, [sym_size_int, 1500, 1])
	        view_693 = torch.ops.aten.view.default(convert_element_type_264, [sym_size_int, 1500, 1])
	        reciprocal_44 = torch.ops.aten.reciprocal.default(view_692);  view_692 = None
	        mul_4324 = torch.ops.aten.mul.Tensor(reciprocal_44, 1.0);  reciprocal_44 = None
	        mul_4327 = torch.ops.aten.mul.Tensor(add_6458, mul_4324);  add_6458 = mul_4324 = None
	        round_90 = torch.ops.aten.round.default(mul_4327);  mul_4327 = None
	        add_6845 = torch.ops.aten.add.Tensor(round_90, view_693);  round_90 = view_693 = None
	        clamp_min_134 = torch.ops.aten.clamp_min.default(add_6845, -128);  add_6845 = None
	        clamp_max_89 = torch.ops.aten.clamp_max.default(clamp_min_134, 127);  clamp_min_134 = None
	        _assert_tensor_metadata_400 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_89, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_400 = None
	        convert_element_type_265 = torch.ops.prims.convert_element_type.default(clamp_max_89, torch.int8);  clamp_max_89 = None
	        view_696 = torch.ops.aten.view.default(clamp_min_132, [sym_size_int, 1500, 1]);  clamp_min_132 = None
	        view_697 = torch.ops.aten.view.default(convert_element_type_264, [sym_size_int, 1500, 1]);  convert_element_type_264 = None
	        _assert_tensor_metadata_401 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_265, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_401 = None
	        convert_element_type_266 = torch.ops.prims.convert_element_type.default(convert_element_type_265, torch.float32);  convert_element_type_265 = None
	        _assert_tensor_metadata_402 = torch.ops.aten._assert_tensor_metadata.default(view_697, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_402 = None
	        convert_element_type_267 = torch.ops.prims.convert_element_type.default(view_697, torch.float32);  view_697 = None
	        sub_2051 = torch.ops.aten.sub.Tensor(convert_element_type_266, convert_element_type_267);  convert_element_type_266 = convert_element_type_267 = None
	        mul_4349 = torch.ops.aten.mul.Tensor(sub_2051, view_696);  sub_2051 = view_696 = None
	        _assert_tensor_metadata_403 = torch.ops.aten._assert_tensor_metadata.default(mul_4349, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_403 = None
	        view_699 = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_700 = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_701 = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_404 = torch.ops.aten._assert_tensor_metadata.default(view_699, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_404 = None
	        convert_element_type_268 = torch.ops.prims.convert_element_type.default(view_699, torch.float32);  view_699 = None
	        _assert_tensor_metadata_405 = torch.ops.aten._assert_tensor_metadata.default(view_701, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_405 = None
	        convert_element_type_269 = torch.ops.prims.convert_element_type.default(view_701, torch.float32);  view_701 = None
	        sub_2055 = torch.ops.aten.sub.Tensor(convert_element_type_268, convert_element_type_269);  convert_element_type_268 = convert_element_type_269 = None
	        mul_4354 = torch.ops.aten.mul.Tensor(sub_2055, view_700);  sub_2055 = view_700 = None
	        view_702 = torch.ops.aten.view.default(mul_4354, [1280, 1280]);  mul_4354 = None
	        _assert_tensor_metadata_406 = torch.ops.aten._assert_tensor_metadata.default(view_702, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_406 = None
	        mul_4359 = sym_size_int * 1500
	        view_703 = torch.ops.aten.view.default(mul_4349, [mul_4359, 1280]);  mul_4349 = mul_4359 = None
	        permute_75 = torch.ops.aten.permute.default(view_702, [1, 0]);  view_702 = None
	        addmm_36 = torch.ops.aten.addmm.default(model_audio_tower_layers_7_self_attn_v_proj_bias, view_703, permute_75);  model_audio_tower_layers_7_self_attn_v_proj_bias = view_703 = permute_75 = None
	        view_704 = torch.ops.aten.view.default(addmm_36, [sym_size_int, 1500, 1280]);  addmm_36 = None
	        view_705 = torch.ops.aten.view.default(view_704, [sym_size_int, -1, 20, 64]);  view_704 = None
	        permute_76 = torch.ops.aten.permute.default(view_705, [0, 2, 1, 3]);  view_705 = None
	        clone_60 = torch.ops.aten.clone.default(permute_76, memory_format = torch.contiguous_format);  permute_76 = None
	        _scaled_dot_product_efficient_attention_7 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_58, clone_59, clone_60, None, False, scale = 1.0);  clone_58 = clone_59 = clone_60 = None
	        getitem_58 = _scaled_dot_product_efficient_attention_7[0];  _scaled_dot_product_efficient_attention_7 = None
	        permute_77 = torch.ops.aten.permute.default(getitem_58, [0, 2, 1, 3]);  getitem_58 = None
	        view_706 = torch.ops.aten.view.default(permute_77, [sym_size_int, 1500, -1]);  permute_77 = None
	        amin_45 = torch.ops.aten.amin.default(view_706, [2])
	        amax_45 = torch.ops.aten.amax.default(view_706, [2])
	        full_90 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_45 = torch.ops.aten.minimum.default(amin_45, full_90);  amin_45 = full_90 = None
	        full_91 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_45 = torch.ops.aten.maximum.default(amax_45, full_91);  amax_45 = full_91 = None
	        sub_2073 = torch.ops.aten.sub.Tensor(maximum_45, minimum_45);  maximum_45 = None
	        div_90 = torch.ops.aten.div.Tensor(sub_2073, 255.0);  sub_2073 = None
	        clamp_min_135 = torch.ops.aten.clamp_min.default(div_90, 1.1920928955078125e-07);  div_90 = None
	        div_91 = torch.ops.aten.div.Tensor(minimum_45, clamp_min_135);  minimum_45 = None
	        round_91 = torch.ops.aten.round.default(div_91);  div_91 = None
	        sub_2079 = torch.ops.aten.sub.Tensor(-128, round_91);  round_91 = None
	        clamp_min_136 = torch.ops.aten.clamp_min.default(sub_2079, -128);  sub_2079 = None
	        clamp_max_90 = torch.ops.aten.clamp_max.default(clamp_min_136, 127);  clamp_min_136 = None
	        _assert_tensor_metadata_407 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_135, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_407 = None
	        _assert_tensor_metadata_408 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_90, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_408 = None
	        convert_element_type_270 = torch.ops.prims.convert_element_type.default(clamp_max_90, torch.int8);  clamp_max_90 = None
	        view_709 = torch.ops.aten.view.default(clamp_min_135, [sym_size_int, 1500, 1])
	        view_710 = torch.ops.aten.view.default(convert_element_type_270, [sym_size_int, 1500, 1])
	        reciprocal_45 = torch.ops.aten.reciprocal.default(view_709);  view_709 = None
	        mul_4429 = torch.ops.aten.mul.Tensor(reciprocal_45, 1.0);  reciprocal_45 = None
	        mul_4432 = torch.ops.aten.mul.Tensor(view_706, mul_4429);  view_706 = mul_4429 = None
	        round_92 = torch.ops.aten.round.default(mul_4432);  mul_4432 = None
	        add_7009 = torch.ops.aten.add.Tensor(round_92, view_710);  round_92 = view_710 = None
	        clamp_min_137 = torch.ops.aten.clamp_min.default(add_7009, -128);  add_7009 = None
	        clamp_max_91 = torch.ops.aten.clamp_max.default(clamp_min_137, 127);  clamp_min_137 = None
	        _assert_tensor_metadata_409 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_91, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_409 = None
	        convert_element_type_271 = torch.ops.prims.convert_element_type.default(clamp_max_91, torch.int8);  clamp_max_91 = None
	        view_713 = torch.ops.aten.view.default(clamp_min_135, [sym_size_int, 1500, 1]);  clamp_min_135 = None
	        view_714 = torch.ops.aten.view.default(convert_element_type_270, [sym_size_int, 1500, 1]);  convert_element_type_270 = None
	        _assert_tensor_metadata_410 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_271, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_410 = None
	        convert_element_type_272 = torch.ops.prims.convert_element_type.default(convert_element_type_271, torch.float32);  convert_element_type_271 = None
	        _assert_tensor_metadata_411 = torch.ops.aten._assert_tensor_metadata.default(view_714, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_411 = None
	        convert_element_type_273 = torch.ops.prims.convert_element_type.default(view_714, torch.float32);  view_714 = None
	        sub_2099 = torch.ops.aten.sub.Tensor(convert_element_type_272, convert_element_type_273);  convert_element_type_272 = convert_element_type_273 = None
	        mul_4454 = torch.ops.aten.mul.Tensor(sub_2099, view_713);  sub_2099 = view_713 = None
	        _assert_tensor_metadata_412 = torch.ops.aten._assert_tensor_metadata.default(mul_4454, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_412 = None
	        view_716 = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_717 = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_718 = torch.ops.aten.view.default(model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_413 = torch.ops.aten._assert_tensor_metadata.default(view_716, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_413 = None
	        convert_element_type_274 = torch.ops.prims.convert_element_type.default(view_716, torch.float32);  view_716 = None
	        _assert_tensor_metadata_414 = torch.ops.aten._assert_tensor_metadata.default(view_718, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_414 = None
	        convert_element_type_275 = torch.ops.prims.convert_element_type.default(view_718, torch.float32);  view_718 = None
	        sub_2103 = torch.ops.aten.sub.Tensor(convert_element_type_274, convert_element_type_275);  convert_element_type_274 = convert_element_type_275 = None
	        mul_4459 = torch.ops.aten.mul.Tensor(sub_2103, view_717);  sub_2103 = view_717 = None
	        view_719 = torch.ops.aten.view.default(mul_4459, [1280, 1280]);  mul_4459 = None
	        _assert_tensor_metadata_415 = torch.ops.aten._assert_tensor_metadata.default(view_719, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_415 = None
	        mul_4464 = sym_size_int * 1500
	        view_720 = torch.ops.aten.view.default(mul_4454, [mul_4464, 1280]);  mul_4454 = mul_4464 = None
	        permute_78 = torch.ops.aten.permute.default(view_719, [1, 0]);  view_719 = None
	        addmm_37 = torch.ops.aten.addmm.default(model_audio_tower_layers_7_self_attn_out_proj_bias, view_720, permute_78);  model_audio_tower_layers_7_self_attn_out_proj_bias = view_720 = permute_78 = None
	        view_721 = torch.ops.aten.view.default(addmm_37, [sym_size_int, 1500, 1280]);  addmm_37 = None
	        add_7072 = torch.ops.aten.add.Tensor(add_6452, view_721);  add_6452 = view_721 = None
	        clone_62 = torch.ops.aten.clone.default(add_7072, memory_format = torch.contiguous_format)
	        var_mean_15 = torch.ops.aten.var_mean.correction(clone_62, [2], correction = 0, keepdim = True)
	        getitem_62 = var_mean_15[0]
	        getitem_63 = var_mean_15[1];  var_mean_15 = None
	        add_7077 = torch.ops.aten.add.Tensor(getitem_62, 1e-05);  getitem_62 = None
	        rsqrt_15 = torch.ops.aten.rsqrt.default(add_7077);  add_7077 = None
	        sub_2109 = torch.ops.aten.sub.Tensor(clone_62, getitem_63);  clone_62 = getitem_63 = None
	        mul_4475 = torch.ops.aten.mul.Tensor(sub_2109, rsqrt_15);  sub_2109 = rsqrt_15 = None
	        mul_4476 = torch.ops.aten.mul.Tensor(mul_4475, model_audio_tower_layers_7_final_layer_norm_weight);  mul_4475 = model_audio_tower_layers_7_final_layer_norm_weight = None
	        add_7078 = torch.ops.aten.add.Tensor(mul_4476, model_audio_tower_layers_7_final_layer_norm_bias);  mul_4476 = model_audio_tower_layers_7_final_layer_norm_bias = None
	        amin_46 = torch.ops.aten.amin.default(add_7078, [2])
	        amax_46 = torch.ops.aten.amax.default(add_7078, [2])
	        full_92 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_46 = torch.ops.aten.minimum.default(amin_46, full_92);  amin_46 = full_92 = None
	        full_93 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_46 = torch.ops.aten.maximum.default(amax_46, full_93);  amax_46 = full_93 = None
	        sub_2120 = torch.ops.aten.sub.Tensor(maximum_46, minimum_46);  maximum_46 = None
	        div_92 = torch.ops.aten.div.Tensor(sub_2120, 255.0);  sub_2120 = None
	        clamp_min_138 = torch.ops.aten.clamp_min.default(div_92, 1.1920928955078125e-07);  div_92 = None
	        div_93 = torch.ops.aten.div.Tensor(minimum_46, clamp_min_138);  minimum_46 = None
	        round_93 = torch.ops.aten.round.default(div_93);  div_93 = None
	        sub_2126 = torch.ops.aten.sub.Tensor(-128, round_93);  round_93 = None
	        clamp_min_139 = torch.ops.aten.clamp_min.default(sub_2126, -128);  sub_2126 = None
	        clamp_max_92 = torch.ops.aten.clamp_max.default(clamp_min_139, 127);  clamp_min_139 = None
	        _assert_tensor_metadata_416 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_138, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_416 = None
	        _assert_tensor_metadata_417 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_92, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_417 = None
	        convert_element_type_276 = torch.ops.prims.convert_element_type.default(clamp_max_92, torch.int8);  clamp_max_92 = None
	        view_724 = torch.ops.aten.view.default(clamp_min_138, [sym_size_int, 1500, 1])
	        view_725 = torch.ops.aten.view.default(convert_element_type_276, [sym_size_int, 1500, 1])
	        reciprocal_46 = torch.ops.aten.reciprocal.default(view_724);  view_724 = None
	        mul_4524 = torch.ops.aten.mul.Tensor(reciprocal_46, 1.0);  reciprocal_46 = None
	        mul_4527 = torch.ops.aten.mul.Tensor(add_7078, mul_4524);  add_7078 = mul_4524 = None
	        round_94 = torch.ops.aten.round.default(mul_4527);  mul_4527 = None
	        add_7165 = torch.ops.aten.add.Tensor(round_94, view_725);  round_94 = view_725 = None
	        clamp_min_140 = torch.ops.aten.clamp_min.default(add_7165, -128);  add_7165 = None
	        clamp_max_93 = torch.ops.aten.clamp_max.default(clamp_min_140, 127);  clamp_min_140 = None
	        _assert_tensor_metadata_418 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_93, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_418 = None
	        convert_element_type_277 = torch.ops.prims.convert_element_type.default(clamp_max_93, torch.int8);  clamp_max_93 = None
	        view_728 = torch.ops.aten.view.default(clamp_min_138, [sym_size_int, 1500, 1]);  clamp_min_138 = None
	        view_729 = torch.ops.aten.view.default(convert_element_type_276, [sym_size_int, 1500, 1]);  convert_element_type_276 = None
	        _assert_tensor_metadata_419 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_277, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_419 = None
	        convert_element_type_278 = torch.ops.prims.convert_element_type.default(convert_element_type_277, torch.float32);  convert_element_type_277 = None
	        _assert_tensor_metadata_420 = torch.ops.aten._assert_tensor_metadata.default(view_729, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_420 = None
	        convert_element_type_279 = torch.ops.prims.convert_element_type.default(view_729, torch.float32);  view_729 = None
	        sub_2146 = torch.ops.aten.sub.Tensor(convert_element_type_278, convert_element_type_279);  convert_element_type_278 = convert_element_type_279 = None
	        mul_4549 = torch.ops.aten.mul.Tensor(sub_2146, view_728);  sub_2146 = view_728 = None
	        _assert_tensor_metadata_421 = torch.ops.aten._assert_tensor_metadata.default(mul_4549, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_421 = None
	        view_731 = torch.ops.aten.view.default(model_audio_tower_layers_7_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_7_fc1_parametrizations_weight_original0 = None
	        view_732 = torch.ops.aten.view.default(model_audio_tower_layers_7_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_7_fc1_parametrizations_weight_original1 = None
	        view_733 = torch.ops.aten.view.default(model_audio_tower_layers_7_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_7_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_422 = torch.ops.aten._assert_tensor_metadata.default(view_731, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_422 = None
	        convert_element_type_280 = torch.ops.prims.convert_element_type.default(view_731, torch.float32);  view_731 = None
	        _assert_tensor_metadata_423 = torch.ops.aten._assert_tensor_metadata.default(view_733, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_423 = None
	        convert_element_type_281 = torch.ops.prims.convert_element_type.default(view_733, torch.float32);  view_733 = None
	        sub_2150 = torch.ops.aten.sub.Tensor(convert_element_type_280, convert_element_type_281);  convert_element_type_280 = convert_element_type_281 = None
	        mul_4554 = torch.ops.aten.mul.Tensor(sub_2150, view_732);  sub_2150 = view_732 = None
	        view_734 = torch.ops.aten.view.default(mul_4554, [5120, 1280]);  mul_4554 = None
	        _assert_tensor_metadata_424 = torch.ops.aten._assert_tensor_metadata.default(view_734, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_424 = None
	        mul_4559 = sym_size_int * 1500
	        view_735 = torch.ops.aten.view.default(mul_4549, [mul_4559, 1280]);  mul_4549 = mul_4559 = None
	        permute_79 = torch.ops.aten.permute.default(view_734, [1, 0]);  view_734 = None
	        addmm_38 = torch.ops.aten.addmm.default(model_audio_tower_layers_7_fc1_bias, view_735, permute_79);  model_audio_tower_layers_7_fc1_bias = view_735 = permute_79 = None
	        view_736 = torch.ops.aten.view.default(addmm_38, [sym_size_int, 1500, 5120]);  addmm_38 = None
	        mul_4566 = torch.ops.aten.mul.Tensor(view_736, 0.5)
	        mul_4567 = torch.ops.aten.mul.Tensor(view_736, 0.7071067811865476);  view_736 = None
	        erf_9 = torch.ops.aten.erf.default(mul_4567);  mul_4567 = None
	        add_7224 = torch.ops.aten.add.Tensor(erf_9, 1);  erf_9 = None
	        mul_4568 = torch.ops.aten.mul.Tensor(mul_4566, add_7224);  mul_4566 = add_7224 = None
	        amin_47 = torch.ops.aten.amin.default(mul_4568, [2])
	        amax_47 = torch.ops.aten.amax.default(mul_4568, [2])
	        full_94 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_47 = torch.ops.aten.minimum.default(amin_47, full_94);  amin_47 = full_94 = None
	        full_95 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_47 = torch.ops.aten.maximum.default(amax_47, full_95);  amax_47 = full_95 = None
	        sub_2163 = torch.ops.aten.sub.Tensor(maximum_47, minimum_47);  maximum_47 = None
	        div_94 = torch.ops.aten.div.Tensor(sub_2163, 255.0);  sub_2163 = None
	        clamp_min_141 = torch.ops.aten.clamp_min.default(div_94, 1.1920928955078125e-07);  div_94 = None
	        div_95 = torch.ops.aten.div.Tensor(minimum_47, clamp_min_141);  minimum_47 = None
	        round_95 = torch.ops.aten.round.default(div_95);  div_95 = None
	        sub_2169 = torch.ops.aten.sub.Tensor(-128, round_95);  round_95 = None
	        clamp_min_142 = torch.ops.aten.clamp_min.default(sub_2169, -128);  sub_2169 = None
	        clamp_max_94 = torch.ops.aten.clamp_max.default(clamp_min_142, 127);  clamp_min_142 = None
	        _assert_tensor_metadata_425 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_141, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_425 = None
	        _assert_tensor_metadata_426 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_94, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_426 = None
	        convert_element_type_282 = torch.ops.prims.convert_element_type.default(clamp_max_94, torch.int8);  clamp_max_94 = None
	        view_739 = torch.ops.aten.view.default(clamp_min_141, [sym_size_int, 1500, 1])
	        view_740 = torch.ops.aten.view.default(convert_element_type_282, [sym_size_int, 1500, 1])
	        reciprocal_47 = torch.ops.aten.reciprocal.default(view_739);  view_739 = None
	        mul_4614 = torch.ops.aten.mul.Tensor(reciprocal_47, 1.0);  reciprocal_47 = None
	        mul_4617 = torch.ops.aten.mul.Tensor(mul_4568, mul_4614);  mul_4568 = mul_4614 = None
	        round_96 = torch.ops.aten.round.default(mul_4617);  mul_4617 = None
	        add_7307 = torch.ops.aten.add.Tensor(round_96, view_740);  round_96 = view_740 = None
	        clamp_min_143 = torch.ops.aten.clamp_min.default(add_7307, -128);  add_7307 = None
	        clamp_max_95 = torch.ops.aten.clamp_max.default(clamp_min_143, 127);  clamp_min_143 = None
	        _assert_tensor_metadata_427 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_95, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_427 = None
	        convert_element_type_283 = torch.ops.prims.convert_element_type.default(clamp_max_95, torch.int8);  clamp_max_95 = None
	        view_743 = torch.ops.aten.view.default(clamp_min_141, [sym_size_int, 1500, 1]);  clamp_min_141 = None
	        view_744 = torch.ops.aten.view.default(convert_element_type_282, [sym_size_int, 1500, 1]);  convert_element_type_282 = None
	        _assert_tensor_metadata_428 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_283, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_428 = None
	        convert_element_type_284 = torch.ops.prims.convert_element_type.default(convert_element_type_283, torch.float32);  convert_element_type_283 = None
	        _assert_tensor_metadata_429 = torch.ops.aten._assert_tensor_metadata.default(view_744, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_429 = None
	        convert_element_type_285 = torch.ops.prims.convert_element_type.default(view_744, torch.float32);  view_744 = None
	        sub_2189 = torch.ops.aten.sub.Tensor(convert_element_type_284, convert_element_type_285);  convert_element_type_284 = convert_element_type_285 = None
	        mul_4639 = torch.ops.aten.mul.Tensor(sub_2189, view_743);  sub_2189 = view_743 = None
	        _assert_tensor_metadata_430 = torch.ops.aten._assert_tensor_metadata.default(mul_4639, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_430 = None
	        view_746 = torch.ops.aten.view.default(model_audio_tower_layers_7_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_7_fc2_parametrizations_weight_original0 = None
	        view_747 = torch.ops.aten.view.default(model_audio_tower_layers_7_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_7_fc2_parametrizations_weight_original1 = None
	        view_748 = torch.ops.aten.view.default(model_audio_tower_layers_7_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_7_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_431 = torch.ops.aten._assert_tensor_metadata.default(view_746, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_431 = None
	        convert_element_type_286 = torch.ops.prims.convert_element_type.default(view_746, torch.float32);  view_746 = None
	        _assert_tensor_metadata_432 = torch.ops.aten._assert_tensor_metadata.default(view_748, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_432 = None
	        convert_element_type_287 = torch.ops.prims.convert_element_type.default(view_748, torch.float32);  view_748 = None
	        sub_2193 = torch.ops.aten.sub.Tensor(convert_element_type_286, convert_element_type_287);  convert_element_type_286 = convert_element_type_287 = None
	        mul_4644 = torch.ops.aten.mul.Tensor(sub_2193, view_747);  sub_2193 = view_747 = None
	        view_749 = torch.ops.aten.view.default(mul_4644, [1280, 5120]);  mul_4644 = None
	        _assert_tensor_metadata_433 = torch.ops.aten._assert_tensor_metadata.default(view_749, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_433 = None
	        mul_4649 = sym_size_int * 1500
	        view_750 = torch.ops.aten.view.default(mul_4639, [mul_4649, 5120]);  mul_4639 = mul_4649 = None
	        permute_80 = torch.ops.aten.permute.default(view_749, [1, 0]);  view_749 = None
	        addmm_39 = torch.ops.aten.addmm.default(model_audio_tower_layers_7_fc2_bias, view_750, permute_80);  model_audio_tower_layers_7_fc2_bias = view_750 = permute_80 = None
	        view_751 = torch.ops.aten.view.default(addmm_39, [sym_size_int, 1500, 1280]);  addmm_39 = None
	        add_7370 = torch.ops.aten.add.Tensor(add_7072, view_751);  add_7072 = view_751 = None
	        clone_65 = torch.ops.aten.clone.default(add_7370, memory_format = torch.contiguous_format)
	        var_mean_16 = torch.ops.aten.var_mean.correction(clone_65, [2], correction = 0, keepdim = True)
	        getitem_64 = var_mean_16[0]
	        getitem_65 = var_mean_16[1];  var_mean_16 = None
	        add_7375 = torch.ops.aten.add.Tensor(getitem_64, 1e-05);  getitem_64 = None
	        rsqrt_16 = torch.ops.aten.rsqrt.default(add_7375);  add_7375 = None
	        sub_2199 = torch.ops.aten.sub.Tensor(clone_65, getitem_65);  clone_65 = getitem_65 = None
	        mul_4660 = torch.ops.aten.mul.Tensor(sub_2199, rsqrt_16);  sub_2199 = rsqrt_16 = None
	        mul_4661 = torch.ops.aten.mul.Tensor(mul_4660, model_audio_tower_layers_8_self_attn_layer_norm_weight);  mul_4660 = model_audio_tower_layers_8_self_attn_layer_norm_weight = None
	        add_7376 = torch.ops.aten.add.Tensor(mul_4661, model_audio_tower_layers_8_self_attn_layer_norm_bias);  mul_4661 = model_audio_tower_layers_8_self_attn_layer_norm_bias = None
	        amin_48 = torch.ops.aten.amin.default(add_7376, [2])
	        amax_48 = torch.ops.aten.amax.default(add_7376, [2])
	        full_96 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_48 = torch.ops.aten.minimum.default(amin_48, full_96);  amin_48 = full_96 = None
	        full_97 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_48 = torch.ops.aten.maximum.default(amax_48, full_97);  amax_48 = full_97 = None
	        sub_2210 = torch.ops.aten.sub.Tensor(maximum_48, minimum_48);  maximum_48 = None
	        div_96 = torch.ops.aten.div.Tensor(sub_2210, 255.0);  sub_2210 = None
	        clamp_min_144 = torch.ops.aten.clamp_min.default(div_96, 1.1920928955078125e-07);  div_96 = None
	        div_97 = torch.ops.aten.div.Tensor(minimum_48, clamp_min_144);  minimum_48 = None
	        round_97 = torch.ops.aten.round.default(div_97);  div_97 = None
	        sub_2216 = torch.ops.aten.sub.Tensor(-128, round_97);  round_97 = None
	        clamp_min_145 = torch.ops.aten.clamp_min.default(sub_2216, -128);  sub_2216 = None
	        clamp_max_96 = torch.ops.aten.clamp_max.default(clamp_min_145, 127);  clamp_min_145 = None
	        _assert_tensor_metadata_434 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_144, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_434 = None
	        _assert_tensor_metadata_435 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_96, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_435 = None
	        convert_element_type_288 = torch.ops.prims.convert_element_type.default(clamp_max_96, torch.int8);  clamp_max_96 = None
	        view_754 = torch.ops.aten.view.default(clamp_min_144, [sym_size_int, 1500, 1])
	        view_755 = torch.ops.aten.view.default(convert_element_type_288, [sym_size_int, 1500, 1])
	        reciprocal_48 = torch.ops.aten.reciprocal.default(view_754);  view_754 = None
	        mul_4709 = torch.ops.aten.mul.Tensor(reciprocal_48, 1.0);  reciprocal_48 = None
	        mul_4712 = torch.ops.aten.mul.Tensor(add_7376, mul_4709);  mul_4709 = None
	        round_98 = torch.ops.aten.round.default(mul_4712);  mul_4712 = None
	        add_7463 = torch.ops.aten.add.Tensor(round_98, view_755);  round_98 = view_755 = None
	        clamp_min_146 = torch.ops.aten.clamp_min.default(add_7463, -128);  add_7463 = None
	        clamp_max_97 = torch.ops.aten.clamp_max.default(clamp_min_146, 127);  clamp_min_146 = None
	        _assert_tensor_metadata_436 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_97, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_436 = None
	        convert_element_type_289 = torch.ops.prims.convert_element_type.default(clamp_max_97, torch.int8);  clamp_max_97 = None
	        view_758 = torch.ops.aten.view.default(clamp_min_144, [sym_size_int, 1500, 1]);  clamp_min_144 = None
	        view_759 = torch.ops.aten.view.default(convert_element_type_288, [sym_size_int, 1500, 1]);  convert_element_type_288 = None
	        _assert_tensor_metadata_437 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_289, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_437 = None
	        convert_element_type_290 = torch.ops.prims.convert_element_type.default(convert_element_type_289, torch.float32);  convert_element_type_289 = None
	        _assert_tensor_metadata_438 = torch.ops.aten._assert_tensor_metadata.default(view_759, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_438 = None
	        convert_element_type_291 = torch.ops.prims.convert_element_type.default(view_759, torch.float32);  view_759 = None
	        sub_2236 = torch.ops.aten.sub.Tensor(convert_element_type_290, convert_element_type_291);  convert_element_type_290 = convert_element_type_291 = None
	        mul_4734 = torch.ops.aten.mul.Tensor(sub_2236, view_758);  sub_2236 = view_758 = None
	        _assert_tensor_metadata_439 = torch.ops.aten._assert_tensor_metadata.default(mul_4734, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_439 = None
	        view_761 = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_762 = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_763 = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_440 = torch.ops.aten._assert_tensor_metadata.default(view_761, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_440 = None
	        convert_element_type_292 = torch.ops.prims.convert_element_type.default(view_761, torch.float32);  view_761 = None
	        _assert_tensor_metadata_441 = torch.ops.aten._assert_tensor_metadata.default(view_763, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_441 = None
	        convert_element_type_293 = torch.ops.prims.convert_element_type.default(view_763, torch.float32);  view_763 = None
	        sub_2240 = torch.ops.aten.sub.Tensor(convert_element_type_292, convert_element_type_293);  convert_element_type_292 = convert_element_type_293 = None
	        mul_4739 = torch.ops.aten.mul.Tensor(sub_2240, view_762);  sub_2240 = view_762 = None
	        view_764 = torch.ops.aten.view.default(mul_4739, [1280, 1280]);  mul_4739 = None
	        _assert_tensor_metadata_442 = torch.ops.aten._assert_tensor_metadata.default(view_764, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_442 = None
	        mul_4744 = sym_size_int * 1500
	        view_765 = torch.ops.aten.view.default(mul_4734, [mul_4744, 1280]);  mul_4734 = mul_4744 = None
	        permute_81 = torch.ops.aten.permute.default(view_764, [1, 0]);  view_764 = None
	        addmm_40 = torch.ops.aten.addmm.default(model_audio_tower_layers_8_self_attn_q_proj_bias, view_765, permute_81);  model_audio_tower_layers_8_self_attn_q_proj_bias = view_765 = permute_81 = None
	        view_766 = torch.ops.aten.view.default(addmm_40, [sym_size_int, 1500, 1280]);  addmm_40 = None
	        mul_4751 = torch.ops.aten.mul.Tensor(view_766, 0.125);  view_766 = None
	        view_767 = torch.ops.aten.view.default(mul_4751, [sym_size_int, 1500, 20, 64]);  mul_4751 = None
	        permute_82 = torch.ops.aten.permute.default(view_767, [0, 2, 1, 3]);  view_767 = None
	        clone_66 = torch.ops.aten.clone.default(permute_82, memory_format = torch.contiguous_format);  permute_82 = None
	        amin_49 = torch.ops.aten.amin.default(add_7376, [2])
	        amax_49 = torch.ops.aten.amax.default(add_7376, [2])
	        full_98 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_49 = torch.ops.aten.minimum.default(amin_49, full_98);  amin_49 = full_98 = None
	        full_99 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_49 = torch.ops.aten.maximum.default(amax_49, full_99);  amax_49 = full_99 = None
	        sub_2255 = torch.ops.aten.sub.Tensor(maximum_49, minimum_49);  maximum_49 = None
	        div_98 = torch.ops.aten.div.Tensor(sub_2255, 255.0);  sub_2255 = None
	        clamp_min_147 = torch.ops.aten.clamp_min.default(div_98, 1.1920928955078125e-07);  div_98 = None
	        div_99 = torch.ops.aten.div.Tensor(minimum_49, clamp_min_147);  minimum_49 = None
	        round_99 = torch.ops.aten.round.default(div_99);  div_99 = None
	        sub_2261 = torch.ops.aten.sub.Tensor(-128, round_99);  round_99 = None
	        clamp_min_148 = torch.ops.aten.clamp_min.default(sub_2261, -128);  sub_2261 = None
	        clamp_max_98 = torch.ops.aten.clamp_max.default(clamp_min_148, 127);  clamp_min_148 = None
	        _assert_tensor_metadata_443 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_147, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_443 = None
	        _assert_tensor_metadata_444 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_98, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_444 = None
	        convert_element_type_294 = torch.ops.prims.convert_element_type.default(clamp_max_98, torch.int8);  clamp_max_98 = None
	        view_770 = torch.ops.aten.view.default(clamp_min_147, [sym_size_int, 1500, 1])
	        view_771 = torch.ops.aten.view.default(convert_element_type_294, [sym_size_int, 1500, 1])
	        reciprocal_49 = torch.ops.aten.reciprocal.default(view_770);  view_770 = None
	        mul_4805 = torch.ops.aten.mul.Tensor(reciprocal_49, 1.0);  reciprocal_49 = None
	        mul_4808 = torch.ops.aten.mul.Tensor(add_7376, mul_4805);  mul_4805 = None
	        round_100 = torch.ops.aten.round.default(mul_4808);  mul_4808 = None
	        add_7615 = torch.ops.aten.add.Tensor(round_100, view_771);  round_100 = view_771 = None
	        clamp_min_149 = torch.ops.aten.clamp_min.default(add_7615, -128);  add_7615 = None
	        clamp_max_99 = torch.ops.aten.clamp_max.default(clamp_min_149, 127);  clamp_min_149 = None
	        _assert_tensor_metadata_445 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_99, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_445 = None
	        convert_element_type_295 = torch.ops.prims.convert_element_type.default(clamp_max_99, torch.int8);  clamp_max_99 = None
	        view_774 = torch.ops.aten.view.default(clamp_min_147, [sym_size_int, 1500, 1]);  clamp_min_147 = None
	        view_775 = torch.ops.aten.view.default(convert_element_type_294, [sym_size_int, 1500, 1]);  convert_element_type_294 = None
	        _assert_tensor_metadata_446 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_295, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_446 = None
	        convert_element_type_296 = torch.ops.prims.convert_element_type.default(convert_element_type_295, torch.float32);  convert_element_type_295 = None
	        _assert_tensor_metadata_447 = torch.ops.aten._assert_tensor_metadata.default(view_775, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_447 = None
	        convert_element_type_297 = torch.ops.prims.convert_element_type.default(view_775, torch.float32);  view_775 = None
	        sub_2281 = torch.ops.aten.sub.Tensor(convert_element_type_296, convert_element_type_297);  convert_element_type_296 = convert_element_type_297 = None
	        mul_4830 = torch.ops.aten.mul.Tensor(sub_2281, view_774);  sub_2281 = view_774 = None
	        _assert_tensor_metadata_448 = torch.ops.aten._assert_tensor_metadata.default(mul_4830, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_448 = None
	        view_777 = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_778 = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_779 = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_449 = torch.ops.aten._assert_tensor_metadata.default(view_777, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_449 = None
	        convert_element_type_298 = torch.ops.prims.convert_element_type.default(view_777, torch.float32);  view_777 = None
	        _assert_tensor_metadata_450 = torch.ops.aten._assert_tensor_metadata.default(view_779, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_450 = None
	        convert_element_type_299 = torch.ops.prims.convert_element_type.default(view_779, torch.float32);  view_779 = None
	        sub_2285 = torch.ops.aten.sub.Tensor(convert_element_type_298, convert_element_type_299);  convert_element_type_298 = convert_element_type_299 = None
	        mul_4835 = torch.ops.aten.mul.Tensor(sub_2285, view_778);  sub_2285 = view_778 = None
	        view_780 = torch.ops.aten.view.default(mul_4835, [1280, 1280]);  mul_4835 = None
	        _assert_tensor_metadata_451 = torch.ops.aten._assert_tensor_metadata.default(view_780, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_451 = None
	        permute_83 = torch.ops.aten.permute.default(view_780, [1, 0]);  view_780 = None
	        mul_4838 = sym_size_int * 1500
	        view_781 = torch.ops.aten.view.default(mul_4830, [mul_4838, 1280]);  mul_4830 = mul_4838 = None
	        mm_8 = torch.ops.aten.mm.default(view_781, permute_83);  view_781 = permute_83 = None
	        view_782 = torch.ops.aten.view.default(mm_8, [sym_size_int, 1500, 1280]);  mm_8 = None
	        view_783 = torch.ops.aten.view.default(view_782, [sym_size_int, -1, 20, 64]);  view_782 = None
	        permute_84 = torch.ops.aten.permute.default(view_783, [0, 2, 1, 3]);  view_783 = None
	        clone_67 = torch.ops.aten.clone.default(permute_84, memory_format = torch.contiguous_format);  permute_84 = None
	        amin_50 = torch.ops.aten.amin.default(add_7376, [2])
	        amax_50 = torch.ops.aten.amax.default(add_7376, [2])
	        full_100 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_50 = torch.ops.aten.minimum.default(amin_50, full_100);  amin_50 = full_100 = None
	        full_101 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_50 = torch.ops.aten.maximum.default(amax_50, full_101);  amax_50 = full_101 = None
	        sub_2299 = torch.ops.aten.sub.Tensor(maximum_50, minimum_50);  maximum_50 = None
	        div_100 = torch.ops.aten.div.Tensor(sub_2299, 255.0);  sub_2299 = None
	        clamp_min_150 = torch.ops.aten.clamp_min.default(div_100, 1.1920928955078125e-07);  div_100 = None
	        div_101 = torch.ops.aten.div.Tensor(minimum_50, clamp_min_150);  minimum_50 = None
	        round_101 = torch.ops.aten.round.default(div_101);  div_101 = None
	        sub_2305 = torch.ops.aten.sub.Tensor(-128, round_101);  round_101 = None
	        clamp_min_151 = torch.ops.aten.clamp_min.default(sub_2305, -128);  sub_2305 = None
	        clamp_max_100 = torch.ops.aten.clamp_max.default(clamp_min_151, 127);  clamp_min_151 = None
	        _assert_tensor_metadata_452 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_150, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_452 = None
	        _assert_tensor_metadata_453 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_100, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_453 = None
	        convert_element_type_300 = torch.ops.prims.convert_element_type.default(clamp_max_100, torch.int8);  clamp_max_100 = None
	        view_786 = torch.ops.aten.view.default(clamp_min_150, [sym_size_int, 1500, 1])
	        view_787 = torch.ops.aten.view.default(convert_element_type_300, [sym_size_int, 1500, 1])
	        reciprocal_50 = torch.ops.aten.reciprocal.default(view_786);  view_786 = None
	        mul_4904 = torch.ops.aten.mul.Tensor(reciprocal_50, 1.0);  reciprocal_50 = None
	        mul_4907 = torch.ops.aten.mul.Tensor(add_7376, mul_4904);  add_7376 = mul_4904 = None
	        round_102 = torch.ops.aten.round.default(mul_4907);  mul_4907 = None
	        add_7763 = torch.ops.aten.add.Tensor(round_102, view_787);  round_102 = view_787 = None
	        clamp_min_152 = torch.ops.aten.clamp_min.default(add_7763, -128);  add_7763 = None
	        clamp_max_101 = torch.ops.aten.clamp_max.default(clamp_min_152, 127);  clamp_min_152 = None
	        _assert_tensor_metadata_454 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_101, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_454 = None
	        convert_element_type_301 = torch.ops.prims.convert_element_type.default(clamp_max_101, torch.int8);  clamp_max_101 = None
	        view_790 = torch.ops.aten.view.default(clamp_min_150, [sym_size_int, 1500, 1]);  clamp_min_150 = None
	        view_791 = torch.ops.aten.view.default(convert_element_type_300, [sym_size_int, 1500, 1]);  convert_element_type_300 = None
	        _assert_tensor_metadata_455 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_301, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_455 = None
	        convert_element_type_302 = torch.ops.prims.convert_element_type.default(convert_element_type_301, torch.float32);  convert_element_type_301 = None
	        _assert_tensor_metadata_456 = torch.ops.aten._assert_tensor_metadata.default(view_791, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_456 = None
	        convert_element_type_303 = torch.ops.prims.convert_element_type.default(view_791, torch.float32);  view_791 = None
	        sub_2325 = torch.ops.aten.sub.Tensor(convert_element_type_302, convert_element_type_303);  convert_element_type_302 = convert_element_type_303 = None
	        mul_4929 = torch.ops.aten.mul.Tensor(sub_2325, view_790);  sub_2325 = view_790 = None
	        _assert_tensor_metadata_457 = torch.ops.aten._assert_tensor_metadata.default(mul_4929, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_457 = None
	        view_793 = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_794 = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_795 = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_458 = torch.ops.aten._assert_tensor_metadata.default(view_793, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_458 = None
	        convert_element_type_304 = torch.ops.prims.convert_element_type.default(view_793, torch.float32);  view_793 = None
	        _assert_tensor_metadata_459 = torch.ops.aten._assert_tensor_metadata.default(view_795, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_459 = None
	        convert_element_type_305 = torch.ops.prims.convert_element_type.default(view_795, torch.float32);  view_795 = None
	        sub_2329 = torch.ops.aten.sub.Tensor(convert_element_type_304, convert_element_type_305);  convert_element_type_304 = convert_element_type_305 = None
	        mul_4934 = torch.ops.aten.mul.Tensor(sub_2329, view_794);  sub_2329 = view_794 = None
	        view_796 = torch.ops.aten.view.default(mul_4934, [1280, 1280]);  mul_4934 = None
	        _assert_tensor_metadata_460 = torch.ops.aten._assert_tensor_metadata.default(view_796, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_460 = None
	        mul_4939 = sym_size_int * 1500
	        view_797 = torch.ops.aten.view.default(mul_4929, [mul_4939, 1280]);  mul_4929 = mul_4939 = None
	        permute_85 = torch.ops.aten.permute.default(view_796, [1, 0]);  view_796 = None
	        addmm_41 = torch.ops.aten.addmm.default(model_audio_tower_layers_8_self_attn_v_proj_bias, view_797, permute_85);  model_audio_tower_layers_8_self_attn_v_proj_bias = view_797 = permute_85 = None
	        view_798 = torch.ops.aten.view.default(addmm_41, [sym_size_int, 1500, 1280]);  addmm_41 = None
	        view_799 = torch.ops.aten.view.default(view_798, [sym_size_int, -1, 20, 64]);  view_798 = None
	        permute_86 = torch.ops.aten.permute.default(view_799, [0, 2, 1, 3]);  view_799 = None
	        clone_68 = torch.ops.aten.clone.default(permute_86, memory_format = torch.contiguous_format);  permute_86 = None
	        _scaled_dot_product_efficient_attention_8 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_66, clone_67, clone_68, None, False, scale = 1.0);  clone_66 = clone_67 = clone_68 = None
	        getitem_66 = _scaled_dot_product_efficient_attention_8[0];  _scaled_dot_product_efficient_attention_8 = None
	        permute_87 = torch.ops.aten.permute.default(getitem_66, [0, 2, 1, 3]);  getitem_66 = None
	        view_800 = torch.ops.aten.view.default(permute_87, [sym_size_int, 1500, -1]);  permute_87 = None
	        amin_51 = torch.ops.aten.amin.default(view_800, [2])
	        amax_51 = torch.ops.aten.amax.default(view_800, [2])
	        full_102 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_51 = torch.ops.aten.minimum.default(amin_51, full_102);  amin_51 = full_102 = None
	        full_103 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_51 = torch.ops.aten.maximum.default(amax_51, full_103);  amax_51 = full_103 = None
	        sub_2347 = torch.ops.aten.sub.Tensor(maximum_51, minimum_51);  maximum_51 = None
	        div_102 = torch.ops.aten.div.Tensor(sub_2347, 255.0);  sub_2347 = None
	        clamp_min_153 = torch.ops.aten.clamp_min.default(div_102, 1.1920928955078125e-07);  div_102 = None
	        div_103 = torch.ops.aten.div.Tensor(minimum_51, clamp_min_153);  minimum_51 = None
	        round_103 = torch.ops.aten.round.default(div_103);  div_103 = None
	        sub_2353 = torch.ops.aten.sub.Tensor(-128, round_103);  round_103 = None
	        clamp_min_154 = torch.ops.aten.clamp_min.default(sub_2353, -128);  sub_2353 = None
	        clamp_max_102 = torch.ops.aten.clamp_max.default(clamp_min_154, 127);  clamp_min_154 = None
	        _assert_tensor_metadata_461 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_153, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_461 = None
	        _assert_tensor_metadata_462 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_102, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_462 = None
	        convert_element_type_306 = torch.ops.prims.convert_element_type.default(clamp_max_102, torch.int8);  clamp_max_102 = None
	        view_803 = torch.ops.aten.view.default(clamp_min_153, [sym_size_int, 1500, 1])
	        view_804 = torch.ops.aten.view.default(convert_element_type_306, [sym_size_int, 1500, 1])
	        reciprocal_51 = torch.ops.aten.reciprocal.default(view_803);  view_803 = None
	        mul_5009 = torch.ops.aten.mul.Tensor(reciprocal_51, 1.0);  reciprocal_51 = None
	        mul_5012 = torch.ops.aten.mul.Tensor(view_800, mul_5009);  view_800 = mul_5009 = None
	        round_104 = torch.ops.aten.round.default(mul_5012);  mul_5012 = None
	        add_7927 = torch.ops.aten.add.Tensor(round_104, view_804);  round_104 = view_804 = None
	        clamp_min_155 = torch.ops.aten.clamp_min.default(add_7927, -128);  add_7927 = None
	        clamp_max_103 = torch.ops.aten.clamp_max.default(clamp_min_155, 127);  clamp_min_155 = None
	        _assert_tensor_metadata_463 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_103, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_463 = None
	        convert_element_type_307 = torch.ops.prims.convert_element_type.default(clamp_max_103, torch.int8);  clamp_max_103 = None
	        view_807 = torch.ops.aten.view.default(clamp_min_153, [sym_size_int, 1500, 1]);  clamp_min_153 = None
	        view_808 = torch.ops.aten.view.default(convert_element_type_306, [sym_size_int, 1500, 1]);  convert_element_type_306 = None
	        _assert_tensor_metadata_464 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_307, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_464 = None
	        convert_element_type_308 = torch.ops.prims.convert_element_type.default(convert_element_type_307, torch.float32);  convert_element_type_307 = None
	        _assert_tensor_metadata_465 = torch.ops.aten._assert_tensor_metadata.default(view_808, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_465 = None
	        convert_element_type_309 = torch.ops.prims.convert_element_type.default(view_808, torch.float32);  view_808 = None
	        sub_2373 = torch.ops.aten.sub.Tensor(convert_element_type_308, convert_element_type_309);  convert_element_type_308 = convert_element_type_309 = None
	        mul_5034 = torch.ops.aten.mul.Tensor(sub_2373, view_807);  sub_2373 = view_807 = None
	        _assert_tensor_metadata_466 = torch.ops.aten._assert_tensor_metadata.default(mul_5034, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_466 = None
	        view_810 = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_811 = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_812 = torch.ops.aten.view.default(model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_467 = torch.ops.aten._assert_tensor_metadata.default(view_810, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_467 = None
	        convert_element_type_310 = torch.ops.prims.convert_element_type.default(view_810, torch.float32);  view_810 = None
	        _assert_tensor_metadata_468 = torch.ops.aten._assert_tensor_metadata.default(view_812, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_468 = None
	        convert_element_type_311 = torch.ops.prims.convert_element_type.default(view_812, torch.float32);  view_812 = None
	        sub_2377 = torch.ops.aten.sub.Tensor(convert_element_type_310, convert_element_type_311);  convert_element_type_310 = convert_element_type_311 = None
	        mul_5039 = torch.ops.aten.mul.Tensor(sub_2377, view_811);  sub_2377 = view_811 = None
	        view_813 = torch.ops.aten.view.default(mul_5039, [1280, 1280]);  mul_5039 = None
	        _assert_tensor_metadata_469 = torch.ops.aten._assert_tensor_metadata.default(view_813, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_469 = None
	        mul_5044 = sym_size_int * 1500
	        view_814 = torch.ops.aten.view.default(mul_5034, [mul_5044, 1280]);  mul_5034 = mul_5044 = None
	        permute_88 = torch.ops.aten.permute.default(view_813, [1, 0]);  view_813 = None
	        addmm_42 = torch.ops.aten.addmm.default(model_audio_tower_layers_8_self_attn_out_proj_bias, view_814, permute_88);  model_audio_tower_layers_8_self_attn_out_proj_bias = view_814 = permute_88 = None
	        view_815 = torch.ops.aten.view.default(addmm_42, [sym_size_int, 1500, 1280]);  addmm_42 = None
	        add_7990 = torch.ops.aten.add.Tensor(add_7370, view_815);  add_7370 = view_815 = None
	        clone_70 = torch.ops.aten.clone.default(add_7990, memory_format = torch.contiguous_format)
	        var_mean_17 = torch.ops.aten.var_mean.correction(clone_70, [2], correction = 0, keepdim = True)
	        getitem_70 = var_mean_17[0]
	        getitem_71 = var_mean_17[1];  var_mean_17 = None
	        add_7995 = torch.ops.aten.add.Tensor(getitem_70, 1e-05);  getitem_70 = None
	        rsqrt_17 = torch.ops.aten.rsqrt.default(add_7995);  add_7995 = None
	        sub_2383 = torch.ops.aten.sub.Tensor(clone_70, getitem_71);  clone_70 = getitem_71 = None
	        mul_5055 = torch.ops.aten.mul.Tensor(sub_2383, rsqrt_17);  sub_2383 = rsqrt_17 = None
	        mul_5056 = torch.ops.aten.mul.Tensor(mul_5055, model_audio_tower_layers_8_final_layer_norm_weight);  mul_5055 = model_audio_tower_layers_8_final_layer_norm_weight = None
	        add_7996 = torch.ops.aten.add.Tensor(mul_5056, model_audio_tower_layers_8_final_layer_norm_bias);  mul_5056 = model_audio_tower_layers_8_final_layer_norm_bias = None
	        amin_52 = torch.ops.aten.amin.default(add_7996, [2])
	        amax_52 = torch.ops.aten.amax.default(add_7996, [2])
	        full_104 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_52 = torch.ops.aten.minimum.default(amin_52, full_104);  amin_52 = full_104 = None
	        full_105 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_52 = torch.ops.aten.maximum.default(amax_52, full_105);  amax_52 = full_105 = None
	        sub_2394 = torch.ops.aten.sub.Tensor(maximum_52, minimum_52);  maximum_52 = None
	        div_104 = torch.ops.aten.div.Tensor(sub_2394, 255.0);  sub_2394 = None
	        clamp_min_156 = torch.ops.aten.clamp_min.default(div_104, 1.1920928955078125e-07);  div_104 = None
	        div_105 = torch.ops.aten.div.Tensor(minimum_52, clamp_min_156);  minimum_52 = None
	        round_105 = torch.ops.aten.round.default(div_105);  div_105 = None
	        sub_2400 = torch.ops.aten.sub.Tensor(-128, round_105);  round_105 = None
	        clamp_min_157 = torch.ops.aten.clamp_min.default(sub_2400, -128);  sub_2400 = None
	        clamp_max_104 = torch.ops.aten.clamp_max.default(clamp_min_157, 127);  clamp_min_157 = None
	        _assert_tensor_metadata_470 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_156, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_470 = None
	        _assert_tensor_metadata_471 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_104, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_471 = None
	        convert_element_type_312 = torch.ops.prims.convert_element_type.default(clamp_max_104, torch.int8);  clamp_max_104 = None
	        view_818 = torch.ops.aten.view.default(clamp_min_156, [sym_size_int, 1500, 1])
	        view_819 = torch.ops.aten.view.default(convert_element_type_312, [sym_size_int, 1500, 1])
	        reciprocal_52 = torch.ops.aten.reciprocal.default(view_818);  view_818 = None
	        mul_5104 = torch.ops.aten.mul.Tensor(reciprocal_52, 1.0);  reciprocal_52 = None
	        mul_5107 = torch.ops.aten.mul.Tensor(add_7996, mul_5104);  add_7996 = mul_5104 = None
	        round_106 = torch.ops.aten.round.default(mul_5107);  mul_5107 = None
	        add_8083 = torch.ops.aten.add.Tensor(round_106, view_819);  round_106 = view_819 = None
	        clamp_min_158 = torch.ops.aten.clamp_min.default(add_8083, -128);  add_8083 = None
	        clamp_max_105 = torch.ops.aten.clamp_max.default(clamp_min_158, 127);  clamp_min_158 = None
	        _assert_tensor_metadata_472 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_105, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_472 = None
	        convert_element_type_313 = torch.ops.prims.convert_element_type.default(clamp_max_105, torch.int8);  clamp_max_105 = None
	        view_822 = torch.ops.aten.view.default(clamp_min_156, [sym_size_int, 1500, 1]);  clamp_min_156 = None
	        view_823 = torch.ops.aten.view.default(convert_element_type_312, [sym_size_int, 1500, 1]);  convert_element_type_312 = None
	        _assert_tensor_metadata_473 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_313, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_473 = None
	        convert_element_type_314 = torch.ops.prims.convert_element_type.default(convert_element_type_313, torch.float32);  convert_element_type_313 = None
	        _assert_tensor_metadata_474 = torch.ops.aten._assert_tensor_metadata.default(view_823, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_474 = None
	        convert_element_type_315 = torch.ops.prims.convert_element_type.default(view_823, torch.float32);  view_823 = None
	        sub_2420 = torch.ops.aten.sub.Tensor(convert_element_type_314, convert_element_type_315);  convert_element_type_314 = convert_element_type_315 = None
	        mul_5129 = torch.ops.aten.mul.Tensor(sub_2420, view_822);  sub_2420 = view_822 = None
	        _assert_tensor_metadata_475 = torch.ops.aten._assert_tensor_metadata.default(mul_5129, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_475 = None
	        view_825 = torch.ops.aten.view.default(model_audio_tower_layers_8_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_8_fc1_parametrizations_weight_original0 = None
	        view_826 = torch.ops.aten.view.default(model_audio_tower_layers_8_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_8_fc1_parametrizations_weight_original1 = None
	        view_827 = torch.ops.aten.view.default(model_audio_tower_layers_8_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_8_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_476 = torch.ops.aten._assert_tensor_metadata.default(view_825, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_476 = None
	        convert_element_type_316 = torch.ops.prims.convert_element_type.default(view_825, torch.float32);  view_825 = None
	        _assert_tensor_metadata_477 = torch.ops.aten._assert_tensor_metadata.default(view_827, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_477 = None
	        convert_element_type_317 = torch.ops.prims.convert_element_type.default(view_827, torch.float32);  view_827 = None
	        sub_2424 = torch.ops.aten.sub.Tensor(convert_element_type_316, convert_element_type_317);  convert_element_type_316 = convert_element_type_317 = None
	        mul_5134 = torch.ops.aten.mul.Tensor(sub_2424, view_826);  sub_2424 = view_826 = None
	        view_828 = torch.ops.aten.view.default(mul_5134, [5120, 1280]);  mul_5134 = None
	        _assert_tensor_metadata_478 = torch.ops.aten._assert_tensor_metadata.default(view_828, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_478 = None
	        mul_5139 = sym_size_int * 1500
	        view_829 = torch.ops.aten.view.default(mul_5129, [mul_5139, 1280]);  mul_5129 = mul_5139 = None
	        permute_89 = torch.ops.aten.permute.default(view_828, [1, 0]);  view_828 = None
	        addmm_43 = torch.ops.aten.addmm.default(model_audio_tower_layers_8_fc1_bias, view_829, permute_89);  model_audio_tower_layers_8_fc1_bias = view_829 = permute_89 = None
	        view_830 = torch.ops.aten.view.default(addmm_43, [sym_size_int, 1500, 5120]);  addmm_43 = None
	        mul_5146 = torch.ops.aten.mul.Tensor(view_830, 0.5)
	        mul_5147 = torch.ops.aten.mul.Tensor(view_830, 0.7071067811865476);  view_830 = None
	        erf_10 = torch.ops.aten.erf.default(mul_5147);  mul_5147 = None
	        add_8142 = torch.ops.aten.add.Tensor(erf_10, 1);  erf_10 = None
	        mul_5148 = torch.ops.aten.mul.Tensor(mul_5146, add_8142);  mul_5146 = add_8142 = None
	        amin_53 = torch.ops.aten.amin.default(mul_5148, [2])
	        amax_53 = torch.ops.aten.amax.default(mul_5148, [2])
	        full_106 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_53 = torch.ops.aten.minimum.default(amin_53, full_106);  amin_53 = full_106 = None
	        full_107 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_53 = torch.ops.aten.maximum.default(amax_53, full_107);  amax_53 = full_107 = None
	        sub_2437 = torch.ops.aten.sub.Tensor(maximum_53, minimum_53);  maximum_53 = None
	        div_106 = torch.ops.aten.div.Tensor(sub_2437, 255.0);  sub_2437 = None
	        clamp_min_159 = torch.ops.aten.clamp_min.default(div_106, 1.1920928955078125e-07);  div_106 = None
	        div_107 = torch.ops.aten.div.Tensor(minimum_53, clamp_min_159);  minimum_53 = None
	        round_107 = torch.ops.aten.round.default(div_107);  div_107 = None
	        sub_2443 = torch.ops.aten.sub.Tensor(-128, round_107);  round_107 = None
	        clamp_min_160 = torch.ops.aten.clamp_min.default(sub_2443, -128);  sub_2443 = None
	        clamp_max_106 = torch.ops.aten.clamp_max.default(clamp_min_160, 127);  clamp_min_160 = None
	        _assert_tensor_metadata_479 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_159, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_479 = None
	        _assert_tensor_metadata_480 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_106, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_480 = None
	        convert_element_type_318 = torch.ops.prims.convert_element_type.default(clamp_max_106, torch.int8);  clamp_max_106 = None
	        view_833 = torch.ops.aten.view.default(clamp_min_159, [sym_size_int, 1500, 1])
	        view_834 = torch.ops.aten.view.default(convert_element_type_318, [sym_size_int, 1500, 1])
	        reciprocal_53 = torch.ops.aten.reciprocal.default(view_833);  view_833 = None
	        mul_5194 = torch.ops.aten.mul.Tensor(reciprocal_53, 1.0);  reciprocal_53 = None
	        mul_5197 = torch.ops.aten.mul.Tensor(mul_5148, mul_5194);  mul_5148 = mul_5194 = None
	        round_108 = torch.ops.aten.round.default(mul_5197);  mul_5197 = None
	        add_8225 = torch.ops.aten.add.Tensor(round_108, view_834);  round_108 = view_834 = None
	        clamp_min_161 = torch.ops.aten.clamp_min.default(add_8225, -128);  add_8225 = None
	        clamp_max_107 = torch.ops.aten.clamp_max.default(clamp_min_161, 127);  clamp_min_161 = None
	        _assert_tensor_metadata_481 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_107, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_481 = None
	        convert_element_type_319 = torch.ops.prims.convert_element_type.default(clamp_max_107, torch.int8);  clamp_max_107 = None
	        view_837 = torch.ops.aten.view.default(clamp_min_159, [sym_size_int, 1500, 1]);  clamp_min_159 = None
	        view_838 = torch.ops.aten.view.default(convert_element_type_318, [sym_size_int, 1500, 1]);  convert_element_type_318 = None
	        _assert_tensor_metadata_482 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_319, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_482 = None
	        convert_element_type_320 = torch.ops.prims.convert_element_type.default(convert_element_type_319, torch.float32);  convert_element_type_319 = None
	        _assert_tensor_metadata_483 = torch.ops.aten._assert_tensor_metadata.default(view_838, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_483 = None
	        convert_element_type_321 = torch.ops.prims.convert_element_type.default(view_838, torch.float32);  view_838 = None
	        sub_2463 = torch.ops.aten.sub.Tensor(convert_element_type_320, convert_element_type_321);  convert_element_type_320 = convert_element_type_321 = None
	        mul_5219 = torch.ops.aten.mul.Tensor(sub_2463, view_837);  sub_2463 = view_837 = None
	        _assert_tensor_metadata_484 = torch.ops.aten._assert_tensor_metadata.default(mul_5219, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_484 = None
	        view_840 = torch.ops.aten.view.default(model_audio_tower_layers_8_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_8_fc2_parametrizations_weight_original0 = None
	        view_841 = torch.ops.aten.view.default(model_audio_tower_layers_8_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_8_fc2_parametrizations_weight_original1 = None
	        view_842 = torch.ops.aten.view.default(model_audio_tower_layers_8_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_8_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_485 = torch.ops.aten._assert_tensor_metadata.default(view_840, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_485 = None
	        convert_element_type_322 = torch.ops.prims.convert_element_type.default(view_840, torch.float32);  view_840 = None
	        _assert_tensor_metadata_486 = torch.ops.aten._assert_tensor_metadata.default(view_842, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_486 = None
	        convert_element_type_323 = torch.ops.prims.convert_element_type.default(view_842, torch.float32);  view_842 = None
	        sub_2467 = torch.ops.aten.sub.Tensor(convert_element_type_322, convert_element_type_323);  convert_element_type_322 = convert_element_type_323 = None
	        mul_5224 = torch.ops.aten.mul.Tensor(sub_2467, view_841);  sub_2467 = view_841 = None
	        view_843 = torch.ops.aten.view.default(mul_5224, [1280, 5120]);  mul_5224 = None
	        _assert_tensor_metadata_487 = torch.ops.aten._assert_tensor_metadata.default(view_843, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_487 = None
	        mul_5229 = sym_size_int * 1500
	        view_844 = torch.ops.aten.view.default(mul_5219, [mul_5229, 5120]);  mul_5219 = mul_5229 = None
	        permute_90 = torch.ops.aten.permute.default(view_843, [1, 0]);  view_843 = None
	        addmm_44 = torch.ops.aten.addmm.default(model_audio_tower_layers_8_fc2_bias, view_844, permute_90);  model_audio_tower_layers_8_fc2_bias = view_844 = permute_90 = None
	        view_845 = torch.ops.aten.view.default(addmm_44, [sym_size_int, 1500, 1280]);  addmm_44 = None
	        add_8288 = torch.ops.aten.add.Tensor(add_7990, view_845);  add_7990 = view_845 = None
	        clone_73 = torch.ops.aten.clone.default(add_8288, memory_format = torch.contiguous_format)
	        var_mean_18 = torch.ops.aten.var_mean.correction(clone_73, [2], correction = 0, keepdim = True)
	        getitem_72 = var_mean_18[0]
	        getitem_73 = var_mean_18[1];  var_mean_18 = None
	        add_8293 = torch.ops.aten.add.Tensor(getitem_72, 1e-05);  getitem_72 = None
	        rsqrt_18 = torch.ops.aten.rsqrt.default(add_8293);  add_8293 = None
	        sub_2473 = torch.ops.aten.sub.Tensor(clone_73, getitem_73);  clone_73 = getitem_73 = None
	        mul_5240 = torch.ops.aten.mul.Tensor(sub_2473, rsqrt_18);  sub_2473 = rsqrt_18 = None
	        mul_5241 = torch.ops.aten.mul.Tensor(mul_5240, model_audio_tower_layers_9_self_attn_layer_norm_weight);  mul_5240 = model_audio_tower_layers_9_self_attn_layer_norm_weight = None
	        add_8294 = torch.ops.aten.add.Tensor(mul_5241, model_audio_tower_layers_9_self_attn_layer_norm_bias);  mul_5241 = model_audio_tower_layers_9_self_attn_layer_norm_bias = None
	        amin_54 = torch.ops.aten.amin.default(add_8294, [2])
	        amax_54 = torch.ops.aten.amax.default(add_8294, [2])
	        full_108 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_54 = torch.ops.aten.minimum.default(amin_54, full_108);  amin_54 = full_108 = None
	        full_109 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_54 = torch.ops.aten.maximum.default(amax_54, full_109);  amax_54 = full_109 = None
	        sub_2484 = torch.ops.aten.sub.Tensor(maximum_54, minimum_54);  maximum_54 = None
	        div_108 = torch.ops.aten.div.Tensor(sub_2484, 255.0);  sub_2484 = None
	        clamp_min_162 = torch.ops.aten.clamp_min.default(div_108, 1.1920928955078125e-07);  div_108 = None
	        div_109 = torch.ops.aten.div.Tensor(minimum_54, clamp_min_162);  minimum_54 = None
	        round_109 = torch.ops.aten.round.default(div_109);  div_109 = None
	        sub_2490 = torch.ops.aten.sub.Tensor(-128, round_109);  round_109 = None
	        clamp_min_163 = torch.ops.aten.clamp_min.default(sub_2490, -128);  sub_2490 = None
	        clamp_max_108 = torch.ops.aten.clamp_max.default(clamp_min_163, 127);  clamp_min_163 = None
	        _assert_tensor_metadata_488 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_162, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_488 = None
	        _assert_tensor_metadata_489 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_108, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_489 = None
	        convert_element_type_324 = torch.ops.prims.convert_element_type.default(clamp_max_108, torch.int8);  clamp_max_108 = None
	        view_848 = torch.ops.aten.view.default(clamp_min_162, [sym_size_int, 1500, 1])
	        view_849 = torch.ops.aten.view.default(convert_element_type_324, [sym_size_int, 1500, 1])
	        reciprocal_54 = torch.ops.aten.reciprocal.default(view_848);  view_848 = None
	        mul_5289 = torch.ops.aten.mul.Tensor(reciprocal_54, 1.0);  reciprocal_54 = None
	        mul_5292 = torch.ops.aten.mul.Tensor(add_8294, mul_5289);  mul_5289 = None
	        round_110 = torch.ops.aten.round.default(mul_5292);  mul_5292 = None
	        add_8381 = torch.ops.aten.add.Tensor(round_110, view_849);  round_110 = view_849 = None
	        clamp_min_164 = torch.ops.aten.clamp_min.default(add_8381, -128);  add_8381 = None
	        clamp_max_109 = torch.ops.aten.clamp_max.default(clamp_min_164, 127);  clamp_min_164 = None
	        _assert_tensor_metadata_490 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_109, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_490 = None
	        convert_element_type_325 = torch.ops.prims.convert_element_type.default(clamp_max_109, torch.int8);  clamp_max_109 = None
	        view_852 = torch.ops.aten.view.default(clamp_min_162, [sym_size_int, 1500, 1]);  clamp_min_162 = None
	        view_853 = torch.ops.aten.view.default(convert_element_type_324, [sym_size_int, 1500, 1]);  convert_element_type_324 = None
	        _assert_tensor_metadata_491 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_325, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_491 = None
	        convert_element_type_326 = torch.ops.prims.convert_element_type.default(convert_element_type_325, torch.float32);  convert_element_type_325 = None
	        _assert_tensor_metadata_492 = torch.ops.aten._assert_tensor_metadata.default(view_853, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_492 = None
	        convert_element_type_327 = torch.ops.prims.convert_element_type.default(view_853, torch.float32);  view_853 = None
	        sub_2510 = torch.ops.aten.sub.Tensor(convert_element_type_326, convert_element_type_327);  convert_element_type_326 = convert_element_type_327 = None
	        mul_5314 = torch.ops.aten.mul.Tensor(sub_2510, view_852);  sub_2510 = view_852 = None
	        _assert_tensor_metadata_493 = torch.ops.aten._assert_tensor_metadata.default(mul_5314, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_493 = None
	        view_855 = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_856 = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_857 = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_494 = torch.ops.aten._assert_tensor_metadata.default(view_855, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_494 = None
	        convert_element_type_328 = torch.ops.prims.convert_element_type.default(view_855, torch.float32);  view_855 = None
	        _assert_tensor_metadata_495 = torch.ops.aten._assert_tensor_metadata.default(view_857, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_495 = None
	        convert_element_type_329 = torch.ops.prims.convert_element_type.default(view_857, torch.float32);  view_857 = None
	        sub_2514 = torch.ops.aten.sub.Tensor(convert_element_type_328, convert_element_type_329);  convert_element_type_328 = convert_element_type_329 = None
	        mul_5319 = torch.ops.aten.mul.Tensor(sub_2514, view_856);  sub_2514 = view_856 = None
	        view_858 = torch.ops.aten.view.default(mul_5319, [1280, 1280]);  mul_5319 = None
	        _assert_tensor_metadata_496 = torch.ops.aten._assert_tensor_metadata.default(view_858, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_496 = None
	        mul_5324 = sym_size_int * 1500
	        view_859 = torch.ops.aten.view.default(mul_5314, [mul_5324, 1280]);  mul_5314 = mul_5324 = None
	        permute_91 = torch.ops.aten.permute.default(view_858, [1, 0]);  view_858 = None
	        addmm_45 = torch.ops.aten.addmm.default(model_audio_tower_layers_9_self_attn_q_proj_bias, view_859, permute_91);  model_audio_tower_layers_9_self_attn_q_proj_bias = view_859 = permute_91 = None
	        view_860 = torch.ops.aten.view.default(addmm_45, [sym_size_int, 1500, 1280]);  addmm_45 = None
	        mul_5331 = torch.ops.aten.mul.Tensor(view_860, 0.125);  view_860 = None
	        view_861 = torch.ops.aten.view.default(mul_5331, [sym_size_int, 1500, 20, 64]);  mul_5331 = None
	        permute_92 = torch.ops.aten.permute.default(view_861, [0, 2, 1, 3]);  view_861 = None
	        clone_74 = torch.ops.aten.clone.default(permute_92, memory_format = torch.contiguous_format);  permute_92 = None
	        amin_55 = torch.ops.aten.amin.default(add_8294, [2])
	        amax_55 = torch.ops.aten.amax.default(add_8294, [2])
	        full_110 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_55 = torch.ops.aten.minimum.default(amin_55, full_110);  amin_55 = full_110 = None
	        full_111 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_55 = torch.ops.aten.maximum.default(amax_55, full_111);  amax_55 = full_111 = None
	        sub_2529 = torch.ops.aten.sub.Tensor(maximum_55, minimum_55);  maximum_55 = None
	        div_110 = torch.ops.aten.div.Tensor(sub_2529, 255.0);  sub_2529 = None
	        clamp_min_165 = torch.ops.aten.clamp_min.default(div_110, 1.1920928955078125e-07);  div_110 = None
	        div_111 = torch.ops.aten.div.Tensor(minimum_55, clamp_min_165);  minimum_55 = None
	        round_111 = torch.ops.aten.round.default(div_111);  div_111 = None
	        sub_2535 = torch.ops.aten.sub.Tensor(-128, round_111);  round_111 = None
	        clamp_min_166 = torch.ops.aten.clamp_min.default(sub_2535, -128);  sub_2535 = None
	        clamp_max_110 = torch.ops.aten.clamp_max.default(clamp_min_166, 127);  clamp_min_166 = None
	        _assert_tensor_metadata_497 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_165, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_497 = None
	        _assert_tensor_metadata_498 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_110, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_498 = None
	        convert_element_type_330 = torch.ops.prims.convert_element_type.default(clamp_max_110, torch.int8);  clamp_max_110 = None
	        view_864 = torch.ops.aten.view.default(clamp_min_165, [sym_size_int, 1500, 1])
	        view_865 = torch.ops.aten.view.default(convert_element_type_330, [sym_size_int, 1500, 1])
	        reciprocal_55 = torch.ops.aten.reciprocal.default(view_864);  view_864 = None
	        mul_5385 = torch.ops.aten.mul.Tensor(reciprocal_55, 1.0);  reciprocal_55 = None
	        mul_5388 = torch.ops.aten.mul.Tensor(add_8294, mul_5385);  mul_5385 = None
	        round_112 = torch.ops.aten.round.default(mul_5388);  mul_5388 = None
	        add_8533 = torch.ops.aten.add.Tensor(round_112, view_865);  round_112 = view_865 = None
	        clamp_min_167 = torch.ops.aten.clamp_min.default(add_8533, -128);  add_8533 = None
	        clamp_max_111 = torch.ops.aten.clamp_max.default(clamp_min_167, 127);  clamp_min_167 = None
	        _assert_tensor_metadata_499 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_111, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_499 = None
	        convert_element_type_331 = torch.ops.prims.convert_element_type.default(clamp_max_111, torch.int8);  clamp_max_111 = None
	        view_868 = torch.ops.aten.view.default(clamp_min_165, [sym_size_int, 1500, 1]);  clamp_min_165 = None
	        view_869 = torch.ops.aten.view.default(convert_element_type_330, [sym_size_int, 1500, 1]);  convert_element_type_330 = None
	        _assert_tensor_metadata_500 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_331, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_500 = None
	        convert_element_type_332 = torch.ops.prims.convert_element_type.default(convert_element_type_331, torch.float32);  convert_element_type_331 = None
	        _assert_tensor_metadata_501 = torch.ops.aten._assert_tensor_metadata.default(view_869, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_501 = None
	        convert_element_type_333 = torch.ops.prims.convert_element_type.default(view_869, torch.float32);  view_869 = None
	        sub_2555 = torch.ops.aten.sub.Tensor(convert_element_type_332, convert_element_type_333);  convert_element_type_332 = convert_element_type_333 = None
	        mul_5410 = torch.ops.aten.mul.Tensor(sub_2555, view_868);  sub_2555 = view_868 = None
	        _assert_tensor_metadata_502 = torch.ops.aten._assert_tensor_metadata.default(mul_5410, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_502 = None
	        view_871 = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_872 = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_873 = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_503 = torch.ops.aten._assert_tensor_metadata.default(view_871, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_503 = None
	        convert_element_type_334 = torch.ops.prims.convert_element_type.default(view_871, torch.float32);  view_871 = None
	        _assert_tensor_metadata_504 = torch.ops.aten._assert_tensor_metadata.default(view_873, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_504 = None
	        convert_element_type_335 = torch.ops.prims.convert_element_type.default(view_873, torch.float32);  view_873 = None
	        sub_2559 = torch.ops.aten.sub.Tensor(convert_element_type_334, convert_element_type_335);  convert_element_type_334 = convert_element_type_335 = None
	        mul_5415 = torch.ops.aten.mul.Tensor(sub_2559, view_872);  sub_2559 = view_872 = None
	        view_874 = torch.ops.aten.view.default(mul_5415, [1280, 1280]);  mul_5415 = None
	        _assert_tensor_metadata_505 = torch.ops.aten._assert_tensor_metadata.default(view_874, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_505 = None
	        permute_93 = torch.ops.aten.permute.default(view_874, [1, 0]);  view_874 = None
	        mul_5418 = sym_size_int * 1500
	        view_875 = torch.ops.aten.view.default(mul_5410, [mul_5418, 1280]);  mul_5410 = mul_5418 = None
	        mm_9 = torch.ops.aten.mm.default(view_875, permute_93);  view_875 = permute_93 = None
	        view_876 = torch.ops.aten.view.default(mm_9, [sym_size_int, 1500, 1280]);  mm_9 = None
	        view_877 = torch.ops.aten.view.default(view_876, [sym_size_int, -1, 20, 64]);  view_876 = None
	        permute_94 = torch.ops.aten.permute.default(view_877, [0, 2, 1, 3]);  view_877 = None
	        clone_75 = torch.ops.aten.clone.default(permute_94, memory_format = torch.contiguous_format);  permute_94 = None
	        amin_56 = torch.ops.aten.amin.default(add_8294, [2])
	        amax_56 = torch.ops.aten.amax.default(add_8294, [2])
	        full_112 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_56 = torch.ops.aten.minimum.default(amin_56, full_112);  amin_56 = full_112 = None
	        full_113 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_56 = torch.ops.aten.maximum.default(amax_56, full_113);  amax_56 = full_113 = None
	        sub_2573 = torch.ops.aten.sub.Tensor(maximum_56, minimum_56);  maximum_56 = None
	        div_112 = torch.ops.aten.div.Tensor(sub_2573, 255.0);  sub_2573 = None
	        clamp_min_168 = torch.ops.aten.clamp_min.default(div_112, 1.1920928955078125e-07);  div_112 = None
	        div_113 = torch.ops.aten.div.Tensor(minimum_56, clamp_min_168);  minimum_56 = None
	        round_113 = torch.ops.aten.round.default(div_113);  div_113 = None
	        sub_2579 = torch.ops.aten.sub.Tensor(-128, round_113);  round_113 = None
	        clamp_min_169 = torch.ops.aten.clamp_min.default(sub_2579, -128);  sub_2579 = None
	        clamp_max_112 = torch.ops.aten.clamp_max.default(clamp_min_169, 127);  clamp_min_169 = None
	        _assert_tensor_metadata_506 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_168, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_506 = None
	        _assert_tensor_metadata_507 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_112, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_507 = None
	        convert_element_type_336 = torch.ops.prims.convert_element_type.default(clamp_max_112, torch.int8);  clamp_max_112 = None
	        view_880 = torch.ops.aten.view.default(clamp_min_168, [sym_size_int, 1500, 1])
	        view_881 = torch.ops.aten.view.default(convert_element_type_336, [sym_size_int, 1500, 1])
	        reciprocal_56 = torch.ops.aten.reciprocal.default(view_880);  view_880 = None
	        mul_5484 = torch.ops.aten.mul.Tensor(reciprocal_56, 1.0);  reciprocal_56 = None
	        mul_5487 = torch.ops.aten.mul.Tensor(add_8294, mul_5484);  add_8294 = mul_5484 = None
	        round_114 = torch.ops.aten.round.default(mul_5487);  mul_5487 = None
	        add_8681 = torch.ops.aten.add.Tensor(round_114, view_881);  round_114 = view_881 = None
	        clamp_min_170 = torch.ops.aten.clamp_min.default(add_8681, -128);  add_8681 = None
	        clamp_max_113 = torch.ops.aten.clamp_max.default(clamp_min_170, 127);  clamp_min_170 = None
	        _assert_tensor_metadata_508 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_113, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_508 = None
	        convert_element_type_337 = torch.ops.prims.convert_element_type.default(clamp_max_113, torch.int8);  clamp_max_113 = None
	        view_884 = torch.ops.aten.view.default(clamp_min_168, [sym_size_int, 1500, 1]);  clamp_min_168 = None
	        view_885 = torch.ops.aten.view.default(convert_element_type_336, [sym_size_int, 1500, 1]);  convert_element_type_336 = None
	        _assert_tensor_metadata_509 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_337, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_509 = None
	        convert_element_type_338 = torch.ops.prims.convert_element_type.default(convert_element_type_337, torch.float32);  convert_element_type_337 = None
	        _assert_tensor_metadata_510 = torch.ops.aten._assert_tensor_metadata.default(view_885, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_510 = None
	        convert_element_type_339 = torch.ops.prims.convert_element_type.default(view_885, torch.float32);  view_885 = None
	        sub_2599 = torch.ops.aten.sub.Tensor(convert_element_type_338, convert_element_type_339);  convert_element_type_338 = convert_element_type_339 = None
	        mul_5509 = torch.ops.aten.mul.Tensor(sub_2599, view_884);  sub_2599 = view_884 = None
	        _assert_tensor_metadata_511 = torch.ops.aten._assert_tensor_metadata.default(mul_5509, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_511 = None
	        view_887 = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_888 = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_889 = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_512 = torch.ops.aten._assert_tensor_metadata.default(view_887, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_512 = None
	        convert_element_type_340 = torch.ops.prims.convert_element_type.default(view_887, torch.float32);  view_887 = None
	        _assert_tensor_metadata_513 = torch.ops.aten._assert_tensor_metadata.default(view_889, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_513 = None
	        convert_element_type_341 = torch.ops.prims.convert_element_type.default(view_889, torch.float32);  view_889 = None
	        sub_2603 = torch.ops.aten.sub.Tensor(convert_element_type_340, convert_element_type_341);  convert_element_type_340 = convert_element_type_341 = None
	        mul_5514 = torch.ops.aten.mul.Tensor(sub_2603, view_888);  sub_2603 = view_888 = None
	        view_890 = torch.ops.aten.view.default(mul_5514, [1280, 1280]);  mul_5514 = None
	        _assert_tensor_metadata_514 = torch.ops.aten._assert_tensor_metadata.default(view_890, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_514 = None
	        mul_5519 = sym_size_int * 1500
	        view_891 = torch.ops.aten.view.default(mul_5509, [mul_5519, 1280]);  mul_5509 = mul_5519 = None
	        permute_95 = torch.ops.aten.permute.default(view_890, [1, 0]);  view_890 = None
	        addmm_46 = torch.ops.aten.addmm.default(model_audio_tower_layers_9_self_attn_v_proj_bias, view_891, permute_95);  model_audio_tower_layers_9_self_attn_v_proj_bias = view_891 = permute_95 = None
	        view_892 = torch.ops.aten.view.default(addmm_46, [sym_size_int, 1500, 1280]);  addmm_46 = None
	        view_893 = torch.ops.aten.view.default(view_892, [sym_size_int, -1, 20, 64]);  view_892 = None
	        permute_96 = torch.ops.aten.permute.default(view_893, [0, 2, 1, 3]);  view_893 = None
	        clone_76 = torch.ops.aten.clone.default(permute_96, memory_format = torch.contiguous_format);  permute_96 = None
	        _scaled_dot_product_efficient_attention_9 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_74, clone_75, clone_76, None, False, scale = 1.0);  clone_74 = clone_75 = clone_76 = None
	        getitem_74 = _scaled_dot_product_efficient_attention_9[0];  _scaled_dot_product_efficient_attention_9 = None
	        permute_97 = torch.ops.aten.permute.default(getitem_74, [0, 2, 1, 3]);  getitem_74 = None
	        view_894 = torch.ops.aten.view.default(permute_97, [sym_size_int, 1500, -1]);  permute_97 = None
	        amin_57 = torch.ops.aten.amin.default(view_894, [2])
	        amax_57 = torch.ops.aten.amax.default(view_894, [2])
	        full_114 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_57 = torch.ops.aten.minimum.default(amin_57, full_114);  amin_57 = full_114 = None
	        full_115 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_57 = torch.ops.aten.maximum.default(amax_57, full_115);  amax_57 = full_115 = None
	        sub_2621 = torch.ops.aten.sub.Tensor(maximum_57, minimum_57);  maximum_57 = None
	        div_114 = torch.ops.aten.div.Tensor(sub_2621, 255.0);  sub_2621 = None
	        clamp_min_171 = torch.ops.aten.clamp_min.default(div_114, 1.1920928955078125e-07);  div_114 = None
	        div_115 = torch.ops.aten.div.Tensor(minimum_57, clamp_min_171);  minimum_57 = None
	        round_115 = torch.ops.aten.round.default(div_115);  div_115 = None
	        sub_2627 = torch.ops.aten.sub.Tensor(-128, round_115);  round_115 = None
	        clamp_min_172 = torch.ops.aten.clamp_min.default(sub_2627, -128);  sub_2627 = None
	        clamp_max_114 = torch.ops.aten.clamp_max.default(clamp_min_172, 127);  clamp_min_172 = None
	        _assert_tensor_metadata_515 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_171, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_515 = None
	        _assert_tensor_metadata_516 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_114, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_516 = None
	        convert_element_type_342 = torch.ops.prims.convert_element_type.default(clamp_max_114, torch.int8);  clamp_max_114 = None
	        view_897 = torch.ops.aten.view.default(clamp_min_171, [sym_size_int, 1500, 1])
	        view_898 = torch.ops.aten.view.default(convert_element_type_342, [sym_size_int, 1500, 1])
	        reciprocal_57 = torch.ops.aten.reciprocal.default(view_897);  view_897 = None
	        mul_5589 = torch.ops.aten.mul.Tensor(reciprocal_57, 1.0);  reciprocal_57 = None
	        mul_5592 = torch.ops.aten.mul.Tensor(view_894, mul_5589);  view_894 = mul_5589 = None
	        round_116 = torch.ops.aten.round.default(mul_5592);  mul_5592 = None
	        add_8845 = torch.ops.aten.add.Tensor(round_116, view_898);  round_116 = view_898 = None
	        clamp_min_173 = torch.ops.aten.clamp_min.default(add_8845, -128);  add_8845 = None
	        clamp_max_115 = torch.ops.aten.clamp_max.default(clamp_min_173, 127);  clamp_min_173 = None
	        _assert_tensor_metadata_517 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_115, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_517 = None
	        convert_element_type_343 = torch.ops.prims.convert_element_type.default(clamp_max_115, torch.int8);  clamp_max_115 = None
	        view_901 = torch.ops.aten.view.default(clamp_min_171, [sym_size_int, 1500, 1]);  clamp_min_171 = None
	        view_902 = torch.ops.aten.view.default(convert_element_type_342, [sym_size_int, 1500, 1]);  convert_element_type_342 = None
	        _assert_tensor_metadata_518 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_343, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_518 = None
	        convert_element_type_344 = torch.ops.prims.convert_element_type.default(convert_element_type_343, torch.float32);  convert_element_type_343 = None
	        _assert_tensor_metadata_519 = torch.ops.aten._assert_tensor_metadata.default(view_902, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_519 = None
	        convert_element_type_345 = torch.ops.prims.convert_element_type.default(view_902, torch.float32);  view_902 = None
	        sub_2647 = torch.ops.aten.sub.Tensor(convert_element_type_344, convert_element_type_345);  convert_element_type_344 = convert_element_type_345 = None
	        mul_5614 = torch.ops.aten.mul.Tensor(sub_2647, view_901);  sub_2647 = view_901 = None
	        _assert_tensor_metadata_520 = torch.ops.aten._assert_tensor_metadata.default(mul_5614, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_520 = None
	        view_904 = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_905 = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_906 = torch.ops.aten.view.default(model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_521 = torch.ops.aten._assert_tensor_metadata.default(view_904, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_521 = None
	        convert_element_type_346 = torch.ops.prims.convert_element_type.default(view_904, torch.float32);  view_904 = None
	        _assert_tensor_metadata_522 = torch.ops.aten._assert_tensor_metadata.default(view_906, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_522 = None
	        convert_element_type_347 = torch.ops.prims.convert_element_type.default(view_906, torch.float32);  view_906 = None
	        sub_2651 = torch.ops.aten.sub.Tensor(convert_element_type_346, convert_element_type_347);  convert_element_type_346 = convert_element_type_347 = None
	        mul_5619 = torch.ops.aten.mul.Tensor(sub_2651, view_905);  sub_2651 = view_905 = None
	        view_907 = torch.ops.aten.view.default(mul_5619, [1280, 1280]);  mul_5619 = None
	        _assert_tensor_metadata_523 = torch.ops.aten._assert_tensor_metadata.default(view_907, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_523 = None
	        mul_5624 = sym_size_int * 1500
	        view_908 = torch.ops.aten.view.default(mul_5614, [mul_5624, 1280]);  mul_5614 = mul_5624 = None
	        permute_98 = torch.ops.aten.permute.default(view_907, [1, 0]);  view_907 = None
	        addmm_47 = torch.ops.aten.addmm.default(model_audio_tower_layers_9_self_attn_out_proj_bias, view_908, permute_98);  model_audio_tower_layers_9_self_attn_out_proj_bias = view_908 = permute_98 = None
	        view_909 = torch.ops.aten.view.default(addmm_47, [sym_size_int, 1500, 1280]);  addmm_47 = None
	        add_8908 = torch.ops.aten.add.Tensor(add_8288, view_909);  add_8288 = view_909 = None
	        clone_78 = torch.ops.aten.clone.default(add_8908, memory_format = torch.contiguous_format)
	        var_mean_19 = torch.ops.aten.var_mean.correction(clone_78, [2], correction = 0, keepdim = True)
	        getitem_78 = var_mean_19[0]
	        getitem_79 = var_mean_19[1];  var_mean_19 = None
	        add_8913 = torch.ops.aten.add.Tensor(getitem_78, 1e-05);  getitem_78 = None
	        rsqrt_19 = torch.ops.aten.rsqrt.default(add_8913);  add_8913 = None
	        sub_2657 = torch.ops.aten.sub.Tensor(clone_78, getitem_79);  clone_78 = getitem_79 = None
	        mul_5635 = torch.ops.aten.mul.Tensor(sub_2657, rsqrt_19);  sub_2657 = rsqrt_19 = None
	        mul_5636 = torch.ops.aten.mul.Tensor(mul_5635, model_audio_tower_layers_9_final_layer_norm_weight);  mul_5635 = model_audio_tower_layers_9_final_layer_norm_weight = None
	        add_8914 = torch.ops.aten.add.Tensor(mul_5636, model_audio_tower_layers_9_final_layer_norm_bias);  mul_5636 = model_audio_tower_layers_9_final_layer_norm_bias = None
	        amin_58 = torch.ops.aten.amin.default(add_8914, [2])
	        amax_58 = torch.ops.aten.amax.default(add_8914, [2])
	        full_116 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_58 = torch.ops.aten.minimum.default(amin_58, full_116);  amin_58 = full_116 = None
	        full_117 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_58 = torch.ops.aten.maximum.default(amax_58, full_117);  amax_58 = full_117 = None
	        sub_2668 = torch.ops.aten.sub.Tensor(maximum_58, minimum_58);  maximum_58 = None
	        div_116 = torch.ops.aten.div.Tensor(sub_2668, 255.0);  sub_2668 = None
	        clamp_min_174 = torch.ops.aten.clamp_min.default(div_116, 1.1920928955078125e-07);  div_116 = None
	        div_117 = torch.ops.aten.div.Tensor(minimum_58, clamp_min_174);  minimum_58 = None
	        round_117 = torch.ops.aten.round.default(div_117);  div_117 = None
	        sub_2674 = torch.ops.aten.sub.Tensor(-128, round_117);  round_117 = None
	        clamp_min_175 = torch.ops.aten.clamp_min.default(sub_2674, -128);  sub_2674 = None
	        clamp_max_116 = torch.ops.aten.clamp_max.default(clamp_min_175, 127);  clamp_min_175 = None
	        _assert_tensor_metadata_524 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_174, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_524 = None
	        _assert_tensor_metadata_525 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_116, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_525 = None
	        convert_element_type_348 = torch.ops.prims.convert_element_type.default(clamp_max_116, torch.int8);  clamp_max_116 = None
	        view_912 = torch.ops.aten.view.default(clamp_min_174, [sym_size_int, 1500, 1])
	        view_913 = torch.ops.aten.view.default(convert_element_type_348, [sym_size_int, 1500, 1])
	        reciprocal_58 = torch.ops.aten.reciprocal.default(view_912);  view_912 = None
	        mul_5684 = torch.ops.aten.mul.Tensor(reciprocal_58, 1.0);  reciprocal_58 = None
	        mul_5687 = torch.ops.aten.mul.Tensor(add_8914, mul_5684);  add_8914 = mul_5684 = None
	        round_118 = torch.ops.aten.round.default(mul_5687);  mul_5687 = None
	        add_9001 = torch.ops.aten.add.Tensor(round_118, view_913);  round_118 = view_913 = None
	        clamp_min_176 = torch.ops.aten.clamp_min.default(add_9001, -128);  add_9001 = None
	        clamp_max_117 = torch.ops.aten.clamp_max.default(clamp_min_176, 127);  clamp_min_176 = None
	        _assert_tensor_metadata_526 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_117, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_526 = None
	        convert_element_type_349 = torch.ops.prims.convert_element_type.default(clamp_max_117, torch.int8);  clamp_max_117 = None
	        view_916 = torch.ops.aten.view.default(clamp_min_174, [sym_size_int, 1500, 1]);  clamp_min_174 = None
	        view_917 = torch.ops.aten.view.default(convert_element_type_348, [sym_size_int, 1500, 1]);  convert_element_type_348 = None
	        _assert_tensor_metadata_527 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_349, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_527 = None
	        convert_element_type_350 = torch.ops.prims.convert_element_type.default(convert_element_type_349, torch.float32);  convert_element_type_349 = None
	        _assert_tensor_metadata_528 = torch.ops.aten._assert_tensor_metadata.default(view_917, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_528 = None
	        convert_element_type_351 = torch.ops.prims.convert_element_type.default(view_917, torch.float32);  view_917 = None
	        sub_2694 = torch.ops.aten.sub.Tensor(convert_element_type_350, convert_element_type_351);  convert_element_type_350 = convert_element_type_351 = None
	        mul_5709 = torch.ops.aten.mul.Tensor(sub_2694, view_916);  sub_2694 = view_916 = None
	        _assert_tensor_metadata_529 = torch.ops.aten._assert_tensor_metadata.default(mul_5709, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_529 = None
	        view_919 = torch.ops.aten.view.default(model_audio_tower_layers_9_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_9_fc1_parametrizations_weight_original0 = None
	        view_920 = torch.ops.aten.view.default(model_audio_tower_layers_9_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_9_fc1_parametrizations_weight_original1 = None
	        view_921 = torch.ops.aten.view.default(model_audio_tower_layers_9_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_9_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_530 = torch.ops.aten._assert_tensor_metadata.default(view_919, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_530 = None
	        convert_element_type_352 = torch.ops.prims.convert_element_type.default(view_919, torch.float32);  view_919 = None
	        _assert_tensor_metadata_531 = torch.ops.aten._assert_tensor_metadata.default(view_921, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_531 = None
	        convert_element_type_353 = torch.ops.prims.convert_element_type.default(view_921, torch.float32);  view_921 = None
	        sub_2698 = torch.ops.aten.sub.Tensor(convert_element_type_352, convert_element_type_353);  convert_element_type_352 = convert_element_type_353 = None
	        mul_5714 = torch.ops.aten.mul.Tensor(sub_2698, view_920);  sub_2698 = view_920 = None
	        view_922 = torch.ops.aten.view.default(mul_5714, [5120, 1280]);  mul_5714 = None
	        _assert_tensor_metadata_532 = torch.ops.aten._assert_tensor_metadata.default(view_922, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_532 = None
	        mul_5719 = sym_size_int * 1500
	        view_923 = torch.ops.aten.view.default(mul_5709, [mul_5719, 1280]);  mul_5709 = mul_5719 = None
	        permute_99 = torch.ops.aten.permute.default(view_922, [1, 0]);  view_922 = None
	        addmm_48 = torch.ops.aten.addmm.default(model_audio_tower_layers_9_fc1_bias, view_923, permute_99);  model_audio_tower_layers_9_fc1_bias = view_923 = permute_99 = None
	        view_924 = torch.ops.aten.view.default(addmm_48, [sym_size_int, 1500, 5120]);  addmm_48 = None
	        mul_5726 = torch.ops.aten.mul.Tensor(view_924, 0.5)
	        mul_5727 = torch.ops.aten.mul.Tensor(view_924, 0.7071067811865476);  view_924 = None
	        erf_11 = torch.ops.aten.erf.default(mul_5727);  mul_5727 = None
	        add_9060 = torch.ops.aten.add.Tensor(erf_11, 1);  erf_11 = None
	        mul_5728 = torch.ops.aten.mul.Tensor(mul_5726, add_9060);  mul_5726 = add_9060 = None
	        amin_59 = torch.ops.aten.amin.default(mul_5728, [2])
	        amax_59 = torch.ops.aten.amax.default(mul_5728, [2])
	        full_118 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_59 = torch.ops.aten.minimum.default(amin_59, full_118);  amin_59 = full_118 = None
	        full_119 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_59 = torch.ops.aten.maximum.default(amax_59, full_119);  amax_59 = full_119 = None
	        sub_2711 = torch.ops.aten.sub.Tensor(maximum_59, minimum_59);  maximum_59 = None
	        div_118 = torch.ops.aten.div.Tensor(sub_2711, 255.0);  sub_2711 = None
	        clamp_min_177 = torch.ops.aten.clamp_min.default(div_118, 1.1920928955078125e-07);  div_118 = None
	        div_119 = torch.ops.aten.div.Tensor(minimum_59, clamp_min_177);  minimum_59 = None
	        round_119 = torch.ops.aten.round.default(div_119);  div_119 = None
	        sub_2717 = torch.ops.aten.sub.Tensor(-128, round_119);  round_119 = None
	        clamp_min_178 = torch.ops.aten.clamp_min.default(sub_2717, -128);  sub_2717 = None
	        clamp_max_118 = torch.ops.aten.clamp_max.default(clamp_min_178, 127);  clamp_min_178 = None
	        _assert_tensor_metadata_533 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_177, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_533 = None
	        _assert_tensor_metadata_534 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_118, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_534 = None
	        convert_element_type_354 = torch.ops.prims.convert_element_type.default(clamp_max_118, torch.int8);  clamp_max_118 = None
	        view_927 = torch.ops.aten.view.default(clamp_min_177, [sym_size_int, 1500, 1])
	        view_928 = torch.ops.aten.view.default(convert_element_type_354, [sym_size_int, 1500, 1])
	        reciprocal_59 = torch.ops.aten.reciprocal.default(view_927);  view_927 = None
	        mul_5774 = torch.ops.aten.mul.Tensor(reciprocal_59, 1.0);  reciprocal_59 = None
	        mul_5777 = torch.ops.aten.mul.Tensor(mul_5728, mul_5774);  mul_5728 = mul_5774 = None
	        round_120 = torch.ops.aten.round.default(mul_5777);  mul_5777 = None
	        add_9143 = torch.ops.aten.add.Tensor(round_120, view_928);  round_120 = view_928 = None
	        clamp_min_179 = torch.ops.aten.clamp_min.default(add_9143, -128);  add_9143 = None
	        clamp_max_119 = torch.ops.aten.clamp_max.default(clamp_min_179, 127);  clamp_min_179 = None
	        _assert_tensor_metadata_535 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_119, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_535 = None
	        convert_element_type_355 = torch.ops.prims.convert_element_type.default(clamp_max_119, torch.int8);  clamp_max_119 = None
	        view_931 = torch.ops.aten.view.default(clamp_min_177, [sym_size_int, 1500, 1]);  clamp_min_177 = None
	        view_932 = torch.ops.aten.view.default(convert_element_type_354, [sym_size_int, 1500, 1]);  convert_element_type_354 = None
	        _assert_tensor_metadata_536 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_355, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_536 = None
	        convert_element_type_356 = torch.ops.prims.convert_element_type.default(convert_element_type_355, torch.float32);  convert_element_type_355 = None
	        _assert_tensor_metadata_537 = torch.ops.aten._assert_tensor_metadata.default(view_932, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_537 = None
	        convert_element_type_357 = torch.ops.prims.convert_element_type.default(view_932, torch.float32);  view_932 = None
	        sub_2737 = torch.ops.aten.sub.Tensor(convert_element_type_356, convert_element_type_357);  convert_element_type_356 = convert_element_type_357 = None
	        mul_5799 = torch.ops.aten.mul.Tensor(sub_2737, view_931);  sub_2737 = view_931 = None
	        _assert_tensor_metadata_538 = torch.ops.aten._assert_tensor_metadata.default(mul_5799, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_538 = None
	        view_934 = torch.ops.aten.view.default(model_audio_tower_layers_9_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_9_fc2_parametrizations_weight_original0 = None
	        view_935 = torch.ops.aten.view.default(model_audio_tower_layers_9_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_9_fc2_parametrizations_weight_original1 = None
	        view_936 = torch.ops.aten.view.default(model_audio_tower_layers_9_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_9_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_539 = torch.ops.aten._assert_tensor_metadata.default(view_934, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_539 = None
	        convert_element_type_358 = torch.ops.prims.convert_element_type.default(view_934, torch.float32);  view_934 = None
	        _assert_tensor_metadata_540 = torch.ops.aten._assert_tensor_metadata.default(view_936, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_540 = None
	        convert_element_type_359 = torch.ops.prims.convert_element_type.default(view_936, torch.float32);  view_936 = None
	        sub_2741 = torch.ops.aten.sub.Tensor(convert_element_type_358, convert_element_type_359);  convert_element_type_358 = convert_element_type_359 = None
	        mul_5804 = torch.ops.aten.mul.Tensor(sub_2741, view_935);  sub_2741 = view_935 = None
	        view_937 = torch.ops.aten.view.default(mul_5804, [1280, 5120]);  mul_5804 = None
	        _assert_tensor_metadata_541 = torch.ops.aten._assert_tensor_metadata.default(view_937, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_541 = None
	        mul_5809 = sym_size_int * 1500
	        view_938 = torch.ops.aten.view.default(mul_5799, [mul_5809, 5120]);  mul_5799 = mul_5809 = None
	        permute_100 = torch.ops.aten.permute.default(view_937, [1, 0]);  view_937 = None
	        addmm_49 = torch.ops.aten.addmm.default(model_audio_tower_layers_9_fc2_bias, view_938, permute_100);  model_audio_tower_layers_9_fc2_bias = view_938 = permute_100 = None
	        view_939 = torch.ops.aten.view.default(addmm_49, [sym_size_int, 1500, 1280]);  addmm_49 = None
	        add_9206 = torch.ops.aten.add.Tensor(add_8908, view_939);  add_8908 = view_939 = None
	        clone_81 = torch.ops.aten.clone.default(add_9206, memory_format = torch.contiguous_format)
	        var_mean_20 = torch.ops.aten.var_mean.correction(clone_81, [2], correction = 0, keepdim = True)
	        getitem_80 = var_mean_20[0]
	        getitem_81 = var_mean_20[1];  var_mean_20 = None
	        add_9211 = torch.ops.aten.add.Tensor(getitem_80, 1e-05);  getitem_80 = None
	        rsqrt_20 = torch.ops.aten.rsqrt.default(add_9211);  add_9211 = None
	        sub_2747 = torch.ops.aten.sub.Tensor(clone_81, getitem_81);  clone_81 = getitem_81 = None
	        mul_5820 = torch.ops.aten.mul.Tensor(sub_2747, rsqrt_20);  sub_2747 = rsqrt_20 = None
	        mul_5821 = torch.ops.aten.mul.Tensor(mul_5820, model_audio_tower_layers_10_self_attn_layer_norm_weight);  mul_5820 = model_audio_tower_layers_10_self_attn_layer_norm_weight = None
	        add_9212 = torch.ops.aten.add.Tensor(mul_5821, model_audio_tower_layers_10_self_attn_layer_norm_bias);  mul_5821 = model_audio_tower_layers_10_self_attn_layer_norm_bias = None
	        amin_60 = torch.ops.aten.amin.default(add_9212, [2])
	        amax_60 = torch.ops.aten.amax.default(add_9212, [2])
	        full_120 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_60 = torch.ops.aten.minimum.default(amin_60, full_120);  amin_60 = full_120 = None
	        full_121 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_60 = torch.ops.aten.maximum.default(amax_60, full_121);  amax_60 = full_121 = None
	        sub_2758 = torch.ops.aten.sub.Tensor(maximum_60, minimum_60);  maximum_60 = None
	        div_120 = torch.ops.aten.div.Tensor(sub_2758, 255.0);  sub_2758 = None
	        clamp_min_180 = torch.ops.aten.clamp_min.default(div_120, 1.1920928955078125e-07);  div_120 = None
	        div_121 = torch.ops.aten.div.Tensor(minimum_60, clamp_min_180);  minimum_60 = None
	        round_121 = torch.ops.aten.round.default(div_121);  div_121 = None
	        sub_2764 = torch.ops.aten.sub.Tensor(-128, round_121);  round_121 = None
	        clamp_min_181 = torch.ops.aten.clamp_min.default(sub_2764, -128);  sub_2764 = None
	        clamp_max_120 = torch.ops.aten.clamp_max.default(clamp_min_181, 127);  clamp_min_181 = None
	        _assert_tensor_metadata_542 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_180, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_542 = None
	        _assert_tensor_metadata_543 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_120, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_543 = None
	        convert_element_type_360 = torch.ops.prims.convert_element_type.default(clamp_max_120, torch.int8);  clamp_max_120 = None
	        view_942 = torch.ops.aten.view.default(clamp_min_180, [sym_size_int, 1500, 1])
	        view_943 = torch.ops.aten.view.default(convert_element_type_360, [sym_size_int, 1500, 1])
	        reciprocal_60 = torch.ops.aten.reciprocal.default(view_942);  view_942 = None
	        mul_5869 = torch.ops.aten.mul.Tensor(reciprocal_60, 1.0);  reciprocal_60 = None
	        mul_5872 = torch.ops.aten.mul.Tensor(add_9212, mul_5869);  mul_5869 = None
	        round_122 = torch.ops.aten.round.default(mul_5872);  mul_5872 = None
	        add_9299 = torch.ops.aten.add.Tensor(round_122, view_943);  round_122 = view_943 = None
	        clamp_min_182 = torch.ops.aten.clamp_min.default(add_9299, -128);  add_9299 = None
	        clamp_max_121 = torch.ops.aten.clamp_max.default(clamp_min_182, 127);  clamp_min_182 = None
	        _assert_tensor_metadata_544 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_121, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_544 = None
	        convert_element_type_361 = torch.ops.prims.convert_element_type.default(clamp_max_121, torch.int8);  clamp_max_121 = None
	        view_946 = torch.ops.aten.view.default(clamp_min_180, [sym_size_int, 1500, 1]);  clamp_min_180 = None
	        view_947 = torch.ops.aten.view.default(convert_element_type_360, [sym_size_int, 1500, 1]);  convert_element_type_360 = None
	        _assert_tensor_metadata_545 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_361, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_545 = None
	        convert_element_type_362 = torch.ops.prims.convert_element_type.default(convert_element_type_361, torch.float32);  convert_element_type_361 = None
	        _assert_tensor_metadata_546 = torch.ops.aten._assert_tensor_metadata.default(view_947, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_546 = None
	        convert_element_type_363 = torch.ops.prims.convert_element_type.default(view_947, torch.float32);  view_947 = None
	        sub_2784 = torch.ops.aten.sub.Tensor(convert_element_type_362, convert_element_type_363);  convert_element_type_362 = convert_element_type_363 = None
	        mul_5894 = torch.ops.aten.mul.Tensor(sub_2784, view_946);  sub_2784 = view_946 = None
	        _assert_tensor_metadata_547 = torch.ops.aten._assert_tensor_metadata.default(mul_5894, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_547 = None
	        view_949 = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_950 = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_951 = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_548 = torch.ops.aten._assert_tensor_metadata.default(view_949, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_548 = None
	        convert_element_type_364 = torch.ops.prims.convert_element_type.default(view_949, torch.float32);  view_949 = None
	        _assert_tensor_metadata_549 = torch.ops.aten._assert_tensor_metadata.default(view_951, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_549 = None
	        convert_element_type_365 = torch.ops.prims.convert_element_type.default(view_951, torch.float32);  view_951 = None
	        sub_2788 = torch.ops.aten.sub.Tensor(convert_element_type_364, convert_element_type_365);  convert_element_type_364 = convert_element_type_365 = None
	        mul_5899 = torch.ops.aten.mul.Tensor(sub_2788, view_950);  sub_2788 = view_950 = None
	        view_952 = torch.ops.aten.view.default(mul_5899, [1280, 1280]);  mul_5899 = None
	        _assert_tensor_metadata_550 = torch.ops.aten._assert_tensor_metadata.default(view_952, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_550 = None
	        mul_5904 = sym_size_int * 1500
	        view_953 = torch.ops.aten.view.default(mul_5894, [mul_5904, 1280]);  mul_5894 = mul_5904 = None
	        permute_101 = torch.ops.aten.permute.default(view_952, [1, 0]);  view_952 = None
	        addmm_50 = torch.ops.aten.addmm.default(model_audio_tower_layers_10_self_attn_q_proj_bias, view_953, permute_101);  model_audio_tower_layers_10_self_attn_q_proj_bias = view_953 = permute_101 = None
	        view_954 = torch.ops.aten.view.default(addmm_50, [sym_size_int, 1500, 1280]);  addmm_50 = None
	        mul_5911 = torch.ops.aten.mul.Tensor(view_954, 0.125);  view_954 = None
	        view_955 = torch.ops.aten.view.default(mul_5911, [sym_size_int, 1500, 20, 64]);  mul_5911 = None
	        permute_102 = torch.ops.aten.permute.default(view_955, [0, 2, 1, 3]);  view_955 = None
	        clone_82 = torch.ops.aten.clone.default(permute_102, memory_format = torch.contiguous_format);  permute_102 = None
	        amin_61 = torch.ops.aten.amin.default(add_9212, [2])
	        amax_61 = torch.ops.aten.amax.default(add_9212, [2])
	        full_122 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_61 = torch.ops.aten.minimum.default(amin_61, full_122);  amin_61 = full_122 = None
	        full_123 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_61 = torch.ops.aten.maximum.default(amax_61, full_123);  amax_61 = full_123 = None
	        sub_2803 = torch.ops.aten.sub.Tensor(maximum_61, minimum_61);  maximum_61 = None
	        div_122 = torch.ops.aten.div.Tensor(sub_2803, 255.0);  sub_2803 = None
	        clamp_min_183 = torch.ops.aten.clamp_min.default(div_122, 1.1920928955078125e-07);  div_122 = None
	        div_123 = torch.ops.aten.div.Tensor(minimum_61, clamp_min_183);  minimum_61 = None
	        round_123 = torch.ops.aten.round.default(div_123);  div_123 = None
	        sub_2809 = torch.ops.aten.sub.Tensor(-128, round_123);  round_123 = None
	        clamp_min_184 = torch.ops.aten.clamp_min.default(sub_2809, -128);  sub_2809 = None
	        clamp_max_122 = torch.ops.aten.clamp_max.default(clamp_min_184, 127);  clamp_min_184 = None
	        _assert_tensor_metadata_551 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_183, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_551 = None
	        _assert_tensor_metadata_552 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_122, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_552 = None
	        convert_element_type_366 = torch.ops.prims.convert_element_type.default(clamp_max_122, torch.int8);  clamp_max_122 = None
	        view_958 = torch.ops.aten.view.default(clamp_min_183, [sym_size_int, 1500, 1])
	        view_959 = torch.ops.aten.view.default(convert_element_type_366, [sym_size_int, 1500, 1])
	        reciprocal_61 = torch.ops.aten.reciprocal.default(view_958);  view_958 = None
	        mul_5965 = torch.ops.aten.mul.Tensor(reciprocal_61, 1.0);  reciprocal_61 = None
	        mul_5968 = torch.ops.aten.mul.Tensor(add_9212, mul_5965);  mul_5965 = None
	        round_124 = torch.ops.aten.round.default(mul_5968);  mul_5968 = None
	        add_9451 = torch.ops.aten.add.Tensor(round_124, view_959);  round_124 = view_959 = None
	        clamp_min_185 = torch.ops.aten.clamp_min.default(add_9451, -128);  add_9451 = None
	        clamp_max_123 = torch.ops.aten.clamp_max.default(clamp_min_185, 127);  clamp_min_185 = None
	        _assert_tensor_metadata_553 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_123, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_553 = None
	        convert_element_type_367 = torch.ops.prims.convert_element_type.default(clamp_max_123, torch.int8);  clamp_max_123 = None
	        view_962 = torch.ops.aten.view.default(clamp_min_183, [sym_size_int, 1500, 1]);  clamp_min_183 = None
	        view_963 = torch.ops.aten.view.default(convert_element_type_366, [sym_size_int, 1500, 1]);  convert_element_type_366 = None
	        _assert_tensor_metadata_554 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_367, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_554 = None
	        convert_element_type_368 = torch.ops.prims.convert_element_type.default(convert_element_type_367, torch.float32);  convert_element_type_367 = None
	        _assert_tensor_metadata_555 = torch.ops.aten._assert_tensor_metadata.default(view_963, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_555 = None
	        convert_element_type_369 = torch.ops.prims.convert_element_type.default(view_963, torch.float32);  view_963 = None
	        sub_2829 = torch.ops.aten.sub.Tensor(convert_element_type_368, convert_element_type_369);  convert_element_type_368 = convert_element_type_369 = None
	        mul_5990 = torch.ops.aten.mul.Tensor(sub_2829, view_962);  sub_2829 = view_962 = None
	        _assert_tensor_metadata_556 = torch.ops.aten._assert_tensor_metadata.default(mul_5990, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_556 = None
	        view_965 = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_966 = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_967 = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_557 = torch.ops.aten._assert_tensor_metadata.default(view_965, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_557 = None
	        convert_element_type_370 = torch.ops.prims.convert_element_type.default(view_965, torch.float32);  view_965 = None
	        _assert_tensor_metadata_558 = torch.ops.aten._assert_tensor_metadata.default(view_967, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_558 = None
	        convert_element_type_371 = torch.ops.prims.convert_element_type.default(view_967, torch.float32);  view_967 = None
	        sub_2833 = torch.ops.aten.sub.Tensor(convert_element_type_370, convert_element_type_371);  convert_element_type_370 = convert_element_type_371 = None
	        mul_5995 = torch.ops.aten.mul.Tensor(sub_2833, view_966);  sub_2833 = view_966 = None
	        view_968 = torch.ops.aten.view.default(mul_5995, [1280, 1280]);  mul_5995 = None
	        _assert_tensor_metadata_559 = torch.ops.aten._assert_tensor_metadata.default(view_968, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_559 = None
	        permute_103 = torch.ops.aten.permute.default(view_968, [1, 0]);  view_968 = None
	        mul_5998 = sym_size_int * 1500
	        view_969 = torch.ops.aten.view.default(mul_5990, [mul_5998, 1280]);  mul_5990 = mul_5998 = None
	        mm_10 = torch.ops.aten.mm.default(view_969, permute_103);  view_969 = permute_103 = None
	        view_970 = torch.ops.aten.view.default(mm_10, [sym_size_int, 1500, 1280]);  mm_10 = None
	        view_971 = torch.ops.aten.view.default(view_970, [sym_size_int, -1, 20, 64]);  view_970 = None
	        permute_104 = torch.ops.aten.permute.default(view_971, [0, 2, 1, 3]);  view_971 = None
	        clone_83 = torch.ops.aten.clone.default(permute_104, memory_format = torch.contiguous_format);  permute_104 = None
	        amin_62 = torch.ops.aten.amin.default(add_9212, [2])
	        amax_62 = torch.ops.aten.amax.default(add_9212, [2])
	        full_124 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_62 = torch.ops.aten.minimum.default(amin_62, full_124);  amin_62 = full_124 = None
	        full_125 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_62 = torch.ops.aten.maximum.default(amax_62, full_125);  amax_62 = full_125 = None
	        sub_2847 = torch.ops.aten.sub.Tensor(maximum_62, minimum_62);  maximum_62 = None
	        div_124 = torch.ops.aten.div.Tensor(sub_2847, 255.0);  sub_2847 = None
	        clamp_min_186 = torch.ops.aten.clamp_min.default(div_124, 1.1920928955078125e-07);  div_124 = None
	        div_125 = torch.ops.aten.div.Tensor(minimum_62, clamp_min_186);  minimum_62 = None
	        round_125 = torch.ops.aten.round.default(div_125);  div_125 = None
	        sub_2853 = torch.ops.aten.sub.Tensor(-128, round_125);  round_125 = None
	        clamp_min_187 = torch.ops.aten.clamp_min.default(sub_2853, -128);  sub_2853 = None
	        clamp_max_124 = torch.ops.aten.clamp_max.default(clamp_min_187, 127);  clamp_min_187 = None
	        _assert_tensor_metadata_560 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_186, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_560 = None
	        _assert_tensor_metadata_561 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_124, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_561 = None
	        convert_element_type_372 = torch.ops.prims.convert_element_type.default(clamp_max_124, torch.int8);  clamp_max_124 = None
	        view_974 = torch.ops.aten.view.default(clamp_min_186, [sym_size_int, 1500, 1])
	        view_975 = torch.ops.aten.view.default(convert_element_type_372, [sym_size_int, 1500, 1])
	        reciprocal_62 = torch.ops.aten.reciprocal.default(view_974);  view_974 = None
	        mul_6064 = torch.ops.aten.mul.Tensor(reciprocal_62, 1.0);  reciprocal_62 = None
	        mul_6067 = torch.ops.aten.mul.Tensor(add_9212, mul_6064);  add_9212 = mul_6064 = None
	        round_126 = torch.ops.aten.round.default(mul_6067);  mul_6067 = None
	        add_9599 = torch.ops.aten.add.Tensor(round_126, view_975);  round_126 = view_975 = None
	        clamp_min_188 = torch.ops.aten.clamp_min.default(add_9599, -128);  add_9599 = None
	        clamp_max_125 = torch.ops.aten.clamp_max.default(clamp_min_188, 127);  clamp_min_188 = None
	        _assert_tensor_metadata_562 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_125, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_562 = None
	        convert_element_type_373 = torch.ops.prims.convert_element_type.default(clamp_max_125, torch.int8);  clamp_max_125 = None
	        view_978 = torch.ops.aten.view.default(clamp_min_186, [sym_size_int, 1500, 1]);  clamp_min_186 = None
	        view_979 = torch.ops.aten.view.default(convert_element_type_372, [sym_size_int, 1500, 1]);  convert_element_type_372 = None
	        _assert_tensor_metadata_563 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_373, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_563 = None
	        convert_element_type_374 = torch.ops.prims.convert_element_type.default(convert_element_type_373, torch.float32);  convert_element_type_373 = None
	        _assert_tensor_metadata_564 = torch.ops.aten._assert_tensor_metadata.default(view_979, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_564 = None
	        convert_element_type_375 = torch.ops.prims.convert_element_type.default(view_979, torch.float32);  view_979 = None
	        sub_2873 = torch.ops.aten.sub.Tensor(convert_element_type_374, convert_element_type_375);  convert_element_type_374 = convert_element_type_375 = None
	        mul_6089 = torch.ops.aten.mul.Tensor(sub_2873, view_978);  sub_2873 = view_978 = None
	        _assert_tensor_metadata_565 = torch.ops.aten._assert_tensor_metadata.default(mul_6089, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_565 = None
	        view_981 = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_982 = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_983 = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_566 = torch.ops.aten._assert_tensor_metadata.default(view_981, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_566 = None
	        convert_element_type_376 = torch.ops.prims.convert_element_type.default(view_981, torch.float32);  view_981 = None
	        _assert_tensor_metadata_567 = torch.ops.aten._assert_tensor_metadata.default(view_983, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_567 = None
	        convert_element_type_377 = torch.ops.prims.convert_element_type.default(view_983, torch.float32);  view_983 = None
	        sub_2877 = torch.ops.aten.sub.Tensor(convert_element_type_376, convert_element_type_377);  convert_element_type_376 = convert_element_type_377 = None
	        mul_6094 = torch.ops.aten.mul.Tensor(sub_2877, view_982);  sub_2877 = view_982 = None
	        view_984 = torch.ops.aten.view.default(mul_6094, [1280, 1280]);  mul_6094 = None
	        _assert_tensor_metadata_568 = torch.ops.aten._assert_tensor_metadata.default(view_984, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_568 = None
	        mul_6099 = sym_size_int * 1500
	        view_985 = torch.ops.aten.view.default(mul_6089, [mul_6099, 1280]);  mul_6089 = mul_6099 = None
	        permute_105 = torch.ops.aten.permute.default(view_984, [1, 0]);  view_984 = None
	        addmm_51 = torch.ops.aten.addmm.default(model_audio_tower_layers_10_self_attn_v_proj_bias, view_985, permute_105);  model_audio_tower_layers_10_self_attn_v_proj_bias = view_985 = permute_105 = None
	        view_986 = torch.ops.aten.view.default(addmm_51, [sym_size_int, 1500, 1280]);  addmm_51 = None
	        view_987 = torch.ops.aten.view.default(view_986, [sym_size_int, -1, 20, 64]);  view_986 = None
	        permute_106 = torch.ops.aten.permute.default(view_987, [0, 2, 1, 3]);  view_987 = None
	        clone_84 = torch.ops.aten.clone.default(permute_106, memory_format = torch.contiguous_format);  permute_106 = None
	        _scaled_dot_product_efficient_attention_10 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_82, clone_83, clone_84, None, False, scale = 1.0);  clone_82 = clone_83 = clone_84 = None
	        getitem_82 = _scaled_dot_product_efficient_attention_10[0];  _scaled_dot_product_efficient_attention_10 = None
	        permute_107 = torch.ops.aten.permute.default(getitem_82, [0, 2, 1, 3]);  getitem_82 = None
	        view_988 = torch.ops.aten.view.default(permute_107, [sym_size_int, 1500, -1]);  permute_107 = None
	        amin_63 = torch.ops.aten.amin.default(view_988, [2])
	        amax_63 = torch.ops.aten.amax.default(view_988, [2])
	        full_126 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_63 = torch.ops.aten.minimum.default(amin_63, full_126);  amin_63 = full_126 = None
	        full_127 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_63 = torch.ops.aten.maximum.default(amax_63, full_127);  amax_63 = full_127 = None
	        sub_2895 = torch.ops.aten.sub.Tensor(maximum_63, minimum_63);  maximum_63 = None
	        div_126 = torch.ops.aten.div.Tensor(sub_2895, 255.0);  sub_2895 = None
	        clamp_min_189 = torch.ops.aten.clamp_min.default(div_126, 1.1920928955078125e-07);  div_126 = None
	        div_127 = torch.ops.aten.div.Tensor(minimum_63, clamp_min_189);  minimum_63 = None
	        round_127 = torch.ops.aten.round.default(div_127);  div_127 = None
	        sub_2901 = torch.ops.aten.sub.Tensor(-128, round_127);  round_127 = None
	        clamp_min_190 = torch.ops.aten.clamp_min.default(sub_2901, -128);  sub_2901 = None
	        clamp_max_126 = torch.ops.aten.clamp_max.default(clamp_min_190, 127);  clamp_min_190 = None
	        _assert_tensor_metadata_569 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_189, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_569 = None
	        _assert_tensor_metadata_570 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_126, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_570 = None
	        convert_element_type_378 = torch.ops.prims.convert_element_type.default(clamp_max_126, torch.int8);  clamp_max_126 = None
	        view_991 = torch.ops.aten.view.default(clamp_min_189, [sym_size_int, 1500, 1])
	        view_992 = torch.ops.aten.view.default(convert_element_type_378, [sym_size_int, 1500, 1])
	        reciprocal_63 = torch.ops.aten.reciprocal.default(view_991);  view_991 = None
	        mul_6169 = torch.ops.aten.mul.Tensor(reciprocal_63, 1.0);  reciprocal_63 = None
	        mul_6172 = torch.ops.aten.mul.Tensor(view_988, mul_6169);  view_988 = mul_6169 = None
	        round_128 = torch.ops.aten.round.default(mul_6172);  mul_6172 = None
	        add_9763 = torch.ops.aten.add.Tensor(round_128, view_992);  round_128 = view_992 = None
	        clamp_min_191 = torch.ops.aten.clamp_min.default(add_9763, -128);  add_9763 = None
	        clamp_max_127 = torch.ops.aten.clamp_max.default(clamp_min_191, 127);  clamp_min_191 = None
	        _assert_tensor_metadata_571 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_127, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_571 = None
	        convert_element_type_379 = torch.ops.prims.convert_element_type.default(clamp_max_127, torch.int8);  clamp_max_127 = None
	        view_995 = torch.ops.aten.view.default(clamp_min_189, [sym_size_int, 1500, 1]);  clamp_min_189 = None
	        view_996 = torch.ops.aten.view.default(convert_element_type_378, [sym_size_int, 1500, 1]);  convert_element_type_378 = None
	        _assert_tensor_metadata_572 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_379, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_572 = None
	        convert_element_type_380 = torch.ops.prims.convert_element_type.default(convert_element_type_379, torch.float32);  convert_element_type_379 = None
	        _assert_tensor_metadata_573 = torch.ops.aten._assert_tensor_metadata.default(view_996, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_573 = None
	        convert_element_type_381 = torch.ops.prims.convert_element_type.default(view_996, torch.float32);  view_996 = None
	        sub_2921 = torch.ops.aten.sub.Tensor(convert_element_type_380, convert_element_type_381);  convert_element_type_380 = convert_element_type_381 = None
	        mul_6194 = torch.ops.aten.mul.Tensor(sub_2921, view_995);  sub_2921 = view_995 = None
	        _assert_tensor_metadata_574 = torch.ops.aten._assert_tensor_metadata.default(mul_6194, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_574 = None
	        view_998 = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_999 = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1000 = torch.ops.aten.view.default(model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_575 = torch.ops.aten._assert_tensor_metadata.default(view_998, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_575 = None
	        convert_element_type_382 = torch.ops.prims.convert_element_type.default(view_998, torch.float32);  view_998 = None
	        _assert_tensor_metadata_576 = torch.ops.aten._assert_tensor_metadata.default(view_1000, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_576 = None
	        convert_element_type_383 = torch.ops.prims.convert_element_type.default(view_1000, torch.float32);  view_1000 = None
	        sub_2925 = torch.ops.aten.sub.Tensor(convert_element_type_382, convert_element_type_383);  convert_element_type_382 = convert_element_type_383 = None
	        mul_6199 = torch.ops.aten.mul.Tensor(sub_2925, view_999);  sub_2925 = view_999 = None
	        view_1001 = torch.ops.aten.view.default(mul_6199, [1280, 1280]);  mul_6199 = None
	        _assert_tensor_metadata_577 = torch.ops.aten._assert_tensor_metadata.default(view_1001, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_577 = None
	        mul_6204 = sym_size_int * 1500
	        view_1002 = torch.ops.aten.view.default(mul_6194, [mul_6204, 1280]);  mul_6194 = mul_6204 = None
	        permute_108 = torch.ops.aten.permute.default(view_1001, [1, 0]);  view_1001 = None
	        addmm_52 = torch.ops.aten.addmm.default(model_audio_tower_layers_10_self_attn_out_proj_bias, view_1002, permute_108);  model_audio_tower_layers_10_self_attn_out_proj_bias = view_1002 = permute_108 = None
	        view_1003 = torch.ops.aten.view.default(addmm_52, [sym_size_int, 1500, 1280]);  addmm_52 = None
	        add_9826 = torch.ops.aten.add.Tensor(add_9206, view_1003);  add_9206 = view_1003 = None
	        clone_86 = torch.ops.aten.clone.default(add_9826, memory_format = torch.contiguous_format)
	        var_mean_21 = torch.ops.aten.var_mean.correction(clone_86, [2], correction = 0, keepdim = True)
	        getitem_86 = var_mean_21[0]
	        getitem_87 = var_mean_21[1];  var_mean_21 = None
	        add_9831 = torch.ops.aten.add.Tensor(getitem_86, 1e-05);  getitem_86 = None
	        rsqrt_21 = torch.ops.aten.rsqrt.default(add_9831);  add_9831 = None
	        sub_2931 = torch.ops.aten.sub.Tensor(clone_86, getitem_87);  clone_86 = getitem_87 = None
	        mul_6215 = torch.ops.aten.mul.Tensor(sub_2931, rsqrt_21);  sub_2931 = rsqrt_21 = None
	        mul_6216 = torch.ops.aten.mul.Tensor(mul_6215, model_audio_tower_layers_10_final_layer_norm_weight);  mul_6215 = model_audio_tower_layers_10_final_layer_norm_weight = None
	        add_9832 = torch.ops.aten.add.Tensor(mul_6216, model_audio_tower_layers_10_final_layer_norm_bias);  mul_6216 = model_audio_tower_layers_10_final_layer_norm_bias = None
	        amin_64 = torch.ops.aten.amin.default(add_9832, [2])
	        amax_64 = torch.ops.aten.amax.default(add_9832, [2])
	        full_128 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_64 = torch.ops.aten.minimum.default(amin_64, full_128);  amin_64 = full_128 = None
	        full_129 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_64 = torch.ops.aten.maximum.default(amax_64, full_129);  amax_64 = full_129 = None
	        sub_2942 = torch.ops.aten.sub.Tensor(maximum_64, minimum_64);  maximum_64 = None
	        div_128 = torch.ops.aten.div.Tensor(sub_2942, 255.0);  sub_2942 = None
	        clamp_min_192 = torch.ops.aten.clamp_min.default(div_128, 1.1920928955078125e-07);  div_128 = None
	        div_129 = torch.ops.aten.div.Tensor(minimum_64, clamp_min_192);  minimum_64 = None
	        round_129 = torch.ops.aten.round.default(div_129);  div_129 = None
	        sub_2948 = torch.ops.aten.sub.Tensor(-128, round_129);  round_129 = None
	        clamp_min_193 = torch.ops.aten.clamp_min.default(sub_2948, -128);  sub_2948 = None
	        clamp_max_128 = torch.ops.aten.clamp_max.default(clamp_min_193, 127);  clamp_min_193 = None
	        _assert_tensor_metadata_578 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_192, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_578 = None
	        _assert_tensor_metadata_579 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_128, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_579 = None
	        convert_element_type_384 = torch.ops.prims.convert_element_type.default(clamp_max_128, torch.int8);  clamp_max_128 = None
	        view_1006 = torch.ops.aten.view.default(clamp_min_192, [sym_size_int, 1500, 1])
	        view_1007 = torch.ops.aten.view.default(convert_element_type_384, [sym_size_int, 1500, 1])
	        reciprocal_64 = torch.ops.aten.reciprocal.default(view_1006);  view_1006 = None
	        mul_6264 = torch.ops.aten.mul.Tensor(reciprocal_64, 1.0);  reciprocal_64 = None
	        mul_6267 = torch.ops.aten.mul.Tensor(add_9832, mul_6264);  add_9832 = mul_6264 = None
	        round_130 = torch.ops.aten.round.default(mul_6267);  mul_6267 = None
	        add_9919 = torch.ops.aten.add.Tensor(round_130, view_1007);  round_130 = view_1007 = None
	        clamp_min_194 = torch.ops.aten.clamp_min.default(add_9919, -128);  add_9919 = None
	        clamp_max_129 = torch.ops.aten.clamp_max.default(clamp_min_194, 127);  clamp_min_194 = None
	        _assert_tensor_metadata_580 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_129, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_580 = None
	        convert_element_type_385 = torch.ops.prims.convert_element_type.default(clamp_max_129, torch.int8);  clamp_max_129 = None
	        view_1010 = torch.ops.aten.view.default(clamp_min_192, [sym_size_int, 1500, 1]);  clamp_min_192 = None
	        view_1011 = torch.ops.aten.view.default(convert_element_type_384, [sym_size_int, 1500, 1]);  convert_element_type_384 = None
	        _assert_tensor_metadata_581 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_385, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_581 = None
	        convert_element_type_386 = torch.ops.prims.convert_element_type.default(convert_element_type_385, torch.float32);  convert_element_type_385 = None
	        _assert_tensor_metadata_582 = torch.ops.aten._assert_tensor_metadata.default(view_1011, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_582 = None
	        convert_element_type_387 = torch.ops.prims.convert_element_type.default(view_1011, torch.float32);  view_1011 = None
	        sub_2968 = torch.ops.aten.sub.Tensor(convert_element_type_386, convert_element_type_387);  convert_element_type_386 = convert_element_type_387 = None
	        mul_6289 = torch.ops.aten.mul.Tensor(sub_2968, view_1010);  sub_2968 = view_1010 = None
	        _assert_tensor_metadata_583 = torch.ops.aten._assert_tensor_metadata.default(mul_6289, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_583 = None
	        view_1013 = torch.ops.aten.view.default(model_audio_tower_layers_10_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_10_fc1_parametrizations_weight_original0 = None
	        view_1014 = torch.ops.aten.view.default(model_audio_tower_layers_10_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_10_fc1_parametrizations_weight_original1 = None
	        view_1015 = torch.ops.aten.view.default(model_audio_tower_layers_10_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_10_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_584 = torch.ops.aten._assert_tensor_metadata.default(view_1013, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_584 = None
	        convert_element_type_388 = torch.ops.prims.convert_element_type.default(view_1013, torch.float32);  view_1013 = None
	        _assert_tensor_metadata_585 = torch.ops.aten._assert_tensor_metadata.default(view_1015, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_585 = None
	        convert_element_type_389 = torch.ops.prims.convert_element_type.default(view_1015, torch.float32);  view_1015 = None
	        sub_2972 = torch.ops.aten.sub.Tensor(convert_element_type_388, convert_element_type_389);  convert_element_type_388 = convert_element_type_389 = None
	        mul_6294 = torch.ops.aten.mul.Tensor(sub_2972, view_1014);  sub_2972 = view_1014 = None
	        view_1016 = torch.ops.aten.view.default(mul_6294, [5120, 1280]);  mul_6294 = None
	        _assert_tensor_metadata_586 = torch.ops.aten._assert_tensor_metadata.default(view_1016, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_586 = None
	        mul_6299 = sym_size_int * 1500
	        view_1017 = torch.ops.aten.view.default(mul_6289, [mul_6299, 1280]);  mul_6289 = mul_6299 = None
	        permute_109 = torch.ops.aten.permute.default(view_1016, [1, 0]);  view_1016 = None
	        addmm_53 = torch.ops.aten.addmm.default(model_audio_tower_layers_10_fc1_bias, view_1017, permute_109);  model_audio_tower_layers_10_fc1_bias = view_1017 = permute_109 = None
	        view_1018 = torch.ops.aten.view.default(addmm_53, [sym_size_int, 1500, 5120]);  addmm_53 = None
	        mul_6306 = torch.ops.aten.mul.Tensor(view_1018, 0.5)
	        mul_6307 = torch.ops.aten.mul.Tensor(view_1018, 0.7071067811865476);  view_1018 = None
	        erf_12 = torch.ops.aten.erf.default(mul_6307);  mul_6307 = None
	        add_9978 = torch.ops.aten.add.Tensor(erf_12, 1);  erf_12 = None
	        mul_6308 = torch.ops.aten.mul.Tensor(mul_6306, add_9978);  mul_6306 = add_9978 = None
	        amin_65 = torch.ops.aten.amin.default(mul_6308, [2])
	        amax_65 = torch.ops.aten.amax.default(mul_6308, [2])
	        full_130 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_65 = torch.ops.aten.minimum.default(amin_65, full_130);  amin_65 = full_130 = None
	        full_131 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_65 = torch.ops.aten.maximum.default(amax_65, full_131);  amax_65 = full_131 = None
	        sub_2985 = torch.ops.aten.sub.Tensor(maximum_65, minimum_65);  maximum_65 = None
	        div_130 = torch.ops.aten.div.Tensor(sub_2985, 255.0);  sub_2985 = None
	        clamp_min_195 = torch.ops.aten.clamp_min.default(div_130, 1.1920928955078125e-07);  div_130 = None
	        div_131 = torch.ops.aten.div.Tensor(minimum_65, clamp_min_195);  minimum_65 = None
	        round_131 = torch.ops.aten.round.default(div_131);  div_131 = None
	        sub_2991 = torch.ops.aten.sub.Tensor(-128, round_131);  round_131 = None
	        clamp_min_196 = torch.ops.aten.clamp_min.default(sub_2991, -128);  sub_2991 = None
	        clamp_max_130 = torch.ops.aten.clamp_max.default(clamp_min_196, 127);  clamp_min_196 = None
	        _assert_tensor_metadata_587 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_195, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_587 = None
	        _assert_tensor_metadata_588 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_130, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_588 = None
	        convert_element_type_390 = torch.ops.prims.convert_element_type.default(clamp_max_130, torch.int8);  clamp_max_130 = None
	        view_1021 = torch.ops.aten.view.default(clamp_min_195, [sym_size_int, 1500, 1])
	        view_1022 = torch.ops.aten.view.default(convert_element_type_390, [sym_size_int, 1500, 1])
	        reciprocal_65 = torch.ops.aten.reciprocal.default(view_1021);  view_1021 = None
	        mul_6354 = torch.ops.aten.mul.Tensor(reciprocal_65, 1.0);  reciprocal_65 = None
	        mul_6357 = torch.ops.aten.mul.Tensor(mul_6308, mul_6354);  mul_6308 = mul_6354 = None
	        round_132 = torch.ops.aten.round.default(mul_6357);  mul_6357 = None
	        add_10061 = torch.ops.aten.add.Tensor(round_132, view_1022);  round_132 = view_1022 = None
	        clamp_min_197 = torch.ops.aten.clamp_min.default(add_10061, -128);  add_10061 = None
	        clamp_max_131 = torch.ops.aten.clamp_max.default(clamp_min_197, 127);  clamp_min_197 = None
	        _assert_tensor_metadata_589 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_131, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_589 = None
	        convert_element_type_391 = torch.ops.prims.convert_element_type.default(clamp_max_131, torch.int8);  clamp_max_131 = None
	        view_1025 = torch.ops.aten.view.default(clamp_min_195, [sym_size_int, 1500, 1]);  clamp_min_195 = None
	        view_1026 = torch.ops.aten.view.default(convert_element_type_390, [sym_size_int, 1500, 1]);  convert_element_type_390 = None
	        _assert_tensor_metadata_590 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_391, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_590 = None
	        convert_element_type_392 = torch.ops.prims.convert_element_type.default(convert_element_type_391, torch.float32);  convert_element_type_391 = None
	        _assert_tensor_metadata_591 = torch.ops.aten._assert_tensor_metadata.default(view_1026, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_591 = None
	        convert_element_type_393 = torch.ops.prims.convert_element_type.default(view_1026, torch.float32);  view_1026 = None
	        sub_3011 = torch.ops.aten.sub.Tensor(convert_element_type_392, convert_element_type_393);  convert_element_type_392 = convert_element_type_393 = None
	        mul_6379 = torch.ops.aten.mul.Tensor(sub_3011, view_1025);  sub_3011 = view_1025 = None
	        _assert_tensor_metadata_592 = torch.ops.aten._assert_tensor_metadata.default(mul_6379, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_592 = None
	        view_1028 = torch.ops.aten.view.default(model_audio_tower_layers_10_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_10_fc2_parametrizations_weight_original0 = None
	        view_1029 = torch.ops.aten.view.default(model_audio_tower_layers_10_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_10_fc2_parametrizations_weight_original1 = None
	        view_1030 = torch.ops.aten.view.default(model_audio_tower_layers_10_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_10_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_593 = torch.ops.aten._assert_tensor_metadata.default(view_1028, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_593 = None
	        convert_element_type_394 = torch.ops.prims.convert_element_type.default(view_1028, torch.float32);  view_1028 = None
	        _assert_tensor_metadata_594 = torch.ops.aten._assert_tensor_metadata.default(view_1030, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_594 = None
	        convert_element_type_395 = torch.ops.prims.convert_element_type.default(view_1030, torch.float32);  view_1030 = None
	        sub_3015 = torch.ops.aten.sub.Tensor(convert_element_type_394, convert_element_type_395);  convert_element_type_394 = convert_element_type_395 = None
	        mul_6384 = torch.ops.aten.mul.Tensor(sub_3015, view_1029);  sub_3015 = view_1029 = None
	        view_1031 = torch.ops.aten.view.default(mul_6384, [1280, 5120]);  mul_6384 = None
	        _assert_tensor_metadata_595 = torch.ops.aten._assert_tensor_metadata.default(view_1031, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_595 = None
	        mul_6389 = sym_size_int * 1500
	        view_1032 = torch.ops.aten.view.default(mul_6379, [mul_6389, 5120]);  mul_6379 = mul_6389 = None
	        permute_110 = torch.ops.aten.permute.default(view_1031, [1, 0]);  view_1031 = None
	        addmm_54 = torch.ops.aten.addmm.default(model_audio_tower_layers_10_fc2_bias, view_1032, permute_110);  model_audio_tower_layers_10_fc2_bias = view_1032 = permute_110 = None
	        view_1033 = torch.ops.aten.view.default(addmm_54, [sym_size_int, 1500, 1280]);  addmm_54 = None
	        add_10124 = torch.ops.aten.add.Tensor(add_9826, view_1033);  add_9826 = view_1033 = None
	        clone_89 = torch.ops.aten.clone.default(add_10124, memory_format = torch.contiguous_format)
	        var_mean_22 = torch.ops.aten.var_mean.correction(clone_89, [2], correction = 0, keepdim = True)
	        getitem_88 = var_mean_22[0]
	        getitem_89 = var_mean_22[1];  var_mean_22 = None
	        add_10129 = torch.ops.aten.add.Tensor(getitem_88, 1e-05);  getitem_88 = None
	        rsqrt_22 = torch.ops.aten.rsqrt.default(add_10129);  add_10129 = None
	        sub_3021 = torch.ops.aten.sub.Tensor(clone_89, getitem_89);  clone_89 = getitem_89 = None
	        mul_6400 = torch.ops.aten.mul.Tensor(sub_3021, rsqrt_22);  sub_3021 = rsqrt_22 = None
	        mul_6401 = torch.ops.aten.mul.Tensor(mul_6400, model_audio_tower_layers_11_self_attn_layer_norm_weight);  mul_6400 = model_audio_tower_layers_11_self_attn_layer_norm_weight = None
	        add_10130 = torch.ops.aten.add.Tensor(mul_6401, model_audio_tower_layers_11_self_attn_layer_norm_bias);  mul_6401 = model_audio_tower_layers_11_self_attn_layer_norm_bias = None
	        amin_66 = torch.ops.aten.amin.default(add_10130, [2])
	        amax_66 = torch.ops.aten.amax.default(add_10130, [2])
	        full_132 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_66 = torch.ops.aten.minimum.default(amin_66, full_132);  amin_66 = full_132 = None
	        full_133 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_66 = torch.ops.aten.maximum.default(amax_66, full_133);  amax_66 = full_133 = None
	        sub_3032 = torch.ops.aten.sub.Tensor(maximum_66, minimum_66);  maximum_66 = None
	        div_132 = torch.ops.aten.div.Tensor(sub_3032, 255.0);  sub_3032 = None
	        clamp_min_198 = torch.ops.aten.clamp_min.default(div_132, 1.1920928955078125e-07);  div_132 = None
	        div_133 = torch.ops.aten.div.Tensor(minimum_66, clamp_min_198);  minimum_66 = None
	        round_133 = torch.ops.aten.round.default(div_133);  div_133 = None
	        sub_3038 = torch.ops.aten.sub.Tensor(-128, round_133);  round_133 = None
	        clamp_min_199 = torch.ops.aten.clamp_min.default(sub_3038, -128);  sub_3038 = None
	        clamp_max_132 = torch.ops.aten.clamp_max.default(clamp_min_199, 127);  clamp_min_199 = None
	        _assert_tensor_metadata_596 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_198, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_596 = None
	        _assert_tensor_metadata_597 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_132, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_597 = None
	        convert_element_type_396 = torch.ops.prims.convert_element_type.default(clamp_max_132, torch.int8);  clamp_max_132 = None
	        view_1036 = torch.ops.aten.view.default(clamp_min_198, [sym_size_int, 1500, 1])
	        view_1037 = torch.ops.aten.view.default(convert_element_type_396, [sym_size_int, 1500, 1])
	        reciprocal_66 = torch.ops.aten.reciprocal.default(view_1036);  view_1036 = None
	        mul_6449 = torch.ops.aten.mul.Tensor(reciprocal_66, 1.0);  reciprocal_66 = None
	        mul_6452 = torch.ops.aten.mul.Tensor(add_10130, mul_6449);  mul_6449 = None
	        round_134 = torch.ops.aten.round.default(mul_6452);  mul_6452 = None
	        add_10217 = torch.ops.aten.add.Tensor(round_134, view_1037);  round_134 = view_1037 = None
	        clamp_min_200 = torch.ops.aten.clamp_min.default(add_10217, -128);  add_10217 = None
	        clamp_max_133 = torch.ops.aten.clamp_max.default(clamp_min_200, 127);  clamp_min_200 = None
	        _assert_tensor_metadata_598 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_133, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_598 = None
	        convert_element_type_397 = torch.ops.prims.convert_element_type.default(clamp_max_133, torch.int8);  clamp_max_133 = None
	        view_1040 = torch.ops.aten.view.default(clamp_min_198, [sym_size_int, 1500, 1]);  clamp_min_198 = None
	        view_1041 = torch.ops.aten.view.default(convert_element_type_396, [sym_size_int, 1500, 1]);  convert_element_type_396 = None
	        _assert_tensor_metadata_599 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_397, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_599 = None
	        convert_element_type_398 = torch.ops.prims.convert_element_type.default(convert_element_type_397, torch.float32);  convert_element_type_397 = None
	        _assert_tensor_metadata_600 = torch.ops.aten._assert_tensor_metadata.default(view_1041, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_600 = None
	        convert_element_type_399 = torch.ops.prims.convert_element_type.default(view_1041, torch.float32);  view_1041 = None
	        sub_3058 = torch.ops.aten.sub.Tensor(convert_element_type_398, convert_element_type_399);  convert_element_type_398 = convert_element_type_399 = None
	        mul_6474 = torch.ops.aten.mul.Tensor(sub_3058, view_1040);  sub_3058 = view_1040 = None
	        _assert_tensor_metadata_601 = torch.ops.aten._assert_tensor_metadata.default(mul_6474, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_601 = None
	        view_1043 = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1044 = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1045 = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_602 = torch.ops.aten._assert_tensor_metadata.default(view_1043, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_602 = None
	        convert_element_type_400 = torch.ops.prims.convert_element_type.default(view_1043, torch.float32);  view_1043 = None
	        _assert_tensor_metadata_603 = torch.ops.aten._assert_tensor_metadata.default(view_1045, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_603 = None
	        convert_element_type_401 = torch.ops.prims.convert_element_type.default(view_1045, torch.float32);  view_1045 = None
	        sub_3062 = torch.ops.aten.sub.Tensor(convert_element_type_400, convert_element_type_401);  convert_element_type_400 = convert_element_type_401 = None
	        mul_6479 = torch.ops.aten.mul.Tensor(sub_3062, view_1044);  sub_3062 = view_1044 = None
	        view_1046 = torch.ops.aten.view.default(mul_6479, [1280, 1280]);  mul_6479 = None
	        _assert_tensor_metadata_604 = torch.ops.aten._assert_tensor_metadata.default(view_1046, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_604 = None
	        mul_6484 = sym_size_int * 1500
	        view_1047 = torch.ops.aten.view.default(mul_6474, [mul_6484, 1280]);  mul_6474 = mul_6484 = None
	        permute_111 = torch.ops.aten.permute.default(view_1046, [1, 0]);  view_1046 = None
	        addmm_55 = torch.ops.aten.addmm.default(model_audio_tower_layers_11_self_attn_q_proj_bias, view_1047, permute_111);  model_audio_tower_layers_11_self_attn_q_proj_bias = view_1047 = permute_111 = None
	        view_1048 = torch.ops.aten.view.default(addmm_55, [sym_size_int, 1500, 1280]);  addmm_55 = None
	        mul_6491 = torch.ops.aten.mul.Tensor(view_1048, 0.125);  view_1048 = None
	        view_1049 = torch.ops.aten.view.default(mul_6491, [sym_size_int, 1500, 20, 64]);  mul_6491 = None
	        permute_112 = torch.ops.aten.permute.default(view_1049, [0, 2, 1, 3]);  view_1049 = None
	        clone_90 = torch.ops.aten.clone.default(permute_112, memory_format = torch.contiguous_format);  permute_112 = None
	        amin_67 = torch.ops.aten.amin.default(add_10130, [2])
	        amax_67 = torch.ops.aten.amax.default(add_10130, [2])
	        full_134 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_67 = torch.ops.aten.minimum.default(amin_67, full_134);  amin_67 = full_134 = None
	        full_135 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_67 = torch.ops.aten.maximum.default(amax_67, full_135);  amax_67 = full_135 = None
	        sub_3077 = torch.ops.aten.sub.Tensor(maximum_67, minimum_67);  maximum_67 = None
	        div_134 = torch.ops.aten.div.Tensor(sub_3077, 255.0);  sub_3077 = None
	        clamp_min_201 = torch.ops.aten.clamp_min.default(div_134, 1.1920928955078125e-07);  div_134 = None
	        div_135 = torch.ops.aten.div.Tensor(minimum_67, clamp_min_201);  minimum_67 = None
	        round_135 = torch.ops.aten.round.default(div_135);  div_135 = None
	        sub_3083 = torch.ops.aten.sub.Tensor(-128, round_135);  round_135 = None
	        clamp_min_202 = torch.ops.aten.clamp_min.default(sub_3083, -128);  sub_3083 = None
	        clamp_max_134 = torch.ops.aten.clamp_max.default(clamp_min_202, 127);  clamp_min_202 = None
	        _assert_tensor_metadata_605 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_201, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_605 = None
	        _assert_tensor_metadata_606 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_134, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_606 = None
	        convert_element_type_402 = torch.ops.prims.convert_element_type.default(clamp_max_134, torch.int8);  clamp_max_134 = None
	        view_1052 = torch.ops.aten.view.default(clamp_min_201, [sym_size_int, 1500, 1])
	        view_1053 = torch.ops.aten.view.default(convert_element_type_402, [sym_size_int, 1500, 1])
	        reciprocal_67 = torch.ops.aten.reciprocal.default(view_1052);  view_1052 = None
	        mul_6545 = torch.ops.aten.mul.Tensor(reciprocal_67, 1.0);  reciprocal_67 = None
	        mul_6548 = torch.ops.aten.mul.Tensor(add_10130, mul_6545);  mul_6545 = None
	        round_136 = torch.ops.aten.round.default(mul_6548);  mul_6548 = None
	        add_10369 = torch.ops.aten.add.Tensor(round_136, view_1053);  round_136 = view_1053 = None
	        clamp_min_203 = torch.ops.aten.clamp_min.default(add_10369, -128);  add_10369 = None
	        clamp_max_135 = torch.ops.aten.clamp_max.default(clamp_min_203, 127);  clamp_min_203 = None
	        _assert_tensor_metadata_607 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_135, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_607 = None
	        convert_element_type_403 = torch.ops.prims.convert_element_type.default(clamp_max_135, torch.int8);  clamp_max_135 = None
	        view_1056 = torch.ops.aten.view.default(clamp_min_201, [sym_size_int, 1500, 1]);  clamp_min_201 = None
	        view_1057 = torch.ops.aten.view.default(convert_element_type_402, [sym_size_int, 1500, 1]);  convert_element_type_402 = None
	        _assert_tensor_metadata_608 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_403, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_608 = None
	        convert_element_type_404 = torch.ops.prims.convert_element_type.default(convert_element_type_403, torch.float32);  convert_element_type_403 = None
	        _assert_tensor_metadata_609 = torch.ops.aten._assert_tensor_metadata.default(view_1057, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_609 = None
	        convert_element_type_405 = torch.ops.prims.convert_element_type.default(view_1057, torch.float32);  view_1057 = None
	        sub_3103 = torch.ops.aten.sub.Tensor(convert_element_type_404, convert_element_type_405);  convert_element_type_404 = convert_element_type_405 = None
	        mul_6570 = torch.ops.aten.mul.Tensor(sub_3103, view_1056);  sub_3103 = view_1056 = None
	        _assert_tensor_metadata_610 = torch.ops.aten._assert_tensor_metadata.default(mul_6570, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_610 = None
	        view_1059 = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1060 = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1061 = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_611 = torch.ops.aten._assert_tensor_metadata.default(view_1059, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_611 = None
	        convert_element_type_406 = torch.ops.prims.convert_element_type.default(view_1059, torch.float32);  view_1059 = None
	        _assert_tensor_metadata_612 = torch.ops.aten._assert_tensor_metadata.default(view_1061, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_612 = None
	        convert_element_type_407 = torch.ops.prims.convert_element_type.default(view_1061, torch.float32);  view_1061 = None
	        sub_3107 = torch.ops.aten.sub.Tensor(convert_element_type_406, convert_element_type_407);  convert_element_type_406 = convert_element_type_407 = None
	        mul_6575 = torch.ops.aten.mul.Tensor(sub_3107, view_1060);  sub_3107 = view_1060 = None
	        view_1062 = torch.ops.aten.view.default(mul_6575, [1280, 1280]);  mul_6575 = None
	        _assert_tensor_metadata_613 = torch.ops.aten._assert_tensor_metadata.default(view_1062, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_613 = None
	        permute_113 = torch.ops.aten.permute.default(view_1062, [1, 0]);  view_1062 = None
	        mul_6578 = sym_size_int * 1500
	        view_1063 = torch.ops.aten.view.default(mul_6570, [mul_6578, 1280]);  mul_6570 = mul_6578 = None
	        mm_11 = torch.ops.aten.mm.default(view_1063, permute_113);  view_1063 = permute_113 = None
	        view_1064 = torch.ops.aten.view.default(mm_11, [sym_size_int, 1500, 1280]);  mm_11 = None
	        view_1065 = torch.ops.aten.view.default(view_1064, [sym_size_int, -1, 20, 64]);  view_1064 = None
	        permute_114 = torch.ops.aten.permute.default(view_1065, [0, 2, 1, 3]);  view_1065 = None
	        clone_91 = torch.ops.aten.clone.default(permute_114, memory_format = torch.contiguous_format);  permute_114 = None
	        amin_68 = torch.ops.aten.amin.default(add_10130, [2])
	        amax_68 = torch.ops.aten.amax.default(add_10130, [2])
	        full_136 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_68 = torch.ops.aten.minimum.default(amin_68, full_136);  amin_68 = full_136 = None
	        full_137 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_68 = torch.ops.aten.maximum.default(amax_68, full_137);  amax_68 = full_137 = None
	        sub_3121 = torch.ops.aten.sub.Tensor(maximum_68, minimum_68);  maximum_68 = None
	        div_136 = torch.ops.aten.div.Tensor(sub_3121, 255.0);  sub_3121 = None
	        clamp_min_204 = torch.ops.aten.clamp_min.default(div_136, 1.1920928955078125e-07);  div_136 = None
	        div_137 = torch.ops.aten.div.Tensor(minimum_68, clamp_min_204);  minimum_68 = None
	        round_137 = torch.ops.aten.round.default(div_137);  div_137 = None
	        sub_3127 = torch.ops.aten.sub.Tensor(-128, round_137);  round_137 = None
	        clamp_min_205 = torch.ops.aten.clamp_min.default(sub_3127, -128);  sub_3127 = None
	        clamp_max_136 = torch.ops.aten.clamp_max.default(clamp_min_205, 127);  clamp_min_205 = None
	        _assert_tensor_metadata_614 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_204, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_614 = None
	        _assert_tensor_metadata_615 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_136, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_615 = None
	        convert_element_type_408 = torch.ops.prims.convert_element_type.default(clamp_max_136, torch.int8);  clamp_max_136 = None
	        view_1068 = torch.ops.aten.view.default(clamp_min_204, [sym_size_int, 1500, 1])
	        view_1069 = torch.ops.aten.view.default(convert_element_type_408, [sym_size_int, 1500, 1])
	        reciprocal_68 = torch.ops.aten.reciprocal.default(view_1068);  view_1068 = None
	        mul_6644 = torch.ops.aten.mul.Tensor(reciprocal_68, 1.0);  reciprocal_68 = None
	        mul_6647 = torch.ops.aten.mul.Tensor(add_10130, mul_6644);  add_10130 = mul_6644 = None
	        round_138 = torch.ops.aten.round.default(mul_6647);  mul_6647 = None
	        add_10517 = torch.ops.aten.add.Tensor(round_138, view_1069);  round_138 = view_1069 = None
	        clamp_min_206 = torch.ops.aten.clamp_min.default(add_10517, -128);  add_10517 = None
	        clamp_max_137 = torch.ops.aten.clamp_max.default(clamp_min_206, 127);  clamp_min_206 = None
	        _assert_tensor_metadata_616 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_137, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_616 = None
	        convert_element_type_409 = torch.ops.prims.convert_element_type.default(clamp_max_137, torch.int8);  clamp_max_137 = None
	        view_1072 = torch.ops.aten.view.default(clamp_min_204, [sym_size_int, 1500, 1]);  clamp_min_204 = None
	        view_1073 = torch.ops.aten.view.default(convert_element_type_408, [sym_size_int, 1500, 1]);  convert_element_type_408 = None
	        _assert_tensor_metadata_617 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_409, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_617 = None
	        convert_element_type_410 = torch.ops.prims.convert_element_type.default(convert_element_type_409, torch.float32);  convert_element_type_409 = None
	        _assert_tensor_metadata_618 = torch.ops.aten._assert_tensor_metadata.default(view_1073, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_618 = None
	        convert_element_type_411 = torch.ops.prims.convert_element_type.default(view_1073, torch.float32);  view_1073 = None
	        sub_3147 = torch.ops.aten.sub.Tensor(convert_element_type_410, convert_element_type_411);  convert_element_type_410 = convert_element_type_411 = None
	        mul_6669 = torch.ops.aten.mul.Tensor(sub_3147, view_1072);  sub_3147 = view_1072 = None
	        _assert_tensor_metadata_619 = torch.ops.aten._assert_tensor_metadata.default(mul_6669, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_619 = None
	        view_1075 = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1076 = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1077 = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_620 = torch.ops.aten._assert_tensor_metadata.default(view_1075, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_620 = None
	        convert_element_type_412 = torch.ops.prims.convert_element_type.default(view_1075, torch.float32);  view_1075 = None
	        _assert_tensor_metadata_621 = torch.ops.aten._assert_tensor_metadata.default(view_1077, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_621 = None
	        convert_element_type_413 = torch.ops.prims.convert_element_type.default(view_1077, torch.float32);  view_1077 = None
	        sub_3151 = torch.ops.aten.sub.Tensor(convert_element_type_412, convert_element_type_413);  convert_element_type_412 = convert_element_type_413 = None
	        mul_6674 = torch.ops.aten.mul.Tensor(sub_3151, view_1076);  sub_3151 = view_1076 = None
	        view_1078 = torch.ops.aten.view.default(mul_6674, [1280, 1280]);  mul_6674 = None
	        _assert_tensor_metadata_622 = torch.ops.aten._assert_tensor_metadata.default(view_1078, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_622 = None
	        mul_6679 = sym_size_int * 1500
	        view_1079 = torch.ops.aten.view.default(mul_6669, [mul_6679, 1280]);  mul_6669 = mul_6679 = None
	        permute_115 = torch.ops.aten.permute.default(view_1078, [1, 0]);  view_1078 = None
	        addmm_56 = torch.ops.aten.addmm.default(model_audio_tower_layers_11_self_attn_v_proj_bias, view_1079, permute_115);  model_audio_tower_layers_11_self_attn_v_proj_bias = view_1079 = permute_115 = None
	        view_1080 = torch.ops.aten.view.default(addmm_56, [sym_size_int, 1500, 1280]);  addmm_56 = None
	        view_1081 = torch.ops.aten.view.default(view_1080, [sym_size_int, -1, 20, 64]);  view_1080 = None
	        permute_116 = torch.ops.aten.permute.default(view_1081, [0, 2, 1, 3]);  view_1081 = None
	        clone_92 = torch.ops.aten.clone.default(permute_116, memory_format = torch.contiguous_format);  permute_116 = None
	        _scaled_dot_product_efficient_attention_11 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_90, clone_91, clone_92, None, False, scale = 1.0);  clone_90 = clone_91 = clone_92 = None
	        getitem_90 = _scaled_dot_product_efficient_attention_11[0];  _scaled_dot_product_efficient_attention_11 = None
	        permute_117 = torch.ops.aten.permute.default(getitem_90, [0, 2, 1, 3]);  getitem_90 = None
	        view_1082 = torch.ops.aten.view.default(permute_117, [sym_size_int, 1500, -1]);  permute_117 = None
	        amin_69 = torch.ops.aten.amin.default(view_1082, [2])
	        amax_69 = torch.ops.aten.amax.default(view_1082, [2])
	        full_138 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_69 = torch.ops.aten.minimum.default(amin_69, full_138);  amin_69 = full_138 = None
	        full_139 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_69 = torch.ops.aten.maximum.default(amax_69, full_139);  amax_69 = full_139 = None
	        sub_3169 = torch.ops.aten.sub.Tensor(maximum_69, minimum_69);  maximum_69 = None
	        div_138 = torch.ops.aten.div.Tensor(sub_3169, 255.0);  sub_3169 = None
	        clamp_min_207 = torch.ops.aten.clamp_min.default(div_138, 1.1920928955078125e-07);  div_138 = None
	        div_139 = torch.ops.aten.div.Tensor(minimum_69, clamp_min_207);  minimum_69 = None
	        round_139 = torch.ops.aten.round.default(div_139);  div_139 = None
	        sub_3175 = torch.ops.aten.sub.Tensor(-128, round_139);  round_139 = None
	        clamp_min_208 = torch.ops.aten.clamp_min.default(sub_3175, -128);  sub_3175 = None
	        clamp_max_138 = torch.ops.aten.clamp_max.default(clamp_min_208, 127);  clamp_min_208 = None
	        _assert_tensor_metadata_623 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_207, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_623 = None
	        _assert_tensor_metadata_624 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_138, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_624 = None
	        convert_element_type_414 = torch.ops.prims.convert_element_type.default(clamp_max_138, torch.int8);  clamp_max_138 = None
	        view_1085 = torch.ops.aten.view.default(clamp_min_207, [sym_size_int, 1500, 1])
	        view_1086 = torch.ops.aten.view.default(convert_element_type_414, [sym_size_int, 1500, 1])
	        reciprocal_69 = torch.ops.aten.reciprocal.default(view_1085);  view_1085 = None
	        mul_6749 = torch.ops.aten.mul.Tensor(reciprocal_69, 1.0);  reciprocal_69 = None
	        mul_6752 = torch.ops.aten.mul.Tensor(view_1082, mul_6749);  view_1082 = mul_6749 = None
	        round_140 = torch.ops.aten.round.default(mul_6752);  mul_6752 = None
	        add_10681 = torch.ops.aten.add.Tensor(round_140, view_1086);  round_140 = view_1086 = None
	        clamp_min_209 = torch.ops.aten.clamp_min.default(add_10681, -128);  add_10681 = None
	        clamp_max_139 = torch.ops.aten.clamp_max.default(clamp_min_209, 127);  clamp_min_209 = None
	        _assert_tensor_metadata_625 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_139, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_625 = None
	        convert_element_type_415 = torch.ops.prims.convert_element_type.default(clamp_max_139, torch.int8);  clamp_max_139 = None
	        view_1089 = torch.ops.aten.view.default(clamp_min_207, [sym_size_int, 1500, 1]);  clamp_min_207 = None
	        view_1090 = torch.ops.aten.view.default(convert_element_type_414, [sym_size_int, 1500, 1]);  convert_element_type_414 = None
	        _assert_tensor_metadata_626 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_415, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_626 = None
	        convert_element_type_416 = torch.ops.prims.convert_element_type.default(convert_element_type_415, torch.float32);  convert_element_type_415 = None
	        _assert_tensor_metadata_627 = torch.ops.aten._assert_tensor_metadata.default(view_1090, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_627 = None
	        convert_element_type_417 = torch.ops.prims.convert_element_type.default(view_1090, torch.float32);  view_1090 = None
	        sub_3195 = torch.ops.aten.sub.Tensor(convert_element_type_416, convert_element_type_417);  convert_element_type_416 = convert_element_type_417 = None
	        mul_6774 = torch.ops.aten.mul.Tensor(sub_3195, view_1089);  sub_3195 = view_1089 = None
	        _assert_tensor_metadata_628 = torch.ops.aten._assert_tensor_metadata.default(mul_6774, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_628 = None
	        view_1092 = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1093 = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1094 = torch.ops.aten.view.default(model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_629 = torch.ops.aten._assert_tensor_metadata.default(view_1092, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_629 = None
	        convert_element_type_418 = torch.ops.prims.convert_element_type.default(view_1092, torch.float32);  view_1092 = None
	        _assert_tensor_metadata_630 = torch.ops.aten._assert_tensor_metadata.default(view_1094, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_630 = None
	        convert_element_type_419 = torch.ops.prims.convert_element_type.default(view_1094, torch.float32);  view_1094 = None
	        sub_3199 = torch.ops.aten.sub.Tensor(convert_element_type_418, convert_element_type_419);  convert_element_type_418 = convert_element_type_419 = None
	        mul_6779 = torch.ops.aten.mul.Tensor(sub_3199, view_1093);  sub_3199 = view_1093 = None
	        view_1095 = torch.ops.aten.view.default(mul_6779, [1280, 1280]);  mul_6779 = None
	        _assert_tensor_metadata_631 = torch.ops.aten._assert_tensor_metadata.default(view_1095, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_631 = None
	        mul_6784 = sym_size_int * 1500
	        view_1096 = torch.ops.aten.view.default(mul_6774, [mul_6784, 1280]);  mul_6774 = mul_6784 = None
	        permute_118 = torch.ops.aten.permute.default(view_1095, [1, 0]);  view_1095 = None
	        addmm_57 = torch.ops.aten.addmm.default(model_audio_tower_layers_11_self_attn_out_proj_bias, view_1096, permute_118);  model_audio_tower_layers_11_self_attn_out_proj_bias = view_1096 = permute_118 = None
	        view_1097 = torch.ops.aten.view.default(addmm_57, [sym_size_int, 1500, 1280]);  addmm_57 = None
	        add_10744 = torch.ops.aten.add.Tensor(add_10124, view_1097);  add_10124 = view_1097 = None
	        clone_94 = torch.ops.aten.clone.default(add_10744, memory_format = torch.contiguous_format)
	        var_mean_23 = torch.ops.aten.var_mean.correction(clone_94, [2], correction = 0, keepdim = True)
	        getitem_94 = var_mean_23[0]
	        getitem_95 = var_mean_23[1];  var_mean_23 = None
	        add_10749 = torch.ops.aten.add.Tensor(getitem_94, 1e-05);  getitem_94 = None
	        rsqrt_23 = torch.ops.aten.rsqrt.default(add_10749);  add_10749 = None
	        sub_3205 = torch.ops.aten.sub.Tensor(clone_94, getitem_95);  clone_94 = getitem_95 = None
	        mul_6795 = torch.ops.aten.mul.Tensor(sub_3205, rsqrt_23);  sub_3205 = rsqrt_23 = None
	        mul_6796 = torch.ops.aten.mul.Tensor(mul_6795, model_audio_tower_layers_11_final_layer_norm_weight);  mul_6795 = model_audio_tower_layers_11_final_layer_norm_weight = None
	        add_10750 = torch.ops.aten.add.Tensor(mul_6796, model_audio_tower_layers_11_final_layer_norm_bias);  mul_6796 = model_audio_tower_layers_11_final_layer_norm_bias = None
	        amin_70 = torch.ops.aten.amin.default(add_10750, [2])
	        amax_70 = torch.ops.aten.amax.default(add_10750, [2])
	        full_140 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_70 = torch.ops.aten.minimum.default(amin_70, full_140);  amin_70 = full_140 = None
	        full_141 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_70 = torch.ops.aten.maximum.default(amax_70, full_141);  amax_70 = full_141 = None
	        sub_3216 = torch.ops.aten.sub.Tensor(maximum_70, minimum_70);  maximum_70 = None
	        div_140 = torch.ops.aten.div.Tensor(sub_3216, 255.0);  sub_3216 = None
	        clamp_min_210 = torch.ops.aten.clamp_min.default(div_140, 1.1920928955078125e-07);  div_140 = None
	        div_141 = torch.ops.aten.div.Tensor(minimum_70, clamp_min_210);  minimum_70 = None
	        round_141 = torch.ops.aten.round.default(div_141);  div_141 = None
	        sub_3222 = torch.ops.aten.sub.Tensor(-128, round_141);  round_141 = None
	        clamp_min_211 = torch.ops.aten.clamp_min.default(sub_3222, -128);  sub_3222 = None
	        clamp_max_140 = torch.ops.aten.clamp_max.default(clamp_min_211, 127);  clamp_min_211 = None
	        _assert_tensor_metadata_632 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_210, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_632 = None
	        _assert_tensor_metadata_633 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_140, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_633 = None
	        convert_element_type_420 = torch.ops.prims.convert_element_type.default(clamp_max_140, torch.int8);  clamp_max_140 = None
	        view_1100 = torch.ops.aten.view.default(clamp_min_210, [sym_size_int, 1500, 1])
	        view_1101 = torch.ops.aten.view.default(convert_element_type_420, [sym_size_int, 1500, 1])
	        reciprocal_70 = torch.ops.aten.reciprocal.default(view_1100);  view_1100 = None
	        mul_6844 = torch.ops.aten.mul.Tensor(reciprocal_70, 1.0);  reciprocal_70 = None
	        mul_6847 = torch.ops.aten.mul.Tensor(add_10750, mul_6844);  add_10750 = mul_6844 = None
	        round_142 = torch.ops.aten.round.default(mul_6847);  mul_6847 = None
	        add_10837 = torch.ops.aten.add.Tensor(round_142, view_1101);  round_142 = view_1101 = None
	        clamp_min_212 = torch.ops.aten.clamp_min.default(add_10837, -128);  add_10837 = None
	        clamp_max_141 = torch.ops.aten.clamp_max.default(clamp_min_212, 127);  clamp_min_212 = None
	        _assert_tensor_metadata_634 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_141, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_634 = None
	        convert_element_type_421 = torch.ops.prims.convert_element_type.default(clamp_max_141, torch.int8);  clamp_max_141 = None
	        view_1104 = torch.ops.aten.view.default(clamp_min_210, [sym_size_int, 1500, 1]);  clamp_min_210 = None
	        view_1105 = torch.ops.aten.view.default(convert_element_type_420, [sym_size_int, 1500, 1]);  convert_element_type_420 = None
	        _assert_tensor_metadata_635 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_421, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_635 = None
	        convert_element_type_422 = torch.ops.prims.convert_element_type.default(convert_element_type_421, torch.float32);  convert_element_type_421 = None
	        _assert_tensor_metadata_636 = torch.ops.aten._assert_tensor_metadata.default(view_1105, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_636 = None
	        convert_element_type_423 = torch.ops.prims.convert_element_type.default(view_1105, torch.float32);  view_1105 = None
	        sub_3242 = torch.ops.aten.sub.Tensor(convert_element_type_422, convert_element_type_423);  convert_element_type_422 = convert_element_type_423 = None
	        mul_6869 = torch.ops.aten.mul.Tensor(sub_3242, view_1104);  sub_3242 = view_1104 = None
	        _assert_tensor_metadata_637 = torch.ops.aten._assert_tensor_metadata.default(mul_6869, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_637 = None
	        view_1107 = torch.ops.aten.view.default(model_audio_tower_layers_11_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_11_fc1_parametrizations_weight_original0 = None
	        view_1108 = torch.ops.aten.view.default(model_audio_tower_layers_11_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_11_fc1_parametrizations_weight_original1 = None
	        view_1109 = torch.ops.aten.view.default(model_audio_tower_layers_11_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_11_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_638 = torch.ops.aten._assert_tensor_metadata.default(view_1107, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_638 = None
	        convert_element_type_424 = torch.ops.prims.convert_element_type.default(view_1107, torch.float32);  view_1107 = None
	        _assert_tensor_metadata_639 = torch.ops.aten._assert_tensor_metadata.default(view_1109, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_639 = None
	        convert_element_type_425 = torch.ops.prims.convert_element_type.default(view_1109, torch.float32);  view_1109 = None
	        sub_3246 = torch.ops.aten.sub.Tensor(convert_element_type_424, convert_element_type_425);  convert_element_type_424 = convert_element_type_425 = None
	        mul_6874 = torch.ops.aten.mul.Tensor(sub_3246, view_1108);  sub_3246 = view_1108 = None
	        view_1110 = torch.ops.aten.view.default(mul_6874, [5120, 1280]);  mul_6874 = None
	        _assert_tensor_metadata_640 = torch.ops.aten._assert_tensor_metadata.default(view_1110, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_640 = None
	        mul_6879 = sym_size_int * 1500
	        view_1111 = torch.ops.aten.view.default(mul_6869, [mul_6879, 1280]);  mul_6869 = mul_6879 = None
	        permute_119 = torch.ops.aten.permute.default(view_1110, [1, 0]);  view_1110 = None
	        addmm_58 = torch.ops.aten.addmm.default(model_audio_tower_layers_11_fc1_bias, view_1111, permute_119);  model_audio_tower_layers_11_fc1_bias = view_1111 = permute_119 = None
	        view_1112 = torch.ops.aten.view.default(addmm_58, [sym_size_int, 1500, 5120]);  addmm_58 = None
	        mul_6886 = torch.ops.aten.mul.Tensor(view_1112, 0.5)
	        mul_6887 = torch.ops.aten.mul.Tensor(view_1112, 0.7071067811865476);  view_1112 = None
	        erf_13 = torch.ops.aten.erf.default(mul_6887);  mul_6887 = None
	        add_10896 = torch.ops.aten.add.Tensor(erf_13, 1);  erf_13 = None
	        mul_6888 = torch.ops.aten.mul.Tensor(mul_6886, add_10896);  mul_6886 = add_10896 = None
	        amin_71 = torch.ops.aten.amin.default(mul_6888, [2])
	        amax_71 = torch.ops.aten.amax.default(mul_6888, [2])
	        full_142 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_71 = torch.ops.aten.minimum.default(amin_71, full_142);  amin_71 = full_142 = None
	        full_143 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_71 = torch.ops.aten.maximum.default(amax_71, full_143);  amax_71 = full_143 = None
	        sub_3259 = torch.ops.aten.sub.Tensor(maximum_71, minimum_71);  maximum_71 = None
	        div_142 = torch.ops.aten.div.Tensor(sub_3259, 255.0);  sub_3259 = None
	        clamp_min_213 = torch.ops.aten.clamp_min.default(div_142, 1.1920928955078125e-07);  div_142 = None
	        div_143 = torch.ops.aten.div.Tensor(minimum_71, clamp_min_213);  minimum_71 = None
	        round_143 = torch.ops.aten.round.default(div_143);  div_143 = None
	        sub_3265 = torch.ops.aten.sub.Tensor(-128, round_143);  round_143 = None
	        clamp_min_214 = torch.ops.aten.clamp_min.default(sub_3265, -128);  sub_3265 = None
	        clamp_max_142 = torch.ops.aten.clamp_max.default(clamp_min_214, 127);  clamp_min_214 = None
	        _assert_tensor_metadata_641 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_213, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_641 = None
	        _assert_tensor_metadata_642 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_142, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_642 = None
	        convert_element_type_426 = torch.ops.prims.convert_element_type.default(clamp_max_142, torch.int8);  clamp_max_142 = None
	        view_1115 = torch.ops.aten.view.default(clamp_min_213, [sym_size_int, 1500, 1])
	        view_1116 = torch.ops.aten.view.default(convert_element_type_426, [sym_size_int, 1500, 1])
	        reciprocal_71 = torch.ops.aten.reciprocal.default(view_1115);  view_1115 = None
	        mul_6934 = torch.ops.aten.mul.Tensor(reciprocal_71, 1.0);  reciprocal_71 = None
	        mul_6937 = torch.ops.aten.mul.Tensor(mul_6888, mul_6934);  mul_6888 = mul_6934 = None
	        round_144 = torch.ops.aten.round.default(mul_6937);  mul_6937 = None
	        add_10979 = torch.ops.aten.add.Tensor(round_144, view_1116);  round_144 = view_1116 = None
	        clamp_min_215 = torch.ops.aten.clamp_min.default(add_10979, -128);  add_10979 = None
	        clamp_max_143 = torch.ops.aten.clamp_max.default(clamp_min_215, 127);  clamp_min_215 = None
	        _assert_tensor_metadata_643 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_143, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_643 = None
	        convert_element_type_427 = torch.ops.prims.convert_element_type.default(clamp_max_143, torch.int8);  clamp_max_143 = None
	        view_1119 = torch.ops.aten.view.default(clamp_min_213, [sym_size_int, 1500, 1]);  clamp_min_213 = None
	        view_1120 = torch.ops.aten.view.default(convert_element_type_426, [sym_size_int, 1500, 1]);  convert_element_type_426 = None
	        _assert_tensor_metadata_644 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_427, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_644 = None
	        convert_element_type_428 = torch.ops.prims.convert_element_type.default(convert_element_type_427, torch.float32);  convert_element_type_427 = None
	        _assert_tensor_metadata_645 = torch.ops.aten._assert_tensor_metadata.default(view_1120, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_645 = None
	        convert_element_type_429 = torch.ops.prims.convert_element_type.default(view_1120, torch.float32);  view_1120 = None
	        sub_3285 = torch.ops.aten.sub.Tensor(convert_element_type_428, convert_element_type_429);  convert_element_type_428 = convert_element_type_429 = None
	        mul_6959 = torch.ops.aten.mul.Tensor(sub_3285, view_1119);  sub_3285 = view_1119 = None
	        _assert_tensor_metadata_646 = torch.ops.aten._assert_tensor_metadata.default(mul_6959, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_646 = None
	        view_1122 = torch.ops.aten.view.default(model_audio_tower_layers_11_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_11_fc2_parametrizations_weight_original0 = None
	        view_1123 = torch.ops.aten.view.default(model_audio_tower_layers_11_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_11_fc2_parametrizations_weight_original1 = None
	        view_1124 = torch.ops.aten.view.default(model_audio_tower_layers_11_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_11_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_647 = torch.ops.aten._assert_tensor_metadata.default(view_1122, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_647 = None
	        convert_element_type_430 = torch.ops.prims.convert_element_type.default(view_1122, torch.float32);  view_1122 = None
	        _assert_tensor_metadata_648 = torch.ops.aten._assert_tensor_metadata.default(view_1124, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_648 = None
	        convert_element_type_431 = torch.ops.prims.convert_element_type.default(view_1124, torch.float32);  view_1124 = None
	        sub_3289 = torch.ops.aten.sub.Tensor(convert_element_type_430, convert_element_type_431);  convert_element_type_430 = convert_element_type_431 = None
	        mul_6964 = torch.ops.aten.mul.Tensor(sub_3289, view_1123);  sub_3289 = view_1123 = None
	        view_1125 = torch.ops.aten.view.default(mul_6964, [1280, 5120]);  mul_6964 = None
	        _assert_tensor_metadata_649 = torch.ops.aten._assert_tensor_metadata.default(view_1125, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_649 = None
	        mul_6969 = sym_size_int * 1500
	        view_1126 = torch.ops.aten.view.default(mul_6959, [mul_6969, 5120]);  mul_6959 = mul_6969 = None
	        permute_120 = torch.ops.aten.permute.default(view_1125, [1, 0]);  view_1125 = None
	        addmm_59 = torch.ops.aten.addmm.default(model_audio_tower_layers_11_fc2_bias, view_1126, permute_120);  model_audio_tower_layers_11_fc2_bias = view_1126 = permute_120 = None
	        view_1127 = torch.ops.aten.view.default(addmm_59, [sym_size_int, 1500, 1280]);  addmm_59 = None
	        add_11042 = torch.ops.aten.add.Tensor(add_10744, view_1127);  add_10744 = view_1127 = None
	        clone_97 = torch.ops.aten.clone.default(add_11042, memory_format = torch.contiguous_format)
	        var_mean_24 = torch.ops.aten.var_mean.correction(clone_97, [2], correction = 0, keepdim = True)
	        getitem_96 = var_mean_24[0]
	        getitem_97 = var_mean_24[1];  var_mean_24 = None
	        add_11047 = torch.ops.aten.add.Tensor(getitem_96, 1e-05);  getitem_96 = None
	        rsqrt_24 = torch.ops.aten.rsqrt.default(add_11047);  add_11047 = None
	        sub_3295 = torch.ops.aten.sub.Tensor(clone_97, getitem_97);  clone_97 = getitem_97 = None
	        mul_6980 = torch.ops.aten.mul.Tensor(sub_3295, rsqrt_24);  sub_3295 = rsqrt_24 = None
	        mul_6981 = torch.ops.aten.mul.Tensor(mul_6980, model_audio_tower_layers_12_self_attn_layer_norm_weight);  mul_6980 = model_audio_tower_layers_12_self_attn_layer_norm_weight = None
	        add_11048 = torch.ops.aten.add.Tensor(mul_6981, model_audio_tower_layers_12_self_attn_layer_norm_bias);  mul_6981 = model_audio_tower_layers_12_self_attn_layer_norm_bias = None
	        amin_72 = torch.ops.aten.amin.default(add_11048, [2])
	        amax_72 = torch.ops.aten.amax.default(add_11048, [2])
	        full_144 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_72 = torch.ops.aten.minimum.default(amin_72, full_144);  amin_72 = full_144 = None
	        full_145 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_72 = torch.ops.aten.maximum.default(amax_72, full_145);  amax_72 = full_145 = None
	        sub_3306 = torch.ops.aten.sub.Tensor(maximum_72, minimum_72);  maximum_72 = None
	        div_144 = torch.ops.aten.div.Tensor(sub_3306, 255.0);  sub_3306 = None
	        clamp_min_216 = torch.ops.aten.clamp_min.default(div_144, 1.1920928955078125e-07);  div_144 = None
	        div_145 = torch.ops.aten.div.Tensor(minimum_72, clamp_min_216);  minimum_72 = None
	        round_145 = torch.ops.aten.round.default(div_145);  div_145 = None
	        sub_3312 = torch.ops.aten.sub.Tensor(-128, round_145);  round_145 = None
	        clamp_min_217 = torch.ops.aten.clamp_min.default(sub_3312, -128);  sub_3312 = None
	        clamp_max_144 = torch.ops.aten.clamp_max.default(clamp_min_217, 127);  clamp_min_217 = None
	        _assert_tensor_metadata_650 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_216, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_650 = None
	        _assert_tensor_metadata_651 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_144, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_651 = None
	        convert_element_type_432 = torch.ops.prims.convert_element_type.default(clamp_max_144, torch.int8);  clamp_max_144 = None
	        view_1130 = torch.ops.aten.view.default(clamp_min_216, [sym_size_int, 1500, 1])
	        view_1131 = torch.ops.aten.view.default(convert_element_type_432, [sym_size_int, 1500, 1])
	        reciprocal_72 = torch.ops.aten.reciprocal.default(view_1130);  view_1130 = None
	        mul_7029 = torch.ops.aten.mul.Tensor(reciprocal_72, 1.0);  reciprocal_72 = None
	        mul_7032 = torch.ops.aten.mul.Tensor(add_11048, mul_7029);  mul_7029 = None
	        round_146 = torch.ops.aten.round.default(mul_7032);  mul_7032 = None
	        add_11135 = torch.ops.aten.add.Tensor(round_146, view_1131);  round_146 = view_1131 = None
	        clamp_min_218 = torch.ops.aten.clamp_min.default(add_11135, -128);  add_11135 = None
	        clamp_max_145 = torch.ops.aten.clamp_max.default(clamp_min_218, 127);  clamp_min_218 = None
	        _assert_tensor_metadata_652 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_145, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_652 = None
	        convert_element_type_433 = torch.ops.prims.convert_element_type.default(clamp_max_145, torch.int8);  clamp_max_145 = None
	        view_1134 = torch.ops.aten.view.default(clamp_min_216, [sym_size_int, 1500, 1]);  clamp_min_216 = None
	        view_1135 = torch.ops.aten.view.default(convert_element_type_432, [sym_size_int, 1500, 1]);  convert_element_type_432 = None
	        _assert_tensor_metadata_653 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_433, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_653 = None
	        convert_element_type_434 = torch.ops.prims.convert_element_type.default(convert_element_type_433, torch.float32);  convert_element_type_433 = None
	        _assert_tensor_metadata_654 = torch.ops.aten._assert_tensor_metadata.default(view_1135, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_654 = None
	        convert_element_type_435 = torch.ops.prims.convert_element_type.default(view_1135, torch.float32);  view_1135 = None
	        sub_3332 = torch.ops.aten.sub.Tensor(convert_element_type_434, convert_element_type_435);  convert_element_type_434 = convert_element_type_435 = None
	        mul_7054 = torch.ops.aten.mul.Tensor(sub_3332, view_1134);  sub_3332 = view_1134 = None
	        _assert_tensor_metadata_655 = torch.ops.aten._assert_tensor_metadata.default(mul_7054, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_655 = None
	        view_1137 = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1138 = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1139 = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_656 = torch.ops.aten._assert_tensor_metadata.default(view_1137, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_656 = None
	        convert_element_type_436 = torch.ops.prims.convert_element_type.default(view_1137, torch.float32);  view_1137 = None
	        _assert_tensor_metadata_657 = torch.ops.aten._assert_tensor_metadata.default(view_1139, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_657 = None
	        convert_element_type_437 = torch.ops.prims.convert_element_type.default(view_1139, torch.float32);  view_1139 = None
	        sub_3336 = torch.ops.aten.sub.Tensor(convert_element_type_436, convert_element_type_437);  convert_element_type_436 = convert_element_type_437 = None
	        mul_7059 = torch.ops.aten.mul.Tensor(sub_3336, view_1138);  sub_3336 = view_1138 = None
	        view_1140 = torch.ops.aten.view.default(mul_7059, [1280, 1280]);  mul_7059 = None
	        _assert_tensor_metadata_658 = torch.ops.aten._assert_tensor_metadata.default(view_1140, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_658 = None
	        mul_7064 = sym_size_int * 1500
	        view_1141 = torch.ops.aten.view.default(mul_7054, [mul_7064, 1280]);  mul_7054 = mul_7064 = None
	        permute_121 = torch.ops.aten.permute.default(view_1140, [1, 0]);  view_1140 = None
	        addmm_60 = torch.ops.aten.addmm.default(model_audio_tower_layers_12_self_attn_q_proj_bias, view_1141, permute_121);  model_audio_tower_layers_12_self_attn_q_proj_bias = view_1141 = permute_121 = None
	        view_1142 = torch.ops.aten.view.default(addmm_60, [sym_size_int, 1500, 1280]);  addmm_60 = None
	        mul_7071 = torch.ops.aten.mul.Tensor(view_1142, 0.125);  view_1142 = None
	        view_1143 = torch.ops.aten.view.default(mul_7071, [sym_size_int, 1500, 20, 64]);  mul_7071 = None
	        permute_122 = torch.ops.aten.permute.default(view_1143, [0, 2, 1, 3]);  view_1143 = None
	        clone_98 = torch.ops.aten.clone.default(permute_122, memory_format = torch.contiguous_format);  permute_122 = None
	        amin_73 = torch.ops.aten.amin.default(add_11048, [2])
	        amax_73 = torch.ops.aten.amax.default(add_11048, [2])
	        full_146 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_73 = torch.ops.aten.minimum.default(amin_73, full_146);  amin_73 = full_146 = None
	        full_147 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_73 = torch.ops.aten.maximum.default(amax_73, full_147);  amax_73 = full_147 = None
	        sub_3351 = torch.ops.aten.sub.Tensor(maximum_73, minimum_73);  maximum_73 = None
	        div_146 = torch.ops.aten.div.Tensor(sub_3351, 255.0);  sub_3351 = None
	        clamp_min_219 = torch.ops.aten.clamp_min.default(div_146, 1.1920928955078125e-07);  div_146 = None
	        div_147 = torch.ops.aten.div.Tensor(minimum_73, clamp_min_219);  minimum_73 = None
	        round_147 = torch.ops.aten.round.default(div_147);  div_147 = None
	        sub_3357 = torch.ops.aten.sub.Tensor(-128, round_147);  round_147 = None
	        clamp_min_220 = torch.ops.aten.clamp_min.default(sub_3357, -128);  sub_3357 = None
	        clamp_max_146 = torch.ops.aten.clamp_max.default(clamp_min_220, 127);  clamp_min_220 = None
	        _assert_tensor_metadata_659 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_219, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_659 = None
	        _assert_tensor_metadata_660 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_146, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_660 = None
	        convert_element_type_438 = torch.ops.prims.convert_element_type.default(clamp_max_146, torch.int8);  clamp_max_146 = None
	        view_1146 = torch.ops.aten.view.default(clamp_min_219, [sym_size_int, 1500, 1])
	        view_1147 = torch.ops.aten.view.default(convert_element_type_438, [sym_size_int, 1500, 1])
	        reciprocal_73 = torch.ops.aten.reciprocal.default(view_1146);  view_1146 = None
	        mul_7125 = torch.ops.aten.mul.Tensor(reciprocal_73, 1.0);  reciprocal_73 = None
	        mul_7128 = torch.ops.aten.mul.Tensor(add_11048, mul_7125);  mul_7125 = None
	        round_148 = torch.ops.aten.round.default(mul_7128);  mul_7128 = None
	        add_11287 = torch.ops.aten.add.Tensor(round_148, view_1147);  round_148 = view_1147 = None
	        clamp_min_221 = torch.ops.aten.clamp_min.default(add_11287, -128);  add_11287 = None
	        clamp_max_147 = torch.ops.aten.clamp_max.default(clamp_min_221, 127);  clamp_min_221 = None
	        _assert_tensor_metadata_661 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_147, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_661 = None
	        convert_element_type_439 = torch.ops.prims.convert_element_type.default(clamp_max_147, torch.int8);  clamp_max_147 = None
	        view_1150 = torch.ops.aten.view.default(clamp_min_219, [sym_size_int, 1500, 1]);  clamp_min_219 = None
	        view_1151 = torch.ops.aten.view.default(convert_element_type_438, [sym_size_int, 1500, 1]);  convert_element_type_438 = None
	        _assert_tensor_metadata_662 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_439, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_662 = None
	        convert_element_type_440 = torch.ops.prims.convert_element_type.default(convert_element_type_439, torch.float32);  convert_element_type_439 = None
	        _assert_tensor_metadata_663 = torch.ops.aten._assert_tensor_metadata.default(view_1151, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_663 = None
	        convert_element_type_441 = torch.ops.prims.convert_element_type.default(view_1151, torch.float32);  view_1151 = None
	        sub_3377 = torch.ops.aten.sub.Tensor(convert_element_type_440, convert_element_type_441);  convert_element_type_440 = convert_element_type_441 = None
	        mul_7150 = torch.ops.aten.mul.Tensor(sub_3377, view_1150);  sub_3377 = view_1150 = None
	        _assert_tensor_metadata_664 = torch.ops.aten._assert_tensor_metadata.default(mul_7150, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_664 = None
	        view_1153 = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1154 = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1155 = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_665 = torch.ops.aten._assert_tensor_metadata.default(view_1153, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_665 = None
	        convert_element_type_442 = torch.ops.prims.convert_element_type.default(view_1153, torch.float32);  view_1153 = None
	        _assert_tensor_metadata_666 = torch.ops.aten._assert_tensor_metadata.default(view_1155, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_666 = None
	        convert_element_type_443 = torch.ops.prims.convert_element_type.default(view_1155, torch.float32);  view_1155 = None
	        sub_3381 = torch.ops.aten.sub.Tensor(convert_element_type_442, convert_element_type_443);  convert_element_type_442 = convert_element_type_443 = None
	        mul_7155 = torch.ops.aten.mul.Tensor(sub_3381, view_1154);  sub_3381 = view_1154 = None
	        view_1156 = torch.ops.aten.view.default(mul_7155, [1280, 1280]);  mul_7155 = None
	        _assert_tensor_metadata_667 = torch.ops.aten._assert_tensor_metadata.default(view_1156, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_667 = None
	        permute_123 = torch.ops.aten.permute.default(view_1156, [1, 0]);  view_1156 = None
	        mul_7158 = sym_size_int * 1500
	        view_1157 = torch.ops.aten.view.default(mul_7150, [mul_7158, 1280]);  mul_7150 = mul_7158 = None
	        mm_12 = torch.ops.aten.mm.default(view_1157, permute_123);  view_1157 = permute_123 = None
	        view_1158 = torch.ops.aten.view.default(mm_12, [sym_size_int, 1500, 1280]);  mm_12 = None
	        view_1159 = torch.ops.aten.view.default(view_1158, [sym_size_int, -1, 20, 64]);  view_1158 = None
	        permute_124 = torch.ops.aten.permute.default(view_1159, [0, 2, 1, 3]);  view_1159 = None
	        clone_99 = torch.ops.aten.clone.default(permute_124, memory_format = torch.contiguous_format);  permute_124 = None
	        amin_74 = torch.ops.aten.amin.default(add_11048, [2])
	        amax_74 = torch.ops.aten.amax.default(add_11048, [2])
	        full_148 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_74 = torch.ops.aten.minimum.default(amin_74, full_148);  amin_74 = full_148 = None
	        full_149 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_74 = torch.ops.aten.maximum.default(amax_74, full_149);  amax_74 = full_149 = None
	        sub_3395 = torch.ops.aten.sub.Tensor(maximum_74, minimum_74);  maximum_74 = None
	        div_148 = torch.ops.aten.div.Tensor(sub_3395, 255.0);  sub_3395 = None
	        clamp_min_222 = torch.ops.aten.clamp_min.default(div_148, 1.1920928955078125e-07);  div_148 = None
	        div_149 = torch.ops.aten.div.Tensor(minimum_74, clamp_min_222);  minimum_74 = None
	        round_149 = torch.ops.aten.round.default(div_149);  div_149 = None
	        sub_3401 = torch.ops.aten.sub.Tensor(-128, round_149);  round_149 = None
	        clamp_min_223 = torch.ops.aten.clamp_min.default(sub_3401, -128);  sub_3401 = None
	        clamp_max_148 = torch.ops.aten.clamp_max.default(clamp_min_223, 127);  clamp_min_223 = None
	        _assert_tensor_metadata_668 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_222, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_668 = None
	        _assert_tensor_metadata_669 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_148, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_669 = None
	        convert_element_type_444 = torch.ops.prims.convert_element_type.default(clamp_max_148, torch.int8);  clamp_max_148 = None
	        view_1162 = torch.ops.aten.view.default(clamp_min_222, [sym_size_int, 1500, 1])
	        view_1163 = torch.ops.aten.view.default(convert_element_type_444, [sym_size_int, 1500, 1])
	        reciprocal_74 = torch.ops.aten.reciprocal.default(view_1162);  view_1162 = None
	        mul_7224 = torch.ops.aten.mul.Tensor(reciprocal_74, 1.0);  reciprocal_74 = None
	        mul_7227 = torch.ops.aten.mul.Tensor(add_11048, mul_7224);  add_11048 = mul_7224 = None
	        round_150 = torch.ops.aten.round.default(mul_7227);  mul_7227 = None
	        add_11435 = torch.ops.aten.add.Tensor(round_150, view_1163);  round_150 = view_1163 = None
	        clamp_min_224 = torch.ops.aten.clamp_min.default(add_11435, -128);  add_11435 = None
	        clamp_max_149 = torch.ops.aten.clamp_max.default(clamp_min_224, 127);  clamp_min_224 = None
	        _assert_tensor_metadata_670 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_149, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_670 = None
	        convert_element_type_445 = torch.ops.prims.convert_element_type.default(clamp_max_149, torch.int8);  clamp_max_149 = None
	        view_1166 = torch.ops.aten.view.default(clamp_min_222, [sym_size_int, 1500, 1]);  clamp_min_222 = None
	        view_1167 = torch.ops.aten.view.default(convert_element_type_444, [sym_size_int, 1500, 1]);  convert_element_type_444 = None
	        _assert_tensor_metadata_671 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_445, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_671 = None
	        convert_element_type_446 = torch.ops.prims.convert_element_type.default(convert_element_type_445, torch.float32);  convert_element_type_445 = None
	        _assert_tensor_metadata_672 = torch.ops.aten._assert_tensor_metadata.default(view_1167, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_672 = None
	        convert_element_type_447 = torch.ops.prims.convert_element_type.default(view_1167, torch.float32);  view_1167 = None
	        sub_3421 = torch.ops.aten.sub.Tensor(convert_element_type_446, convert_element_type_447);  convert_element_type_446 = convert_element_type_447 = None
	        mul_7249 = torch.ops.aten.mul.Tensor(sub_3421, view_1166);  sub_3421 = view_1166 = None
	        _assert_tensor_metadata_673 = torch.ops.aten._assert_tensor_metadata.default(mul_7249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_673 = None
	        view_1169 = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1170 = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1171 = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_674 = torch.ops.aten._assert_tensor_metadata.default(view_1169, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_674 = None
	        convert_element_type_448 = torch.ops.prims.convert_element_type.default(view_1169, torch.float32);  view_1169 = None
	        _assert_tensor_metadata_675 = torch.ops.aten._assert_tensor_metadata.default(view_1171, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_675 = None
	        convert_element_type_449 = torch.ops.prims.convert_element_type.default(view_1171, torch.float32);  view_1171 = None
	        sub_3425 = torch.ops.aten.sub.Tensor(convert_element_type_448, convert_element_type_449);  convert_element_type_448 = convert_element_type_449 = None
	        mul_7254 = torch.ops.aten.mul.Tensor(sub_3425, view_1170);  sub_3425 = view_1170 = None
	        view_1172 = torch.ops.aten.view.default(mul_7254, [1280, 1280]);  mul_7254 = None
	        _assert_tensor_metadata_676 = torch.ops.aten._assert_tensor_metadata.default(view_1172, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_676 = None
	        mul_7259 = sym_size_int * 1500
	        view_1173 = torch.ops.aten.view.default(mul_7249, [mul_7259, 1280]);  mul_7249 = mul_7259 = None
	        permute_125 = torch.ops.aten.permute.default(view_1172, [1, 0]);  view_1172 = None
	        addmm_61 = torch.ops.aten.addmm.default(model_audio_tower_layers_12_self_attn_v_proj_bias, view_1173, permute_125);  model_audio_tower_layers_12_self_attn_v_proj_bias = view_1173 = permute_125 = None
	        view_1174 = torch.ops.aten.view.default(addmm_61, [sym_size_int, 1500, 1280]);  addmm_61 = None
	        view_1175 = torch.ops.aten.view.default(view_1174, [sym_size_int, -1, 20, 64]);  view_1174 = None
	        permute_126 = torch.ops.aten.permute.default(view_1175, [0, 2, 1, 3]);  view_1175 = None
	        clone_100 = torch.ops.aten.clone.default(permute_126, memory_format = torch.contiguous_format);  permute_126 = None
	        _scaled_dot_product_efficient_attention_12 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_98, clone_99, clone_100, None, False, scale = 1.0);  clone_98 = clone_99 = clone_100 = None
	        getitem_98 = _scaled_dot_product_efficient_attention_12[0];  _scaled_dot_product_efficient_attention_12 = None
	        permute_127 = torch.ops.aten.permute.default(getitem_98, [0, 2, 1, 3]);  getitem_98 = None
	        view_1176 = torch.ops.aten.view.default(permute_127, [sym_size_int, 1500, -1]);  permute_127 = None
	        amin_75 = torch.ops.aten.amin.default(view_1176, [2])
	        amax_75 = torch.ops.aten.amax.default(view_1176, [2])
	        full_150 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_75 = torch.ops.aten.minimum.default(amin_75, full_150);  amin_75 = full_150 = None
	        full_151 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_75 = torch.ops.aten.maximum.default(amax_75, full_151);  amax_75 = full_151 = None
	        sub_3443 = torch.ops.aten.sub.Tensor(maximum_75, minimum_75);  maximum_75 = None
	        div_150 = torch.ops.aten.div.Tensor(sub_3443, 255.0);  sub_3443 = None
	        clamp_min_225 = torch.ops.aten.clamp_min.default(div_150, 1.1920928955078125e-07);  div_150 = None
	        div_151 = torch.ops.aten.div.Tensor(minimum_75, clamp_min_225);  minimum_75 = None
	        round_151 = torch.ops.aten.round.default(div_151);  div_151 = None
	        sub_3449 = torch.ops.aten.sub.Tensor(-128, round_151);  round_151 = None
	        clamp_min_226 = torch.ops.aten.clamp_min.default(sub_3449, -128);  sub_3449 = None
	        clamp_max_150 = torch.ops.aten.clamp_max.default(clamp_min_226, 127);  clamp_min_226 = None
	        _assert_tensor_metadata_677 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_225, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_677 = None
	        _assert_tensor_metadata_678 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_150, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_678 = None
	        convert_element_type_450 = torch.ops.prims.convert_element_type.default(clamp_max_150, torch.int8);  clamp_max_150 = None
	        view_1179 = torch.ops.aten.view.default(clamp_min_225, [sym_size_int, 1500, 1])
	        view_1180 = torch.ops.aten.view.default(convert_element_type_450, [sym_size_int, 1500, 1])
	        reciprocal_75 = torch.ops.aten.reciprocal.default(view_1179);  view_1179 = None
	        mul_7329 = torch.ops.aten.mul.Tensor(reciprocal_75, 1.0);  reciprocal_75 = None
	        mul_7332 = torch.ops.aten.mul.Tensor(view_1176, mul_7329);  view_1176 = mul_7329 = None
	        round_152 = torch.ops.aten.round.default(mul_7332);  mul_7332 = None
	        add_11599 = torch.ops.aten.add.Tensor(round_152, view_1180);  round_152 = view_1180 = None
	        clamp_min_227 = torch.ops.aten.clamp_min.default(add_11599, -128);  add_11599 = None
	        clamp_max_151 = torch.ops.aten.clamp_max.default(clamp_min_227, 127);  clamp_min_227 = None
	        _assert_tensor_metadata_679 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_151, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_679 = None
	        convert_element_type_451 = torch.ops.prims.convert_element_type.default(clamp_max_151, torch.int8);  clamp_max_151 = None
	        view_1183 = torch.ops.aten.view.default(clamp_min_225, [sym_size_int, 1500, 1]);  clamp_min_225 = None
	        view_1184 = torch.ops.aten.view.default(convert_element_type_450, [sym_size_int, 1500, 1]);  convert_element_type_450 = None
	        _assert_tensor_metadata_680 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_451, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_680 = None
	        convert_element_type_452 = torch.ops.prims.convert_element_type.default(convert_element_type_451, torch.float32);  convert_element_type_451 = None
	        _assert_tensor_metadata_681 = torch.ops.aten._assert_tensor_metadata.default(view_1184, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_681 = None
	        convert_element_type_453 = torch.ops.prims.convert_element_type.default(view_1184, torch.float32);  view_1184 = None
	        sub_3469 = torch.ops.aten.sub.Tensor(convert_element_type_452, convert_element_type_453);  convert_element_type_452 = convert_element_type_453 = None
	        mul_7354 = torch.ops.aten.mul.Tensor(sub_3469, view_1183);  sub_3469 = view_1183 = None
	        _assert_tensor_metadata_682 = torch.ops.aten._assert_tensor_metadata.default(mul_7354, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_682 = None
	        view_1186 = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1187 = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1188 = torch.ops.aten.view.default(model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_683 = torch.ops.aten._assert_tensor_metadata.default(view_1186, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_683 = None
	        convert_element_type_454 = torch.ops.prims.convert_element_type.default(view_1186, torch.float32);  view_1186 = None
	        _assert_tensor_metadata_684 = torch.ops.aten._assert_tensor_metadata.default(view_1188, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_684 = None
	        convert_element_type_455 = torch.ops.prims.convert_element_type.default(view_1188, torch.float32);  view_1188 = None
	        sub_3473 = torch.ops.aten.sub.Tensor(convert_element_type_454, convert_element_type_455);  convert_element_type_454 = convert_element_type_455 = None
	        mul_7359 = torch.ops.aten.mul.Tensor(sub_3473, view_1187);  sub_3473 = view_1187 = None
	        view_1189 = torch.ops.aten.view.default(mul_7359, [1280, 1280]);  mul_7359 = None
	        _assert_tensor_metadata_685 = torch.ops.aten._assert_tensor_metadata.default(view_1189, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_685 = None
	        mul_7364 = sym_size_int * 1500
	        view_1190 = torch.ops.aten.view.default(mul_7354, [mul_7364, 1280]);  mul_7354 = mul_7364 = None
	        permute_128 = torch.ops.aten.permute.default(view_1189, [1, 0]);  view_1189 = None
	        addmm_62 = torch.ops.aten.addmm.default(model_audio_tower_layers_12_self_attn_out_proj_bias, view_1190, permute_128);  model_audio_tower_layers_12_self_attn_out_proj_bias = view_1190 = permute_128 = None
	        view_1191 = torch.ops.aten.view.default(addmm_62, [sym_size_int, 1500, 1280]);  addmm_62 = None
	        add_11662 = torch.ops.aten.add.Tensor(add_11042, view_1191);  add_11042 = view_1191 = None
	        clone_102 = torch.ops.aten.clone.default(add_11662, memory_format = torch.contiguous_format)
	        var_mean_25 = torch.ops.aten.var_mean.correction(clone_102, [2], correction = 0, keepdim = True)
	        getitem_102 = var_mean_25[0]
	        getitem_103 = var_mean_25[1];  var_mean_25 = None
	        add_11667 = torch.ops.aten.add.Tensor(getitem_102, 1e-05);  getitem_102 = None
	        rsqrt_25 = torch.ops.aten.rsqrt.default(add_11667);  add_11667 = None
	        sub_3479 = torch.ops.aten.sub.Tensor(clone_102, getitem_103);  clone_102 = getitem_103 = None
	        mul_7375 = torch.ops.aten.mul.Tensor(sub_3479, rsqrt_25);  sub_3479 = rsqrt_25 = None
	        mul_7376 = torch.ops.aten.mul.Tensor(mul_7375, model_audio_tower_layers_12_final_layer_norm_weight);  mul_7375 = model_audio_tower_layers_12_final_layer_norm_weight = None
	        add_11668 = torch.ops.aten.add.Tensor(mul_7376, model_audio_tower_layers_12_final_layer_norm_bias);  mul_7376 = model_audio_tower_layers_12_final_layer_norm_bias = None
	        amin_76 = torch.ops.aten.amin.default(add_11668, [2])
	        amax_76 = torch.ops.aten.amax.default(add_11668, [2])
	        full_152 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_76 = torch.ops.aten.minimum.default(amin_76, full_152);  amin_76 = full_152 = None
	        full_153 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_76 = torch.ops.aten.maximum.default(amax_76, full_153);  amax_76 = full_153 = None
	        sub_3490 = torch.ops.aten.sub.Tensor(maximum_76, minimum_76);  maximum_76 = None
	        div_152 = torch.ops.aten.div.Tensor(sub_3490, 255.0);  sub_3490 = None
	        clamp_min_228 = torch.ops.aten.clamp_min.default(div_152, 1.1920928955078125e-07);  div_152 = None
	        div_153 = torch.ops.aten.div.Tensor(minimum_76, clamp_min_228);  minimum_76 = None
	        round_153 = torch.ops.aten.round.default(div_153);  div_153 = None
	        sub_3496 = torch.ops.aten.sub.Tensor(-128, round_153);  round_153 = None
	        clamp_min_229 = torch.ops.aten.clamp_min.default(sub_3496, -128);  sub_3496 = None
	        clamp_max_152 = torch.ops.aten.clamp_max.default(clamp_min_229, 127);  clamp_min_229 = None
	        _assert_tensor_metadata_686 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_228, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_686 = None
	        _assert_tensor_metadata_687 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_152, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_687 = None
	        convert_element_type_456 = torch.ops.prims.convert_element_type.default(clamp_max_152, torch.int8);  clamp_max_152 = None
	        view_1194 = torch.ops.aten.view.default(clamp_min_228, [sym_size_int, 1500, 1])
	        view_1195 = torch.ops.aten.view.default(convert_element_type_456, [sym_size_int, 1500, 1])
	        reciprocal_76 = torch.ops.aten.reciprocal.default(view_1194);  view_1194 = None
	        mul_7424 = torch.ops.aten.mul.Tensor(reciprocal_76, 1.0);  reciprocal_76 = None
	        mul_7427 = torch.ops.aten.mul.Tensor(add_11668, mul_7424);  add_11668 = mul_7424 = None
	        round_154 = torch.ops.aten.round.default(mul_7427);  mul_7427 = None
	        add_11755 = torch.ops.aten.add.Tensor(round_154, view_1195);  round_154 = view_1195 = None
	        clamp_min_230 = torch.ops.aten.clamp_min.default(add_11755, -128);  add_11755 = None
	        clamp_max_153 = torch.ops.aten.clamp_max.default(clamp_min_230, 127);  clamp_min_230 = None
	        _assert_tensor_metadata_688 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_153, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_688 = None
	        convert_element_type_457 = torch.ops.prims.convert_element_type.default(clamp_max_153, torch.int8);  clamp_max_153 = None
	        view_1198 = torch.ops.aten.view.default(clamp_min_228, [sym_size_int, 1500, 1]);  clamp_min_228 = None
	        view_1199 = torch.ops.aten.view.default(convert_element_type_456, [sym_size_int, 1500, 1]);  convert_element_type_456 = None
	        _assert_tensor_metadata_689 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_457, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_689 = None
	        convert_element_type_458 = torch.ops.prims.convert_element_type.default(convert_element_type_457, torch.float32);  convert_element_type_457 = None
	        _assert_tensor_metadata_690 = torch.ops.aten._assert_tensor_metadata.default(view_1199, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_690 = None
	        convert_element_type_459 = torch.ops.prims.convert_element_type.default(view_1199, torch.float32);  view_1199 = None
	        sub_3516 = torch.ops.aten.sub.Tensor(convert_element_type_458, convert_element_type_459);  convert_element_type_458 = convert_element_type_459 = None
	        mul_7449 = torch.ops.aten.mul.Tensor(sub_3516, view_1198);  sub_3516 = view_1198 = None
	        _assert_tensor_metadata_691 = torch.ops.aten._assert_tensor_metadata.default(mul_7449, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_691 = None
	        view_1201 = torch.ops.aten.view.default(model_audio_tower_layers_12_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_12_fc1_parametrizations_weight_original0 = None
	        view_1202 = torch.ops.aten.view.default(model_audio_tower_layers_12_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_12_fc1_parametrizations_weight_original1 = None
	        view_1203 = torch.ops.aten.view.default(model_audio_tower_layers_12_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_12_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_692 = torch.ops.aten._assert_tensor_metadata.default(view_1201, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_692 = None
	        convert_element_type_460 = torch.ops.prims.convert_element_type.default(view_1201, torch.float32);  view_1201 = None
	        _assert_tensor_metadata_693 = torch.ops.aten._assert_tensor_metadata.default(view_1203, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_693 = None
	        convert_element_type_461 = torch.ops.prims.convert_element_type.default(view_1203, torch.float32);  view_1203 = None
	        sub_3520 = torch.ops.aten.sub.Tensor(convert_element_type_460, convert_element_type_461);  convert_element_type_460 = convert_element_type_461 = None
	        mul_7454 = torch.ops.aten.mul.Tensor(sub_3520, view_1202);  sub_3520 = view_1202 = None
	        view_1204 = torch.ops.aten.view.default(mul_7454, [5120, 1280]);  mul_7454 = None
	        _assert_tensor_metadata_694 = torch.ops.aten._assert_tensor_metadata.default(view_1204, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_694 = None
	        mul_7459 = sym_size_int * 1500
	        view_1205 = torch.ops.aten.view.default(mul_7449, [mul_7459, 1280]);  mul_7449 = mul_7459 = None
	        permute_129 = torch.ops.aten.permute.default(view_1204, [1, 0]);  view_1204 = None
	        addmm_63 = torch.ops.aten.addmm.default(model_audio_tower_layers_12_fc1_bias, view_1205, permute_129);  model_audio_tower_layers_12_fc1_bias = view_1205 = permute_129 = None
	        view_1206 = torch.ops.aten.view.default(addmm_63, [sym_size_int, 1500, 5120]);  addmm_63 = None
	        mul_7466 = torch.ops.aten.mul.Tensor(view_1206, 0.5)
	        mul_7467 = torch.ops.aten.mul.Tensor(view_1206, 0.7071067811865476);  view_1206 = None
	        erf_14 = torch.ops.aten.erf.default(mul_7467);  mul_7467 = None
	        add_11814 = torch.ops.aten.add.Tensor(erf_14, 1);  erf_14 = None
	        mul_7468 = torch.ops.aten.mul.Tensor(mul_7466, add_11814);  mul_7466 = add_11814 = None
	        amin_77 = torch.ops.aten.amin.default(mul_7468, [2])
	        amax_77 = torch.ops.aten.amax.default(mul_7468, [2])
	        full_154 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_77 = torch.ops.aten.minimum.default(amin_77, full_154);  amin_77 = full_154 = None
	        full_155 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_77 = torch.ops.aten.maximum.default(amax_77, full_155);  amax_77 = full_155 = None
	        sub_3533 = torch.ops.aten.sub.Tensor(maximum_77, minimum_77);  maximum_77 = None
	        div_154 = torch.ops.aten.div.Tensor(sub_3533, 255.0);  sub_3533 = None
	        clamp_min_231 = torch.ops.aten.clamp_min.default(div_154, 1.1920928955078125e-07);  div_154 = None
	        div_155 = torch.ops.aten.div.Tensor(minimum_77, clamp_min_231);  minimum_77 = None
	        round_155 = torch.ops.aten.round.default(div_155);  div_155 = None
	        sub_3539 = torch.ops.aten.sub.Tensor(-128, round_155);  round_155 = None
	        clamp_min_232 = torch.ops.aten.clamp_min.default(sub_3539, -128);  sub_3539 = None
	        clamp_max_154 = torch.ops.aten.clamp_max.default(clamp_min_232, 127);  clamp_min_232 = None
	        _assert_tensor_metadata_695 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_231, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_695 = None
	        _assert_tensor_metadata_696 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_154, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_696 = None
	        convert_element_type_462 = torch.ops.prims.convert_element_type.default(clamp_max_154, torch.int8);  clamp_max_154 = None
	        view_1209 = torch.ops.aten.view.default(clamp_min_231, [sym_size_int, 1500, 1])
	        view_1210 = torch.ops.aten.view.default(convert_element_type_462, [sym_size_int, 1500, 1])
	        reciprocal_77 = torch.ops.aten.reciprocal.default(view_1209);  view_1209 = None
	        mul_7514 = torch.ops.aten.mul.Tensor(reciprocal_77, 1.0);  reciprocal_77 = None
	        mul_7517 = torch.ops.aten.mul.Tensor(mul_7468, mul_7514);  mul_7468 = mul_7514 = None
	        round_156 = torch.ops.aten.round.default(mul_7517);  mul_7517 = None
	        add_11897 = torch.ops.aten.add.Tensor(round_156, view_1210);  round_156 = view_1210 = None
	        clamp_min_233 = torch.ops.aten.clamp_min.default(add_11897, -128);  add_11897 = None
	        clamp_max_155 = torch.ops.aten.clamp_max.default(clamp_min_233, 127);  clamp_min_233 = None
	        _assert_tensor_metadata_697 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_155, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_697 = None
	        convert_element_type_463 = torch.ops.prims.convert_element_type.default(clamp_max_155, torch.int8);  clamp_max_155 = None
	        view_1213 = torch.ops.aten.view.default(clamp_min_231, [sym_size_int, 1500, 1]);  clamp_min_231 = None
	        view_1214 = torch.ops.aten.view.default(convert_element_type_462, [sym_size_int, 1500, 1]);  convert_element_type_462 = None
	        _assert_tensor_metadata_698 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_463, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_698 = None
	        convert_element_type_464 = torch.ops.prims.convert_element_type.default(convert_element_type_463, torch.float32);  convert_element_type_463 = None
	        _assert_tensor_metadata_699 = torch.ops.aten._assert_tensor_metadata.default(view_1214, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_699 = None
	        convert_element_type_465 = torch.ops.prims.convert_element_type.default(view_1214, torch.float32);  view_1214 = None
	        sub_3559 = torch.ops.aten.sub.Tensor(convert_element_type_464, convert_element_type_465);  convert_element_type_464 = convert_element_type_465 = None
	        mul_7539 = torch.ops.aten.mul.Tensor(sub_3559, view_1213);  sub_3559 = view_1213 = None
	        _assert_tensor_metadata_700 = torch.ops.aten._assert_tensor_metadata.default(mul_7539, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_700 = None
	        view_1216 = torch.ops.aten.view.default(model_audio_tower_layers_12_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_12_fc2_parametrizations_weight_original0 = None
	        view_1217 = torch.ops.aten.view.default(model_audio_tower_layers_12_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_12_fc2_parametrizations_weight_original1 = None
	        view_1218 = torch.ops.aten.view.default(model_audio_tower_layers_12_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_12_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_701 = torch.ops.aten._assert_tensor_metadata.default(view_1216, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_701 = None
	        convert_element_type_466 = torch.ops.prims.convert_element_type.default(view_1216, torch.float32);  view_1216 = None
	        _assert_tensor_metadata_702 = torch.ops.aten._assert_tensor_metadata.default(view_1218, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_702 = None
	        convert_element_type_467 = torch.ops.prims.convert_element_type.default(view_1218, torch.float32);  view_1218 = None
	        sub_3563 = torch.ops.aten.sub.Tensor(convert_element_type_466, convert_element_type_467);  convert_element_type_466 = convert_element_type_467 = None
	        mul_7544 = torch.ops.aten.mul.Tensor(sub_3563, view_1217);  sub_3563 = view_1217 = None
	        view_1219 = torch.ops.aten.view.default(mul_7544, [1280, 5120]);  mul_7544 = None
	        _assert_tensor_metadata_703 = torch.ops.aten._assert_tensor_metadata.default(view_1219, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_703 = None
	        mul_7549 = sym_size_int * 1500
	        view_1220 = torch.ops.aten.view.default(mul_7539, [mul_7549, 5120]);  mul_7539 = mul_7549 = None
	        permute_130 = torch.ops.aten.permute.default(view_1219, [1, 0]);  view_1219 = None
	        addmm_64 = torch.ops.aten.addmm.default(model_audio_tower_layers_12_fc2_bias, view_1220, permute_130);  model_audio_tower_layers_12_fc2_bias = view_1220 = permute_130 = None
	        view_1221 = torch.ops.aten.view.default(addmm_64, [sym_size_int, 1500, 1280]);  addmm_64 = None
	        add_11960 = torch.ops.aten.add.Tensor(add_11662, view_1221);  add_11662 = view_1221 = None
	        clone_105 = torch.ops.aten.clone.default(add_11960, memory_format = torch.contiguous_format)
	        var_mean_26 = torch.ops.aten.var_mean.correction(clone_105, [2], correction = 0, keepdim = True)
	        getitem_104 = var_mean_26[0]
	        getitem_105 = var_mean_26[1];  var_mean_26 = None
	        add_11965 = torch.ops.aten.add.Tensor(getitem_104, 1e-05);  getitem_104 = None
	        rsqrt_26 = torch.ops.aten.rsqrt.default(add_11965);  add_11965 = None
	        sub_3569 = torch.ops.aten.sub.Tensor(clone_105, getitem_105);  clone_105 = getitem_105 = None
	        mul_7560 = torch.ops.aten.mul.Tensor(sub_3569, rsqrt_26);  sub_3569 = rsqrt_26 = None
	        mul_7561 = torch.ops.aten.mul.Tensor(mul_7560, model_audio_tower_layers_13_self_attn_layer_norm_weight);  mul_7560 = model_audio_tower_layers_13_self_attn_layer_norm_weight = None
	        add_11966 = torch.ops.aten.add.Tensor(mul_7561, model_audio_tower_layers_13_self_attn_layer_norm_bias);  mul_7561 = model_audio_tower_layers_13_self_attn_layer_norm_bias = None
	        amin_78 = torch.ops.aten.amin.default(add_11966, [2])
	        amax_78 = torch.ops.aten.amax.default(add_11966, [2])
	        full_156 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_78 = torch.ops.aten.minimum.default(amin_78, full_156);  amin_78 = full_156 = None
	        full_157 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_78 = torch.ops.aten.maximum.default(amax_78, full_157);  amax_78 = full_157 = None
	        sub_3580 = torch.ops.aten.sub.Tensor(maximum_78, minimum_78);  maximum_78 = None
	        div_156 = torch.ops.aten.div.Tensor(sub_3580, 255.0);  sub_3580 = None
	        clamp_min_234 = torch.ops.aten.clamp_min.default(div_156, 1.1920928955078125e-07);  div_156 = None
	        div_157 = torch.ops.aten.div.Tensor(minimum_78, clamp_min_234);  minimum_78 = None
	        round_157 = torch.ops.aten.round.default(div_157);  div_157 = None
	        sub_3586 = torch.ops.aten.sub.Tensor(-128, round_157);  round_157 = None
	        clamp_min_235 = torch.ops.aten.clamp_min.default(sub_3586, -128);  sub_3586 = None
	        clamp_max_156 = torch.ops.aten.clamp_max.default(clamp_min_235, 127);  clamp_min_235 = None
	        _assert_tensor_metadata_704 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_234, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_704 = None
	        _assert_tensor_metadata_705 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_156, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_705 = None
	        convert_element_type_468 = torch.ops.prims.convert_element_type.default(clamp_max_156, torch.int8);  clamp_max_156 = None
	        view_1224 = torch.ops.aten.view.default(clamp_min_234, [sym_size_int, 1500, 1])
	        view_1225 = torch.ops.aten.view.default(convert_element_type_468, [sym_size_int, 1500, 1])
	        reciprocal_78 = torch.ops.aten.reciprocal.default(view_1224);  view_1224 = None
	        mul_7609 = torch.ops.aten.mul.Tensor(reciprocal_78, 1.0);  reciprocal_78 = None
	        mul_7612 = torch.ops.aten.mul.Tensor(add_11966, mul_7609);  mul_7609 = None
	        round_158 = torch.ops.aten.round.default(mul_7612);  mul_7612 = None
	        add_12053 = torch.ops.aten.add.Tensor(round_158, view_1225);  round_158 = view_1225 = None
	        clamp_min_236 = torch.ops.aten.clamp_min.default(add_12053, -128);  add_12053 = None
	        clamp_max_157 = torch.ops.aten.clamp_max.default(clamp_min_236, 127);  clamp_min_236 = None
	        _assert_tensor_metadata_706 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_157, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_706 = None
	        convert_element_type_469 = torch.ops.prims.convert_element_type.default(clamp_max_157, torch.int8);  clamp_max_157 = None
	        view_1228 = torch.ops.aten.view.default(clamp_min_234, [sym_size_int, 1500, 1]);  clamp_min_234 = None
	        view_1229 = torch.ops.aten.view.default(convert_element_type_468, [sym_size_int, 1500, 1]);  convert_element_type_468 = None
	        _assert_tensor_metadata_707 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_469, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_707 = None
	        convert_element_type_470 = torch.ops.prims.convert_element_type.default(convert_element_type_469, torch.float32);  convert_element_type_469 = None
	        _assert_tensor_metadata_708 = torch.ops.aten._assert_tensor_metadata.default(view_1229, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_708 = None
	        convert_element_type_471 = torch.ops.prims.convert_element_type.default(view_1229, torch.float32);  view_1229 = None
	        sub_3606 = torch.ops.aten.sub.Tensor(convert_element_type_470, convert_element_type_471);  convert_element_type_470 = convert_element_type_471 = None
	        mul_7634 = torch.ops.aten.mul.Tensor(sub_3606, view_1228);  sub_3606 = view_1228 = None
	        _assert_tensor_metadata_709 = torch.ops.aten._assert_tensor_metadata.default(mul_7634, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_709 = None
	        view_1231 = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1232 = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1233 = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_710 = torch.ops.aten._assert_tensor_metadata.default(view_1231, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_710 = None
	        convert_element_type_472 = torch.ops.prims.convert_element_type.default(view_1231, torch.float32);  view_1231 = None
	        _assert_tensor_metadata_711 = torch.ops.aten._assert_tensor_metadata.default(view_1233, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_711 = None
	        convert_element_type_473 = torch.ops.prims.convert_element_type.default(view_1233, torch.float32);  view_1233 = None
	        sub_3610 = torch.ops.aten.sub.Tensor(convert_element_type_472, convert_element_type_473);  convert_element_type_472 = convert_element_type_473 = None
	        mul_7639 = torch.ops.aten.mul.Tensor(sub_3610, view_1232);  sub_3610 = view_1232 = None
	        view_1234 = torch.ops.aten.view.default(mul_7639, [1280, 1280]);  mul_7639 = None
	        _assert_tensor_metadata_712 = torch.ops.aten._assert_tensor_metadata.default(view_1234, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_712 = None
	        mul_7644 = sym_size_int * 1500
	        view_1235 = torch.ops.aten.view.default(mul_7634, [mul_7644, 1280]);  mul_7634 = mul_7644 = None
	        permute_131 = torch.ops.aten.permute.default(view_1234, [1, 0]);  view_1234 = None
	        addmm_65 = torch.ops.aten.addmm.default(model_audio_tower_layers_13_self_attn_q_proj_bias, view_1235, permute_131);  model_audio_tower_layers_13_self_attn_q_proj_bias = view_1235 = permute_131 = None
	        view_1236 = torch.ops.aten.view.default(addmm_65, [sym_size_int, 1500, 1280]);  addmm_65 = None
	        mul_7651 = torch.ops.aten.mul.Tensor(view_1236, 0.125);  view_1236 = None
	        view_1237 = torch.ops.aten.view.default(mul_7651, [sym_size_int, 1500, 20, 64]);  mul_7651 = None
	        permute_132 = torch.ops.aten.permute.default(view_1237, [0, 2, 1, 3]);  view_1237 = None
	        clone_106 = torch.ops.aten.clone.default(permute_132, memory_format = torch.contiguous_format);  permute_132 = None
	        amin_79 = torch.ops.aten.amin.default(add_11966, [2])
	        amax_79 = torch.ops.aten.amax.default(add_11966, [2])
	        full_158 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_79 = torch.ops.aten.minimum.default(amin_79, full_158);  amin_79 = full_158 = None
	        full_159 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_79 = torch.ops.aten.maximum.default(amax_79, full_159);  amax_79 = full_159 = None
	        sub_3625 = torch.ops.aten.sub.Tensor(maximum_79, minimum_79);  maximum_79 = None
	        div_158 = torch.ops.aten.div.Tensor(sub_3625, 255.0);  sub_3625 = None
	        clamp_min_237 = torch.ops.aten.clamp_min.default(div_158, 1.1920928955078125e-07);  div_158 = None
	        div_159 = torch.ops.aten.div.Tensor(minimum_79, clamp_min_237);  minimum_79 = None
	        round_159 = torch.ops.aten.round.default(div_159);  div_159 = None
	        sub_3631 = torch.ops.aten.sub.Tensor(-128, round_159);  round_159 = None
	        clamp_min_238 = torch.ops.aten.clamp_min.default(sub_3631, -128);  sub_3631 = None
	        clamp_max_158 = torch.ops.aten.clamp_max.default(clamp_min_238, 127);  clamp_min_238 = None
	        _assert_tensor_metadata_713 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_237, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_713 = None
	        _assert_tensor_metadata_714 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_158, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_714 = None
	        convert_element_type_474 = torch.ops.prims.convert_element_type.default(clamp_max_158, torch.int8);  clamp_max_158 = None
	        view_1240 = torch.ops.aten.view.default(clamp_min_237, [sym_size_int, 1500, 1])
	        view_1241 = torch.ops.aten.view.default(convert_element_type_474, [sym_size_int, 1500, 1])
	        reciprocal_79 = torch.ops.aten.reciprocal.default(view_1240);  view_1240 = None
	        mul_7705 = torch.ops.aten.mul.Tensor(reciprocal_79, 1.0);  reciprocal_79 = None
	        mul_7708 = torch.ops.aten.mul.Tensor(add_11966, mul_7705);  mul_7705 = None
	        round_160 = torch.ops.aten.round.default(mul_7708);  mul_7708 = None
	        add_12205 = torch.ops.aten.add.Tensor(round_160, view_1241);  round_160 = view_1241 = None
	        clamp_min_239 = torch.ops.aten.clamp_min.default(add_12205, -128);  add_12205 = None
	        clamp_max_159 = torch.ops.aten.clamp_max.default(clamp_min_239, 127);  clamp_min_239 = None
	        _assert_tensor_metadata_715 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_159, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_715 = None
	        convert_element_type_475 = torch.ops.prims.convert_element_type.default(clamp_max_159, torch.int8);  clamp_max_159 = None
	        view_1244 = torch.ops.aten.view.default(clamp_min_237, [sym_size_int, 1500, 1]);  clamp_min_237 = None
	        view_1245 = torch.ops.aten.view.default(convert_element_type_474, [sym_size_int, 1500, 1]);  convert_element_type_474 = None
	        _assert_tensor_metadata_716 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_475, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_716 = None
	        convert_element_type_476 = torch.ops.prims.convert_element_type.default(convert_element_type_475, torch.float32);  convert_element_type_475 = None
	        _assert_tensor_metadata_717 = torch.ops.aten._assert_tensor_metadata.default(view_1245, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_717 = None
	        convert_element_type_477 = torch.ops.prims.convert_element_type.default(view_1245, torch.float32);  view_1245 = None
	        sub_3651 = torch.ops.aten.sub.Tensor(convert_element_type_476, convert_element_type_477);  convert_element_type_476 = convert_element_type_477 = None
	        mul_7730 = torch.ops.aten.mul.Tensor(sub_3651, view_1244);  sub_3651 = view_1244 = None
	        _assert_tensor_metadata_718 = torch.ops.aten._assert_tensor_metadata.default(mul_7730, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_718 = None
	        view_1247 = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1248 = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1249 = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_719 = torch.ops.aten._assert_tensor_metadata.default(view_1247, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_719 = None
	        convert_element_type_478 = torch.ops.prims.convert_element_type.default(view_1247, torch.float32);  view_1247 = None
	        _assert_tensor_metadata_720 = torch.ops.aten._assert_tensor_metadata.default(view_1249, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_720 = None
	        convert_element_type_479 = torch.ops.prims.convert_element_type.default(view_1249, torch.float32);  view_1249 = None
	        sub_3655 = torch.ops.aten.sub.Tensor(convert_element_type_478, convert_element_type_479);  convert_element_type_478 = convert_element_type_479 = None
	        mul_7735 = torch.ops.aten.mul.Tensor(sub_3655, view_1248);  sub_3655 = view_1248 = None
	        view_1250 = torch.ops.aten.view.default(mul_7735, [1280, 1280]);  mul_7735 = None
	        _assert_tensor_metadata_721 = torch.ops.aten._assert_tensor_metadata.default(view_1250, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_721 = None
	        permute_133 = torch.ops.aten.permute.default(view_1250, [1, 0]);  view_1250 = None
	        mul_7738 = sym_size_int * 1500
	        view_1251 = torch.ops.aten.view.default(mul_7730, [mul_7738, 1280]);  mul_7730 = mul_7738 = None
	        mm_13 = torch.ops.aten.mm.default(view_1251, permute_133);  view_1251 = permute_133 = None
	        view_1252 = torch.ops.aten.view.default(mm_13, [sym_size_int, 1500, 1280]);  mm_13 = None
	        view_1253 = torch.ops.aten.view.default(view_1252, [sym_size_int, -1, 20, 64]);  view_1252 = None
	        permute_134 = torch.ops.aten.permute.default(view_1253, [0, 2, 1, 3]);  view_1253 = None
	        clone_107 = torch.ops.aten.clone.default(permute_134, memory_format = torch.contiguous_format);  permute_134 = None
	        amin_80 = torch.ops.aten.amin.default(add_11966, [2])
	        amax_80 = torch.ops.aten.amax.default(add_11966, [2])
	        full_160 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_80 = torch.ops.aten.minimum.default(amin_80, full_160);  amin_80 = full_160 = None
	        full_161 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_80 = torch.ops.aten.maximum.default(amax_80, full_161);  amax_80 = full_161 = None
	        sub_3669 = torch.ops.aten.sub.Tensor(maximum_80, minimum_80);  maximum_80 = None
	        div_160 = torch.ops.aten.div.Tensor(sub_3669, 255.0);  sub_3669 = None
	        clamp_min_240 = torch.ops.aten.clamp_min.default(div_160, 1.1920928955078125e-07);  div_160 = None
	        div_161 = torch.ops.aten.div.Tensor(minimum_80, clamp_min_240);  minimum_80 = None
	        round_161 = torch.ops.aten.round.default(div_161);  div_161 = None
	        sub_3675 = torch.ops.aten.sub.Tensor(-128, round_161);  round_161 = None
	        clamp_min_241 = torch.ops.aten.clamp_min.default(sub_3675, -128);  sub_3675 = None
	        clamp_max_160 = torch.ops.aten.clamp_max.default(clamp_min_241, 127);  clamp_min_241 = None
	        _assert_tensor_metadata_722 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_240, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_722 = None
	        _assert_tensor_metadata_723 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_160, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_723 = None
	        convert_element_type_480 = torch.ops.prims.convert_element_type.default(clamp_max_160, torch.int8);  clamp_max_160 = None
	        view_1256 = torch.ops.aten.view.default(clamp_min_240, [sym_size_int, 1500, 1])
	        view_1257 = torch.ops.aten.view.default(convert_element_type_480, [sym_size_int, 1500, 1])
	        reciprocal_80 = torch.ops.aten.reciprocal.default(view_1256);  view_1256 = None
	        mul_7804 = torch.ops.aten.mul.Tensor(reciprocal_80, 1.0);  reciprocal_80 = None
	        mul_7807 = torch.ops.aten.mul.Tensor(add_11966, mul_7804);  add_11966 = mul_7804 = None
	        round_162 = torch.ops.aten.round.default(mul_7807);  mul_7807 = None
	        add_12353 = torch.ops.aten.add.Tensor(round_162, view_1257);  round_162 = view_1257 = None
	        clamp_min_242 = torch.ops.aten.clamp_min.default(add_12353, -128);  add_12353 = None
	        clamp_max_161 = torch.ops.aten.clamp_max.default(clamp_min_242, 127);  clamp_min_242 = None
	        _assert_tensor_metadata_724 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_161, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_724 = None
	        convert_element_type_481 = torch.ops.prims.convert_element_type.default(clamp_max_161, torch.int8);  clamp_max_161 = None
	        view_1260 = torch.ops.aten.view.default(clamp_min_240, [sym_size_int, 1500, 1]);  clamp_min_240 = None
	        view_1261 = torch.ops.aten.view.default(convert_element_type_480, [sym_size_int, 1500, 1]);  convert_element_type_480 = None
	        _assert_tensor_metadata_725 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_481, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_725 = None
	        convert_element_type_482 = torch.ops.prims.convert_element_type.default(convert_element_type_481, torch.float32);  convert_element_type_481 = None
	        _assert_tensor_metadata_726 = torch.ops.aten._assert_tensor_metadata.default(view_1261, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_726 = None
	        convert_element_type_483 = torch.ops.prims.convert_element_type.default(view_1261, torch.float32);  view_1261 = None
	        sub_3695 = torch.ops.aten.sub.Tensor(convert_element_type_482, convert_element_type_483);  convert_element_type_482 = convert_element_type_483 = None
	        mul_7829 = torch.ops.aten.mul.Tensor(sub_3695, view_1260);  sub_3695 = view_1260 = None
	        _assert_tensor_metadata_727 = torch.ops.aten._assert_tensor_metadata.default(mul_7829, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_727 = None
	        view_1263 = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1264 = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1265 = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_728 = torch.ops.aten._assert_tensor_metadata.default(view_1263, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_728 = None
	        convert_element_type_484 = torch.ops.prims.convert_element_type.default(view_1263, torch.float32);  view_1263 = None
	        _assert_tensor_metadata_729 = torch.ops.aten._assert_tensor_metadata.default(view_1265, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_729 = None
	        convert_element_type_485 = torch.ops.prims.convert_element_type.default(view_1265, torch.float32);  view_1265 = None
	        sub_3699 = torch.ops.aten.sub.Tensor(convert_element_type_484, convert_element_type_485);  convert_element_type_484 = convert_element_type_485 = None
	        mul_7834 = torch.ops.aten.mul.Tensor(sub_3699, view_1264);  sub_3699 = view_1264 = None
	        view_1266 = torch.ops.aten.view.default(mul_7834, [1280, 1280]);  mul_7834 = None
	        _assert_tensor_metadata_730 = torch.ops.aten._assert_tensor_metadata.default(view_1266, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_730 = None
	        mul_7839 = sym_size_int * 1500
	        view_1267 = torch.ops.aten.view.default(mul_7829, [mul_7839, 1280]);  mul_7829 = mul_7839 = None
	        permute_135 = torch.ops.aten.permute.default(view_1266, [1, 0]);  view_1266 = None
	        addmm_66 = torch.ops.aten.addmm.default(model_audio_tower_layers_13_self_attn_v_proj_bias, view_1267, permute_135);  model_audio_tower_layers_13_self_attn_v_proj_bias = view_1267 = permute_135 = None
	        view_1268 = torch.ops.aten.view.default(addmm_66, [sym_size_int, 1500, 1280]);  addmm_66 = None
	        view_1269 = torch.ops.aten.view.default(view_1268, [sym_size_int, -1, 20, 64]);  view_1268 = None
	        permute_136 = torch.ops.aten.permute.default(view_1269, [0, 2, 1, 3]);  view_1269 = None
	        clone_108 = torch.ops.aten.clone.default(permute_136, memory_format = torch.contiguous_format);  permute_136 = None
	        _scaled_dot_product_efficient_attention_13 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_106, clone_107, clone_108, None, False, scale = 1.0);  clone_106 = clone_107 = clone_108 = None
	        getitem_106 = _scaled_dot_product_efficient_attention_13[0];  _scaled_dot_product_efficient_attention_13 = None
	        permute_137 = torch.ops.aten.permute.default(getitem_106, [0, 2, 1, 3]);  getitem_106 = None
	        view_1270 = torch.ops.aten.view.default(permute_137, [sym_size_int, 1500, -1]);  permute_137 = None
	        amin_81 = torch.ops.aten.amin.default(view_1270, [2])
	        amax_81 = torch.ops.aten.amax.default(view_1270, [2])
	        full_162 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_81 = torch.ops.aten.minimum.default(amin_81, full_162);  amin_81 = full_162 = None
	        full_163 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_81 = torch.ops.aten.maximum.default(amax_81, full_163);  amax_81 = full_163 = None
	        sub_3717 = torch.ops.aten.sub.Tensor(maximum_81, minimum_81);  maximum_81 = None
	        div_162 = torch.ops.aten.div.Tensor(sub_3717, 255.0);  sub_3717 = None
	        clamp_min_243 = torch.ops.aten.clamp_min.default(div_162, 1.1920928955078125e-07);  div_162 = None
	        div_163 = torch.ops.aten.div.Tensor(minimum_81, clamp_min_243);  minimum_81 = None
	        round_163 = torch.ops.aten.round.default(div_163);  div_163 = None
	        sub_3723 = torch.ops.aten.sub.Tensor(-128, round_163);  round_163 = None
	        clamp_min_244 = torch.ops.aten.clamp_min.default(sub_3723, -128);  sub_3723 = None
	        clamp_max_162 = torch.ops.aten.clamp_max.default(clamp_min_244, 127);  clamp_min_244 = None
	        _assert_tensor_metadata_731 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_243, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_731 = None
	        _assert_tensor_metadata_732 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_162, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_732 = None
	        convert_element_type_486 = torch.ops.prims.convert_element_type.default(clamp_max_162, torch.int8);  clamp_max_162 = None
	        view_1273 = torch.ops.aten.view.default(clamp_min_243, [sym_size_int, 1500, 1])
	        view_1274 = torch.ops.aten.view.default(convert_element_type_486, [sym_size_int, 1500, 1])
	        reciprocal_81 = torch.ops.aten.reciprocal.default(view_1273);  view_1273 = None
	        mul_7909 = torch.ops.aten.mul.Tensor(reciprocal_81, 1.0);  reciprocal_81 = None
	        mul_7912 = torch.ops.aten.mul.Tensor(view_1270, mul_7909);  view_1270 = mul_7909 = None
	        round_164 = torch.ops.aten.round.default(mul_7912);  mul_7912 = None
	        add_12517 = torch.ops.aten.add.Tensor(round_164, view_1274);  round_164 = view_1274 = None
	        clamp_min_245 = torch.ops.aten.clamp_min.default(add_12517, -128);  add_12517 = None
	        clamp_max_163 = torch.ops.aten.clamp_max.default(clamp_min_245, 127);  clamp_min_245 = None
	        _assert_tensor_metadata_733 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_163, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_733 = None
	        convert_element_type_487 = torch.ops.prims.convert_element_type.default(clamp_max_163, torch.int8);  clamp_max_163 = None
	        view_1277 = torch.ops.aten.view.default(clamp_min_243, [sym_size_int, 1500, 1]);  clamp_min_243 = None
	        view_1278 = torch.ops.aten.view.default(convert_element_type_486, [sym_size_int, 1500, 1]);  convert_element_type_486 = None
	        _assert_tensor_metadata_734 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_487, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_734 = None
	        convert_element_type_488 = torch.ops.prims.convert_element_type.default(convert_element_type_487, torch.float32);  convert_element_type_487 = None
	        _assert_tensor_metadata_735 = torch.ops.aten._assert_tensor_metadata.default(view_1278, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_735 = None
	        convert_element_type_489 = torch.ops.prims.convert_element_type.default(view_1278, torch.float32);  view_1278 = None
	        sub_3743 = torch.ops.aten.sub.Tensor(convert_element_type_488, convert_element_type_489);  convert_element_type_488 = convert_element_type_489 = None
	        mul_7934 = torch.ops.aten.mul.Tensor(sub_3743, view_1277);  sub_3743 = view_1277 = None
	        _assert_tensor_metadata_736 = torch.ops.aten._assert_tensor_metadata.default(mul_7934, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_736 = None
	        view_1280 = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1281 = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1282 = torch.ops.aten.view.default(model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_737 = torch.ops.aten._assert_tensor_metadata.default(view_1280, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_737 = None
	        convert_element_type_490 = torch.ops.prims.convert_element_type.default(view_1280, torch.float32);  view_1280 = None
	        _assert_tensor_metadata_738 = torch.ops.aten._assert_tensor_metadata.default(view_1282, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_738 = None
	        convert_element_type_491 = torch.ops.prims.convert_element_type.default(view_1282, torch.float32);  view_1282 = None
	        sub_3747 = torch.ops.aten.sub.Tensor(convert_element_type_490, convert_element_type_491);  convert_element_type_490 = convert_element_type_491 = None
	        mul_7939 = torch.ops.aten.mul.Tensor(sub_3747, view_1281);  sub_3747 = view_1281 = None
	        view_1283 = torch.ops.aten.view.default(mul_7939, [1280, 1280]);  mul_7939 = None
	        _assert_tensor_metadata_739 = torch.ops.aten._assert_tensor_metadata.default(view_1283, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_739 = None
	        mul_7944 = sym_size_int * 1500
	        view_1284 = torch.ops.aten.view.default(mul_7934, [mul_7944, 1280]);  mul_7934 = mul_7944 = None
	        permute_138 = torch.ops.aten.permute.default(view_1283, [1, 0]);  view_1283 = None
	        addmm_67 = torch.ops.aten.addmm.default(model_audio_tower_layers_13_self_attn_out_proj_bias, view_1284, permute_138);  model_audio_tower_layers_13_self_attn_out_proj_bias = view_1284 = permute_138 = None
	        view_1285 = torch.ops.aten.view.default(addmm_67, [sym_size_int, 1500, 1280]);  addmm_67 = None
	        add_12580 = torch.ops.aten.add.Tensor(add_11960, view_1285);  add_11960 = view_1285 = None
	        clone_110 = torch.ops.aten.clone.default(add_12580, memory_format = torch.contiguous_format)
	        var_mean_27 = torch.ops.aten.var_mean.correction(clone_110, [2], correction = 0, keepdim = True)
	        getitem_110 = var_mean_27[0]
	        getitem_111 = var_mean_27[1];  var_mean_27 = None
	        add_12585 = torch.ops.aten.add.Tensor(getitem_110, 1e-05);  getitem_110 = None
	        rsqrt_27 = torch.ops.aten.rsqrt.default(add_12585);  add_12585 = None
	        sub_3753 = torch.ops.aten.sub.Tensor(clone_110, getitem_111);  clone_110 = getitem_111 = None
	        mul_7955 = torch.ops.aten.mul.Tensor(sub_3753, rsqrt_27);  sub_3753 = rsqrt_27 = None
	        mul_7956 = torch.ops.aten.mul.Tensor(mul_7955, model_audio_tower_layers_13_final_layer_norm_weight);  mul_7955 = model_audio_tower_layers_13_final_layer_norm_weight = None
	        add_12586 = torch.ops.aten.add.Tensor(mul_7956, model_audio_tower_layers_13_final_layer_norm_bias);  mul_7956 = model_audio_tower_layers_13_final_layer_norm_bias = None
	        amin_82 = torch.ops.aten.amin.default(add_12586, [2])
	        amax_82 = torch.ops.aten.amax.default(add_12586, [2])
	        full_164 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_82 = torch.ops.aten.minimum.default(amin_82, full_164);  amin_82 = full_164 = None
	        full_165 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_82 = torch.ops.aten.maximum.default(amax_82, full_165);  amax_82 = full_165 = None
	        sub_3764 = torch.ops.aten.sub.Tensor(maximum_82, minimum_82);  maximum_82 = None
	        div_164 = torch.ops.aten.div.Tensor(sub_3764, 255.0);  sub_3764 = None
	        clamp_min_246 = torch.ops.aten.clamp_min.default(div_164, 1.1920928955078125e-07);  div_164 = None
	        div_165 = torch.ops.aten.div.Tensor(minimum_82, clamp_min_246);  minimum_82 = None
	        round_165 = torch.ops.aten.round.default(div_165);  div_165 = None
	        sub_3770 = torch.ops.aten.sub.Tensor(-128, round_165);  round_165 = None
	        clamp_min_247 = torch.ops.aten.clamp_min.default(sub_3770, -128);  sub_3770 = None
	        clamp_max_164 = torch.ops.aten.clamp_max.default(clamp_min_247, 127);  clamp_min_247 = None
	        _assert_tensor_metadata_740 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_246, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_740 = None
	        _assert_tensor_metadata_741 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_164, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_741 = None
	        convert_element_type_492 = torch.ops.prims.convert_element_type.default(clamp_max_164, torch.int8);  clamp_max_164 = None
	        view_1288 = torch.ops.aten.view.default(clamp_min_246, [sym_size_int, 1500, 1])
	        view_1289 = torch.ops.aten.view.default(convert_element_type_492, [sym_size_int, 1500, 1])
	        reciprocal_82 = torch.ops.aten.reciprocal.default(view_1288);  view_1288 = None
	        mul_8004 = torch.ops.aten.mul.Tensor(reciprocal_82, 1.0);  reciprocal_82 = None
	        mul_8007 = torch.ops.aten.mul.Tensor(add_12586, mul_8004);  add_12586 = mul_8004 = None
	        round_166 = torch.ops.aten.round.default(mul_8007);  mul_8007 = None
	        add_12673 = torch.ops.aten.add.Tensor(round_166, view_1289);  round_166 = view_1289 = None
	        clamp_min_248 = torch.ops.aten.clamp_min.default(add_12673, -128);  add_12673 = None
	        clamp_max_165 = torch.ops.aten.clamp_max.default(clamp_min_248, 127);  clamp_min_248 = None
	        _assert_tensor_metadata_742 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_165, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_742 = None
	        convert_element_type_493 = torch.ops.prims.convert_element_type.default(clamp_max_165, torch.int8);  clamp_max_165 = None
	        view_1292 = torch.ops.aten.view.default(clamp_min_246, [sym_size_int, 1500, 1]);  clamp_min_246 = None
	        view_1293 = torch.ops.aten.view.default(convert_element_type_492, [sym_size_int, 1500, 1]);  convert_element_type_492 = None
	        _assert_tensor_metadata_743 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_493, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_743 = None
	        convert_element_type_494 = torch.ops.prims.convert_element_type.default(convert_element_type_493, torch.float32);  convert_element_type_493 = None
	        _assert_tensor_metadata_744 = torch.ops.aten._assert_tensor_metadata.default(view_1293, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_744 = None
	        convert_element_type_495 = torch.ops.prims.convert_element_type.default(view_1293, torch.float32);  view_1293 = None
	        sub_3790 = torch.ops.aten.sub.Tensor(convert_element_type_494, convert_element_type_495);  convert_element_type_494 = convert_element_type_495 = None
	        mul_8029 = torch.ops.aten.mul.Tensor(sub_3790, view_1292);  sub_3790 = view_1292 = None
	        _assert_tensor_metadata_745 = torch.ops.aten._assert_tensor_metadata.default(mul_8029, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_745 = None
	        view_1295 = torch.ops.aten.view.default(model_audio_tower_layers_13_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_13_fc1_parametrizations_weight_original0 = None
	        view_1296 = torch.ops.aten.view.default(model_audio_tower_layers_13_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_13_fc1_parametrizations_weight_original1 = None
	        view_1297 = torch.ops.aten.view.default(model_audio_tower_layers_13_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_13_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_746 = torch.ops.aten._assert_tensor_metadata.default(view_1295, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_746 = None
	        convert_element_type_496 = torch.ops.prims.convert_element_type.default(view_1295, torch.float32);  view_1295 = None
	        _assert_tensor_metadata_747 = torch.ops.aten._assert_tensor_metadata.default(view_1297, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_747 = None
	        convert_element_type_497 = torch.ops.prims.convert_element_type.default(view_1297, torch.float32);  view_1297 = None
	        sub_3794 = torch.ops.aten.sub.Tensor(convert_element_type_496, convert_element_type_497);  convert_element_type_496 = convert_element_type_497 = None
	        mul_8034 = torch.ops.aten.mul.Tensor(sub_3794, view_1296);  sub_3794 = view_1296 = None
	        view_1298 = torch.ops.aten.view.default(mul_8034, [5120, 1280]);  mul_8034 = None
	        _assert_tensor_metadata_748 = torch.ops.aten._assert_tensor_metadata.default(view_1298, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_748 = None
	        mul_8039 = sym_size_int * 1500
	        view_1299 = torch.ops.aten.view.default(mul_8029, [mul_8039, 1280]);  mul_8029 = mul_8039 = None
	        permute_139 = torch.ops.aten.permute.default(view_1298, [1, 0]);  view_1298 = None
	        addmm_68 = torch.ops.aten.addmm.default(model_audio_tower_layers_13_fc1_bias, view_1299, permute_139);  model_audio_tower_layers_13_fc1_bias = view_1299 = permute_139 = None
	        view_1300 = torch.ops.aten.view.default(addmm_68, [sym_size_int, 1500, 5120]);  addmm_68 = None
	        mul_8046 = torch.ops.aten.mul.Tensor(view_1300, 0.5)
	        mul_8047 = torch.ops.aten.mul.Tensor(view_1300, 0.7071067811865476);  view_1300 = None
	        erf_15 = torch.ops.aten.erf.default(mul_8047);  mul_8047 = None
	        add_12732 = torch.ops.aten.add.Tensor(erf_15, 1);  erf_15 = None
	        mul_8048 = torch.ops.aten.mul.Tensor(mul_8046, add_12732);  mul_8046 = add_12732 = None
	        amin_83 = torch.ops.aten.amin.default(mul_8048, [2])
	        amax_83 = torch.ops.aten.amax.default(mul_8048, [2])
	        full_166 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_83 = torch.ops.aten.minimum.default(amin_83, full_166);  amin_83 = full_166 = None
	        full_167 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_83 = torch.ops.aten.maximum.default(amax_83, full_167);  amax_83 = full_167 = None
	        sub_3807 = torch.ops.aten.sub.Tensor(maximum_83, minimum_83);  maximum_83 = None
	        div_166 = torch.ops.aten.div.Tensor(sub_3807, 255.0);  sub_3807 = None
	        clamp_min_249 = torch.ops.aten.clamp_min.default(div_166, 1.1920928955078125e-07);  div_166 = None
	        div_167 = torch.ops.aten.div.Tensor(minimum_83, clamp_min_249);  minimum_83 = None
	        round_167 = torch.ops.aten.round.default(div_167);  div_167 = None
	        sub_3813 = torch.ops.aten.sub.Tensor(-128, round_167);  round_167 = None
	        clamp_min_250 = torch.ops.aten.clamp_min.default(sub_3813, -128);  sub_3813 = None
	        clamp_max_166 = torch.ops.aten.clamp_max.default(clamp_min_250, 127);  clamp_min_250 = None
	        _assert_tensor_metadata_749 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_749 = None
	        _assert_tensor_metadata_750 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_166, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_750 = None
	        convert_element_type_498 = torch.ops.prims.convert_element_type.default(clamp_max_166, torch.int8);  clamp_max_166 = None
	        view_1303 = torch.ops.aten.view.default(clamp_min_249, [sym_size_int, 1500, 1])
	        view_1304 = torch.ops.aten.view.default(convert_element_type_498, [sym_size_int, 1500, 1])
	        reciprocal_83 = torch.ops.aten.reciprocal.default(view_1303);  view_1303 = None
	        mul_8094 = torch.ops.aten.mul.Tensor(reciprocal_83, 1.0);  reciprocal_83 = None
	        mul_8097 = torch.ops.aten.mul.Tensor(mul_8048, mul_8094);  mul_8048 = mul_8094 = None
	        round_168 = torch.ops.aten.round.default(mul_8097);  mul_8097 = None
	        add_12815 = torch.ops.aten.add.Tensor(round_168, view_1304);  round_168 = view_1304 = None
	        clamp_min_251 = torch.ops.aten.clamp_min.default(add_12815, -128);  add_12815 = None
	        clamp_max_167 = torch.ops.aten.clamp_max.default(clamp_min_251, 127);  clamp_min_251 = None
	        _assert_tensor_metadata_751 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_167, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_751 = None
	        convert_element_type_499 = torch.ops.prims.convert_element_type.default(clamp_max_167, torch.int8);  clamp_max_167 = None
	        view_1307 = torch.ops.aten.view.default(clamp_min_249, [sym_size_int, 1500, 1]);  clamp_min_249 = None
	        view_1308 = torch.ops.aten.view.default(convert_element_type_498, [sym_size_int, 1500, 1]);  convert_element_type_498 = None
	        _assert_tensor_metadata_752 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_499, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_752 = None
	        convert_element_type_500 = torch.ops.prims.convert_element_type.default(convert_element_type_499, torch.float32);  convert_element_type_499 = None
	        _assert_tensor_metadata_753 = torch.ops.aten._assert_tensor_metadata.default(view_1308, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_753 = None
	        convert_element_type_501 = torch.ops.prims.convert_element_type.default(view_1308, torch.float32);  view_1308 = None
	        sub_3833 = torch.ops.aten.sub.Tensor(convert_element_type_500, convert_element_type_501);  convert_element_type_500 = convert_element_type_501 = None
	        mul_8119 = torch.ops.aten.mul.Tensor(sub_3833, view_1307);  sub_3833 = view_1307 = None
	        _assert_tensor_metadata_754 = torch.ops.aten._assert_tensor_metadata.default(mul_8119, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_754 = None
	        view_1310 = torch.ops.aten.view.default(model_audio_tower_layers_13_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_13_fc2_parametrizations_weight_original0 = None
	        view_1311 = torch.ops.aten.view.default(model_audio_tower_layers_13_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_13_fc2_parametrizations_weight_original1 = None
	        view_1312 = torch.ops.aten.view.default(model_audio_tower_layers_13_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_13_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_755 = torch.ops.aten._assert_tensor_metadata.default(view_1310, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_755 = None
	        convert_element_type_502 = torch.ops.prims.convert_element_type.default(view_1310, torch.float32);  view_1310 = None
	        _assert_tensor_metadata_756 = torch.ops.aten._assert_tensor_metadata.default(view_1312, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_756 = None
	        convert_element_type_503 = torch.ops.prims.convert_element_type.default(view_1312, torch.float32);  view_1312 = None
	        sub_3837 = torch.ops.aten.sub.Tensor(convert_element_type_502, convert_element_type_503);  convert_element_type_502 = convert_element_type_503 = None
	        mul_8124 = torch.ops.aten.mul.Tensor(sub_3837, view_1311);  sub_3837 = view_1311 = None
	        view_1313 = torch.ops.aten.view.default(mul_8124, [1280, 5120]);  mul_8124 = None
	        _assert_tensor_metadata_757 = torch.ops.aten._assert_tensor_metadata.default(view_1313, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_757 = None
	        mul_8129 = sym_size_int * 1500
	        view_1314 = torch.ops.aten.view.default(mul_8119, [mul_8129, 5120]);  mul_8119 = mul_8129 = None
	        permute_140 = torch.ops.aten.permute.default(view_1313, [1, 0]);  view_1313 = None
	        addmm_69 = torch.ops.aten.addmm.default(model_audio_tower_layers_13_fc2_bias, view_1314, permute_140);  model_audio_tower_layers_13_fc2_bias = view_1314 = permute_140 = None
	        view_1315 = torch.ops.aten.view.default(addmm_69, [sym_size_int, 1500, 1280]);  addmm_69 = None
	        add_12878 = torch.ops.aten.add.Tensor(add_12580, view_1315);  add_12580 = view_1315 = None
	        clone_113 = torch.ops.aten.clone.default(add_12878, memory_format = torch.contiguous_format)
	        var_mean_28 = torch.ops.aten.var_mean.correction(clone_113, [2], correction = 0, keepdim = True)
	        getitem_112 = var_mean_28[0]
	        getitem_113 = var_mean_28[1];  var_mean_28 = None
	        add_12883 = torch.ops.aten.add.Tensor(getitem_112, 1e-05);  getitem_112 = None
	        rsqrt_28 = torch.ops.aten.rsqrt.default(add_12883);  add_12883 = None
	        sub_3843 = torch.ops.aten.sub.Tensor(clone_113, getitem_113);  clone_113 = getitem_113 = None
	        mul_8140 = torch.ops.aten.mul.Tensor(sub_3843, rsqrt_28);  sub_3843 = rsqrt_28 = None
	        mul_8141 = torch.ops.aten.mul.Tensor(mul_8140, model_audio_tower_layers_14_self_attn_layer_norm_weight);  mul_8140 = model_audio_tower_layers_14_self_attn_layer_norm_weight = None
	        add_12884 = torch.ops.aten.add.Tensor(mul_8141, model_audio_tower_layers_14_self_attn_layer_norm_bias);  mul_8141 = model_audio_tower_layers_14_self_attn_layer_norm_bias = None
	        amin_84 = torch.ops.aten.amin.default(add_12884, [2])
	        amax_84 = torch.ops.aten.amax.default(add_12884, [2])
	        full_168 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_84 = torch.ops.aten.minimum.default(amin_84, full_168);  amin_84 = full_168 = None
	        full_169 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_84 = torch.ops.aten.maximum.default(amax_84, full_169);  amax_84 = full_169 = None
	        sub_3854 = torch.ops.aten.sub.Tensor(maximum_84, minimum_84);  maximum_84 = None
	        div_168 = torch.ops.aten.div.Tensor(sub_3854, 255.0);  sub_3854 = None
	        clamp_min_252 = torch.ops.aten.clamp_min.default(div_168, 1.1920928955078125e-07);  div_168 = None
	        div_169 = torch.ops.aten.div.Tensor(minimum_84, clamp_min_252);  minimum_84 = None
	        round_169 = torch.ops.aten.round.default(div_169);  div_169 = None
	        sub_3860 = torch.ops.aten.sub.Tensor(-128, round_169);  round_169 = None
	        clamp_min_253 = torch.ops.aten.clamp_min.default(sub_3860, -128);  sub_3860 = None
	        clamp_max_168 = torch.ops.aten.clamp_max.default(clamp_min_253, 127);  clamp_min_253 = None
	        _assert_tensor_metadata_758 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_252, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_758 = None
	        _assert_tensor_metadata_759 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_168, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_759 = None
	        convert_element_type_504 = torch.ops.prims.convert_element_type.default(clamp_max_168, torch.int8);  clamp_max_168 = None
	        view_1318 = torch.ops.aten.view.default(clamp_min_252, [sym_size_int, 1500, 1])
	        view_1319 = torch.ops.aten.view.default(convert_element_type_504, [sym_size_int, 1500, 1])
	        reciprocal_84 = torch.ops.aten.reciprocal.default(view_1318);  view_1318 = None
	        mul_8189 = torch.ops.aten.mul.Tensor(reciprocal_84, 1.0);  reciprocal_84 = None
	        mul_8192 = torch.ops.aten.mul.Tensor(add_12884, mul_8189);  mul_8189 = None
	        round_170 = torch.ops.aten.round.default(mul_8192);  mul_8192 = None
	        add_12971 = torch.ops.aten.add.Tensor(round_170, view_1319);  round_170 = view_1319 = None
	        clamp_min_254 = torch.ops.aten.clamp_min.default(add_12971, -128);  add_12971 = None
	        clamp_max_169 = torch.ops.aten.clamp_max.default(clamp_min_254, 127);  clamp_min_254 = None
	        _assert_tensor_metadata_760 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_169, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_760 = None
	        convert_element_type_505 = torch.ops.prims.convert_element_type.default(clamp_max_169, torch.int8);  clamp_max_169 = None
	        view_1322 = torch.ops.aten.view.default(clamp_min_252, [sym_size_int, 1500, 1]);  clamp_min_252 = None
	        view_1323 = torch.ops.aten.view.default(convert_element_type_504, [sym_size_int, 1500, 1]);  convert_element_type_504 = None
	        _assert_tensor_metadata_761 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_505, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_761 = None
	        convert_element_type_506 = torch.ops.prims.convert_element_type.default(convert_element_type_505, torch.float32);  convert_element_type_505 = None
	        _assert_tensor_metadata_762 = torch.ops.aten._assert_tensor_metadata.default(view_1323, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_762 = None
	        convert_element_type_507 = torch.ops.prims.convert_element_type.default(view_1323, torch.float32);  view_1323 = None
	        sub_3880 = torch.ops.aten.sub.Tensor(convert_element_type_506, convert_element_type_507);  convert_element_type_506 = convert_element_type_507 = None
	        mul_8214 = torch.ops.aten.mul.Tensor(sub_3880, view_1322);  sub_3880 = view_1322 = None
	        _assert_tensor_metadata_763 = torch.ops.aten._assert_tensor_metadata.default(mul_8214, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_763 = None
	        view_1325 = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1326 = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1327 = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_764 = torch.ops.aten._assert_tensor_metadata.default(view_1325, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_764 = None
	        convert_element_type_508 = torch.ops.prims.convert_element_type.default(view_1325, torch.float32);  view_1325 = None
	        _assert_tensor_metadata_765 = torch.ops.aten._assert_tensor_metadata.default(view_1327, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_765 = None
	        convert_element_type_509 = torch.ops.prims.convert_element_type.default(view_1327, torch.float32);  view_1327 = None
	        sub_3884 = torch.ops.aten.sub.Tensor(convert_element_type_508, convert_element_type_509);  convert_element_type_508 = convert_element_type_509 = None
	        mul_8219 = torch.ops.aten.mul.Tensor(sub_3884, view_1326);  sub_3884 = view_1326 = None
	        view_1328 = torch.ops.aten.view.default(mul_8219, [1280, 1280]);  mul_8219 = None
	        _assert_tensor_metadata_766 = torch.ops.aten._assert_tensor_metadata.default(view_1328, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_766 = None
	        mul_8224 = sym_size_int * 1500
	        view_1329 = torch.ops.aten.view.default(mul_8214, [mul_8224, 1280]);  mul_8214 = mul_8224 = None
	        permute_141 = torch.ops.aten.permute.default(view_1328, [1, 0]);  view_1328 = None
	        addmm_70 = torch.ops.aten.addmm.default(model_audio_tower_layers_14_self_attn_q_proj_bias, view_1329, permute_141);  model_audio_tower_layers_14_self_attn_q_proj_bias = view_1329 = permute_141 = None
	        view_1330 = torch.ops.aten.view.default(addmm_70, [sym_size_int, 1500, 1280]);  addmm_70 = None
	        mul_8231 = torch.ops.aten.mul.Tensor(view_1330, 0.125);  view_1330 = None
	        view_1331 = torch.ops.aten.view.default(mul_8231, [sym_size_int, 1500, 20, 64]);  mul_8231 = None
	        permute_142 = torch.ops.aten.permute.default(view_1331, [0, 2, 1, 3]);  view_1331 = None
	        clone_114 = torch.ops.aten.clone.default(permute_142, memory_format = torch.contiguous_format);  permute_142 = None
	        amin_85 = torch.ops.aten.amin.default(add_12884, [2])
	        amax_85 = torch.ops.aten.amax.default(add_12884, [2])
	        full_170 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_85 = torch.ops.aten.minimum.default(amin_85, full_170);  amin_85 = full_170 = None
	        full_171 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_85 = torch.ops.aten.maximum.default(amax_85, full_171);  amax_85 = full_171 = None
	        sub_3899 = torch.ops.aten.sub.Tensor(maximum_85, minimum_85);  maximum_85 = None
	        div_170 = torch.ops.aten.div.Tensor(sub_3899, 255.0);  sub_3899 = None
	        clamp_min_255 = torch.ops.aten.clamp_min.default(div_170, 1.1920928955078125e-07);  div_170 = None
	        div_171 = torch.ops.aten.div.Tensor(minimum_85, clamp_min_255);  minimum_85 = None
	        round_171 = torch.ops.aten.round.default(div_171);  div_171 = None
	        sub_3905 = torch.ops.aten.sub.Tensor(-128, round_171);  round_171 = None
	        clamp_min_256 = torch.ops.aten.clamp_min.default(sub_3905, -128);  sub_3905 = None
	        clamp_max_170 = torch.ops.aten.clamp_max.default(clamp_min_256, 127);  clamp_min_256 = None
	        _assert_tensor_metadata_767 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_255, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_767 = None
	        _assert_tensor_metadata_768 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_170, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_768 = None
	        convert_element_type_510 = torch.ops.prims.convert_element_type.default(clamp_max_170, torch.int8);  clamp_max_170 = None
	        view_1334 = torch.ops.aten.view.default(clamp_min_255, [sym_size_int, 1500, 1])
	        view_1335 = torch.ops.aten.view.default(convert_element_type_510, [sym_size_int, 1500, 1])
	        reciprocal_85 = torch.ops.aten.reciprocal.default(view_1334);  view_1334 = None
	        mul_8285 = torch.ops.aten.mul.Tensor(reciprocal_85, 1.0);  reciprocal_85 = None
	        mul_8288 = torch.ops.aten.mul.Tensor(add_12884, mul_8285);  mul_8285 = None
	        round_172 = torch.ops.aten.round.default(mul_8288);  mul_8288 = None
	        add_13123 = torch.ops.aten.add.Tensor(round_172, view_1335);  round_172 = view_1335 = None
	        clamp_min_257 = torch.ops.aten.clamp_min.default(add_13123, -128);  add_13123 = None
	        clamp_max_171 = torch.ops.aten.clamp_max.default(clamp_min_257, 127);  clamp_min_257 = None
	        _assert_tensor_metadata_769 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_171, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_769 = None
	        convert_element_type_511 = torch.ops.prims.convert_element_type.default(clamp_max_171, torch.int8);  clamp_max_171 = None
	        view_1338 = torch.ops.aten.view.default(clamp_min_255, [sym_size_int, 1500, 1]);  clamp_min_255 = None
	        view_1339 = torch.ops.aten.view.default(convert_element_type_510, [sym_size_int, 1500, 1]);  convert_element_type_510 = None
	        _assert_tensor_metadata_770 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_511, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_770 = None
	        convert_element_type_512 = torch.ops.prims.convert_element_type.default(convert_element_type_511, torch.float32);  convert_element_type_511 = None
	        _assert_tensor_metadata_771 = torch.ops.aten._assert_tensor_metadata.default(view_1339, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_771 = None
	        convert_element_type_513 = torch.ops.prims.convert_element_type.default(view_1339, torch.float32);  view_1339 = None
	        sub_3925 = torch.ops.aten.sub.Tensor(convert_element_type_512, convert_element_type_513);  convert_element_type_512 = convert_element_type_513 = None
	        mul_8310 = torch.ops.aten.mul.Tensor(sub_3925, view_1338);  sub_3925 = view_1338 = None
	        _assert_tensor_metadata_772 = torch.ops.aten._assert_tensor_metadata.default(mul_8310, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_772 = None
	        view_1341 = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1342 = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1343 = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_773 = torch.ops.aten._assert_tensor_metadata.default(view_1341, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_773 = None
	        convert_element_type_514 = torch.ops.prims.convert_element_type.default(view_1341, torch.float32);  view_1341 = None
	        _assert_tensor_metadata_774 = torch.ops.aten._assert_tensor_metadata.default(view_1343, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_774 = None
	        convert_element_type_515 = torch.ops.prims.convert_element_type.default(view_1343, torch.float32);  view_1343 = None
	        sub_3929 = torch.ops.aten.sub.Tensor(convert_element_type_514, convert_element_type_515);  convert_element_type_514 = convert_element_type_515 = None
	        mul_8315 = torch.ops.aten.mul.Tensor(sub_3929, view_1342);  sub_3929 = view_1342 = None
	        view_1344 = torch.ops.aten.view.default(mul_8315, [1280, 1280]);  mul_8315 = None
	        _assert_tensor_metadata_775 = torch.ops.aten._assert_tensor_metadata.default(view_1344, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_775 = None
	        permute_143 = torch.ops.aten.permute.default(view_1344, [1, 0]);  view_1344 = None
	        mul_8318 = sym_size_int * 1500
	        view_1345 = torch.ops.aten.view.default(mul_8310, [mul_8318, 1280]);  mul_8310 = mul_8318 = None
	        mm_14 = torch.ops.aten.mm.default(view_1345, permute_143);  view_1345 = permute_143 = None
	        view_1346 = torch.ops.aten.view.default(mm_14, [sym_size_int, 1500, 1280]);  mm_14 = None
	        view_1347 = torch.ops.aten.view.default(view_1346, [sym_size_int, -1, 20, 64]);  view_1346 = None
	        permute_144 = torch.ops.aten.permute.default(view_1347, [0, 2, 1, 3]);  view_1347 = None
	        clone_115 = torch.ops.aten.clone.default(permute_144, memory_format = torch.contiguous_format);  permute_144 = None
	        amin_86 = torch.ops.aten.amin.default(add_12884, [2])
	        amax_86 = torch.ops.aten.amax.default(add_12884, [2])
	        full_172 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_86 = torch.ops.aten.minimum.default(amin_86, full_172);  amin_86 = full_172 = None
	        full_173 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_86 = torch.ops.aten.maximum.default(amax_86, full_173);  amax_86 = full_173 = None
	        sub_3943 = torch.ops.aten.sub.Tensor(maximum_86, minimum_86);  maximum_86 = None
	        div_172 = torch.ops.aten.div.Tensor(sub_3943, 255.0);  sub_3943 = None
	        clamp_min_258 = torch.ops.aten.clamp_min.default(div_172, 1.1920928955078125e-07);  div_172 = None
	        div_173 = torch.ops.aten.div.Tensor(minimum_86, clamp_min_258);  minimum_86 = None
	        round_173 = torch.ops.aten.round.default(div_173);  div_173 = None
	        sub_3949 = torch.ops.aten.sub.Tensor(-128, round_173);  round_173 = None
	        clamp_min_259 = torch.ops.aten.clamp_min.default(sub_3949, -128);  sub_3949 = None
	        clamp_max_172 = torch.ops.aten.clamp_max.default(clamp_min_259, 127);  clamp_min_259 = None
	        _assert_tensor_metadata_776 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_258, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_776 = None
	        _assert_tensor_metadata_777 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_172, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_777 = None
	        convert_element_type_516 = torch.ops.prims.convert_element_type.default(clamp_max_172, torch.int8);  clamp_max_172 = None
	        view_1350 = torch.ops.aten.view.default(clamp_min_258, [sym_size_int, 1500, 1])
	        view_1351 = torch.ops.aten.view.default(convert_element_type_516, [sym_size_int, 1500, 1])
	        reciprocal_86 = torch.ops.aten.reciprocal.default(view_1350);  view_1350 = None
	        mul_8384 = torch.ops.aten.mul.Tensor(reciprocal_86, 1.0);  reciprocal_86 = None
	        mul_8387 = torch.ops.aten.mul.Tensor(add_12884, mul_8384);  add_12884 = mul_8384 = None
	        round_174 = torch.ops.aten.round.default(mul_8387);  mul_8387 = None
	        add_13271 = torch.ops.aten.add.Tensor(round_174, view_1351);  round_174 = view_1351 = None
	        clamp_min_260 = torch.ops.aten.clamp_min.default(add_13271, -128);  add_13271 = None
	        clamp_max_173 = torch.ops.aten.clamp_max.default(clamp_min_260, 127);  clamp_min_260 = None
	        _assert_tensor_metadata_778 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_173, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_778 = None
	        convert_element_type_517 = torch.ops.prims.convert_element_type.default(clamp_max_173, torch.int8);  clamp_max_173 = None
	        view_1354 = torch.ops.aten.view.default(clamp_min_258, [sym_size_int, 1500, 1]);  clamp_min_258 = None
	        view_1355 = torch.ops.aten.view.default(convert_element_type_516, [sym_size_int, 1500, 1]);  convert_element_type_516 = None
	        _assert_tensor_metadata_779 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_517, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_779 = None
	        convert_element_type_518 = torch.ops.prims.convert_element_type.default(convert_element_type_517, torch.float32);  convert_element_type_517 = None
	        _assert_tensor_metadata_780 = torch.ops.aten._assert_tensor_metadata.default(view_1355, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_780 = None
	        convert_element_type_519 = torch.ops.prims.convert_element_type.default(view_1355, torch.float32);  view_1355 = None
	        sub_3969 = torch.ops.aten.sub.Tensor(convert_element_type_518, convert_element_type_519);  convert_element_type_518 = convert_element_type_519 = None
	        mul_8409 = torch.ops.aten.mul.Tensor(sub_3969, view_1354);  sub_3969 = view_1354 = None
	        _assert_tensor_metadata_781 = torch.ops.aten._assert_tensor_metadata.default(mul_8409, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_781 = None
	        view_1357 = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1358 = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1359 = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_782 = torch.ops.aten._assert_tensor_metadata.default(view_1357, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_782 = None
	        convert_element_type_520 = torch.ops.prims.convert_element_type.default(view_1357, torch.float32);  view_1357 = None
	        _assert_tensor_metadata_783 = torch.ops.aten._assert_tensor_metadata.default(view_1359, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_783 = None
	        convert_element_type_521 = torch.ops.prims.convert_element_type.default(view_1359, torch.float32);  view_1359 = None
	        sub_3973 = torch.ops.aten.sub.Tensor(convert_element_type_520, convert_element_type_521);  convert_element_type_520 = convert_element_type_521 = None
	        mul_8414 = torch.ops.aten.mul.Tensor(sub_3973, view_1358);  sub_3973 = view_1358 = None
	        view_1360 = torch.ops.aten.view.default(mul_8414, [1280, 1280]);  mul_8414 = None
	        _assert_tensor_metadata_784 = torch.ops.aten._assert_tensor_metadata.default(view_1360, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_784 = None
	        mul_8419 = sym_size_int * 1500
	        view_1361 = torch.ops.aten.view.default(mul_8409, [mul_8419, 1280]);  mul_8409 = mul_8419 = None
	        permute_145 = torch.ops.aten.permute.default(view_1360, [1, 0]);  view_1360 = None
	        addmm_71 = torch.ops.aten.addmm.default(model_audio_tower_layers_14_self_attn_v_proj_bias, view_1361, permute_145);  model_audio_tower_layers_14_self_attn_v_proj_bias = view_1361 = permute_145 = None
	        view_1362 = torch.ops.aten.view.default(addmm_71, [sym_size_int, 1500, 1280]);  addmm_71 = None
	        view_1363 = torch.ops.aten.view.default(view_1362, [sym_size_int, -1, 20, 64]);  view_1362 = None
	        permute_146 = torch.ops.aten.permute.default(view_1363, [0, 2, 1, 3]);  view_1363 = None
	        clone_116 = torch.ops.aten.clone.default(permute_146, memory_format = torch.contiguous_format);  permute_146 = None
	        _scaled_dot_product_efficient_attention_14 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_114, clone_115, clone_116, None, False, scale = 1.0);  clone_114 = clone_115 = clone_116 = None
	        getitem_114 = _scaled_dot_product_efficient_attention_14[0];  _scaled_dot_product_efficient_attention_14 = None
	        permute_147 = torch.ops.aten.permute.default(getitem_114, [0, 2, 1, 3]);  getitem_114 = None
	        view_1364 = torch.ops.aten.view.default(permute_147, [sym_size_int, 1500, -1]);  permute_147 = None
	        amin_87 = torch.ops.aten.amin.default(view_1364, [2])
	        amax_87 = torch.ops.aten.amax.default(view_1364, [2])
	        full_174 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_87 = torch.ops.aten.minimum.default(amin_87, full_174);  amin_87 = full_174 = None
	        full_175 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_87 = torch.ops.aten.maximum.default(amax_87, full_175);  amax_87 = full_175 = None
	        sub_3991 = torch.ops.aten.sub.Tensor(maximum_87, minimum_87);  maximum_87 = None
	        div_174 = torch.ops.aten.div.Tensor(sub_3991, 255.0);  sub_3991 = None
	        clamp_min_261 = torch.ops.aten.clamp_min.default(div_174, 1.1920928955078125e-07);  div_174 = None
	        div_175 = torch.ops.aten.div.Tensor(minimum_87, clamp_min_261);  minimum_87 = None
	        round_175 = torch.ops.aten.round.default(div_175);  div_175 = None
	        sub_3997 = torch.ops.aten.sub.Tensor(-128, round_175);  round_175 = None
	        clamp_min_262 = torch.ops.aten.clamp_min.default(sub_3997, -128);  sub_3997 = None
	        clamp_max_174 = torch.ops.aten.clamp_max.default(clamp_min_262, 127);  clamp_min_262 = None
	        _assert_tensor_metadata_785 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_261, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_785 = None
	        _assert_tensor_metadata_786 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_174, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_786 = None
	        convert_element_type_522 = torch.ops.prims.convert_element_type.default(clamp_max_174, torch.int8);  clamp_max_174 = None
	        view_1367 = torch.ops.aten.view.default(clamp_min_261, [sym_size_int, 1500, 1])
	        view_1368 = torch.ops.aten.view.default(convert_element_type_522, [sym_size_int, 1500, 1])
	        reciprocal_87 = torch.ops.aten.reciprocal.default(view_1367);  view_1367 = None
	        mul_8489 = torch.ops.aten.mul.Tensor(reciprocal_87, 1.0);  reciprocal_87 = None
	        mul_8492 = torch.ops.aten.mul.Tensor(view_1364, mul_8489);  view_1364 = mul_8489 = None
	        round_176 = torch.ops.aten.round.default(mul_8492);  mul_8492 = None
	        add_13435 = torch.ops.aten.add.Tensor(round_176, view_1368);  round_176 = view_1368 = None
	        clamp_min_263 = torch.ops.aten.clamp_min.default(add_13435, -128);  add_13435 = None
	        clamp_max_175 = torch.ops.aten.clamp_max.default(clamp_min_263, 127);  clamp_min_263 = None
	        _assert_tensor_metadata_787 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_175, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_787 = None
	        convert_element_type_523 = torch.ops.prims.convert_element_type.default(clamp_max_175, torch.int8);  clamp_max_175 = None
	        view_1371 = torch.ops.aten.view.default(clamp_min_261, [sym_size_int, 1500, 1]);  clamp_min_261 = None
	        view_1372 = torch.ops.aten.view.default(convert_element_type_522, [sym_size_int, 1500, 1]);  convert_element_type_522 = None
	        _assert_tensor_metadata_788 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_523, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_788 = None
	        convert_element_type_524 = torch.ops.prims.convert_element_type.default(convert_element_type_523, torch.float32);  convert_element_type_523 = None
	        _assert_tensor_metadata_789 = torch.ops.aten._assert_tensor_metadata.default(view_1372, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_789 = None
	        convert_element_type_525 = torch.ops.prims.convert_element_type.default(view_1372, torch.float32);  view_1372 = None
	        sub_4017 = torch.ops.aten.sub.Tensor(convert_element_type_524, convert_element_type_525);  convert_element_type_524 = convert_element_type_525 = None
	        mul_8514 = torch.ops.aten.mul.Tensor(sub_4017, view_1371);  sub_4017 = view_1371 = None
	        _assert_tensor_metadata_790 = torch.ops.aten._assert_tensor_metadata.default(mul_8514, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_790 = None
	        view_1374 = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1375 = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1376 = torch.ops.aten.view.default(model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_791 = torch.ops.aten._assert_tensor_metadata.default(view_1374, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_791 = None
	        convert_element_type_526 = torch.ops.prims.convert_element_type.default(view_1374, torch.float32);  view_1374 = None
	        _assert_tensor_metadata_792 = torch.ops.aten._assert_tensor_metadata.default(view_1376, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_792 = None
	        convert_element_type_527 = torch.ops.prims.convert_element_type.default(view_1376, torch.float32);  view_1376 = None
	        sub_4021 = torch.ops.aten.sub.Tensor(convert_element_type_526, convert_element_type_527);  convert_element_type_526 = convert_element_type_527 = None
	        mul_8519 = torch.ops.aten.mul.Tensor(sub_4021, view_1375);  sub_4021 = view_1375 = None
	        view_1377 = torch.ops.aten.view.default(mul_8519, [1280, 1280]);  mul_8519 = None
	        _assert_tensor_metadata_793 = torch.ops.aten._assert_tensor_metadata.default(view_1377, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_793 = None
	        mul_8524 = sym_size_int * 1500
	        view_1378 = torch.ops.aten.view.default(mul_8514, [mul_8524, 1280]);  mul_8514 = mul_8524 = None
	        permute_148 = torch.ops.aten.permute.default(view_1377, [1, 0]);  view_1377 = None
	        addmm_72 = torch.ops.aten.addmm.default(model_audio_tower_layers_14_self_attn_out_proj_bias, view_1378, permute_148);  model_audio_tower_layers_14_self_attn_out_proj_bias = view_1378 = permute_148 = None
	        view_1379 = torch.ops.aten.view.default(addmm_72, [sym_size_int, 1500, 1280]);  addmm_72 = None
	        add_13498 = torch.ops.aten.add.Tensor(add_12878, view_1379);  add_12878 = view_1379 = None
	        clone_118 = torch.ops.aten.clone.default(add_13498, memory_format = torch.contiguous_format)
	        var_mean_29 = torch.ops.aten.var_mean.correction(clone_118, [2], correction = 0, keepdim = True)
	        getitem_118 = var_mean_29[0]
	        getitem_119 = var_mean_29[1];  var_mean_29 = None
	        add_13503 = torch.ops.aten.add.Tensor(getitem_118, 1e-05);  getitem_118 = None
	        rsqrt_29 = torch.ops.aten.rsqrt.default(add_13503);  add_13503 = None
	        sub_4027 = torch.ops.aten.sub.Tensor(clone_118, getitem_119);  clone_118 = getitem_119 = None
	        mul_8535 = torch.ops.aten.mul.Tensor(sub_4027, rsqrt_29);  sub_4027 = rsqrt_29 = None
	        mul_8536 = torch.ops.aten.mul.Tensor(mul_8535, model_audio_tower_layers_14_final_layer_norm_weight);  mul_8535 = model_audio_tower_layers_14_final_layer_norm_weight = None
	        add_13504 = torch.ops.aten.add.Tensor(mul_8536, model_audio_tower_layers_14_final_layer_norm_bias);  mul_8536 = model_audio_tower_layers_14_final_layer_norm_bias = None
	        amin_88 = torch.ops.aten.amin.default(add_13504, [2])
	        amax_88 = torch.ops.aten.amax.default(add_13504, [2])
	        full_176 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_88 = torch.ops.aten.minimum.default(amin_88, full_176);  amin_88 = full_176 = None
	        full_177 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_88 = torch.ops.aten.maximum.default(amax_88, full_177);  amax_88 = full_177 = None
	        sub_4038 = torch.ops.aten.sub.Tensor(maximum_88, minimum_88);  maximum_88 = None
	        div_176 = torch.ops.aten.div.Tensor(sub_4038, 255.0);  sub_4038 = None
	        clamp_min_264 = torch.ops.aten.clamp_min.default(div_176, 1.1920928955078125e-07);  div_176 = None
	        div_177 = torch.ops.aten.div.Tensor(minimum_88, clamp_min_264);  minimum_88 = None
	        round_177 = torch.ops.aten.round.default(div_177);  div_177 = None
	        sub_4044 = torch.ops.aten.sub.Tensor(-128, round_177);  round_177 = None
	        clamp_min_265 = torch.ops.aten.clamp_min.default(sub_4044, -128);  sub_4044 = None
	        clamp_max_176 = torch.ops.aten.clamp_max.default(clamp_min_265, 127);  clamp_min_265 = None
	        _assert_tensor_metadata_794 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_264, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_794 = None
	        _assert_tensor_metadata_795 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_176, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_795 = None
	        convert_element_type_528 = torch.ops.prims.convert_element_type.default(clamp_max_176, torch.int8);  clamp_max_176 = None
	        view_1382 = torch.ops.aten.view.default(clamp_min_264, [sym_size_int, 1500, 1])
	        view_1383 = torch.ops.aten.view.default(convert_element_type_528, [sym_size_int, 1500, 1])
	        reciprocal_88 = torch.ops.aten.reciprocal.default(view_1382);  view_1382 = None
	        mul_8584 = torch.ops.aten.mul.Tensor(reciprocal_88, 1.0);  reciprocal_88 = None
	        mul_8587 = torch.ops.aten.mul.Tensor(add_13504, mul_8584);  add_13504 = mul_8584 = None
	        round_178 = torch.ops.aten.round.default(mul_8587);  mul_8587 = None
	        add_13591 = torch.ops.aten.add.Tensor(round_178, view_1383);  round_178 = view_1383 = None
	        clamp_min_266 = torch.ops.aten.clamp_min.default(add_13591, -128);  add_13591 = None
	        clamp_max_177 = torch.ops.aten.clamp_max.default(clamp_min_266, 127);  clamp_min_266 = None
	        _assert_tensor_metadata_796 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_177, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_796 = None
	        convert_element_type_529 = torch.ops.prims.convert_element_type.default(clamp_max_177, torch.int8);  clamp_max_177 = None
	        view_1386 = torch.ops.aten.view.default(clamp_min_264, [sym_size_int, 1500, 1]);  clamp_min_264 = None
	        view_1387 = torch.ops.aten.view.default(convert_element_type_528, [sym_size_int, 1500, 1]);  convert_element_type_528 = None
	        _assert_tensor_metadata_797 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_529, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_797 = None
	        convert_element_type_530 = torch.ops.prims.convert_element_type.default(convert_element_type_529, torch.float32);  convert_element_type_529 = None
	        _assert_tensor_metadata_798 = torch.ops.aten._assert_tensor_metadata.default(view_1387, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_798 = None
	        convert_element_type_531 = torch.ops.prims.convert_element_type.default(view_1387, torch.float32);  view_1387 = None
	        sub_4064 = torch.ops.aten.sub.Tensor(convert_element_type_530, convert_element_type_531);  convert_element_type_530 = convert_element_type_531 = None
	        mul_8609 = torch.ops.aten.mul.Tensor(sub_4064, view_1386);  sub_4064 = view_1386 = None
	        _assert_tensor_metadata_799 = torch.ops.aten._assert_tensor_metadata.default(mul_8609, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_799 = None
	        view_1389 = torch.ops.aten.view.default(model_audio_tower_layers_14_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_14_fc1_parametrizations_weight_original0 = None
	        view_1390 = torch.ops.aten.view.default(model_audio_tower_layers_14_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_14_fc1_parametrizations_weight_original1 = None
	        view_1391 = torch.ops.aten.view.default(model_audio_tower_layers_14_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_14_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_800 = torch.ops.aten._assert_tensor_metadata.default(view_1389, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_800 = None
	        convert_element_type_532 = torch.ops.prims.convert_element_type.default(view_1389, torch.float32);  view_1389 = None
	        _assert_tensor_metadata_801 = torch.ops.aten._assert_tensor_metadata.default(view_1391, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_801 = None
	        convert_element_type_533 = torch.ops.prims.convert_element_type.default(view_1391, torch.float32);  view_1391 = None
	        sub_4068 = torch.ops.aten.sub.Tensor(convert_element_type_532, convert_element_type_533);  convert_element_type_532 = convert_element_type_533 = None
	        mul_8614 = torch.ops.aten.mul.Tensor(sub_4068, view_1390);  sub_4068 = view_1390 = None
	        view_1392 = torch.ops.aten.view.default(mul_8614, [5120, 1280]);  mul_8614 = None
	        _assert_tensor_metadata_802 = torch.ops.aten._assert_tensor_metadata.default(view_1392, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_802 = None
	        mul_8619 = sym_size_int * 1500
	        view_1393 = torch.ops.aten.view.default(mul_8609, [mul_8619, 1280]);  mul_8609 = mul_8619 = None
	        permute_149 = torch.ops.aten.permute.default(view_1392, [1, 0]);  view_1392 = None
	        addmm_73 = torch.ops.aten.addmm.default(model_audio_tower_layers_14_fc1_bias, view_1393, permute_149);  model_audio_tower_layers_14_fc1_bias = view_1393 = permute_149 = None
	        view_1394 = torch.ops.aten.view.default(addmm_73, [sym_size_int, 1500, 5120]);  addmm_73 = None
	        mul_8626 = torch.ops.aten.mul.Tensor(view_1394, 0.5)
	        mul_8627 = torch.ops.aten.mul.Tensor(view_1394, 0.7071067811865476);  view_1394 = None
	        erf_16 = torch.ops.aten.erf.default(mul_8627);  mul_8627 = None
	        add_13650 = torch.ops.aten.add.Tensor(erf_16, 1);  erf_16 = None
	        mul_8628 = torch.ops.aten.mul.Tensor(mul_8626, add_13650);  mul_8626 = add_13650 = None
	        amin_89 = torch.ops.aten.amin.default(mul_8628, [2])
	        amax_89 = torch.ops.aten.amax.default(mul_8628, [2])
	        full_178 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_89 = torch.ops.aten.minimum.default(amin_89, full_178);  amin_89 = full_178 = None
	        full_179 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_89 = torch.ops.aten.maximum.default(amax_89, full_179);  amax_89 = full_179 = None
	        sub_4081 = torch.ops.aten.sub.Tensor(maximum_89, minimum_89);  maximum_89 = None
	        div_178 = torch.ops.aten.div.Tensor(sub_4081, 255.0);  sub_4081 = None
	        clamp_min_267 = torch.ops.aten.clamp_min.default(div_178, 1.1920928955078125e-07);  div_178 = None
	        div_179 = torch.ops.aten.div.Tensor(minimum_89, clamp_min_267);  minimum_89 = None
	        round_179 = torch.ops.aten.round.default(div_179);  div_179 = None
	        sub_4087 = torch.ops.aten.sub.Tensor(-128, round_179);  round_179 = None
	        clamp_min_268 = torch.ops.aten.clamp_min.default(sub_4087, -128);  sub_4087 = None
	        clamp_max_178 = torch.ops.aten.clamp_max.default(clamp_min_268, 127);  clamp_min_268 = None
	        _assert_tensor_metadata_803 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_267, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_803 = None
	        _assert_tensor_metadata_804 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_178, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_804 = None
	        convert_element_type_534 = torch.ops.prims.convert_element_type.default(clamp_max_178, torch.int8);  clamp_max_178 = None
	        view_1397 = torch.ops.aten.view.default(clamp_min_267, [sym_size_int, 1500, 1])
	        view_1398 = torch.ops.aten.view.default(convert_element_type_534, [sym_size_int, 1500, 1])
	        reciprocal_89 = torch.ops.aten.reciprocal.default(view_1397);  view_1397 = None
	        mul_8674 = torch.ops.aten.mul.Tensor(reciprocal_89, 1.0);  reciprocal_89 = None
	        mul_8677 = torch.ops.aten.mul.Tensor(mul_8628, mul_8674);  mul_8628 = mul_8674 = None
	        round_180 = torch.ops.aten.round.default(mul_8677);  mul_8677 = None
	        add_13733 = torch.ops.aten.add.Tensor(round_180, view_1398);  round_180 = view_1398 = None
	        clamp_min_269 = torch.ops.aten.clamp_min.default(add_13733, -128);  add_13733 = None
	        clamp_max_179 = torch.ops.aten.clamp_max.default(clamp_min_269, 127);  clamp_min_269 = None
	        _assert_tensor_metadata_805 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_179, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_805 = None
	        convert_element_type_535 = torch.ops.prims.convert_element_type.default(clamp_max_179, torch.int8);  clamp_max_179 = None
	        view_1401 = torch.ops.aten.view.default(clamp_min_267, [sym_size_int, 1500, 1]);  clamp_min_267 = None
	        view_1402 = torch.ops.aten.view.default(convert_element_type_534, [sym_size_int, 1500, 1]);  convert_element_type_534 = None
	        _assert_tensor_metadata_806 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_535, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_806 = None
	        convert_element_type_536 = torch.ops.prims.convert_element_type.default(convert_element_type_535, torch.float32);  convert_element_type_535 = None
	        _assert_tensor_metadata_807 = torch.ops.aten._assert_tensor_metadata.default(view_1402, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_807 = None
	        convert_element_type_537 = torch.ops.prims.convert_element_type.default(view_1402, torch.float32);  view_1402 = None
	        sub_4107 = torch.ops.aten.sub.Tensor(convert_element_type_536, convert_element_type_537);  convert_element_type_536 = convert_element_type_537 = None
	        mul_8699 = torch.ops.aten.mul.Tensor(sub_4107, view_1401);  sub_4107 = view_1401 = None
	        _assert_tensor_metadata_808 = torch.ops.aten._assert_tensor_metadata.default(mul_8699, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_808 = None
	        view_1404 = torch.ops.aten.view.default(model_audio_tower_layers_14_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_14_fc2_parametrizations_weight_original0 = None
	        view_1405 = torch.ops.aten.view.default(model_audio_tower_layers_14_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_14_fc2_parametrizations_weight_original1 = None
	        view_1406 = torch.ops.aten.view.default(model_audio_tower_layers_14_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_14_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_809 = torch.ops.aten._assert_tensor_metadata.default(view_1404, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_809 = None
	        convert_element_type_538 = torch.ops.prims.convert_element_type.default(view_1404, torch.float32);  view_1404 = None
	        _assert_tensor_metadata_810 = torch.ops.aten._assert_tensor_metadata.default(view_1406, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_810 = None
	        convert_element_type_539 = torch.ops.prims.convert_element_type.default(view_1406, torch.float32);  view_1406 = None
	        sub_4111 = torch.ops.aten.sub.Tensor(convert_element_type_538, convert_element_type_539);  convert_element_type_538 = convert_element_type_539 = None
	        mul_8704 = torch.ops.aten.mul.Tensor(sub_4111, view_1405);  sub_4111 = view_1405 = None
	        view_1407 = torch.ops.aten.view.default(mul_8704, [1280, 5120]);  mul_8704 = None
	        _assert_tensor_metadata_811 = torch.ops.aten._assert_tensor_metadata.default(view_1407, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_811 = None
	        mul_8709 = sym_size_int * 1500
	        view_1408 = torch.ops.aten.view.default(mul_8699, [mul_8709, 5120]);  mul_8699 = mul_8709 = None
	        permute_150 = torch.ops.aten.permute.default(view_1407, [1, 0]);  view_1407 = None
	        addmm_74 = torch.ops.aten.addmm.default(model_audio_tower_layers_14_fc2_bias, view_1408, permute_150);  model_audio_tower_layers_14_fc2_bias = view_1408 = permute_150 = None
	        view_1409 = torch.ops.aten.view.default(addmm_74, [sym_size_int, 1500, 1280]);  addmm_74 = None
	        add_13796 = torch.ops.aten.add.Tensor(add_13498, view_1409);  add_13498 = view_1409 = None
	        clone_121 = torch.ops.aten.clone.default(add_13796, memory_format = torch.contiguous_format)
	        var_mean_30 = torch.ops.aten.var_mean.correction(clone_121, [2], correction = 0, keepdim = True)
	        getitem_120 = var_mean_30[0]
	        getitem_121 = var_mean_30[1];  var_mean_30 = None
	        add_13801 = torch.ops.aten.add.Tensor(getitem_120, 1e-05);  getitem_120 = None
	        rsqrt_30 = torch.ops.aten.rsqrt.default(add_13801);  add_13801 = None
	        sub_4117 = torch.ops.aten.sub.Tensor(clone_121, getitem_121);  clone_121 = getitem_121 = None
	        mul_8720 = torch.ops.aten.mul.Tensor(sub_4117, rsqrt_30);  sub_4117 = rsqrt_30 = None
	        mul_8721 = torch.ops.aten.mul.Tensor(mul_8720, model_audio_tower_layers_15_self_attn_layer_norm_weight);  mul_8720 = model_audio_tower_layers_15_self_attn_layer_norm_weight = None
	        add_13802 = torch.ops.aten.add.Tensor(mul_8721, model_audio_tower_layers_15_self_attn_layer_norm_bias);  mul_8721 = model_audio_tower_layers_15_self_attn_layer_norm_bias = None
	        amin_90 = torch.ops.aten.amin.default(add_13802, [2])
	        amax_90 = torch.ops.aten.amax.default(add_13802, [2])
	        full_180 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_90 = torch.ops.aten.minimum.default(amin_90, full_180);  amin_90 = full_180 = None
	        full_181 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_90 = torch.ops.aten.maximum.default(amax_90, full_181);  amax_90 = full_181 = None
	        sub_4128 = torch.ops.aten.sub.Tensor(maximum_90, minimum_90);  maximum_90 = None
	        div_180 = torch.ops.aten.div.Tensor(sub_4128, 255.0);  sub_4128 = None
	        clamp_min_270 = torch.ops.aten.clamp_min.default(div_180, 1.1920928955078125e-07);  div_180 = None
	        div_181 = torch.ops.aten.div.Tensor(minimum_90, clamp_min_270);  minimum_90 = None
	        round_181 = torch.ops.aten.round.default(div_181);  div_181 = None
	        sub_4134 = torch.ops.aten.sub.Tensor(-128, round_181);  round_181 = None
	        clamp_min_271 = torch.ops.aten.clamp_min.default(sub_4134, -128);  sub_4134 = None
	        clamp_max_180 = torch.ops.aten.clamp_max.default(clamp_min_271, 127);  clamp_min_271 = None
	        _assert_tensor_metadata_812 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_270, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_812 = None
	        _assert_tensor_metadata_813 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_180, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_813 = None
	        convert_element_type_540 = torch.ops.prims.convert_element_type.default(clamp_max_180, torch.int8);  clamp_max_180 = None
	        view_1412 = torch.ops.aten.view.default(clamp_min_270, [sym_size_int, 1500, 1])
	        view_1413 = torch.ops.aten.view.default(convert_element_type_540, [sym_size_int, 1500, 1])
	        reciprocal_90 = torch.ops.aten.reciprocal.default(view_1412);  view_1412 = None
	        mul_8769 = torch.ops.aten.mul.Tensor(reciprocal_90, 1.0);  reciprocal_90 = None
	        mul_8772 = torch.ops.aten.mul.Tensor(add_13802, mul_8769);  mul_8769 = None
	        round_182 = torch.ops.aten.round.default(mul_8772);  mul_8772 = None
	        add_13889 = torch.ops.aten.add.Tensor(round_182, view_1413);  round_182 = view_1413 = None
	        clamp_min_272 = torch.ops.aten.clamp_min.default(add_13889, -128);  add_13889 = None
	        clamp_max_181 = torch.ops.aten.clamp_max.default(clamp_min_272, 127);  clamp_min_272 = None
	        _assert_tensor_metadata_814 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_181, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_814 = None
	        convert_element_type_541 = torch.ops.prims.convert_element_type.default(clamp_max_181, torch.int8);  clamp_max_181 = None
	        view_1416 = torch.ops.aten.view.default(clamp_min_270, [sym_size_int, 1500, 1]);  clamp_min_270 = None
	        view_1417 = torch.ops.aten.view.default(convert_element_type_540, [sym_size_int, 1500, 1]);  convert_element_type_540 = None
	        _assert_tensor_metadata_815 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_541, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_815 = None
	        convert_element_type_542 = torch.ops.prims.convert_element_type.default(convert_element_type_541, torch.float32);  convert_element_type_541 = None
	        _assert_tensor_metadata_816 = torch.ops.aten._assert_tensor_metadata.default(view_1417, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_816 = None
	        convert_element_type_543 = torch.ops.prims.convert_element_type.default(view_1417, torch.float32);  view_1417 = None
	        sub_4154 = torch.ops.aten.sub.Tensor(convert_element_type_542, convert_element_type_543);  convert_element_type_542 = convert_element_type_543 = None
	        mul_8794 = torch.ops.aten.mul.Tensor(sub_4154, view_1416);  sub_4154 = view_1416 = None
	        _assert_tensor_metadata_817 = torch.ops.aten._assert_tensor_metadata.default(mul_8794, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_817 = None
	        view_1419 = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1420 = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1421 = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_818 = torch.ops.aten._assert_tensor_metadata.default(view_1419, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_818 = None
	        convert_element_type_544 = torch.ops.prims.convert_element_type.default(view_1419, torch.float32);  view_1419 = None
	        _assert_tensor_metadata_819 = torch.ops.aten._assert_tensor_metadata.default(view_1421, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_819 = None
	        convert_element_type_545 = torch.ops.prims.convert_element_type.default(view_1421, torch.float32);  view_1421 = None
	        sub_4158 = torch.ops.aten.sub.Tensor(convert_element_type_544, convert_element_type_545);  convert_element_type_544 = convert_element_type_545 = None
	        mul_8799 = torch.ops.aten.mul.Tensor(sub_4158, view_1420);  sub_4158 = view_1420 = None
	        view_1422 = torch.ops.aten.view.default(mul_8799, [1280, 1280]);  mul_8799 = None
	        _assert_tensor_metadata_820 = torch.ops.aten._assert_tensor_metadata.default(view_1422, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_820 = None
	        mul_8804 = sym_size_int * 1500
	        view_1423 = torch.ops.aten.view.default(mul_8794, [mul_8804, 1280]);  mul_8794 = mul_8804 = None
	        permute_151 = torch.ops.aten.permute.default(view_1422, [1, 0]);  view_1422 = None
	        addmm_75 = torch.ops.aten.addmm.default(model_audio_tower_layers_15_self_attn_q_proj_bias, view_1423, permute_151);  model_audio_tower_layers_15_self_attn_q_proj_bias = view_1423 = permute_151 = None
	        view_1424 = torch.ops.aten.view.default(addmm_75, [sym_size_int, 1500, 1280]);  addmm_75 = None
	        mul_8811 = torch.ops.aten.mul.Tensor(view_1424, 0.125);  view_1424 = None
	        view_1425 = torch.ops.aten.view.default(mul_8811, [sym_size_int, 1500, 20, 64]);  mul_8811 = None
	        permute_152 = torch.ops.aten.permute.default(view_1425, [0, 2, 1, 3]);  view_1425 = None
	        clone_122 = torch.ops.aten.clone.default(permute_152, memory_format = torch.contiguous_format);  permute_152 = None
	        amin_91 = torch.ops.aten.amin.default(add_13802, [2])
	        amax_91 = torch.ops.aten.amax.default(add_13802, [2])
	        full_182 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_91 = torch.ops.aten.minimum.default(amin_91, full_182);  amin_91 = full_182 = None
	        full_183 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_91 = torch.ops.aten.maximum.default(amax_91, full_183);  amax_91 = full_183 = None
	        sub_4173 = torch.ops.aten.sub.Tensor(maximum_91, minimum_91);  maximum_91 = None
	        div_182 = torch.ops.aten.div.Tensor(sub_4173, 255.0);  sub_4173 = None
	        clamp_min_273 = torch.ops.aten.clamp_min.default(div_182, 1.1920928955078125e-07);  div_182 = None
	        div_183 = torch.ops.aten.div.Tensor(minimum_91, clamp_min_273);  minimum_91 = None
	        round_183 = torch.ops.aten.round.default(div_183);  div_183 = None
	        sub_4179 = torch.ops.aten.sub.Tensor(-128, round_183);  round_183 = None
	        clamp_min_274 = torch.ops.aten.clamp_min.default(sub_4179, -128);  sub_4179 = None
	        clamp_max_182 = torch.ops.aten.clamp_max.default(clamp_min_274, 127);  clamp_min_274 = None
	        _assert_tensor_metadata_821 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_273, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_821 = None
	        _assert_tensor_metadata_822 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_182, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_822 = None
	        convert_element_type_546 = torch.ops.prims.convert_element_type.default(clamp_max_182, torch.int8);  clamp_max_182 = None
	        view_1428 = torch.ops.aten.view.default(clamp_min_273, [sym_size_int, 1500, 1])
	        view_1429 = torch.ops.aten.view.default(convert_element_type_546, [sym_size_int, 1500, 1])
	        reciprocal_91 = torch.ops.aten.reciprocal.default(view_1428);  view_1428 = None
	        mul_8865 = torch.ops.aten.mul.Tensor(reciprocal_91, 1.0);  reciprocal_91 = None
	        mul_8868 = torch.ops.aten.mul.Tensor(add_13802, mul_8865);  mul_8865 = None
	        round_184 = torch.ops.aten.round.default(mul_8868);  mul_8868 = None
	        add_14041 = torch.ops.aten.add.Tensor(round_184, view_1429);  round_184 = view_1429 = None
	        clamp_min_275 = torch.ops.aten.clamp_min.default(add_14041, -128);  add_14041 = None
	        clamp_max_183 = torch.ops.aten.clamp_max.default(clamp_min_275, 127);  clamp_min_275 = None
	        _assert_tensor_metadata_823 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_183, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_823 = None
	        convert_element_type_547 = torch.ops.prims.convert_element_type.default(clamp_max_183, torch.int8);  clamp_max_183 = None
	        view_1432 = torch.ops.aten.view.default(clamp_min_273, [sym_size_int, 1500, 1]);  clamp_min_273 = None
	        view_1433 = torch.ops.aten.view.default(convert_element_type_546, [sym_size_int, 1500, 1]);  convert_element_type_546 = None
	        _assert_tensor_metadata_824 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_547, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_824 = None
	        convert_element_type_548 = torch.ops.prims.convert_element_type.default(convert_element_type_547, torch.float32);  convert_element_type_547 = None
	        _assert_tensor_metadata_825 = torch.ops.aten._assert_tensor_metadata.default(view_1433, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_825 = None
	        convert_element_type_549 = torch.ops.prims.convert_element_type.default(view_1433, torch.float32);  view_1433 = None
	        sub_4199 = torch.ops.aten.sub.Tensor(convert_element_type_548, convert_element_type_549);  convert_element_type_548 = convert_element_type_549 = None
	        mul_8890 = torch.ops.aten.mul.Tensor(sub_4199, view_1432);  sub_4199 = view_1432 = None
	        _assert_tensor_metadata_826 = torch.ops.aten._assert_tensor_metadata.default(mul_8890, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_826 = None
	        view_1435 = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1436 = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1437 = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_827 = torch.ops.aten._assert_tensor_metadata.default(view_1435, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_827 = None
	        convert_element_type_550 = torch.ops.prims.convert_element_type.default(view_1435, torch.float32);  view_1435 = None
	        _assert_tensor_metadata_828 = torch.ops.aten._assert_tensor_metadata.default(view_1437, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_828 = None
	        convert_element_type_551 = torch.ops.prims.convert_element_type.default(view_1437, torch.float32);  view_1437 = None
	        sub_4203 = torch.ops.aten.sub.Tensor(convert_element_type_550, convert_element_type_551);  convert_element_type_550 = convert_element_type_551 = None
	        mul_8895 = torch.ops.aten.mul.Tensor(sub_4203, view_1436);  sub_4203 = view_1436 = None
	        view_1438 = torch.ops.aten.view.default(mul_8895, [1280, 1280]);  mul_8895 = None
	        _assert_tensor_metadata_829 = torch.ops.aten._assert_tensor_metadata.default(view_1438, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_829 = None
	        permute_153 = torch.ops.aten.permute.default(view_1438, [1, 0]);  view_1438 = None
	        mul_8898 = sym_size_int * 1500
	        view_1439 = torch.ops.aten.view.default(mul_8890, [mul_8898, 1280]);  mul_8890 = mul_8898 = None
	        mm_15 = torch.ops.aten.mm.default(view_1439, permute_153);  view_1439 = permute_153 = None
	        view_1440 = torch.ops.aten.view.default(mm_15, [sym_size_int, 1500, 1280]);  mm_15 = None
	        view_1441 = torch.ops.aten.view.default(view_1440, [sym_size_int, -1, 20, 64]);  view_1440 = None
	        permute_154 = torch.ops.aten.permute.default(view_1441, [0, 2, 1, 3]);  view_1441 = None
	        clone_123 = torch.ops.aten.clone.default(permute_154, memory_format = torch.contiguous_format);  permute_154 = None
	        amin_92 = torch.ops.aten.amin.default(add_13802, [2])
	        amax_92 = torch.ops.aten.amax.default(add_13802, [2])
	        full_184 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_92 = torch.ops.aten.minimum.default(amin_92, full_184);  amin_92 = full_184 = None
	        full_185 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_92 = torch.ops.aten.maximum.default(amax_92, full_185);  amax_92 = full_185 = None
	        sub_4217 = torch.ops.aten.sub.Tensor(maximum_92, minimum_92);  maximum_92 = None
	        div_184 = torch.ops.aten.div.Tensor(sub_4217, 255.0);  sub_4217 = None
	        clamp_min_276 = torch.ops.aten.clamp_min.default(div_184, 1.1920928955078125e-07);  div_184 = None
	        div_185 = torch.ops.aten.div.Tensor(minimum_92, clamp_min_276);  minimum_92 = None
	        round_185 = torch.ops.aten.round.default(div_185);  div_185 = None
	        sub_4223 = torch.ops.aten.sub.Tensor(-128, round_185);  round_185 = None
	        clamp_min_277 = torch.ops.aten.clamp_min.default(sub_4223, -128);  sub_4223 = None
	        clamp_max_184 = torch.ops.aten.clamp_max.default(clamp_min_277, 127);  clamp_min_277 = None
	        _assert_tensor_metadata_830 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_276, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_830 = None
	        _assert_tensor_metadata_831 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_184, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_831 = None
	        convert_element_type_552 = torch.ops.prims.convert_element_type.default(clamp_max_184, torch.int8);  clamp_max_184 = None
	        view_1444 = torch.ops.aten.view.default(clamp_min_276, [sym_size_int, 1500, 1])
	        view_1445 = torch.ops.aten.view.default(convert_element_type_552, [sym_size_int, 1500, 1])
	        reciprocal_92 = torch.ops.aten.reciprocal.default(view_1444);  view_1444 = None
	        mul_8964 = torch.ops.aten.mul.Tensor(reciprocal_92, 1.0);  reciprocal_92 = None
	        mul_8967 = torch.ops.aten.mul.Tensor(add_13802, mul_8964);  add_13802 = mul_8964 = None
	        round_186 = torch.ops.aten.round.default(mul_8967);  mul_8967 = None
	        add_14189 = torch.ops.aten.add.Tensor(round_186, view_1445);  round_186 = view_1445 = None
	        clamp_min_278 = torch.ops.aten.clamp_min.default(add_14189, -128);  add_14189 = None
	        clamp_max_185 = torch.ops.aten.clamp_max.default(clamp_min_278, 127);  clamp_min_278 = None
	        _assert_tensor_metadata_832 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_185, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_832 = None
	        convert_element_type_553 = torch.ops.prims.convert_element_type.default(clamp_max_185, torch.int8);  clamp_max_185 = None
	        view_1448 = torch.ops.aten.view.default(clamp_min_276, [sym_size_int, 1500, 1]);  clamp_min_276 = None
	        view_1449 = torch.ops.aten.view.default(convert_element_type_552, [sym_size_int, 1500, 1]);  convert_element_type_552 = None
	        _assert_tensor_metadata_833 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_553, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_833 = None
	        convert_element_type_554 = torch.ops.prims.convert_element_type.default(convert_element_type_553, torch.float32);  convert_element_type_553 = None
	        _assert_tensor_metadata_834 = torch.ops.aten._assert_tensor_metadata.default(view_1449, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_834 = None
	        convert_element_type_555 = torch.ops.prims.convert_element_type.default(view_1449, torch.float32);  view_1449 = None
	        sub_4243 = torch.ops.aten.sub.Tensor(convert_element_type_554, convert_element_type_555);  convert_element_type_554 = convert_element_type_555 = None
	        mul_8989 = torch.ops.aten.mul.Tensor(sub_4243, view_1448);  sub_4243 = view_1448 = None
	        _assert_tensor_metadata_835 = torch.ops.aten._assert_tensor_metadata.default(mul_8989, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_835 = None
	        view_1451 = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1452 = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1453 = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_836 = torch.ops.aten._assert_tensor_metadata.default(view_1451, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_836 = None
	        convert_element_type_556 = torch.ops.prims.convert_element_type.default(view_1451, torch.float32);  view_1451 = None
	        _assert_tensor_metadata_837 = torch.ops.aten._assert_tensor_metadata.default(view_1453, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_837 = None
	        convert_element_type_557 = torch.ops.prims.convert_element_type.default(view_1453, torch.float32);  view_1453 = None
	        sub_4247 = torch.ops.aten.sub.Tensor(convert_element_type_556, convert_element_type_557);  convert_element_type_556 = convert_element_type_557 = None
	        mul_8994 = torch.ops.aten.mul.Tensor(sub_4247, view_1452);  sub_4247 = view_1452 = None
	        view_1454 = torch.ops.aten.view.default(mul_8994, [1280, 1280]);  mul_8994 = None
	        _assert_tensor_metadata_838 = torch.ops.aten._assert_tensor_metadata.default(view_1454, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_838 = None
	        mul_8999 = sym_size_int * 1500
	        view_1455 = torch.ops.aten.view.default(mul_8989, [mul_8999, 1280]);  mul_8989 = mul_8999 = None
	        permute_155 = torch.ops.aten.permute.default(view_1454, [1, 0]);  view_1454 = None
	        addmm_76 = torch.ops.aten.addmm.default(model_audio_tower_layers_15_self_attn_v_proj_bias, view_1455, permute_155);  model_audio_tower_layers_15_self_attn_v_proj_bias = view_1455 = permute_155 = None
	        view_1456 = torch.ops.aten.view.default(addmm_76, [sym_size_int, 1500, 1280]);  addmm_76 = None
	        view_1457 = torch.ops.aten.view.default(view_1456, [sym_size_int, -1, 20, 64]);  view_1456 = None
	        permute_156 = torch.ops.aten.permute.default(view_1457, [0, 2, 1, 3]);  view_1457 = None
	        clone_124 = torch.ops.aten.clone.default(permute_156, memory_format = torch.contiguous_format);  permute_156 = None
	        _scaled_dot_product_efficient_attention_15 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_122, clone_123, clone_124, None, False, scale = 1.0);  clone_122 = clone_123 = clone_124 = None
	        getitem_122 = _scaled_dot_product_efficient_attention_15[0];  _scaled_dot_product_efficient_attention_15 = None
	        permute_157 = torch.ops.aten.permute.default(getitem_122, [0, 2, 1, 3]);  getitem_122 = None
	        view_1458 = torch.ops.aten.view.default(permute_157, [sym_size_int, 1500, -1]);  permute_157 = None
	        amin_93 = torch.ops.aten.amin.default(view_1458, [2])
	        amax_93 = torch.ops.aten.amax.default(view_1458, [2])
	        full_186 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_93 = torch.ops.aten.minimum.default(amin_93, full_186);  amin_93 = full_186 = None
	        full_187 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_93 = torch.ops.aten.maximum.default(amax_93, full_187);  amax_93 = full_187 = None
	        sub_4265 = torch.ops.aten.sub.Tensor(maximum_93, minimum_93);  maximum_93 = None
	        div_186 = torch.ops.aten.div.Tensor(sub_4265, 255.0);  sub_4265 = None
	        clamp_min_279 = torch.ops.aten.clamp_min.default(div_186, 1.1920928955078125e-07);  div_186 = None
	        div_187 = torch.ops.aten.div.Tensor(minimum_93, clamp_min_279);  minimum_93 = None
	        round_187 = torch.ops.aten.round.default(div_187);  div_187 = None
	        sub_4271 = torch.ops.aten.sub.Tensor(-128, round_187);  round_187 = None
	        clamp_min_280 = torch.ops.aten.clamp_min.default(sub_4271, -128);  sub_4271 = None
	        clamp_max_186 = torch.ops.aten.clamp_max.default(clamp_min_280, 127);  clamp_min_280 = None
	        _assert_tensor_metadata_839 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_279, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_839 = None
	        _assert_tensor_metadata_840 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_186, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_840 = None
	        convert_element_type_558 = torch.ops.prims.convert_element_type.default(clamp_max_186, torch.int8);  clamp_max_186 = None
	        view_1461 = torch.ops.aten.view.default(clamp_min_279, [sym_size_int, 1500, 1])
	        view_1462 = torch.ops.aten.view.default(convert_element_type_558, [sym_size_int, 1500, 1])
	        reciprocal_93 = torch.ops.aten.reciprocal.default(view_1461);  view_1461 = None
	        mul_9069 = torch.ops.aten.mul.Tensor(reciprocal_93, 1.0);  reciprocal_93 = None
	        mul_9072 = torch.ops.aten.mul.Tensor(view_1458, mul_9069);  view_1458 = mul_9069 = None
	        round_188 = torch.ops.aten.round.default(mul_9072);  mul_9072 = None
	        add_14353 = torch.ops.aten.add.Tensor(round_188, view_1462);  round_188 = view_1462 = None
	        clamp_min_281 = torch.ops.aten.clamp_min.default(add_14353, -128);  add_14353 = None
	        clamp_max_187 = torch.ops.aten.clamp_max.default(clamp_min_281, 127);  clamp_min_281 = None
	        _assert_tensor_metadata_841 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_187, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_841 = None
	        convert_element_type_559 = torch.ops.prims.convert_element_type.default(clamp_max_187, torch.int8);  clamp_max_187 = None
	        view_1465 = torch.ops.aten.view.default(clamp_min_279, [sym_size_int, 1500, 1]);  clamp_min_279 = None
	        view_1466 = torch.ops.aten.view.default(convert_element_type_558, [sym_size_int, 1500, 1]);  convert_element_type_558 = None
	        _assert_tensor_metadata_842 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_559, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_842 = None
	        convert_element_type_560 = torch.ops.prims.convert_element_type.default(convert_element_type_559, torch.float32);  convert_element_type_559 = None
	        _assert_tensor_metadata_843 = torch.ops.aten._assert_tensor_metadata.default(view_1466, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_843 = None
	        convert_element_type_561 = torch.ops.prims.convert_element_type.default(view_1466, torch.float32);  view_1466 = None
	        sub_4291 = torch.ops.aten.sub.Tensor(convert_element_type_560, convert_element_type_561);  convert_element_type_560 = convert_element_type_561 = None
	        mul_9094 = torch.ops.aten.mul.Tensor(sub_4291, view_1465);  sub_4291 = view_1465 = None
	        _assert_tensor_metadata_844 = torch.ops.aten._assert_tensor_metadata.default(mul_9094, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_844 = None
	        view_1468 = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1469 = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1470 = torch.ops.aten.view.default(model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_845 = torch.ops.aten._assert_tensor_metadata.default(view_1468, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_845 = None
	        convert_element_type_562 = torch.ops.prims.convert_element_type.default(view_1468, torch.float32);  view_1468 = None
	        _assert_tensor_metadata_846 = torch.ops.aten._assert_tensor_metadata.default(view_1470, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_846 = None
	        convert_element_type_563 = torch.ops.prims.convert_element_type.default(view_1470, torch.float32);  view_1470 = None
	        sub_4295 = torch.ops.aten.sub.Tensor(convert_element_type_562, convert_element_type_563);  convert_element_type_562 = convert_element_type_563 = None
	        mul_9099 = torch.ops.aten.mul.Tensor(sub_4295, view_1469);  sub_4295 = view_1469 = None
	        view_1471 = torch.ops.aten.view.default(mul_9099, [1280, 1280]);  mul_9099 = None
	        _assert_tensor_metadata_847 = torch.ops.aten._assert_tensor_metadata.default(view_1471, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_847 = None
	        mul_9104 = sym_size_int * 1500
	        view_1472 = torch.ops.aten.view.default(mul_9094, [mul_9104, 1280]);  mul_9094 = mul_9104 = None
	        permute_158 = torch.ops.aten.permute.default(view_1471, [1, 0]);  view_1471 = None
	        addmm_77 = torch.ops.aten.addmm.default(model_audio_tower_layers_15_self_attn_out_proj_bias, view_1472, permute_158);  model_audio_tower_layers_15_self_attn_out_proj_bias = view_1472 = permute_158 = None
	        view_1473 = torch.ops.aten.view.default(addmm_77, [sym_size_int, 1500, 1280]);  addmm_77 = None
	        add_14416 = torch.ops.aten.add.Tensor(add_13796, view_1473);  add_13796 = view_1473 = None
	        clone_126 = torch.ops.aten.clone.default(add_14416, memory_format = torch.contiguous_format)
	        var_mean_31 = torch.ops.aten.var_mean.correction(clone_126, [2], correction = 0, keepdim = True)
	        getitem_126 = var_mean_31[0]
	        getitem_127 = var_mean_31[1];  var_mean_31 = None
	        add_14421 = torch.ops.aten.add.Tensor(getitem_126, 1e-05);  getitem_126 = None
	        rsqrt_31 = torch.ops.aten.rsqrt.default(add_14421);  add_14421 = None
	        sub_4301 = torch.ops.aten.sub.Tensor(clone_126, getitem_127);  clone_126 = getitem_127 = None
	        mul_9115 = torch.ops.aten.mul.Tensor(sub_4301, rsqrt_31);  sub_4301 = rsqrt_31 = None
	        mul_9116 = torch.ops.aten.mul.Tensor(mul_9115, model_audio_tower_layers_15_final_layer_norm_weight);  mul_9115 = model_audio_tower_layers_15_final_layer_norm_weight = None
	        add_14422 = torch.ops.aten.add.Tensor(mul_9116, model_audio_tower_layers_15_final_layer_norm_bias);  mul_9116 = model_audio_tower_layers_15_final_layer_norm_bias = None
	        amin_94 = torch.ops.aten.amin.default(add_14422, [2])
	        amax_94 = torch.ops.aten.amax.default(add_14422, [2])
	        full_188 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_94 = torch.ops.aten.minimum.default(amin_94, full_188);  amin_94 = full_188 = None
	        full_189 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_94 = torch.ops.aten.maximum.default(amax_94, full_189);  amax_94 = full_189 = None
	        sub_4312 = torch.ops.aten.sub.Tensor(maximum_94, minimum_94);  maximum_94 = None
	        div_188 = torch.ops.aten.div.Tensor(sub_4312, 255.0);  sub_4312 = None
	        clamp_min_282 = torch.ops.aten.clamp_min.default(div_188, 1.1920928955078125e-07);  div_188 = None
	        div_189 = torch.ops.aten.div.Tensor(minimum_94, clamp_min_282);  minimum_94 = None
	        round_189 = torch.ops.aten.round.default(div_189);  div_189 = None
	        sub_4318 = torch.ops.aten.sub.Tensor(-128, round_189);  round_189 = None
	        clamp_min_283 = torch.ops.aten.clamp_min.default(sub_4318, -128);  sub_4318 = None
	        clamp_max_188 = torch.ops.aten.clamp_max.default(clamp_min_283, 127);  clamp_min_283 = None
	        _assert_tensor_metadata_848 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_282, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_848 = None
	        _assert_tensor_metadata_849 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_188, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_849 = None
	        convert_element_type_564 = torch.ops.prims.convert_element_type.default(clamp_max_188, torch.int8);  clamp_max_188 = None
	        view_1476 = torch.ops.aten.view.default(clamp_min_282, [sym_size_int, 1500, 1])
	        view_1477 = torch.ops.aten.view.default(convert_element_type_564, [sym_size_int, 1500, 1])
	        reciprocal_94 = torch.ops.aten.reciprocal.default(view_1476);  view_1476 = None
	        mul_9164 = torch.ops.aten.mul.Tensor(reciprocal_94, 1.0);  reciprocal_94 = None
	        mul_9167 = torch.ops.aten.mul.Tensor(add_14422, mul_9164);  add_14422 = mul_9164 = None
	        round_190 = torch.ops.aten.round.default(mul_9167);  mul_9167 = None
	        add_14509 = torch.ops.aten.add.Tensor(round_190, view_1477);  round_190 = view_1477 = None
	        clamp_min_284 = torch.ops.aten.clamp_min.default(add_14509, -128);  add_14509 = None
	        clamp_max_189 = torch.ops.aten.clamp_max.default(clamp_min_284, 127);  clamp_min_284 = None
	        _assert_tensor_metadata_850 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_189, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_850 = None
	        convert_element_type_565 = torch.ops.prims.convert_element_type.default(clamp_max_189, torch.int8);  clamp_max_189 = None
	        view_1480 = torch.ops.aten.view.default(clamp_min_282, [sym_size_int, 1500, 1]);  clamp_min_282 = None
	        view_1481 = torch.ops.aten.view.default(convert_element_type_564, [sym_size_int, 1500, 1]);  convert_element_type_564 = None
	        _assert_tensor_metadata_851 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_565, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_851 = None
	        convert_element_type_566 = torch.ops.prims.convert_element_type.default(convert_element_type_565, torch.float32);  convert_element_type_565 = None
	        _assert_tensor_metadata_852 = torch.ops.aten._assert_tensor_metadata.default(view_1481, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_852 = None
	        convert_element_type_567 = torch.ops.prims.convert_element_type.default(view_1481, torch.float32);  view_1481 = None
	        sub_4338 = torch.ops.aten.sub.Tensor(convert_element_type_566, convert_element_type_567);  convert_element_type_566 = convert_element_type_567 = None
	        mul_9189 = torch.ops.aten.mul.Tensor(sub_4338, view_1480);  sub_4338 = view_1480 = None
	        _assert_tensor_metadata_853 = torch.ops.aten._assert_tensor_metadata.default(mul_9189, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_853 = None
	        view_1483 = torch.ops.aten.view.default(model_audio_tower_layers_15_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_15_fc1_parametrizations_weight_original0 = None
	        view_1484 = torch.ops.aten.view.default(model_audio_tower_layers_15_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_15_fc1_parametrizations_weight_original1 = None
	        view_1485 = torch.ops.aten.view.default(model_audio_tower_layers_15_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_15_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_854 = torch.ops.aten._assert_tensor_metadata.default(view_1483, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_854 = None
	        convert_element_type_568 = torch.ops.prims.convert_element_type.default(view_1483, torch.float32);  view_1483 = None
	        _assert_tensor_metadata_855 = torch.ops.aten._assert_tensor_metadata.default(view_1485, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_855 = None
	        convert_element_type_569 = torch.ops.prims.convert_element_type.default(view_1485, torch.float32);  view_1485 = None
	        sub_4342 = torch.ops.aten.sub.Tensor(convert_element_type_568, convert_element_type_569);  convert_element_type_568 = convert_element_type_569 = None
	        mul_9194 = torch.ops.aten.mul.Tensor(sub_4342, view_1484);  sub_4342 = view_1484 = None
	        view_1486 = torch.ops.aten.view.default(mul_9194, [5120, 1280]);  mul_9194 = None
	        _assert_tensor_metadata_856 = torch.ops.aten._assert_tensor_metadata.default(view_1486, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_856 = None
	        mul_9199 = sym_size_int * 1500
	        view_1487 = torch.ops.aten.view.default(mul_9189, [mul_9199, 1280]);  mul_9189 = mul_9199 = None
	        permute_159 = torch.ops.aten.permute.default(view_1486, [1, 0]);  view_1486 = None
	        addmm_78 = torch.ops.aten.addmm.default(model_audio_tower_layers_15_fc1_bias, view_1487, permute_159);  model_audio_tower_layers_15_fc1_bias = view_1487 = permute_159 = None
	        view_1488 = torch.ops.aten.view.default(addmm_78, [sym_size_int, 1500, 5120]);  addmm_78 = None
	        mul_9206 = torch.ops.aten.mul.Tensor(view_1488, 0.5)
	        mul_9207 = torch.ops.aten.mul.Tensor(view_1488, 0.7071067811865476);  view_1488 = None
	        erf_17 = torch.ops.aten.erf.default(mul_9207);  mul_9207 = None
	        add_14568 = torch.ops.aten.add.Tensor(erf_17, 1);  erf_17 = None
	        mul_9208 = torch.ops.aten.mul.Tensor(mul_9206, add_14568);  mul_9206 = add_14568 = None
	        amin_95 = torch.ops.aten.amin.default(mul_9208, [2])
	        amax_95 = torch.ops.aten.amax.default(mul_9208, [2])
	        full_190 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_95 = torch.ops.aten.minimum.default(amin_95, full_190);  amin_95 = full_190 = None
	        full_191 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_95 = torch.ops.aten.maximum.default(amax_95, full_191);  amax_95 = full_191 = None
	        sub_4355 = torch.ops.aten.sub.Tensor(maximum_95, minimum_95);  maximum_95 = None
	        div_190 = torch.ops.aten.div.Tensor(sub_4355, 255.0);  sub_4355 = None
	        clamp_min_285 = torch.ops.aten.clamp_min.default(div_190, 1.1920928955078125e-07);  div_190 = None
	        div_191 = torch.ops.aten.div.Tensor(minimum_95, clamp_min_285);  minimum_95 = None
	        round_191 = torch.ops.aten.round.default(div_191);  div_191 = None
	        sub_4361 = torch.ops.aten.sub.Tensor(-128, round_191);  round_191 = None
	        clamp_min_286 = torch.ops.aten.clamp_min.default(sub_4361, -128);  sub_4361 = None
	        clamp_max_190 = torch.ops.aten.clamp_max.default(clamp_min_286, 127);  clamp_min_286 = None
	        _assert_tensor_metadata_857 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_285, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_857 = None
	        _assert_tensor_metadata_858 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_190, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_858 = None
	        convert_element_type_570 = torch.ops.prims.convert_element_type.default(clamp_max_190, torch.int8);  clamp_max_190 = None
	        view_1491 = torch.ops.aten.view.default(clamp_min_285, [sym_size_int, 1500, 1])
	        view_1492 = torch.ops.aten.view.default(convert_element_type_570, [sym_size_int, 1500, 1])
	        reciprocal_95 = torch.ops.aten.reciprocal.default(view_1491);  view_1491 = None
	        mul_9254 = torch.ops.aten.mul.Tensor(reciprocal_95, 1.0);  reciprocal_95 = None
	        mul_9257 = torch.ops.aten.mul.Tensor(mul_9208, mul_9254);  mul_9208 = mul_9254 = None
	        round_192 = torch.ops.aten.round.default(mul_9257);  mul_9257 = None
	        add_14651 = torch.ops.aten.add.Tensor(round_192, view_1492);  round_192 = view_1492 = None
	        clamp_min_287 = torch.ops.aten.clamp_min.default(add_14651, -128);  add_14651 = None
	        clamp_max_191 = torch.ops.aten.clamp_max.default(clamp_min_287, 127);  clamp_min_287 = None
	        _assert_tensor_metadata_859 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_191, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_859 = None
	        convert_element_type_571 = torch.ops.prims.convert_element_type.default(clamp_max_191, torch.int8);  clamp_max_191 = None
	        view_1495 = torch.ops.aten.view.default(clamp_min_285, [sym_size_int, 1500, 1]);  clamp_min_285 = None
	        view_1496 = torch.ops.aten.view.default(convert_element_type_570, [sym_size_int, 1500, 1]);  convert_element_type_570 = None
	        _assert_tensor_metadata_860 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_571, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_860 = None
	        convert_element_type_572 = torch.ops.prims.convert_element_type.default(convert_element_type_571, torch.float32);  convert_element_type_571 = None
	        _assert_tensor_metadata_861 = torch.ops.aten._assert_tensor_metadata.default(view_1496, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_861 = None
	        convert_element_type_573 = torch.ops.prims.convert_element_type.default(view_1496, torch.float32);  view_1496 = None
	        sub_4381 = torch.ops.aten.sub.Tensor(convert_element_type_572, convert_element_type_573);  convert_element_type_572 = convert_element_type_573 = None
	        mul_9279 = torch.ops.aten.mul.Tensor(sub_4381, view_1495);  sub_4381 = view_1495 = None
	        _assert_tensor_metadata_862 = torch.ops.aten._assert_tensor_metadata.default(mul_9279, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_862 = None
	        view_1498 = torch.ops.aten.view.default(model_audio_tower_layers_15_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_15_fc2_parametrizations_weight_original0 = None
	        view_1499 = torch.ops.aten.view.default(model_audio_tower_layers_15_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_15_fc2_parametrizations_weight_original1 = None
	        view_1500 = torch.ops.aten.view.default(model_audio_tower_layers_15_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_15_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_863 = torch.ops.aten._assert_tensor_metadata.default(view_1498, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_863 = None
	        convert_element_type_574 = torch.ops.prims.convert_element_type.default(view_1498, torch.float32);  view_1498 = None
	        _assert_tensor_metadata_864 = torch.ops.aten._assert_tensor_metadata.default(view_1500, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_864 = None
	        convert_element_type_575 = torch.ops.prims.convert_element_type.default(view_1500, torch.float32);  view_1500 = None
	        sub_4385 = torch.ops.aten.sub.Tensor(convert_element_type_574, convert_element_type_575);  convert_element_type_574 = convert_element_type_575 = None
	        mul_9284 = torch.ops.aten.mul.Tensor(sub_4385, view_1499);  sub_4385 = view_1499 = None
	        view_1501 = torch.ops.aten.view.default(mul_9284, [1280, 5120]);  mul_9284 = None
	        _assert_tensor_metadata_865 = torch.ops.aten._assert_tensor_metadata.default(view_1501, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_865 = None
	        mul_9289 = sym_size_int * 1500
	        view_1502 = torch.ops.aten.view.default(mul_9279, [mul_9289, 5120]);  mul_9279 = mul_9289 = None
	        permute_160 = torch.ops.aten.permute.default(view_1501, [1, 0]);  view_1501 = None
	        addmm_79 = torch.ops.aten.addmm.default(model_audio_tower_layers_15_fc2_bias, view_1502, permute_160);  model_audio_tower_layers_15_fc2_bias = view_1502 = permute_160 = None
	        view_1503 = torch.ops.aten.view.default(addmm_79, [sym_size_int, 1500, 1280]);  addmm_79 = None
	        add_14714 = torch.ops.aten.add.Tensor(add_14416, view_1503);  add_14416 = view_1503 = None
	        clone_129 = torch.ops.aten.clone.default(add_14714, memory_format = torch.contiguous_format)
	        var_mean_32 = torch.ops.aten.var_mean.correction(clone_129, [2], correction = 0, keepdim = True)
	        getitem_128 = var_mean_32[0]
	        getitem_129 = var_mean_32[1];  var_mean_32 = None
	        add_14719 = torch.ops.aten.add.Tensor(getitem_128, 1e-05);  getitem_128 = None
	        rsqrt_32 = torch.ops.aten.rsqrt.default(add_14719);  add_14719 = None
	        sub_4391 = torch.ops.aten.sub.Tensor(clone_129, getitem_129);  clone_129 = getitem_129 = None
	        mul_9300 = torch.ops.aten.mul.Tensor(sub_4391, rsqrt_32);  sub_4391 = rsqrt_32 = None
	        mul_9301 = torch.ops.aten.mul.Tensor(mul_9300, model_audio_tower_layers_16_self_attn_layer_norm_weight);  mul_9300 = model_audio_tower_layers_16_self_attn_layer_norm_weight = None
	        add_14720 = torch.ops.aten.add.Tensor(mul_9301, model_audio_tower_layers_16_self_attn_layer_norm_bias);  mul_9301 = model_audio_tower_layers_16_self_attn_layer_norm_bias = None
	        amin_96 = torch.ops.aten.amin.default(add_14720, [2])
	        amax_96 = torch.ops.aten.amax.default(add_14720, [2])
	        full_192 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_96 = torch.ops.aten.minimum.default(amin_96, full_192);  amin_96 = full_192 = None
	        full_193 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_96 = torch.ops.aten.maximum.default(amax_96, full_193);  amax_96 = full_193 = None
	        sub_4402 = torch.ops.aten.sub.Tensor(maximum_96, minimum_96);  maximum_96 = None
	        div_192 = torch.ops.aten.div.Tensor(sub_4402, 255.0);  sub_4402 = None
	        clamp_min_288 = torch.ops.aten.clamp_min.default(div_192, 1.1920928955078125e-07);  div_192 = None
	        div_193 = torch.ops.aten.div.Tensor(minimum_96, clamp_min_288);  minimum_96 = None
	        round_193 = torch.ops.aten.round.default(div_193);  div_193 = None
	        sub_4408 = torch.ops.aten.sub.Tensor(-128, round_193);  round_193 = None
	        clamp_min_289 = torch.ops.aten.clamp_min.default(sub_4408, -128);  sub_4408 = None
	        clamp_max_192 = torch.ops.aten.clamp_max.default(clamp_min_289, 127);  clamp_min_289 = None
	        _assert_tensor_metadata_866 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_288, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_866 = None
	        _assert_tensor_metadata_867 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_192, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_867 = None
	        convert_element_type_576 = torch.ops.prims.convert_element_type.default(clamp_max_192, torch.int8);  clamp_max_192 = None
	        view_1506 = torch.ops.aten.view.default(clamp_min_288, [sym_size_int, 1500, 1])
	        view_1507 = torch.ops.aten.view.default(convert_element_type_576, [sym_size_int, 1500, 1])
	        reciprocal_96 = torch.ops.aten.reciprocal.default(view_1506);  view_1506 = None
	        mul_9349 = torch.ops.aten.mul.Tensor(reciprocal_96, 1.0);  reciprocal_96 = None
	        mul_9352 = torch.ops.aten.mul.Tensor(add_14720, mul_9349);  mul_9349 = None
	        round_194 = torch.ops.aten.round.default(mul_9352);  mul_9352 = None
	        add_14807 = torch.ops.aten.add.Tensor(round_194, view_1507);  round_194 = view_1507 = None
	        clamp_min_290 = torch.ops.aten.clamp_min.default(add_14807, -128);  add_14807 = None
	        clamp_max_193 = torch.ops.aten.clamp_max.default(clamp_min_290, 127);  clamp_min_290 = None
	        _assert_tensor_metadata_868 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_193, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_868 = None
	        convert_element_type_577 = torch.ops.prims.convert_element_type.default(clamp_max_193, torch.int8);  clamp_max_193 = None
	        view_1510 = torch.ops.aten.view.default(clamp_min_288, [sym_size_int, 1500, 1]);  clamp_min_288 = None
	        view_1511 = torch.ops.aten.view.default(convert_element_type_576, [sym_size_int, 1500, 1]);  convert_element_type_576 = None
	        _assert_tensor_metadata_869 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_577, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_869 = None
	        convert_element_type_578 = torch.ops.prims.convert_element_type.default(convert_element_type_577, torch.float32);  convert_element_type_577 = None
	        _assert_tensor_metadata_870 = torch.ops.aten._assert_tensor_metadata.default(view_1511, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_870 = None
	        convert_element_type_579 = torch.ops.prims.convert_element_type.default(view_1511, torch.float32);  view_1511 = None
	        sub_4428 = torch.ops.aten.sub.Tensor(convert_element_type_578, convert_element_type_579);  convert_element_type_578 = convert_element_type_579 = None
	        mul_9374 = torch.ops.aten.mul.Tensor(sub_4428, view_1510);  sub_4428 = view_1510 = None
	        _assert_tensor_metadata_871 = torch.ops.aten._assert_tensor_metadata.default(mul_9374, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_871 = None
	        view_1513 = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1514 = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1515 = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_872 = torch.ops.aten._assert_tensor_metadata.default(view_1513, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_872 = None
	        convert_element_type_580 = torch.ops.prims.convert_element_type.default(view_1513, torch.float32);  view_1513 = None
	        _assert_tensor_metadata_873 = torch.ops.aten._assert_tensor_metadata.default(view_1515, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_873 = None
	        convert_element_type_581 = torch.ops.prims.convert_element_type.default(view_1515, torch.float32);  view_1515 = None
	        sub_4432 = torch.ops.aten.sub.Tensor(convert_element_type_580, convert_element_type_581);  convert_element_type_580 = convert_element_type_581 = None
	        mul_9379 = torch.ops.aten.mul.Tensor(sub_4432, view_1514);  sub_4432 = view_1514 = None
	        view_1516 = torch.ops.aten.view.default(mul_9379, [1280, 1280]);  mul_9379 = None
	        _assert_tensor_metadata_874 = torch.ops.aten._assert_tensor_metadata.default(view_1516, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_874 = None
	        mul_9384 = sym_size_int * 1500
	        view_1517 = torch.ops.aten.view.default(mul_9374, [mul_9384, 1280]);  mul_9374 = mul_9384 = None
	        permute_161 = torch.ops.aten.permute.default(view_1516, [1, 0]);  view_1516 = None
	        addmm_80 = torch.ops.aten.addmm.default(model_audio_tower_layers_16_self_attn_q_proj_bias, view_1517, permute_161);  model_audio_tower_layers_16_self_attn_q_proj_bias = view_1517 = permute_161 = None
	        view_1518 = torch.ops.aten.view.default(addmm_80, [sym_size_int, 1500, 1280]);  addmm_80 = None
	        mul_9391 = torch.ops.aten.mul.Tensor(view_1518, 0.125);  view_1518 = None
	        view_1519 = torch.ops.aten.view.default(mul_9391, [sym_size_int, 1500, 20, 64]);  mul_9391 = None
	        permute_162 = torch.ops.aten.permute.default(view_1519, [0, 2, 1, 3]);  view_1519 = None
	        clone_130 = torch.ops.aten.clone.default(permute_162, memory_format = torch.contiguous_format);  permute_162 = None
	        amin_97 = torch.ops.aten.amin.default(add_14720, [2])
	        amax_97 = torch.ops.aten.amax.default(add_14720, [2])
	        full_194 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_97 = torch.ops.aten.minimum.default(amin_97, full_194);  amin_97 = full_194 = None
	        full_195 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_97 = torch.ops.aten.maximum.default(amax_97, full_195);  amax_97 = full_195 = None
	        sub_4447 = torch.ops.aten.sub.Tensor(maximum_97, minimum_97);  maximum_97 = None
	        div_194 = torch.ops.aten.div.Tensor(sub_4447, 255.0);  sub_4447 = None
	        clamp_min_291 = torch.ops.aten.clamp_min.default(div_194, 1.1920928955078125e-07);  div_194 = None
	        div_195 = torch.ops.aten.div.Tensor(minimum_97, clamp_min_291);  minimum_97 = None
	        round_195 = torch.ops.aten.round.default(div_195);  div_195 = None
	        sub_4453 = torch.ops.aten.sub.Tensor(-128, round_195);  round_195 = None
	        clamp_min_292 = torch.ops.aten.clamp_min.default(sub_4453, -128);  sub_4453 = None
	        clamp_max_194 = torch.ops.aten.clamp_max.default(clamp_min_292, 127);  clamp_min_292 = None
	        _assert_tensor_metadata_875 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_291, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_875 = None
	        _assert_tensor_metadata_876 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_194, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_876 = None
	        convert_element_type_582 = torch.ops.prims.convert_element_type.default(clamp_max_194, torch.int8);  clamp_max_194 = None
	        view_1522 = torch.ops.aten.view.default(clamp_min_291, [sym_size_int, 1500, 1])
	        view_1523 = torch.ops.aten.view.default(convert_element_type_582, [sym_size_int, 1500, 1])
	        reciprocal_97 = torch.ops.aten.reciprocal.default(view_1522);  view_1522 = None
	        mul_9445 = torch.ops.aten.mul.Tensor(reciprocal_97, 1.0);  reciprocal_97 = None
	        mul_9448 = torch.ops.aten.mul.Tensor(add_14720, mul_9445);  mul_9445 = None
	        round_196 = torch.ops.aten.round.default(mul_9448);  mul_9448 = None
	        add_14959 = torch.ops.aten.add.Tensor(round_196, view_1523);  round_196 = view_1523 = None
	        clamp_min_293 = torch.ops.aten.clamp_min.default(add_14959, -128);  add_14959 = None
	        clamp_max_195 = torch.ops.aten.clamp_max.default(clamp_min_293, 127);  clamp_min_293 = None
	        _assert_tensor_metadata_877 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_195, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_877 = None
	        convert_element_type_583 = torch.ops.prims.convert_element_type.default(clamp_max_195, torch.int8);  clamp_max_195 = None
	        view_1526 = torch.ops.aten.view.default(clamp_min_291, [sym_size_int, 1500, 1]);  clamp_min_291 = None
	        view_1527 = torch.ops.aten.view.default(convert_element_type_582, [sym_size_int, 1500, 1]);  convert_element_type_582 = None
	        _assert_tensor_metadata_878 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_583, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_878 = None
	        convert_element_type_584 = torch.ops.prims.convert_element_type.default(convert_element_type_583, torch.float32);  convert_element_type_583 = None
	        _assert_tensor_metadata_879 = torch.ops.aten._assert_tensor_metadata.default(view_1527, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_879 = None
	        convert_element_type_585 = torch.ops.prims.convert_element_type.default(view_1527, torch.float32);  view_1527 = None
	        sub_4473 = torch.ops.aten.sub.Tensor(convert_element_type_584, convert_element_type_585);  convert_element_type_584 = convert_element_type_585 = None
	        mul_9470 = torch.ops.aten.mul.Tensor(sub_4473, view_1526);  sub_4473 = view_1526 = None
	        _assert_tensor_metadata_880 = torch.ops.aten._assert_tensor_metadata.default(mul_9470, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_880 = None
	        view_1529 = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1530 = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1531 = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_881 = torch.ops.aten._assert_tensor_metadata.default(view_1529, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_881 = None
	        convert_element_type_586 = torch.ops.prims.convert_element_type.default(view_1529, torch.float32);  view_1529 = None
	        _assert_tensor_metadata_882 = torch.ops.aten._assert_tensor_metadata.default(view_1531, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_882 = None
	        convert_element_type_587 = torch.ops.prims.convert_element_type.default(view_1531, torch.float32);  view_1531 = None
	        sub_4477 = torch.ops.aten.sub.Tensor(convert_element_type_586, convert_element_type_587);  convert_element_type_586 = convert_element_type_587 = None
	        mul_9475 = torch.ops.aten.mul.Tensor(sub_4477, view_1530);  sub_4477 = view_1530 = None
	        view_1532 = torch.ops.aten.view.default(mul_9475, [1280, 1280]);  mul_9475 = None
	        _assert_tensor_metadata_883 = torch.ops.aten._assert_tensor_metadata.default(view_1532, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_883 = None
	        permute_163 = torch.ops.aten.permute.default(view_1532, [1, 0]);  view_1532 = None
	        mul_9478 = sym_size_int * 1500
	        view_1533 = torch.ops.aten.view.default(mul_9470, [mul_9478, 1280]);  mul_9470 = mul_9478 = None
	        mm_16 = torch.ops.aten.mm.default(view_1533, permute_163);  view_1533 = permute_163 = None
	        view_1534 = torch.ops.aten.view.default(mm_16, [sym_size_int, 1500, 1280]);  mm_16 = None
	        view_1535 = torch.ops.aten.view.default(view_1534, [sym_size_int, -1, 20, 64]);  view_1534 = None
	        permute_164 = torch.ops.aten.permute.default(view_1535, [0, 2, 1, 3]);  view_1535 = None
	        clone_131 = torch.ops.aten.clone.default(permute_164, memory_format = torch.contiguous_format);  permute_164 = None
	        amin_98 = torch.ops.aten.amin.default(add_14720, [2])
	        amax_98 = torch.ops.aten.amax.default(add_14720, [2])
	        full_196 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_98 = torch.ops.aten.minimum.default(amin_98, full_196);  amin_98 = full_196 = None
	        full_197 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_98 = torch.ops.aten.maximum.default(amax_98, full_197);  amax_98 = full_197 = None
	        sub_4491 = torch.ops.aten.sub.Tensor(maximum_98, minimum_98);  maximum_98 = None
	        div_196 = torch.ops.aten.div.Tensor(sub_4491, 255.0);  sub_4491 = None
	        clamp_min_294 = torch.ops.aten.clamp_min.default(div_196, 1.1920928955078125e-07);  div_196 = None
	        div_197 = torch.ops.aten.div.Tensor(minimum_98, clamp_min_294);  minimum_98 = None
	        round_197 = torch.ops.aten.round.default(div_197);  div_197 = None
	        sub_4497 = torch.ops.aten.sub.Tensor(-128, round_197);  round_197 = None
	        clamp_min_295 = torch.ops.aten.clamp_min.default(sub_4497, -128);  sub_4497 = None
	        clamp_max_196 = torch.ops.aten.clamp_max.default(clamp_min_295, 127);  clamp_min_295 = None
	        _assert_tensor_metadata_884 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_884 = None
	        _assert_tensor_metadata_885 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_196, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_885 = None
	        convert_element_type_588 = torch.ops.prims.convert_element_type.default(clamp_max_196, torch.int8);  clamp_max_196 = None
	        view_1538 = torch.ops.aten.view.default(clamp_min_294, [sym_size_int, 1500, 1])
	        view_1539 = torch.ops.aten.view.default(convert_element_type_588, [sym_size_int, 1500, 1])
	        reciprocal_98 = torch.ops.aten.reciprocal.default(view_1538);  view_1538 = None
	        mul_9544 = torch.ops.aten.mul.Tensor(reciprocal_98, 1.0);  reciprocal_98 = None
	        mul_9547 = torch.ops.aten.mul.Tensor(add_14720, mul_9544);  add_14720 = mul_9544 = None
	        round_198 = torch.ops.aten.round.default(mul_9547);  mul_9547 = None
	        add_15107 = torch.ops.aten.add.Tensor(round_198, view_1539);  round_198 = view_1539 = None
	        clamp_min_296 = torch.ops.aten.clamp_min.default(add_15107, -128);  add_15107 = None
	        clamp_max_197 = torch.ops.aten.clamp_max.default(clamp_min_296, 127);  clamp_min_296 = None
	        _assert_tensor_metadata_886 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_197, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_886 = None
	        convert_element_type_589 = torch.ops.prims.convert_element_type.default(clamp_max_197, torch.int8);  clamp_max_197 = None
	        view_1542 = torch.ops.aten.view.default(clamp_min_294, [sym_size_int, 1500, 1]);  clamp_min_294 = None
	        view_1543 = torch.ops.aten.view.default(convert_element_type_588, [sym_size_int, 1500, 1]);  convert_element_type_588 = None
	        _assert_tensor_metadata_887 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_589, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_887 = None
	        convert_element_type_590 = torch.ops.prims.convert_element_type.default(convert_element_type_589, torch.float32);  convert_element_type_589 = None
	        _assert_tensor_metadata_888 = torch.ops.aten._assert_tensor_metadata.default(view_1543, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_888 = None
	        convert_element_type_591 = torch.ops.prims.convert_element_type.default(view_1543, torch.float32);  view_1543 = None
	        sub_4517 = torch.ops.aten.sub.Tensor(convert_element_type_590, convert_element_type_591);  convert_element_type_590 = convert_element_type_591 = None
	        mul_9569 = torch.ops.aten.mul.Tensor(sub_4517, view_1542);  sub_4517 = view_1542 = None
	        _assert_tensor_metadata_889 = torch.ops.aten._assert_tensor_metadata.default(mul_9569, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_889 = None
	        view_1545 = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1546 = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1547 = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_890 = torch.ops.aten._assert_tensor_metadata.default(view_1545, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_890 = None
	        convert_element_type_592 = torch.ops.prims.convert_element_type.default(view_1545, torch.float32);  view_1545 = None
	        _assert_tensor_metadata_891 = torch.ops.aten._assert_tensor_metadata.default(view_1547, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_891 = None
	        convert_element_type_593 = torch.ops.prims.convert_element_type.default(view_1547, torch.float32);  view_1547 = None
	        sub_4521 = torch.ops.aten.sub.Tensor(convert_element_type_592, convert_element_type_593);  convert_element_type_592 = convert_element_type_593 = None
	        mul_9574 = torch.ops.aten.mul.Tensor(sub_4521, view_1546);  sub_4521 = view_1546 = None
	        view_1548 = torch.ops.aten.view.default(mul_9574, [1280, 1280]);  mul_9574 = None
	        _assert_tensor_metadata_892 = torch.ops.aten._assert_tensor_metadata.default(view_1548, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_892 = None
	        mul_9579 = sym_size_int * 1500
	        view_1549 = torch.ops.aten.view.default(mul_9569, [mul_9579, 1280]);  mul_9569 = mul_9579 = None
	        permute_165 = torch.ops.aten.permute.default(view_1548, [1, 0]);  view_1548 = None
	        addmm_81 = torch.ops.aten.addmm.default(model_audio_tower_layers_16_self_attn_v_proj_bias, view_1549, permute_165);  model_audio_tower_layers_16_self_attn_v_proj_bias = view_1549 = permute_165 = None
	        view_1550 = torch.ops.aten.view.default(addmm_81, [sym_size_int, 1500, 1280]);  addmm_81 = None
	        view_1551 = torch.ops.aten.view.default(view_1550, [sym_size_int, -1, 20, 64]);  view_1550 = None
	        permute_166 = torch.ops.aten.permute.default(view_1551, [0, 2, 1, 3]);  view_1551 = None
	        clone_132 = torch.ops.aten.clone.default(permute_166, memory_format = torch.contiguous_format);  permute_166 = None
	        _scaled_dot_product_efficient_attention_16 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_130, clone_131, clone_132, None, False, scale = 1.0);  clone_130 = clone_131 = clone_132 = None
	        getitem_130 = _scaled_dot_product_efficient_attention_16[0];  _scaled_dot_product_efficient_attention_16 = None
	        permute_167 = torch.ops.aten.permute.default(getitem_130, [0, 2, 1, 3]);  getitem_130 = None
	        view_1552 = torch.ops.aten.view.default(permute_167, [sym_size_int, 1500, -1]);  permute_167 = None
	        amin_99 = torch.ops.aten.amin.default(view_1552, [2])
	        amax_99 = torch.ops.aten.amax.default(view_1552, [2])
	        full_198 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_99 = torch.ops.aten.minimum.default(amin_99, full_198);  amin_99 = full_198 = None
	        full_199 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_99 = torch.ops.aten.maximum.default(amax_99, full_199);  amax_99 = full_199 = None
	        sub_4539 = torch.ops.aten.sub.Tensor(maximum_99, minimum_99);  maximum_99 = None
	        div_198 = torch.ops.aten.div.Tensor(sub_4539, 255.0);  sub_4539 = None
	        clamp_min_297 = torch.ops.aten.clamp_min.default(div_198, 1.1920928955078125e-07);  div_198 = None
	        div_199 = torch.ops.aten.div.Tensor(minimum_99, clamp_min_297);  minimum_99 = None
	        round_199 = torch.ops.aten.round.default(div_199);  div_199 = None
	        sub_4545 = torch.ops.aten.sub.Tensor(-128, round_199);  round_199 = None
	        clamp_min_298 = torch.ops.aten.clamp_min.default(sub_4545, -128);  sub_4545 = None
	        clamp_max_198 = torch.ops.aten.clamp_max.default(clamp_min_298, 127);  clamp_min_298 = None
	        _assert_tensor_metadata_893 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_297, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_893 = None
	        _assert_tensor_metadata_894 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_198, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_894 = None
	        convert_element_type_594 = torch.ops.prims.convert_element_type.default(clamp_max_198, torch.int8);  clamp_max_198 = None
	        view_1555 = torch.ops.aten.view.default(clamp_min_297, [sym_size_int, 1500, 1])
	        view_1556 = torch.ops.aten.view.default(convert_element_type_594, [sym_size_int, 1500, 1])
	        reciprocal_99 = torch.ops.aten.reciprocal.default(view_1555);  view_1555 = None
	        mul_9649 = torch.ops.aten.mul.Tensor(reciprocal_99, 1.0);  reciprocal_99 = None
	        mul_9652 = torch.ops.aten.mul.Tensor(view_1552, mul_9649);  view_1552 = mul_9649 = None
	        round_200 = torch.ops.aten.round.default(mul_9652);  mul_9652 = None
	        add_15271 = torch.ops.aten.add.Tensor(round_200, view_1556);  round_200 = view_1556 = None
	        clamp_min_299 = torch.ops.aten.clamp_min.default(add_15271, -128);  add_15271 = None
	        clamp_max_199 = torch.ops.aten.clamp_max.default(clamp_min_299, 127);  clamp_min_299 = None
	        _assert_tensor_metadata_895 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_199, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_895 = None
	        convert_element_type_595 = torch.ops.prims.convert_element_type.default(clamp_max_199, torch.int8);  clamp_max_199 = None
	        view_1559 = torch.ops.aten.view.default(clamp_min_297, [sym_size_int, 1500, 1]);  clamp_min_297 = None
	        view_1560 = torch.ops.aten.view.default(convert_element_type_594, [sym_size_int, 1500, 1]);  convert_element_type_594 = None
	        _assert_tensor_metadata_896 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_595, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_896 = None
	        convert_element_type_596 = torch.ops.prims.convert_element_type.default(convert_element_type_595, torch.float32);  convert_element_type_595 = None
	        _assert_tensor_metadata_897 = torch.ops.aten._assert_tensor_metadata.default(view_1560, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_897 = None
	        convert_element_type_597 = torch.ops.prims.convert_element_type.default(view_1560, torch.float32);  view_1560 = None
	        sub_4565 = torch.ops.aten.sub.Tensor(convert_element_type_596, convert_element_type_597);  convert_element_type_596 = convert_element_type_597 = None
	        mul_9674 = torch.ops.aten.mul.Tensor(sub_4565, view_1559);  sub_4565 = view_1559 = None
	        _assert_tensor_metadata_898 = torch.ops.aten._assert_tensor_metadata.default(mul_9674, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_898 = None
	        view_1562 = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1563 = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1564 = torch.ops.aten.view.default(model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_899 = torch.ops.aten._assert_tensor_metadata.default(view_1562, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_899 = None
	        convert_element_type_598 = torch.ops.prims.convert_element_type.default(view_1562, torch.float32);  view_1562 = None
	        _assert_tensor_metadata_900 = torch.ops.aten._assert_tensor_metadata.default(view_1564, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_900 = None
	        convert_element_type_599 = torch.ops.prims.convert_element_type.default(view_1564, torch.float32);  view_1564 = None
	        sub_4569 = torch.ops.aten.sub.Tensor(convert_element_type_598, convert_element_type_599);  convert_element_type_598 = convert_element_type_599 = None
	        mul_9679 = torch.ops.aten.mul.Tensor(sub_4569, view_1563);  sub_4569 = view_1563 = None
	        view_1565 = torch.ops.aten.view.default(mul_9679, [1280, 1280]);  mul_9679 = None
	        _assert_tensor_metadata_901 = torch.ops.aten._assert_tensor_metadata.default(view_1565, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_901 = None
	        mul_9684 = sym_size_int * 1500
	        view_1566 = torch.ops.aten.view.default(mul_9674, [mul_9684, 1280]);  mul_9674 = mul_9684 = None
	        permute_168 = torch.ops.aten.permute.default(view_1565, [1, 0]);  view_1565 = None
	        addmm_82 = torch.ops.aten.addmm.default(model_audio_tower_layers_16_self_attn_out_proj_bias, view_1566, permute_168);  model_audio_tower_layers_16_self_attn_out_proj_bias = view_1566 = permute_168 = None
	        view_1567 = torch.ops.aten.view.default(addmm_82, [sym_size_int, 1500, 1280]);  addmm_82 = None
	        add_15334 = torch.ops.aten.add.Tensor(add_14714, view_1567);  add_14714 = view_1567 = None
	        clone_134 = torch.ops.aten.clone.default(add_15334, memory_format = torch.contiguous_format)
	        var_mean_33 = torch.ops.aten.var_mean.correction(clone_134, [2], correction = 0, keepdim = True)
	        getitem_134 = var_mean_33[0]
	        getitem_135 = var_mean_33[1];  var_mean_33 = None
	        add_15339 = torch.ops.aten.add.Tensor(getitem_134, 1e-05);  getitem_134 = None
	        rsqrt_33 = torch.ops.aten.rsqrt.default(add_15339);  add_15339 = None
	        sub_4575 = torch.ops.aten.sub.Tensor(clone_134, getitem_135);  clone_134 = getitem_135 = None
	        mul_9695 = torch.ops.aten.mul.Tensor(sub_4575, rsqrt_33);  sub_4575 = rsqrt_33 = None
	        mul_9696 = torch.ops.aten.mul.Tensor(mul_9695, model_audio_tower_layers_16_final_layer_norm_weight);  mul_9695 = model_audio_tower_layers_16_final_layer_norm_weight = None
	        add_15340 = torch.ops.aten.add.Tensor(mul_9696, model_audio_tower_layers_16_final_layer_norm_bias);  mul_9696 = model_audio_tower_layers_16_final_layer_norm_bias = None
	        amin_100 = torch.ops.aten.amin.default(add_15340, [2])
	        amax_100 = torch.ops.aten.amax.default(add_15340, [2])
	        full_200 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_100 = torch.ops.aten.minimum.default(amin_100, full_200);  amin_100 = full_200 = None
	        full_201 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_100 = torch.ops.aten.maximum.default(amax_100, full_201);  amax_100 = full_201 = None
	        sub_4586 = torch.ops.aten.sub.Tensor(maximum_100, minimum_100);  maximum_100 = None
	        div_200 = torch.ops.aten.div.Tensor(sub_4586, 255.0);  sub_4586 = None
	        clamp_min_300 = torch.ops.aten.clamp_min.default(div_200, 1.1920928955078125e-07);  div_200 = None
	        div_201 = torch.ops.aten.div.Tensor(minimum_100, clamp_min_300);  minimum_100 = None
	        round_201 = torch.ops.aten.round.default(div_201);  div_201 = None
	        sub_4592 = torch.ops.aten.sub.Tensor(-128, round_201);  round_201 = None
	        clamp_min_301 = torch.ops.aten.clamp_min.default(sub_4592, -128);  sub_4592 = None
	        clamp_max_200 = torch.ops.aten.clamp_max.default(clamp_min_301, 127);  clamp_min_301 = None
	        _assert_tensor_metadata_902 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_300, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_902 = None
	        _assert_tensor_metadata_903 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_200, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_903 = None
	        convert_element_type_600 = torch.ops.prims.convert_element_type.default(clamp_max_200, torch.int8);  clamp_max_200 = None
	        view_1570 = torch.ops.aten.view.default(clamp_min_300, [sym_size_int, 1500, 1])
	        view_1571 = torch.ops.aten.view.default(convert_element_type_600, [sym_size_int, 1500, 1])
	        reciprocal_100 = torch.ops.aten.reciprocal.default(view_1570);  view_1570 = None
	        mul_9744 = torch.ops.aten.mul.Tensor(reciprocal_100, 1.0);  reciprocal_100 = None
	        mul_9747 = torch.ops.aten.mul.Tensor(add_15340, mul_9744);  add_15340 = mul_9744 = None
	        round_202 = torch.ops.aten.round.default(mul_9747);  mul_9747 = None
	        add_15427 = torch.ops.aten.add.Tensor(round_202, view_1571);  round_202 = view_1571 = None
	        clamp_min_302 = torch.ops.aten.clamp_min.default(add_15427, -128);  add_15427 = None
	        clamp_max_201 = torch.ops.aten.clamp_max.default(clamp_min_302, 127);  clamp_min_302 = None
	        _assert_tensor_metadata_904 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_201, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_904 = None
	        convert_element_type_601 = torch.ops.prims.convert_element_type.default(clamp_max_201, torch.int8);  clamp_max_201 = None
	        view_1574 = torch.ops.aten.view.default(clamp_min_300, [sym_size_int, 1500, 1]);  clamp_min_300 = None
	        view_1575 = torch.ops.aten.view.default(convert_element_type_600, [sym_size_int, 1500, 1]);  convert_element_type_600 = None
	        _assert_tensor_metadata_905 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_601, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_905 = None
	        convert_element_type_602 = torch.ops.prims.convert_element_type.default(convert_element_type_601, torch.float32);  convert_element_type_601 = None
	        _assert_tensor_metadata_906 = torch.ops.aten._assert_tensor_metadata.default(view_1575, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_906 = None
	        convert_element_type_603 = torch.ops.prims.convert_element_type.default(view_1575, torch.float32);  view_1575 = None
	        sub_4612 = torch.ops.aten.sub.Tensor(convert_element_type_602, convert_element_type_603);  convert_element_type_602 = convert_element_type_603 = None
	        mul_9769 = torch.ops.aten.mul.Tensor(sub_4612, view_1574);  sub_4612 = view_1574 = None
	        _assert_tensor_metadata_907 = torch.ops.aten._assert_tensor_metadata.default(mul_9769, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_907 = None
	        view_1577 = torch.ops.aten.view.default(model_audio_tower_layers_16_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_16_fc1_parametrizations_weight_original0 = None
	        view_1578 = torch.ops.aten.view.default(model_audio_tower_layers_16_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_16_fc1_parametrizations_weight_original1 = None
	        view_1579 = torch.ops.aten.view.default(model_audio_tower_layers_16_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_16_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_908 = torch.ops.aten._assert_tensor_metadata.default(view_1577, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_908 = None
	        convert_element_type_604 = torch.ops.prims.convert_element_type.default(view_1577, torch.float32);  view_1577 = None
	        _assert_tensor_metadata_909 = torch.ops.aten._assert_tensor_metadata.default(view_1579, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_909 = None
	        convert_element_type_605 = torch.ops.prims.convert_element_type.default(view_1579, torch.float32);  view_1579 = None
	        sub_4616 = torch.ops.aten.sub.Tensor(convert_element_type_604, convert_element_type_605);  convert_element_type_604 = convert_element_type_605 = None
	        mul_9774 = torch.ops.aten.mul.Tensor(sub_4616, view_1578);  sub_4616 = view_1578 = None
	        view_1580 = torch.ops.aten.view.default(mul_9774, [5120, 1280]);  mul_9774 = None
	        _assert_tensor_metadata_910 = torch.ops.aten._assert_tensor_metadata.default(view_1580, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_910 = None
	        mul_9779 = sym_size_int * 1500
	        view_1581 = torch.ops.aten.view.default(mul_9769, [mul_9779, 1280]);  mul_9769 = mul_9779 = None
	        permute_169 = torch.ops.aten.permute.default(view_1580, [1, 0]);  view_1580 = None
	        addmm_83 = torch.ops.aten.addmm.default(model_audio_tower_layers_16_fc1_bias, view_1581, permute_169);  model_audio_tower_layers_16_fc1_bias = view_1581 = permute_169 = None
	        view_1582 = torch.ops.aten.view.default(addmm_83, [sym_size_int, 1500, 5120]);  addmm_83 = None
	        mul_9786 = torch.ops.aten.mul.Tensor(view_1582, 0.5)
	        mul_9787 = torch.ops.aten.mul.Tensor(view_1582, 0.7071067811865476);  view_1582 = None
	        erf_18 = torch.ops.aten.erf.default(mul_9787);  mul_9787 = None
	        add_15486 = torch.ops.aten.add.Tensor(erf_18, 1);  erf_18 = None
	        mul_9788 = torch.ops.aten.mul.Tensor(mul_9786, add_15486);  mul_9786 = add_15486 = None
	        amin_101 = torch.ops.aten.amin.default(mul_9788, [2])
	        amax_101 = torch.ops.aten.amax.default(mul_9788, [2])
	        full_202 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_101 = torch.ops.aten.minimum.default(amin_101, full_202);  amin_101 = full_202 = None
	        full_203 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_101 = torch.ops.aten.maximum.default(amax_101, full_203);  amax_101 = full_203 = None
	        sub_4629 = torch.ops.aten.sub.Tensor(maximum_101, minimum_101);  maximum_101 = None
	        div_202 = torch.ops.aten.div.Tensor(sub_4629, 255.0);  sub_4629 = None
	        clamp_min_303 = torch.ops.aten.clamp_min.default(div_202, 1.1920928955078125e-07);  div_202 = None
	        div_203 = torch.ops.aten.div.Tensor(minimum_101, clamp_min_303);  minimum_101 = None
	        round_203 = torch.ops.aten.round.default(div_203);  div_203 = None
	        sub_4635 = torch.ops.aten.sub.Tensor(-128, round_203);  round_203 = None
	        clamp_min_304 = torch.ops.aten.clamp_min.default(sub_4635, -128);  sub_4635 = None
	        clamp_max_202 = torch.ops.aten.clamp_max.default(clamp_min_304, 127);  clamp_min_304 = None
	        _assert_tensor_metadata_911 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_303, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_911 = None
	        _assert_tensor_metadata_912 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_202, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_912 = None
	        convert_element_type_606 = torch.ops.prims.convert_element_type.default(clamp_max_202, torch.int8);  clamp_max_202 = None
	        view_1585 = torch.ops.aten.view.default(clamp_min_303, [sym_size_int, 1500, 1])
	        view_1586 = torch.ops.aten.view.default(convert_element_type_606, [sym_size_int, 1500, 1])
	        reciprocal_101 = torch.ops.aten.reciprocal.default(view_1585);  view_1585 = None
	        mul_9834 = torch.ops.aten.mul.Tensor(reciprocal_101, 1.0);  reciprocal_101 = None
	        mul_9837 = torch.ops.aten.mul.Tensor(mul_9788, mul_9834);  mul_9788 = mul_9834 = None
	        round_204 = torch.ops.aten.round.default(mul_9837);  mul_9837 = None
	        add_15569 = torch.ops.aten.add.Tensor(round_204, view_1586);  round_204 = view_1586 = None
	        clamp_min_305 = torch.ops.aten.clamp_min.default(add_15569, -128);  add_15569 = None
	        clamp_max_203 = torch.ops.aten.clamp_max.default(clamp_min_305, 127);  clamp_min_305 = None
	        _assert_tensor_metadata_913 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_203, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_913 = None
	        convert_element_type_607 = torch.ops.prims.convert_element_type.default(clamp_max_203, torch.int8);  clamp_max_203 = None
	        view_1589 = torch.ops.aten.view.default(clamp_min_303, [sym_size_int, 1500, 1]);  clamp_min_303 = None
	        view_1590 = torch.ops.aten.view.default(convert_element_type_606, [sym_size_int, 1500, 1]);  convert_element_type_606 = None
	        _assert_tensor_metadata_914 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_607, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_914 = None
	        convert_element_type_608 = torch.ops.prims.convert_element_type.default(convert_element_type_607, torch.float32);  convert_element_type_607 = None
	        _assert_tensor_metadata_915 = torch.ops.aten._assert_tensor_metadata.default(view_1590, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_915 = None
	        convert_element_type_609 = torch.ops.prims.convert_element_type.default(view_1590, torch.float32);  view_1590 = None
	        sub_4655 = torch.ops.aten.sub.Tensor(convert_element_type_608, convert_element_type_609);  convert_element_type_608 = convert_element_type_609 = None
	        mul_9859 = torch.ops.aten.mul.Tensor(sub_4655, view_1589);  sub_4655 = view_1589 = None
	        _assert_tensor_metadata_916 = torch.ops.aten._assert_tensor_metadata.default(mul_9859, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_916 = None
	        view_1592 = torch.ops.aten.view.default(model_audio_tower_layers_16_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_16_fc2_parametrizations_weight_original0 = None
	        view_1593 = torch.ops.aten.view.default(model_audio_tower_layers_16_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_16_fc2_parametrizations_weight_original1 = None
	        view_1594 = torch.ops.aten.view.default(model_audio_tower_layers_16_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_16_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_917 = torch.ops.aten._assert_tensor_metadata.default(view_1592, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_917 = None
	        convert_element_type_610 = torch.ops.prims.convert_element_type.default(view_1592, torch.float32);  view_1592 = None
	        _assert_tensor_metadata_918 = torch.ops.aten._assert_tensor_metadata.default(view_1594, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_918 = None
	        convert_element_type_611 = torch.ops.prims.convert_element_type.default(view_1594, torch.float32);  view_1594 = None
	        sub_4659 = torch.ops.aten.sub.Tensor(convert_element_type_610, convert_element_type_611);  convert_element_type_610 = convert_element_type_611 = None
	        mul_9864 = torch.ops.aten.mul.Tensor(sub_4659, view_1593);  sub_4659 = view_1593 = None
	        view_1595 = torch.ops.aten.view.default(mul_9864, [1280, 5120]);  mul_9864 = None
	        _assert_tensor_metadata_919 = torch.ops.aten._assert_tensor_metadata.default(view_1595, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_919 = None
	        mul_9869 = sym_size_int * 1500
	        view_1596 = torch.ops.aten.view.default(mul_9859, [mul_9869, 5120]);  mul_9859 = mul_9869 = None
	        permute_170 = torch.ops.aten.permute.default(view_1595, [1, 0]);  view_1595 = None
	        addmm_84 = torch.ops.aten.addmm.default(model_audio_tower_layers_16_fc2_bias, view_1596, permute_170);  model_audio_tower_layers_16_fc2_bias = view_1596 = permute_170 = None
	        view_1597 = torch.ops.aten.view.default(addmm_84, [sym_size_int, 1500, 1280]);  addmm_84 = None
	        add_15632 = torch.ops.aten.add.Tensor(add_15334, view_1597);  add_15334 = view_1597 = None
	        clone_137 = torch.ops.aten.clone.default(add_15632, memory_format = torch.contiguous_format)
	        var_mean_34 = torch.ops.aten.var_mean.correction(clone_137, [2], correction = 0, keepdim = True)
	        getitem_136 = var_mean_34[0]
	        getitem_137 = var_mean_34[1];  var_mean_34 = None
	        add_15637 = torch.ops.aten.add.Tensor(getitem_136, 1e-05);  getitem_136 = None
	        rsqrt_34 = torch.ops.aten.rsqrt.default(add_15637);  add_15637 = None
	        sub_4665 = torch.ops.aten.sub.Tensor(clone_137, getitem_137);  clone_137 = getitem_137 = None
	        mul_9880 = torch.ops.aten.mul.Tensor(sub_4665, rsqrt_34);  sub_4665 = rsqrt_34 = None
	        mul_9881 = torch.ops.aten.mul.Tensor(mul_9880, model_audio_tower_layers_17_self_attn_layer_norm_weight);  mul_9880 = model_audio_tower_layers_17_self_attn_layer_norm_weight = None
	        add_15638 = torch.ops.aten.add.Tensor(mul_9881, model_audio_tower_layers_17_self_attn_layer_norm_bias);  mul_9881 = model_audio_tower_layers_17_self_attn_layer_norm_bias = None
	        amin_102 = torch.ops.aten.amin.default(add_15638, [2])
	        amax_102 = torch.ops.aten.amax.default(add_15638, [2])
	        full_204 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_102 = torch.ops.aten.minimum.default(amin_102, full_204);  amin_102 = full_204 = None
	        full_205 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_102 = torch.ops.aten.maximum.default(amax_102, full_205);  amax_102 = full_205 = None
	        sub_4676 = torch.ops.aten.sub.Tensor(maximum_102, minimum_102);  maximum_102 = None
	        div_204 = torch.ops.aten.div.Tensor(sub_4676, 255.0);  sub_4676 = None
	        clamp_min_306 = torch.ops.aten.clamp_min.default(div_204, 1.1920928955078125e-07);  div_204 = None
	        div_205 = torch.ops.aten.div.Tensor(minimum_102, clamp_min_306);  minimum_102 = None
	        round_205 = torch.ops.aten.round.default(div_205);  div_205 = None
	        sub_4682 = torch.ops.aten.sub.Tensor(-128, round_205);  round_205 = None
	        clamp_min_307 = torch.ops.aten.clamp_min.default(sub_4682, -128);  sub_4682 = None
	        clamp_max_204 = torch.ops.aten.clamp_max.default(clamp_min_307, 127);  clamp_min_307 = None
	        _assert_tensor_metadata_920 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_306, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_920 = None
	        _assert_tensor_metadata_921 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_204, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_921 = None
	        convert_element_type_612 = torch.ops.prims.convert_element_type.default(clamp_max_204, torch.int8);  clamp_max_204 = None
	        view_1600 = torch.ops.aten.view.default(clamp_min_306, [sym_size_int, 1500, 1])
	        view_1601 = torch.ops.aten.view.default(convert_element_type_612, [sym_size_int, 1500, 1])
	        reciprocal_102 = torch.ops.aten.reciprocal.default(view_1600);  view_1600 = None
	        mul_9929 = torch.ops.aten.mul.Tensor(reciprocal_102, 1.0);  reciprocal_102 = None
	        mul_9932 = torch.ops.aten.mul.Tensor(add_15638, mul_9929);  mul_9929 = None
	        round_206 = torch.ops.aten.round.default(mul_9932);  mul_9932 = None
	        add_15725 = torch.ops.aten.add.Tensor(round_206, view_1601);  round_206 = view_1601 = None
	        clamp_min_308 = torch.ops.aten.clamp_min.default(add_15725, -128);  add_15725 = None
	        clamp_max_205 = torch.ops.aten.clamp_max.default(clamp_min_308, 127);  clamp_min_308 = None
	        _assert_tensor_metadata_922 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_205, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_922 = None
	        convert_element_type_613 = torch.ops.prims.convert_element_type.default(clamp_max_205, torch.int8);  clamp_max_205 = None
	        view_1604 = torch.ops.aten.view.default(clamp_min_306, [sym_size_int, 1500, 1]);  clamp_min_306 = None
	        view_1605 = torch.ops.aten.view.default(convert_element_type_612, [sym_size_int, 1500, 1]);  convert_element_type_612 = None
	        _assert_tensor_metadata_923 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_613, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_923 = None
	        convert_element_type_614 = torch.ops.prims.convert_element_type.default(convert_element_type_613, torch.float32);  convert_element_type_613 = None
	        _assert_tensor_metadata_924 = torch.ops.aten._assert_tensor_metadata.default(view_1605, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_924 = None
	        convert_element_type_615 = torch.ops.prims.convert_element_type.default(view_1605, torch.float32);  view_1605 = None
	        sub_4702 = torch.ops.aten.sub.Tensor(convert_element_type_614, convert_element_type_615);  convert_element_type_614 = convert_element_type_615 = None
	        mul_9954 = torch.ops.aten.mul.Tensor(sub_4702, view_1604);  sub_4702 = view_1604 = None
	        _assert_tensor_metadata_925 = torch.ops.aten._assert_tensor_metadata.default(mul_9954, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_925 = None
	        view_1607 = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1608 = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1609 = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_926 = torch.ops.aten._assert_tensor_metadata.default(view_1607, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_926 = None
	        convert_element_type_616 = torch.ops.prims.convert_element_type.default(view_1607, torch.float32);  view_1607 = None
	        _assert_tensor_metadata_927 = torch.ops.aten._assert_tensor_metadata.default(view_1609, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_927 = None
	        convert_element_type_617 = torch.ops.prims.convert_element_type.default(view_1609, torch.float32);  view_1609 = None
	        sub_4706 = torch.ops.aten.sub.Tensor(convert_element_type_616, convert_element_type_617);  convert_element_type_616 = convert_element_type_617 = None
	        mul_9959 = torch.ops.aten.mul.Tensor(sub_4706, view_1608);  sub_4706 = view_1608 = None
	        view_1610 = torch.ops.aten.view.default(mul_9959, [1280, 1280]);  mul_9959 = None
	        _assert_tensor_metadata_928 = torch.ops.aten._assert_tensor_metadata.default(view_1610, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_928 = None
	        mul_9964 = sym_size_int * 1500
	        view_1611 = torch.ops.aten.view.default(mul_9954, [mul_9964, 1280]);  mul_9954 = mul_9964 = None
	        permute_171 = torch.ops.aten.permute.default(view_1610, [1, 0]);  view_1610 = None
	        addmm_85 = torch.ops.aten.addmm.default(model_audio_tower_layers_17_self_attn_q_proj_bias, view_1611, permute_171);  model_audio_tower_layers_17_self_attn_q_proj_bias = view_1611 = permute_171 = None
	        view_1612 = torch.ops.aten.view.default(addmm_85, [sym_size_int, 1500, 1280]);  addmm_85 = None
	        mul_9971 = torch.ops.aten.mul.Tensor(view_1612, 0.125);  view_1612 = None
	        view_1613 = torch.ops.aten.view.default(mul_9971, [sym_size_int, 1500, 20, 64]);  mul_9971 = None
	        permute_172 = torch.ops.aten.permute.default(view_1613, [0, 2, 1, 3]);  view_1613 = None
	        clone_138 = torch.ops.aten.clone.default(permute_172, memory_format = torch.contiguous_format);  permute_172 = None
	        amin_103 = torch.ops.aten.amin.default(add_15638, [2])
	        amax_103 = torch.ops.aten.amax.default(add_15638, [2])
	        full_206 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_103 = torch.ops.aten.minimum.default(amin_103, full_206);  amin_103 = full_206 = None
	        full_207 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_103 = torch.ops.aten.maximum.default(amax_103, full_207);  amax_103 = full_207 = None
	        sub_4721 = torch.ops.aten.sub.Tensor(maximum_103, minimum_103);  maximum_103 = None
	        div_206 = torch.ops.aten.div.Tensor(sub_4721, 255.0);  sub_4721 = None
	        clamp_min_309 = torch.ops.aten.clamp_min.default(div_206, 1.1920928955078125e-07);  div_206 = None
	        div_207 = torch.ops.aten.div.Tensor(minimum_103, clamp_min_309);  minimum_103 = None
	        round_207 = torch.ops.aten.round.default(div_207);  div_207 = None
	        sub_4727 = torch.ops.aten.sub.Tensor(-128, round_207);  round_207 = None
	        clamp_min_310 = torch.ops.aten.clamp_min.default(sub_4727, -128);  sub_4727 = None
	        clamp_max_206 = torch.ops.aten.clamp_max.default(clamp_min_310, 127);  clamp_min_310 = None
	        _assert_tensor_metadata_929 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_309, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_929 = None
	        _assert_tensor_metadata_930 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_206, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_930 = None
	        convert_element_type_618 = torch.ops.prims.convert_element_type.default(clamp_max_206, torch.int8);  clamp_max_206 = None
	        view_1616 = torch.ops.aten.view.default(clamp_min_309, [sym_size_int, 1500, 1])
	        view_1617 = torch.ops.aten.view.default(convert_element_type_618, [sym_size_int, 1500, 1])
	        reciprocal_103 = torch.ops.aten.reciprocal.default(view_1616);  view_1616 = None
	        mul_10025 = torch.ops.aten.mul.Tensor(reciprocal_103, 1.0);  reciprocal_103 = None
	        mul_10028 = torch.ops.aten.mul.Tensor(add_15638, mul_10025);  mul_10025 = None
	        round_208 = torch.ops.aten.round.default(mul_10028);  mul_10028 = None
	        add_15877 = torch.ops.aten.add.Tensor(round_208, view_1617);  round_208 = view_1617 = None
	        clamp_min_311 = torch.ops.aten.clamp_min.default(add_15877, -128);  add_15877 = None
	        clamp_max_207 = torch.ops.aten.clamp_max.default(clamp_min_311, 127);  clamp_min_311 = None
	        _assert_tensor_metadata_931 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_207, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_931 = None
	        convert_element_type_619 = torch.ops.prims.convert_element_type.default(clamp_max_207, torch.int8);  clamp_max_207 = None
	        view_1620 = torch.ops.aten.view.default(clamp_min_309, [sym_size_int, 1500, 1]);  clamp_min_309 = None
	        view_1621 = torch.ops.aten.view.default(convert_element_type_618, [sym_size_int, 1500, 1]);  convert_element_type_618 = None
	        _assert_tensor_metadata_932 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_619, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_932 = None
	        convert_element_type_620 = torch.ops.prims.convert_element_type.default(convert_element_type_619, torch.float32);  convert_element_type_619 = None
	        _assert_tensor_metadata_933 = torch.ops.aten._assert_tensor_metadata.default(view_1621, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_933 = None
	        convert_element_type_621 = torch.ops.prims.convert_element_type.default(view_1621, torch.float32);  view_1621 = None
	        sub_4747 = torch.ops.aten.sub.Tensor(convert_element_type_620, convert_element_type_621);  convert_element_type_620 = convert_element_type_621 = None
	        mul_10050 = torch.ops.aten.mul.Tensor(sub_4747, view_1620);  sub_4747 = view_1620 = None
	        _assert_tensor_metadata_934 = torch.ops.aten._assert_tensor_metadata.default(mul_10050, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_934 = None
	        view_1623 = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1624 = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1625 = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_935 = torch.ops.aten._assert_tensor_metadata.default(view_1623, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_935 = None
	        convert_element_type_622 = torch.ops.prims.convert_element_type.default(view_1623, torch.float32);  view_1623 = None
	        _assert_tensor_metadata_936 = torch.ops.aten._assert_tensor_metadata.default(view_1625, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_936 = None
	        convert_element_type_623 = torch.ops.prims.convert_element_type.default(view_1625, torch.float32);  view_1625 = None
	        sub_4751 = torch.ops.aten.sub.Tensor(convert_element_type_622, convert_element_type_623);  convert_element_type_622 = convert_element_type_623 = None
	        mul_10055 = torch.ops.aten.mul.Tensor(sub_4751, view_1624);  sub_4751 = view_1624 = None
	        view_1626 = torch.ops.aten.view.default(mul_10055, [1280, 1280]);  mul_10055 = None
	        _assert_tensor_metadata_937 = torch.ops.aten._assert_tensor_metadata.default(view_1626, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_937 = None
	        permute_173 = torch.ops.aten.permute.default(view_1626, [1, 0]);  view_1626 = None
	        mul_10058 = sym_size_int * 1500
	        view_1627 = torch.ops.aten.view.default(mul_10050, [mul_10058, 1280]);  mul_10050 = mul_10058 = None
	        mm_17 = torch.ops.aten.mm.default(view_1627, permute_173);  view_1627 = permute_173 = None
	        view_1628 = torch.ops.aten.view.default(mm_17, [sym_size_int, 1500, 1280]);  mm_17 = None
	        view_1629 = torch.ops.aten.view.default(view_1628, [sym_size_int, -1, 20, 64]);  view_1628 = None
	        permute_174 = torch.ops.aten.permute.default(view_1629, [0, 2, 1, 3]);  view_1629 = None
	        clone_139 = torch.ops.aten.clone.default(permute_174, memory_format = torch.contiguous_format);  permute_174 = None
	        amin_104 = torch.ops.aten.amin.default(add_15638, [2])
	        amax_104 = torch.ops.aten.amax.default(add_15638, [2])
	        full_208 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_104 = torch.ops.aten.minimum.default(amin_104, full_208);  amin_104 = full_208 = None
	        full_209 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_104 = torch.ops.aten.maximum.default(amax_104, full_209);  amax_104 = full_209 = None
	        sub_4765 = torch.ops.aten.sub.Tensor(maximum_104, minimum_104);  maximum_104 = None
	        div_208 = torch.ops.aten.div.Tensor(sub_4765, 255.0);  sub_4765 = None
	        clamp_min_312 = torch.ops.aten.clamp_min.default(div_208, 1.1920928955078125e-07);  div_208 = None
	        div_209 = torch.ops.aten.div.Tensor(minimum_104, clamp_min_312);  minimum_104 = None
	        round_209 = torch.ops.aten.round.default(div_209);  div_209 = None
	        sub_4771 = torch.ops.aten.sub.Tensor(-128, round_209);  round_209 = None
	        clamp_min_313 = torch.ops.aten.clamp_min.default(sub_4771, -128);  sub_4771 = None
	        clamp_max_208 = torch.ops.aten.clamp_max.default(clamp_min_313, 127);  clamp_min_313 = None
	        _assert_tensor_metadata_938 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_312, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_938 = None
	        _assert_tensor_metadata_939 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_208, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_939 = None
	        convert_element_type_624 = torch.ops.prims.convert_element_type.default(clamp_max_208, torch.int8);  clamp_max_208 = None
	        view_1632 = torch.ops.aten.view.default(clamp_min_312, [sym_size_int, 1500, 1])
	        view_1633 = torch.ops.aten.view.default(convert_element_type_624, [sym_size_int, 1500, 1])
	        reciprocal_104 = torch.ops.aten.reciprocal.default(view_1632);  view_1632 = None
	        mul_10124 = torch.ops.aten.mul.Tensor(reciprocal_104, 1.0);  reciprocal_104 = None
	        mul_10127 = torch.ops.aten.mul.Tensor(add_15638, mul_10124);  add_15638 = mul_10124 = None
	        round_210 = torch.ops.aten.round.default(mul_10127);  mul_10127 = None
	        add_16025 = torch.ops.aten.add.Tensor(round_210, view_1633);  round_210 = view_1633 = None
	        clamp_min_314 = torch.ops.aten.clamp_min.default(add_16025, -128);  add_16025 = None
	        clamp_max_209 = torch.ops.aten.clamp_max.default(clamp_min_314, 127);  clamp_min_314 = None
	        _assert_tensor_metadata_940 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_209, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_940 = None
	        convert_element_type_625 = torch.ops.prims.convert_element_type.default(clamp_max_209, torch.int8);  clamp_max_209 = None
	        view_1636 = torch.ops.aten.view.default(clamp_min_312, [sym_size_int, 1500, 1]);  clamp_min_312 = None
	        view_1637 = torch.ops.aten.view.default(convert_element_type_624, [sym_size_int, 1500, 1]);  convert_element_type_624 = None
	        _assert_tensor_metadata_941 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_625, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_941 = None
	        convert_element_type_626 = torch.ops.prims.convert_element_type.default(convert_element_type_625, torch.float32);  convert_element_type_625 = None
	        _assert_tensor_metadata_942 = torch.ops.aten._assert_tensor_metadata.default(view_1637, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_942 = None
	        convert_element_type_627 = torch.ops.prims.convert_element_type.default(view_1637, torch.float32);  view_1637 = None
	        sub_4791 = torch.ops.aten.sub.Tensor(convert_element_type_626, convert_element_type_627);  convert_element_type_626 = convert_element_type_627 = None
	        mul_10149 = torch.ops.aten.mul.Tensor(sub_4791, view_1636);  sub_4791 = view_1636 = None
	        _assert_tensor_metadata_943 = torch.ops.aten._assert_tensor_metadata.default(mul_10149, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_943 = None
	        view_1639 = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1640 = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1641 = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_944 = torch.ops.aten._assert_tensor_metadata.default(view_1639, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_944 = None
	        convert_element_type_628 = torch.ops.prims.convert_element_type.default(view_1639, torch.float32);  view_1639 = None
	        _assert_tensor_metadata_945 = torch.ops.aten._assert_tensor_metadata.default(view_1641, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_945 = None
	        convert_element_type_629 = torch.ops.prims.convert_element_type.default(view_1641, torch.float32);  view_1641 = None
	        sub_4795 = torch.ops.aten.sub.Tensor(convert_element_type_628, convert_element_type_629);  convert_element_type_628 = convert_element_type_629 = None
	        mul_10154 = torch.ops.aten.mul.Tensor(sub_4795, view_1640);  sub_4795 = view_1640 = None
	        view_1642 = torch.ops.aten.view.default(mul_10154, [1280, 1280]);  mul_10154 = None
	        _assert_tensor_metadata_946 = torch.ops.aten._assert_tensor_metadata.default(view_1642, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_946 = None
	        mul_10159 = sym_size_int * 1500
	        view_1643 = torch.ops.aten.view.default(mul_10149, [mul_10159, 1280]);  mul_10149 = mul_10159 = None
	        permute_175 = torch.ops.aten.permute.default(view_1642, [1, 0]);  view_1642 = None
	        addmm_86 = torch.ops.aten.addmm.default(model_audio_tower_layers_17_self_attn_v_proj_bias, view_1643, permute_175);  model_audio_tower_layers_17_self_attn_v_proj_bias = view_1643 = permute_175 = None
	        view_1644 = torch.ops.aten.view.default(addmm_86, [sym_size_int, 1500, 1280]);  addmm_86 = None
	        view_1645 = torch.ops.aten.view.default(view_1644, [sym_size_int, -1, 20, 64]);  view_1644 = None
	        permute_176 = torch.ops.aten.permute.default(view_1645, [0, 2, 1, 3]);  view_1645 = None
	        clone_140 = torch.ops.aten.clone.default(permute_176, memory_format = torch.contiguous_format);  permute_176 = None
	        _scaled_dot_product_efficient_attention_17 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_138, clone_139, clone_140, None, False, scale = 1.0);  clone_138 = clone_139 = clone_140 = None
	        getitem_138 = _scaled_dot_product_efficient_attention_17[0];  _scaled_dot_product_efficient_attention_17 = None
	        permute_177 = torch.ops.aten.permute.default(getitem_138, [0, 2, 1, 3]);  getitem_138 = None
	        view_1646 = torch.ops.aten.view.default(permute_177, [sym_size_int, 1500, -1]);  permute_177 = None
	        amin_105 = torch.ops.aten.amin.default(view_1646, [2])
	        amax_105 = torch.ops.aten.amax.default(view_1646, [2])
	        full_210 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_105 = torch.ops.aten.minimum.default(amin_105, full_210);  amin_105 = full_210 = None
	        full_211 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_105 = torch.ops.aten.maximum.default(amax_105, full_211);  amax_105 = full_211 = None
	        sub_4813 = torch.ops.aten.sub.Tensor(maximum_105, minimum_105);  maximum_105 = None
	        div_210 = torch.ops.aten.div.Tensor(sub_4813, 255.0);  sub_4813 = None
	        clamp_min_315 = torch.ops.aten.clamp_min.default(div_210, 1.1920928955078125e-07);  div_210 = None
	        div_211 = torch.ops.aten.div.Tensor(minimum_105, clamp_min_315);  minimum_105 = None
	        round_211 = torch.ops.aten.round.default(div_211);  div_211 = None
	        sub_4819 = torch.ops.aten.sub.Tensor(-128, round_211);  round_211 = None
	        clamp_min_316 = torch.ops.aten.clamp_min.default(sub_4819, -128);  sub_4819 = None
	        clamp_max_210 = torch.ops.aten.clamp_max.default(clamp_min_316, 127);  clamp_min_316 = None
	        _assert_tensor_metadata_947 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_315, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_947 = None
	        _assert_tensor_metadata_948 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_210, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_948 = None
	        convert_element_type_630 = torch.ops.prims.convert_element_type.default(clamp_max_210, torch.int8);  clamp_max_210 = None
	        view_1649 = torch.ops.aten.view.default(clamp_min_315, [sym_size_int, 1500, 1])
	        view_1650 = torch.ops.aten.view.default(convert_element_type_630, [sym_size_int, 1500, 1])
	        reciprocal_105 = torch.ops.aten.reciprocal.default(view_1649);  view_1649 = None
	        mul_10229 = torch.ops.aten.mul.Tensor(reciprocal_105, 1.0);  reciprocal_105 = None
	        mul_10232 = torch.ops.aten.mul.Tensor(view_1646, mul_10229);  view_1646 = mul_10229 = None
	        round_212 = torch.ops.aten.round.default(mul_10232);  mul_10232 = None
	        add_16189 = torch.ops.aten.add.Tensor(round_212, view_1650);  round_212 = view_1650 = None
	        clamp_min_317 = torch.ops.aten.clamp_min.default(add_16189, -128);  add_16189 = None
	        clamp_max_211 = torch.ops.aten.clamp_max.default(clamp_min_317, 127);  clamp_min_317 = None
	        _assert_tensor_metadata_949 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_211, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_949 = None
	        convert_element_type_631 = torch.ops.prims.convert_element_type.default(clamp_max_211, torch.int8);  clamp_max_211 = None
	        view_1653 = torch.ops.aten.view.default(clamp_min_315, [sym_size_int, 1500, 1]);  clamp_min_315 = None
	        view_1654 = torch.ops.aten.view.default(convert_element_type_630, [sym_size_int, 1500, 1]);  convert_element_type_630 = None
	        _assert_tensor_metadata_950 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_631, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_950 = None
	        convert_element_type_632 = torch.ops.prims.convert_element_type.default(convert_element_type_631, torch.float32);  convert_element_type_631 = None
	        _assert_tensor_metadata_951 = torch.ops.aten._assert_tensor_metadata.default(view_1654, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_951 = None
	        convert_element_type_633 = torch.ops.prims.convert_element_type.default(view_1654, torch.float32);  view_1654 = None
	        sub_4839 = torch.ops.aten.sub.Tensor(convert_element_type_632, convert_element_type_633);  convert_element_type_632 = convert_element_type_633 = None
	        mul_10254 = torch.ops.aten.mul.Tensor(sub_4839, view_1653);  sub_4839 = view_1653 = None
	        _assert_tensor_metadata_952 = torch.ops.aten._assert_tensor_metadata.default(mul_10254, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_952 = None
	        view_1656 = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1657 = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1658 = torch.ops.aten.view.default(model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_953 = torch.ops.aten._assert_tensor_metadata.default(view_1656, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_953 = None
	        convert_element_type_634 = torch.ops.prims.convert_element_type.default(view_1656, torch.float32);  view_1656 = None
	        _assert_tensor_metadata_954 = torch.ops.aten._assert_tensor_metadata.default(view_1658, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_954 = None
	        convert_element_type_635 = torch.ops.prims.convert_element_type.default(view_1658, torch.float32);  view_1658 = None
	        sub_4843 = torch.ops.aten.sub.Tensor(convert_element_type_634, convert_element_type_635);  convert_element_type_634 = convert_element_type_635 = None
	        mul_10259 = torch.ops.aten.mul.Tensor(sub_4843, view_1657);  sub_4843 = view_1657 = None
	        view_1659 = torch.ops.aten.view.default(mul_10259, [1280, 1280]);  mul_10259 = None
	        _assert_tensor_metadata_955 = torch.ops.aten._assert_tensor_metadata.default(view_1659, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_955 = None
	        mul_10264 = sym_size_int * 1500
	        view_1660 = torch.ops.aten.view.default(mul_10254, [mul_10264, 1280]);  mul_10254 = mul_10264 = None
	        permute_178 = torch.ops.aten.permute.default(view_1659, [1, 0]);  view_1659 = None
	        addmm_87 = torch.ops.aten.addmm.default(model_audio_tower_layers_17_self_attn_out_proj_bias, view_1660, permute_178);  model_audio_tower_layers_17_self_attn_out_proj_bias = view_1660 = permute_178 = None
	        view_1661 = torch.ops.aten.view.default(addmm_87, [sym_size_int, 1500, 1280]);  addmm_87 = None
	        add_16252 = torch.ops.aten.add.Tensor(add_15632, view_1661);  add_15632 = view_1661 = None
	        clone_142 = torch.ops.aten.clone.default(add_16252, memory_format = torch.contiguous_format)
	        var_mean_35 = torch.ops.aten.var_mean.correction(clone_142, [2], correction = 0, keepdim = True)
	        getitem_142 = var_mean_35[0]
	        getitem_143 = var_mean_35[1];  var_mean_35 = None
	        add_16257 = torch.ops.aten.add.Tensor(getitem_142, 1e-05);  getitem_142 = None
	        rsqrt_35 = torch.ops.aten.rsqrt.default(add_16257);  add_16257 = None
	        sub_4849 = torch.ops.aten.sub.Tensor(clone_142, getitem_143);  clone_142 = getitem_143 = None
	        mul_10275 = torch.ops.aten.mul.Tensor(sub_4849, rsqrt_35);  sub_4849 = rsqrt_35 = None
	        mul_10276 = torch.ops.aten.mul.Tensor(mul_10275, model_audio_tower_layers_17_final_layer_norm_weight);  mul_10275 = model_audio_tower_layers_17_final_layer_norm_weight = None
	        add_16258 = torch.ops.aten.add.Tensor(mul_10276, model_audio_tower_layers_17_final_layer_norm_bias);  mul_10276 = model_audio_tower_layers_17_final_layer_norm_bias = None
	        amin_106 = torch.ops.aten.amin.default(add_16258, [2])
	        amax_106 = torch.ops.aten.amax.default(add_16258, [2])
	        full_212 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_106 = torch.ops.aten.minimum.default(amin_106, full_212);  amin_106 = full_212 = None
	        full_213 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_106 = torch.ops.aten.maximum.default(amax_106, full_213);  amax_106 = full_213 = None
	        sub_4860 = torch.ops.aten.sub.Tensor(maximum_106, minimum_106);  maximum_106 = None
	        div_212 = torch.ops.aten.div.Tensor(sub_4860, 255.0);  sub_4860 = None
	        clamp_min_318 = torch.ops.aten.clamp_min.default(div_212, 1.1920928955078125e-07);  div_212 = None
	        div_213 = torch.ops.aten.div.Tensor(minimum_106, clamp_min_318);  minimum_106 = None
	        round_213 = torch.ops.aten.round.default(div_213);  div_213 = None
	        sub_4866 = torch.ops.aten.sub.Tensor(-128, round_213);  round_213 = None
	        clamp_min_319 = torch.ops.aten.clamp_min.default(sub_4866, -128);  sub_4866 = None
	        clamp_max_212 = torch.ops.aten.clamp_max.default(clamp_min_319, 127);  clamp_min_319 = None
	        _assert_tensor_metadata_956 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_318, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_956 = None
	        _assert_tensor_metadata_957 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_212, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_957 = None
	        convert_element_type_636 = torch.ops.prims.convert_element_type.default(clamp_max_212, torch.int8);  clamp_max_212 = None
	        view_1664 = torch.ops.aten.view.default(clamp_min_318, [sym_size_int, 1500, 1])
	        view_1665 = torch.ops.aten.view.default(convert_element_type_636, [sym_size_int, 1500, 1])
	        reciprocal_106 = torch.ops.aten.reciprocal.default(view_1664);  view_1664 = None
	        mul_10324 = torch.ops.aten.mul.Tensor(reciprocal_106, 1.0);  reciprocal_106 = None
	        mul_10327 = torch.ops.aten.mul.Tensor(add_16258, mul_10324);  add_16258 = mul_10324 = None
	        round_214 = torch.ops.aten.round.default(mul_10327);  mul_10327 = None
	        add_16345 = torch.ops.aten.add.Tensor(round_214, view_1665);  round_214 = view_1665 = None
	        clamp_min_320 = torch.ops.aten.clamp_min.default(add_16345, -128);  add_16345 = None
	        clamp_max_213 = torch.ops.aten.clamp_max.default(clamp_min_320, 127);  clamp_min_320 = None
	        _assert_tensor_metadata_958 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_213, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_958 = None
	        convert_element_type_637 = torch.ops.prims.convert_element_type.default(clamp_max_213, torch.int8);  clamp_max_213 = None
	        view_1668 = torch.ops.aten.view.default(clamp_min_318, [sym_size_int, 1500, 1]);  clamp_min_318 = None
	        view_1669 = torch.ops.aten.view.default(convert_element_type_636, [sym_size_int, 1500, 1]);  convert_element_type_636 = None
	        _assert_tensor_metadata_959 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_637, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_959 = None
	        convert_element_type_638 = torch.ops.prims.convert_element_type.default(convert_element_type_637, torch.float32);  convert_element_type_637 = None
	        _assert_tensor_metadata_960 = torch.ops.aten._assert_tensor_metadata.default(view_1669, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_960 = None
	        convert_element_type_639 = torch.ops.prims.convert_element_type.default(view_1669, torch.float32);  view_1669 = None
	        sub_4886 = torch.ops.aten.sub.Tensor(convert_element_type_638, convert_element_type_639);  convert_element_type_638 = convert_element_type_639 = None
	        mul_10349 = torch.ops.aten.mul.Tensor(sub_4886, view_1668);  sub_4886 = view_1668 = None
	        _assert_tensor_metadata_961 = torch.ops.aten._assert_tensor_metadata.default(mul_10349, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_961 = None
	        view_1671 = torch.ops.aten.view.default(model_audio_tower_layers_17_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_17_fc1_parametrizations_weight_original0 = None
	        view_1672 = torch.ops.aten.view.default(model_audio_tower_layers_17_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_17_fc1_parametrizations_weight_original1 = None
	        view_1673 = torch.ops.aten.view.default(model_audio_tower_layers_17_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_17_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_962 = torch.ops.aten._assert_tensor_metadata.default(view_1671, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_962 = None
	        convert_element_type_640 = torch.ops.prims.convert_element_type.default(view_1671, torch.float32);  view_1671 = None
	        _assert_tensor_metadata_963 = torch.ops.aten._assert_tensor_metadata.default(view_1673, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_963 = None
	        convert_element_type_641 = torch.ops.prims.convert_element_type.default(view_1673, torch.float32);  view_1673 = None
	        sub_4890 = torch.ops.aten.sub.Tensor(convert_element_type_640, convert_element_type_641);  convert_element_type_640 = convert_element_type_641 = None
	        mul_10354 = torch.ops.aten.mul.Tensor(sub_4890, view_1672);  sub_4890 = view_1672 = None
	        view_1674 = torch.ops.aten.view.default(mul_10354, [5120, 1280]);  mul_10354 = None
	        _assert_tensor_metadata_964 = torch.ops.aten._assert_tensor_metadata.default(view_1674, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_964 = None
	        mul_10359 = sym_size_int * 1500
	        view_1675 = torch.ops.aten.view.default(mul_10349, [mul_10359, 1280]);  mul_10349 = mul_10359 = None
	        permute_179 = torch.ops.aten.permute.default(view_1674, [1, 0]);  view_1674 = None
	        addmm_88 = torch.ops.aten.addmm.default(model_audio_tower_layers_17_fc1_bias, view_1675, permute_179);  model_audio_tower_layers_17_fc1_bias = view_1675 = permute_179 = None
	        view_1676 = torch.ops.aten.view.default(addmm_88, [sym_size_int, 1500, 5120]);  addmm_88 = None
	        mul_10366 = torch.ops.aten.mul.Tensor(view_1676, 0.5)
	        mul_10367 = torch.ops.aten.mul.Tensor(view_1676, 0.7071067811865476);  view_1676 = None
	        erf_19 = torch.ops.aten.erf.default(mul_10367);  mul_10367 = None
	        add_16404 = torch.ops.aten.add.Tensor(erf_19, 1);  erf_19 = None
	        mul_10368 = torch.ops.aten.mul.Tensor(mul_10366, add_16404);  mul_10366 = add_16404 = None
	        amin_107 = torch.ops.aten.amin.default(mul_10368, [2])
	        amax_107 = torch.ops.aten.amax.default(mul_10368, [2])
	        full_214 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_107 = torch.ops.aten.minimum.default(amin_107, full_214);  amin_107 = full_214 = None
	        full_215 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_107 = torch.ops.aten.maximum.default(amax_107, full_215);  amax_107 = full_215 = None
	        sub_4903 = torch.ops.aten.sub.Tensor(maximum_107, minimum_107);  maximum_107 = None
	        div_214 = torch.ops.aten.div.Tensor(sub_4903, 255.0);  sub_4903 = None
	        clamp_min_321 = torch.ops.aten.clamp_min.default(div_214, 1.1920928955078125e-07);  div_214 = None
	        div_215 = torch.ops.aten.div.Tensor(minimum_107, clamp_min_321);  minimum_107 = None
	        round_215 = torch.ops.aten.round.default(div_215);  div_215 = None
	        sub_4909 = torch.ops.aten.sub.Tensor(-128, round_215);  round_215 = None
	        clamp_min_322 = torch.ops.aten.clamp_min.default(sub_4909, -128);  sub_4909 = None
	        clamp_max_214 = torch.ops.aten.clamp_max.default(clamp_min_322, 127);  clamp_min_322 = None
	        _assert_tensor_metadata_965 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_321, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_965 = None
	        _assert_tensor_metadata_966 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_214, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_966 = None
	        convert_element_type_642 = torch.ops.prims.convert_element_type.default(clamp_max_214, torch.int8);  clamp_max_214 = None
	        view_1679 = torch.ops.aten.view.default(clamp_min_321, [sym_size_int, 1500, 1])
	        view_1680 = torch.ops.aten.view.default(convert_element_type_642, [sym_size_int, 1500, 1])
	        reciprocal_107 = torch.ops.aten.reciprocal.default(view_1679);  view_1679 = None
	        mul_10414 = torch.ops.aten.mul.Tensor(reciprocal_107, 1.0);  reciprocal_107 = None
	        mul_10417 = torch.ops.aten.mul.Tensor(mul_10368, mul_10414);  mul_10368 = mul_10414 = None
	        round_216 = torch.ops.aten.round.default(mul_10417);  mul_10417 = None
	        add_16487 = torch.ops.aten.add.Tensor(round_216, view_1680);  round_216 = view_1680 = None
	        clamp_min_323 = torch.ops.aten.clamp_min.default(add_16487, -128);  add_16487 = None
	        clamp_max_215 = torch.ops.aten.clamp_max.default(clamp_min_323, 127);  clamp_min_323 = None
	        _assert_tensor_metadata_967 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_215, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_967 = None
	        convert_element_type_643 = torch.ops.prims.convert_element_type.default(clamp_max_215, torch.int8);  clamp_max_215 = None
	        view_1683 = torch.ops.aten.view.default(clamp_min_321, [sym_size_int, 1500, 1]);  clamp_min_321 = None
	        view_1684 = torch.ops.aten.view.default(convert_element_type_642, [sym_size_int, 1500, 1]);  convert_element_type_642 = None
	        _assert_tensor_metadata_968 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_643, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_968 = None
	        convert_element_type_644 = torch.ops.prims.convert_element_type.default(convert_element_type_643, torch.float32);  convert_element_type_643 = None
	        _assert_tensor_metadata_969 = torch.ops.aten._assert_tensor_metadata.default(view_1684, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_969 = None
	        convert_element_type_645 = torch.ops.prims.convert_element_type.default(view_1684, torch.float32);  view_1684 = None
	        sub_4929 = torch.ops.aten.sub.Tensor(convert_element_type_644, convert_element_type_645);  convert_element_type_644 = convert_element_type_645 = None
	        mul_10439 = torch.ops.aten.mul.Tensor(sub_4929, view_1683);  sub_4929 = view_1683 = None
	        _assert_tensor_metadata_970 = torch.ops.aten._assert_tensor_metadata.default(mul_10439, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_970 = None
	        view_1686 = torch.ops.aten.view.default(model_audio_tower_layers_17_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_17_fc2_parametrizations_weight_original0 = None
	        view_1687 = torch.ops.aten.view.default(model_audio_tower_layers_17_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_17_fc2_parametrizations_weight_original1 = None
	        view_1688 = torch.ops.aten.view.default(model_audio_tower_layers_17_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_17_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_971 = torch.ops.aten._assert_tensor_metadata.default(view_1686, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_971 = None
	        convert_element_type_646 = torch.ops.prims.convert_element_type.default(view_1686, torch.float32);  view_1686 = None
	        _assert_tensor_metadata_972 = torch.ops.aten._assert_tensor_metadata.default(view_1688, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_972 = None
	        convert_element_type_647 = torch.ops.prims.convert_element_type.default(view_1688, torch.float32);  view_1688 = None
	        sub_4933 = torch.ops.aten.sub.Tensor(convert_element_type_646, convert_element_type_647);  convert_element_type_646 = convert_element_type_647 = None
	        mul_10444 = torch.ops.aten.mul.Tensor(sub_4933, view_1687);  sub_4933 = view_1687 = None
	        view_1689 = torch.ops.aten.view.default(mul_10444, [1280, 5120]);  mul_10444 = None
	        _assert_tensor_metadata_973 = torch.ops.aten._assert_tensor_metadata.default(view_1689, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_973 = None
	        mul_10449 = sym_size_int * 1500
	        view_1690 = torch.ops.aten.view.default(mul_10439, [mul_10449, 5120]);  mul_10439 = mul_10449 = None
	        permute_180 = torch.ops.aten.permute.default(view_1689, [1, 0]);  view_1689 = None
	        addmm_89 = torch.ops.aten.addmm.default(model_audio_tower_layers_17_fc2_bias, view_1690, permute_180);  model_audio_tower_layers_17_fc2_bias = view_1690 = permute_180 = None
	        view_1691 = torch.ops.aten.view.default(addmm_89, [sym_size_int, 1500, 1280]);  addmm_89 = None
	        add_16550 = torch.ops.aten.add.Tensor(add_16252, view_1691);  add_16252 = view_1691 = None
	        clone_145 = torch.ops.aten.clone.default(add_16550, memory_format = torch.contiguous_format)
	        var_mean_36 = torch.ops.aten.var_mean.correction(clone_145, [2], correction = 0, keepdim = True)
	        getitem_144 = var_mean_36[0]
	        getitem_145 = var_mean_36[1];  var_mean_36 = None
	        add_16555 = torch.ops.aten.add.Tensor(getitem_144, 1e-05);  getitem_144 = None
	        rsqrt_36 = torch.ops.aten.rsqrt.default(add_16555);  add_16555 = None
	        sub_4939 = torch.ops.aten.sub.Tensor(clone_145, getitem_145);  clone_145 = getitem_145 = None
	        mul_10460 = torch.ops.aten.mul.Tensor(sub_4939, rsqrt_36);  sub_4939 = rsqrt_36 = None
	        mul_10461 = torch.ops.aten.mul.Tensor(mul_10460, model_audio_tower_layers_18_self_attn_layer_norm_weight);  mul_10460 = model_audio_tower_layers_18_self_attn_layer_norm_weight = None
	        add_16556 = torch.ops.aten.add.Tensor(mul_10461, model_audio_tower_layers_18_self_attn_layer_norm_bias);  mul_10461 = model_audio_tower_layers_18_self_attn_layer_norm_bias = None
	        amin_108 = torch.ops.aten.amin.default(add_16556, [2])
	        amax_108 = torch.ops.aten.amax.default(add_16556, [2])
	        full_216 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_108 = torch.ops.aten.minimum.default(amin_108, full_216);  amin_108 = full_216 = None
	        full_217 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_108 = torch.ops.aten.maximum.default(amax_108, full_217);  amax_108 = full_217 = None
	        sub_4950 = torch.ops.aten.sub.Tensor(maximum_108, minimum_108);  maximum_108 = None
	        div_216 = torch.ops.aten.div.Tensor(sub_4950, 255.0);  sub_4950 = None
	        clamp_min_324 = torch.ops.aten.clamp_min.default(div_216, 1.1920928955078125e-07);  div_216 = None
	        div_217 = torch.ops.aten.div.Tensor(minimum_108, clamp_min_324);  minimum_108 = None
	        round_217 = torch.ops.aten.round.default(div_217);  div_217 = None
	        sub_4956 = torch.ops.aten.sub.Tensor(-128, round_217);  round_217 = None
	        clamp_min_325 = torch.ops.aten.clamp_min.default(sub_4956, -128);  sub_4956 = None
	        clamp_max_216 = torch.ops.aten.clamp_max.default(clamp_min_325, 127);  clamp_min_325 = None
	        _assert_tensor_metadata_974 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_324, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_974 = None
	        _assert_tensor_metadata_975 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_216, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_975 = None
	        convert_element_type_648 = torch.ops.prims.convert_element_type.default(clamp_max_216, torch.int8);  clamp_max_216 = None
	        view_1694 = torch.ops.aten.view.default(clamp_min_324, [sym_size_int, 1500, 1])
	        view_1695 = torch.ops.aten.view.default(convert_element_type_648, [sym_size_int, 1500, 1])
	        reciprocal_108 = torch.ops.aten.reciprocal.default(view_1694);  view_1694 = None
	        mul_10509 = torch.ops.aten.mul.Tensor(reciprocal_108, 1.0);  reciprocal_108 = None
	        mul_10512 = torch.ops.aten.mul.Tensor(add_16556, mul_10509);  mul_10509 = None
	        round_218 = torch.ops.aten.round.default(mul_10512);  mul_10512 = None
	        add_16643 = torch.ops.aten.add.Tensor(round_218, view_1695);  round_218 = view_1695 = None
	        clamp_min_326 = torch.ops.aten.clamp_min.default(add_16643, -128);  add_16643 = None
	        clamp_max_217 = torch.ops.aten.clamp_max.default(clamp_min_326, 127);  clamp_min_326 = None
	        _assert_tensor_metadata_976 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_217, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_976 = None
	        convert_element_type_649 = torch.ops.prims.convert_element_type.default(clamp_max_217, torch.int8);  clamp_max_217 = None
	        view_1698 = torch.ops.aten.view.default(clamp_min_324, [sym_size_int, 1500, 1]);  clamp_min_324 = None
	        view_1699 = torch.ops.aten.view.default(convert_element_type_648, [sym_size_int, 1500, 1]);  convert_element_type_648 = None
	        _assert_tensor_metadata_977 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_649, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_977 = None
	        convert_element_type_650 = torch.ops.prims.convert_element_type.default(convert_element_type_649, torch.float32);  convert_element_type_649 = None
	        _assert_tensor_metadata_978 = torch.ops.aten._assert_tensor_metadata.default(view_1699, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_978 = None
	        convert_element_type_651 = torch.ops.prims.convert_element_type.default(view_1699, torch.float32);  view_1699 = None
	        sub_4976 = torch.ops.aten.sub.Tensor(convert_element_type_650, convert_element_type_651);  convert_element_type_650 = convert_element_type_651 = None
	        mul_10534 = torch.ops.aten.mul.Tensor(sub_4976, view_1698);  sub_4976 = view_1698 = None
	        _assert_tensor_metadata_979 = torch.ops.aten._assert_tensor_metadata.default(mul_10534, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_979 = None
	        view_1701 = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1702 = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1703 = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_980 = torch.ops.aten._assert_tensor_metadata.default(view_1701, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_980 = None
	        convert_element_type_652 = torch.ops.prims.convert_element_type.default(view_1701, torch.float32);  view_1701 = None
	        _assert_tensor_metadata_981 = torch.ops.aten._assert_tensor_metadata.default(view_1703, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_981 = None
	        convert_element_type_653 = torch.ops.prims.convert_element_type.default(view_1703, torch.float32);  view_1703 = None
	        sub_4980 = torch.ops.aten.sub.Tensor(convert_element_type_652, convert_element_type_653);  convert_element_type_652 = convert_element_type_653 = None
	        mul_10539 = torch.ops.aten.mul.Tensor(sub_4980, view_1702);  sub_4980 = view_1702 = None
	        view_1704 = torch.ops.aten.view.default(mul_10539, [1280, 1280]);  mul_10539 = None
	        _assert_tensor_metadata_982 = torch.ops.aten._assert_tensor_metadata.default(view_1704, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_982 = None
	        mul_10544 = sym_size_int * 1500
	        view_1705 = torch.ops.aten.view.default(mul_10534, [mul_10544, 1280]);  mul_10534 = mul_10544 = None
	        permute_181 = torch.ops.aten.permute.default(view_1704, [1, 0]);  view_1704 = None
	        addmm_90 = torch.ops.aten.addmm.default(model_audio_tower_layers_18_self_attn_q_proj_bias, view_1705, permute_181);  model_audio_tower_layers_18_self_attn_q_proj_bias = view_1705 = permute_181 = None
	        view_1706 = torch.ops.aten.view.default(addmm_90, [sym_size_int, 1500, 1280]);  addmm_90 = None
	        mul_10551 = torch.ops.aten.mul.Tensor(view_1706, 0.125);  view_1706 = None
	        view_1707 = torch.ops.aten.view.default(mul_10551, [sym_size_int, 1500, 20, 64]);  mul_10551 = None
	        permute_182 = torch.ops.aten.permute.default(view_1707, [0, 2, 1, 3]);  view_1707 = None
	        clone_146 = torch.ops.aten.clone.default(permute_182, memory_format = torch.contiguous_format);  permute_182 = None
	        amin_109 = torch.ops.aten.amin.default(add_16556, [2])
	        amax_109 = torch.ops.aten.amax.default(add_16556, [2])
	        full_218 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_109 = torch.ops.aten.minimum.default(amin_109, full_218);  amin_109 = full_218 = None
	        full_219 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_109 = torch.ops.aten.maximum.default(amax_109, full_219);  amax_109 = full_219 = None
	        sub_4995 = torch.ops.aten.sub.Tensor(maximum_109, minimum_109);  maximum_109 = None
	        div_218 = torch.ops.aten.div.Tensor(sub_4995, 255.0);  sub_4995 = None
	        clamp_min_327 = torch.ops.aten.clamp_min.default(div_218, 1.1920928955078125e-07);  div_218 = None
	        div_219 = torch.ops.aten.div.Tensor(minimum_109, clamp_min_327);  minimum_109 = None
	        round_219 = torch.ops.aten.round.default(div_219);  div_219 = None
	        sub_5001 = torch.ops.aten.sub.Tensor(-128, round_219);  round_219 = None
	        clamp_min_328 = torch.ops.aten.clamp_min.default(sub_5001, -128);  sub_5001 = None
	        clamp_max_218 = torch.ops.aten.clamp_max.default(clamp_min_328, 127);  clamp_min_328 = None
	        _assert_tensor_metadata_983 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_327, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_983 = None
	        _assert_tensor_metadata_984 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_218, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_984 = None
	        convert_element_type_654 = torch.ops.prims.convert_element_type.default(clamp_max_218, torch.int8);  clamp_max_218 = None
	        view_1710 = torch.ops.aten.view.default(clamp_min_327, [sym_size_int, 1500, 1])
	        view_1711 = torch.ops.aten.view.default(convert_element_type_654, [sym_size_int, 1500, 1])
	        reciprocal_109 = torch.ops.aten.reciprocal.default(view_1710);  view_1710 = None
	        mul_10605 = torch.ops.aten.mul.Tensor(reciprocal_109, 1.0);  reciprocal_109 = None
	        mul_10608 = torch.ops.aten.mul.Tensor(add_16556, mul_10605);  mul_10605 = None
	        round_220 = torch.ops.aten.round.default(mul_10608);  mul_10608 = None
	        add_16795 = torch.ops.aten.add.Tensor(round_220, view_1711);  round_220 = view_1711 = None
	        clamp_min_329 = torch.ops.aten.clamp_min.default(add_16795, -128);  add_16795 = None
	        clamp_max_219 = torch.ops.aten.clamp_max.default(clamp_min_329, 127);  clamp_min_329 = None
	        _assert_tensor_metadata_985 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_219, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_985 = None
	        convert_element_type_655 = torch.ops.prims.convert_element_type.default(clamp_max_219, torch.int8);  clamp_max_219 = None
	        view_1714 = torch.ops.aten.view.default(clamp_min_327, [sym_size_int, 1500, 1]);  clamp_min_327 = None
	        view_1715 = torch.ops.aten.view.default(convert_element_type_654, [sym_size_int, 1500, 1]);  convert_element_type_654 = None
	        _assert_tensor_metadata_986 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_655, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_986 = None
	        convert_element_type_656 = torch.ops.prims.convert_element_type.default(convert_element_type_655, torch.float32);  convert_element_type_655 = None
	        _assert_tensor_metadata_987 = torch.ops.aten._assert_tensor_metadata.default(view_1715, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_987 = None
	        convert_element_type_657 = torch.ops.prims.convert_element_type.default(view_1715, torch.float32);  view_1715 = None
	        sub_5021 = torch.ops.aten.sub.Tensor(convert_element_type_656, convert_element_type_657);  convert_element_type_656 = convert_element_type_657 = None
	        mul_10630 = torch.ops.aten.mul.Tensor(sub_5021, view_1714);  sub_5021 = view_1714 = None
	        _assert_tensor_metadata_988 = torch.ops.aten._assert_tensor_metadata.default(mul_10630, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_988 = None
	        view_1717 = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1718 = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1719 = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_989 = torch.ops.aten._assert_tensor_metadata.default(view_1717, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_989 = None
	        convert_element_type_658 = torch.ops.prims.convert_element_type.default(view_1717, torch.float32);  view_1717 = None
	        _assert_tensor_metadata_990 = torch.ops.aten._assert_tensor_metadata.default(view_1719, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_990 = None
	        convert_element_type_659 = torch.ops.prims.convert_element_type.default(view_1719, torch.float32);  view_1719 = None
	        sub_5025 = torch.ops.aten.sub.Tensor(convert_element_type_658, convert_element_type_659);  convert_element_type_658 = convert_element_type_659 = None
	        mul_10635 = torch.ops.aten.mul.Tensor(sub_5025, view_1718);  sub_5025 = view_1718 = None
	        view_1720 = torch.ops.aten.view.default(mul_10635, [1280, 1280]);  mul_10635 = None
	        _assert_tensor_metadata_991 = torch.ops.aten._assert_tensor_metadata.default(view_1720, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_991 = None
	        permute_183 = torch.ops.aten.permute.default(view_1720, [1, 0]);  view_1720 = None
	        mul_10638 = sym_size_int * 1500
	        view_1721 = torch.ops.aten.view.default(mul_10630, [mul_10638, 1280]);  mul_10630 = mul_10638 = None
	        mm_18 = torch.ops.aten.mm.default(view_1721, permute_183);  view_1721 = permute_183 = None
	        view_1722 = torch.ops.aten.view.default(mm_18, [sym_size_int, 1500, 1280]);  mm_18 = None
	        view_1723 = torch.ops.aten.view.default(view_1722, [sym_size_int, -1, 20, 64]);  view_1722 = None
	        permute_184 = torch.ops.aten.permute.default(view_1723, [0, 2, 1, 3]);  view_1723 = None
	        clone_147 = torch.ops.aten.clone.default(permute_184, memory_format = torch.contiguous_format);  permute_184 = None
	        amin_110 = torch.ops.aten.amin.default(add_16556, [2])
	        amax_110 = torch.ops.aten.amax.default(add_16556, [2])
	        full_220 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_110 = torch.ops.aten.minimum.default(amin_110, full_220);  amin_110 = full_220 = None
	        full_221 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_110 = torch.ops.aten.maximum.default(amax_110, full_221);  amax_110 = full_221 = None
	        sub_5039 = torch.ops.aten.sub.Tensor(maximum_110, minimum_110);  maximum_110 = None
	        div_220 = torch.ops.aten.div.Tensor(sub_5039, 255.0);  sub_5039 = None
	        clamp_min_330 = torch.ops.aten.clamp_min.default(div_220, 1.1920928955078125e-07);  div_220 = None
	        div_221 = torch.ops.aten.div.Tensor(minimum_110, clamp_min_330);  minimum_110 = None
	        round_221 = torch.ops.aten.round.default(div_221);  div_221 = None
	        sub_5045 = torch.ops.aten.sub.Tensor(-128, round_221);  round_221 = None
	        clamp_min_331 = torch.ops.aten.clamp_min.default(sub_5045, -128);  sub_5045 = None
	        clamp_max_220 = torch.ops.aten.clamp_max.default(clamp_min_331, 127);  clamp_min_331 = None
	        _assert_tensor_metadata_992 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_330, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_992 = None
	        _assert_tensor_metadata_993 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_220, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_993 = None
	        convert_element_type_660 = torch.ops.prims.convert_element_type.default(clamp_max_220, torch.int8);  clamp_max_220 = None
	        view_1726 = torch.ops.aten.view.default(clamp_min_330, [sym_size_int, 1500, 1])
	        view_1727 = torch.ops.aten.view.default(convert_element_type_660, [sym_size_int, 1500, 1])
	        reciprocal_110 = torch.ops.aten.reciprocal.default(view_1726);  view_1726 = None
	        mul_10704 = torch.ops.aten.mul.Tensor(reciprocal_110, 1.0);  reciprocal_110 = None
	        mul_10707 = torch.ops.aten.mul.Tensor(add_16556, mul_10704);  add_16556 = mul_10704 = None
	        round_222 = torch.ops.aten.round.default(mul_10707);  mul_10707 = None
	        add_16943 = torch.ops.aten.add.Tensor(round_222, view_1727);  round_222 = view_1727 = None
	        clamp_min_332 = torch.ops.aten.clamp_min.default(add_16943, -128);  add_16943 = None
	        clamp_max_221 = torch.ops.aten.clamp_max.default(clamp_min_332, 127);  clamp_min_332 = None
	        _assert_tensor_metadata_994 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_221, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_994 = None
	        convert_element_type_661 = torch.ops.prims.convert_element_type.default(clamp_max_221, torch.int8);  clamp_max_221 = None
	        view_1730 = torch.ops.aten.view.default(clamp_min_330, [sym_size_int, 1500, 1]);  clamp_min_330 = None
	        view_1731 = torch.ops.aten.view.default(convert_element_type_660, [sym_size_int, 1500, 1]);  convert_element_type_660 = None
	        _assert_tensor_metadata_995 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_661, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_995 = None
	        convert_element_type_662 = torch.ops.prims.convert_element_type.default(convert_element_type_661, torch.float32);  convert_element_type_661 = None
	        _assert_tensor_metadata_996 = torch.ops.aten._assert_tensor_metadata.default(view_1731, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_996 = None
	        convert_element_type_663 = torch.ops.prims.convert_element_type.default(view_1731, torch.float32);  view_1731 = None
	        sub_5065 = torch.ops.aten.sub.Tensor(convert_element_type_662, convert_element_type_663);  convert_element_type_662 = convert_element_type_663 = None
	        mul_10729 = torch.ops.aten.mul.Tensor(sub_5065, view_1730);  sub_5065 = view_1730 = None
	        _assert_tensor_metadata_997 = torch.ops.aten._assert_tensor_metadata.default(mul_10729, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_997 = None
	        view_1733 = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1734 = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1735 = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_998 = torch.ops.aten._assert_tensor_metadata.default(view_1733, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_998 = None
	        convert_element_type_664 = torch.ops.prims.convert_element_type.default(view_1733, torch.float32);  view_1733 = None
	        _assert_tensor_metadata_999 = torch.ops.aten._assert_tensor_metadata.default(view_1735, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_999 = None
	        convert_element_type_665 = torch.ops.prims.convert_element_type.default(view_1735, torch.float32);  view_1735 = None
	        sub_5069 = torch.ops.aten.sub.Tensor(convert_element_type_664, convert_element_type_665);  convert_element_type_664 = convert_element_type_665 = None
	        mul_10734 = torch.ops.aten.mul.Tensor(sub_5069, view_1734);  sub_5069 = view_1734 = None
	        view_1736 = torch.ops.aten.view.default(mul_10734, [1280, 1280]);  mul_10734 = None
	        _assert_tensor_metadata_1000 = torch.ops.aten._assert_tensor_metadata.default(view_1736, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1000 = None
	        mul_10739 = sym_size_int * 1500
	        view_1737 = torch.ops.aten.view.default(mul_10729, [mul_10739, 1280]);  mul_10729 = mul_10739 = None
	        permute_185 = torch.ops.aten.permute.default(view_1736, [1, 0]);  view_1736 = None
	        addmm_91 = torch.ops.aten.addmm.default(model_audio_tower_layers_18_self_attn_v_proj_bias, view_1737, permute_185);  model_audio_tower_layers_18_self_attn_v_proj_bias = view_1737 = permute_185 = None
	        view_1738 = torch.ops.aten.view.default(addmm_91, [sym_size_int, 1500, 1280]);  addmm_91 = None
	        view_1739 = torch.ops.aten.view.default(view_1738, [sym_size_int, -1, 20, 64]);  view_1738 = None
	        permute_186 = torch.ops.aten.permute.default(view_1739, [0, 2, 1, 3]);  view_1739 = None
	        clone_148 = torch.ops.aten.clone.default(permute_186, memory_format = torch.contiguous_format);  permute_186 = None
	        _scaled_dot_product_efficient_attention_18 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_146, clone_147, clone_148, None, False, scale = 1.0);  clone_146 = clone_147 = clone_148 = None
	        getitem_146 = _scaled_dot_product_efficient_attention_18[0];  _scaled_dot_product_efficient_attention_18 = None
	        permute_187 = torch.ops.aten.permute.default(getitem_146, [0, 2, 1, 3]);  getitem_146 = None
	        view_1740 = torch.ops.aten.view.default(permute_187, [sym_size_int, 1500, -1]);  permute_187 = None
	        amin_111 = torch.ops.aten.amin.default(view_1740, [2])
	        amax_111 = torch.ops.aten.amax.default(view_1740, [2])
	        full_222 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_111 = torch.ops.aten.minimum.default(amin_111, full_222);  amin_111 = full_222 = None
	        full_223 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_111 = torch.ops.aten.maximum.default(amax_111, full_223);  amax_111 = full_223 = None
	        sub_5087 = torch.ops.aten.sub.Tensor(maximum_111, minimum_111);  maximum_111 = None
	        div_222 = torch.ops.aten.div.Tensor(sub_5087, 255.0);  sub_5087 = None
	        clamp_min_333 = torch.ops.aten.clamp_min.default(div_222, 1.1920928955078125e-07);  div_222 = None
	        div_223 = torch.ops.aten.div.Tensor(minimum_111, clamp_min_333);  minimum_111 = None
	        round_223 = torch.ops.aten.round.default(div_223);  div_223 = None
	        sub_5093 = torch.ops.aten.sub.Tensor(-128, round_223);  round_223 = None
	        clamp_min_334 = torch.ops.aten.clamp_min.default(sub_5093, -128);  sub_5093 = None
	        clamp_max_222 = torch.ops.aten.clamp_max.default(clamp_min_334, 127);  clamp_min_334 = None
	        _assert_tensor_metadata_1001 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_333, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1001 = None
	        _assert_tensor_metadata_1002 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_222, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1002 = None
	        convert_element_type_666 = torch.ops.prims.convert_element_type.default(clamp_max_222, torch.int8);  clamp_max_222 = None
	        view_1743 = torch.ops.aten.view.default(clamp_min_333, [sym_size_int, 1500, 1])
	        view_1744 = torch.ops.aten.view.default(convert_element_type_666, [sym_size_int, 1500, 1])
	        reciprocal_111 = torch.ops.aten.reciprocal.default(view_1743);  view_1743 = None
	        mul_10809 = torch.ops.aten.mul.Tensor(reciprocal_111, 1.0);  reciprocal_111 = None
	        mul_10812 = torch.ops.aten.mul.Tensor(view_1740, mul_10809);  view_1740 = mul_10809 = None
	        round_224 = torch.ops.aten.round.default(mul_10812);  mul_10812 = None
	        add_17107 = torch.ops.aten.add.Tensor(round_224, view_1744);  round_224 = view_1744 = None
	        clamp_min_335 = torch.ops.aten.clamp_min.default(add_17107, -128);  add_17107 = None
	        clamp_max_223 = torch.ops.aten.clamp_max.default(clamp_min_335, 127);  clamp_min_335 = None
	        _assert_tensor_metadata_1003 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_223, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1003 = None
	        convert_element_type_667 = torch.ops.prims.convert_element_type.default(clamp_max_223, torch.int8);  clamp_max_223 = None
	        view_1747 = torch.ops.aten.view.default(clamp_min_333, [sym_size_int, 1500, 1]);  clamp_min_333 = None
	        view_1748 = torch.ops.aten.view.default(convert_element_type_666, [sym_size_int, 1500, 1]);  convert_element_type_666 = None
	        _assert_tensor_metadata_1004 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_667, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1004 = None
	        convert_element_type_668 = torch.ops.prims.convert_element_type.default(convert_element_type_667, torch.float32);  convert_element_type_667 = None
	        _assert_tensor_metadata_1005 = torch.ops.aten._assert_tensor_metadata.default(view_1748, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1005 = None
	        convert_element_type_669 = torch.ops.prims.convert_element_type.default(view_1748, torch.float32);  view_1748 = None
	        sub_5113 = torch.ops.aten.sub.Tensor(convert_element_type_668, convert_element_type_669);  convert_element_type_668 = convert_element_type_669 = None
	        mul_10834 = torch.ops.aten.mul.Tensor(sub_5113, view_1747);  sub_5113 = view_1747 = None
	        _assert_tensor_metadata_1006 = torch.ops.aten._assert_tensor_metadata.default(mul_10834, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1006 = None
	        view_1750 = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1751 = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1752 = torch.ops.aten.view.default(model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1007 = torch.ops.aten._assert_tensor_metadata.default(view_1750, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1007 = None
	        convert_element_type_670 = torch.ops.prims.convert_element_type.default(view_1750, torch.float32);  view_1750 = None
	        _assert_tensor_metadata_1008 = torch.ops.aten._assert_tensor_metadata.default(view_1752, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1008 = None
	        convert_element_type_671 = torch.ops.prims.convert_element_type.default(view_1752, torch.float32);  view_1752 = None
	        sub_5117 = torch.ops.aten.sub.Tensor(convert_element_type_670, convert_element_type_671);  convert_element_type_670 = convert_element_type_671 = None
	        mul_10839 = torch.ops.aten.mul.Tensor(sub_5117, view_1751);  sub_5117 = view_1751 = None
	        view_1753 = torch.ops.aten.view.default(mul_10839, [1280, 1280]);  mul_10839 = None
	        _assert_tensor_metadata_1009 = torch.ops.aten._assert_tensor_metadata.default(view_1753, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1009 = None
	        mul_10844 = sym_size_int * 1500
	        view_1754 = torch.ops.aten.view.default(mul_10834, [mul_10844, 1280]);  mul_10834 = mul_10844 = None
	        permute_188 = torch.ops.aten.permute.default(view_1753, [1, 0]);  view_1753 = None
	        addmm_92 = torch.ops.aten.addmm.default(model_audio_tower_layers_18_self_attn_out_proj_bias, view_1754, permute_188);  model_audio_tower_layers_18_self_attn_out_proj_bias = view_1754 = permute_188 = None
	        view_1755 = torch.ops.aten.view.default(addmm_92, [sym_size_int, 1500, 1280]);  addmm_92 = None
	        add_17170 = torch.ops.aten.add.Tensor(add_16550, view_1755);  add_16550 = view_1755 = None
	        clone_150 = torch.ops.aten.clone.default(add_17170, memory_format = torch.contiguous_format)
	        var_mean_37 = torch.ops.aten.var_mean.correction(clone_150, [2], correction = 0, keepdim = True)
	        getitem_150 = var_mean_37[0]
	        getitem_151 = var_mean_37[1];  var_mean_37 = None
	        add_17175 = torch.ops.aten.add.Tensor(getitem_150, 1e-05);  getitem_150 = None
	        rsqrt_37 = torch.ops.aten.rsqrt.default(add_17175);  add_17175 = None
	        sub_5123 = torch.ops.aten.sub.Tensor(clone_150, getitem_151);  clone_150 = getitem_151 = None
	        mul_10855 = torch.ops.aten.mul.Tensor(sub_5123, rsqrt_37);  sub_5123 = rsqrt_37 = None
	        mul_10856 = torch.ops.aten.mul.Tensor(mul_10855, model_audio_tower_layers_18_final_layer_norm_weight);  mul_10855 = model_audio_tower_layers_18_final_layer_norm_weight = None
	        add_17176 = torch.ops.aten.add.Tensor(mul_10856, model_audio_tower_layers_18_final_layer_norm_bias);  mul_10856 = model_audio_tower_layers_18_final_layer_norm_bias = None
	        amin_112 = torch.ops.aten.amin.default(add_17176, [2])
	        amax_112 = torch.ops.aten.amax.default(add_17176, [2])
	        full_224 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_112 = torch.ops.aten.minimum.default(amin_112, full_224);  amin_112 = full_224 = None
	        full_225 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_112 = torch.ops.aten.maximum.default(amax_112, full_225);  amax_112 = full_225 = None
	        sub_5134 = torch.ops.aten.sub.Tensor(maximum_112, minimum_112);  maximum_112 = None
	        div_224 = torch.ops.aten.div.Tensor(sub_5134, 255.0);  sub_5134 = None
	        clamp_min_336 = torch.ops.aten.clamp_min.default(div_224, 1.1920928955078125e-07);  div_224 = None
	        div_225 = torch.ops.aten.div.Tensor(minimum_112, clamp_min_336);  minimum_112 = None
	        round_225 = torch.ops.aten.round.default(div_225);  div_225 = None
	        sub_5140 = torch.ops.aten.sub.Tensor(-128, round_225);  round_225 = None
	        clamp_min_337 = torch.ops.aten.clamp_min.default(sub_5140, -128);  sub_5140 = None
	        clamp_max_224 = torch.ops.aten.clamp_max.default(clamp_min_337, 127);  clamp_min_337 = None
	        _assert_tensor_metadata_1010 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_336, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1010 = None
	        _assert_tensor_metadata_1011 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_224, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1011 = None
	        convert_element_type_672 = torch.ops.prims.convert_element_type.default(clamp_max_224, torch.int8);  clamp_max_224 = None
	        view_1758 = torch.ops.aten.view.default(clamp_min_336, [sym_size_int, 1500, 1])
	        view_1759 = torch.ops.aten.view.default(convert_element_type_672, [sym_size_int, 1500, 1])
	        reciprocal_112 = torch.ops.aten.reciprocal.default(view_1758);  view_1758 = None
	        mul_10904 = torch.ops.aten.mul.Tensor(reciprocal_112, 1.0);  reciprocal_112 = None
	        mul_10907 = torch.ops.aten.mul.Tensor(add_17176, mul_10904);  add_17176 = mul_10904 = None
	        round_226 = torch.ops.aten.round.default(mul_10907);  mul_10907 = None
	        add_17263 = torch.ops.aten.add.Tensor(round_226, view_1759);  round_226 = view_1759 = None
	        clamp_min_338 = torch.ops.aten.clamp_min.default(add_17263, -128);  add_17263 = None
	        clamp_max_225 = torch.ops.aten.clamp_max.default(clamp_min_338, 127);  clamp_min_338 = None
	        _assert_tensor_metadata_1012 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_225, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1012 = None
	        convert_element_type_673 = torch.ops.prims.convert_element_type.default(clamp_max_225, torch.int8);  clamp_max_225 = None
	        view_1762 = torch.ops.aten.view.default(clamp_min_336, [sym_size_int, 1500, 1]);  clamp_min_336 = None
	        view_1763 = torch.ops.aten.view.default(convert_element_type_672, [sym_size_int, 1500, 1]);  convert_element_type_672 = None
	        _assert_tensor_metadata_1013 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_673, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1013 = None
	        convert_element_type_674 = torch.ops.prims.convert_element_type.default(convert_element_type_673, torch.float32);  convert_element_type_673 = None
	        _assert_tensor_metadata_1014 = torch.ops.aten._assert_tensor_metadata.default(view_1763, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1014 = None
	        convert_element_type_675 = torch.ops.prims.convert_element_type.default(view_1763, torch.float32);  view_1763 = None
	        sub_5160 = torch.ops.aten.sub.Tensor(convert_element_type_674, convert_element_type_675);  convert_element_type_674 = convert_element_type_675 = None
	        mul_10929 = torch.ops.aten.mul.Tensor(sub_5160, view_1762);  sub_5160 = view_1762 = None
	        _assert_tensor_metadata_1015 = torch.ops.aten._assert_tensor_metadata.default(mul_10929, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1015 = None
	        view_1765 = torch.ops.aten.view.default(model_audio_tower_layers_18_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_18_fc1_parametrizations_weight_original0 = None
	        view_1766 = torch.ops.aten.view.default(model_audio_tower_layers_18_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_18_fc1_parametrizations_weight_original1 = None
	        view_1767 = torch.ops.aten.view.default(model_audio_tower_layers_18_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_18_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1016 = torch.ops.aten._assert_tensor_metadata.default(view_1765, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1016 = None
	        convert_element_type_676 = torch.ops.prims.convert_element_type.default(view_1765, torch.float32);  view_1765 = None
	        _assert_tensor_metadata_1017 = torch.ops.aten._assert_tensor_metadata.default(view_1767, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1017 = None
	        convert_element_type_677 = torch.ops.prims.convert_element_type.default(view_1767, torch.float32);  view_1767 = None
	        sub_5164 = torch.ops.aten.sub.Tensor(convert_element_type_676, convert_element_type_677);  convert_element_type_676 = convert_element_type_677 = None
	        mul_10934 = torch.ops.aten.mul.Tensor(sub_5164, view_1766);  sub_5164 = view_1766 = None
	        view_1768 = torch.ops.aten.view.default(mul_10934, [5120, 1280]);  mul_10934 = None
	        _assert_tensor_metadata_1018 = torch.ops.aten._assert_tensor_metadata.default(view_1768, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1018 = None
	        mul_10939 = sym_size_int * 1500
	        view_1769 = torch.ops.aten.view.default(mul_10929, [mul_10939, 1280]);  mul_10929 = mul_10939 = None
	        permute_189 = torch.ops.aten.permute.default(view_1768, [1, 0]);  view_1768 = None
	        addmm_93 = torch.ops.aten.addmm.default(model_audio_tower_layers_18_fc1_bias, view_1769, permute_189);  model_audio_tower_layers_18_fc1_bias = view_1769 = permute_189 = None
	        view_1770 = torch.ops.aten.view.default(addmm_93, [sym_size_int, 1500, 5120]);  addmm_93 = None
	        mul_10946 = torch.ops.aten.mul.Tensor(view_1770, 0.5)
	        mul_10947 = torch.ops.aten.mul.Tensor(view_1770, 0.7071067811865476);  view_1770 = None
	        erf_20 = torch.ops.aten.erf.default(mul_10947);  mul_10947 = None
	        add_17322 = torch.ops.aten.add.Tensor(erf_20, 1);  erf_20 = None
	        mul_10948 = torch.ops.aten.mul.Tensor(mul_10946, add_17322);  mul_10946 = add_17322 = None
	        amin_113 = torch.ops.aten.amin.default(mul_10948, [2])
	        amax_113 = torch.ops.aten.amax.default(mul_10948, [2])
	        full_226 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_113 = torch.ops.aten.minimum.default(amin_113, full_226);  amin_113 = full_226 = None
	        full_227 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_113 = torch.ops.aten.maximum.default(amax_113, full_227);  amax_113 = full_227 = None
	        sub_5177 = torch.ops.aten.sub.Tensor(maximum_113, minimum_113);  maximum_113 = None
	        div_226 = torch.ops.aten.div.Tensor(sub_5177, 255.0);  sub_5177 = None
	        clamp_min_339 = torch.ops.aten.clamp_min.default(div_226, 1.1920928955078125e-07);  div_226 = None
	        div_227 = torch.ops.aten.div.Tensor(minimum_113, clamp_min_339);  minimum_113 = None
	        round_227 = torch.ops.aten.round.default(div_227);  div_227 = None
	        sub_5183 = torch.ops.aten.sub.Tensor(-128, round_227);  round_227 = None
	        clamp_min_340 = torch.ops.aten.clamp_min.default(sub_5183, -128);  sub_5183 = None
	        clamp_max_226 = torch.ops.aten.clamp_max.default(clamp_min_340, 127);  clamp_min_340 = None
	        _assert_tensor_metadata_1019 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_339, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1019 = None
	        _assert_tensor_metadata_1020 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_226, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1020 = None
	        convert_element_type_678 = torch.ops.prims.convert_element_type.default(clamp_max_226, torch.int8);  clamp_max_226 = None
	        view_1773 = torch.ops.aten.view.default(clamp_min_339, [sym_size_int, 1500, 1])
	        view_1774 = torch.ops.aten.view.default(convert_element_type_678, [sym_size_int, 1500, 1])
	        reciprocal_113 = torch.ops.aten.reciprocal.default(view_1773);  view_1773 = None
	        mul_10994 = torch.ops.aten.mul.Tensor(reciprocal_113, 1.0);  reciprocal_113 = None
	        mul_10997 = torch.ops.aten.mul.Tensor(mul_10948, mul_10994);  mul_10948 = mul_10994 = None
	        round_228 = torch.ops.aten.round.default(mul_10997);  mul_10997 = None
	        add_17405 = torch.ops.aten.add.Tensor(round_228, view_1774);  round_228 = view_1774 = None
	        clamp_min_341 = torch.ops.aten.clamp_min.default(add_17405, -128);  add_17405 = None
	        clamp_max_227 = torch.ops.aten.clamp_max.default(clamp_min_341, 127);  clamp_min_341 = None
	        _assert_tensor_metadata_1021 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_227, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1021 = None
	        convert_element_type_679 = torch.ops.prims.convert_element_type.default(clamp_max_227, torch.int8);  clamp_max_227 = None
	        view_1777 = torch.ops.aten.view.default(clamp_min_339, [sym_size_int, 1500, 1]);  clamp_min_339 = None
	        view_1778 = torch.ops.aten.view.default(convert_element_type_678, [sym_size_int, 1500, 1]);  convert_element_type_678 = None
	        _assert_tensor_metadata_1022 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_679, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1022 = None
	        convert_element_type_680 = torch.ops.prims.convert_element_type.default(convert_element_type_679, torch.float32);  convert_element_type_679 = None
	        _assert_tensor_metadata_1023 = torch.ops.aten._assert_tensor_metadata.default(view_1778, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1023 = None
	        convert_element_type_681 = torch.ops.prims.convert_element_type.default(view_1778, torch.float32);  view_1778 = None
	        sub_5203 = torch.ops.aten.sub.Tensor(convert_element_type_680, convert_element_type_681);  convert_element_type_680 = convert_element_type_681 = None
	        mul_11019 = torch.ops.aten.mul.Tensor(sub_5203, view_1777);  sub_5203 = view_1777 = None
	        _assert_tensor_metadata_1024 = torch.ops.aten._assert_tensor_metadata.default(mul_11019, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1024 = None
	        view_1780 = torch.ops.aten.view.default(model_audio_tower_layers_18_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_18_fc2_parametrizations_weight_original0 = None
	        view_1781 = torch.ops.aten.view.default(model_audio_tower_layers_18_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_18_fc2_parametrizations_weight_original1 = None
	        view_1782 = torch.ops.aten.view.default(model_audio_tower_layers_18_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_18_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1025 = torch.ops.aten._assert_tensor_metadata.default(view_1780, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1025 = None
	        convert_element_type_682 = torch.ops.prims.convert_element_type.default(view_1780, torch.float32);  view_1780 = None
	        _assert_tensor_metadata_1026 = torch.ops.aten._assert_tensor_metadata.default(view_1782, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1026 = None
	        convert_element_type_683 = torch.ops.prims.convert_element_type.default(view_1782, torch.float32);  view_1782 = None
	        sub_5207 = torch.ops.aten.sub.Tensor(convert_element_type_682, convert_element_type_683);  convert_element_type_682 = convert_element_type_683 = None
	        mul_11024 = torch.ops.aten.mul.Tensor(sub_5207, view_1781);  sub_5207 = view_1781 = None
	        view_1783 = torch.ops.aten.view.default(mul_11024, [1280, 5120]);  mul_11024 = None
	        _assert_tensor_metadata_1027 = torch.ops.aten._assert_tensor_metadata.default(view_1783, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1027 = None
	        mul_11029 = sym_size_int * 1500
	        view_1784 = torch.ops.aten.view.default(mul_11019, [mul_11029, 5120]);  mul_11019 = mul_11029 = None
	        permute_190 = torch.ops.aten.permute.default(view_1783, [1, 0]);  view_1783 = None
	        addmm_94 = torch.ops.aten.addmm.default(model_audio_tower_layers_18_fc2_bias, view_1784, permute_190);  model_audio_tower_layers_18_fc2_bias = view_1784 = permute_190 = None
	        view_1785 = torch.ops.aten.view.default(addmm_94, [sym_size_int, 1500, 1280]);  addmm_94 = None
	        add_17468 = torch.ops.aten.add.Tensor(add_17170, view_1785);  add_17170 = view_1785 = None
	        clone_153 = torch.ops.aten.clone.default(add_17468, memory_format = torch.contiguous_format)
	        var_mean_38 = torch.ops.aten.var_mean.correction(clone_153, [2], correction = 0, keepdim = True)
	        getitem_152 = var_mean_38[0]
	        getitem_153 = var_mean_38[1];  var_mean_38 = None
	        add_17473 = torch.ops.aten.add.Tensor(getitem_152, 1e-05);  getitem_152 = None
	        rsqrt_38 = torch.ops.aten.rsqrt.default(add_17473);  add_17473 = None
	        sub_5213 = torch.ops.aten.sub.Tensor(clone_153, getitem_153);  clone_153 = getitem_153 = None
	        mul_11040 = torch.ops.aten.mul.Tensor(sub_5213, rsqrt_38);  sub_5213 = rsqrt_38 = None
	        mul_11041 = torch.ops.aten.mul.Tensor(mul_11040, model_audio_tower_layers_19_self_attn_layer_norm_weight);  mul_11040 = model_audio_tower_layers_19_self_attn_layer_norm_weight = None
	        add_17474 = torch.ops.aten.add.Tensor(mul_11041, model_audio_tower_layers_19_self_attn_layer_norm_bias);  mul_11041 = model_audio_tower_layers_19_self_attn_layer_norm_bias = None
	        amin_114 = torch.ops.aten.amin.default(add_17474, [2])
	        amax_114 = torch.ops.aten.amax.default(add_17474, [2])
	        full_228 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_114 = torch.ops.aten.minimum.default(amin_114, full_228);  amin_114 = full_228 = None
	        full_229 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_114 = torch.ops.aten.maximum.default(amax_114, full_229);  amax_114 = full_229 = None
	        sub_5224 = torch.ops.aten.sub.Tensor(maximum_114, minimum_114);  maximum_114 = None
	        div_228 = torch.ops.aten.div.Tensor(sub_5224, 255.0);  sub_5224 = None
	        clamp_min_342 = torch.ops.aten.clamp_min.default(div_228, 1.1920928955078125e-07);  div_228 = None
	        div_229 = torch.ops.aten.div.Tensor(minimum_114, clamp_min_342);  minimum_114 = None
	        round_229 = torch.ops.aten.round.default(div_229);  div_229 = None
	        sub_5230 = torch.ops.aten.sub.Tensor(-128, round_229);  round_229 = None
	        clamp_min_343 = torch.ops.aten.clamp_min.default(sub_5230, -128);  sub_5230 = None
	        clamp_max_228 = torch.ops.aten.clamp_max.default(clamp_min_343, 127);  clamp_min_343 = None
	        _assert_tensor_metadata_1028 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_342, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1028 = None
	        _assert_tensor_metadata_1029 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_228, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1029 = None
	        convert_element_type_684 = torch.ops.prims.convert_element_type.default(clamp_max_228, torch.int8);  clamp_max_228 = None
	        view_1788 = torch.ops.aten.view.default(clamp_min_342, [sym_size_int, 1500, 1])
	        view_1789 = torch.ops.aten.view.default(convert_element_type_684, [sym_size_int, 1500, 1])
	        reciprocal_114 = torch.ops.aten.reciprocal.default(view_1788);  view_1788 = None
	        mul_11089 = torch.ops.aten.mul.Tensor(reciprocal_114, 1.0);  reciprocal_114 = None
	        mul_11092 = torch.ops.aten.mul.Tensor(add_17474, mul_11089);  mul_11089 = None
	        round_230 = torch.ops.aten.round.default(mul_11092);  mul_11092 = None
	        add_17561 = torch.ops.aten.add.Tensor(round_230, view_1789);  round_230 = view_1789 = None
	        clamp_min_344 = torch.ops.aten.clamp_min.default(add_17561, -128);  add_17561 = None
	        clamp_max_229 = torch.ops.aten.clamp_max.default(clamp_min_344, 127);  clamp_min_344 = None
	        _assert_tensor_metadata_1030 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_229, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1030 = None
	        convert_element_type_685 = torch.ops.prims.convert_element_type.default(clamp_max_229, torch.int8);  clamp_max_229 = None
	        view_1792 = torch.ops.aten.view.default(clamp_min_342, [sym_size_int, 1500, 1]);  clamp_min_342 = None
	        view_1793 = torch.ops.aten.view.default(convert_element_type_684, [sym_size_int, 1500, 1]);  convert_element_type_684 = None
	        _assert_tensor_metadata_1031 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_685, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1031 = None
	        convert_element_type_686 = torch.ops.prims.convert_element_type.default(convert_element_type_685, torch.float32);  convert_element_type_685 = None
	        _assert_tensor_metadata_1032 = torch.ops.aten._assert_tensor_metadata.default(view_1793, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1032 = None
	        convert_element_type_687 = torch.ops.prims.convert_element_type.default(view_1793, torch.float32);  view_1793 = None
	        sub_5250 = torch.ops.aten.sub.Tensor(convert_element_type_686, convert_element_type_687);  convert_element_type_686 = convert_element_type_687 = None
	        mul_11114 = torch.ops.aten.mul.Tensor(sub_5250, view_1792);  sub_5250 = view_1792 = None
	        _assert_tensor_metadata_1033 = torch.ops.aten._assert_tensor_metadata.default(mul_11114, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1033 = None
	        view_1795 = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1796 = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1797 = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1034 = torch.ops.aten._assert_tensor_metadata.default(view_1795, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1034 = None
	        convert_element_type_688 = torch.ops.prims.convert_element_type.default(view_1795, torch.float32);  view_1795 = None
	        _assert_tensor_metadata_1035 = torch.ops.aten._assert_tensor_metadata.default(view_1797, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1035 = None
	        convert_element_type_689 = torch.ops.prims.convert_element_type.default(view_1797, torch.float32);  view_1797 = None
	        sub_5254 = torch.ops.aten.sub.Tensor(convert_element_type_688, convert_element_type_689);  convert_element_type_688 = convert_element_type_689 = None
	        mul_11119 = torch.ops.aten.mul.Tensor(sub_5254, view_1796);  sub_5254 = view_1796 = None
	        view_1798 = torch.ops.aten.view.default(mul_11119, [1280, 1280]);  mul_11119 = None
	        _assert_tensor_metadata_1036 = torch.ops.aten._assert_tensor_metadata.default(view_1798, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1036 = None
	        mul_11124 = sym_size_int * 1500
	        view_1799 = torch.ops.aten.view.default(mul_11114, [mul_11124, 1280]);  mul_11114 = mul_11124 = None
	        permute_191 = torch.ops.aten.permute.default(view_1798, [1, 0]);  view_1798 = None
	        addmm_95 = torch.ops.aten.addmm.default(model_audio_tower_layers_19_self_attn_q_proj_bias, view_1799, permute_191);  model_audio_tower_layers_19_self_attn_q_proj_bias = view_1799 = permute_191 = None
	        view_1800 = torch.ops.aten.view.default(addmm_95, [sym_size_int, 1500, 1280]);  addmm_95 = None
	        mul_11131 = torch.ops.aten.mul.Tensor(view_1800, 0.125);  view_1800 = None
	        view_1801 = torch.ops.aten.view.default(mul_11131, [sym_size_int, 1500, 20, 64]);  mul_11131 = None
	        permute_192 = torch.ops.aten.permute.default(view_1801, [0, 2, 1, 3]);  view_1801 = None
	        clone_154 = torch.ops.aten.clone.default(permute_192, memory_format = torch.contiguous_format);  permute_192 = None
	        amin_115 = torch.ops.aten.amin.default(add_17474, [2])
	        amax_115 = torch.ops.aten.amax.default(add_17474, [2])
	        full_230 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_115 = torch.ops.aten.minimum.default(amin_115, full_230);  amin_115 = full_230 = None
	        full_231 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_115 = torch.ops.aten.maximum.default(amax_115, full_231);  amax_115 = full_231 = None
	        sub_5269 = torch.ops.aten.sub.Tensor(maximum_115, minimum_115);  maximum_115 = None
	        div_230 = torch.ops.aten.div.Tensor(sub_5269, 255.0);  sub_5269 = None
	        clamp_min_345 = torch.ops.aten.clamp_min.default(div_230, 1.1920928955078125e-07);  div_230 = None
	        div_231 = torch.ops.aten.div.Tensor(minimum_115, clamp_min_345);  minimum_115 = None
	        round_231 = torch.ops.aten.round.default(div_231);  div_231 = None
	        sub_5275 = torch.ops.aten.sub.Tensor(-128, round_231);  round_231 = None
	        clamp_min_346 = torch.ops.aten.clamp_min.default(sub_5275, -128);  sub_5275 = None
	        clamp_max_230 = torch.ops.aten.clamp_max.default(clamp_min_346, 127);  clamp_min_346 = None
	        _assert_tensor_metadata_1037 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_345, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1037 = None
	        _assert_tensor_metadata_1038 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_230, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1038 = None
	        convert_element_type_690 = torch.ops.prims.convert_element_type.default(clamp_max_230, torch.int8);  clamp_max_230 = None
	        view_1804 = torch.ops.aten.view.default(clamp_min_345, [sym_size_int, 1500, 1])
	        view_1805 = torch.ops.aten.view.default(convert_element_type_690, [sym_size_int, 1500, 1])
	        reciprocal_115 = torch.ops.aten.reciprocal.default(view_1804);  view_1804 = None
	        mul_11185 = torch.ops.aten.mul.Tensor(reciprocal_115, 1.0);  reciprocal_115 = None
	        mul_11188 = torch.ops.aten.mul.Tensor(add_17474, mul_11185);  mul_11185 = None
	        round_232 = torch.ops.aten.round.default(mul_11188);  mul_11188 = None
	        add_17713 = torch.ops.aten.add.Tensor(round_232, view_1805);  round_232 = view_1805 = None
	        clamp_min_347 = torch.ops.aten.clamp_min.default(add_17713, -128);  add_17713 = None
	        clamp_max_231 = torch.ops.aten.clamp_max.default(clamp_min_347, 127);  clamp_min_347 = None
	        _assert_tensor_metadata_1039 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_231, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1039 = None
	        convert_element_type_691 = torch.ops.prims.convert_element_type.default(clamp_max_231, torch.int8);  clamp_max_231 = None
	        view_1808 = torch.ops.aten.view.default(clamp_min_345, [sym_size_int, 1500, 1]);  clamp_min_345 = None
	        view_1809 = torch.ops.aten.view.default(convert_element_type_690, [sym_size_int, 1500, 1]);  convert_element_type_690 = None
	        _assert_tensor_metadata_1040 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_691, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1040 = None
	        convert_element_type_692 = torch.ops.prims.convert_element_type.default(convert_element_type_691, torch.float32);  convert_element_type_691 = None
	        _assert_tensor_metadata_1041 = torch.ops.aten._assert_tensor_metadata.default(view_1809, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1041 = None
	        convert_element_type_693 = torch.ops.prims.convert_element_type.default(view_1809, torch.float32);  view_1809 = None
	        sub_5295 = torch.ops.aten.sub.Tensor(convert_element_type_692, convert_element_type_693);  convert_element_type_692 = convert_element_type_693 = None
	        mul_11210 = torch.ops.aten.mul.Tensor(sub_5295, view_1808);  sub_5295 = view_1808 = None
	        _assert_tensor_metadata_1042 = torch.ops.aten._assert_tensor_metadata.default(mul_11210, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1042 = None
	        view_1811 = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1812 = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1813 = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1043 = torch.ops.aten._assert_tensor_metadata.default(view_1811, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1043 = None
	        convert_element_type_694 = torch.ops.prims.convert_element_type.default(view_1811, torch.float32);  view_1811 = None
	        _assert_tensor_metadata_1044 = torch.ops.aten._assert_tensor_metadata.default(view_1813, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1044 = None
	        convert_element_type_695 = torch.ops.prims.convert_element_type.default(view_1813, torch.float32);  view_1813 = None
	        sub_5299 = torch.ops.aten.sub.Tensor(convert_element_type_694, convert_element_type_695);  convert_element_type_694 = convert_element_type_695 = None
	        mul_11215 = torch.ops.aten.mul.Tensor(sub_5299, view_1812);  sub_5299 = view_1812 = None
	        view_1814 = torch.ops.aten.view.default(mul_11215, [1280, 1280]);  mul_11215 = None
	        _assert_tensor_metadata_1045 = torch.ops.aten._assert_tensor_metadata.default(view_1814, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1045 = None
	        permute_193 = torch.ops.aten.permute.default(view_1814, [1, 0]);  view_1814 = None
	        mul_11218 = sym_size_int * 1500
	        view_1815 = torch.ops.aten.view.default(mul_11210, [mul_11218, 1280]);  mul_11210 = mul_11218 = None
	        mm_19 = torch.ops.aten.mm.default(view_1815, permute_193);  view_1815 = permute_193 = None
	        view_1816 = torch.ops.aten.view.default(mm_19, [sym_size_int, 1500, 1280]);  mm_19 = None
	        view_1817 = torch.ops.aten.view.default(view_1816, [sym_size_int, -1, 20, 64]);  view_1816 = None
	        permute_194 = torch.ops.aten.permute.default(view_1817, [0, 2, 1, 3]);  view_1817 = None
	        clone_155 = torch.ops.aten.clone.default(permute_194, memory_format = torch.contiguous_format);  permute_194 = None
	        amin_116 = torch.ops.aten.amin.default(add_17474, [2])
	        amax_116 = torch.ops.aten.amax.default(add_17474, [2])
	        full_232 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_116 = torch.ops.aten.minimum.default(amin_116, full_232);  amin_116 = full_232 = None
	        full_233 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_116 = torch.ops.aten.maximum.default(amax_116, full_233);  amax_116 = full_233 = None
	        sub_5313 = torch.ops.aten.sub.Tensor(maximum_116, minimum_116);  maximum_116 = None
	        div_232 = torch.ops.aten.div.Tensor(sub_5313, 255.0);  sub_5313 = None
	        clamp_min_348 = torch.ops.aten.clamp_min.default(div_232, 1.1920928955078125e-07);  div_232 = None
	        div_233 = torch.ops.aten.div.Tensor(minimum_116, clamp_min_348);  minimum_116 = None
	        round_233 = torch.ops.aten.round.default(div_233);  div_233 = None
	        sub_5319 = torch.ops.aten.sub.Tensor(-128, round_233);  round_233 = None
	        clamp_min_349 = torch.ops.aten.clamp_min.default(sub_5319, -128);  sub_5319 = None
	        clamp_max_232 = torch.ops.aten.clamp_max.default(clamp_min_349, 127);  clamp_min_349 = None
	        _assert_tensor_metadata_1046 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_348, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1046 = None
	        _assert_tensor_metadata_1047 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_232, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1047 = None
	        convert_element_type_696 = torch.ops.prims.convert_element_type.default(clamp_max_232, torch.int8);  clamp_max_232 = None
	        view_1820 = torch.ops.aten.view.default(clamp_min_348, [sym_size_int, 1500, 1])
	        view_1821 = torch.ops.aten.view.default(convert_element_type_696, [sym_size_int, 1500, 1])
	        reciprocal_116 = torch.ops.aten.reciprocal.default(view_1820);  view_1820 = None
	        mul_11284 = torch.ops.aten.mul.Tensor(reciprocal_116, 1.0);  reciprocal_116 = None
	        mul_11287 = torch.ops.aten.mul.Tensor(add_17474, mul_11284);  add_17474 = mul_11284 = None
	        round_234 = torch.ops.aten.round.default(mul_11287);  mul_11287 = None
	        add_17861 = torch.ops.aten.add.Tensor(round_234, view_1821);  round_234 = view_1821 = None
	        clamp_min_350 = torch.ops.aten.clamp_min.default(add_17861, -128);  add_17861 = None
	        clamp_max_233 = torch.ops.aten.clamp_max.default(clamp_min_350, 127);  clamp_min_350 = None
	        _assert_tensor_metadata_1048 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_233, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1048 = None
	        convert_element_type_697 = torch.ops.prims.convert_element_type.default(clamp_max_233, torch.int8);  clamp_max_233 = None
	        view_1824 = torch.ops.aten.view.default(clamp_min_348, [sym_size_int, 1500, 1]);  clamp_min_348 = None
	        view_1825 = torch.ops.aten.view.default(convert_element_type_696, [sym_size_int, 1500, 1]);  convert_element_type_696 = None
	        _assert_tensor_metadata_1049 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_697, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1049 = None
	        convert_element_type_698 = torch.ops.prims.convert_element_type.default(convert_element_type_697, torch.float32);  convert_element_type_697 = None
	        _assert_tensor_metadata_1050 = torch.ops.aten._assert_tensor_metadata.default(view_1825, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1050 = None
	        convert_element_type_699 = torch.ops.prims.convert_element_type.default(view_1825, torch.float32);  view_1825 = None
	        sub_5339 = torch.ops.aten.sub.Tensor(convert_element_type_698, convert_element_type_699);  convert_element_type_698 = convert_element_type_699 = None
	        mul_11309 = torch.ops.aten.mul.Tensor(sub_5339, view_1824);  sub_5339 = view_1824 = None
	        _assert_tensor_metadata_1051 = torch.ops.aten._assert_tensor_metadata.default(mul_11309, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1051 = None
	        view_1827 = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1828 = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1829 = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1052 = torch.ops.aten._assert_tensor_metadata.default(view_1827, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1052 = None
	        convert_element_type_700 = torch.ops.prims.convert_element_type.default(view_1827, torch.float32);  view_1827 = None
	        _assert_tensor_metadata_1053 = torch.ops.aten._assert_tensor_metadata.default(view_1829, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1053 = None
	        convert_element_type_701 = torch.ops.prims.convert_element_type.default(view_1829, torch.float32);  view_1829 = None
	        sub_5343 = torch.ops.aten.sub.Tensor(convert_element_type_700, convert_element_type_701);  convert_element_type_700 = convert_element_type_701 = None
	        mul_11314 = torch.ops.aten.mul.Tensor(sub_5343, view_1828);  sub_5343 = view_1828 = None
	        view_1830 = torch.ops.aten.view.default(mul_11314, [1280, 1280]);  mul_11314 = None
	        _assert_tensor_metadata_1054 = torch.ops.aten._assert_tensor_metadata.default(view_1830, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1054 = None
	        mul_11319 = sym_size_int * 1500
	        view_1831 = torch.ops.aten.view.default(mul_11309, [mul_11319, 1280]);  mul_11309 = mul_11319 = None
	        permute_195 = torch.ops.aten.permute.default(view_1830, [1, 0]);  view_1830 = None
	        addmm_96 = torch.ops.aten.addmm.default(model_audio_tower_layers_19_self_attn_v_proj_bias, view_1831, permute_195);  model_audio_tower_layers_19_self_attn_v_proj_bias = view_1831 = permute_195 = None
	        view_1832 = torch.ops.aten.view.default(addmm_96, [sym_size_int, 1500, 1280]);  addmm_96 = None
	        view_1833 = torch.ops.aten.view.default(view_1832, [sym_size_int, -1, 20, 64]);  view_1832 = None
	        permute_196 = torch.ops.aten.permute.default(view_1833, [0, 2, 1, 3]);  view_1833 = None
	        clone_156 = torch.ops.aten.clone.default(permute_196, memory_format = torch.contiguous_format);  permute_196 = None
	        _scaled_dot_product_efficient_attention_19 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_154, clone_155, clone_156, None, False, scale = 1.0);  clone_154 = clone_155 = clone_156 = None
	        getitem_154 = _scaled_dot_product_efficient_attention_19[0];  _scaled_dot_product_efficient_attention_19 = None
	        permute_197 = torch.ops.aten.permute.default(getitem_154, [0, 2, 1, 3]);  getitem_154 = None
	        view_1834 = torch.ops.aten.view.default(permute_197, [sym_size_int, 1500, -1]);  permute_197 = None
	        amin_117 = torch.ops.aten.amin.default(view_1834, [2])
	        amax_117 = torch.ops.aten.amax.default(view_1834, [2])
	        full_234 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_117 = torch.ops.aten.minimum.default(amin_117, full_234);  amin_117 = full_234 = None
	        full_235 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_117 = torch.ops.aten.maximum.default(amax_117, full_235);  amax_117 = full_235 = None
	        sub_5361 = torch.ops.aten.sub.Tensor(maximum_117, minimum_117);  maximum_117 = None
	        div_234 = torch.ops.aten.div.Tensor(sub_5361, 255.0);  sub_5361 = None
	        clamp_min_351 = torch.ops.aten.clamp_min.default(div_234, 1.1920928955078125e-07);  div_234 = None
	        div_235 = torch.ops.aten.div.Tensor(minimum_117, clamp_min_351);  minimum_117 = None
	        round_235 = torch.ops.aten.round.default(div_235);  div_235 = None
	        sub_5367 = torch.ops.aten.sub.Tensor(-128, round_235);  round_235 = None
	        clamp_min_352 = torch.ops.aten.clamp_min.default(sub_5367, -128);  sub_5367 = None
	        clamp_max_234 = torch.ops.aten.clamp_max.default(clamp_min_352, 127);  clamp_min_352 = None
	        _assert_tensor_metadata_1055 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_351, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1055 = None
	        _assert_tensor_metadata_1056 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_234, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1056 = None
	        convert_element_type_702 = torch.ops.prims.convert_element_type.default(clamp_max_234, torch.int8);  clamp_max_234 = None
	        view_1837 = torch.ops.aten.view.default(clamp_min_351, [sym_size_int, 1500, 1])
	        view_1838 = torch.ops.aten.view.default(convert_element_type_702, [sym_size_int, 1500, 1])
	        reciprocal_117 = torch.ops.aten.reciprocal.default(view_1837);  view_1837 = None
	        mul_11389 = torch.ops.aten.mul.Tensor(reciprocal_117, 1.0);  reciprocal_117 = None
	        mul_11392 = torch.ops.aten.mul.Tensor(view_1834, mul_11389);  view_1834 = mul_11389 = None
	        round_236 = torch.ops.aten.round.default(mul_11392);  mul_11392 = None
	        add_18025 = torch.ops.aten.add.Tensor(round_236, view_1838);  round_236 = view_1838 = None
	        clamp_min_353 = torch.ops.aten.clamp_min.default(add_18025, -128);  add_18025 = None
	        clamp_max_235 = torch.ops.aten.clamp_max.default(clamp_min_353, 127);  clamp_min_353 = None
	        _assert_tensor_metadata_1057 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_235, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1057 = None
	        convert_element_type_703 = torch.ops.prims.convert_element_type.default(clamp_max_235, torch.int8);  clamp_max_235 = None
	        view_1841 = torch.ops.aten.view.default(clamp_min_351, [sym_size_int, 1500, 1]);  clamp_min_351 = None
	        view_1842 = torch.ops.aten.view.default(convert_element_type_702, [sym_size_int, 1500, 1]);  convert_element_type_702 = None
	        _assert_tensor_metadata_1058 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_703, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1058 = None
	        convert_element_type_704 = torch.ops.prims.convert_element_type.default(convert_element_type_703, torch.float32);  convert_element_type_703 = None
	        _assert_tensor_metadata_1059 = torch.ops.aten._assert_tensor_metadata.default(view_1842, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1059 = None
	        convert_element_type_705 = torch.ops.prims.convert_element_type.default(view_1842, torch.float32);  view_1842 = None
	        sub_5387 = torch.ops.aten.sub.Tensor(convert_element_type_704, convert_element_type_705);  convert_element_type_704 = convert_element_type_705 = None
	        mul_11414 = torch.ops.aten.mul.Tensor(sub_5387, view_1841);  sub_5387 = view_1841 = None
	        _assert_tensor_metadata_1060 = torch.ops.aten._assert_tensor_metadata.default(mul_11414, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1060 = None
	        view_1844 = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1845 = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1846 = torch.ops.aten.view.default(model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1061 = torch.ops.aten._assert_tensor_metadata.default(view_1844, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1061 = None
	        convert_element_type_706 = torch.ops.prims.convert_element_type.default(view_1844, torch.float32);  view_1844 = None
	        _assert_tensor_metadata_1062 = torch.ops.aten._assert_tensor_metadata.default(view_1846, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1062 = None
	        convert_element_type_707 = torch.ops.prims.convert_element_type.default(view_1846, torch.float32);  view_1846 = None
	        sub_5391 = torch.ops.aten.sub.Tensor(convert_element_type_706, convert_element_type_707);  convert_element_type_706 = convert_element_type_707 = None
	        mul_11419 = torch.ops.aten.mul.Tensor(sub_5391, view_1845);  sub_5391 = view_1845 = None
	        view_1847 = torch.ops.aten.view.default(mul_11419, [1280, 1280]);  mul_11419 = None
	        _assert_tensor_metadata_1063 = torch.ops.aten._assert_tensor_metadata.default(view_1847, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1063 = None
	        mul_11424 = sym_size_int * 1500
	        view_1848 = torch.ops.aten.view.default(mul_11414, [mul_11424, 1280]);  mul_11414 = mul_11424 = None
	        permute_198 = torch.ops.aten.permute.default(view_1847, [1, 0]);  view_1847 = None
	        addmm_97 = torch.ops.aten.addmm.default(model_audio_tower_layers_19_self_attn_out_proj_bias, view_1848, permute_198);  model_audio_tower_layers_19_self_attn_out_proj_bias = view_1848 = permute_198 = None
	        view_1849 = torch.ops.aten.view.default(addmm_97, [sym_size_int, 1500, 1280]);  addmm_97 = None
	        add_18088 = torch.ops.aten.add.Tensor(add_17468, view_1849);  add_17468 = view_1849 = None
	        clone_158 = torch.ops.aten.clone.default(add_18088, memory_format = torch.contiguous_format)
	        var_mean_39 = torch.ops.aten.var_mean.correction(clone_158, [2], correction = 0, keepdim = True)
	        getitem_158 = var_mean_39[0]
	        getitem_159 = var_mean_39[1];  var_mean_39 = None
	        add_18093 = torch.ops.aten.add.Tensor(getitem_158, 1e-05);  getitem_158 = None
	        rsqrt_39 = torch.ops.aten.rsqrt.default(add_18093);  add_18093 = None
	        sub_5397 = torch.ops.aten.sub.Tensor(clone_158, getitem_159);  clone_158 = getitem_159 = None
	        mul_11435 = torch.ops.aten.mul.Tensor(sub_5397, rsqrt_39);  sub_5397 = rsqrt_39 = None
	        mul_11436 = torch.ops.aten.mul.Tensor(mul_11435, model_audio_tower_layers_19_final_layer_norm_weight);  mul_11435 = model_audio_tower_layers_19_final_layer_norm_weight = None
	        add_18094 = torch.ops.aten.add.Tensor(mul_11436, model_audio_tower_layers_19_final_layer_norm_bias);  mul_11436 = model_audio_tower_layers_19_final_layer_norm_bias = None
	        amin_118 = torch.ops.aten.amin.default(add_18094, [2])
	        amax_118 = torch.ops.aten.amax.default(add_18094, [2])
	        full_236 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_118 = torch.ops.aten.minimum.default(amin_118, full_236);  amin_118 = full_236 = None
	        full_237 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_118 = torch.ops.aten.maximum.default(amax_118, full_237);  amax_118 = full_237 = None
	        sub_5408 = torch.ops.aten.sub.Tensor(maximum_118, minimum_118);  maximum_118 = None
	        div_236 = torch.ops.aten.div.Tensor(sub_5408, 255.0);  sub_5408 = None
	        clamp_min_354 = torch.ops.aten.clamp_min.default(div_236, 1.1920928955078125e-07);  div_236 = None
	        div_237 = torch.ops.aten.div.Tensor(minimum_118, clamp_min_354);  minimum_118 = None
	        round_237 = torch.ops.aten.round.default(div_237);  div_237 = None
	        sub_5414 = torch.ops.aten.sub.Tensor(-128, round_237);  round_237 = None
	        clamp_min_355 = torch.ops.aten.clamp_min.default(sub_5414, -128);  sub_5414 = None
	        clamp_max_236 = torch.ops.aten.clamp_max.default(clamp_min_355, 127);  clamp_min_355 = None
	        _assert_tensor_metadata_1064 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_354, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1064 = None
	        _assert_tensor_metadata_1065 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_236, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1065 = None
	        convert_element_type_708 = torch.ops.prims.convert_element_type.default(clamp_max_236, torch.int8);  clamp_max_236 = None
	        view_1852 = torch.ops.aten.view.default(clamp_min_354, [sym_size_int, 1500, 1])
	        view_1853 = torch.ops.aten.view.default(convert_element_type_708, [sym_size_int, 1500, 1])
	        reciprocal_118 = torch.ops.aten.reciprocal.default(view_1852);  view_1852 = None
	        mul_11484 = torch.ops.aten.mul.Tensor(reciprocal_118, 1.0);  reciprocal_118 = None
	        mul_11487 = torch.ops.aten.mul.Tensor(add_18094, mul_11484);  add_18094 = mul_11484 = None
	        round_238 = torch.ops.aten.round.default(mul_11487);  mul_11487 = None
	        add_18181 = torch.ops.aten.add.Tensor(round_238, view_1853);  round_238 = view_1853 = None
	        clamp_min_356 = torch.ops.aten.clamp_min.default(add_18181, -128);  add_18181 = None
	        clamp_max_237 = torch.ops.aten.clamp_max.default(clamp_min_356, 127);  clamp_min_356 = None
	        _assert_tensor_metadata_1066 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_237, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1066 = None
	        convert_element_type_709 = torch.ops.prims.convert_element_type.default(clamp_max_237, torch.int8);  clamp_max_237 = None
	        view_1856 = torch.ops.aten.view.default(clamp_min_354, [sym_size_int, 1500, 1]);  clamp_min_354 = None
	        view_1857 = torch.ops.aten.view.default(convert_element_type_708, [sym_size_int, 1500, 1]);  convert_element_type_708 = None
	        _assert_tensor_metadata_1067 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_709, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1067 = None
	        convert_element_type_710 = torch.ops.prims.convert_element_type.default(convert_element_type_709, torch.float32);  convert_element_type_709 = None
	        _assert_tensor_metadata_1068 = torch.ops.aten._assert_tensor_metadata.default(view_1857, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1068 = None
	        convert_element_type_711 = torch.ops.prims.convert_element_type.default(view_1857, torch.float32);  view_1857 = None
	        sub_5434 = torch.ops.aten.sub.Tensor(convert_element_type_710, convert_element_type_711);  convert_element_type_710 = convert_element_type_711 = None
	        mul_11509 = torch.ops.aten.mul.Tensor(sub_5434, view_1856);  sub_5434 = view_1856 = None
	        _assert_tensor_metadata_1069 = torch.ops.aten._assert_tensor_metadata.default(mul_11509, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1069 = None
	        view_1859 = torch.ops.aten.view.default(model_audio_tower_layers_19_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_19_fc1_parametrizations_weight_original0 = None
	        view_1860 = torch.ops.aten.view.default(model_audio_tower_layers_19_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_19_fc1_parametrizations_weight_original1 = None
	        view_1861 = torch.ops.aten.view.default(model_audio_tower_layers_19_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_19_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1070 = torch.ops.aten._assert_tensor_metadata.default(view_1859, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1070 = None
	        convert_element_type_712 = torch.ops.prims.convert_element_type.default(view_1859, torch.float32);  view_1859 = None
	        _assert_tensor_metadata_1071 = torch.ops.aten._assert_tensor_metadata.default(view_1861, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1071 = None
	        convert_element_type_713 = torch.ops.prims.convert_element_type.default(view_1861, torch.float32);  view_1861 = None
	        sub_5438 = torch.ops.aten.sub.Tensor(convert_element_type_712, convert_element_type_713);  convert_element_type_712 = convert_element_type_713 = None
	        mul_11514 = torch.ops.aten.mul.Tensor(sub_5438, view_1860);  sub_5438 = view_1860 = None
	        view_1862 = torch.ops.aten.view.default(mul_11514, [5120, 1280]);  mul_11514 = None
	        _assert_tensor_metadata_1072 = torch.ops.aten._assert_tensor_metadata.default(view_1862, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1072 = None
	        mul_11519 = sym_size_int * 1500
	        view_1863 = torch.ops.aten.view.default(mul_11509, [mul_11519, 1280]);  mul_11509 = mul_11519 = None
	        permute_199 = torch.ops.aten.permute.default(view_1862, [1, 0]);  view_1862 = None
	        addmm_98 = torch.ops.aten.addmm.default(model_audio_tower_layers_19_fc1_bias, view_1863, permute_199);  model_audio_tower_layers_19_fc1_bias = view_1863 = permute_199 = None
	        view_1864 = torch.ops.aten.view.default(addmm_98, [sym_size_int, 1500, 5120]);  addmm_98 = None
	        mul_11526 = torch.ops.aten.mul.Tensor(view_1864, 0.5)
	        mul_11527 = torch.ops.aten.mul.Tensor(view_1864, 0.7071067811865476);  view_1864 = None
	        erf_21 = torch.ops.aten.erf.default(mul_11527);  mul_11527 = None
	        add_18240 = torch.ops.aten.add.Tensor(erf_21, 1);  erf_21 = None
	        mul_11528 = torch.ops.aten.mul.Tensor(mul_11526, add_18240);  mul_11526 = add_18240 = None
	        amin_119 = torch.ops.aten.amin.default(mul_11528, [2])
	        amax_119 = torch.ops.aten.amax.default(mul_11528, [2])
	        full_238 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_119 = torch.ops.aten.minimum.default(amin_119, full_238);  amin_119 = full_238 = None
	        full_239 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_119 = torch.ops.aten.maximum.default(amax_119, full_239);  amax_119 = full_239 = None
	        sub_5451 = torch.ops.aten.sub.Tensor(maximum_119, minimum_119);  maximum_119 = None
	        div_238 = torch.ops.aten.div.Tensor(sub_5451, 255.0);  sub_5451 = None
	        clamp_min_357 = torch.ops.aten.clamp_min.default(div_238, 1.1920928955078125e-07);  div_238 = None
	        div_239 = torch.ops.aten.div.Tensor(minimum_119, clamp_min_357);  minimum_119 = None
	        round_239 = torch.ops.aten.round.default(div_239);  div_239 = None
	        sub_5457 = torch.ops.aten.sub.Tensor(-128, round_239);  round_239 = None
	        clamp_min_358 = torch.ops.aten.clamp_min.default(sub_5457, -128);  sub_5457 = None
	        clamp_max_238 = torch.ops.aten.clamp_max.default(clamp_min_358, 127);  clamp_min_358 = None
	        _assert_tensor_metadata_1073 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_357, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1073 = None
	        _assert_tensor_metadata_1074 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_238, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1074 = None
	        convert_element_type_714 = torch.ops.prims.convert_element_type.default(clamp_max_238, torch.int8);  clamp_max_238 = None
	        view_1867 = torch.ops.aten.view.default(clamp_min_357, [sym_size_int, 1500, 1])
	        view_1868 = torch.ops.aten.view.default(convert_element_type_714, [sym_size_int, 1500, 1])
	        reciprocal_119 = torch.ops.aten.reciprocal.default(view_1867);  view_1867 = None
	        mul_11574 = torch.ops.aten.mul.Tensor(reciprocal_119, 1.0);  reciprocal_119 = None
	        mul_11577 = torch.ops.aten.mul.Tensor(mul_11528, mul_11574);  mul_11528 = mul_11574 = None
	        round_240 = torch.ops.aten.round.default(mul_11577);  mul_11577 = None
	        add_18323 = torch.ops.aten.add.Tensor(round_240, view_1868);  round_240 = view_1868 = None
	        clamp_min_359 = torch.ops.aten.clamp_min.default(add_18323, -128);  add_18323 = None
	        clamp_max_239 = torch.ops.aten.clamp_max.default(clamp_min_359, 127);  clamp_min_359 = None
	        _assert_tensor_metadata_1075 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_239, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1075 = None
	        convert_element_type_715 = torch.ops.prims.convert_element_type.default(clamp_max_239, torch.int8);  clamp_max_239 = None
	        view_1871 = torch.ops.aten.view.default(clamp_min_357, [sym_size_int, 1500, 1]);  clamp_min_357 = None
	        view_1872 = torch.ops.aten.view.default(convert_element_type_714, [sym_size_int, 1500, 1]);  convert_element_type_714 = None
	        _assert_tensor_metadata_1076 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_715, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1076 = None
	        convert_element_type_716 = torch.ops.prims.convert_element_type.default(convert_element_type_715, torch.float32);  convert_element_type_715 = None
	        _assert_tensor_metadata_1077 = torch.ops.aten._assert_tensor_metadata.default(view_1872, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1077 = None
	        convert_element_type_717 = torch.ops.prims.convert_element_type.default(view_1872, torch.float32);  view_1872 = None
	        sub_5477 = torch.ops.aten.sub.Tensor(convert_element_type_716, convert_element_type_717);  convert_element_type_716 = convert_element_type_717 = None
	        mul_11599 = torch.ops.aten.mul.Tensor(sub_5477, view_1871);  sub_5477 = view_1871 = None
	        _assert_tensor_metadata_1078 = torch.ops.aten._assert_tensor_metadata.default(mul_11599, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1078 = None
	        view_1874 = torch.ops.aten.view.default(model_audio_tower_layers_19_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_19_fc2_parametrizations_weight_original0 = None
	        view_1875 = torch.ops.aten.view.default(model_audio_tower_layers_19_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_19_fc2_parametrizations_weight_original1 = None
	        view_1876 = torch.ops.aten.view.default(model_audio_tower_layers_19_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_19_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1079 = torch.ops.aten._assert_tensor_metadata.default(view_1874, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1079 = None
	        convert_element_type_718 = torch.ops.prims.convert_element_type.default(view_1874, torch.float32);  view_1874 = None
	        _assert_tensor_metadata_1080 = torch.ops.aten._assert_tensor_metadata.default(view_1876, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1080 = None
	        convert_element_type_719 = torch.ops.prims.convert_element_type.default(view_1876, torch.float32);  view_1876 = None
	        sub_5481 = torch.ops.aten.sub.Tensor(convert_element_type_718, convert_element_type_719);  convert_element_type_718 = convert_element_type_719 = None
	        mul_11604 = torch.ops.aten.mul.Tensor(sub_5481, view_1875);  sub_5481 = view_1875 = None
	        view_1877 = torch.ops.aten.view.default(mul_11604, [1280, 5120]);  mul_11604 = None
	        _assert_tensor_metadata_1081 = torch.ops.aten._assert_tensor_metadata.default(view_1877, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1081 = None
	        mul_11609 = sym_size_int * 1500
	        view_1878 = torch.ops.aten.view.default(mul_11599, [mul_11609, 5120]);  mul_11599 = mul_11609 = None
	        permute_200 = torch.ops.aten.permute.default(view_1877, [1, 0]);  view_1877 = None
	        addmm_99 = torch.ops.aten.addmm.default(model_audio_tower_layers_19_fc2_bias, view_1878, permute_200);  model_audio_tower_layers_19_fc2_bias = view_1878 = permute_200 = None
	        view_1879 = torch.ops.aten.view.default(addmm_99, [sym_size_int, 1500, 1280]);  addmm_99 = None
	        add_18386 = torch.ops.aten.add.Tensor(add_18088, view_1879);  add_18088 = view_1879 = None
	        clone_161 = torch.ops.aten.clone.default(add_18386, memory_format = torch.contiguous_format)
	        var_mean_40 = torch.ops.aten.var_mean.correction(clone_161, [2], correction = 0, keepdim = True)
	        getitem_160 = var_mean_40[0]
	        getitem_161 = var_mean_40[1];  var_mean_40 = None
	        add_18391 = torch.ops.aten.add.Tensor(getitem_160, 1e-05);  getitem_160 = None
	        rsqrt_40 = torch.ops.aten.rsqrt.default(add_18391);  add_18391 = None
	        sub_5487 = torch.ops.aten.sub.Tensor(clone_161, getitem_161);  clone_161 = getitem_161 = None
	        mul_11620 = torch.ops.aten.mul.Tensor(sub_5487, rsqrt_40);  sub_5487 = rsqrt_40 = None
	        mul_11621 = torch.ops.aten.mul.Tensor(mul_11620, model_audio_tower_layers_20_self_attn_layer_norm_weight);  mul_11620 = model_audio_tower_layers_20_self_attn_layer_norm_weight = None
	        add_18392 = torch.ops.aten.add.Tensor(mul_11621, model_audio_tower_layers_20_self_attn_layer_norm_bias);  mul_11621 = model_audio_tower_layers_20_self_attn_layer_norm_bias = None
	        amin_120 = torch.ops.aten.amin.default(add_18392, [2])
	        amax_120 = torch.ops.aten.amax.default(add_18392, [2])
	        full_240 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_120 = torch.ops.aten.minimum.default(amin_120, full_240);  amin_120 = full_240 = None
	        full_241 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_120 = torch.ops.aten.maximum.default(amax_120, full_241);  amax_120 = full_241 = None
	        sub_5498 = torch.ops.aten.sub.Tensor(maximum_120, minimum_120);  maximum_120 = None
	        div_240 = torch.ops.aten.div.Tensor(sub_5498, 255.0);  sub_5498 = None
	        clamp_min_360 = torch.ops.aten.clamp_min.default(div_240, 1.1920928955078125e-07);  div_240 = None
	        div_241 = torch.ops.aten.div.Tensor(minimum_120, clamp_min_360);  minimum_120 = None
	        round_241 = torch.ops.aten.round.default(div_241);  div_241 = None
	        sub_5504 = torch.ops.aten.sub.Tensor(-128, round_241);  round_241 = None
	        clamp_min_361 = torch.ops.aten.clamp_min.default(sub_5504, -128);  sub_5504 = None
	        clamp_max_240 = torch.ops.aten.clamp_max.default(clamp_min_361, 127);  clamp_min_361 = None
	        _assert_tensor_metadata_1082 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_360, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1082 = None
	        _assert_tensor_metadata_1083 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_240, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1083 = None
	        convert_element_type_720 = torch.ops.prims.convert_element_type.default(clamp_max_240, torch.int8);  clamp_max_240 = None
	        view_1882 = torch.ops.aten.view.default(clamp_min_360, [sym_size_int, 1500, 1])
	        view_1883 = torch.ops.aten.view.default(convert_element_type_720, [sym_size_int, 1500, 1])
	        reciprocal_120 = torch.ops.aten.reciprocal.default(view_1882);  view_1882 = None
	        mul_11669 = torch.ops.aten.mul.Tensor(reciprocal_120, 1.0);  reciprocal_120 = None
	        mul_11672 = torch.ops.aten.mul.Tensor(add_18392, mul_11669);  mul_11669 = None
	        round_242 = torch.ops.aten.round.default(mul_11672);  mul_11672 = None
	        add_18479 = torch.ops.aten.add.Tensor(round_242, view_1883);  round_242 = view_1883 = None
	        clamp_min_362 = torch.ops.aten.clamp_min.default(add_18479, -128);  add_18479 = None
	        clamp_max_241 = torch.ops.aten.clamp_max.default(clamp_min_362, 127);  clamp_min_362 = None
	        _assert_tensor_metadata_1084 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_241, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1084 = None
	        convert_element_type_721 = torch.ops.prims.convert_element_type.default(clamp_max_241, torch.int8);  clamp_max_241 = None
	        view_1886 = torch.ops.aten.view.default(clamp_min_360, [sym_size_int, 1500, 1]);  clamp_min_360 = None
	        view_1887 = torch.ops.aten.view.default(convert_element_type_720, [sym_size_int, 1500, 1]);  convert_element_type_720 = None
	        _assert_tensor_metadata_1085 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_721, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1085 = None
	        convert_element_type_722 = torch.ops.prims.convert_element_type.default(convert_element_type_721, torch.float32);  convert_element_type_721 = None
	        _assert_tensor_metadata_1086 = torch.ops.aten._assert_tensor_metadata.default(view_1887, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1086 = None
	        convert_element_type_723 = torch.ops.prims.convert_element_type.default(view_1887, torch.float32);  view_1887 = None
	        sub_5524 = torch.ops.aten.sub.Tensor(convert_element_type_722, convert_element_type_723);  convert_element_type_722 = convert_element_type_723 = None
	        mul_11694 = torch.ops.aten.mul.Tensor(sub_5524, view_1886);  sub_5524 = view_1886 = None
	        _assert_tensor_metadata_1087 = torch.ops.aten._assert_tensor_metadata.default(mul_11694, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1087 = None
	        view_1889 = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1890 = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1891 = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1088 = torch.ops.aten._assert_tensor_metadata.default(view_1889, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1088 = None
	        convert_element_type_724 = torch.ops.prims.convert_element_type.default(view_1889, torch.float32);  view_1889 = None
	        _assert_tensor_metadata_1089 = torch.ops.aten._assert_tensor_metadata.default(view_1891, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1089 = None
	        convert_element_type_725 = torch.ops.prims.convert_element_type.default(view_1891, torch.float32);  view_1891 = None
	        sub_5528 = torch.ops.aten.sub.Tensor(convert_element_type_724, convert_element_type_725);  convert_element_type_724 = convert_element_type_725 = None
	        mul_11699 = torch.ops.aten.mul.Tensor(sub_5528, view_1890);  sub_5528 = view_1890 = None
	        view_1892 = torch.ops.aten.view.default(mul_11699, [1280, 1280]);  mul_11699 = None
	        _assert_tensor_metadata_1090 = torch.ops.aten._assert_tensor_metadata.default(view_1892, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1090 = None
	        mul_11704 = sym_size_int * 1500
	        view_1893 = torch.ops.aten.view.default(mul_11694, [mul_11704, 1280]);  mul_11694 = mul_11704 = None
	        permute_201 = torch.ops.aten.permute.default(view_1892, [1, 0]);  view_1892 = None
	        addmm_100 = torch.ops.aten.addmm.default(model_audio_tower_layers_20_self_attn_q_proj_bias, view_1893, permute_201);  model_audio_tower_layers_20_self_attn_q_proj_bias = view_1893 = permute_201 = None
	        view_1894 = torch.ops.aten.view.default(addmm_100, [sym_size_int, 1500, 1280]);  addmm_100 = None
	        mul_11711 = torch.ops.aten.mul.Tensor(view_1894, 0.125);  view_1894 = None
	        view_1895 = torch.ops.aten.view.default(mul_11711, [sym_size_int, 1500, 20, 64]);  mul_11711 = None
	        permute_202 = torch.ops.aten.permute.default(view_1895, [0, 2, 1, 3]);  view_1895 = None
	        clone_162 = torch.ops.aten.clone.default(permute_202, memory_format = torch.contiguous_format);  permute_202 = None
	        amin_121 = torch.ops.aten.amin.default(add_18392, [2])
	        amax_121 = torch.ops.aten.amax.default(add_18392, [2])
	        full_242 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_121 = torch.ops.aten.minimum.default(amin_121, full_242);  amin_121 = full_242 = None
	        full_243 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_121 = torch.ops.aten.maximum.default(amax_121, full_243);  amax_121 = full_243 = None
	        sub_5543 = torch.ops.aten.sub.Tensor(maximum_121, minimum_121);  maximum_121 = None
	        div_242 = torch.ops.aten.div.Tensor(sub_5543, 255.0);  sub_5543 = None
	        clamp_min_363 = torch.ops.aten.clamp_min.default(div_242, 1.1920928955078125e-07);  div_242 = None
	        div_243 = torch.ops.aten.div.Tensor(minimum_121, clamp_min_363);  minimum_121 = None
	        round_243 = torch.ops.aten.round.default(div_243);  div_243 = None
	        sub_5549 = torch.ops.aten.sub.Tensor(-128, round_243);  round_243 = None
	        clamp_min_364 = torch.ops.aten.clamp_min.default(sub_5549, -128);  sub_5549 = None
	        clamp_max_242 = torch.ops.aten.clamp_max.default(clamp_min_364, 127);  clamp_min_364 = None
	        _assert_tensor_metadata_1091 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_363, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1091 = None
	        _assert_tensor_metadata_1092 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_242, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1092 = None
	        convert_element_type_726 = torch.ops.prims.convert_element_type.default(clamp_max_242, torch.int8);  clamp_max_242 = None
	        view_1898 = torch.ops.aten.view.default(clamp_min_363, [sym_size_int, 1500, 1])
	        view_1899 = torch.ops.aten.view.default(convert_element_type_726, [sym_size_int, 1500, 1])
	        reciprocal_121 = torch.ops.aten.reciprocal.default(view_1898);  view_1898 = None
	        mul_11765 = torch.ops.aten.mul.Tensor(reciprocal_121, 1.0);  reciprocal_121 = None
	        mul_11768 = torch.ops.aten.mul.Tensor(add_18392, mul_11765);  mul_11765 = None
	        round_244 = torch.ops.aten.round.default(mul_11768);  mul_11768 = None
	        add_18631 = torch.ops.aten.add.Tensor(round_244, view_1899);  round_244 = view_1899 = None
	        clamp_min_365 = torch.ops.aten.clamp_min.default(add_18631, -128);  add_18631 = None
	        clamp_max_243 = torch.ops.aten.clamp_max.default(clamp_min_365, 127);  clamp_min_365 = None
	        _assert_tensor_metadata_1093 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_243, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1093 = None
	        convert_element_type_727 = torch.ops.prims.convert_element_type.default(clamp_max_243, torch.int8);  clamp_max_243 = None
	        view_1902 = torch.ops.aten.view.default(clamp_min_363, [sym_size_int, 1500, 1]);  clamp_min_363 = None
	        view_1903 = torch.ops.aten.view.default(convert_element_type_726, [sym_size_int, 1500, 1]);  convert_element_type_726 = None
	        _assert_tensor_metadata_1094 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_727, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1094 = None
	        convert_element_type_728 = torch.ops.prims.convert_element_type.default(convert_element_type_727, torch.float32);  convert_element_type_727 = None
	        _assert_tensor_metadata_1095 = torch.ops.aten._assert_tensor_metadata.default(view_1903, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1095 = None
	        convert_element_type_729 = torch.ops.prims.convert_element_type.default(view_1903, torch.float32);  view_1903 = None
	        sub_5569 = torch.ops.aten.sub.Tensor(convert_element_type_728, convert_element_type_729);  convert_element_type_728 = convert_element_type_729 = None
	        mul_11790 = torch.ops.aten.mul.Tensor(sub_5569, view_1902);  sub_5569 = view_1902 = None
	        _assert_tensor_metadata_1096 = torch.ops.aten._assert_tensor_metadata.default(mul_11790, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1096 = None
	        view_1905 = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1906 = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1907 = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1097 = torch.ops.aten._assert_tensor_metadata.default(view_1905, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1097 = None
	        convert_element_type_730 = torch.ops.prims.convert_element_type.default(view_1905, torch.float32);  view_1905 = None
	        _assert_tensor_metadata_1098 = torch.ops.aten._assert_tensor_metadata.default(view_1907, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1098 = None
	        convert_element_type_731 = torch.ops.prims.convert_element_type.default(view_1907, torch.float32);  view_1907 = None
	        sub_5573 = torch.ops.aten.sub.Tensor(convert_element_type_730, convert_element_type_731);  convert_element_type_730 = convert_element_type_731 = None
	        mul_11795 = torch.ops.aten.mul.Tensor(sub_5573, view_1906);  sub_5573 = view_1906 = None
	        view_1908 = torch.ops.aten.view.default(mul_11795, [1280, 1280]);  mul_11795 = None
	        _assert_tensor_metadata_1099 = torch.ops.aten._assert_tensor_metadata.default(view_1908, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1099 = None
	        permute_203 = torch.ops.aten.permute.default(view_1908, [1, 0]);  view_1908 = None
	        mul_11798 = sym_size_int * 1500
	        view_1909 = torch.ops.aten.view.default(mul_11790, [mul_11798, 1280]);  mul_11790 = mul_11798 = None
	        mm_20 = torch.ops.aten.mm.default(view_1909, permute_203);  view_1909 = permute_203 = None
	        view_1910 = torch.ops.aten.view.default(mm_20, [sym_size_int, 1500, 1280]);  mm_20 = None
	        view_1911 = torch.ops.aten.view.default(view_1910, [sym_size_int, -1, 20, 64]);  view_1910 = None
	        permute_204 = torch.ops.aten.permute.default(view_1911, [0, 2, 1, 3]);  view_1911 = None
	        clone_163 = torch.ops.aten.clone.default(permute_204, memory_format = torch.contiguous_format);  permute_204 = None
	        amin_122 = torch.ops.aten.amin.default(add_18392, [2])
	        amax_122 = torch.ops.aten.amax.default(add_18392, [2])
	        full_244 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_122 = torch.ops.aten.minimum.default(amin_122, full_244);  amin_122 = full_244 = None
	        full_245 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_122 = torch.ops.aten.maximum.default(amax_122, full_245);  amax_122 = full_245 = None
	        sub_5587 = torch.ops.aten.sub.Tensor(maximum_122, minimum_122);  maximum_122 = None
	        div_244 = torch.ops.aten.div.Tensor(sub_5587, 255.0);  sub_5587 = None
	        clamp_min_366 = torch.ops.aten.clamp_min.default(div_244, 1.1920928955078125e-07);  div_244 = None
	        div_245 = torch.ops.aten.div.Tensor(minimum_122, clamp_min_366);  minimum_122 = None
	        round_245 = torch.ops.aten.round.default(div_245);  div_245 = None
	        sub_5593 = torch.ops.aten.sub.Tensor(-128, round_245);  round_245 = None
	        clamp_min_367 = torch.ops.aten.clamp_min.default(sub_5593, -128);  sub_5593 = None
	        clamp_max_244 = torch.ops.aten.clamp_max.default(clamp_min_367, 127);  clamp_min_367 = None
	        _assert_tensor_metadata_1100 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_366, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1100 = None
	        _assert_tensor_metadata_1101 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_244, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1101 = None
	        convert_element_type_732 = torch.ops.prims.convert_element_type.default(clamp_max_244, torch.int8);  clamp_max_244 = None
	        view_1914 = torch.ops.aten.view.default(clamp_min_366, [sym_size_int, 1500, 1])
	        view_1915 = torch.ops.aten.view.default(convert_element_type_732, [sym_size_int, 1500, 1])
	        reciprocal_122 = torch.ops.aten.reciprocal.default(view_1914);  view_1914 = None
	        mul_11864 = torch.ops.aten.mul.Tensor(reciprocal_122, 1.0);  reciprocal_122 = None
	        mul_11867 = torch.ops.aten.mul.Tensor(add_18392, mul_11864);  add_18392 = mul_11864 = None
	        round_246 = torch.ops.aten.round.default(mul_11867);  mul_11867 = None
	        add_18779 = torch.ops.aten.add.Tensor(round_246, view_1915);  round_246 = view_1915 = None
	        clamp_min_368 = torch.ops.aten.clamp_min.default(add_18779, -128);  add_18779 = None
	        clamp_max_245 = torch.ops.aten.clamp_max.default(clamp_min_368, 127);  clamp_min_368 = None
	        _assert_tensor_metadata_1102 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_245, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1102 = None
	        convert_element_type_733 = torch.ops.prims.convert_element_type.default(clamp_max_245, torch.int8);  clamp_max_245 = None
	        view_1918 = torch.ops.aten.view.default(clamp_min_366, [sym_size_int, 1500, 1]);  clamp_min_366 = None
	        view_1919 = torch.ops.aten.view.default(convert_element_type_732, [sym_size_int, 1500, 1]);  convert_element_type_732 = None
	        _assert_tensor_metadata_1103 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_733, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1103 = None
	        convert_element_type_734 = torch.ops.prims.convert_element_type.default(convert_element_type_733, torch.float32);  convert_element_type_733 = None
	        _assert_tensor_metadata_1104 = torch.ops.aten._assert_tensor_metadata.default(view_1919, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1104 = None
	        convert_element_type_735 = torch.ops.prims.convert_element_type.default(view_1919, torch.float32);  view_1919 = None
	        sub_5613 = torch.ops.aten.sub.Tensor(convert_element_type_734, convert_element_type_735);  convert_element_type_734 = convert_element_type_735 = None
	        mul_11889 = torch.ops.aten.mul.Tensor(sub_5613, view_1918);  sub_5613 = view_1918 = None
	        _assert_tensor_metadata_1105 = torch.ops.aten._assert_tensor_metadata.default(mul_11889, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1105 = None
	        view_1921 = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1922 = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1923 = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1106 = torch.ops.aten._assert_tensor_metadata.default(view_1921, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1106 = None
	        convert_element_type_736 = torch.ops.prims.convert_element_type.default(view_1921, torch.float32);  view_1921 = None
	        _assert_tensor_metadata_1107 = torch.ops.aten._assert_tensor_metadata.default(view_1923, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1107 = None
	        convert_element_type_737 = torch.ops.prims.convert_element_type.default(view_1923, torch.float32);  view_1923 = None
	        sub_5617 = torch.ops.aten.sub.Tensor(convert_element_type_736, convert_element_type_737);  convert_element_type_736 = convert_element_type_737 = None
	        mul_11894 = torch.ops.aten.mul.Tensor(sub_5617, view_1922);  sub_5617 = view_1922 = None
	        view_1924 = torch.ops.aten.view.default(mul_11894, [1280, 1280]);  mul_11894 = None
	        _assert_tensor_metadata_1108 = torch.ops.aten._assert_tensor_metadata.default(view_1924, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1108 = None
	        mul_11899 = sym_size_int * 1500
	        view_1925 = torch.ops.aten.view.default(mul_11889, [mul_11899, 1280]);  mul_11889 = mul_11899 = None
	        permute_205 = torch.ops.aten.permute.default(view_1924, [1, 0]);  view_1924 = None
	        addmm_101 = torch.ops.aten.addmm.default(model_audio_tower_layers_20_self_attn_v_proj_bias, view_1925, permute_205);  model_audio_tower_layers_20_self_attn_v_proj_bias = view_1925 = permute_205 = None
	        view_1926 = torch.ops.aten.view.default(addmm_101, [sym_size_int, 1500, 1280]);  addmm_101 = None
	        view_1927 = torch.ops.aten.view.default(view_1926, [sym_size_int, -1, 20, 64]);  view_1926 = None
	        permute_206 = torch.ops.aten.permute.default(view_1927, [0, 2, 1, 3]);  view_1927 = None
	        clone_164 = torch.ops.aten.clone.default(permute_206, memory_format = torch.contiguous_format);  permute_206 = None
	        _scaled_dot_product_efficient_attention_20 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_162, clone_163, clone_164, None, False, scale = 1.0);  clone_162 = clone_163 = clone_164 = None
	        getitem_162 = _scaled_dot_product_efficient_attention_20[0];  _scaled_dot_product_efficient_attention_20 = None
	        permute_207 = torch.ops.aten.permute.default(getitem_162, [0, 2, 1, 3]);  getitem_162 = None
	        view_1928 = torch.ops.aten.view.default(permute_207, [sym_size_int, 1500, -1]);  permute_207 = None
	        amin_123 = torch.ops.aten.amin.default(view_1928, [2])
	        amax_123 = torch.ops.aten.amax.default(view_1928, [2])
	        full_246 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_123 = torch.ops.aten.minimum.default(amin_123, full_246);  amin_123 = full_246 = None
	        full_247 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_123 = torch.ops.aten.maximum.default(amax_123, full_247);  amax_123 = full_247 = None
	        sub_5635 = torch.ops.aten.sub.Tensor(maximum_123, minimum_123);  maximum_123 = None
	        div_246 = torch.ops.aten.div.Tensor(sub_5635, 255.0);  sub_5635 = None
	        clamp_min_369 = torch.ops.aten.clamp_min.default(div_246, 1.1920928955078125e-07);  div_246 = None
	        div_247 = torch.ops.aten.div.Tensor(minimum_123, clamp_min_369);  minimum_123 = None
	        round_247 = torch.ops.aten.round.default(div_247);  div_247 = None
	        sub_5641 = torch.ops.aten.sub.Tensor(-128, round_247);  round_247 = None
	        clamp_min_370 = torch.ops.aten.clamp_min.default(sub_5641, -128);  sub_5641 = None
	        clamp_max_246 = torch.ops.aten.clamp_max.default(clamp_min_370, 127);  clamp_min_370 = None
	        _assert_tensor_metadata_1109 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_369, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1109 = None
	        _assert_tensor_metadata_1110 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_246, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1110 = None
	        convert_element_type_738 = torch.ops.prims.convert_element_type.default(clamp_max_246, torch.int8);  clamp_max_246 = None
	        view_1931 = torch.ops.aten.view.default(clamp_min_369, [sym_size_int, 1500, 1])
	        view_1932 = torch.ops.aten.view.default(convert_element_type_738, [sym_size_int, 1500, 1])
	        reciprocal_123 = torch.ops.aten.reciprocal.default(view_1931);  view_1931 = None
	        mul_11969 = torch.ops.aten.mul.Tensor(reciprocal_123, 1.0);  reciprocal_123 = None
	        mul_11972 = torch.ops.aten.mul.Tensor(view_1928, mul_11969);  view_1928 = mul_11969 = None
	        round_248 = torch.ops.aten.round.default(mul_11972);  mul_11972 = None
	        add_18943 = torch.ops.aten.add.Tensor(round_248, view_1932);  round_248 = view_1932 = None
	        clamp_min_371 = torch.ops.aten.clamp_min.default(add_18943, -128);  add_18943 = None
	        clamp_max_247 = torch.ops.aten.clamp_max.default(clamp_min_371, 127);  clamp_min_371 = None
	        _assert_tensor_metadata_1111 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_247, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1111 = None
	        convert_element_type_739 = torch.ops.prims.convert_element_type.default(clamp_max_247, torch.int8);  clamp_max_247 = None
	        view_1935 = torch.ops.aten.view.default(clamp_min_369, [sym_size_int, 1500, 1]);  clamp_min_369 = None
	        view_1936 = torch.ops.aten.view.default(convert_element_type_738, [sym_size_int, 1500, 1]);  convert_element_type_738 = None
	        _assert_tensor_metadata_1112 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_739, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1112 = None
	        convert_element_type_740 = torch.ops.prims.convert_element_type.default(convert_element_type_739, torch.float32);  convert_element_type_739 = None
	        _assert_tensor_metadata_1113 = torch.ops.aten._assert_tensor_metadata.default(view_1936, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1113 = None
	        convert_element_type_741 = torch.ops.prims.convert_element_type.default(view_1936, torch.float32);  view_1936 = None
	        sub_5661 = torch.ops.aten.sub.Tensor(convert_element_type_740, convert_element_type_741);  convert_element_type_740 = convert_element_type_741 = None
	        mul_11994 = torch.ops.aten.mul.Tensor(sub_5661, view_1935);  sub_5661 = view_1935 = None
	        _assert_tensor_metadata_1114 = torch.ops.aten._assert_tensor_metadata.default(mul_11994, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1114 = None
	        view_1938 = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1939 = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1940 = torch.ops.aten.view.default(model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1115 = torch.ops.aten._assert_tensor_metadata.default(view_1938, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1115 = None
	        convert_element_type_742 = torch.ops.prims.convert_element_type.default(view_1938, torch.float32);  view_1938 = None
	        _assert_tensor_metadata_1116 = torch.ops.aten._assert_tensor_metadata.default(view_1940, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1116 = None
	        convert_element_type_743 = torch.ops.prims.convert_element_type.default(view_1940, torch.float32);  view_1940 = None
	        sub_5665 = torch.ops.aten.sub.Tensor(convert_element_type_742, convert_element_type_743);  convert_element_type_742 = convert_element_type_743 = None
	        mul_11999 = torch.ops.aten.mul.Tensor(sub_5665, view_1939);  sub_5665 = view_1939 = None
	        view_1941 = torch.ops.aten.view.default(mul_11999, [1280, 1280]);  mul_11999 = None
	        _assert_tensor_metadata_1117 = torch.ops.aten._assert_tensor_metadata.default(view_1941, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1117 = None
	        mul_12004 = sym_size_int * 1500
	        view_1942 = torch.ops.aten.view.default(mul_11994, [mul_12004, 1280]);  mul_11994 = mul_12004 = None
	        permute_208 = torch.ops.aten.permute.default(view_1941, [1, 0]);  view_1941 = None
	        addmm_102 = torch.ops.aten.addmm.default(model_audio_tower_layers_20_self_attn_out_proj_bias, view_1942, permute_208);  model_audio_tower_layers_20_self_attn_out_proj_bias = view_1942 = permute_208 = None
	        view_1943 = torch.ops.aten.view.default(addmm_102, [sym_size_int, 1500, 1280]);  addmm_102 = None
	        add_19006 = torch.ops.aten.add.Tensor(add_18386, view_1943);  add_18386 = view_1943 = None
	        clone_166 = torch.ops.aten.clone.default(add_19006, memory_format = torch.contiguous_format)
	        var_mean_41 = torch.ops.aten.var_mean.correction(clone_166, [2], correction = 0, keepdim = True)
	        getitem_166 = var_mean_41[0]
	        getitem_167 = var_mean_41[1];  var_mean_41 = None
	        add_19011 = torch.ops.aten.add.Tensor(getitem_166, 1e-05);  getitem_166 = None
	        rsqrt_41 = torch.ops.aten.rsqrt.default(add_19011);  add_19011 = None
	        sub_5671 = torch.ops.aten.sub.Tensor(clone_166, getitem_167);  clone_166 = getitem_167 = None
	        mul_12015 = torch.ops.aten.mul.Tensor(sub_5671, rsqrt_41);  sub_5671 = rsqrt_41 = None
	        mul_12016 = torch.ops.aten.mul.Tensor(mul_12015, model_audio_tower_layers_20_final_layer_norm_weight);  mul_12015 = model_audio_tower_layers_20_final_layer_norm_weight = None
	        add_19012 = torch.ops.aten.add.Tensor(mul_12016, model_audio_tower_layers_20_final_layer_norm_bias);  mul_12016 = model_audio_tower_layers_20_final_layer_norm_bias = None
	        amin_124 = torch.ops.aten.amin.default(add_19012, [2])
	        amax_124 = torch.ops.aten.amax.default(add_19012, [2])
	        full_248 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_124 = torch.ops.aten.minimum.default(amin_124, full_248);  amin_124 = full_248 = None
	        full_249 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_124 = torch.ops.aten.maximum.default(amax_124, full_249);  amax_124 = full_249 = None
	        sub_5682 = torch.ops.aten.sub.Tensor(maximum_124, minimum_124);  maximum_124 = None
	        div_248 = torch.ops.aten.div.Tensor(sub_5682, 255.0);  sub_5682 = None
	        clamp_min_372 = torch.ops.aten.clamp_min.default(div_248, 1.1920928955078125e-07);  div_248 = None
	        div_249 = torch.ops.aten.div.Tensor(minimum_124, clamp_min_372);  minimum_124 = None
	        round_249 = torch.ops.aten.round.default(div_249);  div_249 = None
	        sub_5688 = torch.ops.aten.sub.Tensor(-128, round_249);  round_249 = None
	        clamp_min_373 = torch.ops.aten.clamp_min.default(sub_5688, -128);  sub_5688 = None
	        clamp_max_248 = torch.ops.aten.clamp_max.default(clamp_min_373, 127);  clamp_min_373 = None
	        _assert_tensor_metadata_1118 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_372, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1118 = None
	        _assert_tensor_metadata_1119 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_248, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1119 = None
	        convert_element_type_744 = torch.ops.prims.convert_element_type.default(clamp_max_248, torch.int8);  clamp_max_248 = None
	        view_1946 = torch.ops.aten.view.default(clamp_min_372, [sym_size_int, 1500, 1])
	        view_1947 = torch.ops.aten.view.default(convert_element_type_744, [sym_size_int, 1500, 1])
	        reciprocal_124 = torch.ops.aten.reciprocal.default(view_1946);  view_1946 = None
	        mul_12064 = torch.ops.aten.mul.Tensor(reciprocal_124, 1.0);  reciprocal_124 = None
	        mul_12067 = torch.ops.aten.mul.Tensor(add_19012, mul_12064);  add_19012 = mul_12064 = None
	        round_250 = torch.ops.aten.round.default(mul_12067);  mul_12067 = None
	        add_19099 = torch.ops.aten.add.Tensor(round_250, view_1947);  round_250 = view_1947 = None
	        clamp_min_374 = torch.ops.aten.clamp_min.default(add_19099, -128);  add_19099 = None
	        clamp_max_249 = torch.ops.aten.clamp_max.default(clamp_min_374, 127);  clamp_min_374 = None
	        _assert_tensor_metadata_1120 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1120 = None
	        convert_element_type_745 = torch.ops.prims.convert_element_type.default(clamp_max_249, torch.int8);  clamp_max_249 = None
	        view_1950 = torch.ops.aten.view.default(clamp_min_372, [sym_size_int, 1500, 1]);  clamp_min_372 = None
	        view_1951 = torch.ops.aten.view.default(convert_element_type_744, [sym_size_int, 1500, 1]);  convert_element_type_744 = None
	        _assert_tensor_metadata_1121 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_745, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1121 = None
	        convert_element_type_746 = torch.ops.prims.convert_element_type.default(convert_element_type_745, torch.float32);  convert_element_type_745 = None
	        _assert_tensor_metadata_1122 = torch.ops.aten._assert_tensor_metadata.default(view_1951, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1122 = None
	        convert_element_type_747 = torch.ops.prims.convert_element_type.default(view_1951, torch.float32);  view_1951 = None
	        sub_5708 = torch.ops.aten.sub.Tensor(convert_element_type_746, convert_element_type_747);  convert_element_type_746 = convert_element_type_747 = None
	        mul_12089 = torch.ops.aten.mul.Tensor(sub_5708, view_1950);  sub_5708 = view_1950 = None
	        _assert_tensor_metadata_1123 = torch.ops.aten._assert_tensor_metadata.default(mul_12089, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1123 = None
	        view_1953 = torch.ops.aten.view.default(model_audio_tower_layers_20_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_20_fc1_parametrizations_weight_original0 = None
	        view_1954 = torch.ops.aten.view.default(model_audio_tower_layers_20_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_20_fc1_parametrizations_weight_original1 = None
	        view_1955 = torch.ops.aten.view.default(model_audio_tower_layers_20_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_20_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1124 = torch.ops.aten._assert_tensor_metadata.default(view_1953, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1124 = None
	        convert_element_type_748 = torch.ops.prims.convert_element_type.default(view_1953, torch.float32);  view_1953 = None
	        _assert_tensor_metadata_1125 = torch.ops.aten._assert_tensor_metadata.default(view_1955, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1125 = None
	        convert_element_type_749 = torch.ops.prims.convert_element_type.default(view_1955, torch.float32);  view_1955 = None
	        sub_5712 = torch.ops.aten.sub.Tensor(convert_element_type_748, convert_element_type_749);  convert_element_type_748 = convert_element_type_749 = None
	        mul_12094 = torch.ops.aten.mul.Tensor(sub_5712, view_1954);  sub_5712 = view_1954 = None
	        view_1956 = torch.ops.aten.view.default(mul_12094, [5120, 1280]);  mul_12094 = None
	        _assert_tensor_metadata_1126 = torch.ops.aten._assert_tensor_metadata.default(view_1956, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1126 = None
	        mul_12099 = sym_size_int * 1500
	        view_1957 = torch.ops.aten.view.default(mul_12089, [mul_12099, 1280]);  mul_12089 = mul_12099 = None
	        permute_209 = torch.ops.aten.permute.default(view_1956, [1, 0]);  view_1956 = None
	        addmm_103 = torch.ops.aten.addmm.default(model_audio_tower_layers_20_fc1_bias, view_1957, permute_209);  model_audio_tower_layers_20_fc1_bias = view_1957 = permute_209 = None
	        view_1958 = torch.ops.aten.view.default(addmm_103, [sym_size_int, 1500, 5120]);  addmm_103 = None
	        mul_12106 = torch.ops.aten.mul.Tensor(view_1958, 0.5)
	        mul_12107 = torch.ops.aten.mul.Tensor(view_1958, 0.7071067811865476);  view_1958 = None
	        erf_22 = torch.ops.aten.erf.default(mul_12107);  mul_12107 = None
	        add_19158 = torch.ops.aten.add.Tensor(erf_22, 1);  erf_22 = None
	        mul_12108 = torch.ops.aten.mul.Tensor(mul_12106, add_19158);  mul_12106 = add_19158 = None
	        amin_125 = torch.ops.aten.amin.default(mul_12108, [2])
	        amax_125 = torch.ops.aten.amax.default(mul_12108, [2])
	        full_250 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_125 = torch.ops.aten.minimum.default(amin_125, full_250);  amin_125 = full_250 = None
	        full_251 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_125 = torch.ops.aten.maximum.default(amax_125, full_251);  amax_125 = full_251 = None
	        sub_5725 = torch.ops.aten.sub.Tensor(maximum_125, minimum_125);  maximum_125 = None
	        div_250 = torch.ops.aten.div.Tensor(sub_5725, 255.0);  sub_5725 = None
	        clamp_min_375 = torch.ops.aten.clamp_min.default(div_250, 1.1920928955078125e-07);  div_250 = None
	        div_251 = torch.ops.aten.div.Tensor(minimum_125, clamp_min_375);  minimum_125 = None
	        round_251 = torch.ops.aten.round.default(div_251);  div_251 = None
	        sub_5731 = torch.ops.aten.sub.Tensor(-128, round_251);  round_251 = None
	        clamp_min_376 = torch.ops.aten.clamp_min.default(sub_5731, -128);  sub_5731 = None
	        clamp_max_250 = torch.ops.aten.clamp_max.default(clamp_min_376, 127);  clamp_min_376 = None
	        _assert_tensor_metadata_1127 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_375, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1127 = None
	        _assert_tensor_metadata_1128 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_250, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1128 = None
	        convert_element_type_750 = torch.ops.prims.convert_element_type.default(clamp_max_250, torch.int8);  clamp_max_250 = None
	        view_1961 = torch.ops.aten.view.default(clamp_min_375, [sym_size_int, 1500, 1])
	        view_1962 = torch.ops.aten.view.default(convert_element_type_750, [sym_size_int, 1500, 1])
	        reciprocal_125 = torch.ops.aten.reciprocal.default(view_1961);  view_1961 = None
	        mul_12154 = torch.ops.aten.mul.Tensor(reciprocal_125, 1.0);  reciprocal_125 = None
	        mul_12157 = torch.ops.aten.mul.Tensor(mul_12108, mul_12154);  mul_12108 = mul_12154 = None
	        round_252 = torch.ops.aten.round.default(mul_12157);  mul_12157 = None
	        add_19241 = torch.ops.aten.add.Tensor(round_252, view_1962);  round_252 = view_1962 = None
	        clamp_min_377 = torch.ops.aten.clamp_min.default(add_19241, -128);  add_19241 = None
	        clamp_max_251 = torch.ops.aten.clamp_max.default(clamp_min_377, 127);  clamp_min_377 = None
	        _assert_tensor_metadata_1129 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_251, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1129 = None
	        convert_element_type_751 = torch.ops.prims.convert_element_type.default(clamp_max_251, torch.int8);  clamp_max_251 = None
	        view_1965 = torch.ops.aten.view.default(clamp_min_375, [sym_size_int, 1500, 1]);  clamp_min_375 = None
	        view_1966 = torch.ops.aten.view.default(convert_element_type_750, [sym_size_int, 1500, 1]);  convert_element_type_750 = None
	        _assert_tensor_metadata_1130 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_751, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1130 = None
	        convert_element_type_752 = torch.ops.prims.convert_element_type.default(convert_element_type_751, torch.float32);  convert_element_type_751 = None
	        _assert_tensor_metadata_1131 = torch.ops.aten._assert_tensor_metadata.default(view_1966, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1131 = None
	        convert_element_type_753 = torch.ops.prims.convert_element_type.default(view_1966, torch.float32);  view_1966 = None
	        sub_5751 = torch.ops.aten.sub.Tensor(convert_element_type_752, convert_element_type_753);  convert_element_type_752 = convert_element_type_753 = None
	        mul_12179 = torch.ops.aten.mul.Tensor(sub_5751, view_1965);  sub_5751 = view_1965 = None
	        _assert_tensor_metadata_1132 = torch.ops.aten._assert_tensor_metadata.default(mul_12179, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1132 = None
	        view_1968 = torch.ops.aten.view.default(model_audio_tower_layers_20_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_20_fc2_parametrizations_weight_original0 = None
	        view_1969 = torch.ops.aten.view.default(model_audio_tower_layers_20_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_20_fc2_parametrizations_weight_original1 = None
	        view_1970 = torch.ops.aten.view.default(model_audio_tower_layers_20_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_20_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1133 = torch.ops.aten._assert_tensor_metadata.default(view_1968, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1133 = None
	        convert_element_type_754 = torch.ops.prims.convert_element_type.default(view_1968, torch.float32);  view_1968 = None
	        _assert_tensor_metadata_1134 = torch.ops.aten._assert_tensor_metadata.default(view_1970, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1134 = None
	        convert_element_type_755 = torch.ops.prims.convert_element_type.default(view_1970, torch.float32);  view_1970 = None
	        sub_5755 = torch.ops.aten.sub.Tensor(convert_element_type_754, convert_element_type_755);  convert_element_type_754 = convert_element_type_755 = None
	        mul_12184 = torch.ops.aten.mul.Tensor(sub_5755, view_1969);  sub_5755 = view_1969 = None
	        view_1971 = torch.ops.aten.view.default(mul_12184, [1280, 5120]);  mul_12184 = None
	        _assert_tensor_metadata_1135 = torch.ops.aten._assert_tensor_metadata.default(view_1971, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1135 = None
	        mul_12189 = sym_size_int * 1500
	        view_1972 = torch.ops.aten.view.default(mul_12179, [mul_12189, 5120]);  mul_12179 = mul_12189 = None
	        permute_210 = torch.ops.aten.permute.default(view_1971, [1, 0]);  view_1971 = None
	        addmm_104 = torch.ops.aten.addmm.default(model_audio_tower_layers_20_fc2_bias, view_1972, permute_210);  model_audio_tower_layers_20_fc2_bias = view_1972 = permute_210 = None
	        view_1973 = torch.ops.aten.view.default(addmm_104, [sym_size_int, 1500, 1280]);  addmm_104 = None
	        add_19304 = torch.ops.aten.add.Tensor(add_19006, view_1973);  add_19006 = view_1973 = None
	        clone_169 = torch.ops.aten.clone.default(add_19304, memory_format = torch.contiguous_format)
	        var_mean_42 = torch.ops.aten.var_mean.correction(clone_169, [2], correction = 0, keepdim = True)
	        getitem_168 = var_mean_42[0]
	        getitem_169 = var_mean_42[1];  var_mean_42 = None
	        add_19309 = torch.ops.aten.add.Tensor(getitem_168, 1e-05);  getitem_168 = None
	        rsqrt_42 = torch.ops.aten.rsqrt.default(add_19309);  add_19309 = None
	        sub_5761 = torch.ops.aten.sub.Tensor(clone_169, getitem_169);  clone_169 = getitem_169 = None
	        mul_12200 = torch.ops.aten.mul.Tensor(sub_5761, rsqrt_42);  sub_5761 = rsqrt_42 = None
	        mul_12201 = torch.ops.aten.mul.Tensor(mul_12200, model_audio_tower_layers_21_self_attn_layer_norm_weight);  mul_12200 = model_audio_tower_layers_21_self_attn_layer_norm_weight = None
	        add_19310 = torch.ops.aten.add.Tensor(mul_12201, model_audio_tower_layers_21_self_attn_layer_norm_bias);  mul_12201 = model_audio_tower_layers_21_self_attn_layer_norm_bias = None
	        amin_126 = torch.ops.aten.amin.default(add_19310, [2])
	        amax_126 = torch.ops.aten.amax.default(add_19310, [2])
	        full_252 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_126 = torch.ops.aten.minimum.default(amin_126, full_252);  amin_126 = full_252 = None
	        full_253 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_126 = torch.ops.aten.maximum.default(amax_126, full_253);  amax_126 = full_253 = None
	        sub_5772 = torch.ops.aten.sub.Tensor(maximum_126, minimum_126);  maximum_126 = None
	        div_252 = torch.ops.aten.div.Tensor(sub_5772, 255.0);  sub_5772 = None
	        clamp_min_378 = torch.ops.aten.clamp_min.default(div_252, 1.1920928955078125e-07);  div_252 = None
	        div_253 = torch.ops.aten.div.Tensor(minimum_126, clamp_min_378);  minimum_126 = None
	        round_253 = torch.ops.aten.round.default(div_253);  div_253 = None
	        sub_5778 = torch.ops.aten.sub.Tensor(-128, round_253);  round_253 = None
	        clamp_min_379 = torch.ops.aten.clamp_min.default(sub_5778, -128);  sub_5778 = None
	        clamp_max_252 = torch.ops.aten.clamp_max.default(clamp_min_379, 127);  clamp_min_379 = None
	        _assert_tensor_metadata_1136 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_378, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1136 = None
	        _assert_tensor_metadata_1137 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_252, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1137 = None
	        convert_element_type_756 = torch.ops.prims.convert_element_type.default(clamp_max_252, torch.int8);  clamp_max_252 = None
	        view_1976 = torch.ops.aten.view.default(clamp_min_378, [sym_size_int, 1500, 1])
	        view_1977 = torch.ops.aten.view.default(convert_element_type_756, [sym_size_int, 1500, 1])
	        reciprocal_126 = torch.ops.aten.reciprocal.default(view_1976);  view_1976 = None
	        mul_12249 = torch.ops.aten.mul.Tensor(reciprocal_126, 1.0);  reciprocal_126 = None
	        mul_12252 = torch.ops.aten.mul.Tensor(add_19310, mul_12249);  mul_12249 = None
	        round_254 = torch.ops.aten.round.default(mul_12252);  mul_12252 = None
	        add_19397 = torch.ops.aten.add.Tensor(round_254, view_1977);  round_254 = view_1977 = None
	        clamp_min_380 = torch.ops.aten.clamp_min.default(add_19397, -128);  add_19397 = None
	        clamp_max_253 = torch.ops.aten.clamp_max.default(clamp_min_380, 127);  clamp_min_380 = None
	        _assert_tensor_metadata_1138 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_253, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1138 = None
	        convert_element_type_757 = torch.ops.prims.convert_element_type.default(clamp_max_253, torch.int8);  clamp_max_253 = None
	        view_1980 = torch.ops.aten.view.default(clamp_min_378, [sym_size_int, 1500, 1]);  clamp_min_378 = None
	        view_1981 = torch.ops.aten.view.default(convert_element_type_756, [sym_size_int, 1500, 1]);  convert_element_type_756 = None
	        _assert_tensor_metadata_1139 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_757, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1139 = None
	        convert_element_type_758 = torch.ops.prims.convert_element_type.default(convert_element_type_757, torch.float32);  convert_element_type_757 = None
	        _assert_tensor_metadata_1140 = torch.ops.aten._assert_tensor_metadata.default(view_1981, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1140 = None
	        convert_element_type_759 = torch.ops.prims.convert_element_type.default(view_1981, torch.float32);  view_1981 = None
	        sub_5798 = torch.ops.aten.sub.Tensor(convert_element_type_758, convert_element_type_759);  convert_element_type_758 = convert_element_type_759 = None
	        mul_12274 = torch.ops.aten.mul.Tensor(sub_5798, view_1980);  sub_5798 = view_1980 = None
	        _assert_tensor_metadata_1141 = torch.ops.aten._assert_tensor_metadata.default(mul_12274, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1141 = None
	        view_1983 = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1984 = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1985 = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1142 = torch.ops.aten._assert_tensor_metadata.default(view_1983, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1142 = None
	        convert_element_type_760 = torch.ops.prims.convert_element_type.default(view_1983, torch.float32);  view_1983 = None
	        _assert_tensor_metadata_1143 = torch.ops.aten._assert_tensor_metadata.default(view_1985, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1143 = None
	        convert_element_type_761 = torch.ops.prims.convert_element_type.default(view_1985, torch.float32);  view_1985 = None
	        sub_5802 = torch.ops.aten.sub.Tensor(convert_element_type_760, convert_element_type_761);  convert_element_type_760 = convert_element_type_761 = None
	        mul_12279 = torch.ops.aten.mul.Tensor(sub_5802, view_1984);  sub_5802 = view_1984 = None
	        view_1986 = torch.ops.aten.view.default(mul_12279, [1280, 1280]);  mul_12279 = None
	        _assert_tensor_metadata_1144 = torch.ops.aten._assert_tensor_metadata.default(view_1986, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1144 = None
	        mul_12284 = sym_size_int * 1500
	        view_1987 = torch.ops.aten.view.default(mul_12274, [mul_12284, 1280]);  mul_12274 = mul_12284 = None
	        permute_211 = torch.ops.aten.permute.default(view_1986, [1, 0]);  view_1986 = None
	        addmm_105 = torch.ops.aten.addmm.default(model_audio_tower_layers_21_self_attn_q_proj_bias, view_1987, permute_211);  model_audio_tower_layers_21_self_attn_q_proj_bias = view_1987 = permute_211 = None
	        view_1988 = torch.ops.aten.view.default(addmm_105, [sym_size_int, 1500, 1280]);  addmm_105 = None
	        mul_12291 = torch.ops.aten.mul.Tensor(view_1988, 0.125);  view_1988 = None
	        view_1989 = torch.ops.aten.view.default(mul_12291, [sym_size_int, 1500, 20, 64]);  mul_12291 = None
	        permute_212 = torch.ops.aten.permute.default(view_1989, [0, 2, 1, 3]);  view_1989 = None
	        clone_170 = torch.ops.aten.clone.default(permute_212, memory_format = torch.contiguous_format);  permute_212 = None
	        amin_127 = torch.ops.aten.amin.default(add_19310, [2])
	        amax_127 = torch.ops.aten.amax.default(add_19310, [2])
	        full_254 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_127 = torch.ops.aten.minimum.default(amin_127, full_254);  amin_127 = full_254 = None
	        full_255 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_127 = torch.ops.aten.maximum.default(amax_127, full_255);  amax_127 = full_255 = None
	        sub_5817 = torch.ops.aten.sub.Tensor(maximum_127, minimum_127);  maximum_127 = None
	        div_254 = torch.ops.aten.div.Tensor(sub_5817, 255.0);  sub_5817 = None
	        clamp_min_381 = torch.ops.aten.clamp_min.default(div_254, 1.1920928955078125e-07);  div_254 = None
	        div_255 = torch.ops.aten.div.Tensor(minimum_127, clamp_min_381);  minimum_127 = None
	        round_255 = torch.ops.aten.round.default(div_255);  div_255 = None
	        sub_5823 = torch.ops.aten.sub.Tensor(-128, round_255);  round_255 = None
	        clamp_min_382 = torch.ops.aten.clamp_min.default(sub_5823, -128);  sub_5823 = None
	        clamp_max_254 = torch.ops.aten.clamp_max.default(clamp_min_382, 127);  clamp_min_382 = None
	        _assert_tensor_metadata_1145 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_381, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1145 = None
	        _assert_tensor_metadata_1146 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_254, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1146 = None
	        convert_element_type_762 = torch.ops.prims.convert_element_type.default(clamp_max_254, torch.int8);  clamp_max_254 = None
	        view_1992 = torch.ops.aten.view.default(clamp_min_381, [sym_size_int, 1500, 1])
	        view_1993 = torch.ops.aten.view.default(convert_element_type_762, [sym_size_int, 1500, 1])
	        reciprocal_127 = torch.ops.aten.reciprocal.default(view_1992);  view_1992 = None
	        mul_12345 = torch.ops.aten.mul.Tensor(reciprocal_127, 1.0);  reciprocal_127 = None
	        mul_12348 = torch.ops.aten.mul.Tensor(add_19310, mul_12345);  mul_12345 = None
	        round_256 = torch.ops.aten.round.default(mul_12348);  mul_12348 = None
	        add_19549 = torch.ops.aten.add.Tensor(round_256, view_1993);  round_256 = view_1993 = None
	        clamp_min_383 = torch.ops.aten.clamp_min.default(add_19549, -128);  add_19549 = None
	        clamp_max_255 = torch.ops.aten.clamp_max.default(clamp_min_383, 127);  clamp_min_383 = None
	        _assert_tensor_metadata_1147 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_255, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1147 = None
	        convert_element_type_763 = torch.ops.prims.convert_element_type.default(clamp_max_255, torch.int8);  clamp_max_255 = None
	        view_1996 = torch.ops.aten.view.default(clamp_min_381, [sym_size_int, 1500, 1]);  clamp_min_381 = None
	        view_1997 = torch.ops.aten.view.default(convert_element_type_762, [sym_size_int, 1500, 1]);  convert_element_type_762 = None
	        _assert_tensor_metadata_1148 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_763, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1148 = None
	        convert_element_type_764 = torch.ops.prims.convert_element_type.default(convert_element_type_763, torch.float32);  convert_element_type_763 = None
	        _assert_tensor_metadata_1149 = torch.ops.aten._assert_tensor_metadata.default(view_1997, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1149 = None
	        convert_element_type_765 = torch.ops.prims.convert_element_type.default(view_1997, torch.float32);  view_1997 = None
	        sub_5843 = torch.ops.aten.sub.Tensor(convert_element_type_764, convert_element_type_765);  convert_element_type_764 = convert_element_type_765 = None
	        mul_12370 = torch.ops.aten.mul.Tensor(sub_5843, view_1996);  sub_5843 = view_1996 = None
	        _assert_tensor_metadata_1150 = torch.ops.aten._assert_tensor_metadata.default(mul_12370, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1150 = None
	        view_1999 = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2000 = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2001 = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1151 = torch.ops.aten._assert_tensor_metadata.default(view_1999, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1151 = None
	        convert_element_type_766 = torch.ops.prims.convert_element_type.default(view_1999, torch.float32);  view_1999 = None
	        _assert_tensor_metadata_1152 = torch.ops.aten._assert_tensor_metadata.default(view_2001, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1152 = None
	        convert_element_type_767 = torch.ops.prims.convert_element_type.default(view_2001, torch.float32);  view_2001 = None
	        sub_5847 = torch.ops.aten.sub.Tensor(convert_element_type_766, convert_element_type_767);  convert_element_type_766 = convert_element_type_767 = None
	        mul_12375 = torch.ops.aten.mul.Tensor(sub_5847, view_2000);  sub_5847 = view_2000 = None
	        view_2002 = torch.ops.aten.view.default(mul_12375, [1280, 1280]);  mul_12375 = None
	        _assert_tensor_metadata_1153 = torch.ops.aten._assert_tensor_metadata.default(view_2002, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1153 = None
	        permute_213 = torch.ops.aten.permute.default(view_2002, [1, 0]);  view_2002 = None
	        mul_12378 = sym_size_int * 1500
	        view_2003 = torch.ops.aten.view.default(mul_12370, [mul_12378, 1280]);  mul_12370 = mul_12378 = None
	        mm_21 = torch.ops.aten.mm.default(view_2003, permute_213);  view_2003 = permute_213 = None
	        view_2004 = torch.ops.aten.view.default(mm_21, [sym_size_int, 1500, 1280]);  mm_21 = None
	        view_2005 = torch.ops.aten.view.default(view_2004, [sym_size_int, -1, 20, 64]);  view_2004 = None
	        permute_214 = torch.ops.aten.permute.default(view_2005, [0, 2, 1, 3]);  view_2005 = None
	        clone_171 = torch.ops.aten.clone.default(permute_214, memory_format = torch.contiguous_format);  permute_214 = None
	        amin_128 = torch.ops.aten.amin.default(add_19310, [2])
	        amax_128 = torch.ops.aten.amax.default(add_19310, [2])
	        full_256 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_128 = torch.ops.aten.minimum.default(amin_128, full_256);  amin_128 = full_256 = None
	        full_257 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_128 = torch.ops.aten.maximum.default(amax_128, full_257);  amax_128 = full_257 = None
	        sub_5861 = torch.ops.aten.sub.Tensor(maximum_128, minimum_128);  maximum_128 = None
	        div_256 = torch.ops.aten.div.Tensor(sub_5861, 255.0);  sub_5861 = None
	        clamp_min_384 = torch.ops.aten.clamp_min.default(div_256, 1.1920928955078125e-07);  div_256 = None
	        div_257 = torch.ops.aten.div.Tensor(minimum_128, clamp_min_384);  minimum_128 = None
	        round_257 = torch.ops.aten.round.default(div_257);  div_257 = None
	        sub_5867 = torch.ops.aten.sub.Tensor(-128, round_257);  round_257 = None
	        clamp_min_385 = torch.ops.aten.clamp_min.default(sub_5867, -128);  sub_5867 = None
	        clamp_max_256 = torch.ops.aten.clamp_max.default(clamp_min_385, 127);  clamp_min_385 = None
	        _assert_tensor_metadata_1154 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_384, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1154 = None
	        _assert_tensor_metadata_1155 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_256, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1155 = None
	        convert_element_type_768 = torch.ops.prims.convert_element_type.default(clamp_max_256, torch.int8);  clamp_max_256 = None
	        view_2008 = torch.ops.aten.view.default(clamp_min_384, [sym_size_int, 1500, 1])
	        view_2009 = torch.ops.aten.view.default(convert_element_type_768, [sym_size_int, 1500, 1])
	        reciprocal_128 = torch.ops.aten.reciprocal.default(view_2008);  view_2008 = None
	        mul_12444 = torch.ops.aten.mul.Tensor(reciprocal_128, 1.0);  reciprocal_128 = None
	        mul_12447 = torch.ops.aten.mul.Tensor(add_19310, mul_12444);  add_19310 = mul_12444 = None
	        round_258 = torch.ops.aten.round.default(mul_12447);  mul_12447 = None
	        add_19697 = torch.ops.aten.add.Tensor(round_258, view_2009);  round_258 = view_2009 = None
	        clamp_min_386 = torch.ops.aten.clamp_min.default(add_19697, -128);  add_19697 = None
	        clamp_max_257 = torch.ops.aten.clamp_max.default(clamp_min_386, 127);  clamp_min_386 = None
	        _assert_tensor_metadata_1156 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_257, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1156 = None
	        convert_element_type_769 = torch.ops.prims.convert_element_type.default(clamp_max_257, torch.int8);  clamp_max_257 = None
	        view_2012 = torch.ops.aten.view.default(clamp_min_384, [sym_size_int, 1500, 1]);  clamp_min_384 = None
	        view_2013 = torch.ops.aten.view.default(convert_element_type_768, [sym_size_int, 1500, 1]);  convert_element_type_768 = None
	        _assert_tensor_metadata_1157 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_769, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1157 = None
	        convert_element_type_770 = torch.ops.prims.convert_element_type.default(convert_element_type_769, torch.float32);  convert_element_type_769 = None
	        _assert_tensor_metadata_1158 = torch.ops.aten._assert_tensor_metadata.default(view_2013, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1158 = None
	        convert_element_type_771 = torch.ops.prims.convert_element_type.default(view_2013, torch.float32);  view_2013 = None
	        sub_5887 = torch.ops.aten.sub.Tensor(convert_element_type_770, convert_element_type_771);  convert_element_type_770 = convert_element_type_771 = None
	        mul_12469 = torch.ops.aten.mul.Tensor(sub_5887, view_2012);  sub_5887 = view_2012 = None
	        _assert_tensor_metadata_1159 = torch.ops.aten._assert_tensor_metadata.default(mul_12469, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1159 = None
	        view_2015 = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2016 = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2017 = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1160 = torch.ops.aten._assert_tensor_metadata.default(view_2015, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1160 = None
	        convert_element_type_772 = torch.ops.prims.convert_element_type.default(view_2015, torch.float32);  view_2015 = None
	        _assert_tensor_metadata_1161 = torch.ops.aten._assert_tensor_metadata.default(view_2017, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1161 = None
	        convert_element_type_773 = torch.ops.prims.convert_element_type.default(view_2017, torch.float32);  view_2017 = None
	        sub_5891 = torch.ops.aten.sub.Tensor(convert_element_type_772, convert_element_type_773);  convert_element_type_772 = convert_element_type_773 = None
	        mul_12474 = torch.ops.aten.mul.Tensor(sub_5891, view_2016);  sub_5891 = view_2016 = None
	        view_2018 = torch.ops.aten.view.default(mul_12474, [1280, 1280]);  mul_12474 = None
	        _assert_tensor_metadata_1162 = torch.ops.aten._assert_tensor_metadata.default(view_2018, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1162 = None
	        mul_12479 = sym_size_int * 1500
	        view_2019 = torch.ops.aten.view.default(mul_12469, [mul_12479, 1280]);  mul_12469 = mul_12479 = None
	        permute_215 = torch.ops.aten.permute.default(view_2018, [1, 0]);  view_2018 = None
	        addmm_106 = torch.ops.aten.addmm.default(model_audio_tower_layers_21_self_attn_v_proj_bias, view_2019, permute_215);  model_audio_tower_layers_21_self_attn_v_proj_bias = view_2019 = permute_215 = None
	        view_2020 = torch.ops.aten.view.default(addmm_106, [sym_size_int, 1500, 1280]);  addmm_106 = None
	        view_2021 = torch.ops.aten.view.default(view_2020, [sym_size_int, -1, 20, 64]);  view_2020 = None
	        permute_216 = torch.ops.aten.permute.default(view_2021, [0, 2, 1, 3]);  view_2021 = None
	        clone_172 = torch.ops.aten.clone.default(permute_216, memory_format = torch.contiguous_format);  permute_216 = None
	        _scaled_dot_product_efficient_attention_21 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_170, clone_171, clone_172, None, False, scale = 1.0);  clone_170 = clone_171 = clone_172 = None
	        getitem_170 = _scaled_dot_product_efficient_attention_21[0];  _scaled_dot_product_efficient_attention_21 = None
	        permute_217 = torch.ops.aten.permute.default(getitem_170, [0, 2, 1, 3]);  getitem_170 = None
	        view_2022 = torch.ops.aten.view.default(permute_217, [sym_size_int, 1500, -1]);  permute_217 = None
	        amin_129 = torch.ops.aten.amin.default(view_2022, [2])
	        amax_129 = torch.ops.aten.amax.default(view_2022, [2])
	        full_258 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_129 = torch.ops.aten.minimum.default(amin_129, full_258);  amin_129 = full_258 = None
	        full_259 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_129 = torch.ops.aten.maximum.default(amax_129, full_259);  amax_129 = full_259 = None
	        sub_5909 = torch.ops.aten.sub.Tensor(maximum_129, minimum_129);  maximum_129 = None
	        div_258 = torch.ops.aten.div.Tensor(sub_5909, 255.0);  sub_5909 = None
	        clamp_min_387 = torch.ops.aten.clamp_min.default(div_258, 1.1920928955078125e-07);  div_258 = None
	        div_259 = torch.ops.aten.div.Tensor(minimum_129, clamp_min_387);  minimum_129 = None
	        round_259 = torch.ops.aten.round.default(div_259);  div_259 = None
	        sub_5915 = torch.ops.aten.sub.Tensor(-128, round_259);  round_259 = None
	        clamp_min_388 = torch.ops.aten.clamp_min.default(sub_5915, -128);  sub_5915 = None
	        clamp_max_258 = torch.ops.aten.clamp_max.default(clamp_min_388, 127);  clamp_min_388 = None
	        _assert_tensor_metadata_1163 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_387, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1163 = None
	        _assert_tensor_metadata_1164 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_258, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1164 = None
	        convert_element_type_774 = torch.ops.prims.convert_element_type.default(clamp_max_258, torch.int8);  clamp_max_258 = None
	        view_2025 = torch.ops.aten.view.default(clamp_min_387, [sym_size_int, 1500, 1])
	        view_2026 = torch.ops.aten.view.default(convert_element_type_774, [sym_size_int, 1500, 1])
	        reciprocal_129 = torch.ops.aten.reciprocal.default(view_2025);  view_2025 = None
	        mul_12549 = torch.ops.aten.mul.Tensor(reciprocal_129, 1.0);  reciprocal_129 = None
	        mul_12552 = torch.ops.aten.mul.Tensor(view_2022, mul_12549);  view_2022 = mul_12549 = None
	        round_260 = torch.ops.aten.round.default(mul_12552);  mul_12552 = None
	        add_19861 = torch.ops.aten.add.Tensor(round_260, view_2026);  round_260 = view_2026 = None
	        clamp_min_389 = torch.ops.aten.clamp_min.default(add_19861, -128);  add_19861 = None
	        clamp_max_259 = torch.ops.aten.clamp_max.default(clamp_min_389, 127);  clamp_min_389 = None
	        _assert_tensor_metadata_1165 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_259, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1165 = None
	        convert_element_type_775 = torch.ops.prims.convert_element_type.default(clamp_max_259, torch.int8);  clamp_max_259 = None
	        view_2029 = torch.ops.aten.view.default(clamp_min_387, [sym_size_int, 1500, 1]);  clamp_min_387 = None
	        view_2030 = torch.ops.aten.view.default(convert_element_type_774, [sym_size_int, 1500, 1]);  convert_element_type_774 = None
	        _assert_tensor_metadata_1166 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_775, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1166 = None
	        convert_element_type_776 = torch.ops.prims.convert_element_type.default(convert_element_type_775, torch.float32);  convert_element_type_775 = None
	        _assert_tensor_metadata_1167 = torch.ops.aten._assert_tensor_metadata.default(view_2030, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1167 = None
	        convert_element_type_777 = torch.ops.prims.convert_element_type.default(view_2030, torch.float32);  view_2030 = None
	        sub_5935 = torch.ops.aten.sub.Tensor(convert_element_type_776, convert_element_type_777);  convert_element_type_776 = convert_element_type_777 = None
	        mul_12574 = torch.ops.aten.mul.Tensor(sub_5935, view_2029);  sub_5935 = view_2029 = None
	        _assert_tensor_metadata_1168 = torch.ops.aten._assert_tensor_metadata.default(mul_12574, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1168 = None
	        view_2032 = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2033 = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2034 = torch.ops.aten.view.default(model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1169 = torch.ops.aten._assert_tensor_metadata.default(view_2032, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1169 = None
	        convert_element_type_778 = torch.ops.prims.convert_element_type.default(view_2032, torch.float32);  view_2032 = None
	        _assert_tensor_metadata_1170 = torch.ops.aten._assert_tensor_metadata.default(view_2034, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1170 = None
	        convert_element_type_779 = torch.ops.prims.convert_element_type.default(view_2034, torch.float32);  view_2034 = None
	        sub_5939 = torch.ops.aten.sub.Tensor(convert_element_type_778, convert_element_type_779);  convert_element_type_778 = convert_element_type_779 = None
	        mul_12579 = torch.ops.aten.mul.Tensor(sub_5939, view_2033);  sub_5939 = view_2033 = None
	        view_2035 = torch.ops.aten.view.default(mul_12579, [1280, 1280]);  mul_12579 = None
	        _assert_tensor_metadata_1171 = torch.ops.aten._assert_tensor_metadata.default(view_2035, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1171 = None
	        mul_12584 = sym_size_int * 1500
	        view_2036 = torch.ops.aten.view.default(mul_12574, [mul_12584, 1280]);  mul_12574 = mul_12584 = None
	        permute_218 = torch.ops.aten.permute.default(view_2035, [1, 0]);  view_2035 = None
	        addmm_107 = torch.ops.aten.addmm.default(model_audio_tower_layers_21_self_attn_out_proj_bias, view_2036, permute_218);  model_audio_tower_layers_21_self_attn_out_proj_bias = view_2036 = permute_218 = None
	        view_2037 = torch.ops.aten.view.default(addmm_107, [sym_size_int, 1500, 1280]);  addmm_107 = None
	        add_19924 = torch.ops.aten.add.Tensor(add_19304, view_2037);  add_19304 = view_2037 = None
	        clone_174 = torch.ops.aten.clone.default(add_19924, memory_format = torch.contiguous_format)
	        var_mean_43 = torch.ops.aten.var_mean.correction(clone_174, [2], correction = 0, keepdim = True)
	        getitem_174 = var_mean_43[0]
	        getitem_175 = var_mean_43[1];  var_mean_43 = None
	        add_19929 = torch.ops.aten.add.Tensor(getitem_174, 1e-05);  getitem_174 = None
	        rsqrt_43 = torch.ops.aten.rsqrt.default(add_19929);  add_19929 = None
	        sub_5945 = torch.ops.aten.sub.Tensor(clone_174, getitem_175);  clone_174 = getitem_175 = None
	        mul_12595 = torch.ops.aten.mul.Tensor(sub_5945, rsqrt_43);  sub_5945 = rsqrt_43 = None
	        mul_12596 = torch.ops.aten.mul.Tensor(mul_12595, model_audio_tower_layers_21_final_layer_norm_weight);  mul_12595 = model_audio_tower_layers_21_final_layer_norm_weight = None
	        add_19930 = torch.ops.aten.add.Tensor(mul_12596, model_audio_tower_layers_21_final_layer_norm_bias);  mul_12596 = model_audio_tower_layers_21_final_layer_norm_bias = None
	        amin_130 = torch.ops.aten.amin.default(add_19930, [2])
	        amax_130 = torch.ops.aten.amax.default(add_19930, [2])
	        full_260 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_130 = torch.ops.aten.minimum.default(amin_130, full_260);  amin_130 = full_260 = None
	        full_261 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_130 = torch.ops.aten.maximum.default(amax_130, full_261);  amax_130 = full_261 = None
	        sub_5956 = torch.ops.aten.sub.Tensor(maximum_130, minimum_130);  maximum_130 = None
	        div_260 = torch.ops.aten.div.Tensor(sub_5956, 255.0);  sub_5956 = None
	        clamp_min_390 = torch.ops.aten.clamp_min.default(div_260, 1.1920928955078125e-07);  div_260 = None
	        div_261 = torch.ops.aten.div.Tensor(minimum_130, clamp_min_390);  minimum_130 = None
	        round_261 = torch.ops.aten.round.default(div_261);  div_261 = None
	        sub_5962 = torch.ops.aten.sub.Tensor(-128, round_261);  round_261 = None
	        clamp_min_391 = torch.ops.aten.clamp_min.default(sub_5962, -128);  sub_5962 = None
	        clamp_max_260 = torch.ops.aten.clamp_max.default(clamp_min_391, 127);  clamp_min_391 = None
	        _assert_tensor_metadata_1172 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_390, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1172 = None
	        _assert_tensor_metadata_1173 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_260, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1173 = None
	        convert_element_type_780 = torch.ops.prims.convert_element_type.default(clamp_max_260, torch.int8);  clamp_max_260 = None
	        view_2040 = torch.ops.aten.view.default(clamp_min_390, [sym_size_int, 1500, 1])
	        view_2041 = torch.ops.aten.view.default(convert_element_type_780, [sym_size_int, 1500, 1])
	        reciprocal_130 = torch.ops.aten.reciprocal.default(view_2040);  view_2040 = None
	        mul_12644 = torch.ops.aten.mul.Tensor(reciprocal_130, 1.0);  reciprocal_130 = None
	        mul_12647 = torch.ops.aten.mul.Tensor(add_19930, mul_12644);  add_19930 = mul_12644 = None
	        round_262 = torch.ops.aten.round.default(mul_12647);  mul_12647 = None
	        add_20017 = torch.ops.aten.add.Tensor(round_262, view_2041);  round_262 = view_2041 = None
	        clamp_min_392 = torch.ops.aten.clamp_min.default(add_20017, -128);  add_20017 = None
	        clamp_max_261 = torch.ops.aten.clamp_max.default(clamp_min_392, 127);  clamp_min_392 = None
	        _assert_tensor_metadata_1174 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_261, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1174 = None
	        convert_element_type_781 = torch.ops.prims.convert_element_type.default(clamp_max_261, torch.int8);  clamp_max_261 = None
	        view_2044 = torch.ops.aten.view.default(clamp_min_390, [sym_size_int, 1500, 1]);  clamp_min_390 = None
	        view_2045 = torch.ops.aten.view.default(convert_element_type_780, [sym_size_int, 1500, 1]);  convert_element_type_780 = None
	        _assert_tensor_metadata_1175 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_781, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1175 = None
	        convert_element_type_782 = torch.ops.prims.convert_element_type.default(convert_element_type_781, torch.float32);  convert_element_type_781 = None
	        _assert_tensor_metadata_1176 = torch.ops.aten._assert_tensor_metadata.default(view_2045, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1176 = None
	        convert_element_type_783 = torch.ops.prims.convert_element_type.default(view_2045, torch.float32);  view_2045 = None
	        sub_5982 = torch.ops.aten.sub.Tensor(convert_element_type_782, convert_element_type_783);  convert_element_type_782 = convert_element_type_783 = None
	        mul_12669 = torch.ops.aten.mul.Tensor(sub_5982, view_2044);  sub_5982 = view_2044 = None
	        _assert_tensor_metadata_1177 = torch.ops.aten._assert_tensor_metadata.default(mul_12669, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1177 = None
	        view_2047 = torch.ops.aten.view.default(model_audio_tower_layers_21_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_21_fc1_parametrizations_weight_original0 = None
	        view_2048 = torch.ops.aten.view.default(model_audio_tower_layers_21_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_21_fc1_parametrizations_weight_original1 = None
	        view_2049 = torch.ops.aten.view.default(model_audio_tower_layers_21_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_21_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1178 = torch.ops.aten._assert_tensor_metadata.default(view_2047, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1178 = None
	        convert_element_type_784 = torch.ops.prims.convert_element_type.default(view_2047, torch.float32);  view_2047 = None
	        _assert_tensor_metadata_1179 = torch.ops.aten._assert_tensor_metadata.default(view_2049, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1179 = None
	        convert_element_type_785 = torch.ops.prims.convert_element_type.default(view_2049, torch.float32);  view_2049 = None
	        sub_5986 = torch.ops.aten.sub.Tensor(convert_element_type_784, convert_element_type_785);  convert_element_type_784 = convert_element_type_785 = None
	        mul_12674 = torch.ops.aten.mul.Tensor(sub_5986, view_2048);  sub_5986 = view_2048 = None
	        view_2050 = torch.ops.aten.view.default(mul_12674, [5120, 1280]);  mul_12674 = None
	        _assert_tensor_metadata_1180 = torch.ops.aten._assert_tensor_metadata.default(view_2050, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1180 = None
	        mul_12679 = sym_size_int * 1500
	        view_2051 = torch.ops.aten.view.default(mul_12669, [mul_12679, 1280]);  mul_12669 = mul_12679 = None
	        permute_219 = torch.ops.aten.permute.default(view_2050, [1, 0]);  view_2050 = None
	        addmm_108 = torch.ops.aten.addmm.default(model_audio_tower_layers_21_fc1_bias, view_2051, permute_219);  model_audio_tower_layers_21_fc1_bias = view_2051 = permute_219 = None
	        view_2052 = torch.ops.aten.view.default(addmm_108, [sym_size_int, 1500, 5120]);  addmm_108 = None
	        mul_12686 = torch.ops.aten.mul.Tensor(view_2052, 0.5)
	        mul_12687 = torch.ops.aten.mul.Tensor(view_2052, 0.7071067811865476);  view_2052 = None
	        erf_23 = torch.ops.aten.erf.default(mul_12687);  mul_12687 = None
	        add_20076 = torch.ops.aten.add.Tensor(erf_23, 1);  erf_23 = None
	        mul_12688 = torch.ops.aten.mul.Tensor(mul_12686, add_20076);  mul_12686 = add_20076 = None
	        amin_131 = torch.ops.aten.amin.default(mul_12688, [2])
	        amax_131 = torch.ops.aten.amax.default(mul_12688, [2])
	        full_262 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_131 = torch.ops.aten.minimum.default(amin_131, full_262);  amin_131 = full_262 = None
	        full_263 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_131 = torch.ops.aten.maximum.default(amax_131, full_263);  amax_131 = full_263 = None
	        sub_5999 = torch.ops.aten.sub.Tensor(maximum_131, minimum_131);  maximum_131 = None
	        div_262 = torch.ops.aten.div.Tensor(sub_5999, 255.0);  sub_5999 = None
	        clamp_min_393 = torch.ops.aten.clamp_min.default(div_262, 1.1920928955078125e-07);  div_262 = None
	        div_263 = torch.ops.aten.div.Tensor(minimum_131, clamp_min_393);  minimum_131 = None
	        round_263 = torch.ops.aten.round.default(div_263);  div_263 = None
	        sub_6005 = torch.ops.aten.sub.Tensor(-128, round_263);  round_263 = None
	        clamp_min_394 = torch.ops.aten.clamp_min.default(sub_6005, -128);  sub_6005 = None
	        clamp_max_262 = torch.ops.aten.clamp_max.default(clamp_min_394, 127);  clamp_min_394 = None
	        _assert_tensor_metadata_1181 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_393, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1181 = None
	        _assert_tensor_metadata_1182 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_262, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1182 = None
	        convert_element_type_786 = torch.ops.prims.convert_element_type.default(clamp_max_262, torch.int8);  clamp_max_262 = None
	        view_2055 = torch.ops.aten.view.default(clamp_min_393, [sym_size_int, 1500, 1])
	        view_2056 = torch.ops.aten.view.default(convert_element_type_786, [sym_size_int, 1500, 1])
	        reciprocal_131 = torch.ops.aten.reciprocal.default(view_2055);  view_2055 = None
	        mul_12734 = torch.ops.aten.mul.Tensor(reciprocal_131, 1.0);  reciprocal_131 = None
	        mul_12737 = torch.ops.aten.mul.Tensor(mul_12688, mul_12734);  mul_12688 = mul_12734 = None
	        round_264 = torch.ops.aten.round.default(mul_12737);  mul_12737 = None
	        add_20159 = torch.ops.aten.add.Tensor(round_264, view_2056);  round_264 = view_2056 = None
	        clamp_min_395 = torch.ops.aten.clamp_min.default(add_20159, -128);  add_20159 = None
	        clamp_max_263 = torch.ops.aten.clamp_max.default(clamp_min_395, 127);  clamp_min_395 = None
	        _assert_tensor_metadata_1183 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_263, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1183 = None
	        convert_element_type_787 = torch.ops.prims.convert_element_type.default(clamp_max_263, torch.int8);  clamp_max_263 = None
	        view_2059 = torch.ops.aten.view.default(clamp_min_393, [sym_size_int, 1500, 1]);  clamp_min_393 = None
	        view_2060 = torch.ops.aten.view.default(convert_element_type_786, [sym_size_int, 1500, 1]);  convert_element_type_786 = None
	        _assert_tensor_metadata_1184 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_787, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1184 = None
	        convert_element_type_788 = torch.ops.prims.convert_element_type.default(convert_element_type_787, torch.float32);  convert_element_type_787 = None
	        _assert_tensor_metadata_1185 = torch.ops.aten._assert_tensor_metadata.default(view_2060, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1185 = None
	        convert_element_type_789 = torch.ops.prims.convert_element_type.default(view_2060, torch.float32);  view_2060 = None
	        sub_6025 = torch.ops.aten.sub.Tensor(convert_element_type_788, convert_element_type_789);  convert_element_type_788 = convert_element_type_789 = None
	        mul_12759 = torch.ops.aten.mul.Tensor(sub_6025, view_2059);  sub_6025 = view_2059 = None
	        _assert_tensor_metadata_1186 = torch.ops.aten._assert_tensor_metadata.default(mul_12759, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1186 = None
	        view_2062 = torch.ops.aten.view.default(model_audio_tower_layers_21_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_21_fc2_parametrizations_weight_original0 = None
	        view_2063 = torch.ops.aten.view.default(model_audio_tower_layers_21_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_21_fc2_parametrizations_weight_original1 = None
	        view_2064 = torch.ops.aten.view.default(model_audio_tower_layers_21_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_21_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1187 = torch.ops.aten._assert_tensor_metadata.default(view_2062, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1187 = None
	        convert_element_type_790 = torch.ops.prims.convert_element_type.default(view_2062, torch.float32);  view_2062 = None
	        _assert_tensor_metadata_1188 = torch.ops.aten._assert_tensor_metadata.default(view_2064, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1188 = None
	        convert_element_type_791 = torch.ops.prims.convert_element_type.default(view_2064, torch.float32);  view_2064 = None
	        sub_6029 = torch.ops.aten.sub.Tensor(convert_element_type_790, convert_element_type_791);  convert_element_type_790 = convert_element_type_791 = None
	        mul_12764 = torch.ops.aten.mul.Tensor(sub_6029, view_2063);  sub_6029 = view_2063 = None
	        view_2065 = torch.ops.aten.view.default(mul_12764, [1280, 5120]);  mul_12764 = None
	        _assert_tensor_metadata_1189 = torch.ops.aten._assert_tensor_metadata.default(view_2065, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1189 = None
	        mul_12769 = sym_size_int * 1500
	        view_2066 = torch.ops.aten.view.default(mul_12759, [mul_12769, 5120]);  mul_12759 = mul_12769 = None
	        permute_220 = torch.ops.aten.permute.default(view_2065, [1, 0]);  view_2065 = None
	        addmm_109 = torch.ops.aten.addmm.default(model_audio_tower_layers_21_fc2_bias, view_2066, permute_220);  model_audio_tower_layers_21_fc2_bias = view_2066 = permute_220 = None
	        view_2067 = torch.ops.aten.view.default(addmm_109, [sym_size_int, 1500, 1280]);  addmm_109 = None
	        add_20222 = torch.ops.aten.add.Tensor(add_19924, view_2067);  add_19924 = view_2067 = None
	        clone_177 = torch.ops.aten.clone.default(add_20222, memory_format = torch.contiguous_format)
	        var_mean_44 = torch.ops.aten.var_mean.correction(clone_177, [2], correction = 0, keepdim = True)
	        getitem_176 = var_mean_44[0]
	        getitem_177 = var_mean_44[1];  var_mean_44 = None
	        add_20227 = torch.ops.aten.add.Tensor(getitem_176, 1e-05);  getitem_176 = None
	        rsqrt_44 = torch.ops.aten.rsqrt.default(add_20227);  add_20227 = None
	        sub_6035 = torch.ops.aten.sub.Tensor(clone_177, getitem_177);  clone_177 = getitem_177 = None
	        mul_12780 = torch.ops.aten.mul.Tensor(sub_6035, rsqrt_44);  sub_6035 = rsqrt_44 = None
	        mul_12781 = torch.ops.aten.mul.Tensor(mul_12780, model_audio_tower_layers_22_self_attn_layer_norm_weight);  mul_12780 = model_audio_tower_layers_22_self_attn_layer_norm_weight = None
	        add_20228 = torch.ops.aten.add.Tensor(mul_12781, model_audio_tower_layers_22_self_attn_layer_norm_bias);  mul_12781 = model_audio_tower_layers_22_self_attn_layer_norm_bias = None
	        amin_132 = torch.ops.aten.amin.default(add_20228, [2])
	        amax_132 = torch.ops.aten.amax.default(add_20228, [2])
	        full_264 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_132 = torch.ops.aten.minimum.default(amin_132, full_264);  amin_132 = full_264 = None
	        full_265 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_132 = torch.ops.aten.maximum.default(amax_132, full_265);  amax_132 = full_265 = None
	        sub_6046 = torch.ops.aten.sub.Tensor(maximum_132, minimum_132);  maximum_132 = None
	        div_264 = torch.ops.aten.div.Tensor(sub_6046, 255.0);  sub_6046 = None
	        clamp_min_396 = torch.ops.aten.clamp_min.default(div_264, 1.1920928955078125e-07);  div_264 = None
	        div_265 = torch.ops.aten.div.Tensor(minimum_132, clamp_min_396);  minimum_132 = None
	        round_265 = torch.ops.aten.round.default(div_265);  div_265 = None
	        sub_6052 = torch.ops.aten.sub.Tensor(-128, round_265);  round_265 = None
	        clamp_min_397 = torch.ops.aten.clamp_min.default(sub_6052, -128);  sub_6052 = None
	        clamp_max_264 = torch.ops.aten.clamp_max.default(clamp_min_397, 127);  clamp_min_397 = None
	        _assert_tensor_metadata_1190 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_396, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1190 = None
	        _assert_tensor_metadata_1191 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_264, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1191 = None
	        convert_element_type_792 = torch.ops.prims.convert_element_type.default(clamp_max_264, torch.int8);  clamp_max_264 = None
	        view_2070 = torch.ops.aten.view.default(clamp_min_396, [sym_size_int, 1500, 1])
	        view_2071 = torch.ops.aten.view.default(convert_element_type_792, [sym_size_int, 1500, 1])
	        reciprocal_132 = torch.ops.aten.reciprocal.default(view_2070);  view_2070 = None
	        mul_12829 = torch.ops.aten.mul.Tensor(reciprocal_132, 1.0);  reciprocal_132 = None
	        mul_12832 = torch.ops.aten.mul.Tensor(add_20228, mul_12829);  mul_12829 = None
	        round_266 = torch.ops.aten.round.default(mul_12832);  mul_12832 = None
	        add_20315 = torch.ops.aten.add.Tensor(round_266, view_2071);  round_266 = view_2071 = None
	        clamp_min_398 = torch.ops.aten.clamp_min.default(add_20315, -128);  add_20315 = None
	        clamp_max_265 = torch.ops.aten.clamp_max.default(clamp_min_398, 127);  clamp_min_398 = None
	        _assert_tensor_metadata_1192 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_265, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1192 = None
	        convert_element_type_793 = torch.ops.prims.convert_element_type.default(clamp_max_265, torch.int8);  clamp_max_265 = None
	        view_2074 = torch.ops.aten.view.default(clamp_min_396, [sym_size_int, 1500, 1]);  clamp_min_396 = None
	        view_2075 = torch.ops.aten.view.default(convert_element_type_792, [sym_size_int, 1500, 1]);  convert_element_type_792 = None
	        _assert_tensor_metadata_1193 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_793, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1193 = None
	        convert_element_type_794 = torch.ops.prims.convert_element_type.default(convert_element_type_793, torch.float32);  convert_element_type_793 = None
	        _assert_tensor_metadata_1194 = torch.ops.aten._assert_tensor_metadata.default(view_2075, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1194 = None
	        convert_element_type_795 = torch.ops.prims.convert_element_type.default(view_2075, torch.float32);  view_2075 = None
	        sub_6072 = torch.ops.aten.sub.Tensor(convert_element_type_794, convert_element_type_795);  convert_element_type_794 = convert_element_type_795 = None
	        mul_12854 = torch.ops.aten.mul.Tensor(sub_6072, view_2074);  sub_6072 = view_2074 = None
	        _assert_tensor_metadata_1195 = torch.ops.aten._assert_tensor_metadata.default(mul_12854, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1195 = None
	        view_2077 = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2078 = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2079 = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1196 = torch.ops.aten._assert_tensor_metadata.default(view_2077, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1196 = None
	        convert_element_type_796 = torch.ops.prims.convert_element_type.default(view_2077, torch.float32);  view_2077 = None
	        _assert_tensor_metadata_1197 = torch.ops.aten._assert_tensor_metadata.default(view_2079, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1197 = None
	        convert_element_type_797 = torch.ops.prims.convert_element_type.default(view_2079, torch.float32);  view_2079 = None
	        sub_6076 = torch.ops.aten.sub.Tensor(convert_element_type_796, convert_element_type_797);  convert_element_type_796 = convert_element_type_797 = None
	        mul_12859 = torch.ops.aten.mul.Tensor(sub_6076, view_2078);  sub_6076 = view_2078 = None
	        view_2080 = torch.ops.aten.view.default(mul_12859, [1280, 1280]);  mul_12859 = None
	        _assert_tensor_metadata_1198 = torch.ops.aten._assert_tensor_metadata.default(view_2080, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1198 = None
	        mul_12864 = sym_size_int * 1500
	        view_2081 = torch.ops.aten.view.default(mul_12854, [mul_12864, 1280]);  mul_12854 = mul_12864 = None
	        permute_221 = torch.ops.aten.permute.default(view_2080, [1, 0]);  view_2080 = None
	        addmm_110 = torch.ops.aten.addmm.default(model_audio_tower_layers_22_self_attn_q_proj_bias, view_2081, permute_221);  model_audio_tower_layers_22_self_attn_q_proj_bias = view_2081 = permute_221 = None
	        view_2082 = torch.ops.aten.view.default(addmm_110, [sym_size_int, 1500, 1280]);  addmm_110 = None
	        mul_12871 = torch.ops.aten.mul.Tensor(view_2082, 0.125);  view_2082 = None
	        view_2083 = torch.ops.aten.view.default(mul_12871, [sym_size_int, 1500, 20, 64]);  mul_12871 = None
	        permute_222 = torch.ops.aten.permute.default(view_2083, [0, 2, 1, 3]);  view_2083 = None
	        clone_178 = torch.ops.aten.clone.default(permute_222, memory_format = torch.contiguous_format);  permute_222 = None
	        amin_133 = torch.ops.aten.amin.default(add_20228, [2])
	        amax_133 = torch.ops.aten.amax.default(add_20228, [2])
	        full_266 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_133 = torch.ops.aten.minimum.default(amin_133, full_266);  amin_133 = full_266 = None
	        full_267 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_133 = torch.ops.aten.maximum.default(amax_133, full_267);  amax_133 = full_267 = None
	        sub_6091 = torch.ops.aten.sub.Tensor(maximum_133, minimum_133);  maximum_133 = None
	        div_266 = torch.ops.aten.div.Tensor(sub_6091, 255.0);  sub_6091 = None
	        clamp_min_399 = torch.ops.aten.clamp_min.default(div_266, 1.1920928955078125e-07);  div_266 = None
	        div_267 = torch.ops.aten.div.Tensor(minimum_133, clamp_min_399);  minimum_133 = None
	        round_267 = torch.ops.aten.round.default(div_267);  div_267 = None
	        sub_6097 = torch.ops.aten.sub.Tensor(-128, round_267);  round_267 = None
	        clamp_min_400 = torch.ops.aten.clamp_min.default(sub_6097, -128);  sub_6097 = None
	        clamp_max_266 = torch.ops.aten.clamp_max.default(clamp_min_400, 127);  clamp_min_400 = None
	        _assert_tensor_metadata_1199 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_399, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1199 = None
	        _assert_tensor_metadata_1200 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_266, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1200 = None
	        convert_element_type_798 = torch.ops.prims.convert_element_type.default(clamp_max_266, torch.int8);  clamp_max_266 = None
	        view_2086 = torch.ops.aten.view.default(clamp_min_399, [sym_size_int, 1500, 1])
	        view_2087 = torch.ops.aten.view.default(convert_element_type_798, [sym_size_int, 1500, 1])
	        reciprocal_133 = torch.ops.aten.reciprocal.default(view_2086);  view_2086 = None
	        mul_12925 = torch.ops.aten.mul.Tensor(reciprocal_133, 1.0);  reciprocal_133 = None
	        mul_12928 = torch.ops.aten.mul.Tensor(add_20228, mul_12925);  mul_12925 = None
	        round_268 = torch.ops.aten.round.default(mul_12928);  mul_12928 = None
	        add_20467 = torch.ops.aten.add.Tensor(round_268, view_2087);  round_268 = view_2087 = None
	        clamp_min_401 = torch.ops.aten.clamp_min.default(add_20467, -128);  add_20467 = None
	        clamp_max_267 = torch.ops.aten.clamp_max.default(clamp_min_401, 127);  clamp_min_401 = None
	        _assert_tensor_metadata_1201 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_267, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1201 = None
	        convert_element_type_799 = torch.ops.prims.convert_element_type.default(clamp_max_267, torch.int8);  clamp_max_267 = None
	        view_2090 = torch.ops.aten.view.default(clamp_min_399, [sym_size_int, 1500, 1]);  clamp_min_399 = None
	        view_2091 = torch.ops.aten.view.default(convert_element_type_798, [sym_size_int, 1500, 1]);  convert_element_type_798 = None
	        _assert_tensor_metadata_1202 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_799, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1202 = None
	        convert_element_type_800 = torch.ops.prims.convert_element_type.default(convert_element_type_799, torch.float32);  convert_element_type_799 = None
	        _assert_tensor_metadata_1203 = torch.ops.aten._assert_tensor_metadata.default(view_2091, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1203 = None
	        convert_element_type_801 = torch.ops.prims.convert_element_type.default(view_2091, torch.float32);  view_2091 = None
	        sub_6117 = torch.ops.aten.sub.Tensor(convert_element_type_800, convert_element_type_801);  convert_element_type_800 = convert_element_type_801 = None
	        mul_12950 = torch.ops.aten.mul.Tensor(sub_6117, view_2090);  sub_6117 = view_2090 = None
	        _assert_tensor_metadata_1204 = torch.ops.aten._assert_tensor_metadata.default(mul_12950, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1204 = None
	        view_2093 = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2094 = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2095 = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1205 = torch.ops.aten._assert_tensor_metadata.default(view_2093, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1205 = None
	        convert_element_type_802 = torch.ops.prims.convert_element_type.default(view_2093, torch.float32);  view_2093 = None
	        _assert_tensor_metadata_1206 = torch.ops.aten._assert_tensor_metadata.default(view_2095, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1206 = None
	        convert_element_type_803 = torch.ops.prims.convert_element_type.default(view_2095, torch.float32);  view_2095 = None
	        sub_6121 = torch.ops.aten.sub.Tensor(convert_element_type_802, convert_element_type_803);  convert_element_type_802 = convert_element_type_803 = None
	        mul_12955 = torch.ops.aten.mul.Tensor(sub_6121, view_2094);  sub_6121 = view_2094 = None
	        view_2096 = torch.ops.aten.view.default(mul_12955, [1280, 1280]);  mul_12955 = None
	        _assert_tensor_metadata_1207 = torch.ops.aten._assert_tensor_metadata.default(view_2096, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1207 = None
	        permute_223 = torch.ops.aten.permute.default(view_2096, [1, 0]);  view_2096 = None
	        mul_12958 = sym_size_int * 1500
	        view_2097 = torch.ops.aten.view.default(mul_12950, [mul_12958, 1280]);  mul_12950 = mul_12958 = None
	        mm_22 = torch.ops.aten.mm.default(view_2097, permute_223);  view_2097 = permute_223 = None
	        view_2098 = torch.ops.aten.view.default(mm_22, [sym_size_int, 1500, 1280]);  mm_22 = None
	        view_2099 = torch.ops.aten.view.default(view_2098, [sym_size_int, -1, 20, 64]);  view_2098 = None
	        permute_224 = torch.ops.aten.permute.default(view_2099, [0, 2, 1, 3]);  view_2099 = None
	        clone_179 = torch.ops.aten.clone.default(permute_224, memory_format = torch.contiguous_format);  permute_224 = None
	        amin_134 = torch.ops.aten.amin.default(add_20228, [2])
	        amax_134 = torch.ops.aten.amax.default(add_20228, [2])
	        full_268 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_134 = torch.ops.aten.minimum.default(amin_134, full_268);  amin_134 = full_268 = None
	        full_269 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_134 = torch.ops.aten.maximum.default(amax_134, full_269);  amax_134 = full_269 = None
	        sub_6135 = torch.ops.aten.sub.Tensor(maximum_134, minimum_134);  maximum_134 = None
	        div_268 = torch.ops.aten.div.Tensor(sub_6135, 255.0);  sub_6135 = None
	        clamp_min_402 = torch.ops.aten.clamp_min.default(div_268, 1.1920928955078125e-07);  div_268 = None
	        div_269 = torch.ops.aten.div.Tensor(minimum_134, clamp_min_402);  minimum_134 = None
	        round_269 = torch.ops.aten.round.default(div_269);  div_269 = None
	        sub_6141 = torch.ops.aten.sub.Tensor(-128, round_269);  round_269 = None
	        clamp_min_403 = torch.ops.aten.clamp_min.default(sub_6141, -128);  sub_6141 = None
	        clamp_max_268 = torch.ops.aten.clamp_max.default(clamp_min_403, 127);  clamp_min_403 = None
	        _assert_tensor_metadata_1208 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_402, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1208 = None
	        _assert_tensor_metadata_1209 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_268, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1209 = None
	        convert_element_type_804 = torch.ops.prims.convert_element_type.default(clamp_max_268, torch.int8);  clamp_max_268 = None
	        view_2102 = torch.ops.aten.view.default(clamp_min_402, [sym_size_int, 1500, 1])
	        view_2103 = torch.ops.aten.view.default(convert_element_type_804, [sym_size_int, 1500, 1])
	        reciprocal_134 = torch.ops.aten.reciprocal.default(view_2102);  view_2102 = None
	        mul_13024 = torch.ops.aten.mul.Tensor(reciprocal_134, 1.0);  reciprocal_134 = None
	        mul_13027 = torch.ops.aten.mul.Tensor(add_20228, mul_13024);  add_20228 = mul_13024 = None
	        round_270 = torch.ops.aten.round.default(mul_13027);  mul_13027 = None
	        add_20615 = torch.ops.aten.add.Tensor(round_270, view_2103);  round_270 = view_2103 = None
	        clamp_min_404 = torch.ops.aten.clamp_min.default(add_20615, -128);  add_20615 = None
	        clamp_max_269 = torch.ops.aten.clamp_max.default(clamp_min_404, 127);  clamp_min_404 = None
	        _assert_tensor_metadata_1210 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_269, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1210 = None
	        convert_element_type_805 = torch.ops.prims.convert_element_type.default(clamp_max_269, torch.int8);  clamp_max_269 = None
	        view_2106 = torch.ops.aten.view.default(clamp_min_402, [sym_size_int, 1500, 1]);  clamp_min_402 = None
	        view_2107 = torch.ops.aten.view.default(convert_element_type_804, [sym_size_int, 1500, 1]);  convert_element_type_804 = None
	        _assert_tensor_metadata_1211 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_805, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1211 = None
	        convert_element_type_806 = torch.ops.prims.convert_element_type.default(convert_element_type_805, torch.float32);  convert_element_type_805 = None
	        _assert_tensor_metadata_1212 = torch.ops.aten._assert_tensor_metadata.default(view_2107, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1212 = None
	        convert_element_type_807 = torch.ops.prims.convert_element_type.default(view_2107, torch.float32);  view_2107 = None
	        sub_6161 = torch.ops.aten.sub.Tensor(convert_element_type_806, convert_element_type_807);  convert_element_type_806 = convert_element_type_807 = None
	        mul_13049 = torch.ops.aten.mul.Tensor(sub_6161, view_2106);  sub_6161 = view_2106 = None
	        _assert_tensor_metadata_1213 = torch.ops.aten._assert_tensor_metadata.default(mul_13049, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1213 = None
	        view_2109 = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2110 = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2111 = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1214 = torch.ops.aten._assert_tensor_metadata.default(view_2109, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1214 = None
	        convert_element_type_808 = torch.ops.prims.convert_element_type.default(view_2109, torch.float32);  view_2109 = None
	        _assert_tensor_metadata_1215 = torch.ops.aten._assert_tensor_metadata.default(view_2111, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1215 = None
	        convert_element_type_809 = torch.ops.prims.convert_element_type.default(view_2111, torch.float32);  view_2111 = None
	        sub_6165 = torch.ops.aten.sub.Tensor(convert_element_type_808, convert_element_type_809);  convert_element_type_808 = convert_element_type_809 = None
	        mul_13054 = torch.ops.aten.mul.Tensor(sub_6165, view_2110);  sub_6165 = view_2110 = None
	        view_2112 = torch.ops.aten.view.default(mul_13054, [1280, 1280]);  mul_13054 = None
	        _assert_tensor_metadata_1216 = torch.ops.aten._assert_tensor_metadata.default(view_2112, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1216 = None
	        mul_13059 = sym_size_int * 1500
	        view_2113 = torch.ops.aten.view.default(mul_13049, [mul_13059, 1280]);  mul_13049 = mul_13059 = None
	        permute_225 = torch.ops.aten.permute.default(view_2112, [1, 0]);  view_2112 = None
	        addmm_111 = torch.ops.aten.addmm.default(model_audio_tower_layers_22_self_attn_v_proj_bias, view_2113, permute_225);  model_audio_tower_layers_22_self_attn_v_proj_bias = view_2113 = permute_225 = None
	        view_2114 = torch.ops.aten.view.default(addmm_111, [sym_size_int, 1500, 1280]);  addmm_111 = None
	        view_2115 = torch.ops.aten.view.default(view_2114, [sym_size_int, -1, 20, 64]);  view_2114 = None
	        permute_226 = torch.ops.aten.permute.default(view_2115, [0, 2, 1, 3]);  view_2115 = None
	        clone_180 = torch.ops.aten.clone.default(permute_226, memory_format = torch.contiguous_format);  permute_226 = None
	        _scaled_dot_product_efficient_attention_22 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_178, clone_179, clone_180, None, False, scale = 1.0);  clone_178 = clone_179 = clone_180 = None
	        getitem_178 = _scaled_dot_product_efficient_attention_22[0];  _scaled_dot_product_efficient_attention_22 = None
	        permute_227 = torch.ops.aten.permute.default(getitem_178, [0, 2, 1, 3]);  getitem_178 = None
	        view_2116 = torch.ops.aten.view.default(permute_227, [sym_size_int, 1500, -1]);  permute_227 = None
	        amin_135 = torch.ops.aten.amin.default(view_2116, [2])
	        amax_135 = torch.ops.aten.amax.default(view_2116, [2])
	        full_270 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_135 = torch.ops.aten.minimum.default(amin_135, full_270);  amin_135 = full_270 = None
	        full_271 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_135 = torch.ops.aten.maximum.default(amax_135, full_271);  amax_135 = full_271 = None
	        sub_6183 = torch.ops.aten.sub.Tensor(maximum_135, minimum_135);  maximum_135 = None
	        div_270 = torch.ops.aten.div.Tensor(sub_6183, 255.0);  sub_6183 = None
	        clamp_min_405 = torch.ops.aten.clamp_min.default(div_270, 1.1920928955078125e-07);  div_270 = None
	        div_271 = torch.ops.aten.div.Tensor(minimum_135, clamp_min_405);  minimum_135 = None
	        round_271 = torch.ops.aten.round.default(div_271);  div_271 = None
	        sub_6189 = torch.ops.aten.sub.Tensor(-128, round_271);  round_271 = None
	        clamp_min_406 = torch.ops.aten.clamp_min.default(sub_6189, -128);  sub_6189 = None
	        clamp_max_270 = torch.ops.aten.clamp_max.default(clamp_min_406, 127);  clamp_min_406 = None
	        _assert_tensor_metadata_1217 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_405, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1217 = None
	        _assert_tensor_metadata_1218 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_270, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1218 = None
	        convert_element_type_810 = torch.ops.prims.convert_element_type.default(clamp_max_270, torch.int8);  clamp_max_270 = None
	        view_2119 = torch.ops.aten.view.default(clamp_min_405, [sym_size_int, 1500, 1])
	        view_2120 = torch.ops.aten.view.default(convert_element_type_810, [sym_size_int, 1500, 1])
	        reciprocal_135 = torch.ops.aten.reciprocal.default(view_2119);  view_2119 = None
	        mul_13129 = torch.ops.aten.mul.Tensor(reciprocal_135, 1.0);  reciprocal_135 = None
	        mul_13132 = torch.ops.aten.mul.Tensor(view_2116, mul_13129);  view_2116 = mul_13129 = None
	        round_272 = torch.ops.aten.round.default(mul_13132);  mul_13132 = None
	        add_20779 = torch.ops.aten.add.Tensor(round_272, view_2120);  round_272 = view_2120 = None
	        clamp_min_407 = torch.ops.aten.clamp_min.default(add_20779, -128);  add_20779 = None
	        clamp_max_271 = torch.ops.aten.clamp_max.default(clamp_min_407, 127);  clamp_min_407 = None
	        _assert_tensor_metadata_1219 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_271, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1219 = None
	        convert_element_type_811 = torch.ops.prims.convert_element_type.default(clamp_max_271, torch.int8);  clamp_max_271 = None
	        view_2123 = torch.ops.aten.view.default(clamp_min_405, [sym_size_int, 1500, 1]);  clamp_min_405 = None
	        view_2124 = torch.ops.aten.view.default(convert_element_type_810, [sym_size_int, 1500, 1]);  convert_element_type_810 = None
	        _assert_tensor_metadata_1220 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_811, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1220 = None
	        convert_element_type_812 = torch.ops.prims.convert_element_type.default(convert_element_type_811, torch.float32);  convert_element_type_811 = None
	        _assert_tensor_metadata_1221 = torch.ops.aten._assert_tensor_metadata.default(view_2124, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1221 = None
	        convert_element_type_813 = torch.ops.prims.convert_element_type.default(view_2124, torch.float32);  view_2124 = None
	        sub_6209 = torch.ops.aten.sub.Tensor(convert_element_type_812, convert_element_type_813);  convert_element_type_812 = convert_element_type_813 = None
	        mul_13154 = torch.ops.aten.mul.Tensor(sub_6209, view_2123);  sub_6209 = view_2123 = None
	        _assert_tensor_metadata_1222 = torch.ops.aten._assert_tensor_metadata.default(mul_13154, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1222 = None
	        view_2126 = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2127 = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2128 = torch.ops.aten.view.default(model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1223 = torch.ops.aten._assert_tensor_metadata.default(view_2126, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1223 = None
	        convert_element_type_814 = torch.ops.prims.convert_element_type.default(view_2126, torch.float32);  view_2126 = None
	        _assert_tensor_metadata_1224 = torch.ops.aten._assert_tensor_metadata.default(view_2128, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1224 = None
	        convert_element_type_815 = torch.ops.prims.convert_element_type.default(view_2128, torch.float32);  view_2128 = None
	        sub_6213 = torch.ops.aten.sub.Tensor(convert_element_type_814, convert_element_type_815);  convert_element_type_814 = convert_element_type_815 = None
	        mul_13159 = torch.ops.aten.mul.Tensor(sub_6213, view_2127);  sub_6213 = view_2127 = None
	        view_2129 = torch.ops.aten.view.default(mul_13159, [1280, 1280]);  mul_13159 = None
	        _assert_tensor_metadata_1225 = torch.ops.aten._assert_tensor_metadata.default(view_2129, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1225 = None
	        mul_13164 = sym_size_int * 1500
	        view_2130 = torch.ops.aten.view.default(mul_13154, [mul_13164, 1280]);  mul_13154 = mul_13164 = None
	        permute_228 = torch.ops.aten.permute.default(view_2129, [1, 0]);  view_2129 = None
	        addmm_112 = torch.ops.aten.addmm.default(model_audio_tower_layers_22_self_attn_out_proj_bias, view_2130, permute_228);  model_audio_tower_layers_22_self_attn_out_proj_bias = view_2130 = permute_228 = None
	        view_2131 = torch.ops.aten.view.default(addmm_112, [sym_size_int, 1500, 1280]);  addmm_112 = None
	        add_20842 = torch.ops.aten.add.Tensor(add_20222, view_2131);  add_20222 = view_2131 = None
	        clone_182 = torch.ops.aten.clone.default(add_20842, memory_format = torch.contiguous_format)
	        var_mean_45 = torch.ops.aten.var_mean.correction(clone_182, [2], correction = 0, keepdim = True)
	        getitem_182 = var_mean_45[0]
	        getitem_183 = var_mean_45[1];  var_mean_45 = None
	        add_20847 = torch.ops.aten.add.Tensor(getitem_182, 1e-05);  getitem_182 = None
	        rsqrt_45 = torch.ops.aten.rsqrt.default(add_20847);  add_20847 = None
	        sub_6219 = torch.ops.aten.sub.Tensor(clone_182, getitem_183);  clone_182 = getitem_183 = None
	        mul_13175 = torch.ops.aten.mul.Tensor(sub_6219, rsqrt_45);  sub_6219 = rsqrt_45 = None
	        mul_13176 = torch.ops.aten.mul.Tensor(mul_13175, model_audio_tower_layers_22_final_layer_norm_weight);  mul_13175 = model_audio_tower_layers_22_final_layer_norm_weight = None
	        add_20848 = torch.ops.aten.add.Tensor(mul_13176, model_audio_tower_layers_22_final_layer_norm_bias);  mul_13176 = model_audio_tower_layers_22_final_layer_norm_bias = None
	        amin_136 = torch.ops.aten.amin.default(add_20848, [2])
	        amax_136 = torch.ops.aten.amax.default(add_20848, [2])
	        full_272 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_136 = torch.ops.aten.minimum.default(amin_136, full_272);  amin_136 = full_272 = None
	        full_273 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_136 = torch.ops.aten.maximum.default(amax_136, full_273);  amax_136 = full_273 = None
	        sub_6230 = torch.ops.aten.sub.Tensor(maximum_136, minimum_136);  maximum_136 = None
	        div_272 = torch.ops.aten.div.Tensor(sub_6230, 255.0);  sub_6230 = None
	        clamp_min_408 = torch.ops.aten.clamp_min.default(div_272, 1.1920928955078125e-07);  div_272 = None
	        div_273 = torch.ops.aten.div.Tensor(minimum_136, clamp_min_408);  minimum_136 = None
	        round_273 = torch.ops.aten.round.default(div_273);  div_273 = None
	        sub_6236 = torch.ops.aten.sub.Tensor(-128, round_273);  round_273 = None
	        clamp_min_409 = torch.ops.aten.clamp_min.default(sub_6236, -128);  sub_6236 = None
	        clamp_max_272 = torch.ops.aten.clamp_max.default(clamp_min_409, 127);  clamp_min_409 = None
	        _assert_tensor_metadata_1226 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_408, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1226 = None
	        _assert_tensor_metadata_1227 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_272, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1227 = None
	        convert_element_type_816 = torch.ops.prims.convert_element_type.default(clamp_max_272, torch.int8);  clamp_max_272 = None
	        view_2134 = torch.ops.aten.view.default(clamp_min_408, [sym_size_int, 1500, 1])
	        view_2135 = torch.ops.aten.view.default(convert_element_type_816, [sym_size_int, 1500, 1])
	        reciprocal_136 = torch.ops.aten.reciprocal.default(view_2134);  view_2134 = None
	        mul_13224 = torch.ops.aten.mul.Tensor(reciprocal_136, 1.0);  reciprocal_136 = None
	        mul_13227 = torch.ops.aten.mul.Tensor(add_20848, mul_13224);  add_20848 = mul_13224 = None
	        round_274 = torch.ops.aten.round.default(mul_13227);  mul_13227 = None
	        add_20935 = torch.ops.aten.add.Tensor(round_274, view_2135);  round_274 = view_2135 = None
	        clamp_min_410 = torch.ops.aten.clamp_min.default(add_20935, -128);  add_20935 = None
	        clamp_max_273 = torch.ops.aten.clamp_max.default(clamp_min_410, 127);  clamp_min_410 = None
	        _assert_tensor_metadata_1228 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_273, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1228 = None
	        convert_element_type_817 = torch.ops.prims.convert_element_type.default(clamp_max_273, torch.int8);  clamp_max_273 = None
	        view_2138 = torch.ops.aten.view.default(clamp_min_408, [sym_size_int, 1500, 1]);  clamp_min_408 = None
	        view_2139 = torch.ops.aten.view.default(convert_element_type_816, [sym_size_int, 1500, 1]);  convert_element_type_816 = None
	        _assert_tensor_metadata_1229 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_817, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1229 = None
	        convert_element_type_818 = torch.ops.prims.convert_element_type.default(convert_element_type_817, torch.float32);  convert_element_type_817 = None
	        _assert_tensor_metadata_1230 = torch.ops.aten._assert_tensor_metadata.default(view_2139, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1230 = None
	        convert_element_type_819 = torch.ops.prims.convert_element_type.default(view_2139, torch.float32);  view_2139 = None
	        sub_6256 = torch.ops.aten.sub.Tensor(convert_element_type_818, convert_element_type_819);  convert_element_type_818 = convert_element_type_819 = None
	        mul_13249 = torch.ops.aten.mul.Tensor(sub_6256, view_2138);  sub_6256 = view_2138 = None
	        _assert_tensor_metadata_1231 = torch.ops.aten._assert_tensor_metadata.default(mul_13249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1231 = None
	        view_2141 = torch.ops.aten.view.default(model_audio_tower_layers_22_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_22_fc1_parametrizations_weight_original0 = None
	        view_2142 = torch.ops.aten.view.default(model_audio_tower_layers_22_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_22_fc1_parametrizations_weight_original1 = None
	        view_2143 = torch.ops.aten.view.default(model_audio_tower_layers_22_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_22_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1232 = torch.ops.aten._assert_tensor_metadata.default(view_2141, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1232 = None
	        convert_element_type_820 = torch.ops.prims.convert_element_type.default(view_2141, torch.float32);  view_2141 = None
	        _assert_tensor_metadata_1233 = torch.ops.aten._assert_tensor_metadata.default(view_2143, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1233 = None
	        convert_element_type_821 = torch.ops.prims.convert_element_type.default(view_2143, torch.float32);  view_2143 = None
	        sub_6260 = torch.ops.aten.sub.Tensor(convert_element_type_820, convert_element_type_821);  convert_element_type_820 = convert_element_type_821 = None
	        mul_13254 = torch.ops.aten.mul.Tensor(sub_6260, view_2142);  sub_6260 = view_2142 = None
	        view_2144 = torch.ops.aten.view.default(mul_13254, [5120, 1280]);  mul_13254 = None
	        _assert_tensor_metadata_1234 = torch.ops.aten._assert_tensor_metadata.default(view_2144, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1234 = None
	        mul_13259 = sym_size_int * 1500
	        view_2145 = torch.ops.aten.view.default(mul_13249, [mul_13259, 1280]);  mul_13249 = mul_13259 = None
	        permute_229 = torch.ops.aten.permute.default(view_2144, [1, 0]);  view_2144 = None
	        addmm_113 = torch.ops.aten.addmm.default(model_audio_tower_layers_22_fc1_bias, view_2145, permute_229);  model_audio_tower_layers_22_fc1_bias = view_2145 = permute_229 = None
	        view_2146 = torch.ops.aten.view.default(addmm_113, [sym_size_int, 1500, 5120]);  addmm_113 = None
	        mul_13266 = torch.ops.aten.mul.Tensor(view_2146, 0.5)
	        mul_13267 = torch.ops.aten.mul.Tensor(view_2146, 0.7071067811865476);  view_2146 = None
	        erf_24 = torch.ops.aten.erf.default(mul_13267);  mul_13267 = None
	        add_20994 = torch.ops.aten.add.Tensor(erf_24, 1);  erf_24 = None
	        mul_13268 = torch.ops.aten.mul.Tensor(mul_13266, add_20994);  mul_13266 = add_20994 = None
	        amin_137 = torch.ops.aten.amin.default(mul_13268, [2])
	        amax_137 = torch.ops.aten.amax.default(mul_13268, [2])
	        full_274 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_137 = torch.ops.aten.minimum.default(amin_137, full_274);  amin_137 = full_274 = None
	        full_275 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_137 = torch.ops.aten.maximum.default(amax_137, full_275);  amax_137 = full_275 = None
	        sub_6273 = torch.ops.aten.sub.Tensor(maximum_137, minimum_137);  maximum_137 = None
	        div_274 = torch.ops.aten.div.Tensor(sub_6273, 255.0);  sub_6273 = None
	        clamp_min_411 = torch.ops.aten.clamp_min.default(div_274, 1.1920928955078125e-07);  div_274 = None
	        div_275 = torch.ops.aten.div.Tensor(minimum_137, clamp_min_411);  minimum_137 = None
	        round_275 = torch.ops.aten.round.default(div_275);  div_275 = None
	        sub_6279 = torch.ops.aten.sub.Tensor(-128, round_275);  round_275 = None
	        clamp_min_412 = torch.ops.aten.clamp_min.default(sub_6279, -128);  sub_6279 = None
	        clamp_max_274 = torch.ops.aten.clamp_max.default(clamp_min_412, 127);  clamp_min_412 = None
	        _assert_tensor_metadata_1235 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_411, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1235 = None
	        _assert_tensor_metadata_1236 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_274, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1236 = None
	        convert_element_type_822 = torch.ops.prims.convert_element_type.default(clamp_max_274, torch.int8);  clamp_max_274 = None
	        view_2149 = torch.ops.aten.view.default(clamp_min_411, [sym_size_int, 1500, 1])
	        view_2150 = torch.ops.aten.view.default(convert_element_type_822, [sym_size_int, 1500, 1])
	        reciprocal_137 = torch.ops.aten.reciprocal.default(view_2149);  view_2149 = None
	        mul_13314 = torch.ops.aten.mul.Tensor(reciprocal_137, 1.0);  reciprocal_137 = None
	        mul_13317 = torch.ops.aten.mul.Tensor(mul_13268, mul_13314);  mul_13268 = mul_13314 = None
	        round_276 = torch.ops.aten.round.default(mul_13317);  mul_13317 = None
	        add_21077 = torch.ops.aten.add.Tensor(round_276, view_2150);  round_276 = view_2150 = None
	        clamp_min_413 = torch.ops.aten.clamp_min.default(add_21077, -128);  add_21077 = None
	        clamp_max_275 = torch.ops.aten.clamp_max.default(clamp_min_413, 127);  clamp_min_413 = None
	        _assert_tensor_metadata_1237 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_275, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1237 = None
	        convert_element_type_823 = torch.ops.prims.convert_element_type.default(clamp_max_275, torch.int8);  clamp_max_275 = None
	        view_2153 = torch.ops.aten.view.default(clamp_min_411, [sym_size_int, 1500, 1]);  clamp_min_411 = None
	        view_2154 = torch.ops.aten.view.default(convert_element_type_822, [sym_size_int, 1500, 1]);  convert_element_type_822 = None
	        _assert_tensor_metadata_1238 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_823, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1238 = None
	        convert_element_type_824 = torch.ops.prims.convert_element_type.default(convert_element_type_823, torch.float32);  convert_element_type_823 = None
	        _assert_tensor_metadata_1239 = torch.ops.aten._assert_tensor_metadata.default(view_2154, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1239 = None
	        convert_element_type_825 = torch.ops.prims.convert_element_type.default(view_2154, torch.float32);  view_2154 = None
	        sub_6299 = torch.ops.aten.sub.Tensor(convert_element_type_824, convert_element_type_825);  convert_element_type_824 = convert_element_type_825 = None
	        mul_13339 = torch.ops.aten.mul.Tensor(sub_6299, view_2153);  sub_6299 = view_2153 = None
	        _assert_tensor_metadata_1240 = torch.ops.aten._assert_tensor_metadata.default(mul_13339, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1240 = None
	        view_2156 = torch.ops.aten.view.default(model_audio_tower_layers_22_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_22_fc2_parametrizations_weight_original0 = None
	        view_2157 = torch.ops.aten.view.default(model_audio_tower_layers_22_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_22_fc2_parametrizations_weight_original1 = None
	        view_2158 = torch.ops.aten.view.default(model_audio_tower_layers_22_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_22_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1241 = torch.ops.aten._assert_tensor_metadata.default(view_2156, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1241 = None
	        convert_element_type_826 = torch.ops.prims.convert_element_type.default(view_2156, torch.float32);  view_2156 = None
	        _assert_tensor_metadata_1242 = torch.ops.aten._assert_tensor_metadata.default(view_2158, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1242 = None
	        convert_element_type_827 = torch.ops.prims.convert_element_type.default(view_2158, torch.float32);  view_2158 = None
	        sub_6303 = torch.ops.aten.sub.Tensor(convert_element_type_826, convert_element_type_827);  convert_element_type_826 = convert_element_type_827 = None
	        mul_13344 = torch.ops.aten.mul.Tensor(sub_6303, view_2157);  sub_6303 = view_2157 = None
	        view_2159 = torch.ops.aten.view.default(mul_13344, [1280, 5120]);  mul_13344 = None
	        _assert_tensor_metadata_1243 = torch.ops.aten._assert_tensor_metadata.default(view_2159, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1243 = None
	        mul_13349 = sym_size_int * 1500
	        view_2160 = torch.ops.aten.view.default(mul_13339, [mul_13349, 5120]);  mul_13339 = mul_13349 = None
	        permute_230 = torch.ops.aten.permute.default(view_2159, [1, 0]);  view_2159 = None
	        addmm_114 = torch.ops.aten.addmm.default(model_audio_tower_layers_22_fc2_bias, view_2160, permute_230);  model_audio_tower_layers_22_fc2_bias = view_2160 = permute_230 = None
	        view_2161 = torch.ops.aten.view.default(addmm_114, [sym_size_int, 1500, 1280]);  addmm_114 = None
	        add_21140 = torch.ops.aten.add.Tensor(add_20842, view_2161);  add_20842 = view_2161 = None
	        clone_185 = torch.ops.aten.clone.default(add_21140, memory_format = torch.contiguous_format)
	        var_mean_46 = torch.ops.aten.var_mean.correction(clone_185, [2], correction = 0, keepdim = True)
	        getitem_184 = var_mean_46[0]
	        getitem_185 = var_mean_46[1];  var_mean_46 = None
	        add_21145 = torch.ops.aten.add.Tensor(getitem_184, 1e-05);  getitem_184 = None
	        rsqrt_46 = torch.ops.aten.rsqrt.default(add_21145);  add_21145 = None
	        sub_6309 = torch.ops.aten.sub.Tensor(clone_185, getitem_185);  clone_185 = getitem_185 = None
	        mul_13360 = torch.ops.aten.mul.Tensor(sub_6309, rsqrt_46);  sub_6309 = rsqrt_46 = None
	        mul_13361 = torch.ops.aten.mul.Tensor(mul_13360, model_audio_tower_layers_23_self_attn_layer_norm_weight);  mul_13360 = model_audio_tower_layers_23_self_attn_layer_norm_weight = None
	        add_21146 = torch.ops.aten.add.Tensor(mul_13361, model_audio_tower_layers_23_self_attn_layer_norm_bias);  mul_13361 = model_audio_tower_layers_23_self_attn_layer_norm_bias = None
	        amin_138 = torch.ops.aten.amin.default(add_21146, [2])
	        amax_138 = torch.ops.aten.amax.default(add_21146, [2])
	        full_276 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_138 = torch.ops.aten.minimum.default(amin_138, full_276);  amin_138 = full_276 = None
	        full_277 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_138 = torch.ops.aten.maximum.default(amax_138, full_277);  amax_138 = full_277 = None
	        sub_6320 = torch.ops.aten.sub.Tensor(maximum_138, minimum_138);  maximum_138 = None
	        div_276 = torch.ops.aten.div.Tensor(sub_6320, 255.0);  sub_6320 = None
	        clamp_min_414 = torch.ops.aten.clamp_min.default(div_276, 1.1920928955078125e-07);  div_276 = None
	        div_277 = torch.ops.aten.div.Tensor(minimum_138, clamp_min_414);  minimum_138 = None
	        round_277 = torch.ops.aten.round.default(div_277);  div_277 = None
	        sub_6326 = torch.ops.aten.sub.Tensor(-128, round_277);  round_277 = None
	        clamp_min_415 = torch.ops.aten.clamp_min.default(sub_6326, -128);  sub_6326 = None
	        clamp_max_276 = torch.ops.aten.clamp_max.default(clamp_min_415, 127);  clamp_min_415 = None
	        _assert_tensor_metadata_1244 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_414, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1244 = None
	        _assert_tensor_metadata_1245 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_276, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1245 = None
	        convert_element_type_828 = torch.ops.prims.convert_element_type.default(clamp_max_276, torch.int8);  clamp_max_276 = None
	        view_2164 = torch.ops.aten.view.default(clamp_min_414, [sym_size_int, 1500, 1])
	        view_2165 = torch.ops.aten.view.default(convert_element_type_828, [sym_size_int, 1500, 1])
	        reciprocal_138 = torch.ops.aten.reciprocal.default(view_2164);  view_2164 = None
	        mul_13409 = torch.ops.aten.mul.Tensor(reciprocal_138, 1.0);  reciprocal_138 = None
	        mul_13412 = torch.ops.aten.mul.Tensor(add_21146, mul_13409);  mul_13409 = None
	        round_278 = torch.ops.aten.round.default(mul_13412);  mul_13412 = None
	        add_21233 = torch.ops.aten.add.Tensor(round_278, view_2165);  round_278 = view_2165 = None
	        clamp_min_416 = torch.ops.aten.clamp_min.default(add_21233, -128);  add_21233 = None
	        clamp_max_277 = torch.ops.aten.clamp_max.default(clamp_min_416, 127);  clamp_min_416 = None
	        _assert_tensor_metadata_1246 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_277, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1246 = None
	        convert_element_type_829 = torch.ops.prims.convert_element_type.default(clamp_max_277, torch.int8);  clamp_max_277 = None
	        view_2168 = torch.ops.aten.view.default(clamp_min_414, [sym_size_int, 1500, 1]);  clamp_min_414 = None
	        view_2169 = torch.ops.aten.view.default(convert_element_type_828, [sym_size_int, 1500, 1]);  convert_element_type_828 = None
	        _assert_tensor_metadata_1247 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_829, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1247 = None
	        convert_element_type_830 = torch.ops.prims.convert_element_type.default(convert_element_type_829, torch.float32);  convert_element_type_829 = None
	        _assert_tensor_metadata_1248 = torch.ops.aten._assert_tensor_metadata.default(view_2169, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1248 = None
	        convert_element_type_831 = torch.ops.prims.convert_element_type.default(view_2169, torch.float32);  view_2169 = None
	        sub_6346 = torch.ops.aten.sub.Tensor(convert_element_type_830, convert_element_type_831);  convert_element_type_830 = convert_element_type_831 = None
	        mul_13434 = torch.ops.aten.mul.Tensor(sub_6346, view_2168);  sub_6346 = view_2168 = None
	        _assert_tensor_metadata_1249 = torch.ops.aten._assert_tensor_metadata.default(mul_13434, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1249 = None
	        view_2171 = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2172 = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2173 = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1250 = torch.ops.aten._assert_tensor_metadata.default(view_2171, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1250 = None
	        convert_element_type_832 = torch.ops.prims.convert_element_type.default(view_2171, torch.float32);  view_2171 = None
	        _assert_tensor_metadata_1251 = torch.ops.aten._assert_tensor_metadata.default(view_2173, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1251 = None
	        convert_element_type_833 = torch.ops.prims.convert_element_type.default(view_2173, torch.float32);  view_2173 = None
	        sub_6350 = torch.ops.aten.sub.Tensor(convert_element_type_832, convert_element_type_833);  convert_element_type_832 = convert_element_type_833 = None
	        mul_13439 = torch.ops.aten.mul.Tensor(sub_6350, view_2172);  sub_6350 = view_2172 = None
	        view_2174 = torch.ops.aten.view.default(mul_13439, [1280, 1280]);  mul_13439 = None
	        _assert_tensor_metadata_1252 = torch.ops.aten._assert_tensor_metadata.default(view_2174, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1252 = None
	        mul_13444 = sym_size_int * 1500
	        view_2175 = torch.ops.aten.view.default(mul_13434, [mul_13444, 1280]);  mul_13434 = mul_13444 = None
	        permute_231 = torch.ops.aten.permute.default(view_2174, [1, 0]);  view_2174 = None
	        addmm_115 = torch.ops.aten.addmm.default(model_audio_tower_layers_23_self_attn_q_proj_bias, view_2175, permute_231);  model_audio_tower_layers_23_self_attn_q_proj_bias = view_2175 = permute_231 = None
	        view_2176 = torch.ops.aten.view.default(addmm_115, [sym_size_int, 1500, 1280]);  addmm_115 = None
	        mul_13451 = torch.ops.aten.mul.Tensor(view_2176, 0.125);  view_2176 = None
	        view_2177 = torch.ops.aten.view.default(mul_13451, [sym_size_int, 1500, 20, 64]);  mul_13451 = None
	        permute_232 = torch.ops.aten.permute.default(view_2177, [0, 2, 1, 3]);  view_2177 = None
	        clone_186 = torch.ops.aten.clone.default(permute_232, memory_format = torch.contiguous_format);  permute_232 = None
	        amin_139 = torch.ops.aten.amin.default(add_21146, [2])
	        amax_139 = torch.ops.aten.amax.default(add_21146, [2])
	        full_278 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_139 = torch.ops.aten.minimum.default(amin_139, full_278);  amin_139 = full_278 = None
	        full_279 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_139 = torch.ops.aten.maximum.default(amax_139, full_279);  amax_139 = full_279 = None
	        sub_6365 = torch.ops.aten.sub.Tensor(maximum_139, minimum_139);  maximum_139 = None
	        div_278 = torch.ops.aten.div.Tensor(sub_6365, 255.0);  sub_6365 = None
	        clamp_min_417 = torch.ops.aten.clamp_min.default(div_278, 1.1920928955078125e-07);  div_278 = None
	        div_279 = torch.ops.aten.div.Tensor(minimum_139, clamp_min_417);  minimum_139 = None
	        round_279 = torch.ops.aten.round.default(div_279);  div_279 = None
	        sub_6371 = torch.ops.aten.sub.Tensor(-128, round_279);  round_279 = None
	        clamp_min_418 = torch.ops.aten.clamp_min.default(sub_6371, -128);  sub_6371 = None
	        clamp_max_278 = torch.ops.aten.clamp_max.default(clamp_min_418, 127);  clamp_min_418 = None
	        _assert_tensor_metadata_1253 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_417, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1253 = None
	        _assert_tensor_metadata_1254 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_278, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1254 = None
	        convert_element_type_834 = torch.ops.prims.convert_element_type.default(clamp_max_278, torch.int8);  clamp_max_278 = None
	        view_2180 = torch.ops.aten.view.default(clamp_min_417, [sym_size_int, 1500, 1])
	        view_2181 = torch.ops.aten.view.default(convert_element_type_834, [sym_size_int, 1500, 1])
	        reciprocal_139 = torch.ops.aten.reciprocal.default(view_2180);  view_2180 = None
	        mul_13505 = torch.ops.aten.mul.Tensor(reciprocal_139, 1.0);  reciprocal_139 = None
	        mul_13508 = torch.ops.aten.mul.Tensor(add_21146, mul_13505);  mul_13505 = None
	        round_280 = torch.ops.aten.round.default(mul_13508);  mul_13508 = None
	        add_21385 = torch.ops.aten.add.Tensor(round_280, view_2181);  round_280 = view_2181 = None
	        clamp_min_419 = torch.ops.aten.clamp_min.default(add_21385, -128);  add_21385 = None
	        clamp_max_279 = torch.ops.aten.clamp_max.default(clamp_min_419, 127);  clamp_min_419 = None
	        _assert_tensor_metadata_1255 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_279, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1255 = None
	        convert_element_type_835 = torch.ops.prims.convert_element_type.default(clamp_max_279, torch.int8);  clamp_max_279 = None
	        view_2184 = torch.ops.aten.view.default(clamp_min_417, [sym_size_int, 1500, 1]);  clamp_min_417 = None
	        view_2185 = torch.ops.aten.view.default(convert_element_type_834, [sym_size_int, 1500, 1]);  convert_element_type_834 = None
	        _assert_tensor_metadata_1256 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_835, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1256 = None
	        convert_element_type_836 = torch.ops.prims.convert_element_type.default(convert_element_type_835, torch.float32);  convert_element_type_835 = None
	        _assert_tensor_metadata_1257 = torch.ops.aten._assert_tensor_metadata.default(view_2185, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1257 = None
	        convert_element_type_837 = torch.ops.prims.convert_element_type.default(view_2185, torch.float32);  view_2185 = None
	        sub_6391 = torch.ops.aten.sub.Tensor(convert_element_type_836, convert_element_type_837);  convert_element_type_836 = convert_element_type_837 = None
	        mul_13530 = torch.ops.aten.mul.Tensor(sub_6391, view_2184);  sub_6391 = view_2184 = None
	        _assert_tensor_metadata_1258 = torch.ops.aten._assert_tensor_metadata.default(mul_13530, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1258 = None
	        view_2187 = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2188 = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2189 = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1259 = torch.ops.aten._assert_tensor_metadata.default(view_2187, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1259 = None
	        convert_element_type_838 = torch.ops.prims.convert_element_type.default(view_2187, torch.float32);  view_2187 = None
	        _assert_tensor_metadata_1260 = torch.ops.aten._assert_tensor_metadata.default(view_2189, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1260 = None
	        convert_element_type_839 = torch.ops.prims.convert_element_type.default(view_2189, torch.float32);  view_2189 = None
	        sub_6395 = torch.ops.aten.sub.Tensor(convert_element_type_838, convert_element_type_839);  convert_element_type_838 = convert_element_type_839 = None
	        mul_13535 = torch.ops.aten.mul.Tensor(sub_6395, view_2188);  sub_6395 = view_2188 = None
	        view_2190 = torch.ops.aten.view.default(mul_13535, [1280, 1280]);  mul_13535 = None
	        _assert_tensor_metadata_1261 = torch.ops.aten._assert_tensor_metadata.default(view_2190, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1261 = None
	        permute_233 = torch.ops.aten.permute.default(view_2190, [1, 0]);  view_2190 = None
	        mul_13538 = sym_size_int * 1500
	        view_2191 = torch.ops.aten.view.default(mul_13530, [mul_13538, 1280]);  mul_13530 = mul_13538 = None
	        mm_23 = torch.ops.aten.mm.default(view_2191, permute_233);  view_2191 = permute_233 = None
	        view_2192 = torch.ops.aten.view.default(mm_23, [sym_size_int, 1500, 1280]);  mm_23 = None
	        view_2193 = torch.ops.aten.view.default(view_2192, [sym_size_int, -1, 20, 64]);  view_2192 = None
	        permute_234 = torch.ops.aten.permute.default(view_2193, [0, 2, 1, 3]);  view_2193 = None
	        clone_187 = torch.ops.aten.clone.default(permute_234, memory_format = torch.contiguous_format);  permute_234 = None
	        amin_140 = torch.ops.aten.amin.default(add_21146, [2])
	        amax_140 = torch.ops.aten.amax.default(add_21146, [2])
	        full_280 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_140 = torch.ops.aten.minimum.default(amin_140, full_280);  amin_140 = full_280 = None
	        full_281 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_140 = torch.ops.aten.maximum.default(amax_140, full_281);  amax_140 = full_281 = None
	        sub_6409 = torch.ops.aten.sub.Tensor(maximum_140, minimum_140);  maximum_140 = None
	        div_280 = torch.ops.aten.div.Tensor(sub_6409, 255.0);  sub_6409 = None
	        clamp_min_420 = torch.ops.aten.clamp_min.default(div_280, 1.1920928955078125e-07);  div_280 = None
	        div_281 = torch.ops.aten.div.Tensor(minimum_140, clamp_min_420);  minimum_140 = None
	        round_281 = torch.ops.aten.round.default(div_281);  div_281 = None
	        sub_6415 = torch.ops.aten.sub.Tensor(-128, round_281);  round_281 = None
	        clamp_min_421 = torch.ops.aten.clamp_min.default(sub_6415, -128);  sub_6415 = None
	        clamp_max_280 = torch.ops.aten.clamp_max.default(clamp_min_421, 127);  clamp_min_421 = None
	        _assert_tensor_metadata_1262 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_420, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1262 = None
	        _assert_tensor_metadata_1263 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_280, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1263 = None
	        convert_element_type_840 = torch.ops.prims.convert_element_type.default(clamp_max_280, torch.int8);  clamp_max_280 = None
	        view_2196 = torch.ops.aten.view.default(clamp_min_420, [sym_size_int, 1500, 1])
	        view_2197 = torch.ops.aten.view.default(convert_element_type_840, [sym_size_int, 1500, 1])
	        reciprocal_140 = torch.ops.aten.reciprocal.default(view_2196);  view_2196 = None
	        mul_13604 = torch.ops.aten.mul.Tensor(reciprocal_140, 1.0);  reciprocal_140 = None
	        mul_13607 = torch.ops.aten.mul.Tensor(add_21146, mul_13604);  add_21146 = mul_13604 = None
	        round_282 = torch.ops.aten.round.default(mul_13607);  mul_13607 = None
	        add_21533 = torch.ops.aten.add.Tensor(round_282, view_2197);  round_282 = view_2197 = None
	        clamp_min_422 = torch.ops.aten.clamp_min.default(add_21533, -128);  add_21533 = None
	        clamp_max_281 = torch.ops.aten.clamp_max.default(clamp_min_422, 127);  clamp_min_422 = None
	        _assert_tensor_metadata_1264 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_281, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1264 = None
	        convert_element_type_841 = torch.ops.prims.convert_element_type.default(clamp_max_281, torch.int8);  clamp_max_281 = None
	        view_2200 = torch.ops.aten.view.default(clamp_min_420, [sym_size_int, 1500, 1]);  clamp_min_420 = None
	        view_2201 = torch.ops.aten.view.default(convert_element_type_840, [sym_size_int, 1500, 1]);  convert_element_type_840 = None
	        _assert_tensor_metadata_1265 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_841, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1265 = None
	        convert_element_type_842 = torch.ops.prims.convert_element_type.default(convert_element_type_841, torch.float32);  convert_element_type_841 = None
	        _assert_tensor_metadata_1266 = torch.ops.aten._assert_tensor_metadata.default(view_2201, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1266 = None
	        convert_element_type_843 = torch.ops.prims.convert_element_type.default(view_2201, torch.float32);  view_2201 = None
	        sub_6435 = torch.ops.aten.sub.Tensor(convert_element_type_842, convert_element_type_843);  convert_element_type_842 = convert_element_type_843 = None
	        mul_13629 = torch.ops.aten.mul.Tensor(sub_6435, view_2200);  sub_6435 = view_2200 = None
	        _assert_tensor_metadata_1267 = torch.ops.aten._assert_tensor_metadata.default(mul_13629, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1267 = None
	        view_2203 = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2204 = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2205 = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1268 = torch.ops.aten._assert_tensor_metadata.default(view_2203, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1268 = None
	        convert_element_type_844 = torch.ops.prims.convert_element_type.default(view_2203, torch.float32);  view_2203 = None
	        _assert_tensor_metadata_1269 = torch.ops.aten._assert_tensor_metadata.default(view_2205, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1269 = None
	        convert_element_type_845 = torch.ops.prims.convert_element_type.default(view_2205, torch.float32);  view_2205 = None
	        sub_6439 = torch.ops.aten.sub.Tensor(convert_element_type_844, convert_element_type_845);  convert_element_type_844 = convert_element_type_845 = None
	        mul_13634 = torch.ops.aten.mul.Tensor(sub_6439, view_2204);  sub_6439 = view_2204 = None
	        view_2206 = torch.ops.aten.view.default(mul_13634, [1280, 1280]);  mul_13634 = None
	        _assert_tensor_metadata_1270 = torch.ops.aten._assert_tensor_metadata.default(view_2206, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1270 = None
	        mul_13639 = sym_size_int * 1500
	        view_2207 = torch.ops.aten.view.default(mul_13629, [mul_13639, 1280]);  mul_13629 = mul_13639 = None
	        permute_235 = torch.ops.aten.permute.default(view_2206, [1, 0]);  view_2206 = None
	        addmm_116 = torch.ops.aten.addmm.default(model_audio_tower_layers_23_self_attn_v_proj_bias, view_2207, permute_235);  model_audio_tower_layers_23_self_attn_v_proj_bias = view_2207 = permute_235 = None
	        view_2208 = torch.ops.aten.view.default(addmm_116, [sym_size_int, 1500, 1280]);  addmm_116 = None
	        view_2209 = torch.ops.aten.view.default(view_2208, [sym_size_int, -1, 20, 64]);  view_2208 = None
	        permute_236 = torch.ops.aten.permute.default(view_2209, [0, 2, 1, 3]);  view_2209 = None
	        clone_188 = torch.ops.aten.clone.default(permute_236, memory_format = torch.contiguous_format);  permute_236 = None
	        _scaled_dot_product_efficient_attention_23 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_186, clone_187, clone_188, None, False, scale = 1.0);  clone_186 = clone_187 = clone_188 = None
	        getitem_186 = _scaled_dot_product_efficient_attention_23[0];  _scaled_dot_product_efficient_attention_23 = None
	        permute_237 = torch.ops.aten.permute.default(getitem_186, [0, 2, 1, 3]);  getitem_186 = None
	        view_2210 = torch.ops.aten.view.default(permute_237, [sym_size_int, 1500, -1]);  permute_237 = None
	        amin_141 = torch.ops.aten.amin.default(view_2210, [2])
	        amax_141 = torch.ops.aten.amax.default(view_2210, [2])
	        full_282 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_141 = torch.ops.aten.minimum.default(amin_141, full_282);  amin_141 = full_282 = None
	        full_283 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_141 = torch.ops.aten.maximum.default(amax_141, full_283);  amax_141 = full_283 = None
	        sub_6457 = torch.ops.aten.sub.Tensor(maximum_141, minimum_141);  maximum_141 = None
	        div_282 = torch.ops.aten.div.Tensor(sub_6457, 255.0);  sub_6457 = None
	        clamp_min_423 = torch.ops.aten.clamp_min.default(div_282, 1.1920928955078125e-07);  div_282 = None
	        div_283 = torch.ops.aten.div.Tensor(minimum_141, clamp_min_423);  minimum_141 = None
	        round_283 = torch.ops.aten.round.default(div_283);  div_283 = None
	        sub_6463 = torch.ops.aten.sub.Tensor(-128, round_283);  round_283 = None
	        clamp_min_424 = torch.ops.aten.clamp_min.default(sub_6463, -128);  sub_6463 = None
	        clamp_max_282 = torch.ops.aten.clamp_max.default(clamp_min_424, 127);  clamp_min_424 = None
	        _assert_tensor_metadata_1271 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_423, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1271 = None
	        _assert_tensor_metadata_1272 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_282, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1272 = None
	        convert_element_type_846 = torch.ops.prims.convert_element_type.default(clamp_max_282, torch.int8);  clamp_max_282 = None
	        view_2213 = torch.ops.aten.view.default(clamp_min_423, [sym_size_int, 1500, 1])
	        view_2214 = torch.ops.aten.view.default(convert_element_type_846, [sym_size_int, 1500, 1])
	        reciprocal_141 = torch.ops.aten.reciprocal.default(view_2213);  view_2213 = None
	        mul_13709 = torch.ops.aten.mul.Tensor(reciprocal_141, 1.0);  reciprocal_141 = None
	        mul_13712 = torch.ops.aten.mul.Tensor(view_2210, mul_13709);  view_2210 = mul_13709 = None
	        round_284 = torch.ops.aten.round.default(mul_13712);  mul_13712 = None
	        add_21697 = torch.ops.aten.add.Tensor(round_284, view_2214);  round_284 = view_2214 = None
	        clamp_min_425 = torch.ops.aten.clamp_min.default(add_21697, -128);  add_21697 = None
	        clamp_max_283 = torch.ops.aten.clamp_max.default(clamp_min_425, 127);  clamp_min_425 = None
	        _assert_tensor_metadata_1273 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_283, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1273 = None
	        convert_element_type_847 = torch.ops.prims.convert_element_type.default(clamp_max_283, torch.int8);  clamp_max_283 = None
	        view_2217 = torch.ops.aten.view.default(clamp_min_423, [sym_size_int, 1500, 1]);  clamp_min_423 = None
	        view_2218 = torch.ops.aten.view.default(convert_element_type_846, [sym_size_int, 1500, 1]);  convert_element_type_846 = None
	        _assert_tensor_metadata_1274 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_847, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1274 = None
	        convert_element_type_848 = torch.ops.prims.convert_element_type.default(convert_element_type_847, torch.float32);  convert_element_type_847 = None
	        _assert_tensor_metadata_1275 = torch.ops.aten._assert_tensor_metadata.default(view_2218, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1275 = None
	        convert_element_type_849 = torch.ops.prims.convert_element_type.default(view_2218, torch.float32);  view_2218 = None
	        sub_6483 = torch.ops.aten.sub.Tensor(convert_element_type_848, convert_element_type_849);  convert_element_type_848 = convert_element_type_849 = None
	        mul_13734 = torch.ops.aten.mul.Tensor(sub_6483, view_2217);  sub_6483 = view_2217 = None
	        _assert_tensor_metadata_1276 = torch.ops.aten._assert_tensor_metadata.default(mul_13734, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1276 = None
	        view_2220 = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2221 = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2222 = torch.ops.aten.view.default(model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1277 = torch.ops.aten._assert_tensor_metadata.default(view_2220, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1277 = None
	        convert_element_type_850 = torch.ops.prims.convert_element_type.default(view_2220, torch.float32);  view_2220 = None
	        _assert_tensor_metadata_1278 = torch.ops.aten._assert_tensor_metadata.default(view_2222, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1278 = None
	        convert_element_type_851 = torch.ops.prims.convert_element_type.default(view_2222, torch.float32);  view_2222 = None
	        sub_6487 = torch.ops.aten.sub.Tensor(convert_element_type_850, convert_element_type_851);  convert_element_type_850 = convert_element_type_851 = None
	        mul_13739 = torch.ops.aten.mul.Tensor(sub_6487, view_2221);  sub_6487 = view_2221 = None
	        view_2223 = torch.ops.aten.view.default(mul_13739, [1280, 1280]);  mul_13739 = None
	        _assert_tensor_metadata_1279 = torch.ops.aten._assert_tensor_metadata.default(view_2223, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1279 = None
	        mul_13744 = sym_size_int * 1500
	        view_2224 = torch.ops.aten.view.default(mul_13734, [mul_13744, 1280]);  mul_13734 = mul_13744 = None
	        permute_238 = torch.ops.aten.permute.default(view_2223, [1, 0]);  view_2223 = None
	        addmm_117 = torch.ops.aten.addmm.default(model_audio_tower_layers_23_self_attn_out_proj_bias, view_2224, permute_238);  model_audio_tower_layers_23_self_attn_out_proj_bias = view_2224 = permute_238 = None
	        view_2225 = torch.ops.aten.view.default(addmm_117, [sym_size_int, 1500, 1280]);  addmm_117 = None
	        add_21760 = torch.ops.aten.add.Tensor(add_21140, view_2225);  add_21140 = view_2225 = None
	        clone_190 = torch.ops.aten.clone.default(add_21760, memory_format = torch.contiguous_format)
	        var_mean_47 = torch.ops.aten.var_mean.correction(clone_190, [2], correction = 0, keepdim = True)
	        getitem_190 = var_mean_47[0]
	        getitem_191 = var_mean_47[1];  var_mean_47 = None
	        add_21765 = torch.ops.aten.add.Tensor(getitem_190, 1e-05);  getitem_190 = None
	        rsqrt_47 = torch.ops.aten.rsqrt.default(add_21765);  add_21765 = None
	        sub_6493 = torch.ops.aten.sub.Tensor(clone_190, getitem_191);  clone_190 = getitem_191 = None
	        mul_13755 = torch.ops.aten.mul.Tensor(sub_6493, rsqrt_47);  sub_6493 = rsqrt_47 = None
	        mul_13756 = torch.ops.aten.mul.Tensor(mul_13755, model_audio_tower_layers_23_final_layer_norm_weight);  mul_13755 = model_audio_tower_layers_23_final_layer_norm_weight = None
	        add_21766 = torch.ops.aten.add.Tensor(mul_13756, model_audio_tower_layers_23_final_layer_norm_bias);  mul_13756 = model_audio_tower_layers_23_final_layer_norm_bias = None
	        amin_142 = torch.ops.aten.amin.default(add_21766, [2])
	        amax_142 = torch.ops.aten.amax.default(add_21766, [2])
	        full_284 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_142 = torch.ops.aten.minimum.default(amin_142, full_284);  amin_142 = full_284 = None
	        full_285 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_142 = torch.ops.aten.maximum.default(amax_142, full_285);  amax_142 = full_285 = None
	        sub_6504 = torch.ops.aten.sub.Tensor(maximum_142, minimum_142);  maximum_142 = None
	        div_284 = torch.ops.aten.div.Tensor(sub_6504, 255.0);  sub_6504 = None
	        clamp_min_426 = torch.ops.aten.clamp_min.default(div_284, 1.1920928955078125e-07);  div_284 = None
	        div_285 = torch.ops.aten.div.Tensor(minimum_142, clamp_min_426);  minimum_142 = None
	        round_285 = torch.ops.aten.round.default(div_285);  div_285 = None
	        sub_6510 = torch.ops.aten.sub.Tensor(-128, round_285);  round_285 = None
	        clamp_min_427 = torch.ops.aten.clamp_min.default(sub_6510, -128);  sub_6510 = None
	        clamp_max_284 = torch.ops.aten.clamp_max.default(clamp_min_427, 127);  clamp_min_427 = None
	        _assert_tensor_metadata_1280 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_426, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1280 = None
	        _assert_tensor_metadata_1281 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_284, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1281 = None
	        convert_element_type_852 = torch.ops.prims.convert_element_type.default(clamp_max_284, torch.int8);  clamp_max_284 = None
	        view_2228 = torch.ops.aten.view.default(clamp_min_426, [sym_size_int, 1500, 1])
	        view_2229 = torch.ops.aten.view.default(convert_element_type_852, [sym_size_int, 1500, 1])
	        reciprocal_142 = torch.ops.aten.reciprocal.default(view_2228);  view_2228 = None
	        mul_13804 = torch.ops.aten.mul.Tensor(reciprocal_142, 1.0);  reciprocal_142 = None
	        mul_13807 = torch.ops.aten.mul.Tensor(add_21766, mul_13804);  add_21766 = mul_13804 = None
	        round_286 = torch.ops.aten.round.default(mul_13807);  mul_13807 = None
	        add_21853 = torch.ops.aten.add.Tensor(round_286, view_2229);  round_286 = view_2229 = None
	        clamp_min_428 = torch.ops.aten.clamp_min.default(add_21853, -128);  add_21853 = None
	        clamp_max_285 = torch.ops.aten.clamp_max.default(clamp_min_428, 127);  clamp_min_428 = None
	        _assert_tensor_metadata_1282 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_285, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1282 = None
	        convert_element_type_853 = torch.ops.prims.convert_element_type.default(clamp_max_285, torch.int8);  clamp_max_285 = None
	        view_2232 = torch.ops.aten.view.default(clamp_min_426, [sym_size_int, 1500, 1]);  clamp_min_426 = None
	        view_2233 = torch.ops.aten.view.default(convert_element_type_852, [sym_size_int, 1500, 1]);  convert_element_type_852 = None
	        _assert_tensor_metadata_1283 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_853, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1283 = None
	        convert_element_type_854 = torch.ops.prims.convert_element_type.default(convert_element_type_853, torch.float32);  convert_element_type_853 = None
	        _assert_tensor_metadata_1284 = torch.ops.aten._assert_tensor_metadata.default(view_2233, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1284 = None
	        convert_element_type_855 = torch.ops.prims.convert_element_type.default(view_2233, torch.float32);  view_2233 = None
	        sub_6530 = torch.ops.aten.sub.Tensor(convert_element_type_854, convert_element_type_855);  convert_element_type_854 = convert_element_type_855 = None
	        mul_13829 = torch.ops.aten.mul.Tensor(sub_6530, view_2232);  sub_6530 = view_2232 = None
	        _assert_tensor_metadata_1285 = torch.ops.aten._assert_tensor_metadata.default(mul_13829, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1285 = None
	        view_2235 = torch.ops.aten.view.default(model_audio_tower_layers_23_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_23_fc1_parametrizations_weight_original0 = None
	        view_2236 = torch.ops.aten.view.default(model_audio_tower_layers_23_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_23_fc1_parametrizations_weight_original1 = None
	        view_2237 = torch.ops.aten.view.default(model_audio_tower_layers_23_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_23_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1286 = torch.ops.aten._assert_tensor_metadata.default(view_2235, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1286 = None
	        convert_element_type_856 = torch.ops.prims.convert_element_type.default(view_2235, torch.float32);  view_2235 = None
	        _assert_tensor_metadata_1287 = torch.ops.aten._assert_tensor_metadata.default(view_2237, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1287 = None
	        convert_element_type_857 = torch.ops.prims.convert_element_type.default(view_2237, torch.float32);  view_2237 = None
	        sub_6534 = torch.ops.aten.sub.Tensor(convert_element_type_856, convert_element_type_857);  convert_element_type_856 = convert_element_type_857 = None
	        mul_13834 = torch.ops.aten.mul.Tensor(sub_6534, view_2236);  sub_6534 = view_2236 = None
	        view_2238 = torch.ops.aten.view.default(mul_13834, [5120, 1280]);  mul_13834 = None
	        _assert_tensor_metadata_1288 = torch.ops.aten._assert_tensor_metadata.default(view_2238, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1288 = None
	        mul_13839 = sym_size_int * 1500
	        view_2239 = torch.ops.aten.view.default(mul_13829, [mul_13839, 1280]);  mul_13829 = mul_13839 = None
	        permute_239 = torch.ops.aten.permute.default(view_2238, [1, 0]);  view_2238 = None
	        addmm_118 = torch.ops.aten.addmm.default(model_audio_tower_layers_23_fc1_bias, view_2239, permute_239);  model_audio_tower_layers_23_fc1_bias = view_2239 = permute_239 = None
	        view_2240 = torch.ops.aten.view.default(addmm_118, [sym_size_int, 1500, 5120]);  addmm_118 = None
	        mul_13846 = torch.ops.aten.mul.Tensor(view_2240, 0.5)
	        mul_13847 = torch.ops.aten.mul.Tensor(view_2240, 0.7071067811865476);  view_2240 = None
	        erf_25 = torch.ops.aten.erf.default(mul_13847);  mul_13847 = None
	        add_21912 = torch.ops.aten.add.Tensor(erf_25, 1);  erf_25 = None
	        mul_13848 = torch.ops.aten.mul.Tensor(mul_13846, add_21912);  mul_13846 = add_21912 = None
	        amin_143 = torch.ops.aten.amin.default(mul_13848, [2])
	        amax_143 = torch.ops.aten.amax.default(mul_13848, [2])
	        full_286 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_143 = torch.ops.aten.minimum.default(amin_143, full_286);  amin_143 = full_286 = None
	        full_287 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_143 = torch.ops.aten.maximum.default(amax_143, full_287);  amax_143 = full_287 = None
	        sub_6547 = torch.ops.aten.sub.Tensor(maximum_143, minimum_143);  maximum_143 = None
	        div_286 = torch.ops.aten.div.Tensor(sub_6547, 255.0);  sub_6547 = None
	        clamp_min_429 = torch.ops.aten.clamp_min.default(div_286, 1.1920928955078125e-07);  div_286 = None
	        div_287 = torch.ops.aten.div.Tensor(minimum_143, clamp_min_429);  minimum_143 = None
	        round_287 = torch.ops.aten.round.default(div_287);  div_287 = None
	        sub_6553 = torch.ops.aten.sub.Tensor(-128, round_287);  round_287 = None
	        clamp_min_430 = torch.ops.aten.clamp_min.default(sub_6553, -128);  sub_6553 = None
	        clamp_max_286 = torch.ops.aten.clamp_max.default(clamp_min_430, 127);  clamp_min_430 = None
	        _assert_tensor_metadata_1289 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_429, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1289 = None
	        _assert_tensor_metadata_1290 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_286, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1290 = None
	        convert_element_type_858 = torch.ops.prims.convert_element_type.default(clamp_max_286, torch.int8);  clamp_max_286 = None
	        view_2243 = torch.ops.aten.view.default(clamp_min_429, [sym_size_int, 1500, 1])
	        view_2244 = torch.ops.aten.view.default(convert_element_type_858, [sym_size_int, 1500, 1])
	        reciprocal_143 = torch.ops.aten.reciprocal.default(view_2243);  view_2243 = None
	        mul_13894 = torch.ops.aten.mul.Tensor(reciprocal_143, 1.0);  reciprocal_143 = None
	        mul_13897 = torch.ops.aten.mul.Tensor(mul_13848, mul_13894);  mul_13848 = mul_13894 = None
	        round_288 = torch.ops.aten.round.default(mul_13897);  mul_13897 = None
	        add_21995 = torch.ops.aten.add.Tensor(round_288, view_2244);  round_288 = view_2244 = None
	        clamp_min_431 = torch.ops.aten.clamp_min.default(add_21995, -128);  add_21995 = None
	        clamp_max_287 = torch.ops.aten.clamp_max.default(clamp_min_431, 127);  clamp_min_431 = None
	        _assert_tensor_metadata_1291 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_287, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1291 = None
	        convert_element_type_859 = torch.ops.prims.convert_element_type.default(clamp_max_287, torch.int8);  clamp_max_287 = None
	        view_2247 = torch.ops.aten.view.default(clamp_min_429, [sym_size_int, 1500, 1]);  clamp_min_429 = None
	        view_2248 = torch.ops.aten.view.default(convert_element_type_858, [sym_size_int, 1500, 1]);  convert_element_type_858 = None
	        _assert_tensor_metadata_1292 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_859, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1292 = None
	        convert_element_type_860 = torch.ops.prims.convert_element_type.default(convert_element_type_859, torch.float32);  convert_element_type_859 = None
	        _assert_tensor_metadata_1293 = torch.ops.aten._assert_tensor_metadata.default(view_2248, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1293 = None
	        convert_element_type_861 = torch.ops.prims.convert_element_type.default(view_2248, torch.float32);  view_2248 = None
	        sub_6573 = torch.ops.aten.sub.Tensor(convert_element_type_860, convert_element_type_861);  convert_element_type_860 = convert_element_type_861 = None
	        mul_13919 = torch.ops.aten.mul.Tensor(sub_6573, view_2247);  sub_6573 = view_2247 = None
	        _assert_tensor_metadata_1294 = torch.ops.aten._assert_tensor_metadata.default(mul_13919, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1294 = None
	        view_2250 = torch.ops.aten.view.default(model_audio_tower_layers_23_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_23_fc2_parametrizations_weight_original0 = None
	        view_2251 = torch.ops.aten.view.default(model_audio_tower_layers_23_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_23_fc2_parametrizations_weight_original1 = None
	        view_2252 = torch.ops.aten.view.default(model_audio_tower_layers_23_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_23_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1295 = torch.ops.aten._assert_tensor_metadata.default(view_2250, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1295 = None
	        convert_element_type_862 = torch.ops.prims.convert_element_type.default(view_2250, torch.float32);  view_2250 = None
	        _assert_tensor_metadata_1296 = torch.ops.aten._assert_tensor_metadata.default(view_2252, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1296 = None
	        convert_element_type_863 = torch.ops.prims.convert_element_type.default(view_2252, torch.float32);  view_2252 = None
	        sub_6577 = torch.ops.aten.sub.Tensor(convert_element_type_862, convert_element_type_863);  convert_element_type_862 = convert_element_type_863 = None
	        mul_13924 = torch.ops.aten.mul.Tensor(sub_6577, view_2251);  sub_6577 = view_2251 = None
	        view_2253 = torch.ops.aten.view.default(mul_13924, [1280, 5120]);  mul_13924 = None
	        _assert_tensor_metadata_1297 = torch.ops.aten._assert_tensor_metadata.default(view_2253, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1297 = None
	        mul_13929 = sym_size_int * 1500
	        view_2254 = torch.ops.aten.view.default(mul_13919, [mul_13929, 5120]);  mul_13919 = mul_13929 = None
	        permute_240 = torch.ops.aten.permute.default(view_2253, [1, 0]);  view_2253 = None
	        addmm_119 = torch.ops.aten.addmm.default(model_audio_tower_layers_23_fc2_bias, view_2254, permute_240);  model_audio_tower_layers_23_fc2_bias = view_2254 = permute_240 = None
	        view_2255 = torch.ops.aten.view.default(addmm_119, [sym_size_int, 1500, 1280]);  addmm_119 = None
	        add_22058 = torch.ops.aten.add.Tensor(add_21760, view_2255);  add_21760 = view_2255 = None
	        clone_193 = torch.ops.aten.clone.default(add_22058, memory_format = torch.contiguous_format)
	        var_mean_48 = torch.ops.aten.var_mean.correction(clone_193, [2], correction = 0, keepdim = True)
	        getitem_192 = var_mean_48[0]
	        getitem_193 = var_mean_48[1];  var_mean_48 = None
	        add_22063 = torch.ops.aten.add.Tensor(getitem_192, 1e-05);  getitem_192 = None
	        rsqrt_48 = torch.ops.aten.rsqrt.default(add_22063);  add_22063 = None
	        sub_6583 = torch.ops.aten.sub.Tensor(clone_193, getitem_193);  clone_193 = getitem_193 = None
	        mul_13940 = torch.ops.aten.mul.Tensor(sub_6583, rsqrt_48);  sub_6583 = rsqrt_48 = None
	        mul_13941 = torch.ops.aten.mul.Tensor(mul_13940, model_audio_tower_layers_24_self_attn_layer_norm_weight);  mul_13940 = model_audio_tower_layers_24_self_attn_layer_norm_weight = None
	        add_22064 = torch.ops.aten.add.Tensor(mul_13941, model_audio_tower_layers_24_self_attn_layer_norm_bias);  mul_13941 = model_audio_tower_layers_24_self_attn_layer_norm_bias = None
	        amin_144 = torch.ops.aten.amin.default(add_22064, [2])
	        amax_144 = torch.ops.aten.amax.default(add_22064, [2])
	        full_288 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_144 = torch.ops.aten.minimum.default(amin_144, full_288);  amin_144 = full_288 = None
	        full_289 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_144 = torch.ops.aten.maximum.default(amax_144, full_289);  amax_144 = full_289 = None
	        sub_6594 = torch.ops.aten.sub.Tensor(maximum_144, minimum_144);  maximum_144 = None
	        div_288 = torch.ops.aten.div.Tensor(sub_6594, 255.0);  sub_6594 = None
	        clamp_min_432 = torch.ops.aten.clamp_min.default(div_288, 1.1920928955078125e-07);  div_288 = None
	        div_289 = torch.ops.aten.div.Tensor(minimum_144, clamp_min_432);  minimum_144 = None
	        round_289 = torch.ops.aten.round.default(div_289);  div_289 = None
	        sub_6600 = torch.ops.aten.sub.Tensor(-128, round_289);  round_289 = None
	        clamp_min_433 = torch.ops.aten.clamp_min.default(sub_6600, -128);  sub_6600 = None
	        clamp_max_288 = torch.ops.aten.clamp_max.default(clamp_min_433, 127);  clamp_min_433 = None
	        _assert_tensor_metadata_1298 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_432, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1298 = None
	        _assert_tensor_metadata_1299 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_288, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1299 = None
	        convert_element_type_864 = torch.ops.prims.convert_element_type.default(clamp_max_288, torch.int8);  clamp_max_288 = None
	        view_2258 = torch.ops.aten.view.default(clamp_min_432, [sym_size_int, 1500, 1])
	        view_2259 = torch.ops.aten.view.default(convert_element_type_864, [sym_size_int, 1500, 1])
	        reciprocal_144 = torch.ops.aten.reciprocal.default(view_2258);  view_2258 = None
	        mul_13989 = torch.ops.aten.mul.Tensor(reciprocal_144, 1.0);  reciprocal_144 = None
	        mul_13992 = torch.ops.aten.mul.Tensor(add_22064, mul_13989);  mul_13989 = None
	        round_290 = torch.ops.aten.round.default(mul_13992);  mul_13992 = None
	        add_22151 = torch.ops.aten.add.Tensor(round_290, view_2259);  round_290 = view_2259 = None
	        clamp_min_434 = torch.ops.aten.clamp_min.default(add_22151, -128);  add_22151 = None
	        clamp_max_289 = torch.ops.aten.clamp_max.default(clamp_min_434, 127);  clamp_min_434 = None
	        _assert_tensor_metadata_1300 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_289, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1300 = None
	        convert_element_type_865 = torch.ops.prims.convert_element_type.default(clamp_max_289, torch.int8);  clamp_max_289 = None
	        view_2262 = torch.ops.aten.view.default(clamp_min_432, [sym_size_int, 1500, 1]);  clamp_min_432 = None
	        view_2263 = torch.ops.aten.view.default(convert_element_type_864, [sym_size_int, 1500, 1]);  convert_element_type_864 = None
	        _assert_tensor_metadata_1301 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_865, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1301 = None
	        convert_element_type_866 = torch.ops.prims.convert_element_type.default(convert_element_type_865, torch.float32);  convert_element_type_865 = None
	        _assert_tensor_metadata_1302 = torch.ops.aten._assert_tensor_metadata.default(view_2263, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1302 = None
	        convert_element_type_867 = torch.ops.prims.convert_element_type.default(view_2263, torch.float32);  view_2263 = None
	        sub_6620 = torch.ops.aten.sub.Tensor(convert_element_type_866, convert_element_type_867);  convert_element_type_866 = convert_element_type_867 = None
	        mul_14014 = torch.ops.aten.mul.Tensor(sub_6620, view_2262);  sub_6620 = view_2262 = None
	        _assert_tensor_metadata_1303 = torch.ops.aten._assert_tensor_metadata.default(mul_14014, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1303 = None
	        view_2265 = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2266 = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2267 = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1304 = torch.ops.aten._assert_tensor_metadata.default(view_2265, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1304 = None
	        convert_element_type_868 = torch.ops.prims.convert_element_type.default(view_2265, torch.float32);  view_2265 = None
	        _assert_tensor_metadata_1305 = torch.ops.aten._assert_tensor_metadata.default(view_2267, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1305 = None
	        convert_element_type_869 = torch.ops.prims.convert_element_type.default(view_2267, torch.float32);  view_2267 = None
	        sub_6624 = torch.ops.aten.sub.Tensor(convert_element_type_868, convert_element_type_869);  convert_element_type_868 = convert_element_type_869 = None
	        mul_14019 = torch.ops.aten.mul.Tensor(sub_6624, view_2266);  sub_6624 = view_2266 = None
	        view_2268 = torch.ops.aten.view.default(mul_14019, [1280, 1280]);  mul_14019 = None
	        _assert_tensor_metadata_1306 = torch.ops.aten._assert_tensor_metadata.default(view_2268, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1306 = None
	        mul_14024 = sym_size_int * 1500
	        view_2269 = torch.ops.aten.view.default(mul_14014, [mul_14024, 1280]);  mul_14014 = mul_14024 = None
	        permute_241 = torch.ops.aten.permute.default(view_2268, [1, 0]);  view_2268 = None
	        addmm_120 = torch.ops.aten.addmm.default(model_audio_tower_layers_24_self_attn_q_proj_bias, view_2269, permute_241);  model_audio_tower_layers_24_self_attn_q_proj_bias = view_2269 = permute_241 = None
	        view_2270 = torch.ops.aten.view.default(addmm_120, [sym_size_int, 1500, 1280]);  addmm_120 = None
	        mul_14031 = torch.ops.aten.mul.Tensor(view_2270, 0.125);  view_2270 = None
	        view_2271 = torch.ops.aten.view.default(mul_14031, [sym_size_int, 1500, 20, 64]);  mul_14031 = None
	        permute_242 = torch.ops.aten.permute.default(view_2271, [0, 2, 1, 3]);  view_2271 = None
	        clone_194 = torch.ops.aten.clone.default(permute_242, memory_format = torch.contiguous_format);  permute_242 = None
	        amin_145 = torch.ops.aten.amin.default(add_22064, [2])
	        amax_145 = torch.ops.aten.amax.default(add_22064, [2])
	        full_290 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_145 = torch.ops.aten.minimum.default(amin_145, full_290);  amin_145 = full_290 = None
	        full_291 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_145 = torch.ops.aten.maximum.default(amax_145, full_291);  amax_145 = full_291 = None
	        sub_6639 = torch.ops.aten.sub.Tensor(maximum_145, minimum_145);  maximum_145 = None
	        div_290 = torch.ops.aten.div.Tensor(sub_6639, 255.0);  sub_6639 = None
	        clamp_min_435 = torch.ops.aten.clamp_min.default(div_290, 1.1920928955078125e-07);  div_290 = None
	        div_291 = torch.ops.aten.div.Tensor(minimum_145, clamp_min_435);  minimum_145 = None
	        round_291 = torch.ops.aten.round.default(div_291);  div_291 = None
	        sub_6645 = torch.ops.aten.sub.Tensor(-128, round_291);  round_291 = None
	        clamp_min_436 = torch.ops.aten.clamp_min.default(sub_6645, -128);  sub_6645 = None
	        clamp_max_290 = torch.ops.aten.clamp_max.default(clamp_min_436, 127);  clamp_min_436 = None
	        _assert_tensor_metadata_1307 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_435, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1307 = None
	        _assert_tensor_metadata_1308 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_290, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1308 = None
	        convert_element_type_870 = torch.ops.prims.convert_element_type.default(clamp_max_290, torch.int8);  clamp_max_290 = None
	        view_2274 = torch.ops.aten.view.default(clamp_min_435, [sym_size_int, 1500, 1])
	        view_2275 = torch.ops.aten.view.default(convert_element_type_870, [sym_size_int, 1500, 1])
	        reciprocal_145 = torch.ops.aten.reciprocal.default(view_2274);  view_2274 = None
	        mul_14085 = torch.ops.aten.mul.Tensor(reciprocal_145, 1.0);  reciprocal_145 = None
	        mul_14088 = torch.ops.aten.mul.Tensor(add_22064, mul_14085);  mul_14085 = None
	        round_292 = torch.ops.aten.round.default(mul_14088);  mul_14088 = None
	        add_22303 = torch.ops.aten.add.Tensor(round_292, view_2275);  round_292 = view_2275 = None
	        clamp_min_437 = torch.ops.aten.clamp_min.default(add_22303, -128);  add_22303 = None
	        clamp_max_291 = torch.ops.aten.clamp_max.default(clamp_min_437, 127);  clamp_min_437 = None
	        _assert_tensor_metadata_1309 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_291, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1309 = None
	        convert_element_type_871 = torch.ops.prims.convert_element_type.default(clamp_max_291, torch.int8);  clamp_max_291 = None
	        view_2278 = torch.ops.aten.view.default(clamp_min_435, [sym_size_int, 1500, 1]);  clamp_min_435 = None
	        view_2279 = torch.ops.aten.view.default(convert_element_type_870, [sym_size_int, 1500, 1]);  convert_element_type_870 = None
	        _assert_tensor_metadata_1310 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_871, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1310 = None
	        convert_element_type_872 = torch.ops.prims.convert_element_type.default(convert_element_type_871, torch.float32);  convert_element_type_871 = None
	        _assert_tensor_metadata_1311 = torch.ops.aten._assert_tensor_metadata.default(view_2279, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1311 = None
	        convert_element_type_873 = torch.ops.prims.convert_element_type.default(view_2279, torch.float32);  view_2279 = None
	        sub_6665 = torch.ops.aten.sub.Tensor(convert_element_type_872, convert_element_type_873);  convert_element_type_872 = convert_element_type_873 = None
	        mul_14110 = torch.ops.aten.mul.Tensor(sub_6665, view_2278);  sub_6665 = view_2278 = None
	        _assert_tensor_metadata_1312 = torch.ops.aten._assert_tensor_metadata.default(mul_14110, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1312 = None
	        view_2281 = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2282 = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2283 = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1313 = torch.ops.aten._assert_tensor_metadata.default(view_2281, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1313 = None
	        convert_element_type_874 = torch.ops.prims.convert_element_type.default(view_2281, torch.float32);  view_2281 = None
	        _assert_tensor_metadata_1314 = torch.ops.aten._assert_tensor_metadata.default(view_2283, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1314 = None
	        convert_element_type_875 = torch.ops.prims.convert_element_type.default(view_2283, torch.float32);  view_2283 = None
	        sub_6669 = torch.ops.aten.sub.Tensor(convert_element_type_874, convert_element_type_875);  convert_element_type_874 = convert_element_type_875 = None
	        mul_14115 = torch.ops.aten.mul.Tensor(sub_6669, view_2282);  sub_6669 = view_2282 = None
	        view_2284 = torch.ops.aten.view.default(mul_14115, [1280, 1280]);  mul_14115 = None
	        _assert_tensor_metadata_1315 = torch.ops.aten._assert_tensor_metadata.default(view_2284, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1315 = None
	        permute_243 = torch.ops.aten.permute.default(view_2284, [1, 0]);  view_2284 = None
	        mul_14118 = sym_size_int * 1500
	        view_2285 = torch.ops.aten.view.default(mul_14110, [mul_14118, 1280]);  mul_14110 = mul_14118 = None
	        mm_24 = torch.ops.aten.mm.default(view_2285, permute_243);  view_2285 = permute_243 = None
	        view_2286 = torch.ops.aten.view.default(mm_24, [sym_size_int, 1500, 1280]);  mm_24 = None
	        view_2287 = torch.ops.aten.view.default(view_2286, [sym_size_int, -1, 20, 64]);  view_2286 = None
	        permute_244 = torch.ops.aten.permute.default(view_2287, [0, 2, 1, 3]);  view_2287 = None
	        clone_195 = torch.ops.aten.clone.default(permute_244, memory_format = torch.contiguous_format);  permute_244 = None
	        amin_146 = torch.ops.aten.amin.default(add_22064, [2])
	        amax_146 = torch.ops.aten.amax.default(add_22064, [2])
	        full_292 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_146 = torch.ops.aten.minimum.default(amin_146, full_292);  amin_146 = full_292 = None
	        full_293 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_146 = torch.ops.aten.maximum.default(amax_146, full_293);  amax_146 = full_293 = None
	        sub_6683 = torch.ops.aten.sub.Tensor(maximum_146, minimum_146);  maximum_146 = None
	        div_292 = torch.ops.aten.div.Tensor(sub_6683, 255.0);  sub_6683 = None
	        clamp_min_438 = torch.ops.aten.clamp_min.default(div_292, 1.1920928955078125e-07);  div_292 = None
	        div_293 = torch.ops.aten.div.Tensor(minimum_146, clamp_min_438);  minimum_146 = None
	        round_293 = torch.ops.aten.round.default(div_293);  div_293 = None
	        sub_6689 = torch.ops.aten.sub.Tensor(-128, round_293);  round_293 = None
	        clamp_min_439 = torch.ops.aten.clamp_min.default(sub_6689, -128);  sub_6689 = None
	        clamp_max_292 = torch.ops.aten.clamp_max.default(clamp_min_439, 127);  clamp_min_439 = None
	        _assert_tensor_metadata_1316 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_438, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1316 = None
	        _assert_tensor_metadata_1317 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_292, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1317 = None
	        convert_element_type_876 = torch.ops.prims.convert_element_type.default(clamp_max_292, torch.int8);  clamp_max_292 = None
	        view_2290 = torch.ops.aten.view.default(clamp_min_438, [sym_size_int, 1500, 1])
	        view_2291 = torch.ops.aten.view.default(convert_element_type_876, [sym_size_int, 1500, 1])
	        reciprocal_146 = torch.ops.aten.reciprocal.default(view_2290);  view_2290 = None
	        mul_14184 = torch.ops.aten.mul.Tensor(reciprocal_146, 1.0);  reciprocal_146 = None
	        mul_14187 = torch.ops.aten.mul.Tensor(add_22064, mul_14184);  add_22064 = mul_14184 = None
	        round_294 = torch.ops.aten.round.default(mul_14187);  mul_14187 = None
	        add_22451 = torch.ops.aten.add.Tensor(round_294, view_2291);  round_294 = view_2291 = None
	        clamp_min_440 = torch.ops.aten.clamp_min.default(add_22451, -128);  add_22451 = None
	        clamp_max_293 = torch.ops.aten.clamp_max.default(clamp_min_440, 127);  clamp_min_440 = None
	        _assert_tensor_metadata_1318 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_293, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1318 = None
	        convert_element_type_877 = torch.ops.prims.convert_element_type.default(clamp_max_293, torch.int8);  clamp_max_293 = None
	        view_2294 = torch.ops.aten.view.default(clamp_min_438, [sym_size_int, 1500, 1]);  clamp_min_438 = None
	        view_2295 = torch.ops.aten.view.default(convert_element_type_876, [sym_size_int, 1500, 1]);  convert_element_type_876 = None
	        _assert_tensor_metadata_1319 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_877, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1319 = None
	        convert_element_type_878 = torch.ops.prims.convert_element_type.default(convert_element_type_877, torch.float32);  convert_element_type_877 = None
	        _assert_tensor_metadata_1320 = torch.ops.aten._assert_tensor_metadata.default(view_2295, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1320 = None
	        convert_element_type_879 = torch.ops.prims.convert_element_type.default(view_2295, torch.float32);  view_2295 = None
	        sub_6709 = torch.ops.aten.sub.Tensor(convert_element_type_878, convert_element_type_879);  convert_element_type_878 = convert_element_type_879 = None
	        mul_14209 = torch.ops.aten.mul.Tensor(sub_6709, view_2294);  sub_6709 = view_2294 = None
	        _assert_tensor_metadata_1321 = torch.ops.aten._assert_tensor_metadata.default(mul_14209, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1321 = None
	        view_2297 = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2298 = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2299 = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1322 = torch.ops.aten._assert_tensor_metadata.default(view_2297, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1322 = None
	        convert_element_type_880 = torch.ops.prims.convert_element_type.default(view_2297, torch.float32);  view_2297 = None
	        _assert_tensor_metadata_1323 = torch.ops.aten._assert_tensor_metadata.default(view_2299, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1323 = None
	        convert_element_type_881 = torch.ops.prims.convert_element_type.default(view_2299, torch.float32);  view_2299 = None
	        sub_6713 = torch.ops.aten.sub.Tensor(convert_element_type_880, convert_element_type_881);  convert_element_type_880 = convert_element_type_881 = None
	        mul_14214 = torch.ops.aten.mul.Tensor(sub_6713, view_2298);  sub_6713 = view_2298 = None
	        view_2300 = torch.ops.aten.view.default(mul_14214, [1280, 1280]);  mul_14214 = None
	        _assert_tensor_metadata_1324 = torch.ops.aten._assert_tensor_metadata.default(view_2300, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1324 = None
	        mul_14219 = sym_size_int * 1500
	        view_2301 = torch.ops.aten.view.default(mul_14209, [mul_14219, 1280]);  mul_14209 = mul_14219 = None
	        permute_245 = torch.ops.aten.permute.default(view_2300, [1, 0]);  view_2300 = None
	        addmm_121 = torch.ops.aten.addmm.default(model_audio_tower_layers_24_self_attn_v_proj_bias, view_2301, permute_245);  model_audio_tower_layers_24_self_attn_v_proj_bias = view_2301 = permute_245 = None
	        view_2302 = torch.ops.aten.view.default(addmm_121, [sym_size_int, 1500, 1280]);  addmm_121 = None
	        view_2303 = torch.ops.aten.view.default(view_2302, [sym_size_int, -1, 20, 64]);  view_2302 = None
	        permute_246 = torch.ops.aten.permute.default(view_2303, [0, 2, 1, 3]);  view_2303 = None
	        clone_196 = torch.ops.aten.clone.default(permute_246, memory_format = torch.contiguous_format);  permute_246 = None
	        _scaled_dot_product_efficient_attention_24 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_194, clone_195, clone_196, None, False, scale = 1.0);  clone_194 = clone_195 = clone_196 = None
	        getitem_194 = _scaled_dot_product_efficient_attention_24[0];  _scaled_dot_product_efficient_attention_24 = None
	        permute_247 = torch.ops.aten.permute.default(getitem_194, [0, 2, 1, 3]);  getitem_194 = None
	        view_2304 = torch.ops.aten.view.default(permute_247, [sym_size_int, 1500, -1]);  permute_247 = None
	        amin_147 = torch.ops.aten.amin.default(view_2304, [2])
	        amax_147 = torch.ops.aten.amax.default(view_2304, [2])
	        full_294 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_147 = torch.ops.aten.minimum.default(amin_147, full_294);  amin_147 = full_294 = None
	        full_295 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_147 = torch.ops.aten.maximum.default(amax_147, full_295);  amax_147 = full_295 = None
	        sub_6731 = torch.ops.aten.sub.Tensor(maximum_147, minimum_147);  maximum_147 = None
	        div_294 = torch.ops.aten.div.Tensor(sub_6731, 255.0);  sub_6731 = None
	        clamp_min_441 = torch.ops.aten.clamp_min.default(div_294, 1.1920928955078125e-07);  div_294 = None
	        div_295 = torch.ops.aten.div.Tensor(minimum_147, clamp_min_441);  minimum_147 = None
	        round_295 = torch.ops.aten.round.default(div_295);  div_295 = None
	        sub_6737 = torch.ops.aten.sub.Tensor(-128, round_295);  round_295 = None
	        clamp_min_442 = torch.ops.aten.clamp_min.default(sub_6737, -128);  sub_6737 = None
	        clamp_max_294 = torch.ops.aten.clamp_max.default(clamp_min_442, 127);  clamp_min_442 = None
	        _assert_tensor_metadata_1325 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_441, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1325 = None
	        _assert_tensor_metadata_1326 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1326 = None
	        convert_element_type_882 = torch.ops.prims.convert_element_type.default(clamp_max_294, torch.int8);  clamp_max_294 = None
	        view_2307 = torch.ops.aten.view.default(clamp_min_441, [sym_size_int, 1500, 1])
	        view_2308 = torch.ops.aten.view.default(convert_element_type_882, [sym_size_int, 1500, 1])
	        reciprocal_147 = torch.ops.aten.reciprocal.default(view_2307);  view_2307 = None
	        mul_14289 = torch.ops.aten.mul.Tensor(reciprocal_147, 1.0);  reciprocal_147 = None
	        mul_14292 = torch.ops.aten.mul.Tensor(view_2304, mul_14289);  view_2304 = mul_14289 = None
	        round_296 = torch.ops.aten.round.default(mul_14292);  mul_14292 = None
	        add_22615 = torch.ops.aten.add.Tensor(round_296, view_2308);  round_296 = view_2308 = None
	        clamp_min_443 = torch.ops.aten.clamp_min.default(add_22615, -128);  add_22615 = None
	        clamp_max_295 = torch.ops.aten.clamp_max.default(clamp_min_443, 127);  clamp_min_443 = None
	        _assert_tensor_metadata_1327 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_295, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1327 = None
	        convert_element_type_883 = torch.ops.prims.convert_element_type.default(clamp_max_295, torch.int8);  clamp_max_295 = None
	        view_2311 = torch.ops.aten.view.default(clamp_min_441, [sym_size_int, 1500, 1]);  clamp_min_441 = None
	        view_2312 = torch.ops.aten.view.default(convert_element_type_882, [sym_size_int, 1500, 1]);  convert_element_type_882 = None
	        _assert_tensor_metadata_1328 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_883, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1328 = None
	        convert_element_type_884 = torch.ops.prims.convert_element_type.default(convert_element_type_883, torch.float32);  convert_element_type_883 = None
	        _assert_tensor_metadata_1329 = torch.ops.aten._assert_tensor_metadata.default(view_2312, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1329 = None
	        convert_element_type_885 = torch.ops.prims.convert_element_type.default(view_2312, torch.float32);  view_2312 = None
	        sub_6757 = torch.ops.aten.sub.Tensor(convert_element_type_884, convert_element_type_885);  convert_element_type_884 = convert_element_type_885 = None
	        mul_14314 = torch.ops.aten.mul.Tensor(sub_6757, view_2311);  sub_6757 = view_2311 = None
	        _assert_tensor_metadata_1330 = torch.ops.aten._assert_tensor_metadata.default(mul_14314, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1330 = None
	        view_2314 = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2315 = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2316 = torch.ops.aten.view.default(model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1331 = torch.ops.aten._assert_tensor_metadata.default(view_2314, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1331 = None
	        convert_element_type_886 = torch.ops.prims.convert_element_type.default(view_2314, torch.float32);  view_2314 = None
	        _assert_tensor_metadata_1332 = torch.ops.aten._assert_tensor_metadata.default(view_2316, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1332 = None
	        convert_element_type_887 = torch.ops.prims.convert_element_type.default(view_2316, torch.float32);  view_2316 = None
	        sub_6761 = torch.ops.aten.sub.Tensor(convert_element_type_886, convert_element_type_887);  convert_element_type_886 = convert_element_type_887 = None
	        mul_14319 = torch.ops.aten.mul.Tensor(sub_6761, view_2315);  sub_6761 = view_2315 = None
	        view_2317 = torch.ops.aten.view.default(mul_14319, [1280, 1280]);  mul_14319 = None
	        _assert_tensor_metadata_1333 = torch.ops.aten._assert_tensor_metadata.default(view_2317, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1333 = None
	        mul_14324 = sym_size_int * 1500
	        view_2318 = torch.ops.aten.view.default(mul_14314, [mul_14324, 1280]);  mul_14314 = mul_14324 = None
	        permute_248 = torch.ops.aten.permute.default(view_2317, [1, 0]);  view_2317 = None
	        addmm_122 = torch.ops.aten.addmm.default(model_audio_tower_layers_24_self_attn_out_proj_bias, view_2318, permute_248);  model_audio_tower_layers_24_self_attn_out_proj_bias = view_2318 = permute_248 = None
	        view_2319 = torch.ops.aten.view.default(addmm_122, [sym_size_int, 1500, 1280]);  addmm_122 = None
	        add_22678 = torch.ops.aten.add.Tensor(add_22058, view_2319);  add_22058 = view_2319 = None
	        clone_198 = torch.ops.aten.clone.default(add_22678, memory_format = torch.contiguous_format)
	        var_mean_49 = torch.ops.aten.var_mean.correction(clone_198, [2], correction = 0, keepdim = True)
	        getitem_198 = var_mean_49[0]
	        getitem_199 = var_mean_49[1];  var_mean_49 = None
	        add_22683 = torch.ops.aten.add.Tensor(getitem_198, 1e-05);  getitem_198 = None
	        rsqrt_49 = torch.ops.aten.rsqrt.default(add_22683);  add_22683 = None
	        sub_6767 = torch.ops.aten.sub.Tensor(clone_198, getitem_199);  clone_198 = getitem_199 = None
	        mul_14335 = torch.ops.aten.mul.Tensor(sub_6767, rsqrt_49);  sub_6767 = rsqrt_49 = None
	        mul_14336 = torch.ops.aten.mul.Tensor(mul_14335, model_audio_tower_layers_24_final_layer_norm_weight);  mul_14335 = model_audio_tower_layers_24_final_layer_norm_weight = None
	        add_22684 = torch.ops.aten.add.Tensor(mul_14336, model_audio_tower_layers_24_final_layer_norm_bias);  mul_14336 = model_audio_tower_layers_24_final_layer_norm_bias = None
	        amin_148 = torch.ops.aten.amin.default(add_22684, [2])
	        amax_148 = torch.ops.aten.amax.default(add_22684, [2])
	        full_296 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_148 = torch.ops.aten.minimum.default(amin_148, full_296);  amin_148 = full_296 = None
	        full_297 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_148 = torch.ops.aten.maximum.default(amax_148, full_297);  amax_148 = full_297 = None
	        sub_6778 = torch.ops.aten.sub.Tensor(maximum_148, minimum_148);  maximum_148 = None
	        div_296 = torch.ops.aten.div.Tensor(sub_6778, 255.0);  sub_6778 = None
	        clamp_min_444 = torch.ops.aten.clamp_min.default(div_296, 1.1920928955078125e-07);  div_296 = None
	        div_297 = torch.ops.aten.div.Tensor(minimum_148, clamp_min_444);  minimum_148 = None
	        round_297 = torch.ops.aten.round.default(div_297);  div_297 = None
	        sub_6784 = torch.ops.aten.sub.Tensor(-128, round_297);  round_297 = None
	        clamp_min_445 = torch.ops.aten.clamp_min.default(sub_6784, -128);  sub_6784 = None
	        clamp_max_296 = torch.ops.aten.clamp_max.default(clamp_min_445, 127);  clamp_min_445 = None
	        _assert_tensor_metadata_1334 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_444, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1334 = None
	        _assert_tensor_metadata_1335 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_296, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1335 = None
	        convert_element_type_888 = torch.ops.prims.convert_element_type.default(clamp_max_296, torch.int8);  clamp_max_296 = None
	        view_2322 = torch.ops.aten.view.default(clamp_min_444, [sym_size_int, 1500, 1])
	        view_2323 = torch.ops.aten.view.default(convert_element_type_888, [sym_size_int, 1500, 1])
	        reciprocal_148 = torch.ops.aten.reciprocal.default(view_2322);  view_2322 = None
	        mul_14384 = torch.ops.aten.mul.Tensor(reciprocal_148, 1.0);  reciprocal_148 = None
	        mul_14387 = torch.ops.aten.mul.Tensor(add_22684, mul_14384);  add_22684 = mul_14384 = None
	        round_298 = torch.ops.aten.round.default(mul_14387);  mul_14387 = None
	        add_22771 = torch.ops.aten.add.Tensor(round_298, view_2323);  round_298 = view_2323 = None
	        clamp_min_446 = torch.ops.aten.clamp_min.default(add_22771, -128);  add_22771 = None
	        clamp_max_297 = torch.ops.aten.clamp_max.default(clamp_min_446, 127);  clamp_min_446 = None
	        _assert_tensor_metadata_1336 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_297, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1336 = None
	        convert_element_type_889 = torch.ops.prims.convert_element_type.default(clamp_max_297, torch.int8);  clamp_max_297 = None
	        view_2326 = torch.ops.aten.view.default(clamp_min_444, [sym_size_int, 1500, 1]);  clamp_min_444 = None
	        view_2327 = torch.ops.aten.view.default(convert_element_type_888, [sym_size_int, 1500, 1]);  convert_element_type_888 = None
	        _assert_tensor_metadata_1337 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_889, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1337 = None
	        convert_element_type_890 = torch.ops.prims.convert_element_type.default(convert_element_type_889, torch.float32);  convert_element_type_889 = None
	        _assert_tensor_metadata_1338 = torch.ops.aten._assert_tensor_metadata.default(view_2327, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1338 = None
	        convert_element_type_891 = torch.ops.prims.convert_element_type.default(view_2327, torch.float32);  view_2327 = None
	        sub_6804 = torch.ops.aten.sub.Tensor(convert_element_type_890, convert_element_type_891);  convert_element_type_890 = convert_element_type_891 = None
	        mul_14409 = torch.ops.aten.mul.Tensor(sub_6804, view_2326);  sub_6804 = view_2326 = None
	        _assert_tensor_metadata_1339 = torch.ops.aten._assert_tensor_metadata.default(mul_14409, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1339 = None
	        view_2329 = torch.ops.aten.view.default(model_audio_tower_layers_24_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_24_fc1_parametrizations_weight_original0 = None
	        view_2330 = torch.ops.aten.view.default(model_audio_tower_layers_24_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_24_fc1_parametrizations_weight_original1 = None
	        view_2331 = torch.ops.aten.view.default(model_audio_tower_layers_24_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_24_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1340 = torch.ops.aten._assert_tensor_metadata.default(view_2329, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1340 = None
	        convert_element_type_892 = torch.ops.prims.convert_element_type.default(view_2329, torch.float32);  view_2329 = None
	        _assert_tensor_metadata_1341 = torch.ops.aten._assert_tensor_metadata.default(view_2331, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1341 = None
	        convert_element_type_893 = torch.ops.prims.convert_element_type.default(view_2331, torch.float32);  view_2331 = None
	        sub_6808 = torch.ops.aten.sub.Tensor(convert_element_type_892, convert_element_type_893);  convert_element_type_892 = convert_element_type_893 = None
	        mul_14414 = torch.ops.aten.mul.Tensor(sub_6808, view_2330);  sub_6808 = view_2330 = None
	        view_2332 = torch.ops.aten.view.default(mul_14414, [5120, 1280]);  mul_14414 = None
	        _assert_tensor_metadata_1342 = torch.ops.aten._assert_tensor_metadata.default(view_2332, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1342 = None
	        mul_14419 = sym_size_int * 1500
	        view_2333 = torch.ops.aten.view.default(mul_14409, [mul_14419, 1280]);  mul_14409 = mul_14419 = None
	        permute_249 = torch.ops.aten.permute.default(view_2332, [1, 0]);  view_2332 = None
	        addmm_123 = torch.ops.aten.addmm.default(model_audio_tower_layers_24_fc1_bias, view_2333, permute_249);  model_audio_tower_layers_24_fc1_bias = view_2333 = permute_249 = None
	        view_2334 = torch.ops.aten.view.default(addmm_123, [sym_size_int, 1500, 5120]);  addmm_123 = None
	        mul_14426 = torch.ops.aten.mul.Tensor(view_2334, 0.5)
	        mul_14427 = torch.ops.aten.mul.Tensor(view_2334, 0.7071067811865476);  view_2334 = None
	        erf_26 = torch.ops.aten.erf.default(mul_14427);  mul_14427 = None
	        add_22830 = torch.ops.aten.add.Tensor(erf_26, 1);  erf_26 = None
	        mul_14428 = torch.ops.aten.mul.Tensor(mul_14426, add_22830);  mul_14426 = add_22830 = None
	        amin_149 = torch.ops.aten.amin.default(mul_14428, [2])
	        amax_149 = torch.ops.aten.amax.default(mul_14428, [2])
	        full_298 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_149 = torch.ops.aten.minimum.default(amin_149, full_298);  amin_149 = full_298 = None
	        full_299 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_149 = torch.ops.aten.maximum.default(amax_149, full_299);  amax_149 = full_299 = None
	        sub_6821 = torch.ops.aten.sub.Tensor(maximum_149, minimum_149);  maximum_149 = None
	        div_298 = torch.ops.aten.div.Tensor(sub_6821, 255.0);  sub_6821 = None
	        clamp_min_447 = torch.ops.aten.clamp_min.default(div_298, 1.1920928955078125e-07);  div_298 = None
	        div_299 = torch.ops.aten.div.Tensor(minimum_149, clamp_min_447);  minimum_149 = None
	        round_299 = torch.ops.aten.round.default(div_299);  div_299 = None
	        sub_6827 = torch.ops.aten.sub.Tensor(-128, round_299);  round_299 = None
	        clamp_min_448 = torch.ops.aten.clamp_min.default(sub_6827, -128);  sub_6827 = None
	        clamp_max_298 = torch.ops.aten.clamp_max.default(clamp_min_448, 127);  clamp_min_448 = None
	        _assert_tensor_metadata_1343 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_447, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1343 = None
	        _assert_tensor_metadata_1344 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_298, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1344 = None
	        convert_element_type_894 = torch.ops.prims.convert_element_type.default(clamp_max_298, torch.int8);  clamp_max_298 = None
	        view_2337 = torch.ops.aten.view.default(clamp_min_447, [sym_size_int, 1500, 1])
	        view_2338 = torch.ops.aten.view.default(convert_element_type_894, [sym_size_int, 1500, 1])
	        reciprocal_149 = torch.ops.aten.reciprocal.default(view_2337);  view_2337 = None
	        mul_14474 = torch.ops.aten.mul.Tensor(reciprocal_149, 1.0);  reciprocal_149 = None
	        mul_14477 = torch.ops.aten.mul.Tensor(mul_14428, mul_14474);  mul_14428 = mul_14474 = None
	        round_300 = torch.ops.aten.round.default(mul_14477);  mul_14477 = None
	        add_22913 = torch.ops.aten.add.Tensor(round_300, view_2338);  round_300 = view_2338 = None
	        clamp_min_449 = torch.ops.aten.clamp_min.default(add_22913, -128);  add_22913 = None
	        clamp_max_299 = torch.ops.aten.clamp_max.default(clamp_min_449, 127);  clamp_min_449 = None
	        _assert_tensor_metadata_1345 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_299, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1345 = None
	        convert_element_type_895 = torch.ops.prims.convert_element_type.default(clamp_max_299, torch.int8);  clamp_max_299 = None
	        view_2341 = torch.ops.aten.view.default(clamp_min_447, [sym_size_int, 1500, 1]);  clamp_min_447 = None
	        view_2342 = torch.ops.aten.view.default(convert_element_type_894, [sym_size_int, 1500, 1]);  convert_element_type_894 = None
	        _assert_tensor_metadata_1346 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_895, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1346 = None
	        convert_element_type_896 = torch.ops.prims.convert_element_type.default(convert_element_type_895, torch.float32);  convert_element_type_895 = None
	        _assert_tensor_metadata_1347 = torch.ops.aten._assert_tensor_metadata.default(view_2342, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1347 = None
	        convert_element_type_897 = torch.ops.prims.convert_element_type.default(view_2342, torch.float32);  view_2342 = None
	        sub_6847 = torch.ops.aten.sub.Tensor(convert_element_type_896, convert_element_type_897);  convert_element_type_896 = convert_element_type_897 = None
	        mul_14499 = torch.ops.aten.mul.Tensor(sub_6847, view_2341);  sub_6847 = view_2341 = None
	        _assert_tensor_metadata_1348 = torch.ops.aten._assert_tensor_metadata.default(mul_14499, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1348 = None
	        view_2344 = torch.ops.aten.view.default(model_audio_tower_layers_24_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_24_fc2_parametrizations_weight_original0 = None
	        view_2345 = torch.ops.aten.view.default(model_audio_tower_layers_24_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_24_fc2_parametrizations_weight_original1 = None
	        view_2346 = torch.ops.aten.view.default(model_audio_tower_layers_24_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_24_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1349 = torch.ops.aten._assert_tensor_metadata.default(view_2344, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1349 = None
	        convert_element_type_898 = torch.ops.prims.convert_element_type.default(view_2344, torch.float32);  view_2344 = None
	        _assert_tensor_metadata_1350 = torch.ops.aten._assert_tensor_metadata.default(view_2346, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1350 = None
	        convert_element_type_899 = torch.ops.prims.convert_element_type.default(view_2346, torch.float32);  view_2346 = None
	        sub_6851 = torch.ops.aten.sub.Tensor(convert_element_type_898, convert_element_type_899);  convert_element_type_898 = convert_element_type_899 = None
	        mul_14504 = torch.ops.aten.mul.Tensor(sub_6851, view_2345);  sub_6851 = view_2345 = None
	        view_2347 = torch.ops.aten.view.default(mul_14504, [1280, 5120]);  mul_14504 = None
	        _assert_tensor_metadata_1351 = torch.ops.aten._assert_tensor_metadata.default(view_2347, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1351 = None
	        mul_14509 = sym_size_int * 1500
	        view_2348 = torch.ops.aten.view.default(mul_14499, [mul_14509, 5120]);  mul_14499 = mul_14509 = None
	        permute_250 = torch.ops.aten.permute.default(view_2347, [1, 0]);  view_2347 = None
	        addmm_124 = torch.ops.aten.addmm.default(model_audio_tower_layers_24_fc2_bias, view_2348, permute_250);  model_audio_tower_layers_24_fc2_bias = view_2348 = permute_250 = None
	        view_2349 = torch.ops.aten.view.default(addmm_124, [sym_size_int, 1500, 1280]);  addmm_124 = None
	        add_22976 = torch.ops.aten.add.Tensor(add_22678, view_2349);  add_22678 = view_2349 = None
	        clone_201 = torch.ops.aten.clone.default(add_22976, memory_format = torch.contiguous_format)
	        var_mean_50 = torch.ops.aten.var_mean.correction(clone_201, [2], correction = 0, keepdim = True)
	        getitem_200 = var_mean_50[0]
	        getitem_201 = var_mean_50[1];  var_mean_50 = None
	        add_22981 = torch.ops.aten.add.Tensor(getitem_200, 1e-05);  getitem_200 = None
	        rsqrt_50 = torch.ops.aten.rsqrt.default(add_22981);  add_22981 = None
	        sub_6857 = torch.ops.aten.sub.Tensor(clone_201, getitem_201);  clone_201 = getitem_201 = None
	        mul_14520 = torch.ops.aten.mul.Tensor(sub_6857, rsqrt_50);  sub_6857 = rsqrt_50 = None
	        mul_14521 = torch.ops.aten.mul.Tensor(mul_14520, model_audio_tower_layers_25_self_attn_layer_norm_weight);  mul_14520 = model_audio_tower_layers_25_self_attn_layer_norm_weight = None
	        add_22982 = torch.ops.aten.add.Tensor(mul_14521, model_audio_tower_layers_25_self_attn_layer_norm_bias);  mul_14521 = model_audio_tower_layers_25_self_attn_layer_norm_bias = None
	        amin_150 = torch.ops.aten.amin.default(add_22982, [2])
	        amax_150 = torch.ops.aten.amax.default(add_22982, [2])
	        full_300 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_150 = torch.ops.aten.minimum.default(amin_150, full_300);  amin_150 = full_300 = None
	        full_301 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_150 = torch.ops.aten.maximum.default(amax_150, full_301);  amax_150 = full_301 = None
	        sub_6868 = torch.ops.aten.sub.Tensor(maximum_150, minimum_150);  maximum_150 = None
	        div_300 = torch.ops.aten.div.Tensor(sub_6868, 255.0);  sub_6868 = None
	        clamp_min_450 = torch.ops.aten.clamp_min.default(div_300, 1.1920928955078125e-07);  div_300 = None
	        div_301 = torch.ops.aten.div.Tensor(minimum_150, clamp_min_450);  minimum_150 = None
	        round_301 = torch.ops.aten.round.default(div_301);  div_301 = None
	        sub_6874 = torch.ops.aten.sub.Tensor(-128, round_301);  round_301 = None
	        clamp_min_451 = torch.ops.aten.clamp_min.default(sub_6874, -128);  sub_6874 = None
	        clamp_max_300 = torch.ops.aten.clamp_max.default(clamp_min_451, 127);  clamp_min_451 = None
	        _assert_tensor_metadata_1352 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_450, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1352 = None
	        _assert_tensor_metadata_1353 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_300, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1353 = None
	        convert_element_type_900 = torch.ops.prims.convert_element_type.default(clamp_max_300, torch.int8);  clamp_max_300 = None
	        view_2352 = torch.ops.aten.view.default(clamp_min_450, [sym_size_int, 1500, 1])
	        view_2353 = torch.ops.aten.view.default(convert_element_type_900, [sym_size_int, 1500, 1])
	        reciprocal_150 = torch.ops.aten.reciprocal.default(view_2352);  view_2352 = None
	        mul_14569 = torch.ops.aten.mul.Tensor(reciprocal_150, 1.0);  reciprocal_150 = None
	        mul_14572 = torch.ops.aten.mul.Tensor(add_22982, mul_14569);  mul_14569 = None
	        round_302 = torch.ops.aten.round.default(mul_14572);  mul_14572 = None
	        add_23069 = torch.ops.aten.add.Tensor(round_302, view_2353);  round_302 = view_2353 = None
	        clamp_min_452 = torch.ops.aten.clamp_min.default(add_23069, -128);  add_23069 = None
	        clamp_max_301 = torch.ops.aten.clamp_max.default(clamp_min_452, 127);  clamp_min_452 = None
	        _assert_tensor_metadata_1354 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_301, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1354 = None
	        convert_element_type_901 = torch.ops.prims.convert_element_type.default(clamp_max_301, torch.int8);  clamp_max_301 = None
	        view_2356 = torch.ops.aten.view.default(clamp_min_450, [sym_size_int, 1500, 1]);  clamp_min_450 = None
	        view_2357 = torch.ops.aten.view.default(convert_element_type_900, [sym_size_int, 1500, 1]);  convert_element_type_900 = None
	        _assert_tensor_metadata_1355 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_901, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1355 = None
	        convert_element_type_902 = torch.ops.prims.convert_element_type.default(convert_element_type_901, torch.float32);  convert_element_type_901 = None
	        _assert_tensor_metadata_1356 = torch.ops.aten._assert_tensor_metadata.default(view_2357, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1356 = None
	        convert_element_type_903 = torch.ops.prims.convert_element_type.default(view_2357, torch.float32);  view_2357 = None
	        sub_6894 = torch.ops.aten.sub.Tensor(convert_element_type_902, convert_element_type_903);  convert_element_type_902 = convert_element_type_903 = None
	        mul_14594 = torch.ops.aten.mul.Tensor(sub_6894, view_2356);  sub_6894 = view_2356 = None
	        _assert_tensor_metadata_1357 = torch.ops.aten._assert_tensor_metadata.default(mul_14594, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1357 = None
	        view_2359 = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2360 = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2361 = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1358 = torch.ops.aten._assert_tensor_metadata.default(view_2359, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1358 = None
	        convert_element_type_904 = torch.ops.prims.convert_element_type.default(view_2359, torch.float32);  view_2359 = None
	        _assert_tensor_metadata_1359 = torch.ops.aten._assert_tensor_metadata.default(view_2361, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1359 = None
	        convert_element_type_905 = torch.ops.prims.convert_element_type.default(view_2361, torch.float32);  view_2361 = None
	        sub_6898 = torch.ops.aten.sub.Tensor(convert_element_type_904, convert_element_type_905);  convert_element_type_904 = convert_element_type_905 = None
	        mul_14599 = torch.ops.aten.mul.Tensor(sub_6898, view_2360);  sub_6898 = view_2360 = None
	        view_2362 = torch.ops.aten.view.default(mul_14599, [1280, 1280]);  mul_14599 = None
	        _assert_tensor_metadata_1360 = torch.ops.aten._assert_tensor_metadata.default(view_2362, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1360 = None
	        mul_14604 = sym_size_int * 1500
	        view_2363 = torch.ops.aten.view.default(mul_14594, [mul_14604, 1280]);  mul_14594 = mul_14604 = None
	        permute_251 = torch.ops.aten.permute.default(view_2362, [1, 0]);  view_2362 = None
	        addmm_125 = torch.ops.aten.addmm.default(model_audio_tower_layers_25_self_attn_q_proj_bias, view_2363, permute_251);  model_audio_tower_layers_25_self_attn_q_proj_bias = view_2363 = permute_251 = None
	        view_2364 = torch.ops.aten.view.default(addmm_125, [sym_size_int, 1500, 1280]);  addmm_125 = None
	        mul_14611 = torch.ops.aten.mul.Tensor(view_2364, 0.125);  view_2364 = None
	        view_2365 = torch.ops.aten.view.default(mul_14611, [sym_size_int, 1500, 20, 64]);  mul_14611 = None
	        permute_252 = torch.ops.aten.permute.default(view_2365, [0, 2, 1, 3]);  view_2365 = None
	        clone_202 = torch.ops.aten.clone.default(permute_252, memory_format = torch.contiguous_format);  permute_252 = None
	        amin_151 = torch.ops.aten.amin.default(add_22982, [2])
	        amax_151 = torch.ops.aten.amax.default(add_22982, [2])
	        full_302 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_151 = torch.ops.aten.minimum.default(amin_151, full_302);  amin_151 = full_302 = None
	        full_303 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_151 = torch.ops.aten.maximum.default(amax_151, full_303);  amax_151 = full_303 = None
	        sub_6913 = torch.ops.aten.sub.Tensor(maximum_151, minimum_151);  maximum_151 = None
	        div_302 = torch.ops.aten.div.Tensor(sub_6913, 255.0);  sub_6913 = None
	        clamp_min_453 = torch.ops.aten.clamp_min.default(div_302, 1.1920928955078125e-07);  div_302 = None
	        div_303 = torch.ops.aten.div.Tensor(minimum_151, clamp_min_453);  minimum_151 = None
	        round_303 = torch.ops.aten.round.default(div_303);  div_303 = None
	        sub_6919 = torch.ops.aten.sub.Tensor(-128, round_303);  round_303 = None
	        clamp_min_454 = torch.ops.aten.clamp_min.default(sub_6919, -128);  sub_6919 = None
	        clamp_max_302 = torch.ops.aten.clamp_max.default(clamp_min_454, 127);  clamp_min_454 = None
	        _assert_tensor_metadata_1361 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_453, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1361 = None
	        _assert_tensor_metadata_1362 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_302, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1362 = None
	        convert_element_type_906 = torch.ops.prims.convert_element_type.default(clamp_max_302, torch.int8);  clamp_max_302 = None
	        view_2368 = torch.ops.aten.view.default(clamp_min_453, [sym_size_int, 1500, 1])
	        view_2369 = torch.ops.aten.view.default(convert_element_type_906, [sym_size_int, 1500, 1])
	        reciprocal_151 = torch.ops.aten.reciprocal.default(view_2368);  view_2368 = None
	        mul_14665 = torch.ops.aten.mul.Tensor(reciprocal_151, 1.0);  reciprocal_151 = None
	        mul_14668 = torch.ops.aten.mul.Tensor(add_22982, mul_14665);  mul_14665 = None
	        round_304 = torch.ops.aten.round.default(mul_14668);  mul_14668 = None
	        add_23221 = torch.ops.aten.add.Tensor(round_304, view_2369);  round_304 = view_2369 = None
	        clamp_min_455 = torch.ops.aten.clamp_min.default(add_23221, -128);  add_23221 = None
	        clamp_max_303 = torch.ops.aten.clamp_max.default(clamp_min_455, 127);  clamp_min_455 = None
	        _assert_tensor_metadata_1363 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_303, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1363 = None
	        convert_element_type_907 = torch.ops.prims.convert_element_type.default(clamp_max_303, torch.int8);  clamp_max_303 = None
	        view_2372 = torch.ops.aten.view.default(clamp_min_453, [sym_size_int, 1500, 1]);  clamp_min_453 = None
	        view_2373 = torch.ops.aten.view.default(convert_element_type_906, [sym_size_int, 1500, 1]);  convert_element_type_906 = None
	        _assert_tensor_metadata_1364 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_907, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1364 = None
	        convert_element_type_908 = torch.ops.prims.convert_element_type.default(convert_element_type_907, torch.float32);  convert_element_type_907 = None
	        _assert_tensor_metadata_1365 = torch.ops.aten._assert_tensor_metadata.default(view_2373, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1365 = None
	        convert_element_type_909 = torch.ops.prims.convert_element_type.default(view_2373, torch.float32);  view_2373 = None
	        sub_6939 = torch.ops.aten.sub.Tensor(convert_element_type_908, convert_element_type_909);  convert_element_type_908 = convert_element_type_909 = None
	        mul_14690 = torch.ops.aten.mul.Tensor(sub_6939, view_2372);  sub_6939 = view_2372 = None
	        _assert_tensor_metadata_1366 = torch.ops.aten._assert_tensor_metadata.default(mul_14690, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1366 = None
	        view_2375 = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2376 = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2377 = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1367 = torch.ops.aten._assert_tensor_metadata.default(view_2375, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1367 = None
	        convert_element_type_910 = torch.ops.prims.convert_element_type.default(view_2375, torch.float32);  view_2375 = None
	        _assert_tensor_metadata_1368 = torch.ops.aten._assert_tensor_metadata.default(view_2377, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1368 = None
	        convert_element_type_911 = torch.ops.prims.convert_element_type.default(view_2377, torch.float32);  view_2377 = None
	        sub_6943 = torch.ops.aten.sub.Tensor(convert_element_type_910, convert_element_type_911);  convert_element_type_910 = convert_element_type_911 = None
	        mul_14695 = torch.ops.aten.mul.Tensor(sub_6943, view_2376);  sub_6943 = view_2376 = None
	        view_2378 = torch.ops.aten.view.default(mul_14695, [1280, 1280]);  mul_14695 = None
	        _assert_tensor_metadata_1369 = torch.ops.aten._assert_tensor_metadata.default(view_2378, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1369 = None
	        permute_253 = torch.ops.aten.permute.default(view_2378, [1, 0]);  view_2378 = None
	        mul_14698 = sym_size_int * 1500
	        view_2379 = torch.ops.aten.view.default(mul_14690, [mul_14698, 1280]);  mul_14690 = mul_14698 = None
	        mm_25 = torch.ops.aten.mm.default(view_2379, permute_253);  view_2379 = permute_253 = None
	        view_2380 = torch.ops.aten.view.default(mm_25, [sym_size_int, 1500, 1280]);  mm_25 = None
	        view_2381 = torch.ops.aten.view.default(view_2380, [sym_size_int, -1, 20, 64]);  view_2380 = None
	        permute_254 = torch.ops.aten.permute.default(view_2381, [0, 2, 1, 3]);  view_2381 = None
	        clone_203 = torch.ops.aten.clone.default(permute_254, memory_format = torch.contiguous_format);  permute_254 = None
	        amin_152 = torch.ops.aten.amin.default(add_22982, [2])
	        amax_152 = torch.ops.aten.amax.default(add_22982, [2])
	        full_304 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_152 = torch.ops.aten.minimum.default(amin_152, full_304);  amin_152 = full_304 = None
	        full_305 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_152 = torch.ops.aten.maximum.default(amax_152, full_305);  amax_152 = full_305 = None
	        sub_6957 = torch.ops.aten.sub.Tensor(maximum_152, minimum_152);  maximum_152 = None
	        div_304 = torch.ops.aten.div.Tensor(sub_6957, 255.0);  sub_6957 = None
	        clamp_min_456 = torch.ops.aten.clamp_min.default(div_304, 1.1920928955078125e-07);  div_304 = None
	        div_305 = torch.ops.aten.div.Tensor(minimum_152, clamp_min_456);  minimum_152 = None
	        round_305 = torch.ops.aten.round.default(div_305);  div_305 = None
	        sub_6963 = torch.ops.aten.sub.Tensor(-128, round_305);  round_305 = None
	        clamp_min_457 = torch.ops.aten.clamp_min.default(sub_6963, -128);  sub_6963 = None
	        clamp_max_304 = torch.ops.aten.clamp_max.default(clamp_min_457, 127);  clamp_min_457 = None
	        _assert_tensor_metadata_1370 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_456, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1370 = None
	        _assert_tensor_metadata_1371 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_304, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1371 = None
	        convert_element_type_912 = torch.ops.prims.convert_element_type.default(clamp_max_304, torch.int8);  clamp_max_304 = None
	        view_2384 = torch.ops.aten.view.default(clamp_min_456, [sym_size_int, 1500, 1])
	        view_2385 = torch.ops.aten.view.default(convert_element_type_912, [sym_size_int, 1500, 1])
	        reciprocal_152 = torch.ops.aten.reciprocal.default(view_2384);  view_2384 = None
	        mul_14764 = torch.ops.aten.mul.Tensor(reciprocal_152, 1.0);  reciprocal_152 = None
	        mul_14767 = torch.ops.aten.mul.Tensor(add_22982, mul_14764);  add_22982 = mul_14764 = None
	        round_306 = torch.ops.aten.round.default(mul_14767);  mul_14767 = None
	        add_23369 = torch.ops.aten.add.Tensor(round_306, view_2385);  round_306 = view_2385 = None
	        clamp_min_458 = torch.ops.aten.clamp_min.default(add_23369, -128);  add_23369 = None
	        clamp_max_305 = torch.ops.aten.clamp_max.default(clamp_min_458, 127);  clamp_min_458 = None
	        _assert_tensor_metadata_1372 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_305, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1372 = None
	        convert_element_type_913 = torch.ops.prims.convert_element_type.default(clamp_max_305, torch.int8);  clamp_max_305 = None
	        view_2388 = torch.ops.aten.view.default(clamp_min_456, [sym_size_int, 1500, 1]);  clamp_min_456 = None
	        view_2389 = torch.ops.aten.view.default(convert_element_type_912, [sym_size_int, 1500, 1]);  convert_element_type_912 = None
	        _assert_tensor_metadata_1373 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_913, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1373 = None
	        convert_element_type_914 = torch.ops.prims.convert_element_type.default(convert_element_type_913, torch.float32);  convert_element_type_913 = None
	        _assert_tensor_metadata_1374 = torch.ops.aten._assert_tensor_metadata.default(view_2389, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1374 = None
	        convert_element_type_915 = torch.ops.prims.convert_element_type.default(view_2389, torch.float32);  view_2389 = None
	        sub_6983 = torch.ops.aten.sub.Tensor(convert_element_type_914, convert_element_type_915);  convert_element_type_914 = convert_element_type_915 = None
	        mul_14789 = torch.ops.aten.mul.Tensor(sub_6983, view_2388);  sub_6983 = view_2388 = None
	        _assert_tensor_metadata_1375 = torch.ops.aten._assert_tensor_metadata.default(mul_14789, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1375 = None
	        view_2391 = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2392 = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2393 = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1376 = torch.ops.aten._assert_tensor_metadata.default(view_2391, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1376 = None
	        convert_element_type_916 = torch.ops.prims.convert_element_type.default(view_2391, torch.float32);  view_2391 = None
	        _assert_tensor_metadata_1377 = torch.ops.aten._assert_tensor_metadata.default(view_2393, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1377 = None
	        convert_element_type_917 = torch.ops.prims.convert_element_type.default(view_2393, torch.float32);  view_2393 = None
	        sub_6987 = torch.ops.aten.sub.Tensor(convert_element_type_916, convert_element_type_917);  convert_element_type_916 = convert_element_type_917 = None
	        mul_14794 = torch.ops.aten.mul.Tensor(sub_6987, view_2392);  sub_6987 = view_2392 = None
	        view_2394 = torch.ops.aten.view.default(mul_14794, [1280, 1280]);  mul_14794 = None
	        _assert_tensor_metadata_1378 = torch.ops.aten._assert_tensor_metadata.default(view_2394, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1378 = None
	        mul_14799 = sym_size_int * 1500
	        view_2395 = torch.ops.aten.view.default(mul_14789, [mul_14799, 1280]);  mul_14789 = mul_14799 = None
	        permute_255 = torch.ops.aten.permute.default(view_2394, [1, 0]);  view_2394 = None
	        addmm_126 = torch.ops.aten.addmm.default(model_audio_tower_layers_25_self_attn_v_proj_bias, view_2395, permute_255);  model_audio_tower_layers_25_self_attn_v_proj_bias = view_2395 = permute_255 = None
	        view_2396 = torch.ops.aten.view.default(addmm_126, [sym_size_int, 1500, 1280]);  addmm_126 = None
	        view_2397 = torch.ops.aten.view.default(view_2396, [sym_size_int, -1, 20, 64]);  view_2396 = None
	        permute_256 = torch.ops.aten.permute.default(view_2397, [0, 2, 1, 3]);  view_2397 = None
	        clone_204 = torch.ops.aten.clone.default(permute_256, memory_format = torch.contiguous_format);  permute_256 = None
	        _scaled_dot_product_efficient_attention_25 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_202, clone_203, clone_204, None, False, scale = 1.0);  clone_202 = clone_203 = clone_204 = None
	        getitem_202 = _scaled_dot_product_efficient_attention_25[0];  _scaled_dot_product_efficient_attention_25 = None
	        permute_257 = torch.ops.aten.permute.default(getitem_202, [0, 2, 1, 3]);  getitem_202 = None
	        view_2398 = torch.ops.aten.view.default(permute_257, [sym_size_int, 1500, -1]);  permute_257 = None
	        amin_153 = torch.ops.aten.amin.default(view_2398, [2])
	        amax_153 = torch.ops.aten.amax.default(view_2398, [2])
	        full_306 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_153 = torch.ops.aten.minimum.default(amin_153, full_306);  amin_153 = full_306 = None
	        full_307 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_153 = torch.ops.aten.maximum.default(amax_153, full_307);  amax_153 = full_307 = None
	        sub_7005 = torch.ops.aten.sub.Tensor(maximum_153, minimum_153);  maximum_153 = None
	        div_306 = torch.ops.aten.div.Tensor(sub_7005, 255.0);  sub_7005 = None
	        clamp_min_459 = torch.ops.aten.clamp_min.default(div_306, 1.1920928955078125e-07);  div_306 = None
	        div_307 = torch.ops.aten.div.Tensor(minimum_153, clamp_min_459);  minimum_153 = None
	        round_307 = torch.ops.aten.round.default(div_307);  div_307 = None
	        sub_7011 = torch.ops.aten.sub.Tensor(-128, round_307);  round_307 = None
	        clamp_min_460 = torch.ops.aten.clamp_min.default(sub_7011, -128);  sub_7011 = None
	        clamp_max_306 = torch.ops.aten.clamp_max.default(clamp_min_460, 127);  clamp_min_460 = None
	        _assert_tensor_metadata_1379 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_459, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1379 = None
	        _assert_tensor_metadata_1380 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_306, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1380 = None
	        convert_element_type_918 = torch.ops.prims.convert_element_type.default(clamp_max_306, torch.int8);  clamp_max_306 = None
	        view_2401 = torch.ops.aten.view.default(clamp_min_459, [sym_size_int, 1500, 1])
	        view_2402 = torch.ops.aten.view.default(convert_element_type_918, [sym_size_int, 1500, 1])
	        reciprocal_153 = torch.ops.aten.reciprocal.default(view_2401);  view_2401 = None
	        mul_14869 = torch.ops.aten.mul.Tensor(reciprocal_153, 1.0);  reciprocal_153 = None
	        mul_14872 = torch.ops.aten.mul.Tensor(view_2398, mul_14869);  view_2398 = mul_14869 = None
	        round_308 = torch.ops.aten.round.default(mul_14872);  mul_14872 = None
	        add_23533 = torch.ops.aten.add.Tensor(round_308, view_2402);  round_308 = view_2402 = None
	        clamp_min_461 = torch.ops.aten.clamp_min.default(add_23533, -128);  add_23533 = None
	        clamp_max_307 = torch.ops.aten.clamp_max.default(clamp_min_461, 127);  clamp_min_461 = None
	        _assert_tensor_metadata_1381 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_307, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1381 = None
	        convert_element_type_919 = torch.ops.prims.convert_element_type.default(clamp_max_307, torch.int8);  clamp_max_307 = None
	        view_2405 = torch.ops.aten.view.default(clamp_min_459, [sym_size_int, 1500, 1]);  clamp_min_459 = None
	        view_2406 = torch.ops.aten.view.default(convert_element_type_918, [sym_size_int, 1500, 1]);  convert_element_type_918 = None
	        _assert_tensor_metadata_1382 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_919, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1382 = None
	        convert_element_type_920 = torch.ops.prims.convert_element_type.default(convert_element_type_919, torch.float32);  convert_element_type_919 = None
	        _assert_tensor_metadata_1383 = torch.ops.aten._assert_tensor_metadata.default(view_2406, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1383 = None
	        convert_element_type_921 = torch.ops.prims.convert_element_type.default(view_2406, torch.float32);  view_2406 = None
	        sub_7031 = torch.ops.aten.sub.Tensor(convert_element_type_920, convert_element_type_921);  convert_element_type_920 = convert_element_type_921 = None
	        mul_14894 = torch.ops.aten.mul.Tensor(sub_7031, view_2405);  sub_7031 = view_2405 = None
	        _assert_tensor_metadata_1384 = torch.ops.aten._assert_tensor_metadata.default(mul_14894, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1384 = None
	        view_2408 = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2409 = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2410 = torch.ops.aten.view.default(model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1385 = torch.ops.aten._assert_tensor_metadata.default(view_2408, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1385 = None
	        convert_element_type_922 = torch.ops.prims.convert_element_type.default(view_2408, torch.float32);  view_2408 = None
	        _assert_tensor_metadata_1386 = torch.ops.aten._assert_tensor_metadata.default(view_2410, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1386 = None
	        convert_element_type_923 = torch.ops.prims.convert_element_type.default(view_2410, torch.float32);  view_2410 = None
	        sub_7035 = torch.ops.aten.sub.Tensor(convert_element_type_922, convert_element_type_923);  convert_element_type_922 = convert_element_type_923 = None
	        mul_14899 = torch.ops.aten.mul.Tensor(sub_7035, view_2409);  sub_7035 = view_2409 = None
	        view_2411 = torch.ops.aten.view.default(mul_14899, [1280, 1280]);  mul_14899 = None
	        _assert_tensor_metadata_1387 = torch.ops.aten._assert_tensor_metadata.default(view_2411, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1387 = None
	        mul_14904 = sym_size_int * 1500
	        view_2412 = torch.ops.aten.view.default(mul_14894, [mul_14904, 1280]);  mul_14894 = mul_14904 = None
	        permute_258 = torch.ops.aten.permute.default(view_2411, [1, 0]);  view_2411 = None
	        addmm_127 = torch.ops.aten.addmm.default(model_audio_tower_layers_25_self_attn_out_proj_bias, view_2412, permute_258);  model_audio_tower_layers_25_self_attn_out_proj_bias = view_2412 = permute_258 = None
	        view_2413 = torch.ops.aten.view.default(addmm_127, [sym_size_int, 1500, 1280]);  addmm_127 = None
	        add_23596 = torch.ops.aten.add.Tensor(add_22976, view_2413);  add_22976 = view_2413 = None
	        clone_206 = torch.ops.aten.clone.default(add_23596, memory_format = torch.contiguous_format)
	        var_mean_51 = torch.ops.aten.var_mean.correction(clone_206, [2], correction = 0, keepdim = True)
	        getitem_206 = var_mean_51[0]
	        getitem_207 = var_mean_51[1];  var_mean_51 = None
	        add_23601 = torch.ops.aten.add.Tensor(getitem_206, 1e-05);  getitem_206 = None
	        rsqrt_51 = torch.ops.aten.rsqrt.default(add_23601);  add_23601 = None
	        sub_7041 = torch.ops.aten.sub.Tensor(clone_206, getitem_207);  clone_206 = getitem_207 = None
	        mul_14915 = torch.ops.aten.mul.Tensor(sub_7041, rsqrt_51);  sub_7041 = rsqrt_51 = None
	        mul_14916 = torch.ops.aten.mul.Tensor(mul_14915, model_audio_tower_layers_25_final_layer_norm_weight);  mul_14915 = model_audio_tower_layers_25_final_layer_norm_weight = None
	        add_23602 = torch.ops.aten.add.Tensor(mul_14916, model_audio_tower_layers_25_final_layer_norm_bias);  mul_14916 = model_audio_tower_layers_25_final_layer_norm_bias = None
	        amin_154 = torch.ops.aten.amin.default(add_23602, [2])
	        amax_154 = torch.ops.aten.amax.default(add_23602, [2])
	        full_308 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_154 = torch.ops.aten.minimum.default(amin_154, full_308);  amin_154 = full_308 = None
	        full_309 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_154 = torch.ops.aten.maximum.default(amax_154, full_309);  amax_154 = full_309 = None
	        sub_7052 = torch.ops.aten.sub.Tensor(maximum_154, minimum_154);  maximum_154 = None
	        div_308 = torch.ops.aten.div.Tensor(sub_7052, 255.0);  sub_7052 = None
	        clamp_min_462 = torch.ops.aten.clamp_min.default(div_308, 1.1920928955078125e-07);  div_308 = None
	        div_309 = torch.ops.aten.div.Tensor(minimum_154, clamp_min_462);  minimum_154 = None
	        round_309 = torch.ops.aten.round.default(div_309);  div_309 = None
	        sub_7058 = torch.ops.aten.sub.Tensor(-128, round_309);  round_309 = None
	        clamp_min_463 = torch.ops.aten.clamp_min.default(sub_7058, -128);  sub_7058 = None
	        clamp_max_308 = torch.ops.aten.clamp_max.default(clamp_min_463, 127);  clamp_min_463 = None
	        _assert_tensor_metadata_1388 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_462, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1388 = None
	        _assert_tensor_metadata_1389 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_308, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1389 = None
	        convert_element_type_924 = torch.ops.prims.convert_element_type.default(clamp_max_308, torch.int8);  clamp_max_308 = None
	        view_2416 = torch.ops.aten.view.default(clamp_min_462, [sym_size_int, 1500, 1])
	        view_2417 = torch.ops.aten.view.default(convert_element_type_924, [sym_size_int, 1500, 1])
	        reciprocal_154 = torch.ops.aten.reciprocal.default(view_2416);  view_2416 = None
	        mul_14964 = torch.ops.aten.mul.Tensor(reciprocal_154, 1.0);  reciprocal_154 = None
	        mul_14967 = torch.ops.aten.mul.Tensor(add_23602, mul_14964);  add_23602 = mul_14964 = None
	        round_310 = torch.ops.aten.round.default(mul_14967);  mul_14967 = None
	        add_23689 = torch.ops.aten.add.Tensor(round_310, view_2417);  round_310 = view_2417 = None
	        clamp_min_464 = torch.ops.aten.clamp_min.default(add_23689, -128);  add_23689 = None
	        clamp_max_309 = torch.ops.aten.clamp_max.default(clamp_min_464, 127);  clamp_min_464 = None
	        _assert_tensor_metadata_1390 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_309, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1390 = None
	        convert_element_type_925 = torch.ops.prims.convert_element_type.default(clamp_max_309, torch.int8);  clamp_max_309 = None
	        view_2420 = torch.ops.aten.view.default(clamp_min_462, [sym_size_int, 1500, 1]);  clamp_min_462 = None
	        view_2421 = torch.ops.aten.view.default(convert_element_type_924, [sym_size_int, 1500, 1]);  convert_element_type_924 = None
	        _assert_tensor_metadata_1391 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_925, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1391 = None
	        convert_element_type_926 = torch.ops.prims.convert_element_type.default(convert_element_type_925, torch.float32);  convert_element_type_925 = None
	        _assert_tensor_metadata_1392 = torch.ops.aten._assert_tensor_metadata.default(view_2421, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1392 = None
	        convert_element_type_927 = torch.ops.prims.convert_element_type.default(view_2421, torch.float32);  view_2421 = None
	        sub_7078 = torch.ops.aten.sub.Tensor(convert_element_type_926, convert_element_type_927);  convert_element_type_926 = convert_element_type_927 = None
	        mul_14989 = torch.ops.aten.mul.Tensor(sub_7078, view_2420);  sub_7078 = view_2420 = None
	        _assert_tensor_metadata_1393 = torch.ops.aten._assert_tensor_metadata.default(mul_14989, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1393 = None
	        view_2423 = torch.ops.aten.view.default(model_audio_tower_layers_25_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_25_fc1_parametrizations_weight_original0 = None
	        view_2424 = torch.ops.aten.view.default(model_audio_tower_layers_25_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_25_fc1_parametrizations_weight_original1 = None
	        view_2425 = torch.ops.aten.view.default(model_audio_tower_layers_25_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_25_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1394 = torch.ops.aten._assert_tensor_metadata.default(view_2423, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1394 = None
	        convert_element_type_928 = torch.ops.prims.convert_element_type.default(view_2423, torch.float32);  view_2423 = None
	        _assert_tensor_metadata_1395 = torch.ops.aten._assert_tensor_metadata.default(view_2425, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1395 = None
	        convert_element_type_929 = torch.ops.prims.convert_element_type.default(view_2425, torch.float32);  view_2425 = None
	        sub_7082 = torch.ops.aten.sub.Tensor(convert_element_type_928, convert_element_type_929);  convert_element_type_928 = convert_element_type_929 = None
	        mul_14994 = torch.ops.aten.mul.Tensor(sub_7082, view_2424);  sub_7082 = view_2424 = None
	        view_2426 = torch.ops.aten.view.default(mul_14994, [5120, 1280]);  mul_14994 = None
	        _assert_tensor_metadata_1396 = torch.ops.aten._assert_tensor_metadata.default(view_2426, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1396 = None
	        mul_14999 = sym_size_int * 1500
	        view_2427 = torch.ops.aten.view.default(mul_14989, [mul_14999, 1280]);  mul_14989 = mul_14999 = None
	        permute_259 = torch.ops.aten.permute.default(view_2426, [1, 0]);  view_2426 = None
	        addmm_128 = torch.ops.aten.addmm.default(model_audio_tower_layers_25_fc1_bias, view_2427, permute_259);  model_audio_tower_layers_25_fc1_bias = view_2427 = permute_259 = None
	        view_2428 = torch.ops.aten.view.default(addmm_128, [sym_size_int, 1500, 5120]);  addmm_128 = None
	        mul_15006 = torch.ops.aten.mul.Tensor(view_2428, 0.5)
	        mul_15007 = torch.ops.aten.mul.Tensor(view_2428, 0.7071067811865476);  view_2428 = None
	        erf_27 = torch.ops.aten.erf.default(mul_15007);  mul_15007 = None
	        add_23748 = torch.ops.aten.add.Tensor(erf_27, 1);  erf_27 = None
	        mul_15008 = torch.ops.aten.mul.Tensor(mul_15006, add_23748);  mul_15006 = add_23748 = None
	        amin_155 = torch.ops.aten.amin.default(mul_15008, [2])
	        amax_155 = torch.ops.aten.amax.default(mul_15008, [2])
	        full_310 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_155 = torch.ops.aten.minimum.default(amin_155, full_310);  amin_155 = full_310 = None
	        full_311 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_155 = torch.ops.aten.maximum.default(amax_155, full_311);  amax_155 = full_311 = None
	        sub_7095 = torch.ops.aten.sub.Tensor(maximum_155, minimum_155);  maximum_155 = None
	        div_310 = torch.ops.aten.div.Tensor(sub_7095, 255.0);  sub_7095 = None
	        clamp_min_465 = torch.ops.aten.clamp_min.default(div_310, 1.1920928955078125e-07);  div_310 = None
	        div_311 = torch.ops.aten.div.Tensor(minimum_155, clamp_min_465);  minimum_155 = None
	        round_311 = torch.ops.aten.round.default(div_311);  div_311 = None
	        sub_7101 = torch.ops.aten.sub.Tensor(-128, round_311);  round_311 = None
	        clamp_min_466 = torch.ops.aten.clamp_min.default(sub_7101, -128);  sub_7101 = None
	        clamp_max_310 = torch.ops.aten.clamp_max.default(clamp_min_466, 127);  clamp_min_466 = None
	        _assert_tensor_metadata_1397 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_465, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1397 = None
	        _assert_tensor_metadata_1398 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_310, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1398 = None
	        convert_element_type_930 = torch.ops.prims.convert_element_type.default(clamp_max_310, torch.int8);  clamp_max_310 = None
	        view_2431 = torch.ops.aten.view.default(clamp_min_465, [sym_size_int, 1500, 1])
	        view_2432 = torch.ops.aten.view.default(convert_element_type_930, [sym_size_int, 1500, 1])
	        reciprocal_155 = torch.ops.aten.reciprocal.default(view_2431);  view_2431 = None
	        mul_15054 = torch.ops.aten.mul.Tensor(reciprocal_155, 1.0);  reciprocal_155 = None
	        mul_15057 = torch.ops.aten.mul.Tensor(mul_15008, mul_15054);  mul_15008 = mul_15054 = None
	        round_312 = torch.ops.aten.round.default(mul_15057);  mul_15057 = None
	        add_23831 = torch.ops.aten.add.Tensor(round_312, view_2432);  round_312 = view_2432 = None
	        clamp_min_467 = torch.ops.aten.clamp_min.default(add_23831, -128);  add_23831 = None
	        clamp_max_311 = torch.ops.aten.clamp_max.default(clamp_min_467, 127);  clamp_min_467 = None
	        _assert_tensor_metadata_1399 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_311, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1399 = None
	        convert_element_type_931 = torch.ops.prims.convert_element_type.default(clamp_max_311, torch.int8);  clamp_max_311 = None
	        view_2435 = torch.ops.aten.view.default(clamp_min_465, [sym_size_int, 1500, 1]);  clamp_min_465 = None
	        view_2436 = torch.ops.aten.view.default(convert_element_type_930, [sym_size_int, 1500, 1]);  convert_element_type_930 = None
	        _assert_tensor_metadata_1400 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_931, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1400 = None
	        convert_element_type_932 = torch.ops.prims.convert_element_type.default(convert_element_type_931, torch.float32);  convert_element_type_931 = None
	        _assert_tensor_metadata_1401 = torch.ops.aten._assert_tensor_metadata.default(view_2436, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1401 = None
	        convert_element_type_933 = torch.ops.prims.convert_element_type.default(view_2436, torch.float32);  view_2436 = None
	        sub_7121 = torch.ops.aten.sub.Tensor(convert_element_type_932, convert_element_type_933);  convert_element_type_932 = convert_element_type_933 = None
	        mul_15079 = torch.ops.aten.mul.Tensor(sub_7121, view_2435);  sub_7121 = view_2435 = None
	        _assert_tensor_metadata_1402 = torch.ops.aten._assert_tensor_metadata.default(mul_15079, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1402 = None
	        view_2438 = torch.ops.aten.view.default(model_audio_tower_layers_25_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_25_fc2_parametrizations_weight_original0 = None
	        view_2439 = torch.ops.aten.view.default(model_audio_tower_layers_25_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_25_fc2_parametrizations_weight_original1 = None
	        view_2440 = torch.ops.aten.view.default(model_audio_tower_layers_25_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_25_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1403 = torch.ops.aten._assert_tensor_metadata.default(view_2438, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1403 = None
	        convert_element_type_934 = torch.ops.prims.convert_element_type.default(view_2438, torch.float32);  view_2438 = None
	        _assert_tensor_metadata_1404 = torch.ops.aten._assert_tensor_metadata.default(view_2440, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1404 = None
	        convert_element_type_935 = torch.ops.prims.convert_element_type.default(view_2440, torch.float32);  view_2440 = None
	        sub_7125 = torch.ops.aten.sub.Tensor(convert_element_type_934, convert_element_type_935);  convert_element_type_934 = convert_element_type_935 = None
	        mul_15084 = torch.ops.aten.mul.Tensor(sub_7125, view_2439);  sub_7125 = view_2439 = None
	        view_2441 = torch.ops.aten.view.default(mul_15084, [1280, 5120]);  mul_15084 = None
	        _assert_tensor_metadata_1405 = torch.ops.aten._assert_tensor_metadata.default(view_2441, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1405 = None
	        mul_15089 = sym_size_int * 1500
	        view_2442 = torch.ops.aten.view.default(mul_15079, [mul_15089, 5120]);  mul_15079 = mul_15089 = None
	        permute_260 = torch.ops.aten.permute.default(view_2441, [1, 0]);  view_2441 = None
	        addmm_129 = torch.ops.aten.addmm.default(model_audio_tower_layers_25_fc2_bias, view_2442, permute_260);  model_audio_tower_layers_25_fc2_bias = view_2442 = permute_260 = None
	        view_2443 = torch.ops.aten.view.default(addmm_129, [sym_size_int, 1500, 1280]);  addmm_129 = None
	        add_23894 = torch.ops.aten.add.Tensor(add_23596, view_2443);  add_23596 = view_2443 = None
	        clone_209 = torch.ops.aten.clone.default(add_23894, memory_format = torch.contiguous_format)
	        var_mean_52 = torch.ops.aten.var_mean.correction(clone_209, [2], correction = 0, keepdim = True)
	        getitem_208 = var_mean_52[0]
	        getitem_209 = var_mean_52[1];  var_mean_52 = None
	        add_23899 = torch.ops.aten.add.Tensor(getitem_208, 1e-05);  getitem_208 = None
	        rsqrt_52 = torch.ops.aten.rsqrt.default(add_23899);  add_23899 = None
	        sub_7131 = torch.ops.aten.sub.Tensor(clone_209, getitem_209);  clone_209 = getitem_209 = None
	        mul_15100 = torch.ops.aten.mul.Tensor(sub_7131, rsqrt_52);  sub_7131 = rsqrt_52 = None
	        mul_15101 = torch.ops.aten.mul.Tensor(mul_15100, model_audio_tower_layers_26_self_attn_layer_norm_weight);  mul_15100 = model_audio_tower_layers_26_self_attn_layer_norm_weight = None
	        add_23900 = torch.ops.aten.add.Tensor(mul_15101, model_audio_tower_layers_26_self_attn_layer_norm_bias);  mul_15101 = model_audio_tower_layers_26_self_attn_layer_norm_bias = None
	        amin_156 = torch.ops.aten.amin.default(add_23900, [2])
	        amax_156 = torch.ops.aten.amax.default(add_23900, [2])
	        full_312 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_156 = torch.ops.aten.minimum.default(amin_156, full_312);  amin_156 = full_312 = None
	        full_313 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_156 = torch.ops.aten.maximum.default(amax_156, full_313);  amax_156 = full_313 = None
	        sub_7142 = torch.ops.aten.sub.Tensor(maximum_156, minimum_156);  maximum_156 = None
	        div_312 = torch.ops.aten.div.Tensor(sub_7142, 255.0);  sub_7142 = None
	        clamp_min_468 = torch.ops.aten.clamp_min.default(div_312, 1.1920928955078125e-07);  div_312 = None
	        div_313 = torch.ops.aten.div.Tensor(minimum_156, clamp_min_468);  minimum_156 = None
	        round_313 = torch.ops.aten.round.default(div_313);  div_313 = None
	        sub_7148 = torch.ops.aten.sub.Tensor(-128, round_313);  round_313 = None
	        clamp_min_469 = torch.ops.aten.clamp_min.default(sub_7148, -128);  sub_7148 = None
	        clamp_max_312 = torch.ops.aten.clamp_max.default(clamp_min_469, 127);  clamp_min_469 = None
	        _assert_tensor_metadata_1406 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_468, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1406 = None
	        _assert_tensor_metadata_1407 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_312, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1407 = None
	        convert_element_type_936 = torch.ops.prims.convert_element_type.default(clamp_max_312, torch.int8);  clamp_max_312 = None
	        view_2446 = torch.ops.aten.view.default(clamp_min_468, [sym_size_int, 1500, 1])
	        view_2447 = torch.ops.aten.view.default(convert_element_type_936, [sym_size_int, 1500, 1])
	        reciprocal_156 = torch.ops.aten.reciprocal.default(view_2446);  view_2446 = None
	        mul_15149 = torch.ops.aten.mul.Tensor(reciprocal_156, 1.0);  reciprocal_156 = None
	        mul_15152 = torch.ops.aten.mul.Tensor(add_23900, mul_15149);  mul_15149 = None
	        round_314 = torch.ops.aten.round.default(mul_15152);  mul_15152 = None
	        add_23987 = torch.ops.aten.add.Tensor(round_314, view_2447);  round_314 = view_2447 = None
	        clamp_min_470 = torch.ops.aten.clamp_min.default(add_23987, -128);  add_23987 = None
	        clamp_max_313 = torch.ops.aten.clamp_max.default(clamp_min_470, 127);  clamp_min_470 = None
	        _assert_tensor_metadata_1408 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_313, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1408 = None
	        convert_element_type_937 = torch.ops.prims.convert_element_type.default(clamp_max_313, torch.int8);  clamp_max_313 = None
	        view_2450 = torch.ops.aten.view.default(clamp_min_468, [sym_size_int, 1500, 1]);  clamp_min_468 = None
	        view_2451 = torch.ops.aten.view.default(convert_element_type_936, [sym_size_int, 1500, 1]);  convert_element_type_936 = None
	        _assert_tensor_metadata_1409 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_937, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1409 = None
	        convert_element_type_938 = torch.ops.prims.convert_element_type.default(convert_element_type_937, torch.float32);  convert_element_type_937 = None
	        _assert_tensor_metadata_1410 = torch.ops.aten._assert_tensor_metadata.default(view_2451, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1410 = None
	        convert_element_type_939 = torch.ops.prims.convert_element_type.default(view_2451, torch.float32);  view_2451 = None
	        sub_7168 = torch.ops.aten.sub.Tensor(convert_element_type_938, convert_element_type_939);  convert_element_type_938 = convert_element_type_939 = None
	        mul_15174 = torch.ops.aten.mul.Tensor(sub_7168, view_2450);  sub_7168 = view_2450 = None
	        _assert_tensor_metadata_1411 = torch.ops.aten._assert_tensor_metadata.default(mul_15174, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1411 = None
	        view_2453 = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2454 = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2455 = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1412 = torch.ops.aten._assert_tensor_metadata.default(view_2453, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1412 = None
	        convert_element_type_940 = torch.ops.prims.convert_element_type.default(view_2453, torch.float32);  view_2453 = None
	        _assert_tensor_metadata_1413 = torch.ops.aten._assert_tensor_metadata.default(view_2455, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1413 = None
	        convert_element_type_941 = torch.ops.prims.convert_element_type.default(view_2455, torch.float32);  view_2455 = None
	        sub_7172 = torch.ops.aten.sub.Tensor(convert_element_type_940, convert_element_type_941);  convert_element_type_940 = convert_element_type_941 = None
	        mul_15179 = torch.ops.aten.mul.Tensor(sub_7172, view_2454);  sub_7172 = view_2454 = None
	        view_2456 = torch.ops.aten.view.default(mul_15179, [1280, 1280]);  mul_15179 = None
	        _assert_tensor_metadata_1414 = torch.ops.aten._assert_tensor_metadata.default(view_2456, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1414 = None
	        mul_15184 = sym_size_int * 1500
	        view_2457 = torch.ops.aten.view.default(mul_15174, [mul_15184, 1280]);  mul_15174 = mul_15184 = None
	        permute_261 = torch.ops.aten.permute.default(view_2456, [1, 0]);  view_2456 = None
	        addmm_130 = torch.ops.aten.addmm.default(model_audio_tower_layers_26_self_attn_q_proj_bias, view_2457, permute_261);  model_audio_tower_layers_26_self_attn_q_proj_bias = view_2457 = permute_261 = None
	        view_2458 = torch.ops.aten.view.default(addmm_130, [sym_size_int, 1500, 1280]);  addmm_130 = None
	        mul_15191 = torch.ops.aten.mul.Tensor(view_2458, 0.125);  view_2458 = None
	        view_2459 = torch.ops.aten.view.default(mul_15191, [sym_size_int, 1500, 20, 64]);  mul_15191 = None
	        permute_262 = torch.ops.aten.permute.default(view_2459, [0, 2, 1, 3]);  view_2459 = None
	        clone_210 = torch.ops.aten.clone.default(permute_262, memory_format = torch.contiguous_format);  permute_262 = None
	        amin_157 = torch.ops.aten.amin.default(add_23900, [2])
	        amax_157 = torch.ops.aten.amax.default(add_23900, [2])
	        full_314 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_157 = torch.ops.aten.minimum.default(amin_157, full_314);  amin_157 = full_314 = None
	        full_315 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_157 = torch.ops.aten.maximum.default(amax_157, full_315);  amax_157 = full_315 = None
	        sub_7187 = torch.ops.aten.sub.Tensor(maximum_157, minimum_157);  maximum_157 = None
	        div_314 = torch.ops.aten.div.Tensor(sub_7187, 255.0);  sub_7187 = None
	        clamp_min_471 = torch.ops.aten.clamp_min.default(div_314, 1.1920928955078125e-07);  div_314 = None
	        div_315 = torch.ops.aten.div.Tensor(minimum_157, clamp_min_471);  minimum_157 = None
	        round_315 = torch.ops.aten.round.default(div_315);  div_315 = None
	        sub_7193 = torch.ops.aten.sub.Tensor(-128, round_315);  round_315 = None
	        clamp_min_472 = torch.ops.aten.clamp_min.default(sub_7193, -128);  sub_7193 = None
	        clamp_max_314 = torch.ops.aten.clamp_max.default(clamp_min_472, 127);  clamp_min_472 = None
	        _assert_tensor_metadata_1415 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_471, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1415 = None
	        _assert_tensor_metadata_1416 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_314, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1416 = None
	        convert_element_type_942 = torch.ops.prims.convert_element_type.default(clamp_max_314, torch.int8);  clamp_max_314 = None
	        view_2462 = torch.ops.aten.view.default(clamp_min_471, [sym_size_int, 1500, 1])
	        view_2463 = torch.ops.aten.view.default(convert_element_type_942, [sym_size_int, 1500, 1])
	        reciprocal_157 = torch.ops.aten.reciprocal.default(view_2462);  view_2462 = None
	        mul_15245 = torch.ops.aten.mul.Tensor(reciprocal_157, 1.0);  reciprocal_157 = None
	        mul_15248 = torch.ops.aten.mul.Tensor(add_23900, mul_15245);  mul_15245 = None
	        round_316 = torch.ops.aten.round.default(mul_15248);  mul_15248 = None
	        add_24139 = torch.ops.aten.add.Tensor(round_316, view_2463);  round_316 = view_2463 = None
	        clamp_min_473 = torch.ops.aten.clamp_min.default(add_24139, -128);  add_24139 = None
	        clamp_max_315 = torch.ops.aten.clamp_max.default(clamp_min_473, 127);  clamp_min_473 = None
	        _assert_tensor_metadata_1417 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_315, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1417 = None
	        convert_element_type_943 = torch.ops.prims.convert_element_type.default(clamp_max_315, torch.int8);  clamp_max_315 = None
	        view_2466 = torch.ops.aten.view.default(clamp_min_471, [sym_size_int, 1500, 1]);  clamp_min_471 = None
	        view_2467 = torch.ops.aten.view.default(convert_element_type_942, [sym_size_int, 1500, 1]);  convert_element_type_942 = None
	        _assert_tensor_metadata_1418 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_943, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1418 = None
	        convert_element_type_944 = torch.ops.prims.convert_element_type.default(convert_element_type_943, torch.float32);  convert_element_type_943 = None
	        _assert_tensor_metadata_1419 = torch.ops.aten._assert_tensor_metadata.default(view_2467, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1419 = None
	        convert_element_type_945 = torch.ops.prims.convert_element_type.default(view_2467, torch.float32);  view_2467 = None
	        sub_7213 = torch.ops.aten.sub.Tensor(convert_element_type_944, convert_element_type_945);  convert_element_type_944 = convert_element_type_945 = None
	        mul_15270 = torch.ops.aten.mul.Tensor(sub_7213, view_2466);  sub_7213 = view_2466 = None
	        _assert_tensor_metadata_1420 = torch.ops.aten._assert_tensor_metadata.default(mul_15270, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1420 = None
	        view_2469 = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2470 = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2471 = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1421 = torch.ops.aten._assert_tensor_metadata.default(view_2469, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1421 = None
	        convert_element_type_946 = torch.ops.prims.convert_element_type.default(view_2469, torch.float32);  view_2469 = None
	        _assert_tensor_metadata_1422 = torch.ops.aten._assert_tensor_metadata.default(view_2471, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1422 = None
	        convert_element_type_947 = torch.ops.prims.convert_element_type.default(view_2471, torch.float32);  view_2471 = None
	        sub_7217 = torch.ops.aten.sub.Tensor(convert_element_type_946, convert_element_type_947);  convert_element_type_946 = convert_element_type_947 = None
	        mul_15275 = torch.ops.aten.mul.Tensor(sub_7217, view_2470);  sub_7217 = view_2470 = None
	        view_2472 = torch.ops.aten.view.default(mul_15275, [1280, 1280]);  mul_15275 = None
	        _assert_tensor_metadata_1423 = torch.ops.aten._assert_tensor_metadata.default(view_2472, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1423 = None
	        permute_263 = torch.ops.aten.permute.default(view_2472, [1, 0]);  view_2472 = None
	        mul_15278 = sym_size_int * 1500
	        view_2473 = torch.ops.aten.view.default(mul_15270, [mul_15278, 1280]);  mul_15270 = mul_15278 = None
	        mm_26 = torch.ops.aten.mm.default(view_2473, permute_263);  view_2473 = permute_263 = None
	        view_2474 = torch.ops.aten.view.default(mm_26, [sym_size_int, 1500, 1280]);  mm_26 = None
	        view_2475 = torch.ops.aten.view.default(view_2474, [sym_size_int, -1, 20, 64]);  view_2474 = None
	        permute_264 = torch.ops.aten.permute.default(view_2475, [0, 2, 1, 3]);  view_2475 = None
	        clone_211 = torch.ops.aten.clone.default(permute_264, memory_format = torch.contiguous_format);  permute_264 = None
	        amin_158 = torch.ops.aten.amin.default(add_23900, [2])
	        amax_158 = torch.ops.aten.amax.default(add_23900, [2])
	        full_316 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_158 = torch.ops.aten.minimum.default(amin_158, full_316);  amin_158 = full_316 = None
	        full_317 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_158 = torch.ops.aten.maximum.default(amax_158, full_317);  amax_158 = full_317 = None
	        sub_7231 = torch.ops.aten.sub.Tensor(maximum_158, minimum_158);  maximum_158 = None
	        div_316 = torch.ops.aten.div.Tensor(sub_7231, 255.0);  sub_7231 = None
	        clamp_min_474 = torch.ops.aten.clamp_min.default(div_316, 1.1920928955078125e-07);  div_316 = None
	        div_317 = torch.ops.aten.div.Tensor(minimum_158, clamp_min_474);  minimum_158 = None
	        round_317 = torch.ops.aten.round.default(div_317);  div_317 = None
	        sub_7237 = torch.ops.aten.sub.Tensor(-128, round_317);  round_317 = None
	        clamp_min_475 = torch.ops.aten.clamp_min.default(sub_7237, -128);  sub_7237 = None
	        clamp_max_316 = torch.ops.aten.clamp_max.default(clamp_min_475, 127);  clamp_min_475 = None
	        _assert_tensor_metadata_1424 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_474, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1424 = None
	        _assert_tensor_metadata_1425 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_316, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1425 = None
	        convert_element_type_948 = torch.ops.prims.convert_element_type.default(clamp_max_316, torch.int8);  clamp_max_316 = None
	        view_2478 = torch.ops.aten.view.default(clamp_min_474, [sym_size_int, 1500, 1])
	        view_2479 = torch.ops.aten.view.default(convert_element_type_948, [sym_size_int, 1500, 1])
	        reciprocal_158 = torch.ops.aten.reciprocal.default(view_2478);  view_2478 = None
	        mul_15344 = torch.ops.aten.mul.Tensor(reciprocal_158, 1.0);  reciprocal_158 = None
	        mul_15347 = torch.ops.aten.mul.Tensor(add_23900, mul_15344);  add_23900 = mul_15344 = None
	        round_318 = torch.ops.aten.round.default(mul_15347);  mul_15347 = None
	        add_24287 = torch.ops.aten.add.Tensor(round_318, view_2479);  round_318 = view_2479 = None
	        clamp_min_476 = torch.ops.aten.clamp_min.default(add_24287, -128);  add_24287 = None
	        clamp_max_317 = torch.ops.aten.clamp_max.default(clamp_min_476, 127);  clamp_min_476 = None
	        _assert_tensor_metadata_1426 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_317, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1426 = None
	        convert_element_type_949 = torch.ops.prims.convert_element_type.default(clamp_max_317, torch.int8);  clamp_max_317 = None
	        view_2482 = torch.ops.aten.view.default(clamp_min_474, [sym_size_int, 1500, 1]);  clamp_min_474 = None
	        view_2483 = torch.ops.aten.view.default(convert_element_type_948, [sym_size_int, 1500, 1]);  convert_element_type_948 = None
	        _assert_tensor_metadata_1427 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_949, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1427 = None
	        convert_element_type_950 = torch.ops.prims.convert_element_type.default(convert_element_type_949, torch.float32);  convert_element_type_949 = None
	        _assert_tensor_metadata_1428 = torch.ops.aten._assert_tensor_metadata.default(view_2483, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1428 = None
	        convert_element_type_951 = torch.ops.prims.convert_element_type.default(view_2483, torch.float32);  view_2483 = None
	        sub_7257 = torch.ops.aten.sub.Tensor(convert_element_type_950, convert_element_type_951);  convert_element_type_950 = convert_element_type_951 = None
	        mul_15369 = torch.ops.aten.mul.Tensor(sub_7257, view_2482);  sub_7257 = view_2482 = None
	        _assert_tensor_metadata_1429 = torch.ops.aten._assert_tensor_metadata.default(mul_15369, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1429 = None
	        view_2485 = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2486 = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2487 = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1430 = torch.ops.aten._assert_tensor_metadata.default(view_2485, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1430 = None
	        convert_element_type_952 = torch.ops.prims.convert_element_type.default(view_2485, torch.float32);  view_2485 = None
	        _assert_tensor_metadata_1431 = torch.ops.aten._assert_tensor_metadata.default(view_2487, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1431 = None
	        convert_element_type_953 = torch.ops.prims.convert_element_type.default(view_2487, torch.float32);  view_2487 = None
	        sub_7261 = torch.ops.aten.sub.Tensor(convert_element_type_952, convert_element_type_953);  convert_element_type_952 = convert_element_type_953 = None
	        mul_15374 = torch.ops.aten.mul.Tensor(sub_7261, view_2486);  sub_7261 = view_2486 = None
	        view_2488 = torch.ops.aten.view.default(mul_15374, [1280, 1280]);  mul_15374 = None
	        _assert_tensor_metadata_1432 = torch.ops.aten._assert_tensor_metadata.default(view_2488, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1432 = None
	        mul_15379 = sym_size_int * 1500
	        view_2489 = torch.ops.aten.view.default(mul_15369, [mul_15379, 1280]);  mul_15369 = mul_15379 = None
	        permute_265 = torch.ops.aten.permute.default(view_2488, [1, 0]);  view_2488 = None
	        addmm_131 = torch.ops.aten.addmm.default(model_audio_tower_layers_26_self_attn_v_proj_bias, view_2489, permute_265);  model_audio_tower_layers_26_self_attn_v_proj_bias = view_2489 = permute_265 = None
	        view_2490 = torch.ops.aten.view.default(addmm_131, [sym_size_int, 1500, 1280]);  addmm_131 = None
	        view_2491 = torch.ops.aten.view.default(view_2490, [sym_size_int, -1, 20, 64]);  view_2490 = None
	        permute_266 = torch.ops.aten.permute.default(view_2491, [0, 2, 1, 3]);  view_2491 = None
	        clone_212 = torch.ops.aten.clone.default(permute_266, memory_format = torch.contiguous_format);  permute_266 = None
	        _scaled_dot_product_efficient_attention_26 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_210, clone_211, clone_212, None, False, scale = 1.0);  clone_210 = clone_211 = clone_212 = None
	        getitem_210 = _scaled_dot_product_efficient_attention_26[0];  _scaled_dot_product_efficient_attention_26 = None
	        permute_267 = torch.ops.aten.permute.default(getitem_210, [0, 2, 1, 3]);  getitem_210 = None
	        view_2492 = torch.ops.aten.view.default(permute_267, [sym_size_int, 1500, -1]);  permute_267 = None
	        amin_159 = torch.ops.aten.amin.default(view_2492, [2])
	        amax_159 = torch.ops.aten.amax.default(view_2492, [2])
	        full_318 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_159 = torch.ops.aten.minimum.default(amin_159, full_318);  amin_159 = full_318 = None
	        full_319 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_159 = torch.ops.aten.maximum.default(amax_159, full_319);  amax_159 = full_319 = None
	        sub_7279 = torch.ops.aten.sub.Tensor(maximum_159, minimum_159);  maximum_159 = None
	        div_318 = torch.ops.aten.div.Tensor(sub_7279, 255.0);  sub_7279 = None
	        clamp_min_477 = torch.ops.aten.clamp_min.default(div_318, 1.1920928955078125e-07);  div_318 = None
	        div_319 = torch.ops.aten.div.Tensor(minimum_159, clamp_min_477);  minimum_159 = None
	        round_319 = torch.ops.aten.round.default(div_319);  div_319 = None
	        sub_7285 = torch.ops.aten.sub.Tensor(-128, round_319);  round_319 = None
	        clamp_min_478 = torch.ops.aten.clamp_min.default(sub_7285, -128);  sub_7285 = None
	        clamp_max_318 = torch.ops.aten.clamp_max.default(clamp_min_478, 127);  clamp_min_478 = None
	        _assert_tensor_metadata_1433 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_477, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1433 = None
	        _assert_tensor_metadata_1434 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_318, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1434 = None
	        convert_element_type_954 = torch.ops.prims.convert_element_type.default(clamp_max_318, torch.int8);  clamp_max_318 = None
	        view_2495 = torch.ops.aten.view.default(clamp_min_477, [sym_size_int, 1500, 1])
	        view_2496 = torch.ops.aten.view.default(convert_element_type_954, [sym_size_int, 1500, 1])
	        reciprocal_159 = torch.ops.aten.reciprocal.default(view_2495);  view_2495 = None
	        mul_15449 = torch.ops.aten.mul.Tensor(reciprocal_159, 1.0);  reciprocal_159 = None
	        mul_15452 = torch.ops.aten.mul.Tensor(view_2492, mul_15449);  view_2492 = mul_15449 = None
	        round_320 = torch.ops.aten.round.default(mul_15452);  mul_15452 = None
	        add_24451 = torch.ops.aten.add.Tensor(round_320, view_2496);  round_320 = view_2496 = None
	        clamp_min_479 = torch.ops.aten.clamp_min.default(add_24451, -128);  add_24451 = None
	        clamp_max_319 = torch.ops.aten.clamp_max.default(clamp_min_479, 127);  clamp_min_479 = None
	        _assert_tensor_metadata_1435 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_319, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1435 = None
	        convert_element_type_955 = torch.ops.prims.convert_element_type.default(clamp_max_319, torch.int8);  clamp_max_319 = None
	        view_2499 = torch.ops.aten.view.default(clamp_min_477, [sym_size_int, 1500, 1]);  clamp_min_477 = None
	        view_2500 = torch.ops.aten.view.default(convert_element_type_954, [sym_size_int, 1500, 1]);  convert_element_type_954 = None
	        _assert_tensor_metadata_1436 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_955, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1436 = None
	        convert_element_type_956 = torch.ops.prims.convert_element_type.default(convert_element_type_955, torch.float32);  convert_element_type_955 = None
	        _assert_tensor_metadata_1437 = torch.ops.aten._assert_tensor_metadata.default(view_2500, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1437 = None
	        convert_element_type_957 = torch.ops.prims.convert_element_type.default(view_2500, torch.float32);  view_2500 = None
	        sub_7305 = torch.ops.aten.sub.Tensor(convert_element_type_956, convert_element_type_957);  convert_element_type_956 = convert_element_type_957 = None
	        mul_15474 = torch.ops.aten.mul.Tensor(sub_7305, view_2499);  sub_7305 = view_2499 = None
	        _assert_tensor_metadata_1438 = torch.ops.aten._assert_tensor_metadata.default(mul_15474, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1438 = None
	        view_2502 = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2503 = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2504 = torch.ops.aten.view.default(model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1439 = torch.ops.aten._assert_tensor_metadata.default(view_2502, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1439 = None
	        convert_element_type_958 = torch.ops.prims.convert_element_type.default(view_2502, torch.float32);  view_2502 = None
	        _assert_tensor_metadata_1440 = torch.ops.aten._assert_tensor_metadata.default(view_2504, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1440 = None
	        convert_element_type_959 = torch.ops.prims.convert_element_type.default(view_2504, torch.float32);  view_2504 = None
	        sub_7309 = torch.ops.aten.sub.Tensor(convert_element_type_958, convert_element_type_959);  convert_element_type_958 = convert_element_type_959 = None
	        mul_15479 = torch.ops.aten.mul.Tensor(sub_7309, view_2503);  sub_7309 = view_2503 = None
	        view_2505 = torch.ops.aten.view.default(mul_15479, [1280, 1280]);  mul_15479 = None
	        _assert_tensor_metadata_1441 = torch.ops.aten._assert_tensor_metadata.default(view_2505, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1441 = None
	        mul_15484 = sym_size_int * 1500
	        view_2506 = torch.ops.aten.view.default(mul_15474, [mul_15484, 1280]);  mul_15474 = mul_15484 = None
	        permute_268 = torch.ops.aten.permute.default(view_2505, [1, 0]);  view_2505 = None
	        addmm_132 = torch.ops.aten.addmm.default(model_audio_tower_layers_26_self_attn_out_proj_bias, view_2506, permute_268);  model_audio_tower_layers_26_self_attn_out_proj_bias = view_2506 = permute_268 = None
	        view_2507 = torch.ops.aten.view.default(addmm_132, [sym_size_int, 1500, 1280]);  addmm_132 = None
	        add_24514 = torch.ops.aten.add.Tensor(add_23894, view_2507);  add_23894 = view_2507 = None
	        clone_214 = torch.ops.aten.clone.default(add_24514, memory_format = torch.contiguous_format)
	        var_mean_53 = torch.ops.aten.var_mean.correction(clone_214, [2], correction = 0, keepdim = True)
	        getitem_214 = var_mean_53[0]
	        getitem_215 = var_mean_53[1];  var_mean_53 = None
	        add_24519 = torch.ops.aten.add.Tensor(getitem_214, 1e-05);  getitem_214 = None
	        rsqrt_53 = torch.ops.aten.rsqrt.default(add_24519);  add_24519 = None
	        sub_7315 = torch.ops.aten.sub.Tensor(clone_214, getitem_215);  clone_214 = getitem_215 = None
	        mul_15495 = torch.ops.aten.mul.Tensor(sub_7315, rsqrt_53);  sub_7315 = rsqrt_53 = None
	        mul_15496 = torch.ops.aten.mul.Tensor(mul_15495, model_audio_tower_layers_26_final_layer_norm_weight);  mul_15495 = model_audio_tower_layers_26_final_layer_norm_weight = None
	        add_24520 = torch.ops.aten.add.Tensor(mul_15496, model_audio_tower_layers_26_final_layer_norm_bias);  mul_15496 = model_audio_tower_layers_26_final_layer_norm_bias = None
	        amin_160 = torch.ops.aten.amin.default(add_24520, [2])
	        amax_160 = torch.ops.aten.amax.default(add_24520, [2])
	        full_320 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_160 = torch.ops.aten.minimum.default(amin_160, full_320);  amin_160 = full_320 = None
	        full_321 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_160 = torch.ops.aten.maximum.default(amax_160, full_321);  amax_160 = full_321 = None
	        sub_7326 = torch.ops.aten.sub.Tensor(maximum_160, minimum_160);  maximum_160 = None
	        div_320 = torch.ops.aten.div.Tensor(sub_7326, 255.0);  sub_7326 = None
	        clamp_min_480 = torch.ops.aten.clamp_min.default(div_320, 1.1920928955078125e-07);  div_320 = None
	        div_321 = torch.ops.aten.div.Tensor(minimum_160, clamp_min_480);  minimum_160 = None
	        round_321 = torch.ops.aten.round.default(div_321);  div_321 = None
	        sub_7332 = torch.ops.aten.sub.Tensor(-128, round_321);  round_321 = None
	        clamp_min_481 = torch.ops.aten.clamp_min.default(sub_7332, -128);  sub_7332 = None
	        clamp_max_320 = torch.ops.aten.clamp_max.default(clamp_min_481, 127);  clamp_min_481 = None
	        _assert_tensor_metadata_1442 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_480, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1442 = None
	        _assert_tensor_metadata_1443 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_320, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1443 = None
	        convert_element_type_960 = torch.ops.prims.convert_element_type.default(clamp_max_320, torch.int8);  clamp_max_320 = None
	        view_2510 = torch.ops.aten.view.default(clamp_min_480, [sym_size_int, 1500, 1])
	        view_2511 = torch.ops.aten.view.default(convert_element_type_960, [sym_size_int, 1500, 1])
	        reciprocal_160 = torch.ops.aten.reciprocal.default(view_2510);  view_2510 = None
	        mul_15544 = torch.ops.aten.mul.Tensor(reciprocal_160, 1.0);  reciprocal_160 = None
	        mul_15547 = torch.ops.aten.mul.Tensor(add_24520, mul_15544);  add_24520 = mul_15544 = None
	        round_322 = torch.ops.aten.round.default(mul_15547);  mul_15547 = None
	        add_24607 = torch.ops.aten.add.Tensor(round_322, view_2511);  round_322 = view_2511 = None
	        clamp_min_482 = torch.ops.aten.clamp_min.default(add_24607, -128);  add_24607 = None
	        clamp_max_321 = torch.ops.aten.clamp_max.default(clamp_min_482, 127);  clamp_min_482 = None
	        _assert_tensor_metadata_1444 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_321, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1444 = None
	        convert_element_type_961 = torch.ops.prims.convert_element_type.default(clamp_max_321, torch.int8);  clamp_max_321 = None
	        view_2514 = torch.ops.aten.view.default(clamp_min_480, [sym_size_int, 1500, 1]);  clamp_min_480 = None
	        view_2515 = torch.ops.aten.view.default(convert_element_type_960, [sym_size_int, 1500, 1]);  convert_element_type_960 = None
	        _assert_tensor_metadata_1445 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_961, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1445 = None
	        convert_element_type_962 = torch.ops.prims.convert_element_type.default(convert_element_type_961, torch.float32);  convert_element_type_961 = None
	        _assert_tensor_metadata_1446 = torch.ops.aten._assert_tensor_metadata.default(view_2515, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1446 = None
	        convert_element_type_963 = torch.ops.prims.convert_element_type.default(view_2515, torch.float32);  view_2515 = None
	        sub_7352 = torch.ops.aten.sub.Tensor(convert_element_type_962, convert_element_type_963);  convert_element_type_962 = convert_element_type_963 = None
	        mul_15569 = torch.ops.aten.mul.Tensor(sub_7352, view_2514);  sub_7352 = view_2514 = None
	        _assert_tensor_metadata_1447 = torch.ops.aten._assert_tensor_metadata.default(mul_15569, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1447 = None
	        view_2517 = torch.ops.aten.view.default(model_audio_tower_layers_26_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_26_fc1_parametrizations_weight_original0 = None
	        view_2518 = torch.ops.aten.view.default(model_audio_tower_layers_26_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_26_fc1_parametrizations_weight_original1 = None
	        view_2519 = torch.ops.aten.view.default(model_audio_tower_layers_26_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_26_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1448 = torch.ops.aten._assert_tensor_metadata.default(view_2517, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1448 = None
	        convert_element_type_964 = torch.ops.prims.convert_element_type.default(view_2517, torch.float32);  view_2517 = None
	        _assert_tensor_metadata_1449 = torch.ops.aten._assert_tensor_metadata.default(view_2519, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1449 = None
	        convert_element_type_965 = torch.ops.prims.convert_element_type.default(view_2519, torch.float32);  view_2519 = None
	        sub_7356 = torch.ops.aten.sub.Tensor(convert_element_type_964, convert_element_type_965);  convert_element_type_964 = convert_element_type_965 = None
	        mul_15574 = torch.ops.aten.mul.Tensor(sub_7356, view_2518);  sub_7356 = view_2518 = None
	        view_2520 = torch.ops.aten.view.default(mul_15574, [5120, 1280]);  mul_15574 = None
	        _assert_tensor_metadata_1450 = torch.ops.aten._assert_tensor_metadata.default(view_2520, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1450 = None
	        mul_15579 = sym_size_int * 1500
	        view_2521 = torch.ops.aten.view.default(mul_15569, [mul_15579, 1280]);  mul_15569 = mul_15579 = None
	        permute_269 = torch.ops.aten.permute.default(view_2520, [1, 0]);  view_2520 = None
	        addmm_133 = torch.ops.aten.addmm.default(model_audio_tower_layers_26_fc1_bias, view_2521, permute_269);  model_audio_tower_layers_26_fc1_bias = view_2521 = permute_269 = None
	        view_2522 = torch.ops.aten.view.default(addmm_133, [sym_size_int, 1500, 5120]);  addmm_133 = None
	        mul_15586 = torch.ops.aten.mul.Tensor(view_2522, 0.5)
	        mul_15587 = torch.ops.aten.mul.Tensor(view_2522, 0.7071067811865476);  view_2522 = None
	        erf_28 = torch.ops.aten.erf.default(mul_15587);  mul_15587 = None
	        add_24666 = torch.ops.aten.add.Tensor(erf_28, 1);  erf_28 = None
	        mul_15588 = torch.ops.aten.mul.Tensor(mul_15586, add_24666);  mul_15586 = add_24666 = None
	        amin_161 = torch.ops.aten.amin.default(mul_15588, [2])
	        amax_161 = torch.ops.aten.amax.default(mul_15588, [2])
	        full_322 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_161 = torch.ops.aten.minimum.default(amin_161, full_322);  amin_161 = full_322 = None
	        full_323 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_161 = torch.ops.aten.maximum.default(amax_161, full_323);  amax_161 = full_323 = None
	        sub_7369 = torch.ops.aten.sub.Tensor(maximum_161, minimum_161);  maximum_161 = None
	        div_322 = torch.ops.aten.div.Tensor(sub_7369, 255.0);  sub_7369 = None
	        clamp_min_483 = torch.ops.aten.clamp_min.default(div_322, 1.1920928955078125e-07);  div_322 = None
	        div_323 = torch.ops.aten.div.Tensor(minimum_161, clamp_min_483);  minimum_161 = None
	        round_323 = torch.ops.aten.round.default(div_323);  div_323 = None
	        sub_7375 = torch.ops.aten.sub.Tensor(-128, round_323);  round_323 = None
	        clamp_min_484 = torch.ops.aten.clamp_min.default(sub_7375, -128);  sub_7375 = None
	        clamp_max_322 = torch.ops.aten.clamp_max.default(clamp_min_484, 127);  clamp_min_484 = None
	        _assert_tensor_metadata_1451 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_483, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1451 = None
	        _assert_tensor_metadata_1452 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_322, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1452 = None
	        convert_element_type_966 = torch.ops.prims.convert_element_type.default(clamp_max_322, torch.int8);  clamp_max_322 = None
	        view_2525 = torch.ops.aten.view.default(clamp_min_483, [sym_size_int, 1500, 1])
	        view_2526 = torch.ops.aten.view.default(convert_element_type_966, [sym_size_int, 1500, 1])
	        reciprocal_161 = torch.ops.aten.reciprocal.default(view_2525);  view_2525 = None
	        mul_15634 = torch.ops.aten.mul.Tensor(reciprocal_161, 1.0);  reciprocal_161 = None
	        mul_15637 = torch.ops.aten.mul.Tensor(mul_15588, mul_15634);  mul_15588 = mul_15634 = None
	        round_324 = torch.ops.aten.round.default(mul_15637);  mul_15637 = None
	        add_24749 = torch.ops.aten.add.Tensor(round_324, view_2526);  round_324 = view_2526 = None
	        clamp_min_485 = torch.ops.aten.clamp_min.default(add_24749, -128);  add_24749 = None
	        clamp_max_323 = torch.ops.aten.clamp_max.default(clamp_min_485, 127);  clamp_min_485 = None
	        _assert_tensor_metadata_1453 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_323, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1453 = None
	        convert_element_type_967 = torch.ops.prims.convert_element_type.default(clamp_max_323, torch.int8);  clamp_max_323 = None
	        view_2529 = torch.ops.aten.view.default(clamp_min_483, [sym_size_int, 1500, 1]);  clamp_min_483 = None
	        view_2530 = torch.ops.aten.view.default(convert_element_type_966, [sym_size_int, 1500, 1]);  convert_element_type_966 = None
	        _assert_tensor_metadata_1454 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_967, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1454 = None
	        convert_element_type_968 = torch.ops.prims.convert_element_type.default(convert_element_type_967, torch.float32);  convert_element_type_967 = None
	        _assert_tensor_metadata_1455 = torch.ops.aten._assert_tensor_metadata.default(view_2530, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1455 = None
	        convert_element_type_969 = torch.ops.prims.convert_element_type.default(view_2530, torch.float32);  view_2530 = None
	        sub_7395 = torch.ops.aten.sub.Tensor(convert_element_type_968, convert_element_type_969);  convert_element_type_968 = convert_element_type_969 = None
	        mul_15659 = torch.ops.aten.mul.Tensor(sub_7395, view_2529);  sub_7395 = view_2529 = None
	        _assert_tensor_metadata_1456 = torch.ops.aten._assert_tensor_metadata.default(mul_15659, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1456 = None
	        view_2532 = torch.ops.aten.view.default(model_audio_tower_layers_26_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_26_fc2_parametrizations_weight_original0 = None
	        view_2533 = torch.ops.aten.view.default(model_audio_tower_layers_26_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_26_fc2_parametrizations_weight_original1 = None
	        view_2534 = torch.ops.aten.view.default(model_audio_tower_layers_26_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_26_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1457 = torch.ops.aten._assert_tensor_metadata.default(view_2532, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1457 = None
	        convert_element_type_970 = torch.ops.prims.convert_element_type.default(view_2532, torch.float32);  view_2532 = None
	        _assert_tensor_metadata_1458 = torch.ops.aten._assert_tensor_metadata.default(view_2534, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1458 = None
	        convert_element_type_971 = torch.ops.prims.convert_element_type.default(view_2534, torch.float32);  view_2534 = None
	        sub_7399 = torch.ops.aten.sub.Tensor(convert_element_type_970, convert_element_type_971);  convert_element_type_970 = convert_element_type_971 = None
	        mul_15664 = torch.ops.aten.mul.Tensor(sub_7399, view_2533);  sub_7399 = view_2533 = None
	        view_2535 = torch.ops.aten.view.default(mul_15664, [1280, 5120]);  mul_15664 = None
	        _assert_tensor_metadata_1459 = torch.ops.aten._assert_tensor_metadata.default(view_2535, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1459 = None
	        mul_15669 = sym_size_int * 1500
	        view_2536 = torch.ops.aten.view.default(mul_15659, [mul_15669, 5120]);  mul_15659 = mul_15669 = None
	        permute_270 = torch.ops.aten.permute.default(view_2535, [1, 0]);  view_2535 = None
	        addmm_134 = torch.ops.aten.addmm.default(model_audio_tower_layers_26_fc2_bias, view_2536, permute_270);  model_audio_tower_layers_26_fc2_bias = view_2536 = permute_270 = None
	        view_2537 = torch.ops.aten.view.default(addmm_134, [sym_size_int, 1500, 1280]);  addmm_134 = None
	        add_24812 = torch.ops.aten.add.Tensor(add_24514, view_2537);  add_24514 = view_2537 = None
	        clone_217 = torch.ops.aten.clone.default(add_24812, memory_format = torch.contiguous_format)
	        var_mean_54 = torch.ops.aten.var_mean.correction(clone_217, [2], correction = 0, keepdim = True)
	        getitem_216 = var_mean_54[0]
	        getitem_217 = var_mean_54[1];  var_mean_54 = None
	        add_24817 = torch.ops.aten.add.Tensor(getitem_216, 1e-05);  getitem_216 = None
	        rsqrt_54 = torch.ops.aten.rsqrt.default(add_24817);  add_24817 = None
	        sub_7405 = torch.ops.aten.sub.Tensor(clone_217, getitem_217);  clone_217 = getitem_217 = None
	        mul_15680 = torch.ops.aten.mul.Tensor(sub_7405, rsqrt_54);  sub_7405 = rsqrt_54 = None
	        mul_15681 = torch.ops.aten.mul.Tensor(mul_15680, model_audio_tower_layers_27_self_attn_layer_norm_weight);  mul_15680 = model_audio_tower_layers_27_self_attn_layer_norm_weight = None
	        add_24818 = torch.ops.aten.add.Tensor(mul_15681, model_audio_tower_layers_27_self_attn_layer_norm_bias);  mul_15681 = model_audio_tower_layers_27_self_attn_layer_norm_bias = None
	        amin_162 = torch.ops.aten.amin.default(add_24818, [2])
	        amax_162 = torch.ops.aten.amax.default(add_24818, [2])
	        full_324 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_162 = torch.ops.aten.minimum.default(amin_162, full_324);  amin_162 = full_324 = None
	        full_325 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_162 = torch.ops.aten.maximum.default(amax_162, full_325);  amax_162 = full_325 = None
	        sub_7416 = torch.ops.aten.sub.Tensor(maximum_162, minimum_162);  maximum_162 = None
	        div_324 = torch.ops.aten.div.Tensor(sub_7416, 255.0);  sub_7416 = None
	        clamp_min_486 = torch.ops.aten.clamp_min.default(div_324, 1.1920928955078125e-07);  div_324 = None
	        div_325 = torch.ops.aten.div.Tensor(minimum_162, clamp_min_486);  minimum_162 = None
	        round_325 = torch.ops.aten.round.default(div_325);  div_325 = None
	        sub_7422 = torch.ops.aten.sub.Tensor(-128, round_325);  round_325 = None
	        clamp_min_487 = torch.ops.aten.clamp_min.default(sub_7422, -128);  sub_7422 = None
	        clamp_max_324 = torch.ops.aten.clamp_max.default(clamp_min_487, 127);  clamp_min_487 = None
	        _assert_tensor_metadata_1460 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_486, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1460 = None
	        _assert_tensor_metadata_1461 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_324, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1461 = None
	        convert_element_type_972 = torch.ops.prims.convert_element_type.default(clamp_max_324, torch.int8);  clamp_max_324 = None
	        view_2540 = torch.ops.aten.view.default(clamp_min_486, [sym_size_int, 1500, 1])
	        view_2541 = torch.ops.aten.view.default(convert_element_type_972, [sym_size_int, 1500, 1])
	        reciprocal_162 = torch.ops.aten.reciprocal.default(view_2540);  view_2540 = None
	        mul_15729 = torch.ops.aten.mul.Tensor(reciprocal_162, 1.0);  reciprocal_162 = None
	        mul_15732 = torch.ops.aten.mul.Tensor(add_24818, mul_15729);  mul_15729 = None
	        round_326 = torch.ops.aten.round.default(mul_15732);  mul_15732 = None
	        add_24905 = torch.ops.aten.add.Tensor(round_326, view_2541);  round_326 = view_2541 = None
	        clamp_min_488 = torch.ops.aten.clamp_min.default(add_24905, -128);  add_24905 = None
	        clamp_max_325 = torch.ops.aten.clamp_max.default(clamp_min_488, 127);  clamp_min_488 = None
	        _assert_tensor_metadata_1462 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_325, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1462 = None
	        convert_element_type_973 = torch.ops.prims.convert_element_type.default(clamp_max_325, torch.int8);  clamp_max_325 = None
	        view_2544 = torch.ops.aten.view.default(clamp_min_486, [sym_size_int, 1500, 1]);  clamp_min_486 = None
	        view_2545 = torch.ops.aten.view.default(convert_element_type_972, [sym_size_int, 1500, 1]);  convert_element_type_972 = None
	        _assert_tensor_metadata_1463 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_973, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1463 = None
	        convert_element_type_974 = torch.ops.prims.convert_element_type.default(convert_element_type_973, torch.float32);  convert_element_type_973 = None
	        _assert_tensor_metadata_1464 = torch.ops.aten._assert_tensor_metadata.default(view_2545, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1464 = None
	        convert_element_type_975 = torch.ops.prims.convert_element_type.default(view_2545, torch.float32);  view_2545 = None
	        sub_7442 = torch.ops.aten.sub.Tensor(convert_element_type_974, convert_element_type_975);  convert_element_type_974 = convert_element_type_975 = None
	        mul_15754 = torch.ops.aten.mul.Tensor(sub_7442, view_2544);  sub_7442 = view_2544 = None
	        _assert_tensor_metadata_1465 = torch.ops.aten._assert_tensor_metadata.default(mul_15754, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1465 = None
	        view_2547 = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2548 = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2549 = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1466 = torch.ops.aten._assert_tensor_metadata.default(view_2547, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1466 = None
	        convert_element_type_976 = torch.ops.prims.convert_element_type.default(view_2547, torch.float32);  view_2547 = None
	        _assert_tensor_metadata_1467 = torch.ops.aten._assert_tensor_metadata.default(view_2549, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1467 = None
	        convert_element_type_977 = torch.ops.prims.convert_element_type.default(view_2549, torch.float32);  view_2549 = None
	        sub_7446 = torch.ops.aten.sub.Tensor(convert_element_type_976, convert_element_type_977);  convert_element_type_976 = convert_element_type_977 = None
	        mul_15759 = torch.ops.aten.mul.Tensor(sub_7446, view_2548);  sub_7446 = view_2548 = None
	        view_2550 = torch.ops.aten.view.default(mul_15759, [1280, 1280]);  mul_15759 = None
	        _assert_tensor_metadata_1468 = torch.ops.aten._assert_tensor_metadata.default(view_2550, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1468 = None
	        mul_15764 = sym_size_int * 1500
	        view_2551 = torch.ops.aten.view.default(mul_15754, [mul_15764, 1280]);  mul_15754 = mul_15764 = None
	        permute_271 = torch.ops.aten.permute.default(view_2550, [1, 0]);  view_2550 = None
	        addmm_135 = torch.ops.aten.addmm.default(model_audio_tower_layers_27_self_attn_q_proj_bias, view_2551, permute_271);  model_audio_tower_layers_27_self_attn_q_proj_bias = view_2551 = permute_271 = None
	        view_2552 = torch.ops.aten.view.default(addmm_135, [sym_size_int, 1500, 1280]);  addmm_135 = None
	        mul_15771 = torch.ops.aten.mul.Tensor(view_2552, 0.125);  view_2552 = None
	        view_2553 = torch.ops.aten.view.default(mul_15771, [sym_size_int, 1500, 20, 64]);  mul_15771 = None
	        permute_272 = torch.ops.aten.permute.default(view_2553, [0, 2, 1, 3]);  view_2553 = None
	        clone_218 = torch.ops.aten.clone.default(permute_272, memory_format = torch.contiguous_format);  permute_272 = None
	        amin_163 = torch.ops.aten.amin.default(add_24818, [2])
	        amax_163 = torch.ops.aten.amax.default(add_24818, [2])
	        full_326 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_163 = torch.ops.aten.minimum.default(amin_163, full_326);  amin_163 = full_326 = None
	        full_327 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_163 = torch.ops.aten.maximum.default(amax_163, full_327);  amax_163 = full_327 = None
	        sub_7461 = torch.ops.aten.sub.Tensor(maximum_163, minimum_163);  maximum_163 = None
	        div_326 = torch.ops.aten.div.Tensor(sub_7461, 255.0);  sub_7461 = None
	        clamp_min_489 = torch.ops.aten.clamp_min.default(div_326, 1.1920928955078125e-07);  div_326 = None
	        div_327 = torch.ops.aten.div.Tensor(minimum_163, clamp_min_489);  minimum_163 = None
	        round_327 = torch.ops.aten.round.default(div_327);  div_327 = None
	        sub_7467 = torch.ops.aten.sub.Tensor(-128, round_327);  round_327 = None
	        clamp_min_490 = torch.ops.aten.clamp_min.default(sub_7467, -128);  sub_7467 = None
	        clamp_max_326 = torch.ops.aten.clamp_max.default(clamp_min_490, 127);  clamp_min_490 = None
	        _assert_tensor_metadata_1469 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_489, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1469 = None
	        _assert_tensor_metadata_1470 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_326, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1470 = None
	        convert_element_type_978 = torch.ops.prims.convert_element_type.default(clamp_max_326, torch.int8);  clamp_max_326 = None
	        view_2556 = torch.ops.aten.view.default(clamp_min_489, [sym_size_int, 1500, 1])
	        view_2557 = torch.ops.aten.view.default(convert_element_type_978, [sym_size_int, 1500, 1])
	        reciprocal_163 = torch.ops.aten.reciprocal.default(view_2556);  view_2556 = None
	        mul_15825 = torch.ops.aten.mul.Tensor(reciprocal_163, 1.0);  reciprocal_163 = None
	        mul_15828 = torch.ops.aten.mul.Tensor(add_24818, mul_15825);  mul_15825 = None
	        round_328 = torch.ops.aten.round.default(mul_15828);  mul_15828 = None
	        add_25057 = torch.ops.aten.add.Tensor(round_328, view_2557);  round_328 = view_2557 = None
	        clamp_min_491 = torch.ops.aten.clamp_min.default(add_25057, -128);  add_25057 = None
	        clamp_max_327 = torch.ops.aten.clamp_max.default(clamp_min_491, 127);  clamp_min_491 = None
	        _assert_tensor_metadata_1471 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_327, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1471 = None
	        convert_element_type_979 = torch.ops.prims.convert_element_type.default(clamp_max_327, torch.int8);  clamp_max_327 = None
	        view_2560 = torch.ops.aten.view.default(clamp_min_489, [sym_size_int, 1500, 1]);  clamp_min_489 = None
	        view_2561 = torch.ops.aten.view.default(convert_element_type_978, [sym_size_int, 1500, 1]);  convert_element_type_978 = None
	        _assert_tensor_metadata_1472 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_979, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1472 = None
	        convert_element_type_980 = torch.ops.prims.convert_element_type.default(convert_element_type_979, torch.float32);  convert_element_type_979 = None
	        _assert_tensor_metadata_1473 = torch.ops.aten._assert_tensor_metadata.default(view_2561, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1473 = None
	        convert_element_type_981 = torch.ops.prims.convert_element_type.default(view_2561, torch.float32);  view_2561 = None
	        sub_7487 = torch.ops.aten.sub.Tensor(convert_element_type_980, convert_element_type_981);  convert_element_type_980 = convert_element_type_981 = None
	        mul_15850 = torch.ops.aten.mul.Tensor(sub_7487, view_2560);  sub_7487 = view_2560 = None
	        _assert_tensor_metadata_1474 = torch.ops.aten._assert_tensor_metadata.default(mul_15850, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1474 = None
	        view_2563 = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2564 = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2565 = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1475 = torch.ops.aten._assert_tensor_metadata.default(view_2563, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1475 = None
	        convert_element_type_982 = torch.ops.prims.convert_element_type.default(view_2563, torch.float32);  view_2563 = None
	        _assert_tensor_metadata_1476 = torch.ops.aten._assert_tensor_metadata.default(view_2565, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1476 = None
	        convert_element_type_983 = torch.ops.prims.convert_element_type.default(view_2565, torch.float32);  view_2565 = None
	        sub_7491 = torch.ops.aten.sub.Tensor(convert_element_type_982, convert_element_type_983);  convert_element_type_982 = convert_element_type_983 = None
	        mul_15855 = torch.ops.aten.mul.Tensor(sub_7491, view_2564);  sub_7491 = view_2564 = None
	        view_2566 = torch.ops.aten.view.default(mul_15855, [1280, 1280]);  mul_15855 = None
	        _assert_tensor_metadata_1477 = torch.ops.aten._assert_tensor_metadata.default(view_2566, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1477 = None
	        permute_273 = torch.ops.aten.permute.default(view_2566, [1, 0]);  view_2566 = None
	        mul_15858 = sym_size_int * 1500
	        view_2567 = torch.ops.aten.view.default(mul_15850, [mul_15858, 1280]);  mul_15850 = mul_15858 = None
	        mm_27 = torch.ops.aten.mm.default(view_2567, permute_273);  view_2567 = permute_273 = None
	        view_2568 = torch.ops.aten.view.default(mm_27, [sym_size_int, 1500, 1280]);  mm_27 = None
	        view_2569 = torch.ops.aten.view.default(view_2568, [sym_size_int, -1, 20, 64]);  view_2568 = None
	        permute_274 = torch.ops.aten.permute.default(view_2569, [0, 2, 1, 3]);  view_2569 = None
	        clone_219 = torch.ops.aten.clone.default(permute_274, memory_format = torch.contiguous_format);  permute_274 = None
	        amin_164 = torch.ops.aten.amin.default(add_24818, [2])
	        amax_164 = torch.ops.aten.amax.default(add_24818, [2])
	        full_328 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_164 = torch.ops.aten.minimum.default(amin_164, full_328);  amin_164 = full_328 = None
	        full_329 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_164 = torch.ops.aten.maximum.default(amax_164, full_329);  amax_164 = full_329 = None
	        sub_7505 = torch.ops.aten.sub.Tensor(maximum_164, minimum_164);  maximum_164 = None
	        div_328 = torch.ops.aten.div.Tensor(sub_7505, 255.0);  sub_7505 = None
	        clamp_min_492 = torch.ops.aten.clamp_min.default(div_328, 1.1920928955078125e-07);  div_328 = None
	        div_329 = torch.ops.aten.div.Tensor(minimum_164, clamp_min_492);  minimum_164 = None
	        round_329 = torch.ops.aten.round.default(div_329);  div_329 = None
	        sub_7511 = torch.ops.aten.sub.Tensor(-128, round_329);  round_329 = None
	        clamp_min_493 = torch.ops.aten.clamp_min.default(sub_7511, -128);  sub_7511 = None
	        clamp_max_328 = torch.ops.aten.clamp_max.default(clamp_min_493, 127);  clamp_min_493 = None
	        _assert_tensor_metadata_1478 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_492, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1478 = None
	        _assert_tensor_metadata_1479 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_328, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1479 = None
	        convert_element_type_984 = torch.ops.prims.convert_element_type.default(clamp_max_328, torch.int8);  clamp_max_328 = None
	        view_2572 = torch.ops.aten.view.default(clamp_min_492, [sym_size_int, 1500, 1])
	        view_2573 = torch.ops.aten.view.default(convert_element_type_984, [sym_size_int, 1500, 1])
	        reciprocal_164 = torch.ops.aten.reciprocal.default(view_2572);  view_2572 = None
	        mul_15924 = torch.ops.aten.mul.Tensor(reciprocal_164, 1.0);  reciprocal_164 = None
	        mul_15927 = torch.ops.aten.mul.Tensor(add_24818, mul_15924);  add_24818 = mul_15924 = None
	        round_330 = torch.ops.aten.round.default(mul_15927);  mul_15927 = None
	        add_25205 = torch.ops.aten.add.Tensor(round_330, view_2573);  round_330 = view_2573 = None
	        clamp_min_494 = torch.ops.aten.clamp_min.default(add_25205, -128);  add_25205 = None
	        clamp_max_329 = torch.ops.aten.clamp_max.default(clamp_min_494, 127);  clamp_min_494 = None
	        _assert_tensor_metadata_1480 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_329, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1480 = None
	        convert_element_type_985 = torch.ops.prims.convert_element_type.default(clamp_max_329, torch.int8);  clamp_max_329 = None
	        view_2576 = torch.ops.aten.view.default(clamp_min_492, [sym_size_int, 1500, 1]);  clamp_min_492 = None
	        view_2577 = torch.ops.aten.view.default(convert_element_type_984, [sym_size_int, 1500, 1]);  convert_element_type_984 = None
	        _assert_tensor_metadata_1481 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_985, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1481 = None
	        convert_element_type_986 = torch.ops.prims.convert_element_type.default(convert_element_type_985, torch.float32);  convert_element_type_985 = None
	        _assert_tensor_metadata_1482 = torch.ops.aten._assert_tensor_metadata.default(view_2577, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1482 = None
	        convert_element_type_987 = torch.ops.prims.convert_element_type.default(view_2577, torch.float32);  view_2577 = None
	        sub_7531 = torch.ops.aten.sub.Tensor(convert_element_type_986, convert_element_type_987);  convert_element_type_986 = convert_element_type_987 = None
	        mul_15949 = torch.ops.aten.mul.Tensor(sub_7531, view_2576);  sub_7531 = view_2576 = None
	        _assert_tensor_metadata_1483 = torch.ops.aten._assert_tensor_metadata.default(mul_15949, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1483 = None
	        view_2579 = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2580 = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2581 = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1484 = torch.ops.aten._assert_tensor_metadata.default(view_2579, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1484 = None
	        convert_element_type_988 = torch.ops.prims.convert_element_type.default(view_2579, torch.float32);  view_2579 = None
	        _assert_tensor_metadata_1485 = torch.ops.aten._assert_tensor_metadata.default(view_2581, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1485 = None
	        convert_element_type_989 = torch.ops.prims.convert_element_type.default(view_2581, torch.float32);  view_2581 = None
	        sub_7535 = torch.ops.aten.sub.Tensor(convert_element_type_988, convert_element_type_989);  convert_element_type_988 = convert_element_type_989 = None
	        mul_15954 = torch.ops.aten.mul.Tensor(sub_7535, view_2580);  sub_7535 = view_2580 = None
	        view_2582 = torch.ops.aten.view.default(mul_15954, [1280, 1280]);  mul_15954 = None
	        _assert_tensor_metadata_1486 = torch.ops.aten._assert_tensor_metadata.default(view_2582, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1486 = None
	        mul_15959 = sym_size_int * 1500
	        view_2583 = torch.ops.aten.view.default(mul_15949, [mul_15959, 1280]);  mul_15949 = mul_15959 = None
	        permute_275 = torch.ops.aten.permute.default(view_2582, [1, 0]);  view_2582 = None
	        addmm_136 = torch.ops.aten.addmm.default(model_audio_tower_layers_27_self_attn_v_proj_bias, view_2583, permute_275);  model_audio_tower_layers_27_self_attn_v_proj_bias = view_2583 = permute_275 = None
	        view_2584 = torch.ops.aten.view.default(addmm_136, [sym_size_int, 1500, 1280]);  addmm_136 = None
	        view_2585 = torch.ops.aten.view.default(view_2584, [sym_size_int, -1, 20, 64]);  view_2584 = None
	        permute_276 = torch.ops.aten.permute.default(view_2585, [0, 2, 1, 3]);  view_2585 = None
	        clone_220 = torch.ops.aten.clone.default(permute_276, memory_format = torch.contiguous_format);  permute_276 = None
	        _scaled_dot_product_efficient_attention_27 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_218, clone_219, clone_220, None, False, scale = 1.0);  clone_218 = clone_219 = clone_220 = None
	        getitem_218 = _scaled_dot_product_efficient_attention_27[0];  _scaled_dot_product_efficient_attention_27 = None
	        permute_277 = torch.ops.aten.permute.default(getitem_218, [0, 2, 1, 3]);  getitem_218 = None
	        view_2586 = torch.ops.aten.view.default(permute_277, [sym_size_int, 1500, -1]);  permute_277 = None
	        amin_165 = torch.ops.aten.amin.default(view_2586, [2])
	        amax_165 = torch.ops.aten.amax.default(view_2586, [2])
	        full_330 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_165 = torch.ops.aten.minimum.default(amin_165, full_330);  amin_165 = full_330 = None
	        full_331 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_165 = torch.ops.aten.maximum.default(amax_165, full_331);  amax_165 = full_331 = None
	        sub_7553 = torch.ops.aten.sub.Tensor(maximum_165, minimum_165);  maximum_165 = None
	        div_330 = torch.ops.aten.div.Tensor(sub_7553, 255.0);  sub_7553 = None
	        clamp_min_495 = torch.ops.aten.clamp_min.default(div_330, 1.1920928955078125e-07);  div_330 = None
	        div_331 = torch.ops.aten.div.Tensor(minimum_165, clamp_min_495);  minimum_165 = None
	        round_331 = torch.ops.aten.round.default(div_331);  div_331 = None
	        sub_7559 = torch.ops.aten.sub.Tensor(-128, round_331);  round_331 = None
	        clamp_min_496 = torch.ops.aten.clamp_min.default(sub_7559, -128);  sub_7559 = None
	        clamp_max_330 = torch.ops.aten.clamp_max.default(clamp_min_496, 127);  clamp_min_496 = None
	        _assert_tensor_metadata_1487 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_495, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1487 = None
	        _assert_tensor_metadata_1488 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_330, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1488 = None
	        convert_element_type_990 = torch.ops.prims.convert_element_type.default(clamp_max_330, torch.int8);  clamp_max_330 = None
	        view_2589 = torch.ops.aten.view.default(clamp_min_495, [sym_size_int, 1500, 1])
	        view_2590 = torch.ops.aten.view.default(convert_element_type_990, [sym_size_int, 1500, 1])
	        reciprocal_165 = torch.ops.aten.reciprocal.default(view_2589);  view_2589 = None
	        mul_16029 = torch.ops.aten.mul.Tensor(reciprocal_165, 1.0);  reciprocal_165 = None
	        mul_16032 = torch.ops.aten.mul.Tensor(view_2586, mul_16029);  view_2586 = mul_16029 = None
	        round_332 = torch.ops.aten.round.default(mul_16032);  mul_16032 = None
	        add_25369 = torch.ops.aten.add.Tensor(round_332, view_2590);  round_332 = view_2590 = None
	        clamp_min_497 = torch.ops.aten.clamp_min.default(add_25369, -128);  add_25369 = None
	        clamp_max_331 = torch.ops.aten.clamp_max.default(clamp_min_497, 127);  clamp_min_497 = None
	        _assert_tensor_metadata_1489 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_331, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1489 = None
	        convert_element_type_991 = torch.ops.prims.convert_element_type.default(clamp_max_331, torch.int8);  clamp_max_331 = None
	        view_2593 = torch.ops.aten.view.default(clamp_min_495, [sym_size_int, 1500, 1]);  clamp_min_495 = None
	        view_2594 = torch.ops.aten.view.default(convert_element_type_990, [sym_size_int, 1500, 1]);  convert_element_type_990 = None
	        _assert_tensor_metadata_1490 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_991, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1490 = None
	        convert_element_type_992 = torch.ops.prims.convert_element_type.default(convert_element_type_991, torch.float32);  convert_element_type_991 = None
	        _assert_tensor_metadata_1491 = torch.ops.aten._assert_tensor_metadata.default(view_2594, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1491 = None
	        convert_element_type_993 = torch.ops.prims.convert_element_type.default(view_2594, torch.float32);  view_2594 = None
	        sub_7579 = torch.ops.aten.sub.Tensor(convert_element_type_992, convert_element_type_993);  convert_element_type_992 = convert_element_type_993 = None
	        mul_16054 = torch.ops.aten.mul.Tensor(sub_7579, view_2593);  sub_7579 = view_2593 = None
	        _assert_tensor_metadata_1492 = torch.ops.aten._assert_tensor_metadata.default(mul_16054, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1492 = None
	        view_2596 = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2597 = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2598 = torch.ops.aten.view.default(model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1493 = torch.ops.aten._assert_tensor_metadata.default(view_2596, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1493 = None
	        convert_element_type_994 = torch.ops.prims.convert_element_type.default(view_2596, torch.float32);  view_2596 = None
	        _assert_tensor_metadata_1494 = torch.ops.aten._assert_tensor_metadata.default(view_2598, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1494 = None
	        convert_element_type_995 = torch.ops.prims.convert_element_type.default(view_2598, torch.float32);  view_2598 = None
	        sub_7583 = torch.ops.aten.sub.Tensor(convert_element_type_994, convert_element_type_995);  convert_element_type_994 = convert_element_type_995 = None
	        mul_16059 = torch.ops.aten.mul.Tensor(sub_7583, view_2597);  sub_7583 = view_2597 = None
	        view_2599 = torch.ops.aten.view.default(mul_16059, [1280, 1280]);  mul_16059 = None
	        _assert_tensor_metadata_1495 = torch.ops.aten._assert_tensor_metadata.default(view_2599, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1495 = None
	        mul_16064 = sym_size_int * 1500
	        view_2600 = torch.ops.aten.view.default(mul_16054, [mul_16064, 1280]);  mul_16054 = mul_16064 = None
	        permute_278 = torch.ops.aten.permute.default(view_2599, [1, 0]);  view_2599 = None
	        addmm_137 = torch.ops.aten.addmm.default(model_audio_tower_layers_27_self_attn_out_proj_bias, view_2600, permute_278);  model_audio_tower_layers_27_self_attn_out_proj_bias = view_2600 = permute_278 = None
	        view_2601 = torch.ops.aten.view.default(addmm_137, [sym_size_int, 1500, 1280]);  addmm_137 = None
	        add_25432 = torch.ops.aten.add.Tensor(add_24812, view_2601);  add_24812 = view_2601 = None
	        clone_222 = torch.ops.aten.clone.default(add_25432, memory_format = torch.contiguous_format)
	        var_mean_55 = torch.ops.aten.var_mean.correction(clone_222, [2], correction = 0, keepdim = True)
	        getitem_222 = var_mean_55[0]
	        getitem_223 = var_mean_55[1];  var_mean_55 = None
	        add_25437 = torch.ops.aten.add.Tensor(getitem_222, 1e-05);  getitem_222 = None
	        rsqrt_55 = torch.ops.aten.rsqrt.default(add_25437);  add_25437 = None
	        sub_7589 = torch.ops.aten.sub.Tensor(clone_222, getitem_223);  clone_222 = getitem_223 = None
	        mul_16075 = torch.ops.aten.mul.Tensor(sub_7589, rsqrt_55);  sub_7589 = rsqrt_55 = None
	        mul_16076 = torch.ops.aten.mul.Tensor(mul_16075, model_audio_tower_layers_27_final_layer_norm_weight);  mul_16075 = model_audio_tower_layers_27_final_layer_norm_weight = None
	        add_25438 = torch.ops.aten.add.Tensor(mul_16076, model_audio_tower_layers_27_final_layer_norm_bias);  mul_16076 = model_audio_tower_layers_27_final_layer_norm_bias = None
	        amin_166 = torch.ops.aten.amin.default(add_25438, [2])
	        amax_166 = torch.ops.aten.amax.default(add_25438, [2])
	        full_332 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_166 = torch.ops.aten.minimum.default(amin_166, full_332);  amin_166 = full_332 = None
	        full_333 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_166 = torch.ops.aten.maximum.default(amax_166, full_333);  amax_166 = full_333 = None
	        sub_7600 = torch.ops.aten.sub.Tensor(maximum_166, minimum_166);  maximum_166 = None
	        div_332 = torch.ops.aten.div.Tensor(sub_7600, 255.0);  sub_7600 = None
	        clamp_min_498 = torch.ops.aten.clamp_min.default(div_332, 1.1920928955078125e-07);  div_332 = None
	        div_333 = torch.ops.aten.div.Tensor(minimum_166, clamp_min_498);  minimum_166 = None
	        round_333 = torch.ops.aten.round.default(div_333);  div_333 = None
	        sub_7606 = torch.ops.aten.sub.Tensor(-128, round_333);  round_333 = None
	        clamp_min_499 = torch.ops.aten.clamp_min.default(sub_7606, -128);  sub_7606 = None
	        clamp_max_332 = torch.ops.aten.clamp_max.default(clamp_min_499, 127);  clamp_min_499 = None
	        _assert_tensor_metadata_1496 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_498, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1496 = None
	        _assert_tensor_metadata_1497 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_332, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1497 = None
	        convert_element_type_996 = torch.ops.prims.convert_element_type.default(clamp_max_332, torch.int8);  clamp_max_332 = None
	        view_2604 = torch.ops.aten.view.default(clamp_min_498, [sym_size_int, 1500, 1])
	        view_2605 = torch.ops.aten.view.default(convert_element_type_996, [sym_size_int, 1500, 1])
	        reciprocal_166 = torch.ops.aten.reciprocal.default(view_2604);  view_2604 = None
	        mul_16124 = torch.ops.aten.mul.Tensor(reciprocal_166, 1.0);  reciprocal_166 = None
	        mul_16127 = torch.ops.aten.mul.Tensor(add_25438, mul_16124);  add_25438 = mul_16124 = None
	        round_334 = torch.ops.aten.round.default(mul_16127);  mul_16127 = None
	        add_25525 = torch.ops.aten.add.Tensor(round_334, view_2605);  round_334 = view_2605 = None
	        clamp_min_500 = torch.ops.aten.clamp_min.default(add_25525, -128);  add_25525 = None
	        clamp_max_333 = torch.ops.aten.clamp_max.default(clamp_min_500, 127);  clamp_min_500 = None
	        _assert_tensor_metadata_1498 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_333, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1498 = None
	        convert_element_type_997 = torch.ops.prims.convert_element_type.default(clamp_max_333, torch.int8);  clamp_max_333 = None
	        view_2608 = torch.ops.aten.view.default(clamp_min_498, [sym_size_int, 1500, 1]);  clamp_min_498 = None
	        view_2609 = torch.ops.aten.view.default(convert_element_type_996, [sym_size_int, 1500, 1]);  convert_element_type_996 = None
	        _assert_tensor_metadata_1499 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_997, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1499 = None
	        convert_element_type_998 = torch.ops.prims.convert_element_type.default(convert_element_type_997, torch.float32);  convert_element_type_997 = None
	        _assert_tensor_metadata_1500 = torch.ops.aten._assert_tensor_metadata.default(view_2609, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1500 = None
	        convert_element_type_999 = torch.ops.prims.convert_element_type.default(view_2609, torch.float32);  view_2609 = None
	        sub_7626 = torch.ops.aten.sub.Tensor(convert_element_type_998, convert_element_type_999);  convert_element_type_998 = convert_element_type_999 = None
	        mul_16149 = torch.ops.aten.mul.Tensor(sub_7626, view_2608);  sub_7626 = view_2608 = None
	        _assert_tensor_metadata_1501 = torch.ops.aten._assert_tensor_metadata.default(mul_16149, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1501 = None
	        view_2611 = torch.ops.aten.view.default(model_audio_tower_layers_27_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_27_fc1_parametrizations_weight_original0 = None
	        view_2612 = torch.ops.aten.view.default(model_audio_tower_layers_27_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_27_fc1_parametrizations_weight_original1 = None
	        view_2613 = torch.ops.aten.view.default(model_audio_tower_layers_27_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_27_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1502 = torch.ops.aten._assert_tensor_metadata.default(view_2611, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1502 = None
	        convert_element_type_1000 = torch.ops.prims.convert_element_type.default(view_2611, torch.float32);  view_2611 = None
	        _assert_tensor_metadata_1503 = torch.ops.aten._assert_tensor_metadata.default(view_2613, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1503 = None
	        convert_element_type_1001 = torch.ops.prims.convert_element_type.default(view_2613, torch.float32);  view_2613 = None
	        sub_7630 = torch.ops.aten.sub.Tensor(convert_element_type_1000, convert_element_type_1001);  convert_element_type_1000 = convert_element_type_1001 = None
	        mul_16154 = torch.ops.aten.mul.Tensor(sub_7630, view_2612);  sub_7630 = view_2612 = None
	        view_2614 = torch.ops.aten.view.default(mul_16154, [5120, 1280]);  mul_16154 = None
	        _assert_tensor_metadata_1504 = torch.ops.aten._assert_tensor_metadata.default(view_2614, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1504 = None
	        mul_16159 = sym_size_int * 1500
	        view_2615 = torch.ops.aten.view.default(mul_16149, [mul_16159, 1280]);  mul_16149 = mul_16159 = None
	        permute_279 = torch.ops.aten.permute.default(view_2614, [1, 0]);  view_2614 = None
	        addmm_138 = torch.ops.aten.addmm.default(model_audio_tower_layers_27_fc1_bias, view_2615, permute_279);  model_audio_tower_layers_27_fc1_bias = view_2615 = permute_279 = None
	        view_2616 = torch.ops.aten.view.default(addmm_138, [sym_size_int, 1500, 5120]);  addmm_138 = None
	        mul_16166 = torch.ops.aten.mul.Tensor(view_2616, 0.5)
	        mul_16167 = torch.ops.aten.mul.Tensor(view_2616, 0.7071067811865476);  view_2616 = None
	        erf_29 = torch.ops.aten.erf.default(mul_16167);  mul_16167 = None
	        add_25584 = torch.ops.aten.add.Tensor(erf_29, 1);  erf_29 = None
	        mul_16168 = torch.ops.aten.mul.Tensor(mul_16166, add_25584);  mul_16166 = add_25584 = None
	        amin_167 = torch.ops.aten.amin.default(mul_16168, [2])
	        amax_167 = torch.ops.aten.amax.default(mul_16168, [2])
	        full_334 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_167 = torch.ops.aten.minimum.default(amin_167, full_334);  amin_167 = full_334 = None
	        full_335 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_167 = torch.ops.aten.maximum.default(amax_167, full_335);  amax_167 = full_335 = None
	        sub_7643 = torch.ops.aten.sub.Tensor(maximum_167, minimum_167);  maximum_167 = None
	        div_334 = torch.ops.aten.div.Tensor(sub_7643, 255.0);  sub_7643 = None
	        clamp_min_501 = torch.ops.aten.clamp_min.default(div_334, 1.1920928955078125e-07);  div_334 = None
	        div_335 = torch.ops.aten.div.Tensor(minimum_167, clamp_min_501);  minimum_167 = None
	        round_335 = torch.ops.aten.round.default(div_335);  div_335 = None
	        sub_7649 = torch.ops.aten.sub.Tensor(-128, round_335);  round_335 = None
	        clamp_min_502 = torch.ops.aten.clamp_min.default(sub_7649, -128);  sub_7649 = None
	        clamp_max_334 = torch.ops.aten.clamp_max.default(clamp_min_502, 127);  clamp_min_502 = None
	        _assert_tensor_metadata_1505 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_501, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1505 = None
	        _assert_tensor_metadata_1506 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_334, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1506 = None
	        convert_element_type_1002 = torch.ops.prims.convert_element_type.default(clamp_max_334, torch.int8);  clamp_max_334 = None
	        view_2619 = torch.ops.aten.view.default(clamp_min_501, [sym_size_int, 1500, 1])
	        view_2620 = torch.ops.aten.view.default(convert_element_type_1002, [sym_size_int, 1500, 1])
	        reciprocal_167 = torch.ops.aten.reciprocal.default(view_2619);  view_2619 = None
	        mul_16214 = torch.ops.aten.mul.Tensor(reciprocal_167, 1.0);  reciprocal_167 = None
	        mul_16217 = torch.ops.aten.mul.Tensor(mul_16168, mul_16214);  mul_16168 = mul_16214 = None
	        round_336 = torch.ops.aten.round.default(mul_16217);  mul_16217 = None
	        add_25667 = torch.ops.aten.add.Tensor(round_336, view_2620);  round_336 = view_2620 = None
	        clamp_min_503 = torch.ops.aten.clamp_min.default(add_25667, -128);  add_25667 = None
	        clamp_max_335 = torch.ops.aten.clamp_max.default(clamp_min_503, 127);  clamp_min_503 = None
	        _assert_tensor_metadata_1507 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_335, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1507 = None
	        convert_element_type_1003 = torch.ops.prims.convert_element_type.default(clamp_max_335, torch.int8);  clamp_max_335 = None
	        view_2623 = torch.ops.aten.view.default(clamp_min_501, [sym_size_int, 1500, 1]);  clamp_min_501 = None
	        view_2624 = torch.ops.aten.view.default(convert_element_type_1002, [sym_size_int, 1500, 1]);  convert_element_type_1002 = None
	        _assert_tensor_metadata_1508 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1003, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1508 = None
	        convert_element_type_1004 = torch.ops.prims.convert_element_type.default(convert_element_type_1003, torch.float32);  convert_element_type_1003 = None
	        _assert_tensor_metadata_1509 = torch.ops.aten._assert_tensor_metadata.default(view_2624, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1509 = None
	        convert_element_type_1005 = torch.ops.prims.convert_element_type.default(view_2624, torch.float32);  view_2624 = None
	        sub_7669 = torch.ops.aten.sub.Tensor(convert_element_type_1004, convert_element_type_1005);  convert_element_type_1004 = convert_element_type_1005 = None
	        mul_16239 = torch.ops.aten.mul.Tensor(sub_7669, view_2623);  sub_7669 = view_2623 = None
	        _assert_tensor_metadata_1510 = torch.ops.aten._assert_tensor_metadata.default(mul_16239, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1510 = None
	        view_2626 = torch.ops.aten.view.default(model_audio_tower_layers_27_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_27_fc2_parametrizations_weight_original0 = None
	        view_2627 = torch.ops.aten.view.default(model_audio_tower_layers_27_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_27_fc2_parametrizations_weight_original1 = None
	        view_2628 = torch.ops.aten.view.default(model_audio_tower_layers_27_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_27_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1511 = torch.ops.aten._assert_tensor_metadata.default(view_2626, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1511 = None
	        convert_element_type_1006 = torch.ops.prims.convert_element_type.default(view_2626, torch.float32);  view_2626 = None
	        _assert_tensor_metadata_1512 = torch.ops.aten._assert_tensor_metadata.default(view_2628, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1512 = None
	        convert_element_type_1007 = torch.ops.prims.convert_element_type.default(view_2628, torch.float32);  view_2628 = None
	        sub_7673 = torch.ops.aten.sub.Tensor(convert_element_type_1006, convert_element_type_1007);  convert_element_type_1006 = convert_element_type_1007 = None
	        mul_16244 = torch.ops.aten.mul.Tensor(sub_7673, view_2627);  sub_7673 = view_2627 = None
	        view_2629 = torch.ops.aten.view.default(mul_16244, [1280, 5120]);  mul_16244 = None
	        _assert_tensor_metadata_1513 = torch.ops.aten._assert_tensor_metadata.default(view_2629, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1513 = None
	        mul_16249 = sym_size_int * 1500
	        view_2630 = torch.ops.aten.view.default(mul_16239, [mul_16249, 5120]);  mul_16239 = mul_16249 = None
	        permute_280 = torch.ops.aten.permute.default(view_2629, [1, 0]);  view_2629 = None
	        addmm_139 = torch.ops.aten.addmm.default(model_audio_tower_layers_27_fc2_bias, view_2630, permute_280);  model_audio_tower_layers_27_fc2_bias = view_2630 = permute_280 = None
	        view_2631 = torch.ops.aten.view.default(addmm_139, [sym_size_int, 1500, 1280]);  addmm_139 = None
	        add_25730 = torch.ops.aten.add.Tensor(add_25432, view_2631);  add_25432 = view_2631 = None
	        clone_225 = torch.ops.aten.clone.default(add_25730, memory_format = torch.contiguous_format)
	        var_mean_56 = torch.ops.aten.var_mean.correction(clone_225, [2], correction = 0, keepdim = True)
	        getitem_224 = var_mean_56[0]
	        getitem_225 = var_mean_56[1];  var_mean_56 = None
	        add_25735 = torch.ops.aten.add.Tensor(getitem_224, 1e-05);  getitem_224 = None
	        rsqrt_56 = torch.ops.aten.rsqrt.default(add_25735);  add_25735 = None
	        sub_7679 = torch.ops.aten.sub.Tensor(clone_225, getitem_225);  clone_225 = getitem_225 = None
	        mul_16260 = torch.ops.aten.mul.Tensor(sub_7679, rsqrt_56);  sub_7679 = rsqrt_56 = None
	        mul_16261 = torch.ops.aten.mul.Tensor(mul_16260, model_audio_tower_layers_28_self_attn_layer_norm_weight);  mul_16260 = model_audio_tower_layers_28_self_attn_layer_norm_weight = None
	        add_25736 = torch.ops.aten.add.Tensor(mul_16261, model_audio_tower_layers_28_self_attn_layer_norm_bias);  mul_16261 = model_audio_tower_layers_28_self_attn_layer_norm_bias = None
	        amin_168 = torch.ops.aten.amin.default(add_25736, [2])
	        amax_168 = torch.ops.aten.amax.default(add_25736, [2])
	        full_336 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_168 = torch.ops.aten.minimum.default(amin_168, full_336);  amin_168 = full_336 = None
	        full_337 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_168 = torch.ops.aten.maximum.default(amax_168, full_337);  amax_168 = full_337 = None
	        sub_7690 = torch.ops.aten.sub.Tensor(maximum_168, minimum_168);  maximum_168 = None
	        div_336 = torch.ops.aten.div.Tensor(sub_7690, 255.0);  sub_7690 = None
	        clamp_min_504 = torch.ops.aten.clamp_min.default(div_336, 1.1920928955078125e-07);  div_336 = None
	        div_337 = torch.ops.aten.div.Tensor(minimum_168, clamp_min_504);  minimum_168 = None
	        round_337 = torch.ops.aten.round.default(div_337);  div_337 = None
	        sub_7696 = torch.ops.aten.sub.Tensor(-128, round_337);  round_337 = None
	        clamp_min_505 = torch.ops.aten.clamp_min.default(sub_7696, -128);  sub_7696 = None
	        clamp_max_336 = torch.ops.aten.clamp_max.default(clamp_min_505, 127);  clamp_min_505 = None
	        _assert_tensor_metadata_1514 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_504, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1514 = None
	        _assert_tensor_metadata_1515 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_336, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1515 = None
	        convert_element_type_1008 = torch.ops.prims.convert_element_type.default(clamp_max_336, torch.int8);  clamp_max_336 = None
	        view_2634 = torch.ops.aten.view.default(clamp_min_504, [sym_size_int, 1500, 1])
	        view_2635 = torch.ops.aten.view.default(convert_element_type_1008, [sym_size_int, 1500, 1])
	        reciprocal_168 = torch.ops.aten.reciprocal.default(view_2634);  view_2634 = None
	        mul_16309 = torch.ops.aten.mul.Tensor(reciprocal_168, 1.0);  reciprocal_168 = None
	        mul_16312 = torch.ops.aten.mul.Tensor(add_25736, mul_16309);  mul_16309 = None
	        round_338 = torch.ops.aten.round.default(mul_16312);  mul_16312 = None
	        add_25823 = torch.ops.aten.add.Tensor(round_338, view_2635);  round_338 = view_2635 = None
	        clamp_min_506 = torch.ops.aten.clamp_min.default(add_25823, -128);  add_25823 = None
	        clamp_max_337 = torch.ops.aten.clamp_max.default(clamp_min_506, 127);  clamp_min_506 = None
	        _assert_tensor_metadata_1516 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_337, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1516 = None
	        convert_element_type_1009 = torch.ops.prims.convert_element_type.default(clamp_max_337, torch.int8);  clamp_max_337 = None
	        view_2638 = torch.ops.aten.view.default(clamp_min_504, [sym_size_int, 1500, 1]);  clamp_min_504 = None
	        view_2639 = torch.ops.aten.view.default(convert_element_type_1008, [sym_size_int, 1500, 1]);  convert_element_type_1008 = None
	        _assert_tensor_metadata_1517 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1009, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1517 = None
	        convert_element_type_1010 = torch.ops.prims.convert_element_type.default(convert_element_type_1009, torch.float32);  convert_element_type_1009 = None
	        _assert_tensor_metadata_1518 = torch.ops.aten._assert_tensor_metadata.default(view_2639, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1518 = None
	        convert_element_type_1011 = torch.ops.prims.convert_element_type.default(view_2639, torch.float32);  view_2639 = None
	        sub_7716 = torch.ops.aten.sub.Tensor(convert_element_type_1010, convert_element_type_1011);  convert_element_type_1010 = convert_element_type_1011 = None
	        mul_16334 = torch.ops.aten.mul.Tensor(sub_7716, view_2638);  sub_7716 = view_2638 = None
	        _assert_tensor_metadata_1519 = torch.ops.aten._assert_tensor_metadata.default(mul_16334, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1519 = None
	        view_2641 = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2642 = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2643 = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1520 = torch.ops.aten._assert_tensor_metadata.default(view_2641, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1520 = None
	        convert_element_type_1012 = torch.ops.prims.convert_element_type.default(view_2641, torch.float32);  view_2641 = None
	        _assert_tensor_metadata_1521 = torch.ops.aten._assert_tensor_metadata.default(view_2643, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1521 = None
	        convert_element_type_1013 = torch.ops.prims.convert_element_type.default(view_2643, torch.float32);  view_2643 = None
	        sub_7720 = torch.ops.aten.sub.Tensor(convert_element_type_1012, convert_element_type_1013);  convert_element_type_1012 = convert_element_type_1013 = None
	        mul_16339 = torch.ops.aten.mul.Tensor(sub_7720, view_2642);  sub_7720 = view_2642 = None
	        view_2644 = torch.ops.aten.view.default(mul_16339, [1280, 1280]);  mul_16339 = None
	        _assert_tensor_metadata_1522 = torch.ops.aten._assert_tensor_metadata.default(view_2644, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1522 = None
	        mul_16344 = sym_size_int * 1500
	        view_2645 = torch.ops.aten.view.default(mul_16334, [mul_16344, 1280]);  mul_16334 = mul_16344 = None
	        permute_281 = torch.ops.aten.permute.default(view_2644, [1, 0]);  view_2644 = None
	        addmm_140 = torch.ops.aten.addmm.default(model_audio_tower_layers_28_self_attn_q_proj_bias, view_2645, permute_281);  model_audio_tower_layers_28_self_attn_q_proj_bias = view_2645 = permute_281 = None
	        view_2646 = torch.ops.aten.view.default(addmm_140, [sym_size_int, 1500, 1280]);  addmm_140 = None
	        mul_16351 = torch.ops.aten.mul.Tensor(view_2646, 0.125);  view_2646 = None
	        view_2647 = torch.ops.aten.view.default(mul_16351, [sym_size_int, 1500, 20, 64]);  mul_16351 = None
	        permute_282 = torch.ops.aten.permute.default(view_2647, [0, 2, 1, 3]);  view_2647 = None
	        clone_226 = torch.ops.aten.clone.default(permute_282, memory_format = torch.contiguous_format);  permute_282 = None
	        amin_169 = torch.ops.aten.amin.default(add_25736, [2])
	        amax_169 = torch.ops.aten.amax.default(add_25736, [2])
	        full_338 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_169 = torch.ops.aten.minimum.default(amin_169, full_338);  amin_169 = full_338 = None
	        full_339 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_169 = torch.ops.aten.maximum.default(amax_169, full_339);  amax_169 = full_339 = None
	        sub_7735 = torch.ops.aten.sub.Tensor(maximum_169, minimum_169);  maximum_169 = None
	        div_338 = torch.ops.aten.div.Tensor(sub_7735, 255.0);  sub_7735 = None
	        clamp_min_507 = torch.ops.aten.clamp_min.default(div_338, 1.1920928955078125e-07);  div_338 = None
	        div_339 = torch.ops.aten.div.Tensor(minimum_169, clamp_min_507);  minimum_169 = None
	        round_339 = torch.ops.aten.round.default(div_339);  div_339 = None
	        sub_7741 = torch.ops.aten.sub.Tensor(-128, round_339);  round_339 = None
	        clamp_min_508 = torch.ops.aten.clamp_min.default(sub_7741, -128);  sub_7741 = None
	        clamp_max_338 = torch.ops.aten.clamp_max.default(clamp_min_508, 127);  clamp_min_508 = None
	        _assert_tensor_metadata_1523 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_507, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1523 = None
	        _assert_tensor_metadata_1524 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_338, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1524 = None
	        convert_element_type_1014 = torch.ops.prims.convert_element_type.default(clamp_max_338, torch.int8);  clamp_max_338 = None
	        view_2650 = torch.ops.aten.view.default(clamp_min_507, [sym_size_int, 1500, 1])
	        view_2651 = torch.ops.aten.view.default(convert_element_type_1014, [sym_size_int, 1500, 1])
	        reciprocal_169 = torch.ops.aten.reciprocal.default(view_2650);  view_2650 = None
	        mul_16405 = torch.ops.aten.mul.Tensor(reciprocal_169, 1.0);  reciprocal_169 = None
	        mul_16408 = torch.ops.aten.mul.Tensor(add_25736, mul_16405);  mul_16405 = None
	        round_340 = torch.ops.aten.round.default(mul_16408);  mul_16408 = None
	        add_25975 = torch.ops.aten.add.Tensor(round_340, view_2651);  round_340 = view_2651 = None
	        clamp_min_509 = torch.ops.aten.clamp_min.default(add_25975, -128);  add_25975 = None
	        clamp_max_339 = torch.ops.aten.clamp_max.default(clamp_min_509, 127);  clamp_min_509 = None
	        _assert_tensor_metadata_1525 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_339, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1525 = None
	        convert_element_type_1015 = torch.ops.prims.convert_element_type.default(clamp_max_339, torch.int8);  clamp_max_339 = None
	        view_2654 = torch.ops.aten.view.default(clamp_min_507, [sym_size_int, 1500, 1]);  clamp_min_507 = None
	        view_2655 = torch.ops.aten.view.default(convert_element_type_1014, [sym_size_int, 1500, 1]);  convert_element_type_1014 = None
	        _assert_tensor_metadata_1526 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1015, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1526 = None
	        convert_element_type_1016 = torch.ops.prims.convert_element_type.default(convert_element_type_1015, torch.float32);  convert_element_type_1015 = None
	        _assert_tensor_metadata_1527 = torch.ops.aten._assert_tensor_metadata.default(view_2655, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1527 = None
	        convert_element_type_1017 = torch.ops.prims.convert_element_type.default(view_2655, torch.float32);  view_2655 = None
	        sub_7761 = torch.ops.aten.sub.Tensor(convert_element_type_1016, convert_element_type_1017);  convert_element_type_1016 = convert_element_type_1017 = None
	        mul_16430 = torch.ops.aten.mul.Tensor(sub_7761, view_2654);  sub_7761 = view_2654 = None
	        _assert_tensor_metadata_1528 = torch.ops.aten._assert_tensor_metadata.default(mul_16430, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1528 = None
	        view_2657 = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2658 = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2659 = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1529 = torch.ops.aten._assert_tensor_metadata.default(view_2657, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1529 = None
	        convert_element_type_1018 = torch.ops.prims.convert_element_type.default(view_2657, torch.float32);  view_2657 = None
	        _assert_tensor_metadata_1530 = torch.ops.aten._assert_tensor_metadata.default(view_2659, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1530 = None
	        convert_element_type_1019 = torch.ops.prims.convert_element_type.default(view_2659, torch.float32);  view_2659 = None
	        sub_7765 = torch.ops.aten.sub.Tensor(convert_element_type_1018, convert_element_type_1019);  convert_element_type_1018 = convert_element_type_1019 = None
	        mul_16435 = torch.ops.aten.mul.Tensor(sub_7765, view_2658);  sub_7765 = view_2658 = None
	        view_2660 = torch.ops.aten.view.default(mul_16435, [1280, 1280]);  mul_16435 = None
	        _assert_tensor_metadata_1531 = torch.ops.aten._assert_tensor_metadata.default(view_2660, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1531 = None
	        permute_283 = torch.ops.aten.permute.default(view_2660, [1, 0]);  view_2660 = None
	        mul_16438 = sym_size_int * 1500
	        view_2661 = torch.ops.aten.view.default(mul_16430, [mul_16438, 1280]);  mul_16430 = mul_16438 = None
	        mm_28 = torch.ops.aten.mm.default(view_2661, permute_283);  view_2661 = permute_283 = None
	        view_2662 = torch.ops.aten.view.default(mm_28, [sym_size_int, 1500, 1280]);  mm_28 = None
	        view_2663 = torch.ops.aten.view.default(view_2662, [sym_size_int, -1, 20, 64]);  view_2662 = None
	        permute_284 = torch.ops.aten.permute.default(view_2663, [0, 2, 1, 3]);  view_2663 = None
	        clone_227 = torch.ops.aten.clone.default(permute_284, memory_format = torch.contiguous_format);  permute_284 = None
	        amin_170 = torch.ops.aten.amin.default(add_25736, [2])
	        amax_170 = torch.ops.aten.amax.default(add_25736, [2])
	        full_340 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_170 = torch.ops.aten.minimum.default(amin_170, full_340);  amin_170 = full_340 = None
	        full_341 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_170 = torch.ops.aten.maximum.default(amax_170, full_341);  amax_170 = full_341 = None
	        sub_7779 = torch.ops.aten.sub.Tensor(maximum_170, minimum_170);  maximum_170 = None
	        div_340 = torch.ops.aten.div.Tensor(sub_7779, 255.0);  sub_7779 = None
	        clamp_min_510 = torch.ops.aten.clamp_min.default(div_340, 1.1920928955078125e-07);  div_340 = None
	        div_341 = torch.ops.aten.div.Tensor(minimum_170, clamp_min_510);  minimum_170 = None
	        round_341 = torch.ops.aten.round.default(div_341);  div_341 = None
	        sub_7785 = torch.ops.aten.sub.Tensor(-128, round_341);  round_341 = None
	        clamp_min_511 = torch.ops.aten.clamp_min.default(sub_7785, -128);  sub_7785 = None
	        clamp_max_340 = torch.ops.aten.clamp_max.default(clamp_min_511, 127);  clamp_min_511 = None
	        _assert_tensor_metadata_1532 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_510, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1532 = None
	        _assert_tensor_metadata_1533 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_340, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1533 = None
	        convert_element_type_1020 = torch.ops.prims.convert_element_type.default(clamp_max_340, torch.int8);  clamp_max_340 = None
	        view_2666 = torch.ops.aten.view.default(clamp_min_510, [sym_size_int, 1500, 1])
	        view_2667 = torch.ops.aten.view.default(convert_element_type_1020, [sym_size_int, 1500, 1])
	        reciprocal_170 = torch.ops.aten.reciprocal.default(view_2666);  view_2666 = None
	        mul_16504 = torch.ops.aten.mul.Tensor(reciprocal_170, 1.0);  reciprocal_170 = None
	        mul_16507 = torch.ops.aten.mul.Tensor(add_25736, mul_16504);  add_25736 = mul_16504 = None
	        round_342 = torch.ops.aten.round.default(mul_16507);  mul_16507 = None
	        add_26123 = torch.ops.aten.add.Tensor(round_342, view_2667);  round_342 = view_2667 = None
	        clamp_min_512 = torch.ops.aten.clamp_min.default(add_26123, -128);  add_26123 = None
	        clamp_max_341 = torch.ops.aten.clamp_max.default(clamp_min_512, 127);  clamp_min_512 = None
	        _assert_tensor_metadata_1534 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_341, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1534 = None
	        convert_element_type_1021 = torch.ops.prims.convert_element_type.default(clamp_max_341, torch.int8);  clamp_max_341 = None
	        view_2670 = torch.ops.aten.view.default(clamp_min_510, [sym_size_int, 1500, 1]);  clamp_min_510 = None
	        view_2671 = torch.ops.aten.view.default(convert_element_type_1020, [sym_size_int, 1500, 1]);  convert_element_type_1020 = None
	        _assert_tensor_metadata_1535 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1021, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1535 = None
	        convert_element_type_1022 = torch.ops.prims.convert_element_type.default(convert_element_type_1021, torch.float32);  convert_element_type_1021 = None
	        _assert_tensor_metadata_1536 = torch.ops.aten._assert_tensor_metadata.default(view_2671, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1536 = None
	        convert_element_type_1023 = torch.ops.prims.convert_element_type.default(view_2671, torch.float32);  view_2671 = None
	        sub_7805 = torch.ops.aten.sub.Tensor(convert_element_type_1022, convert_element_type_1023);  convert_element_type_1022 = convert_element_type_1023 = None
	        mul_16529 = torch.ops.aten.mul.Tensor(sub_7805, view_2670);  sub_7805 = view_2670 = None
	        _assert_tensor_metadata_1537 = torch.ops.aten._assert_tensor_metadata.default(mul_16529, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1537 = None
	        view_2673 = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2674 = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2675 = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1538 = torch.ops.aten._assert_tensor_metadata.default(view_2673, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1538 = None
	        convert_element_type_1024 = torch.ops.prims.convert_element_type.default(view_2673, torch.float32);  view_2673 = None
	        _assert_tensor_metadata_1539 = torch.ops.aten._assert_tensor_metadata.default(view_2675, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1539 = None
	        convert_element_type_1025 = torch.ops.prims.convert_element_type.default(view_2675, torch.float32);  view_2675 = None
	        sub_7809 = torch.ops.aten.sub.Tensor(convert_element_type_1024, convert_element_type_1025);  convert_element_type_1024 = convert_element_type_1025 = None
	        mul_16534 = torch.ops.aten.mul.Tensor(sub_7809, view_2674);  sub_7809 = view_2674 = None
	        view_2676 = torch.ops.aten.view.default(mul_16534, [1280, 1280]);  mul_16534 = None
	        _assert_tensor_metadata_1540 = torch.ops.aten._assert_tensor_metadata.default(view_2676, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1540 = None
	        mul_16539 = sym_size_int * 1500
	        view_2677 = torch.ops.aten.view.default(mul_16529, [mul_16539, 1280]);  mul_16529 = mul_16539 = None
	        permute_285 = torch.ops.aten.permute.default(view_2676, [1, 0]);  view_2676 = None
	        addmm_141 = torch.ops.aten.addmm.default(model_audio_tower_layers_28_self_attn_v_proj_bias, view_2677, permute_285);  model_audio_tower_layers_28_self_attn_v_proj_bias = view_2677 = permute_285 = None
	        view_2678 = torch.ops.aten.view.default(addmm_141, [sym_size_int, 1500, 1280]);  addmm_141 = None
	        view_2679 = torch.ops.aten.view.default(view_2678, [sym_size_int, -1, 20, 64]);  view_2678 = None
	        permute_286 = torch.ops.aten.permute.default(view_2679, [0, 2, 1, 3]);  view_2679 = None
	        clone_228 = torch.ops.aten.clone.default(permute_286, memory_format = torch.contiguous_format);  permute_286 = None
	        _scaled_dot_product_efficient_attention_28 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_226, clone_227, clone_228, None, False, scale = 1.0);  clone_226 = clone_227 = clone_228 = None
	        getitem_226 = _scaled_dot_product_efficient_attention_28[0];  _scaled_dot_product_efficient_attention_28 = None
	        permute_287 = torch.ops.aten.permute.default(getitem_226, [0, 2, 1, 3]);  getitem_226 = None
	        view_2680 = torch.ops.aten.view.default(permute_287, [sym_size_int, 1500, -1]);  permute_287 = None
	        amin_171 = torch.ops.aten.amin.default(view_2680, [2])
	        amax_171 = torch.ops.aten.amax.default(view_2680, [2])
	        full_342 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_171 = torch.ops.aten.minimum.default(amin_171, full_342);  amin_171 = full_342 = None
	        full_343 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_171 = torch.ops.aten.maximum.default(amax_171, full_343);  amax_171 = full_343 = None
	        sub_7827 = torch.ops.aten.sub.Tensor(maximum_171, minimum_171);  maximum_171 = None
	        div_342 = torch.ops.aten.div.Tensor(sub_7827, 255.0);  sub_7827 = None
	        clamp_min_513 = torch.ops.aten.clamp_min.default(div_342, 1.1920928955078125e-07);  div_342 = None
	        div_343 = torch.ops.aten.div.Tensor(minimum_171, clamp_min_513);  minimum_171 = None
	        round_343 = torch.ops.aten.round.default(div_343);  div_343 = None
	        sub_7833 = torch.ops.aten.sub.Tensor(-128, round_343);  round_343 = None
	        clamp_min_514 = torch.ops.aten.clamp_min.default(sub_7833, -128);  sub_7833 = None
	        clamp_max_342 = torch.ops.aten.clamp_max.default(clamp_min_514, 127);  clamp_min_514 = None
	        _assert_tensor_metadata_1541 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_513, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1541 = None
	        _assert_tensor_metadata_1542 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_342, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1542 = None
	        convert_element_type_1026 = torch.ops.prims.convert_element_type.default(clamp_max_342, torch.int8);  clamp_max_342 = None
	        view_2683 = torch.ops.aten.view.default(clamp_min_513, [sym_size_int, 1500, 1])
	        view_2684 = torch.ops.aten.view.default(convert_element_type_1026, [sym_size_int, 1500, 1])
	        reciprocal_171 = torch.ops.aten.reciprocal.default(view_2683);  view_2683 = None
	        mul_16609 = torch.ops.aten.mul.Tensor(reciprocal_171, 1.0);  reciprocal_171 = None
	        mul_16612 = torch.ops.aten.mul.Tensor(view_2680, mul_16609);  view_2680 = mul_16609 = None
	        round_344 = torch.ops.aten.round.default(mul_16612);  mul_16612 = None
	        add_26287 = torch.ops.aten.add.Tensor(round_344, view_2684);  round_344 = view_2684 = None
	        clamp_min_515 = torch.ops.aten.clamp_min.default(add_26287, -128);  add_26287 = None
	        clamp_max_343 = torch.ops.aten.clamp_max.default(clamp_min_515, 127);  clamp_min_515 = None
	        _assert_tensor_metadata_1543 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_343, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1543 = None
	        convert_element_type_1027 = torch.ops.prims.convert_element_type.default(clamp_max_343, torch.int8);  clamp_max_343 = None
	        view_2687 = torch.ops.aten.view.default(clamp_min_513, [sym_size_int, 1500, 1]);  clamp_min_513 = None
	        view_2688 = torch.ops.aten.view.default(convert_element_type_1026, [sym_size_int, 1500, 1]);  convert_element_type_1026 = None
	        _assert_tensor_metadata_1544 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1027, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1544 = None
	        convert_element_type_1028 = torch.ops.prims.convert_element_type.default(convert_element_type_1027, torch.float32);  convert_element_type_1027 = None
	        _assert_tensor_metadata_1545 = torch.ops.aten._assert_tensor_metadata.default(view_2688, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1545 = None
	        convert_element_type_1029 = torch.ops.prims.convert_element_type.default(view_2688, torch.float32);  view_2688 = None
	        sub_7853 = torch.ops.aten.sub.Tensor(convert_element_type_1028, convert_element_type_1029);  convert_element_type_1028 = convert_element_type_1029 = None
	        mul_16634 = torch.ops.aten.mul.Tensor(sub_7853, view_2687);  sub_7853 = view_2687 = None
	        _assert_tensor_metadata_1546 = torch.ops.aten._assert_tensor_metadata.default(mul_16634, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1546 = None
	        view_2690 = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2691 = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2692 = torch.ops.aten.view.default(model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1547 = torch.ops.aten._assert_tensor_metadata.default(view_2690, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1547 = None
	        convert_element_type_1030 = torch.ops.prims.convert_element_type.default(view_2690, torch.float32);  view_2690 = None
	        _assert_tensor_metadata_1548 = torch.ops.aten._assert_tensor_metadata.default(view_2692, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1548 = None
	        convert_element_type_1031 = torch.ops.prims.convert_element_type.default(view_2692, torch.float32);  view_2692 = None
	        sub_7857 = torch.ops.aten.sub.Tensor(convert_element_type_1030, convert_element_type_1031);  convert_element_type_1030 = convert_element_type_1031 = None
	        mul_16639 = torch.ops.aten.mul.Tensor(sub_7857, view_2691);  sub_7857 = view_2691 = None
	        view_2693 = torch.ops.aten.view.default(mul_16639, [1280, 1280]);  mul_16639 = None
	        _assert_tensor_metadata_1549 = torch.ops.aten._assert_tensor_metadata.default(view_2693, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1549 = None
	        mul_16644 = sym_size_int * 1500
	        view_2694 = torch.ops.aten.view.default(mul_16634, [mul_16644, 1280]);  mul_16634 = mul_16644 = None
	        permute_288 = torch.ops.aten.permute.default(view_2693, [1, 0]);  view_2693 = None
	        addmm_142 = torch.ops.aten.addmm.default(model_audio_tower_layers_28_self_attn_out_proj_bias, view_2694, permute_288);  model_audio_tower_layers_28_self_attn_out_proj_bias = view_2694 = permute_288 = None
	        view_2695 = torch.ops.aten.view.default(addmm_142, [sym_size_int, 1500, 1280]);  addmm_142 = None
	        add_26350 = torch.ops.aten.add.Tensor(add_25730, view_2695);  add_25730 = view_2695 = None
	        clone_230 = torch.ops.aten.clone.default(add_26350, memory_format = torch.contiguous_format)
	        var_mean_57 = torch.ops.aten.var_mean.correction(clone_230, [2], correction = 0, keepdim = True)
	        getitem_230 = var_mean_57[0]
	        getitem_231 = var_mean_57[1];  var_mean_57 = None
	        add_26355 = torch.ops.aten.add.Tensor(getitem_230, 1e-05);  getitem_230 = None
	        rsqrt_57 = torch.ops.aten.rsqrt.default(add_26355);  add_26355 = None
	        sub_7863 = torch.ops.aten.sub.Tensor(clone_230, getitem_231);  clone_230 = getitem_231 = None
	        mul_16655 = torch.ops.aten.mul.Tensor(sub_7863, rsqrt_57);  sub_7863 = rsqrt_57 = None
	        mul_16656 = torch.ops.aten.mul.Tensor(mul_16655, model_audio_tower_layers_28_final_layer_norm_weight);  mul_16655 = model_audio_tower_layers_28_final_layer_norm_weight = None
	        add_26356 = torch.ops.aten.add.Tensor(mul_16656, model_audio_tower_layers_28_final_layer_norm_bias);  mul_16656 = model_audio_tower_layers_28_final_layer_norm_bias = None
	        amin_172 = torch.ops.aten.amin.default(add_26356, [2])
	        amax_172 = torch.ops.aten.amax.default(add_26356, [2])
	        full_344 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_172 = torch.ops.aten.minimum.default(amin_172, full_344);  amin_172 = full_344 = None
	        full_345 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_172 = torch.ops.aten.maximum.default(amax_172, full_345);  amax_172 = full_345 = None
	        sub_7874 = torch.ops.aten.sub.Tensor(maximum_172, minimum_172);  maximum_172 = None
	        div_344 = torch.ops.aten.div.Tensor(sub_7874, 255.0);  sub_7874 = None
	        clamp_min_516 = torch.ops.aten.clamp_min.default(div_344, 1.1920928955078125e-07);  div_344 = None
	        div_345 = torch.ops.aten.div.Tensor(minimum_172, clamp_min_516);  minimum_172 = None
	        round_345 = torch.ops.aten.round.default(div_345);  div_345 = None
	        sub_7880 = torch.ops.aten.sub.Tensor(-128, round_345);  round_345 = None
	        clamp_min_517 = torch.ops.aten.clamp_min.default(sub_7880, -128);  sub_7880 = None
	        clamp_max_344 = torch.ops.aten.clamp_max.default(clamp_min_517, 127);  clamp_min_517 = None
	        _assert_tensor_metadata_1550 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_516, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1550 = None
	        _assert_tensor_metadata_1551 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_344, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1551 = None
	        convert_element_type_1032 = torch.ops.prims.convert_element_type.default(clamp_max_344, torch.int8);  clamp_max_344 = None
	        view_2698 = torch.ops.aten.view.default(clamp_min_516, [sym_size_int, 1500, 1])
	        view_2699 = torch.ops.aten.view.default(convert_element_type_1032, [sym_size_int, 1500, 1])
	        reciprocal_172 = torch.ops.aten.reciprocal.default(view_2698);  view_2698 = None
	        mul_16704 = torch.ops.aten.mul.Tensor(reciprocal_172, 1.0);  reciprocal_172 = None
	        mul_16707 = torch.ops.aten.mul.Tensor(add_26356, mul_16704);  add_26356 = mul_16704 = None
	        round_346 = torch.ops.aten.round.default(mul_16707);  mul_16707 = None
	        add_26443 = torch.ops.aten.add.Tensor(round_346, view_2699);  round_346 = view_2699 = None
	        clamp_min_518 = torch.ops.aten.clamp_min.default(add_26443, -128);  add_26443 = None
	        clamp_max_345 = torch.ops.aten.clamp_max.default(clamp_min_518, 127);  clamp_min_518 = None
	        _assert_tensor_metadata_1552 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_345, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1552 = None
	        convert_element_type_1033 = torch.ops.prims.convert_element_type.default(clamp_max_345, torch.int8);  clamp_max_345 = None
	        view_2702 = torch.ops.aten.view.default(clamp_min_516, [sym_size_int, 1500, 1]);  clamp_min_516 = None
	        view_2703 = torch.ops.aten.view.default(convert_element_type_1032, [sym_size_int, 1500, 1]);  convert_element_type_1032 = None
	        _assert_tensor_metadata_1553 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1033, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1553 = None
	        convert_element_type_1034 = torch.ops.prims.convert_element_type.default(convert_element_type_1033, torch.float32);  convert_element_type_1033 = None
	        _assert_tensor_metadata_1554 = torch.ops.aten._assert_tensor_metadata.default(view_2703, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1554 = None
	        convert_element_type_1035 = torch.ops.prims.convert_element_type.default(view_2703, torch.float32);  view_2703 = None
	        sub_7900 = torch.ops.aten.sub.Tensor(convert_element_type_1034, convert_element_type_1035);  convert_element_type_1034 = convert_element_type_1035 = None
	        mul_16729 = torch.ops.aten.mul.Tensor(sub_7900, view_2702);  sub_7900 = view_2702 = None
	        _assert_tensor_metadata_1555 = torch.ops.aten._assert_tensor_metadata.default(mul_16729, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1555 = None
	        view_2705 = torch.ops.aten.view.default(model_audio_tower_layers_28_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_28_fc1_parametrizations_weight_original0 = None
	        view_2706 = torch.ops.aten.view.default(model_audio_tower_layers_28_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_28_fc1_parametrizations_weight_original1 = None
	        view_2707 = torch.ops.aten.view.default(model_audio_tower_layers_28_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_28_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1556 = torch.ops.aten._assert_tensor_metadata.default(view_2705, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1556 = None
	        convert_element_type_1036 = torch.ops.prims.convert_element_type.default(view_2705, torch.float32);  view_2705 = None
	        _assert_tensor_metadata_1557 = torch.ops.aten._assert_tensor_metadata.default(view_2707, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1557 = None
	        convert_element_type_1037 = torch.ops.prims.convert_element_type.default(view_2707, torch.float32);  view_2707 = None
	        sub_7904 = torch.ops.aten.sub.Tensor(convert_element_type_1036, convert_element_type_1037);  convert_element_type_1036 = convert_element_type_1037 = None
	        mul_16734 = torch.ops.aten.mul.Tensor(sub_7904, view_2706);  sub_7904 = view_2706 = None
	        view_2708 = torch.ops.aten.view.default(mul_16734, [5120, 1280]);  mul_16734 = None
	        _assert_tensor_metadata_1558 = torch.ops.aten._assert_tensor_metadata.default(view_2708, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1558 = None
	        mul_16739 = sym_size_int * 1500
	        view_2709 = torch.ops.aten.view.default(mul_16729, [mul_16739, 1280]);  mul_16729 = mul_16739 = None
	        permute_289 = torch.ops.aten.permute.default(view_2708, [1, 0]);  view_2708 = None
	        addmm_143 = torch.ops.aten.addmm.default(model_audio_tower_layers_28_fc1_bias, view_2709, permute_289);  model_audio_tower_layers_28_fc1_bias = view_2709 = permute_289 = None
	        view_2710 = torch.ops.aten.view.default(addmm_143, [sym_size_int, 1500, 5120]);  addmm_143 = None
	        mul_16746 = torch.ops.aten.mul.Tensor(view_2710, 0.5)
	        mul_16747 = torch.ops.aten.mul.Tensor(view_2710, 0.7071067811865476);  view_2710 = None
	        erf_30 = torch.ops.aten.erf.default(mul_16747);  mul_16747 = None
	        add_26502 = torch.ops.aten.add.Tensor(erf_30, 1);  erf_30 = None
	        mul_16748 = torch.ops.aten.mul.Tensor(mul_16746, add_26502);  mul_16746 = add_26502 = None
	        amin_173 = torch.ops.aten.amin.default(mul_16748, [2])
	        amax_173 = torch.ops.aten.amax.default(mul_16748, [2])
	        full_346 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_173 = torch.ops.aten.minimum.default(amin_173, full_346);  amin_173 = full_346 = None
	        full_347 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_173 = torch.ops.aten.maximum.default(amax_173, full_347);  amax_173 = full_347 = None
	        sub_7917 = torch.ops.aten.sub.Tensor(maximum_173, minimum_173);  maximum_173 = None
	        div_346 = torch.ops.aten.div.Tensor(sub_7917, 255.0);  sub_7917 = None
	        clamp_min_519 = torch.ops.aten.clamp_min.default(div_346, 1.1920928955078125e-07);  div_346 = None
	        div_347 = torch.ops.aten.div.Tensor(minimum_173, clamp_min_519);  minimum_173 = None
	        round_347 = torch.ops.aten.round.default(div_347);  div_347 = None
	        sub_7923 = torch.ops.aten.sub.Tensor(-128, round_347);  round_347 = None
	        clamp_min_520 = torch.ops.aten.clamp_min.default(sub_7923, -128);  sub_7923 = None
	        clamp_max_346 = torch.ops.aten.clamp_max.default(clamp_min_520, 127);  clamp_min_520 = None
	        _assert_tensor_metadata_1559 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_519, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1559 = None
	        _assert_tensor_metadata_1560 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_346, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1560 = None
	        convert_element_type_1038 = torch.ops.prims.convert_element_type.default(clamp_max_346, torch.int8);  clamp_max_346 = None
	        view_2713 = torch.ops.aten.view.default(clamp_min_519, [sym_size_int, 1500, 1])
	        view_2714 = torch.ops.aten.view.default(convert_element_type_1038, [sym_size_int, 1500, 1])
	        reciprocal_173 = torch.ops.aten.reciprocal.default(view_2713);  view_2713 = None
	        mul_16794 = torch.ops.aten.mul.Tensor(reciprocal_173, 1.0);  reciprocal_173 = None
	        mul_16797 = torch.ops.aten.mul.Tensor(mul_16748, mul_16794);  mul_16748 = mul_16794 = None
	        round_348 = torch.ops.aten.round.default(mul_16797);  mul_16797 = None
	        add_26585 = torch.ops.aten.add.Tensor(round_348, view_2714);  round_348 = view_2714 = None
	        clamp_min_521 = torch.ops.aten.clamp_min.default(add_26585, -128);  add_26585 = None
	        clamp_max_347 = torch.ops.aten.clamp_max.default(clamp_min_521, 127);  clamp_min_521 = None
	        _assert_tensor_metadata_1561 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_347, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1561 = None
	        convert_element_type_1039 = torch.ops.prims.convert_element_type.default(clamp_max_347, torch.int8);  clamp_max_347 = None
	        view_2717 = torch.ops.aten.view.default(clamp_min_519, [sym_size_int, 1500, 1]);  clamp_min_519 = None
	        view_2718 = torch.ops.aten.view.default(convert_element_type_1038, [sym_size_int, 1500, 1]);  convert_element_type_1038 = None
	        _assert_tensor_metadata_1562 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1039, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1562 = None
	        convert_element_type_1040 = torch.ops.prims.convert_element_type.default(convert_element_type_1039, torch.float32);  convert_element_type_1039 = None
	        _assert_tensor_metadata_1563 = torch.ops.aten._assert_tensor_metadata.default(view_2718, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1563 = None
	        convert_element_type_1041 = torch.ops.prims.convert_element_type.default(view_2718, torch.float32);  view_2718 = None
	        sub_7943 = torch.ops.aten.sub.Tensor(convert_element_type_1040, convert_element_type_1041);  convert_element_type_1040 = convert_element_type_1041 = None
	        mul_16819 = torch.ops.aten.mul.Tensor(sub_7943, view_2717);  sub_7943 = view_2717 = None
	        _assert_tensor_metadata_1564 = torch.ops.aten._assert_tensor_metadata.default(mul_16819, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1564 = None
	        view_2720 = torch.ops.aten.view.default(model_audio_tower_layers_28_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_28_fc2_parametrizations_weight_original0 = None
	        view_2721 = torch.ops.aten.view.default(model_audio_tower_layers_28_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_28_fc2_parametrizations_weight_original1 = None
	        view_2722 = torch.ops.aten.view.default(model_audio_tower_layers_28_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_28_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1565 = torch.ops.aten._assert_tensor_metadata.default(view_2720, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1565 = None
	        convert_element_type_1042 = torch.ops.prims.convert_element_type.default(view_2720, torch.float32);  view_2720 = None
	        _assert_tensor_metadata_1566 = torch.ops.aten._assert_tensor_metadata.default(view_2722, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1566 = None
	        convert_element_type_1043 = torch.ops.prims.convert_element_type.default(view_2722, torch.float32);  view_2722 = None
	        sub_7947 = torch.ops.aten.sub.Tensor(convert_element_type_1042, convert_element_type_1043);  convert_element_type_1042 = convert_element_type_1043 = None
	        mul_16824 = torch.ops.aten.mul.Tensor(sub_7947, view_2721);  sub_7947 = view_2721 = None
	        view_2723 = torch.ops.aten.view.default(mul_16824, [1280, 5120]);  mul_16824 = None
	        _assert_tensor_metadata_1567 = torch.ops.aten._assert_tensor_metadata.default(view_2723, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1567 = None
	        mul_16829 = sym_size_int * 1500
	        view_2724 = torch.ops.aten.view.default(mul_16819, [mul_16829, 5120]);  mul_16819 = mul_16829 = None
	        permute_290 = torch.ops.aten.permute.default(view_2723, [1, 0]);  view_2723 = None
	        addmm_144 = torch.ops.aten.addmm.default(model_audio_tower_layers_28_fc2_bias, view_2724, permute_290);  model_audio_tower_layers_28_fc2_bias = view_2724 = permute_290 = None
	        view_2725 = torch.ops.aten.view.default(addmm_144, [sym_size_int, 1500, 1280]);  addmm_144 = None
	        add_26648 = torch.ops.aten.add.Tensor(add_26350, view_2725);  add_26350 = view_2725 = None
	        clone_233 = torch.ops.aten.clone.default(add_26648, memory_format = torch.contiguous_format)
	        var_mean_58 = torch.ops.aten.var_mean.correction(clone_233, [2], correction = 0, keepdim = True)
	        getitem_232 = var_mean_58[0]
	        getitem_233 = var_mean_58[1];  var_mean_58 = None
	        add_26653 = torch.ops.aten.add.Tensor(getitem_232, 1e-05);  getitem_232 = None
	        rsqrt_58 = torch.ops.aten.rsqrt.default(add_26653);  add_26653 = None
	        sub_7953 = torch.ops.aten.sub.Tensor(clone_233, getitem_233);  clone_233 = getitem_233 = None
	        mul_16840 = torch.ops.aten.mul.Tensor(sub_7953, rsqrt_58);  sub_7953 = rsqrt_58 = None
	        mul_16841 = torch.ops.aten.mul.Tensor(mul_16840, model_audio_tower_layers_29_self_attn_layer_norm_weight);  mul_16840 = model_audio_tower_layers_29_self_attn_layer_norm_weight = None
	        add_26654 = torch.ops.aten.add.Tensor(mul_16841, model_audio_tower_layers_29_self_attn_layer_norm_bias);  mul_16841 = model_audio_tower_layers_29_self_attn_layer_norm_bias = None
	        amin_174 = torch.ops.aten.amin.default(add_26654, [2])
	        amax_174 = torch.ops.aten.amax.default(add_26654, [2])
	        full_348 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_174 = torch.ops.aten.minimum.default(amin_174, full_348);  amin_174 = full_348 = None
	        full_349 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_174 = torch.ops.aten.maximum.default(amax_174, full_349);  amax_174 = full_349 = None
	        sub_7964 = torch.ops.aten.sub.Tensor(maximum_174, minimum_174);  maximum_174 = None
	        div_348 = torch.ops.aten.div.Tensor(sub_7964, 255.0);  sub_7964 = None
	        clamp_min_522 = torch.ops.aten.clamp_min.default(div_348, 1.1920928955078125e-07);  div_348 = None
	        div_349 = torch.ops.aten.div.Tensor(minimum_174, clamp_min_522);  minimum_174 = None
	        round_349 = torch.ops.aten.round.default(div_349);  div_349 = None
	        sub_7970 = torch.ops.aten.sub.Tensor(-128, round_349);  round_349 = None
	        clamp_min_523 = torch.ops.aten.clamp_min.default(sub_7970, -128);  sub_7970 = None
	        clamp_max_348 = torch.ops.aten.clamp_max.default(clamp_min_523, 127);  clamp_min_523 = None
	        _assert_tensor_metadata_1568 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_522, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1568 = None
	        _assert_tensor_metadata_1569 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_348, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1569 = None
	        convert_element_type_1044 = torch.ops.prims.convert_element_type.default(clamp_max_348, torch.int8);  clamp_max_348 = None
	        view_2728 = torch.ops.aten.view.default(clamp_min_522, [sym_size_int, 1500, 1])
	        view_2729 = torch.ops.aten.view.default(convert_element_type_1044, [sym_size_int, 1500, 1])
	        reciprocal_174 = torch.ops.aten.reciprocal.default(view_2728);  view_2728 = None
	        mul_16889 = torch.ops.aten.mul.Tensor(reciprocal_174, 1.0);  reciprocal_174 = None
	        mul_16892 = torch.ops.aten.mul.Tensor(add_26654, mul_16889);  mul_16889 = None
	        round_350 = torch.ops.aten.round.default(mul_16892);  mul_16892 = None
	        add_26741 = torch.ops.aten.add.Tensor(round_350, view_2729);  round_350 = view_2729 = None
	        clamp_min_524 = torch.ops.aten.clamp_min.default(add_26741, -128);  add_26741 = None
	        clamp_max_349 = torch.ops.aten.clamp_max.default(clamp_min_524, 127);  clamp_min_524 = None
	        _assert_tensor_metadata_1570 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_349, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1570 = None
	        convert_element_type_1045 = torch.ops.prims.convert_element_type.default(clamp_max_349, torch.int8);  clamp_max_349 = None
	        view_2732 = torch.ops.aten.view.default(clamp_min_522, [sym_size_int, 1500, 1]);  clamp_min_522 = None
	        view_2733 = torch.ops.aten.view.default(convert_element_type_1044, [sym_size_int, 1500, 1]);  convert_element_type_1044 = None
	        _assert_tensor_metadata_1571 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1045, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1571 = None
	        convert_element_type_1046 = torch.ops.prims.convert_element_type.default(convert_element_type_1045, torch.float32);  convert_element_type_1045 = None
	        _assert_tensor_metadata_1572 = torch.ops.aten._assert_tensor_metadata.default(view_2733, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1572 = None
	        convert_element_type_1047 = torch.ops.prims.convert_element_type.default(view_2733, torch.float32);  view_2733 = None
	        sub_7990 = torch.ops.aten.sub.Tensor(convert_element_type_1046, convert_element_type_1047);  convert_element_type_1046 = convert_element_type_1047 = None
	        mul_16914 = torch.ops.aten.mul.Tensor(sub_7990, view_2732);  sub_7990 = view_2732 = None
	        _assert_tensor_metadata_1573 = torch.ops.aten._assert_tensor_metadata.default(mul_16914, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1573 = None
	        view_2735 = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2736 = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2737 = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1574 = torch.ops.aten._assert_tensor_metadata.default(view_2735, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1574 = None
	        convert_element_type_1048 = torch.ops.prims.convert_element_type.default(view_2735, torch.float32);  view_2735 = None
	        _assert_tensor_metadata_1575 = torch.ops.aten._assert_tensor_metadata.default(view_2737, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1575 = None
	        convert_element_type_1049 = torch.ops.prims.convert_element_type.default(view_2737, torch.float32);  view_2737 = None
	        sub_7994 = torch.ops.aten.sub.Tensor(convert_element_type_1048, convert_element_type_1049);  convert_element_type_1048 = convert_element_type_1049 = None
	        mul_16919 = torch.ops.aten.mul.Tensor(sub_7994, view_2736);  sub_7994 = view_2736 = None
	        view_2738 = torch.ops.aten.view.default(mul_16919, [1280, 1280]);  mul_16919 = None
	        _assert_tensor_metadata_1576 = torch.ops.aten._assert_tensor_metadata.default(view_2738, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1576 = None
	        mul_16924 = sym_size_int * 1500
	        view_2739 = torch.ops.aten.view.default(mul_16914, [mul_16924, 1280]);  mul_16914 = mul_16924 = None
	        permute_291 = torch.ops.aten.permute.default(view_2738, [1, 0]);  view_2738 = None
	        addmm_145 = torch.ops.aten.addmm.default(model_audio_tower_layers_29_self_attn_q_proj_bias, view_2739, permute_291);  model_audio_tower_layers_29_self_attn_q_proj_bias = view_2739 = permute_291 = None
	        view_2740 = torch.ops.aten.view.default(addmm_145, [sym_size_int, 1500, 1280]);  addmm_145 = None
	        mul_16931 = torch.ops.aten.mul.Tensor(view_2740, 0.125);  view_2740 = None
	        view_2741 = torch.ops.aten.view.default(mul_16931, [sym_size_int, 1500, 20, 64]);  mul_16931 = None
	        permute_292 = torch.ops.aten.permute.default(view_2741, [0, 2, 1, 3]);  view_2741 = None
	        clone_234 = torch.ops.aten.clone.default(permute_292, memory_format = torch.contiguous_format);  permute_292 = None
	        amin_175 = torch.ops.aten.amin.default(add_26654, [2])
	        amax_175 = torch.ops.aten.amax.default(add_26654, [2])
	        full_350 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_175 = torch.ops.aten.minimum.default(amin_175, full_350);  amin_175 = full_350 = None
	        full_351 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_175 = torch.ops.aten.maximum.default(amax_175, full_351);  amax_175 = full_351 = None
	        sub_8009 = torch.ops.aten.sub.Tensor(maximum_175, minimum_175);  maximum_175 = None
	        div_350 = torch.ops.aten.div.Tensor(sub_8009, 255.0);  sub_8009 = None
	        clamp_min_525 = torch.ops.aten.clamp_min.default(div_350, 1.1920928955078125e-07);  div_350 = None
	        div_351 = torch.ops.aten.div.Tensor(minimum_175, clamp_min_525);  minimum_175 = None
	        round_351 = torch.ops.aten.round.default(div_351);  div_351 = None
	        sub_8015 = torch.ops.aten.sub.Tensor(-128, round_351);  round_351 = None
	        clamp_min_526 = torch.ops.aten.clamp_min.default(sub_8015, -128);  sub_8015 = None
	        clamp_max_350 = torch.ops.aten.clamp_max.default(clamp_min_526, 127);  clamp_min_526 = None
	        _assert_tensor_metadata_1577 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_525, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1577 = None
	        _assert_tensor_metadata_1578 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_350, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1578 = None
	        convert_element_type_1050 = torch.ops.prims.convert_element_type.default(clamp_max_350, torch.int8);  clamp_max_350 = None
	        view_2744 = torch.ops.aten.view.default(clamp_min_525, [sym_size_int, 1500, 1])
	        view_2745 = torch.ops.aten.view.default(convert_element_type_1050, [sym_size_int, 1500, 1])
	        reciprocal_175 = torch.ops.aten.reciprocal.default(view_2744);  view_2744 = None
	        mul_16985 = torch.ops.aten.mul.Tensor(reciprocal_175, 1.0);  reciprocal_175 = None
	        mul_16988 = torch.ops.aten.mul.Tensor(add_26654, mul_16985);  mul_16985 = None
	        round_352 = torch.ops.aten.round.default(mul_16988);  mul_16988 = None
	        add_26893 = torch.ops.aten.add.Tensor(round_352, view_2745);  round_352 = view_2745 = None
	        clamp_min_527 = torch.ops.aten.clamp_min.default(add_26893, -128);  add_26893 = None
	        clamp_max_351 = torch.ops.aten.clamp_max.default(clamp_min_527, 127);  clamp_min_527 = None
	        _assert_tensor_metadata_1579 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_351, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1579 = None
	        convert_element_type_1051 = torch.ops.prims.convert_element_type.default(clamp_max_351, torch.int8);  clamp_max_351 = None
	        view_2748 = torch.ops.aten.view.default(clamp_min_525, [sym_size_int, 1500, 1]);  clamp_min_525 = None
	        view_2749 = torch.ops.aten.view.default(convert_element_type_1050, [sym_size_int, 1500, 1]);  convert_element_type_1050 = None
	        _assert_tensor_metadata_1580 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1051, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1580 = None
	        convert_element_type_1052 = torch.ops.prims.convert_element_type.default(convert_element_type_1051, torch.float32);  convert_element_type_1051 = None
	        _assert_tensor_metadata_1581 = torch.ops.aten._assert_tensor_metadata.default(view_2749, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1581 = None
	        convert_element_type_1053 = torch.ops.prims.convert_element_type.default(view_2749, torch.float32);  view_2749 = None
	        sub_8035 = torch.ops.aten.sub.Tensor(convert_element_type_1052, convert_element_type_1053);  convert_element_type_1052 = convert_element_type_1053 = None
	        mul_17010 = torch.ops.aten.mul.Tensor(sub_8035, view_2748);  sub_8035 = view_2748 = None
	        _assert_tensor_metadata_1582 = torch.ops.aten._assert_tensor_metadata.default(mul_17010, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1582 = None
	        view_2751 = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2752 = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2753 = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1583 = torch.ops.aten._assert_tensor_metadata.default(view_2751, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1583 = None
	        convert_element_type_1054 = torch.ops.prims.convert_element_type.default(view_2751, torch.float32);  view_2751 = None
	        _assert_tensor_metadata_1584 = torch.ops.aten._assert_tensor_metadata.default(view_2753, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1584 = None
	        convert_element_type_1055 = torch.ops.prims.convert_element_type.default(view_2753, torch.float32);  view_2753 = None
	        sub_8039 = torch.ops.aten.sub.Tensor(convert_element_type_1054, convert_element_type_1055);  convert_element_type_1054 = convert_element_type_1055 = None
	        mul_17015 = torch.ops.aten.mul.Tensor(sub_8039, view_2752);  sub_8039 = view_2752 = None
	        view_2754 = torch.ops.aten.view.default(mul_17015, [1280, 1280]);  mul_17015 = None
	        _assert_tensor_metadata_1585 = torch.ops.aten._assert_tensor_metadata.default(view_2754, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1585 = None
	        permute_293 = torch.ops.aten.permute.default(view_2754, [1, 0]);  view_2754 = None
	        mul_17018 = sym_size_int * 1500
	        view_2755 = torch.ops.aten.view.default(mul_17010, [mul_17018, 1280]);  mul_17010 = mul_17018 = None
	        mm_29 = torch.ops.aten.mm.default(view_2755, permute_293);  view_2755 = permute_293 = None
	        view_2756 = torch.ops.aten.view.default(mm_29, [sym_size_int, 1500, 1280]);  mm_29 = None
	        view_2757 = torch.ops.aten.view.default(view_2756, [sym_size_int, -1, 20, 64]);  view_2756 = None
	        permute_294 = torch.ops.aten.permute.default(view_2757, [0, 2, 1, 3]);  view_2757 = None
	        clone_235 = torch.ops.aten.clone.default(permute_294, memory_format = torch.contiguous_format);  permute_294 = None
	        amin_176 = torch.ops.aten.amin.default(add_26654, [2])
	        amax_176 = torch.ops.aten.amax.default(add_26654, [2])
	        full_352 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_176 = torch.ops.aten.minimum.default(amin_176, full_352);  amin_176 = full_352 = None
	        full_353 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_176 = torch.ops.aten.maximum.default(amax_176, full_353);  amax_176 = full_353 = None
	        sub_8053 = torch.ops.aten.sub.Tensor(maximum_176, minimum_176);  maximum_176 = None
	        div_352 = torch.ops.aten.div.Tensor(sub_8053, 255.0);  sub_8053 = None
	        clamp_min_528 = torch.ops.aten.clamp_min.default(div_352, 1.1920928955078125e-07);  div_352 = None
	        div_353 = torch.ops.aten.div.Tensor(minimum_176, clamp_min_528);  minimum_176 = None
	        round_353 = torch.ops.aten.round.default(div_353);  div_353 = None
	        sub_8059 = torch.ops.aten.sub.Tensor(-128, round_353);  round_353 = None
	        clamp_min_529 = torch.ops.aten.clamp_min.default(sub_8059, -128);  sub_8059 = None
	        clamp_max_352 = torch.ops.aten.clamp_max.default(clamp_min_529, 127);  clamp_min_529 = None
	        _assert_tensor_metadata_1586 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_528, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1586 = None
	        _assert_tensor_metadata_1587 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_352, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1587 = None
	        convert_element_type_1056 = torch.ops.prims.convert_element_type.default(clamp_max_352, torch.int8);  clamp_max_352 = None
	        view_2760 = torch.ops.aten.view.default(clamp_min_528, [sym_size_int, 1500, 1])
	        view_2761 = torch.ops.aten.view.default(convert_element_type_1056, [sym_size_int, 1500, 1])
	        reciprocal_176 = torch.ops.aten.reciprocal.default(view_2760);  view_2760 = None
	        mul_17084 = torch.ops.aten.mul.Tensor(reciprocal_176, 1.0);  reciprocal_176 = None
	        mul_17087 = torch.ops.aten.mul.Tensor(add_26654, mul_17084);  add_26654 = mul_17084 = None
	        round_354 = torch.ops.aten.round.default(mul_17087);  mul_17087 = None
	        add_27041 = torch.ops.aten.add.Tensor(round_354, view_2761);  round_354 = view_2761 = None
	        clamp_min_530 = torch.ops.aten.clamp_min.default(add_27041, -128);  add_27041 = None
	        clamp_max_353 = torch.ops.aten.clamp_max.default(clamp_min_530, 127);  clamp_min_530 = None
	        _assert_tensor_metadata_1588 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_353, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1588 = None
	        convert_element_type_1057 = torch.ops.prims.convert_element_type.default(clamp_max_353, torch.int8);  clamp_max_353 = None
	        view_2764 = torch.ops.aten.view.default(clamp_min_528, [sym_size_int, 1500, 1]);  clamp_min_528 = None
	        view_2765 = torch.ops.aten.view.default(convert_element_type_1056, [sym_size_int, 1500, 1]);  convert_element_type_1056 = None
	        _assert_tensor_metadata_1589 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1057, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1589 = None
	        convert_element_type_1058 = torch.ops.prims.convert_element_type.default(convert_element_type_1057, torch.float32);  convert_element_type_1057 = None
	        _assert_tensor_metadata_1590 = torch.ops.aten._assert_tensor_metadata.default(view_2765, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1590 = None
	        convert_element_type_1059 = torch.ops.prims.convert_element_type.default(view_2765, torch.float32);  view_2765 = None
	        sub_8079 = torch.ops.aten.sub.Tensor(convert_element_type_1058, convert_element_type_1059);  convert_element_type_1058 = convert_element_type_1059 = None
	        mul_17109 = torch.ops.aten.mul.Tensor(sub_8079, view_2764);  sub_8079 = view_2764 = None
	        _assert_tensor_metadata_1591 = torch.ops.aten._assert_tensor_metadata.default(mul_17109, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1591 = None
	        view_2767 = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2768 = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2769 = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1592 = torch.ops.aten._assert_tensor_metadata.default(view_2767, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1592 = None
	        convert_element_type_1060 = torch.ops.prims.convert_element_type.default(view_2767, torch.float32);  view_2767 = None
	        _assert_tensor_metadata_1593 = torch.ops.aten._assert_tensor_metadata.default(view_2769, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1593 = None
	        convert_element_type_1061 = torch.ops.prims.convert_element_type.default(view_2769, torch.float32);  view_2769 = None
	        sub_8083 = torch.ops.aten.sub.Tensor(convert_element_type_1060, convert_element_type_1061);  convert_element_type_1060 = convert_element_type_1061 = None
	        mul_17114 = torch.ops.aten.mul.Tensor(sub_8083, view_2768);  sub_8083 = view_2768 = None
	        view_2770 = torch.ops.aten.view.default(mul_17114, [1280, 1280]);  mul_17114 = None
	        _assert_tensor_metadata_1594 = torch.ops.aten._assert_tensor_metadata.default(view_2770, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1594 = None
	        mul_17119 = sym_size_int * 1500
	        view_2771 = torch.ops.aten.view.default(mul_17109, [mul_17119, 1280]);  mul_17109 = mul_17119 = None
	        permute_295 = torch.ops.aten.permute.default(view_2770, [1, 0]);  view_2770 = None
	        addmm_146 = torch.ops.aten.addmm.default(model_audio_tower_layers_29_self_attn_v_proj_bias, view_2771, permute_295);  model_audio_tower_layers_29_self_attn_v_proj_bias = view_2771 = permute_295 = None
	        view_2772 = torch.ops.aten.view.default(addmm_146, [sym_size_int, 1500, 1280]);  addmm_146 = None
	        view_2773 = torch.ops.aten.view.default(view_2772, [sym_size_int, -1, 20, 64]);  view_2772 = None
	        permute_296 = torch.ops.aten.permute.default(view_2773, [0, 2, 1, 3]);  view_2773 = None
	        clone_236 = torch.ops.aten.clone.default(permute_296, memory_format = torch.contiguous_format);  permute_296 = None
	        _scaled_dot_product_efficient_attention_29 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_234, clone_235, clone_236, None, False, scale = 1.0);  clone_234 = clone_235 = clone_236 = None
	        getitem_234 = _scaled_dot_product_efficient_attention_29[0];  _scaled_dot_product_efficient_attention_29 = None
	        permute_297 = torch.ops.aten.permute.default(getitem_234, [0, 2, 1, 3]);  getitem_234 = None
	        view_2774 = torch.ops.aten.view.default(permute_297, [sym_size_int, 1500, -1]);  permute_297 = None
	        amin_177 = torch.ops.aten.amin.default(view_2774, [2])
	        amax_177 = torch.ops.aten.amax.default(view_2774, [2])
	        full_354 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_177 = torch.ops.aten.minimum.default(amin_177, full_354);  amin_177 = full_354 = None
	        full_355 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_177 = torch.ops.aten.maximum.default(amax_177, full_355);  amax_177 = full_355 = None
	        sub_8101 = torch.ops.aten.sub.Tensor(maximum_177, minimum_177);  maximum_177 = None
	        div_354 = torch.ops.aten.div.Tensor(sub_8101, 255.0);  sub_8101 = None
	        clamp_min_531 = torch.ops.aten.clamp_min.default(div_354, 1.1920928955078125e-07);  div_354 = None
	        div_355 = torch.ops.aten.div.Tensor(minimum_177, clamp_min_531);  minimum_177 = None
	        round_355 = torch.ops.aten.round.default(div_355);  div_355 = None
	        sub_8107 = torch.ops.aten.sub.Tensor(-128, round_355);  round_355 = None
	        clamp_min_532 = torch.ops.aten.clamp_min.default(sub_8107, -128);  sub_8107 = None
	        clamp_max_354 = torch.ops.aten.clamp_max.default(clamp_min_532, 127);  clamp_min_532 = None
	        _assert_tensor_metadata_1595 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_531, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1595 = None
	        _assert_tensor_metadata_1596 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_354, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1596 = None
	        convert_element_type_1062 = torch.ops.prims.convert_element_type.default(clamp_max_354, torch.int8);  clamp_max_354 = None
	        view_2777 = torch.ops.aten.view.default(clamp_min_531, [sym_size_int, 1500, 1])
	        view_2778 = torch.ops.aten.view.default(convert_element_type_1062, [sym_size_int, 1500, 1])
	        reciprocal_177 = torch.ops.aten.reciprocal.default(view_2777);  view_2777 = None
	        mul_17189 = torch.ops.aten.mul.Tensor(reciprocal_177, 1.0);  reciprocal_177 = None
	        mul_17192 = torch.ops.aten.mul.Tensor(view_2774, mul_17189);  view_2774 = mul_17189 = None
	        round_356 = torch.ops.aten.round.default(mul_17192);  mul_17192 = None
	        add_27205 = torch.ops.aten.add.Tensor(round_356, view_2778);  round_356 = view_2778 = None
	        clamp_min_533 = torch.ops.aten.clamp_min.default(add_27205, -128);  add_27205 = None
	        clamp_max_355 = torch.ops.aten.clamp_max.default(clamp_min_533, 127);  clamp_min_533 = None
	        _assert_tensor_metadata_1597 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_355, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1597 = None
	        convert_element_type_1063 = torch.ops.prims.convert_element_type.default(clamp_max_355, torch.int8);  clamp_max_355 = None
	        view_2781 = torch.ops.aten.view.default(clamp_min_531, [sym_size_int, 1500, 1]);  clamp_min_531 = None
	        view_2782 = torch.ops.aten.view.default(convert_element_type_1062, [sym_size_int, 1500, 1]);  convert_element_type_1062 = None
	        _assert_tensor_metadata_1598 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1063, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1598 = None
	        convert_element_type_1064 = torch.ops.prims.convert_element_type.default(convert_element_type_1063, torch.float32);  convert_element_type_1063 = None
	        _assert_tensor_metadata_1599 = torch.ops.aten._assert_tensor_metadata.default(view_2782, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1599 = None
	        convert_element_type_1065 = torch.ops.prims.convert_element_type.default(view_2782, torch.float32);  view_2782 = None
	        sub_8127 = torch.ops.aten.sub.Tensor(convert_element_type_1064, convert_element_type_1065);  convert_element_type_1064 = convert_element_type_1065 = None
	        mul_17214 = torch.ops.aten.mul.Tensor(sub_8127, view_2781);  sub_8127 = view_2781 = None
	        _assert_tensor_metadata_1600 = torch.ops.aten._assert_tensor_metadata.default(mul_17214, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1600 = None
	        view_2784 = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2785 = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2786 = torch.ops.aten.view.default(model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1601 = torch.ops.aten._assert_tensor_metadata.default(view_2784, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1601 = None
	        convert_element_type_1066 = torch.ops.prims.convert_element_type.default(view_2784, torch.float32);  view_2784 = None
	        _assert_tensor_metadata_1602 = torch.ops.aten._assert_tensor_metadata.default(view_2786, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1602 = None
	        convert_element_type_1067 = torch.ops.prims.convert_element_type.default(view_2786, torch.float32);  view_2786 = None
	        sub_8131 = torch.ops.aten.sub.Tensor(convert_element_type_1066, convert_element_type_1067);  convert_element_type_1066 = convert_element_type_1067 = None
	        mul_17219 = torch.ops.aten.mul.Tensor(sub_8131, view_2785);  sub_8131 = view_2785 = None
	        view_2787 = torch.ops.aten.view.default(mul_17219, [1280, 1280]);  mul_17219 = None
	        _assert_tensor_metadata_1603 = torch.ops.aten._assert_tensor_metadata.default(view_2787, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1603 = None
	        mul_17224 = sym_size_int * 1500
	        view_2788 = torch.ops.aten.view.default(mul_17214, [mul_17224, 1280]);  mul_17214 = mul_17224 = None
	        permute_298 = torch.ops.aten.permute.default(view_2787, [1, 0]);  view_2787 = None
	        addmm_147 = torch.ops.aten.addmm.default(model_audio_tower_layers_29_self_attn_out_proj_bias, view_2788, permute_298);  model_audio_tower_layers_29_self_attn_out_proj_bias = view_2788 = permute_298 = None
	        view_2789 = torch.ops.aten.view.default(addmm_147, [sym_size_int, 1500, 1280]);  addmm_147 = None
	        add_27268 = torch.ops.aten.add.Tensor(add_26648, view_2789);  add_26648 = view_2789 = None
	        clone_238 = torch.ops.aten.clone.default(add_27268, memory_format = torch.contiguous_format)
	        var_mean_59 = torch.ops.aten.var_mean.correction(clone_238, [2], correction = 0, keepdim = True)
	        getitem_238 = var_mean_59[0]
	        getitem_239 = var_mean_59[1];  var_mean_59 = None
	        add_27273 = torch.ops.aten.add.Tensor(getitem_238, 1e-05);  getitem_238 = None
	        rsqrt_59 = torch.ops.aten.rsqrt.default(add_27273);  add_27273 = None
	        sub_8137 = torch.ops.aten.sub.Tensor(clone_238, getitem_239);  clone_238 = getitem_239 = None
	        mul_17235 = torch.ops.aten.mul.Tensor(sub_8137, rsqrt_59);  sub_8137 = rsqrt_59 = None
	        mul_17236 = torch.ops.aten.mul.Tensor(mul_17235, model_audio_tower_layers_29_final_layer_norm_weight);  mul_17235 = model_audio_tower_layers_29_final_layer_norm_weight = None
	        add_27274 = torch.ops.aten.add.Tensor(mul_17236, model_audio_tower_layers_29_final_layer_norm_bias);  mul_17236 = model_audio_tower_layers_29_final_layer_norm_bias = None
	        amin_178 = torch.ops.aten.amin.default(add_27274, [2])
	        amax_178 = torch.ops.aten.amax.default(add_27274, [2])
	        full_356 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_178 = torch.ops.aten.minimum.default(amin_178, full_356);  amin_178 = full_356 = None
	        full_357 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_178 = torch.ops.aten.maximum.default(amax_178, full_357);  amax_178 = full_357 = None
	        sub_8148 = torch.ops.aten.sub.Tensor(maximum_178, minimum_178);  maximum_178 = None
	        div_356 = torch.ops.aten.div.Tensor(sub_8148, 255.0);  sub_8148 = None
	        clamp_min_534 = torch.ops.aten.clamp_min.default(div_356, 1.1920928955078125e-07);  div_356 = None
	        div_357 = torch.ops.aten.div.Tensor(minimum_178, clamp_min_534);  minimum_178 = None
	        round_357 = torch.ops.aten.round.default(div_357);  div_357 = None
	        sub_8154 = torch.ops.aten.sub.Tensor(-128, round_357);  round_357 = None
	        clamp_min_535 = torch.ops.aten.clamp_min.default(sub_8154, -128);  sub_8154 = None
	        clamp_max_356 = torch.ops.aten.clamp_max.default(clamp_min_535, 127);  clamp_min_535 = None
	        _assert_tensor_metadata_1604 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_534, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1604 = None
	        _assert_tensor_metadata_1605 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_356, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1605 = None
	        convert_element_type_1068 = torch.ops.prims.convert_element_type.default(clamp_max_356, torch.int8);  clamp_max_356 = None
	        view_2792 = torch.ops.aten.view.default(clamp_min_534, [sym_size_int, 1500, 1])
	        view_2793 = torch.ops.aten.view.default(convert_element_type_1068, [sym_size_int, 1500, 1])
	        reciprocal_178 = torch.ops.aten.reciprocal.default(view_2792);  view_2792 = None
	        mul_17284 = torch.ops.aten.mul.Tensor(reciprocal_178, 1.0);  reciprocal_178 = None
	        mul_17287 = torch.ops.aten.mul.Tensor(add_27274, mul_17284);  add_27274 = mul_17284 = None
	        round_358 = torch.ops.aten.round.default(mul_17287);  mul_17287 = None
	        add_27361 = torch.ops.aten.add.Tensor(round_358, view_2793);  round_358 = view_2793 = None
	        clamp_min_536 = torch.ops.aten.clamp_min.default(add_27361, -128);  add_27361 = None
	        clamp_max_357 = torch.ops.aten.clamp_max.default(clamp_min_536, 127);  clamp_min_536 = None
	        _assert_tensor_metadata_1606 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_357, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1606 = None
	        convert_element_type_1069 = torch.ops.prims.convert_element_type.default(clamp_max_357, torch.int8);  clamp_max_357 = None
	        view_2796 = torch.ops.aten.view.default(clamp_min_534, [sym_size_int, 1500, 1]);  clamp_min_534 = None
	        view_2797 = torch.ops.aten.view.default(convert_element_type_1068, [sym_size_int, 1500, 1]);  convert_element_type_1068 = None
	        _assert_tensor_metadata_1607 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1069, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1607 = None
	        convert_element_type_1070 = torch.ops.prims.convert_element_type.default(convert_element_type_1069, torch.float32);  convert_element_type_1069 = None
	        _assert_tensor_metadata_1608 = torch.ops.aten._assert_tensor_metadata.default(view_2797, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1608 = None
	        convert_element_type_1071 = torch.ops.prims.convert_element_type.default(view_2797, torch.float32);  view_2797 = None
	        sub_8174 = torch.ops.aten.sub.Tensor(convert_element_type_1070, convert_element_type_1071);  convert_element_type_1070 = convert_element_type_1071 = None
	        mul_17309 = torch.ops.aten.mul.Tensor(sub_8174, view_2796);  sub_8174 = view_2796 = None
	        _assert_tensor_metadata_1609 = torch.ops.aten._assert_tensor_metadata.default(mul_17309, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1609 = None
	        view_2799 = torch.ops.aten.view.default(model_audio_tower_layers_29_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_29_fc1_parametrizations_weight_original0 = None
	        view_2800 = torch.ops.aten.view.default(model_audio_tower_layers_29_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_29_fc1_parametrizations_weight_original1 = None
	        view_2801 = torch.ops.aten.view.default(model_audio_tower_layers_29_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_29_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1610 = torch.ops.aten._assert_tensor_metadata.default(view_2799, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1610 = None
	        convert_element_type_1072 = torch.ops.prims.convert_element_type.default(view_2799, torch.float32);  view_2799 = None
	        _assert_tensor_metadata_1611 = torch.ops.aten._assert_tensor_metadata.default(view_2801, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1611 = None
	        convert_element_type_1073 = torch.ops.prims.convert_element_type.default(view_2801, torch.float32);  view_2801 = None
	        sub_8178 = torch.ops.aten.sub.Tensor(convert_element_type_1072, convert_element_type_1073);  convert_element_type_1072 = convert_element_type_1073 = None
	        mul_17314 = torch.ops.aten.mul.Tensor(sub_8178, view_2800);  sub_8178 = view_2800 = None
	        view_2802 = torch.ops.aten.view.default(mul_17314, [5120, 1280]);  mul_17314 = None
	        _assert_tensor_metadata_1612 = torch.ops.aten._assert_tensor_metadata.default(view_2802, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1612 = None
	        mul_17319 = sym_size_int * 1500
	        view_2803 = torch.ops.aten.view.default(mul_17309, [mul_17319, 1280]);  mul_17309 = mul_17319 = None
	        permute_299 = torch.ops.aten.permute.default(view_2802, [1, 0]);  view_2802 = None
	        addmm_148 = torch.ops.aten.addmm.default(model_audio_tower_layers_29_fc1_bias, view_2803, permute_299);  model_audio_tower_layers_29_fc1_bias = view_2803 = permute_299 = None
	        view_2804 = torch.ops.aten.view.default(addmm_148, [sym_size_int, 1500, 5120]);  addmm_148 = None
	        mul_17326 = torch.ops.aten.mul.Tensor(view_2804, 0.5)
	        mul_17327 = torch.ops.aten.mul.Tensor(view_2804, 0.7071067811865476);  view_2804 = None
	        erf_31 = torch.ops.aten.erf.default(mul_17327);  mul_17327 = None
	        add_27420 = torch.ops.aten.add.Tensor(erf_31, 1);  erf_31 = None
	        mul_17328 = torch.ops.aten.mul.Tensor(mul_17326, add_27420);  mul_17326 = add_27420 = None
	        amin_179 = torch.ops.aten.amin.default(mul_17328, [2])
	        amax_179 = torch.ops.aten.amax.default(mul_17328, [2])
	        full_358 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_179 = torch.ops.aten.minimum.default(amin_179, full_358);  amin_179 = full_358 = None
	        full_359 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_179 = torch.ops.aten.maximum.default(amax_179, full_359);  amax_179 = full_359 = None
	        sub_8191 = torch.ops.aten.sub.Tensor(maximum_179, minimum_179);  maximum_179 = None
	        div_358 = torch.ops.aten.div.Tensor(sub_8191, 255.0);  sub_8191 = None
	        clamp_min_537 = torch.ops.aten.clamp_min.default(div_358, 1.1920928955078125e-07);  div_358 = None
	        div_359 = torch.ops.aten.div.Tensor(minimum_179, clamp_min_537);  minimum_179 = None
	        round_359 = torch.ops.aten.round.default(div_359);  div_359 = None
	        sub_8197 = torch.ops.aten.sub.Tensor(-128, round_359);  round_359 = None
	        clamp_min_538 = torch.ops.aten.clamp_min.default(sub_8197, -128);  sub_8197 = None
	        clamp_max_358 = torch.ops.aten.clamp_max.default(clamp_min_538, 127);  clamp_min_538 = None
	        _assert_tensor_metadata_1613 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_537, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1613 = None
	        _assert_tensor_metadata_1614 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_358, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1614 = None
	        convert_element_type_1074 = torch.ops.prims.convert_element_type.default(clamp_max_358, torch.int8);  clamp_max_358 = None
	        view_2807 = torch.ops.aten.view.default(clamp_min_537, [sym_size_int, 1500, 1])
	        view_2808 = torch.ops.aten.view.default(convert_element_type_1074, [sym_size_int, 1500, 1])
	        reciprocal_179 = torch.ops.aten.reciprocal.default(view_2807);  view_2807 = None
	        mul_17374 = torch.ops.aten.mul.Tensor(reciprocal_179, 1.0);  reciprocal_179 = None
	        mul_17377 = torch.ops.aten.mul.Tensor(mul_17328, mul_17374);  mul_17328 = mul_17374 = None
	        round_360 = torch.ops.aten.round.default(mul_17377);  mul_17377 = None
	        add_27503 = torch.ops.aten.add.Tensor(round_360, view_2808);  round_360 = view_2808 = None
	        clamp_min_539 = torch.ops.aten.clamp_min.default(add_27503, -128);  add_27503 = None
	        clamp_max_359 = torch.ops.aten.clamp_max.default(clamp_min_539, 127);  clamp_min_539 = None
	        _assert_tensor_metadata_1615 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_359, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1615 = None
	        convert_element_type_1075 = torch.ops.prims.convert_element_type.default(clamp_max_359, torch.int8);  clamp_max_359 = None
	        view_2811 = torch.ops.aten.view.default(clamp_min_537, [sym_size_int, 1500, 1]);  clamp_min_537 = None
	        view_2812 = torch.ops.aten.view.default(convert_element_type_1074, [sym_size_int, 1500, 1]);  convert_element_type_1074 = None
	        _assert_tensor_metadata_1616 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1075, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1616 = None
	        convert_element_type_1076 = torch.ops.prims.convert_element_type.default(convert_element_type_1075, torch.float32);  convert_element_type_1075 = None
	        _assert_tensor_metadata_1617 = torch.ops.aten._assert_tensor_metadata.default(view_2812, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1617 = None
	        convert_element_type_1077 = torch.ops.prims.convert_element_type.default(view_2812, torch.float32);  view_2812 = None
	        sub_8217 = torch.ops.aten.sub.Tensor(convert_element_type_1076, convert_element_type_1077);  convert_element_type_1076 = convert_element_type_1077 = None
	        mul_17399 = torch.ops.aten.mul.Tensor(sub_8217, view_2811);  sub_8217 = view_2811 = None
	        _assert_tensor_metadata_1618 = torch.ops.aten._assert_tensor_metadata.default(mul_17399, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1618 = None
	        view_2814 = torch.ops.aten.view.default(model_audio_tower_layers_29_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_29_fc2_parametrizations_weight_original0 = None
	        view_2815 = torch.ops.aten.view.default(model_audio_tower_layers_29_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_29_fc2_parametrizations_weight_original1 = None
	        view_2816 = torch.ops.aten.view.default(model_audio_tower_layers_29_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_29_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1619 = torch.ops.aten._assert_tensor_metadata.default(view_2814, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1619 = None
	        convert_element_type_1078 = torch.ops.prims.convert_element_type.default(view_2814, torch.float32);  view_2814 = None
	        _assert_tensor_metadata_1620 = torch.ops.aten._assert_tensor_metadata.default(view_2816, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1620 = None
	        convert_element_type_1079 = torch.ops.prims.convert_element_type.default(view_2816, torch.float32);  view_2816 = None
	        sub_8221 = torch.ops.aten.sub.Tensor(convert_element_type_1078, convert_element_type_1079);  convert_element_type_1078 = convert_element_type_1079 = None
	        mul_17404 = torch.ops.aten.mul.Tensor(sub_8221, view_2815);  sub_8221 = view_2815 = None
	        view_2817 = torch.ops.aten.view.default(mul_17404, [1280, 5120]);  mul_17404 = None
	        _assert_tensor_metadata_1621 = torch.ops.aten._assert_tensor_metadata.default(view_2817, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1621 = None
	        mul_17409 = sym_size_int * 1500
	        view_2818 = torch.ops.aten.view.default(mul_17399, [mul_17409, 5120]);  mul_17399 = mul_17409 = None
	        permute_300 = torch.ops.aten.permute.default(view_2817, [1, 0]);  view_2817 = None
	        addmm_149 = torch.ops.aten.addmm.default(model_audio_tower_layers_29_fc2_bias, view_2818, permute_300);  model_audio_tower_layers_29_fc2_bias = view_2818 = permute_300 = None
	        view_2819 = torch.ops.aten.view.default(addmm_149, [sym_size_int, 1500, 1280]);  addmm_149 = None
	        add_27566 = torch.ops.aten.add.Tensor(add_27268, view_2819);  add_27268 = view_2819 = None
	        clone_241 = torch.ops.aten.clone.default(add_27566, memory_format = torch.contiguous_format)
	        var_mean_60 = torch.ops.aten.var_mean.correction(clone_241, [2], correction = 0, keepdim = True)
	        getitem_240 = var_mean_60[0]
	        getitem_241 = var_mean_60[1];  var_mean_60 = None
	        add_27571 = torch.ops.aten.add.Tensor(getitem_240, 1e-05);  getitem_240 = None
	        rsqrt_60 = torch.ops.aten.rsqrt.default(add_27571);  add_27571 = None
	        sub_8227 = torch.ops.aten.sub.Tensor(clone_241, getitem_241);  clone_241 = getitem_241 = None
	        mul_17420 = torch.ops.aten.mul.Tensor(sub_8227, rsqrt_60);  sub_8227 = rsqrt_60 = None
	        mul_17421 = torch.ops.aten.mul.Tensor(mul_17420, model_audio_tower_layers_30_self_attn_layer_norm_weight);  mul_17420 = model_audio_tower_layers_30_self_attn_layer_norm_weight = None
	        add_27572 = torch.ops.aten.add.Tensor(mul_17421, model_audio_tower_layers_30_self_attn_layer_norm_bias);  mul_17421 = model_audio_tower_layers_30_self_attn_layer_norm_bias = None
	        amin_180 = torch.ops.aten.amin.default(add_27572, [2])
	        amax_180 = torch.ops.aten.amax.default(add_27572, [2])
	        full_360 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_180 = torch.ops.aten.minimum.default(amin_180, full_360);  amin_180 = full_360 = None
	        full_361 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_180 = torch.ops.aten.maximum.default(amax_180, full_361);  amax_180 = full_361 = None
	        sub_8238 = torch.ops.aten.sub.Tensor(maximum_180, minimum_180);  maximum_180 = None
	        div_360 = torch.ops.aten.div.Tensor(sub_8238, 255.0);  sub_8238 = None
	        clamp_min_540 = torch.ops.aten.clamp_min.default(div_360, 1.1920928955078125e-07);  div_360 = None
	        div_361 = torch.ops.aten.div.Tensor(minimum_180, clamp_min_540);  minimum_180 = None
	        round_361 = torch.ops.aten.round.default(div_361);  div_361 = None
	        sub_8244 = torch.ops.aten.sub.Tensor(-128, round_361);  round_361 = None
	        clamp_min_541 = torch.ops.aten.clamp_min.default(sub_8244, -128);  sub_8244 = None
	        clamp_max_360 = torch.ops.aten.clamp_max.default(clamp_min_541, 127);  clamp_min_541 = None
	        _assert_tensor_metadata_1622 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_540, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1622 = None
	        _assert_tensor_metadata_1623 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_360, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1623 = None
	        convert_element_type_1080 = torch.ops.prims.convert_element_type.default(clamp_max_360, torch.int8);  clamp_max_360 = None
	        view_2822 = torch.ops.aten.view.default(clamp_min_540, [sym_size_int, 1500, 1])
	        view_2823 = torch.ops.aten.view.default(convert_element_type_1080, [sym_size_int, 1500, 1])
	        reciprocal_180 = torch.ops.aten.reciprocal.default(view_2822);  view_2822 = None
	        mul_17469 = torch.ops.aten.mul.Tensor(reciprocal_180, 1.0);  reciprocal_180 = None
	        mul_17472 = torch.ops.aten.mul.Tensor(add_27572, mul_17469);  mul_17469 = None
	        round_362 = torch.ops.aten.round.default(mul_17472);  mul_17472 = None
	        add_27659 = torch.ops.aten.add.Tensor(round_362, view_2823);  round_362 = view_2823 = None
	        clamp_min_542 = torch.ops.aten.clamp_min.default(add_27659, -128);  add_27659 = None
	        clamp_max_361 = torch.ops.aten.clamp_max.default(clamp_min_542, 127);  clamp_min_542 = None
	        _assert_tensor_metadata_1624 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_361, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1624 = None
	        convert_element_type_1081 = torch.ops.prims.convert_element_type.default(clamp_max_361, torch.int8);  clamp_max_361 = None
	        view_2826 = torch.ops.aten.view.default(clamp_min_540, [sym_size_int, 1500, 1]);  clamp_min_540 = None
	        view_2827 = torch.ops.aten.view.default(convert_element_type_1080, [sym_size_int, 1500, 1]);  convert_element_type_1080 = None
	        _assert_tensor_metadata_1625 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1081, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1625 = None
	        convert_element_type_1082 = torch.ops.prims.convert_element_type.default(convert_element_type_1081, torch.float32);  convert_element_type_1081 = None
	        _assert_tensor_metadata_1626 = torch.ops.aten._assert_tensor_metadata.default(view_2827, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1626 = None
	        convert_element_type_1083 = torch.ops.prims.convert_element_type.default(view_2827, torch.float32);  view_2827 = None
	        sub_8264 = torch.ops.aten.sub.Tensor(convert_element_type_1082, convert_element_type_1083);  convert_element_type_1082 = convert_element_type_1083 = None
	        mul_17494 = torch.ops.aten.mul.Tensor(sub_8264, view_2826);  sub_8264 = view_2826 = None
	        _assert_tensor_metadata_1627 = torch.ops.aten._assert_tensor_metadata.default(mul_17494, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1627 = None
	        view_2829 = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2830 = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2831 = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1628 = torch.ops.aten._assert_tensor_metadata.default(view_2829, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1628 = None
	        convert_element_type_1084 = torch.ops.prims.convert_element_type.default(view_2829, torch.float32);  view_2829 = None
	        _assert_tensor_metadata_1629 = torch.ops.aten._assert_tensor_metadata.default(view_2831, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1629 = None
	        convert_element_type_1085 = torch.ops.prims.convert_element_type.default(view_2831, torch.float32);  view_2831 = None
	        sub_8268 = torch.ops.aten.sub.Tensor(convert_element_type_1084, convert_element_type_1085);  convert_element_type_1084 = convert_element_type_1085 = None
	        mul_17499 = torch.ops.aten.mul.Tensor(sub_8268, view_2830);  sub_8268 = view_2830 = None
	        view_2832 = torch.ops.aten.view.default(mul_17499, [1280, 1280]);  mul_17499 = None
	        _assert_tensor_metadata_1630 = torch.ops.aten._assert_tensor_metadata.default(view_2832, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1630 = None
	        mul_17504 = sym_size_int * 1500
	        view_2833 = torch.ops.aten.view.default(mul_17494, [mul_17504, 1280]);  mul_17494 = mul_17504 = None
	        permute_301 = torch.ops.aten.permute.default(view_2832, [1, 0]);  view_2832 = None
	        addmm_150 = torch.ops.aten.addmm.default(model_audio_tower_layers_30_self_attn_q_proj_bias, view_2833, permute_301);  model_audio_tower_layers_30_self_attn_q_proj_bias = view_2833 = permute_301 = None
	        view_2834 = torch.ops.aten.view.default(addmm_150, [sym_size_int, 1500, 1280]);  addmm_150 = None
	        mul_17511 = torch.ops.aten.mul.Tensor(view_2834, 0.125);  view_2834 = None
	        view_2835 = torch.ops.aten.view.default(mul_17511, [sym_size_int, 1500, 20, 64]);  mul_17511 = None
	        permute_302 = torch.ops.aten.permute.default(view_2835, [0, 2, 1, 3]);  view_2835 = None
	        clone_242 = torch.ops.aten.clone.default(permute_302, memory_format = torch.contiguous_format);  permute_302 = None
	        amin_181 = torch.ops.aten.amin.default(add_27572, [2])
	        amax_181 = torch.ops.aten.amax.default(add_27572, [2])
	        full_362 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_181 = torch.ops.aten.minimum.default(amin_181, full_362);  amin_181 = full_362 = None
	        full_363 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_181 = torch.ops.aten.maximum.default(amax_181, full_363);  amax_181 = full_363 = None
	        sub_8283 = torch.ops.aten.sub.Tensor(maximum_181, minimum_181);  maximum_181 = None
	        div_362 = torch.ops.aten.div.Tensor(sub_8283, 255.0);  sub_8283 = None
	        clamp_min_543 = torch.ops.aten.clamp_min.default(div_362, 1.1920928955078125e-07);  div_362 = None
	        div_363 = torch.ops.aten.div.Tensor(minimum_181, clamp_min_543);  minimum_181 = None
	        round_363 = torch.ops.aten.round.default(div_363);  div_363 = None
	        sub_8289 = torch.ops.aten.sub.Tensor(-128, round_363);  round_363 = None
	        clamp_min_544 = torch.ops.aten.clamp_min.default(sub_8289, -128);  sub_8289 = None
	        clamp_max_362 = torch.ops.aten.clamp_max.default(clamp_min_544, 127);  clamp_min_544 = None
	        _assert_tensor_metadata_1631 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_543, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1631 = None
	        _assert_tensor_metadata_1632 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_362, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1632 = None
	        convert_element_type_1086 = torch.ops.prims.convert_element_type.default(clamp_max_362, torch.int8);  clamp_max_362 = None
	        view_2838 = torch.ops.aten.view.default(clamp_min_543, [sym_size_int, 1500, 1])
	        view_2839 = torch.ops.aten.view.default(convert_element_type_1086, [sym_size_int, 1500, 1])
	        reciprocal_181 = torch.ops.aten.reciprocal.default(view_2838);  view_2838 = None
	        mul_17565 = torch.ops.aten.mul.Tensor(reciprocal_181, 1.0);  reciprocal_181 = None
	        mul_17568 = torch.ops.aten.mul.Tensor(add_27572, mul_17565);  mul_17565 = None
	        round_364 = torch.ops.aten.round.default(mul_17568);  mul_17568 = None
	        add_27811 = torch.ops.aten.add.Tensor(round_364, view_2839);  round_364 = view_2839 = None
	        clamp_min_545 = torch.ops.aten.clamp_min.default(add_27811, -128);  add_27811 = None
	        clamp_max_363 = torch.ops.aten.clamp_max.default(clamp_min_545, 127);  clamp_min_545 = None
	        _assert_tensor_metadata_1633 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_363, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1633 = None
	        convert_element_type_1087 = torch.ops.prims.convert_element_type.default(clamp_max_363, torch.int8);  clamp_max_363 = None
	        view_2842 = torch.ops.aten.view.default(clamp_min_543, [sym_size_int, 1500, 1]);  clamp_min_543 = None
	        view_2843 = torch.ops.aten.view.default(convert_element_type_1086, [sym_size_int, 1500, 1]);  convert_element_type_1086 = None
	        _assert_tensor_metadata_1634 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1087, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1634 = None
	        convert_element_type_1088 = torch.ops.prims.convert_element_type.default(convert_element_type_1087, torch.float32);  convert_element_type_1087 = None
	        _assert_tensor_metadata_1635 = torch.ops.aten._assert_tensor_metadata.default(view_2843, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1635 = None
	        convert_element_type_1089 = torch.ops.prims.convert_element_type.default(view_2843, torch.float32);  view_2843 = None
	        sub_8309 = torch.ops.aten.sub.Tensor(convert_element_type_1088, convert_element_type_1089);  convert_element_type_1088 = convert_element_type_1089 = None
	        mul_17590 = torch.ops.aten.mul.Tensor(sub_8309, view_2842);  sub_8309 = view_2842 = None
	        _assert_tensor_metadata_1636 = torch.ops.aten._assert_tensor_metadata.default(mul_17590, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1636 = None
	        view_2845 = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2846 = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2847 = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1637 = torch.ops.aten._assert_tensor_metadata.default(view_2845, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1637 = None
	        convert_element_type_1090 = torch.ops.prims.convert_element_type.default(view_2845, torch.float32);  view_2845 = None
	        _assert_tensor_metadata_1638 = torch.ops.aten._assert_tensor_metadata.default(view_2847, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1638 = None
	        convert_element_type_1091 = torch.ops.prims.convert_element_type.default(view_2847, torch.float32);  view_2847 = None
	        sub_8313 = torch.ops.aten.sub.Tensor(convert_element_type_1090, convert_element_type_1091);  convert_element_type_1090 = convert_element_type_1091 = None
	        mul_17595 = torch.ops.aten.mul.Tensor(sub_8313, view_2846);  sub_8313 = view_2846 = None
	        view_2848 = torch.ops.aten.view.default(mul_17595, [1280, 1280]);  mul_17595 = None
	        _assert_tensor_metadata_1639 = torch.ops.aten._assert_tensor_metadata.default(view_2848, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1639 = None
	        permute_303 = torch.ops.aten.permute.default(view_2848, [1, 0]);  view_2848 = None
	        mul_17598 = sym_size_int * 1500
	        view_2849 = torch.ops.aten.view.default(mul_17590, [mul_17598, 1280]);  mul_17590 = mul_17598 = None
	        mm_30 = torch.ops.aten.mm.default(view_2849, permute_303);  view_2849 = permute_303 = None
	        view_2850 = torch.ops.aten.view.default(mm_30, [sym_size_int, 1500, 1280]);  mm_30 = None
	        view_2851 = torch.ops.aten.view.default(view_2850, [sym_size_int, -1, 20, 64]);  view_2850 = None
	        permute_304 = torch.ops.aten.permute.default(view_2851, [0, 2, 1, 3]);  view_2851 = None
	        clone_243 = torch.ops.aten.clone.default(permute_304, memory_format = torch.contiguous_format);  permute_304 = None
	        amin_182 = torch.ops.aten.amin.default(add_27572, [2])
	        amax_182 = torch.ops.aten.amax.default(add_27572, [2])
	        full_364 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_182 = torch.ops.aten.minimum.default(amin_182, full_364);  amin_182 = full_364 = None
	        full_365 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_182 = torch.ops.aten.maximum.default(amax_182, full_365);  amax_182 = full_365 = None
	        sub_8327 = torch.ops.aten.sub.Tensor(maximum_182, minimum_182);  maximum_182 = None
	        div_364 = torch.ops.aten.div.Tensor(sub_8327, 255.0);  sub_8327 = None
	        clamp_min_546 = torch.ops.aten.clamp_min.default(div_364, 1.1920928955078125e-07);  div_364 = None
	        div_365 = torch.ops.aten.div.Tensor(minimum_182, clamp_min_546);  minimum_182 = None
	        round_365 = torch.ops.aten.round.default(div_365);  div_365 = None
	        sub_8333 = torch.ops.aten.sub.Tensor(-128, round_365);  round_365 = None
	        clamp_min_547 = torch.ops.aten.clamp_min.default(sub_8333, -128);  sub_8333 = None
	        clamp_max_364 = torch.ops.aten.clamp_max.default(clamp_min_547, 127);  clamp_min_547 = None
	        _assert_tensor_metadata_1640 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_546, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1640 = None
	        _assert_tensor_metadata_1641 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_364, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1641 = None
	        convert_element_type_1092 = torch.ops.prims.convert_element_type.default(clamp_max_364, torch.int8);  clamp_max_364 = None
	        view_2854 = torch.ops.aten.view.default(clamp_min_546, [sym_size_int, 1500, 1])
	        view_2855 = torch.ops.aten.view.default(convert_element_type_1092, [sym_size_int, 1500, 1])
	        reciprocal_182 = torch.ops.aten.reciprocal.default(view_2854);  view_2854 = None
	        mul_17664 = torch.ops.aten.mul.Tensor(reciprocal_182, 1.0);  reciprocal_182 = None
	        mul_17667 = torch.ops.aten.mul.Tensor(add_27572, mul_17664);  add_27572 = mul_17664 = None
	        round_366 = torch.ops.aten.round.default(mul_17667);  mul_17667 = None
	        add_27959 = torch.ops.aten.add.Tensor(round_366, view_2855);  round_366 = view_2855 = None
	        clamp_min_548 = torch.ops.aten.clamp_min.default(add_27959, -128);  add_27959 = None
	        clamp_max_365 = torch.ops.aten.clamp_max.default(clamp_min_548, 127);  clamp_min_548 = None
	        _assert_tensor_metadata_1642 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_365, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1642 = None
	        convert_element_type_1093 = torch.ops.prims.convert_element_type.default(clamp_max_365, torch.int8);  clamp_max_365 = None
	        view_2858 = torch.ops.aten.view.default(clamp_min_546, [sym_size_int, 1500, 1]);  clamp_min_546 = None
	        view_2859 = torch.ops.aten.view.default(convert_element_type_1092, [sym_size_int, 1500, 1]);  convert_element_type_1092 = None
	        _assert_tensor_metadata_1643 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1093, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1643 = None
	        convert_element_type_1094 = torch.ops.prims.convert_element_type.default(convert_element_type_1093, torch.float32);  convert_element_type_1093 = None
	        _assert_tensor_metadata_1644 = torch.ops.aten._assert_tensor_metadata.default(view_2859, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1644 = None
	        convert_element_type_1095 = torch.ops.prims.convert_element_type.default(view_2859, torch.float32);  view_2859 = None
	        sub_8353 = torch.ops.aten.sub.Tensor(convert_element_type_1094, convert_element_type_1095);  convert_element_type_1094 = convert_element_type_1095 = None
	        mul_17689 = torch.ops.aten.mul.Tensor(sub_8353, view_2858);  sub_8353 = view_2858 = None
	        _assert_tensor_metadata_1645 = torch.ops.aten._assert_tensor_metadata.default(mul_17689, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1645 = None
	        view_2861 = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2862 = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2863 = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1646 = torch.ops.aten._assert_tensor_metadata.default(view_2861, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1646 = None
	        convert_element_type_1096 = torch.ops.prims.convert_element_type.default(view_2861, torch.float32);  view_2861 = None
	        _assert_tensor_metadata_1647 = torch.ops.aten._assert_tensor_metadata.default(view_2863, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1647 = None
	        convert_element_type_1097 = torch.ops.prims.convert_element_type.default(view_2863, torch.float32);  view_2863 = None
	        sub_8357 = torch.ops.aten.sub.Tensor(convert_element_type_1096, convert_element_type_1097);  convert_element_type_1096 = convert_element_type_1097 = None
	        mul_17694 = torch.ops.aten.mul.Tensor(sub_8357, view_2862);  sub_8357 = view_2862 = None
	        view_2864 = torch.ops.aten.view.default(mul_17694, [1280, 1280]);  mul_17694 = None
	        _assert_tensor_metadata_1648 = torch.ops.aten._assert_tensor_metadata.default(view_2864, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1648 = None
	        mul_17699 = sym_size_int * 1500
	        view_2865 = torch.ops.aten.view.default(mul_17689, [mul_17699, 1280]);  mul_17689 = mul_17699 = None
	        permute_305 = torch.ops.aten.permute.default(view_2864, [1, 0]);  view_2864 = None
	        addmm_151 = torch.ops.aten.addmm.default(model_audio_tower_layers_30_self_attn_v_proj_bias, view_2865, permute_305);  model_audio_tower_layers_30_self_attn_v_proj_bias = view_2865 = permute_305 = None
	        view_2866 = torch.ops.aten.view.default(addmm_151, [sym_size_int, 1500, 1280]);  addmm_151 = None
	        view_2867 = torch.ops.aten.view.default(view_2866, [sym_size_int, -1, 20, 64]);  view_2866 = None
	        permute_306 = torch.ops.aten.permute.default(view_2867, [0, 2, 1, 3]);  view_2867 = None
	        clone_244 = torch.ops.aten.clone.default(permute_306, memory_format = torch.contiguous_format);  permute_306 = None
	        _scaled_dot_product_efficient_attention_30 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_242, clone_243, clone_244, None, False, scale = 1.0);  clone_242 = clone_243 = clone_244 = None
	        getitem_242 = _scaled_dot_product_efficient_attention_30[0];  _scaled_dot_product_efficient_attention_30 = None
	        permute_307 = torch.ops.aten.permute.default(getitem_242, [0, 2, 1, 3]);  getitem_242 = None
	        view_2868 = torch.ops.aten.view.default(permute_307, [sym_size_int, 1500, -1]);  permute_307 = None
	        amin_183 = torch.ops.aten.amin.default(view_2868, [2])
	        amax_183 = torch.ops.aten.amax.default(view_2868, [2])
	        full_366 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_183 = torch.ops.aten.minimum.default(amin_183, full_366);  amin_183 = full_366 = None
	        full_367 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_183 = torch.ops.aten.maximum.default(amax_183, full_367);  amax_183 = full_367 = None
	        sub_8375 = torch.ops.aten.sub.Tensor(maximum_183, minimum_183);  maximum_183 = None
	        div_366 = torch.ops.aten.div.Tensor(sub_8375, 255.0);  sub_8375 = None
	        clamp_min_549 = torch.ops.aten.clamp_min.default(div_366, 1.1920928955078125e-07);  div_366 = None
	        div_367 = torch.ops.aten.div.Tensor(minimum_183, clamp_min_549);  minimum_183 = None
	        round_367 = torch.ops.aten.round.default(div_367);  div_367 = None
	        sub_8381 = torch.ops.aten.sub.Tensor(-128, round_367);  round_367 = None
	        clamp_min_550 = torch.ops.aten.clamp_min.default(sub_8381, -128);  sub_8381 = None
	        clamp_max_366 = torch.ops.aten.clamp_max.default(clamp_min_550, 127);  clamp_min_550 = None
	        _assert_tensor_metadata_1649 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_549, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1649 = None
	        _assert_tensor_metadata_1650 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_366, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1650 = None
	        convert_element_type_1098 = torch.ops.prims.convert_element_type.default(clamp_max_366, torch.int8);  clamp_max_366 = None
	        view_2871 = torch.ops.aten.view.default(clamp_min_549, [sym_size_int, 1500, 1])
	        view_2872 = torch.ops.aten.view.default(convert_element_type_1098, [sym_size_int, 1500, 1])
	        reciprocal_183 = torch.ops.aten.reciprocal.default(view_2871);  view_2871 = None
	        mul_17769 = torch.ops.aten.mul.Tensor(reciprocal_183, 1.0);  reciprocal_183 = None
	        mul_17772 = torch.ops.aten.mul.Tensor(view_2868, mul_17769);  view_2868 = mul_17769 = None
	        round_368 = torch.ops.aten.round.default(mul_17772);  mul_17772 = None
	        add_28123 = torch.ops.aten.add.Tensor(round_368, view_2872);  round_368 = view_2872 = None
	        clamp_min_551 = torch.ops.aten.clamp_min.default(add_28123, -128);  add_28123 = None
	        clamp_max_367 = torch.ops.aten.clamp_max.default(clamp_min_551, 127);  clamp_min_551 = None
	        _assert_tensor_metadata_1651 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_367, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1651 = None
	        convert_element_type_1099 = torch.ops.prims.convert_element_type.default(clamp_max_367, torch.int8);  clamp_max_367 = None
	        view_2875 = torch.ops.aten.view.default(clamp_min_549, [sym_size_int, 1500, 1]);  clamp_min_549 = None
	        view_2876 = torch.ops.aten.view.default(convert_element_type_1098, [sym_size_int, 1500, 1]);  convert_element_type_1098 = None
	        _assert_tensor_metadata_1652 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1099, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1652 = None
	        convert_element_type_1100 = torch.ops.prims.convert_element_type.default(convert_element_type_1099, torch.float32);  convert_element_type_1099 = None
	        _assert_tensor_metadata_1653 = torch.ops.aten._assert_tensor_metadata.default(view_2876, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1653 = None
	        convert_element_type_1101 = torch.ops.prims.convert_element_type.default(view_2876, torch.float32);  view_2876 = None
	        sub_8401 = torch.ops.aten.sub.Tensor(convert_element_type_1100, convert_element_type_1101);  convert_element_type_1100 = convert_element_type_1101 = None
	        mul_17794 = torch.ops.aten.mul.Tensor(sub_8401, view_2875);  sub_8401 = view_2875 = None
	        _assert_tensor_metadata_1654 = torch.ops.aten._assert_tensor_metadata.default(mul_17794, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1654 = None
	        view_2878 = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2879 = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2880 = torch.ops.aten.view.default(model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1655 = torch.ops.aten._assert_tensor_metadata.default(view_2878, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1655 = None
	        convert_element_type_1102 = torch.ops.prims.convert_element_type.default(view_2878, torch.float32);  view_2878 = None
	        _assert_tensor_metadata_1656 = torch.ops.aten._assert_tensor_metadata.default(view_2880, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1656 = None
	        convert_element_type_1103 = torch.ops.prims.convert_element_type.default(view_2880, torch.float32);  view_2880 = None
	        sub_8405 = torch.ops.aten.sub.Tensor(convert_element_type_1102, convert_element_type_1103);  convert_element_type_1102 = convert_element_type_1103 = None
	        mul_17799 = torch.ops.aten.mul.Tensor(sub_8405, view_2879);  sub_8405 = view_2879 = None
	        view_2881 = torch.ops.aten.view.default(mul_17799, [1280, 1280]);  mul_17799 = None
	        _assert_tensor_metadata_1657 = torch.ops.aten._assert_tensor_metadata.default(view_2881, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1657 = None
	        mul_17804 = sym_size_int * 1500
	        view_2882 = torch.ops.aten.view.default(mul_17794, [mul_17804, 1280]);  mul_17794 = mul_17804 = None
	        permute_308 = torch.ops.aten.permute.default(view_2881, [1, 0]);  view_2881 = None
	        addmm_152 = torch.ops.aten.addmm.default(model_audio_tower_layers_30_self_attn_out_proj_bias, view_2882, permute_308);  model_audio_tower_layers_30_self_attn_out_proj_bias = view_2882 = permute_308 = None
	        view_2883 = torch.ops.aten.view.default(addmm_152, [sym_size_int, 1500, 1280]);  addmm_152 = None
	        add_28186 = torch.ops.aten.add.Tensor(add_27566, view_2883);  add_27566 = view_2883 = None
	        clone_246 = torch.ops.aten.clone.default(add_28186, memory_format = torch.contiguous_format)
	        var_mean_61 = torch.ops.aten.var_mean.correction(clone_246, [2], correction = 0, keepdim = True)
	        getitem_246 = var_mean_61[0]
	        getitem_247 = var_mean_61[1];  var_mean_61 = None
	        add_28191 = torch.ops.aten.add.Tensor(getitem_246, 1e-05);  getitem_246 = None
	        rsqrt_61 = torch.ops.aten.rsqrt.default(add_28191);  add_28191 = None
	        sub_8411 = torch.ops.aten.sub.Tensor(clone_246, getitem_247);  clone_246 = getitem_247 = None
	        mul_17815 = torch.ops.aten.mul.Tensor(sub_8411, rsqrt_61);  sub_8411 = rsqrt_61 = None
	        mul_17816 = torch.ops.aten.mul.Tensor(mul_17815, model_audio_tower_layers_30_final_layer_norm_weight);  mul_17815 = model_audio_tower_layers_30_final_layer_norm_weight = None
	        add_28192 = torch.ops.aten.add.Tensor(mul_17816, model_audio_tower_layers_30_final_layer_norm_bias);  mul_17816 = model_audio_tower_layers_30_final_layer_norm_bias = None
	        amin_184 = torch.ops.aten.amin.default(add_28192, [2])
	        amax_184 = torch.ops.aten.amax.default(add_28192, [2])
	        full_368 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_184 = torch.ops.aten.minimum.default(amin_184, full_368);  amin_184 = full_368 = None
	        full_369 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_184 = torch.ops.aten.maximum.default(amax_184, full_369);  amax_184 = full_369 = None
	        sub_8422 = torch.ops.aten.sub.Tensor(maximum_184, minimum_184);  maximum_184 = None
	        div_368 = torch.ops.aten.div.Tensor(sub_8422, 255.0);  sub_8422 = None
	        clamp_min_552 = torch.ops.aten.clamp_min.default(div_368, 1.1920928955078125e-07);  div_368 = None
	        div_369 = torch.ops.aten.div.Tensor(minimum_184, clamp_min_552);  minimum_184 = None
	        round_369 = torch.ops.aten.round.default(div_369);  div_369 = None
	        sub_8428 = torch.ops.aten.sub.Tensor(-128, round_369);  round_369 = None
	        clamp_min_553 = torch.ops.aten.clamp_min.default(sub_8428, -128);  sub_8428 = None
	        clamp_max_368 = torch.ops.aten.clamp_max.default(clamp_min_553, 127);  clamp_min_553 = None
	        _assert_tensor_metadata_1658 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_552, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1658 = None
	        _assert_tensor_metadata_1659 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_368, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1659 = None
	        convert_element_type_1104 = torch.ops.prims.convert_element_type.default(clamp_max_368, torch.int8);  clamp_max_368 = None
	        view_2886 = torch.ops.aten.view.default(clamp_min_552, [sym_size_int, 1500, 1])
	        view_2887 = torch.ops.aten.view.default(convert_element_type_1104, [sym_size_int, 1500, 1])
	        reciprocal_184 = torch.ops.aten.reciprocal.default(view_2886);  view_2886 = None
	        mul_17864 = torch.ops.aten.mul.Tensor(reciprocal_184, 1.0);  reciprocal_184 = None
	        mul_17867 = torch.ops.aten.mul.Tensor(add_28192, mul_17864);  add_28192 = mul_17864 = None
	        round_370 = torch.ops.aten.round.default(mul_17867);  mul_17867 = None
	        add_28279 = torch.ops.aten.add.Tensor(round_370, view_2887);  round_370 = view_2887 = None
	        clamp_min_554 = torch.ops.aten.clamp_min.default(add_28279, -128);  add_28279 = None
	        clamp_max_369 = torch.ops.aten.clamp_max.default(clamp_min_554, 127);  clamp_min_554 = None
	        _assert_tensor_metadata_1660 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_369, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1660 = None
	        convert_element_type_1105 = torch.ops.prims.convert_element_type.default(clamp_max_369, torch.int8);  clamp_max_369 = None
	        view_2890 = torch.ops.aten.view.default(clamp_min_552, [sym_size_int, 1500, 1]);  clamp_min_552 = None
	        view_2891 = torch.ops.aten.view.default(convert_element_type_1104, [sym_size_int, 1500, 1]);  convert_element_type_1104 = None
	        _assert_tensor_metadata_1661 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1105, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1661 = None
	        convert_element_type_1106 = torch.ops.prims.convert_element_type.default(convert_element_type_1105, torch.float32);  convert_element_type_1105 = None
	        _assert_tensor_metadata_1662 = torch.ops.aten._assert_tensor_metadata.default(view_2891, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1662 = None
	        convert_element_type_1107 = torch.ops.prims.convert_element_type.default(view_2891, torch.float32);  view_2891 = None
	        sub_8448 = torch.ops.aten.sub.Tensor(convert_element_type_1106, convert_element_type_1107);  convert_element_type_1106 = convert_element_type_1107 = None
	        mul_17889 = torch.ops.aten.mul.Tensor(sub_8448, view_2890);  sub_8448 = view_2890 = None
	        _assert_tensor_metadata_1663 = torch.ops.aten._assert_tensor_metadata.default(mul_17889, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1663 = None
	        view_2893 = torch.ops.aten.view.default(model_audio_tower_layers_30_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_30_fc1_parametrizations_weight_original0 = None
	        view_2894 = torch.ops.aten.view.default(model_audio_tower_layers_30_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_30_fc1_parametrizations_weight_original1 = None
	        view_2895 = torch.ops.aten.view.default(model_audio_tower_layers_30_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_30_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1664 = torch.ops.aten._assert_tensor_metadata.default(view_2893, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1664 = None
	        convert_element_type_1108 = torch.ops.prims.convert_element_type.default(view_2893, torch.float32);  view_2893 = None
	        _assert_tensor_metadata_1665 = torch.ops.aten._assert_tensor_metadata.default(view_2895, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1665 = None
	        convert_element_type_1109 = torch.ops.prims.convert_element_type.default(view_2895, torch.float32);  view_2895 = None
	        sub_8452 = torch.ops.aten.sub.Tensor(convert_element_type_1108, convert_element_type_1109);  convert_element_type_1108 = convert_element_type_1109 = None
	        mul_17894 = torch.ops.aten.mul.Tensor(sub_8452, view_2894);  sub_8452 = view_2894 = None
	        view_2896 = torch.ops.aten.view.default(mul_17894, [5120, 1280]);  mul_17894 = None
	        _assert_tensor_metadata_1666 = torch.ops.aten._assert_tensor_metadata.default(view_2896, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1666 = None
	        mul_17899 = sym_size_int * 1500
	        view_2897 = torch.ops.aten.view.default(mul_17889, [mul_17899, 1280]);  mul_17889 = mul_17899 = None
	        permute_309 = torch.ops.aten.permute.default(view_2896, [1, 0]);  view_2896 = None
	        addmm_153 = torch.ops.aten.addmm.default(model_audio_tower_layers_30_fc1_bias, view_2897, permute_309);  model_audio_tower_layers_30_fc1_bias = view_2897 = permute_309 = None
	        view_2898 = torch.ops.aten.view.default(addmm_153, [sym_size_int, 1500, 5120]);  addmm_153 = None
	        mul_17906 = torch.ops.aten.mul.Tensor(view_2898, 0.5)
	        mul_17907 = torch.ops.aten.mul.Tensor(view_2898, 0.7071067811865476);  view_2898 = None
	        erf_32 = torch.ops.aten.erf.default(mul_17907);  mul_17907 = None
	        add_28338 = torch.ops.aten.add.Tensor(erf_32, 1);  erf_32 = None
	        mul_17908 = torch.ops.aten.mul.Tensor(mul_17906, add_28338);  mul_17906 = add_28338 = None
	        amin_185 = torch.ops.aten.amin.default(mul_17908, [2])
	        amax_185 = torch.ops.aten.amax.default(mul_17908, [2])
	        full_370 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_185 = torch.ops.aten.minimum.default(amin_185, full_370);  amin_185 = full_370 = None
	        full_371 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_185 = torch.ops.aten.maximum.default(amax_185, full_371);  amax_185 = full_371 = None
	        sub_8465 = torch.ops.aten.sub.Tensor(maximum_185, minimum_185);  maximum_185 = None
	        div_370 = torch.ops.aten.div.Tensor(sub_8465, 255.0);  sub_8465 = None
	        clamp_min_555 = torch.ops.aten.clamp_min.default(div_370, 1.1920928955078125e-07);  div_370 = None
	        div_371 = torch.ops.aten.div.Tensor(minimum_185, clamp_min_555);  minimum_185 = None
	        round_371 = torch.ops.aten.round.default(div_371);  div_371 = None
	        sub_8471 = torch.ops.aten.sub.Tensor(-128, round_371);  round_371 = None
	        clamp_min_556 = torch.ops.aten.clamp_min.default(sub_8471, -128);  sub_8471 = None
	        clamp_max_370 = torch.ops.aten.clamp_max.default(clamp_min_556, 127);  clamp_min_556 = None
	        _assert_tensor_metadata_1667 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_555, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1667 = None
	        _assert_tensor_metadata_1668 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_370, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1668 = None
	        convert_element_type_1110 = torch.ops.prims.convert_element_type.default(clamp_max_370, torch.int8);  clamp_max_370 = None
	        view_2901 = torch.ops.aten.view.default(clamp_min_555, [sym_size_int, 1500, 1])
	        view_2902 = torch.ops.aten.view.default(convert_element_type_1110, [sym_size_int, 1500, 1])
	        reciprocal_185 = torch.ops.aten.reciprocal.default(view_2901);  view_2901 = None
	        mul_17954 = torch.ops.aten.mul.Tensor(reciprocal_185, 1.0);  reciprocal_185 = None
	        mul_17957 = torch.ops.aten.mul.Tensor(mul_17908, mul_17954);  mul_17908 = mul_17954 = None
	        round_372 = torch.ops.aten.round.default(mul_17957);  mul_17957 = None
	        add_28421 = torch.ops.aten.add.Tensor(round_372, view_2902);  round_372 = view_2902 = None
	        clamp_min_557 = torch.ops.aten.clamp_min.default(add_28421, -128);  add_28421 = None
	        clamp_max_371 = torch.ops.aten.clamp_max.default(clamp_min_557, 127);  clamp_min_557 = None
	        _assert_tensor_metadata_1669 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_371, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1669 = None
	        convert_element_type_1111 = torch.ops.prims.convert_element_type.default(clamp_max_371, torch.int8);  clamp_max_371 = None
	        view_2905 = torch.ops.aten.view.default(clamp_min_555, [sym_size_int, 1500, 1]);  clamp_min_555 = None
	        view_2906 = torch.ops.aten.view.default(convert_element_type_1110, [sym_size_int, 1500, 1]);  convert_element_type_1110 = None
	        _assert_tensor_metadata_1670 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1111, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1670 = None
	        convert_element_type_1112 = torch.ops.prims.convert_element_type.default(convert_element_type_1111, torch.float32);  convert_element_type_1111 = None
	        _assert_tensor_metadata_1671 = torch.ops.aten._assert_tensor_metadata.default(view_2906, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1671 = None
	        convert_element_type_1113 = torch.ops.prims.convert_element_type.default(view_2906, torch.float32);  view_2906 = None
	        sub_8491 = torch.ops.aten.sub.Tensor(convert_element_type_1112, convert_element_type_1113);  convert_element_type_1112 = convert_element_type_1113 = None
	        mul_17979 = torch.ops.aten.mul.Tensor(sub_8491, view_2905);  sub_8491 = view_2905 = None
	        _assert_tensor_metadata_1672 = torch.ops.aten._assert_tensor_metadata.default(mul_17979, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1672 = None
	        view_2908 = torch.ops.aten.view.default(model_audio_tower_layers_30_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_30_fc2_parametrizations_weight_original0 = None
	        view_2909 = torch.ops.aten.view.default(model_audio_tower_layers_30_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_30_fc2_parametrizations_weight_original1 = None
	        view_2910 = torch.ops.aten.view.default(model_audio_tower_layers_30_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_30_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1673 = torch.ops.aten._assert_tensor_metadata.default(view_2908, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1673 = None
	        convert_element_type_1114 = torch.ops.prims.convert_element_type.default(view_2908, torch.float32);  view_2908 = None
	        _assert_tensor_metadata_1674 = torch.ops.aten._assert_tensor_metadata.default(view_2910, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1674 = None
	        convert_element_type_1115 = torch.ops.prims.convert_element_type.default(view_2910, torch.float32);  view_2910 = None
	        sub_8495 = torch.ops.aten.sub.Tensor(convert_element_type_1114, convert_element_type_1115);  convert_element_type_1114 = convert_element_type_1115 = None
	        mul_17984 = torch.ops.aten.mul.Tensor(sub_8495, view_2909);  sub_8495 = view_2909 = None
	        view_2911 = torch.ops.aten.view.default(mul_17984, [1280, 5120]);  mul_17984 = None
	        _assert_tensor_metadata_1675 = torch.ops.aten._assert_tensor_metadata.default(view_2911, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1675 = None
	        mul_17989 = sym_size_int * 1500
	        view_2912 = torch.ops.aten.view.default(mul_17979, [mul_17989, 5120]);  mul_17979 = mul_17989 = None
	        permute_310 = torch.ops.aten.permute.default(view_2911, [1, 0]);  view_2911 = None
	        addmm_154 = torch.ops.aten.addmm.default(model_audio_tower_layers_30_fc2_bias, view_2912, permute_310);  model_audio_tower_layers_30_fc2_bias = view_2912 = permute_310 = None
	        view_2913 = torch.ops.aten.view.default(addmm_154, [sym_size_int, 1500, 1280]);  addmm_154 = None
	        add_28484 = torch.ops.aten.add.Tensor(add_28186, view_2913);  add_28186 = view_2913 = None
	        clone_249 = torch.ops.aten.clone.default(add_28484, memory_format = torch.contiguous_format)
	        var_mean_62 = torch.ops.aten.var_mean.correction(clone_249, [2], correction = 0, keepdim = True)
	        getitem_248 = var_mean_62[0]
	        getitem_249 = var_mean_62[1];  var_mean_62 = None
	        add_28489 = torch.ops.aten.add.Tensor(getitem_248, 1e-05);  getitem_248 = None
	        rsqrt_62 = torch.ops.aten.rsqrt.default(add_28489);  add_28489 = None
	        sub_8501 = torch.ops.aten.sub.Tensor(clone_249, getitem_249);  clone_249 = getitem_249 = None
	        mul_18000 = torch.ops.aten.mul.Tensor(sub_8501, rsqrt_62);  sub_8501 = rsqrt_62 = None
	        mul_18001 = torch.ops.aten.mul.Tensor(mul_18000, model_audio_tower_layers_31_self_attn_layer_norm_weight);  mul_18000 = model_audio_tower_layers_31_self_attn_layer_norm_weight = None
	        add_28490 = torch.ops.aten.add.Tensor(mul_18001, model_audio_tower_layers_31_self_attn_layer_norm_bias);  mul_18001 = model_audio_tower_layers_31_self_attn_layer_norm_bias = None
	        amin_186 = torch.ops.aten.amin.default(add_28490, [2])
	        amax_186 = torch.ops.aten.amax.default(add_28490, [2])
	        full_372 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_186 = torch.ops.aten.minimum.default(amin_186, full_372);  amin_186 = full_372 = None
	        full_373 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_186 = torch.ops.aten.maximum.default(amax_186, full_373);  amax_186 = full_373 = None
	        sub_8512 = torch.ops.aten.sub.Tensor(maximum_186, minimum_186);  maximum_186 = None
	        div_372 = torch.ops.aten.div.Tensor(sub_8512, 255.0);  sub_8512 = None
	        clamp_min_558 = torch.ops.aten.clamp_min.default(div_372, 1.1920928955078125e-07);  div_372 = None
	        div_373 = torch.ops.aten.div.Tensor(minimum_186, clamp_min_558);  minimum_186 = None
	        round_373 = torch.ops.aten.round.default(div_373);  div_373 = None
	        sub_8518 = torch.ops.aten.sub.Tensor(-128, round_373);  round_373 = None
	        clamp_min_559 = torch.ops.aten.clamp_min.default(sub_8518, -128);  sub_8518 = None
	        clamp_max_372 = torch.ops.aten.clamp_max.default(clamp_min_559, 127);  clamp_min_559 = None
	        _assert_tensor_metadata_1676 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_558, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1676 = None
	        _assert_tensor_metadata_1677 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_372, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1677 = None
	        convert_element_type_1116 = torch.ops.prims.convert_element_type.default(clamp_max_372, torch.int8);  clamp_max_372 = None
	        view_2916 = torch.ops.aten.view.default(clamp_min_558, [sym_size_int, 1500, 1])
	        view_2917 = torch.ops.aten.view.default(convert_element_type_1116, [sym_size_int, 1500, 1])
	        reciprocal_186 = torch.ops.aten.reciprocal.default(view_2916);  view_2916 = None
	        mul_18049 = torch.ops.aten.mul.Tensor(reciprocal_186, 1.0);  reciprocal_186 = None
	        mul_18052 = torch.ops.aten.mul.Tensor(add_28490, mul_18049);  mul_18049 = None
	        round_374 = torch.ops.aten.round.default(mul_18052);  mul_18052 = None
	        add_28577 = torch.ops.aten.add.Tensor(round_374, view_2917);  round_374 = view_2917 = None
	        clamp_min_560 = torch.ops.aten.clamp_min.default(add_28577, -128);  add_28577 = None
	        clamp_max_373 = torch.ops.aten.clamp_max.default(clamp_min_560, 127);  clamp_min_560 = None
	        _assert_tensor_metadata_1678 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_373, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1678 = None
	        convert_element_type_1117 = torch.ops.prims.convert_element_type.default(clamp_max_373, torch.int8);  clamp_max_373 = None
	        view_2920 = torch.ops.aten.view.default(clamp_min_558, [sym_size_int, 1500, 1]);  clamp_min_558 = None
	        view_2921 = torch.ops.aten.view.default(convert_element_type_1116, [sym_size_int, 1500, 1]);  convert_element_type_1116 = None
	        _assert_tensor_metadata_1679 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1117, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1679 = None
	        convert_element_type_1118 = torch.ops.prims.convert_element_type.default(convert_element_type_1117, torch.float32);  convert_element_type_1117 = None
	        _assert_tensor_metadata_1680 = torch.ops.aten._assert_tensor_metadata.default(view_2921, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1680 = None
	        convert_element_type_1119 = torch.ops.prims.convert_element_type.default(view_2921, torch.float32);  view_2921 = None
	        sub_8538 = torch.ops.aten.sub.Tensor(convert_element_type_1118, convert_element_type_1119);  convert_element_type_1118 = convert_element_type_1119 = None
	        mul_18074 = torch.ops.aten.mul.Tensor(sub_8538, view_2920);  sub_8538 = view_2920 = None
	        _assert_tensor_metadata_1681 = torch.ops.aten._assert_tensor_metadata.default(mul_18074, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1681 = None
	        view_2923 = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2924 = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2925 = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1682 = torch.ops.aten._assert_tensor_metadata.default(view_2923, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1682 = None
	        convert_element_type_1120 = torch.ops.prims.convert_element_type.default(view_2923, torch.float32);  view_2923 = None
	        _assert_tensor_metadata_1683 = torch.ops.aten._assert_tensor_metadata.default(view_2925, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1683 = None
	        convert_element_type_1121 = torch.ops.prims.convert_element_type.default(view_2925, torch.float32);  view_2925 = None
	        sub_8542 = torch.ops.aten.sub.Tensor(convert_element_type_1120, convert_element_type_1121);  convert_element_type_1120 = convert_element_type_1121 = None
	        mul_18079 = torch.ops.aten.mul.Tensor(sub_8542, view_2924);  sub_8542 = view_2924 = None
	        view_2926 = torch.ops.aten.view.default(mul_18079, [1280, 1280]);  mul_18079 = None
	        _assert_tensor_metadata_1684 = torch.ops.aten._assert_tensor_metadata.default(view_2926, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1684 = None
	        mul_18084 = sym_size_int * 1500
	        view_2927 = torch.ops.aten.view.default(mul_18074, [mul_18084, 1280]);  mul_18074 = mul_18084 = None
	        permute_311 = torch.ops.aten.permute.default(view_2926, [1, 0]);  view_2926 = None
	        addmm_155 = torch.ops.aten.addmm.default(model_audio_tower_layers_31_self_attn_q_proj_bias, view_2927, permute_311);  model_audio_tower_layers_31_self_attn_q_proj_bias = view_2927 = permute_311 = None
	        view_2928 = torch.ops.aten.view.default(addmm_155, [sym_size_int, 1500, 1280]);  addmm_155 = None
	        mul_18091 = torch.ops.aten.mul.Tensor(view_2928, 0.125);  view_2928 = None
	        view_2929 = torch.ops.aten.view.default(mul_18091, [sym_size_int, 1500, 20, 64]);  mul_18091 = None
	        permute_312 = torch.ops.aten.permute.default(view_2929, [0, 2, 1, 3]);  view_2929 = None
	        clone_250 = torch.ops.aten.clone.default(permute_312, memory_format = torch.contiguous_format);  permute_312 = None
	        amin_187 = torch.ops.aten.amin.default(add_28490, [2])
	        amax_187 = torch.ops.aten.amax.default(add_28490, [2])
	        full_374 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_187 = torch.ops.aten.minimum.default(amin_187, full_374);  amin_187 = full_374 = None
	        full_375 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_187 = torch.ops.aten.maximum.default(amax_187, full_375);  amax_187 = full_375 = None
	        sub_8557 = torch.ops.aten.sub.Tensor(maximum_187, minimum_187);  maximum_187 = None
	        div_374 = torch.ops.aten.div.Tensor(sub_8557, 255.0);  sub_8557 = None
	        clamp_min_561 = torch.ops.aten.clamp_min.default(div_374, 1.1920928955078125e-07);  div_374 = None
	        div_375 = torch.ops.aten.div.Tensor(minimum_187, clamp_min_561);  minimum_187 = None
	        round_375 = torch.ops.aten.round.default(div_375);  div_375 = None
	        sub_8563 = torch.ops.aten.sub.Tensor(-128, round_375);  round_375 = None
	        clamp_min_562 = torch.ops.aten.clamp_min.default(sub_8563, -128);  sub_8563 = None
	        clamp_max_374 = torch.ops.aten.clamp_max.default(clamp_min_562, 127);  clamp_min_562 = None
	        _assert_tensor_metadata_1685 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_561, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1685 = None
	        _assert_tensor_metadata_1686 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_374, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1686 = None
	        convert_element_type_1122 = torch.ops.prims.convert_element_type.default(clamp_max_374, torch.int8);  clamp_max_374 = None
	        view_2932 = torch.ops.aten.view.default(clamp_min_561, [sym_size_int, 1500, 1])
	        view_2933 = torch.ops.aten.view.default(convert_element_type_1122, [sym_size_int, 1500, 1])
	        reciprocal_187 = torch.ops.aten.reciprocal.default(view_2932);  view_2932 = None
	        mul_18145 = torch.ops.aten.mul.Tensor(reciprocal_187, 1.0);  reciprocal_187 = None
	        mul_18148 = torch.ops.aten.mul.Tensor(add_28490, mul_18145);  mul_18145 = None
	        round_376 = torch.ops.aten.round.default(mul_18148);  mul_18148 = None
	        add_28729 = torch.ops.aten.add.Tensor(round_376, view_2933);  round_376 = view_2933 = None
	        clamp_min_563 = torch.ops.aten.clamp_min.default(add_28729, -128);  add_28729 = None
	        clamp_max_375 = torch.ops.aten.clamp_max.default(clamp_min_563, 127);  clamp_min_563 = None
	        _assert_tensor_metadata_1687 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_375, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1687 = None
	        convert_element_type_1123 = torch.ops.prims.convert_element_type.default(clamp_max_375, torch.int8);  clamp_max_375 = None
	        view_2936 = torch.ops.aten.view.default(clamp_min_561, [sym_size_int, 1500, 1]);  clamp_min_561 = None
	        view_2937 = torch.ops.aten.view.default(convert_element_type_1122, [sym_size_int, 1500, 1]);  convert_element_type_1122 = None
	        _assert_tensor_metadata_1688 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1123, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1688 = None
	        convert_element_type_1124 = torch.ops.prims.convert_element_type.default(convert_element_type_1123, torch.float32);  convert_element_type_1123 = None
	        _assert_tensor_metadata_1689 = torch.ops.aten._assert_tensor_metadata.default(view_2937, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1689 = None
	        convert_element_type_1125 = torch.ops.prims.convert_element_type.default(view_2937, torch.float32);  view_2937 = None
	        sub_8583 = torch.ops.aten.sub.Tensor(convert_element_type_1124, convert_element_type_1125);  convert_element_type_1124 = convert_element_type_1125 = None
	        mul_18170 = torch.ops.aten.mul.Tensor(sub_8583, view_2936);  sub_8583 = view_2936 = None
	        _assert_tensor_metadata_1690 = torch.ops.aten._assert_tensor_metadata.default(mul_18170, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1690 = None
	        view_2939 = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2940 = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2941 = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1691 = torch.ops.aten._assert_tensor_metadata.default(view_2939, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1691 = None
	        convert_element_type_1126 = torch.ops.prims.convert_element_type.default(view_2939, torch.float32);  view_2939 = None
	        _assert_tensor_metadata_1692 = torch.ops.aten._assert_tensor_metadata.default(view_2941, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1692 = None
	        convert_element_type_1127 = torch.ops.prims.convert_element_type.default(view_2941, torch.float32);  view_2941 = None
	        sub_8587 = torch.ops.aten.sub.Tensor(convert_element_type_1126, convert_element_type_1127);  convert_element_type_1126 = convert_element_type_1127 = None
	        mul_18175 = torch.ops.aten.mul.Tensor(sub_8587, view_2940);  sub_8587 = view_2940 = None
	        view_2942 = torch.ops.aten.view.default(mul_18175, [1280, 1280]);  mul_18175 = None
	        _assert_tensor_metadata_1693 = torch.ops.aten._assert_tensor_metadata.default(view_2942, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1693 = None
	        permute_313 = torch.ops.aten.permute.default(view_2942, [1, 0]);  view_2942 = None
	        mul_18178 = sym_size_int * 1500
	        view_2943 = torch.ops.aten.view.default(mul_18170, [mul_18178, 1280]);  mul_18170 = mul_18178 = None
	        mm_31 = torch.ops.aten.mm.default(view_2943, permute_313);  view_2943 = permute_313 = None
	        view_2944 = torch.ops.aten.view.default(mm_31, [sym_size_int, 1500, 1280]);  mm_31 = None
	        view_2945 = torch.ops.aten.view.default(view_2944, [sym_size_int, -1, 20, 64]);  view_2944 = None
	        permute_314 = torch.ops.aten.permute.default(view_2945, [0, 2, 1, 3]);  view_2945 = None
	        clone_251 = torch.ops.aten.clone.default(permute_314, memory_format = torch.contiguous_format);  permute_314 = None
	        amin_188 = torch.ops.aten.amin.default(add_28490, [2])
	        amax_188 = torch.ops.aten.amax.default(add_28490, [2])
	        full_376 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_188 = torch.ops.aten.minimum.default(amin_188, full_376);  amin_188 = full_376 = None
	        full_377 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_188 = torch.ops.aten.maximum.default(amax_188, full_377);  amax_188 = full_377 = None
	        sub_8601 = torch.ops.aten.sub.Tensor(maximum_188, minimum_188);  maximum_188 = None
	        div_376 = torch.ops.aten.div.Tensor(sub_8601, 255.0);  sub_8601 = None
	        clamp_min_564 = torch.ops.aten.clamp_min.default(div_376, 1.1920928955078125e-07);  div_376 = None
	        div_377 = torch.ops.aten.div.Tensor(minimum_188, clamp_min_564);  minimum_188 = None
	        round_377 = torch.ops.aten.round.default(div_377);  div_377 = None
	        sub_8607 = torch.ops.aten.sub.Tensor(-128, round_377);  round_377 = None
	        clamp_min_565 = torch.ops.aten.clamp_min.default(sub_8607, -128);  sub_8607 = None
	        clamp_max_376 = torch.ops.aten.clamp_max.default(clamp_min_565, 127);  clamp_min_565 = None
	        _assert_tensor_metadata_1694 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_564, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1694 = None
	        _assert_tensor_metadata_1695 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_376, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1695 = None
	        convert_element_type_1128 = torch.ops.prims.convert_element_type.default(clamp_max_376, torch.int8);  clamp_max_376 = None
	        view_2948 = torch.ops.aten.view.default(clamp_min_564, [sym_size_int, 1500, 1])
	        view_2949 = torch.ops.aten.view.default(convert_element_type_1128, [sym_size_int, 1500, 1])
	        reciprocal_188 = torch.ops.aten.reciprocal.default(view_2948);  view_2948 = None
	        mul_18244 = torch.ops.aten.mul.Tensor(reciprocal_188, 1.0);  reciprocal_188 = None
	        mul_18247 = torch.ops.aten.mul.Tensor(add_28490, mul_18244);  add_28490 = mul_18244 = None
	        round_378 = torch.ops.aten.round.default(mul_18247);  mul_18247 = None
	        add_28877 = torch.ops.aten.add.Tensor(round_378, view_2949);  round_378 = view_2949 = None
	        clamp_min_566 = torch.ops.aten.clamp_min.default(add_28877, -128);  add_28877 = None
	        clamp_max_377 = torch.ops.aten.clamp_max.default(clamp_min_566, 127);  clamp_min_566 = None
	        _assert_tensor_metadata_1696 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_377, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1696 = None
	        convert_element_type_1129 = torch.ops.prims.convert_element_type.default(clamp_max_377, torch.int8);  clamp_max_377 = None
	        view_2952 = torch.ops.aten.view.default(clamp_min_564, [sym_size_int, 1500, 1]);  clamp_min_564 = None
	        view_2953 = torch.ops.aten.view.default(convert_element_type_1128, [sym_size_int, 1500, 1]);  convert_element_type_1128 = None
	        _assert_tensor_metadata_1697 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1129, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1697 = None
	        convert_element_type_1130 = torch.ops.prims.convert_element_type.default(convert_element_type_1129, torch.float32);  convert_element_type_1129 = None
	        _assert_tensor_metadata_1698 = torch.ops.aten._assert_tensor_metadata.default(view_2953, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1698 = None
	        convert_element_type_1131 = torch.ops.prims.convert_element_type.default(view_2953, torch.float32);  view_2953 = None
	        sub_8627 = torch.ops.aten.sub.Tensor(convert_element_type_1130, convert_element_type_1131);  convert_element_type_1130 = convert_element_type_1131 = None
	        mul_18269 = torch.ops.aten.mul.Tensor(sub_8627, view_2952);  sub_8627 = view_2952 = None
	        _assert_tensor_metadata_1699 = torch.ops.aten._assert_tensor_metadata.default(mul_18269, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1699 = None
	        view_2955 = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2956 = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2957 = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1700 = torch.ops.aten._assert_tensor_metadata.default(view_2955, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1700 = None
	        convert_element_type_1132 = torch.ops.prims.convert_element_type.default(view_2955, torch.float32);  view_2955 = None
	        _assert_tensor_metadata_1701 = torch.ops.aten._assert_tensor_metadata.default(view_2957, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1701 = None
	        convert_element_type_1133 = torch.ops.prims.convert_element_type.default(view_2957, torch.float32);  view_2957 = None
	        sub_8631 = torch.ops.aten.sub.Tensor(convert_element_type_1132, convert_element_type_1133);  convert_element_type_1132 = convert_element_type_1133 = None
	        mul_18274 = torch.ops.aten.mul.Tensor(sub_8631, view_2956);  sub_8631 = view_2956 = None
	        view_2958 = torch.ops.aten.view.default(mul_18274, [1280, 1280]);  mul_18274 = None
	        _assert_tensor_metadata_1702 = torch.ops.aten._assert_tensor_metadata.default(view_2958, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1702 = None
	        mul_18279 = sym_size_int * 1500
	        view_2959 = torch.ops.aten.view.default(mul_18269, [mul_18279, 1280]);  mul_18269 = mul_18279 = None
	        permute_315 = torch.ops.aten.permute.default(view_2958, [1, 0]);  view_2958 = None
	        addmm_156 = torch.ops.aten.addmm.default(model_audio_tower_layers_31_self_attn_v_proj_bias, view_2959, permute_315);  model_audio_tower_layers_31_self_attn_v_proj_bias = view_2959 = permute_315 = None
	        view_2960 = torch.ops.aten.view.default(addmm_156, [sym_size_int, 1500, 1280]);  addmm_156 = None
	        view_2961 = torch.ops.aten.view.default(view_2960, [sym_size_int, -1, 20, 64]);  view_2960 = None
	        permute_316 = torch.ops.aten.permute.default(view_2961, [0, 2, 1, 3]);  view_2961 = None
	        clone_252 = torch.ops.aten.clone.default(permute_316, memory_format = torch.contiguous_format);  permute_316 = None
	        _scaled_dot_product_efficient_attention_31 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_250, clone_251, clone_252, None, False, scale = 1.0);  clone_250 = clone_251 = clone_252 = None
	        getitem_250 = _scaled_dot_product_efficient_attention_31[0];  _scaled_dot_product_efficient_attention_31 = None
	        permute_317 = torch.ops.aten.permute.default(getitem_250, [0, 2, 1, 3]);  getitem_250 = None
	        view_2962 = torch.ops.aten.view.default(permute_317, [sym_size_int, 1500, -1]);  permute_317 = None
	        amin_189 = torch.ops.aten.amin.default(view_2962, [2])
	        amax_189 = torch.ops.aten.amax.default(view_2962, [2])
	        full_378 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_189 = torch.ops.aten.minimum.default(amin_189, full_378);  amin_189 = full_378 = None
	        full_379 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_189 = torch.ops.aten.maximum.default(amax_189, full_379);  amax_189 = full_379 = None
	        sub_8649 = torch.ops.aten.sub.Tensor(maximum_189, minimum_189);  maximum_189 = None
	        div_378 = torch.ops.aten.div.Tensor(sub_8649, 255.0);  sub_8649 = None
	        clamp_min_567 = torch.ops.aten.clamp_min.default(div_378, 1.1920928955078125e-07);  div_378 = None
	        div_379 = torch.ops.aten.div.Tensor(minimum_189, clamp_min_567);  minimum_189 = None
	        round_379 = torch.ops.aten.round.default(div_379);  div_379 = None
	        sub_8655 = torch.ops.aten.sub.Tensor(-128, round_379);  round_379 = None
	        clamp_min_568 = torch.ops.aten.clamp_min.default(sub_8655, -128);  sub_8655 = None
	        clamp_max_378 = torch.ops.aten.clamp_max.default(clamp_min_568, 127);  clamp_min_568 = None
	        _assert_tensor_metadata_1703 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_567, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1703 = None
	        _assert_tensor_metadata_1704 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_378, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1704 = None
	        convert_element_type_1134 = torch.ops.prims.convert_element_type.default(clamp_max_378, torch.int8);  clamp_max_378 = None
	        view_2965 = torch.ops.aten.view.default(clamp_min_567, [sym_size_int, 1500, 1])
	        view_2966 = torch.ops.aten.view.default(convert_element_type_1134, [sym_size_int, 1500, 1])
	        reciprocal_189 = torch.ops.aten.reciprocal.default(view_2965);  view_2965 = None
	        mul_18349 = torch.ops.aten.mul.Tensor(reciprocal_189, 1.0);  reciprocal_189 = None
	        mul_18352 = torch.ops.aten.mul.Tensor(view_2962, mul_18349);  view_2962 = mul_18349 = None
	        round_380 = torch.ops.aten.round.default(mul_18352);  mul_18352 = None
	        add_29041 = torch.ops.aten.add.Tensor(round_380, view_2966);  round_380 = view_2966 = None
	        clamp_min_569 = torch.ops.aten.clamp_min.default(add_29041, -128);  add_29041 = None
	        clamp_max_379 = torch.ops.aten.clamp_max.default(clamp_min_569, 127);  clamp_min_569 = None
	        _assert_tensor_metadata_1705 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_379, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1705 = None
	        convert_element_type_1135 = torch.ops.prims.convert_element_type.default(clamp_max_379, torch.int8);  clamp_max_379 = None
	        view_2969 = torch.ops.aten.view.default(clamp_min_567, [sym_size_int, 1500, 1]);  clamp_min_567 = None
	        view_2970 = torch.ops.aten.view.default(convert_element_type_1134, [sym_size_int, 1500, 1]);  convert_element_type_1134 = None
	        _assert_tensor_metadata_1706 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1135, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1706 = None
	        convert_element_type_1136 = torch.ops.prims.convert_element_type.default(convert_element_type_1135, torch.float32);  convert_element_type_1135 = None
	        _assert_tensor_metadata_1707 = torch.ops.aten._assert_tensor_metadata.default(view_2970, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1707 = None
	        convert_element_type_1137 = torch.ops.prims.convert_element_type.default(view_2970, torch.float32);  view_2970 = None
	        sub_8675 = torch.ops.aten.sub.Tensor(convert_element_type_1136, convert_element_type_1137);  convert_element_type_1136 = convert_element_type_1137 = None
	        mul_18374 = torch.ops.aten.mul.Tensor(sub_8675, view_2969);  sub_8675 = view_2969 = None
	        _assert_tensor_metadata_1708 = torch.ops.aten._assert_tensor_metadata.default(mul_18374, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1708 = None
	        view_2972 = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2973 = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2974 = torch.ops.aten.view.default(model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1709 = torch.ops.aten._assert_tensor_metadata.default(view_2972, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1709 = None
	        convert_element_type_1138 = torch.ops.prims.convert_element_type.default(view_2972, torch.float32);  view_2972 = None
	        _assert_tensor_metadata_1710 = torch.ops.aten._assert_tensor_metadata.default(view_2974, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1710 = None
	        convert_element_type_1139 = torch.ops.prims.convert_element_type.default(view_2974, torch.float32);  view_2974 = None
	        sub_8679 = torch.ops.aten.sub.Tensor(convert_element_type_1138, convert_element_type_1139);  convert_element_type_1138 = convert_element_type_1139 = None
	        mul_18379 = torch.ops.aten.mul.Tensor(sub_8679, view_2973);  sub_8679 = view_2973 = None
	        view_2975 = torch.ops.aten.view.default(mul_18379, [1280, 1280]);  mul_18379 = None
	        _assert_tensor_metadata_1711 = torch.ops.aten._assert_tensor_metadata.default(view_2975, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1711 = None
	        mul_18384 = sym_size_int * 1500
	        view_2976 = torch.ops.aten.view.default(mul_18374, [mul_18384, 1280]);  mul_18374 = mul_18384 = None
	        permute_318 = torch.ops.aten.permute.default(view_2975, [1, 0]);  view_2975 = None
	        addmm_157 = torch.ops.aten.addmm.default(model_audio_tower_layers_31_self_attn_out_proj_bias, view_2976, permute_318);  model_audio_tower_layers_31_self_attn_out_proj_bias = view_2976 = permute_318 = None
	        view_2977 = torch.ops.aten.view.default(addmm_157, [sym_size_int, 1500, 1280]);  addmm_157 = None
	        add_29104 = torch.ops.aten.add.Tensor(add_28484, view_2977);  add_28484 = view_2977 = None
	        clone_254 = torch.ops.aten.clone.default(add_29104, memory_format = torch.contiguous_format)
	        var_mean_63 = torch.ops.aten.var_mean.correction(clone_254, [2], correction = 0, keepdim = True)
	        getitem_254 = var_mean_63[0]
	        getitem_255 = var_mean_63[1];  var_mean_63 = None
	        add_29109 = torch.ops.aten.add.Tensor(getitem_254, 1e-05);  getitem_254 = None
	        rsqrt_63 = torch.ops.aten.rsqrt.default(add_29109);  add_29109 = None
	        sub_8685 = torch.ops.aten.sub.Tensor(clone_254, getitem_255);  clone_254 = getitem_255 = None
	        mul_18395 = torch.ops.aten.mul.Tensor(sub_8685, rsqrt_63);  sub_8685 = rsqrt_63 = None
	        mul_18396 = torch.ops.aten.mul.Tensor(mul_18395, model_audio_tower_layers_31_final_layer_norm_weight);  mul_18395 = model_audio_tower_layers_31_final_layer_norm_weight = None
	        add_29110 = torch.ops.aten.add.Tensor(mul_18396, model_audio_tower_layers_31_final_layer_norm_bias);  mul_18396 = model_audio_tower_layers_31_final_layer_norm_bias = None
	        amin_190 = torch.ops.aten.amin.default(add_29110, [2])
	        amax_190 = torch.ops.aten.amax.default(add_29110, [2])
	        full_380 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_190 = torch.ops.aten.minimum.default(amin_190, full_380);  amin_190 = full_380 = None
	        full_381 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_190 = torch.ops.aten.maximum.default(amax_190, full_381);  amax_190 = full_381 = None
	        sub_8696 = torch.ops.aten.sub.Tensor(maximum_190, minimum_190);  maximum_190 = None
	        div_380 = torch.ops.aten.div.Tensor(sub_8696, 255.0);  sub_8696 = None
	        clamp_min_570 = torch.ops.aten.clamp_min.default(div_380, 1.1920928955078125e-07);  div_380 = None
	        div_381 = torch.ops.aten.div.Tensor(minimum_190, clamp_min_570);  minimum_190 = None
	        round_381 = torch.ops.aten.round.default(div_381);  div_381 = None
	        sub_8702 = torch.ops.aten.sub.Tensor(-128, round_381);  round_381 = None
	        clamp_min_571 = torch.ops.aten.clamp_min.default(sub_8702, -128);  sub_8702 = None
	        clamp_max_380 = torch.ops.aten.clamp_max.default(clamp_min_571, 127);  clamp_min_571 = None
	        _assert_tensor_metadata_1712 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_570, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1712 = None
	        _assert_tensor_metadata_1713 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_380, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1713 = None
	        convert_element_type_1140 = torch.ops.prims.convert_element_type.default(clamp_max_380, torch.int8);  clamp_max_380 = None
	        view_2980 = torch.ops.aten.view.default(clamp_min_570, [sym_size_int, 1500, 1])
	        view_2981 = torch.ops.aten.view.default(convert_element_type_1140, [sym_size_int, 1500, 1])
	        reciprocal_190 = torch.ops.aten.reciprocal.default(view_2980);  view_2980 = None
	        mul_18444 = torch.ops.aten.mul.Tensor(reciprocal_190, 1.0);  reciprocal_190 = None
	        mul_18447 = torch.ops.aten.mul.Tensor(add_29110, mul_18444);  add_29110 = mul_18444 = None
	        round_382 = torch.ops.aten.round.default(mul_18447);  mul_18447 = None
	        add_29197 = torch.ops.aten.add.Tensor(round_382, view_2981);  round_382 = view_2981 = None
	        clamp_min_572 = torch.ops.aten.clamp_min.default(add_29197, -128);  add_29197 = None
	        clamp_max_381 = torch.ops.aten.clamp_max.default(clamp_min_572, 127);  clamp_min_572 = None
	        _assert_tensor_metadata_1714 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_381, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1714 = None
	        convert_element_type_1141 = torch.ops.prims.convert_element_type.default(clamp_max_381, torch.int8);  clamp_max_381 = None
	        view_2984 = torch.ops.aten.view.default(clamp_min_570, [sym_size_int, 1500, 1]);  clamp_min_570 = None
	        view_2985 = torch.ops.aten.view.default(convert_element_type_1140, [sym_size_int, 1500, 1]);  convert_element_type_1140 = None
	        _assert_tensor_metadata_1715 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1141, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1715 = None
	        convert_element_type_1142 = torch.ops.prims.convert_element_type.default(convert_element_type_1141, torch.float32);  convert_element_type_1141 = None
	        _assert_tensor_metadata_1716 = torch.ops.aten._assert_tensor_metadata.default(view_2985, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1716 = None
	        convert_element_type_1143 = torch.ops.prims.convert_element_type.default(view_2985, torch.float32);  view_2985 = None
	        sub_8722 = torch.ops.aten.sub.Tensor(convert_element_type_1142, convert_element_type_1143);  convert_element_type_1142 = convert_element_type_1143 = None
	        mul_18469 = torch.ops.aten.mul.Tensor(sub_8722, view_2984);  sub_8722 = view_2984 = None
	        _assert_tensor_metadata_1717 = torch.ops.aten._assert_tensor_metadata.default(mul_18469, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1717 = None
	        view_2987 = torch.ops.aten.view.default(model_audio_tower_layers_31_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_31_fc1_parametrizations_weight_original0 = None
	        view_2988 = torch.ops.aten.view.default(model_audio_tower_layers_31_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_31_fc1_parametrizations_weight_original1 = None
	        view_2989 = torch.ops.aten.view.default(model_audio_tower_layers_31_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_31_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1718 = torch.ops.aten._assert_tensor_metadata.default(view_2987, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1718 = None
	        convert_element_type_1144 = torch.ops.prims.convert_element_type.default(view_2987, torch.float32);  view_2987 = None
	        _assert_tensor_metadata_1719 = torch.ops.aten._assert_tensor_metadata.default(view_2989, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1719 = None
	        convert_element_type_1145 = torch.ops.prims.convert_element_type.default(view_2989, torch.float32);  view_2989 = None
	        sub_8726 = torch.ops.aten.sub.Tensor(convert_element_type_1144, convert_element_type_1145);  convert_element_type_1144 = convert_element_type_1145 = None
	        mul_18474 = torch.ops.aten.mul.Tensor(sub_8726, view_2988);  sub_8726 = view_2988 = None
	        view_2990 = torch.ops.aten.view.default(mul_18474, [5120, 1280]);  mul_18474 = None
	        _assert_tensor_metadata_1720 = torch.ops.aten._assert_tensor_metadata.default(view_2990, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1720 = None
	        mul_18479 = sym_size_int * 1500
	        view_2991 = torch.ops.aten.view.default(mul_18469, [mul_18479, 1280]);  mul_18469 = mul_18479 = None
	        permute_319 = torch.ops.aten.permute.default(view_2990, [1, 0]);  view_2990 = None
	        addmm_158 = torch.ops.aten.addmm.default(model_audio_tower_layers_31_fc1_bias, view_2991, permute_319);  model_audio_tower_layers_31_fc1_bias = view_2991 = permute_319 = None
	        view_2992 = torch.ops.aten.view.default(addmm_158, [sym_size_int, 1500, 5120]);  addmm_158 = None
	        mul_18486 = torch.ops.aten.mul.Tensor(view_2992, 0.5)
	        mul_18487 = torch.ops.aten.mul.Tensor(view_2992, 0.7071067811865476);  view_2992 = None
	        erf_33 = torch.ops.aten.erf.default(mul_18487);  mul_18487 = None
	        add_29256 = torch.ops.aten.add.Tensor(erf_33, 1);  erf_33 = None
	        mul_18488 = torch.ops.aten.mul.Tensor(mul_18486, add_29256);  mul_18486 = add_29256 = None
	        amin_191 = torch.ops.aten.amin.default(mul_18488, [2])
	        amax_191 = torch.ops.aten.amax.default(mul_18488, [2])
	        full_382 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_191 = torch.ops.aten.minimum.default(amin_191, full_382);  amin_191 = full_382 = None
	        full_383 = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_191 = torch.ops.aten.maximum.default(amax_191, full_383);  amax_191 = full_383 = None
	        sub_8739 = torch.ops.aten.sub.Tensor(maximum_191, minimum_191);  maximum_191 = None
	        div_382 = torch.ops.aten.div.Tensor(sub_8739, 255.0);  sub_8739 = None
	        clamp_min_573 = torch.ops.aten.clamp_min.default(div_382, 1.1920928955078125e-07);  div_382 = None
	        div_383 = torch.ops.aten.div.Tensor(minimum_191, clamp_min_573);  minimum_191 = None
	        round_383 = torch.ops.aten.round.default(div_383);  div_383 = None
	        sub_8745 = torch.ops.aten.sub.Tensor(-128, round_383);  round_383 = None
	        clamp_min_574 = torch.ops.aten.clamp_min.default(sub_8745, -128);  sub_8745 = None
	        clamp_max_382 = torch.ops.aten.clamp_max.default(clamp_min_574, 127);  clamp_min_574 = None
	        _assert_tensor_metadata_1721 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_573, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1721 = None
	        _assert_tensor_metadata_1722 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_382, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1722 = None
	        convert_element_type_1146 = torch.ops.prims.convert_element_type.default(clamp_max_382, torch.int8);  clamp_max_382 = None
	        view_2995 = torch.ops.aten.view.default(clamp_min_573, [sym_size_int, 1500, 1])
	        view_2996 = torch.ops.aten.view.default(convert_element_type_1146, [sym_size_int, 1500, 1])
	        reciprocal_191 = torch.ops.aten.reciprocal.default(view_2995);  view_2995 = None
	        mul_18534 = torch.ops.aten.mul.Tensor(reciprocal_191, 1.0);  reciprocal_191 = None
	        mul_18537 = torch.ops.aten.mul.Tensor(mul_18488, mul_18534);  mul_18488 = mul_18534 = None
	        round_384 = torch.ops.aten.round.default(mul_18537);  mul_18537 = None
	        add_29339 = torch.ops.aten.add.Tensor(round_384, view_2996);  round_384 = view_2996 = None
	        clamp_min_575 = torch.ops.aten.clamp_min.default(add_29339, -128);  add_29339 = None
	        clamp_max_383 = torch.ops.aten.clamp_max.default(clamp_min_575, 127);  clamp_min_575 = None
	        _assert_tensor_metadata_1723 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_383, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1723 = None
	        convert_element_type_1147 = torch.ops.prims.convert_element_type.default(clamp_max_383, torch.int8);  clamp_max_383 = None
	        view_2999 = torch.ops.aten.view.default(clamp_min_573, [sym_size_int, 1500, 1]);  clamp_min_573 = None
	        view_3000 = torch.ops.aten.view.default(convert_element_type_1146, [sym_size_int, 1500, 1]);  convert_element_type_1146 = None
	        _assert_tensor_metadata_1724 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1147, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1724 = None
	        convert_element_type_1148 = torch.ops.prims.convert_element_type.default(convert_element_type_1147, torch.float32);  convert_element_type_1147 = None
	        _assert_tensor_metadata_1725 = torch.ops.aten._assert_tensor_metadata.default(view_3000, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1725 = None
	        convert_element_type_1149 = torch.ops.prims.convert_element_type.default(view_3000, torch.float32);  view_3000 = None
	        sub_8765 = torch.ops.aten.sub.Tensor(convert_element_type_1148, convert_element_type_1149);  convert_element_type_1148 = convert_element_type_1149 = None
	        mul_18559 = torch.ops.aten.mul.Tensor(sub_8765, view_2999);  sub_8765 = view_2999 = None
	        _assert_tensor_metadata_1726 = torch.ops.aten._assert_tensor_metadata.default(mul_18559, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1726 = None
	        view_3002 = torch.ops.aten.view.default(model_audio_tower_layers_31_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_31_fc2_parametrizations_weight_original0 = None
	        view_3003 = torch.ops.aten.view.default(model_audio_tower_layers_31_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_31_fc2_parametrizations_weight_original1 = None
	        view_3004 = torch.ops.aten.view.default(model_audio_tower_layers_31_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_31_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1727 = torch.ops.aten._assert_tensor_metadata.default(view_3002, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1727 = None
	        convert_element_type_1150 = torch.ops.prims.convert_element_type.default(view_3002, torch.float32);  view_3002 = None
	        _assert_tensor_metadata_1728 = torch.ops.aten._assert_tensor_metadata.default(view_3004, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1728 = None
	        convert_element_type_1151 = torch.ops.prims.convert_element_type.default(view_3004, torch.float32);  view_3004 = None
	        sub_8769 = torch.ops.aten.sub.Tensor(convert_element_type_1150, convert_element_type_1151);  convert_element_type_1150 = convert_element_type_1151 = None
	        mul_18564 = torch.ops.aten.mul.Tensor(sub_8769, view_3003);  sub_8769 = view_3003 = None
	        view_3005 = torch.ops.aten.view.default(mul_18564, [1280, 5120]);  mul_18564 = None
	        _assert_tensor_metadata_1729 = torch.ops.aten._assert_tensor_metadata.default(view_3005, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1729 = None
	        mul_18569 = sym_size_int * 1500
	        view_3006 = torch.ops.aten.view.default(mul_18559, [mul_18569, 5120]);  mul_18559 = mul_18569 = None
	        permute_320 = torch.ops.aten.permute.default(view_3005, [1, 0]);  view_3005 = None
	        addmm_159 = torch.ops.aten.addmm.default(model_audio_tower_layers_31_fc2_bias, view_3006, permute_320);  model_audio_tower_layers_31_fc2_bias = view_3006 = permute_320 = None
	        view_3007 = torch.ops.aten.view.default(addmm_159, [sym_size_int, 1500, 1280]);  addmm_159 = sym_size_int = None
	        add_29402 = torch.ops.aten.add.Tensor(add_29104, view_3007);  add_29104 = view_3007 = None
	        clone_257 = torch.ops.aten.clone.default(add_29402, memory_format = torch.contiguous_format);  add_29402 = None
	        var_mean_64 = torch.ops.aten.var_mean.correction(clone_257, [2], correction = 0, keepdim = True)
	        getitem_256 = var_mean_64[0]
	        getitem_257 = var_mean_64[1];  var_mean_64 = None
	        add_29407 = torch.ops.aten.add.Tensor(getitem_256, 1e-05);  getitem_256 = None
	        rsqrt_64 = torch.ops.aten.rsqrt.default(add_29407);  add_29407 = None
	        sub_8775 = torch.ops.aten.sub.Tensor(clone_257, getitem_257);  clone_257 = getitem_257 = None
	        mul_18580 = torch.ops.aten.mul.Tensor(sub_8775, rsqrt_64);  sub_8775 = rsqrt_64 = None
	        mul_18581 = torch.ops.aten.mul.Tensor(mul_18580, model_audio_tower_layer_norm_weight);  mul_18580 = model_audio_tower_layer_norm_weight = None
	        add_29408 = torch.ops.aten.add.Tensor(mul_18581, model_audio_tower_layer_norm_bias);  mul_18581 = model_audio_tower_layer_norm_bias = None
	        view_3008 = torch.ops.aten.view.default(add_29408, [-1, 5120]);  add_29408 = None
	        sym_size_int_193 = torch.ops.aten.sym_size.int(view_3008, 0)
	        amin_192 = torch.ops.aten.amin.default(view_3008, [1])
	        amax_192 = torch.ops.aten.amax.default(view_3008, [1])
	        full_384 = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_192 = torch.ops.aten.minimum.default(amin_192, full_384);  amin_192 = full_384 = None
	        full_385 = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_192 = torch.ops.aten.maximum.default(amax_192, full_385);  amax_192 = full_385 = None
	        sub_8787 = torch.ops.aten.sub.Tensor(maximum_192, minimum_192);  maximum_192 = None
	        div_384 = torch.ops.aten.div.Tensor(sub_8787, 255.0);  sub_8787 = None
	        clamp_min_576 = torch.ops.aten.clamp_min.default(div_384, 1.1920928955078125e-07);  div_384 = None
	        div_385 = torch.ops.aten.div.Tensor(minimum_192, clamp_min_576);  minimum_192 = None
	        round_385 = torch.ops.aten.round.default(div_385);  div_385 = None
	        sub_8793 = torch.ops.aten.sub.Tensor(-128, round_385);  round_385 = None
	        clamp_min_577 = torch.ops.aten.clamp_min.default(sub_8793, -128);  sub_8793 = None
	        clamp_max_384 = torch.ops.aten.clamp_max.default(clamp_min_577, 127);  clamp_min_577 = None
	        _assert_tensor_metadata_1730 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_576, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1730 = None
	        _assert_tensor_metadata_1731 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_384, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1731 = None
	        convert_element_type_1152 = torch.ops.prims.convert_element_type.default(clamp_max_384, torch.int8);  clamp_max_384 = None
	        view_3011 = torch.ops.aten.view.default(clamp_min_576, [sym_size_int_193, 1])
	        view_3012 = torch.ops.aten.view.default(convert_element_type_1152, [sym_size_int_193, 1])
	        reciprocal_192 = torch.ops.aten.reciprocal.default(view_3011);  view_3011 = None
	        mul_18613 = torch.ops.aten.mul.Tensor(reciprocal_192, 1.0);  reciprocal_192 = None
	        mul_18615 = torch.ops.aten.mul.Tensor(view_3008, mul_18613);  view_3008 = mul_18613 = None
	        round_386 = torch.ops.aten.round.default(mul_18615);  mul_18615 = None
	        add_29476 = torch.ops.aten.add.Tensor(round_386, view_3012);  round_386 = view_3012 = None
	        clamp_min_578 = torch.ops.aten.clamp_min.default(add_29476, -128);  add_29476 = None
	        clamp_max_385 = torch.ops.aten.clamp_max.default(clamp_min_578, 127);  clamp_min_578 = None
	        _assert_tensor_metadata_1732 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_385, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1732 = None
	        convert_element_type_1153 = torch.ops.prims.convert_element_type.default(clamp_max_385, torch.int8);  clamp_max_385 = None
	        view_3015 = torch.ops.aten.view.default(clamp_min_576, [sym_size_int_193, 1]);  clamp_min_576 = None
	        view_3016 = torch.ops.aten.view.default(convert_element_type_1152, [sym_size_int_193, 1]);  convert_element_type_1152 = None
	        _assert_tensor_metadata_1733 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1153, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1733 = None
	        convert_element_type_1154 = torch.ops.prims.convert_element_type.default(convert_element_type_1153, torch.float32);  convert_element_type_1153 = None
	        _assert_tensor_metadata_1734 = torch.ops.aten._assert_tensor_metadata.default(view_3016, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1734 = None
	        convert_element_type_1155 = torch.ops.prims.convert_element_type.default(view_3016, torch.float32);  view_3016 = None
	        sub_8813 = torch.ops.aten.sub.Tensor(convert_element_type_1154, convert_element_type_1155);  convert_element_type_1154 = convert_element_type_1155 = None
	        mul_18634 = torch.ops.aten.mul.Tensor(sub_8813, view_3015);  sub_8813 = view_3015 = None
	        _assert_tensor_metadata_1735 = torch.ops.aten._assert_tensor_metadata.default(mul_18634, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1735 = None
	        view_3018 = torch.ops.aten.view.default(model_multi_modal_projector_linear_1_parametrizations_weight_original0, [3072, 160, 32]);  model_multi_modal_projector_linear_1_parametrizations_weight_original0 = None
	        view_3019 = torch.ops.aten.view.default(model_multi_modal_projector_linear_1_parametrizations_weight_original1, [3072, 160, 1]);  model_multi_modal_projector_linear_1_parametrizations_weight_original1 = None
	        view_3020 = torch.ops.aten.view.default(model_multi_modal_projector_linear_1_parametrizations_weight_original2, [3072, 160, 1]);  model_multi_modal_projector_linear_1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1736 = torch.ops.aten._assert_tensor_metadata.default(view_3018, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1736 = None
	        convert_element_type_1156 = torch.ops.prims.convert_element_type.default(view_3018, torch.float32);  view_3018 = None
	        _assert_tensor_metadata_1737 = torch.ops.aten._assert_tensor_metadata.default(view_3020, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1737 = None
	        convert_element_type_1157 = torch.ops.prims.convert_element_type.default(view_3020, torch.float32);  view_3020 = None
	        sub_8817 = torch.ops.aten.sub.Tensor(convert_element_type_1156, convert_element_type_1157);  convert_element_type_1156 = convert_element_type_1157 = None
	        mul_18639 = torch.ops.aten.mul.Tensor(sub_8817, view_3019);  sub_8817 = view_3019 = None
	        view_3021 = torch.ops.aten.view.default(mul_18639, [3072, 5120]);  mul_18639 = None
	        _assert_tensor_metadata_1738 = torch.ops.aten._assert_tensor_metadata.default(view_3021, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1738 = None
	        permute_321 = torch.ops.aten.permute.default(view_3021, [1, 0]);  view_3021 = None
	        mm_32 = torch.ops.aten.mm.default(mul_18634, permute_321);  mul_18634 = permute_321 = None
	        mul_18642 = torch.ops.aten.mul.Tensor(mm_32, 0.5)
	        mul_18643 = torch.ops.aten.mul.Tensor(mm_32, 0.7071067811865476);  mm_32 = None
	        erf_34 = torch.ops.aten.erf.default(mul_18643);  mul_18643 = None
	        add_29516 = torch.ops.aten.add.Tensor(erf_34, 1);  erf_34 = None
	        mul_18644 = torch.ops.aten.mul.Tensor(mul_18642, add_29516);  mul_18642 = add_29516 = None
	        amin_193 = torch.ops.aten.amin.default(mul_18644, [1])
	        amax_193 = torch.ops.aten.amax.default(mul_18644, [1])
	        full_386 = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_193 = torch.ops.aten.minimum.default(amin_193, full_386);  amin_193 = full_386 = None
	        full_387 = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_193 = torch.ops.aten.maximum.default(amax_193, full_387);  amax_193 = full_387 = None
	        sub_8827 = torch.ops.aten.sub.Tensor(maximum_193, minimum_193);  maximum_193 = None
	        div_386 = torch.ops.aten.div.Tensor(sub_8827, 255.0);  sub_8827 = None
	        clamp_min_579 = torch.ops.aten.clamp_min.default(div_386, 1.1920928955078125e-07);  div_386 = None
	        div_387 = torch.ops.aten.div.Tensor(minimum_193, clamp_min_579);  minimum_193 = None
	        round_387 = torch.ops.aten.round.default(div_387);  div_387 = None
	        sub_8833 = torch.ops.aten.sub.Tensor(-128, round_387);  round_387 = None
	        clamp_min_580 = torch.ops.aten.clamp_min.default(sub_8833, -128);  sub_8833 = None
	        clamp_max_386 = torch.ops.aten.clamp_max.default(clamp_min_580, 127);  clamp_min_580 = None
	        _assert_tensor_metadata_1739 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_579, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1739 = None
	        _assert_tensor_metadata_1740 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_386, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1740 = None
	        convert_element_type_1158 = torch.ops.prims.convert_element_type.default(clamp_max_386, torch.int8);  clamp_max_386 = None
	        view_3024 = torch.ops.aten.view.default(clamp_min_579, [sym_size_int_193, 1])
	        view_3025 = torch.ops.aten.view.default(convert_element_type_1158, [sym_size_int_193, 1])
	        reciprocal_193 = torch.ops.aten.reciprocal.default(view_3024);  view_3024 = None
	        mul_18666 = torch.ops.aten.mul.Tensor(reciprocal_193, 1.0);  reciprocal_193 = None
	        mul_18668 = torch.ops.aten.mul.Tensor(mul_18644, mul_18666);  mul_18644 = mul_18666 = None
	        round_388 = torch.ops.aten.round.default(mul_18668);  mul_18668 = None
	        add_29572 = torch.ops.aten.add.Tensor(round_388, view_3025);  round_388 = view_3025 = None
	        clamp_min_581 = torch.ops.aten.clamp_min.default(add_29572, -128);  add_29572 = None
	        clamp_max_387 = torch.ops.aten.clamp_max.default(clamp_min_581, 127);  clamp_min_581 = None
	        _assert_tensor_metadata_1741 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_387, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1741 = None
	        convert_element_type_1159 = torch.ops.prims.convert_element_type.default(clamp_max_387, torch.int8);  clamp_max_387 = None
	        view_3028 = torch.ops.aten.view.default(clamp_min_579, [sym_size_int_193, 1]);  clamp_min_579 = None
	        view_3029 = torch.ops.aten.view.default(convert_element_type_1158, [sym_size_int_193, 1]);  convert_element_type_1158 = sym_size_int_193 = None
	        _assert_tensor_metadata_1742 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1159, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1742 = None
	        convert_element_type_1160 = torch.ops.prims.convert_element_type.default(convert_element_type_1159, torch.float32);  convert_element_type_1159 = None
	        _assert_tensor_metadata_1743 = torch.ops.aten._assert_tensor_metadata.default(view_3029, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1743 = None
	        convert_element_type_1161 = torch.ops.prims.convert_element_type.default(view_3029, torch.float32);  view_3029 = None
	        sub_8853 = torch.ops.aten.sub.Tensor(convert_element_type_1160, convert_element_type_1161);  convert_element_type_1160 = convert_element_type_1161 = None
	        mul_18687 = torch.ops.aten.mul.Tensor(sub_8853, view_3028);  sub_8853 = view_3028 = None
	        _assert_tensor_metadata_1744 = torch.ops.aten._assert_tensor_metadata.default(mul_18687, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1744 = None
	        view_3031 = torch.ops.aten.view.default(model_multi_modal_projector_linear_2_parametrizations_weight_original0, [3072, 96, 32]);  model_multi_modal_projector_linear_2_parametrizations_weight_original0 = None
	        view_3032 = torch.ops.aten.view.default(model_multi_modal_projector_linear_2_parametrizations_weight_original1, [3072, 96, 1]);  model_multi_modal_projector_linear_2_parametrizations_weight_original1 = None
	        view_3033 = torch.ops.aten.view.default(model_multi_modal_projector_linear_2_parametrizations_weight_original2, [3072, 96, 1]);  model_multi_modal_projector_linear_2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1745 = torch.ops.aten._assert_tensor_metadata.default(view_3031, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1745 = None
	        convert_element_type_1162 = torch.ops.prims.convert_element_type.default(view_3031, torch.float32);  view_3031 = None
	        _assert_tensor_metadata_1746 = torch.ops.aten._assert_tensor_metadata.default(view_3033, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1746 = None
	        convert_element_type_1163 = torch.ops.prims.convert_element_type.default(view_3033, torch.float32);  view_3033 = None
	        sub_8857 = torch.ops.aten.sub.Tensor(convert_element_type_1162, convert_element_type_1163);  convert_element_type_1162 = convert_element_type_1163 = None
	        mul_18692 = torch.ops.aten.mul.Tensor(sub_8857, view_3032);  sub_8857 = view_3032 = None
	        view_3034 = torch.ops.aten.view.default(mul_18692, [3072, 3072]);  mul_18692 = None
	        _assert_tensor_metadata_1747 = torch.ops.aten._assert_tensor_metadata.default(view_3034, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1747 = None
	        permute_322 = torch.ops.aten.permute.default(view_3034, [1, 0]);  view_3034 = None
	        mm_33 = torch.ops.aten.mm.default(mul_18687, permute_322);  mul_18687 = permute_322 = None
	        unsqueeze = torch.ops.aten.unsqueeze.default(mm_33, 0);  mm_33 = None
	        return (unsqueeze,)
	        
	def load_args(reader):
	    buf0 = reader.storage(None, 1536000*s6, device=device(type='cuda', index=0))
	    reader.tensor(buf0, (s6, 128, 3000), is_leaf=True)  # arg877_1
	load_args._version = 0
	mod = Repro()
	if __name__ == '__main__':
	    from torch._dynamo.repro.after_aot import run_repro
	    with torch.no_grad():
	        run_repro(mod, load_args, accuracy=False, command='run', save_dir=None, tracing_mode='symbolic', check_str=None)
	        # To run it separately, do 
	        # mod, args = run_repro(mod, load_args, accuracy=False, command='get_args', save_dir=None, tracing_mode='symbolic', check_str=None)
	        # mod(*args)
V0910 09:42:47.010000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8e3f8f855389beb0d6b1f0e4f356cb7c"}
	{
	"name": "additional_fake_tensor_prop",
	"ts": 1757522567010360.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:56.080000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a3dde9268f8224a4b9b72f79df52ec7e"}
	{
	"name": "additional_fake_tensor_prop",
	"ts": 1757522576080064.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:42:56.708000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_inductor/compile_fx.py:1264] {"artifact": {"name": "before_post_grad_graph", "encoding": "string"}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2657, "name": "compile_fx", "filename": 19, "loc": "return inference_compiler(unlifted_gm, example_inputs_)"}, {"line": 1267, "name": "__call__", "filename": 33, "loc": "return self.compiler_fn(gm, example_inputs)"}, {"line": 2543, "name": "fw_compiler_base", "filename": 19, "loc": "return compile_fx_forward("}, {"line": 2260, "name": "compile_fx_forward", "filename": 19, "loc": "return inner_compile("}, {"line": 81, "name": "inner", "filename": 34, "loc": "return func(*args, **kwds)"}, {"line": 781, "name": "compile_fx_inner", "filename": 19, "loc": "return wrap_compiler_debug(_compile_fx_inner, compiler_name=\"inductor\")("}, {"line": 144, "name": "debug_wrapper", "filename": 35, "loc": "inner_compiled_fn = compiler_fn(gm, example_inputs)"}, {"line": 167, "name": "newFunction", "filename": 36, "loc": "return old_func(*args, **kwargs)"}, {"line": 962, "name": "_compile_fx_inner", "filename": 19, "loc": "mb_compiled_graph = fx_codegen_and_compile("}, {"line": 1694, "name": "fx_codegen_and_compile", "filename": 19, "loc": "return scheme.codegen_and_compile(gm, example_inputs, inputs_to_check, graph_kwargs)"}, {"line": 1264, "name": "codegen_and_compile", "filename": 19, "loc": "trace_structured("}], "has_payload": "c6e8885cb9368bbdb1752733c849484f"}
	class <lambda>(torch.nn.Module):
	    def forward(self):
	        arg877_1: "f32[s6, 128, 3000][384000, 3000, 1]cuda:0"; 
	    
	        arg877_1, = fx_pytree.tree_flatten_spec([], self._in_spec)
	        # No stacktrace found for following nodes
	        model_audio_tower_embed_positions_weight: "f32[1500, 1280][1280, 1]cuda:0" = self.model.audio_tower.embed_positions.weight
	        model_audio_tower_conv1_weight: "f32[1280, 128, 3][384, 3, 1]cuda:0" = self.model.audio_tower.conv1.weight
	        model_audio_tower_conv1_bias: "f32[1280][1]cuda:0" = self.model.audio_tower.conv1.bias
	        model_audio_tower_conv2_weight: "f32[1280, 1280, 3][3840, 3, 1]cuda:0" = self.model.audio_tower.conv2.weight
	        model_audio_tower_conv2_bias: "f32[1280][1]cuda:0" = self.model.audio_tower.conv2.bias
	        model_audio_tower_layers_0_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn_layer_norm.weight
	        model_audio_tower_layers_0_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn_layer_norm.bias
	        model_audio_tower_layers_0_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.bias
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.bias
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.bias
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").final_layer_norm.weight
	        model_audio_tower_layers_0_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").final_layer_norm.bias
	        model_audio_tower_layers_0_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.bias
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_0_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.bias
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn_layer_norm.weight
	        model_audio_tower_layers_1_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn_layer_norm.bias
	        model_audio_tower_layers_1_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.bias
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.bias
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.bias
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").final_layer_norm.weight
	        model_audio_tower_layers_1_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").final_layer_norm.bias
	        model_audio_tower_layers_1_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.bias
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_1_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.bias
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn_layer_norm.weight
	        model_audio_tower_layers_2_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn_layer_norm.bias
	        model_audio_tower_layers_2_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.bias
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.bias
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.bias
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").final_layer_norm.weight
	        model_audio_tower_layers_2_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").final_layer_norm.bias
	        model_audio_tower_layers_2_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.bias
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_2_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.bias
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn_layer_norm.weight
	        model_audio_tower_layers_3_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn_layer_norm.bias
	        model_audio_tower_layers_3_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.bias
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.bias
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.bias
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").final_layer_norm.weight
	        model_audio_tower_layers_3_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").final_layer_norm.bias
	        model_audio_tower_layers_3_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.bias
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_3_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.bias
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn_layer_norm.weight
	        model_audio_tower_layers_4_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn_layer_norm.bias
	        model_audio_tower_layers_4_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.bias
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.bias
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.bias
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").final_layer_norm.weight
	        model_audio_tower_layers_4_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").final_layer_norm.bias
	        model_audio_tower_layers_4_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.bias
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_4_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.bias
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn_layer_norm.weight
	        model_audio_tower_layers_5_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn_layer_norm.bias
	        model_audio_tower_layers_5_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.bias
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.bias
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.bias
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").final_layer_norm.weight
	        model_audio_tower_layers_5_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").final_layer_norm.bias
	        model_audio_tower_layers_5_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.bias
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_5_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.bias
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn_layer_norm.weight
	        model_audio_tower_layers_6_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn_layer_norm.bias
	        model_audio_tower_layers_6_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.bias
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.bias
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.bias
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").final_layer_norm.weight
	        model_audio_tower_layers_6_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").final_layer_norm.bias
	        model_audio_tower_layers_6_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.bias
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_6_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.bias
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn_layer_norm.weight
	        model_audio_tower_layers_7_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn_layer_norm.bias
	        model_audio_tower_layers_7_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.bias
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.bias
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.bias
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").final_layer_norm.weight
	        model_audio_tower_layers_7_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").final_layer_norm.bias
	        model_audio_tower_layers_7_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.bias
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_7_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.bias
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn_layer_norm.weight
	        model_audio_tower_layers_8_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn_layer_norm.bias
	        model_audio_tower_layers_8_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.bias
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.bias
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.bias
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").final_layer_norm.weight
	        model_audio_tower_layers_8_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").final_layer_norm.bias
	        model_audio_tower_layers_8_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.bias
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_8_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.bias
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn_layer_norm.weight
	        model_audio_tower_layers_9_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn_layer_norm.bias
	        model_audio_tower_layers_9_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.bias
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.bias
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.bias
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").final_layer_norm.weight
	        model_audio_tower_layers_9_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").final_layer_norm.bias
	        model_audio_tower_layers_9_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.bias
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_9_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.bias
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn_layer_norm.weight
	        model_audio_tower_layers_10_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn_layer_norm.bias
	        model_audio_tower_layers_10_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.bias
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.bias
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.bias
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").final_layer_norm.weight
	        model_audio_tower_layers_10_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").final_layer_norm.bias
	        model_audio_tower_layers_10_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.bias
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_10_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.bias
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn_layer_norm.weight
	        model_audio_tower_layers_11_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn_layer_norm.bias
	        model_audio_tower_layers_11_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.bias
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.bias
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.bias
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").final_layer_norm.weight
	        model_audio_tower_layers_11_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").final_layer_norm.bias
	        model_audio_tower_layers_11_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.bias
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_11_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.bias
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn_layer_norm.weight
	        model_audio_tower_layers_12_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn_layer_norm.bias
	        model_audio_tower_layers_12_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.bias
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.bias
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.bias
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").final_layer_norm.weight
	        model_audio_tower_layers_12_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").final_layer_norm.bias
	        model_audio_tower_layers_12_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.bias
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_12_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.bias
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn_layer_norm.weight
	        model_audio_tower_layers_13_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn_layer_norm.bias
	        model_audio_tower_layers_13_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.bias
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.bias
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.bias
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").final_layer_norm.weight
	        model_audio_tower_layers_13_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").final_layer_norm.bias
	        model_audio_tower_layers_13_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.bias
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_13_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.bias
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn_layer_norm.weight
	        model_audio_tower_layers_14_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn_layer_norm.bias
	        model_audio_tower_layers_14_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.bias
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.bias
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.bias
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").final_layer_norm.weight
	        model_audio_tower_layers_14_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").final_layer_norm.bias
	        model_audio_tower_layers_14_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.bias
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_14_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.bias
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn_layer_norm.weight
	        model_audio_tower_layers_15_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn_layer_norm.bias
	        model_audio_tower_layers_15_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.bias
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.bias
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.bias
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").final_layer_norm.weight
	        model_audio_tower_layers_15_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").final_layer_norm.bias
	        model_audio_tower_layers_15_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.bias
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_15_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.bias
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn_layer_norm.weight
	        model_audio_tower_layers_16_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn_layer_norm.bias
	        model_audio_tower_layers_16_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.bias
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.bias
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.bias
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").final_layer_norm.weight
	        model_audio_tower_layers_16_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").final_layer_norm.bias
	        model_audio_tower_layers_16_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.bias
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_16_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.bias
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn_layer_norm.weight
	        model_audio_tower_layers_17_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn_layer_norm.bias
	        model_audio_tower_layers_17_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.bias
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.bias
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.bias
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").final_layer_norm.weight
	        model_audio_tower_layers_17_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").final_layer_norm.bias
	        model_audio_tower_layers_17_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.bias
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_17_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.bias
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn_layer_norm.weight
	        model_audio_tower_layers_18_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn_layer_norm.bias
	        model_audio_tower_layers_18_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.bias
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.bias
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.bias
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").final_layer_norm.weight
	        model_audio_tower_layers_18_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").final_layer_norm.bias
	        model_audio_tower_layers_18_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.bias
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_18_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.bias
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn_layer_norm.weight
	        model_audio_tower_layers_19_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn_layer_norm.bias
	        model_audio_tower_layers_19_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.bias
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.bias
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.bias
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").final_layer_norm.weight
	        model_audio_tower_layers_19_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").final_layer_norm.bias
	        model_audio_tower_layers_19_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.bias
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_19_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.bias
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn_layer_norm.weight
	        model_audio_tower_layers_20_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn_layer_norm.bias
	        model_audio_tower_layers_20_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.bias
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.bias
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.bias
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").final_layer_norm.weight
	        model_audio_tower_layers_20_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").final_layer_norm.bias
	        model_audio_tower_layers_20_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.bias
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_20_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.bias
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn_layer_norm.weight
	        model_audio_tower_layers_21_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn_layer_norm.bias
	        model_audio_tower_layers_21_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.bias
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.bias
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.bias
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").final_layer_norm.weight
	        model_audio_tower_layers_21_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").final_layer_norm.bias
	        model_audio_tower_layers_21_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.bias
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_21_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.bias
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn_layer_norm.weight
	        model_audio_tower_layers_22_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn_layer_norm.bias
	        model_audio_tower_layers_22_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.bias
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.bias
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.bias
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").final_layer_norm.weight
	        model_audio_tower_layers_22_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").final_layer_norm.bias
	        model_audio_tower_layers_22_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.bias
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_22_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.bias
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn_layer_norm.weight
	        model_audio_tower_layers_23_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn_layer_norm.bias
	        model_audio_tower_layers_23_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.bias
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.bias
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.bias
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").final_layer_norm.weight
	        model_audio_tower_layers_23_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").final_layer_norm.bias
	        model_audio_tower_layers_23_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.bias
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_23_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.bias
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn_layer_norm.weight
	        model_audio_tower_layers_24_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn_layer_norm.bias
	        model_audio_tower_layers_24_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.bias
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.bias
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.bias
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").final_layer_norm.weight
	        model_audio_tower_layers_24_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").final_layer_norm.bias
	        model_audio_tower_layers_24_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.bias
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_24_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.bias
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn_layer_norm.weight
	        model_audio_tower_layers_25_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn_layer_norm.bias
	        model_audio_tower_layers_25_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.bias
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.bias
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.bias
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").final_layer_norm.weight
	        model_audio_tower_layers_25_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").final_layer_norm.bias
	        model_audio_tower_layers_25_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.bias
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_25_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.bias
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn_layer_norm.weight
	        model_audio_tower_layers_26_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn_layer_norm.bias
	        model_audio_tower_layers_26_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.bias
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.bias
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.bias
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").final_layer_norm.weight
	        model_audio_tower_layers_26_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").final_layer_norm.bias
	        model_audio_tower_layers_26_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.bias
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_26_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.bias
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn_layer_norm.weight
	        model_audio_tower_layers_27_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn_layer_norm.bias
	        model_audio_tower_layers_27_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.bias
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.bias
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.bias
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").final_layer_norm.weight
	        model_audio_tower_layers_27_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").final_layer_norm.bias
	        model_audio_tower_layers_27_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.bias
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_27_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.bias
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn_layer_norm.weight
	        model_audio_tower_layers_28_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn_layer_norm.bias
	        model_audio_tower_layers_28_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.bias
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.bias
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.bias
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").final_layer_norm.weight
	        model_audio_tower_layers_28_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").final_layer_norm.bias
	        model_audio_tower_layers_28_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.bias
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_28_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.bias
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn_layer_norm.weight
	        model_audio_tower_layers_29_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn_layer_norm.bias
	        model_audio_tower_layers_29_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.bias
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.bias
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.bias
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").final_layer_norm.weight
	        model_audio_tower_layers_29_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").final_layer_norm.bias
	        model_audio_tower_layers_29_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.bias
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_29_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.bias
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn_layer_norm.weight
	        model_audio_tower_layers_30_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn_layer_norm.bias
	        model_audio_tower_layers_30_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.bias
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.bias
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.bias
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").final_layer_norm.weight
	        model_audio_tower_layers_30_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").final_layer_norm.bias
	        model_audio_tower_layers_30_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.bias
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_30_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.bias
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn_layer_norm.weight
	        model_audio_tower_layers_31_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn_layer_norm.bias
	        model_audio_tower_layers_31_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.bias
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.bias
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.bias
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").final_layer_norm.weight
	        model_audio_tower_layers_31_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").final_layer_norm.bias
	        model_audio_tower_layers_31_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.bias
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_31_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.bias
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original2
	        model_audio_tower_layer_norm_weight: "f32[1280][1]cuda:0" = self.model.audio_tower.layer_norm.weight
	        model_audio_tower_layer_norm_bias: "f32[1280][1]cuda:0" = self.model.audio_tower.layer_norm.bias
	        model_multi_modal_projector_linear_1_parametrizations_weight_original0: "i8[3072, 5120][5120, 1]cuda:0" = self.model.multi_modal_projector.linear_1.parametrizations.weight.original0
	        model_multi_modal_projector_linear_1_parametrizations_weight_original1: "f32[3072, 160][160, 1]cuda:0" = self.model.multi_modal_projector.linear_1.parametrizations.weight.original1
	        model_multi_modal_projector_linear_1_parametrizations_weight_original2: "i8[3072, 160][160, 1]cuda:0" = self.model.multi_modal_projector.linear_1.parametrizations.weight.original2
	        model_multi_modal_projector_linear_2_parametrizations_weight_original0: "i8[3072, 3072][3072, 1]cuda:0" = self.model.multi_modal_projector.linear_2.parametrizations.weight.original0
	        model_multi_modal_projector_linear_2_parametrizations_weight_original1: "f32[3072, 96][96, 1]cuda:0" = self.model.multi_modal_projector.linear_2.parametrizations.weight.original1
	        model_multi_modal_projector_linear_2_parametrizations_weight_original2: "i8[3072, 96][96, 1]cuda:0" = self.model.multi_modal_projector.linear_2.parametrizations.weight.original2
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:348 in forward, code: input_features = input_features.to(dtype=self.conv1.weight.dtype, device=self.conv1.weight.device)
	        _assert_tensor_metadata = torch.ops.aten._assert_tensor_metadata.default(arg877_1, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:349 in forward, code: inputs_embeds = nn.functional.gelu(self.conv1(input_features))
	        convolution: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.convolution.default(arg877_1, model_audio_tower_conv1_weight, model_audio_tower_conv1_bias, [1], [1], [1], False, [0], 1);  model_audio_tower_conv1_weight = model_audio_tower_conv1_bias = None
	        sym_size_int: "Sym(s6)" = torch.ops.aten.sym_size.int(arg877_1, 0);  arg877_1 = None
	        mul_2: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution, 0.5)
	        mul_3: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution, 0.7071067811865476);  convolution = None
	        erf: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.erf.default(mul_3);  mul_3 = None
	        add_4: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.add.Tensor(erf, 1);  erf = None
	        mul_4: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2, add_4);  mul_2 = add_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:350 in forward, code: inputs_embeds = nn.functional.gelu(self.conv2(inputs_embeds))
	        convolution_1: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.convolution.default(mul_4, model_audio_tower_conv2_weight, model_audio_tower_conv2_bias, [2], [1], [1], False, [0], 1);  mul_4 = model_audio_tower_conv2_weight = model_audio_tower_conv2_bias = None
	        mul_9: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution_1, 0.5)
	        mul_10: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution_1, 0.7071067811865476);  convolution_1 = None
	        erf_1: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.erf.default(mul_10);  mul_10 = None
	        add_13: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_1, 1);  erf_1 = None
	        mul_11: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9, add_13);  mul_9 = add_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:351 in forward, code: inputs_embeds = inputs_embeds.permute(0, 2, 1)
	        permute: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.permute.default(mul_11, [0, 2, 1]);  mul_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:354 in forward, code: hidden_states = (inputs_embeds + embed_pos).to(inputs_embeds.dtype)
	        add_22: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(permute, model_audio_tower_embed_positions_weight);  permute = model_audio_tower_embed_positions_weight = None
	        _assert_tensor_metadata_1 = torch.ops.aten._assert_tensor_metadata.default(add_22, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_22, memory_format = torch.contiguous_format)
	        var_mean = torch.ops.aten.var_mean.correction(clone_1, [2], correction = 0, keepdim = True)
	        getitem: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean[0]
	        getitem_1: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean[1];  var_mean = None
	        add_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem, 1e-05);  getitem = None
	        rsqrt: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_31);  add_31 = None
	        sub_7: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_1, getitem_1);  clone_1 = getitem_1 = None
	        mul_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7, rsqrt);  sub_7 = rsqrt = None
	        mul_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_20, model_audio_tower_layers_0_self_attn_layer_norm_weight);  mul_20 = model_audio_tower_layers_0_self_attn_layer_norm_weight = None
	        add_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_21, model_audio_tower_layers_0_self_attn_layer_norm_bias);  mul_21 = model_audio_tower_layers_0_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_32, [2])
	        amax: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_32, [2])
	        full: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin, full);  amin = full = None
	        full_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax, full_1);  amax = full_1 = None
	        sub_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum, minimum);  maximum = None
	        div: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_18, 255.0);  sub_18 = None
	        clamp_min: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div, 1.1920928955078125e-07);  div = None
	        div_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum, clamp_min);  minimum = None
	        round_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_1);  div_1 = None
	        sub_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_1);  round_1 = None
	        clamp_min_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_24, -128);  sub_24 = None
	        clamp_max: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_1, 127);  clamp_min_1 = None
	        _assert_tensor_metadata_2 = torch.ops.aten._assert_tensor_metadata.default(clamp_min, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_2 = None
	        _assert_tensor_metadata_3 = torch.ops.aten._assert_tensor_metadata.default(clamp_max, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_3 = None
	        convert_element_type: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max, torch.int8);  clamp_max = None
	        view_2: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min, [sym_size_int, 1500, 1])
	        view_3: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type, [sym_size_int, 1500, 1])
	        reciprocal: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2);  view_2 = None
	        mul_69: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal, 1.0);  reciprocal = None
	        mul_72: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_32, mul_69);  mul_69 = None
	        round_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_72);  mul_72 = None
	        add_119: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_2, view_3);  round_2 = view_3 = None
	        clamp_min_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_119, -128);  add_119 = None
	        clamp_max_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_2, 127);  clamp_min_2 = None
	        _assert_tensor_metadata_4 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_1, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_4 = None
	        convert_element_type_1: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_1, torch.int8);  clamp_max_1 = None
	        view_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min, [sym_size_int, 1500, 1]);  clamp_min = None
	        view_7: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type, [sym_size_int, 1500, 1]);  convert_element_type = None
	        _assert_tensor_metadata_5 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_5 = None
	        convert_element_type_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1, torch.float32);  convert_element_type_1 = None
	        _assert_tensor_metadata_6 = torch.ops.aten._assert_tensor_metadata.default(view_7, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_6 = None
	        convert_element_type_3: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_7, torch.float32);  view_7 = None
	        sub_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_2, convert_element_type_3);  convert_element_type_2 = convert_element_type_3 = None
	        mul_94: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_44, view_6);  sub_44 = view_6 = None
	        _assert_tensor_metadata_7 = torch.ops.aten._assert_tensor_metadata.default(mul_94, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_7 = None
	        view_9: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_10: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_11: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_8 = torch.ops.aten._assert_tensor_metadata.default(view_9, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_8 = None
	        convert_element_type_4: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_9, torch.float32);  view_9 = None
	        _assert_tensor_metadata_9 = torch.ops.aten._assert_tensor_metadata.default(view_11, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_9 = None
	        convert_element_type_5: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_11, torch.float32);  view_11 = None
	        sub_48: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_4, convert_element_type_5);  convert_element_type_4 = convert_element_type_5 = None
	        mul_99: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_48, view_10);  sub_48 = view_10 = None
	        view_12: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_99, [1280, 1280]);  mul_99 = None
	        _assert_tensor_metadata_10 = torch.ops.aten._assert_tensor_metadata.default(view_12, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_10 = None
	        mul_104: "Sym(1500*s6)" = sym_size_int * 1500
	        view_13: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_94, [mul_104, 1280]);  mul_94 = mul_104 = None
	        permute_1: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_12, [1, 0]);  view_12 = None
	        addmm: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_0_self_attn_q_proj_bias, view_13, permute_1);  model_audio_tower_layers_0_self_attn_q_proj_bias = view_13 = permute_1 = None
	        view_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm, [sym_size_int, 1500, 1280]);  addmm = None
	        mul_111: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_14, 0.125);  view_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_15: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_111, [sym_size_int, 1500, 20, 64]);  mul_111 = None
	        permute_2: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_15, [0, 2, 1, 3]);  view_15 = None
	        clone_2: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_2, memory_format = torch.contiguous_format);  permute_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_32, [2])
	        amax_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_32, [2])
	        full_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_1, full_2);  amin_1 = full_2 = None
	        full_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_1, full_3);  amax_1 = full_3 = None
	        sub_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_1, minimum_1);  maximum_1 = None
	        div_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_63, 255.0);  sub_63 = None
	        clamp_min_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_2, 1.1920928955078125e-07);  div_2 = None
	        div_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_1, clamp_min_3);  minimum_1 = None
	        round_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_3);  div_3 = None
	        sub_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_3);  round_3 = None
	        clamp_min_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_69, -128);  sub_69 = None
	        clamp_max_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_4, 127);  clamp_min_4 = None
	        _assert_tensor_metadata_11 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_3, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_11 = None
	        _assert_tensor_metadata_12 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_2, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_12 = None
	        convert_element_type_6: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_2, torch.int8);  clamp_max_2 = None
	        view_18: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_3, [sym_size_int, 1500, 1])
	        view_19: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_6, [sym_size_int, 1500, 1])
	        reciprocal_1: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_18);  view_18 = None
	        mul_165: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_1, 1.0);  reciprocal_1 = None
	        mul_168: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_32, mul_165);  mul_165 = None
	        round_4: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_168);  mul_168 = None
	        add_271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_4, view_19);  round_4 = view_19 = None
	        clamp_min_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_271, -128);  add_271 = None
	        clamp_max_3: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_5, 127);  clamp_min_5 = None
	        _assert_tensor_metadata_13 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_3, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_13 = None
	        convert_element_type_7: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_3, torch.int8);  clamp_max_3 = None
	        view_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_3, [sym_size_int, 1500, 1]);  clamp_min_3 = None
	        view_23: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_6, [sym_size_int, 1500, 1]);  convert_element_type_6 = None
	        _assert_tensor_metadata_14 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_7, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_14 = None
	        convert_element_type_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_7, torch.float32);  convert_element_type_7 = None
	        _assert_tensor_metadata_15 = torch.ops.aten._assert_tensor_metadata.default(view_23, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_15 = None
	        convert_element_type_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_23, torch.float32);  view_23 = None
	        sub_89: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_8, convert_element_type_9);  convert_element_type_8 = convert_element_type_9 = None
	        mul_190: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_89, view_22);  sub_89 = view_22 = None
	        _assert_tensor_metadata_16 = torch.ops.aten._assert_tensor_metadata.default(mul_190, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_16 = None
	        view_25: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_26: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_27: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_17 = torch.ops.aten._assert_tensor_metadata.default(view_25, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_17 = None
	        convert_element_type_10: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_25, torch.float32);  view_25 = None
	        _assert_tensor_metadata_18 = torch.ops.aten._assert_tensor_metadata.default(view_27, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_18 = None
	        convert_element_type_11: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_27, torch.float32);  view_27 = None
	        sub_93: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_10, convert_element_type_11);  convert_element_type_10 = convert_element_type_11 = None
	        mul_195: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_93, view_26);  sub_93 = view_26 = None
	        view_28: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_195, [1280, 1280]);  mul_195 = None
	        _assert_tensor_metadata_19 = torch.ops.aten._assert_tensor_metadata.default(view_28, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_19 = None
	        permute_3: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_28, [1, 0]);  view_28 = None
	        mul_198: "Sym(1500*s6)" = sym_size_int * 1500
	        view_29: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_190, [mul_198, 1280]);  mul_190 = mul_198 = None
	        mm: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_29, permute_3);  view_29 = permute_3 = None
	        view_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm, [sym_size_int, 1500, 1280]);  mm = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_31: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_30, [sym_size_int, -1, 20, 64]);  view_30 = None
	        permute_4: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_31, [0, 2, 1, 3]);  view_31 = None
	        clone_3: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_4, memory_format = torch.contiguous_format);  permute_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_32, [2])
	        amax_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_32, [2])
	        full_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_2, full_4);  amin_2 = full_4 = None
	        full_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_2, full_5);  amax_2 = full_5 = None
	        sub_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_2, minimum_2);  maximum_2 = None
	        div_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_107, 255.0);  sub_107 = None
	        clamp_min_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_4, 1.1920928955078125e-07);  div_4 = None
	        div_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_2, clamp_min_6);  minimum_2 = None
	        round_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_5);  div_5 = None
	        sub_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_5);  round_5 = None
	        clamp_min_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_113, -128);  sub_113 = None
	        clamp_max_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_7, 127);  clamp_min_7 = None
	        _assert_tensor_metadata_20 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_6, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_20 = None
	        _assert_tensor_metadata_21 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_4, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_21 = None
	        convert_element_type_12: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_4, torch.int8);  clamp_max_4 = None
	        view_34: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_6, [sym_size_int, 1500, 1])
	        view_35: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_12, [sym_size_int, 1500, 1])
	        reciprocal_2: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_34);  view_34 = None
	        mul_264: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_2, 1.0);  reciprocal_2 = None
	        mul_267: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_32, mul_264);  add_32 = mul_264 = None
	        round_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_267);  mul_267 = None
	        add_419: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_6, view_35);  round_6 = view_35 = None
	        clamp_min_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_419, -128);  add_419 = None
	        clamp_max_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_8, 127);  clamp_min_8 = None
	        _assert_tensor_metadata_22 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_5, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_22 = None
	        convert_element_type_13: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_5, torch.int8);  clamp_max_5 = None
	        view_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_6, [sym_size_int, 1500, 1]);  clamp_min_6 = None
	        view_39: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_12, [sym_size_int, 1500, 1]);  convert_element_type_12 = None
	        _assert_tensor_metadata_23 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_13, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_23 = None
	        convert_element_type_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_13, torch.float32);  convert_element_type_13 = None
	        _assert_tensor_metadata_24 = torch.ops.aten._assert_tensor_metadata.default(view_39, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_24 = None
	        convert_element_type_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_39, torch.float32);  view_39 = None
	        sub_133: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_14, convert_element_type_15);  convert_element_type_14 = convert_element_type_15 = None
	        mul_289: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_133, view_38);  sub_133 = view_38 = None
	        _assert_tensor_metadata_25 = torch.ops.aten._assert_tensor_metadata.default(mul_289, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_25 = None
	        view_41: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_42: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_43: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_26 = torch.ops.aten._assert_tensor_metadata.default(view_41, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_26 = None
	        convert_element_type_16: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_41, torch.float32);  view_41 = None
	        _assert_tensor_metadata_27 = torch.ops.aten._assert_tensor_metadata.default(view_43, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_27 = None
	        convert_element_type_17: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_43, torch.float32);  view_43 = None
	        sub_137: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_16, convert_element_type_17);  convert_element_type_16 = convert_element_type_17 = None
	        mul_294: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_137, view_42);  sub_137 = view_42 = None
	        view_44: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_294, [1280, 1280]);  mul_294 = None
	        _assert_tensor_metadata_28 = torch.ops.aten._assert_tensor_metadata.default(view_44, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_28 = None
	        mul_299: "Sym(1500*s6)" = sym_size_int * 1500
	        view_45: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_289, [mul_299, 1280]);  mul_289 = mul_299 = None
	        permute_5: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_44, [1, 0]);  view_44 = None
	        addmm_1: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_0_self_attn_v_proj_bias, view_45, permute_5);  model_audio_tower_layers_0_self_attn_v_proj_bias = view_45 = permute_5 = None
	        view_46: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_1, [sym_size_int, 1500, 1280]);  addmm_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_47: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_46, [sym_size_int, -1, 20, 64]);  view_46 = None
	        permute_6: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_47, [0, 2, 1, 3]);  view_47 = None
	        clone_4: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_6, memory_format = torch.contiguous_format);  permute_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_2, clone_3, clone_4, None, False, scale = 1.0);  clone_2 = clone_3 = clone_4 = None
	        getitem_2: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention[0];  _scaled_dot_product_efficient_attention = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_7: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_2, [0, 2, 1, 3]);  getitem_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_48: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_7, [sym_size_int, 1500, -1]);  permute_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_48, [2])
	        amax_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_48, [2])
	        full_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_3, full_6);  amin_3 = full_6 = None
	        full_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_3, full_7);  amax_3 = full_7 = None
	        sub_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_3, minimum_3);  maximum_3 = None
	        div_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_155, 255.0);  sub_155 = None
	        clamp_min_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_6, 1.1920928955078125e-07);  div_6 = None
	        div_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_3, clamp_min_9);  minimum_3 = None
	        round_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_7);  div_7 = None
	        sub_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_7);  round_7 = None
	        clamp_min_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_161, -128);  sub_161 = None
	        clamp_max_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_10, 127);  clamp_min_10 = None
	        _assert_tensor_metadata_29 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_9, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_29 = None
	        _assert_tensor_metadata_30 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_6, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_30 = None
	        convert_element_type_18: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_6, torch.int8);  clamp_max_6 = None
	        view_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_9, [sym_size_int, 1500, 1])
	        view_52: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_18, [sym_size_int, 1500, 1])
	        reciprocal_3: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_51);  view_51 = None
	        mul_369: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_3, 1.0);  reciprocal_3 = None
	        mul_372: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_48, mul_369);  view_48 = mul_369 = None
	        round_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_372);  mul_372 = None
	        add_583: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_8, view_52);  round_8 = view_52 = None
	        clamp_min_11: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_583, -128);  add_583 = None
	        clamp_max_7: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_11, 127);  clamp_min_11 = None
	        _assert_tensor_metadata_31 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_7, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_31 = None
	        convert_element_type_19: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_7, torch.int8);  clamp_max_7 = None
	        view_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_9, [sym_size_int, 1500, 1]);  clamp_min_9 = None
	        view_56: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_18, [sym_size_int, 1500, 1]);  convert_element_type_18 = None
	        _assert_tensor_metadata_32 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_19, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_32 = None
	        convert_element_type_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_19, torch.float32);  convert_element_type_19 = None
	        _assert_tensor_metadata_33 = torch.ops.aten._assert_tensor_metadata.default(view_56, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_33 = None
	        convert_element_type_21: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_56, torch.float32);  view_56 = None
	        sub_181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_20, convert_element_type_21);  convert_element_type_20 = convert_element_type_21 = None
	        mul_394: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_181, view_55);  sub_181 = view_55 = None
	        _assert_tensor_metadata_34 = torch.ops.aten._assert_tensor_metadata.default(mul_394, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_34 = None
	        view_58: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_59: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_60: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_35 = torch.ops.aten._assert_tensor_metadata.default(view_58, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_35 = None
	        convert_element_type_22: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_58, torch.float32);  view_58 = None
	        _assert_tensor_metadata_36 = torch.ops.aten._assert_tensor_metadata.default(view_60, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_36 = None
	        convert_element_type_23: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_60, torch.float32);  view_60 = None
	        sub_185: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_22, convert_element_type_23);  convert_element_type_22 = convert_element_type_23 = None
	        mul_399: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_185, view_59);  sub_185 = view_59 = None
	        view_61: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_399, [1280, 1280]);  mul_399 = None
	        _assert_tensor_metadata_37 = torch.ops.aten._assert_tensor_metadata.default(view_61, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_37 = None
	        mul_404: "Sym(1500*s6)" = sym_size_int * 1500
	        view_62: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_394, [mul_404, 1280]);  mul_394 = mul_404 = None
	        permute_8: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_61, [1, 0]);  view_61 = None
	        addmm_2: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_0_self_attn_out_proj_bias, view_62, permute_8);  model_audio_tower_layers_0_self_attn_out_proj_bias = view_62 = permute_8 = None
	        view_63: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_2, [sym_size_int, 1500, 1280]);  addmm_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_646: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_22, view_63);  add_22 = view_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_646, memory_format = torch.contiguous_format)
	        var_mean_1 = torch.ops.aten.var_mean.correction(clone_6, [2], correction = 0, keepdim = True)
	        getitem_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_1[0]
	        getitem_7: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_1[1];  var_mean_1 = None
	        add_651: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_6, 1e-05);  getitem_6 = None
	        rsqrt_1: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_651);  add_651 = None
	        sub_191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_6, getitem_7);  clone_6 = getitem_7 = None
	        mul_415: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_191, rsqrt_1);  sub_191 = rsqrt_1 = None
	        mul_416: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_415, model_audio_tower_layers_0_final_layer_norm_weight);  mul_415 = model_audio_tower_layers_0_final_layer_norm_weight = None
	        add_652: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_416, model_audio_tower_layers_0_final_layer_norm_bias);  mul_416 = model_audio_tower_layers_0_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_652, [2])
	        amax_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_652, [2])
	        full_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_4, full_8);  amin_4 = full_8 = None
	        full_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_4, full_9);  amax_4 = full_9 = None
	        sub_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_4, minimum_4);  maximum_4 = None
	        div_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_202, 255.0);  sub_202 = None
	        clamp_min_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_8, 1.1920928955078125e-07);  div_8 = None
	        div_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_4, clamp_min_12);  minimum_4 = None
	        round_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_9);  div_9 = None
	        sub_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_9);  round_9 = None
	        clamp_min_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_208, -128);  sub_208 = None
	        clamp_max_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_13, 127);  clamp_min_13 = None
	        _assert_tensor_metadata_38 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_12, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_38 = None
	        _assert_tensor_metadata_39 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_8, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_39 = None
	        convert_element_type_24: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_8, torch.int8);  clamp_max_8 = None
	        view_66: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_12, [sym_size_int, 1500, 1])
	        view_67: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_24, [sym_size_int, 1500, 1])
	        reciprocal_4: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_66);  view_66 = None
	        mul_464: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_4, 1.0);  reciprocal_4 = None
	        mul_467: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_652, mul_464);  add_652 = mul_464 = None
	        round_10: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_467);  mul_467 = None
	        add_739: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_10, view_67);  round_10 = view_67 = None
	        clamp_min_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_739, -128);  add_739 = None
	        clamp_max_9: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_14, 127);  clamp_min_14 = None
	        _assert_tensor_metadata_40 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_9, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_40 = None
	        convert_element_type_25: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_9, torch.int8);  clamp_max_9 = None
	        view_70: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_12, [sym_size_int, 1500, 1]);  clamp_min_12 = None
	        view_71: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_24, [sym_size_int, 1500, 1]);  convert_element_type_24 = None
	        _assert_tensor_metadata_41 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_25, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_41 = None
	        convert_element_type_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_25, torch.float32);  convert_element_type_25 = None
	        _assert_tensor_metadata_42 = torch.ops.aten._assert_tensor_metadata.default(view_71, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_42 = None
	        convert_element_type_27: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_71, torch.float32);  view_71 = None
	        sub_228: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_26, convert_element_type_27);  convert_element_type_26 = convert_element_type_27 = None
	        mul_489: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_228, view_70);  sub_228 = view_70 = None
	        _assert_tensor_metadata_43 = torch.ops.aten._assert_tensor_metadata.default(mul_489, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_43 = None
	        view_73: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_0_fc1_parametrizations_weight_original0 = None
	        view_74: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_0_fc1_parametrizations_weight_original1 = None
	        view_75: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_0_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_44 = torch.ops.aten._assert_tensor_metadata.default(view_73, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_44 = None
	        convert_element_type_28: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_73, torch.float32);  view_73 = None
	        _assert_tensor_metadata_45 = torch.ops.aten._assert_tensor_metadata.default(view_75, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_45 = None
	        convert_element_type_29: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_75, torch.float32);  view_75 = None
	        sub_232: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_28, convert_element_type_29);  convert_element_type_28 = convert_element_type_29 = None
	        mul_494: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_232, view_74);  sub_232 = view_74 = None
	        view_76: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_494, [5120, 1280]);  mul_494 = None
	        _assert_tensor_metadata_46 = torch.ops.aten._assert_tensor_metadata.default(view_76, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_46 = None
	        mul_499: "Sym(1500*s6)" = sym_size_int * 1500
	        view_77: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_489, [mul_499, 1280]);  mul_489 = mul_499 = None
	        permute_9: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_76, [1, 0]);  view_76 = None
	        addmm_3: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_0_fc1_bias, view_77, permute_9);  model_audio_tower_layers_0_fc1_bias = view_77 = permute_9 = None
	        view_78: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_3, [sym_size_int, 1500, 5120]);  addmm_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_506: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_78, 0.5)
	        mul_507: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_78, 0.7071067811865476);  view_78 = None
	        erf_2: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_507);  mul_507 = None
	        add_798: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_2, 1);  erf_2 = None
	        mul_508: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_506, add_798);  mul_506 = add_798 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_508, [2])
	        amax_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_508, [2])
	        full_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_5, full_10);  amin_5 = full_10 = None
	        full_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_5, full_11);  amax_5 = full_11 = None
	        sub_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_5, minimum_5);  maximum_5 = None
	        div_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_245, 255.0);  sub_245 = None
	        clamp_min_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_10, 1.1920928955078125e-07);  div_10 = None
	        div_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_5, clamp_min_15);  minimum_5 = None
	        round_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_11);  div_11 = None
	        sub_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_11);  round_11 = None
	        clamp_min_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_251, -128);  sub_251 = None
	        clamp_max_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_16, 127);  clamp_min_16 = None
	        _assert_tensor_metadata_47 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_15, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_47 = None
	        _assert_tensor_metadata_48 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_10, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_48 = None
	        convert_element_type_30: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_10, torch.int8);  clamp_max_10 = None
	        view_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_15, [sym_size_int, 1500, 1])
	        view_82: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_30, [sym_size_int, 1500, 1])
	        reciprocal_5: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_81);  view_81 = None
	        mul_554: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_5, 1.0);  reciprocal_5 = None
	        mul_557: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_508, mul_554);  mul_508 = mul_554 = None
	        round_12: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_557);  mul_557 = None
	        add_881: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_12, view_82);  round_12 = view_82 = None
	        clamp_min_17: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_881, -128);  add_881 = None
	        clamp_max_11: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_17, 127);  clamp_min_17 = None
	        _assert_tensor_metadata_49 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_11, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_49 = None
	        convert_element_type_31: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_11, torch.int8);  clamp_max_11 = None
	        view_85: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_15, [sym_size_int, 1500, 1]);  clamp_min_15 = None
	        view_86: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_30, [sym_size_int, 1500, 1]);  convert_element_type_30 = None
	        _assert_tensor_metadata_50 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_31, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_50 = None
	        convert_element_type_32: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_31, torch.float32);  convert_element_type_31 = None
	        _assert_tensor_metadata_51 = torch.ops.aten._assert_tensor_metadata.default(view_86, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_51 = None
	        convert_element_type_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_86, torch.float32);  view_86 = None
	        sub_271: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_32, convert_element_type_33);  convert_element_type_32 = convert_element_type_33 = None
	        mul_579: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_271, view_85);  sub_271 = view_85 = None
	        _assert_tensor_metadata_52 = torch.ops.aten._assert_tensor_metadata.default(mul_579, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_52 = None
	        view_88: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_0_fc2_parametrizations_weight_original0 = None
	        view_89: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_0_fc2_parametrizations_weight_original1 = None
	        view_90: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_0_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_53 = torch.ops.aten._assert_tensor_metadata.default(view_88, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_53 = None
	        convert_element_type_34: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_88, torch.float32);  view_88 = None
	        _assert_tensor_metadata_54 = torch.ops.aten._assert_tensor_metadata.default(view_90, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_54 = None
	        convert_element_type_35: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_90, torch.float32);  view_90 = None
	        sub_275: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_34, convert_element_type_35);  convert_element_type_34 = convert_element_type_35 = None
	        mul_584: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_275, view_89);  sub_275 = view_89 = None
	        view_91: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_584, [1280, 5120]);  mul_584 = None
	        _assert_tensor_metadata_55 = torch.ops.aten._assert_tensor_metadata.default(view_91, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_55 = None
	        mul_589: "Sym(1500*s6)" = sym_size_int * 1500
	        view_92: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_579, [mul_589, 5120]);  mul_579 = mul_589 = None
	        permute_10: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_91, [1, 0]);  view_91 = None
	        addmm_4: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_0_fc2_bias, view_92, permute_10);  model_audio_tower_layers_0_fc2_bias = view_92 = permute_10 = None
	        view_93: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_4, [sym_size_int, 1500, 1280]);  addmm_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_944: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_646, view_93);  add_646 = view_93 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_9: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_944, memory_format = torch.contiguous_format)
	        var_mean_2 = torch.ops.aten.var_mean.correction(clone_9, [2], correction = 0, keepdim = True)
	        getitem_8: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_2[0]
	        getitem_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_2[1];  var_mean_2 = None
	        add_949: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_8, 1e-05);  getitem_8 = None
	        rsqrt_2: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_949);  add_949 = None
	        sub_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_9, getitem_9);  clone_9 = getitem_9 = None
	        mul_600: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_281, rsqrt_2);  sub_281 = rsqrt_2 = None
	        mul_601: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_600, model_audio_tower_layers_1_self_attn_layer_norm_weight);  mul_600 = model_audio_tower_layers_1_self_attn_layer_norm_weight = None
	        add_950: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_601, model_audio_tower_layers_1_self_attn_layer_norm_bias);  mul_601 = model_audio_tower_layers_1_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_950, [2])
	        amax_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_950, [2])
	        full_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_6, full_12);  amin_6 = full_12 = None
	        full_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_6, full_13);  amax_6 = full_13 = None
	        sub_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_6, minimum_6);  maximum_6 = None
	        div_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_292, 255.0);  sub_292 = None
	        clamp_min_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_12, 1.1920928955078125e-07);  div_12 = None
	        div_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_6, clamp_min_18);  minimum_6 = None
	        round_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_13);  div_13 = None
	        sub_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_13);  round_13 = None
	        clamp_min_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_298, -128);  sub_298 = None
	        clamp_max_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_19, 127);  clamp_min_19 = None
	        _assert_tensor_metadata_56 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_18, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_56 = None
	        _assert_tensor_metadata_57 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_12, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_57 = None
	        convert_element_type_36: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_12, torch.int8);  clamp_max_12 = None
	        view_96: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_18, [sym_size_int, 1500, 1])
	        view_97: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_36, [sym_size_int, 1500, 1])
	        reciprocal_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_96);  view_96 = None
	        mul_649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_6, 1.0);  reciprocal_6 = None
	        mul_652: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_950, mul_649);  mul_649 = None
	        round_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_652);  mul_652 = None
	        add_1037: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_14, view_97);  round_14 = view_97 = None
	        clamp_min_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1037, -128);  add_1037 = None
	        clamp_max_13: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_20, 127);  clamp_min_20 = None
	        _assert_tensor_metadata_58 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_13, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_58 = None
	        convert_element_type_37: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_13, torch.int8);  clamp_max_13 = None
	        view_100: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_18, [sym_size_int, 1500, 1]);  clamp_min_18 = None
	        view_101: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_36, [sym_size_int, 1500, 1]);  convert_element_type_36 = None
	        _assert_tensor_metadata_59 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_37, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_59 = None
	        convert_element_type_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_37, torch.float32);  convert_element_type_37 = None
	        _assert_tensor_metadata_60 = torch.ops.aten._assert_tensor_metadata.default(view_101, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_60 = None
	        convert_element_type_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_101, torch.float32);  view_101 = None
	        sub_318: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_38, convert_element_type_39);  convert_element_type_38 = convert_element_type_39 = None
	        mul_674: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_318, view_100);  sub_318 = view_100 = None
	        _assert_tensor_metadata_61 = torch.ops.aten._assert_tensor_metadata.default(mul_674, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_61 = None
	        view_103: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_104: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_105: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_62 = torch.ops.aten._assert_tensor_metadata.default(view_103, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_62 = None
	        convert_element_type_40: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_103, torch.float32);  view_103 = None
	        _assert_tensor_metadata_63 = torch.ops.aten._assert_tensor_metadata.default(view_105, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_63 = None
	        convert_element_type_41: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_105, torch.float32);  view_105 = None
	        sub_322: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_40, convert_element_type_41);  convert_element_type_40 = convert_element_type_41 = None
	        mul_679: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_322, view_104);  sub_322 = view_104 = None
	        view_106: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_679, [1280, 1280]);  mul_679 = None
	        _assert_tensor_metadata_64 = torch.ops.aten._assert_tensor_metadata.default(view_106, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_64 = None
	        mul_684: "Sym(1500*s6)" = sym_size_int * 1500
	        view_107: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_674, [mul_684, 1280]);  mul_674 = mul_684 = None
	        permute_11: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_106, [1, 0]);  view_106 = None
	        addmm_5: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_1_self_attn_q_proj_bias, view_107, permute_11);  model_audio_tower_layers_1_self_attn_q_proj_bias = view_107 = permute_11 = None
	        view_108: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_5, [sym_size_int, 1500, 1280]);  addmm_5 = None
	        mul_691: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_108, 0.125);  view_108 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_109: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_691, [sym_size_int, 1500, 20, 64]);  mul_691 = None
	        permute_12: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_109, [0, 2, 1, 3]);  view_109 = None
	        clone_10: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_12, memory_format = torch.contiguous_format);  permute_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_950, [2])
	        amax_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_950, [2])
	        full_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_7, full_14);  amin_7 = full_14 = None
	        full_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_7, full_15);  amax_7 = full_15 = None
	        sub_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_7, minimum_7);  maximum_7 = None
	        div_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_337, 255.0);  sub_337 = None
	        clamp_min_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_14, 1.1920928955078125e-07);  div_14 = None
	        div_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_7, clamp_min_21);  minimum_7 = None
	        round_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_15);  div_15 = None
	        sub_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_15);  round_15 = None
	        clamp_min_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_343, -128);  sub_343 = None
	        clamp_max_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_22, 127);  clamp_min_22 = None
	        _assert_tensor_metadata_65 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_21, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_65 = None
	        _assert_tensor_metadata_66 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_14, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_66 = None
	        convert_element_type_42: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_14, torch.int8);  clamp_max_14 = None
	        view_112: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_21, [sym_size_int, 1500, 1])
	        view_113: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_42, [sym_size_int, 1500, 1])
	        reciprocal_7: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_112);  view_112 = None
	        mul_745: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_7, 1.0);  reciprocal_7 = None
	        mul_748: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_950, mul_745);  mul_745 = None
	        round_16: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_748);  mul_748 = None
	        add_1189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_16, view_113);  round_16 = view_113 = None
	        clamp_min_23: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1189, -128);  add_1189 = None
	        clamp_max_15: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_23, 127);  clamp_min_23 = None
	        _assert_tensor_metadata_67 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_15, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_67 = None
	        convert_element_type_43: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_15, torch.int8);  clamp_max_15 = None
	        view_116: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_21, [sym_size_int, 1500, 1]);  clamp_min_21 = None
	        view_117: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_42, [sym_size_int, 1500, 1]);  convert_element_type_42 = None
	        _assert_tensor_metadata_68 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_43, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_68 = None
	        convert_element_type_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_43, torch.float32);  convert_element_type_43 = None
	        _assert_tensor_metadata_69 = torch.ops.aten._assert_tensor_metadata.default(view_117, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_69 = None
	        convert_element_type_45: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_117, torch.float32);  view_117 = None
	        sub_363: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_44, convert_element_type_45);  convert_element_type_44 = convert_element_type_45 = None
	        mul_770: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_363, view_116);  sub_363 = view_116 = None
	        _assert_tensor_metadata_70 = torch.ops.aten._assert_tensor_metadata.default(mul_770, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_70 = None
	        view_119: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_120: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_121: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_71 = torch.ops.aten._assert_tensor_metadata.default(view_119, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_71 = None
	        convert_element_type_46: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_119, torch.float32);  view_119 = None
	        _assert_tensor_metadata_72 = torch.ops.aten._assert_tensor_metadata.default(view_121, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_72 = None
	        convert_element_type_47: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_121, torch.float32);  view_121 = None
	        sub_367: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_46, convert_element_type_47);  convert_element_type_46 = convert_element_type_47 = None
	        mul_775: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_367, view_120);  sub_367 = view_120 = None
	        view_122: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_775, [1280, 1280]);  mul_775 = None
	        _assert_tensor_metadata_73 = torch.ops.aten._assert_tensor_metadata.default(view_122, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_73 = None
	        permute_13: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_122, [1, 0]);  view_122 = None
	        mul_778: "Sym(1500*s6)" = sym_size_int * 1500
	        view_123: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_770, [mul_778, 1280]);  mul_770 = mul_778 = None
	        mm_1: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_123, permute_13);  view_123 = permute_13 = None
	        view_124: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_1, [sym_size_int, 1500, 1280]);  mm_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_125: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_124, [sym_size_int, -1, 20, 64]);  view_124 = None
	        permute_14: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_125, [0, 2, 1, 3]);  view_125 = None
	        clone_11: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_14, memory_format = torch.contiguous_format);  permute_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_950, [2])
	        amax_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_950, [2])
	        full_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_8, full_16);  amin_8 = full_16 = None
	        full_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_8, full_17);  amax_8 = full_17 = None
	        sub_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_8, minimum_8);  maximum_8 = None
	        div_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_381, 255.0);  sub_381 = None
	        clamp_min_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_16, 1.1920928955078125e-07);  div_16 = None
	        div_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_8, clamp_min_24);  minimum_8 = None
	        round_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_17);  div_17 = None
	        sub_387: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_17);  round_17 = None
	        clamp_min_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_387, -128);  sub_387 = None
	        clamp_max_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_25, 127);  clamp_min_25 = None
	        _assert_tensor_metadata_74 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_24, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_74 = None
	        _assert_tensor_metadata_75 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_16, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_75 = None
	        convert_element_type_48: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_16, torch.int8);  clamp_max_16 = None
	        view_128: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_24, [sym_size_int, 1500, 1])
	        view_129: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_48, [sym_size_int, 1500, 1])
	        reciprocal_8: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_128);  view_128 = None
	        mul_844: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_8, 1.0);  reciprocal_8 = None
	        mul_847: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_950, mul_844);  add_950 = mul_844 = None
	        round_18: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_847);  mul_847 = None
	        add_1337: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_18, view_129);  round_18 = view_129 = None
	        clamp_min_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1337, -128);  add_1337 = None
	        clamp_max_17: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_26, 127);  clamp_min_26 = None
	        _assert_tensor_metadata_76 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_17, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_76 = None
	        convert_element_type_49: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_17, torch.int8);  clamp_max_17 = None
	        view_132: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_24, [sym_size_int, 1500, 1]);  clamp_min_24 = None
	        view_133: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_48, [sym_size_int, 1500, 1]);  convert_element_type_48 = None
	        _assert_tensor_metadata_77 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_49, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_77 = None
	        convert_element_type_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_49, torch.float32);  convert_element_type_49 = None
	        _assert_tensor_metadata_78 = torch.ops.aten._assert_tensor_metadata.default(view_133, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_78 = None
	        convert_element_type_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_133, torch.float32);  view_133 = None
	        sub_407: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_50, convert_element_type_51);  convert_element_type_50 = convert_element_type_51 = None
	        mul_869: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_407, view_132);  sub_407 = view_132 = None
	        _assert_tensor_metadata_79 = torch.ops.aten._assert_tensor_metadata.default(mul_869, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_79 = None
	        view_135: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_136: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_137: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_80 = torch.ops.aten._assert_tensor_metadata.default(view_135, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_80 = None
	        convert_element_type_52: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_135, torch.float32);  view_135 = None
	        _assert_tensor_metadata_81 = torch.ops.aten._assert_tensor_metadata.default(view_137, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_81 = None
	        convert_element_type_53: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_137, torch.float32);  view_137 = None
	        sub_411: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_52, convert_element_type_53);  convert_element_type_52 = convert_element_type_53 = None
	        mul_874: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_411, view_136);  sub_411 = view_136 = None
	        view_138: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_874, [1280, 1280]);  mul_874 = None
	        _assert_tensor_metadata_82 = torch.ops.aten._assert_tensor_metadata.default(view_138, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_82 = None
	        mul_879: "Sym(1500*s6)" = sym_size_int * 1500
	        view_139: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_869, [mul_879, 1280]);  mul_869 = mul_879 = None
	        permute_15: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_138, [1, 0]);  view_138 = None
	        addmm_6: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_1_self_attn_v_proj_bias, view_139, permute_15);  model_audio_tower_layers_1_self_attn_v_proj_bias = view_139 = permute_15 = None
	        view_140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_6, [sym_size_int, 1500, 1280]);  addmm_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_141: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_140, [sym_size_int, -1, 20, 64]);  view_140 = None
	        permute_16: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_141, [0, 2, 1, 3]);  view_141 = None
	        clone_12: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_16, memory_format = torch.contiguous_format);  permute_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_1 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_10, clone_11, clone_12, None, False, scale = 1.0);  clone_10 = clone_11 = clone_12 = None
	        getitem_10: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_1[0];  _scaled_dot_product_efficient_attention_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_17: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_10, [0, 2, 1, 3]);  getitem_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_17, [sym_size_int, 1500, -1]);  permute_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_142, [2])
	        amax_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_142, [2])
	        full_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_9, full_18);  amin_9 = full_18 = None
	        full_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_9, full_19);  amax_9 = full_19 = None
	        sub_429: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_9, minimum_9);  maximum_9 = None
	        div_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_429, 255.0);  sub_429 = None
	        clamp_min_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_18, 1.1920928955078125e-07);  div_18 = None
	        div_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_9, clamp_min_27);  minimum_9 = None
	        round_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_19);  div_19 = None
	        sub_435: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_19);  round_19 = None
	        clamp_min_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_435, -128);  sub_435 = None
	        clamp_max_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_28, 127);  clamp_min_28 = None
	        _assert_tensor_metadata_83 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_27, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_83 = None
	        _assert_tensor_metadata_84 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_18, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_84 = None
	        convert_element_type_54: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_18, torch.int8);  clamp_max_18 = None
	        view_145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_27, [sym_size_int, 1500, 1])
	        view_146: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_54, [sym_size_int, 1500, 1])
	        reciprocal_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_145);  view_145 = None
	        mul_949: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_9, 1.0);  reciprocal_9 = None
	        mul_952: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_142, mul_949);  view_142 = mul_949 = None
	        round_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_952);  mul_952 = None
	        add_1501: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_20, view_146);  round_20 = view_146 = None
	        clamp_min_29: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1501, -128);  add_1501 = None
	        clamp_max_19: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_29, 127);  clamp_min_29 = None
	        _assert_tensor_metadata_85 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_19, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_85 = None
	        convert_element_type_55: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_19, torch.int8);  clamp_max_19 = None
	        view_149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_27, [sym_size_int, 1500, 1]);  clamp_min_27 = None
	        view_150: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_54, [sym_size_int, 1500, 1]);  convert_element_type_54 = None
	        _assert_tensor_metadata_86 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_55, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_86 = None
	        convert_element_type_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_55, torch.float32);  convert_element_type_55 = None
	        _assert_tensor_metadata_87 = torch.ops.aten._assert_tensor_metadata.default(view_150, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_87 = None
	        convert_element_type_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_150, torch.float32);  view_150 = None
	        sub_455: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_56, convert_element_type_57);  convert_element_type_56 = convert_element_type_57 = None
	        mul_974: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_455, view_149);  sub_455 = view_149 = None
	        _assert_tensor_metadata_88 = torch.ops.aten._assert_tensor_metadata.default(mul_974, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_88 = None
	        view_152: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_153: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_154: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_89 = torch.ops.aten._assert_tensor_metadata.default(view_152, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_89 = None
	        convert_element_type_58: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_152, torch.float32);  view_152 = None
	        _assert_tensor_metadata_90 = torch.ops.aten._assert_tensor_metadata.default(view_154, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_90 = None
	        convert_element_type_59: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_154, torch.float32);  view_154 = None
	        sub_459: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_58, convert_element_type_59);  convert_element_type_58 = convert_element_type_59 = None
	        mul_979: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_459, view_153);  sub_459 = view_153 = None
	        view_155: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_979, [1280, 1280]);  mul_979 = None
	        _assert_tensor_metadata_91 = torch.ops.aten._assert_tensor_metadata.default(view_155, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_91 = None
	        mul_984: "Sym(1500*s6)" = sym_size_int * 1500
	        view_156: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_974, [mul_984, 1280]);  mul_974 = mul_984 = None
	        permute_18: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_155, [1, 0]);  view_155 = None
	        addmm_7: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_1_self_attn_out_proj_bias, view_156, permute_18);  model_audio_tower_layers_1_self_attn_out_proj_bias = view_156 = permute_18 = None
	        view_157: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_7, [sym_size_int, 1500, 1280]);  addmm_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_1564: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_944, view_157);  add_944 = view_157 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_1564, memory_format = torch.contiguous_format)
	        var_mean_3 = torch.ops.aten.var_mean.correction(clone_14, [2], correction = 0, keepdim = True)
	        getitem_14: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_3[0]
	        getitem_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_3[1];  var_mean_3 = None
	        add_1569: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_14, 1e-05);  getitem_14 = None
	        rsqrt_3: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_1569);  add_1569 = None
	        sub_465: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_14, getitem_15);  clone_14 = getitem_15 = None
	        mul_995: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_465, rsqrt_3);  sub_465 = rsqrt_3 = None
	        mul_996: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_995, model_audio_tower_layers_1_final_layer_norm_weight);  mul_995 = model_audio_tower_layers_1_final_layer_norm_weight = None
	        add_1570: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_996, model_audio_tower_layers_1_final_layer_norm_bias);  mul_996 = model_audio_tower_layers_1_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_1570, [2])
	        amax_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_1570, [2])
	        full_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_10, full_20);  amin_10 = full_20 = None
	        full_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_10, full_21);  amax_10 = full_21 = None
	        sub_476: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_10, minimum_10);  maximum_10 = None
	        div_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_476, 255.0);  sub_476 = None
	        clamp_min_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_20, 1.1920928955078125e-07);  div_20 = None
	        div_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_10, clamp_min_30);  minimum_10 = None
	        round_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_21);  div_21 = None
	        sub_482: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_21);  round_21 = None
	        clamp_min_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_482, -128);  sub_482 = None
	        clamp_max_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_31, 127);  clamp_min_31 = None
	        _assert_tensor_metadata_92 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_30, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_92 = None
	        _assert_tensor_metadata_93 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_20, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_93 = None
	        convert_element_type_60: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_20, torch.int8);  clamp_max_20 = None
	        view_160: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_30, [sym_size_int, 1500, 1])
	        view_161: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_60, [sym_size_int, 1500, 1])
	        reciprocal_10: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_160);  view_160 = None
	        mul_1044: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_10, 1.0);  reciprocal_10 = None
	        mul_1047: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_1570, mul_1044);  add_1570 = mul_1044 = None
	        round_22: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1047);  mul_1047 = None
	        add_1657: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_22, view_161);  round_22 = view_161 = None
	        clamp_min_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1657, -128);  add_1657 = None
	        clamp_max_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_32, 127);  clamp_min_32 = None
	        _assert_tensor_metadata_94 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_21, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_94 = None
	        convert_element_type_61: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_21, torch.int8);  clamp_max_21 = None
	        view_164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_30, [sym_size_int, 1500, 1]);  clamp_min_30 = None
	        view_165: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_60, [sym_size_int, 1500, 1]);  convert_element_type_60 = None
	        _assert_tensor_metadata_95 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_61, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_95 = None
	        convert_element_type_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_61, torch.float32);  convert_element_type_61 = None
	        _assert_tensor_metadata_96 = torch.ops.aten._assert_tensor_metadata.default(view_165, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_96 = None
	        convert_element_type_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_165, torch.float32);  view_165 = None
	        sub_502: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_62, convert_element_type_63);  convert_element_type_62 = convert_element_type_63 = None
	        mul_1069: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_502, view_164);  sub_502 = view_164 = None
	        _assert_tensor_metadata_97 = torch.ops.aten._assert_tensor_metadata.default(mul_1069, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_97 = None
	        view_167: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_1_fc1_parametrizations_weight_original0 = None
	        view_168: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_1_fc1_parametrizations_weight_original1 = None
	        view_169: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_1_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_98 = torch.ops.aten._assert_tensor_metadata.default(view_167, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_98 = None
	        convert_element_type_64: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_167, torch.float32);  view_167 = None
	        _assert_tensor_metadata_99 = torch.ops.aten._assert_tensor_metadata.default(view_169, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_99 = None
	        convert_element_type_65: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_169, torch.float32);  view_169 = None
	        sub_506: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_64, convert_element_type_65);  convert_element_type_64 = convert_element_type_65 = None
	        mul_1074: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_506, view_168);  sub_506 = view_168 = None
	        view_170: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1074, [5120, 1280]);  mul_1074 = None
	        _assert_tensor_metadata_100 = torch.ops.aten._assert_tensor_metadata.default(view_170, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_100 = None
	        mul_1079: "Sym(1500*s6)" = sym_size_int * 1500
	        view_171: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1069, [mul_1079, 1280]);  mul_1069 = mul_1079 = None
	        permute_19: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_170, [1, 0]);  view_170 = None
	        addmm_8: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_1_fc1_bias, view_171, permute_19);  model_audio_tower_layers_1_fc1_bias = view_171 = permute_19 = None
	        view_172: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_8, [sym_size_int, 1500, 5120]);  addmm_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_1086: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_172, 0.5)
	        mul_1087: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_172, 0.7071067811865476);  view_172 = None
	        erf_3: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_1087);  mul_1087 = None
	        add_1716: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_3, 1);  erf_3 = None
	        mul_1088: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1086, add_1716);  mul_1086 = add_1716 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_1088, [2])
	        amax_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_1088, [2])
	        full_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_11, full_22);  amin_11 = full_22 = None
	        full_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_11, full_23);  amax_11 = full_23 = None
	        sub_519: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_11, minimum_11);  maximum_11 = None
	        div_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_519, 255.0);  sub_519 = None
	        clamp_min_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_22, 1.1920928955078125e-07);  div_22 = None
	        div_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_11, clamp_min_33);  minimum_11 = None
	        round_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_23);  div_23 = None
	        sub_525: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_23);  round_23 = None
	        clamp_min_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_525, -128);  sub_525 = None
	        clamp_max_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_34, 127);  clamp_min_34 = None
	        _assert_tensor_metadata_101 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_33, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_101 = None
	        _assert_tensor_metadata_102 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_22, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_102 = None
	        convert_element_type_66: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_22, torch.int8);  clamp_max_22 = None
	        view_175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_33, [sym_size_int, 1500, 1])
	        view_176: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_66, [sym_size_int, 1500, 1])
	        reciprocal_11: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_175);  view_175 = None
	        mul_1134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_11, 1.0);  reciprocal_11 = None
	        mul_1137: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1088, mul_1134);  mul_1088 = mul_1134 = None
	        round_24: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_1137);  mul_1137 = None
	        add_1799: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_24, view_176);  round_24 = view_176 = None
	        clamp_min_35: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1799, -128);  add_1799 = None
	        clamp_max_23: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_35, 127);  clamp_min_35 = None
	        _assert_tensor_metadata_103 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_23, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_103 = None
	        convert_element_type_67: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_23, torch.int8);  clamp_max_23 = None
	        view_179: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_33, [sym_size_int, 1500, 1]);  clamp_min_33 = None
	        view_180: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_66, [sym_size_int, 1500, 1]);  convert_element_type_66 = None
	        _assert_tensor_metadata_104 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_67, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_104 = None
	        convert_element_type_68: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_67, torch.float32);  convert_element_type_67 = None
	        _assert_tensor_metadata_105 = torch.ops.aten._assert_tensor_metadata.default(view_180, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_105 = None
	        convert_element_type_69: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_180, torch.float32);  view_180 = None
	        sub_545: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_68, convert_element_type_69);  convert_element_type_68 = convert_element_type_69 = None
	        mul_1159: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_545, view_179);  sub_545 = view_179 = None
	        _assert_tensor_metadata_106 = torch.ops.aten._assert_tensor_metadata.default(mul_1159, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_106 = None
	        view_182: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_1_fc2_parametrizations_weight_original0 = None
	        view_183: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_1_fc2_parametrizations_weight_original1 = None
	        view_184: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_1_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_107 = torch.ops.aten._assert_tensor_metadata.default(view_182, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_107 = None
	        convert_element_type_70: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_182, torch.float32);  view_182 = None
	        _assert_tensor_metadata_108 = torch.ops.aten._assert_tensor_metadata.default(view_184, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_108 = None
	        convert_element_type_71: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_184, torch.float32);  view_184 = None
	        sub_549: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_70, convert_element_type_71);  convert_element_type_70 = convert_element_type_71 = None
	        mul_1164: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_549, view_183);  sub_549 = view_183 = None
	        view_185: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1164, [1280, 5120]);  mul_1164 = None
	        _assert_tensor_metadata_109 = torch.ops.aten._assert_tensor_metadata.default(view_185, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_109 = None
	        mul_1169: "Sym(1500*s6)" = sym_size_int * 1500
	        view_186: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1159, [mul_1169, 5120]);  mul_1159 = mul_1169 = None
	        permute_20: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_185, [1, 0]);  view_185 = None
	        addmm_9: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_1_fc2_bias, view_186, permute_20);  model_audio_tower_layers_1_fc2_bias = view_186 = permute_20 = None
	        view_187: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_9, [sym_size_int, 1500, 1280]);  addmm_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_1862: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_1564, view_187);  add_1564 = view_187 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_17: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_1862, memory_format = torch.contiguous_format)
	        var_mean_4 = torch.ops.aten.var_mean.correction(clone_17, [2], correction = 0, keepdim = True)
	        getitem_16: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_4[0]
	        getitem_17: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_4[1];  var_mean_4 = None
	        add_1867: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_16, 1e-05);  getitem_16 = None
	        rsqrt_4: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_1867);  add_1867 = None
	        sub_555: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_17, getitem_17);  clone_17 = getitem_17 = None
	        mul_1180: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_555, rsqrt_4);  sub_555 = rsqrt_4 = None
	        mul_1181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1180, model_audio_tower_layers_2_self_attn_layer_norm_weight);  mul_1180 = model_audio_tower_layers_2_self_attn_layer_norm_weight = None
	        add_1868: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_1181, model_audio_tower_layers_2_self_attn_layer_norm_bias);  mul_1181 = model_audio_tower_layers_2_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_1868, [2])
	        amax_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_1868, [2])
	        full_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_12, full_24);  amin_12 = full_24 = None
	        full_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_12, full_25);  amax_12 = full_25 = None
	        sub_566: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_12, minimum_12);  maximum_12 = None
	        div_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_566, 255.0);  sub_566 = None
	        clamp_min_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_24, 1.1920928955078125e-07);  div_24 = None
	        div_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_12, clamp_min_36);  minimum_12 = None
	        round_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_25);  div_25 = None
	        sub_572: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_25);  round_25 = None
	        clamp_min_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_572, -128);  sub_572 = None
	        clamp_max_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_37, 127);  clamp_min_37 = None
	        _assert_tensor_metadata_110 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_36, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_110 = None
	        _assert_tensor_metadata_111 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_24, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_111 = None
	        convert_element_type_72: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_24, torch.int8);  clamp_max_24 = None
	        view_190: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_36, [sym_size_int, 1500, 1])
	        view_191: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_72, [sym_size_int, 1500, 1])
	        reciprocal_12: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_190);  view_190 = None
	        mul_1229: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_12, 1.0);  reciprocal_12 = None
	        mul_1232: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_1868, mul_1229);  mul_1229 = None
	        round_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1232);  mul_1232 = None
	        add_1955: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_26, view_191);  round_26 = view_191 = None
	        clamp_min_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1955, -128);  add_1955 = None
	        clamp_max_25: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_38, 127);  clamp_min_38 = None
	        _assert_tensor_metadata_112 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_25, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_112 = None
	        convert_element_type_73: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_25, torch.int8);  clamp_max_25 = None
	        view_194: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_36, [sym_size_int, 1500, 1]);  clamp_min_36 = None
	        view_195: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_72, [sym_size_int, 1500, 1]);  convert_element_type_72 = None
	        _assert_tensor_metadata_113 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_73, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_113 = None
	        convert_element_type_74: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_73, torch.float32);  convert_element_type_73 = None
	        _assert_tensor_metadata_114 = torch.ops.aten._assert_tensor_metadata.default(view_195, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_114 = None
	        convert_element_type_75: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_195, torch.float32);  view_195 = None
	        sub_592: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_74, convert_element_type_75);  convert_element_type_74 = convert_element_type_75 = None
	        mul_1254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_592, view_194);  sub_592 = view_194 = None
	        _assert_tensor_metadata_115 = torch.ops.aten._assert_tensor_metadata.default(mul_1254, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_115 = None
	        view_197: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_198: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_199: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_116 = torch.ops.aten._assert_tensor_metadata.default(view_197, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_116 = None
	        convert_element_type_76: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_197, torch.float32);  view_197 = None
	        _assert_tensor_metadata_117 = torch.ops.aten._assert_tensor_metadata.default(view_199, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_117 = None
	        convert_element_type_77: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_199, torch.float32);  view_199 = None
	        sub_596: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_76, convert_element_type_77);  convert_element_type_76 = convert_element_type_77 = None
	        mul_1259: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_596, view_198);  sub_596 = view_198 = None
	        view_200: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1259, [1280, 1280]);  mul_1259 = None
	        _assert_tensor_metadata_118 = torch.ops.aten._assert_tensor_metadata.default(view_200, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_118 = None
	        mul_1264: "Sym(1500*s6)" = sym_size_int * 1500
	        view_201: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1254, [mul_1264, 1280]);  mul_1254 = mul_1264 = None
	        permute_21: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_200, [1, 0]);  view_200 = None
	        addmm_10: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_2_self_attn_q_proj_bias, view_201, permute_21);  model_audio_tower_layers_2_self_attn_q_proj_bias = view_201 = permute_21 = None
	        view_202: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_10, [sym_size_int, 1500, 1280]);  addmm_10 = None
	        mul_1271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_202, 0.125);  view_202 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_203: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1271, [sym_size_int, 1500, 20, 64]);  mul_1271 = None
	        permute_22: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_203, [0, 2, 1, 3]);  view_203 = None
	        clone_18: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_22, memory_format = torch.contiguous_format);  permute_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_1868, [2])
	        amax_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_1868, [2])
	        full_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_13, full_26);  amin_13 = full_26 = None
	        full_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_13, full_27);  amax_13 = full_27 = None
	        sub_611: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_13, minimum_13);  maximum_13 = None
	        div_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_611, 255.0);  sub_611 = None
	        clamp_min_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_26, 1.1920928955078125e-07);  div_26 = None
	        div_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_13, clamp_min_39);  minimum_13 = None
	        round_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_27);  div_27 = None
	        sub_617: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_27);  round_27 = None
	        clamp_min_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_617, -128);  sub_617 = None
	        clamp_max_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_40, 127);  clamp_min_40 = None
	        _assert_tensor_metadata_119 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_39, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_119 = None
	        _assert_tensor_metadata_120 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_26, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_120 = None
	        convert_element_type_78: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_26, torch.int8);  clamp_max_26 = None
	        view_206: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_39, [sym_size_int, 1500, 1])
	        view_207: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_78, [sym_size_int, 1500, 1])
	        reciprocal_13: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_206);  view_206 = None
	        mul_1325: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_13, 1.0);  reciprocal_13 = None
	        mul_1328: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_1868, mul_1325);  mul_1325 = None
	        round_28: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1328);  mul_1328 = None
	        add_2107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_28, view_207);  round_28 = view_207 = None
	        clamp_min_41: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2107, -128);  add_2107 = None
	        clamp_max_27: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_41, 127);  clamp_min_41 = None
	        _assert_tensor_metadata_121 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_27, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_121 = None
	        convert_element_type_79: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_27, torch.int8);  clamp_max_27 = None
	        view_210: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_39, [sym_size_int, 1500, 1]);  clamp_min_39 = None
	        view_211: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_78, [sym_size_int, 1500, 1]);  convert_element_type_78 = None
	        _assert_tensor_metadata_122 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_79, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_122 = None
	        convert_element_type_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_79, torch.float32);  convert_element_type_79 = None
	        _assert_tensor_metadata_123 = torch.ops.aten._assert_tensor_metadata.default(view_211, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_123 = None
	        convert_element_type_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_211, torch.float32);  view_211 = None
	        sub_637: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_80, convert_element_type_81);  convert_element_type_80 = convert_element_type_81 = None
	        mul_1350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_637, view_210);  sub_637 = view_210 = None
	        _assert_tensor_metadata_124 = torch.ops.aten._assert_tensor_metadata.default(mul_1350, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_124 = None
	        view_213: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_214: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_215: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_125 = torch.ops.aten._assert_tensor_metadata.default(view_213, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_125 = None
	        convert_element_type_82: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_213, torch.float32);  view_213 = None
	        _assert_tensor_metadata_126 = torch.ops.aten._assert_tensor_metadata.default(view_215, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_126 = None
	        convert_element_type_83: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_215, torch.float32);  view_215 = None
	        sub_641: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_82, convert_element_type_83);  convert_element_type_82 = convert_element_type_83 = None
	        mul_1355: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_641, view_214);  sub_641 = view_214 = None
	        view_216: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1355, [1280, 1280]);  mul_1355 = None
	        _assert_tensor_metadata_127 = torch.ops.aten._assert_tensor_metadata.default(view_216, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_127 = None
	        permute_23: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_216, [1, 0]);  view_216 = None
	        mul_1358: "Sym(1500*s6)" = sym_size_int * 1500
	        view_217: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1350, [mul_1358, 1280]);  mul_1350 = mul_1358 = None
	        mm_2: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_217, permute_23);  view_217 = permute_23 = None
	        view_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_2, [sym_size_int, 1500, 1280]);  mm_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_219: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_218, [sym_size_int, -1, 20, 64]);  view_218 = None
	        permute_24: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_219, [0, 2, 1, 3]);  view_219 = None
	        clone_19: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_24, memory_format = torch.contiguous_format);  permute_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_1868, [2])
	        amax_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_1868, [2])
	        full_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_14, full_28);  amin_14 = full_28 = None
	        full_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_14, full_29);  amax_14 = full_29 = None
	        sub_655: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_14, minimum_14);  maximum_14 = None
	        div_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_655, 255.0);  sub_655 = None
	        clamp_min_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_28, 1.1920928955078125e-07);  div_28 = None
	        div_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_14, clamp_min_42);  minimum_14 = None
	        round_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_29);  div_29 = None
	        sub_661: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_29);  round_29 = None
	        clamp_min_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_661, -128);  sub_661 = None
	        clamp_max_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_43, 127);  clamp_min_43 = None
	        _assert_tensor_metadata_128 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_42, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_128 = None
	        _assert_tensor_metadata_129 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_28, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_129 = None
	        convert_element_type_84: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_28, torch.int8);  clamp_max_28 = None
	        view_222: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_42, [sym_size_int, 1500, 1])
	        view_223: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_84, [sym_size_int, 1500, 1])
	        reciprocal_14: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_222);  view_222 = None
	        mul_1424: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_14, 1.0);  reciprocal_14 = None
	        mul_1427: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_1868, mul_1424);  add_1868 = mul_1424 = None
	        round_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1427);  mul_1427 = None
	        add_2255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_30, view_223);  round_30 = view_223 = None
	        clamp_min_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2255, -128);  add_2255 = None
	        clamp_max_29: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_44, 127);  clamp_min_44 = None
	        _assert_tensor_metadata_130 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_29, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_130 = None
	        convert_element_type_85: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_29, torch.int8);  clamp_max_29 = None
	        view_226: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_42, [sym_size_int, 1500, 1]);  clamp_min_42 = None
	        view_227: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_84, [sym_size_int, 1500, 1]);  convert_element_type_84 = None
	        _assert_tensor_metadata_131 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_85, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_131 = None
	        convert_element_type_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_85, torch.float32);  convert_element_type_85 = None
	        _assert_tensor_metadata_132 = torch.ops.aten._assert_tensor_metadata.default(view_227, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_132 = None
	        convert_element_type_87: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_227, torch.float32);  view_227 = None
	        sub_681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_86, convert_element_type_87);  convert_element_type_86 = convert_element_type_87 = None
	        mul_1449: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_681, view_226);  sub_681 = view_226 = None
	        _assert_tensor_metadata_133 = torch.ops.aten._assert_tensor_metadata.default(mul_1449, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_133 = None
	        view_229: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_230: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_231: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_134 = torch.ops.aten._assert_tensor_metadata.default(view_229, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_134 = None
	        convert_element_type_88: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_229, torch.float32);  view_229 = None
	        _assert_tensor_metadata_135 = torch.ops.aten._assert_tensor_metadata.default(view_231, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_135 = None
	        convert_element_type_89: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_231, torch.float32);  view_231 = None
	        sub_685: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_88, convert_element_type_89);  convert_element_type_88 = convert_element_type_89 = None
	        mul_1454: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_685, view_230);  sub_685 = view_230 = None
	        view_232: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1454, [1280, 1280]);  mul_1454 = None
	        _assert_tensor_metadata_136 = torch.ops.aten._assert_tensor_metadata.default(view_232, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_136 = None
	        mul_1459: "Sym(1500*s6)" = sym_size_int * 1500
	        view_233: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1449, [mul_1459, 1280]);  mul_1449 = mul_1459 = None
	        permute_25: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_232, [1, 0]);  view_232 = None
	        addmm_11: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_2_self_attn_v_proj_bias, view_233, permute_25);  model_audio_tower_layers_2_self_attn_v_proj_bias = view_233 = permute_25 = None
	        view_234: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_11, [sym_size_int, 1500, 1280]);  addmm_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_235: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_234, [sym_size_int, -1, 20, 64]);  view_234 = None
	        permute_26: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_235, [0, 2, 1, 3]);  view_235 = None
	        clone_20: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_26, memory_format = torch.contiguous_format);  permute_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_2 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_18, clone_19, clone_20, None, False, scale = 1.0);  clone_18 = clone_19 = clone_20 = None
	        getitem_18: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_2[0];  _scaled_dot_product_efficient_attention_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_27: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_18, [0, 2, 1, 3]);  getitem_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_27, [sym_size_int, 1500, -1]);  permute_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_236, [2])
	        amax_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_236, [2])
	        full_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_15, full_30);  amin_15 = full_30 = None
	        full_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_15, full_31);  amax_15 = full_31 = None
	        sub_703: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_15, minimum_15);  maximum_15 = None
	        div_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_703, 255.0);  sub_703 = None
	        clamp_min_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_30, 1.1920928955078125e-07);  div_30 = None
	        div_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_15, clamp_min_45);  minimum_15 = None
	        round_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_31);  div_31 = None
	        sub_709: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_31);  round_31 = None
	        clamp_min_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_709, -128);  sub_709 = None
	        clamp_max_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_46, 127);  clamp_min_46 = None
	        _assert_tensor_metadata_137 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_45, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_137 = None
	        _assert_tensor_metadata_138 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_30, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_138 = None
	        convert_element_type_90: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_30, torch.int8);  clamp_max_30 = None
	        view_239: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_45, [sym_size_int, 1500, 1])
	        view_240: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_90, [sym_size_int, 1500, 1])
	        reciprocal_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_239);  view_239 = None
	        mul_1529: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_15, 1.0);  reciprocal_15 = None
	        mul_1532: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_236, mul_1529);  view_236 = mul_1529 = None
	        round_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1532);  mul_1532 = None
	        add_2419: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_32, view_240);  round_32 = view_240 = None
	        clamp_min_47: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2419, -128);  add_2419 = None
	        clamp_max_31: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_47, 127);  clamp_min_47 = None
	        _assert_tensor_metadata_139 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_31, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_139 = None
	        convert_element_type_91: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_31, torch.int8);  clamp_max_31 = None
	        view_243: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_45, [sym_size_int, 1500, 1]);  clamp_min_45 = None
	        view_244: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_90, [sym_size_int, 1500, 1]);  convert_element_type_90 = None
	        _assert_tensor_metadata_140 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_91, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_140 = None
	        convert_element_type_92: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_91, torch.float32);  convert_element_type_91 = None
	        _assert_tensor_metadata_141 = torch.ops.aten._assert_tensor_metadata.default(view_244, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_141 = None
	        convert_element_type_93: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_244, torch.float32);  view_244 = None
	        sub_729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_92, convert_element_type_93);  convert_element_type_92 = convert_element_type_93 = None
	        mul_1554: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_729, view_243);  sub_729 = view_243 = None
	        _assert_tensor_metadata_142 = torch.ops.aten._assert_tensor_metadata.default(mul_1554, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_142 = None
	        view_246: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_247: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_248: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_143 = torch.ops.aten._assert_tensor_metadata.default(view_246, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_143 = None
	        convert_element_type_94: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_246, torch.float32);  view_246 = None
	        _assert_tensor_metadata_144 = torch.ops.aten._assert_tensor_metadata.default(view_248, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_144 = None
	        convert_element_type_95: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_248, torch.float32);  view_248 = None
	        sub_733: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_94, convert_element_type_95);  convert_element_type_94 = convert_element_type_95 = None
	        mul_1559: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_733, view_247);  sub_733 = view_247 = None
	        view_249: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1559, [1280, 1280]);  mul_1559 = None
	        _assert_tensor_metadata_145 = torch.ops.aten._assert_tensor_metadata.default(view_249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_145 = None
	        mul_1564: "Sym(1500*s6)" = sym_size_int * 1500
	        view_250: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1554, [mul_1564, 1280]);  mul_1554 = mul_1564 = None
	        permute_28: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_249, [1, 0]);  view_249 = None
	        addmm_12: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_2_self_attn_out_proj_bias, view_250, permute_28);  model_audio_tower_layers_2_self_attn_out_proj_bias = view_250 = permute_28 = None
	        view_251: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_12, [sym_size_int, 1500, 1280]);  addmm_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_2482: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_1862, view_251);  add_1862 = view_251 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_22: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_2482, memory_format = torch.contiguous_format)
	        var_mean_5 = torch.ops.aten.var_mean.correction(clone_22, [2], correction = 0, keepdim = True)
	        getitem_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_5[0]
	        getitem_23: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_5[1];  var_mean_5 = None
	        add_2487: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_22, 1e-05);  getitem_22 = None
	        rsqrt_5: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_2487);  add_2487 = None
	        sub_739: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_22, getitem_23);  clone_22 = getitem_23 = None
	        mul_1575: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_739, rsqrt_5);  sub_739 = rsqrt_5 = None
	        mul_1576: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1575, model_audio_tower_layers_2_final_layer_norm_weight);  mul_1575 = model_audio_tower_layers_2_final_layer_norm_weight = None
	        add_2488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_1576, model_audio_tower_layers_2_final_layer_norm_bias);  mul_1576 = model_audio_tower_layers_2_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_2488, [2])
	        amax_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_2488, [2])
	        full_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_16, full_32);  amin_16 = full_32 = None
	        full_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_16, full_33);  amax_16 = full_33 = None
	        sub_750: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_16, minimum_16);  maximum_16 = None
	        div_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_750, 255.0);  sub_750 = None
	        clamp_min_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_32, 1.1920928955078125e-07);  div_32 = None
	        div_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_16, clamp_min_48);  minimum_16 = None
	        round_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_33);  div_33 = None
	        sub_756: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_33);  round_33 = None
	        clamp_min_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_756, -128);  sub_756 = None
	        clamp_max_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_49, 127);  clamp_min_49 = None
	        _assert_tensor_metadata_146 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_48, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_146 = None
	        _assert_tensor_metadata_147 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_32, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_147 = None
	        convert_element_type_96: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_32, torch.int8);  clamp_max_32 = None
	        view_254: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_48, [sym_size_int, 1500, 1])
	        view_255: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_96, [sym_size_int, 1500, 1])
	        reciprocal_16: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_254);  view_254 = None
	        mul_1624: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_16, 1.0);  reciprocal_16 = None
	        mul_1627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_2488, mul_1624);  add_2488 = mul_1624 = None
	        round_34: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1627);  mul_1627 = None
	        add_2575: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_34, view_255);  round_34 = view_255 = None
	        clamp_min_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2575, -128);  add_2575 = None
	        clamp_max_33: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_50, 127);  clamp_min_50 = None
	        _assert_tensor_metadata_148 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_33, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_148 = None
	        convert_element_type_97: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_33, torch.int8);  clamp_max_33 = None
	        view_258: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_48, [sym_size_int, 1500, 1]);  clamp_min_48 = None
	        view_259: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_96, [sym_size_int, 1500, 1]);  convert_element_type_96 = None
	        _assert_tensor_metadata_149 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_97, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_149 = None
	        convert_element_type_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_97, torch.float32);  convert_element_type_97 = None
	        _assert_tensor_metadata_150 = torch.ops.aten._assert_tensor_metadata.default(view_259, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_150 = None
	        convert_element_type_99: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_259, torch.float32);  view_259 = None
	        sub_776: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_98, convert_element_type_99);  convert_element_type_98 = convert_element_type_99 = None
	        mul_1649: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_776, view_258);  sub_776 = view_258 = None
	        _assert_tensor_metadata_151 = torch.ops.aten._assert_tensor_metadata.default(mul_1649, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_151 = None
	        view_261: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_2_fc1_parametrizations_weight_original0 = None
	        view_262: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_2_fc1_parametrizations_weight_original1 = None
	        view_263: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_2_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_152 = torch.ops.aten._assert_tensor_metadata.default(view_261, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_152 = None
	        convert_element_type_100: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_261, torch.float32);  view_261 = None
	        _assert_tensor_metadata_153 = torch.ops.aten._assert_tensor_metadata.default(view_263, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_153 = None
	        convert_element_type_101: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_263, torch.float32);  view_263 = None
	        sub_780: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_100, convert_element_type_101);  convert_element_type_100 = convert_element_type_101 = None
	        mul_1654: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_780, view_262);  sub_780 = view_262 = None
	        view_264: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1654, [5120, 1280]);  mul_1654 = None
	        _assert_tensor_metadata_154 = torch.ops.aten._assert_tensor_metadata.default(view_264, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_154 = None
	        mul_1659: "Sym(1500*s6)" = sym_size_int * 1500
	        view_265: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1649, [mul_1659, 1280]);  mul_1649 = mul_1659 = None
	        permute_29: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_264, [1, 0]);  view_264 = None
	        addmm_13: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_2_fc1_bias, view_265, permute_29);  model_audio_tower_layers_2_fc1_bias = view_265 = permute_29 = None
	        view_266: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_13, [sym_size_int, 1500, 5120]);  addmm_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_1666: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_266, 0.5)
	        mul_1667: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_266, 0.7071067811865476);  view_266 = None
	        erf_4: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_1667);  mul_1667 = None
	        add_2634: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_4, 1);  erf_4 = None
	        mul_1668: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1666, add_2634);  mul_1666 = add_2634 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_1668, [2])
	        amax_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_1668, [2])
	        full_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_17, full_34);  amin_17 = full_34 = None
	        full_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_17, full_35);  amax_17 = full_35 = None
	        sub_793: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_17, minimum_17);  maximum_17 = None
	        div_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_793, 255.0);  sub_793 = None
	        clamp_min_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_34, 1.1920928955078125e-07);  div_34 = None
	        div_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_17, clamp_min_51);  minimum_17 = None
	        round_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_35);  div_35 = None
	        sub_799: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_35);  round_35 = None
	        clamp_min_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_799, -128);  sub_799 = None
	        clamp_max_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_52, 127);  clamp_min_52 = None
	        _assert_tensor_metadata_155 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_51, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_155 = None
	        _assert_tensor_metadata_156 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_34, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_156 = None
	        convert_element_type_102: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_34, torch.int8);  clamp_max_34 = None
	        view_269: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_51, [sym_size_int, 1500, 1])
	        view_270: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_102, [sym_size_int, 1500, 1])
	        reciprocal_17: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_269);  view_269 = None
	        mul_1714: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_17, 1.0);  reciprocal_17 = None
	        mul_1717: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1668, mul_1714);  mul_1668 = mul_1714 = None
	        round_36: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_1717);  mul_1717 = None
	        add_2717: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_36, view_270);  round_36 = view_270 = None
	        clamp_min_53: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2717, -128);  add_2717 = None
	        clamp_max_35: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_53, 127);  clamp_min_53 = None
	        _assert_tensor_metadata_157 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_35, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_157 = None
	        convert_element_type_103: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_35, torch.int8);  clamp_max_35 = None
	        view_273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_51, [sym_size_int, 1500, 1]);  clamp_min_51 = None
	        view_274: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_102, [sym_size_int, 1500, 1]);  convert_element_type_102 = None
	        _assert_tensor_metadata_158 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_103, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_158 = None
	        convert_element_type_104: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_103, torch.float32);  convert_element_type_103 = None
	        _assert_tensor_metadata_159 = torch.ops.aten._assert_tensor_metadata.default(view_274, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_159 = None
	        convert_element_type_105: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_274, torch.float32);  view_274 = None
	        sub_819: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_104, convert_element_type_105);  convert_element_type_104 = convert_element_type_105 = None
	        mul_1739: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_819, view_273);  sub_819 = view_273 = None
	        _assert_tensor_metadata_160 = torch.ops.aten._assert_tensor_metadata.default(mul_1739, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_160 = None
	        view_276: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_2_fc2_parametrizations_weight_original0 = None
	        view_277: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_2_fc2_parametrizations_weight_original1 = None
	        view_278: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_2_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_161 = torch.ops.aten._assert_tensor_metadata.default(view_276, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_161 = None
	        convert_element_type_106: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_276, torch.float32);  view_276 = None
	        _assert_tensor_metadata_162 = torch.ops.aten._assert_tensor_metadata.default(view_278, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_162 = None
	        convert_element_type_107: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_278, torch.float32);  view_278 = None
	        sub_823: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_106, convert_element_type_107);  convert_element_type_106 = convert_element_type_107 = None
	        mul_1744: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_823, view_277);  sub_823 = view_277 = None
	        view_279: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1744, [1280, 5120]);  mul_1744 = None
	        _assert_tensor_metadata_163 = torch.ops.aten._assert_tensor_metadata.default(view_279, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_163 = None
	        mul_1749: "Sym(1500*s6)" = sym_size_int * 1500
	        view_280: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1739, [mul_1749, 5120]);  mul_1739 = mul_1749 = None
	        permute_30: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_279, [1, 0]);  view_279 = None
	        addmm_14: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_2_fc2_bias, view_280, permute_30);  model_audio_tower_layers_2_fc2_bias = view_280 = permute_30 = None
	        view_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_14, [sym_size_int, 1500, 1280]);  addmm_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_2780: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_2482, view_281);  add_2482 = view_281 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_25: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_2780, memory_format = torch.contiguous_format)
	        var_mean_6 = torch.ops.aten.var_mean.correction(clone_25, [2], correction = 0, keepdim = True)
	        getitem_24: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_6[0]
	        getitem_25: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_6[1];  var_mean_6 = None
	        add_2785: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_24, 1e-05);  getitem_24 = None
	        rsqrt_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_2785);  add_2785 = None
	        sub_829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_25, getitem_25);  clone_25 = getitem_25 = None
	        mul_1760: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_829, rsqrt_6);  sub_829 = rsqrt_6 = None
	        mul_1761: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1760, model_audio_tower_layers_3_self_attn_layer_norm_weight);  mul_1760 = model_audio_tower_layers_3_self_attn_layer_norm_weight = None
	        add_2786: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_1761, model_audio_tower_layers_3_self_attn_layer_norm_bias);  mul_1761 = model_audio_tower_layers_3_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_2786, [2])
	        amax_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_2786, [2])
	        full_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_18, full_36);  amin_18 = full_36 = None
	        full_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_18, full_37);  amax_18 = full_37 = None
	        sub_840: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_18, minimum_18);  maximum_18 = None
	        div_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_840, 255.0);  sub_840 = None
	        clamp_min_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_36, 1.1920928955078125e-07);  div_36 = None
	        div_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_18, clamp_min_54);  minimum_18 = None
	        round_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_37);  div_37 = None
	        sub_846: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_37);  round_37 = None
	        clamp_min_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_846, -128);  sub_846 = None
	        clamp_max_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_55, 127);  clamp_min_55 = None
	        _assert_tensor_metadata_164 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_54, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_164 = None
	        _assert_tensor_metadata_165 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_36, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_165 = None
	        convert_element_type_108: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_36, torch.int8);  clamp_max_36 = None
	        view_284: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_54, [sym_size_int, 1500, 1])
	        view_285: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_108, [sym_size_int, 1500, 1])
	        reciprocal_18: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_284);  view_284 = None
	        mul_1809: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_18, 1.0);  reciprocal_18 = None
	        mul_1812: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_2786, mul_1809);  mul_1809 = None
	        round_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1812);  mul_1812 = None
	        add_2873: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_38, view_285);  round_38 = view_285 = None
	        clamp_min_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2873, -128);  add_2873 = None
	        clamp_max_37: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_56, 127);  clamp_min_56 = None
	        _assert_tensor_metadata_166 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_37, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_166 = None
	        convert_element_type_109: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_37, torch.int8);  clamp_max_37 = None
	        view_288: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_54, [sym_size_int, 1500, 1]);  clamp_min_54 = None
	        view_289: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_108, [sym_size_int, 1500, 1]);  convert_element_type_108 = None
	        _assert_tensor_metadata_167 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_109, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_167 = None
	        convert_element_type_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_109, torch.float32);  convert_element_type_109 = None
	        _assert_tensor_metadata_168 = torch.ops.aten._assert_tensor_metadata.default(view_289, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_168 = None
	        convert_element_type_111: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_289, torch.float32);  view_289 = None
	        sub_866: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_110, convert_element_type_111);  convert_element_type_110 = convert_element_type_111 = None
	        mul_1834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_866, view_288);  sub_866 = view_288 = None
	        _assert_tensor_metadata_169 = torch.ops.aten._assert_tensor_metadata.default(mul_1834, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_169 = None
	        view_291: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_292: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_293: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_170 = torch.ops.aten._assert_tensor_metadata.default(view_291, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_170 = None
	        convert_element_type_112: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_291, torch.float32);  view_291 = None
	        _assert_tensor_metadata_171 = torch.ops.aten._assert_tensor_metadata.default(view_293, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_171 = None
	        convert_element_type_113: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_293, torch.float32);  view_293 = None
	        sub_870: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_112, convert_element_type_113);  convert_element_type_112 = convert_element_type_113 = None
	        mul_1839: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_870, view_292);  sub_870 = view_292 = None
	        view_294: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1839, [1280, 1280]);  mul_1839 = None
	        _assert_tensor_metadata_172 = torch.ops.aten._assert_tensor_metadata.default(view_294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_172 = None
	        mul_1844: "Sym(1500*s6)" = sym_size_int * 1500
	        view_295: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1834, [mul_1844, 1280]);  mul_1834 = mul_1844 = None
	        permute_31: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_294, [1, 0]);  view_294 = None
	        addmm_15: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_3_self_attn_q_proj_bias, view_295, permute_31);  model_audio_tower_layers_3_self_attn_q_proj_bias = view_295 = permute_31 = None
	        view_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_15, [sym_size_int, 1500, 1280]);  addmm_15 = None
	        mul_1851: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_296, 0.125);  view_296 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_297: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1851, [sym_size_int, 1500, 20, 64]);  mul_1851 = None
	        permute_32: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_297, [0, 2, 1, 3]);  view_297 = None
	        clone_26: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_32, memory_format = torch.contiguous_format);  permute_32 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_2786, [2])
	        amax_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_2786, [2])
	        full_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_19, full_38);  amin_19 = full_38 = None
	        full_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_19, full_39);  amax_19 = full_39 = None
	        sub_885: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_19, minimum_19);  maximum_19 = None
	        div_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_885, 255.0);  sub_885 = None
	        clamp_min_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_38, 1.1920928955078125e-07);  div_38 = None
	        div_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_19, clamp_min_57);  minimum_19 = None
	        round_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_39);  div_39 = None
	        sub_891: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_39);  round_39 = None
	        clamp_min_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_891, -128);  sub_891 = None
	        clamp_max_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_58, 127);  clamp_min_58 = None
	        _assert_tensor_metadata_173 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_57, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_173 = None
	        _assert_tensor_metadata_174 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_38, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_174 = None
	        convert_element_type_114: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_38, torch.int8);  clamp_max_38 = None
	        view_300: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_57, [sym_size_int, 1500, 1])
	        view_301: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_114, [sym_size_int, 1500, 1])
	        reciprocal_19: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_300);  view_300 = None
	        mul_1905: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_19, 1.0);  reciprocal_19 = None
	        mul_1908: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_2786, mul_1905);  mul_1905 = None
	        round_40: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1908);  mul_1908 = None
	        add_3025: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_40, view_301);  round_40 = view_301 = None
	        clamp_min_59: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3025, -128);  add_3025 = None
	        clamp_max_39: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_59, 127);  clamp_min_59 = None
	        _assert_tensor_metadata_175 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_39, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_175 = None
	        convert_element_type_115: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_39, torch.int8);  clamp_max_39 = None
	        view_304: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_57, [sym_size_int, 1500, 1]);  clamp_min_57 = None
	        view_305: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_114, [sym_size_int, 1500, 1]);  convert_element_type_114 = None
	        _assert_tensor_metadata_176 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_115, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_176 = None
	        convert_element_type_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_115, torch.float32);  convert_element_type_115 = None
	        _assert_tensor_metadata_177 = torch.ops.aten._assert_tensor_metadata.default(view_305, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_177 = None
	        convert_element_type_117: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_305, torch.float32);  view_305 = None
	        sub_911: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_116, convert_element_type_117);  convert_element_type_116 = convert_element_type_117 = None
	        mul_1930: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_911, view_304);  sub_911 = view_304 = None
	        _assert_tensor_metadata_178 = torch.ops.aten._assert_tensor_metadata.default(mul_1930, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_178 = None
	        view_307: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_308: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_309: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_179 = torch.ops.aten._assert_tensor_metadata.default(view_307, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_179 = None
	        convert_element_type_118: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_307, torch.float32);  view_307 = None
	        _assert_tensor_metadata_180 = torch.ops.aten._assert_tensor_metadata.default(view_309, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_180 = None
	        convert_element_type_119: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_309, torch.float32);  view_309 = None
	        sub_915: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_118, convert_element_type_119);  convert_element_type_118 = convert_element_type_119 = None
	        mul_1935: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_915, view_308);  sub_915 = view_308 = None
	        view_310: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1935, [1280, 1280]);  mul_1935 = None
	        _assert_tensor_metadata_181 = torch.ops.aten._assert_tensor_metadata.default(view_310, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_181 = None
	        permute_33: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_310, [1, 0]);  view_310 = None
	        mul_1938: "Sym(1500*s6)" = sym_size_int * 1500
	        view_311: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1930, [mul_1938, 1280]);  mul_1930 = mul_1938 = None
	        mm_3: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_311, permute_33);  view_311 = permute_33 = None
	        view_312: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_3, [sym_size_int, 1500, 1280]);  mm_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_313: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_312, [sym_size_int, -1, 20, 64]);  view_312 = None
	        permute_34: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_313, [0, 2, 1, 3]);  view_313 = None
	        clone_27: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_34, memory_format = torch.contiguous_format);  permute_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_2786, [2])
	        amax_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_2786, [2])
	        full_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_20, full_40);  amin_20 = full_40 = None
	        full_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_20, full_41);  amax_20 = full_41 = None
	        sub_929: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_20, minimum_20);  maximum_20 = None
	        div_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_929, 255.0);  sub_929 = None
	        clamp_min_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_40, 1.1920928955078125e-07);  div_40 = None
	        div_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_20, clamp_min_60);  minimum_20 = None
	        round_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_41);  div_41 = None
	        sub_935: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_41);  round_41 = None
	        clamp_min_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_935, -128);  sub_935 = None
	        clamp_max_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_61, 127);  clamp_min_61 = None
	        _assert_tensor_metadata_182 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_60, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_182 = None
	        _assert_tensor_metadata_183 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_40, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_183 = None
	        convert_element_type_120: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_40, torch.int8);  clamp_max_40 = None
	        view_316: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_60, [sym_size_int, 1500, 1])
	        view_317: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_120, [sym_size_int, 1500, 1])
	        reciprocal_20: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_316);  view_316 = None
	        mul_2004: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_20, 1.0);  reciprocal_20 = None
	        mul_2007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_2786, mul_2004);  add_2786 = mul_2004 = None
	        round_42: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2007);  mul_2007 = None
	        add_3173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_42, view_317);  round_42 = view_317 = None
	        clamp_min_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3173, -128);  add_3173 = None
	        clamp_max_41: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_62, 127);  clamp_min_62 = None
	        _assert_tensor_metadata_184 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_41, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_184 = None
	        convert_element_type_121: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_41, torch.int8);  clamp_max_41 = None
	        view_320: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_60, [sym_size_int, 1500, 1]);  clamp_min_60 = None
	        view_321: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_120, [sym_size_int, 1500, 1]);  convert_element_type_120 = None
	        _assert_tensor_metadata_185 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_121, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_185 = None
	        convert_element_type_122: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_121, torch.float32);  convert_element_type_121 = None
	        _assert_tensor_metadata_186 = torch.ops.aten._assert_tensor_metadata.default(view_321, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_186 = None
	        convert_element_type_123: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_321, torch.float32);  view_321 = None
	        sub_955: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_122, convert_element_type_123);  convert_element_type_122 = convert_element_type_123 = None
	        mul_2029: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_955, view_320);  sub_955 = view_320 = None
	        _assert_tensor_metadata_187 = torch.ops.aten._assert_tensor_metadata.default(mul_2029, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_187 = None
	        view_323: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_324: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_325: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_188 = torch.ops.aten._assert_tensor_metadata.default(view_323, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_188 = None
	        convert_element_type_124: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_323, torch.float32);  view_323 = None
	        _assert_tensor_metadata_189 = torch.ops.aten._assert_tensor_metadata.default(view_325, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_189 = None
	        convert_element_type_125: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_325, torch.float32);  view_325 = None
	        sub_959: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_124, convert_element_type_125);  convert_element_type_124 = convert_element_type_125 = None
	        mul_2034: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_959, view_324);  sub_959 = view_324 = None
	        view_326: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2034, [1280, 1280]);  mul_2034 = None
	        _assert_tensor_metadata_190 = torch.ops.aten._assert_tensor_metadata.default(view_326, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_190 = None
	        mul_2039: "Sym(1500*s6)" = sym_size_int * 1500
	        view_327: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2029, [mul_2039, 1280]);  mul_2029 = mul_2039 = None
	        permute_35: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_326, [1, 0]);  view_326 = None
	        addmm_16: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_3_self_attn_v_proj_bias, view_327, permute_35);  model_audio_tower_layers_3_self_attn_v_proj_bias = view_327 = permute_35 = None
	        view_328: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_16, [sym_size_int, 1500, 1280]);  addmm_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_329: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_328, [sym_size_int, -1, 20, 64]);  view_328 = None
	        permute_36: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_329, [0, 2, 1, 3]);  view_329 = None
	        clone_28: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_36, memory_format = torch.contiguous_format);  permute_36 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_3 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_26, clone_27, clone_28, None, False, scale = 1.0);  clone_26 = clone_27 = clone_28 = None
	        getitem_26: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_3[0];  _scaled_dot_product_efficient_attention_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_37: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_26, [0, 2, 1, 3]);  getitem_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_330: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_37, [sym_size_int, 1500, -1]);  permute_37 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_330, [2])
	        amax_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_330, [2])
	        full_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_21, full_42);  amin_21 = full_42 = None
	        full_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_21, full_43);  amax_21 = full_43 = None
	        sub_977: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_21, minimum_21);  maximum_21 = None
	        div_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_977, 255.0);  sub_977 = None
	        clamp_min_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_42, 1.1920928955078125e-07);  div_42 = None
	        div_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_21, clamp_min_63);  minimum_21 = None
	        round_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_43);  div_43 = None
	        sub_983: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_43);  round_43 = None
	        clamp_min_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_983, -128);  sub_983 = None
	        clamp_max_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_64, 127);  clamp_min_64 = None
	        _assert_tensor_metadata_191 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_63, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_191 = None
	        _assert_tensor_metadata_192 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_42, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_192 = None
	        convert_element_type_126: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_42, torch.int8);  clamp_max_42 = None
	        view_333: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_63, [sym_size_int, 1500, 1])
	        view_334: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_126, [sym_size_int, 1500, 1])
	        reciprocal_21: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_333);  view_333 = None
	        mul_2109: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_21, 1.0);  reciprocal_21 = None
	        mul_2112: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_330, mul_2109);  view_330 = mul_2109 = None
	        round_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2112);  mul_2112 = None
	        add_3337: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_44, view_334);  round_44 = view_334 = None
	        clamp_min_65: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3337, -128);  add_3337 = None
	        clamp_max_43: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_65, 127);  clamp_min_65 = None
	        _assert_tensor_metadata_193 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_43, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_193 = None
	        convert_element_type_127: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_43, torch.int8);  clamp_max_43 = None
	        view_337: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_63, [sym_size_int, 1500, 1]);  clamp_min_63 = None
	        view_338: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_126, [sym_size_int, 1500, 1]);  convert_element_type_126 = None
	        _assert_tensor_metadata_194 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_127, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_194 = None
	        convert_element_type_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_127, torch.float32);  convert_element_type_127 = None
	        _assert_tensor_metadata_195 = torch.ops.aten._assert_tensor_metadata.default(view_338, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_195 = None
	        convert_element_type_129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_338, torch.float32);  view_338 = None
	        sub_1003: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_128, convert_element_type_129);  convert_element_type_128 = convert_element_type_129 = None
	        mul_2134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1003, view_337);  sub_1003 = view_337 = None
	        _assert_tensor_metadata_196 = torch.ops.aten._assert_tensor_metadata.default(mul_2134, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_196 = None
	        view_340: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_341: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_342: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_197 = torch.ops.aten._assert_tensor_metadata.default(view_340, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_197 = None
	        convert_element_type_130: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_340, torch.float32);  view_340 = None
	        _assert_tensor_metadata_198 = torch.ops.aten._assert_tensor_metadata.default(view_342, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_198 = None
	        convert_element_type_131: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_342, torch.float32);  view_342 = None
	        sub_1007: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_130, convert_element_type_131);  convert_element_type_130 = convert_element_type_131 = None
	        mul_2139: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1007, view_341);  sub_1007 = view_341 = None
	        view_343: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2139, [1280, 1280]);  mul_2139 = None
	        _assert_tensor_metadata_199 = torch.ops.aten._assert_tensor_metadata.default(view_343, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_199 = None
	        mul_2144: "Sym(1500*s6)" = sym_size_int * 1500
	        view_344: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2134, [mul_2144, 1280]);  mul_2134 = mul_2144 = None
	        permute_38: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_343, [1, 0]);  view_343 = None
	        addmm_17: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_3_self_attn_out_proj_bias, view_344, permute_38);  model_audio_tower_layers_3_self_attn_out_proj_bias = view_344 = permute_38 = None
	        view_345: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_17, [sym_size_int, 1500, 1280]);  addmm_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_3400: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_2780, view_345);  add_2780 = view_345 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_3400, memory_format = torch.contiguous_format)
	        var_mean_7 = torch.ops.aten.var_mean.correction(clone_30, [2], correction = 0, keepdim = True)
	        getitem_30: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_7[0]
	        getitem_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_7[1];  var_mean_7 = None
	        add_3405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_30, 1e-05);  getitem_30 = None
	        rsqrt_7: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_3405);  add_3405 = None
	        sub_1013: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_30, getitem_31);  clone_30 = getitem_31 = None
	        mul_2155: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1013, rsqrt_7);  sub_1013 = rsqrt_7 = None
	        mul_2156: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2155, model_audio_tower_layers_3_final_layer_norm_weight);  mul_2155 = model_audio_tower_layers_3_final_layer_norm_weight = None
	        add_3406: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2156, model_audio_tower_layers_3_final_layer_norm_bias);  mul_2156 = model_audio_tower_layers_3_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_3406, [2])
	        amax_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_3406, [2])
	        full_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_22, full_44);  amin_22 = full_44 = None
	        full_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_22, full_45);  amax_22 = full_45 = None
	        sub_1024: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_22, minimum_22);  maximum_22 = None
	        div_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1024, 255.0);  sub_1024 = None
	        clamp_min_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_44, 1.1920928955078125e-07);  div_44 = None
	        div_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_22, clamp_min_66);  minimum_22 = None
	        round_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_45);  div_45 = None
	        sub_1030: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_45);  round_45 = None
	        clamp_min_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1030, -128);  sub_1030 = None
	        clamp_max_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_67, 127);  clamp_min_67 = None
	        _assert_tensor_metadata_200 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_66, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_200 = None
	        _assert_tensor_metadata_201 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_44, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_201 = None
	        convert_element_type_132: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_44, torch.int8);  clamp_max_44 = None
	        view_348: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_66, [sym_size_int, 1500, 1])
	        view_349: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_132, [sym_size_int, 1500, 1])
	        reciprocal_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_348);  view_348 = None
	        mul_2204: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_22, 1.0);  reciprocal_22 = None
	        mul_2207: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_3406, mul_2204);  add_3406 = mul_2204 = None
	        round_46: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2207);  mul_2207 = None
	        add_3493: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_46, view_349);  round_46 = view_349 = None
	        clamp_min_68: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3493, -128);  add_3493 = None
	        clamp_max_45: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_68, 127);  clamp_min_68 = None
	        _assert_tensor_metadata_202 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_45, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_202 = None
	        convert_element_type_133: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_45, torch.int8);  clamp_max_45 = None
	        view_352: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_66, [sym_size_int, 1500, 1]);  clamp_min_66 = None
	        view_353: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_132, [sym_size_int, 1500, 1]);  convert_element_type_132 = None
	        _assert_tensor_metadata_203 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_133, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_203 = None
	        convert_element_type_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_133, torch.float32);  convert_element_type_133 = None
	        _assert_tensor_metadata_204 = torch.ops.aten._assert_tensor_metadata.default(view_353, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_204 = None
	        convert_element_type_135: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_353, torch.float32);  view_353 = None
	        sub_1050: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_134, convert_element_type_135);  convert_element_type_134 = convert_element_type_135 = None
	        mul_2229: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1050, view_352);  sub_1050 = view_352 = None
	        _assert_tensor_metadata_205 = torch.ops.aten._assert_tensor_metadata.default(mul_2229, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_205 = None
	        view_355: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_3_fc1_parametrizations_weight_original0 = None
	        view_356: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_3_fc1_parametrizations_weight_original1 = None
	        view_357: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_3_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_206 = torch.ops.aten._assert_tensor_metadata.default(view_355, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_206 = None
	        convert_element_type_136: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_355, torch.float32);  view_355 = None
	        _assert_tensor_metadata_207 = torch.ops.aten._assert_tensor_metadata.default(view_357, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_207 = None
	        convert_element_type_137: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_357, torch.float32);  view_357 = None
	        sub_1054: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_136, convert_element_type_137);  convert_element_type_136 = convert_element_type_137 = None
	        mul_2234: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1054, view_356);  sub_1054 = view_356 = None
	        view_358: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2234, [5120, 1280]);  mul_2234 = None
	        _assert_tensor_metadata_208 = torch.ops.aten._assert_tensor_metadata.default(view_358, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_208 = None
	        mul_2239: "Sym(1500*s6)" = sym_size_int * 1500
	        view_359: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2229, [mul_2239, 1280]);  mul_2229 = mul_2239 = None
	        permute_39: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_358, [1, 0]);  view_358 = None
	        addmm_18: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_3_fc1_bias, view_359, permute_39);  model_audio_tower_layers_3_fc1_bias = view_359 = permute_39 = None
	        view_360: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_18, [sym_size_int, 1500, 5120]);  addmm_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_2246: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_360, 0.5)
	        mul_2247: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_360, 0.7071067811865476);  view_360 = None
	        erf_5: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_2247);  mul_2247 = None
	        add_3552: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_5, 1);  erf_5 = None
	        mul_2248: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2246, add_3552);  mul_2246 = add_3552 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_2248, [2])
	        amax_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_2248, [2])
	        full_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_23, full_46);  amin_23 = full_46 = None
	        full_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_23, full_47);  amax_23 = full_47 = None
	        sub_1067: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_23, minimum_23);  maximum_23 = None
	        div_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1067, 255.0);  sub_1067 = None
	        clamp_min_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_46, 1.1920928955078125e-07);  div_46 = None
	        div_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_23, clamp_min_69);  minimum_23 = None
	        round_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_47);  div_47 = None
	        sub_1073: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_47);  round_47 = None
	        clamp_min_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1073, -128);  sub_1073 = None
	        clamp_max_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_70, 127);  clamp_min_70 = None
	        _assert_tensor_metadata_209 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_69, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_209 = None
	        _assert_tensor_metadata_210 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_46, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_210 = None
	        convert_element_type_138: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_46, torch.int8);  clamp_max_46 = None
	        view_363: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_69, [sym_size_int, 1500, 1])
	        view_364: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_138, [sym_size_int, 1500, 1])
	        reciprocal_23: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_363);  view_363 = None
	        mul_2294: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_23, 1.0);  reciprocal_23 = None
	        mul_2297: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2248, mul_2294);  mul_2248 = mul_2294 = None
	        round_48: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_2297);  mul_2297 = None
	        add_3635: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_48, view_364);  round_48 = view_364 = None
	        clamp_min_71: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3635, -128);  add_3635 = None
	        clamp_max_47: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_71, 127);  clamp_min_71 = None
	        _assert_tensor_metadata_211 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_47, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_211 = None
	        convert_element_type_139: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_47, torch.int8);  clamp_max_47 = None
	        view_367: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_69, [sym_size_int, 1500, 1]);  clamp_min_69 = None
	        view_368: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_138, [sym_size_int, 1500, 1]);  convert_element_type_138 = None
	        _assert_tensor_metadata_212 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_139, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_212 = None
	        convert_element_type_140: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_139, torch.float32);  convert_element_type_139 = None
	        _assert_tensor_metadata_213 = torch.ops.aten._assert_tensor_metadata.default(view_368, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_213 = None
	        convert_element_type_141: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_368, torch.float32);  view_368 = None
	        sub_1093: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_140, convert_element_type_141);  convert_element_type_140 = convert_element_type_141 = None
	        mul_2319: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1093, view_367);  sub_1093 = view_367 = None
	        _assert_tensor_metadata_214 = torch.ops.aten._assert_tensor_metadata.default(mul_2319, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_214 = None
	        view_370: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_3_fc2_parametrizations_weight_original0 = None
	        view_371: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_3_fc2_parametrizations_weight_original1 = None
	        view_372: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_3_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_215 = torch.ops.aten._assert_tensor_metadata.default(view_370, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_215 = None
	        convert_element_type_142: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_370, torch.float32);  view_370 = None
	        _assert_tensor_metadata_216 = torch.ops.aten._assert_tensor_metadata.default(view_372, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_216 = None
	        convert_element_type_143: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_372, torch.float32);  view_372 = None
	        sub_1097: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_142, convert_element_type_143);  convert_element_type_142 = convert_element_type_143 = None
	        mul_2324: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1097, view_371);  sub_1097 = view_371 = None
	        view_373: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2324, [1280, 5120]);  mul_2324 = None
	        _assert_tensor_metadata_217 = torch.ops.aten._assert_tensor_metadata.default(view_373, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_217 = None
	        mul_2329: "Sym(1500*s6)" = sym_size_int * 1500
	        view_374: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2319, [mul_2329, 5120]);  mul_2319 = mul_2329 = None
	        permute_40: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_373, [1, 0]);  view_373 = None
	        addmm_19: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_3_fc2_bias, view_374, permute_40);  model_audio_tower_layers_3_fc2_bias = view_374 = permute_40 = None
	        view_375: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_19, [sym_size_int, 1500, 1280]);  addmm_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_3698: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_3400, view_375);  add_3400 = view_375 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_33: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_3698, memory_format = torch.contiguous_format)
	        var_mean_8 = torch.ops.aten.var_mean.correction(clone_33, [2], correction = 0, keepdim = True)
	        getitem_32: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_8[0]
	        getitem_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_8[1];  var_mean_8 = None
	        add_3703: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_32, 1e-05);  getitem_32 = None
	        rsqrt_8: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_3703);  add_3703 = None
	        sub_1103: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_33, getitem_33);  clone_33 = getitem_33 = None
	        mul_2340: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1103, rsqrt_8);  sub_1103 = rsqrt_8 = None
	        mul_2341: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2340, model_audio_tower_layers_4_self_attn_layer_norm_weight);  mul_2340 = model_audio_tower_layers_4_self_attn_layer_norm_weight = None
	        add_3704: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2341, model_audio_tower_layers_4_self_attn_layer_norm_bias);  mul_2341 = model_audio_tower_layers_4_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_3704, [2])
	        amax_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_3704, [2])
	        full_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_24, full_48);  amin_24 = full_48 = None
	        full_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_24, full_49);  amax_24 = full_49 = None
	        sub_1114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_24, minimum_24);  maximum_24 = None
	        div_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1114, 255.0);  sub_1114 = None
	        clamp_min_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_48, 1.1920928955078125e-07);  div_48 = None
	        div_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_24, clamp_min_72);  minimum_24 = None
	        round_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_49);  div_49 = None
	        sub_1120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_49);  round_49 = None
	        clamp_min_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1120, -128);  sub_1120 = None
	        clamp_max_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_73, 127);  clamp_min_73 = None
	        _assert_tensor_metadata_218 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_72, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_218 = None
	        _assert_tensor_metadata_219 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_48, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_219 = None
	        convert_element_type_144: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_48, torch.int8);  clamp_max_48 = None
	        view_378: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_72, [sym_size_int, 1500, 1])
	        view_379: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_144, [sym_size_int, 1500, 1])
	        reciprocal_24: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_378);  view_378 = None
	        mul_2389: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_24, 1.0);  reciprocal_24 = None
	        mul_2392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_3704, mul_2389);  mul_2389 = None
	        round_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2392);  mul_2392 = None
	        add_3791: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_50, view_379);  round_50 = view_379 = None
	        clamp_min_74: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3791, -128);  add_3791 = None
	        clamp_max_49: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_74, 127);  clamp_min_74 = None
	        _assert_tensor_metadata_220 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_49, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_220 = None
	        convert_element_type_145: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_49, torch.int8);  clamp_max_49 = None
	        view_382: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_72, [sym_size_int, 1500, 1]);  clamp_min_72 = None
	        view_383: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_144, [sym_size_int, 1500, 1]);  convert_element_type_144 = None
	        _assert_tensor_metadata_221 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_145, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_221 = None
	        convert_element_type_146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_145, torch.float32);  convert_element_type_145 = None
	        _assert_tensor_metadata_222 = torch.ops.aten._assert_tensor_metadata.default(view_383, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_222 = None
	        convert_element_type_147: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_383, torch.float32);  view_383 = None
	        sub_1140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_146, convert_element_type_147);  convert_element_type_146 = convert_element_type_147 = None
	        mul_2414: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1140, view_382);  sub_1140 = view_382 = None
	        _assert_tensor_metadata_223 = torch.ops.aten._assert_tensor_metadata.default(mul_2414, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_223 = None
	        view_385: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_386: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_387: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_224 = torch.ops.aten._assert_tensor_metadata.default(view_385, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_224 = None
	        convert_element_type_148: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_385, torch.float32);  view_385 = None
	        _assert_tensor_metadata_225 = torch.ops.aten._assert_tensor_metadata.default(view_387, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_225 = None
	        convert_element_type_149: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_387, torch.float32);  view_387 = None
	        sub_1144: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_148, convert_element_type_149);  convert_element_type_148 = convert_element_type_149 = None
	        mul_2419: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1144, view_386);  sub_1144 = view_386 = None
	        view_388: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2419, [1280, 1280]);  mul_2419 = None
	        _assert_tensor_metadata_226 = torch.ops.aten._assert_tensor_metadata.default(view_388, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_226 = None
	        mul_2424: "Sym(1500*s6)" = sym_size_int * 1500
	        view_389: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2414, [mul_2424, 1280]);  mul_2414 = mul_2424 = None
	        permute_41: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_388, [1, 0]);  view_388 = None
	        addmm_20: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_4_self_attn_q_proj_bias, view_389, permute_41);  model_audio_tower_layers_4_self_attn_q_proj_bias = view_389 = permute_41 = None
	        view_390: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_20, [sym_size_int, 1500, 1280]);  addmm_20 = None
	        mul_2431: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_390, 0.125);  view_390 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_391: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2431, [sym_size_int, 1500, 20, 64]);  mul_2431 = None
	        permute_42: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_391, [0, 2, 1, 3]);  view_391 = None
	        clone_34: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_42, memory_format = torch.contiguous_format);  permute_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_3704, [2])
	        amax_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_3704, [2])
	        full_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_25, full_50);  amin_25 = full_50 = None
	        full_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_25, full_51);  amax_25 = full_51 = None
	        sub_1159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_25, minimum_25);  maximum_25 = None
	        div_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1159, 255.0);  sub_1159 = None
	        clamp_min_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_50, 1.1920928955078125e-07);  div_50 = None
	        div_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_25, clamp_min_75);  minimum_25 = None
	        round_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_51);  div_51 = None
	        sub_1165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_51);  round_51 = None
	        clamp_min_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1165, -128);  sub_1165 = None
	        clamp_max_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_76, 127);  clamp_min_76 = None
	        _assert_tensor_metadata_227 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_75, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_227 = None
	        _assert_tensor_metadata_228 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_50, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_228 = None
	        convert_element_type_150: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_50, torch.int8);  clamp_max_50 = None
	        view_394: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_75, [sym_size_int, 1500, 1])
	        view_395: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_150, [sym_size_int, 1500, 1])
	        reciprocal_25: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_394);  view_394 = None
	        mul_2485: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_25, 1.0);  reciprocal_25 = None
	        mul_2488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_3704, mul_2485);  mul_2485 = None
	        round_52: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2488);  mul_2488 = None
	        add_3943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_52, view_395);  round_52 = view_395 = None
	        clamp_min_77: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3943, -128);  add_3943 = None
	        clamp_max_51: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_77, 127);  clamp_min_77 = None
	        _assert_tensor_metadata_229 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_51, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_229 = None
	        convert_element_type_151: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_51, torch.int8);  clamp_max_51 = None
	        view_398: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_75, [sym_size_int, 1500, 1]);  clamp_min_75 = None
	        view_399: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_150, [sym_size_int, 1500, 1]);  convert_element_type_150 = None
	        _assert_tensor_metadata_230 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_151, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_230 = None
	        convert_element_type_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_151, torch.float32);  convert_element_type_151 = None
	        _assert_tensor_metadata_231 = torch.ops.aten._assert_tensor_metadata.default(view_399, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_231 = None
	        convert_element_type_153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_399, torch.float32);  view_399 = None
	        sub_1185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_152, convert_element_type_153);  convert_element_type_152 = convert_element_type_153 = None
	        mul_2510: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1185, view_398);  sub_1185 = view_398 = None
	        _assert_tensor_metadata_232 = torch.ops.aten._assert_tensor_metadata.default(mul_2510, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_232 = None
	        view_401: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_402: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_403: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_233 = torch.ops.aten._assert_tensor_metadata.default(view_401, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_233 = None
	        convert_element_type_154: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_401, torch.float32);  view_401 = None
	        _assert_tensor_metadata_234 = torch.ops.aten._assert_tensor_metadata.default(view_403, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_234 = None
	        convert_element_type_155: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_403, torch.float32);  view_403 = None
	        sub_1189: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_154, convert_element_type_155);  convert_element_type_154 = convert_element_type_155 = None
	        mul_2515: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1189, view_402);  sub_1189 = view_402 = None
	        view_404: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2515, [1280, 1280]);  mul_2515 = None
	        _assert_tensor_metadata_235 = torch.ops.aten._assert_tensor_metadata.default(view_404, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_235 = None
	        permute_43: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_404, [1, 0]);  view_404 = None
	        mul_2518: "Sym(1500*s6)" = sym_size_int * 1500
	        view_405: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2510, [mul_2518, 1280]);  mul_2510 = mul_2518 = None
	        mm_4: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_405, permute_43);  view_405 = permute_43 = None
	        view_406: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_4, [sym_size_int, 1500, 1280]);  mm_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_407: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_406, [sym_size_int, -1, 20, 64]);  view_406 = None
	        permute_44: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_407, [0, 2, 1, 3]);  view_407 = None
	        clone_35: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_44, memory_format = torch.contiguous_format);  permute_44 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_3704, [2])
	        amax_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_3704, [2])
	        full_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_26, full_52);  amin_26 = full_52 = None
	        full_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_26, full_53);  amax_26 = full_53 = None
	        sub_1203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_26, minimum_26);  maximum_26 = None
	        div_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1203, 255.0);  sub_1203 = None
	        clamp_min_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_52, 1.1920928955078125e-07);  div_52 = None
	        div_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_26, clamp_min_78);  minimum_26 = None
	        round_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_53);  div_53 = None
	        sub_1209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_53);  round_53 = None
	        clamp_min_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1209, -128);  sub_1209 = None
	        clamp_max_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_79, 127);  clamp_min_79 = None
	        _assert_tensor_metadata_236 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_78, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_236 = None
	        _assert_tensor_metadata_237 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_52, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_237 = None
	        convert_element_type_156: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_52, torch.int8);  clamp_max_52 = None
	        view_410: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_78, [sym_size_int, 1500, 1])
	        view_411: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_156, [sym_size_int, 1500, 1])
	        reciprocal_26: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_410);  view_410 = None
	        mul_2584: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_26, 1.0);  reciprocal_26 = None
	        mul_2587: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_3704, mul_2584);  add_3704 = mul_2584 = None
	        round_54: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2587);  mul_2587 = None
	        add_4091: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_54, view_411);  round_54 = view_411 = None
	        clamp_min_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4091, -128);  add_4091 = None
	        clamp_max_53: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_80, 127);  clamp_min_80 = None
	        _assert_tensor_metadata_238 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_53, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_238 = None
	        convert_element_type_157: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_53, torch.int8);  clamp_max_53 = None
	        view_414: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_78, [sym_size_int, 1500, 1]);  clamp_min_78 = None
	        view_415: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_156, [sym_size_int, 1500, 1]);  convert_element_type_156 = None
	        _assert_tensor_metadata_239 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_157, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_239 = None
	        convert_element_type_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_157, torch.float32);  convert_element_type_157 = None
	        _assert_tensor_metadata_240 = torch.ops.aten._assert_tensor_metadata.default(view_415, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_240 = None
	        convert_element_type_159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_415, torch.float32);  view_415 = None
	        sub_1229: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_158, convert_element_type_159);  convert_element_type_158 = convert_element_type_159 = None
	        mul_2609: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1229, view_414);  sub_1229 = view_414 = None
	        _assert_tensor_metadata_241 = torch.ops.aten._assert_tensor_metadata.default(mul_2609, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_241 = None
	        view_417: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_418: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_419: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_242 = torch.ops.aten._assert_tensor_metadata.default(view_417, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_242 = None
	        convert_element_type_160: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_417, torch.float32);  view_417 = None
	        _assert_tensor_metadata_243 = torch.ops.aten._assert_tensor_metadata.default(view_419, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_243 = None
	        convert_element_type_161: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_419, torch.float32);  view_419 = None
	        sub_1233: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_160, convert_element_type_161);  convert_element_type_160 = convert_element_type_161 = None
	        mul_2614: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1233, view_418);  sub_1233 = view_418 = None
	        view_420: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2614, [1280, 1280]);  mul_2614 = None
	        _assert_tensor_metadata_244 = torch.ops.aten._assert_tensor_metadata.default(view_420, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_244 = None
	        mul_2619: "Sym(1500*s6)" = sym_size_int * 1500
	        view_421: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2609, [mul_2619, 1280]);  mul_2609 = mul_2619 = None
	        permute_45: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_420, [1, 0]);  view_420 = None
	        addmm_21: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_4_self_attn_v_proj_bias, view_421, permute_45);  model_audio_tower_layers_4_self_attn_v_proj_bias = view_421 = permute_45 = None
	        view_422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_21, [sym_size_int, 1500, 1280]);  addmm_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_423: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_422, [sym_size_int, -1, 20, 64]);  view_422 = None
	        permute_46: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_423, [0, 2, 1, 3]);  view_423 = None
	        clone_36: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_46, memory_format = torch.contiguous_format);  permute_46 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_4 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_34, clone_35, clone_36, None, False, scale = 1.0);  clone_34 = clone_35 = clone_36 = None
	        getitem_34: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_4[0];  _scaled_dot_product_efficient_attention_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_47: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_34, [0, 2, 1, 3]);  getitem_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_424: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_47, [sym_size_int, 1500, -1]);  permute_47 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_424, [2])
	        amax_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_424, [2])
	        full_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_27, full_54);  amin_27 = full_54 = None
	        full_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_27, full_55);  amax_27 = full_55 = None
	        sub_1251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_27, minimum_27);  maximum_27 = None
	        div_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1251, 255.0);  sub_1251 = None
	        clamp_min_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_54, 1.1920928955078125e-07);  div_54 = None
	        div_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_27, clamp_min_81);  minimum_27 = None
	        round_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_55);  div_55 = None
	        sub_1257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_55);  round_55 = None
	        clamp_min_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1257, -128);  sub_1257 = None
	        clamp_max_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_82, 127);  clamp_min_82 = None
	        _assert_tensor_metadata_245 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_81, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_245 = None
	        _assert_tensor_metadata_246 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_54, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_246 = None
	        convert_element_type_162: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_54, torch.int8);  clamp_max_54 = None
	        view_427: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_81, [sym_size_int, 1500, 1])
	        view_428: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_162, [sym_size_int, 1500, 1])
	        reciprocal_27: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_427);  view_427 = None
	        mul_2689: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_27, 1.0);  reciprocal_27 = None
	        mul_2692: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_424, mul_2689);  view_424 = mul_2689 = None
	        round_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2692);  mul_2692 = None
	        add_4255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_56, view_428);  round_56 = view_428 = None
	        clamp_min_83: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4255, -128);  add_4255 = None
	        clamp_max_55: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_83, 127);  clamp_min_83 = None
	        _assert_tensor_metadata_247 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_55, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_247 = None
	        convert_element_type_163: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_55, torch.int8);  clamp_max_55 = None
	        view_431: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_81, [sym_size_int, 1500, 1]);  clamp_min_81 = None
	        view_432: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_162, [sym_size_int, 1500, 1]);  convert_element_type_162 = None
	        _assert_tensor_metadata_248 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_163, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_248 = None
	        convert_element_type_164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_163, torch.float32);  convert_element_type_163 = None
	        _assert_tensor_metadata_249 = torch.ops.aten._assert_tensor_metadata.default(view_432, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_249 = None
	        convert_element_type_165: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_432, torch.float32);  view_432 = None
	        sub_1277: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_164, convert_element_type_165);  convert_element_type_164 = convert_element_type_165 = None
	        mul_2714: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1277, view_431);  sub_1277 = view_431 = None
	        _assert_tensor_metadata_250 = torch.ops.aten._assert_tensor_metadata.default(mul_2714, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_250 = None
	        view_434: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_435: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_436: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_251 = torch.ops.aten._assert_tensor_metadata.default(view_434, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_251 = None
	        convert_element_type_166: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_434, torch.float32);  view_434 = None
	        _assert_tensor_metadata_252 = torch.ops.aten._assert_tensor_metadata.default(view_436, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_252 = None
	        convert_element_type_167: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_436, torch.float32);  view_436 = None
	        sub_1281: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_166, convert_element_type_167);  convert_element_type_166 = convert_element_type_167 = None
	        mul_2719: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1281, view_435);  sub_1281 = view_435 = None
	        view_437: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2719, [1280, 1280]);  mul_2719 = None
	        _assert_tensor_metadata_253 = torch.ops.aten._assert_tensor_metadata.default(view_437, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_253 = None
	        mul_2724: "Sym(1500*s6)" = sym_size_int * 1500
	        view_438: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2714, [mul_2724, 1280]);  mul_2714 = mul_2724 = None
	        permute_48: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_437, [1, 0]);  view_437 = None
	        addmm_22: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_4_self_attn_out_proj_bias, view_438, permute_48);  model_audio_tower_layers_4_self_attn_out_proj_bias = view_438 = permute_48 = None
	        view_439: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_22, [sym_size_int, 1500, 1280]);  addmm_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_4318: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_3698, view_439);  add_3698 = view_439 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_4318, memory_format = torch.contiguous_format)
	        var_mean_9 = torch.ops.aten.var_mean.correction(clone_38, [2], correction = 0, keepdim = True)
	        getitem_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_9[0]
	        getitem_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_9[1];  var_mean_9 = None
	        add_4323: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_38, 1e-05);  getitem_38 = None
	        rsqrt_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_4323);  add_4323 = None
	        sub_1287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_38, getitem_39);  clone_38 = getitem_39 = None
	        mul_2735: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1287, rsqrt_9);  sub_1287 = rsqrt_9 = None
	        mul_2736: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2735, model_audio_tower_layers_4_final_layer_norm_weight);  mul_2735 = model_audio_tower_layers_4_final_layer_norm_weight = None
	        add_4324: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2736, model_audio_tower_layers_4_final_layer_norm_bias);  mul_2736 = model_audio_tower_layers_4_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_4324, [2])
	        amax_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_4324, [2])
	        full_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_28, full_56);  amin_28 = full_56 = None
	        full_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_28, full_57);  amax_28 = full_57 = None
	        sub_1298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_28, minimum_28);  maximum_28 = None
	        div_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1298, 255.0);  sub_1298 = None
	        clamp_min_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_56, 1.1920928955078125e-07);  div_56 = None
	        div_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_28, clamp_min_84);  minimum_28 = None
	        round_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_57);  div_57 = None
	        sub_1304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_57);  round_57 = None
	        clamp_min_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1304, -128);  sub_1304 = None
	        clamp_max_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_85, 127);  clamp_min_85 = None
	        _assert_tensor_metadata_254 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_84, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_254 = None
	        _assert_tensor_metadata_255 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_56, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_255 = None
	        convert_element_type_168: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_56, torch.int8);  clamp_max_56 = None
	        view_442: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_84, [sym_size_int, 1500, 1])
	        view_443: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_168, [sym_size_int, 1500, 1])
	        reciprocal_28: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_442);  view_442 = None
	        mul_2784: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_28, 1.0);  reciprocal_28 = None
	        mul_2787: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_4324, mul_2784);  add_4324 = mul_2784 = None
	        round_58: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2787);  mul_2787 = None
	        add_4411: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_58, view_443);  round_58 = view_443 = None
	        clamp_min_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4411, -128);  add_4411 = None
	        clamp_max_57: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_86, 127);  clamp_min_86 = None
	        _assert_tensor_metadata_256 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_57, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_256 = None
	        convert_element_type_169: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_57, torch.int8);  clamp_max_57 = None
	        view_446: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_84, [sym_size_int, 1500, 1]);  clamp_min_84 = None
	        view_447: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_168, [sym_size_int, 1500, 1]);  convert_element_type_168 = None
	        _assert_tensor_metadata_257 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_169, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_257 = None
	        convert_element_type_170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_169, torch.float32);  convert_element_type_169 = None
	        _assert_tensor_metadata_258 = torch.ops.aten._assert_tensor_metadata.default(view_447, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_258 = None
	        convert_element_type_171: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_447, torch.float32);  view_447 = None
	        sub_1324: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_170, convert_element_type_171);  convert_element_type_170 = convert_element_type_171 = None
	        mul_2809: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1324, view_446);  sub_1324 = view_446 = None
	        _assert_tensor_metadata_259 = torch.ops.aten._assert_tensor_metadata.default(mul_2809, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_259 = None
	        view_449: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_4_fc1_parametrizations_weight_original0 = None
	        view_450: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_4_fc1_parametrizations_weight_original1 = None
	        view_451: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_4_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_260 = torch.ops.aten._assert_tensor_metadata.default(view_449, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_260 = None
	        convert_element_type_172: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_449, torch.float32);  view_449 = None
	        _assert_tensor_metadata_261 = torch.ops.aten._assert_tensor_metadata.default(view_451, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_261 = None
	        convert_element_type_173: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_451, torch.float32);  view_451 = None
	        sub_1328: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_172, convert_element_type_173);  convert_element_type_172 = convert_element_type_173 = None
	        mul_2814: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1328, view_450);  sub_1328 = view_450 = None
	        view_452: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2814, [5120, 1280]);  mul_2814 = None
	        _assert_tensor_metadata_262 = torch.ops.aten._assert_tensor_metadata.default(view_452, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_262 = None
	        mul_2819: "Sym(1500*s6)" = sym_size_int * 1500
	        view_453: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2809, [mul_2819, 1280]);  mul_2809 = mul_2819 = None
	        permute_49: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_452, [1, 0]);  view_452 = None
	        addmm_23: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_4_fc1_bias, view_453, permute_49);  model_audio_tower_layers_4_fc1_bias = view_453 = permute_49 = None
	        view_454: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_23, [sym_size_int, 1500, 5120]);  addmm_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_2826: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_454, 0.5)
	        mul_2827: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_454, 0.7071067811865476);  view_454 = None
	        erf_6: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_2827);  mul_2827 = None
	        add_4470: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_6, 1);  erf_6 = None
	        mul_2828: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2826, add_4470);  mul_2826 = add_4470 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_2828, [2])
	        amax_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_2828, [2])
	        full_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_29, full_58);  amin_29 = full_58 = None
	        full_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_29, full_59);  amax_29 = full_59 = None
	        sub_1341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_29, minimum_29);  maximum_29 = None
	        div_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1341, 255.0);  sub_1341 = None
	        clamp_min_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_58, 1.1920928955078125e-07);  div_58 = None
	        div_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_29, clamp_min_87);  minimum_29 = None
	        round_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_59);  div_59 = None
	        sub_1347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_59);  round_59 = None
	        clamp_min_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1347, -128);  sub_1347 = None
	        clamp_max_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_88, 127);  clamp_min_88 = None
	        _assert_tensor_metadata_263 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_87, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_263 = None
	        _assert_tensor_metadata_264 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_58, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_264 = None
	        convert_element_type_174: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_58, torch.int8);  clamp_max_58 = None
	        view_457: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_87, [sym_size_int, 1500, 1])
	        view_458: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_174, [sym_size_int, 1500, 1])
	        reciprocal_29: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_457);  view_457 = None
	        mul_2874: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_29, 1.0);  reciprocal_29 = None
	        mul_2877: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2828, mul_2874);  mul_2828 = mul_2874 = None
	        round_60: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_2877);  mul_2877 = None
	        add_4553: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_60, view_458);  round_60 = view_458 = None
	        clamp_min_89: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4553, -128);  add_4553 = None
	        clamp_max_59: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_89, 127);  clamp_min_89 = None
	        _assert_tensor_metadata_265 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_59, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_265 = None
	        convert_element_type_175: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_59, torch.int8);  clamp_max_59 = None
	        view_461: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_87, [sym_size_int, 1500, 1]);  clamp_min_87 = None
	        view_462: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_174, [sym_size_int, 1500, 1]);  convert_element_type_174 = None
	        _assert_tensor_metadata_266 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_175, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_266 = None
	        convert_element_type_176: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_175, torch.float32);  convert_element_type_175 = None
	        _assert_tensor_metadata_267 = torch.ops.aten._assert_tensor_metadata.default(view_462, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_267 = None
	        convert_element_type_177: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_462, torch.float32);  view_462 = None
	        sub_1367: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_176, convert_element_type_177);  convert_element_type_176 = convert_element_type_177 = None
	        mul_2899: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1367, view_461);  sub_1367 = view_461 = None
	        _assert_tensor_metadata_268 = torch.ops.aten._assert_tensor_metadata.default(mul_2899, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_268 = None
	        view_464: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_4_fc2_parametrizations_weight_original0 = None
	        view_465: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_4_fc2_parametrizations_weight_original1 = None
	        view_466: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_4_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_269 = torch.ops.aten._assert_tensor_metadata.default(view_464, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_269 = None
	        convert_element_type_178: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_464, torch.float32);  view_464 = None
	        _assert_tensor_metadata_270 = torch.ops.aten._assert_tensor_metadata.default(view_466, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_270 = None
	        convert_element_type_179: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_466, torch.float32);  view_466 = None
	        sub_1371: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_178, convert_element_type_179);  convert_element_type_178 = convert_element_type_179 = None
	        mul_2904: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1371, view_465);  sub_1371 = view_465 = None
	        view_467: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2904, [1280, 5120]);  mul_2904 = None
	        _assert_tensor_metadata_271 = torch.ops.aten._assert_tensor_metadata.default(view_467, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_271 = None
	        mul_2909: "Sym(1500*s6)" = sym_size_int * 1500
	        view_468: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2899, [mul_2909, 5120]);  mul_2899 = mul_2909 = None
	        permute_50: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_467, [1, 0]);  view_467 = None
	        addmm_24: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_4_fc2_bias, view_468, permute_50);  model_audio_tower_layers_4_fc2_bias = view_468 = permute_50 = None
	        view_469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_24, [sym_size_int, 1500, 1280]);  addmm_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_4616: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_4318, view_469);  add_4318 = view_469 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_41: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_4616, memory_format = torch.contiguous_format)
	        var_mean_10 = torch.ops.aten.var_mean.correction(clone_41, [2], correction = 0, keepdim = True)
	        getitem_40: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_10[0]
	        getitem_41: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_10[1];  var_mean_10 = None
	        add_4621: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_40, 1e-05);  getitem_40 = None
	        rsqrt_10: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_4621);  add_4621 = None
	        sub_1377: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_41, getitem_41);  clone_41 = getitem_41 = None
	        mul_2920: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1377, rsqrt_10);  sub_1377 = rsqrt_10 = None
	        mul_2921: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2920, model_audio_tower_layers_5_self_attn_layer_norm_weight);  mul_2920 = model_audio_tower_layers_5_self_attn_layer_norm_weight = None
	        add_4622: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2921, model_audio_tower_layers_5_self_attn_layer_norm_bias);  mul_2921 = model_audio_tower_layers_5_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_4622, [2])
	        amax_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_4622, [2])
	        full_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_30, full_60);  amin_30 = full_60 = None
	        full_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_30, full_61);  amax_30 = full_61 = None
	        sub_1388: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_30, minimum_30);  maximum_30 = None
	        div_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1388, 255.0);  sub_1388 = None
	        clamp_min_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_60, 1.1920928955078125e-07);  div_60 = None
	        div_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_30, clamp_min_90);  minimum_30 = None
	        round_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_61);  div_61 = None
	        sub_1394: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_61);  round_61 = None
	        clamp_min_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1394, -128);  sub_1394 = None
	        clamp_max_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_91, 127);  clamp_min_91 = None
	        _assert_tensor_metadata_272 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_90, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_272 = None
	        _assert_tensor_metadata_273 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_60, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_273 = None
	        convert_element_type_180: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_60, torch.int8);  clamp_max_60 = None
	        view_472: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_90, [sym_size_int, 1500, 1])
	        view_473: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_180, [sym_size_int, 1500, 1])
	        reciprocal_30: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_472);  view_472 = None
	        mul_2969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_30, 1.0);  reciprocal_30 = None
	        mul_2972: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_4622, mul_2969);  mul_2969 = None
	        round_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2972);  mul_2972 = None
	        add_4709: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_62, view_473);  round_62 = view_473 = None
	        clamp_min_92: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4709, -128);  add_4709 = None
	        clamp_max_61: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_92, 127);  clamp_min_92 = None
	        _assert_tensor_metadata_274 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_61, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_274 = None
	        convert_element_type_181: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_61, torch.int8);  clamp_max_61 = None
	        view_476: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_90, [sym_size_int, 1500, 1]);  clamp_min_90 = None
	        view_477: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_180, [sym_size_int, 1500, 1]);  convert_element_type_180 = None
	        _assert_tensor_metadata_275 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_181, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_275 = None
	        convert_element_type_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_181, torch.float32);  convert_element_type_181 = None
	        _assert_tensor_metadata_276 = torch.ops.aten._assert_tensor_metadata.default(view_477, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_276 = None
	        convert_element_type_183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_477, torch.float32);  view_477 = None
	        sub_1414: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_182, convert_element_type_183);  convert_element_type_182 = convert_element_type_183 = None
	        mul_2994: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1414, view_476);  sub_1414 = view_476 = None
	        _assert_tensor_metadata_277 = torch.ops.aten._assert_tensor_metadata.default(mul_2994, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_277 = None
	        view_479: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_480: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_481: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_278 = torch.ops.aten._assert_tensor_metadata.default(view_479, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_278 = None
	        convert_element_type_184: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_479, torch.float32);  view_479 = None
	        _assert_tensor_metadata_279 = torch.ops.aten._assert_tensor_metadata.default(view_481, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_279 = None
	        convert_element_type_185: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_481, torch.float32);  view_481 = None
	        sub_1418: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_184, convert_element_type_185);  convert_element_type_184 = convert_element_type_185 = None
	        mul_2999: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1418, view_480);  sub_1418 = view_480 = None
	        view_482: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2999, [1280, 1280]);  mul_2999 = None
	        _assert_tensor_metadata_280 = torch.ops.aten._assert_tensor_metadata.default(view_482, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_280 = None
	        mul_3004: "Sym(1500*s6)" = sym_size_int * 1500
	        view_483: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2994, [mul_3004, 1280]);  mul_2994 = mul_3004 = None
	        permute_51: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_482, [1, 0]);  view_482 = None
	        addmm_25: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_5_self_attn_q_proj_bias, view_483, permute_51);  model_audio_tower_layers_5_self_attn_q_proj_bias = view_483 = permute_51 = None
	        view_484: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_25, [sym_size_int, 1500, 1280]);  addmm_25 = None
	        mul_3011: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_484, 0.125);  view_484 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_485: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3011, [sym_size_int, 1500, 20, 64]);  mul_3011 = None
	        permute_52: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_485, [0, 2, 1, 3]);  view_485 = None
	        clone_42: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_52, memory_format = torch.contiguous_format);  permute_52 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_4622, [2])
	        amax_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_4622, [2])
	        full_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_31, full_62);  amin_31 = full_62 = None
	        full_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_31, full_63);  amax_31 = full_63 = None
	        sub_1433: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_31, minimum_31);  maximum_31 = None
	        div_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1433, 255.0);  sub_1433 = None
	        clamp_min_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_62, 1.1920928955078125e-07);  div_62 = None
	        div_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_31, clamp_min_93);  minimum_31 = None
	        round_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_63);  div_63 = None
	        sub_1439: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_63);  round_63 = None
	        clamp_min_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1439, -128);  sub_1439 = None
	        clamp_max_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_94, 127);  clamp_min_94 = None
	        _assert_tensor_metadata_281 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_93, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_281 = None
	        _assert_tensor_metadata_282 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_62, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_282 = None
	        convert_element_type_186: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_62, torch.int8);  clamp_max_62 = None
	        view_488: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_93, [sym_size_int, 1500, 1])
	        view_489: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_186, [sym_size_int, 1500, 1])
	        reciprocal_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_488);  view_488 = None
	        mul_3065: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_31, 1.0);  reciprocal_31 = None
	        mul_3068: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_4622, mul_3065);  mul_3065 = None
	        round_64: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3068);  mul_3068 = None
	        add_4861: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_64, view_489);  round_64 = view_489 = None
	        clamp_min_95: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4861, -128);  add_4861 = None
	        clamp_max_63: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_95, 127);  clamp_min_95 = None
	        _assert_tensor_metadata_283 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_63, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_283 = None
	        convert_element_type_187: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_63, torch.int8);  clamp_max_63 = None
	        view_492: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_93, [sym_size_int, 1500, 1]);  clamp_min_93 = None
	        view_493: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_186, [sym_size_int, 1500, 1]);  convert_element_type_186 = None
	        _assert_tensor_metadata_284 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_187, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_284 = None
	        convert_element_type_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_187, torch.float32);  convert_element_type_187 = None
	        _assert_tensor_metadata_285 = torch.ops.aten._assert_tensor_metadata.default(view_493, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_285 = None
	        convert_element_type_189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_493, torch.float32);  view_493 = None
	        sub_1459: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_188, convert_element_type_189);  convert_element_type_188 = convert_element_type_189 = None
	        mul_3090: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1459, view_492);  sub_1459 = view_492 = None
	        _assert_tensor_metadata_286 = torch.ops.aten._assert_tensor_metadata.default(mul_3090, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_286 = None
	        view_495: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_496: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_497: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_287 = torch.ops.aten._assert_tensor_metadata.default(view_495, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_287 = None
	        convert_element_type_190: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_495, torch.float32);  view_495 = None
	        _assert_tensor_metadata_288 = torch.ops.aten._assert_tensor_metadata.default(view_497, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_288 = None
	        convert_element_type_191: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_497, torch.float32);  view_497 = None
	        sub_1463: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_190, convert_element_type_191);  convert_element_type_190 = convert_element_type_191 = None
	        mul_3095: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1463, view_496);  sub_1463 = view_496 = None
	        view_498: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3095, [1280, 1280]);  mul_3095 = None
	        _assert_tensor_metadata_289 = torch.ops.aten._assert_tensor_metadata.default(view_498, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_289 = None
	        permute_53: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_498, [1, 0]);  view_498 = None
	        mul_3098: "Sym(1500*s6)" = sym_size_int * 1500
	        view_499: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3090, [mul_3098, 1280]);  mul_3090 = mul_3098 = None
	        mm_5: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_499, permute_53);  view_499 = permute_53 = None
	        view_500: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_5, [sym_size_int, 1500, 1280]);  mm_5 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_501: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_500, [sym_size_int, -1, 20, 64]);  view_500 = None
	        permute_54: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_501, [0, 2, 1, 3]);  view_501 = None
	        clone_43: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_54, memory_format = torch.contiguous_format);  permute_54 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_4622, [2])
	        amax_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_4622, [2])
	        full_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_32, full_64);  amin_32 = full_64 = None
	        full_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_32, full_65);  amax_32 = full_65 = None
	        sub_1477: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_32, minimum_32);  maximum_32 = None
	        div_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1477, 255.0);  sub_1477 = None
	        clamp_min_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_64, 1.1920928955078125e-07);  div_64 = None
	        div_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_32, clamp_min_96);  minimum_32 = None
	        round_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_65);  div_65 = None
	        sub_1483: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_65);  round_65 = None
	        clamp_min_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1483, -128);  sub_1483 = None
	        clamp_max_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_97, 127);  clamp_min_97 = None
	        _assert_tensor_metadata_290 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_96, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_290 = None
	        _assert_tensor_metadata_291 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_64, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_291 = None
	        convert_element_type_192: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_64, torch.int8);  clamp_max_64 = None
	        view_504: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_96, [sym_size_int, 1500, 1])
	        view_505: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_192, [sym_size_int, 1500, 1])
	        reciprocal_32: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_504);  view_504 = None
	        mul_3164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_32, 1.0);  reciprocal_32 = None
	        mul_3167: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_4622, mul_3164);  add_4622 = mul_3164 = None
	        round_66: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3167);  mul_3167 = None
	        add_5009: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_66, view_505);  round_66 = view_505 = None
	        clamp_min_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5009, -128);  add_5009 = None
	        clamp_max_65: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_98, 127);  clamp_min_98 = None
	        _assert_tensor_metadata_292 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_65, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_292 = None
	        convert_element_type_193: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_65, torch.int8);  clamp_max_65 = None
	        view_508: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_96, [sym_size_int, 1500, 1]);  clamp_min_96 = None
	        view_509: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_192, [sym_size_int, 1500, 1]);  convert_element_type_192 = None
	        _assert_tensor_metadata_293 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_193, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_293 = None
	        convert_element_type_194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_193, torch.float32);  convert_element_type_193 = None
	        _assert_tensor_metadata_294 = torch.ops.aten._assert_tensor_metadata.default(view_509, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_294 = None
	        convert_element_type_195: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_509, torch.float32);  view_509 = None
	        sub_1503: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_194, convert_element_type_195);  convert_element_type_194 = convert_element_type_195 = None
	        mul_3189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1503, view_508);  sub_1503 = view_508 = None
	        _assert_tensor_metadata_295 = torch.ops.aten._assert_tensor_metadata.default(mul_3189, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_295 = None
	        view_511: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_512: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_513: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_296 = torch.ops.aten._assert_tensor_metadata.default(view_511, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_296 = None
	        convert_element_type_196: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_511, torch.float32);  view_511 = None
	        _assert_tensor_metadata_297 = torch.ops.aten._assert_tensor_metadata.default(view_513, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_297 = None
	        convert_element_type_197: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_513, torch.float32);  view_513 = None
	        sub_1507: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_196, convert_element_type_197);  convert_element_type_196 = convert_element_type_197 = None
	        mul_3194: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1507, view_512);  sub_1507 = view_512 = None
	        view_514: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3194, [1280, 1280]);  mul_3194 = None
	        _assert_tensor_metadata_298 = torch.ops.aten._assert_tensor_metadata.default(view_514, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_298 = None
	        mul_3199: "Sym(1500*s6)" = sym_size_int * 1500
	        view_515: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3189, [mul_3199, 1280]);  mul_3189 = mul_3199 = None
	        permute_55: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_514, [1, 0]);  view_514 = None
	        addmm_26: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_5_self_attn_v_proj_bias, view_515, permute_55);  model_audio_tower_layers_5_self_attn_v_proj_bias = view_515 = permute_55 = None
	        view_516: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_26, [sym_size_int, 1500, 1280]);  addmm_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_517: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_516, [sym_size_int, -1, 20, 64]);  view_516 = None
	        permute_56: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_517, [0, 2, 1, 3]);  view_517 = None
	        clone_44: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_56, memory_format = torch.contiguous_format);  permute_56 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_5 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_42, clone_43, clone_44, None, False, scale = 1.0);  clone_42 = clone_43 = clone_44 = None
	        getitem_42: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_5[0];  _scaled_dot_product_efficient_attention_5 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_57: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_42, [0, 2, 1, 3]);  getitem_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_57, [sym_size_int, 1500, -1]);  permute_57 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_518, [2])
	        amax_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_518, [2])
	        full_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_33, full_66);  amin_33 = full_66 = None
	        full_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_33, full_67);  amax_33 = full_67 = None
	        sub_1525: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_33, minimum_33);  maximum_33 = None
	        div_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1525, 255.0);  sub_1525 = None
	        clamp_min_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_66, 1.1920928955078125e-07);  div_66 = None
	        div_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_33, clamp_min_99);  minimum_33 = None
	        round_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_67);  div_67 = None
	        sub_1531: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_67);  round_67 = None
	        clamp_min_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1531, -128);  sub_1531 = None
	        clamp_max_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_100, 127);  clamp_min_100 = None
	        _assert_tensor_metadata_299 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_99, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_299 = None
	        _assert_tensor_metadata_300 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_66, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_300 = None
	        convert_element_type_198: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_66, torch.int8);  clamp_max_66 = None
	        view_521: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_99, [sym_size_int, 1500, 1])
	        view_522: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_198, [sym_size_int, 1500, 1])
	        reciprocal_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_521);  view_521 = None
	        mul_3269: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_33, 1.0);  reciprocal_33 = None
	        mul_3272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_518, mul_3269);  view_518 = mul_3269 = None
	        round_68: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3272);  mul_3272 = None
	        add_5173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_68, view_522);  round_68 = view_522 = None
	        clamp_min_101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5173, -128);  add_5173 = None
	        clamp_max_67: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_101, 127);  clamp_min_101 = None
	        _assert_tensor_metadata_301 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_67, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_301 = None
	        convert_element_type_199: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_67, torch.int8);  clamp_max_67 = None
	        view_525: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_99, [sym_size_int, 1500, 1]);  clamp_min_99 = None
	        view_526: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_198, [sym_size_int, 1500, 1]);  convert_element_type_198 = None
	        _assert_tensor_metadata_302 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_199, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_302 = None
	        convert_element_type_200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_199, torch.float32);  convert_element_type_199 = None
	        _assert_tensor_metadata_303 = torch.ops.aten._assert_tensor_metadata.default(view_526, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_303 = None
	        convert_element_type_201: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_526, torch.float32);  view_526 = None
	        sub_1551: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_200, convert_element_type_201);  convert_element_type_200 = convert_element_type_201 = None
	        mul_3294: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1551, view_525);  sub_1551 = view_525 = None
	        _assert_tensor_metadata_304 = torch.ops.aten._assert_tensor_metadata.default(mul_3294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_304 = None
	        view_528: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_529: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_530: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_305 = torch.ops.aten._assert_tensor_metadata.default(view_528, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_305 = None
	        convert_element_type_202: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_528, torch.float32);  view_528 = None
	        _assert_tensor_metadata_306 = torch.ops.aten._assert_tensor_metadata.default(view_530, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_306 = None
	        convert_element_type_203: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_530, torch.float32);  view_530 = None
	        sub_1555: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_202, convert_element_type_203);  convert_element_type_202 = convert_element_type_203 = None
	        mul_3299: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1555, view_529);  sub_1555 = view_529 = None
	        view_531: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3299, [1280, 1280]);  mul_3299 = None
	        _assert_tensor_metadata_307 = torch.ops.aten._assert_tensor_metadata.default(view_531, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_307 = None
	        mul_3304: "Sym(1500*s6)" = sym_size_int * 1500
	        view_532: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3294, [mul_3304, 1280]);  mul_3294 = mul_3304 = None
	        permute_58: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_531, [1, 0]);  view_531 = None
	        addmm_27: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_5_self_attn_out_proj_bias, view_532, permute_58);  model_audio_tower_layers_5_self_attn_out_proj_bias = view_532 = permute_58 = None
	        view_533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_27, [sym_size_int, 1500, 1280]);  addmm_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_5236: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_4616, view_533);  add_4616 = view_533 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_46: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_5236, memory_format = torch.contiguous_format)
	        var_mean_11 = torch.ops.aten.var_mean.correction(clone_46, [2], correction = 0, keepdim = True)
	        getitem_46: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_11[0]
	        getitem_47: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_11[1];  var_mean_11 = None
	        add_5241: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_46, 1e-05);  getitem_46 = None
	        rsqrt_11: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_5241);  add_5241 = None
	        sub_1561: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_46, getitem_47);  clone_46 = getitem_47 = None
	        mul_3315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1561, rsqrt_11);  sub_1561 = rsqrt_11 = None
	        mul_3316: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3315, model_audio_tower_layers_5_final_layer_norm_weight);  mul_3315 = model_audio_tower_layers_5_final_layer_norm_weight = None
	        add_5242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_3316, model_audio_tower_layers_5_final_layer_norm_bias);  mul_3316 = model_audio_tower_layers_5_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_5242, [2])
	        amax_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_5242, [2])
	        full_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_34, full_68);  amin_34 = full_68 = None
	        full_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_34, full_69);  amax_34 = full_69 = None
	        sub_1572: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_34, minimum_34);  maximum_34 = None
	        div_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1572, 255.0);  sub_1572 = None
	        clamp_min_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_68, 1.1920928955078125e-07);  div_68 = None
	        div_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_34, clamp_min_102);  minimum_34 = None
	        round_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_69);  div_69 = None
	        sub_1578: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_69);  round_69 = None
	        clamp_min_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1578, -128);  sub_1578 = None
	        clamp_max_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_103, 127);  clamp_min_103 = None
	        _assert_tensor_metadata_308 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_102, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_308 = None
	        _assert_tensor_metadata_309 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_68, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_309 = None
	        convert_element_type_204: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_68, torch.int8);  clamp_max_68 = None
	        view_536: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_102, [sym_size_int, 1500, 1])
	        view_537: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_204, [sym_size_int, 1500, 1])
	        reciprocal_34: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_536);  view_536 = None
	        mul_3364: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_34, 1.0);  reciprocal_34 = None
	        mul_3367: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_5242, mul_3364);  add_5242 = mul_3364 = None
	        round_70: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3367);  mul_3367 = None
	        add_5329: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_70, view_537);  round_70 = view_537 = None
	        clamp_min_104: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5329, -128);  add_5329 = None
	        clamp_max_69: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_104, 127);  clamp_min_104 = None
	        _assert_tensor_metadata_310 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_69, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_310 = None
	        convert_element_type_205: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_69, torch.int8);  clamp_max_69 = None
	        view_540: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_102, [sym_size_int, 1500, 1]);  clamp_min_102 = None
	        view_541: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_204, [sym_size_int, 1500, 1]);  convert_element_type_204 = None
	        _assert_tensor_metadata_311 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_205, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_311 = None
	        convert_element_type_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_205, torch.float32);  convert_element_type_205 = None
	        _assert_tensor_metadata_312 = torch.ops.aten._assert_tensor_metadata.default(view_541, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_312 = None
	        convert_element_type_207: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_541, torch.float32);  view_541 = None
	        sub_1598: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_206, convert_element_type_207);  convert_element_type_206 = convert_element_type_207 = None
	        mul_3389: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1598, view_540);  sub_1598 = view_540 = None
	        _assert_tensor_metadata_313 = torch.ops.aten._assert_tensor_metadata.default(mul_3389, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_313 = None
	        view_543: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_5_fc1_parametrizations_weight_original0 = None
	        view_544: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_5_fc1_parametrizations_weight_original1 = None
	        view_545: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_5_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_314 = torch.ops.aten._assert_tensor_metadata.default(view_543, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_314 = None
	        convert_element_type_208: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_543, torch.float32);  view_543 = None
	        _assert_tensor_metadata_315 = torch.ops.aten._assert_tensor_metadata.default(view_545, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_315 = None
	        convert_element_type_209: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_545, torch.float32);  view_545 = None
	        sub_1602: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_208, convert_element_type_209);  convert_element_type_208 = convert_element_type_209 = None
	        mul_3394: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1602, view_544);  sub_1602 = view_544 = None
	        view_546: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3394, [5120, 1280]);  mul_3394 = None
	        _assert_tensor_metadata_316 = torch.ops.aten._assert_tensor_metadata.default(view_546, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_316 = None
	        mul_3399: "Sym(1500*s6)" = sym_size_int * 1500
	        view_547: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3389, [mul_3399, 1280]);  mul_3389 = mul_3399 = None
	        permute_59: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_546, [1, 0]);  view_546 = None
	        addmm_28: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_5_fc1_bias, view_547, permute_59);  model_audio_tower_layers_5_fc1_bias = view_547 = permute_59 = None
	        view_548: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_28, [sym_size_int, 1500, 5120]);  addmm_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_3406: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_548, 0.5)
	        mul_3407: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_548, 0.7071067811865476);  view_548 = None
	        erf_7: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_3407);  mul_3407 = None
	        add_5388: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_7, 1);  erf_7 = None
	        mul_3408: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3406, add_5388);  mul_3406 = add_5388 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_3408, [2])
	        amax_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_3408, [2])
	        full_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_35, full_70);  amin_35 = full_70 = None
	        full_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_35, full_71);  amax_35 = full_71 = None
	        sub_1615: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_35, minimum_35);  maximum_35 = None
	        div_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1615, 255.0);  sub_1615 = None
	        clamp_min_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_70, 1.1920928955078125e-07);  div_70 = None
	        div_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_35, clamp_min_105);  minimum_35 = None
	        round_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_71);  div_71 = None
	        sub_1621: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_71);  round_71 = None
	        clamp_min_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1621, -128);  sub_1621 = None
	        clamp_max_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_106, 127);  clamp_min_106 = None
	        _assert_tensor_metadata_317 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_105, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_317 = None
	        _assert_tensor_metadata_318 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_70, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_318 = None
	        convert_element_type_210: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_70, torch.int8);  clamp_max_70 = None
	        view_551: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_105, [sym_size_int, 1500, 1])
	        view_552: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_210, [sym_size_int, 1500, 1])
	        reciprocal_35: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_551);  view_551 = None
	        mul_3454: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_35, 1.0);  reciprocal_35 = None
	        mul_3457: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3408, mul_3454);  mul_3408 = mul_3454 = None
	        round_72: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_3457);  mul_3457 = None
	        add_5471: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_72, view_552);  round_72 = view_552 = None
	        clamp_min_107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5471, -128);  add_5471 = None
	        clamp_max_71: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_107, 127);  clamp_min_107 = None
	        _assert_tensor_metadata_319 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_71, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_319 = None
	        convert_element_type_211: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_71, torch.int8);  clamp_max_71 = None
	        view_555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_105, [sym_size_int, 1500, 1]);  clamp_min_105 = None
	        view_556: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_210, [sym_size_int, 1500, 1]);  convert_element_type_210 = None
	        _assert_tensor_metadata_320 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_211, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_320 = None
	        convert_element_type_212: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_211, torch.float32);  convert_element_type_211 = None
	        _assert_tensor_metadata_321 = torch.ops.aten._assert_tensor_metadata.default(view_556, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_321 = None
	        convert_element_type_213: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_556, torch.float32);  view_556 = None
	        sub_1641: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_212, convert_element_type_213);  convert_element_type_212 = convert_element_type_213 = None
	        mul_3479: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1641, view_555);  sub_1641 = view_555 = None
	        _assert_tensor_metadata_322 = torch.ops.aten._assert_tensor_metadata.default(mul_3479, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_322 = None
	        view_558: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_5_fc2_parametrizations_weight_original0 = None
	        view_559: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_5_fc2_parametrizations_weight_original1 = None
	        view_560: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_5_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_323 = torch.ops.aten._assert_tensor_metadata.default(view_558, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_323 = None
	        convert_element_type_214: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_558, torch.float32);  view_558 = None
	        _assert_tensor_metadata_324 = torch.ops.aten._assert_tensor_metadata.default(view_560, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_324 = None
	        convert_element_type_215: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_560, torch.float32);  view_560 = None
	        sub_1645: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_214, convert_element_type_215);  convert_element_type_214 = convert_element_type_215 = None
	        mul_3484: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1645, view_559);  sub_1645 = view_559 = None
	        view_561: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3484, [1280, 5120]);  mul_3484 = None
	        _assert_tensor_metadata_325 = torch.ops.aten._assert_tensor_metadata.default(view_561, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_325 = None
	        mul_3489: "Sym(1500*s6)" = sym_size_int * 1500
	        view_562: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3479, [mul_3489, 5120]);  mul_3479 = mul_3489 = None
	        permute_60: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_561, [1, 0]);  view_561 = None
	        addmm_29: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_5_fc2_bias, view_562, permute_60);  model_audio_tower_layers_5_fc2_bias = view_562 = permute_60 = None
	        view_563: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_29, [sym_size_int, 1500, 1280]);  addmm_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_5534: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_5236, view_563);  add_5236 = view_563 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_49: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_5534, memory_format = torch.contiguous_format)
	        var_mean_12 = torch.ops.aten.var_mean.correction(clone_49, [2], correction = 0, keepdim = True)
	        getitem_48: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_12[0]
	        getitem_49: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_12[1];  var_mean_12 = None
	        add_5539: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_48, 1e-05);  getitem_48 = None
	        rsqrt_12: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_5539);  add_5539 = None
	        sub_1651: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_49, getitem_49);  clone_49 = getitem_49 = None
	        mul_3500: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1651, rsqrt_12);  sub_1651 = rsqrt_12 = None
	        mul_3501: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3500, model_audio_tower_layers_6_self_attn_layer_norm_weight);  mul_3500 = model_audio_tower_layers_6_self_attn_layer_norm_weight = None
	        add_5540: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_3501, model_audio_tower_layers_6_self_attn_layer_norm_bias);  mul_3501 = model_audio_tower_layers_6_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_5540, [2])
	        amax_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_5540, [2])
	        full_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_36, full_72);  amin_36 = full_72 = None
	        full_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_36, full_73);  amax_36 = full_73 = None
	        sub_1662: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_36, minimum_36);  maximum_36 = None
	        div_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1662, 255.0);  sub_1662 = None
	        clamp_min_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_72, 1.1920928955078125e-07);  div_72 = None
	        div_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_36, clamp_min_108);  minimum_36 = None
	        round_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_73);  div_73 = None
	        sub_1668: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_73);  round_73 = None
	        clamp_min_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1668, -128);  sub_1668 = None
	        clamp_max_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_109, 127);  clamp_min_109 = None
	        _assert_tensor_metadata_326 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_108, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_326 = None
	        _assert_tensor_metadata_327 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_72, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_327 = None
	        convert_element_type_216: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_72, torch.int8);  clamp_max_72 = None
	        view_566: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_108, [sym_size_int, 1500, 1])
	        view_567: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_216, [sym_size_int, 1500, 1])
	        reciprocal_36: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_566);  view_566 = None
	        mul_3549: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_36, 1.0);  reciprocal_36 = None
	        mul_3552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_5540, mul_3549);  mul_3549 = None
	        round_74: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3552);  mul_3552 = None
	        add_5627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_74, view_567);  round_74 = view_567 = None
	        clamp_min_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5627, -128);  add_5627 = None
	        clamp_max_73: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_110, 127);  clamp_min_110 = None
	        _assert_tensor_metadata_328 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_73, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_328 = None
	        convert_element_type_217: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_73, torch.int8);  clamp_max_73 = None
	        view_570: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_108, [sym_size_int, 1500, 1]);  clamp_min_108 = None
	        view_571: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_216, [sym_size_int, 1500, 1]);  convert_element_type_216 = None
	        _assert_tensor_metadata_329 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_217, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_329 = None
	        convert_element_type_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_217, torch.float32);  convert_element_type_217 = None
	        _assert_tensor_metadata_330 = torch.ops.aten._assert_tensor_metadata.default(view_571, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_330 = None
	        convert_element_type_219: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_571, torch.float32);  view_571 = None
	        sub_1688: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_218, convert_element_type_219);  convert_element_type_218 = convert_element_type_219 = None
	        mul_3574: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1688, view_570);  sub_1688 = view_570 = None
	        _assert_tensor_metadata_331 = torch.ops.aten._assert_tensor_metadata.default(mul_3574, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_331 = None
	        view_573: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_574: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_575: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_332 = torch.ops.aten._assert_tensor_metadata.default(view_573, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_332 = None
	        convert_element_type_220: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_573, torch.float32);  view_573 = None
	        _assert_tensor_metadata_333 = torch.ops.aten._assert_tensor_metadata.default(view_575, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_333 = None
	        convert_element_type_221: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_575, torch.float32);  view_575 = None
	        sub_1692: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_220, convert_element_type_221);  convert_element_type_220 = convert_element_type_221 = None
	        mul_3579: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1692, view_574);  sub_1692 = view_574 = None
	        view_576: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3579, [1280, 1280]);  mul_3579 = None
	        _assert_tensor_metadata_334 = torch.ops.aten._assert_tensor_metadata.default(view_576, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_334 = None
	        mul_3584: "Sym(1500*s6)" = sym_size_int * 1500
	        view_577: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3574, [mul_3584, 1280]);  mul_3574 = mul_3584 = None
	        permute_61: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_576, [1, 0]);  view_576 = None
	        addmm_30: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_6_self_attn_q_proj_bias, view_577, permute_61);  model_audio_tower_layers_6_self_attn_q_proj_bias = view_577 = permute_61 = None
	        view_578: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_30, [sym_size_int, 1500, 1280]);  addmm_30 = None
	        mul_3591: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_578, 0.125);  view_578 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_579: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3591, [sym_size_int, 1500, 20, 64]);  mul_3591 = None
	        permute_62: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_579, [0, 2, 1, 3]);  view_579 = None
	        clone_50: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_62, memory_format = torch.contiguous_format);  permute_62 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_5540, [2])
	        amax_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_5540, [2])
	        full_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_37, full_74);  amin_37 = full_74 = None
	        full_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_37, full_75);  amax_37 = full_75 = None
	        sub_1707: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_37, minimum_37);  maximum_37 = None
	        div_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1707, 255.0);  sub_1707 = None
	        clamp_min_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_74, 1.1920928955078125e-07);  div_74 = None
	        div_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_37, clamp_min_111);  minimum_37 = None
	        round_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_75);  div_75 = None
	        sub_1713: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_75);  round_75 = None
	        clamp_min_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1713, -128);  sub_1713 = None
	        clamp_max_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_112, 127);  clamp_min_112 = None
	        _assert_tensor_metadata_335 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_111, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_335 = None
	        _assert_tensor_metadata_336 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_74, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_336 = None
	        convert_element_type_222: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_74, torch.int8);  clamp_max_74 = None
	        view_582: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_111, [sym_size_int, 1500, 1])
	        view_583: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_222, [sym_size_int, 1500, 1])
	        reciprocal_37: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_582);  view_582 = None
	        mul_3645: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_37, 1.0);  reciprocal_37 = None
	        mul_3648: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_5540, mul_3645);  mul_3645 = None
	        round_76: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3648);  mul_3648 = None
	        add_5779: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_76, view_583);  round_76 = view_583 = None
	        clamp_min_113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5779, -128);  add_5779 = None
	        clamp_max_75: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_113, 127);  clamp_min_113 = None
	        _assert_tensor_metadata_337 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_75, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_337 = None
	        convert_element_type_223: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_75, torch.int8);  clamp_max_75 = None
	        view_586: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_111, [sym_size_int, 1500, 1]);  clamp_min_111 = None
	        view_587: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_222, [sym_size_int, 1500, 1]);  convert_element_type_222 = None
	        _assert_tensor_metadata_338 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_223, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_338 = None
	        convert_element_type_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_223, torch.float32);  convert_element_type_223 = None
	        _assert_tensor_metadata_339 = torch.ops.aten._assert_tensor_metadata.default(view_587, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_339 = None
	        convert_element_type_225: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_587, torch.float32);  view_587 = None
	        sub_1733: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_224, convert_element_type_225);  convert_element_type_224 = convert_element_type_225 = None
	        mul_3670: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1733, view_586);  sub_1733 = view_586 = None
	        _assert_tensor_metadata_340 = torch.ops.aten._assert_tensor_metadata.default(mul_3670, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_340 = None
	        view_589: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_590: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_591: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_341 = torch.ops.aten._assert_tensor_metadata.default(view_589, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_341 = None
	        convert_element_type_226: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_589, torch.float32);  view_589 = None
	        _assert_tensor_metadata_342 = torch.ops.aten._assert_tensor_metadata.default(view_591, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_342 = None
	        convert_element_type_227: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_591, torch.float32);  view_591 = None
	        sub_1737: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_226, convert_element_type_227);  convert_element_type_226 = convert_element_type_227 = None
	        mul_3675: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1737, view_590);  sub_1737 = view_590 = None
	        view_592: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3675, [1280, 1280]);  mul_3675 = None
	        _assert_tensor_metadata_343 = torch.ops.aten._assert_tensor_metadata.default(view_592, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_343 = None
	        permute_63: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_592, [1, 0]);  view_592 = None
	        mul_3678: "Sym(1500*s6)" = sym_size_int * 1500
	        view_593: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3670, [mul_3678, 1280]);  mul_3670 = mul_3678 = None
	        mm_6: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_593, permute_63);  view_593 = permute_63 = None
	        view_594: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_6, [sym_size_int, 1500, 1280]);  mm_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_595: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_594, [sym_size_int, -1, 20, 64]);  view_594 = None
	        permute_64: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_595, [0, 2, 1, 3]);  view_595 = None
	        clone_51: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_64, memory_format = torch.contiguous_format);  permute_64 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_5540, [2])
	        amax_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_5540, [2])
	        full_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_38, full_76);  amin_38 = full_76 = None
	        full_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_38, full_77);  amax_38 = full_77 = None
	        sub_1751: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_38, minimum_38);  maximum_38 = None
	        div_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1751, 255.0);  sub_1751 = None
	        clamp_min_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_76, 1.1920928955078125e-07);  div_76 = None
	        div_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_38, clamp_min_114);  minimum_38 = None
	        round_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_77);  div_77 = None
	        sub_1757: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_77);  round_77 = None
	        clamp_min_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1757, -128);  sub_1757 = None
	        clamp_max_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_115, 127);  clamp_min_115 = None
	        _assert_tensor_metadata_344 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_114, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_344 = None
	        _assert_tensor_metadata_345 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_76, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_345 = None
	        convert_element_type_228: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_76, torch.int8);  clamp_max_76 = None
	        view_598: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_114, [sym_size_int, 1500, 1])
	        view_599: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_228, [sym_size_int, 1500, 1])
	        reciprocal_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_598);  view_598 = None
	        mul_3744: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_38, 1.0);  reciprocal_38 = None
	        mul_3747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_5540, mul_3744);  add_5540 = mul_3744 = None
	        round_78: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3747);  mul_3747 = None
	        add_5927: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_78, view_599);  round_78 = view_599 = None
	        clamp_min_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5927, -128);  add_5927 = None
	        clamp_max_77: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_116, 127);  clamp_min_116 = None
	        _assert_tensor_metadata_346 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_77, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_346 = None
	        convert_element_type_229: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_77, torch.int8);  clamp_max_77 = None
	        view_602: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_114, [sym_size_int, 1500, 1]);  clamp_min_114 = None
	        view_603: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_228, [sym_size_int, 1500, 1]);  convert_element_type_228 = None
	        _assert_tensor_metadata_347 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_229, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_347 = None
	        convert_element_type_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_229, torch.float32);  convert_element_type_229 = None
	        _assert_tensor_metadata_348 = torch.ops.aten._assert_tensor_metadata.default(view_603, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_348 = None
	        convert_element_type_231: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_603, torch.float32);  view_603 = None
	        sub_1777: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_230, convert_element_type_231);  convert_element_type_230 = convert_element_type_231 = None
	        mul_3769: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1777, view_602);  sub_1777 = view_602 = None
	        _assert_tensor_metadata_349 = torch.ops.aten._assert_tensor_metadata.default(mul_3769, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_349 = None
	        view_605: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_606: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_607: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_350 = torch.ops.aten._assert_tensor_metadata.default(view_605, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_350 = None
	        convert_element_type_232: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_605, torch.float32);  view_605 = None
	        _assert_tensor_metadata_351 = torch.ops.aten._assert_tensor_metadata.default(view_607, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_351 = None
	        convert_element_type_233: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_607, torch.float32);  view_607 = None
	        sub_1781: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_232, convert_element_type_233);  convert_element_type_232 = convert_element_type_233 = None
	        mul_3774: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1781, view_606);  sub_1781 = view_606 = None
	        view_608: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3774, [1280, 1280]);  mul_3774 = None
	        _assert_tensor_metadata_352 = torch.ops.aten._assert_tensor_metadata.default(view_608, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_352 = None
	        mul_3779: "Sym(1500*s6)" = sym_size_int * 1500
	        view_609: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3769, [mul_3779, 1280]);  mul_3769 = mul_3779 = None
	        permute_65: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_608, [1, 0]);  view_608 = None
	        addmm_31: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_6_self_attn_v_proj_bias, view_609, permute_65);  model_audio_tower_layers_6_self_attn_v_proj_bias = view_609 = permute_65 = None
	        view_610: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_31, [sym_size_int, 1500, 1280]);  addmm_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_611: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_610, [sym_size_int, -1, 20, 64]);  view_610 = None
	        permute_66: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_611, [0, 2, 1, 3]);  view_611 = None
	        clone_52: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_66, memory_format = torch.contiguous_format);  permute_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_6 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_50, clone_51, clone_52, None, False, scale = 1.0);  clone_50 = clone_51 = clone_52 = None
	        getitem_50: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_6[0];  _scaled_dot_product_efficient_attention_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_67: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_50, [0, 2, 1, 3]);  getitem_50 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_67, [sym_size_int, 1500, -1]);  permute_67 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_612, [2])
	        amax_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_612, [2])
	        full_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_39, full_78);  amin_39 = full_78 = None
	        full_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_39, full_79);  amax_39 = full_79 = None
	        sub_1799: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_39, minimum_39);  maximum_39 = None
	        div_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1799, 255.0);  sub_1799 = None
	        clamp_min_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_78, 1.1920928955078125e-07);  div_78 = None
	        div_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_39, clamp_min_117);  minimum_39 = None
	        round_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_79);  div_79 = None
	        sub_1805: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_79);  round_79 = None
	        clamp_min_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1805, -128);  sub_1805 = None
	        clamp_max_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_118, 127);  clamp_min_118 = None
	        _assert_tensor_metadata_353 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_117, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_353 = None
	        _assert_tensor_metadata_354 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_78, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_354 = None
	        convert_element_type_234: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_78, torch.int8);  clamp_max_78 = None
	        view_615: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_117, [sym_size_int, 1500, 1])
	        view_616: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_234, [sym_size_int, 1500, 1])
	        reciprocal_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_615);  view_615 = None
	        mul_3849: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_39, 1.0);  reciprocal_39 = None
	        mul_3852: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_612, mul_3849);  view_612 = mul_3849 = None
	        round_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3852);  mul_3852 = None
	        add_6091: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_80, view_616);  round_80 = view_616 = None
	        clamp_min_119: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6091, -128);  add_6091 = None
	        clamp_max_79: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_119, 127);  clamp_min_119 = None
	        _assert_tensor_metadata_355 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_79, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_355 = None
	        convert_element_type_235: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_79, torch.int8);  clamp_max_79 = None
	        view_619: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_117, [sym_size_int, 1500, 1]);  clamp_min_117 = None
	        view_620: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_234, [sym_size_int, 1500, 1]);  convert_element_type_234 = None
	        _assert_tensor_metadata_356 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_235, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_356 = None
	        convert_element_type_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_235, torch.float32);  convert_element_type_235 = None
	        _assert_tensor_metadata_357 = torch.ops.aten._assert_tensor_metadata.default(view_620, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_357 = None
	        convert_element_type_237: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_620, torch.float32);  view_620 = None
	        sub_1825: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_236, convert_element_type_237);  convert_element_type_236 = convert_element_type_237 = None
	        mul_3874: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1825, view_619);  sub_1825 = view_619 = None
	        _assert_tensor_metadata_358 = torch.ops.aten._assert_tensor_metadata.default(mul_3874, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_358 = None
	        view_622: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_623: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_624: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_359 = torch.ops.aten._assert_tensor_metadata.default(view_622, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_359 = None
	        convert_element_type_238: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_622, torch.float32);  view_622 = None
	        _assert_tensor_metadata_360 = torch.ops.aten._assert_tensor_metadata.default(view_624, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_360 = None
	        convert_element_type_239: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_624, torch.float32);  view_624 = None
	        sub_1829: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_238, convert_element_type_239);  convert_element_type_238 = convert_element_type_239 = None
	        mul_3879: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1829, view_623);  sub_1829 = view_623 = None
	        view_625: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3879, [1280, 1280]);  mul_3879 = None
	        _assert_tensor_metadata_361 = torch.ops.aten._assert_tensor_metadata.default(view_625, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_361 = None
	        mul_3884: "Sym(1500*s6)" = sym_size_int * 1500
	        view_626: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3874, [mul_3884, 1280]);  mul_3874 = mul_3884 = None
	        permute_68: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_625, [1, 0]);  view_625 = None
	        addmm_32: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_6_self_attn_out_proj_bias, view_626, permute_68);  model_audio_tower_layers_6_self_attn_out_proj_bias = view_626 = permute_68 = None
	        view_627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_32, [sym_size_int, 1500, 1280]);  addmm_32 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_6154: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_5534, view_627);  add_5534 = view_627 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_54: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_6154, memory_format = torch.contiguous_format)
	        var_mean_13 = torch.ops.aten.var_mean.correction(clone_54, [2], correction = 0, keepdim = True)
	        getitem_54: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_13[0]
	        getitem_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_13[1];  var_mean_13 = None
	        add_6159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_54, 1e-05);  getitem_54 = None
	        rsqrt_13: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_6159);  add_6159 = None
	        sub_1835: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_54, getitem_55);  clone_54 = getitem_55 = None
	        mul_3895: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1835, rsqrt_13);  sub_1835 = rsqrt_13 = None
	        mul_3896: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3895, model_audio_tower_layers_6_final_layer_norm_weight);  mul_3895 = model_audio_tower_layers_6_final_layer_norm_weight = None
	        add_6160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_3896, model_audio_tower_layers_6_final_layer_norm_bias);  mul_3896 = model_audio_tower_layers_6_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_6160, [2])
	        amax_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_6160, [2])
	        full_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_40, full_80);  amin_40 = full_80 = None
	        full_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_40, full_81);  amax_40 = full_81 = None
	        sub_1846: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_40, minimum_40);  maximum_40 = None
	        div_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1846, 255.0);  sub_1846 = None
	        clamp_min_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_80, 1.1920928955078125e-07);  div_80 = None
	        div_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_40, clamp_min_120);  minimum_40 = None
	        round_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_81);  div_81 = None
	        sub_1852: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_81);  round_81 = None
	        clamp_min_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1852, -128);  sub_1852 = None
	        clamp_max_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_121, 127);  clamp_min_121 = None
	        _assert_tensor_metadata_362 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_120, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_362 = None
	        _assert_tensor_metadata_363 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_80, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_363 = None
	        convert_element_type_240: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_80, torch.int8);  clamp_max_80 = None
	        view_630: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_120, [sym_size_int, 1500, 1])
	        view_631: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_240, [sym_size_int, 1500, 1])
	        reciprocal_40: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_630);  view_630 = None
	        mul_3944: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_40, 1.0);  reciprocal_40 = None
	        mul_3947: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_6160, mul_3944);  add_6160 = mul_3944 = None
	        round_82: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3947);  mul_3947 = None
	        add_6247: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_82, view_631);  round_82 = view_631 = None
	        clamp_min_122: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6247, -128);  add_6247 = None
	        clamp_max_81: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_122, 127);  clamp_min_122 = None
	        _assert_tensor_metadata_364 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_81, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_364 = None
	        convert_element_type_241: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_81, torch.int8);  clamp_max_81 = None
	        view_634: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_120, [sym_size_int, 1500, 1]);  clamp_min_120 = None
	        view_635: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_240, [sym_size_int, 1500, 1]);  convert_element_type_240 = None
	        _assert_tensor_metadata_365 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_241, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_365 = None
	        convert_element_type_242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_241, torch.float32);  convert_element_type_241 = None
	        _assert_tensor_metadata_366 = torch.ops.aten._assert_tensor_metadata.default(view_635, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_366 = None
	        convert_element_type_243: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_635, torch.float32);  view_635 = None
	        sub_1872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_242, convert_element_type_243);  convert_element_type_242 = convert_element_type_243 = None
	        mul_3969: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1872, view_634);  sub_1872 = view_634 = None
	        _assert_tensor_metadata_367 = torch.ops.aten._assert_tensor_metadata.default(mul_3969, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_367 = None
	        view_637: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_6_fc1_parametrizations_weight_original0 = None
	        view_638: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_6_fc1_parametrizations_weight_original1 = None
	        view_639: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_6_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_368 = torch.ops.aten._assert_tensor_metadata.default(view_637, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_368 = None
	        convert_element_type_244: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_637, torch.float32);  view_637 = None
	        _assert_tensor_metadata_369 = torch.ops.aten._assert_tensor_metadata.default(view_639, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_369 = None
	        convert_element_type_245: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_639, torch.float32);  view_639 = None
	        sub_1876: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_244, convert_element_type_245);  convert_element_type_244 = convert_element_type_245 = None
	        mul_3974: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1876, view_638);  sub_1876 = view_638 = None
	        view_640: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3974, [5120, 1280]);  mul_3974 = None
	        _assert_tensor_metadata_370 = torch.ops.aten._assert_tensor_metadata.default(view_640, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_370 = None
	        mul_3979: "Sym(1500*s6)" = sym_size_int * 1500
	        view_641: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3969, [mul_3979, 1280]);  mul_3969 = mul_3979 = None
	        permute_69: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_640, [1, 0]);  view_640 = None
	        addmm_33: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_6_fc1_bias, view_641, permute_69);  model_audio_tower_layers_6_fc1_bias = view_641 = permute_69 = None
	        view_642: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_33, [sym_size_int, 1500, 5120]);  addmm_33 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_3986: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_642, 0.5)
	        mul_3987: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_642, 0.7071067811865476);  view_642 = None
	        erf_8: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_3987);  mul_3987 = None
	        add_6306: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_8, 1);  erf_8 = None
	        mul_3988: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3986, add_6306);  mul_3986 = add_6306 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_3988, [2])
	        amax_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_3988, [2])
	        full_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_41, full_82);  amin_41 = full_82 = None
	        full_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_41, full_83);  amax_41 = full_83 = None
	        sub_1889: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_41, minimum_41);  maximum_41 = None
	        div_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1889, 255.0);  sub_1889 = None
	        clamp_min_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_82, 1.1920928955078125e-07);  div_82 = None
	        div_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_41, clamp_min_123);  minimum_41 = None
	        round_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_83);  div_83 = None
	        sub_1895: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_83);  round_83 = None
	        clamp_min_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1895, -128);  sub_1895 = None
	        clamp_max_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_124, 127);  clamp_min_124 = None
	        _assert_tensor_metadata_371 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_123, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_371 = None
	        _assert_tensor_metadata_372 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_82, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_372 = None
	        convert_element_type_246: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_82, torch.int8);  clamp_max_82 = None
	        view_645: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_123, [sym_size_int, 1500, 1])
	        view_646: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_246, [sym_size_int, 1500, 1])
	        reciprocal_41: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_645);  view_645 = None
	        mul_4034: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_41, 1.0);  reciprocal_41 = None
	        mul_4037: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3988, mul_4034);  mul_3988 = mul_4034 = None
	        round_84: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_4037);  mul_4037 = None
	        add_6389: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_84, view_646);  round_84 = view_646 = None
	        clamp_min_125: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6389, -128);  add_6389 = None
	        clamp_max_83: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_125, 127);  clamp_min_125 = None
	        _assert_tensor_metadata_373 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_83, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_373 = None
	        convert_element_type_247: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_83, torch.int8);  clamp_max_83 = None
	        view_649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_123, [sym_size_int, 1500, 1]);  clamp_min_123 = None
	        view_650: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_246, [sym_size_int, 1500, 1]);  convert_element_type_246 = None
	        _assert_tensor_metadata_374 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_247, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_374 = None
	        convert_element_type_248: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_247, torch.float32);  convert_element_type_247 = None
	        _assert_tensor_metadata_375 = torch.ops.aten._assert_tensor_metadata.default(view_650, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_375 = None
	        convert_element_type_249: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_650, torch.float32);  view_650 = None
	        sub_1915: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_248, convert_element_type_249);  convert_element_type_248 = convert_element_type_249 = None
	        mul_4059: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1915, view_649);  sub_1915 = view_649 = None
	        _assert_tensor_metadata_376 = torch.ops.aten._assert_tensor_metadata.default(mul_4059, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_376 = None
	        view_652: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_6_fc2_parametrizations_weight_original0 = None
	        view_653: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_6_fc2_parametrizations_weight_original1 = None
	        view_654: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_6_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_377 = torch.ops.aten._assert_tensor_metadata.default(view_652, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_377 = None
	        convert_element_type_250: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_652, torch.float32);  view_652 = None
	        _assert_tensor_metadata_378 = torch.ops.aten._assert_tensor_metadata.default(view_654, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_378 = None
	        convert_element_type_251: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_654, torch.float32);  view_654 = None
	        sub_1919: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_250, convert_element_type_251);  convert_element_type_250 = convert_element_type_251 = None
	        mul_4064: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1919, view_653);  sub_1919 = view_653 = None
	        view_655: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4064, [1280, 5120]);  mul_4064 = None
	        _assert_tensor_metadata_379 = torch.ops.aten._assert_tensor_metadata.default(view_655, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_379 = None
	        mul_4069: "Sym(1500*s6)" = sym_size_int * 1500
	        view_656: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4059, [mul_4069, 5120]);  mul_4059 = mul_4069 = None
	        permute_70: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_655, [1, 0]);  view_655 = None
	        addmm_34: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_6_fc2_bias, view_656, permute_70);  model_audio_tower_layers_6_fc2_bias = view_656 = permute_70 = None
	        view_657: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_34, [sym_size_int, 1500, 1280]);  addmm_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_6452: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_6154, view_657);  add_6154 = view_657 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_57: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_6452, memory_format = torch.contiguous_format)
	        var_mean_14 = torch.ops.aten.var_mean.correction(clone_57, [2], correction = 0, keepdim = True)
	        getitem_56: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_14[0]
	        getitem_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_14[1];  var_mean_14 = None
	        add_6457: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_56, 1e-05);  getitem_56 = None
	        rsqrt_14: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_6457);  add_6457 = None
	        sub_1925: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_57, getitem_57);  clone_57 = getitem_57 = None
	        mul_4080: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1925, rsqrt_14);  sub_1925 = rsqrt_14 = None
	        mul_4081: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4080, model_audio_tower_layers_7_self_attn_layer_norm_weight);  mul_4080 = model_audio_tower_layers_7_self_attn_layer_norm_weight = None
	        add_6458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_4081, model_audio_tower_layers_7_self_attn_layer_norm_bias);  mul_4081 = model_audio_tower_layers_7_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_6458, [2])
	        amax_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_6458, [2])
	        full_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_42, full_84);  amin_42 = full_84 = None
	        full_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_42, full_85);  amax_42 = full_85 = None
	        sub_1936: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_42, minimum_42);  maximum_42 = None
	        div_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1936, 255.0);  sub_1936 = None
	        clamp_min_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_84, 1.1920928955078125e-07);  div_84 = None
	        div_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_42, clamp_min_126);  minimum_42 = None
	        round_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_85);  div_85 = None
	        sub_1942: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_85);  round_85 = None
	        clamp_min_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1942, -128);  sub_1942 = None
	        clamp_max_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_127, 127);  clamp_min_127 = None
	        _assert_tensor_metadata_380 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_126, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_380 = None
	        _assert_tensor_metadata_381 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_84, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_381 = None
	        convert_element_type_252: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_84, torch.int8);  clamp_max_84 = None
	        view_660: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_126, [sym_size_int, 1500, 1])
	        view_661: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_252, [sym_size_int, 1500, 1])
	        reciprocal_42: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_660);  view_660 = None
	        mul_4129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_42, 1.0);  reciprocal_42 = None
	        mul_4132: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_6458, mul_4129);  mul_4129 = None
	        round_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4132);  mul_4132 = None
	        add_6545: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_86, view_661);  round_86 = view_661 = None
	        clamp_min_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6545, -128);  add_6545 = None
	        clamp_max_85: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_128, 127);  clamp_min_128 = None
	        _assert_tensor_metadata_382 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_85, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_382 = None
	        convert_element_type_253: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_85, torch.int8);  clamp_max_85 = None
	        view_664: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_126, [sym_size_int, 1500, 1]);  clamp_min_126 = None
	        view_665: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_252, [sym_size_int, 1500, 1]);  convert_element_type_252 = None
	        _assert_tensor_metadata_383 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_253, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_383 = None
	        convert_element_type_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_253, torch.float32);  convert_element_type_253 = None
	        _assert_tensor_metadata_384 = torch.ops.aten._assert_tensor_metadata.default(view_665, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_384 = None
	        convert_element_type_255: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_665, torch.float32);  view_665 = None
	        sub_1962: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_254, convert_element_type_255);  convert_element_type_254 = convert_element_type_255 = None
	        mul_4154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1962, view_664);  sub_1962 = view_664 = None
	        _assert_tensor_metadata_385 = torch.ops.aten._assert_tensor_metadata.default(mul_4154, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_385 = None
	        view_667: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_668: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_669: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_386 = torch.ops.aten._assert_tensor_metadata.default(view_667, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_386 = None
	        convert_element_type_256: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_667, torch.float32);  view_667 = None
	        _assert_tensor_metadata_387 = torch.ops.aten._assert_tensor_metadata.default(view_669, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_387 = None
	        convert_element_type_257: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_669, torch.float32);  view_669 = None
	        sub_1966: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_256, convert_element_type_257);  convert_element_type_256 = convert_element_type_257 = None
	        mul_4159: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1966, view_668);  sub_1966 = view_668 = None
	        view_670: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4159, [1280, 1280]);  mul_4159 = None
	        _assert_tensor_metadata_388 = torch.ops.aten._assert_tensor_metadata.default(view_670, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_388 = None
	        mul_4164: "Sym(1500*s6)" = sym_size_int * 1500
	        view_671: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4154, [mul_4164, 1280]);  mul_4154 = mul_4164 = None
	        permute_71: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_670, [1, 0]);  view_670 = None
	        addmm_35: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_7_self_attn_q_proj_bias, view_671, permute_71);  model_audio_tower_layers_7_self_attn_q_proj_bias = view_671 = permute_71 = None
	        view_672: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_35, [sym_size_int, 1500, 1280]);  addmm_35 = None
	        mul_4171: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_672, 0.125);  view_672 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_673: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4171, [sym_size_int, 1500, 20, 64]);  mul_4171 = None
	        permute_72: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_673, [0, 2, 1, 3]);  view_673 = None
	        clone_58: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_72, memory_format = torch.contiguous_format);  permute_72 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_6458, [2])
	        amax_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_6458, [2])
	        full_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_43, full_86);  amin_43 = full_86 = None
	        full_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_43, full_87);  amax_43 = full_87 = None
	        sub_1981: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_43, minimum_43);  maximum_43 = None
	        div_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1981, 255.0);  sub_1981 = None
	        clamp_min_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_86, 1.1920928955078125e-07);  div_86 = None
	        div_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_43, clamp_min_129);  minimum_43 = None
	        round_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_87);  div_87 = None
	        sub_1987: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_87);  round_87 = None
	        clamp_min_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1987, -128);  sub_1987 = None
	        clamp_max_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_130, 127);  clamp_min_130 = None
	        _assert_tensor_metadata_389 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_129, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_389 = None
	        _assert_tensor_metadata_390 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_86, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_390 = None
	        convert_element_type_258: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_86, torch.int8);  clamp_max_86 = None
	        view_676: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_129, [sym_size_int, 1500, 1])
	        view_677: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_258, [sym_size_int, 1500, 1])
	        reciprocal_43: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_676);  view_676 = None
	        mul_4225: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_43, 1.0);  reciprocal_43 = None
	        mul_4228: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_6458, mul_4225);  mul_4225 = None
	        round_88: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4228);  mul_4228 = None
	        add_6697: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_88, view_677);  round_88 = view_677 = None
	        clamp_min_131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6697, -128);  add_6697 = None
	        clamp_max_87: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_131, 127);  clamp_min_131 = None
	        _assert_tensor_metadata_391 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_87, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_391 = None
	        convert_element_type_259: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_87, torch.int8);  clamp_max_87 = None
	        view_680: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_129, [sym_size_int, 1500, 1]);  clamp_min_129 = None
	        view_681: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_258, [sym_size_int, 1500, 1]);  convert_element_type_258 = None
	        _assert_tensor_metadata_392 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_259, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_392 = None
	        convert_element_type_260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_259, torch.float32);  convert_element_type_259 = None
	        _assert_tensor_metadata_393 = torch.ops.aten._assert_tensor_metadata.default(view_681, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_393 = None
	        convert_element_type_261: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_681, torch.float32);  view_681 = None
	        sub_2007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_260, convert_element_type_261);  convert_element_type_260 = convert_element_type_261 = None
	        mul_4250: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2007, view_680);  sub_2007 = view_680 = None
	        _assert_tensor_metadata_394 = torch.ops.aten._assert_tensor_metadata.default(mul_4250, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_394 = None
	        view_683: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_684: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_685: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_395 = torch.ops.aten._assert_tensor_metadata.default(view_683, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_395 = None
	        convert_element_type_262: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_683, torch.float32);  view_683 = None
	        _assert_tensor_metadata_396 = torch.ops.aten._assert_tensor_metadata.default(view_685, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_396 = None
	        convert_element_type_263: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_685, torch.float32);  view_685 = None
	        sub_2011: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_262, convert_element_type_263);  convert_element_type_262 = convert_element_type_263 = None
	        mul_4255: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2011, view_684);  sub_2011 = view_684 = None
	        view_686: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4255, [1280, 1280]);  mul_4255 = None
	        _assert_tensor_metadata_397 = torch.ops.aten._assert_tensor_metadata.default(view_686, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_397 = None
	        permute_73: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_686, [1, 0]);  view_686 = None
	        mul_4258: "Sym(1500*s6)" = sym_size_int * 1500
	        view_687: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4250, [mul_4258, 1280]);  mul_4250 = mul_4258 = None
	        mm_7: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_687, permute_73);  view_687 = permute_73 = None
	        view_688: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_7, [sym_size_int, 1500, 1280]);  mm_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_689: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_688, [sym_size_int, -1, 20, 64]);  view_688 = None
	        permute_74: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_689, [0, 2, 1, 3]);  view_689 = None
	        clone_59: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_74, memory_format = torch.contiguous_format);  permute_74 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_6458, [2])
	        amax_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_6458, [2])
	        full_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_44, full_88);  amin_44 = full_88 = None
	        full_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_44, full_89);  amax_44 = full_89 = None
	        sub_2025: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_44, minimum_44);  maximum_44 = None
	        div_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2025, 255.0);  sub_2025 = None
	        clamp_min_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_88, 1.1920928955078125e-07);  div_88 = None
	        div_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_44, clamp_min_132);  minimum_44 = None
	        round_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_89);  div_89 = None
	        sub_2031: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_89);  round_89 = None
	        clamp_min_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2031, -128);  sub_2031 = None
	        clamp_max_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_133, 127);  clamp_min_133 = None
	        _assert_tensor_metadata_398 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_132, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_398 = None
	        _assert_tensor_metadata_399 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_88, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_399 = None
	        convert_element_type_264: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_88, torch.int8);  clamp_max_88 = None
	        view_692: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_132, [sym_size_int, 1500, 1])
	        view_693: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_264, [sym_size_int, 1500, 1])
	        reciprocal_44: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_692);  view_692 = None
	        mul_4324: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_44, 1.0);  reciprocal_44 = None
	        mul_4327: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_6458, mul_4324);  add_6458 = mul_4324 = None
	        round_90: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4327);  mul_4327 = None
	        add_6845: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_90, view_693);  round_90 = view_693 = None
	        clamp_min_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6845, -128);  add_6845 = None
	        clamp_max_89: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_134, 127);  clamp_min_134 = None
	        _assert_tensor_metadata_400 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_89, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_400 = None
	        convert_element_type_265: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_89, torch.int8);  clamp_max_89 = None
	        view_696: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_132, [sym_size_int, 1500, 1]);  clamp_min_132 = None
	        view_697: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_264, [sym_size_int, 1500, 1]);  convert_element_type_264 = None
	        _assert_tensor_metadata_401 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_265, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_401 = None
	        convert_element_type_266: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_265, torch.float32);  convert_element_type_265 = None
	        _assert_tensor_metadata_402 = torch.ops.aten._assert_tensor_metadata.default(view_697, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_402 = None
	        convert_element_type_267: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_697, torch.float32);  view_697 = None
	        sub_2051: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_266, convert_element_type_267);  convert_element_type_266 = convert_element_type_267 = None
	        mul_4349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2051, view_696);  sub_2051 = view_696 = None
	        _assert_tensor_metadata_403 = torch.ops.aten._assert_tensor_metadata.default(mul_4349, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_403 = None
	        view_699: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_700: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_701: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_404 = torch.ops.aten._assert_tensor_metadata.default(view_699, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_404 = None
	        convert_element_type_268: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_699, torch.float32);  view_699 = None
	        _assert_tensor_metadata_405 = torch.ops.aten._assert_tensor_metadata.default(view_701, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_405 = None
	        convert_element_type_269: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_701, torch.float32);  view_701 = None
	        sub_2055: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_268, convert_element_type_269);  convert_element_type_268 = convert_element_type_269 = None
	        mul_4354: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2055, view_700);  sub_2055 = view_700 = None
	        view_702: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4354, [1280, 1280]);  mul_4354 = None
	        _assert_tensor_metadata_406 = torch.ops.aten._assert_tensor_metadata.default(view_702, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_406 = None
	        mul_4359: "Sym(1500*s6)" = sym_size_int * 1500
	        view_703: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4349, [mul_4359, 1280]);  mul_4349 = mul_4359 = None
	        permute_75: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_702, [1, 0]);  view_702 = None
	        addmm_36: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_7_self_attn_v_proj_bias, view_703, permute_75);  model_audio_tower_layers_7_self_attn_v_proj_bias = view_703 = permute_75 = None
	        view_704: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_36, [sym_size_int, 1500, 1280]);  addmm_36 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_705: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_704, [sym_size_int, -1, 20, 64]);  view_704 = None
	        permute_76: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_705, [0, 2, 1, 3]);  view_705 = None
	        clone_60: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_76, memory_format = torch.contiguous_format);  permute_76 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_7 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_58, clone_59, clone_60, None, False, scale = 1.0);  clone_58 = clone_59 = clone_60 = None
	        getitem_58: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_7[0];  _scaled_dot_product_efficient_attention_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_77: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_58, [0, 2, 1, 3]);  getitem_58 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_706: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_77, [sym_size_int, 1500, -1]);  permute_77 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_706, [2])
	        amax_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_706, [2])
	        full_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_45, full_90);  amin_45 = full_90 = None
	        full_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_45, full_91);  amax_45 = full_91 = None
	        sub_2073: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_45, minimum_45);  maximum_45 = None
	        div_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2073, 255.0);  sub_2073 = None
	        clamp_min_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_90, 1.1920928955078125e-07);  div_90 = None
	        div_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_45, clamp_min_135);  minimum_45 = None
	        round_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_91);  div_91 = None
	        sub_2079: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_91);  round_91 = None
	        clamp_min_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2079, -128);  sub_2079 = None
	        clamp_max_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_136, 127);  clamp_min_136 = None
	        _assert_tensor_metadata_407 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_135, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_407 = None
	        _assert_tensor_metadata_408 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_90, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_408 = None
	        convert_element_type_270: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_90, torch.int8);  clamp_max_90 = None
	        view_709: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_135, [sym_size_int, 1500, 1])
	        view_710: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_270, [sym_size_int, 1500, 1])
	        reciprocal_45: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_709);  view_709 = None
	        mul_4429: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_45, 1.0);  reciprocal_45 = None
	        mul_4432: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_706, mul_4429);  view_706 = mul_4429 = None
	        round_92: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4432);  mul_4432 = None
	        add_7009: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_92, view_710);  round_92 = view_710 = None
	        clamp_min_137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7009, -128);  add_7009 = None
	        clamp_max_91: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_137, 127);  clamp_min_137 = None
	        _assert_tensor_metadata_409 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_91, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_409 = None
	        convert_element_type_271: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_91, torch.int8);  clamp_max_91 = None
	        view_713: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_135, [sym_size_int, 1500, 1]);  clamp_min_135 = None
	        view_714: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_270, [sym_size_int, 1500, 1]);  convert_element_type_270 = None
	        _assert_tensor_metadata_410 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_271, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_410 = None
	        convert_element_type_272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_271, torch.float32);  convert_element_type_271 = None
	        _assert_tensor_metadata_411 = torch.ops.aten._assert_tensor_metadata.default(view_714, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_411 = None
	        convert_element_type_273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_714, torch.float32);  view_714 = None
	        sub_2099: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_272, convert_element_type_273);  convert_element_type_272 = convert_element_type_273 = None
	        mul_4454: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2099, view_713);  sub_2099 = view_713 = None
	        _assert_tensor_metadata_412 = torch.ops.aten._assert_tensor_metadata.default(mul_4454, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_412 = None
	        view_716: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_717: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_718: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_413 = torch.ops.aten._assert_tensor_metadata.default(view_716, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_413 = None
	        convert_element_type_274: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_716, torch.float32);  view_716 = None
	        _assert_tensor_metadata_414 = torch.ops.aten._assert_tensor_metadata.default(view_718, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_414 = None
	        convert_element_type_275: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_718, torch.float32);  view_718 = None
	        sub_2103: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_274, convert_element_type_275);  convert_element_type_274 = convert_element_type_275 = None
	        mul_4459: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2103, view_717);  sub_2103 = view_717 = None
	        view_719: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4459, [1280, 1280]);  mul_4459 = None
	        _assert_tensor_metadata_415 = torch.ops.aten._assert_tensor_metadata.default(view_719, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_415 = None
	        mul_4464: "Sym(1500*s6)" = sym_size_int * 1500
	        view_720: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4454, [mul_4464, 1280]);  mul_4454 = mul_4464 = None
	        permute_78: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_719, [1, 0]);  view_719 = None
	        addmm_37: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_7_self_attn_out_proj_bias, view_720, permute_78);  model_audio_tower_layers_7_self_attn_out_proj_bias = view_720 = permute_78 = None
	        view_721: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_37, [sym_size_int, 1500, 1280]);  addmm_37 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_7072: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_6452, view_721);  add_6452 = view_721 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_7072, memory_format = torch.contiguous_format)
	        var_mean_15 = torch.ops.aten.var_mean.correction(clone_62, [2], correction = 0, keepdim = True)
	        getitem_62: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_15[0]
	        getitem_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_15[1];  var_mean_15 = None
	        add_7077: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_62, 1e-05);  getitem_62 = None
	        rsqrt_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_7077);  add_7077 = None
	        sub_2109: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_62, getitem_63);  clone_62 = getitem_63 = None
	        mul_4475: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2109, rsqrt_15);  sub_2109 = rsqrt_15 = None
	        mul_4476: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4475, model_audio_tower_layers_7_final_layer_norm_weight);  mul_4475 = model_audio_tower_layers_7_final_layer_norm_weight = None
	        add_7078: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_4476, model_audio_tower_layers_7_final_layer_norm_bias);  mul_4476 = model_audio_tower_layers_7_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_7078, [2])
	        amax_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_7078, [2])
	        full_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_46, full_92);  amin_46 = full_92 = None
	        full_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_46, full_93);  amax_46 = full_93 = None
	        sub_2120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_46, minimum_46);  maximum_46 = None
	        div_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2120, 255.0);  sub_2120 = None
	        clamp_min_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_92, 1.1920928955078125e-07);  div_92 = None
	        div_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_46, clamp_min_138);  minimum_46 = None
	        round_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_93);  div_93 = None
	        sub_2126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_93);  round_93 = None
	        clamp_min_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2126, -128);  sub_2126 = None
	        clamp_max_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_139, 127);  clamp_min_139 = None
	        _assert_tensor_metadata_416 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_138, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_416 = None
	        _assert_tensor_metadata_417 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_92, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_417 = None
	        convert_element_type_276: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_92, torch.int8);  clamp_max_92 = None
	        view_724: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_138, [sym_size_int, 1500, 1])
	        view_725: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_276, [sym_size_int, 1500, 1])
	        reciprocal_46: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_724);  view_724 = None
	        mul_4524: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_46, 1.0);  reciprocal_46 = None
	        mul_4527: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_7078, mul_4524);  add_7078 = mul_4524 = None
	        round_94: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4527);  mul_4527 = None
	        add_7165: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_94, view_725);  round_94 = view_725 = None
	        clamp_min_140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7165, -128);  add_7165 = None
	        clamp_max_93: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_140, 127);  clamp_min_140 = None
	        _assert_tensor_metadata_418 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_93, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_418 = None
	        convert_element_type_277: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_93, torch.int8);  clamp_max_93 = None
	        view_728: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_138, [sym_size_int, 1500, 1]);  clamp_min_138 = None
	        view_729: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_276, [sym_size_int, 1500, 1]);  convert_element_type_276 = None
	        _assert_tensor_metadata_419 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_277, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_419 = None
	        convert_element_type_278: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_277, torch.float32);  convert_element_type_277 = None
	        _assert_tensor_metadata_420 = torch.ops.aten._assert_tensor_metadata.default(view_729, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_420 = None
	        convert_element_type_279: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_729, torch.float32);  view_729 = None
	        sub_2146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_278, convert_element_type_279);  convert_element_type_278 = convert_element_type_279 = None
	        mul_4549: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2146, view_728);  sub_2146 = view_728 = None
	        _assert_tensor_metadata_421 = torch.ops.aten._assert_tensor_metadata.default(mul_4549, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_421 = None
	        view_731: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_7_fc1_parametrizations_weight_original0 = None
	        view_732: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_7_fc1_parametrizations_weight_original1 = None
	        view_733: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_7_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_422 = torch.ops.aten._assert_tensor_metadata.default(view_731, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_422 = None
	        convert_element_type_280: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_731, torch.float32);  view_731 = None
	        _assert_tensor_metadata_423 = torch.ops.aten._assert_tensor_metadata.default(view_733, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_423 = None
	        convert_element_type_281: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_733, torch.float32);  view_733 = None
	        sub_2150: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_280, convert_element_type_281);  convert_element_type_280 = convert_element_type_281 = None
	        mul_4554: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2150, view_732);  sub_2150 = view_732 = None
	        view_734: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4554, [5120, 1280]);  mul_4554 = None
	        _assert_tensor_metadata_424 = torch.ops.aten._assert_tensor_metadata.default(view_734, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_424 = None
	        mul_4559: "Sym(1500*s6)" = sym_size_int * 1500
	        view_735: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4549, [mul_4559, 1280]);  mul_4549 = mul_4559 = None
	        permute_79: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_734, [1, 0]);  view_734 = None
	        addmm_38: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_7_fc1_bias, view_735, permute_79);  model_audio_tower_layers_7_fc1_bias = view_735 = permute_79 = None
	        view_736: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_38, [sym_size_int, 1500, 5120]);  addmm_38 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_4566: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_736, 0.5)
	        mul_4567: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_736, 0.7071067811865476);  view_736 = None
	        erf_9: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_4567);  mul_4567 = None
	        add_7224: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_9, 1);  erf_9 = None
	        mul_4568: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4566, add_7224);  mul_4566 = add_7224 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_4568, [2])
	        amax_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_4568, [2])
	        full_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_47, full_94);  amin_47 = full_94 = None
	        full_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_47, full_95);  amax_47 = full_95 = None
	        sub_2163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_47, minimum_47);  maximum_47 = None
	        div_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2163, 255.0);  sub_2163 = None
	        clamp_min_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_94, 1.1920928955078125e-07);  div_94 = None
	        div_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_47, clamp_min_141);  minimum_47 = None
	        round_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_95);  div_95 = None
	        sub_2169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_95);  round_95 = None
	        clamp_min_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2169, -128);  sub_2169 = None
	        clamp_max_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_142, 127);  clamp_min_142 = None
	        _assert_tensor_metadata_425 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_141, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_425 = None
	        _assert_tensor_metadata_426 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_94, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_426 = None
	        convert_element_type_282: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_94, torch.int8);  clamp_max_94 = None
	        view_739: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_141, [sym_size_int, 1500, 1])
	        view_740: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_282, [sym_size_int, 1500, 1])
	        reciprocal_47: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_739);  view_739 = None
	        mul_4614: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_47, 1.0);  reciprocal_47 = None
	        mul_4617: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4568, mul_4614);  mul_4568 = mul_4614 = None
	        round_96: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_4617);  mul_4617 = None
	        add_7307: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_96, view_740);  round_96 = view_740 = None
	        clamp_min_143: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7307, -128);  add_7307 = None
	        clamp_max_95: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_143, 127);  clamp_min_143 = None
	        _assert_tensor_metadata_427 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_95, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_427 = None
	        convert_element_type_283: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_95, torch.int8);  clamp_max_95 = None
	        view_743: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_141, [sym_size_int, 1500, 1]);  clamp_min_141 = None
	        view_744: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_282, [sym_size_int, 1500, 1]);  convert_element_type_282 = None
	        _assert_tensor_metadata_428 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_283, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_428 = None
	        convert_element_type_284: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_283, torch.float32);  convert_element_type_283 = None
	        _assert_tensor_metadata_429 = torch.ops.aten._assert_tensor_metadata.default(view_744, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_429 = None
	        convert_element_type_285: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_744, torch.float32);  view_744 = None
	        sub_2189: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_284, convert_element_type_285);  convert_element_type_284 = convert_element_type_285 = None
	        mul_4639: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2189, view_743);  sub_2189 = view_743 = None
	        _assert_tensor_metadata_430 = torch.ops.aten._assert_tensor_metadata.default(mul_4639, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_430 = None
	        view_746: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_7_fc2_parametrizations_weight_original0 = None
	        view_747: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_7_fc2_parametrizations_weight_original1 = None
	        view_748: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_7_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_431 = torch.ops.aten._assert_tensor_metadata.default(view_746, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_431 = None
	        convert_element_type_286: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_746, torch.float32);  view_746 = None
	        _assert_tensor_metadata_432 = torch.ops.aten._assert_tensor_metadata.default(view_748, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_432 = None
	        convert_element_type_287: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_748, torch.float32);  view_748 = None
	        sub_2193: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_286, convert_element_type_287);  convert_element_type_286 = convert_element_type_287 = None
	        mul_4644: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2193, view_747);  sub_2193 = view_747 = None
	        view_749: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4644, [1280, 5120]);  mul_4644 = None
	        _assert_tensor_metadata_433 = torch.ops.aten._assert_tensor_metadata.default(view_749, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_433 = None
	        mul_4649: "Sym(1500*s6)" = sym_size_int * 1500
	        view_750: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4639, [mul_4649, 5120]);  mul_4639 = mul_4649 = None
	        permute_80: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_749, [1, 0]);  view_749 = None
	        addmm_39: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_7_fc2_bias, view_750, permute_80);  model_audio_tower_layers_7_fc2_bias = view_750 = permute_80 = None
	        view_751: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_39, [sym_size_int, 1500, 1280]);  addmm_39 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_7370: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_7072, view_751);  add_7072 = view_751 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_65: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_7370, memory_format = torch.contiguous_format)
	        var_mean_16 = torch.ops.aten.var_mean.correction(clone_65, [2], correction = 0, keepdim = True)
	        getitem_64: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_16[0]
	        getitem_65: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_16[1];  var_mean_16 = None
	        add_7375: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_64, 1e-05);  getitem_64 = None
	        rsqrt_16: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_7375);  add_7375 = None
	        sub_2199: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_65, getitem_65);  clone_65 = getitem_65 = None
	        mul_4660: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2199, rsqrt_16);  sub_2199 = rsqrt_16 = None
	        mul_4661: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4660, model_audio_tower_layers_8_self_attn_layer_norm_weight);  mul_4660 = model_audio_tower_layers_8_self_attn_layer_norm_weight = None
	        add_7376: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_4661, model_audio_tower_layers_8_self_attn_layer_norm_bias);  mul_4661 = model_audio_tower_layers_8_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_7376, [2])
	        amax_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_7376, [2])
	        full_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_48, full_96);  amin_48 = full_96 = None
	        full_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_48, full_97);  amax_48 = full_97 = None
	        sub_2210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_48, minimum_48);  maximum_48 = None
	        div_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2210, 255.0);  sub_2210 = None
	        clamp_min_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_96, 1.1920928955078125e-07);  div_96 = None
	        div_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_48, clamp_min_144);  minimum_48 = None
	        round_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_97);  div_97 = None
	        sub_2216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_97);  round_97 = None
	        clamp_min_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2216, -128);  sub_2216 = None
	        clamp_max_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_145, 127);  clamp_min_145 = None
	        _assert_tensor_metadata_434 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_144, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_434 = None
	        _assert_tensor_metadata_435 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_96, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_435 = None
	        convert_element_type_288: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_96, torch.int8);  clamp_max_96 = None
	        view_754: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_144, [sym_size_int, 1500, 1])
	        view_755: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_288, [sym_size_int, 1500, 1])
	        reciprocal_48: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_754);  view_754 = None
	        mul_4709: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_48, 1.0);  reciprocal_48 = None
	        mul_4712: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_7376, mul_4709);  mul_4709 = None
	        round_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4712);  mul_4712 = None
	        add_7463: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_98, view_755);  round_98 = view_755 = None
	        clamp_min_146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7463, -128);  add_7463 = None
	        clamp_max_97: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_146, 127);  clamp_min_146 = None
	        _assert_tensor_metadata_436 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_97, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_436 = None
	        convert_element_type_289: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_97, torch.int8);  clamp_max_97 = None
	        view_758: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_144, [sym_size_int, 1500, 1]);  clamp_min_144 = None
	        view_759: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_288, [sym_size_int, 1500, 1]);  convert_element_type_288 = None
	        _assert_tensor_metadata_437 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_289, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_437 = None
	        convert_element_type_290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_289, torch.float32);  convert_element_type_289 = None
	        _assert_tensor_metadata_438 = torch.ops.aten._assert_tensor_metadata.default(view_759, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_438 = None
	        convert_element_type_291: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_759, torch.float32);  view_759 = None
	        sub_2236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_290, convert_element_type_291);  convert_element_type_290 = convert_element_type_291 = None
	        mul_4734: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2236, view_758);  sub_2236 = view_758 = None
	        _assert_tensor_metadata_439 = torch.ops.aten._assert_tensor_metadata.default(mul_4734, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_439 = None
	        view_761: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_762: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_763: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_440 = torch.ops.aten._assert_tensor_metadata.default(view_761, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_440 = None
	        convert_element_type_292: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_761, torch.float32);  view_761 = None
	        _assert_tensor_metadata_441 = torch.ops.aten._assert_tensor_metadata.default(view_763, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_441 = None
	        convert_element_type_293: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_763, torch.float32);  view_763 = None
	        sub_2240: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_292, convert_element_type_293);  convert_element_type_292 = convert_element_type_293 = None
	        mul_4739: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2240, view_762);  sub_2240 = view_762 = None
	        view_764: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4739, [1280, 1280]);  mul_4739 = None
	        _assert_tensor_metadata_442 = torch.ops.aten._assert_tensor_metadata.default(view_764, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_442 = None
	        mul_4744: "Sym(1500*s6)" = sym_size_int * 1500
	        view_765: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4734, [mul_4744, 1280]);  mul_4734 = mul_4744 = None
	        permute_81: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_764, [1, 0]);  view_764 = None
	        addmm_40: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_8_self_attn_q_proj_bias, view_765, permute_81);  model_audio_tower_layers_8_self_attn_q_proj_bias = view_765 = permute_81 = None
	        view_766: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_40, [sym_size_int, 1500, 1280]);  addmm_40 = None
	        mul_4751: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_766, 0.125);  view_766 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_767: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4751, [sym_size_int, 1500, 20, 64]);  mul_4751 = None
	        permute_82: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_767, [0, 2, 1, 3]);  view_767 = None
	        clone_66: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_82, memory_format = torch.contiguous_format);  permute_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_7376, [2])
	        amax_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_7376, [2])
	        full_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_49, full_98);  amin_49 = full_98 = None
	        full_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_49, full_99);  amax_49 = full_99 = None
	        sub_2255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_49, minimum_49);  maximum_49 = None
	        div_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2255, 255.0);  sub_2255 = None
	        clamp_min_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_98, 1.1920928955078125e-07);  div_98 = None
	        div_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_49, clamp_min_147);  minimum_49 = None
	        round_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_99);  div_99 = None
	        sub_2261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_99);  round_99 = None
	        clamp_min_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2261, -128);  sub_2261 = None
	        clamp_max_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_148, 127);  clamp_min_148 = None
	        _assert_tensor_metadata_443 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_147, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_443 = None
	        _assert_tensor_metadata_444 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_98, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_444 = None
	        convert_element_type_294: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_98, torch.int8);  clamp_max_98 = None
	        view_770: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_147, [sym_size_int, 1500, 1])
	        view_771: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_294, [sym_size_int, 1500, 1])
	        reciprocal_49: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_770);  view_770 = None
	        mul_4805: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_49, 1.0);  reciprocal_49 = None
	        mul_4808: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_7376, mul_4805);  mul_4805 = None
	        round_100: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4808);  mul_4808 = None
	        add_7615: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_100, view_771);  round_100 = view_771 = None
	        clamp_min_149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7615, -128);  add_7615 = None
	        clamp_max_99: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_149, 127);  clamp_min_149 = None
	        _assert_tensor_metadata_445 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_99, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_445 = None
	        convert_element_type_295: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_99, torch.int8);  clamp_max_99 = None
	        view_774: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_147, [sym_size_int, 1500, 1]);  clamp_min_147 = None
	        view_775: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_294, [sym_size_int, 1500, 1]);  convert_element_type_294 = None
	        _assert_tensor_metadata_446 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_295, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_446 = None
	        convert_element_type_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_295, torch.float32);  convert_element_type_295 = None
	        _assert_tensor_metadata_447 = torch.ops.aten._assert_tensor_metadata.default(view_775, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_447 = None
	        convert_element_type_297: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_775, torch.float32);  view_775 = None
	        sub_2281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_296, convert_element_type_297);  convert_element_type_296 = convert_element_type_297 = None
	        mul_4830: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2281, view_774);  sub_2281 = view_774 = None
	        _assert_tensor_metadata_448 = torch.ops.aten._assert_tensor_metadata.default(mul_4830, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_448 = None
	        view_777: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_778: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_779: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_449 = torch.ops.aten._assert_tensor_metadata.default(view_777, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_449 = None
	        convert_element_type_298: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_777, torch.float32);  view_777 = None
	        _assert_tensor_metadata_450 = torch.ops.aten._assert_tensor_metadata.default(view_779, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_450 = None
	        convert_element_type_299: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_779, torch.float32);  view_779 = None
	        sub_2285: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_298, convert_element_type_299);  convert_element_type_298 = convert_element_type_299 = None
	        mul_4835: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2285, view_778);  sub_2285 = view_778 = None
	        view_780: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4835, [1280, 1280]);  mul_4835 = None
	        _assert_tensor_metadata_451 = torch.ops.aten._assert_tensor_metadata.default(view_780, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_451 = None
	        permute_83: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_780, [1, 0]);  view_780 = None
	        mul_4838: "Sym(1500*s6)" = sym_size_int * 1500
	        view_781: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4830, [mul_4838, 1280]);  mul_4830 = mul_4838 = None
	        mm_8: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_781, permute_83);  view_781 = permute_83 = None
	        view_782: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_8, [sym_size_int, 1500, 1280]);  mm_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_783: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_782, [sym_size_int, -1, 20, 64]);  view_782 = None
	        permute_84: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_783, [0, 2, 1, 3]);  view_783 = None
	        clone_67: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_84, memory_format = torch.contiguous_format);  permute_84 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_7376, [2])
	        amax_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_7376, [2])
	        full_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_50, full_100);  amin_50 = full_100 = None
	        full_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_50, full_101);  amax_50 = full_101 = None
	        sub_2299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_50, minimum_50);  maximum_50 = None
	        div_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2299, 255.0);  sub_2299 = None
	        clamp_min_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_100, 1.1920928955078125e-07);  div_100 = None
	        div_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_50, clamp_min_150);  minimum_50 = None
	        round_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_101);  div_101 = None
	        sub_2305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_101);  round_101 = None
	        clamp_min_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2305, -128);  sub_2305 = None
	        clamp_max_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_151, 127);  clamp_min_151 = None
	        _assert_tensor_metadata_452 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_150, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_452 = None
	        _assert_tensor_metadata_453 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_100, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_453 = None
	        convert_element_type_300: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_100, torch.int8);  clamp_max_100 = None
	        view_786: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_150, [sym_size_int, 1500, 1])
	        view_787: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_300, [sym_size_int, 1500, 1])
	        reciprocal_50: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_786);  view_786 = None
	        mul_4904: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_50, 1.0);  reciprocal_50 = None
	        mul_4907: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_7376, mul_4904);  add_7376 = mul_4904 = None
	        round_102: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4907);  mul_4907 = None
	        add_7763: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_102, view_787);  round_102 = view_787 = None
	        clamp_min_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7763, -128);  add_7763 = None
	        clamp_max_101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_152, 127);  clamp_min_152 = None
	        _assert_tensor_metadata_454 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_101, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_454 = None
	        convert_element_type_301: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_101, torch.int8);  clamp_max_101 = None
	        view_790: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_150, [sym_size_int, 1500, 1]);  clamp_min_150 = None
	        view_791: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_300, [sym_size_int, 1500, 1]);  convert_element_type_300 = None
	        _assert_tensor_metadata_455 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_301, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_455 = None
	        convert_element_type_302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_301, torch.float32);  convert_element_type_301 = None
	        _assert_tensor_metadata_456 = torch.ops.aten._assert_tensor_metadata.default(view_791, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_456 = None
	        convert_element_type_303: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_791, torch.float32);  view_791 = None
	        sub_2325: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_302, convert_element_type_303);  convert_element_type_302 = convert_element_type_303 = None
	        mul_4929: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2325, view_790);  sub_2325 = view_790 = None
	        _assert_tensor_metadata_457 = torch.ops.aten._assert_tensor_metadata.default(mul_4929, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_457 = None
	        view_793: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_794: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_795: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_458 = torch.ops.aten._assert_tensor_metadata.default(view_793, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_458 = None
	        convert_element_type_304: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_793, torch.float32);  view_793 = None
	        _assert_tensor_metadata_459 = torch.ops.aten._assert_tensor_metadata.default(view_795, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_459 = None
	        convert_element_type_305: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_795, torch.float32);  view_795 = None
	        sub_2329: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_304, convert_element_type_305);  convert_element_type_304 = convert_element_type_305 = None
	        mul_4934: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2329, view_794);  sub_2329 = view_794 = None
	        view_796: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4934, [1280, 1280]);  mul_4934 = None
	        _assert_tensor_metadata_460 = torch.ops.aten._assert_tensor_metadata.default(view_796, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_460 = None
	        mul_4939: "Sym(1500*s6)" = sym_size_int * 1500
	        view_797: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4929, [mul_4939, 1280]);  mul_4929 = mul_4939 = None
	        permute_85: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_796, [1, 0]);  view_796 = None
	        addmm_41: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_8_self_attn_v_proj_bias, view_797, permute_85);  model_audio_tower_layers_8_self_attn_v_proj_bias = view_797 = permute_85 = None
	        view_798: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_41, [sym_size_int, 1500, 1280]);  addmm_41 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_799: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_798, [sym_size_int, -1, 20, 64]);  view_798 = None
	        permute_86: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_799, [0, 2, 1, 3]);  view_799 = None
	        clone_68: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_86, memory_format = torch.contiguous_format);  permute_86 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_8 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_66, clone_67, clone_68, None, False, scale = 1.0);  clone_66 = clone_67 = clone_68 = None
	        getitem_66: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_8[0];  _scaled_dot_product_efficient_attention_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_87: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_66, [0, 2, 1, 3]);  getitem_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_800: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_87, [sym_size_int, 1500, -1]);  permute_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_800, [2])
	        amax_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_800, [2])
	        full_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_51, full_102);  amin_51 = full_102 = None
	        full_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_51, full_103);  amax_51 = full_103 = None
	        sub_2347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_51, minimum_51);  maximum_51 = None
	        div_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2347, 255.0);  sub_2347 = None
	        clamp_min_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_102, 1.1920928955078125e-07);  div_102 = None
	        div_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_51, clamp_min_153);  minimum_51 = None
	        round_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_103);  div_103 = None
	        sub_2353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_103);  round_103 = None
	        clamp_min_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2353, -128);  sub_2353 = None
	        clamp_max_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_154, 127);  clamp_min_154 = None
	        _assert_tensor_metadata_461 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_153, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_461 = None
	        _assert_tensor_metadata_462 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_102, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_462 = None
	        convert_element_type_306: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_102, torch.int8);  clamp_max_102 = None
	        view_803: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_153, [sym_size_int, 1500, 1])
	        view_804: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_306, [sym_size_int, 1500, 1])
	        reciprocal_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_803);  view_803 = None
	        mul_5009: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_51, 1.0);  reciprocal_51 = None
	        mul_5012: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_800, mul_5009);  view_800 = mul_5009 = None
	        round_104: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5012);  mul_5012 = None
	        add_7927: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_104, view_804);  round_104 = view_804 = None
	        clamp_min_155: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7927, -128);  add_7927 = None
	        clamp_max_103: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_155, 127);  clamp_min_155 = None
	        _assert_tensor_metadata_463 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_103, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_463 = None
	        convert_element_type_307: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_103, torch.int8);  clamp_max_103 = None
	        view_807: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_153, [sym_size_int, 1500, 1]);  clamp_min_153 = None
	        view_808: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_306, [sym_size_int, 1500, 1]);  convert_element_type_306 = None
	        _assert_tensor_metadata_464 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_307, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_464 = None
	        convert_element_type_308: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_307, torch.float32);  convert_element_type_307 = None
	        _assert_tensor_metadata_465 = torch.ops.aten._assert_tensor_metadata.default(view_808, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_465 = None
	        convert_element_type_309: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_808, torch.float32);  view_808 = None
	        sub_2373: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_308, convert_element_type_309);  convert_element_type_308 = convert_element_type_309 = None
	        mul_5034: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2373, view_807);  sub_2373 = view_807 = None
	        _assert_tensor_metadata_466 = torch.ops.aten._assert_tensor_metadata.default(mul_5034, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_466 = None
	        view_810: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_811: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_812: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_467 = torch.ops.aten._assert_tensor_metadata.default(view_810, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_467 = None
	        convert_element_type_310: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_810, torch.float32);  view_810 = None
	        _assert_tensor_metadata_468 = torch.ops.aten._assert_tensor_metadata.default(view_812, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_468 = None
	        convert_element_type_311: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_812, torch.float32);  view_812 = None
	        sub_2377: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_310, convert_element_type_311);  convert_element_type_310 = convert_element_type_311 = None
	        mul_5039: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2377, view_811);  sub_2377 = view_811 = None
	        view_813: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5039, [1280, 1280]);  mul_5039 = None
	        _assert_tensor_metadata_469 = torch.ops.aten._assert_tensor_metadata.default(view_813, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_469 = None
	        mul_5044: "Sym(1500*s6)" = sym_size_int * 1500
	        view_814: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5034, [mul_5044, 1280]);  mul_5034 = mul_5044 = None
	        permute_88: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_813, [1, 0]);  view_813 = None
	        addmm_42: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_8_self_attn_out_proj_bias, view_814, permute_88);  model_audio_tower_layers_8_self_attn_out_proj_bias = view_814 = permute_88 = None
	        view_815: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_42, [sym_size_int, 1500, 1280]);  addmm_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_7990: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_7370, view_815);  add_7370 = view_815 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_70: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_7990, memory_format = torch.contiguous_format)
	        var_mean_17 = torch.ops.aten.var_mean.correction(clone_70, [2], correction = 0, keepdim = True)
	        getitem_70: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_17[0]
	        getitem_71: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_17[1];  var_mean_17 = None
	        add_7995: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_70, 1e-05);  getitem_70 = None
	        rsqrt_17: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_7995);  add_7995 = None
	        sub_2383: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_70, getitem_71);  clone_70 = getitem_71 = None
	        mul_5055: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2383, rsqrt_17);  sub_2383 = rsqrt_17 = None
	        mul_5056: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5055, model_audio_tower_layers_8_final_layer_norm_weight);  mul_5055 = model_audio_tower_layers_8_final_layer_norm_weight = None
	        add_7996: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5056, model_audio_tower_layers_8_final_layer_norm_bias);  mul_5056 = model_audio_tower_layers_8_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_7996, [2])
	        amax_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_7996, [2])
	        full_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_52, full_104);  amin_52 = full_104 = None
	        full_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_52, full_105);  amax_52 = full_105 = None
	        sub_2394: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_52, minimum_52);  maximum_52 = None
	        div_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2394, 255.0);  sub_2394 = None
	        clamp_min_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_104, 1.1920928955078125e-07);  div_104 = None
	        div_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_52, clamp_min_156);  minimum_52 = None
	        round_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_105);  div_105 = None
	        sub_2400: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_105);  round_105 = None
	        clamp_min_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2400, -128);  sub_2400 = None
	        clamp_max_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_157, 127);  clamp_min_157 = None
	        _assert_tensor_metadata_470 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_156, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_470 = None
	        _assert_tensor_metadata_471 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_104, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_471 = None
	        convert_element_type_312: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_104, torch.int8);  clamp_max_104 = None
	        view_818: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_156, [sym_size_int, 1500, 1])
	        view_819: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_312, [sym_size_int, 1500, 1])
	        reciprocal_52: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_818);  view_818 = None
	        mul_5104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_52, 1.0);  reciprocal_52 = None
	        mul_5107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_7996, mul_5104);  add_7996 = mul_5104 = None
	        round_106: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5107);  mul_5107 = None
	        add_8083: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_106, view_819);  round_106 = view_819 = None
	        clamp_min_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8083, -128);  add_8083 = None
	        clamp_max_105: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_158, 127);  clamp_min_158 = None
	        _assert_tensor_metadata_472 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_105, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_472 = None
	        convert_element_type_313: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_105, torch.int8);  clamp_max_105 = None
	        view_822: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_156, [sym_size_int, 1500, 1]);  clamp_min_156 = None
	        view_823: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_312, [sym_size_int, 1500, 1]);  convert_element_type_312 = None
	        _assert_tensor_metadata_473 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_313, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_473 = None
	        convert_element_type_314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_313, torch.float32);  convert_element_type_313 = None
	        _assert_tensor_metadata_474 = torch.ops.aten._assert_tensor_metadata.default(view_823, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_474 = None
	        convert_element_type_315: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_823, torch.float32);  view_823 = None
	        sub_2420: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_314, convert_element_type_315);  convert_element_type_314 = convert_element_type_315 = None
	        mul_5129: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2420, view_822);  sub_2420 = view_822 = None
	        _assert_tensor_metadata_475 = torch.ops.aten._assert_tensor_metadata.default(mul_5129, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_475 = None
	        view_825: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_8_fc1_parametrizations_weight_original0 = None
	        view_826: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_8_fc1_parametrizations_weight_original1 = None
	        view_827: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_8_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_476 = torch.ops.aten._assert_tensor_metadata.default(view_825, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_476 = None
	        convert_element_type_316: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_825, torch.float32);  view_825 = None
	        _assert_tensor_metadata_477 = torch.ops.aten._assert_tensor_metadata.default(view_827, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_477 = None
	        convert_element_type_317: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_827, torch.float32);  view_827 = None
	        sub_2424: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_316, convert_element_type_317);  convert_element_type_316 = convert_element_type_317 = None
	        mul_5134: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2424, view_826);  sub_2424 = view_826 = None
	        view_828: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5134, [5120, 1280]);  mul_5134 = None
	        _assert_tensor_metadata_478 = torch.ops.aten._assert_tensor_metadata.default(view_828, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_478 = None
	        mul_5139: "Sym(1500*s6)" = sym_size_int * 1500
	        view_829: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5129, [mul_5139, 1280]);  mul_5129 = mul_5139 = None
	        permute_89: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_828, [1, 0]);  view_828 = None
	        addmm_43: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_8_fc1_bias, view_829, permute_89);  model_audio_tower_layers_8_fc1_bias = view_829 = permute_89 = None
	        view_830: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_43, [sym_size_int, 1500, 5120]);  addmm_43 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_5146: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_830, 0.5)
	        mul_5147: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_830, 0.7071067811865476);  view_830 = None
	        erf_10: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_5147);  mul_5147 = None
	        add_8142: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_10, 1);  erf_10 = None
	        mul_5148: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5146, add_8142);  mul_5146 = add_8142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_5148, [2])
	        amax_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_5148, [2])
	        full_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_53, full_106);  amin_53 = full_106 = None
	        full_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_53, full_107);  amax_53 = full_107 = None
	        sub_2437: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_53, minimum_53);  maximum_53 = None
	        div_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2437, 255.0);  sub_2437 = None
	        clamp_min_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_106, 1.1920928955078125e-07);  div_106 = None
	        div_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_53, clamp_min_159);  minimum_53 = None
	        round_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_107);  div_107 = None
	        sub_2443: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_107);  round_107 = None
	        clamp_min_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2443, -128);  sub_2443 = None
	        clamp_max_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_160, 127);  clamp_min_160 = None
	        _assert_tensor_metadata_479 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_159, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_479 = None
	        _assert_tensor_metadata_480 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_106, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_480 = None
	        convert_element_type_318: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_106, torch.int8);  clamp_max_106 = None
	        view_833: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_159, [sym_size_int, 1500, 1])
	        view_834: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_318, [sym_size_int, 1500, 1])
	        reciprocal_53: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_833);  view_833 = None
	        mul_5194: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_53, 1.0);  reciprocal_53 = None
	        mul_5197: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5148, mul_5194);  mul_5148 = mul_5194 = None
	        round_108: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_5197);  mul_5197 = None
	        add_8225: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_108, view_834);  round_108 = view_834 = None
	        clamp_min_161: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8225, -128);  add_8225 = None
	        clamp_max_107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_161, 127);  clamp_min_161 = None
	        _assert_tensor_metadata_481 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_107, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_481 = None
	        convert_element_type_319: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_107, torch.int8);  clamp_max_107 = None
	        view_837: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_159, [sym_size_int, 1500, 1]);  clamp_min_159 = None
	        view_838: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_318, [sym_size_int, 1500, 1]);  convert_element_type_318 = None
	        _assert_tensor_metadata_482 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_319, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_482 = None
	        convert_element_type_320: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_319, torch.float32);  convert_element_type_319 = None
	        _assert_tensor_metadata_483 = torch.ops.aten._assert_tensor_metadata.default(view_838, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_483 = None
	        convert_element_type_321: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_838, torch.float32);  view_838 = None
	        sub_2463: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_320, convert_element_type_321);  convert_element_type_320 = convert_element_type_321 = None
	        mul_5219: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2463, view_837);  sub_2463 = view_837 = None
	        _assert_tensor_metadata_484 = torch.ops.aten._assert_tensor_metadata.default(mul_5219, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_484 = None
	        view_840: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_8_fc2_parametrizations_weight_original0 = None
	        view_841: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_8_fc2_parametrizations_weight_original1 = None
	        view_842: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_8_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_485 = torch.ops.aten._assert_tensor_metadata.default(view_840, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_485 = None
	        convert_element_type_322: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_840, torch.float32);  view_840 = None
	        _assert_tensor_metadata_486 = torch.ops.aten._assert_tensor_metadata.default(view_842, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_486 = None
	        convert_element_type_323: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_842, torch.float32);  view_842 = None
	        sub_2467: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_322, convert_element_type_323);  convert_element_type_322 = convert_element_type_323 = None
	        mul_5224: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2467, view_841);  sub_2467 = view_841 = None
	        view_843: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5224, [1280, 5120]);  mul_5224 = None
	        _assert_tensor_metadata_487 = torch.ops.aten._assert_tensor_metadata.default(view_843, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_487 = None
	        mul_5229: "Sym(1500*s6)" = sym_size_int * 1500
	        view_844: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5219, [mul_5229, 5120]);  mul_5219 = mul_5229 = None
	        permute_90: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_843, [1, 0]);  view_843 = None
	        addmm_44: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_8_fc2_bias, view_844, permute_90);  model_audio_tower_layers_8_fc2_bias = view_844 = permute_90 = None
	        view_845: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_44, [sym_size_int, 1500, 1280]);  addmm_44 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_8288: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_7990, view_845);  add_7990 = view_845 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_73: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_8288, memory_format = torch.contiguous_format)
	        var_mean_18 = torch.ops.aten.var_mean.correction(clone_73, [2], correction = 0, keepdim = True)
	        getitem_72: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_18[0]
	        getitem_73: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_18[1];  var_mean_18 = None
	        add_8293: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_72, 1e-05);  getitem_72 = None
	        rsqrt_18: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_8293);  add_8293 = None
	        sub_2473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_73, getitem_73);  clone_73 = getitem_73 = None
	        mul_5240: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2473, rsqrt_18);  sub_2473 = rsqrt_18 = None
	        mul_5241: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5240, model_audio_tower_layers_9_self_attn_layer_norm_weight);  mul_5240 = model_audio_tower_layers_9_self_attn_layer_norm_weight = None
	        add_8294: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5241, model_audio_tower_layers_9_self_attn_layer_norm_bias);  mul_5241 = model_audio_tower_layers_9_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_8294, [2])
	        amax_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_8294, [2])
	        full_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_54, full_108);  amin_54 = full_108 = None
	        full_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_54, full_109);  amax_54 = full_109 = None
	        sub_2484: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_54, minimum_54);  maximum_54 = None
	        div_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2484, 255.0);  sub_2484 = None
	        clamp_min_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_108, 1.1920928955078125e-07);  div_108 = None
	        div_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_54, clamp_min_162);  minimum_54 = None
	        round_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_109);  div_109 = None
	        sub_2490: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_109);  round_109 = None
	        clamp_min_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2490, -128);  sub_2490 = None
	        clamp_max_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_163, 127);  clamp_min_163 = None
	        _assert_tensor_metadata_488 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_162, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_488 = None
	        _assert_tensor_metadata_489 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_108, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_489 = None
	        convert_element_type_324: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_108, torch.int8);  clamp_max_108 = None
	        view_848: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_162, [sym_size_int, 1500, 1])
	        view_849: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_324, [sym_size_int, 1500, 1])
	        reciprocal_54: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_848);  view_848 = None
	        mul_5289: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_54, 1.0);  reciprocal_54 = None
	        mul_5292: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_8294, mul_5289);  mul_5289 = None
	        round_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5292);  mul_5292 = None
	        add_8381: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_110, view_849);  round_110 = view_849 = None
	        clamp_min_164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8381, -128);  add_8381 = None
	        clamp_max_109: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_164, 127);  clamp_min_164 = None
	        _assert_tensor_metadata_490 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_109, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_490 = None
	        convert_element_type_325: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_109, torch.int8);  clamp_max_109 = None
	        view_852: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_162, [sym_size_int, 1500, 1]);  clamp_min_162 = None
	        view_853: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_324, [sym_size_int, 1500, 1]);  convert_element_type_324 = None
	        _assert_tensor_metadata_491 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_325, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_491 = None
	        convert_element_type_326: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_325, torch.float32);  convert_element_type_325 = None
	        _assert_tensor_metadata_492 = torch.ops.aten._assert_tensor_metadata.default(view_853, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_492 = None
	        convert_element_type_327: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_853, torch.float32);  view_853 = None
	        sub_2510: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_326, convert_element_type_327);  convert_element_type_326 = convert_element_type_327 = None
	        mul_5314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2510, view_852);  sub_2510 = view_852 = None
	        _assert_tensor_metadata_493 = torch.ops.aten._assert_tensor_metadata.default(mul_5314, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_493 = None
	        view_855: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_856: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_857: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_494 = torch.ops.aten._assert_tensor_metadata.default(view_855, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_494 = None
	        convert_element_type_328: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_855, torch.float32);  view_855 = None
	        _assert_tensor_metadata_495 = torch.ops.aten._assert_tensor_metadata.default(view_857, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_495 = None
	        convert_element_type_329: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_857, torch.float32);  view_857 = None
	        sub_2514: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_328, convert_element_type_329);  convert_element_type_328 = convert_element_type_329 = None
	        mul_5319: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2514, view_856);  sub_2514 = view_856 = None
	        view_858: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5319, [1280, 1280]);  mul_5319 = None
	        _assert_tensor_metadata_496 = torch.ops.aten._assert_tensor_metadata.default(view_858, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_496 = None
	        mul_5324: "Sym(1500*s6)" = sym_size_int * 1500
	        view_859: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5314, [mul_5324, 1280]);  mul_5314 = mul_5324 = None
	        permute_91: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_858, [1, 0]);  view_858 = None
	        addmm_45: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_9_self_attn_q_proj_bias, view_859, permute_91);  model_audio_tower_layers_9_self_attn_q_proj_bias = view_859 = permute_91 = None
	        view_860: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_45, [sym_size_int, 1500, 1280]);  addmm_45 = None
	        mul_5331: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_860, 0.125);  view_860 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_861: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5331, [sym_size_int, 1500, 20, 64]);  mul_5331 = None
	        permute_92: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_861, [0, 2, 1, 3]);  view_861 = None
	        clone_74: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_92, memory_format = torch.contiguous_format);  permute_92 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_8294, [2])
	        amax_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_8294, [2])
	        full_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_55, full_110);  amin_55 = full_110 = None
	        full_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_55, full_111);  amax_55 = full_111 = None
	        sub_2529: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_55, minimum_55);  maximum_55 = None
	        div_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2529, 255.0);  sub_2529 = None
	        clamp_min_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_110, 1.1920928955078125e-07);  div_110 = None
	        div_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_55, clamp_min_165);  minimum_55 = None
	        round_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_111);  div_111 = None
	        sub_2535: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_111);  round_111 = None
	        clamp_min_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2535, -128);  sub_2535 = None
	        clamp_max_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_166, 127);  clamp_min_166 = None
	        _assert_tensor_metadata_497 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_165, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_497 = None
	        _assert_tensor_metadata_498 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_110, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_498 = None
	        convert_element_type_330: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_110, torch.int8);  clamp_max_110 = None
	        view_864: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_165, [sym_size_int, 1500, 1])
	        view_865: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_330, [sym_size_int, 1500, 1])
	        reciprocal_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_864);  view_864 = None
	        mul_5385: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_55, 1.0);  reciprocal_55 = None
	        mul_5388: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_8294, mul_5385);  mul_5385 = None
	        round_112: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5388);  mul_5388 = None
	        add_8533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_112, view_865);  round_112 = view_865 = None
	        clamp_min_167: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8533, -128);  add_8533 = None
	        clamp_max_111: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_167, 127);  clamp_min_167 = None
	        _assert_tensor_metadata_499 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_111, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_499 = None
	        convert_element_type_331: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_111, torch.int8);  clamp_max_111 = None
	        view_868: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_165, [sym_size_int, 1500, 1]);  clamp_min_165 = None
	        view_869: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_330, [sym_size_int, 1500, 1]);  convert_element_type_330 = None
	        _assert_tensor_metadata_500 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_331, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_500 = None
	        convert_element_type_332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_331, torch.float32);  convert_element_type_331 = None
	        _assert_tensor_metadata_501 = torch.ops.aten._assert_tensor_metadata.default(view_869, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_501 = None
	        convert_element_type_333: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_869, torch.float32);  view_869 = None
	        sub_2555: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_332, convert_element_type_333);  convert_element_type_332 = convert_element_type_333 = None
	        mul_5410: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2555, view_868);  sub_2555 = view_868 = None
	        _assert_tensor_metadata_502 = torch.ops.aten._assert_tensor_metadata.default(mul_5410, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_502 = None
	        view_871: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_872: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_873: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_503 = torch.ops.aten._assert_tensor_metadata.default(view_871, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_503 = None
	        convert_element_type_334: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_871, torch.float32);  view_871 = None
	        _assert_tensor_metadata_504 = torch.ops.aten._assert_tensor_metadata.default(view_873, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_504 = None
	        convert_element_type_335: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_873, torch.float32);  view_873 = None
	        sub_2559: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_334, convert_element_type_335);  convert_element_type_334 = convert_element_type_335 = None
	        mul_5415: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2559, view_872);  sub_2559 = view_872 = None
	        view_874: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5415, [1280, 1280]);  mul_5415 = None
	        _assert_tensor_metadata_505 = torch.ops.aten._assert_tensor_metadata.default(view_874, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_505 = None
	        permute_93: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_874, [1, 0]);  view_874 = None
	        mul_5418: "Sym(1500*s6)" = sym_size_int * 1500
	        view_875: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5410, [mul_5418, 1280]);  mul_5410 = mul_5418 = None
	        mm_9: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_875, permute_93);  view_875 = permute_93 = None
	        view_876: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_9, [sym_size_int, 1500, 1280]);  mm_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_877: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_876, [sym_size_int, -1, 20, 64]);  view_876 = None
	        permute_94: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_877, [0, 2, 1, 3]);  view_877 = None
	        clone_75: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_94, memory_format = torch.contiguous_format);  permute_94 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_8294, [2])
	        amax_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_8294, [2])
	        full_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_56, full_112);  amin_56 = full_112 = None
	        full_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_56, full_113);  amax_56 = full_113 = None
	        sub_2573: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_56, minimum_56);  maximum_56 = None
	        div_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2573, 255.0);  sub_2573 = None
	        clamp_min_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_112, 1.1920928955078125e-07);  div_112 = None
	        div_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_56, clamp_min_168);  minimum_56 = None
	        round_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_113);  div_113 = None
	        sub_2579: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_113);  round_113 = None
	        clamp_min_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2579, -128);  sub_2579 = None
	        clamp_max_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_169, 127);  clamp_min_169 = None
	        _assert_tensor_metadata_506 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_168, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_506 = None
	        _assert_tensor_metadata_507 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_112, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_507 = None
	        convert_element_type_336: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_112, torch.int8);  clamp_max_112 = None
	        view_880: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_168, [sym_size_int, 1500, 1])
	        view_881: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_336, [sym_size_int, 1500, 1])
	        reciprocal_56: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_880);  view_880 = None
	        mul_5484: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_56, 1.0);  reciprocal_56 = None
	        mul_5487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_8294, mul_5484);  add_8294 = mul_5484 = None
	        round_114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5487);  mul_5487 = None
	        add_8681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_114, view_881);  round_114 = view_881 = None
	        clamp_min_170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8681, -128);  add_8681 = None
	        clamp_max_113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_170, 127);  clamp_min_170 = None
	        _assert_tensor_metadata_508 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_113, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_508 = None
	        convert_element_type_337: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_113, torch.int8);  clamp_max_113 = None
	        view_884: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_168, [sym_size_int, 1500, 1]);  clamp_min_168 = None
	        view_885: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_336, [sym_size_int, 1500, 1]);  convert_element_type_336 = None
	        _assert_tensor_metadata_509 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_337, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_509 = None
	        convert_element_type_338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_337, torch.float32);  convert_element_type_337 = None
	        _assert_tensor_metadata_510 = torch.ops.aten._assert_tensor_metadata.default(view_885, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_510 = None
	        convert_element_type_339: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_885, torch.float32);  view_885 = None
	        sub_2599: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_338, convert_element_type_339);  convert_element_type_338 = convert_element_type_339 = None
	        mul_5509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2599, view_884);  sub_2599 = view_884 = None
	        _assert_tensor_metadata_511 = torch.ops.aten._assert_tensor_metadata.default(mul_5509, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_511 = None
	        view_887: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_888: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_889: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_512 = torch.ops.aten._assert_tensor_metadata.default(view_887, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_512 = None
	        convert_element_type_340: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_887, torch.float32);  view_887 = None
	        _assert_tensor_metadata_513 = torch.ops.aten._assert_tensor_metadata.default(view_889, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_513 = None
	        convert_element_type_341: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_889, torch.float32);  view_889 = None
	        sub_2603: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_340, convert_element_type_341);  convert_element_type_340 = convert_element_type_341 = None
	        mul_5514: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2603, view_888);  sub_2603 = view_888 = None
	        view_890: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5514, [1280, 1280]);  mul_5514 = None
	        _assert_tensor_metadata_514 = torch.ops.aten._assert_tensor_metadata.default(view_890, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_514 = None
	        mul_5519: "Sym(1500*s6)" = sym_size_int * 1500
	        view_891: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5509, [mul_5519, 1280]);  mul_5509 = mul_5519 = None
	        permute_95: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_890, [1, 0]);  view_890 = None
	        addmm_46: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_9_self_attn_v_proj_bias, view_891, permute_95);  model_audio_tower_layers_9_self_attn_v_proj_bias = view_891 = permute_95 = None
	        view_892: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_46, [sym_size_int, 1500, 1280]);  addmm_46 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_893: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_892, [sym_size_int, -1, 20, 64]);  view_892 = None
	        permute_96: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_893, [0, 2, 1, 3]);  view_893 = None
	        clone_76: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_96, memory_format = torch.contiguous_format);  permute_96 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_9 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_74, clone_75, clone_76, None, False, scale = 1.0);  clone_74 = clone_75 = clone_76 = None
	        getitem_74: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_9[0];  _scaled_dot_product_efficient_attention_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_97: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_74, [0, 2, 1, 3]);  getitem_74 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_97, [sym_size_int, 1500, -1]);  permute_97 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_894, [2])
	        amax_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_894, [2])
	        full_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_57, full_114);  amin_57 = full_114 = None
	        full_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_57, full_115);  amax_57 = full_115 = None
	        sub_2621: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_57, minimum_57);  maximum_57 = None
	        div_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2621, 255.0);  sub_2621 = None
	        clamp_min_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_114, 1.1920928955078125e-07);  div_114 = None
	        div_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_57, clamp_min_171);  minimum_57 = None
	        round_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_115);  div_115 = None
	        sub_2627: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_115);  round_115 = None
	        clamp_min_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2627, -128);  sub_2627 = None
	        clamp_max_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_172, 127);  clamp_min_172 = None
	        _assert_tensor_metadata_515 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_171, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_515 = None
	        _assert_tensor_metadata_516 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_114, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_516 = None
	        convert_element_type_342: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_114, torch.int8);  clamp_max_114 = None
	        view_897: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_171, [sym_size_int, 1500, 1])
	        view_898: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_342, [sym_size_int, 1500, 1])
	        reciprocal_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_897);  view_897 = None
	        mul_5589: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_57, 1.0);  reciprocal_57 = None
	        mul_5592: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_894, mul_5589);  view_894 = mul_5589 = None
	        round_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5592);  mul_5592 = None
	        add_8845: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_116, view_898);  round_116 = view_898 = None
	        clamp_min_173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8845, -128);  add_8845 = None
	        clamp_max_115: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_173, 127);  clamp_min_173 = None
	        _assert_tensor_metadata_517 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_115, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_517 = None
	        convert_element_type_343: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_115, torch.int8);  clamp_max_115 = None
	        view_901: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_171, [sym_size_int, 1500, 1]);  clamp_min_171 = None
	        view_902: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_342, [sym_size_int, 1500, 1]);  convert_element_type_342 = None
	        _assert_tensor_metadata_518 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_343, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_518 = None
	        convert_element_type_344: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_343, torch.float32);  convert_element_type_343 = None
	        _assert_tensor_metadata_519 = torch.ops.aten._assert_tensor_metadata.default(view_902, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_519 = None
	        convert_element_type_345: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_902, torch.float32);  view_902 = None
	        sub_2647: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_344, convert_element_type_345);  convert_element_type_344 = convert_element_type_345 = None
	        mul_5614: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2647, view_901);  sub_2647 = view_901 = None
	        _assert_tensor_metadata_520 = torch.ops.aten._assert_tensor_metadata.default(mul_5614, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_520 = None
	        view_904: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_905: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_906: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_521 = torch.ops.aten._assert_tensor_metadata.default(view_904, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_521 = None
	        convert_element_type_346: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_904, torch.float32);  view_904 = None
	        _assert_tensor_metadata_522 = torch.ops.aten._assert_tensor_metadata.default(view_906, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_522 = None
	        convert_element_type_347: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_906, torch.float32);  view_906 = None
	        sub_2651: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_346, convert_element_type_347);  convert_element_type_346 = convert_element_type_347 = None
	        mul_5619: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2651, view_905);  sub_2651 = view_905 = None
	        view_907: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5619, [1280, 1280]);  mul_5619 = None
	        _assert_tensor_metadata_523 = torch.ops.aten._assert_tensor_metadata.default(view_907, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_523 = None
	        mul_5624: "Sym(1500*s6)" = sym_size_int * 1500
	        view_908: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5614, [mul_5624, 1280]);  mul_5614 = mul_5624 = None
	        permute_98: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_907, [1, 0]);  view_907 = None
	        addmm_47: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_9_self_attn_out_proj_bias, view_908, permute_98);  model_audio_tower_layers_9_self_attn_out_proj_bias = view_908 = permute_98 = None
	        view_909: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_47, [sym_size_int, 1500, 1280]);  addmm_47 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_8908: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_8288, view_909);  add_8288 = view_909 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_78: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_8908, memory_format = torch.contiguous_format)
	        var_mean_19 = torch.ops.aten.var_mean.correction(clone_78, [2], correction = 0, keepdim = True)
	        getitem_78: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_19[0]
	        getitem_79: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_19[1];  var_mean_19 = None
	        add_8913: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_78, 1e-05);  getitem_78 = None
	        rsqrt_19: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_8913);  add_8913 = None
	        sub_2657: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_78, getitem_79);  clone_78 = getitem_79 = None
	        mul_5635: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2657, rsqrt_19);  sub_2657 = rsqrt_19 = None
	        mul_5636: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5635, model_audio_tower_layers_9_final_layer_norm_weight);  mul_5635 = model_audio_tower_layers_9_final_layer_norm_weight = None
	        add_8914: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5636, model_audio_tower_layers_9_final_layer_norm_bias);  mul_5636 = model_audio_tower_layers_9_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_8914, [2])
	        amax_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_8914, [2])
	        full_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_58, full_116);  amin_58 = full_116 = None
	        full_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_58, full_117);  amax_58 = full_117 = None
	        sub_2668: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_58, minimum_58);  maximum_58 = None
	        div_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2668, 255.0);  sub_2668 = None
	        clamp_min_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_116, 1.1920928955078125e-07);  div_116 = None
	        div_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_58, clamp_min_174);  minimum_58 = None
	        round_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_117);  div_117 = None
	        sub_2674: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_117);  round_117 = None
	        clamp_min_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2674, -128);  sub_2674 = None
	        clamp_max_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_175, 127);  clamp_min_175 = None
	        _assert_tensor_metadata_524 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_174, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_524 = None
	        _assert_tensor_metadata_525 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_116, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_525 = None
	        convert_element_type_348: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_116, torch.int8);  clamp_max_116 = None
	        view_912: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_174, [sym_size_int, 1500, 1])
	        view_913: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_348, [sym_size_int, 1500, 1])
	        reciprocal_58: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_912);  view_912 = None
	        mul_5684: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_58, 1.0);  reciprocal_58 = None
	        mul_5687: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_8914, mul_5684);  add_8914 = mul_5684 = None
	        round_118: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5687);  mul_5687 = None
	        add_9001: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_118, view_913);  round_118 = view_913 = None
	        clamp_min_176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9001, -128);  add_9001 = None
	        clamp_max_117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_176, 127);  clamp_min_176 = None
	        _assert_tensor_metadata_526 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_117, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_526 = None
	        convert_element_type_349: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_117, torch.int8);  clamp_max_117 = None
	        view_916: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_174, [sym_size_int, 1500, 1]);  clamp_min_174 = None
	        view_917: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_348, [sym_size_int, 1500, 1]);  convert_element_type_348 = None
	        _assert_tensor_metadata_527 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_349, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_527 = None
	        convert_element_type_350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_349, torch.float32);  convert_element_type_349 = None
	        _assert_tensor_metadata_528 = torch.ops.aten._assert_tensor_metadata.default(view_917, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_528 = None
	        convert_element_type_351: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_917, torch.float32);  view_917 = None
	        sub_2694: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_350, convert_element_type_351);  convert_element_type_350 = convert_element_type_351 = None
	        mul_5709: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2694, view_916);  sub_2694 = view_916 = None
	        _assert_tensor_metadata_529 = torch.ops.aten._assert_tensor_metadata.default(mul_5709, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_529 = None
	        view_919: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_9_fc1_parametrizations_weight_original0 = None
	        view_920: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_9_fc1_parametrizations_weight_original1 = None
	        view_921: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_9_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_530 = torch.ops.aten._assert_tensor_metadata.default(view_919, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_530 = None
	        convert_element_type_352: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_919, torch.float32);  view_919 = None
	        _assert_tensor_metadata_531 = torch.ops.aten._assert_tensor_metadata.default(view_921, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_531 = None
	        convert_element_type_353: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_921, torch.float32);  view_921 = None
	        sub_2698: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_352, convert_element_type_353);  convert_element_type_352 = convert_element_type_353 = None
	        mul_5714: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2698, view_920);  sub_2698 = view_920 = None
	        view_922: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5714, [5120, 1280]);  mul_5714 = None
	        _assert_tensor_metadata_532 = torch.ops.aten._assert_tensor_metadata.default(view_922, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_532 = None
	        mul_5719: "Sym(1500*s6)" = sym_size_int * 1500
	        view_923: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5709, [mul_5719, 1280]);  mul_5709 = mul_5719 = None
	        permute_99: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_922, [1, 0]);  view_922 = None
	        addmm_48: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_9_fc1_bias, view_923, permute_99);  model_audio_tower_layers_9_fc1_bias = view_923 = permute_99 = None
	        view_924: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_48, [sym_size_int, 1500, 5120]);  addmm_48 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_5726: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_924, 0.5)
	        mul_5727: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_924, 0.7071067811865476);  view_924 = None
	        erf_11: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_5727);  mul_5727 = None
	        add_9060: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_11, 1);  erf_11 = None
	        mul_5728: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5726, add_9060);  mul_5726 = add_9060 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_5728, [2])
	        amax_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_5728, [2])
	        full_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_59, full_118);  amin_59 = full_118 = None
	        full_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_59, full_119);  amax_59 = full_119 = None
	        sub_2711: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_59, minimum_59);  maximum_59 = None
	        div_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2711, 255.0);  sub_2711 = None
	        clamp_min_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_118, 1.1920928955078125e-07);  div_118 = None
	        div_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_59, clamp_min_177);  minimum_59 = None
	        round_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_119);  div_119 = None
	        sub_2717: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_119);  round_119 = None
	        clamp_min_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2717, -128);  sub_2717 = None
	        clamp_max_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_178, 127);  clamp_min_178 = None
	        _assert_tensor_metadata_533 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_177, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_533 = None
	        _assert_tensor_metadata_534 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_118, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_534 = None
	        convert_element_type_354: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_118, torch.int8);  clamp_max_118 = None
	        view_927: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_177, [sym_size_int, 1500, 1])
	        view_928: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_354, [sym_size_int, 1500, 1])
	        reciprocal_59: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_927);  view_927 = None
	        mul_5774: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_59, 1.0);  reciprocal_59 = None
	        mul_5777: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5728, mul_5774);  mul_5728 = mul_5774 = None
	        round_120: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_5777);  mul_5777 = None
	        add_9143: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_120, view_928);  round_120 = view_928 = None
	        clamp_min_179: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9143, -128);  add_9143 = None
	        clamp_max_119: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_179, 127);  clamp_min_179 = None
	        _assert_tensor_metadata_535 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_119, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_535 = None
	        convert_element_type_355: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_119, torch.int8);  clamp_max_119 = None
	        view_931: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_177, [sym_size_int, 1500, 1]);  clamp_min_177 = None
	        view_932: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_354, [sym_size_int, 1500, 1]);  convert_element_type_354 = None
	        _assert_tensor_metadata_536 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_355, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_536 = None
	        convert_element_type_356: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_355, torch.float32);  convert_element_type_355 = None
	        _assert_tensor_metadata_537 = torch.ops.aten._assert_tensor_metadata.default(view_932, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_537 = None
	        convert_element_type_357: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_932, torch.float32);  view_932 = None
	        sub_2737: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_356, convert_element_type_357);  convert_element_type_356 = convert_element_type_357 = None
	        mul_5799: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2737, view_931);  sub_2737 = view_931 = None
	        _assert_tensor_metadata_538 = torch.ops.aten._assert_tensor_metadata.default(mul_5799, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_538 = None
	        view_934: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_9_fc2_parametrizations_weight_original0 = None
	        view_935: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_9_fc2_parametrizations_weight_original1 = None
	        view_936: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_9_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_539 = torch.ops.aten._assert_tensor_metadata.default(view_934, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_539 = None
	        convert_element_type_358: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_934, torch.float32);  view_934 = None
	        _assert_tensor_metadata_540 = torch.ops.aten._assert_tensor_metadata.default(view_936, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_540 = None
	        convert_element_type_359: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_936, torch.float32);  view_936 = None
	        sub_2741: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_358, convert_element_type_359);  convert_element_type_358 = convert_element_type_359 = None
	        mul_5804: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2741, view_935);  sub_2741 = view_935 = None
	        view_937: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5804, [1280, 5120]);  mul_5804 = None
	        _assert_tensor_metadata_541 = torch.ops.aten._assert_tensor_metadata.default(view_937, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_541 = None
	        mul_5809: "Sym(1500*s6)" = sym_size_int * 1500
	        view_938: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5799, [mul_5809, 5120]);  mul_5799 = mul_5809 = None
	        permute_100: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_937, [1, 0]);  view_937 = None
	        addmm_49: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_9_fc2_bias, view_938, permute_100);  model_audio_tower_layers_9_fc2_bias = view_938 = permute_100 = None
	        view_939: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_49, [sym_size_int, 1500, 1280]);  addmm_49 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_9206: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_8908, view_939);  add_8908 = view_939 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_81: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_9206, memory_format = torch.contiguous_format)
	        var_mean_20 = torch.ops.aten.var_mean.correction(clone_81, [2], correction = 0, keepdim = True)
	        getitem_80: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_20[0]
	        getitem_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_20[1];  var_mean_20 = None
	        add_9211: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_80, 1e-05);  getitem_80 = None
	        rsqrt_20: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_9211);  add_9211 = None
	        sub_2747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_81, getitem_81);  clone_81 = getitem_81 = None
	        mul_5820: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2747, rsqrt_20);  sub_2747 = rsqrt_20 = None
	        mul_5821: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5820, model_audio_tower_layers_10_self_attn_layer_norm_weight);  mul_5820 = model_audio_tower_layers_10_self_attn_layer_norm_weight = None
	        add_9212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5821, model_audio_tower_layers_10_self_attn_layer_norm_bias);  mul_5821 = model_audio_tower_layers_10_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_9212, [2])
	        amax_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_9212, [2])
	        full_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_60, full_120);  amin_60 = full_120 = None
	        full_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_60, full_121);  amax_60 = full_121 = None
	        sub_2758: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_60, minimum_60);  maximum_60 = None
	        div_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2758, 255.0);  sub_2758 = None
	        clamp_min_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_120, 1.1920928955078125e-07);  div_120 = None
	        div_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_60, clamp_min_180);  minimum_60 = None
	        round_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_121);  div_121 = None
	        sub_2764: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_121);  round_121 = None
	        clamp_min_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2764, -128);  sub_2764 = None
	        clamp_max_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_181, 127);  clamp_min_181 = None
	        _assert_tensor_metadata_542 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_180, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_542 = None
	        _assert_tensor_metadata_543 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_120, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_543 = None
	        convert_element_type_360: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_120, torch.int8);  clamp_max_120 = None
	        view_942: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_180, [sym_size_int, 1500, 1])
	        view_943: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_360, [sym_size_int, 1500, 1])
	        reciprocal_60: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_942);  view_942 = None
	        mul_5869: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_60, 1.0);  reciprocal_60 = None
	        mul_5872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_9212, mul_5869);  mul_5869 = None
	        round_122: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5872);  mul_5872 = None
	        add_9299: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_122, view_943);  round_122 = view_943 = None
	        clamp_min_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9299, -128);  add_9299 = None
	        clamp_max_121: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_182, 127);  clamp_min_182 = None
	        _assert_tensor_metadata_544 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_121, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_544 = None
	        convert_element_type_361: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_121, torch.int8);  clamp_max_121 = None
	        view_946: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_180, [sym_size_int, 1500, 1]);  clamp_min_180 = None
	        view_947: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_360, [sym_size_int, 1500, 1]);  convert_element_type_360 = None
	        _assert_tensor_metadata_545 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_361, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_545 = None
	        convert_element_type_362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_361, torch.float32);  convert_element_type_361 = None
	        _assert_tensor_metadata_546 = torch.ops.aten._assert_tensor_metadata.default(view_947, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_546 = None
	        convert_element_type_363: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_947, torch.float32);  view_947 = None
	        sub_2784: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_362, convert_element_type_363);  convert_element_type_362 = convert_element_type_363 = None
	        mul_5894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2784, view_946);  sub_2784 = view_946 = None
	        _assert_tensor_metadata_547 = torch.ops.aten._assert_tensor_metadata.default(mul_5894, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_547 = None
	        view_949: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_950: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_951: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_548 = torch.ops.aten._assert_tensor_metadata.default(view_949, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_548 = None
	        convert_element_type_364: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_949, torch.float32);  view_949 = None
	        _assert_tensor_metadata_549 = torch.ops.aten._assert_tensor_metadata.default(view_951, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_549 = None
	        convert_element_type_365: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_951, torch.float32);  view_951 = None
	        sub_2788: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_364, convert_element_type_365);  convert_element_type_364 = convert_element_type_365 = None
	        mul_5899: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2788, view_950);  sub_2788 = view_950 = None
	        view_952: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5899, [1280, 1280]);  mul_5899 = None
	        _assert_tensor_metadata_550 = torch.ops.aten._assert_tensor_metadata.default(view_952, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_550 = None
	        mul_5904: "Sym(1500*s6)" = sym_size_int * 1500
	        view_953: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5894, [mul_5904, 1280]);  mul_5894 = mul_5904 = None
	        permute_101: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_952, [1, 0]);  view_952 = None
	        addmm_50: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_10_self_attn_q_proj_bias, view_953, permute_101);  model_audio_tower_layers_10_self_attn_q_proj_bias = view_953 = permute_101 = None
	        view_954: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_50, [sym_size_int, 1500, 1280]);  addmm_50 = None
	        mul_5911: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_954, 0.125);  view_954 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_955: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5911, [sym_size_int, 1500, 20, 64]);  mul_5911 = None
	        permute_102: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_955, [0, 2, 1, 3]);  view_955 = None
	        clone_82: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_102, memory_format = torch.contiguous_format);  permute_102 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_9212, [2])
	        amax_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_9212, [2])
	        full_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_61, full_122);  amin_61 = full_122 = None
	        full_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_61, full_123);  amax_61 = full_123 = None
	        sub_2803: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_61, minimum_61);  maximum_61 = None
	        div_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2803, 255.0);  sub_2803 = None
	        clamp_min_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_122, 1.1920928955078125e-07);  div_122 = None
	        div_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_61, clamp_min_183);  minimum_61 = None
	        round_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_123);  div_123 = None
	        sub_2809: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_123);  round_123 = None
	        clamp_min_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2809, -128);  sub_2809 = None
	        clamp_max_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_184, 127);  clamp_min_184 = None
	        _assert_tensor_metadata_551 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_183, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_551 = None
	        _assert_tensor_metadata_552 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_122, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_552 = None
	        convert_element_type_366: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_122, torch.int8);  clamp_max_122 = None
	        view_958: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_183, [sym_size_int, 1500, 1])
	        view_959: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_366, [sym_size_int, 1500, 1])
	        reciprocal_61: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_958);  view_958 = None
	        mul_5965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_61, 1.0);  reciprocal_61 = None
	        mul_5968: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_9212, mul_5965);  mul_5965 = None
	        round_124: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5968);  mul_5968 = None
	        add_9451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_124, view_959);  round_124 = view_959 = None
	        clamp_min_185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9451, -128);  add_9451 = None
	        clamp_max_123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_185, 127);  clamp_min_185 = None
	        _assert_tensor_metadata_553 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_123, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_553 = None
	        convert_element_type_367: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_123, torch.int8);  clamp_max_123 = None
	        view_962: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_183, [sym_size_int, 1500, 1]);  clamp_min_183 = None
	        view_963: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_366, [sym_size_int, 1500, 1]);  convert_element_type_366 = None
	        _assert_tensor_metadata_554 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_367, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_554 = None
	        convert_element_type_368: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_367, torch.float32);  convert_element_type_367 = None
	        _assert_tensor_metadata_555 = torch.ops.aten._assert_tensor_metadata.default(view_963, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_555 = None
	        convert_element_type_369: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_963, torch.float32);  view_963 = None
	        sub_2829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_368, convert_element_type_369);  convert_element_type_368 = convert_element_type_369 = None
	        mul_5990: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2829, view_962);  sub_2829 = view_962 = None
	        _assert_tensor_metadata_556 = torch.ops.aten._assert_tensor_metadata.default(mul_5990, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_556 = None
	        view_965: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_966: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_967: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_557 = torch.ops.aten._assert_tensor_metadata.default(view_965, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_557 = None
	        convert_element_type_370: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_965, torch.float32);  view_965 = None
	        _assert_tensor_metadata_558 = torch.ops.aten._assert_tensor_metadata.default(view_967, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_558 = None
	        convert_element_type_371: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_967, torch.float32);  view_967 = None
	        sub_2833: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_370, convert_element_type_371);  convert_element_type_370 = convert_element_type_371 = None
	        mul_5995: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2833, view_966);  sub_2833 = view_966 = None
	        view_968: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5995, [1280, 1280]);  mul_5995 = None
	        _assert_tensor_metadata_559 = torch.ops.aten._assert_tensor_metadata.default(view_968, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_559 = None
	        permute_103: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_968, [1, 0]);  view_968 = None
	        mul_5998: "Sym(1500*s6)" = sym_size_int * 1500
	        view_969: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5990, [mul_5998, 1280]);  mul_5990 = mul_5998 = None
	        mm_10: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_969, permute_103);  view_969 = permute_103 = None
	        view_970: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_10, [sym_size_int, 1500, 1280]);  mm_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_971: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_970, [sym_size_int, -1, 20, 64]);  view_970 = None
	        permute_104: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_971, [0, 2, 1, 3]);  view_971 = None
	        clone_83: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_104, memory_format = torch.contiguous_format);  permute_104 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_9212, [2])
	        amax_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_9212, [2])
	        full_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_62, full_124);  amin_62 = full_124 = None
	        full_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_62, full_125);  amax_62 = full_125 = None
	        sub_2847: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_62, minimum_62);  maximum_62 = None
	        div_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2847, 255.0);  sub_2847 = None
	        clamp_min_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_124, 1.1920928955078125e-07);  div_124 = None
	        div_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_62, clamp_min_186);  minimum_62 = None
	        round_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_125);  div_125 = None
	        sub_2853: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_125);  round_125 = None
	        clamp_min_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2853, -128);  sub_2853 = None
	        clamp_max_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_187, 127);  clamp_min_187 = None
	        _assert_tensor_metadata_560 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_186, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_560 = None
	        _assert_tensor_metadata_561 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_124, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_561 = None
	        convert_element_type_372: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_124, torch.int8);  clamp_max_124 = None
	        view_974: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_186, [sym_size_int, 1500, 1])
	        view_975: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_372, [sym_size_int, 1500, 1])
	        reciprocal_62: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_974);  view_974 = None
	        mul_6064: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_62, 1.0);  reciprocal_62 = None
	        mul_6067: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_9212, mul_6064);  add_9212 = mul_6064 = None
	        round_126: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6067);  mul_6067 = None
	        add_9599: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_126, view_975);  round_126 = view_975 = None
	        clamp_min_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9599, -128);  add_9599 = None
	        clamp_max_125: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_188, 127);  clamp_min_188 = None
	        _assert_tensor_metadata_562 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_125, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_562 = None
	        convert_element_type_373: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_125, torch.int8);  clamp_max_125 = None
	        view_978: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_186, [sym_size_int, 1500, 1]);  clamp_min_186 = None
	        view_979: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_372, [sym_size_int, 1500, 1]);  convert_element_type_372 = None
	        _assert_tensor_metadata_563 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_373, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_563 = None
	        convert_element_type_374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_373, torch.float32);  convert_element_type_373 = None
	        _assert_tensor_metadata_564 = torch.ops.aten._assert_tensor_metadata.default(view_979, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_564 = None
	        convert_element_type_375: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_979, torch.float32);  view_979 = None
	        sub_2873: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_374, convert_element_type_375);  convert_element_type_374 = convert_element_type_375 = None
	        mul_6089: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2873, view_978);  sub_2873 = view_978 = None
	        _assert_tensor_metadata_565 = torch.ops.aten._assert_tensor_metadata.default(mul_6089, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_565 = None
	        view_981: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_982: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_983: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_566 = torch.ops.aten._assert_tensor_metadata.default(view_981, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_566 = None
	        convert_element_type_376: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_981, torch.float32);  view_981 = None
	        _assert_tensor_metadata_567 = torch.ops.aten._assert_tensor_metadata.default(view_983, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_567 = None
	        convert_element_type_377: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_983, torch.float32);  view_983 = None
	        sub_2877: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_376, convert_element_type_377);  convert_element_type_376 = convert_element_type_377 = None
	        mul_6094: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2877, view_982);  sub_2877 = view_982 = None
	        view_984: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6094, [1280, 1280]);  mul_6094 = None
	        _assert_tensor_metadata_568 = torch.ops.aten._assert_tensor_metadata.default(view_984, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_568 = None
	        mul_6099: "Sym(1500*s6)" = sym_size_int * 1500
	        view_985: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6089, [mul_6099, 1280]);  mul_6089 = mul_6099 = None
	        permute_105: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_984, [1, 0]);  view_984 = None
	        addmm_51: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_10_self_attn_v_proj_bias, view_985, permute_105);  model_audio_tower_layers_10_self_attn_v_proj_bias = view_985 = permute_105 = None
	        view_986: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_51, [sym_size_int, 1500, 1280]);  addmm_51 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_987: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_986, [sym_size_int, -1, 20, 64]);  view_986 = None
	        permute_106: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_987, [0, 2, 1, 3]);  view_987 = None
	        clone_84: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_106, memory_format = torch.contiguous_format);  permute_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_10 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_82, clone_83, clone_84, None, False, scale = 1.0);  clone_82 = clone_83 = clone_84 = None
	        getitem_82: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_10[0];  _scaled_dot_product_efficient_attention_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_107: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_82, [0, 2, 1, 3]);  getitem_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_988: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_107, [sym_size_int, 1500, -1]);  permute_107 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_988, [2])
	        amax_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_988, [2])
	        full_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_63, full_126);  amin_63 = full_126 = None
	        full_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_63, full_127);  amax_63 = full_127 = None
	        sub_2895: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_63, minimum_63);  maximum_63 = None
	        div_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2895, 255.0);  sub_2895 = None
	        clamp_min_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_126, 1.1920928955078125e-07);  div_126 = None
	        div_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_63, clamp_min_189);  minimum_63 = None
	        round_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_127);  div_127 = None
	        sub_2901: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_127);  round_127 = None
	        clamp_min_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2901, -128);  sub_2901 = None
	        clamp_max_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_190, 127);  clamp_min_190 = None
	        _assert_tensor_metadata_569 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_189, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_569 = None
	        _assert_tensor_metadata_570 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_126, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_570 = None
	        convert_element_type_378: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_126, torch.int8);  clamp_max_126 = None
	        view_991: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_189, [sym_size_int, 1500, 1])
	        view_992: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_378, [sym_size_int, 1500, 1])
	        reciprocal_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_991);  view_991 = None
	        mul_6169: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_63, 1.0);  reciprocal_63 = None
	        mul_6172: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_988, mul_6169);  view_988 = mul_6169 = None
	        round_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6172);  mul_6172 = None
	        add_9763: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_128, view_992);  round_128 = view_992 = None
	        clamp_min_191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9763, -128);  add_9763 = None
	        clamp_max_127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_191, 127);  clamp_min_191 = None
	        _assert_tensor_metadata_571 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_127, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_571 = None
	        convert_element_type_379: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_127, torch.int8);  clamp_max_127 = None
	        view_995: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_189, [sym_size_int, 1500, 1]);  clamp_min_189 = None
	        view_996: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_378, [sym_size_int, 1500, 1]);  convert_element_type_378 = None
	        _assert_tensor_metadata_572 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_379, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_572 = None
	        convert_element_type_380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_379, torch.float32);  convert_element_type_379 = None
	        _assert_tensor_metadata_573 = torch.ops.aten._assert_tensor_metadata.default(view_996, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_573 = None
	        convert_element_type_381: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_996, torch.float32);  view_996 = None
	        sub_2921: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_380, convert_element_type_381);  convert_element_type_380 = convert_element_type_381 = None
	        mul_6194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2921, view_995);  sub_2921 = view_995 = None
	        _assert_tensor_metadata_574 = torch.ops.aten._assert_tensor_metadata.default(mul_6194, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_574 = None
	        view_998: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_999: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1000: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_575 = torch.ops.aten._assert_tensor_metadata.default(view_998, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_575 = None
	        convert_element_type_382: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_998, torch.float32);  view_998 = None
	        _assert_tensor_metadata_576 = torch.ops.aten._assert_tensor_metadata.default(view_1000, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_576 = None
	        convert_element_type_383: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1000, torch.float32);  view_1000 = None
	        sub_2925: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_382, convert_element_type_383);  convert_element_type_382 = convert_element_type_383 = None
	        mul_6199: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2925, view_999);  sub_2925 = view_999 = None
	        view_1001: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6199, [1280, 1280]);  mul_6199 = None
	        _assert_tensor_metadata_577 = torch.ops.aten._assert_tensor_metadata.default(view_1001, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_577 = None
	        mul_6204: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1002: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6194, [mul_6204, 1280]);  mul_6194 = mul_6204 = None
	        permute_108: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1001, [1, 0]);  view_1001 = None
	        addmm_52: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_10_self_attn_out_proj_bias, view_1002, permute_108);  model_audio_tower_layers_10_self_attn_out_proj_bias = view_1002 = permute_108 = None
	        view_1003: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_52, [sym_size_int, 1500, 1280]);  addmm_52 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_9826: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_9206, view_1003);  add_9206 = view_1003 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_9826, memory_format = torch.contiguous_format)
	        var_mean_21 = torch.ops.aten.var_mean.correction(clone_86, [2], correction = 0, keepdim = True)
	        getitem_86: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_21[0]
	        getitem_87: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_21[1];  var_mean_21 = None
	        add_9831: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_86, 1e-05);  getitem_86 = None
	        rsqrt_21: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_9831);  add_9831 = None
	        sub_2931: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_86, getitem_87);  clone_86 = getitem_87 = None
	        mul_6215: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2931, rsqrt_21);  sub_2931 = rsqrt_21 = None
	        mul_6216: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6215, model_audio_tower_layers_10_final_layer_norm_weight);  mul_6215 = model_audio_tower_layers_10_final_layer_norm_weight = None
	        add_9832: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6216, model_audio_tower_layers_10_final_layer_norm_bias);  mul_6216 = model_audio_tower_layers_10_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_9832, [2])
	        amax_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_9832, [2])
	        full_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_64, full_128);  amin_64 = full_128 = None
	        full_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_64, full_129);  amax_64 = full_129 = None
	        sub_2942: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_64, minimum_64);  maximum_64 = None
	        div_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2942, 255.0);  sub_2942 = None
	        clamp_min_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_128, 1.1920928955078125e-07);  div_128 = None
	        div_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_64, clamp_min_192);  minimum_64 = None
	        round_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_129);  div_129 = None
	        sub_2948: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_129);  round_129 = None
	        clamp_min_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2948, -128);  sub_2948 = None
	        clamp_max_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_193, 127);  clamp_min_193 = None
	        _assert_tensor_metadata_578 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_192, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_578 = None
	        _assert_tensor_metadata_579 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_128, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_579 = None
	        convert_element_type_384: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_128, torch.int8);  clamp_max_128 = None
	        view_1006: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_192, [sym_size_int, 1500, 1])
	        view_1007: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_384, [sym_size_int, 1500, 1])
	        reciprocal_64: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1006);  view_1006 = None
	        mul_6264: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_64, 1.0);  reciprocal_64 = None
	        mul_6267: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_9832, mul_6264);  add_9832 = mul_6264 = None
	        round_130: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6267);  mul_6267 = None
	        add_9919: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_130, view_1007);  round_130 = view_1007 = None
	        clamp_min_194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9919, -128);  add_9919 = None
	        clamp_max_129: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_194, 127);  clamp_min_194 = None
	        _assert_tensor_metadata_580 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_129, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_580 = None
	        convert_element_type_385: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_129, torch.int8);  clamp_max_129 = None
	        view_1010: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_192, [sym_size_int, 1500, 1]);  clamp_min_192 = None
	        view_1011: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_384, [sym_size_int, 1500, 1]);  convert_element_type_384 = None
	        _assert_tensor_metadata_581 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_385, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_581 = None
	        convert_element_type_386: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_385, torch.float32);  convert_element_type_385 = None
	        _assert_tensor_metadata_582 = torch.ops.aten._assert_tensor_metadata.default(view_1011, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_582 = None
	        convert_element_type_387: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1011, torch.float32);  view_1011 = None
	        sub_2968: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_386, convert_element_type_387);  convert_element_type_386 = convert_element_type_387 = None
	        mul_6289: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2968, view_1010);  sub_2968 = view_1010 = None
	        _assert_tensor_metadata_583 = torch.ops.aten._assert_tensor_metadata.default(mul_6289, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_583 = None
	        view_1013: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_10_fc1_parametrizations_weight_original0 = None
	        view_1014: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_10_fc1_parametrizations_weight_original1 = None
	        view_1015: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_10_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_584 = torch.ops.aten._assert_tensor_metadata.default(view_1013, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_584 = None
	        convert_element_type_388: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1013, torch.float32);  view_1013 = None
	        _assert_tensor_metadata_585 = torch.ops.aten._assert_tensor_metadata.default(view_1015, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_585 = None
	        convert_element_type_389: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1015, torch.float32);  view_1015 = None
	        sub_2972: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_388, convert_element_type_389);  convert_element_type_388 = convert_element_type_389 = None
	        mul_6294: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2972, view_1014);  sub_2972 = view_1014 = None
	        view_1016: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6294, [5120, 1280]);  mul_6294 = None
	        _assert_tensor_metadata_586 = torch.ops.aten._assert_tensor_metadata.default(view_1016, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_586 = None
	        mul_6299: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1017: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6289, [mul_6299, 1280]);  mul_6289 = mul_6299 = None
	        permute_109: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1016, [1, 0]);  view_1016 = None
	        addmm_53: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_10_fc1_bias, view_1017, permute_109);  model_audio_tower_layers_10_fc1_bias = view_1017 = permute_109 = None
	        view_1018: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_53, [sym_size_int, 1500, 5120]);  addmm_53 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_6306: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1018, 0.5)
	        mul_6307: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1018, 0.7071067811865476);  view_1018 = None
	        erf_12: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_6307);  mul_6307 = None
	        add_9978: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_12, 1);  erf_12 = None
	        mul_6308: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6306, add_9978);  mul_6306 = add_9978 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_6308, [2])
	        amax_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_6308, [2])
	        full_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_65, full_130);  amin_65 = full_130 = None
	        full_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_65, full_131);  amax_65 = full_131 = None
	        sub_2985: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_65, minimum_65);  maximum_65 = None
	        div_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2985, 255.0);  sub_2985 = None
	        clamp_min_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_130, 1.1920928955078125e-07);  div_130 = None
	        div_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_65, clamp_min_195);  minimum_65 = None
	        round_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_131);  div_131 = None
	        sub_2991: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_131);  round_131 = None
	        clamp_min_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2991, -128);  sub_2991 = None
	        clamp_max_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_196, 127);  clamp_min_196 = None
	        _assert_tensor_metadata_587 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_195, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_587 = None
	        _assert_tensor_metadata_588 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_130, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_588 = None
	        convert_element_type_390: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_130, torch.int8);  clamp_max_130 = None
	        view_1021: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_195, [sym_size_int, 1500, 1])
	        view_1022: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_390, [sym_size_int, 1500, 1])
	        reciprocal_65: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1021);  view_1021 = None
	        mul_6354: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_65, 1.0);  reciprocal_65 = None
	        mul_6357: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6308, mul_6354);  mul_6308 = mul_6354 = None
	        round_132: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_6357);  mul_6357 = None
	        add_10061: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_132, view_1022);  round_132 = view_1022 = None
	        clamp_min_197: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10061, -128);  add_10061 = None
	        clamp_max_131: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_197, 127);  clamp_min_197 = None
	        _assert_tensor_metadata_589 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_131, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_589 = None
	        convert_element_type_391: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_131, torch.int8);  clamp_max_131 = None
	        view_1025: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_195, [sym_size_int, 1500, 1]);  clamp_min_195 = None
	        view_1026: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_390, [sym_size_int, 1500, 1]);  convert_element_type_390 = None
	        _assert_tensor_metadata_590 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_391, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_590 = None
	        convert_element_type_392: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_391, torch.float32);  convert_element_type_391 = None
	        _assert_tensor_metadata_591 = torch.ops.aten._assert_tensor_metadata.default(view_1026, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_591 = None
	        convert_element_type_393: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1026, torch.float32);  view_1026 = None
	        sub_3011: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_392, convert_element_type_393);  convert_element_type_392 = convert_element_type_393 = None
	        mul_6379: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3011, view_1025);  sub_3011 = view_1025 = None
	        _assert_tensor_metadata_592 = torch.ops.aten._assert_tensor_metadata.default(mul_6379, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_592 = None
	        view_1028: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_10_fc2_parametrizations_weight_original0 = None
	        view_1029: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_10_fc2_parametrizations_weight_original1 = None
	        view_1030: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_10_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_593 = torch.ops.aten._assert_tensor_metadata.default(view_1028, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_593 = None
	        convert_element_type_394: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1028, torch.float32);  view_1028 = None
	        _assert_tensor_metadata_594 = torch.ops.aten._assert_tensor_metadata.default(view_1030, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_594 = None
	        convert_element_type_395: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1030, torch.float32);  view_1030 = None
	        sub_3015: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_394, convert_element_type_395);  convert_element_type_394 = convert_element_type_395 = None
	        mul_6384: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3015, view_1029);  sub_3015 = view_1029 = None
	        view_1031: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6384, [1280, 5120]);  mul_6384 = None
	        _assert_tensor_metadata_595 = torch.ops.aten._assert_tensor_metadata.default(view_1031, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_595 = None
	        mul_6389: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1032: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6379, [mul_6389, 5120]);  mul_6379 = mul_6389 = None
	        permute_110: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1031, [1, 0]);  view_1031 = None
	        addmm_54: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_10_fc2_bias, view_1032, permute_110);  model_audio_tower_layers_10_fc2_bias = view_1032 = permute_110 = None
	        view_1033: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_54, [sym_size_int, 1500, 1280]);  addmm_54 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_10124: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_9826, view_1033);  add_9826 = view_1033 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_89: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_10124, memory_format = torch.contiguous_format)
	        var_mean_22 = torch.ops.aten.var_mean.correction(clone_89, [2], correction = 0, keepdim = True)
	        getitem_88: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_22[0]
	        getitem_89: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_22[1];  var_mean_22 = None
	        add_10129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_88, 1e-05);  getitem_88 = None
	        rsqrt_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_10129);  add_10129 = None
	        sub_3021: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_89, getitem_89);  clone_89 = getitem_89 = None
	        mul_6400: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3021, rsqrt_22);  sub_3021 = rsqrt_22 = None
	        mul_6401: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6400, model_audio_tower_layers_11_self_attn_layer_norm_weight);  mul_6400 = model_audio_tower_layers_11_self_attn_layer_norm_weight = None
	        add_10130: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6401, model_audio_tower_layers_11_self_attn_layer_norm_bias);  mul_6401 = model_audio_tower_layers_11_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_10130, [2])
	        amax_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_10130, [2])
	        full_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_66, full_132);  amin_66 = full_132 = None
	        full_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_66, full_133);  amax_66 = full_133 = None
	        sub_3032: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_66, minimum_66);  maximum_66 = None
	        div_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3032, 255.0);  sub_3032 = None
	        clamp_min_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_132, 1.1920928955078125e-07);  div_132 = None
	        div_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_66, clamp_min_198);  minimum_66 = None
	        round_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_133);  div_133 = None
	        sub_3038: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_133);  round_133 = None
	        clamp_min_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3038, -128);  sub_3038 = None
	        clamp_max_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_199, 127);  clamp_min_199 = None
	        _assert_tensor_metadata_596 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_198, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_596 = None
	        _assert_tensor_metadata_597 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_132, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_597 = None
	        convert_element_type_396: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_132, torch.int8);  clamp_max_132 = None
	        view_1036: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_198, [sym_size_int, 1500, 1])
	        view_1037: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_396, [sym_size_int, 1500, 1])
	        reciprocal_66: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1036);  view_1036 = None
	        mul_6449: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_66, 1.0);  reciprocal_66 = None
	        mul_6452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_10130, mul_6449);  mul_6449 = None
	        round_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6452);  mul_6452 = None
	        add_10217: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_134, view_1037);  round_134 = view_1037 = None
	        clamp_min_200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10217, -128);  add_10217 = None
	        clamp_max_133: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_200, 127);  clamp_min_200 = None
	        _assert_tensor_metadata_598 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_133, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_598 = None
	        convert_element_type_397: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_133, torch.int8);  clamp_max_133 = None
	        view_1040: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_198, [sym_size_int, 1500, 1]);  clamp_min_198 = None
	        view_1041: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_396, [sym_size_int, 1500, 1]);  convert_element_type_396 = None
	        _assert_tensor_metadata_599 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_397, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_599 = None
	        convert_element_type_398: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_397, torch.float32);  convert_element_type_397 = None
	        _assert_tensor_metadata_600 = torch.ops.aten._assert_tensor_metadata.default(view_1041, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_600 = None
	        convert_element_type_399: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1041, torch.float32);  view_1041 = None
	        sub_3058: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_398, convert_element_type_399);  convert_element_type_398 = convert_element_type_399 = None
	        mul_6474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3058, view_1040);  sub_3058 = view_1040 = None
	        _assert_tensor_metadata_601 = torch.ops.aten._assert_tensor_metadata.default(mul_6474, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_601 = None
	        view_1043: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1044: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1045: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_602 = torch.ops.aten._assert_tensor_metadata.default(view_1043, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_602 = None
	        convert_element_type_400: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1043, torch.float32);  view_1043 = None
	        _assert_tensor_metadata_603 = torch.ops.aten._assert_tensor_metadata.default(view_1045, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_603 = None
	        convert_element_type_401: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1045, torch.float32);  view_1045 = None
	        sub_3062: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_400, convert_element_type_401);  convert_element_type_400 = convert_element_type_401 = None
	        mul_6479: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3062, view_1044);  sub_3062 = view_1044 = None
	        view_1046: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6479, [1280, 1280]);  mul_6479 = None
	        _assert_tensor_metadata_604 = torch.ops.aten._assert_tensor_metadata.default(view_1046, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_604 = None
	        mul_6484: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1047: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6474, [mul_6484, 1280]);  mul_6474 = mul_6484 = None
	        permute_111: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1046, [1, 0]);  view_1046 = None
	        addmm_55: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_11_self_attn_q_proj_bias, view_1047, permute_111);  model_audio_tower_layers_11_self_attn_q_proj_bias = view_1047 = permute_111 = None
	        view_1048: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_55, [sym_size_int, 1500, 1280]);  addmm_55 = None
	        mul_6491: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1048, 0.125);  view_1048 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1049: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6491, [sym_size_int, 1500, 20, 64]);  mul_6491 = None
	        permute_112: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1049, [0, 2, 1, 3]);  view_1049 = None
	        clone_90: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_112, memory_format = torch.contiguous_format);  permute_112 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_10130, [2])
	        amax_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_10130, [2])
	        full_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_67, full_134);  amin_67 = full_134 = None
	        full_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_67, full_135);  amax_67 = full_135 = None
	        sub_3077: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_67, minimum_67);  maximum_67 = None
	        div_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3077, 255.0);  sub_3077 = None
	        clamp_min_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_134, 1.1920928955078125e-07);  div_134 = None
	        div_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_67, clamp_min_201);  minimum_67 = None
	        round_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_135);  div_135 = None
	        sub_3083: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_135);  round_135 = None
	        clamp_min_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3083, -128);  sub_3083 = None
	        clamp_max_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_202, 127);  clamp_min_202 = None
	        _assert_tensor_metadata_605 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_201, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_605 = None
	        _assert_tensor_metadata_606 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_134, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_606 = None
	        convert_element_type_402: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_134, torch.int8);  clamp_max_134 = None
	        view_1052: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_201, [sym_size_int, 1500, 1])
	        view_1053: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_402, [sym_size_int, 1500, 1])
	        reciprocal_67: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1052);  view_1052 = None
	        mul_6545: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_67, 1.0);  reciprocal_67 = None
	        mul_6548: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_10130, mul_6545);  mul_6545 = None
	        round_136: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6548);  mul_6548 = None
	        add_10369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_136, view_1053);  round_136 = view_1053 = None
	        clamp_min_203: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10369, -128);  add_10369 = None
	        clamp_max_135: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_203, 127);  clamp_min_203 = None
	        _assert_tensor_metadata_607 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_135, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_607 = None
	        convert_element_type_403: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_135, torch.int8);  clamp_max_135 = None
	        view_1056: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_201, [sym_size_int, 1500, 1]);  clamp_min_201 = None
	        view_1057: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_402, [sym_size_int, 1500, 1]);  convert_element_type_402 = None
	        _assert_tensor_metadata_608 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_403, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_608 = None
	        convert_element_type_404: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_403, torch.float32);  convert_element_type_403 = None
	        _assert_tensor_metadata_609 = torch.ops.aten._assert_tensor_metadata.default(view_1057, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_609 = None
	        convert_element_type_405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1057, torch.float32);  view_1057 = None
	        sub_3103: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_404, convert_element_type_405);  convert_element_type_404 = convert_element_type_405 = None
	        mul_6570: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3103, view_1056);  sub_3103 = view_1056 = None
	        _assert_tensor_metadata_610 = torch.ops.aten._assert_tensor_metadata.default(mul_6570, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_610 = None
	        view_1059: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1060: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1061: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_611 = torch.ops.aten._assert_tensor_metadata.default(view_1059, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_611 = None
	        convert_element_type_406: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1059, torch.float32);  view_1059 = None
	        _assert_tensor_metadata_612 = torch.ops.aten._assert_tensor_metadata.default(view_1061, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_612 = None
	        convert_element_type_407: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1061, torch.float32);  view_1061 = None
	        sub_3107: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_406, convert_element_type_407);  convert_element_type_406 = convert_element_type_407 = None
	        mul_6575: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3107, view_1060);  sub_3107 = view_1060 = None
	        view_1062: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6575, [1280, 1280]);  mul_6575 = None
	        _assert_tensor_metadata_613 = torch.ops.aten._assert_tensor_metadata.default(view_1062, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_613 = None
	        permute_113: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1062, [1, 0]);  view_1062 = None
	        mul_6578: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1063: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6570, [mul_6578, 1280]);  mul_6570 = mul_6578 = None
	        mm_11: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1063, permute_113);  view_1063 = permute_113 = None
	        view_1064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_11, [sym_size_int, 1500, 1280]);  mm_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1065: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1064, [sym_size_int, -1, 20, 64]);  view_1064 = None
	        permute_114: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1065, [0, 2, 1, 3]);  view_1065 = None
	        clone_91: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_114, memory_format = torch.contiguous_format);  permute_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_10130, [2])
	        amax_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_10130, [2])
	        full_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_68, full_136);  amin_68 = full_136 = None
	        full_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_68, full_137);  amax_68 = full_137 = None
	        sub_3121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_68, minimum_68);  maximum_68 = None
	        div_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3121, 255.0);  sub_3121 = None
	        clamp_min_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_136, 1.1920928955078125e-07);  div_136 = None
	        div_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_68, clamp_min_204);  minimum_68 = None
	        round_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_137);  div_137 = None
	        sub_3127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_137);  round_137 = None
	        clamp_min_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3127, -128);  sub_3127 = None
	        clamp_max_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_205, 127);  clamp_min_205 = None
	        _assert_tensor_metadata_614 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_204, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_614 = None
	        _assert_tensor_metadata_615 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_136, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_615 = None
	        convert_element_type_408: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_136, torch.int8);  clamp_max_136 = None
	        view_1068: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_204, [sym_size_int, 1500, 1])
	        view_1069: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_408, [sym_size_int, 1500, 1])
	        reciprocal_68: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1068);  view_1068 = None
	        mul_6644: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_68, 1.0);  reciprocal_68 = None
	        mul_6647: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_10130, mul_6644);  add_10130 = mul_6644 = None
	        round_138: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6647);  mul_6647 = None
	        add_10517: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_138, view_1069);  round_138 = view_1069 = None
	        clamp_min_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10517, -128);  add_10517 = None
	        clamp_max_137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_206, 127);  clamp_min_206 = None
	        _assert_tensor_metadata_616 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_137, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_616 = None
	        convert_element_type_409: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_137, torch.int8);  clamp_max_137 = None
	        view_1072: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_204, [sym_size_int, 1500, 1]);  clamp_min_204 = None
	        view_1073: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_408, [sym_size_int, 1500, 1]);  convert_element_type_408 = None
	        _assert_tensor_metadata_617 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_409, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_617 = None
	        convert_element_type_410: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_409, torch.float32);  convert_element_type_409 = None
	        _assert_tensor_metadata_618 = torch.ops.aten._assert_tensor_metadata.default(view_1073, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_618 = None
	        convert_element_type_411: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1073, torch.float32);  view_1073 = None
	        sub_3147: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_410, convert_element_type_411);  convert_element_type_410 = convert_element_type_411 = None
	        mul_6669: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3147, view_1072);  sub_3147 = view_1072 = None
	        _assert_tensor_metadata_619 = torch.ops.aten._assert_tensor_metadata.default(mul_6669, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_619 = None
	        view_1075: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1076: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1077: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_620 = torch.ops.aten._assert_tensor_metadata.default(view_1075, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_620 = None
	        convert_element_type_412: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1075, torch.float32);  view_1075 = None
	        _assert_tensor_metadata_621 = torch.ops.aten._assert_tensor_metadata.default(view_1077, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_621 = None
	        convert_element_type_413: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1077, torch.float32);  view_1077 = None
	        sub_3151: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_412, convert_element_type_413);  convert_element_type_412 = convert_element_type_413 = None
	        mul_6674: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3151, view_1076);  sub_3151 = view_1076 = None
	        view_1078: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6674, [1280, 1280]);  mul_6674 = None
	        _assert_tensor_metadata_622 = torch.ops.aten._assert_tensor_metadata.default(view_1078, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_622 = None
	        mul_6679: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1079: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6669, [mul_6679, 1280]);  mul_6669 = mul_6679 = None
	        permute_115: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1078, [1, 0]);  view_1078 = None
	        addmm_56: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_11_self_attn_v_proj_bias, view_1079, permute_115);  model_audio_tower_layers_11_self_attn_v_proj_bias = view_1079 = permute_115 = None
	        view_1080: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_56, [sym_size_int, 1500, 1280]);  addmm_56 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1081: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1080, [sym_size_int, -1, 20, 64]);  view_1080 = None
	        permute_116: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1081, [0, 2, 1, 3]);  view_1081 = None
	        clone_92: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_116, memory_format = torch.contiguous_format);  permute_116 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_11 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_90, clone_91, clone_92, None, False, scale = 1.0);  clone_90 = clone_91 = clone_92 = None
	        getitem_90: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_11[0];  _scaled_dot_product_efficient_attention_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_117: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_90, [0, 2, 1, 3]);  getitem_90 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1082: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_117, [sym_size_int, 1500, -1]);  permute_117 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1082, [2])
	        amax_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1082, [2])
	        full_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_69, full_138);  amin_69 = full_138 = None
	        full_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_69, full_139);  amax_69 = full_139 = None
	        sub_3169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_69, minimum_69);  maximum_69 = None
	        div_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3169, 255.0);  sub_3169 = None
	        clamp_min_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_138, 1.1920928955078125e-07);  div_138 = None
	        div_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_69, clamp_min_207);  minimum_69 = None
	        round_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_139);  div_139 = None
	        sub_3175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_139);  round_139 = None
	        clamp_min_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3175, -128);  sub_3175 = None
	        clamp_max_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_208, 127);  clamp_min_208 = None
	        _assert_tensor_metadata_623 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_207, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_623 = None
	        _assert_tensor_metadata_624 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_138, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_624 = None
	        convert_element_type_414: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_138, torch.int8);  clamp_max_138 = None
	        view_1085: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_207, [sym_size_int, 1500, 1])
	        view_1086: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_414, [sym_size_int, 1500, 1])
	        reciprocal_69: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1085);  view_1085 = None
	        mul_6749: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_69, 1.0);  reciprocal_69 = None
	        mul_6752: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1082, mul_6749);  view_1082 = mul_6749 = None
	        round_140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6752);  mul_6752 = None
	        add_10681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_140, view_1086);  round_140 = view_1086 = None
	        clamp_min_209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10681, -128);  add_10681 = None
	        clamp_max_139: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_209, 127);  clamp_min_209 = None
	        _assert_tensor_metadata_625 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_139, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_625 = None
	        convert_element_type_415: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_139, torch.int8);  clamp_max_139 = None
	        view_1089: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_207, [sym_size_int, 1500, 1]);  clamp_min_207 = None
	        view_1090: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_414, [sym_size_int, 1500, 1]);  convert_element_type_414 = None
	        _assert_tensor_metadata_626 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_415, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_626 = None
	        convert_element_type_416: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_415, torch.float32);  convert_element_type_415 = None
	        _assert_tensor_metadata_627 = torch.ops.aten._assert_tensor_metadata.default(view_1090, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_627 = None
	        convert_element_type_417: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1090, torch.float32);  view_1090 = None
	        sub_3195: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_416, convert_element_type_417);  convert_element_type_416 = convert_element_type_417 = None
	        mul_6774: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3195, view_1089);  sub_3195 = view_1089 = None
	        _assert_tensor_metadata_628 = torch.ops.aten._assert_tensor_metadata.default(mul_6774, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_628 = None
	        view_1092: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1093: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1094: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_629 = torch.ops.aten._assert_tensor_metadata.default(view_1092, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_629 = None
	        convert_element_type_418: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1092, torch.float32);  view_1092 = None
	        _assert_tensor_metadata_630 = torch.ops.aten._assert_tensor_metadata.default(view_1094, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_630 = None
	        convert_element_type_419: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1094, torch.float32);  view_1094 = None
	        sub_3199: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_418, convert_element_type_419);  convert_element_type_418 = convert_element_type_419 = None
	        mul_6779: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3199, view_1093);  sub_3199 = view_1093 = None
	        view_1095: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6779, [1280, 1280]);  mul_6779 = None
	        _assert_tensor_metadata_631 = torch.ops.aten._assert_tensor_metadata.default(view_1095, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_631 = None
	        mul_6784: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1096: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6774, [mul_6784, 1280]);  mul_6774 = mul_6784 = None
	        permute_118: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1095, [1, 0]);  view_1095 = None
	        addmm_57: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_11_self_attn_out_proj_bias, view_1096, permute_118);  model_audio_tower_layers_11_self_attn_out_proj_bias = view_1096 = permute_118 = None
	        view_1097: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_57, [sym_size_int, 1500, 1280]);  addmm_57 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_10744: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_10124, view_1097);  add_10124 = view_1097 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_94: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_10744, memory_format = torch.contiguous_format)
	        var_mean_23 = torch.ops.aten.var_mean.correction(clone_94, [2], correction = 0, keepdim = True)
	        getitem_94: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_23[0]
	        getitem_95: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_23[1];  var_mean_23 = None
	        add_10749: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_94, 1e-05);  getitem_94 = None
	        rsqrt_23: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_10749);  add_10749 = None
	        sub_3205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_94, getitem_95);  clone_94 = getitem_95 = None
	        mul_6795: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3205, rsqrt_23);  sub_3205 = rsqrt_23 = None
	        mul_6796: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6795, model_audio_tower_layers_11_final_layer_norm_weight);  mul_6795 = model_audio_tower_layers_11_final_layer_norm_weight = None
	        add_10750: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6796, model_audio_tower_layers_11_final_layer_norm_bias);  mul_6796 = model_audio_tower_layers_11_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_10750, [2])
	        amax_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_10750, [2])
	        full_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_70, full_140);  amin_70 = full_140 = None
	        full_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_70, full_141);  amax_70 = full_141 = None
	        sub_3216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_70, minimum_70);  maximum_70 = None
	        div_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3216, 255.0);  sub_3216 = None
	        clamp_min_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_140, 1.1920928955078125e-07);  div_140 = None
	        div_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_70, clamp_min_210);  minimum_70 = None
	        round_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_141);  div_141 = None
	        sub_3222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_141);  round_141 = None
	        clamp_min_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3222, -128);  sub_3222 = None
	        clamp_max_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_211, 127);  clamp_min_211 = None
	        _assert_tensor_metadata_632 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_210, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_632 = None
	        _assert_tensor_metadata_633 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_140, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_633 = None
	        convert_element_type_420: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_140, torch.int8);  clamp_max_140 = None
	        view_1100: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_210, [sym_size_int, 1500, 1])
	        view_1101: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_420, [sym_size_int, 1500, 1])
	        reciprocal_70: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1100);  view_1100 = None
	        mul_6844: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_70, 1.0);  reciprocal_70 = None
	        mul_6847: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_10750, mul_6844);  add_10750 = mul_6844 = None
	        round_142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6847);  mul_6847 = None
	        add_10837: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_142, view_1101);  round_142 = view_1101 = None
	        clamp_min_212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10837, -128);  add_10837 = None
	        clamp_max_141: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_212, 127);  clamp_min_212 = None
	        _assert_tensor_metadata_634 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_141, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_634 = None
	        convert_element_type_421: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_141, torch.int8);  clamp_max_141 = None
	        view_1104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_210, [sym_size_int, 1500, 1]);  clamp_min_210 = None
	        view_1105: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_420, [sym_size_int, 1500, 1]);  convert_element_type_420 = None
	        _assert_tensor_metadata_635 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_421, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_635 = None
	        convert_element_type_422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_421, torch.float32);  convert_element_type_421 = None
	        _assert_tensor_metadata_636 = torch.ops.aten._assert_tensor_metadata.default(view_1105, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_636 = None
	        convert_element_type_423: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1105, torch.float32);  view_1105 = None
	        sub_3242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_422, convert_element_type_423);  convert_element_type_422 = convert_element_type_423 = None
	        mul_6869: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3242, view_1104);  sub_3242 = view_1104 = None
	        _assert_tensor_metadata_637 = torch.ops.aten._assert_tensor_metadata.default(mul_6869, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_637 = None
	        view_1107: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_11_fc1_parametrizations_weight_original0 = None
	        view_1108: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_11_fc1_parametrizations_weight_original1 = None
	        view_1109: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_11_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_638 = torch.ops.aten._assert_tensor_metadata.default(view_1107, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_638 = None
	        convert_element_type_424: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1107, torch.float32);  view_1107 = None
	        _assert_tensor_metadata_639 = torch.ops.aten._assert_tensor_metadata.default(view_1109, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_639 = None
	        convert_element_type_425: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1109, torch.float32);  view_1109 = None
	        sub_3246: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_424, convert_element_type_425);  convert_element_type_424 = convert_element_type_425 = None
	        mul_6874: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3246, view_1108);  sub_3246 = view_1108 = None
	        view_1110: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6874, [5120, 1280]);  mul_6874 = None
	        _assert_tensor_metadata_640 = torch.ops.aten._assert_tensor_metadata.default(view_1110, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_640 = None
	        mul_6879: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1111: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6869, [mul_6879, 1280]);  mul_6869 = mul_6879 = None
	        permute_119: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1110, [1, 0]);  view_1110 = None
	        addmm_58: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_11_fc1_bias, view_1111, permute_119);  model_audio_tower_layers_11_fc1_bias = view_1111 = permute_119 = None
	        view_1112: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_58, [sym_size_int, 1500, 5120]);  addmm_58 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_6886: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1112, 0.5)
	        mul_6887: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1112, 0.7071067811865476);  view_1112 = None
	        erf_13: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_6887);  mul_6887 = None
	        add_10896: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_13, 1);  erf_13 = None
	        mul_6888: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6886, add_10896);  mul_6886 = add_10896 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_6888, [2])
	        amax_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_6888, [2])
	        full_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_71, full_142);  amin_71 = full_142 = None
	        full_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_71, full_143);  amax_71 = full_143 = None
	        sub_3259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_71, minimum_71);  maximum_71 = None
	        div_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3259, 255.0);  sub_3259 = None
	        clamp_min_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_142, 1.1920928955078125e-07);  div_142 = None
	        div_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_71, clamp_min_213);  minimum_71 = None
	        round_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_143);  div_143 = None
	        sub_3265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_143);  round_143 = None
	        clamp_min_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3265, -128);  sub_3265 = None
	        clamp_max_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_214, 127);  clamp_min_214 = None
	        _assert_tensor_metadata_641 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_213, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_641 = None
	        _assert_tensor_metadata_642 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_142, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_642 = None
	        convert_element_type_426: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_142, torch.int8);  clamp_max_142 = None
	        view_1115: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_213, [sym_size_int, 1500, 1])
	        view_1116: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_426, [sym_size_int, 1500, 1])
	        reciprocal_71: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1115);  view_1115 = None
	        mul_6934: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_71, 1.0);  reciprocal_71 = None
	        mul_6937: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6888, mul_6934);  mul_6888 = mul_6934 = None
	        round_144: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_6937);  mul_6937 = None
	        add_10979: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_144, view_1116);  round_144 = view_1116 = None
	        clamp_min_215: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10979, -128);  add_10979 = None
	        clamp_max_143: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_215, 127);  clamp_min_215 = None
	        _assert_tensor_metadata_643 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_143, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_643 = None
	        convert_element_type_427: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_143, torch.int8);  clamp_max_143 = None
	        view_1119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_213, [sym_size_int, 1500, 1]);  clamp_min_213 = None
	        view_1120: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_426, [sym_size_int, 1500, 1]);  convert_element_type_426 = None
	        _assert_tensor_metadata_644 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_427, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_644 = None
	        convert_element_type_428: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_427, torch.float32);  convert_element_type_427 = None
	        _assert_tensor_metadata_645 = torch.ops.aten._assert_tensor_metadata.default(view_1120, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_645 = None
	        convert_element_type_429: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1120, torch.float32);  view_1120 = None
	        sub_3285: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_428, convert_element_type_429);  convert_element_type_428 = convert_element_type_429 = None
	        mul_6959: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3285, view_1119);  sub_3285 = view_1119 = None
	        _assert_tensor_metadata_646 = torch.ops.aten._assert_tensor_metadata.default(mul_6959, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_646 = None
	        view_1122: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_11_fc2_parametrizations_weight_original0 = None
	        view_1123: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_11_fc2_parametrizations_weight_original1 = None
	        view_1124: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_11_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_647 = torch.ops.aten._assert_tensor_metadata.default(view_1122, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_647 = None
	        convert_element_type_430: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1122, torch.float32);  view_1122 = None
	        _assert_tensor_metadata_648 = torch.ops.aten._assert_tensor_metadata.default(view_1124, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_648 = None
	        convert_element_type_431: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1124, torch.float32);  view_1124 = None
	        sub_3289: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_430, convert_element_type_431);  convert_element_type_430 = convert_element_type_431 = None
	        mul_6964: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3289, view_1123);  sub_3289 = view_1123 = None
	        view_1125: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6964, [1280, 5120]);  mul_6964 = None
	        _assert_tensor_metadata_649 = torch.ops.aten._assert_tensor_metadata.default(view_1125, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_649 = None
	        mul_6969: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1126: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6959, [mul_6969, 5120]);  mul_6959 = mul_6969 = None
	        permute_120: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1125, [1, 0]);  view_1125 = None
	        addmm_59: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_11_fc2_bias, view_1126, permute_120);  model_audio_tower_layers_11_fc2_bias = view_1126 = permute_120 = None
	        view_1127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_59, [sym_size_int, 1500, 1280]);  addmm_59 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_11042: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_10744, view_1127);  add_10744 = view_1127 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_97: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_11042, memory_format = torch.contiguous_format)
	        var_mean_24 = torch.ops.aten.var_mean.correction(clone_97, [2], correction = 0, keepdim = True)
	        getitem_96: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_24[0]
	        getitem_97: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_24[1];  var_mean_24 = None
	        add_11047: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_96, 1e-05);  getitem_96 = None
	        rsqrt_24: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_11047);  add_11047 = None
	        sub_3295: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_97, getitem_97);  clone_97 = getitem_97 = None
	        mul_6980: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3295, rsqrt_24);  sub_3295 = rsqrt_24 = None
	        mul_6981: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6980, model_audio_tower_layers_12_self_attn_layer_norm_weight);  mul_6980 = model_audio_tower_layers_12_self_attn_layer_norm_weight = None
	        add_11048: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6981, model_audio_tower_layers_12_self_attn_layer_norm_bias);  mul_6981 = model_audio_tower_layers_12_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11048, [2])
	        amax_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11048, [2])
	        full_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_72, full_144);  amin_72 = full_144 = None
	        full_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_72, full_145);  amax_72 = full_145 = None
	        sub_3306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_72, minimum_72);  maximum_72 = None
	        div_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3306, 255.0);  sub_3306 = None
	        clamp_min_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_144, 1.1920928955078125e-07);  div_144 = None
	        div_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_72, clamp_min_216);  minimum_72 = None
	        round_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_145);  div_145 = None
	        sub_3312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_145);  round_145 = None
	        clamp_min_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3312, -128);  sub_3312 = None
	        clamp_max_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_217, 127);  clamp_min_217 = None
	        _assert_tensor_metadata_650 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_216, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_650 = None
	        _assert_tensor_metadata_651 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_144, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_651 = None
	        convert_element_type_432: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_144, torch.int8);  clamp_max_144 = None
	        view_1130: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_216, [sym_size_int, 1500, 1])
	        view_1131: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_432, [sym_size_int, 1500, 1])
	        reciprocal_72: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1130);  view_1130 = None
	        mul_7029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_72, 1.0);  reciprocal_72 = None
	        mul_7032: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11048, mul_7029);  mul_7029 = None
	        round_146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7032);  mul_7032 = None
	        add_11135: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_146, view_1131);  round_146 = view_1131 = None
	        clamp_min_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11135, -128);  add_11135 = None
	        clamp_max_145: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_218, 127);  clamp_min_218 = None
	        _assert_tensor_metadata_652 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_145, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_652 = None
	        convert_element_type_433: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_145, torch.int8);  clamp_max_145 = None
	        view_1134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_216, [sym_size_int, 1500, 1]);  clamp_min_216 = None
	        view_1135: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_432, [sym_size_int, 1500, 1]);  convert_element_type_432 = None
	        _assert_tensor_metadata_653 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_433, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_653 = None
	        convert_element_type_434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_433, torch.float32);  convert_element_type_433 = None
	        _assert_tensor_metadata_654 = torch.ops.aten._assert_tensor_metadata.default(view_1135, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_654 = None
	        convert_element_type_435: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1135, torch.float32);  view_1135 = None
	        sub_3332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_434, convert_element_type_435);  convert_element_type_434 = convert_element_type_435 = None
	        mul_7054: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3332, view_1134);  sub_3332 = view_1134 = None
	        _assert_tensor_metadata_655 = torch.ops.aten._assert_tensor_metadata.default(mul_7054, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_655 = None
	        view_1137: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1138: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1139: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_656 = torch.ops.aten._assert_tensor_metadata.default(view_1137, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_656 = None
	        convert_element_type_436: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1137, torch.float32);  view_1137 = None
	        _assert_tensor_metadata_657 = torch.ops.aten._assert_tensor_metadata.default(view_1139, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_657 = None
	        convert_element_type_437: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1139, torch.float32);  view_1139 = None
	        sub_3336: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_436, convert_element_type_437);  convert_element_type_436 = convert_element_type_437 = None
	        mul_7059: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3336, view_1138);  sub_3336 = view_1138 = None
	        view_1140: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7059, [1280, 1280]);  mul_7059 = None
	        _assert_tensor_metadata_658 = torch.ops.aten._assert_tensor_metadata.default(view_1140, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_658 = None
	        mul_7064: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1141: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7054, [mul_7064, 1280]);  mul_7054 = mul_7064 = None
	        permute_121: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1140, [1, 0]);  view_1140 = None
	        addmm_60: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_12_self_attn_q_proj_bias, view_1141, permute_121);  model_audio_tower_layers_12_self_attn_q_proj_bias = view_1141 = permute_121 = None
	        view_1142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_60, [sym_size_int, 1500, 1280]);  addmm_60 = None
	        mul_7071: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1142, 0.125);  view_1142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1143: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7071, [sym_size_int, 1500, 20, 64]);  mul_7071 = None
	        permute_122: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1143, [0, 2, 1, 3]);  view_1143 = None
	        clone_98: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_122, memory_format = torch.contiguous_format);  permute_122 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11048, [2])
	        amax_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11048, [2])
	        full_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_73, full_146);  amin_73 = full_146 = None
	        full_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_73, full_147);  amax_73 = full_147 = None
	        sub_3351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_73, minimum_73);  maximum_73 = None
	        div_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3351, 255.0);  sub_3351 = None
	        clamp_min_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_146, 1.1920928955078125e-07);  div_146 = None
	        div_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_73, clamp_min_219);  minimum_73 = None
	        round_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_147);  div_147 = None
	        sub_3357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_147);  round_147 = None
	        clamp_min_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3357, -128);  sub_3357 = None
	        clamp_max_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_220, 127);  clamp_min_220 = None
	        _assert_tensor_metadata_659 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_219, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_659 = None
	        _assert_tensor_metadata_660 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_146, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_660 = None
	        convert_element_type_438: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_146, torch.int8);  clamp_max_146 = None
	        view_1146: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_219, [sym_size_int, 1500, 1])
	        view_1147: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_438, [sym_size_int, 1500, 1])
	        reciprocal_73: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1146);  view_1146 = None
	        mul_7125: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_73, 1.0);  reciprocal_73 = None
	        mul_7128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11048, mul_7125);  mul_7125 = None
	        round_148: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7128);  mul_7128 = None
	        add_11287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_148, view_1147);  round_148 = view_1147 = None
	        clamp_min_221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11287, -128);  add_11287 = None
	        clamp_max_147: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_221, 127);  clamp_min_221 = None
	        _assert_tensor_metadata_661 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_147, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_661 = None
	        convert_element_type_439: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_147, torch.int8);  clamp_max_147 = None
	        view_1150: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_219, [sym_size_int, 1500, 1]);  clamp_min_219 = None
	        view_1151: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_438, [sym_size_int, 1500, 1]);  convert_element_type_438 = None
	        _assert_tensor_metadata_662 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_439, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_662 = None
	        convert_element_type_440: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_439, torch.float32);  convert_element_type_439 = None
	        _assert_tensor_metadata_663 = torch.ops.aten._assert_tensor_metadata.default(view_1151, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_663 = None
	        convert_element_type_441: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1151, torch.float32);  view_1151 = None
	        sub_3377: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_440, convert_element_type_441);  convert_element_type_440 = convert_element_type_441 = None
	        mul_7150: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3377, view_1150);  sub_3377 = view_1150 = None
	        _assert_tensor_metadata_664 = torch.ops.aten._assert_tensor_metadata.default(mul_7150, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_664 = None
	        view_1153: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1154: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1155: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_665 = torch.ops.aten._assert_tensor_metadata.default(view_1153, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_665 = None
	        convert_element_type_442: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1153, torch.float32);  view_1153 = None
	        _assert_tensor_metadata_666 = torch.ops.aten._assert_tensor_metadata.default(view_1155, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_666 = None
	        convert_element_type_443: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1155, torch.float32);  view_1155 = None
	        sub_3381: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_442, convert_element_type_443);  convert_element_type_442 = convert_element_type_443 = None
	        mul_7155: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3381, view_1154);  sub_3381 = view_1154 = None
	        view_1156: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7155, [1280, 1280]);  mul_7155 = None
	        _assert_tensor_metadata_667 = torch.ops.aten._assert_tensor_metadata.default(view_1156, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_667 = None
	        permute_123: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1156, [1, 0]);  view_1156 = None
	        mul_7158: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1157: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7150, [mul_7158, 1280]);  mul_7150 = mul_7158 = None
	        mm_12: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1157, permute_123);  view_1157 = permute_123 = None
	        view_1158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_12, [sym_size_int, 1500, 1280]);  mm_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1159: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1158, [sym_size_int, -1, 20, 64]);  view_1158 = None
	        permute_124: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1159, [0, 2, 1, 3]);  view_1159 = None
	        clone_99: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_124, memory_format = torch.contiguous_format);  permute_124 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11048, [2])
	        amax_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11048, [2])
	        full_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_74, full_148);  amin_74 = full_148 = None
	        full_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_74, full_149);  amax_74 = full_149 = None
	        sub_3395: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_74, minimum_74);  maximum_74 = None
	        div_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3395, 255.0);  sub_3395 = None
	        clamp_min_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_148, 1.1920928955078125e-07);  div_148 = None
	        div_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_74, clamp_min_222);  minimum_74 = None
	        round_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_149);  div_149 = None
	        sub_3401: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_149);  round_149 = None
	        clamp_min_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3401, -128);  sub_3401 = None
	        clamp_max_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_223, 127);  clamp_min_223 = None
	        _assert_tensor_metadata_668 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_222, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_668 = None
	        _assert_tensor_metadata_669 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_148, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_669 = None
	        convert_element_type_444: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_148, torch.int8);  clamp_max_148 = None
	        view_1162: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_222, [sym_size_int, 1500, 1])
	        view_1163: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_444, [sym_size_int, 1500, 1])
	        reciprocal_74: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1162);  view_1162 = None
	        mul_7224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_74, 1.0);  reciprocal_74 = None
	        mul_7227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11048, mul_7224);  add_11048 = mul_7224 = None
	        round_150: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7227);  mul_7227 = None
	        add_11435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_150, view_1163);  round_150 = view_1163 = None
	        clamp_min_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11435, -128);  add_11435 = None
	        clamp_max_149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_224, 127);  clamp_min_224 = None
	        _assert_tensor_metadata_670 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_149, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_670 = None
	        convert_element_type_445: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_149, torch.int8);  clamp_max_149 = None
	        view_1166: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_222, [sym_size_int, 1500, 1]);  clamp_min_222 = None
	        view_1167: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_444, [sym_size_int, 1500, 1]);  convert_element_type_444 = None
	        _assert_tensor_metadata_671 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_445, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_671 = None
	        convert_element_type_446: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_445, torch.float32);  convert_element_type_445 = None
	        _assert_tensor_metadata_672 = torch.ops.aten._assert_tensor_metadata.default(view_1167, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_672 = None
	        convert_element_type_447: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1167, torch.float32);  view_1167 = None
	        sub_3421: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_446, convert_element_type_447);  convert_element_type_446 = convert_element_type_447 = None
	        mul_7249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3421, view_1166);  sub_3421 = view_1166 = None
	        _assert_tensor_metadata_673 = torch.ops.aten._assert_tensor_metadata.default(mul_7249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_673 = None
	        view_1169: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1170: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1171: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_674 = torch.ops.aten._assert_tensor_metadata.default(view_1169, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_674 = None
	        convert_element_type_448: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1169, torch.float32);  view_1169 = None
	        _assert_tensor_metadata_675 = torch.ops.aten._assert_tensor_metadata.default(view_1171, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_675 = None
	        convert_element_type_449: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1171, torch.float32);  view_1171 = None
	        sub_3425: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_448, convert_element_type_449);  convert_element_type_448 = convert_element_type_449 = None
	        mul_7254: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3425, view_1170);  sub_3425 = view_1170 = None
	        view_1172: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7254, [1280, 1280]);  mul_7254 = None
	        _assert_tensor_metadata_676 = torch.ops.aten._assert_tensor_metadata.default(view_1172, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_676 = None
	        mul_7259: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1173: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7249, [mul_7259, 1280]);  mul_7249 = mul_7259 = None
	        permute_125: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1172, [1, 0]);  view_1172 = None
	        addmm_61: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_12_self_attn_v_proj_bias, view_1173, permute_125);  model_audio_tower_layers_12_self_attn_v_proj_bias = view_1173 = permute_125 = None
	        view_1174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_61, [sym_size_int, 1500, 1280]);  addmm_61 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1175: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1174, [sym_size_int, -1, 20, 64]);  view_1174 = None
	        permute_126: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1175, [0, 2, 1, 3]);  view_1175 = None
	        clone_100: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_126, memory_format = torch.contiguous_format);  permute_126 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_12 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_98, clone_99, clone_100, None, False, scale = 1.0);  clone_98 = clone_99 = clone_100 = None
	        getitem_98: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_12[0];  _scaled_dot_product_efficient_attention_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_127: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_98, [0, 2, 1, 3]);  getitem_98 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_127, [sym_size_int, 1500, -1]);  permute_127 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1176, [2])
	        amax_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1176, [2])
	        full_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_75, full_150);  amin_75 = full_150 = None
	        full_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_75, full_151);  amax_75 = full_151 = None
	        sub_3443: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_75, minimum_75);  maximum_75 = None
	        div_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3443, 255.0);  sub_3443 = None
	        clamp_min_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_150, 1.1920928955078125e-07);  div_150 = None
	        div_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_75, clamp_min_225);  minimum_75 = None
	        round_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_151);  div_151 = None
	        sub_3449: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_151);  round_151 = None
	        clamp_min_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3449, -128);  sub_3449 = None
	        clamp_max_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_226, 127);  clamp_min_226 = None
	        _assert_tensor_metadata_677 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_225, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_677 = None
	        _assert_tensor_metadata_678 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_150, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_678 = None
	        convert_element_type_450: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_150, torch.int8);  clamp_max_150 = None
	        view_1179: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_225, [sym_size_int, 1500, 1])
	        view_1180: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_450, [sym_size_int, 1500, 1])
	        reciprocal_75: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1179);  view_1179 = None
	        mul_7329: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_75, 1.0);  reciprocal_75 = None
	        mul_7332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1176, mul_7329);  view_1176 = mul_7329 = None
	        round_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7332);  mul_7332 = None
	        add_11599: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_152, view_1180);  round_152 = view_1180 = None
	        clamp_min_227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11599, -128);  add_11599 = None
	        clamp_max_151: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_227, 127);  clamp_min_227 = None
	        _assert_tensor_metadata_679 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_151, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_679 = None
	        convert_element_type_451: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_151, torch.int8);  clamp_max_151 = None
	        view_1183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_225, [sym_size_int, 1500, 1]);  clamp_min_225 = None
	        view_1184: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_450, [sym_size_int, 1500, 1]);  convert_element_type_450 = None
	        _assert_tensor_metadata_680 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_451, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_680 = None
	        convert_element_type_452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_451, torch.float32);  convert_element_type_451 = None
	        _assert_tensor_metadata_681 = torch.ops.aten._assert_tensor_metadata.default(view_1184, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_681 = None
	        convert_element_type_453: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1184, torch.float32);  view_1184 = None
	        sub_3469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_452, convert_element_type_453);  convert_element_type_452 = convert_element_type_453 = None
	        mul_7354: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3469, view_1183);  sub_3469 = view_1183 = None
	        _assert_tensor_metadata_682 = torch.ops.aten._assert_tensor_metadata.default(mul_7354, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_682 = None
	        view_1186: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1187: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1188: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_683 = torch.ops.aten._assert_tensor_metadata.default(view_1186, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_683 = None
	        convert_element_type_454: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1186, torch.float32);  view_1186 = None
	        _assert_tensor_metadata_684 = torch.ops.aten._assert_tensor_metadata.default(view_1188, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_684 = None
	        convert_element_type_455: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1188, torch.float32);  view_1188 = None
	        sub_3473: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_454, convert_element_type_455);  convert_element_type_454 = convert_element_type_455 = None
	        mul_7359: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3473, view_1187);  sub_3473 = view_1187 = None
	        view_1189: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7359, [1280, 1280]);  mul_7359 = None
	        _assert_tensor_metadata_685 = torch.ops.aten._assert_tensor_metadata.default(view_1189, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_685 = None
	        mul_7364: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1190: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7354, [mul_7364, 1280]);  mul_7354 = mul_7364 = None
	        permute_128: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1189, [1, 0]);  view_1189 = None
	        addmm_62: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_12_self_attn_out_proj_bias, view_1190, permute_128);  model_audio_tower_layers_12_self_attn_out_proj_bias = view_1190 = permute_128 = None
	        view_1191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_62, [sym_size_int, 1500, 1280]);  addmm_62 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_11662: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_11042, view_1191);  add_11042 = view_1191 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_102: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_11662, memory_format = torch.contiguous_format)
	        var_mean_25 = torch.ops.aten.var_mean.correction(clone_102, [2], correction = 0, keepdim = True)
	        getitem_102: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_25[0]
	        getitem_103: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_25[1];  var_mean_25 = None
	        add_11667: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_102, 1e-05);  getitem_102 = None
	        rsqrt_25: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_11667);  add_11667 = None
	        sub_3479: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_102, getitem_103);  clone_102 = getitem_103 = None
	        mul_7375: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3479, rsqrt_25);  sub_3479 = rsqrt_25 = None
	        mul_7376: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7375, model_audio_tower_layers_12_final_layer_norm_weight);  mul_7375 = model_audio_tower_layers_12_final_layer_norm_weight = None
	        add_11668: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_7376, model_audio_tower_layers_12_final_layer_norm_bias);  mul_7376 = model_audio_tower_layers_12_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11668, [2])
	        amax_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11668, [2])
	        full_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_76, full_152);  amin_76 = full_152 = None
	        full_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_76, full_153);  amax_76 = full_153 = None
	        sub_3490: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_76, minimum_76);  maximum_76 = None
	        div_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3490, 255.0);  sub_3490 = None
	        clamp_min_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_152, 1.1920928955078125e-07);  div_152 = None
	        div_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_76, clamp_min_228);  minimum_76 = None
	        round_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_153);  div_153 = None
	        sub_3496: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_153);  round_153 = None
	        clamp_min_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3496, -128);  sub_3496 = None
	        clamp_max_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_229, 127);  clamp_min_229 = None
	        _assert_tensor_metadata_686 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_228, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_686 = None
	        _assert_tensor_metadata_687 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_152, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_687 = None
	        convert_element_type_456: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_152, torch.int8);  clamp_max_152 = None
	        view_1194: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_228, [sym_size_int, 1500, 1])
	        view_1195: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_456, [sym_size_int, 1500, 1])
	        reciprocal_76: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1194);  view_1194 = None
	        mul_7424: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_76, 1.0);  reciprocal_76 = None
	        mul_7427: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11668, mul_7424);  add_11668 = mul_7424 = None
	        round_154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7427);  mul_7427 = None
	        add_11755: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_154, view_1195);  round_154 = view_1195 = None
	        clamp_min_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11755, -128);  add_11755 = None
	        clamp_max_153: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_230, 127);  clamp_min_230 = None
	        _assert_tensor_metadata_688 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_153, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_688 = None
	        convert_element_type_457: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_153, torch.int8);  clamp_max_153 = None
	        view_1198: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_228, [sym_size_int, 1500, 1]);  clamp_min_228 = None
	        view_1199: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_456, [sym_size_int, 1500, 1]);  convert_element_type_456 = None
	        _assert_tensor_metadata_689 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_457, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_689 = None
	        convert_element_type_458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_457, torch.float32);  convert_element_type_457 = None
	        _assert_tensor_metadata_690 = torch.ops.aten._assert_tensor_metadata.default(view_1199, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_690 = None
	        convert_element_type_459: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1199, torch.float32);  view_1199 = None
	        sub_3516: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_458, convert_element_type_459);  convert_element_type_458 = convert_element_type_459 = None
	        mul_7449: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3516, view_1198);  sub_3516 = view_1198 = None
	        _assert_tensor_metadata_691 = torch.ops.aten._assert_tensor_metadata.default(mul_7449, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_691 = None
	        view_1201: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_12_fc1_parametrizations_weight_original0 = None
	        view_1202: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_12_fc1_parametrizations_weight_original1 = None
	        view_1203: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_12_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_692 = torch.ops.aten._assert_tensor_metadata.default(view_1201, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_692 = None
	        convert_element_type_460: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1201, torch.float32);  view_1201 = None
	        _assert_tensor_metadata_693 = torch.ops.aten._assert_tensor_metadata.default(view_1203, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_693 = None
	        convert_element_type_461: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1203, torch.float32);  view_1203 = None
	        sub_3520: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_460, convert_element_type_461);  convert_element_type_460 = convert_element_type_461 = None
	        mul_7454: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3520, view_1202);  sub_3520 = view_1202 = None
	        view_1204: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7454, [5120, 1280]);  mul_7454 = None
	        _assert_tensor_metadata_694 = torch.ops.aten._assert_tensor_metadata.default(view_1204, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_694 = None
	        mul_7459: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1205: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7449, [mul_7459, 1280]);  mul_7449 = mul_7459 = None
	        permute_129: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1204, [1, 0]);  view_1204 = None
	        addmm_63: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_12_fc1_bias, view_1205, permute_129);  model_audio_tower_layers_12_fc1_bias = view_1205 = permute_129 = None
	        view_1206: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_63, [sym_size_int, 1500, 5120]);  addmm_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_7466: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1206, 0.5)
	        mul_7467: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1206, 0.7071067811865476);  view_1206 = None
	        erf_14: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_7467);  mul_7467 = None
	        add_11814: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_14, 1);  erf_14 = None
	        mul_7468: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7466, add_11814);  mul_7466 = add_11814 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_7468, [2])
	        amax_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_7468, [2])
	        full_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_77, full_154);  amin_77 = full_154 = None
	        full_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_77, full_155);  amax_77 = full_155 = None
	        sub_3533: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_77, minimum_77);  maximum_77 = None
	        div_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3533, 255.0);  sub_3533 = None
	        clamp_min_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_154, 1.1920928955078125e-07);  div_154 = None
	        div_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_77, clamp_min_231);  minimum_77 = None
	        round_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_155);  div_155 = None
	        sub_3539: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_155);  round_155 = None
	        clamp_min_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3539, -128);  sub_3539 = None
	        clamp_max_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_232, 127);  clamp_min_232 = None
	        _assert_tensor_metadata_695 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_231, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_695 = None
	        _assert_tensor_metadata_696 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_154, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_696 = None
	        convert_element_type_462: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_154, torch.int8);  clamp_max_154 = None
	        view_1209: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_231, [sym_size_int, 1500, 1])
	        view_1210: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_462, [sym_size_int, 1500, 1])
	        reciprocal_77: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1209);  view_1209 = None
	        mul_7514: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_77, 1.0);  reciprocal_77 = None
	        mul_7517: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7468, mul_7514);  mul_7468 = mul_7514 = None
	        round_156: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_7517);  mul_7517 = None
	        add_11897: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_156, view_1210);  round_156 = view_1210 = None
	        clamp_min_233: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11897, -128);  add_11897 = None
	        clamp_max_155: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_233, 127);  clamp_min_233 = None
	        _assert_tensor_metadata_697 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_155, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_697 = None
	        convert_element_type_463: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_155, torch.int8);  clamp_max_155 = None
	        view_1213: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_231, [sym_size_int, 1500, 1]);  clamp_min_231 = None
	        view_1214: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_462, [sym_size_int, 1500, 1]);  convert_element_type_462 = None
	        _assert_tensor_metadata_698 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_463, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_698 = None
	        convert_element_type_464: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_463, torch.float32);  convert_element_type_463 = None
	        _assert_tensor_metadata_699 = torch.ops.aten._assert_tensor_metadata.default(view_1214, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_699 = None
	        convert_element_type_465: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1214, torch.float32);  view_1214 = None
	        sub_3559: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_464, convert_element_type_465);  convert_element_type_464 = convert_element_type_465 = None
	        mul_7539: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3559, view_1213);  sub_3559 = view_1213 = None
	        _assert_tensor_metadata_700 = torch.ops.aten._assert_tensor_metadata.default(mul_7539, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_700 = None
	        view_1216: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_12_fc2_parametrizations_weight_original0 = None
	        view_1217: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_12_fc2_parametrizations_weight_original1 = None
	        view_1218: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_12_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_701 = torch.ops.aten._assert_tensor_metadata.default(view_1216, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_701 = None
	        convert_element_type_466: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1216, torch.float32);  view_1216 = None
	        _assert_tensor_metadata_702 = torch.ops.aten._assert_tensor_metadata.default(view_1218, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_702 = None
	        convert_element_type_467: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1218, torch.float32);  view_1218 = None
	        sub_3563: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_466, convert_element_type_467);  convert_element_type_466 = convert_element_type_467 = None
	        mul_7544: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3563, view_1217);  sub_3563 = view_1217 = None
	        view_1219: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7544, [1280, 5120]);  mul_7544 = None
	        _assert_tensor_metadata_703 = torch.ops.aten._assert_tensor_metadata.default(view_1219, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_703 = None
	        mul_7549: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1220: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7539, [mul_7549, 5120]);  mul_7539 = mul_7549 = None
	        permute_130: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1219, [1, 0]);  view_1219 = None
	        addmm_64: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_12_fc2_bias, view_1220, permute_130);  model_audio_tower_layers_12_fc2_bias = view_1220 = permute_130 = None
	        view_1221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_64, [sym_size_int, 1500, 1280]);  addmm_64 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_11960: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_11662, view_1221);  add_11662 = view_1221 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_105: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_11960, memory_format = torch.contiguous_format)
	        var_mean_26 = torch.ops.aten.var_mean.correction(clone_105, [2], correction = 0, keepdim = True)
	        getitem_104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_26[0]
	        getitem_105: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_26[1];  var_mean_26 = None
	        add_11965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_104, 1e-05);  getitem_104 = None
	        rsqrt_26: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_11965);  add_11965 = None
	        sub_3569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_105, getitem_105);  clone_105 = getitem_105 = None
	        mul_7560: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3569, rsqrt_26);  sub_3569 = rsqrt_26 = None
	        mul_7561: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7560, model_audio_tower_layers_13_self_attn_layer_norm_weight);  mul_7560 = model_audio_tower_layers_13_self_attn_layer_norm_weight = None
	        add_11966: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_7561, model_audio_tower_layers_13_self_attn_layer_norm_bias);  mul_7561 = model_audio_tower_layers_13_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11966, [2])
	        amax_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11966, [2])
	        full_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_78, full_156);  amin_78 = full_156 = None
	        full_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_78, full_157);  amax_78 = full_157 = None
	        sub_3580: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_78, minimum_78);  maximum_78 = None
	        div_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3580, 255.0);  sub_3580 = None
	        clamp_min_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_156, 1.1920928955078125e-07);  div_156 = None
	        div_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_78, clamp_min_234);  minimum_78 = None
	        round_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_157);  div_157 = None
	        sub_3586: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_157);  round_157 = None
	        clamp_min_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3586, -128);  sub_3586 = None
	        clamp_max_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_235, 127);  clamp_min_235 = None
	        _assert_tensor_metadata_704 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_234, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_704 = None
	        _assert_tensor_metadata_705 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_156, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_705 = None
	        convert_element_type_468: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_156, torch.int8);  clamp_max_156 = None
	        view_1224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_234, [sym_size_int, 1500, 1])
	        view_1225: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_468, [sym_size_int, 1500, 1])
	        reciprocal_78: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1224);  view_1224 = None
	        mul_7609: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_78, 1.0);  reciprocal_78 = None
	        mul_7612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11966, mul_7609);  mul_7609 = None
	        round_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7612);  mul_7612 = None
	        add_12053: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_158, view_1225);  round_158 = view_1225 = None
	        clamp_min_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12053, -128);  add_12053 = None
	        clamp_max_157: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_236, 127);  clamp_min_236 = None
	        _assert_tensor_metadata_706 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_157, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_706 = None
	        convert_element_type_469: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_157, torch.int8);  clamp_max_157 = None
	        view_1228: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_234, [sym_size_int, 1500, 1]);  clamp_min_234 = None
	        view_1229: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_468, [sym_size_int, 1500, 1]);  convert_element_type_468 = None
	        _assert_tensor_metadata_707 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_469, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_707 = None
	        convert_element_type_470: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_469, torch.float32);  convert_element_type_469 = None
	        _assert_tensor_metadata_708 = torch.ops.aten._assert_tensor_metadata.default(view_1229, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_708 = None
	        convert_element_type_471: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1229, torch.float32);  view_1229 = None
	        sub_3606: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_470, convert_element_type_471);  convert_element_type_470 = convert_element_type_471 = None
	        mul_7634: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3606, view_1228);  sub_3606 = view_1228 = None
	        _assert_tensor_metadata_709 = torch.ops.aten._assert_tensor_metadata.default(mul_7634, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_709 = None
	        view_1231: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1232: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1233: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_710 = torch.ops.aten._assert_tensor_metadata.default(view_1231, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_710 = None
	        convert_element_type_472: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1231, torch.float32);  view_1231 = None
	        _assert_tensor_metadata_711 = torch.ops.aten._assert_tensor_metadata.default(view_1233, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_711 = None
	        convert_element_type_473: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1233, torch.float32);  view_1233 = None
	        sub_3610: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_472, convert_element_type_473);  convert_element_type_472 = convert_element_type_473 = None
	        mul_7639: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3610, view_1232);  sub_3610 = view_1232 = None
	        view_1234: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7639, [1280, 1280]);  mul_7639 = None
	        _assert_tensor_metadata_712 = torch.ops.aten._assert_tensor_metadata.default(view_1234, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_712 = None
	        mul_7644: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1235: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7634, [mul_7644, 1280]);  mul_7634 = mul_7644 = None
	        permute_131: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1234, [1, 0]);  view_1234 = None
	        addmm_65: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_13_self_attn_q_proj_bias, view_1235, permute_131);  model_audio_tower_layers_13_self_attn_q_proj_bias = view_1235 = permute_131 = None
	        view_1236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_65, [sym_size_int, 1500, 1280]);  addmm_65 = None
	        mul_7651: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1236, 0.125);  view_1236 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1237: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7651, [sym_size_int, 1500, 20, 64]);  mul_7651 = None
	        permute_132: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1237, [0, 2, 1, 3]);  view_1237 = None
	        clone_106: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_132, memory_format = torch.contiguous_format);  permute_132 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11966, [2])
	        amax_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11966, [2])
	        full_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_79, full_158);  amin_79 = full_158 = None
	        full_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_79, full_159);  amax_79 = full_159 = None
	        sub_3625: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_79, minimum_79);  maximum_79 = None
	        div_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3625, 255.0);  sub_3625 = None
	        clamp_min_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_158, 1.1920928955078125e-07);  div_158 = None
	        div_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_79, clamp_min_237);  minimum_79 = None
	        round_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_159);  div_159 = None
	        sub_3631: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_159);  round_159 = None
	        clamp_min_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3631, -128);  sub_3631 = None
	        clamp_max_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_238, 127);  clamp_min_238 = None
	        _assert_tensor_metadata_713 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_237, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_713 = None
	        _assert_tensor_metadata_714 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_158, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_714 = None
	        convert_element_type_474: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_158, torch.int8);  clamp_max_158 = None
	        view_1240: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_237, [sym_size_int, 1500, 1])
	        view_1241: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_474, [sym_size_int, 1500, 1])
	        reciprocal_79: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1240);  view_1240 = None
	        mul_7705: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_79, 1.0);  reciprocal_79 = None
	        mul_7708: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11966, mul_7705);  mul_7705 = None
	        round_160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7708);  mul_7708 = None
	        add_12205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_160, view_1241);  round_160 = view_1241 = None
	        clamp_min_239: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12205, -128);  add_12205 = None
	        clamp_max_159: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_239, 127);  clamp_min_239 = None
	        _assert_tensor_metadata_715 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_159, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_715 = None
	        convert_element_type_475: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_159, torch.int8);  clamp_max_159 = None
	        view_1244: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_237, [sym_size_int, 1500, 1]);  clamp_min_237 = None
	        view_1245: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_474, [sym_size_int, 1500, 1]);  convert_element_type_474 = None
	        _assert_tensor_metadata_716 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_475, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_716 = None
	        convert_element_type_476: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_475, torch.float32);  convert_element_type_475 = None
	        _assert_tensor_metadata_717 = torch.ops.aten._assert_tensor_metadata.default(view_1245, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_717 = None
	        convert_element_type_477: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1245, torch.float32);  view_1245 = None
	        sub_3651: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_476, convert_element_type_477);  convert_element_type_476 = convert_element_type_477 = None
	        mul_7730: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3651, view_1244);  sub_3651 = view_1244 = None
	        _assert_tensor_metadata_718 = torch.ops.aten._assert_tensor_metadata.default(mul_7730, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_718 = None
	        view_1247: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1248: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1249: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_719 = torch.ops.aten._assert_tensor_metadata.default(view_1247, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_719 = None
	        convert_element_type_478: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1247, torch.float32);  view_1247 = None
	        _assert_tensor_metadata_720 = torch.ops.aten._assert_tensor_metadata.default(view_1249, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_720 = None
	        convert_element_type_479: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1249, torch.float32);  view_1249 = None
	        sub_3655: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_478, convert_element_type_479);  convert_element_type_478 = convert_element_type_479 = None
	        mul_7735: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3655, view_1248);  sub_3655 = view_1248 = None
	        view_1250: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7735, [1280, 1280]);  mul_7735 = None
	        _assert_tensor_metadata_721 = torch.ops.aten._assert_tensor_metadata.default(view_1250, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_721 = None
	        permute_133: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1250, [1, 0]);  view_1250 = None
	        mul_7738: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1251: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7730, [mul_7738, 1280]);  mul_7730 = mul_7738 = None
	        mm_13: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1251, permute_133);  view_1251 = permute_133 = None
	        view_1252: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_13, [sym_size_int, 1500, 1280]);  mm_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1253: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1252, [sym_size_int, -1, 20, 64]);  view_1252 = None
	        permute_134: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1253, [0, 2, 1, 3]);  view_1253 = None
	        clone_107: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_134, memory_format = torch.contiguous_format);  permute_134 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11966, [2])
	        amax_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11966, [2])
	        full_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_80, full_160);  amin_80 = full_160 = None
	        full_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_80, full_161);  amax_80 = full_161 = None
	        sub_3669: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_80, minimum_80);  maximum_80 = None
	        div_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3669, 255.0);  sub_3669 = None
	        clamp_min_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_160, 1.1920928955078125e-07);  div_160 = None
	        div_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_80, clamp_min_240);  minimum_80 = None
	        round_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_161);  div_161 = None
	        sub_3675: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_161);  round_161 = None
	        clamp_min_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3675, -128);  sub_3675 = None
	        clamp_max_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_241, 127);  clamp_min_241 = None
	        _assert_tensor_metadata_722 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_240, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_722 = None
	        _assert_tensor_metadata_723 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_160, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_723 = None
	        convert_element_type_480: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_160, torch.int8);  clamp_max_160 = None
	        view_1256: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_240, [sym_size_int, 1500, 1])
	        view_1257: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_480, [sym_size_int, 1500, 1])
	        reciprocal_80: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1256);  view_1256 = None
	        mul_7804: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_80, 1.0);  reciprocal_80 = None
	        mul_7807: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11966, mul_7804);  add_11966 = mul_7804 = None
	        round_162: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7807);  mul_7807 = None
	        add_12353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_162, view_1257);  round_162 = view_1257 = None
	        clamp_min_242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12353, -128);  add_12353 = None
	        clamp_max_161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_242, 127);  clamp_min_242 = None
	        _assert_tensor_metadata_724 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_161, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_724 = None
	        convert_element_type_481: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_161, torch.int8);  clamp_max_161 = None
	        view_1260: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_240, [sym_size_int, 1500, 1]);  clamp_min_240 = None
	        view_1261: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_480, [sym_size_int, 1500, 1]);  convert_element_type_480 = None
	        _assert_tensor_metadata_725 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_481, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_725 = None
	        convert_element_type_482: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_481, torch.float32);  convert_element_type_481 = None
	        _assert_tensor_metadata_726 = torch.ops.aten._assert_tensor_metadata.default(view_1261, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_726 = None
	        convert_element_type_483: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1261, torch.float32);  view_1261 = None
	        sub_3695: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_482, convert_element_type_483);  convert_element_type_482 = convert_element_type_483 = None
	        mul_7829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3695, view_1260);  sub_3695 = view_1260 = None
	        _assert_tensor_metadata_727 = torch.ops.aten._assert_tensor_metadata.default(mul_7829, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_727 = None
	        view_1263: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1264: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1265: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_728 = torch.ops.aten._assert_tensor_metadata.default(view_1263, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_728 = None
	        convert_element_type_484: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1263, torch.float32);  view_1263 = None
	        _assert_tensor_metadata_729 = torch.ops.aten._assert_tensor_metadata.default(view_1265, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_729 = None
	        convert_element_type_485: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1265, torch.float32);  view_1265 = None
	        sub_3699: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_484, convert_element_type_485);  convert_element_type_484 = convert_element_type_485 = None
	        mul_7834: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3699, view_1264);  sub_3699 = view_1264 = None
	        view_1266: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7834, [1280, 1280]);  mul_7834 = None
	        _assert_tensor_metadata_730 = torch.ops.aten._assert_tensor_metadata.default(view_1266, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_730 = None
	        mul_7839: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1267: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7829, [mul_7839, 1280]);  mul_7829 = mul_7839 = None
	        permute_135: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1266, [1, 0]);  view_1266 = None
	        addmm_66: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_13_self_attn_v_proj_bias, view_1267, permute_135);  model_audio_tower_layers_13_self_attn_v_proj_bias = view_1267 = permute_135 = None
	        view_1268: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_66, [sym_size_int, 1500, 1280]);  addmm_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1269: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1268, [sym_size_int, -1, 20, 64]);  view_1268 = None
	        permute_136: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1269, [0, 2, 1, 3]);  view_1269 = None
	        clone_108: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_136, memory_format = torch.contiguous_format);  permute_136 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_13 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_106, clone_107, clone_108, None, False, scale = 1.0);  clone_106 = clone_107 = clone_108 = None
	        getitem_106: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_13[0];  _scaled_dot_product_efficient_attention_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_137: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_106, [0, 2, 1, 3]);  getitem_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_137, [sym_size_int, 1500, -1]);  permute_137 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1270, [2])
	        amax_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1270, [2])
	        full_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_81, full_162);  amin_81 = full_162 = None
	        full_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_81, full_163);  amax_81 = full_163 = None
	        sub_3717: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_81, minimum_81);  maximum_81 = None
	        div_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3717, 255.0);  sub_3717 = None
	        clamp_min_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_162, 1.1920928955078125e-07);  div_162 = None
	        div_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_81, clamp_min_243);  minimum_81 = None
	        round_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_163);  div_163 = None
	        sub_3723: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_163);  round_163 = None
	        clamp_min_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3723, -128);  sub_3723 = None
	        clamp_max_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_244, 127);  clamp_min_244 = None
	        _assert_tensor_metadata_731 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_243, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_731 = None
	        _assert_tensor_metadata_732 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_162, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_732 = None
	        convert_element_type_486: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_162, torch.int8);  clamp_max_162 = None
	        view_1273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_243, [sym_size_int, 1500, 1])
	        view_1274: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_486, [sym_size_int, 1500, 1])
	        reciprocal_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1273);  view_1273 = None
	        mul_7909: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_81, 1.0);  reciprocal_81 = None
	        mul_7912: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1270, mul_7909);  view_1270 = mul_7909 = None
	        round_164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7912);  mul_7912 = None
	        add_12517: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_164, view_1274);  round_164 = view_1274 = None
	        clamp_min_245: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12517, -128);  add_12517 = None
	        clamp_max_163: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_245, 127);  clamp_min_245 = None
	        _assert_tensor_metadata_733 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_163, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_733 = None
	        convert_element_type_487: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_163, torch.int8);  clamp_max_163 = None
	        view_1277: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_243, [sym_size_int, 1500, 1]);  clamp_min_243 = None
	        view_1278: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_486, [sym_size_int, 1500, 1]);  convert_element_type_486 = None
	        _assert_tensor_metadata_734 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_487, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_734 = None
	        convert_element_type_488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_487, torch.float32);  convert_element_type_487 = None
	        _assert_tensor_metadata_735 = torch.ops.aten._assert_tensor_metadata.default(view_1278, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_735 = None
	        convert_element_type_489: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1278, torch.float32);  view_1278 = None
	        sub_3743: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_488, convert_element_type_489);  convert_element_type_488 = convert_element_type_489 = None
	        mul_7934: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3743, view_1277);  sub_3743 = view_1277 = None
	        _assert_tensor_metadata_736 = torch.ops.aten._assert_tensor_metadata.default(mul_7934, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_736 = None
	        view_1280: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1281: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1282: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_737 = torch.ops.aten._assert_tensor_metadata.default(view_1280, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_737 = None
	        convert_element_type_490: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1280, torch.float32);  view_1280 = None
	        _assert_tensor_metadata_738 = torch.ops.aten._assert_tensor_metadata.default(view_1282, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_738 = None
	        convert_element_type_491: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1282, torch.float32);  view_1282 = None
	        sub_3747: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_490, convert_element_type_491);  convert_element_type_490 = convert_element_type_491 = None
	        mul_7939: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3747, view_1281);  sub_3747 = view_1281 = None
	        view_1283: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7939, [1280, 1280]);  mul_7939 = None
	        _assert_tensor_metadata_739 = torch.ops.aten._assert_tensor_metadata.default(view_1283, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_739 = None
	        mul_7944: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1284: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7934, [mul_7944, 1280]);  mul_7934 = mul_7944 = None
	        permute_138: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1283, [1, 0]);  view_1283 = None
	        addmm_67: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_13_self_attn_out_proj_bias, view_1284, permute_138);  model_audio_tower_layers_13_self_attn_out_proj_bias = view_1284 = permute_138 = None
	        view_1285: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_67, [sym_size_int, 1500, 1280]);  addmm_67 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_12580: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_11960, view_1285);  add_11960 = view_1285 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_12580, memory_format = torch.contiguous_format)
	        var_mean_27 = torch.ops.aten.var_mean.correction(clone_110, [2], correction = 0, keepdim = True)
	        getitem_110: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_27[0]
	        getitem_111: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_27[1];  var_mean_27 = None
	        add_12585: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_110, 1e-05);  getitem_110 = None
	        rsqrt_27: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_12585);  add_12585 = None
	        sub_3753: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_110, getitem_111);  clone_110 = getitem_111 = None
	        mul_7955: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3753, rsqrt_27);  sub_3753 = rsqrt_27 = None
	        mul_7956: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7955, model_audio_tower_layers_13_final_layer_norm_weight);  mul_7955 = model_audio_tower_layers_13_final_layer_norm_weight = None
	        add_12586: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_7956, model_audio_tower_layers_13_final_layer_norm_bias);  mul_7956 = model_audio_tower_layers_13_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_12586, [2])
	        amax_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_12586, [2])
	        full_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_82, full_164);  amin_82 = full_164 = None
	        full_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_82, full_165);  amax_82 = full_165 = None
	        sub_3764: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_82, minimum_82);  maximum_82 = None
	        div_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3764, 255.0);  sub_3764 = None
	        clamp_min_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_164, 1.1920928955078125e-07);  div_164 = None
	        div_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_82, clamp_min_246);  minimum_82 = None
	        round_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_165);  div_165 = None
	        sub_3770: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_165);  round_165 = None
	        clamp_min_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3770, -128);  sub_3770 = None
	        clamp_max_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_247, 127);  clamp_min_247 = None
	        _assert_tensor_metadata_740 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_246, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_740 = None
	        _assert_tensor_metadata_741 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_164, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_741 = None
	        convert_element_type_492: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_164, torch.int8);  clamp_max_164 = None
	        view_1288: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_246, [sym_size_int, 1500, 1])
	        view_1289: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_492, [sym_size_int, 1500, 1])
	        reciprocal_82: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1288);  view_1288 = None
	        mul_8004: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_82, 1.0);  reciprocal_82 = None
	        mul_8007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_12586, mul_8004);  add_12586 = mul_8004 = None
	        round_166: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8007);  mul_8007 = None
	        add_12673: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_166, view_1289);  round_166 = view_1289 = None
	        clamp_min_248: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12673, -128);  add_12673 = None
	        clamp_max_165: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_248, 127);  clamp_min_248 = None
	        _assert_tensor_metadata_742 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_165, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_742 = None
	        convert_element_type_493: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_165, torch.int8);  clamp_max_165 = None
	        view_1292: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_246, [sym_size_int, 1500, 1]);  clamp_min_246 = None
	        view_1293: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_492, [sym_size_int, 1500, 1]);  convert_element_type_492 = None
	        _assert_tensor_metadata_743 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_493, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_743 = None
	        convert_element_type_494: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_493, torch.float32);  convert_element_type_493 = None
	        _assert_tensor_metadata_744 = torch.ops.aten._assert_tensor_metadata.default(view_1293, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_744 = None
	        convert_element_type_495: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1293, torch.float32);  view_1293 = None
	        sub_3790: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_494, convert_element_type_495);  convert_element_type_494 = convert_element_type_495 = None
	        mul_8029: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3790, view_1292);  sub_3790 = view_1292 = None
	        _assert_tensor_metadata_745 = torch.ops.aten._assert_tensor_metadata.default(mul_8029, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_745 = None
	        view_1295: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_13_fc1_parametrizations_weight_original0 = None
	        view_1296: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_13_fc1_parametrizations_weight_original1 = None
	        view_1297: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_13_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_746 = torch.ops.aten._assert_tensor_metadata.default(view_1295, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_746 = None
	        convert_element_type_496: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1295, torch.float32);  view_1295 = None
	        _assert_tensor_metadata_747 = torch.ops.aten._assert_tensor_metadata.default(view_1297, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_747 = None
	        convert_element_type_497: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1297, torch.float32);  view_1297 = None
	        sub_3794: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_496, convert_element_type_497);  convert_element_type_496 = convert_element_type_497 = None
	        mul_8034: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3794, view_1296);  sub_3794 = view_1296 = None
	        view_1298: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8034, [5120, 1280]);  mul_8034 = None
	        _assert_tensor_metadata_748 = torch.ops.aten._assert_tensor_metadata.default(view_1298, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_748 = None
	        mul_8039: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1299: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8029, [mul_8039, 1280]);  mul_8029 = mul_8039 = None
	        permute_139: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1298, [1, 0]);  view_1298 = None
	        addmm_68: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_13_fc1_bias, view_1299, permute_139);  model_audio_tower_layers_13_fc1_bias = view_1299 = permute_139 = None
	        view_1300: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_68, [sym_size_int, 1500, 5120]);  addmm_68 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_8046: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1300, 0.5)
	        mul_8047: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1300, 0.7071067811865476);  view_1300 = None
	        erf_15: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_8047);  mul_8047 = None
	        add_12732: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_15, 1);  erf_15 = None
	        mul_8048: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8046, add_12732);  mul_8046 = add_12732 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_8048, [2])
	        amax_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_8048, [2])
	        full_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_83, full_166);  amin_83 = full_166 = None
	        full_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_83, full_167);  amax_83 = full_167 = None
	        sub_3807: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_83, minimum_83);  maximum_83 = None
	        div_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3807, 255.0);  sub_3807 = None
	        clamp_min_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_166, 1.1920928955078125e-07);  div_166 = None
	        div_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_83, clamp_min_249);  minimum_83 = None
	        round_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_167);  div_167 = None
	        sub_3813: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_167);  round_167 = None
	        clamp_min_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3813, -128);  sub_3813 = None
	        clamp_max_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_250, 127);  clamp_min_250 = None
	        _assert_tensor_metadata_749 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_749 = None
	        _assert_tensor_metadata_750 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_166, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_750 = None
	        convert_element_type_498: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_166, torch.int8);  clamp_max_166 = None
	        view_1303: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_249, [sym_size_int, 1500, 1])
	        view_1304: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_498, [sym_size_int, 1500, 1])
	        reciprocal_83: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1303);  view_1303 = None
	        mul_8094: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_83, 1.0);  reciprocal_83 = None
	        mul_8097: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8048, mul_8094);  mul_8048 = mul_8094 = None
	        round_168: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_8097);  mul_8097 = None
	        add_12815: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_168, view_1304);  round_168 = view_1304 = None
	        clamp_min_251: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12815, -128);  add_12815 = None
	        clamp_max_167: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_251, 127);  clamp_min_251 = None
	        _assert_tensor_metadata_751 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_167, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_751 = None
	        convert_element_type_499: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_167, torch.int8);  clamp_max_167 = None
	        view_1307: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_249, [sym_size_int, 1500, 1]);  clamp_min_249 = None
	        view_1308: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_498, [sym_size_int, 1500, 1]);  convert_element_type_498 = None
	        _assert_tensor_metadata_752 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_499, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_752 = None
	        convert_element_type_500: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_499, torch.float32);  convert_element_type_499 = None
	        _assert_tensor_metadata_753 = torch.ops.aten._assert_tensor_metadata.default(view_1308, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_753 = None
	        convert_element_type_501: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1308, torch.float32);  view_1308 = None
	        sub_3833: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_500, convert_element_type_501);  convert_element_type_500 = convert_element_type_501 = None
	        mul_8119: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3833, view_1307);  sub_3833 = view_1307 = None
	        _assert_tensor_metadata_754 = torch.ops.aten._assert_tensor_metadata.default(mul_8119, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_754 = None
	        view_1310: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_13_fc2_parametrizations_weight_original0 = None
	        view_1311: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_13_fc2_parametrizations_weight_original1 = None
	        view_1312: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_13_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_755 = torch.ops.aten._assert_tensor_metadata.default(view_1310, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_755 = None
	        convert_element_type_502: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1310, torch.float32);  view_1310 = None
	        _assert_tensor_metadata_756 = torch.ops.aten._assert_tensor_metadata.default(view_1312, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_756 = None
	        convert_element_type_503: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1312, torch.float32);  view_1312 = None
	        sub_3837: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_502, convert_element_type_503);  convert_element_type_502 = convert_element_type_503 = None
	        mul_8124: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3837, view_1311);  sub_3837 = view_1311 = None
	        view_1313: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8124, [1280, 5120]);  mul_8124 = None
	        _assert_tensor_metadata_757 = torch.ops.aten._assert_tensor_metadata.default(view_1313, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_757 = None
	        mul_8129: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1314: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8119, [mul_8129, 5120]);  mul_8119 = mul_8129 = None
	        permute_140: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1313, [1, 0]);  view_1313 = None
	        addmm_69: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_13_fc2_bias, view_1314, permute_140);  model_audio_tower_layers_13_fc2_bias = view_1314 = permute_140 = None
	        view_1315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_69, [sym_size_int, 1500, 1280]);  addmm_69 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_12878: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_12580, view_1315);  add_12580 = view_1315 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_12878, memory_format = torch.contiguous_format)
	        var_mean_28 = torch.ops.aten.var_mean.correction(clone_113, [2], correction = 0, keepdim = True)
	        getitem_112: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_28[0]
	        getitem_113: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_28[1];  var_mean_28 = None
	        add_12883: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_112, 1e-05);  getitem_112 = None
	        rsqrt_28: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_12883);  add_12883 = None
	        sub_3843: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_113, getitem_113);  clone_113 = getitem_113 = None
	        mul_8140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3843, rsqrt_28);  sub_3843 = rsqrt_28 = None
	        mul_8141: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8140, model_audio_tower_layers_14_self_attn_layer_norm_weight);  mul_8140 = model_audio_tower_layers_14_self_attn_layer_norm_weight = None
	        add_12884: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_8141, model_audio_tower_layers_14_self_attn_layer_norm_bias);  mul_8141 = model_audio_tower_layers_14_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_12884, [2])
	        amax_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_12884, [2])
	        full_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_84, full_168);  amin_84 = full_168 = None
	        full_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_84, full_169);  amax_84 = full_169 = None
	        sub_3854: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_84, minimum_84);  maximum_84 = None
	        div_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3854, 255.0);  sub_3854 = None
	        clamp_min_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_168, 1.1920928955078125e-07);  div_168 = None
	        div_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_84, clamp_min_252);  minimum_84 = None
	        round_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_169);  div_169 = None
	        sub_3860: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_169);  round_169 = None
	        clamp_min_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3860, -128);  sub_3860 = None
	        clamp_max_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_253, 127);  clamp_min_253 = None
	        _assert_tensor_metadata_758 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_252, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_758 = None
	        _assert_tensor_metadata_759 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_168, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_759 = None
	        convert_element_type_504: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_168, torch.int8);  clamp_max_168 = None
	        view_1318: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_252, [sym_size_int, 1500, 1])
	        view_1319: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_504, [sym_size_int, 1500, 1])
	        reciprocal_84: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1318);  view_1318 = None
	        mul_8189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_84, 1.0);  reciprocal_84 = None
	        mul_8192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_12884, mul_8189);  mul_8189 = None
	        round_170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8192);  mul_8192 = None
	        add_12971: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_170, view_1319);  round_170 = view_1319 = None
	        clamp_min_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12971, -128);  add_12971 = None
	        clamp_max_169: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_254, 127);  clamp_min_254 = None
	        _assert_tensor_metadata_760 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_169, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_760 = None
	        convert_element_type_505: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_169, torch.int8);  clamp_max_169 = None
	        view_1322: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_252, [sym_size_int, 1500, 1]);  clamp_min_252 = None
	        view_1323: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_504, [sym_size_int, 1500, 1]);  convert_element_type_504 = None
	        _assert_tensor_metadata_761 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_505, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_761 = None
	        convert_element_type_506: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_505, torch.float32);  convert_element_type_505 = None
	        _assert_tensor_metadata_762 = torch.ops.aten._assert_tensor_metadata.default(view_1323, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_762 = None
	        convert_element_type_507: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1323, torch.float32);  view_1323 = None
	        sub_3880: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_506, convert_element_type_507);  convert_element_type_506 = convert_element_type_507 = None
	        mul_8214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3880, view_1322);  sub_3880 = view_1322 = None
	        _assert_tensor_metadata_763 = torch.ops.aten._assert_tensor_metadata.default(mul_8214, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_763 = None
	        view_1325: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1326: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1327: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_764 = torch.ops.aten._assert_tensor_metadata.default(view_1325, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_764 = None
	        convert_element_type_508: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1325, torch.float32);  view_1325 = None
	        _assert_tensor_metadata_765 = torch.ops.aten._assert_tensor_metadata.default(view_1327, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_765 = None
	        convert_element_type_509: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1327, torch.float32);  view_1327 = None
	        sub_3884: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_508, convert_element_type_509);  convert_element_type_508 = convert_element_type_509 = None
	        mul_8219: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3884, view_1326);  sub_3884 = view_1326 = None
	        view_1328: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8219, [1280, 1280]);  mul_8219 = None
	        _assert_tensor_metadata_766 = torch.ops.aten._assert_tensor_metadata.default(view_1328, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_766 = None
	        mul_8224: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1329: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8214, [mul_8224, 1280]);  mul_8214 = mul_8224 = None
	        permute_141: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1328, [1, 0]);  view_1328 = None
	        addmm_70: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_14_self_attn_q_proj_bias, view_1329, permute_141);  model_audio_tower_layers_14_self_attn_q_proj_bias = view_1329 = permute_141 = None
	        view_1330: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_70, [sym_size_int, 1500, 1280]);  addmm_70 = None
	        mul_8231: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1330, 0.125);  view_1330 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1331: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8231, [sym_size_int, 1500, 20, 64]);  mul_8231 = None
	        permute_142: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1331, [0, 2, 1, 3]);  view_1331 = None
	        clone_114: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_142, memory_format = torch.contiguous_format);  permute_142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_12884, [2])
	        amax_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_12884, [2])
	        full_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_85, full_170);  amin_85 = full_170 = None
	        full_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_85, full_171);  amax_85 = full_171 = None
	        sub_3899: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_85, minimum_85);  maximum_85 = None
	        div_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3899, 255.0);  sub_3899 = None
	        clamp_min_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_170, 1.1920928955078125e-07);  div_170 = None
	        div_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_85, clamp_min_255);  minimum_85 = None
	        round_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_171);  div_171 = None
	        sub_3905: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_171);  round_171 = None
	        clamp_min_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3905, -128);  sub_3905 = None
	        clamp_max_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_256, 127);  clamp_min_256 = None
	        _assert_tensor_metadata_767 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_255, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_767 = None
	        _assert_tensor_metadata_768 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_170, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_768 = None
	        convert_element_type_510: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_170, torch.int8);  clamp_max_170 = None
	        view_1334: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_255, [sym_size_int, 1500, 1])
	        view_1335: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_510, [sym_size_int, 1500, 1])
	        reciprocal_85: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1334);  view_1334 = None
	        mul_8285: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_85, 1.0);  reciprocal_85 = None
	        mul_8288: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_12884, mul_8285);  mul_8285 = None
	        round_172: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8288);  mul_8288 = None
	        add_13123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_172, view_1335);  round_172 = view_1335 = None
	        clamp_min_257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13123, -128);  add_13123 = None
	        clamp_max_171: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_257, 127);  clamp_min_257 = None
	        _assert_tensor_metadata_769 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_171, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_769 = None
	        convert_element_type_511: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_171, torch.int8);  clamp_max_171 = None
	        view_1338: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_255, [sym_size_int, 1500, 1]);  clamp_min_255 = None
	        view_1339: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_510, [sym_size_int, 1500, 1]);  convert_element_type_510 = None
	        _assert_tensor_metadata_770 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_511, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_770 = None
	        convert_element_type_512: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_511, torch.float32);  convert_element_type_511 = None
	        _assert_tensor_metadata_771 = torch.ops.aten._assert_tensor_metadata.default(view_1339, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_771 = None
	        convert_element_type_513: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1339, torch.float32);  view_1339 = None
	        sub_3925: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_512, convert_element_type_513);  convert_element_type_512 = convert_element_type_513 = None
	        mul_8310: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3925, view_1338);  sub_3925 = view_1338 = None
	        _assert_tensor_metadata_772 = torch.ops.aten._assert_tensor_metadata.default(mul_8310, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_772 = None
	        view_1341: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1342: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1343: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_773 = torch.ops.aten._assert_tensor_metadata.default(view_1341, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_773 = None
	        convert_element_type_514: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1341, torch.float32);  view_1341 = None
	        _assert_tensor_metadata_774 = torch.ops.aten._assert_tensor_metadata.default(view_1343, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_774 = None
	        convert_element_type_515: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1343, torch.float32);  view_1343 = None
	        sub_3929: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_514, convert_element_type_515);  convert_element_type_514 = convert_element_type_515 = None
	        mul_8315: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3929, view_1342);  sub_3929 = view_1342 = None
	        view_1344: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8315, [1280, 1280]);  mul_8315 = None
	        _assert_tensor_metadata_775 = torch.ops.aten._assert_tensor_metadata.default(view_1344, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_775 = None
	        permute_143: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1344, [1, 0]);  view_1344 = None
	        mul_8318: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1345: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8310, [mul_8318, 1280]);  mul_8310 = mul_8318 = None
	        mm_14: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1345, permute_143);  view_1345 = permute_143 = None
	        view_1346: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_14, [sym_size_int, 1500, 1280]);  mm_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1347: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1346, [sym_size_int, -1, 20, 64]);  view_1346 = None
	        permute_144: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1347, [0, 2, 1, 3]);  view_1347 = None
	        clone_115: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_144, memory_format = torch.contiguous_format);  permute_144 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_12884, [2])
	        amax_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_12884, [2])
	        full_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_86, full_172);  amin_86 = full_172 = None
	        full_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_86, full_173);  amax_86 = full_173 = None
	        sub_3943: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_86, minimum_86);  maximum_86 = None
	        div_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3943, 255.0);  sub_3943 = None
	        clamp_min_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_172, 1.1920928955078125e-07);  div_172 = None
	        div_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_86, clamp_min_258);  minimum_86 = None
	        round_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_173);  div_173 = None
	        sub_3949: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_173);  round_173 = None
	        clamp_min_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3949, -128);  sub_3949 = None
	        clamp_max_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_259, 127);  clamp_min_259 = None
	        _assert_tensor_metadata_776 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_258, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_776 = None
	        _assert_tensor_metadata_777 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_172, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_777 = None
	        convert_element_type_516: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_172, torch.int8);  clamp_max_172 = None
	        view_1350: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_258, [sym_size_int, 1500, 1])
	        view_1351: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_516, [sym_size_int, 1500, 1])
	        reciprocal_86: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1350);  view_1350 = None
	        mul_8384: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_86, 1.0);  reciprocal_86 = None
	        mul_8387: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_12884, mul_8384);  add_12884 = mul_8384 = None
	        round_174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8387);  mul_8387 = None
	        add_13271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_174, view_1351);  round_174 = view_1351 = None
	        clamp_min_260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13271, -128);  add_13271 = None
	        clamp_max_173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_260, 127);  clamp_min_260 = None
	        _assert_tensor_metadata_778 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_173, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_778 = None
	        convert_element_type_517: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_173, torch.int8);  clamp_max_173 = None
	        view_1354: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_258, [sym_size_int, 1500, 1]);  clamp_min_258 = None
	        view_1355: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_516, [sym_size_int, 1500, 1]);  convert_element_type_516 = None
	        _assert_tensor_metadata_779 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_517, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_779 = None
	        convert_element_type_518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_517, torch.float32);  convert_element_type_517 = None
	        _assert_tensor_metadata_780 = torch.ops.aten._assert_tensor_metadata.default(view_1355, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_780 = None
	        convert_element_type_519: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1355, torch.float32);  view_1355 = None
	        sub_3969: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_518, convert_element_type_519);  convert_element_type_518 = convert_element_type_519 = None
	        mul_8409: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3969, view_1354);  sub_3969 = view_1354 = None
	        _assert_tensor_metadata_781 = torch.ops.aten._assert_tensor_metadata.default(mul_8409, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_781 = None
	        view_1357: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1358: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1359: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_782 = torch.ops.aten._assert_tensor_metadata.default(view_1357, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_782 = None
	        convert_element_type_520: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1357, torch.float32);  view_1357 = None
	        _assert_tensor_metadata_783 = torch.ops.aten._assert_tensor_metadata.default(view_1359, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_783 = None
	        convert_element_type_521: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1359, torch.float32);  view_1359 = None
	        sub_3973: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_520, convert_element_type_521);  convert_element_type_520 = convert_element_type_521 = None
	        mul_8414: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3973, view_1358);  sub_3973 = view_1358 = None
	        view_1360: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8414, [1280, 1280]);  mul_8414 = None
	        _assert_tensor_metadata_784 = torch.ops.aten._assert_tensor_metadata.default(view_1360, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_784 = None
	        mul_8419: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1361: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8409, [mul_8419, 1280]);  mul_8409 = mul_8419 = None
	        permute_145: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1360, [1, 0]);  view_1360 = None
	        addmm_71: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_14_self_attn_v_proj_bias, view_1361, permute_145);  model_audio_tower_layers_14_self_attn_v_proj_bias = view_1361 = permute_145 = None
	        view_1362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_71, [sym_size_int, 1500, 1280]);  addmm_71 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1363: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1362, [sym_size_int, -1, 20, 64]);  view_1362 = None
	        permute_146: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1363, [0, 2, 1, 3]);  view_1363 = None
	        clone_116: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_146, memory_format = torch.contiguous_format);  permute_146 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_14 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_114, clone_115, clone_116, None, False, scale = 1.0);  clone_114 = clone_115 = clone_116 = None
	        getitem_114: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_14[0];  _scaled_dot_product_efficient_attention_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_147: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_114, [0, 2, 1, 3]);  getitem_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1364: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_147, [sym_size_int, 1500, -1]);  permute_147 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1364, [2])
	        amax_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1364, [2])
	        full_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_87, full_174);  amin_87 = full_174 = None
	        full_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_87, full_175);  amax_87 = full_175 = None
	        sub_3991: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_87, minimum_87);  maximum_87 = None
	        div_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3991, 255.0);  sub_3991 = None
	        clamp_min_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_174, 1.1920928955078125e-07);  div_174 = None
	        div_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_87, clamp_min_261);  minimum_87 = None
	        round_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_175);  div_175 = None
	        sub_3997: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_175);  round_175 = None
	        clamp_min_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3997, -128);  sub_3997 = None
	        clamp_max_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_262, 127);  clamp_min_262 = None
	        _assert_tensor_metadata_785 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_261, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_785 = None
	        _assert_tensor_metadata_786 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_174, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_786 = None
	        convert_element_type_522: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_174, torch.int8);  clamp_max_174 = None
	        view_1367: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_261, [sym_size_int, 1500, 1])
	        view_1368: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_522, [sym_size_int, 1500, 1])
	        reciprocal_87: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1367);  view_1367 = None
	        mul_8489: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_87, 1.0);  reciprocal_87 = None
	        mul_8492: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1364, mul_8489);  view_1364 = mul_8489 = None
	        round_176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8492);  mul_8492 = None
	        add_13435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_176, view_1368);  round_176 = view_1368 = None
	        clamp_min_263: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13435, -128);  add_13435 = None
	        clamp_max_175: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_263, 127);  clamp_min_263 = None
	        _assert_tensor_metadata_787 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_175, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_787 = None
	        convert_element_type_523: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_175, torch.int8);  clamp_max_175 = None
	        view_1371: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_261, [sym_size_int, 1500, 1]);  clamp_min_261 = None
	        view_1372: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_522, [sym_size_int, 1500, 1]);  convert_element_type_522 = None
	        _assert_tensor_metadata_788 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_523, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_788 = None
	        convert_element_type_524: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_523, torch.float32);  convert_element_type_523 = None
	        _assert_tensor_metadata_789 = torch.ops.aten._assert_tensor_metadata.default(view_1372, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_789 = None
	        convert_element_type_525: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1372, torch.float32);  view_1372 = None
	        sub_4017: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_524, convert_element_type_525);  convert_element_type_524 = convert_element_type_525 = None
	        mul_8514: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4017, view_1371);  sub_4017 = view_1371 = None
	        _assert_tensor_metadata_790 = torch.ops.aten._assert_tensor_metadata.default(mul_8514, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_790 = None
	        view_1374: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1375: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1376: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_791 = torch.ops.aten._assert_tensor_metadata.default(view_1374, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_791 = None
	        convert_element_type_526: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1374, torch.float32);  view_1374 = None
	        _assert_tensor_metadata_792 = torch.ops.aten._assert_tensor_metadata.default(view_1376, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_792 = None
	        convert_element_type_527: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1376, torch.float32);  view_1376 = None
	        sub_4021: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_526, convert_element_type_527);  convert_element_type_526 = convert_element_type_527 = None
	        mul_8519: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4021, view_1375);  sub_4021 = view_1375 = None
	        view_1377: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8519, [1280, 1280]);  mul_8519 = None
	        _assert_tensor_metadata_793 = torch.ops.aten._assert_tensor_metadata.default(view_1377, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_793 = None
	        mul_8524: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1378: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8514, [mul_8524, 1280]);  mul_8514 = mul_8524 = None
	        permute_148: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1377, [1, 0]);  view_1377 = None
	        addmm_72: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_14_self_attn_out_proj_bias, view_1378, permute_148);  model_audio_tower_layers_14_self_attn_out_proj_bias = view_1378 = permute_148 = None
	        view_1379: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_72, [sym_size_int, 1500, 1280]);  addmm_72 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_13498: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_12878, view_1379);  add_12878 = view_1379 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_118: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_13498, memory_format = torch.contiguous_format)
	        var_mean_29 = torch.ops.aten.var_mean.correction(clone_118, [2], correction = 0, keepdim = True)
	        getitem_118: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_29[0]
	        getitem_119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_29[1];  var_mean_29 = None
	        add_13503: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_118, 1e-05);  getitem_118 = None
	        rsqrt_29: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_13503);  add_13503 = None
	        sub_4027: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_118, getitem_119);  clone_118 = getitem_119 = None
	        mul_8535: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4027, rsqrt_29);  sub_4027 = rsqrt_29 = None
	        mul_8536: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8535, model_audio_tower_layers_14_final_layer_norm_weight);  mul_8535 = model_audio_tower_layers_14_final_layer_norm_weight = None
	        add_13504: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_8536, model_audio_tower_layers_14_final_layer_norm_bias);  mul_8536 = model_audio_tower_layers_14_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_13504, [2])
	        amax_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_13504, [2])
	        full_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_88, full_176);  amin_88 = full_176 = None
	        full_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_88, full_177);  amax_88 = full_177 = None
	        sub_4038: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_88, minimum_88);  maximum_88 = None
	        div_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4038, 255.0);  sub_4038 = None
	        clamp_min_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_176, 1.1920928955078125e-07);  div_176 = None
	        div_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_88, clamp_min_264);  minimum_88 = None
	        round_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_177);  div_177 = None
	        sub_4044: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_177);  round_177 = None
	        clamp_min_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4044, -128);  sub_4044 = None
	        clamp_max_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_265, 127);  clamp_min_265 = None
	        _assert_tensor_metadata_794 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_264, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_794 = None
	        _assert_tensor_metadata_795 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_176, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_795 = None
	        convert_element_type_528: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_176, torch.int8);  clamp_max_176 = None
	        view_1382: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_264, [sym_size_int, 1500, 1])
	        view_1383: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_528, [sym_size_int, 1500, 1])
	        reciprocal_88: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1382);  view_1382 = None
	        mul_8584: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_88, 1.0);  reciprocal_88 = None
	        mul_8587: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_13504, mul_8584);  add_13504 = mul_8584 = None
	        round_178: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8587);  mul_8587 = None
	        add_13591: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_178, view_1383);  round_178 = view_1383 = None
	        clamp_min_266: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13591, -128);  add_13591 = None
	        clamp_max_177: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_266, 127);  clamp_min_266 = None
	        _assert_tensor_metadata_796 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_177, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_796 = None
	        convert_element_type_529: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_177, torch.int8);  clamp_max_177 = None
	        view_1386: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_264, [sym_size_int, 1500, 1]);  clamp_min_264 = None
	        view_1387: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_528, [sym_size_int, 1500, 1]);  convert_element_type_528 = None
	        _assert_tensor_metadata_797 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_529, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_797 = None
	        convert_element_type_530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_529, torch.float32);  convert_element_type_529 = None
	        _assert_tensor_metadata_798 = torch.ops.aten._assert_tensor_metadata.default(view_1387, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_798 = None
	        convert_element_type_531: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1387, torch.float32);  view_1387 = None
	        sub_4064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_530, convert_element_type_531);  convert_element_type_530 = convert_element_type_531 = None
	        mul_8609: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4064, view_1386);  sub_4064 = view_1386 = None
	        _assert_tensor_metadata_799 = torch.ops.aten._assert_tensor_metadata.default(mul_8609, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_799 = None
	        view_1389: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_14_fc1_parametrizations_weight_original0 = None
	        view_1390: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_14_fc1_parametrizations_weight_original1 = None
	        view_1391: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_14_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_800 = torch.ops.aten._assert_tensor_metadata.default(view_1389, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_800 = None
	        convert_element_type_532: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1389, torch.float32);  view_1389 = None
	        _assert_tensor_metadata_801 = torch.ops.aten._assert_tensor_metadata.default(view_1391, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_801 = None
	        convert_element_type_533: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1391, torch.float32);  view_1391 = None
	        sub_4068: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_532, convert_element_type_533);  convert_element_type_532 = convert_element_type_533 = None
	        mul_8614: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4068, view_1390);  sub_4068 = view_1390 = None
	        view_1392: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8614, [5120, 1280]);  mul_8614 = None
	        _assert_tensor_metadata_802 = torch.ops.aten._assert_tensor_metadata.default(view_1392, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_802 = None
	        mul_8619: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1393: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8609, [mul_8619, 1280]);  mul_8609 = mul_8619 = None
	        permute_149: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1392, [1, 0]);  view_1392 = None
	        addmm_73: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_14_fc1_bias, view_1393, permute_149);  model_audio_tower_layers_14_fc1_bias = view_1393 = permute_149 = None
	        view_1394: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_73, [sym_size_int, 1500, 5120]);  addmm_73 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_8626: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1394, 0.5)
	        mul_8627: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1394, 0.7071067811865476);  view_1394 = None
	        erf_16: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_8627);  mul_8627 = None
	        add_13650: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_16, 1);  erf_16 = None
	        mul_8628: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8626, add_13650);  mul_8626 = add_13650 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_8628, [2])
	        amax_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_8628, [2])
	        full_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_89, full_178);  amin_89 = full_178 = None
	        full_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_89, full_179);  amax_89 = full_179 = None
	        sub_4081: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_89, minimum_89);  maximum_89 = None
	        div_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4081, 255.0);  sub_4081 = None
	        clamp_min_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_178, 1.1920928955078125e-07);  div_178 = None
	        div_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_89, clamp_min_267);  minimum_89 = None
	        round_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_179);  div_179 = None
	        sub_4087: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_179);  round_179 = None
	        clamp_min_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4087, -128);  sub_4087 = None
	        clamp_max_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_268, 127);  clamp_min_268 = None
	        _assert_tensor_metadata_803 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_267, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_803 = None
	        _assert_tensor_metadata_804 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_178, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_804 = None
	        convert_element_type_534: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_178, torch.int8);  clamp_max_178 = None
	        view_1397: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_267, [sym_size_int, 1500, 1])
	        view_1398: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_534, [sym_size_int, 1500, 1])
	        reciprocal_89: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1397);  view_1397 = None
	        mul_8674: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_89, 1.0);  reciprocal_89 = None
	        mul_8677: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8628, mul_8674);  mul_8628 = mul_8674 = None
	        round_180: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_8677);  mul_8677 = None
	        add_13733: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_180, view_1398);  round_180 = view_1398 = None
	        clamp_min_269: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13733, -128);  add_13733 = None
	        clamp_max_179: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_269, 127);  clamp_min_269 = None
	        _assert_tensor_metadata_805 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_179, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_805 = None
	        convert_element_type_535: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_179, torch.int8);  clamp_max_179 = None
	        view_1401: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_267, [sym_size_int, 1500, 1]);  clamp_min_267 = None
	        view_1402: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_534, [sym_size_int, 1500, 1]);  convert_element_type_534 = None
	        _assert_tensor_metadata_806 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_535, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_806 = None
	        convert_element_type_536: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_535, torch.float32);  convert_element_type_535 = None
	        _assert_tensor_metadata_807 = torch.ops.aten._assert_tensor_metadata.default(view_1402, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_807 = None
	        convert_element_type_537: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1402, torch.float32);  view_1402 = None
	        sub_4107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_536, convert_element_type_537);  convert_element_type_536 = convert_element_type_537 = None
	        mul_8699: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4107, view_1401);  sub_4107 = view_1401 = None
	        _assert_tensor_metadata_808 = torch.ops.aten._assert_tensor_metadata.default(mul_8699, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_808 = None
	        view_1404: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_14_fc2_parametrizations_weight_original0 = None
	        view_1405: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_14_fc2_parametrizations_weight_original1 = None
	        view_1406: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_14_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_809 = torch.ops.aten._assert_tensor_metadata.default(view_1404, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_809 = None
	        convert_element_type_538: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1404, torch.float32);  view_1404 = None
	        _assert_tensor_metadata_810 = torch.ops.aten._assert_tensor_metadata.default(view_1406, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_810 = None
	        convert_element_type_539: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1406, torch.float32);  view_1406 = None
	        sub_4111: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_538, convert_element_type_539);  convert_element_type_538 = convert_element_type_539 = None
	        mul_8704: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4111, view_1405);  sub_4111 = view_1405 = None
	        view_1407: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8704, [1280, 5120]);  mul_8704 = None
	        _assert_tensor_metadata_811 = torch.ops.aten._assert_tensor_metadata.default(view_1407, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_811 = None
	        mul_8709: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1408: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8699, [mul_8709, 5120]);  mul_8699 = mul_8709 = None
	        permute_150: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1407, [1, 0]);  view_1407 = None
	        addmm_74: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_14_fc2_bias, view_1408, permute_150);  model_audio_tower_layers_14_fc2_bias = view_1408 = permute_150 = None
	        view_1409: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_74, [sym_size_int, 1500, 1280]);  addmm_74 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_13796: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_13498, view_1409);  add_13498 = view_1409 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_121: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_13796, memory_format = torch.contiguous_format)
	        var_mean_30 = torch.ops.aten.var_mean.correction(clone_121, [2], correction = 0, keepdim = True)
	        getitem_120: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_30[0]
	        getitem_121: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_30[1];  var_mean_30 = None
	        add_13801: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_120, 1e-05);  getitem_120 = None
	        rsqrt_30: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_13801);  add_13801 = None
	        sub_4117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_121, getitem_121);  clone_121 = getitem_121 = None
	        mul_8720: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4117, rsqrt_30);  sub_4117 = rsqrt_30 = None
	        mul_8721: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8720, model_audio_tower_layers_15_self_attn_layer_norm_weight);  mul_8720 = model_audio_tower_layers_15_self_attn_layer_norm_weight = None
	        add_13802: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_8721, model_audio_tower_layers_15_self_attn_layer_norm_bias);  mul_8721 = model_audio_tower_layers_15_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_13802, [2])
	        amax_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_13802, [2])
	        full_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_90, full_180);  amin_90 = full_180 = None
	        full_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_90, full_181);  amax_90 = full_181 = None
	        sub_4128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_90, minimum_90);  maximum_90 = None
	        div_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4128, 255.0);  sub_4128 = None
	        clamp_min_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_180, 1.1920928955078125e-07);  div_180 = None
	        div_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_90, clamp_min_270);  minimum_90 = None
	        round_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_181);  div_181 = None
	        sub_4134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_181);  round_181 = None
	        clamp_min_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4134, -128);  sub_4134 = None
	        clamp_max_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_271, 127);  clamp_min_271 = None
	        _assert_tensor_metadata_812 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_270, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_812 = None
	        _assert_tensor_metadata_813 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_180, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_813 = None
	        convert_element_type_540: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_180, torch.int8);  clamp_max_180 = None
	        view_1412: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_270, [sym_size_int, 1500, 1])
	        view_1413: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_540, [sym_size_int, 1500, 1])
	        reciprocal_90: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1412);  view_1412 = None
	        mul_8769: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_90, 1.0);  reciprocal_90 = None
	        mul_8772: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_13802, mul_8769);  mul_8769 = None
	        round_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8772);  mul_8772 = None
	        add_13889: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_182, view_1413);  round_182 = view_1413 = None
	        clamp_min_272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13889, -128);  add_13889 = None
	        clamp_max_181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_272, 127);  clamp_min_272 = None
	        _assert_tensor_metadata_814 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_181, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_814 = None
	        convert_element_type_541: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_181, torch.int8);  clamp_max_181 = None
	        view_1416: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_270, [sym_size_int, 1500, 1]);  clamp_min_270 = None
	        view_1417: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_540, [sym_size_int, 1500, 1]);  convert_element_type_540 = None
	        _assert_tensor_metadata_815 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_541, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_815 = None
	        convert_element_type_542: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_541, torch.float32);  convert_element_type_541 = None
	        _assert_tensor_metadata_816 = torch.ops.aten._assert_tensor_metadata.default(view_1417, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_816 = None
	        convert_element_type_543: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1417, torch.float32);  view_1417 = None
	        sub_4154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_542, convert_element_type_543);  convert_element_type_542 = convert_element_type_543 = None
	        mul_8794: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4154, view_1416);  sub_4154 = view_1416 = None
	        _assert_tensor_metadata_817 = torch.ops.aten._assert_tensor_metadata.default(mul_8794, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_817 = None
	        view_1419: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1420: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1421: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_818 = torch.ops.aten._assert_tensor_metadata.default(view_1419, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_818 = None
	        convert_element_type_544: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1419, torch.float32);  view_1419 = None
	        _assert_tensor_metadata_819 = torch.ops.aten._assert_tensor_metadata.default(view_1421, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_819 = None
	        convert_element_type_545: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1421, torch.float32);  view_1421 = None
	        sub_4158: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_544, convert_element_type_545);  convert_element_type_544 = convert_element_type_545 = None
	        mul_8799: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4158, view_1420);  sub_4158 = view_1420 = None
	        view_1422: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8799, [1280, 1280]);  mul_8799 = None
	        _assert_tensor_metadata_820 = torch.ops.aten._assert_tensor_metadata.default(view_1422, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_820 = None
	        mul_8804: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1423: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8794, [mul_8804, 1280]);  mul_8794 = mul_8804 = None
	        permute_151: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1422, [1, 0]);  view_1422 = None
	        addmm_75: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_15_self_attn_q_proj_bias, view_1423, permute_151);  model_audio_tower_layers_15_self_attn_q_proj_bias = view_1423 = permute_151 = None
	        view_1424: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_75, [sym_size_int, 1500, 1280]);  addmm_75 = None
	        mul_8811: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1424, 0.125);  view_1424 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1425: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8811, [sym_size_int, 1500, 20, 64]);  mul_8811 = None
	        permute_152: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1425, [0, 2, 1, 3]);  view_1425 = None
	        clone_122: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_152, memory_format = torch.contiguous_format);  permute_152 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_13802, [2])
	        amax_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_13802, [2])
	        full_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_91, full_182);  amin_91 = full_182 = None
	        full_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_91, full_183);  amax_91 = full_183 = None
	        sub_4173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_91, minimum_91);  maximum_91 = None
	        div_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4173, 255.0);  sub_4173 = None
	        clamp_min_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_182, 1.1920928955078125e-07);  div_182 = None
	        div_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_91, clamp_min_273);  minimum_91 = None
	        round_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_183);  div_183 = None
	        sub_4179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_183);  round_183 = None
	        clamp_min_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4179, -128);  sub_4179 = None
	        clamp_max_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_274, 127);  clamp_min_274 = None
	        _assert_tensor_metadata_821 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_273, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_821 = None
	        _assert_tensor_metadata_822 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_182, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_822 = None
	        convert_element_type_546: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_182, torch.int8);  clamp_max_182 = None
	        view_1428: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_273, [sym_size_int, 1500, 1])
	        view_1429: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_546, [sym_size_int, 1500, 1])
	        reciprocal_91: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1428);  view_1428 = None
	        mul_8865: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_91, 1.0);  reciprocal_91 = None
	        mul_8868: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_13802, mul_8865);  mul_8865 = None
	        round_184: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8868);  mul_8868 = None
	        add_14041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_184, view_1429);  round_184 = view_1429 = None
	        clamp_min_275: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14041, -128);  add_14041 = None
	        clamp_max_183: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_275, 127);  clamp_min_275 = None
	        _assert_tensor_metadata_823 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_183, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_823 = None
	        convert_element_type_547: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_183, torch.int8);  clamp_max_183 = None
	        view_1432: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_273, [sym_size_int, 1500, 1]);  clamp_min_273 = None
	        view_1433: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_546, [sym_size_int, 1500, 1]);  convert_element_type_546 = None
	        _assert_tensor_metadata_824 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_547, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_824 = None
	        convert_element_type_548: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_547, torch.float32);  convert_element_type_547 = None
	        _assert_tensor_metadata_825 = torch.ops.aten._assert_tensor_metadata.default(view_1433, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_825 = None
	        convert_element_type_549: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1433, torch.float32);  view_1433 = None
	        sub_4199: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_548, convert_element_type_549);  convert_element_type_548 = convert_element_type_549 = None
	        mul_8890: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4199, view_1432);  sub_4199 = view_1432 = None
	        _assert_tensor_metadata_826 = torch.ops.aten._assert_tensor_metadata.default(mul_8890, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_826 = None
	        view_1435: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1436: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1437: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_827 = torch.ops.aten._assert_tensor_metadata.default(view_1435, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_827 = None
	        convert_element_type_550: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1435, torch.float32);  view_1435 = None
	        _assert_tensor_metadata_828 = torch.ops.aten._assert_tensor_metadata.default(view_1437, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_828 = None
	        convert_element_type_551: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1437, torch.float32);  view_1437 = None
	        sub_4203: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_550, convert_element_type_551);  convert_element_type_550 = convert_element_type_551 = None
	        mul_8895: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4203, view_1436);  sub_4203 = view_1436 = None
	        view_1438: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8895, [1280, 1280]);  mul_8895 = None
	        _assert_tensor_metadata_829 = torch.ops.aten._assert_tensor_metadata.default(view_1438, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_829 = None
	        permute_153: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1438, [1, 0]);  view_1438 = None
	        mul_8898: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1439: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8890, [mul_8898, 1280]);  mul_8890 = mul_8898 = None
	        mm_15: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1439, permute_153);  view_1439 = permute_153 = None
	        view_1440: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_15, [sym_size_int, 1500, 1280]);  mm_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1441: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1440, [sym_size_int, -1, 20, 64]);  view_1440 = None
	        permute_154: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1441, [0, 2, 1, 3]);  view_1441 = None
	        clone_123: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_154, memory_format = torch.contiguous_format);  permute_154 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_13802, [2])
	        amax_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_13802, [2])
	        full_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_92, full_184);  amin_92 = full_184 = None
	        full_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_92, full_185);  amax_92 = full_185 = None
	        sub_4217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_92, minimum_92);  maximum_92 = None
	        div_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4217, 255.0);  sub_4217 = None
	        clamp_min_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_184, 1.1920928955078125e-07);  div_184 = None
	        div_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_92, clamp_min_276);  minimum_92 = None
	        round_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_185);  div_185 = None
	        sub_4223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_185);  round_185 = None
	        clamp_min_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4223, -128);  sub_4223 = None
	        clamp_max_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_277, 127);  clamp_min_277 = None
	        _assert_tensor_metadata_830 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_276, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_830 = None
	        _assert_tensor_metadata_831 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_184, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_831 = None
	        convert_element_type_552: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_184, torch.int8);  clamp_max_184 = None
	        view_1444: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_276, [sym_size_int, 1500, 1])
	        view_1445: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_552, [sym_size_int, 1500, 1])
	        reciprocal_92: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1444);  view_1444 = None
	        mul_8964: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_92, 1.0);  reciprocal_92 = None
	        mul_8967: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_13802, mul_8964);  add_13802 = mul_8964 = None
	        round_186: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8967);  mul_8967 = None
	        add_14189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_186, view_1445);  round_186 = view_1445 = None
	        clamp_min_278: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14189, -128);  add_14189 = None
	        clamp_max_185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_278, 127);  clamp_min_278 = None
	        _assert_tensor_metadata_832 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_185, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_832 = None
	        convert_element_type_553: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_185, torch.int8);  clamp_max_185 = None
	        view_1448: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_276, [sym_size_int, 1500, 1]);  clamp_min_276 = None
	        view_1449: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_552, [sym_size_int, 1500, 1]);  convert_element_type_552 = None
	        _assert_tensor_metadata_833 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_553, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_833 = None
	        convert_element_type_554: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_553, torch.float32);  convert_element_type_553 = None
	        _assert_tensor_metadata_834 = torch.ops.aten._assert_tensor_metadata.default(view_1449, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_834 = None
	        convert_element_type_555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1449, torch.float32);  view_1449 = None
	        sub_4243: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_554, convert_element_type_555);  convert_element_type_554 = convert_element_type_555 = None
	        mul_8989: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4243, view_1448);  sub_4243 = view_1448 = None
	        _assert_tensor_metadata_835 = torch.ops.aten._assert_tensor_metadata.default(mul_8989, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_835 = None
	        view_1451: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1452: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1453: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_836 = torch.ops.aten._assert_tensor_metadata.default(view_1451, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_836 = None
	        convert_element_type_556: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1451, torch.float32);  view_1451 = None
	        _assert_tensor_metadata_837 = torch.ops.aten._assert_tensor_metadata.default(view_1453, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_837 = None
	        convert_element_type_557: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1453, torch.float32);  view_1453 = None
	        sub_4247: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_556, convert_element_type_557);  convert_element_type_556 = convert_element_type_557 = None
	        mul_8994: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4247, view_1452);  sub_4247 = view_1452 = None
	        view_1454: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8994, [1280, 1280]);  mul_8994 = None
	        _assert_tensor_metadata_838 = torch.ops.aten._assert_tensor_metadata.default(view_1454, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_838 = None
	        mul_8999: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1455: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8989, [mul_8999, 1280]);  mul_8989 = mul_8999 = None
	        permute_155: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1454, [1, 0]);  view_1454 = None
	        addmm_76: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_15_self_attn_v_proj_bias, view_1455, permute_155);  model_audio_tower_layers_15_self_attn_v_proj_bias = view_1455 = permute_155 = None
	        view_1456: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_76, [sym_size_int, 1500, 1280]);  addmm_76 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1457: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1456, [sym_size_int, -1, 20, 64]);  view_1456 = None
	        permute_156: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1457, [0, 2, 1, 3]);  view_1457 = None
	        clone_124: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_156, memory_format = torch.contiguous_format);  permute_156 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_15 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_122, clone_123, clone_124, None, False, scale = 1.0);  clone_122 = clone_123 = clone_124 = None
	        getitem_122: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_15[0];  _scaled_dot_product_efficient_attention_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_157: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_122, [0, 2, 1, 3]);  getitem_122 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_157, [sym_size_int, 1500, -1]);  permute_157 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1458, [2])
	        amax_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1458, [2])
	        full_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_93, full_186);  amin_93 = full_186 = None
	        full_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_93, full_187);  amax_93 = full_187 = None
	        sub_4265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_93, minimum_93);  maximum_93 = None
	        div_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4265, 255.0);  sub_4265 = None
	        clamp_min_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_186, 1.1920928955078125e-07);  div_186 = None
	        div_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_93, clamp_min_279);  minimum_93 = None
	        round_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_187);  div_187 = None
	        sub_4271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_187);  round_187 = None
	        clamp_min_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4271, -128);  sub_4271 = None
	        clamp_max_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_280, 127);  clamp_min_280 = None
	        _assert_tensor_metadata_839 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_279, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_839 = None
	        _assert_tensor_metadata_840 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_186, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_840 = None
	        convert_element_type_558: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_186, torch.int8);  clamp_max_186 = None
	        view_1461: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_279, [sym_size_int, 1500, 1])
	        view_1462: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_558, [sym_size_int, 1500, 1])
	        reciprocal_93: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1461);  view_1461 = None
	        mul_9069: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_93, 1.0);  reciprocal_93 = None
	        mul_9072: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1458, mul_9069);  view_1458 = mul_9069 = None
	        round_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9072);  mul_9072 = None
	        add_14353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_188, view_1462);  round_188 = view_1462 = None
	        clamp_min_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14353, -128);  add_14353 = None
	        clamp_max_187: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_281, 127);  clamp_min_281 = None
	        _assert_tensor_metadata_841 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_187, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_841 = None
	        convert_element_type_559: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_187, torch.int8);  clamp_max_187 = None
	        view_1465: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_279, [sym_size_int, 1500, 1]);  clamp_min_279 = None
	        view_1466: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_558, [sym_size_int, 1500, 1]);  convert_element_type_558 = None
	        _assert_tensor_metadata_842 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_559, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_842 = None
	        convert_element_type_560: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_559, torch.float32);  convert_element_type_559 = None
	        _assert_tensor_metadata_843 = torch.ops.aten._assert_tensor_metadata.default(view_1466, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_843 = None
	        convert_element_type_561: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1466, torch.float32);  view_1466 = None
	        sub_4291: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_560, convert_element_type_561);  convert_element_type_560 = convert_element_type_561 = None
	        mul_9094: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4291, view_1465);  sub_4291 = view_1465 = None
	        _assert_tensor_metadata_844 = torch.ops.aten._assert_tensor_metadata.default(mul_9094, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_844 = None
	        view_1468: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1469: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1470: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_845 = torch.ops.aten._assert_tensor_metadata.default(view_1468, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_845 = None
	        convert_element_type_562: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1468, torch.float32);  view_1468 = None
	        _assert_tensor_metadata_846 = torch.ops.aten._assert_tensor_metadata.default(view_1470, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_846 = None
	        convert_element_type_563: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1470, torch.float32);  view_1470 = None
	        sub_4295: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_562, convert_element_type_563);  convert_element_type_562 = convert_element_type_563 = None
	        mul_9099: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4295, view_1469);  sub_4295 = view_1469 = None
	        view_1471: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9099, [1280, 1280]);  mul_9099 = None
	        _assert_tensor_metadata_847 = torch.ops.aten._assert_tensor_metadata.default(view_1471, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_847 = None
	        mul_9104: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1472: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9094, [mul_9104, 1280]);  mul_9094 = mul_9104 = None
	        permute_158: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1471, [1, 0]);  view_1471 = None
	        addmm_77: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_15_self_attn_out_proj_bias, view_1472, permute_158);  model_audio_tower_layers_15_self_attn_out_proj_bias = view_1472 = permute_158 = None
	        view_1473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_77, [sym_size_int, 1500, 1280]);  addmm_77 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_14416: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_13796, view_1473);  add_13796 = view_1473 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_126: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_14416, memory_format = torch.contiguous_format)
	        var_mean_31 = torch.ops.aten.var_mean.correction(clone_126, [2], correction = 0, keepdim = True)
	        getitem_126: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_31[0]
	        getitem_127: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_31[1];  var_mean_31 = None
	        add_14421: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_126, 1e-05);  getitem_126 = None
	        rsqrt_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_14421);  add_14421 = None
	        sub_4301: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_126, getitem_127);  clone_126 = getitem_127 = None
	        mul_9115: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4301, rsqrt_31);  sub_4301 = rsqrt_31 = None
	        mul_9116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9115, model_audio_tower_layers_15_final_layer_norm_weight);  mul_9115 = model_audio_tower_layers_15_final_layer_norm_weight = None
	        add_14422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9116, model_audio_tower_layers_15_final_layer_norm_bias);  mul_9116 = model_audio_tower_layers_15_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_14422, [2])
	        amax_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_14422, [2])
	        full_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_94, full_188);  amin_94 = full_188 = None
	        full_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_94, full_189);  amax_94 = full_189 = None
	        sub_4312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_94, minimum_94);  maximum_94 = None
	        div_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4312, 255.0);  sub_4312 = None
	        clamp_min_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_188, 1.1920928955078125e-07);  div_188 = None
	        div_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_94, clamp_min_282);  minimum_94 = None
	        round_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_189);  div_189 = None
	        sub_4318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_189);  round_189 = None
	        clamp_min_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4318, -128);  sub_4318 = None
	        clamp_max_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_283, 127);  clamp_min_283 = None
	        _assert_tensor_metadata_848 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_282, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_848 = None
	        _assert_tensor_metadata_849 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_188, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_849 = None
	        convert_element_type_564: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_188, torch.int8);  clamp_max_188 = None
	        view_1476: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_282, [sym_size_int, 1500, 1])
	        view_1477: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_564, [sym_size_int, 1500, 1])
	        reciprocal_94: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1476);  view_1476 = None
	        mul_9164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_94, 1.0);  reciprocal_94 = None
	        mul_9167: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_14422, mul_9164);  add_14422 = mul_9164 = None
	        round_190: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9167);  mul_9167 = None
	        add_14509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_190, view_1477);  round_190 = view_1477 = None
	        clamp_min_284: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14509, -128);  add_14509 = None
	        clamp_max_189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_284, 127);  clamp_min_284 = None
	        _assert_tensor_metadata_850 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_189, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_850 = None
	        convert_element_type_565: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_189, torch.int8);  clamp_max_189 = None
	        view_1480: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_282, [sym_size_int, 1500, 1]);  clamp_min_282 = None
	        view_1481: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_564, [sym_size_int, 1500, 1]);  convert_element_type_564 = None
	        _assert_tensor_metadata_851 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_565, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_851 = None
	        convert_element_type_566: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_565, torch.float32);  convert_element_type_565 = None
	        _assert_tensor_metadata_852 = torch.ops.aten._assert_tensor_metadata.default(view_1481, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_852 = None
	        convert_element_type_567: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1481, torch.float32);  view_1481 = None
	        sub_4338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_566, convert_element_type_567);  convert_element_type_566 = convert_element_type_567 = None
	        mul_9189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4338, view_1480);  sub_4338 = view_1480 = None
	        _assert_tensor_metadata_853 = torch.ops.aten._assert_tensor_metadata.default(mul_9189, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_853 = None
	        view_1483: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_15_fc1_parametrizations_weight_original0 = None
	        view_1484: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_15_fc1_parametrizations_weight_original1 = None
	        view_1485: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_15_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_854 = torch.ops.aten._assert_tensor_metadata.default(view_1483, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_854 = None
	        convert_element_type_568: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1483, torch.float32);  view_1483 = None
	        _assert_tensor_metadata_855 = torch.ops.aten._assert_tensor_metadata.default(view_1485, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_855 = None
	        convert_element_type_569: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1485, torch.float32);  view_1485 = None
	        sub_4342: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_568, convert_element_type_569);  convert_element_type_568 = convert_element_type_569 = None
	        mul_9194: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4342, view_1484);  sub_4342 = view_1484 = None
	        view_1486: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9194, [5120, 1280]);  mul_9194 = None
	        _assert_tensor_metadata_856 = torch.ops.aten._assert_tensor_metadata.default(view_1486, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_856 = None
	        mul_9199: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1487: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9189, [mul_9199, 1280]);  mul_9189 = mul_9199 = None
	        permute_159: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1486, [1, 0]);  view_1486 = None
	        addmm_78: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_15_fc1_bias, view_1487, permute_159);  model_audio_tower_layers_15_fc1_bias = view_1487 = permute_159 = None
	        view_1488: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_78, [sym_size_int, 1500, 5120]);  addmm_78 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_9206: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1488, 0.5)
	        mul_9207: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1488, 0.7071067811865476);  view_1488 = None
	        erf_17: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_9207);  mul_9207 = None
	        add_14568: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_17, 1);  erf_17 = None
	        mul_9208: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9206, add_14568);  mul_9206 = add_14568 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_9208, [2])
	        amax_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_9208, [2])
	        full_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_95, full_190);  amin_95 = full_190 = None
	        full_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_95, full_191);  amax_95 = full_191 = None
	        sub_4355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_95, minimum_95);  maximum_95 = None
	        div_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4355, 255.0);  sub_4355 = None
	        clamp_min_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_190, 1.1920928955078125e-07);  div_190 = None
	        div_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_95, clamp_min_285);  minimum_95 = None
	        round_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_191);  div_191 = None
	        sub_4361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_191);  round_191 = None
	        clamp_min_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4361, -128);  sub_4361 = None
	        clamp_max_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_286, 127);  clamp_min_286 = None
	        _assert_tensor_metadata_857 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_285, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_857 = None
	        _assert_tensor_metadata_858 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_190, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_858 = None
	        convert_element_type_570: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_190, torch.int8);  clamp_max_190 = None
	        view_1491: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_285, [sym_size_int, 1500, 1])
	        view_1492: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_570, [sym_size_int, 1500, 1])
	        reciprocal_95: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1491);  view_1491 = None
	        mul_9254: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_95, 1.0);  reciprocal_95 = None
	        mul_9257: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9208, mul_9254);  mul_9208 = mul_9254 = None
	        round_192: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_9257);  mul_9257 = None
	        add_14651: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_192, view_1492);  round_192 = view_1492 = None
	        clamp_min_287: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14651, -128);  add_14651 = None
	        clamp_max_191: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_287, 127);  clamp_min_287 = None
	        _assert_tensor_metadata_859 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_191, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_859 = None
	        convert_element_type_571: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_191, torch.int8);  clamp_max_191 = None
	        view_1495: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_285, [sym_size_int, 1500, 1]);  clamp_min_285 = None
	        view_1496: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_570, [sym_size_int, 1500, 1]);  convert_element_type_570 = None
	        _assert_tensor_metadata_860 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_571, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_860 = None
	        convert_element_type_572: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_571, torch.float32);  convert_element_type_571 = None
	        _assert_tensor_metadata_861 = torch.ops.aten._assert_tensor_metadata.default(view_1496, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_861 = None
	        convert_element_type_573: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1496, torch.float32);  view_1496 = None
	        sub_4381: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_572, convert_element_type_573);  convert_element_type_572 = convert_element_type_573 = None
	        mul_9279: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4381, view_1495);  sub_4381 = view_1495 = None
	        _assert_tensor_metadata_862 = torch.ops.aten._assert_tensor_metadata.default(mul_9279, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_862 = None
	        view_1498: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_15_fc2_parametrizations_weight_original0 = None
	        view_1499: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_15_fc2_parametrizations_weight_original1 = None
	        view_1500: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_15_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_863 = torch.ops.aten._assert_tensor_metadata.default(view_1498, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_863 = None
	        convert_element_type_574: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1498, torch.float32);  view_1498 = None
	        _assert_tensor_metadata_864 = torch.ops.aten._assert_tensor_metadata.default(view_1500, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_864 = None
	        convert_element_type_575: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1500, torch.float32);  view_1500 = None
	        sub_4385: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_574, convert_element_type_575);  convert_element_type_574 = convert_element_type_575 = None
	        mul_9284: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4385, view_1499);  sub_4385 = view_1499 = None
	        view_1501: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9284, [1280, 5120]);  mul_9284 = None
	        _assert_tensor_metadata_865 = torch.ops.aten._assert_tensor_metadata.default(view_1501, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_865 = None
	        mul_9289: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1502: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9279, [mul_9289, 5120]);  mul_9279 = mul_9289 = None
	        permute_160: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1501, [1, 0]);  view_1501 = None
	        addmm_79: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_15_fc2_bias, view_1502, permute_160);  model_audio_tower_layers_15_fc2_bias = view_1502 = permute_160 = None
	        view_1503: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_79, [sym_size_int, 1500, 1280]);  addmm_79 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_14714: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_14416, view_1503);  add_14416 = view_1503 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_129: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_14714, memory_format = torch.contiguous_format)
	        var_mean_32 = torch.ops.aten.var_mean.correction(clone_129, [2], correction = 0, keepdim = True)
	        getitem_128: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_32[0]
	        getitem_129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_32[1];  var_mean_32 = None
	        add_14719: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_128, 1e-05);  getitem_128 = None
	        rsqrt_32: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_14719);  add_14719 = None
	        sub_4391: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_129, getitem_129);  clone_129 = getitem_129 = None
	        mul_9300: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4391, rsqrt_32);  sub_4391 = rsqrt_32 = None
	        mul_9301: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9300, model_audio_tower_layers_16_self_attn_layer_norm_weight);  mul_9300 = model_audio_tower_layers_16_self_attn_layer_norm_weight = None
	        add_14720: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9301, model_audio_tower_layers_16_self_attn_layer_norm_bias);  mul_9301 = model_audio_tower_layers_16_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_14720, [2])
	        amax_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_14720, [2])
	        full_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_96, full_192);  amin_96 = full_192 = None
	        full_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_96, full_193);  amax_96 = full_193 = None
	        sub_4402: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_96, minimum_96);  maximum_96 = None
	        div_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4402, 255.0);  sub_4402 = None
	        clamp_min_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_192, 1.1920928955078125e-07);  div_192 = None
	        div_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_96, clamp_min_288);  minimum_96 = None
	        round_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_193);  div_193 = None
	        sub_4408: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_193);  round_193 = None
	        clamp_min_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4408, -128);  sub_4408 = None
	        clamp_max_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_289, 127);  clamp_min_289 = None
	        _assert_tensor_metadata_866 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_288, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_866 = None
	        _assert_tensor_metadata_867 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_192, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_867 = None
	        convert_element_type_576: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_192, torch.int8);  clamp_max_192 = None
	        view_1506: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_288, [sym_size_int, 1500, 1])
	        view_1507: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_576, [sym_size_int, 1500, 1])
	        reciprocal_96: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1506);  view_1506 = None
	        mul_9349: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_96, 1.0);  reciprocal_96 = None
	        mul_9352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_14720, mul_9349);  mul_9349 = None
	        round_194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9352);  mul_9352 = None
	        add_14807: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_194, view_1507);  round_194 = view_1507 = None
	        clamp_min_290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14807, -128);  add_14807 = None
	        clamp_max_193: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_290, 127);  clamp_min_290 = None
	        _assert_tensor_metadata_868 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_193, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_868 = None
	        convert_element_type_577: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_193, torch.int8);  clamp_max_193 = None
	        view_1510: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_288, [sym_size_int, 1500, 1]);  clamp_min_288 = None
	        view_1511: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_576, [sym_size_int, 1500, 1]);  convert_element_type_576 = None
	        _assert_tensor_metadata_869 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_577, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_869 = None
	        convert_element_type_578: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_577, torch.float32);  convert_element_type_577 = None
	        _assert_tensor_metadata_870 = torch.ops.aten._assert_tensor_metadata.default(view_1511, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_870 = None
	        convert_element_type_579: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1511, torch.float32);  view_1511 = None
	        sub_4428: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_578, convert_element_type_579);  convert_element_type_578 = convert_element_type_579 = None
	        mul_9374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4428, view_1510);  sub_4428 = view_1510 = None
	        _assert_tensor_metadata_871 = torch.ops.aten._assert_tensor_metadata.default(mul_9374, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_871 = None
	        view_1513: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1514: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1515: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_872 = torch.ops.aten._assert_tensor_metadata.default(view_1513, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_872 = None
	        convert_element_type_580: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1513, torch.float32);  view_1513 = None
	        _assert_tensor_metadata_873 = torch.ops.aten._assert_tensor_metadata.default(view_1515, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_873 = None
	        convert_element_type_581: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1515, torch.float32);  view_1515 = None
	        sub_4432: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_580, convert_element_type_581);  convert_element_type_580 = convert_element_type_581 = None
	        mul_9379: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4432, view_1514);  sub_4432 = view_1514 = None
	        view_1516: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9379, [1280, 1280]);  mul_9379 = None
	        _assert_tensor_metadata_874 = torch.ops.aten._assert_tensor_metadata.default(view_1516, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_874 = None
	        mul_9384: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1517: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9374, [mul_9384, 1280]);  mul_9374 = mul_9384 = None
	        permute_161: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1516, [1, 0]);  view_1516 = None
	        addmm_80: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_16_self_attn_q_proj_bias, view_1517, permute_161);  model_audio_tower_layers_16_self_attn_q_proj_bias = view_1517 = permute_161 = None
	        view_1518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_80, [sym_size_int, 1500, 1280]);  addmm_80 = None
	        mul_9391: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1518, 0.125);  view_1518 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1519: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9391, [sym_size_int, 1500, 20, 64]);  mul_9391 = None
	        permute_162: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1519, [0, 2, 1, 3]);  view_1519 = None
	        clone_130: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_162, memory_format = torch.contiguous_format);  permute_162 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_14720, [2])
	        amax_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_14720, [2])
	        full_194: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_97, full_194);  amin_97 = full_194 = None
	        full_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_97, full_195);  amax_97 = full_195 = None
	        sub_4447: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_97, minimum_97);  maximum_97 = None
	        div_194: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4447, 255.0);  sub_4447 = None
	        clamp_min_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_194, 1.1920928955078125e-07);  div_194 = None
	        div_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_97, clamp_min_291);  minimum_97 = None
	        round_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_195);  div_195 = None
	        sub_4453: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_195);  round_195 = None
	        clamp_min_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4453, -128);  sub_4453 = None
	        clamp_max_194: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_292, 127);  clamp_min_292 = None
	        _assert_tensor_metadata_875 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_291, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_875 = None
	        _assert_tensor_metadata_876 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_194, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_876 = None
	        convert_element_type_582: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_194, torch.int8);  clamp_max_194 = None
	        view_1522: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_291, [sym_size_int, 1500, 1])
	        view_1523: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_582, [sym_size_int, 1500, 1])
	        reciprocal_97: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1522);  view_1522 = None
	        mul_9445: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_97, 1.0);  reciprocal_97 = None
	        mul_9448: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_14720, mul_9445);  mul_9445 = None
	        round_196: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9448);  mul_9448 = None
	        add_14959: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_196, view_1523);  round_196 = view_1523 = None
	        clamp_min_293: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14959, -128);  add_14959 = None
	        clamp_max_195: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_293, 127);  clamp_min_293 = None
	        _assert_tensor_metadata_877 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_195, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_877 = None
	        convert_element_type_583: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_195, torch.int8);  clamp_max_195 = None
	        view_1526: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_291, [sym_size_int, 1500, 1]);  clamp_min_291 = None
	        view_1527: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_582, [sym_size_int, 1500, 1]);  convert_element_type_582 = None
	        _assert_tensor_metadata_878 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_583, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_878 = None
	        convert_element_type_584: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_583, torch.float32);  convert_element_type_583 = None
	        _assert_tensor_metadata_879 = torch.ops.aten._assert_tensor_metadata.default(view_1527, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_879 = None
	        convert_element_type_585: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1527, torch.float32);  view_1527 = None
	        sub_4473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_584, convert_element_type_585);  convert_element_type_584 = convert_element_type_585 = None
	        mul_9470: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4473, view_1526);  sub_4473 = view_1526 = None
	        _assert_tensor_metadata_880 = torch.ops.aten._assert_tensor_metadata.default(mul_9470, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_880 = None
	        view_1529: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1530: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1531: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_881 = torch.ops.aten._assert_tensor_metadata.default(view_1529, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_881 = None
	        convert_element_type_586: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1529, torch.float32);  view_1529 = None
	        _assert_tensor_metadata_882 = torch.ops.aten._assert_tensor_metadata.default(view_1531, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_882 = None
	        convert_element_type_587: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1531, torch.float32);  view_1531 = None
	        sub_4477: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_586, convert_element_type_587);  convert_element_type_586 = convert_element_type_587 = None
	        mul_9475: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4477, view_1530);  sub_4477 = view_1530 = None
	        view_1532: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9475, [1280, 1280]);  mul_9475 = None
	        _assert_tensor_metadata_883 = torch.ops.aten._assert_tensor_metadata.default(view_1532, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_883 = None
	        permute_163: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1532, [1, 0]);  view_1532 = None
	        mul_9478: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1533: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9470, [mul_9478, 1280]);  mul_9470 = mul_9478 = None
	        mm_16: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1533, permute_163);  view_1533 = permute_163 = None
	        view_1534: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_16, [sym_size_int, 1500, 1280]);  mm_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1535: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1534, [sym_size_int, -1, 20, 64]);  view_1534 = None
	        permute_164: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1535, [0, 2, 1, 3]);  view_1535 = None
	        clone_131: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_164, memory_format = torch.contiguous_format);  permute_164 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_14720, [2])
	        amax_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_14720, [2])
	        full_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_98, full_196);  amin_98 = full_196 = None
	        full_197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_98, full_197);  amax_98 = full_197 = None
	        sub_4491: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_98, minimum_98);  maximum_98 = None
	        div_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4491, 255.0);  sub_4491 = None
	        clamp_min_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_196, 1.1920928955078125e-07);  div_196 = None
	        div_197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_98, clamp_min_294);  minimum_98 = None
	        round_197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_197);  div_197 = None
	        sub_4497: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_197);  round_197 = None
	        clamp_min_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4497, -128);  sub_4497 = None
	        clamp_max_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_295, 127);  clamp_min_295 = None
	        _assert_tensor_metadata_884 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_884 = None
	        _assert_tensor_metadata_885 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_196, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_885 = None
	        convert_element_type_588: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_196, torch.int8);  clamp_max_196 = None
	        view_1538: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_294, [sym_size_int, 1500, 1])
	        view_1539: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_588, [sym_size_int, 1500, 1])
	        reciprocal_98: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1538);  view_1538 = None
	        mul_9544: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_98, 1.0);  reciprocal_98 = None
	        mul_9547: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_14720, mul_9544);  add_14720 = mul_9544 = None
	        round_198: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9547);  mul_9547 = None
	        add_15107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_198, view_1539);  round_198 = view_1539 = None
	        clamp_min_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15107, -128);  add_15107 = None
	        clamp_max_197: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_296, 127);  clamp_min_296 = None
	        _assert_tensor_metadata_886 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_197, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_886 = None
	        convert_element_type_589: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_197, torch.int8);  clamp_max_197 = None
	        view_1542: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_294, [sym_size_int, 1500, 1]);  clamp_min_294 = None
	        view_1543: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_588, [sym_size_int, 1500, 1]);  convert_element_type_588 = None
	        _assert_tensor_metadata_887 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_589, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_887 = None
	        convert_element_type_590: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_589, torch.float32);  convert_element_type_589 = None
	        _assert_tensor_metadata_888 = torch.ops.aten._assert_tensor_metadata.default(view_1543, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_888 = None
	        convert_element_type_591: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1543, torch.float32);  view_1543 = None
	        sub_4517: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_590, convert_element_type_591);  convert_element_type_590 = convert_element_type_591 = None
	        mul_9569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4517, view_1542);  sub_4517 = view_1542 = None
	        _assert_tensor_metadata_889 = torch.ops.aten._assert_tensor_metadata.default(mul_9569, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_889 = None
	        view_1545: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1546: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1547: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_890 = torch.ops.aten._assert_tensor_metadata.default(view_1545, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_890 = None
	        convert_element_type_592: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1545, torch.float32);  view_1545 = None
	        _assert_tensor_metadata_891 = torch.ops.aten._assert_tensor_metadata.default(view_1547, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_891 = None
	        convert_element_type_593: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1547, torch.float32);  view_1547 = None
	        sub_4521: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_592, convert_element_type_593);  convert_element_type_592 = convert_element_type_593 = None
	        mul_9574: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4521, view_1546);  sub_4521 = view_1546 = None
	        view_1548: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9574, [1280, 1280]);  mul_9574 = None
	        _assert_tensor_metadata_892 = torch.ops.aten._assert_tensor_metadata.default(view_1548, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_892 = None
	        mul_9579: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1549: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9569, [mul_9579, 1280]);  mul_9569 = mul_9579 = None
	        permute_165: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1548, [1, 0]);  view_1548 = None
	        addmm_81: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_16_self_attn_v_proj_bias, view_1549, permute_165);  model_audio_tower_layers_16_self_attn_v_proj_bias = view_1549 = permute_165 = None
	        view_1550: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_81, [sym_size_int, 1500, 1280]);  addmm_81 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1551: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1550, [sym_size_int, -1, 20, 64]);  view_1550 = None
	        permute_166: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1551, [0, 2, 1, 3]);  view_1551 = None
	        clone_132: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_166, memory_format = torch.contiguous_format);  permute_166 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_16 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_130, clone_131, clone_132, None, False, scale = 1.0);  clone_130 = clone_131 = clone_132 = None
	        getitem_130: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_16[0];  _scaled_dot_product_efficient_attention_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_167: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_130, [0, 2, 1, 3]);  getitem_130 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_167, [sym_size_int, 1500, -1]);  permute_167 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1552, [2])
	        amax_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1552, [2])
	        full_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_99, full_198);  amin_99 = full_198 = None
	        full_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_99, full_199);  amax_99 = full_199 = None
	        sub_4539: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_99, minimum_99);  maximum_99 = None
	        div_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4539, 255.0);  sub_4539 = None
	        clamp_min_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_198, 1.1920928955078125e-07);  div_198 = None
	        div_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_99, clamp_min_297);  minimum_99 = None
	        round_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_199);  div_199 = None
	        sub_4545: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_199);  round_199 = None
	        clamp_min_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4545, -128);  sub_4545 = None
	        clamp_max_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_298, 127);  clamp_min_298 = None
	        _assert_tensor_metadata_893 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_297, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_893 = None
	        _assert_tensor_metadata_894 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_198, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_894 = None
	        convert_element_type_594: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_198, torch.int8);  clamp_max_198 = None
	        view_1555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_297, [sym_size_int, 1500, 1])
	        view_1556: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_594, [sym_size_int, 1500, 1])
	        reciprocal_99: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1555);  view_1555 = None
	        mul_9649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_99, 1.0);  reciprocal_99 = None
	        mul_9652: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1552, mul_9649);  view_1552 = mul_9649 = None
	        round_200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9652);  mul_9652 = None
	        add_15271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_200, view_1556);  round_200 = view_1556 = None
	        clamp_min_299: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15271, -128);  add_15271 = None
	        clamp_max_199: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_299, 127);  clamp_min_299 = None
	        _assert_tensor_metadata_895 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_199, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_895 = None
	        convert_element_type_595: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_199, torch.int8);  clamp_max_199 = None
	        view_1559: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_297, [sym_size_int, 1500, 1]);  clamp_min_297 = None
	        view_1560: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_594, [sym_size_int, 1500, 1]);  convert_element_type_594 = None
	        _assert_tensor_metadata_896 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_595, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_896 = None
	        convert_element_type_596: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_595, torch.float32);  convert_element_type_595 = None
	        _assert_tensor_metadata_897 = torch.ops.aten._assert_tensor_metadata.default(view_1560, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_897 = None
	        convert_element_type_597: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1560, torch.float32);  view_1560 = None
	        sub_4565: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_596, convert_element_type_597);  convert_element_type_596 = convert_element_type_597 = None
	        mul_9674: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4565, view_1559);  sub_4565 = view_1559 = None
	        _assert_tensor_metadata_898 = torch.ops.aten._assert_tensor_metadata.default(mul_9674, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_898 = None
	        view_1562: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1563: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1564: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_899 = torch.ops.aten._assert_tensor_metadata.default(view_1562, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_899 = None
	        convert_element_type_598: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1562, torch.float32);  view_1562 = None
	        _assert_tensor_metadata_900 = torch.ops.aten._assert_tensor_metadata.default(view_1564, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_900 = None
	        convert_element_type_599: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1564, torch.float32);  view_1564 = None
	        sub_4569: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_598, convert_element_type_599);  convert_element_type_598 = convert_element_type_599 = None
	        mul_9679: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4569, view_1563);  sub_4569 = view_1563 = None
	        view_1565: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9679, [1280, 1280]);  mul_9679 = None
	        _assert_tensor_metadata_901 = torch.ops.aten._assert_tensor_metadata.default(view_1565, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_901 = None
	        mul_9684: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1566: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9674, [mul_9684, 1280]);  mul_9674 = mul_9684 = None
	        permute_168: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1565, [1, 0]);  view_1565 = None
	        addmm_82: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_16_self_attn_out_proj_bias, view_1566, permute_168);  model_audio_tower_layers_16_self_attn_out_proj_bias = view_1566 = permute_168 = None
	        view_1567: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_82, [sym_size_int, 1500, 1280]);  addmm_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_15334: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_14714, view_1567);  add_14714 = view_1567 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_15334, memory_format = torch.contiguous_format)
	        var_mean_33 = torch.ops.aten.var_mean.correction(clone_134, [2], correction = 0, keepdim = True)
	        getitem_134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_33[0]
	        getitem_135: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_33[1];  var_mean_33 = None
	        add_15339: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_134, 1e-05);  getitem_134 = None
	        rsqrt_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_15339);  add_15339 = None
	        sub_4575: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_134, getitem_135);  clone_134 = getitem_135 = None
	        mul_9695: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4575, rsqrt_33);  sub_4575 = rsqrt_33 = None
	        mul_9696: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9695, model_audio_tower_layers_16_final_layer_norm_weight);  mul_9695 = model_audio_tower_layers_16_final_layer_norm_weight = None
	        add_15340: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9696, model_audio_tower_layers_16_final_layer_norm_bias);  mul_9696 = model_audio_tower_layers_16_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_15340, [2])
	        amax_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_15340, [2])
	        full_200: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_100, full_200);  amin_100 = full_200 = None
	        full_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_100, full_201);  amax_100 = full_201 = None
	        sub_4586: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_100, minimum_100);  maximum_100 = None
	        div_200: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4586, 255.0);  sub_4586 = None
	        clamp_min_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_200, 1.1920928955078125e-07);  div_200 = None
	        div_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_100, clamp_min_300);  minimum_100 = None
	        round_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_201);  div_201 = None
	        sub_4592: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_201);  round_201 = None
	        clamp_min_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4592, -128);  sub_4592 = None
	        clamp_max_200: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_301, 127);  clamp_min_301 = None
	        _assert_tensor_metadata_902 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_300, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_902 = None
	        _assert_tensor_metadata_903 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_200, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_903 = None
	        convert_element_type_600: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_200, torch.int8);  clamp_max_200 = None
	        view_1570: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_300, [sym_size_int, 1500, 1])
	        view_1571: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_600, [sym_size_int, 1500, 1])
	        reciprocal_100: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1570);  view_1570 = None
	        mul_9744: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_100, 1.0);  reciprocal_100 = None
	        mul_9747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_15340, mul_9744);  add_15340 = mul_9744 = None
	        round_202: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9747);  mul_9747 = None
	        add_15427: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_202, view_1571);  round_202 = view_1571 = None
	        clamp_min_302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15427, -128);  add_15427 = None
	        clamp_max_201: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_302, 127);  clamp_min_302 = None
	        _assert_tensor_metadata_904 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_201, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_904 = None
	        convert_element_type_601: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_201, torch.int8);  clamp_max_201 = None
	        view_1574: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_300, [sym_size_int, 1500, 1]);  clamp_min_300 = None
	        view_1575: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_600, [sym_size_int, 1500, 1]);  convert_element_type_600 = None
	        _assert_tensor_metadata_905 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_601, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_905 = None
	        convert_element_type_602: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_601, torch.float32);  convert_element_type_601 = None
	        _assert_tensor_metadata_906 = torch.ops.aten._assert_tensor_metadata.default(view_1575, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_906 = None
	        convert_element_type_603: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1575, torch.float32);  view_1575 = None
	        sub_4612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_602, convert_element_type_603);  convert_element_type_602 = convert_element_type_603 = None
	        mul_9769: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4612, view_1574);  sub_4612 = view_1574 = None
	        _assert_tensor_metadata_907 = torch.ops.aten._assert_tensor_metadata.default(mul_9769, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_907 = None
	        view_1577: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_16_fc1_parametrizations_weight_original0 = None
	        view_1578: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_16_fc1_parametrizations_weight_original1 = None
	        view_1579: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_16_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_908 = torch.ops.aten._assert_tensor_metadata.default(view_1577, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_908 = None
	        convert_element_type_604: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1577, torch.float32);  view_1577 = None
	        _assert_tensor_metadata_909 = torch.ops.aten._assert_tensor_metadata.default(view_1579, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_909 = None
	        convert_element_type_605: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1579, torch.float32);  view_1579 = None
	        sub_4616: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_604, convert_element_type_605);  convert_element_type_604 = convert_element_type_605 = None
	        mul_9774: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4616, view_1578);  sub_4616 = view_1578 = None
	        view_1580: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9774, [5120, 1280]);  mul_9774 = None
	        _assert_tensor_metadata_910 = torch.ops.aten._assert_tensor_metadata.default(view_1580, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_910 = None
	        mul_9779: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1581: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9769, [mul_9779, 1280]);  mul_9769 = mul_9779 = None
	        permute_169: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1580, [1, 0]);  view_1580 = None
	        addmm_83: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_16_fc1_bias, view_1581, permute_169);  model_audio_tower_layers_16_fc1_bias = view_1581 = permute_169 = None
	        view_1582: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_83, [sym_size_int, 1500, 5120]);  addmm_83 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_9786: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1582, 0.5)
	        mul_9787: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1582, 0.7071067811865476);  view_1582 = None
	        erf_18: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_9787);  mul_9787 = None
	        add_15486: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_18, 1);  erf_18 = None
	        mul_9788: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9786, add_15486);  mul_9786 = add_15486 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_9788, [2])
	        amax_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_9788, [2])
	        full_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_101, full_202);  amin_101 = full_202 = None
	        full_203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_101, full_203);  amax_101 = full_203 = None
	        sub_4629: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_101, minimum_101);  maximum_101 = None
	        div_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4629, 255.0);  sub_4629 = None
	        clamp_min_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_202, 1.1920928955078125e-07);  div_202 = None
	        div_203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_101, clamp_min_303);  minimum_101 = None
	        round_203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_203);  div_203 = None
	        sub_4635: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_203);  round_203 = None
	        clamp_min_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4635, -128);  sub_4635 = None
	        clamp_max_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_304, 127);  clamp_min_304 = None
	        _assert_tensor_metadata_911 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_303, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_911 = None
	        _assert_tensor_metadata_912 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_202, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_912 = None
	        convert_element_type_606: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_202, torch.int8);  clamp_max_202 = None
	        view_1585: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_303, [sym_size_int, 1500, 1])
	        view_1586: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_606, [sym_size_int, 1500, 1])
	        reciprocal_101: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1585);  view_1585 = None
	        mul_9834: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_101, 1.0);  reciprocal_101 = None
	        mul_9837: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9788, mul_9834);  mul_9788 = mul_9834 = None
	        round_204: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_9837);  mul_9837 = None
	        add_15569: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_204, view_1586);  round_204 = view_1586 = None
	        clamp_min_305: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15569, -128);  add_15569 = None
	        clamp_max_203: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_305, 127);  clamp_min_305 = None
	        _assert_tensor_metadata_913 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_203, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_913 = None
	        convert_element_type_607: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_203, torch.int8);  clamp_max_203 = None
	        view_1589: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_303, [sym_size_int, 1500, 1]);  clamp_min_303 = None
	        view_1590: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_606, [sym_size_int, 1500, 1]);  convert_element_type_606 = None
	        _assert_tensor_metadata_914 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_607, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_914 = None
	        convert_element_type_608: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_607, torch.float32);  convert_element_type_607 = None
	        _assert_tensor_metadata_915 = torch.ops.aten._assert_tensor_metadata.default(view_1590, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_915 = None
	        convert_element_type_609: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1590, torch.float32);  view_1590 = None
	        sub_4655: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_608, convert_element_type_609);  convert_element_type_608 = convert_element_type_609 = None
	        mul_9859: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4655, view_1589);  sub_4655 = view_1589 = None
	        _assert_tensor_metadata_916 = torch.ops.aten._assert_tensor_metadata.default(mul_9859, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_916 = None
	        view_1592: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_16_fc2_parametrizations_weight_original0 = None
	        view_1593: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_16_fc2_parametrizations_weight_original1 = None
	        view_1594: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_16_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_917 = torch.ops.aten._assert_tensor_metadata.default(view_1592, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_917 = None
	        convert_element_type_610: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1592, torch.float32);  view_1592 = None
	        _assert_tensor_metadata_918 = torch.ops.aten._assert_tensor_metadata.default(view_1594, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_918 = None
	        convert_element_type_611: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1594, torch.float32);  view_1594 = None
	        sub_4659: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_610, convert_element_type_611);  convert_element_type_610 = convert_element_type_611 = None
	        mul_9864: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4659, view_1593);  sub_4659 = view_1593 = None
	        view_1595: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9864, [1280, 5120]);  mul_9864 = None
	        _assert_tensor_metadata_919 = torch.ops.aten._assert_tensor_metadata.default(view_1595, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_919 = None
	        mul_9869: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1596: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9859, [mul_9869, 5120]);  mul_9859 = mul_9869 = None
	        permute_170: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1595, [1, 0]);  view_1595 = None
	        addmm_84: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_16_fc2_bias, view_1596, permute_170);  model_audio_tower_layers_16_fc2_bias = view_1596 = permute_170 = None
	        view_1597: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_84, [sym_size_int, 1500, 1280]);  addmm_84 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_15632: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_15334, view_1597);  add_15334 = view_1597 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_15632, memory_format = torch.contiguous_format)
	        var_mean_34 = torch.ops.aten.var_mean.correction(clone_137, [2], correction = 0, keepdim = True)
	        getitem_136: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_34[0]
	        getitem_137: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_34[1];  var_mean_34 = None
	        add_15637: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_136, 1e-05);  getitem_136 = None
	        rsqrt_34: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_15637);  add_15637 = None
	        sub_4665: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_137, getitem_137);  clone_137 = getitem_137 = None
	        mul_9880: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4665, rsqrt_34);  sub_4665 = rsqrt_34 = None
	        mul_9881: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9880, model_audio_tower_layers_17_self_attn_layer_norm_weight);  mul_9880 = model_audio_tower_layers_17_self_attn_layer_norm_weight = None
	        add_15638: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9881, model_audio_tower_layers_17_self_attn_layer_norm_bias);  mul_9881 = model_audio_tower_layers_17_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_15638, [2])
	        amax_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_15638, [2])
	        full_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_102, full_204);  amin_102 = full_204 = None
	        full_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_102, full_205);  amax_102 = full_205 = None
	        sub_4676: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_102, minimum_102);  maximum_102 = None
	        div_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4676, 255.0);  sub_4676 = None
	        clamp_min_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_204, 1.1920928955078125e-07);  div_204 = None
	        div_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_102, clamp_min_306);  minimum_102 = None
	        round_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_205);  div_205 = None
	        sub_4682: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_205);  round_205 = None
	        clamp_min_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4682, -128);  sub_4682 = None
	        clamp_max_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_307, 127);  clamp_min_307 = None
	        _assert_tensor_metadata_920 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_306, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_920 = None
	        _assert_tensor_metadata_921 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_204, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_921 = None
	        convert_element_type_612: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_204, torch.int8);  clamp_max_204 = None
	        view_1600: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_306, [sym_size_int, 1500, 1])
	        view_1601: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_612, [sym_size_int, 1500, 1])
	        reciprocal_102: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1600);  view_1600 = None
	        mul_9929: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_102, 1.0);  reciprocal_102 = None
	        mul_9932: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_15638, mul_9929);  mul_9929 = None
	        round_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9932);  mul_9932 = None
	        add_15725: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_206, view_1601);  round_206 = view_1601 = None
	        clamp_min_308: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15725, -128);  add_15725 = None
	        clamp_max_205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_308, 127);  clamp_min_308 = None
	        _assert_tensor_metadata_922 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_205, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_922 = None
	        convert_element_type_613: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_205, torch.int8);  clamp_max_205 = None
	        view_1604: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_306, [sym_size_int, 1500, 1]);  clamp_min_306 = None
	        view_1605: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_612, [sym_size_int, 1500, 1]);  convert_element_type_612 = None
	        _assert_tensor_metadata_923 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_613, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_923 = None
	        convert_element_type_614: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_613, torch.float32);  convert_element_type_613 = None
	        _assert_tensor_metadata_924 = torch.ops.aten._assert_tensor_metadata.default(view_1605, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_924 = None
	        convert_element_type_615: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1605, torch.float32);  view_1605 = None
	        sub_4702: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_614, convert_element_type_615);  convert_element_type_614 = convert_element_type_615 = None
	        mul_9954: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4702, view_1604);  sub_4702 = view_1604 = None
	        _assert_tensor_metadata_925 = torch.ops.aten._assert_tensor_metadata.default(mul_9954, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_925 = None
	        view_1607: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1608: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1609: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_926 = torch.ops.aten._assert_tensor_metadata.default(view_1607, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_926 = None
	        convert_element_type_616: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1607, torch.float32);  view_1607 = None
	        _assert_tensor_metadata_927 = torch.ops.aten._assert_tensor_metadata.default(view_1609, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_927 = None
	        convert_element_type_617: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1609, torch.float32);  view_1609 = None
	        sub_4706: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_616, convert_element_type_617);  convert_element_type_616 = convert_element_type_617 = None
	        mul_9959: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4706, view_1608);  sub_4706 = view_1608 = None
	        view_1610: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9959, [1280, 1280]);  mul_9959 = None
	        _assert_tensor_metadata_928 = torch.ops.aten._assert_tensor_metadata.default(view_1610, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_928 = None
	        mul_9964: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1611: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9954, [mul_9964, 1280]);  mul_9954 = mul_9964 = None
	        permute_171: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1610, [1, 0]);  view_1610 = None
	        addmm_85: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_17_self_attn_q_proj_bias, view_1611, permute_171);  model_audio_tower_layers_17_self_attn_q_proj_bias = view_1611 = permute_171 = None
	        view_1612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_85, [sym_size_int, 1500, 1280]);  addmm_85 = None
	        mul_9971: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1612, 0.125);  view_1612 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1613: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9971, [sym_size_int, 1500, 20, 64]);  mul_9971 = None
	        permute_172: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1613, [0, 2, 1, 3]);  view_1613 = None
	        clone_138: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_172, memory_format = torch.contiguous_format);  permute_172 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_15638, [2])
	        amax_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_15638, [2])
	        full_206: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_103, full_206);  amin_103 = full_206 = None
	        full_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_103, full_207);  amax_103 = full_207 = None
	        sub_4721: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_103, minimum_103);  maximum_103 = None
	        div_206: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4721, 255.0);  sub_4721 = None
	        clamp_min_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_206, 1.1920928955078125e-07);  div_206 = None
	        div_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_103, clamp_min_309);  minimum_103 = None
	        round_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_207);  div_207 = None
	        sub_4727: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_207);  round_207 = None
	        clamp_min_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4727, -128);  sub_4727 = None
	        clamp_max_206: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_310, 127);  clamp_min_310 = None
	        _assert_tensor_metadata_929 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_309, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_929 = None
	        _assert_tensor_metadata_930 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_206, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_930 = None
	        convert_element_type_618: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_206, torch.int8);  clamp_max_206 = None
	        view_1616: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_309, [sym_size_int, 1500, 1])
	        view_1617: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_618, [sym_size_int, 1500, 1])
	        reciprocal_103: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1616);  view_1616 = None
	        mul_10025: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_103, 1.0);  reciprocal_103 = None
	        mul_10028: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_15638, mul_10025);  mul_10025 = None
	        round_208: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10028);  mul_10028 = None
	        add_15877: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_208, view_1617);  round_208 = view_1617 = None
	        clamp_min_311: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15877, -128);  add_15877 = None
	        clamp_max_207: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_311, 127);  clamp_min_311 = None
	        _assert_tensor_metadata_931 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_207, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_931 = None
	        convert_element_type_619: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_207, torch.int8);  clamp_max_207 = None
	        view_1620: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_309, [sym_size_int, 1500, 1]);  clamp_min_309 = None
	        view_1621: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_618, [sym_size_int, 1500, 1]);  convert_element_type_618 = None
	        _assert_tensor_metadata_932 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_619, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_932 = None
	        convert_element_type_620: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_619, torch.float32);  convert_element_type_619 = None
	        _assert_tensor_metadata_933 = torch.ops.aten._assert_tensor_metadata.default(view_1621, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_933 = None
	        convert_element_type_621: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1621, torch.float32);  view_1621 = None
	        sub_4747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_620, convert_element_type_621);  convert_element_type_620 = convert_element_type_621 = None
	        mul_10050: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4747, view_1620);  sub_4747 = view_1620 = None
	        _assert_tensor_metadata_934 = torch.ops.aten._assert_tensor_metadata.default(mul_10050, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_934 = None
	        view_1623: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1624: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1625: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_935 = torch.ops.aten._assert_tensor_metadata.default(view_1623, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_935 = None
	        convert_element_type_622: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1623, torch.float32);  view_1623 = None
	        _assert_tensor_metadata_936 = torch.ops.aten._assert_tensor_metadata.default(view_1625, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_936 = None
	        convert_element_type_623: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1625, torch.float32);  view_1625 = None
	        sub_4751: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_622, convert_element_type_623);  convert_element_type_622 = convert_element_type_623 = None
	        mul_10055: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4751, view_1624);  sub_4751 = view_1624 = None
	        view_1626: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10055, [1280, 1280]);  mul_10055 = None
	        _assert_tensor_metadata_937 = torch.ops.aten._assert_tensor_metadata.default(view_1626, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_937 = None
	        permute_173: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1626, [1, 0]);  view_1626 = None
	        mul_10058: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1627: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10050, [mul_10058, 1280]);  mul_10050 = mul_10058 = None
	        mm_17: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1627, permute_173);  view_1627 = permute_173 = None
	        view_1628: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_17, [sym_size_int, 1500, 1280]);  mm_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1629: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1628, [sym_size_int, -1, 20, 64]);  view_1628 = None
	        permute_174: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1629, [0, 2, 1, 3]);  view_1629 = None
	        clone_139: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_174, memory_format = torch.contiguous_format);  permute_174 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_15638, [2])
	        amax_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_15638, [2])
	        full_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_104, full_208);  amin_104 = full_208 = None
	        full_209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_104, full_209);  amax_104 = full_209 = None
	        sub_4765: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_104, minimum_104);  maximum_104 = None
	        div_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4765, 255.0);  sub_4765 = None
	        clamp_min_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_208, 1.1920928955078125e-07);  div_208 = None
	        div_209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_104, clamp_min_312);  minimum_104 = None
	        round_209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_209);  div_209 = None
	        sub_4771: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_209);  round_209 = None
	        clamp_min_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4771, -128);  sub_4771 = None
	        clamp_max_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_313, 127);  clamp_min_313 = None
	        _assert_tensor_metadata_938 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_312, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_938 = None
	        _assert_tensor_metadata_939 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_208, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_939 = None
	        convert_element_type_624: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_208, torch.int8);  clamp_max_208 = None
	        view_1632: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_312, [sym_size_int, 1500, 1])
	        view_1633: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_624, [sym_size_int, 1500, 1])
	        reciprocal_104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1632);  view_1632 = None
	        mul_10124: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_104, 1.0);  reciprocal_104 = None
	        mul_10127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_15638, mul_10124);  add_15638 = mul_10124 = None
	        round_210: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10127);  mul_10127 = None
	        add_16025: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_210, view_1633);  round_210 = view_1633 = None
	        clamp_min_314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16025, -128);  add_16025 = None
	        clamp_max_209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_314, 127);  clamp_min_314 = None
	        _assert_tensor_metadata_940 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_209, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_940 = None
	        convert_element_type_625: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_209, torch.int8);  clamp_max_209 = None
	        view_1636: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_312, [sym_size_int, 1500, 1]);  clamp_min_312 = None
	        view_1637: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_624, [sym_size_int, 1500, 1]);  convert_element_type_624 = None
	        _assert_tensor_metadata_941 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_625, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_941 = None
	        convert_element_type_626: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_625, torch.float32);  convert_element_type_625 = None
	        _assert_tensor_metadata_942 = torch.ops.aten._assert_tensor_metadata.default(view_1637, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_942 = None
	        convert_element_type_627: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1637, torch.float32);  view_1637 = None
	        sub_4791: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_626, convert_element_type_627);  convert_element_type_626 = convert_element_type_627 = None
	        mul_10149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4791, view_1636);  sub_4791 = view_1636 = None
	        _assert_tensor_metadata_943 = torch.ops.aten._assert_tensor_metadata.default(mul_10149, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_943 = None
	        view_1639: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1640: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1641: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_944 = torch.ops.aten._assert_tensor_metadata.default(view_1639, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_944 = None
	        convert_element_type_628: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1639, torch.float32);  view_1639 = None
	        _assert_tensor_metadata_945 = torch.ops.aten._assert_tensor_metadata.default(view_1641, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_945 = None
	        convert_element_type_629: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1641, torch.float32);  view_1641 = None
	        sub_4795: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_628, convert_element_type_629);  convert_element_type_628 = convert_element_type_629 = None
	        mul_10154: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4795, view_1640);  sub_4795 = view_1640 = None
	        view_1642: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10154, [1280, 1280]);  mul_10154 = None
	        _assert_tensor_metadata_946 = torch.ops.aten._assert_tensor_metadata.default(view_1642, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_946 = None
	        mul_10159: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1643: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10149, [mul_10159, 1280]);  mul_10149 = mul_10159 = None
	        permute_175: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1642, [1, 0]);  view_1642 = None
	        addmm_86: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_17_self_attn_v_proj_bias, view_1643, permute_175);  model_audio_tower_layers_17_self_attn_v_proj_bias = view_1643 = permute_175 = None
	        view_1644: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_86, [sym_size_int, 1500, 1280]);  addmm_86 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1645: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1644, [sym_size_int, -1, 20, 64]);  view_1644 = None
	        permute_176: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1645, [0, 2, 1, 3]);  view_1645 = None
	        clone_140: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_176, memory_format = torch.contiguous_format);  permute_176 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_17 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_138, clone_139, clone_140, None, False, scale = 1.0);  clone_138 = clone_139 = clone_140 = None
	        getitem_138: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_17[0];  _scaled_dot_product_efficient_attention_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_177: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_138, [0, 2, 1, 3]);  getitem_138 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1646: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_177, [sym_size_int, 1500, -1]);  permute_177 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1646, [2])
	        amax_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1646, [2])
	        full_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_105, full_210);  amin_105 = full_210 = None
	        full_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_105, full_211);  amax_105 = full_211 = None
	        sub_4813: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_105, minimum_105);  maximum_105 = None
	        div_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4813, 255.0);  sub_4813 = None
	        clamp_min_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_210, 1.1920928955078125e-07);  div_210 = None
	        div_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_105, clamp_min_315);  minimum_105 = None
	        round_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_211);  div_211 = None
	        sub_4819: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_211);  round_211 = None
	        clamp_min_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4819, -128);  sub_4819 = None
	        clamp_max_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_316, 127);  clamp_min_316 = None
	        _assert_tensor_metadata_947 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_315, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_947 = None
	        _assert_tensor_metadata_948 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_210, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_948 = None
	        convert_element_type_630: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_210, torch.int8);  clamp_max_210 = None
	        view_1649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_315, [sym_size_int, 1500, 1])
	        view_1650: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_630, [sym_size_int, 1500, 1])
	        reciprocal_105: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1649);  view_1649 = None
	        mul_10229: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_105, 1.0);  reciprocal_105 = None
	        mul_10232: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1646, mul_10229);  view_1646 = mul_10229 = None
	        round_212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10232);  mul_10232 = None
	        add_16189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_212, view_1650);  round_212 = view_1650 = None
	        clamp_min_317: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16189, -128);  add_16189 = None
	        clamp_max_211: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_317, 127);  clamp_min_317 = None
	        _assert_tensor_metadata_949 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_211, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_949 = None
	        convert_element_type_631: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_211, torch.int8);  clamp_max_211 = None
	        view_1653: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_315, [sym_size_int, 1500, 1]);  clamp_min_315 = None
	        view_1654: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_630, [sym_size_int, 1500, 1]);  convert_element_type_630 = None
	        _assert_tensor_metadata_950 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_631, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_950 = None
	        convert_element_type_632: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_631, torch.float32);  convert_element_type_631 = None
	        _assert_tensor_metadata_951 = torch.ops.aten._assert_tensor_metadata.default(view_1654, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_951 = None
	        convert_element_type_633: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1654, torch.float32);  view_1654 = None
	        sub_4839: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_632, convert_element_type_633);  convert_element_type_632 = convert_element_type_633 = None
	        mul_10254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4839, view_1653);  sub_4839 = view_1653 = None
	        _assert_tensor_metadata_952 = torch.ops.aten._assert_tensor_metadata.default(mul_10254, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_952 = None
	        view_1656: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1657: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1658: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_953 = torch.ops.aten._assert_tensor_metadata.default(view_1656, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_953 = None
	        convert_element_type_634: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1656, torch.float32);  view_1656 = None
	        _assert_tensor_metadata_954 = torch.ops.aten._assert_tensor_metadata.default(view_1658, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_954 = None
	        convert_element_type_635: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1658, torch.float32);  view_1658 = None
	        sub_4843: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_634, convert_element_type_635);  convert_element_type_634 = convert_element_type_635 = None
	        mul_10259: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4843, view_1657);  sub_4843 = view_1657 = None
	        view_1659: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10259, [1280, 1280]);  mul_10259 = None
	        _assert_tensor_metadata_955 = torch.ops.aten._assert_tensor_metadata.default(view_1659, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_955 = None
	        mul_10264: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1660: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10254, [mul_10264, 1280]);  mul_10254 = mul_10264 = None
	        permute_178: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1659, [1, 0]);  view_1659 = None
	        addmm_87: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_17_self_attn_out_proj_bias, view_1660, permute_178);  model_audio_tower_layers_17_self_attn_out_proj_bias = view_1660 = permute_178 = None
	        view_1661: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_87, [sym_size_int, 1500, 1280]);  addmm_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_16252: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_15632, view_1661);  add_15632 = view_1661 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_16252, memory_format = torch.contiguous_format)
	        var_mean_35 = torch.ops.aten.var_mean.correction(clone_142, [2], correction = 0, keepdim = True)
	        getitem_142: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_35[0]
	        getitem_143: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_35[1];  var_mean_35 = None
	        add_16257: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_142, 1e-05);  getitem_142 = None
	        rsqrt_35: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_16257);  add_16257 = None
	        sub_4849: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_142, getitem_143);  clone_142 = getitem_143 = None
	        mul_10275: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4849, rsqrt_35);  sub_4849 = rsqrt_35 = None
	        mul_10276: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10275, model_audio_tower_layers_17_final_layer_norm_weight);  mul_10275 = model_audio_tower_layers_17_final_layer_norm_weight = None
	        add_16258: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_10276, model_audio_tower_layers_17_final_layer_norm_bias);  mul_10276 = model_audio_tower_layers_17_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_16258, [2])
	        amax_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_16258, [2])
	        full_212: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_106, full_212);  amin_106 = full_212 = None
	        full_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_106, full_213);  amax_106 = full_213 = None
	        sub_4860: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_106, minimum_106);  maximum_106 = None
	        div_212: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4860, 255.0);  sub_4860 = None
	        clamp_min_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_212, 1.1920928955078125e-07);  div_212 = None
	        div_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_106, clamp_min_318);  minimum_106 = None
	        round_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_213);  div_213 = None
	        sub_4866: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_213);  round_213 = None
	        clamp_min_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4866, -128);  sub_4866 = None
	        clamp_max_212: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_319, 127);  clamp_min_319 = None
	        _assert_tensor_metadata_956 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_318, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_956 = None
	        _assert_tensor_metadata_957 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_212, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_957 = None
	        convert_element_type_636: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_212, torch.int8);  clamp_max_212 = None
	        view_1664: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_318, [sym_size_int, 1500, 1])
	        view_1665: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_636, [sym_size_int, 1500, 1])
	        reciprocal_106: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1664);  view_1664 = None
	        mul_10324: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_106, 1.0);  reciprocal_106 = None
	        mul_10327: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_16258, mul_10324);  add_16258 = mul_10324 = None
	        round_214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10327);  mul_10327 = None
	        add_16345: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_214, view_1665);  round_214 = view_1665 = None
	        clamp_min_320: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16345, -128);  add_16345 = None
	        clamp_max_213: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_320, 127);  clamp_min_320 = None
	        _assert_tensor_metadata_958 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_213, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_958 = None
	        convert_element_type_637: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_213, torch.int8);  clamp_max_213 = None
	        view_1668: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_318, [sym_size_int, 1500, 1]);  clamp_min_318 = None
	        view_1669: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_636, [sym_size_int, 1500, 1]);  convert_element_type_636 = None
	        _assert_tensor_metadata_959 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_637, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_959 = None
	        convert_element_type_638: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_637, torch.float32);  convert_element_type_637 = None
	        _assert_tensor_metadata_960 = torch.ops.aten._assert_tensor_metadata.default(view_1669, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_960 = None
	        convert_element_type_639: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1669, torch.float32);  view_1669 = None
	        sub_4886: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_638, convert_element_type_639);  convert_element_type_638 = convert_element_type_639 = None
	        mul_10349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4886, view_1668);  sub_4886 = view_1668 = None
	        _assert_tensor_metadata_961 = torch.ops.aten._assert_tensor_metadata.default(mul_10349, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_961 = None
	        view_1671: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_17_fc1_parametrizations_weight_original0 = None
	        view_1672: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_17_fc1_parametrizations_weight_original1 = None
	        view_1673: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_17_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_962 = torch.ops.aten._assert_tensor_metadata.default(view_1671, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_962 = None
	        convert_element_type_640: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1671, torch.float32);  view_1671 = None
	        _assert_tensor_metadata_963 = torch.ops.aten._assert_tensor_metadata.default(view_1673, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_963 = None
	        convert_element_type_641: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1673, torch.float32);  view_1673 = None
	        sub_4890: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_640, convert_element_type_641);  convert_element_type_640 = convert_element_type_641 = None
	        mul_10354: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4890, view_1672);  sub_4890 = view_1672 = None
	        view_1674: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10354, [5120, 1280]);  mul_10354 = None
	        _assert_tensor_metadata_964 = torch.ops.aten._assert_tensor_metadata.default(view_1674, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_964 = None
	        mul_10359: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1675: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10349, [mul_10359, 1280]);  mul_10349 = mul_10359 = None
	        permute_179: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1674, [1, 0]);  view_1674 = None
	        addmm_88: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_17_fc1_bias, view_1675, permute_179);  model_audio_tower_layers_17_fc1_bias = view_1675 = permute_179 = None
	        view_1676: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_88, [sym_size_int, 1500, 5120]);  addmm_88 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_10366: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1676, 0.5)
	        mul_10367: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1676, 0.7071067811865476);  view_1676 = None
	        erf_19: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_10367);  mul_10367 = None
	        add_16404: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_19, 1);  erf_19 = None
	        mul_10368: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10366, add_16404);  mul_10366 = add_16404 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_10368, [2])
	        amax_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_10368, [2])
	        full_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_107, full_214);  amin_107 = full_214 = None
	        full_215: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_107, full_215);  amax_107 = full_215 = None
	        sub_4903: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_107, minimum_107);  maximum_107 = None
	        div_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4903, 255.0);  sub_4903 = None
	        clamp_min_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_214, 1.1920928955078125e-07);  div_214 = None
	        div_215: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_107, clamp_min_321);  minimum_107 = None
	        round_215: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_215);  div_215 = None
	        sub_4909: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_215);  round_215 = None
	        clamp_min_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4909, -128);  sub_4909 = None
	        clamp_max_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_322, 127);  clamp_min_322 = None
	        _assert_tensor_metadata_965 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_321, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_965 = None
	        _assert_tensor_metadata_966 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_214, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_966 = None
	        convert_element_type_642: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_214, torch.int8);  clamp_max_214 = None
	        view_1679: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_321, [sym_size_int, 1500, 1])
	        view_1680: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_642, [sym_size_int, 1500, 1])
	        reciprocal_107: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1679);  view_1679 = None
	        mul_10414: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_107, 1.0);  reciprocal_107 = None
	        mul_10417: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10368, mul_10414);  mul_10368 = mul_10414 = None
	        round_216: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_10417);  mul_10417 = None
	        add_16487: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_216, view_1680);  round_216 = view_1680 = None
	        clamp_min_323: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16487, -128);  add_16487 = None
	        clamp_max_215: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_323, 127);  clamp_min_323 = None
	        _assert_tensor_metadata_967 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_215, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_967 = None
	        convert_element_type_643: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_215, torch.int8);  clamp_max_215 = None
	        view_1683: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_321, [sym_size_int, 1500, 1]);  clamp_min_321 = None
	        view_1684: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_642, [sym_size_int, 1500, 1]);  convert_element_type_642 = None
	        _assert_tensor_metadata_968 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_643, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_968 = None
	        convert_element_type_644: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_643, torch.float32);  convert_element_type_643 = None
	        _assert_tensor_metadata_969 = torch.ops.aten._assert_tensor_metadata.default(view_1684, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_969 = None
	        convert_element_type_645: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1684, torch.float32);  view_1684 = None
	        sub_4929: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_644, convert_element_type_645);  convert_element_type_644 = convert_element_type_645 = None
	        mul_10439: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4929, view_1683);  sub_4929 = view_1683 = None
	        _assert_tensor_metadata_970 = torch.ops.aten._assert_tensor_metadata.default(mul_10439, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_970 = None
	        view_1686: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_17_fc2_parametrizations_weight_original0 = None
	        view_1687: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_17_fc2_parametrizations_weight_original1 = None
	        view_1688: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_17_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_971 = torch.ops.aten._assert_tensor_metadata.default(view_1686, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_971 = None
	        convert_element_type_646: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1686, torch.float32);  view_1686 = None
	        _assert_tensor_metadata_972 = torch.ops.aten._assert_tensor_metadata.default(view_1688, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_972 = None
	        convert_element_type_647: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1688, torch.float32);  view_1688 = None
	        sub_4933: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_646, convert_element_type_647);  convert_element_type_646 = convert_element_type_647 = None
	        mul_10444: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4933, view_1687);  sub_4933 = view_1687 = None
	        view_1689: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10444, [1280, 5120]);  mul_10444 = None
	        _assert_tensor_metadata_973 = torch.ops.aten._assert_tensor_metadata.default(view_1689, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_973 = None
	        mul_10449: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1690: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10439, [mul_10449, 5120]);  mul_10439 = mul_10449 = None
	        permute_180: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1689, [1, 0]);  view_1689 = None
	        addmm_89: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_17_fc2_bias, view_1690, permute_180);  model_audio_tower_layers_17_fc2_bias = view_1690 = permute_180 = None
	        view_1691: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_89, [sym_size_int, 1500, 1280]);  addmm_89 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_16550: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_16252, view_1691);  add_16252 = view_1691 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_145: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_16550, memory_format = torch.contiguous_format)
	        var_mean_36 = torch.ops.aten.var_mean.correction(clone_145, [2], correction = 0, keepdim = True)
	        getitem_144: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_36[0]
	        getitem_145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_36[1];  var_mean_36 = None
	        add_16555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_144, 1e-05);  getitem_144 = None
	        rsqrt_36: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_16555);  add_16555 = None
	        sub_4939: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_145, getitem_145);  clone_145 = getitem_145 = None
	        mul_10460: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4939, rsqrt_36);  sub_4939 = rsqrt_36 = None
	        mul_10461: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10460, model_audio_tower_layers_18_self_attn_layer_norm_weight);  mul_10460 = model_audio_tower_layers_18_self_attn_layer_norm_weight = None
	        add_16556: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_10461, model_audio_tower_layers_18_self_attn_layer_norm_bias);  mul_10461 = model_audio_tower_layers_18_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_16556, [2])
	        amax_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_16556, [2])
	        full_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_108, full_216);  amin_108 = full_216 = None
	        full_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_108, full_217);  amax_108 = full_217 = None
	        sub_4950: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_108, minimum_108);  maximum_108 = None
	        div_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4950, 255.0);  sub_4950 = None
	        clamp_min_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_216, 1.1920928955078125e-07);  div_216 = None
	        div_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_108, clamp_min_324);  minimum_108 = None
	        round_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_217);  div_217 = None
	        sub_4956: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_217);  round_217 = None
	        clamp_min_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4956, -128);  sub_4956 = None
	        clamp_max_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_325, 127);  clamp_min_325 = None
	        _assert_tensor_metadata_974 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_324, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_974 = None
	        _assert_tensor_metadata_975 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_216, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_975 = None
	        convert_element_type_648: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_216, torch.int8);  clamp_max_216 = None
	        view_1694: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_324, [sym_size_int, 1500, 1])
	        view_1695: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_648, [sym_size_int, 1500, 1])
	        reciprocal_108: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1694);  view_1694 = None
	        mul_10509: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_108, 1.0);  reciprocal_108 = None
	        mul_10512: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_16556, mul_10509);  mul_10509 = None
	        round_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10512);  mul_10512 = None
	        add_16643: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_218, view_1695);  round_218 = view_1695 = None
	        clamp_min_326: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16643, -128);  add_16643 = None
	        clamp_max_217: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_326, 127);  clamp_min_326 = None
	        _assert_tensor_metadata_976 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_217, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_976 = None
	        convert_element_type_649: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_217, torch.int8);  clamp_max_217 = None
	        view_1698: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_324, [sym_size_int, 1500, 1]);  clamp_min_324 = None
	        view_1699: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_648, [sym_size_int, 1500, 1]);  convert_element_type_648 = None
	        _assert_tensor_metadata_977 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_649, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_977 = None
	        convert_element_type_650: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_649, torch.float32);  convert_element_type_649 = None
	        _assert_tensor_metadata_978 = torch.ops.aten._assert_tensor_metadata.default(view_1699, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_978 = None
	        convert_element_type_651: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1699, torch.float32);  view_1699 = None
	        sub_4976: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_650, convert_element_type_651);  convert_element_type_650 = convert_element_type_651 = None
	        mul_10534: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4976, view_1698);  sub_4976 = view_1698 = None
	        _assert_tensor_metadata_979 = torch.ops.aten._assert_tensor_metadata.default(mul_10534, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_979 = None
	        view_1701: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1702: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1703: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_980 = torch.ops.aten._assert_tensor_metadata.default(view_1701, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_980 = None
	        convert_element_type_652: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1701, torch.float32);  view_1701 = None
	        _assert_tensor_metadata_981 = torch.ops.aten._assert_tensor_metadata.default(view_1703, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_981 = None
	        convert_element_type_653: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1703, torch.float32);  view_1703 = None
	        sub_4980: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_652, convert_element_type_653);  convert_element_type_652 = convert_element_type_653 = None
	        mul_10539: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4980, view_1702);  sub_4980 = view_1702 = None
	        view_1704: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10539, [1280, 1280]);  mul_10539 = None
	        _assert_tensor_metadata_982 = torch.ops.aten._assert_tensor_metadata.default(view_1704, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_982 = None
	        mul_10544: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1705: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10534, [mul_10544, 1280]);  mul_10534 = mul_10544 = None
	        permute_181: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1704, [1, 0]);  view_1704 = None
	        addmm_90: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_18_self_attn_q_proj_bias, view_1705, permute_181);  model_audio_tower_layers_18_self_attn_q_proj_bias = view_1705 = permute_181 = None
	        view_1706: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_90, [sym_size_int, 1500, 1280]);  addmm_90 = None
	        mul_10551: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1706, 0.125);  view_1706 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1707: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10551, [sym_size_int, 1500, 20, 64]);  mul_10551 = None
	        permute_182: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1707, [0, 2, 1, 3]);  view_1707 = None
	        clone_146: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_182, memory_format = torch.contiguous_format);  permute_182 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_16556, [2])
	        amax_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_16556, [2])
	        full_218: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_109, full_218);  amin_109 = full_218 = None
	        full_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_109, full_219);  amax_109 = full_219 = None
	        sub_4995: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_109, minimum_109);  maximum_109 = None
	        div_218: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4995, 255.0);  sub_4995 = None
	        clamp_min_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_218, 1.1920928955078125e-07);  div_218 = None
	        div_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_109, clamp_min_327);  minimum_109 = None
	        round_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_219);  div_219 = None
	        sub_5001: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_219);  round_219 = None
	        clamp_min_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5001, -128);  sub_5001 = None
	        clamp_max_218: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_328, 127);  clamp_min_328 = None
	        _assert_tensor_metadata_983 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_327, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_983 = None
	        _assert_tensor_metadata_984 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_218, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_984 = None
	        convert_element_type_654: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_218, torch.int8);  clamp_max_218 = None
	        view_1710: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_327, [sym_size_int, 1500, 1])
	        view_1711: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_654, [sym_size_int, 1500, 1])
	        reciprocal_109: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1710);  view_1710 = None
	        mul_10605: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_109, 1.0);  reciprocal_109 = None
	        mul_10608: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_16556, mul_10605);  mul_10605 = None
	        round_220: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10608);  mul_10608 = None
	        add_16795: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_220, view_1711);  round_220 = view_1711 = None
	        clamp_min_329: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16795, -128);  add_16795 = None
	        clamp_max_219: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_329, 127);  clamp_min_329 = None
	        _assert_tensor_metadata_985 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_219, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_985 = None
	        convert_element_type_655: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_219, torch.int8);  clamp_max_219 = None
	        view_1714: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_327, [sym_size_int, 1500, 1]);  clamp_min_327 = None
	        view_1715: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_654, [sym_size_int, 1500, 1]);  convert_element_type_654 = None
	        _assert_tensor_metadata_986 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_655, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_986 = None
	        convert_element_type_656: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_655, torch.float32);  convert_element_type_655 = None
	        _assert_tensor_metadata_987 = torch.ops.aten._assert_tensor_metadata.default(view_1715, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_987 = None
	        convert_element_type_657: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1715, torch.float32);  view_1715 = None
	        sub_5021: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_656, convert_element_type_657);  convert_element_type_656 = convert_element_type_657 = None
	        mul_10630: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5021, view_1714);  sub_5021 = view_1714 = None
	        _assert_tensor_metadata_988 = torch.ops.aten._assert_tensor_metadata.default(mul_10630, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_988 = None
	        view_1717: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1718: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1719: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_989 = torch.ops.aten._assert_tensor_metadata.default(view_1717, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_989 = None
	        convert_element_type_658: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1717, torch.float32);  view_1717 = None
	        _assert_tensor_metadata_990 = torch.ops.aten._assert_tensor_metadata.default(view_1719, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_990 = None
	        convert_element_type_659: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1719, torch.float32);  view_1719 = None
	        sub_5025: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_658, convert_element_type_659);  convert_element_type_658 = convert_element_type_659 = None
	        mul_10635: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5025, view_1718);  sub_5025 = view_1718 = None
	        view_1720: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10635, [1280, 1280]);  mul_10635 = None
	        _assert_tensor_metadata_991 = torch.ops.aten._assert_tensor_metadata.default(view_1720, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_991 = None
	        permute_183: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1720, [1, 0]);  view_1720 = None
	        mul_10638: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1721: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10630, [mul_10638, 1280]);  mul_10630 = mul_10638 = None
	        mm_18: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1721, permute_183);  view_1721 = permute_183 = None
	        view_1722: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_18, [sym_size_int, 1500, 1280]);  mm_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1723: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1722, [sym_size_int, -1, 20, 64]);  view_1722 = None
	        permute_184: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1723, [0, 2, 1, 3]);  view_1723 = None
	        clone_147: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_184, memory_format = torch.contiguous_format);  permute_184 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_16556, [2])
	        amax_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_16556, [2])
	        full_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_110, full_220);  amin_110 = full_220 = None
	        full_221: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_110, full_221);  amax_110 = full_221 = None
	        sub_5039: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_110, minimum_110);  maximum_110 = None
	        div_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5039, 255.0);  sub_5039 = None
	        clamp_min_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_220, 1.1920928955078125e-07);  div_220 = None
	        div_221: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_110, clamp_min_330);  minimum_110 = None
	        round_221: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_221);  div_221 = None
	        sub_5045: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_221);  round_221 = None
	        clamp_min_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5045, -128);  sub_5045 = None
	        clamp_max_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_331, 127);  clamp_min_331 = None
	        _assert_tensor_metadata_992 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_330, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_992 = None
	        _assert_tensor_metadata_993 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_220, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_993 = None
	        convert_element_type_660: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_220, torch.int8);  clamp_max_220 = None
	        view_1726: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_330, [sym_size_int, 1500, 1])
	        view_1727: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_660, [sym_size_int, 1500, 1])
	        reciprocal_110: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1726);  view_1726 = None
	        mul_10704: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_110, 1.0);  reciprocal_110 = None
	        mul_10707: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_16556, mul_10704);  add_16556 = mul_10704 = None
	        round_222: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10707);  mul_10707 = None
	        add_16943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_222, view_1727);  round_222 = view_1727 = None
	        clamp_min_332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16943, -128);  add_16943 = None
	        clamp_max_221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_332, 127);  clamp_min_332 = None
	        _assert_tensor_metadata_994 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_221, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_994 = None
	        convert_element_type_661: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_221, torch.int8);  clamp_max_221 = None
	        view_1730: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_330, [sym_size_int, 1500, 1]);  clamp_min_330 = None
	        view_1731: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_660, [sym_size_int, 1500, 1]);  convert_element_type_660 = None
	        _assert_tensor_metadata_995 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_661, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_995 = None
	        convert_element_type_662: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_661, torch.float32);  convert_element_type_661 = None
	        _assert_tensor_metadata_996 = torch.ops.aten._assert_tensor_metadata.default(view_1731, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_996 = None
	        convert_element_type_663: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1731, torch.float32);  view_1731 = None
	        sub_5065: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_662, convert_element_type_663);  convert_element_type_662 = convert_element_type_663 = None
	        mul_10729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5065, view_1730);  sub_5065 = view_1730 = None
	        _assert_tensor_metadata_997 = torch.ops.aten._assert_tensor_metadata.default(mul_10729, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_997 = None
	        view_1733: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1734: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1735: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_998 = torch.ops.aten._assert_tensor_metadata.default(view_1733, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_998 = None
	        convert_element_type_664: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1733, torch.float32);  view_1733 = None
	        _assert_tensor_metadata_999 = torch.ops.aten._assert_tensor_metadata.default(view_1735, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_999 = None
	        convert_element_type_665: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1735, torch.float32);  view_1735 = None
	        sub_5069: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_664, convert_element_type_665);  convert_element_type_664 = convert_element_type_665 = None
	        mul_10734: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5069, view_1734);  sub_5069 = view_1734 = None
	        view_1736: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10734, [1280, 1280]);  mul_10734 = None
	        _assert_tensor_metadata_1000 = torch.ops.aten._assert_tensor_metadata.default(view_1736, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1000 = None
	        mul_10739: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1737: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10729, [mul_10739, 1280]);  mul_10729 = mul_10739 = None
	        permute_185: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1736, [1, 0]);  view_1736 = None
	        addmm_91: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_18_self_attn_v_proj_bias, view_1737, permute_185);  model_audio_tower_layers_18_self_attn_v_proj_bias = view_1737 = permute_185 = None
	        view_1738: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_91, [sym_size_int, 1500, 1280]);  addmm_91 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1739: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1738, [sym_size_int, -1, 20, 64]);  view_1738 = None
	        permute_186: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1739, [0, 2, 1, 3]);  view_1739 = None
	        clone_148: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_186, memory_format = torch.contiguous_format);  permute_186 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_18 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_146, clone_147, clone_148, None, False, scale = 1.0);  clone_146 = clone_147 = clone_148 = None
	        getitem_146: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_18[0];  _scaled_dot_product_efficient_attention_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_187: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_146, [0, 2, 1, 3]);  getitem_146 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1740: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_187, [sym_size_int, 1500, -1]);  permute_187 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1740, [2])
	        amax_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1740, [2])
	        full_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_111, full_222);  amin_111 = full_222 = None
	        full_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_111, full_223);  amax_111 = full_223 = None
	        sub_5087: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_111, minimum_111);  maximum_111 = None
	        div_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5087, 255.0);  sub_5087 = None
	        clamp_min_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_222, 1.1920928955078125e-07);  div_222 = None
	        div_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_111, clamp_min_333);  minimum_111 = None
	        round_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_223);  div_223 = None
	        sub_5093: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_223);  round_223 = None
	        clamp_min_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5093, -128);  sub_5093 = None
	        clamp_max_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_334, 127);  clamp_min_334 = None
	        _assert_tensor_metadata_1001 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_333, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1001 = None
	        _assert_tensor_metadata_1002 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_222, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1002 = None
	        convert_element_type_666: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_222, torch.int8);  clamp_max_222 = None
	        view_1743: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_333, [sym_size_int, 1500, 1])
	        view_1744: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_666, [sym_size_int, 1500, 1])
	        reciprocal_111: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1743);  view_1743 = None
	        mul_10809: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_111, 1.0);  reciprocal_111 = None
	        mul_10812: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1740, mul_10809);  view_1740 = mul_10809 = None
	        round_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10812);  mul_10812 = None
	        add_17107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_224, view_1744);  round_224 = view_1744 = None
	        clamp_min_335: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17107, -128);  add_17107 = None
	        clamp_max_223: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_335, 127);  clamp_min_335 = None
	        _assert_tensor_metadata_1003 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_223, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1003 = None
	        convert_element_type_667: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_223, torch.int8);  clamp_max_223 = None
	        view_1747: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_333, [sym_size_int, 1500, 1]);  clamp_min_333 = None
	        view_1748: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_666, [sym_size_int, 1500, 1]);  convert_element_type_666 = None
	        _assert_tensor_metadata_1004 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_667, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1004 = None
	        convert_element_type_668: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_667, torch.float32);  convert_element_type_667 = None
	        _assert_tensor_metadata_1005 = torch.ops.aten._assert_tensor_metadata.default(view_1748, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1005 = None
	        convert_element_type_669: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1748, torch.float32);  view_1748 = None
	        sub_5113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_668, convert_element_type_669);  convert_element_type_668 = convert_element_type_669 = None
	        mul_10834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5113, view_1747);  sub_5113 = view_1747 = None
	        _assert_tensor_metadata_1006 = torch.ops.aten._assert_tensor_metadata.default(mul_10834, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1006 = None
	        view_1750: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1751: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1752: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1007 = torch.ops.aten._assert_tensor_metadata.default(view_1750, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1007 = None
	        convert_element_type_670: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1750, torch.float32);  view_1750 = None
	        _assert_tensor_metadata_1008 = torch.ops.aten._assert_tensor_metadata.default(view_1752, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1008 = None
	        convert_element_type_671: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1752, torch.float32);  view_1752 = None
	        sub_5117: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_670, convert_element_type_671);  convert_element_type_670 = convert_element_type_671 = None
	        mul_10839: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5117, view_1751);  sub_5117 = view_1751 = None
	        view_1753: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10839, [1280, 1280]);  mul_10839 = None
	        _assert_tensor_metadata_1009 = torch.ops.aten._assert_tensor_metadata.default(view_1753, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1009 = None
	        mul_10844: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1754: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10834, [mul_10844, 1280]);  mul_10834 = mul_10844 = None
	        permute_188: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1753, [1, 0]);  view_1753 = None
	        addmm_92: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_18_self_attn_out_proj_bias, view_1754, permute_188);  model_audio_tower_layers_18_self_attn_out_proj_bias = view_1754 = permute_188 = None
	        view_1755: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_92, [sym_size_int, 1500, 1280]);  addmm_92 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_17170: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_16550, view_1755);  add_16550 = view_1755 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_150: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_17170, memory_format = torch.contiguous_format)
	        var_mean_37 = torch.ops.aten.var_mean.correction(clone_150, [2], correction = 0, keepdim = True)
	        getitem_150: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_37[0]
	        getitem_151: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_37[1];  var_mean_37 = None
	        add_17175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_150, 1e-05);  getitem_150 = None
	        rsqrt_37: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_17175);  add_17175 = None
	        sub_5123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_150, getitem_151);  clone_150 = getitem_151 = None
	        mul_10855: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5123, rsqrt_37);  sub_5123 = rsqrt_37 = None
	        mul_10856: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10855, model_audio_tower_layers_18_final_layer_norm_weight);  mul_10855 = model_audio_tower_layers_18_final_layer_norm_weight = None
	        add_17176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_10856, model_audio_tower_layers_18_final_layer_norm_bias);  mul_10856 = model_audio_tower_layers_18_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_17176, [2])
	        amax_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_17176, [2])
	        full_224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_112, full_224);  amin_112 = full_224 = None
	        full_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_112, full_225);  amax_112 = full_225 = None
	        sub_5134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_112, minimum_112);  maximum_112 = None
	        div_224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5134, 255.0);  sub_5134 = None
	        clamp_min_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_224, 1.1920928955078125e-07);  div_224 = None
	        div_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_112, clamp_min_336);  minimum_112 = None
	        round_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_225);  div_225 = None
	        sub_5140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_225);  round_225 = None
	        clamp_min_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5140, -128);  sub_5140 = None
	        clamp_max_224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_337, 127);  clamp_min_337 = None
	        _assert_tensor_metadata_1010 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_336, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1010 = None
	        _assert_tensor_metadata_1011 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_224, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1011 = None
	        convert_element_type_672: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_224, torch.int8);  clamp_max_224 = None
	        view_1758: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_336, [sym_size_int, 1500, 1])
	        view_1759: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_672, [sym_size_int, 1500, 1])
	        reciprocal_112: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1758);  view_1758 = None
	        mul_10904: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_112, 1.0);  reciprocal_112 = None
	        mul_10907: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_17176, mul_10904);  add_17176 = mul_10904 = None
	        round_226: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10907);  mul_10907 = None
	        add_17263: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_226, view_1759);  round_226 = view_1759 = None
	        clamp_min_338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17263, -128);  add_17263 = None
	        clamp_max_225: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_338, 127);  clamp_min_338 = None
	        _assert_tensor_metadata_1012 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_225, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1012 = None
	        convert_element_type_673: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_225, torch.int8);  clamp_max_225 = None
	        view_1762: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_336, [sym_size_int, 1500, 1]);  clamp_min_336 = None
	        view_1763: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_672, [sym_size_int, 1500, 1]);  convert_element_type_672 = None
	        _assert_tensor_metadata_1013 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_673, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1013 = None
	        convert_element_type_674: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_673, torch.float32);  convert_element_type_673 = None
	        _assert_tensor_metadata_1014 = torch.ops.aten._assert_tensor_metadata.default(view_1763, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1014 = None
	        convert_element_type_675: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1763, torch.float32);  view_1763 = None
	        sub_5160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_674, convert_element_type_675);  convert_element_type_674 = convert_element_type_675 = None
	        mul_10929: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5160, view_1762);  sub_5160 = view_1762 = None
	        _assert_tensor_metadata_1015 = torch.ops.aten._assert_tensor_metadata.default(mul_10929, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1015 = None
	        view_1765: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_18_fc1_parametrizations_weight_original0 = None
	        view_1766: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_18_fc1_parametrizations_weight_original1 = None
	        view_1767: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_18_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1016 = torch.ops.aten._assert_tensor_metadata.default(view_1765, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1016 = None
	        convert_element_type_676: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1765, torch.float32);  view_1765 = None
	        _assert_tensor_metadata_1017 = torch.ops.aten._assert_tensor_metadata.default(view_1767, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1017 = None
	        convert_element_type_677: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1767, torch.float32);  view_1767 = None
	        sub_5164: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_676, convert_element_type_677);  convert_element_type_676 = convert_element_type_677 = None
	        mul_10934: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5164, view_1766);  sub_5164 = view_1766 = None
	        view_1768: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10934, [5120, 1280]);  mul_10934 = None
	        _assert_tensor_metadata_1018 = torch.ops.aten._assert_tensor_metadata.default(view_1768, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1018 = None
	        mul_10939: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1769: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10929, [mul_10939, 1280]);  mul_10929 = mul_10939 = None
	        permute_189: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1768, [1, 0]);  view_1768 = None
	        addmm_93: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_18_fc1_bias, view_1769, permute_189);  model_audio_tower_layers_18_fc1_bias = view_1769 = permute_189 = None
	        view_1770: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_93, [sym_size_int, 1500, 5120]);  addmm_93 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_10946: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1770, 0.5)
	        mul_10947: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1770, 0.7071067811865476);  view_1770 = None
	        erf_20: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_10947);  mul_10947 = None
	        add_17322: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_20, 1);  erf_20 = None
	        mul_10948: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10946, add_17322);  mul_10946 = add_17322 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_10948, [2])
	        amax_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_10948, [2])
	        full_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_113, full_226);  amin_113 = full_226 = None
	        full_227: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_113, full_227);  amax_113 = full_227 = None
	        sub_5177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_113, minimum_113);  maximum_113 = None
	        div_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5177, 255.0);  sub_5177 = None
	        clamp_min_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_226, 1.1920928955078125e-07);  div_226 = None
	        div_227: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_113, clamp_min_339);  minimum_113 = None
	        round_227: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_227);  div_227 = None
	        sub_5183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_227);  round_227 = None
	        clamp_min_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5183, -128);  sub_5183 = None
	        clamp_max_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_340, 127);  clamp_min_340 = None
	        _assert_tensor_metadata_1019 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_339, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1019 = None
	        _assert_tensor_metadata_1020 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_226, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1020 = None
	        convert_element_type_678: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_226, torch.int8);  clamp_max_226 = None
	        view_1773: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_339, [sym_size_int, 1500, 1])
	        view_1774: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_678, [sym_size_int, 1500, 1])
	        reciprocal_113: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1773);  view_1773 = None
	        mul_10994: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_113, 1.0);  reciprocal_113 = None
	        mul_10997: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10948, mul_10994);  mul_10948 = mul_10994 = None
	        round_228: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_10997);  mul_10997 = None
	        add_17405: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_228, view_1774);  round_228 = view_1774 = None
	        clamp_min_341: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17405, -128);  add_17405 = None
	        clamp_max_227: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_341, 127);  clamp_min_341 = None
	        _assert_tensor_metadata_1021 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_227, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1021 = None
	        convert_element_type_679: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_227, torch.int8);  clamp_max_227 = None
	        view_1777: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_339, [sym_size_int, 1500, 1]);  clamp_min_339 = None
	        view_1778: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_678, [sym_size_int, 1500, 1]);  convert_element_type_678 = None
	        _assert_tensor_metadata_1022 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_679, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1022 = None
	        convert_element_type_680: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_679, torch.float32);  convert_element_type_679 = None
	        _assert_tensor_metadata_1023 = torch.ops.aten._assert_tensor_metadata.default(view_1778, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1023 = None
	        convert_element_type_681: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1778, torch.float32);  view_1778 = None
	        sub_5203: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_680, convert_element_type_681);  convert_element_type_680 = convert_element_type_681 = None
	        mul_11019: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5203, view_1777);  sub_5203 = view_1777 = None
	        _assert_tensor_metadata_1024 = torch.ops.aten._assert_tensor_metadata.default(mul_11019, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1024 = None
	        view_1780: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_18_fc2_parametrizations_weight_original0 = None
	        view_1781: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_18_fc2_parametrizations_weight_original1 = None
	        view_1782: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_18_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1025 = torch.ops.aten._assert_tensor_metadata.default(view_1780, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1025 = None
	        convert_element_type_682: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1780, torch.float32);  view_1780 = None
	        _assert_tensor_metadata_1026 = torch.ops.aten._assert_tensor_metadata.default(view_1782, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1026 = None
	        convert_element_type_683: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1782, torch.float32);  view_1782 = None
	        sub_5207: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_682, convert_element_type_683);  convert_element_type_682 = convert_element_type_683 = None
	        mul_11024: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5207, view_1781);  sub_5207 = view_1781 = None
	        view_1783: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11024, [1280, 5120]);  mul_11024 = None
	        _assert_tensor_metadata_1027 = torch.ops.aten._assert_tensor_metadata.default(view_1783, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1027 = None
	        mul_11029: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1784: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11019, [mul_11029, 5120]);  mul_11019 = mul_11029 = None
	        permute_190: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1783, [1, 0]);  view_1783 = None
	        addmm_94: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_18_fc2_bias, view_1784, permute_190);  model_audio_tower_layers_18_fc2_bias = view_1784 = permute_190 = None
	        view_1785: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_94, [sym_size_int, 1500, 1280]);  addmm_94 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_17468: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_17170, view_1785);  add_17170 = view_1785 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_153: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_17468, memory_format = torch.contiguous_format)
	        var_mean_38 = torch.ops.aten.var_mean.correction(clone_153, [2], correction = 0, keepdim = True)
	        getitem_152: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_38[0]
	        getitem_153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_38[1];  var_mean_38 = None
	        add_17473: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_152, 1e-05);  getitem_152 = None
	        rsqrt_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_17473);  add_17473 = None
	        sub_5213: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_153, getitem_153);  clone_153 = getitem_153 = None
	        mul_11040: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5213, rsqrt_38);  sub_5213 = rsqrt_38 = None
	        mul_11041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11040, model_audio_tower_layers_19_self_attn_layer_norm_weight);  mul_11040 = model_audio_tower_layers_19_self_attn_layer_norm_weight = None
	        add_17474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_11041, model_audio_tower_layers_19_self_attn_layer_norm_bias);  mul_11041 = model_audio_tower_layers_19_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_17474, [2])
	        amax_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_17474, [2])
	        full_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_114, full_228);  amin_114 = full_228 = None
	        full_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_114, full_229);  amax_114 = full_229 = None
	        sub_5224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_114, minimum_114);  maximum_114 = None
	        div_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5224, 255.0);  sub_5224 = None
	        clamp_min_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_228, 1.1920928955078125e-07);  div_228 = None
	        div_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_114, clamp_min_342);  minimum_114 = None
	        round_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_229);  div_229 = None
	        sub_5230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_229);  round_229 = None
	        clamp_min_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5230, -128);  sub_5230 = None
	        clamp_max_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_343, 127);  clamp_min_343 = None
	        _assert_tensor_metadata_1028 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_342, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1028 = None
	        _assert_tensor_metadata_1029 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_228, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1029 = None
	        convert_element_type_684: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_228, torch.int8);  clamp_max_228 = None
	        view_1788: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_342, [sym_size_int, 1500, 1])
	        view_1789: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_684, [sym_size_int, 1500, 1])
	        reciprocal_114: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1788);  view_1788 = None
	        mul_11089: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_114, 1.0);  reciprocal_114 = None
	        mul_11092: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_17474, mul_11089);  mul_11089 = None
	        round_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11092);  mul_11092 = None
	        add_17561: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_230, view_1789);  round_230 = view_1789 = None
	        clamp_min_344: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17561, -128);  add_17561 = None
	        clamp_max_229: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_344, 127);  clamp_min_344 = None
	        _assert_tensor_metadata_1030 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_229, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1030 = None
	        convert_element_type_685: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_229, torch.int8);  clamp_max_229 = None
	        view_1792: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_342, [sym_size_int, 1500, 1]);  clamp_min_342 = None
	        view_1793: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_684, [sym_size_int, 1500, 1]);  convert_element_type_684 = None
	        _assert_tensor_metadata_1031 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_685, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1031 = None
	        convert_element_type_686: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_685, torch.float32);  convert_element_type_685 = None
	        _assert_tensor_metadata_1032 = torch.ops.aten._assert_tensor_metadata.default(view_1793, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1032 = None
	        convert_element_type_687: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1793, torch.float32);  view_1793 = None
	        sub_5250: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_686, convert_element_type_687);  convert_element_type_686 = convert_element_type_687 = None
	        mul_11114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5250, view_1792);  sub_5250 = view_1792 = None
	        _assert_tensor_metadata_1033 = torch.ops.aten._assert_tensor_metadata.default(mul_11114, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1033 = None
	        view_1795: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1796: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1797: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1034 = torch.ops.aten._assert_tensor_metadata.default(view_1795, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1034 = None
	        convert_element_type_688: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1795, torch.float32);  view_1795 = None
	        _assert_tensor_metadata_1035 = torch.ops.aten._assert_tensor_metadata.default(view_1797, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1035 = None
	        convert_element_type_689: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1797, torch.float32);  view_1797 = None
	        sub_5254: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_688, convert_element_type_689);  convert_element_type_688 = convert_element_type_689 = None
	        mul_11119: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5254, view_1796);  sub_5254 = view_1796 = None
	        view_1798: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11119, [1280, 1280]);  mul_11119 = None
	        _assert_tensor_metadata_1036 = torch.ops.aten._assert_tensor_metadata.default(view_1798, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1036 = None
	        mul_11124: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1799: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11114, [mul_11124, 1280]);  mul_11114 = mul_11124 = None
	        permute_191: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1798, [1, 0]);  view_1798 = None
	        addmm_95: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_19_self_attn_q_proj_bias, view_1799, permute_191);  model_audio_tower_layers_19_self_attn_q_proj_bias = view_1799 = permute_191 = None
	        view_1800: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_95, [sym_size_int, 1500, 1280]);  addmm_95 = None
	        mul_11131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1800, 0.125);  view_1800 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1801: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11131, [sym_size_int, 1500, 20, 64]);  mul_11131 = None
	        permute_192: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1801, [0, 2, 1, 3]);  view_1801 = None
	        clone_154: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_192, memory_format = torch.contiguous_format);  permute_192 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_17474, [2])
	        amax_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_17474, [2])
	        full_230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_115, full_230);  amin_115 = full_230 = None
	        full_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_115, full_231);  amax_115 = full_231 = None
	        sub_5269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_115, minimum_115);  maximum_115 = None
	        div_230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5269, 255.0);  sub_5269 = None
	        clamp_min_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_230, 1.1920928955078125e-07);  div_230 = None
	        div_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_115, clamp_min_345);  minimum_115 = None
	        round_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_231);  div_231 = None
	        sub_5275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_231);  round_231 = None
	        clamp_min_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5275, -128);  sub_5275 = None
	        clamp_max_230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_346, 127);  clamp_min_346 = None
	        _assert_tensor_metadata_1037 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_345, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1037 = None
	        _assert_tensor_metadata_1038 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_230, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1038 = None
	        convert_element_type_690: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_230, torch.int8);  clamp_max_230 = None
	        view_1804: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_345, [sym_size_int, 1500, 1])
	        view_1805: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_690, [sym_size_int, 1500, 1])
	        reciprocal_115: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1804);  view_1804 = None
	        mul_11185: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_115, 1.0);  reciprocal_115 = None
	        mul_11188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_17474, mul_11185);  mul_11185 = None
	        round_232: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11188);  mul_11188 = None
	        add_17713: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_232, view_1805);  round_232 = view_1805 = None
	        clamp_min_347: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17713, -128);  add_17713 = None
	        clamp_max_231: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_347, 127);  clamp_min_347 = None
	        _assert_tensor_metadata_1039 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_231, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1039 = None
	        convert_element_type_691: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_231, torch.int8);  clamp_max_231 = None
	        view_1808: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_345, [sym_size_int, 1500, 1]);  clamp_min_345 = None
	        view_1809: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_690, [sym_size_int, 1500, 1]);  convert_element_type_690 = None
	        _assert_tensor_metadata_1040 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_691, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1040 = None
	        convert_element_type_692: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_691, torch.float32);  convert_element_type_691 = None
	        _assert_tensor_metadata_1041 = torch.ops.aten._assert_tensor_metadata.default(view_1809, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1041 = None
	        convert_element_type_693: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1809, torch.float32);  view_1809 = None
	        sub_5295: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_692, convert_element_type_693);  convert_element_type_692 = convert_element_type_693 = None
	        mul_11210: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5295, view_1808);  sub_5295 = view_1808 = None
	        _assert_tensor_metadata_1042 = torch.ops.aten._assert_tensor_metadata.default(mul_11210, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1042 = None
	        view_1811: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1812: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1813: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1043 = torch.ops.aten._assert_tensor_metadata.default(view_1811, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1043 = None
	        convert_element_type_694: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1811, torch.float32);  view_1811 = None
	        _assert_tensor_metadata_1044 = torch.ops.aten._assert_tensor_metadata.default(view_1813, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1044 = None
	        convert_element_type_695: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1813, torch.float32);  view_1813 = None
	        sub_5299: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_694, convert_element_type_695);  convert_element_type_694 = convert_element_type_695 = None
	        mul_11215: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5299, view_1812);  sub_5299 = view_1812 = None
	        view_1814: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11215, [1280, 1280]);  mul_11215 = None
	        _assert_tensor_metadata_1045 = torch.ops.aten._assert_tensor_metadata.default(view_1814, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1045 = None
	        permute_193: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1814, [1, 0]);  view_1814 = None
	        mul_11218: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1815: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11210, [mul_11218, 1280]);  mul_11210 = mul_11218 = None
	        mm_19: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1815, permute_193);  view_1815 = permute_193 = None
	        view_1816: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_19, [sym_size_int, 1500, 1280]);  mm_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1817: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1816, [sym_size_int, -1, 20, 64]);  view_1816 = None
	        permute_194: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1817, [0, 2, 1, 3]);  view_1817 = None
	        clone_155: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_194, memory_format = torch.contiguous_format);  permute_194 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_17474, [2])
	        amax_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_17474, [2])
	        full_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_116, full_232);  amin_116 = full_232 = None
	        full_233: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_116, full_233);  amax_116 = full_233 = None
	        sub_5313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_116, minimum_116);  maximum_116 = None
	        div_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5313, 255.0);  sub_5313 = None
	        clamp_min_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_232, 1.1920928955078125e-07);  div_232 = None
	        div_233: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_116, clamp_min_348);  minimum_116 = None
	        round_233: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_233);  div_233 = None
	        sub_5319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_233);  round_233 = None
	        clamp_min_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5319, -128);  sub_5319 = None
	        clamp_max_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_349, 127);  clamp_min_349 = None
	        _assert_tensor_metadata_1046 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_348, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1046 = None
	        _assert_tensor_metadata_1047 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_232, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1047 = None
	        convert_element_type_696: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_232, torch.int8);  clamp_max_232 = None
	        view_1820: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_348, [sym_size_int, 1500, 1])
	        view_1821: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_696, [sym_size_int, 1500, 1])
	        reciprocal_116: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1820);  view_1820 = None
	        mul_11284: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_116, 1.0);  reciprocal_116 = None
	        mul_11287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_17474, mul_11284);  add_17474 = mul_11284 = None
	        round_234: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11287);  mul_11287 = None
	        add_17861: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_234, view_1821);  round_234 = view_1821 = None
	        clamp_min_350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17861, -128);  add_17861 = None
	        clamp_max_233: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_350, 127);  clamp_min_350 = None
	        _assert_tensor_metadata_1048 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_233, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1048 = None
	        convert_element_type_697: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_233, torch.int8);  clamp_max_233 = None
	        view_1824: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_348, [sym_size_int, 1500, 1]);  clamp_min_348 = None
	        view_1825: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_696, [sym_size_int, 1500, 1]);  convert_element_type_696 = None
	        _assert_tensor_metadata_1049 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_697, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1049 = None
	        convert_element_type_698: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_697, torch.float32);  convert_element_type_697 = None
	        _assert_tensor_metadata_1050 = torch.ops.aten._assert_tensor_metadata.default(view_1825, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1050 = None
	        convert_element_type_699: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1825, torch.float32);  view_1825 = None
	        sub_5339: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_698, convert_element_type_699);  convert_element_type_698 = convert_element_type_699 = None
	        mul_11309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5339, view_1824);  sub_5339 = view_1824 = None
	        _assert_tensor_metadata_1051 = torch.ops.aten._assert_tensor_metadata.default(mul_11309, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1051 = None
	        view_1827: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1828: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1829: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1052 = torch.ops.aten._assert_tensor_metadata.default(view_1827, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1052 = None
	        convert_element_type_700: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1827, torch.float32);  view_1827 = None
	        _assert_tensor_metadata_1053 = torch.ops.aten._assert_tensor_metadata.default(view_1829, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1053 = None
	        convert_element_type_701: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1829, torch.float32);  view_1829 = None
	        sub_5343: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_700, convert_element_type_701);  convert_element_type_700 = convert_element_type_701 = None
	        mul_11314: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5343, view_1828);  sub_5343 = view_1828 = None
	        view_1830: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11314, [1280, 1280]);  mul_11314 = None
	        _assert_tensor_metadata_1054 = torch.ops.aten._assert_tensor_metadata.default(view_1830, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1054 = None
	        mul_11319: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1831: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11309, [mul_11319, 1280]);  mul_11309 = mul_11319 = None
	        permute_195: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1830, [1, 0]);  view_1830 = None
	        addmm_96: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_19_self_attn_v_proj_bias, view_1831, permute_195);  model_audio_tower_layers_19_self_attn_v_proj_bias = view_1831 = permute_195 = None
	        view_1832: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_96, [sym_size_int, 1500, 1280]);  addmm_96 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1833: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1832, [sym_size_int, -1, 20, 64]);  view_1832 = None
	        permute_196: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1833, [0, 2, 1, 3]);  view_1833 = None
	        clone_156: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_196, memory_format = torch.contiguous_format);  permute_196 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_19 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_154, clone_155, clone_156, None, False, scale = 1.0);  clone_154 = clone_155 = clone_156 = None
	        getitem_154: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_19[0];  _scaled_dot_product_efficient_attention_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_197: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_154, [0, 2, 1, 3]);  getitem_154 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_197, [sym_size_int, 1500, -1]);  permute_197 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1834, [2])
	        amax_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1834, [2])
	        full_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_117, full_234);  amin_117 = full_234 = None
	        full_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_117, full_235);  amax_117 = full_235 = None
	        sub_5361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_117, minimum_117);  maximum_117 = None
	        div_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5361, 255.0);  sub_5361 = None
	        clamp_min_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_234, 1.1920928955078125e-07);  div_234 = None
	        div_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_117, clamp_min_351);  minimum_117 = None
	        round_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_235);  div_235 = None
	        sub_5367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_235);  round_235 = None
	        clamp_min_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5367, -128);  sub_5367 = None
	        clamp_max_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_352, 127);  clamp_min_352 = None
	        _assert_tensor_metadata_1055 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_351, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1055 = None
	        _assert_tensor_metadata_1056 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_234, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1056 = None
	        convert_element_type_702: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_234, torch.int8);  clamp_max_234 = None
	        view_1837: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_351, [sym_size_int, 1500, 1])
	        view_1838: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_702, [sym_size_int, 1500, 1])
	        reciprocal_117: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1837);  view_1837 = None
	        mul_11389: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_117, 1.0);  reciprocal_117 = None
	        mul_11392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1834, mul_11389);  view_1834 = mul_11389 = None
	        round_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11392);  mul_11392 = None
	        add_18025: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_236, view_1838);  round_236 = view_1838 = None
	        clamp_min_353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18025, -128);  add_18025 = None
	        clamp_max_235: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_353, 127);  clamp_min_353 = None
	        _assert_tensor_metadata_1057 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_235, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1057 = None
	        convert_element_type_703: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_235, torch.int8);  clamp_max_235 = None
	        view_1841: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_351, [sym_size_int, 1500, 1]);  clamp_min_351 = None
	        view_1842: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_702, [sym_size_int, 1500, 1]);  convert_element_type_702 = None
	        _assert_tensor_metadata_1058 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_703, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1058 = None
	        convert_element_type_704: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_703, torch.float32);  convert_element_type_703 = None
	        _assert_tensor_metadata_1059 = torch.ops.aten._assert_tensor_metadata.default(view_1842, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1059 = None
	        convert_element_type_705: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1842, torch.float32);  view_1842 = None
	        sub_5387: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_704, convert_element_type_705);  convert_element_type_704 = convert_element_type_705 = None
	        mul_11414: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5387, view_1841);  sub_5387 = view_1841 = None
	        _assert_tensor_metadata_1060 = torch.ops.aten._assert_tensor_metadata.default(mul_11414, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1060 = None
	        view_1844: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1845: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1846: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1061 = torch.ops.aten._assert_tensor_metadata.default(view_1844, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1061 = None
	        convert_element_type_706: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1844, torch.float32);  view_1844 = None
	        _assert_tensor_metadata_1062 = torch.ops.aten._assert_tensor_metadata.default(view_1846, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1062 = None
	        convert_element_type_707: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1846, torch.float32);  view_1846 = None
	        sub_5391: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_706, convert_element_type_707);  convert_element_type_706 = convert_element_type_707 = None
	        mul_11419: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5391, view_1845);  sub_5391 = view_1845 = None
	        view_1847: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11419, [1280, 1280]);  mul_11419 = None
	        _assert_tensor_metadata_1063 = torch.ops.aten._assert_tensor_metadata.default(view_1847, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1063 = None
	        mul_11424: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1848: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11414, [mul_11424, 1280]);  mul_11414 = mul_11424 = None
	        permute_198: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1847, [1, 0]);  view_1847 = None
	        addmm_97: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_19_self_attn_out_proj_bias, view_1848, permute_198);  model_audio_tower_layers_19_self_attn_out_proj_bias = view_1848 = permute_198 = None
	        view_1849: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_97, [sym_size_int, 1500, 1280]);  addmm_97 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_18088: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_17468, view_1849);  add_17468 = view_1849 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_18088, memory_format = torch.contiguous_format)
	        var_mean_39 = torch.ops.aten.var_mean.correction(clone_158, [2], correction = 0, keepdim = True)
	        getitem_158: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_39[0]
	        getitem_159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_39[1];  var_mean_39 = None
	        add_18093: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_158, 1e-05);  getitem_158 = None
	        rsqrt_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_18093);  add_18093 = None
	        sub_5397: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_158, getitem_159);  clone_158 = getitem_159 = None
	        mul_11435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5397, rsqrt_39);  sub_5397 = rsqrt_39 = None
	        mul_11436: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11435, model_audio_tower_layers_19_final_layer_norm_weight);  mul_11435 = model_audio_tower_layers_19_final_layer_norm_weight = None
	        add_18094: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_11436, model_audio_tower_layers_19_final_layer_norm_bias);  mul_11436 = model_audio_tower_layers_19_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_18094, [2])
	        amax_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_18094, [2])
	        full_236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_118, full_236);  amin_118 = full_236 = None
	        full_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_118, full_237);  amax_118 = full_237 = None
	        sub_5408: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_118, minimum_118);  maximum_118 = None
	        div_236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5408, 255.0);  sub_5408 = None
	        clamp_min_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_236, 1.1920928955078125e-07);  div_236 = None
	        div_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_118, clamp_min_354);  minimum_118 = None
	        round_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_237);  div_237 = None
	        sub_5414: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_237);  round_237 = None
	        clamp_min_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5414, -128);  sub_5414 = None
	        clamp_max_236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_355, 127);  clamp_min_355 = None
	        _assert_tensor_metadata_1064 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_354, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1064 = None
	        _assert_tensor_metadata_1065 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_236, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1065 = None
	        convert_element_type_708: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_236, torch.int8);  clamp_max_236 = None
	        view_1852: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_354, [sym_size_int, 1500, 1])
	        view_1853: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_708, [sym_size_int, 1500, 1])
	        reciprocal_118: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1852);  view_1852 = None
	        mul_11484: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_118, 1.0);  reciprocal_118 = None
	        mul_11487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_18094, mul_11484);  add_18094 = mul_11484 = None
	        round_238: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11487);  mul_11487 = None
	        add_18181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_238, view_1853);  round_238 = view_1853 = None
	        clamp_min_356: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18181, -128);  add_18181 = None
	        clamp_max_237: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_356, 127);  clamp_min_356 = None
	        _assert_tensor_metadata_1066 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_237, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1066 = None
	        convert_element_type_709: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_237, torch.int8);  clamp_max_237 = None
	        view_1856: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_354, [sym_size_int, 1500, 1]);  clamp_min_354 = None
	        view_1857: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_708, [sym_size_int, 1500, 1]);  convert_element_type_708 = None
	        _assert_tensor_metadata_1067 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_709, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1067 = None
	        convert_element_type_710: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_709, torch.float32);  convert_element_type_709 = None
	        _assert_tensor_metadata_1068 = torch.ops.aten._assert_tensor_metadata.default(view_1857, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1068 = None
	        convert_element_type_711: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1857, torch.float32);  view_1857 = None
	        sub_5434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_710, convert_element_type_711);  convert_element_type_710 = convert_element_type_711 = None
	        mul_11509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5434, view_1856);  sub_5434 = view_1856 = None
	        _assert_tensor_metadata_1069 = torch.ops.aten._assert_tensor_metadata.default(mul_11509, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1069 = None
	        view_1859: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_19_fc1_parametrizations_weight_original0 = None
	        view_1860: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_19_fc1_parametrizations_weight_original1 = None
	        view_1861: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_19_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1070 = torch.ops.aten._assert_tensor_metadata.default(view_1859, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1070 = None
	        convert_element_type_712: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1859, torch.float32);  view_1859 = None
	        _assert_tensor_metadata_1071 = torch.ops.aten._assert_tensor_metadata.default(view_1861, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1071 = None
	        convert_element_type_713: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1861, torch.float32);  view_1861 = None
	        sub_5438: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_712, convert_element_type_713);  convert_element_type_712 = convert_element_type_713 = None
	        mul_11514: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5438, view_1860);  sub_5438 = view_1860 = None
	        view_1862: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11514, [5120, 1280]);  mul_11514 = None
	        _assert_tensor_metadata_1072 = torch.ops.aten._assert_tensor_metadata.default(view_1862, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1072 = None
	        mul_11519: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1863: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11509, [mul_11519, 1280]);  mul_11509 = mul_11519 = None
	        permute_199: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1862, [1, 0]);  view_1862 = None
	        addmm_98: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_19_fc1_bias, view_1863, permute_199);  model_audio_tower_layers_19_fc1_bias = view_1863 = permute_199 = None
	        view_1864: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_98, [sym_size_int, 1500, 5120]);  addmm_98 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_11526: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1864, 0.5)
	        mul_11527: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1864, 0.7071067811865476);  view_1864 = None
	        erf_21: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_11527);  mul_11527 = None
	        add_18240: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_21, 1);  erf_21 = None
	        mul_11528: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11526, add_18240);  mul_11526 = add_18240 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_11528, [2])
	        amax_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_11528, [2])
	        full_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_119, full_238);  amin_119 = full_238 = None
	        full_239: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_119, full_239);  amax_119 = full_239 = None
	        sub_5451: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_119, minimum_119);  maximum_119 = None
	        div_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5451, 255.0);  sub_5451 = None
	        clamp_min_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_238, 1.1920928955078125e-07);  div_238 = None
	        div_239: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_119, clamp_min_357);  minimum_119 = None
	        round_239: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_239);  div_239 = None
	        sub_5457: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_239);  round_239 = None
	        clamp_min_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5457, -128);  sub_5457 = None
	        clamp_max_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_358, 127);  clamp_min_358 = None
	        _assert_tensor_metadata_1073 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_357, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1073 = None
	        _assert_tensor_metadata_1074 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_238, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1074 = None
	        convert_element_type_714: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_238, torch.int8);  clamp_max_238 = None
	        view_1867: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_357, [sym_size_int, 1500, 1])
	        view_1868: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_714, [sym_size_int, 1500, 1])
	        reciprocal_119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1867);  view_1867 = None
	        mul_11574: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_119, 1.0);  reciprocal_119 = None
	        mul_11577: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11528, mul_11574);  mul_11528 = mul_11574 = None
	        round_240: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_11577);  mul_11577 = None
	        add_18323: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_240, view_1868);  round_240 = view_1868 = None
	        clamp_min_359: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18323, -128);  add_18323 = None
	        clamp_max_239: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_359, 127);  clamp_min_359 = None
	        _assert_tensor_metadata_1075 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_239, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1075 = None
	        convert_element_type_715: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_239, torch.int8);  clamp_max_239 = None
	        view_1871: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_357, [sym_size_int, 1500, 1]);  clamp_min_357 = None
	        view_1872: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_714, [sym_size_int, 1500, 1]);  convert_element_type_714 = None
	        _assert_tensor_metadata_1076 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_715, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1076 = None
	        convert_element_type_716: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_715, torch.float32);  convert_element_type_715 = None
	        _assert_tensor_metadata_1077 = torch.ops.aten._assert_tensor_metadata.default(view_1872, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1077 = None
	        convert_element_type_717: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1872, torch.float32);  view_1872 = None
	        sub_5477: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_716, convert_element_type_717);  convert_element_type_716 = convert_element_type_717 = None
	        mul_11599: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5477, view_1871);  sub_5477 = view_1871 = None
	        _assert_tensor_metadata_1078 = torch.ops.aten._assert_tensor_metadata.default(mul_11599, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1078 = None
	        view_1874: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_19_fc2_parametrizations_weight_original0 = None
	        view_1875: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_19_fc2_parametrizations_weight_original1 = None
	        view_1876: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_19_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1079 = torch.ops.aten._assert_tensor_metadata.default(view_1874, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1079 = None
	        convert_element_type_718: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1874, torch.float32);  view_1874 = None
	        _assert_tensor_metadata_1080 = torch.ops.aten._assert_tensor_metadata.default(view_1876, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1080 = None
	        convert_element_type_719: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1876, torch.float32);  view_1876 = None
	        sub_5481: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_718, convert_element_type_719);  convert_element_type_718 = convert_element_type_719 = None
	        mul_11604: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5481, view_1875);  sub_5481 = view_1875 = None
	        view_1877: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11604, [1280, 5120]);  mul_11604 = None
	        _assert_tensor_metadata_1081 = torch.ops.aten._assert_tensor_metadata.default(view_1877, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1081 = None
	        mul_11609: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1878: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11599, [mul_11609, 5120]);  mul_11599 = mul_11609 = None
	        permute_200: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1877, [1, 0]);  view_1877 = None
	        addmm_99: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_19_fc2_bias, view_1878, permute_200);  model_audio_tower_layers_19_fc2_bias = view_1878 = permute_200 = None
	        view_1879: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_99, [sym_size_int, 1500, 1280]);  addmm_99 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_18386: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_18088, view_1879);  add_18088 = view_1879 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_18386, memory_format = torch.contiguous_format)
	        var_mean_40 = torch.ops.aten.var_mean.correction(clone_161, [2], correction = 0, keepdim = True)
	        getitem_160: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_40[0]
	        getitem_161: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_40[1];  var_mean_40 = None
	        add_18391: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_160, 1e-05);  getitem_160 = None
	        rsqrt_40: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_18391);  add_18391 = None
	        sub_5487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_161, getitem_161);  clone_161 = getitem_161 = None
	        mul_11620: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5487, rsqrt_40);  sub_5487 = rsqrt_40 = None
	        mul_11621: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11620, model_audio_tower_layers_20_self_attn_layer_norm_weight);  mul_11620 = model_audio_tower_layers_20_self_attn_layer_norm_weight = None
	        add_18392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_11621, model_audio_tower_layers_20_self_attn_layer_norm_bias);  mul_11621 = model_audio_tower_layers_20_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_18392, [2])
	        amax_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_18392, [2])
	        full_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_120, full_240);  amin_120 = full_240 = None
	        full_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_120, full_241);  amax_120 = full_241 = None
	        sub_5498: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_120, minimum_120);  maximum_120 = None
	        div_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5498, 255.0);  sub_5498 = None
	        clamp_min_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_240, 1.1920928955078125e-07);  div_240 = None
	        div_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_120, clamp_min_360);  minimum_120 = None
	        round_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_241);  div_241 = None
	        sub_5504: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_241);  round_241 = None
	        clamp_min_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5504, -128);  sub_5504 = None
	        clamp_max_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_361, 127);  clamp_min_361 = None
	        _assert_tensor_metadata_1082 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_360, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1082 = None
	        _assert_tensor_metadata_1083 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_240, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1083 = None
	        convert_element_type_720: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_240, torch.int8);  clamp_max_240 = None
	        view_1882: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_360, [sym_size_int, 1500, 1])
	        view_1883: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_720, [sym_size_int, 1500, 1])
	        reciprocal_120: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1882);  view_1882 = None
	        mul_11669: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_120, 1.0);  reciprocal_120 = None
	        mul_11672: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_18392, mul_11669);  mul_11669 = None
	        round_242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11672);  mul_11672 = None
	        add_18479: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_242, view_1883);  round_242 = view_1883 = None
	        clamp_min_362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18479, -128);  add_18479 = None
	        clamp_max_241: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_362, 127);  clamp_min_362 = None
	        _assert_tensor_metadata_1084 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_241, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1084 = None
	        convert_element_type_721: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_241, torch.int8);  clamp_max_241 = None
	        view_1886: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_360, [sym_size_int, 1500, 1]);  clamp_min_360 = None
	        view_1887: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_720, [sym_size_int, 1500, 1]);  convert_element_type_720 = None
	        _assert_tensor_metadata_1085 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_721, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1085 = None
	        convert_element_type_722: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_721, torch.float32);  convert_element_type_721 = None
	        _assert_tensor_metadata_1086 = torch.ops.aten._assert_tensor_metadata.default(view_1887, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1086 = None
	        convert_element_type_723: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1887, torch.float32);  view_1887 = None
	        sub_5524: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_722, convert_element_type_723);  convert_element_type_722 = convert_element_type_723 = None
	        mul_11694: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5524, view_1886);  sub_5524 = view_1886 = None
	        _assert_tensor_metadata_1087 = torch.ops.aten._assert_tensor_metadata.default(mul_11694, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1087 = None
	        view_1889: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1890: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1891: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1088 = torch.ops.aten._assert_tensor_metadata.default(view_1889, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1088 = None
	        convert_element_type_724: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1889, torch.float32);  view_1889 = None
	        _assert_tensor_metadata_1089 = torch.ops.aten._assert_tensor_metadata.default(view_1891, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1089 = None
	        convert_element_type_725: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1891, torch.float32);  view_1891 = None
	        sub_5528: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_724, convert_element_type_725);  convert_element_type_724 = convert_element_type_725 = None
	        mul_11699: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5528, view_1890);  sub_5528 = view_1890 = None
	        view_1892: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11699, [1280, 1280]);  mul_11699 = None
	        _assert_tensor_metadata_1090 = torch.ops.aten._assert_tensor_metadata.default(view_1892, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1090 = None
	        mul_11704: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1893: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11694, [mul_11704, 1280]);  mul_11694 = mul_11704 = None
	        permute_201: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1892, [1, 0]);  view_1892 = None
	        addmm_100: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_20_self_attn_q_proj_bias, view_1893, permute_201);  model_audio_tower_layers_20_self_attn_q_proj_bias = view_1893 = permute_201 = None
	        view_1894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_100, [sym_size_int, 1500, 1280]);  addmm_100 = None
	        mul_11711: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1894, 0.125);  view_1894 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1895: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11711, [sym_size_int, 1500, 20, 64]);  mul_11711 = None
	        permute_202: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1895, [0, 2, 1, 3]);  view_1895 = None
	        clone_162: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_202, memory_format = torch.contiguous_format);  permute_202 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_18392, [2])
	        amax_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_18392, [2])
	        full_242: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_121, full_242);  amin_121 = full_242 = None
	        full_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_121, full_243);  amax_121 = full_243 = None
	        sub_5543: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_121, minimum_121);  maximum_121 = None
	        div_242: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5543, 255.0);  sub_5543 = None
	        clamp_min_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_242, 1.1920928955078125e-07);  div_242 = None
	        div_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_121, clamp_min_363);  minimum_121 = None
	        round_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_243);  div_243 = None
	        sub_5549: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_243);  round_243 = None
	        clamp_min_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5549, -128);  sub_5549 = None
	        clamp_max_242: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_364, 127);  clamp_min_364 = None
	        _assert_tensor_metadata_1091 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_363, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1091 = None
	        _assert_tensor_metadata_1092 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_242, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1092 = None
	        convert_element_type_726: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_242, torch.int8);  clamp_max_242 = None
	        view_1898: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_363, [sym_size_int, 1500, 1])
	        view_1899: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_726, [sym_size_int, 1500, 1])
	        reciprocal_121: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1898);  view_1898 = None
	        mul_11765: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_121, 1.0);  reciprocal_121 = None
	        mul_11768: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_18392, mul_11765);  mul_11765 = None
	        round_244: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11768);  mul_11768 = None
	        add_18631: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_244, view_1899);  round_244 = view_1899 = None
	        clamp_min_365: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18631, -128);  add_18631 = None
	        clamp_max_243: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_365, 127);  clamp_min_365 = None
	        _assert_tensor_metadata_1093 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_243, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1093 = None
	        convert_element_type_727: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_243, torch.int8);  clamp_max_243 = None
	        view_1902: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_363, [sym_size_int, 1500, 1]);  clamp_min_363 = None
	        view_1903: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_726, [sym_size_int, 1500, 1]);  convert_element_type_726 = None
	        _assert_tensor_metadata_1094 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_727, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1094 = None
	        convert_element_type_728: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_727, torch.float32);  convert_element_type_727 = None
	        _assert_tensor_metadata_1095 = torch.ops.aten._assert_tensor_metadata.default(view_1903, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1095 = None
	        convert_element_type_729: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1903, torch.float32);  view_1903 = None
	        sub_5569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_728, convert_element_type_729);  convert_element_type_728 = convert_element_type_729 = None
	        mul_11790: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5569, view_1902);  sub_5569 = view_1902 = None
	        _assert_tensor_metadata_1096 = torch.ops.aten._assert_tensor_metadata.default(mul_11790, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1096 = None
	        view_1905: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1906: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_1907: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1097 = torch.ops.aten._assert_tensor_metadata.default(view_1905, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1097 = None
	        convert_element_type_730: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1905, torch.float32);  view_1905 = None
	        _assert_tensor_metadata_1098 = torch.ops.aten._assert_tensor_metadata.default(view_1907, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1098 = None
	        convert_element_type_731: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1907, torch.float32);  view_1907 = None
	        sub_5573: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_730, convert_element_type_731);  convert_element_type_730 = convert_element_type_731 = None
	        mul_11795: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5573, view_1906);  sub_5573 = view_1906 = None
	        view_1908: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11795, [1280, 1280]);  mul_11795 = None
	        _assert_tensor_metadata_1099 = torch.ops.aten._assert_tensor_metadata.default(view_1908, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1099 = None
	        permute_203: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1908, [1, 0]);  view_1908 = None
	        mul_11798: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1909: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11790, [mul_11798, 1280]);  mul_11790 = mul_11798 = None
	        mm_20: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1909, permute_203);  view_1909 = permute_203 = None
	        view_1910: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_20, [sym_size_int, 1500, 1280]);  mm_20 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1911: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1910, [sym_size_int, -1, 20, 64]);  view_1910 = None
	        permute_204: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1911, [0, 2, 1, 3]);  view_1911 = None
	        clone_163: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_204, memory_format = torch.contiguous_format);  permute_204 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_18392, [2])
	        amax_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_18392, [2])
	        full_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_122, full_244);  amin_122 = full_244 = None
	        full_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_122, full_245);  amax_122 = full_245 = None
	        sub_5587: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_122, minimum_122);  maximum_122 = None
	        div_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5587, 255.0);  sub_5587 = None
	        clamp_min_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_244, 1.1920928955078125e-07);  div_244 = None
	        div_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_122, clamp_min_366);  minimum_122 = None
	        round_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_245);  div_245 = None
	        sub_5593: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_245);  round_245 = None
	        clamp_min_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5593, -128);  sub_5593 = None
	        clamp_max_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_367, 127);  clamp_min_367 = None
	        _assert_tensor_metadata_1100 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_366, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1100 = None
	        _assert_tensor_metadata_1101 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_244, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1101 = None
	        convert_element_type_732: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_244, torch.int8);  clamp_max_244 = None
	        view_1914: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_366, [sym_size_int, 1500, 1])
	        view_1915: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_732, [sym_size_int, 1500, 1])
	        reciprocal_122: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1914);  view_1914 = None
	        mul_11864: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_122, 1.0);  reciprocal_122 = None
	        mul_11867: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_18392, mul_11864);  add_18392 = mul_11864 = None
	        round_246: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11867);  mul_11867 = None
	        add_18779: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_246, view_1915);  round_246 = view_1915 = None
	        clamp_min_368: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18779, -128);  add_18779 = None
	        clamp_max_245: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_368, 127);  clamp_min_368 = None
	        _assert_tensor_metadata_1102 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_245, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1102 = None
	        convert_element_type_733: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_245, torch.int8);  clamp_max_245 = None
	        view_1918: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_366, [sym_size_int, 1500, 1]);  clamp_min_366 = None
	        view_1919: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_732, [sym_size_int, 1500, 1]);  convert_element_type_732 = None
	        _assert_tensor_metadata_1103 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_733, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1103 = None
	        convert_element_type_734: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_733, torch.float32);  convert_element_type_733 = None
	        _assert_tensor_metadata_1104 = torch.ops.aten._assert_tensor_metadata.default(view_1919, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1104 = None
	        convert_element_type_735: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1919, torch.float32);  view_1919 = None
	        sub_5613: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_734, convert_element_type_735);  convert_element_type_734 = convert_element_type_735 = None
	        mul_11889: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5613, view_1918);  sub_5613 = view_1918 = None
	        _assert_tensor_metadata_1105 = torch.ops.aten._assert_tensor_metadata.default(mul_11889, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1105 = None
	        view_1921: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1922: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_1923: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1106 = torch.ops.aten._assert_tensor_metadata.default(view_1921, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1106 = None
	        convert_element_type_736: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1921, torch.float32);  view_1921 = None
	        _assert_tensor_metadata_1107 = torch.ops.aten._assert_tensor_metadata.default(view_1923, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1107 = None
	        convert_element_type_737: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1923, torch.float32);  view_1923 = None
	        sub_5617: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_736, convert_element_type_737);  convert_element_type_736 = convert_element_type_737 = None
	        mul_11894: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5617, view_1922);  sub_5617 = view_1922 = None
	        view_1924: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11894, [1280, 1280]);  mul_11894 = None
	        _assert_tensor_metadata_1108 = torch.ops.aten._assert_tensor_metadata.default(view_1924, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1108 = None
	        mul_11899: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1925: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11889, [mul_11899, 1280]);  mul_11889 = mul_11899 = None
	        permute_205: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1924, [1, 0]);  view_1924 = None
	        addmm_101: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_20_self_attn_v_proj_bias, view_1925, permute_205);  model_audio_tower_layers_20_self_attn_v_proj_bias = view_1925 = permute_205 = None
	        view_1926: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_101, [sym_size_int, 1500, 1280]);  addmm_101 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1927: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1926, [sym_size_int, -1, 20, 64]);  view_1926 = None
	        permute_206: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1927, [0, 2, 1, 3]);  view_1927 = None
	        clone_164: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_206, memory_format = torch.contiguous_format);  permute_206 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_20 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_162, clone_163, clone_164, None, False, scale = 1.0);  clone_162 = clone_163 = clone_164 = None
	        getitem_162: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_20[0];  _scaled_dot_product_efficient_attention_20 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_207: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_162, [0, 2, 1, 3]);  getitem_162 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1928: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_207, [sym_size_int, 1500, -1]);  permute_207 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1928, [2])
	        amax_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1928, [2])
	        full_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_123, full_246);  amin_123 = full_246 = None
	        full_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_123, full_247);  amax_123 = full_247 = None
	        sub_5635: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_123, minimum_123);  maximum_123 = None
	        div_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5635, 255.0);  sub_5635 = None
	        clamp_min_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_246, 1.1920928955078125e-07);  div_246 = None
	        div_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_123, clamp_min_369);  minimum_123 = None
	        round_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_247);  div_247 = None
	        sub_5641: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_247);  round_247 = None
	        clamp_min_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5641, -128);  sub_5641 = None
	        clamp_max_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_370, 127);  clamp_min_370 = None
	        _assert_tensor_metadata_1109 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_369, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1109 = None
	        _assert_tensor_metadata_1110 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_246, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1110 = None
	        convert_element_type_738: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_246, torch.int8);  clamp_max_246 = None
	        view_1931: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_369, [sym_size_int, 1500, 1])
	        view_1932: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_738, [sym_size_int, 1500, 1])
	        reciprocal_123: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1931);  view_1931 = None
	        mul_11969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_123, 1.0);  reciprocal_123 = None
	        mul_11972: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1928, mul_11969);  view_1928 = mul_11969 = None
	        round_248: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11972);  mul_11972 = None
	        add_18943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_248, view_1932);  round_248 = view_1932 = None
	        clamp_min_371: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18943, -128);  add_18943 = None
	        clamp_max_247: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_371, 127);  clamp_min_371 = None
	        _assert_tensor_metadata_1111 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_247, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1111 = None
	        convert_element_type_739: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_247, torch.int8);  clamp_max_247 = None
	        view_1935: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_369, [sym_size_int, 1500, 1]);  clamp_min_369 = None
	        view_1936: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_738, [sym_size_int, 1500, 1]);  convert_element_type_738 = None
	        _assert_tensor_metadata_1112 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_739, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1112 = None
	        convert_element_type_740: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_739, torch.float32);  convert_element_type_739 = None
	        _assert_tensor_metadata_1113 = torch.ops.aten._assert_tensor_metadata.default(view_1936, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1113 = None
	        convert_element_type_741: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1936, torch.float32);  view_1936 = None
	        sub_5661: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_740, convert_element_type_741);  convert_element_type_740 = convert_element_type_741 = None
	        mul_11994: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5661, view_1935);  sub_5661 = view_1935 = None
	        _assert_tensor_metadata_1114 = torch.ops.aten._assert_tensor_metadata.default(mul_11994, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1114 = None
	        view_1938: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1939: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_1940: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1115 = torch.ops.aten._assert_tensor_metadata.default(view_1938, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1115 = None
	        convert_element_type_742: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1938, torch.float32);  view_1938 = None
	        _assert_tensor_metadata_1116 = torch.ops.aten._assert_tensor_metadata.default(view_1940, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1116 = None
	        convert_element_type_743: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1940, torch.float32);  view_1940 = None
	        sub_5665: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_742, convert_element_type_743);  convert_element_type_742 = convert_element_type_743 = None
	        mul_11999: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5665, view_1939);  sub_5665 = view_1939 = None
	        view_1941: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11999, [1280, 1280]);  mul_11999 = None
	        _assert_tensor_metadata_1117 = torch.ops.aten._assert_tensor_metadata.default(view_1941, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1117 = None
	        mul_12004: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1942: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11994, [mul_12004, 1280]);  mul_11994 = mul_12004 = None
	        permute_208: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1941, [1, 0]);  view_1941 = None
	        addmm_102: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_20_self_attn_out_proj_bias, view_1942, permute_208);  model_audio_tower_layers_20_self_attn_out_proj_bias = view_1942 = permute_208 = None
	        view_1943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_102, [sym_size_int, 1500, 1280]);  addmm_102 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_19006: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_18386, view_1943);  add_18386 = view_1943 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_166: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_19006, memory_format = torch.contiguous_format)
	        var_mean_41 = torch.ops.aten.var_mean.correction(clone_166, [2], correction = 0, keepdim = True)
	        getitem_166: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_41[0]
	        getitem_167: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_41[1];  var_mean_41 = None
	        add_19011: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_166, 1e-05);  getitem_166 = None
	        rsqrt_41: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_19011);  add_19011 = None
	        sub_5671: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_166, getitem_167);  clone_166 = getitem_167 = None
	        mul_12015: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5671, rsqrt_41);  sub_5671 = rsqrt_41 = None
	        mul_12016: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12015, model_audio_tower_layers_20_final_layer_norm_weight);  mul_12015 = model_audio_tower_layers_20_final_layer_norm_weight = None
	        add_19012: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12016, model_audio_tower_layers_20_final_layer_norm_bias);  mul_12016 = model_audio_tower_layers_20_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_19012, [2])
	        amax_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_19012, [2])
	        full_248: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_124, full_248);  amin_124 = full_248 = None
	        full_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_124, full_249);  amax_124 = full_249 = None
	        sub_5682: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_124, minimum_124);  maximum_124 = None
	        div_248: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5682, 255.0);  sub_5682 = None
	        clamp_min_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_248, 1.1920928955078125e-07);  div_248 = None
	        div_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_124, clamp_min_372);  minimum_124 = None
	        round_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_249);  div_249 = None
	        sub_5688: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_249);  round_249 = None
	        clamp_min_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5688, -128);  sub_5688 = None
	        clamp_max_248: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_373, 127);  clamp_min_373 = None
	        _assert_tensor_metadata_1118 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_372, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1118 = None
	        _assert_tensor_metadata_1119 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_248, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1119 = None
	        convert_element_type_744: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_248, torch.int8);  clamp_max_248 = None
	        view_1946: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_372, [sym_size_int, 1500, 1])
	        view_1947: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_744, [sym_size_int, 1500, 1])
	        reciprocal_124: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1946);  view_1946 = None
	        mul_12064: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_124, 1.0);  reciprocal_124 = None
	        mul_12067: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_19012, mul_12064);  add_19012 = mul_12064 = None
	        round_250: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12067);  mul_12067 = None
	        add_19099: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_250, view_1947);  round_250 = view_1947 = None
	        clamp_min_374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19099, -128);  add_19099 = None
	        clamp_max_249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_374, 127);  clamp_min_374 = None
	        _assert_tensor_metadata_1120 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1120 = None
	        convert_element_type_745: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_249, torch.int8);  clamp_max_249 = None
	        view_1950: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_372, [sym_size_int, 1500, 1]);  clamp_min_372 = None
	        view_1951: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_744, [sym_size_int, 1500, 1]);  convert_element_type_744 = None
	        _assert_tensor_metadata_1121 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_745, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1121 = None
	        convert_element_type_746: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_745, torch.float32);  convert_element_type_745 = None
	        _assert_tensor_metadata_1122 = torch.ops.aten._assert_tensor_metadata.default(view_1951, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1122 = None
	        convert_element_type_747: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1951, torch.float32);  view_1951 = None
	        sub_5708: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_746, convert_element_type_747);  convert_element_type_746 = convert_element_type_747 = None
	        mul_12089: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5708, view_1950);  sub_5708 = view_1950 = None
	        _assert_tensor_metadata_1123 = torch.ops.aten._assert_tensor_metadata.default(mul_12089, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1123 = None
	        view_1953: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_20_fc1_parametrizations_weight_original0 = None
	        view_1954: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_20_fc1_parametrizations_weight_original1 = None
	        view_1955: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_20_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1124 = torch.ops.aten._assert_tensor_metadata.default(view_1953, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1124 = None
	        convert_element_type_748: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1953, torch.float32);  view_1953 = None
	        _assert_tensor_metadata_1125 = torch.ops.aten._assert_tensor_metadata.default(view_1955, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1125 = None
	        convert_element_type_749: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1955, torch.float32);  view_1955 = None
	        sub_5712: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_748, convert_element_type_749);  convert_element_type_748 = convert_element_type_749 = None
	        mul_12094: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5712, view_1954);  sub_5712 = view_1954 = None
	        view_1956: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12094, [5120, 1280]);  mul_12094 = None
	        _assert_tensor_metadata_1126 = torch.ops.aten._assert_tensor_metadata.default(view_1956, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1126 = None
	        mul_12099: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1957: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12089, [mul_12099, 1280]);  mul_12089 = mul_12099 = None
	        permute_209: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1956, [1, 0]);  view_1956 = None
	        addmm_103: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_20_fc1_bias, view_1957, permute_209);  model_audio_tower_layers_20_fc1_bias = view_1957 = permute_209 = None
	        view_1958: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_103, [sym_size_int, 1500, 5120]);  addmm_103 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_12106: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1958, 0.5)
	        mul_12107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1958, 0.7071067811865476);  view_1958 = None
	        erf_22: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_12107);  mul_12107 = None
	        add_19158: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_22, 1);  erf_22 = None
	        mul_12108: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12106, add_19158);  mul_12106 = add_19158 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_12108, [2])
	        amax_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_12108, [2])
	        full_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_125, full_250);  amin_125 = full_250 = None
	        full_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_125, full_251);  amax_125 = full_251 = None
	        sub_5725: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_125, minimum_125);  maximum_125 = None
	        div_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5725, 255.0);  sub_5725 = None
	        clamp_min_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_250, 1.1920928955078125e-07);  div_250 = None
	        div_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_125, clamp_min_375);  minimum_125 = None
	        round_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_251);  div_251 = None
	        sub_5731: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_251);  round_251 = None
	        clamp_min_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5731, -128);  sub_5731 = None
	        clamp_max_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_376, 127);  clamp_min_376 = None
	        _assert_tensor_metadata_1127 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_375, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1127 = None
	        _assert_tensor_metadata_1128 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_250, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1128 = None
	        convert_element_type_750: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_250, torch.int8);  clamp_max_250 = None
	        view_1961: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_375, [sym_size_int, 1500, 1])
	        view_1962: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_750, [sym_size_int, 1500, 1])
	        reciprocal_125: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1961);  view_1961 = None
	        mul_12154: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_125, 1.0);  reciprocal_125 = None
	        mul_12157: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12108, mul_12154);  mul_12108 = mul_12154 = None
	        round_252: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_12157);  mul_12157 = None
	        add_19241: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_252, view_1962);  round_252 = view_1962 = None
	        clamp_min_377: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19241, -128);  add_19241 = None
	        clamp_max_251: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_377, 127);  clamp_min_377 = None
	        _assert_tensor_metadata_1129 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_251, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1129 = None
	        convert_element_type_751: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_251, torch.int8);  clamp_max_251 = None
	        view_1965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_375, [sym_size_int, 1500, 1]);  clamp_min_375 = None
	        view_1966: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_750, [sym_size_int, 1500, 1]);  convert_element_type_750 = None
	        _assert_tensor_metadata_1130 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_751, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1130 = None
	        convert_element_type_752: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_751, torch.float32);  convert_element_type_751 = None
	        _assert_tensor_metadata_1131 = torch.ops.aten._assert_tensor_metadata.default(view_1966, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1131 = None
	        convert_element_type_753: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1966, torch.float32);  view_1966 = None
	        sub_5751: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_752, convert_element_type_753);  convert_element_type_752 = convert_element_type_753 = None
	        mul_12179: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5751, view_1965);  sub_5751 = view_1965 = None
	        _assert_tensor_metadata_1132 = torch.ops.aten._assert_tensor_metadata.default(mul_12179, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1132 = None
	        view_1968: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_20_fc2_parametrizations_weight_original0 = None
	        view_1969: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_20_fc2_parametrizations_weight_original1 = None
	        view_1970: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_20_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1133 = torch.ops.aten._assert_tensor_metadata.default(view_1968, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1133 = None
	        convert_element_type_754: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1968, torch.float32);  view_1968 = None
	        _assert_tensor_metadata_1134 = torch.ops.aten._assert_tensor_metadata.default(view_1970, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1134 = None
	        convert_element_type_755: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1970, torch.float32);  view_1970 = None
	        sub_5755: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_754, convert_element_type_755);  convert_element_type_754 = convert_element_type_755 = None
	        mul_12184: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5755, view_1969);  sub_5755 = view_1969 = None
	        view_1971: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12184, [1280, 5120]);  mul_12184 = None
	        _assert_tensor_metadata_1135 = torch.ops.aten._assert_tensor_metadata.default(view_1971, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1135 = None
	        mul_12189: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1972: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12179, [mul_12189, 5120]);  mul_12179 = mul_12189 = None
	        permute_210: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1971, [1, 0]);  view_1971 = None
	        addmm_104: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_20_fc2_bias, view_1972, permute_210);  model_audio_tower_layers_20_fc2_bias = view_1972 = permute_210 = None
	        view_1973: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_104, [sym_size_int, 1500, 1280]);  addmm_104 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_19304: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_19006, view_1973);  add_19006 = view_1973 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_169: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_19304, memory_format = torch.contiguous_format)
	        var_mean_42 = torch.ops.aten.var_mean.correction(clone_169, [2], correction = 0, keepdim = True)
	        getitem_168: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_42[0]
	        getitem_169: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_42[1];  var_mean_42 = None
	        add_19309: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_168, 1e-05);  getitem_168 = None
	        rsqrt_42: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_19309);  add_19309 = None
	        sub_5761: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_169, getitem_169);  clone_169 = getitem_169 = None
	        mul_12200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5761, rsqrt_42);  sub_5761 = rsqrt_42 = None
	        mul_12201: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12200, model_audio_tower_layers_21_self_attn_layer_norm_weight);  mul_12200 = model_audio_tower_layers_21_self_attn_layer_norm_weight = None
	        add_19310: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12201, model_audio_tower_layers_21_self_attn_layer_norm_bias);  mul_12201 = model_audio_tower_layers_21_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_19310, [2])
	        amax_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_19310, [2])
	        full_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_126, full_252);  amin_126 = full_252 = None
	        full_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_126, full_253);  amax_126 = full_253 = None
	        sub_5772: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_126, minimum_126);  maximum_126 = None
	        div_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5772, 255.0);  sub_5772 = None
	        clamp_min_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_252, 1.1920928955078125e-07);  div_252 = None
	        div_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_126, clamp_min_378);  minimum_126 = None
	        round_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_253);  div_253 = None
	        sub_5778: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_253);  round_253 = None
	        clamp_min_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5778, -128);  sub_5778 = None
	        clamp_max_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_379, 127);  clamp_min_379 = None
	        _assert_tensor_metadata_1136 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_378, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1136 = None
	        _assert_tensor_metadata_1137 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_252, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1137 = None
	        convert_element_type_756: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_252, torch.int8);  clamp_max_252 = None
	        view_1976: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_378, [sym_size_int, 1500, 1])
	        view_1977: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_756, [sym_size_int, 1500, 1])
	        reciprocal_126: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1976);  view_1976 = None
	        mul_12249: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_126, 1.0);  reciprocal_126 = None
	        mul_12252: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_19310, mul_12249);  mul_12249 = None
	        round_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12252);  mul_12252 = None
	        add_19397: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_254, view_1977);  round_254 = view_1977 = None
	        clamp_min_380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19397, -128);  add_19397 = None
	        clamp_max_253: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_380, 127);  clamp_min_380 = None
	        _assert_tensor_metadata_1138 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_253, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1138 = None
	        convert_element_type_757: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_253, torch.int8);  clamp_max_253 = None
	        view_1980: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_378, [sym_size_int, 1500, 1]);  clamp_min_378 = None
	        view_1981: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_756, [sym_size_int, 1500, 1]);  convert_element_type_756 = None
	        _assert_tensor_metadata_1139 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_757, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1139 = None
	        convert_element_type_758: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_757, torch.float32);  convert_element_type_757 = None
	        _assert_tensor_metadata_1140 = torch.ops.aten._assert_tensor_metadata.default(view_1981, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1140 = None
	        convert_element_type_759: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1981, torch.float32);  view_1981 = None
	        sub_5798: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_758, convert_element_type_759);  convert_element_type_758 = convert_element_type_759 = None
	        mul_12274: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5798, view_1980);  sub_5798 = view_1980 = None
	        _assert_tensor_metadata_1141 = torch.ops.aten._assert_tensor_metadata.default(mul_12274, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1141 = None
	        view_1983: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1984: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_1985: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1142 = torch.ops.aten._assert_tensor_metadata.default(view_1983, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1142 = None
	        convert_element_type_760: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1983, torch.float32);  view_1983 = None
	        _assert_tensor_metadata_1143 = torch.ops.aten._assert_tensor_metadata.default(view_1985, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1143 = None
	        convert_element_type_761: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1985, torch.float32);  view_1985 = None
	        sub_5802: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_760, convert_element_type_761);  convert_element_type_760 = convert_element_type_761 = None
	        mul_12279: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5802, view_1984);  sub_5802 = view_1984 = None
	        view_1986: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12279, [1280, 1280]);  mul_12279 = None
	        _assert_tensor_metadata_1144 = torch.ops.aten._assert_tensor_metadata.default(view_1986, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1144 = None
	        mul_12284: "Sym(1500*s6)" = sym_size_int * 1500
	        view_1987: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12274, [mul_12284, 1280]);  mul_12274 = mul_12284 = None
	        permute_211: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1986, [1, 0]);  view_1986 = None
	        addmm_105: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_21_self_attn_q_proj_bias, view_1987, permute_211);  model_audio_tower_layers_21_self_attn_q_proj_bias = view_1987 = permute_211 = None
	        view_1988: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_105, [sym_size_int, 1500, 1280]);  addmm_105 = None
	        mul_12291: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1988, 0.125);  view_1988 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1989: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12291, [sym_size_int, 1500, 20, 64]);  mul_12291 = None
	        permute_212: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1989, [0, 2, 1, 3]);  view_1989 = None
	        clone_170: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_212, memory_format = torch.contiguous_format);  permute_212 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_19310, [2])
	        amax_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_19310, [2])
	        full_254: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_127, full_254);  amin_127 = full_254 = None
	        full_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_127, full_255);  amax_127 = full_255 = None
	        sub_5817: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_127, minimum_127);  maximum_127 = None
	        div_254: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5817, 255.0);  sub_5817 = None
	        clamp_min_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_254, 1.1920928955078125e-07);  div_254 = None
	        div_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_127, clamp_min_381);  minimum_127 = None
	        round_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_255);  div_255 = None
	        sub_5823: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_255);  round_255 = None
	        clamp_min_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5823, -128);  sub_5823 = None
	        clamp_max_254: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_382, 127);  clamp_min_382 = None
	        _assert_tensor_metadata_1145 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_381, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1145 = None
	        _assert_tensor_metadata_1146 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_254, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1146 = None
	        convert_element_type_762: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_254, torch.int8);  clamp_max_254 = None
	        view_1992: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_381, [sym_size_int, 1500, 1])
	        view_1993: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_762, [sym_size_int, 1500, 1])
	        reciprocal_127: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1992);  view_1992 = None
	        mul_12345: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_127, 1.0);  reciprocal_127 = None
	        mul_12348: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_19310, mul_12345);  mul_12345 = None
	        round_256: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12348);  mul_12348 = None
	        add_19549: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_256, view_1993);  round_256 = view_1993 = None
	        clamp_min_383: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19549, -128);  add_19549 = None
	        clamp_max_255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_383, 127);  clamp_min_383 = None
	        _assert_tensor_metadata_1147 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_255, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1147 = None
	        convert_element_type_763: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_255, torch.int8);  clamp_max_255 = None
	        view_1996: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_381, [sym_size_int, 1500, 1]);  clamp_min_381 = None
	        view_1997: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_762, [sym_size_int, 1500, 1]);  convert_element_type_762 = None
	        _assert_tensor_metadata_1148 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_763, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1148 = None
	        convert_element_type_764: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_763, torch.float32);  convert_element_type_763 = None
	        _assert_tensor_metadata_1149 = torch.ops.aten._assert_tensor_metadata.default(view_1997, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1149 = None
	        convert_element_type_765: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1997, torch.float32);  view_1997 = None
	        sub_5843: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_764, convert_element_type_765);  convert_element_type_764 = convert_element_type_765 = None
	        mul_12370: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5843, view_1996);  sub_5843 = view_1996 = None
	        _assert_tensor_metadata_1150 = torch.ops.aten._assert_tensor_metadata.default(mul_12370, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1150 = None
	        view_1999: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2000: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2001: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1151 = torch.ops.aten._assert_tensor_metadata.default(view_1999, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1151 = None
	        convert_element_type_766: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1999, torch.float32);  view_1999 = None
	        _assert_tensor_metadata_1152 = torch.ops.aten._assert_tensor_metadata.default(view_2001, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1152 = None
	        convert_element_type_767: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2001, torch.float32);  view_2001 = None
	        sub_5847: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_766, convert_element_type_767);  convert_element_type_766 = convert_element_type_767 = None
	        mul_12375: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5847, view_2000);  sub_5847 = view_2000 = None
	        view_2002: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12375, [1280, 1280]);  mul_12375 = None
	        _assert_tensor_metadata_1153 = torch.ops.aten._assert_tensor_metadata.default(view_2002, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1153 = None
	        permute_213: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2002, [1, 0]);  view_2002 = None
	        mul_12378: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2003: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12370, [mul_12378, 1280]);  mul_12370 = mul_12378 = None
	        mm_21: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2003, permute_213);  view_2003 = permute_213 = None
	        view_2004: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_21, [sym_size_int, 1500, 1280]);  mm_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2005: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2004, [sym_size_int, -1, 20, 64]);  view_2004 = None
	        permute_214: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2005, [0, 2, 1, 3]);  view_2005 = None
	        clone_171: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_214, memory_format = torch.contiguous_format);  permute_214 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_19310, [2])
	        amax_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_19310, [2])
	        full_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_128, full_256);  amin_128 = full_256 = None
	        full_257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_128, full_257);  amax_128 = full_257 = None
	        sub_5861: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_128, minimum_128);  maximum_128 = None
	        div_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5861, 255.0);  sub_5861 = None
	        clamp_min_384: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_256, 1.1920928955078125e-07);  div_256 = None
	        div_257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_128, clamp_min_384);  minimum_128 = None
	        round_257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_257);  div_257 = None
	        sub_5867: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_257);  round_257 = None
	        clamp_min_385: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5867, -128);  sub_5867 = None
	        clamp_max_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_385, 127);  clamp_min_385 = None
	        _assert_tensor_metadata_1154 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_384, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1154 = None
	        _assert_tensor_metadata_1155 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_256, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1155 = None
	        convert_element_type_768: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_256, torch.int8);  clamp_max_256 = None
	        view_2008: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_384, [sym_size_int, 1500, 1])
	        view_2009: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_768, [sym_size_int, 1500, 1])
	        reciprocal_128: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2008);  view_2008 = None
	        mul_12444: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_128, 1.0);  reciprocal_128 = None
	        mul_12447: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_19310, mul_12444);  add_19310 = mul_12444 = None
	        round_258: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12447);  mul_12447 = None
	        add_19697: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_258, view_2009);  round_258 = view_2009 = None
	        clamp_min_386: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19697, -128);  add_19697 = None
	        clamp_max_257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_386, 127);  clamp_min_386 = None
	        _assert_tensor_metadata_1156 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_257, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1156 = None
	        convert_element_type_769: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_257, torch.int8);  clamp_max_257 = None
	        view_2012: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_384, [sym_size_int, 1500, 1]);  clamp_min_384 = None
	        view_2013: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_768, [sym_size_int, 1500, 1]);  convert_element_type_768 = None
	        _assert_tensor_metadata_1157 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_769, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1157 = None
	        convert_element_type_770: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_769, torch.float32);  convert_element_type_769 = None
	        _assert_tensor_metadata_1158 = torch.ops.aten._assert_tensor_metadata.default(view_2013, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1158 = None
	        convert_element_type_771: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2013, torch.float32);  view_2013 = None
	        sub_5887: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_770, convert_element_type_771);  convert_element_type_770 = convert_element_type_771 = None
	        mul_12469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5887, view_2012);  sub_5887 = view_2012 = None
	        _assert_tensor_metadata_1159 = torch.ops.aten._assert_tensor_metadata.default(mul_12469, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1159 = None
	        view_2015: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2016: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2017: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1160 = torch.ops.aten._assert_tensor_metadata.default(view_2015, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1160 = None
	        convert_element_type_772: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2015, torch.float32);  view_2015 = None
	        _assert_tensor_metadata_1161 = torch.ops.aten._assert_tensor_metadata.default(view_2017, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1161 = None
	        convert_element_type_773: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2017, torch.float32);  view_2017 = None
	        sub_5891: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_772, convert_element_type_773);  convert_element_type_772 = convert_element_type_773 = None
	        mul_12474: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5891, view_2016);  sub_5891 = view_2016 = None
	        view_2018: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12474, [1280, 1280]);  mul_12474 = None
	        _assert_tensor_metadata_1162 = torch.ops.aten._assert_tensor_metadata.default(view_2018, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1162 = None
	        mul_12479: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2019: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12469, [mul_12479, 1280]);  mul_12469 = mul_12479 = None
	        permute_215: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2018, [1, 0]);  view_2018 = None
	        addmm_106: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_21_self_attn_v_proj_bias, view_2019, permute_215);  model_audio_tower_layers_21_self_attn_v_proj_bias = view_2019 = permute_215 = None
	        view_2020: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_106, [sym_size_int, 1500, 1280]);  addmm_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2021: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2020, [sym_size_int, -1, 20, 64]);  view_2020 = None
	        permute_216: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2021, [0, 2, 1, 3]);  view_2021 = None
	        clone_172: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_216, memory_format = torch.contiguous_format);  permute_216 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_21 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_170, clone_171, clone_172, None, False, scale = 1.0);  clone_170 = clone_171 = clone_172 = None
	        getitem_170: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_21[0];  _scaled_dot_product_efficient_attention_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_217: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_170, [0, 2, 1, 3]);  getitem_170 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2022: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_217, [sym_size_int, 1500, -1]);  permute_217 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2022, [2])
	        amax_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2022, [2])
	        full_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_129, full_258);  amin_129 = full_258 = None
	        full_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_129, full_259);  amax_129 = full_259 = None
	        sub_5909: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_129, minimum_129);  maximum_129 = None
	        div_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5909, 255.0);  sub_5909 = None
	        clamp_min_387: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_258, 1.1920928955078125e-07);  div_258 = None
	        div_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_129, clamp_min_387);  minimum_129 = None
	        round_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_259);  div_259 = None
	        sub_5915: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_259);  round_259 = None
	        clamp_min_388: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5915, -128);  sub_5915 = None
	        clamp_max_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_388, 127);  clamp_min_388 = None
	        _assert_tensor_metadata_1163 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_387, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1163 = None
	        _assert_tensor_metadata_1164 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_258, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1164 = None
	        convert_element_type_774: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_258, torch.int8);  clamp_max_258 = None
	        view_2025: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_387, [sym_size_int, 1500, 1])
	        view_2026: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_774, [sym_size_int, 1500, 1])
	        reciprocal_129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2025);  view_2025 = None
	        mul_12549: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_129, 1.0);  reciprocal_129 = None
	        mul_12552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2022, mul_12549);  view_2022 = mul_12549 = None
	        round_260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12552);  mul_12552 = None
	        add_19861: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_260, view_2026);  round_260 = view_2026 = None
	        clamp_min_389: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19861, -128);  add_19861 = None
	        clamp_max_259: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_389, 127);  clamp_min_389 = None
	        _assert_tensor_metadata_1165 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_259, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1165 = None
	        convert_element_type_775: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_259, torch.int8);  clamp_max_259 = None
	        view_2029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_387, [sym_size_int, 1500, 1]);  clamp_min_387 = None
	        view_2030: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_774, [sym_size_int, 1500, 1]);  convert_element_type_774 = None
	        _assert_tensor_metadata_1166 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_775, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1166 = None
	        convert_element_type_776: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_775, torch.float32);  convert_element_type_775 = None
	        _assert_tensor_metadata_1167 = torch.ops.aten._assert_tensor_metadata.default(view_2030, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1167 = None
	        convert_element_type_777: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2030, torch.float32);  view_2030 = None
	        sub_5935: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_776, convert_element_type_777);  convert_element_type_776 = convert_element_type_777 = None
	        mul_12574: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5935, view_2029);  sub_5935 = view_2029 = None
	        _assert_tensor_metadata_1168 = torch.ops.aten._assert_tensor_metadata.default(mul_12574, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1168 = None
	        view_2032: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2033: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2034: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1169 = torch.ops.aten._assert_tensor_metadata.default(view_2032, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1169 = None
	        convert_element_type_778: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2032, torch.float32);  view_2032 = None
	        _assert_tensor_metadata_1170 = torch.ops.aten._assert_tensor_metadata.default(view_2034, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1170 = None
	        convert_element_type_779: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2034, torch.float32);  view_2034 = None
	        sub_5939: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_778, convert_element_type_779);  convert_element_type_778 = convert_element_type_779 = None
	        mul_12579: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5939, view_2033);  sub_5939 = view_2033 = None
	        view_2035: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12579, [1280, 1280]);  mul_12579 = None
	        _assert_tensor_metadata_1171 = torch.ops.aten._assert_tensor_metadata.default(view_2035, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1171 = None
	        mul_12584: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2036: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12574, [mul_12584, 1280]);  mul_12574 = mul_12584 = None
	        permute_218: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2035, [1, 0]);  view_2035 = None
	        addmm_107: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_21_self_attn_out_proj_bias, view_2036, permute_218);  model_audio_tower_layers_21_self_attn_out_proj_bias = view_2036 = permute_218 = None
	        view_2037: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_107, [sym_size_int, 1500, 1280]);  addmm_107 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_19924: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_19304, view_2037);  add_19304 = view_2037 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_19924, memory_format = torch.contiguous_format)
	        var_mean_43 = torch.ops.aten.var_mean.correction(clone_174, [2], correction = 0, keepdim = True)
	        getitem_174: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_43[0]
	        getitem_175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_43[1];  var_mean_43 = None
	        add_19929: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_174, 1e-05);  getitem_174 = None
	        rsqrt_43: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_19929);  add_19929 = None
	        sub_5945: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_174, getitem_175);  clone_174 = getitem_175 = None
	        mul_12595: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5945, rsqrt_43);  sub_5945 = rsqrt_43 = None
	        mul_12596: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12595, model_audio_tower_layers_21_final_layer_norm_weight);  mul_12595 = model_audio_tower_layers_21_final_layer_norm_weight = None
	        add_19930: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12596, model_audio_tower_layers_21_final_layer_norm_bias);  mul_12596 = model_audio_tower_layers_21_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_19930, [2])
	        amax_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_19930, [2])
	        full_260: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_130, full_260);  amin_130 = full_260 = None
	        full_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_130, full_261);  amax_130 = full_261 = None
	        sub_5956: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_130, minimum_130);  maximum_130 = None
	        div_260: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5956, 255.0);  sub_5956 = None
	        clamp_min_390: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_260, 1.1920928955078125e-07);  div_260 = None
	        div_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_130, clamp_min_390);  minimum_130 = None
	        round_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_261);  div_261 = None
	        sub_5962: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_261);  round_261 = None
	        clamp_min_391: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5962, -128);  sub_5962 = None
	        clamp_max_260: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_391, 127);  clamp_min_391 = None
	        _assert_tensor_metadata_1172 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_390, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1172 = None
	        _assert_tensor_metadata_1173 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_260, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1173 = None
	        convert_element_type_780: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_260, torch.int8);  clamp_max_260 = None
	        view_2040: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_390, [sym_size_int, 1500, 1])
	        view_2041: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_780, [sym_size_int, 1500, 1])
	        reciprocal_130: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2040);  view_2040 = None
	        mul_12644: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_130, 1.0);  reciprocal_130 = None
	        mul_12647: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_19930, mul_12644);  add_19930 = mul_12644 = None
	        round_262: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12647);  mul_12647 = None
	        add_20017: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_262, view_2041);  round_262 = view_2041 = None
	        clamp_min_392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20017, -128);  add_20017 = None
	        clamp_max_261: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_392, 127);  clamp_min_392 = None
	        _assert_tensor_metadata_1174 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_261, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1174 = None
	        convert_element_type_781: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_261, torch.int8);  clamp_max_261 = None
	        view_2044: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_390, [sym_size_int, 1500, 1]);  clamp_min_390 = None
	        view_2045: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_780, [sym_size_int, 1500, 1]);  convert_element_type_780 = None
	        _assert_tensor_metadata_1175 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_781, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1175 = None
	        convert_element_type_782: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_781, torch.float32);  convert_element_type_781 = None
	        _assert_tensor_metadata_1176 = torch.ops.aten._assert_tensor_metadata.default(view_2045, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1176 = None
	        convert_element_type_783: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2045, torch.float32);  view_2045 = None
	        sub_5982: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_782, convert_element_type_783);  convert_element_type_782 = convert_element_type_783 = None
	        mul_12669: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5982, view_2044);  sub_5982 = view_2044 = None
	        _assert_tensor_metadata_1177 = torch.ops.aten._assert_tensor_metadata.default(mul_12669, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1177 = None
	        view_2047: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_21_fc1_parametrizations_weight_original0 = None
	        view_2048: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_21_fc1_parametrizations_weight_original1 = None
	        view_2049: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_21_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1178 = torch.ops.aten._assert_tensor_metadata.default(view_2047, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1178 = None
	        convert_element_type_784: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2047, torch.float32);  view_2047 = None
	        _assert_tensor_metadata_1179 = torch.ops.aten._assert_tensor_metadata.default(view_2049, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1179 = None
	        convert_element_type_785: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2049, torch.float32);  view_2049 = None
	        sub_5986: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_784, convert_element_type_785);  convert_element_type_784 = convert_element_type_785 = None
	        mul_12674: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5986, view_2048);  sub_5986 = view_2048 = None
	        view_2050: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12674, [5120, 1280]);  mul_12674 = None
	        _assert_tensor_metadata_1180 = torch.ops.aten._assert_tensor_metadata.default(view_2050, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1180 = None
	        mul_12679: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2051: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12669, [mul_12679, 1280]);  mul_12669 = mul_12679 = None
	        permute_219: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2050, [1, 0]);  view_2050 = None
	        addmm_108: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_21_fc1_bias, view_2051, permute_219);  model_audio_tower_layers_21_fc1_bias = view_2051 = permute_219 = None
	        view_2052: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_108, [sym_size_int, 1500, 5120]);  addmm_108 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_12686: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2052, 0.5)
	        mul_12687: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2052, 0.7071067811865476);  view_2052 = None
	        erf_23: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_12687);  mul_12687 = None
	        add_20076: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_23, 1);  erf_23 = None
	        mul_12688: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12686, add_20076);  mul_12686 = add_20076 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_12688, [2])
	        amax_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_12688, [2])
	        full_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_131, full_262);  amin_131 = full_262 = None
	        full_263: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_131, full_263);  amax_131 = full_263 = None
	        sub_5999: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_131, minimum_131);  maximum_131 = None
	        div_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5999, 255.0);  sub_5999 = None
	        clamp_min_393: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_262, 1.1920928955078125e-07);  div_262 = None
	        div_263: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_131, clamp_min_393);  minimum_131 = None
	        round_263: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_263);  div_263 = None
	        sub_6005: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_263);  round_263 = None
	        clamp_min_394: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6005, -128);  sub_6005 = None
	        clamp_max_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_394, 127);  clamp_min_394 = None
	        _assert_tensor_metadata_1181 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_393, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1181 = None
	        _assert_tensor_metadata_1182 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_262, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1182 = None
	        convert_element_type_786: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_262, torch.int8);  clamp_max_262 = None
	        view_2055: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_393, [sym_size_int, 1500, 1])
	        view_2056: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_786, [sym_size_int, 1500, 1])
	        reciprocal_131: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2055);  view_2055 = None
	        mul_12734: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_131, 1.0);  reciprocal_131 = None
	        mul_12737: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12688, mul_12734);  mul_12688 = mul_12734 = None
	        round_264: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_12737);  mul_12737 = None
	        add_20159: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_264, view_2056);  round_264 = view_2056 = None
	        clamp_min_395: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20159, -128);  add_20159 = None
	        clamp_max_263: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_395, 127);  clamp_min_395 = None
	        _assert_tensor_metadata_1183 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_263, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1183 = None
	        convert_element_type_787: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_263, torch.int8);  clamp_max_263 = None
	        view_2059: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_393, [sym_size_int, 1500, 1]);  clamp_min_393 = None
	        view_2060: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_786, [sym_size_int, 1500, 1]);  convert_element_type_786 = None
	        _assert_tensor_metadata_1184 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_787, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1184 = None
	        convert_element_type_788: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_787, torch.float32);  convert_element_type_787 = None
	        _assert_tensor_metadata_1185 = torch.ops.aten._assert_tensor_metadata.default(view_2060, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1185 = None
	        convert_element_type_789: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2060, torch.float32);  view_2060 = None
	        sub_6025: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_788, convert_element_type_789);  convert_element_type_788 = convert_element_type_789 = None
	        mul_12759: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6025, view_2059);  sub_6025 = view_2059 = None
	        _assert_tensor_metadata_1186 = torch.ops.aten._assert_tensor_metadata.default(mul_12759, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1186 = None
	        view_2062: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_21_fc2_parametrizations_weight_original0 = None
	        view_2063: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_21_fc2_parametrizations_weight_original1 = None
	        view_2064: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_21_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1187 = torch.ops.aten._assert_tensor_metadata.default(view_2062, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1187 = None
	        convert_element_type_790: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2062, torch.float32);  view_2062 = None
	        _assert_tensor_metadata_1188 = torch.ops.aten._assert_tensor_metadata.default(view_2064, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1188 = None
	        convert_element_type_791: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2064, torch.float32);  view_2064 = None
	        sub_6029: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_790, convert_element_type_791);  convert_element_type_790 = convert_element_type_791 = None
	        mul_12764: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6029, view_2063);  sub_6029 = view_2063 = None
	        view_2065: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12764, [1280, 5120]);  mul_12764 = None
	        _assert_tensor_metadata_1189 = torch.ops.aten._assert_tensor_metadata.default(view_2065, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1189 = None
	        mul_12769: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2066: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12759, [mul_12769, 5120]);  mul_12759 = mul_12769 = None
	        permute_220: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2065, [1, 0]);  view_2065 = None
	        addmm_109: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_21_fc2_bias, view_2066, permute_220);  model_audio_tower_layers_21_fc2_bias = view_2066 = permute_220 = None
	        view_2067: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_109, [sym_size_int, 1500, 1280]);  addmm_109 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_20222: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_19924, view_2067);  add_19924 = view_2067 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_177: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_20222, memory_format = torch.contiguous_format)
	        var_mean_44 = torch.ops.aten.var_mean.correction(clone_177, [2], correction = 0, keepdim = True)
	        getitem_176: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_44[0]
	        getitem_177: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_44[1];  var_mean_44 = None
	        add_20227: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_176, 1e-05);  getitem_176 = None
	        rsqrt_44: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_20227);  add_20227 = None
	        sub_6035: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_177, getitem_177);  clone_177 = getitem_177 = None
	        mul_12780: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6035, rsqrt_44);  sub_6035 = rsqrt_44 = None
	        mul_12781: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12780, model_audio_tower_layers_22_self_attn_layer_norm_weight);  mul_12780 = model_audio_tower_layers_22_self_attn_layer_norm_weight = None
	        add_20228: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12781, model_audio_tower_layers_22_self_attn_layer_norm_bias);  mul_12781 = model_audio_tower_layers_22_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_20228, [2])
	        amax_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_20228, [2])
	        full_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_132, full_264);  amin_132 = full_264 = None
	        full_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_132, full_265);  amax_132 = full_265 = None
	        sub_6046: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_132, minimum_132);  maximum_132 = None
	        div_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6046, 255.0);  sub_6046 = None
	        clamp_min_396: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_264, 1.1920928955078125e-07);  div_264 = None
	        div_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_132, clamp_min_396);  minimum_132 = None
	        round_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_265);  div_265 = None
	        sub_6052: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_265);  round_265 = None
	        clamp_min_397: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6052, -128);  sub_6052 = None
	        clamp_max_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_397, 127);  clamp_min_397 = None
	        _assert_tensor_metadata_1190 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_396, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1190 = None
	        _assert_tensor_metadata_1191 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_264, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1191 = None
	        convert_element_type_792: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_264, torch.int8);  clamp_max_264 = None
	        view_2070: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_396, [sym_size_int, 1500, 1])
	        view_2071: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_792, [sym_size_int, 1500, 1])
	        reciprocal_132: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2070);  view_2070 = None
	        mul_12829: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_132, 1.0);  reciprocal_132 = None
	        mul_12832: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_20228, mul_12829);  mul_12829 = None
	        round_266: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12832);  mul_12832 = None
	        add_20315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_266, view_2071);  round_266 = view_2071 = None
	        clamp_min_398: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20315, -128);  add_20315 = None
	        clamp_max_265: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_398, 127);  clamp_min_398 = None
	        _assert_tensor_metadata_1192 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_265, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1192 = None
	        convert_element_type_793: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_265, torch.int8);  clamp_max_265 = None
	        view_2074: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_396, [sym_size_int, 1500, 1]);  clamp_min_396 = None
	        view_2075: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_792, [sym_size_int, 1500, 1]);  convert_element_type_792 = None
	        _assert_tensor_metadata_1193 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_793, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1193 = None
	        convert_element_type_794: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_793, torch.float32);  convert_element_type_793 = None
	        _assert_tensor_metadata_1194 = torch.ops.aten._assert_tensor_metadata.default(view_2075, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1194 = None
	        convert_element_type_795: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2075, torch.float32);  view_2075 = None
	        sub_6072: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_794, convert_element_type_795);  convert_element_type_794 = convert_element_type_795 = None
	        mul_12854: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6072, view_2074);  sub_6072 = view_2074 = None
	        _assert_tensor_metadata_1195 = torch.ops.aten._assert_tensor_metadata.default(mul_12854, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1195 = None
	        view_2077: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2078: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2079: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1196 = torch.ops.aten._assert_tensor_metadata.default(view_2077, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1196 = None
	        convert_element_type_796: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2077, torch.float32);  view_2077 = None
	        _assert_tensor_metadata_1197 = torch.ops.aten._assert_tensor_metadata.default(view_2079, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1197 = None
	        convert_element_type_797: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2079, torch.float32);  view_2079 = None
	        sub_6076: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_796, convert_element_type_797);  convert_element_type_796 = convert_element_type_797 = None
	        mul_12859: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6076, view_2078);  sub_6076 = view_2078 = None
	        view_2080: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12859, [1280, 1280]);  mul_12859 = None
	        _assert_tensor_metadata_1198 = torch.ops.aten._assert_tensor_metadata.default(view_2080, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1198 = None
	        mul_12864: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2081: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12854, [mul_12864, 1280]);  mul_12854 = mul_12864 = None
	        permute_221: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2080, [1, 0]);  view_2080 = None
	        addmm_110: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_22_self_attn_q_proj_bias, view_2081, permute_221);  model_audio_tower_layers_22_self_attn_q_proj_bias = view_2081 = permute_221 = None
	        view_2082: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_110, [sym_size_int, 1500, 1280]);  addmm_110 = None
	        mul_12871: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2082, 0.125);  view_2082 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2083: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12871, [sym_size_int, 1500, 20, 64]);  mul_12871 = None
	        permute_222: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2083, [0, 2, 1, 3]);  view_2083 = None
	        clone_178: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_222, memory_format = torch.contiguous_format);  permute_222 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_20228, [2])
	        amax_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_20228, [2])
	        full_266: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_133, full_266);  amin_133 = full_266 = None
	        full_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_133, full_267);  amax_133 = full_267 = None
	        sub_6091: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_133, minimum_133);  maximum_133 = None
	        div_266: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6091, 255.0);  sub_6091 = None
	        clamp_min_399: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_266, 1.1920928955078125e-07);  div_266 = None
	        div_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_133, clamp_min_399);  minimum_133 = None
	        round_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_267);  div_267 = None
	        sub_6097: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_267);  round_267 = None
	        clamp_min_400: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6097, -128);  sub_6097 = None
	        clamp_max_266: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_400, 127);  clamp_min_400 = None
	        _assert_tensor_metadata_1199 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_399, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1199 = None
	        _assert_tensor_metadata_1200 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_266, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1200 = None
	        convert_element_type_798: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_266, torch.int8);  clamp_max_266 = None
	        view_2086: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_399, [sym_size_int, 1500, 1])
	        view_2087: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_798, [sym_size_int, 1500, 1])
	        reciprocal_133: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2086);  view_2086 = None
	        mul_12925: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_133, 1.0);  reciprocal_133 = None
	        mul_12928: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_20228, mul_12925);  mul_12925 = None
	        round_268: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12928);  mul_12928 = None
	        add_20467: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_268, view_2087);  round_268 = view_2087 = None
	        clamp_min_401: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20467, -128);  add_20467 = None
	        clamp_max_267: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_401, 127);  clamp_min_401 = None
	        _assert_tensor_metadata_1201 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_267, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1201 = None
	        convert_element_type_799: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_267, torch.int8);  clamp_max_267 = None
	        view_2090: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_399, [sym_size_int, 1500, 1]);  clamp_min_399 = None
	        view_2091: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_798, [sym_size_int, 1500, 1]);  convert_element_type_798 = None
	        _assert_tensor_metadata_1202 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_799, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1202 = None
	        convert_element_type_800: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_799, torch.float32);  convert_element_type_799 = None
	        _assert_tensor_metadata_1203 = torch.ops.aten._assert_tensor_metadata.default(view_2091, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1203 = None
	        convert_element_type_801: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2091, torch.float32);  view_2091 = None
	        sub_6117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_800, convert_element_type_801);  convert_element_type_800 = convert_element_type_801 = None
	        mul_12950: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6117, view_2090);  sub_6117 = view_2090 = None
	        _assert_tensor_metadata_1204 = torch.ops.aten._assert_tensor_metadata.default(mul_12950, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1204 = None
	        view_2093: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2094: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2095: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1205 = torch.ops.aten._assert_tensor_metadata.default(view_2093, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1205 = None
	        convert_element_type_802: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2093, torch.float32);  view_2093 = None
	        _assert_tensor_metadata_1206 = torch.ops.aten._assert_tensor_metadata.default(view_2095, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1206 = None
	        convert_element_type_803: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2095, torch.float32);  view_2095 = None
	        sub_6121: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_802, convert_element_type_803);  convert_element_type_802 = convert_element_type_803 = None
	        mul_12955: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6121, view_2094);  sub_6121 = view_2094 = None
	        view_2096: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12955, [1280, 1280]);  mul_12955 = None
	        _assert_tensor_metadata_1207 = torch.ops.aten._assert_tensor_metadata.default(view_2096, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1207 = None
	        permute_223: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2096, [1, 0]);  view_2096 = None
	        mul_12958: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2097: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12950, [mul_12958, 1280]);  mul_12950 = mul_12958 = None
	        mm_22: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2097, permute_223);  view_2097 = permute_223 = None
	        view_2098: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_22, [sym_size_int, 1500, 1280]);  mm_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2099: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2098, [sym_size_int, -1, 20, 64]);  view_2098 = None
	        permute_224: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2099, [0, 2, 1, 3]);  view_2099 = None
	        clone_179: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_224, memory_format = torch.contiguous_format);  permute_224 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_20228, [2])
	        amax_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_20228, [2])
	        full_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_134, full_268);  amin_134 = full_268 = None
	        full_269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_134, full_269);  amax_134 = full_269 = None
	        sub_6135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_134, minimum_134);  maximum_134 = None
	        div_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6135, 255.0);  sub_6135 = None
	        clamp_min_402: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_268, 1.1920928955078125e-07);  div_268 = None
	        div_269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_134, clamp_min_402);  minimum_134 = None
	        round_269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_269);  div_269 = None
	        sub_6141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_269);  round_269 = None
	        clamp_min_403: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6141, -128);  sub_6141 = None
	        clamp_max_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_403, 127);  clamp_min_403 = None
	        _assert_tensor_metadata_1208 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_402, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1208 = None
	        _assert_tensor_metadata_1209 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_268, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1209 = None
	        convert_element_type_804: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_268, torch.int8);  clamp_max_268 = None
	        view_2102: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_402, [sym_size_int, 1500, 1])
	        view_2103: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_804, [sym_size_int, 1500, 1])
	        reciprocal_134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2102);  view_2102 = None
	        mul_13024: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_134, 1.0);  reciprocal_134 = None
	        mul_13027: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_20228, mul_13024);  add_20228 = mul_13024 = None
	        round_270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13027);  mul_13027 = None
	        add_20615: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_270, view_2103);  round_270 = view_2103 = None
	        clamp_min_404: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20615, -128);  add_20615 = None
	        clamp_max_269: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_404, 127);  clamp_min_404 = None
	        _assert_tensor_metadata_1210 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_269, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1210 = None
	        convert_element_type_805: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_269, torch.int8);  clamp_max_269 = None
	        view_2106: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_402, [sym_size_int, 1500, 1]);  clamp_min_402 = None
	        view_2107: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_804, [sym_size_int, 1500, 1]);  convert_element_type_804 = None
	        _assert_tensor_metadata_1211 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_805, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1211 = None
	        convert_element_type_806: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_805, torch.float32);  convert_element_type_805 = None
	        _assert_tensor_metadata_1212 = torch.ops.aten._assert_tensor_metadata.default(view_2107, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1212 = None
	        convert_element_type_807: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2107, torch.float32);  view_2107 = None
	        sub_6161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_806, convert_element_type_807);  convert_element_type_806 = convert_element_type_807 = None
	        mul_13049: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6161, view_2106);  sub_6161 = view_2106 = None
	        _assert_tensor_metadata_1213 = torch.ops.aten._assert_tensor_metadata.default(mul_13049, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1213 = None
	        view_2109: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2110: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2111: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1214 = torch.ops.aten._assert_tensor_metadata.default(view_2109, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1214 = None
	        convert_element_type_808: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2109, torch.float32);  view_2109 = None
	        _assert_tensor_metadata_1215 = torch.ops.aten._assert_tensor_metadata.default(view_2111, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1215 = None
	        convert_element_type_809: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2111, torch.float32);  view_2111 = None
	        sub_6165: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_808, convert_element_type_809);  convert_element_type_808 = convert_element_type_809 = None
	        mul_13054: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6165, view_2110);  sub_6165 = view_2110 = None
	        view_2112: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13054, [1280, 1280]);  mul_13054 = None
	        _assert_tensor_metadata_1216 = torch.ops.aten._assert_tensor_metadata.default(view_2112, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1216 = None
	        mul_13059: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2113: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13049, [mul_13059, 1280]);  mul_13049 = mul_13059 = None
	        permute_225: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2112, [1, 0]);  view_2112 = None
	        addmm_111: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_22_self_attn_v_proj_bias, view_2113, permute_225);  model_audio_tower_layers_22_self_attn_v_proj_bias = view_2113 = permute_225 = None
	        view_2114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_111, [sym_size_int, 1500, 1280]);  addmm_111 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2115: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2114, [sym_size_int, -1, 20, 64]);  view_2114 = None
	        permute_226: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2115, [0, 2, 1, 3]);  view_2115 = None
	        clone_180: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_226, memory_format = torch.contiguous_format);  permute_226 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_22 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_178, clone_179, clone_180, None, False, scale = 1.0);  clone_178 = clone_179 = clone_180 = None
	        getitem_178: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_22[0];  _scaled_dot_product_efficient_attention_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_227: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_178, [0, 2, 1, 3]);  getitem_178 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_227, [sym_size_int, 1500, -1]);  permute_227 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2116, [2])
	        amax_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2116, [2])
	        full_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_135, full_270);  amin_135 = full_270 = None
	        full_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_135, full_271);  amax_135 = full_271 = None
	        sub_6183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_135, minimum_135);  maximum_135 = None
	        div_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6183, 255.0);  sub_6183 = None
	        clamp_min_405: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_270, 1.1920928955078125e-07);  div_270 = None
	        div_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_135, clamp_min_405);  minimum_135 = None
	        round_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_271);  div_271 = None
	        sub_6189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_271);  round_271 = None
	        clamp_min_406: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6189, -128);  sub_6189 = None
	        clamp_max_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_406, 127);  clamp_min_406 = None
	        _assert_tensor_metadata_1217 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_405, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1217 = None
	        _assert_tensor_metadata_1218 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_270, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1218 = None
	        convert_element_type_810: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_270, torch.int8);  clamp_max_270 = None
	        view_2119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_405, [sym_size_int, 1500, 1])
	        view_2120: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_810, [sym_size_int, 1500, 1])
	        reciprocal_135: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2119);  view_2119 = None
	        mul_13129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_135, 1.0);  reciprocal_135 = None
	        mul_13132: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2116, mul_13129);  view_2116 = mul_13129 = None
	        round_272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13132);  mul_13132 = None
	        add_20779: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_272, view_2120);  round_272 = view_2120 = None
	        clamp_min_407: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20779, -128);  add_20779 = None
	        clamp_max_271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_407, 127);  clamp_min_407 = None
	        _assert_tensor_metadata_1219 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_271, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1219 = None
	        convert_element_type_811: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_271, torch.int8);  clamp_max_271 = None
	        view_2123: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_405, [sym_size_int, 1500, 1]);  clamp_min_405 = None
	        view_2124: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_810, [sym_size_int, 1500, 1]);  convert_element_type_810 = None
	        _assert_tensor_metadata_1220 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_811, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1220 = None
	        convert_element_type_812: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_811, torch.float32);  convert_element_type_811 = None
	        _assert_tensor_metadata_1221 = torch.ops.aten._assert_tensor_metadata.default(view_2124, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1221 = None
	        convert_element_type_813: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2124, torch.float32);  view_2124 = None
	        sub_6209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_812, convert_element_type_813);  convert_element_type_812 = convert_element_type_813 = None
	        mul_13154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6209, view_2123);  sub_6209 = view_2123 = None
	        _assert_tensor_metadata_1222 = torch.ops.aten._assert_tensor_metadata.default(mul_13154, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1222 = None
	        view_2126: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2127: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2128: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1223 = torch.ops.aten._assert_tensor_metadata.default(view_2126, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1223 = None
	        convert_element_type_814: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2126, torch.float32);  view_2126 = None
	        _assert_tensor_metadata_1224 = torch.ops.aten._assert_tensor_metadata.default(view_2128, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1224 = None
	        convert_element_type_815: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2128, torch.float32);  view_2128 = None
	        sub_6213: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_814, convert_element_type_815);  convert_element_type_814 = convert_element_type_815 = None
	        mul_13159: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6213, view_2127);  sub_6213 = view_2127 = None
	        view_2129: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13159, [1280, 1280]);  mul_13159 = None
	        _assert_tensor_metadata_1225 = torch.ops.aten._assert_tensor_metadata.default(view_2129, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1225 = None
	        mul_13164: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2130: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13154, [mul_13164, 1280]);  mul_13154 = mul_13164 = None
	        permute_228: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2129, [1, 0]);  view_2129 = None
	        addmm_112: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_22_self_attn_out_proj_bias, view_2130, permute_228);  model_audio_tower_layers_22_self_attn_out_proj_bias = view_2130 = permute_228 = None
	        view_2131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_112, [sym_size_int, 1500, 1280]);  addmm_112 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_20842: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_20222, view_2131);  add_20222 = view_2131 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_20842, memory_format = torch.contiguous_format)
	        var_mean_45 = torch.ops.aten.var_mean.correction(clone_182, [2], correction = 0, keepdim = True)
	        getitem_182: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_45[0]
	        getitem_183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_45[1];  var_mean_45 = None
	        add_20847: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_182, 1e-05);  getitem_182 = None
	        rsqrt_45: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_20847);  add_20847 = None
	        sub_6219: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_182, getitem_183);  clone_182 = getitem_183 = None
	        mul_13175: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6219, rsqrt_45);  sub_6219 = rsqrt_45 = None
	        mul_13176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13175, model_audio_tower_layers_22_final_layer_norm_weight);  mul_13175 = model_audio_tower_layers_22_final_layer_norm_weight = None
	        add_20848: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13176, model_audio_tower_layers_22_final_layer_norm_bias);  mul_13176 = model_audio_tower_layers_22_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_20848, [2])
	        amax_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_20848, [2])
	        full_272: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_136, full_272);  amin_136 = full_272 = None
	        full_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_136, full_273);  amax_136 = full_273 = None
	        sub_6230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_136, minimum_136);  maximum_136 = None
	        div_272: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6230, 255.0);  sub_6230 = None
	        clamp_min_408: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_272, 1.1920928955078125e-07);  div_272 = None
	        div_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_136, clamp_min_408);  minimum_136 = None
	        round_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_273);  div_273 = None
	        sub_6236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_273);  round_273 = None
	        clamp_min_409: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6236, -128);  sub_6236 = None
	        clamp_max_272: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_409, 127);  clamp_min_409 = None
	        _assert_tensor_metadata_1226 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_408, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1226 = None
	        _assert_tensor_metadata_1227 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_272, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1227 = None
	        convert_element_type_816: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_272, torch.int8);  clamp_max_272 = None
	        view_2134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_408, [sym_size_int, 1500, 1])
	        view_2135: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_816, [sym_size_int, 1500, 1])
	        reciprocal_136: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2134);  view_2134 = None
	        mul_13224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_136, 1.0);  reciprocal_136 = None
	        mul_13227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_20848, mul_13224);  add_20848 = mul_13224 = None
	        round_274: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13227);  mul_13227 = None
	        add_20935: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_274, view_2135);  round_274 = view_2135 = None
	        clamp_min_410: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20935, -128);  add_20935 = None
	        clamp_max_273: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_410, 127);  clamp_min_410 = None
	        _assert_tensor_metadata_1228 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_273, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1228 = None
	        convert_element_type_817: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_273, torch.int8);  clamp_max_273 = None
	        view_2138: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_408, [sym_size_int, 1500, 1]);  clamp_min_408 = None
	        view_2139: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_816, [sym_size_int, 1500, 1]);  convert_element_type_816 = None
	        _assert_tensor_metadata_1229 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_817, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1229 = None
	        convert_element_type_818: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_817, torch.float32);  convert_element_type_817 = None
	        _assert_tensor_metadata_1230 = torch.ops.aten._assert_tensor_metadata.default(view_2139, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1230 = None
	        convert_element_type_819: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2139, torch.float32);  view_2139 = None
	        sub_6256: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_818, convert_element_type_819);  convert_element_type_818 = convert_element_type_819 = None
	        mul_13249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6256, view_2138);  sub_6256 = view_2138 = None
	        _assert_tensor_metadata_1231 = torch.ops.aten._assert_tensor_metadata.default(mul_13249, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1231 = None
	        view_2141: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_22_fc1_parametrizations_weight_original0 = None
	        view_2142: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_22_fc1_parametrizations_weight_original1 = None
	        view_2143: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_22_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1232 = torch.ops.aten._assert_tensor_metadata.default(view_2141, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1232 = None
	        convert_element_type_820: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2141, torch.float32);  view_2141 = None
	        _assert_tensor_metadata_1233 = torch.ops.aten._assert_tensor_metadata.default(view_2143, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1233 = None
	        convert_element_type_821: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2143, torch.float32);  view_2143 = None
	        sub_6260: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_820, convert_element_type_821);  convert_element_type_820 = convert_element_type_821 = None
	        mul_13254: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6260, view_2142);  sub_6260 = view_2142 = None
	        view_2144: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13254, [5120, 1280]);  mul_13254 = None
	        _assert_tensor_metadata_1234 = torch.ops.aten._assert_tensor_metadata.default(view_2144, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1234 = None
	        mul_13259: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2145: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13249, [mul_13259, 1280]);  mul_13249 = mul_13259 = None
	        permute_229: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2144, [1, 0]);  view_2144 = None
	        addmm_113: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_22_fc1_bias, view_2145, permute_229);  model_audio_tower_layers_22_fc1_bias = view_2145 = permute_229 = None
	        view_2146: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_113, [sym_size_int, 1500, 5120]);  addmm_113 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_13266: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2146, 0.5)
	        mul_13267: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2146, 0.7071067811865476);  view_2146 = None
	        erf_24: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_13267);  mul_13267 = None
	        add_20994: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_24, 1);  erf_24 = None
	        mul_13268: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13266, add_20994);  mul_13266 = add_20994 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_13268, [2])
	        amax_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_13268, [2])
	        full_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_137, full_274);  amin_137 = full_274 = None
	        full_275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_137, full_275);  amax_137 = full_275 = None
	        sub_6273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_137, minimum_137);  maximum_137 = None
	        div_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6273, 255.0);  sub_6273 = None
	        clamp_min_411: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_274, 1.1920928955078125e-07);  div_274 = None
	        div_275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_137, clamp_min_411);  minimum_137 = None
	        round_275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_275);  div_275 = None
	        sub_6279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_275);  round_275 = None
	        clamp_min_412: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6279, -128);  sub_6279 = None
	        clamp_max_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_412, 127);  clamp_min_412 = None
	        _assert_tensor_metadata_1235 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_411, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1235 = None
	        _assert_tensor_metadata_1236 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_274, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1236 = None
	        convert_element_type_822: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_274, torch.int8);  clamp_max_274 = None
	        view_2149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_411, [sym_size_int, 1500, 1])
	        view_2150: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_822, [sym_size_int, 1500, 1])
	        reciprocal_137: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2149);  view_2149 = None
	        mul_13314: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_137, 1.0);  reciprocal_137 = None
	        mul_13317: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13268, mul_13314);  mul_13268 = mul_13314 = None
	        round_276: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_13317);  mul_13317 = None
	        add_21077: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_276, view_2150);  round_276 = view_2150 = None
	        clamp_min_413: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21077, -128);  add_21077 = None
	        clamp_max_275: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_413, 127);  clamp_min_413 = None
	        _assert_tensor_metadata_1237 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_275, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1237 = None
	        convert_element_type_823: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_275, torch.int8);  clamp_max_275 = None
	        view_2153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_411, [sym_size_int, 1500, 1]);  clamp_min_411 = None
	        view_2154: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_822, [sym_size_int, 1500, 1]);  convert_element_type_822 = None
	        _assert_tensor_metadata_1238 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_823, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1238 = None
	        convert_element_type_824: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_823, torch.float32);  convert_element_type_823 = None
	        _assert_tensor_metadata_1239 = torch.ops.aten._assert_tensor_metadata.default(view_2154, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1239 = None
	        convert_element_type_825: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2154, torch.float32);  view_2154 = None
	        sub_6299: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_824, convert_element_type_825);  convert_element_type_824 = convert_element_type_825 = None
	        mul_13339: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6299, view_2153);  sub_6299 = view_2153 = None
	        _assert_tensor_metadata_1240 = torch.ops.aten._assert_tensor_metadata.default(mul_13339, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1240 = None
	        view_2156: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_22_fc2_parametrizations_weight_original0 = None
	        view_2157: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_22_fc2_parametrizations_weight_original1 = None
	        view_2158: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_22_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1241 = torch.ops.aten._assert_tensor_metadata.default(view_2156, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1241 = None
	        convert_element_type_826: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2156, torch.float32);  view_2156 = None
	        _assert_tensor_metadata_1242 = torch.ops.aten._assert_tensor_metadata.default(view_2158, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1242 = None
	        convert_element_type_827: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2158, torch.float32);  view_2158 = None
	        sub_6303: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_826, convert_element_type_827);  convert_element_type_826 = convert_element_type_827 = None
	        mul_13344: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6303, view_2157);  sub_6303 = view_2157 = None
	        view_2159: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13344, [1280, 5120]);  mul_13344 = None
	        _assert_tensor_metadata_1243 = torch.ops.aten._assert_tensor_metadata.default(view_2159, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1243 = None
	        mul_13349: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2160: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13339, [mul_13349, 5120]);  mul_13339 = mul_13349 = None
	        permute_230: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2159, [1, 0]);  view_2159 = None
	        addmm_114: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_22_fc2_bias, view_2160, permute_230);  model_audio_tower_layers_22_fc2_bias = view_2160 = permute_230 = None
	        view_2161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_114, [sym_size_int, 1500, 1280]);  addmm_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_21140: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_20842, view_2161);  add_20842 = view_2161 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_21140, memory_format = torch.contiguous_format)
	        var_mean_46 = torch.ops.aten.var_mean.correction(clone_185, [2], correction = 0, keepdim = True)
	        getitem_184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_46[0]
	        getitem_185: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_46[1];  var_mean_46 = None
	        add_21145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_184, 1e-05);  getitem_184 = None
	        rsqrt_46: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_21145);  add_21145 = None
	        sub_6309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_185, getitem_185);  clone_185 = getitem_185 = None
	        mul_13360: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6309, rsqrt_46);  sub_6309 = rsqrt_46 = None
	        mul_13361: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13360, model_audio_tower_layers_23_self_attn_layer_norm_weight);  mul_13360 = model_audio_tower_layers_23_self_attn_layer_norm_weight = None
	        add_21146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13361, model_audio_tower_layers_23_self_attn_layer_norm_bias);  mul_13361 = model_audio_tower_layers_23_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_21146, [2])
	        amax_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_21146, [2])
	        full_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_138, full_276);  amin_138 = full_276 = None
	        full_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_138, full_277);  amax_138 = full_277 = None
	        sub_6320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_138, minimum_138);  maximum_138 = None
	        div_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6320, 255.0);  sub_6320 = None
	        clamp_min_414: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_276, 1.1920928955078125e-07);  div_276 = None
	        div_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_138, clamp_min_414);  minimum_138 = None
	        round_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_277);  div_277 = None
	        sub_6326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_277);  round_277 = None
	        clamp_min_415: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6326, -128);  sub_6326 = None
	        clamp_max_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_415, 127);  clamp_min_415 = None
	        _assert_tensor_metadata_1244 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_414, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1244 = None
	        _assert_tensor_metadata_1245 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_276, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1245 = None
	        convert_element_type_828: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_276, torch.int8);  clamp_max_276 = None
	        view_2164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_414, [sym_size_int, 1500, 1])
	        view_2165: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_828, [sym_size_int, 1500, 1])
	        reciprocal_138: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2164);  view_2164 = None
	        mul_13409: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_138, 1.0);  reciprocal_138 = None
	        mul_13412: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_21146, mul_13409);  mul_13409 = None
	        round_278: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13412);  mul_13412 = None
	        add_21233: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_278, view_2165);  round_278 = view_2165 = None
	        clamp_min_416: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21233, -128);  add_21233 = None
	        clamp_max_277: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_416, 127);  clamp_min_416 = None
	        _assert_tensor_metadata_1246 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_277, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1246 = None
	        convert_element_type_829: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_277, torch.int8);  clamp_max_277 = None
	        view_2168: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_414, [sym_size_int, 1500, 1]);  clamp_min_414 = None
	        view_2169: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_828, [sym_size_int, 1500, 1]);  convert_element_type_828 = None
	        _assert_tensor_metadata_1247 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_829, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1247 = None
	        convert_element_type_830: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_829, torch.float32);  convert_element_type_829 = None
	        _assert_tensor_metadata_1248 = torch.ops.aten._assert_tensor_metadata.default(view_2169, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1248 = None
	        convert_element_type_831: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2169, torch.float32);  view_2169 = None
	        sub_6346: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_830, convert_element_type_831);  convert_element_type_830 = convert_element_type_831 = None
	        mul_13434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6346, view_2168);  sub_6346 = view_2168 = None
	        _assert_tensor_metadata_1249 = torch.ops.aten._assert_tensor_metadata.default(mul_13434, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1249 = None
	        view_2171: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2172: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2173: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1250 = torch.ops.aten._assert_tensor_metadata.default(view_2171, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1250 = None
	        convert_element_type_832: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2171, torch.float32);  view_2171 = None
	        _assert_tensor_metadata_1251 = torch.ops.aten._assert_tensor_metadata.default(view_2173, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1251 = None
	        convert_element_type_833: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2173, torch.float32);  view_2173 = None
	        sub_6350: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_832, convert_element_type_833);  convert_element_type_832 = convert_element_type_833 = None
	        mul_13439: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6350, view_2172);  sub_6350 = view_2172 = None
	        view_2174: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13439, [1280, 1280]);  mul_13439 = None
	        _assert_tensor_metadata_1252 = torch.ops.aten._assert_tensor_metadata.default(view_2174, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1252 = None
	        mul_13444: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2175: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13434, [mul_13444, 1280]);  mul_13434 = mul_13444 = None
	        permute_231: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2174, [1, 0]);  view_2174 = None
	        addmm_115: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_23_self_attn_q_proj_bias, view_2175, permute_231);  model_audio_tower_layers_23_self_attn_q_proj_bias = view_2175 = permute_231 = None
	        view_2176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_115, [sym_size_int, 1500, 1280]);  addmm_115 = None
	        mul_13451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2176, 0.125);  view_2176 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2177: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13451, [sym_size_int, 1500, 20, 64]);  mul_13451 = None
	        permute_232: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2177, [0, 2, 1, 3]);  view_2177 = None
	        clone_186: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_232, memory_format = torch.contiguous_format);  permute_232 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_21146, [2])
	        amax_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_21146, [2])
	        full_278: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_139, full_278);  amin_139 = full_278 = None
	        full_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_139, full_279);  amax_139 = full_279 = None
	        sub_6365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_139, minimum_139);  maximum_139 = None
	        div_278: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6365, 255.0);  sub_6365 = None
	        clamp_min_417: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_278, 1.1920928955078125e-07);  div_278 = None
	        div_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_139, clamp_min_417);  minimum_139 = None
	        round_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_279);  div_279 = None
	        sub_6371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_279);  round_279 = None
	        clamp_min_418: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6371, -128);  sub_6371 = None
	        clamp_max_278: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_418, 127);  clamp_min_418 = None
	        _assert_tensor_metadata_1253 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_417, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1253 = None
	        _assert_tensor_metadata_1254 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_278, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1254 = None
	        convert_element_type_834: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_278, torch.int8);  clamp_max_278 = None
	        view_2180: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_417, [sym_size_int, 1500, 1])
	        view_2181: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_834, [sym_size_int, 1500, 1])
	        reciprocal_139: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2180);  view_2180 = None
	        mul_13505: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_139, 1.0);  reciprocal_139 = None
	        mul_13508: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_21146, mul_13505);  mul_13505 = None
	        round_280: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13508);  mul_13508 = None
	        add_21385: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_280, view_2181);  round_280 = view_2181 = None
	        clamp_min_419: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21385, -128);  add_21385 = None
	        clamp_max_279: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_419, 127);  clamp_min_419 = None
	        _assert_tensor_metadata_1255 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_279, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1255 = None
	        convert_element_type_835: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_279, torch.int8);  clamp_max_279 = None
	        view_2184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_417, [sym_size_int, 1500, 1]);  clamp_min_417 = None
	        view_2185: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_834, [sym_size_int, 1500, 1]);  convert_element_type_834 = None
	        _assert_tensor_metadata_1256 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_835, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1256 = None
	        convert_element_type_836: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_835, torch.float32);  convert_element_type_835 = None
	        _assert_tensor_metadata_1257 = torch.ops.aten._assert_tensor_metadata.default(view_2185, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1257 = None
	        convert_element_type_837: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2185, torch.float32);  view_2185 = None
	        sub_6391: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_836, convert_element_type_837);  convert_element_type_836 = convert_element_type_837 = None
	        mul_13530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6391, view_2184);  sub_6391 = view_2184 = None
	        _assert_tensor_metadata_1258 = torch.ops.aten._assert_tensor_metadata.default(mul_13530, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1258 = None
	        view_2187: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2188: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2189: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1259 = torch.ops.aten._assert_tensor_metadata.default(view_2187, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1259 = None
	        convert_element_type_838: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2187, torch.float32);  view_2187 = None
	        _assert_tensor_metadata_1260 = torch.ops.aten._assert_tensor_metadata.default(view_2189, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1260 = None
	        convert_element_type_839: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2189, torch.float32);  view_2189 = None
	        sub_6395: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_838, convert_element_type_839);  convert_element_type_838 = convert_element_type_839 = None
	        mul_13535: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6395, view_2188);  sub_6395 = view_2188 = None
	        view_2190: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13535, [1280, 1280]);  mul_13535 = None
	        _assert_tensor_metadata_1261 = torch.ops.aten._assert_tensor_metadata.default(view_2190, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1261 = None
	        permute_233: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2190, [1, 0]);  view_2190 = None
	        mul_13538: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2191: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13530, [mul_13538, 1280]);  mul_13530 = mul_13538 = None
	        mm_23: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2191, permute_233);  view_2191 = permute_233 = None
	        view_2192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_23, [sym_size_int, 1500, 1280]);  mm_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2193: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2192, [sym_size_int, -1, 20, 64]);  view_2192 = None
	        permute_234: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2193, [0, 2, 1, 3]);  view_2193 = None
	        clone_187: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_234, memory_format = torch.contiguous_format);  permute_234 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_21146, [2])
	        amax_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_21146, [2])
	        full_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_140, full_280);  amin_140 = full_280 = None
	        full_281: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_140, full_281);  amax_140 = full_281 = None
	        sub_6409: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_140, minimum_140);  maximum_140 = None
	        div_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6409, 255.0);  sub_6409 = None
	        clamp_min_420: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_280, 1.1920928955078125e-07);  div_280 = None
	        div_281: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_140, clamp_min_420);  minimum_140 = None
	        round_281: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_281);  div_281 = None
	        sub_6415: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_281);  round_281 = None
	        clamp_min_421: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6415, -128);  sub_6415 = None
	        clamp_max_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_421, 127);  clamp_min_421 = None
	        _assert_tensor_metadata_1262 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_420, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1262 = None
	        _assert_tensor_metadata_1263 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_280, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1263 = None
	        convert_element_type_840: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_280, torch.int8);  clamp_max_280 = None
	        view_2196: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_420, [sym_size_int, 1500, 1])
	        view_2197: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_840, [sym_size_int, 1500, 1])
	        reciprocal_140: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2196);  view_2196 = None
	        mul_13604: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_140, 1.0);  reciprocal_140 = None
	        mul_13607: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_21146, mul_13604);  add_21146 = mul_13604 = None
	        round_282: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13607);  mul_13607 = None
	        add_21533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_282, view_2197);  round_282 = view_2197 = None
	        clamp_min_422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21533, -128);  add_21533 = None
	        clamp_max_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_422, 127);  clamp_min_422 = None
	        _assert_tensor_metadata_1264 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_281, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1264 = None
	        convert_element_type_841: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_281, torch.int8);  clamp_max_281 = None
	        view_2200: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_420, [sym_size_int, 1500, 1]);  clamp_min_420 = None
	        view_2201: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_840, [sym_size_int, 1500, 1]);  convert_element_type_840 = None
	        _assert_tensor_metadata_1265 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_841, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1265 = None
	        convert_element_type_842: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_841, torch.float32);  convert_element_type_841 = None
	        _assert_tensor_metadata_1266 = torch.ops.aten._assert_tensor_metadata.default(view_2201, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1266 = None
	        convert_element_type_843: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2201, torch.float32);  view_2201 = None
	        sub_6435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_842, convert_element_type_843);  convert_element_type_842 = convert_element_type_843 = None
	        mul_13629: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6435, view_2200);  sub_6435 = view_2200 = None
	        _assert_tensor_metadata_1267 = torch.ops.aten._assert_tensor_metadata.default(mul_13629, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1267 = None
	        view_2203: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2204: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2205: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1268 = torch.ops.aten._assert_tensor_metadata.default(view_2203, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1268 = None
	        convert_element_type_844: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2203, torch.float32);  view_2203 = None
	        _assert_tensor_metadata_1269 = torch.ops.aten._assert_tensor_metadata.default(view_2205, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1269 = None
	        convert_element_type_845: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2205, torch.float32);  view_2205 = None
	        sub_6439: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_844, convert_element_type_845);  convert_element_type_844 = convert_element_type_845 = None
	        mul_13634: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6439, view_2204);  sub_6439 = view_2204 = None
	        view_2206: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13634, [1280, 1280]);  mul_13634 = None
	        _assert_tensor_metadata_1270 = torch.ops.aten._assert_tensor_metadata.default(view_2206, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1270 = None
	        mul_13639: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2207: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13629, [mul_13639, 1280]);  mul_13629 = mul_13639 = None
	        permute_235: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2206, [1, 0]);  view_2206 = None
	        addmm_116: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_23_self_attn_v_proj_bias, view_2207, permute_235);  model_audio_tower_layers_23_self_attn_v_proj_bias = view_2207 = permute_235 = None
	        view_2208: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_116, [sym_size_int, 1500, 1280]);  addmm_116 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2209: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2208, [sym_size_int, -1, 20, 64]);  view_2208 = None
	        permute_236: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2209, [0, 2, 1, 3]);  view_2209 = None
	        clone_188: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_236, memory_format = torch.contiguous_format);  permute_236 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_23 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_186, clone_187, clone_188, None, False, scale = 1.0);  clone_186 = clone_187 = clone_188 = None
	        getitem_186: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_23[0];  _scaled_dot_product_efficient_attention_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_237: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_186, [0, 2, 1, 3]);  getitem_186 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2210: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_237, [sym_size_int, 1500, -1]);  permute_237 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2210, [2])
	        amax_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2210, [2])
	        full_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_141, full_282);  amin_141 = full_282 = None
	        full_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_141, full_283);  amax_141 = full_283 = None
	        sub_6457: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_141, minimum_141);  maximum_141 = None
	        div_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6457, 255.0);  sub_6457 = None
	        clamp_min_423: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_282, 1.1920928955078125e-07);  div_282 = None
	        div_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_141, clamp_min_423);  minimum_141 = None
	        round_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_283);  div_283 = None
	        sub_6463: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_283);  round_283 = None
	        clamp_min_424: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6463, -128);  sub_6463 = None
	        clamp_max_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_424, 127);  clamp_min_424 = None
	        _assert_tensor_metadata_1271 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_423, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1271 = None
	        _assert_tensor_metadata_1272 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_282, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1272 = None
	        convert_element_type_846: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_282, torch.int8);  clamp_max_282 = None
	        view_2213: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_423, [sym_size_int, 1500, 1])
	        view_2214: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_846, [sym_size_int, 1500, 1])
	        reciprocal_141: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2213);  view_2213 = None
	        mul_13709: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_141, 1.0);  reciprocal_141 = None
	        mul_13712: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2210, mul_13709);  view_2210 = mul_13709 = None
	        round_284: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13712);  mul_13712 = None
	        add_21697: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_284, view_2214);  round_284 = view_2214 = None
	        clamp_min_425: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21697, -128);  add_21697 = None
	        clamp_max_283: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_425, 127);  clamp_min_425 = None
	        _assert_tensor_metadata_1273 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_283, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1273 = None
	        convert_element_type_847: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_283, torch.int8);  clamp_max_283 = None
	        view_2217: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_423, [sym_size_int, 1500, 1]);  clamp_min_423 = None
	        view_2218: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_846, [sym_size_int, 1500, 1]);  convert_element_type_846 = None
	        _assert_tensor_metadata_1274 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_847, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1274 = None
	        convert_element_type_848: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_847, torch.float32);  convert_element_type_847 = None
	        _assert_tensor_metadata_1275 = torch.ops.aten._assert_tensor_metadata.default(view_2218, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1275 = None
	        convert_element_type_849: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2218, torch.float32);  view_2218 = None
	        sub_6483: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_848, convert_element_type_849);  convert_element_type_848 = convert_element_type_849 = None
	        mul_13734: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6483, view_2217);  sub_6483 = view_2217 = None
	        _assert_tensor_metadata_1276 = torch.ops.aten._assert_tensor_metadata.default(mul_13734, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1276 = None
	        view_2220: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2221: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2222: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1277 = torch.ops.aten._assert_tensor_metadata.default(view_2220, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1277 = None
	        convert_element_type_850: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2220, torch.float32);  view_2220 = None
	        _assert_tensor_metadata_1278 = torch.ops.aten._assert_tensor_metadata.default(view_2222, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1278 = None
	        convert_element_type_851: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2222, torch.float32);  view_2222 = None
	        sub_6487: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_850, convert_element_type_851);  convert_element_type_850 = convert_element_type_851 = None
	        mul_13739: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6487, view_2221);  sub_6487 = view_2221 = None
	        view_2223: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13739, [1280, 1280]);  mul_13739 = None
	        _assert_tensor_metadata_1279 = torch.ops.aten._assert_tensor_metadata.default(view_2223, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1279 = None
	        mul_13744: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2224: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13734, [mul_13744, 1280]);  mul_13734 = mul_13744 = None
	        permute_238: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2223, [1, 0]);  view_2223 = None
	        addmm_117: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_23_self_attn_out_proj_bias, view_2224, permute_238);  model_audio_tower_layers_23_self_attn_out_proj_bias = view_2224 = permute_238 = None
	        view_2225: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_117, [sym_size_int, 1500, 1280]);  addmm_117 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_21760: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_21140, view_2225);  add_21140 = view_2225 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_190: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_21760, memory_format = torch.contiguous_format)
	        var_mean_47 = torch.ops.aten.var_mean.correction(clone_190, [2], correction = 0, keepdim = True)
	        getitem_190: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_47[0]
	        getitem_191: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_47[1];  var_mean_47 = None
	        add_21765: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_190, 1e-05);  getitem_190 = None
	        rsqrt_47: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_21765);  add_21765 = None
	        sub_6493: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_190, getitem_191);  clone_190 = getitem_191 = None
	        mul_13755: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6493, rsqrt_47);  sub_6493 = rsqrt_47 = None
	        mul_13756: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13755, model_audio_tower_layers_23_final_layer_norm_weight);  mul_13755 = model_audio_tower_layers_23_final_layer_norm_weight = None
	        add_21766: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13756, model_audio_tower_layers_23_final_layer_norm_bias);  mul_13756 = model_audio_tower_layers_23_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_21766, [2])
	        amax_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_21766, [2])
	        full_284: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_142, full_284);  amin_142 = full_284 = None
	        full_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_142, full_285);  amax_142 = full_285 = None
	        sub_6504: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_142, minimum_142);  maximum_142 = None
	        div_284: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6504, 255.0);  sub_6504 = None
	        clamp_min_426: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_284, 1.1920928955078125e-07);  div_284 = None
	        div_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_142, clamp_min_426);  minimum_142 = None
	        round_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_285);  div_285 = None
	        sub_6510: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_285);  round_285 = None
	        clamp_min_427: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6510, -128);  sub_6510 = None
	        clamp_max_284: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_427, 127);  clamp_min_427 = None
	        _assert_tensor_metadata_1280 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_426, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1280 = None
	        _assert_tensor_metadata_1281 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_284, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1281 = None
	        convert_element_type_852: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_284, torch.int8);  clamp_max_284 = None
	        view_2228: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_426, [sym_size_int, 1500, 1])
	        view_2229: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_852, [sym_size_int, 1500, 1])
	        reciprocal_142: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2228);  view_2228 = None
	        mul_13804: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_142, 1.0);  reciprocal_142 = None
	        mul_13807: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_21766, mul_13804);  add_21766 = mul_13804 = None
	        round_286: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13807);  mul_13807 = None
	        add_21853: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_286, view_2229);  round_286 = view_2229 = None
	        clamp_min_428: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21853, -128);  add_21853 = None
	        clamp_max_285: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_428, 127);  clamp_min_428 = None
	        _assert_tensor_metadata_1282 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_285, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1282 = None
	        convert_element_type_853: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_285, torch.int8);  clamp_max_285 = None
	        view_2232: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_426, [sym_size_int, 1500, 1]);  clamp_min_426 = None
	        view_2233: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_852, [sym_size_int, 1500, 1]);  convert_element_type_852 = None
	        _assert_tensor_metadata_1283 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_853, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1283 = None
	        convert_element_type_854: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_853, torch.float32);  convert_element_type_853 = None
	        _assert_tensor_metadata_1284 = torch.ops.aten._assert_tensor_metadata.default(view_2233, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1284 = None
	        convert_element_type_855: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2233, torch.float32);  view_2233 = None
	        sub_6530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_854, convert_element_type_855);  convert_element_type_854 = convert_element_type_855 = None
	        mul_13829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6530, view_2232);  sub_6530 = view_2232 = None
	        _assert_tensor_metadata_1285 = torch.ops.aten._assert_tensor_metadata.default(mul_13829, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1285 = None
	        view_2235: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_23_fc1_parametrizations_weight_original0 = None
	        view_2236: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_23_fc1_parametrizations_weight_original1 = None
	        view_2237: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_23_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1286 = torch.ops.aten._assert_tensor_metadata.default(view_2235, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1286 = None
	        convert_element_type_856: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2235, torch.float32);  view_2235 = None
	        _assert_tensor_metadata_1287 = torch.ops.aten._assert_tensor_metadata.default(view_2237, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1287 = None
	        convert_element_type_857: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2237, torch.float32);  view_2237 = None
	        sub_6534: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_856, convert_element_type_857);  convert_element_type_856 = convert_element_type_857 = None
	        mul_13834: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6534, view_2236);  sub_6534 = view_2236 = None
	        view_2238: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13834, [5120, 1280]);  mul_13834 = None
	        _assert_tensor_metadata_1288 = torch.ops.aten._assert_tensor_metadata.default(view_2238, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1288 = None
	        mul_13839: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2239: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13829, [mul_13839, 1280]);  mul_13829 = mul_13839 = None
	        permute_239: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2238, [1, 0]);  view_2238 = None
	        addmm_118: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_23_fc1_bias, view_2239, permute_239);  model_audio_tower_layers_23_fc1_bias = view_2239 = permute_239 = None
	        view_2240: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_118, [sym_size_int, 1500, 5120]);  addmm_118 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_13846: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2240, 0.5)
	        mul_13847: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2240, 0.7071067811865476);  view_2240 = None
	        erf_25: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_13847);  mul_13847 = None
	        add_21912: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_25, 1);  erf_25 = None
	        mul_13848: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13846, add_21912);  mul_13846 = add_21912 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_13848, [2])
	        amax_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_13848, [2])
	        full_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_143, full_286);  amin_143 = full_286 = None
	        full_287: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_143, full_287);  amax_143 = full_287 = None
	        sub_6547: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_143, minimum_143);  maximum_143 = None
	        div_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6547, 255.0);  sub_6547 = None
	        clamp_min_429: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_286, 1.1920928955078125e-07);  div_286 = None
	        div_287: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_143, clamp_min_429);  minimum_143 = None
	        round_287: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_287);  div_287 = None
	        sub_6553: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_287);  round_287 = None
	        clamp_min_430: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6553, -128);  sub_6553 = None
	        clamp_max_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_430, 127);  clamp_min_430 = None
	        _assert_tensor_metadata_1289 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_429, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1289 = None
	        _assert_tensor_metadata_1290 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_286, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1290 = None
	        convert_element_type_858: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_286, torch.int8);  clamp_max_286 = None
	        view_2243: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_429, [sym_size_int, 1500, 1])
	        view_2244: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_858, [sym_size_int, 1500, 1])
	        reciprocal_143: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2243);  view_2243 = None
	        mul_13894: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_143, 1.0);  reciprocal_143 = None
	        mul_13897: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13848, mul_13894);  mul_13848 = mul_13894 = None
	        round_288: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_13897);  mul_13897 = None
	        add_21995: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_288, view_2244);  round_288 = view_2244 = None
	        clamp_min_431: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21995, -128);  add_21995 = None
	        clamp_max_287: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_431, 127);  clamp_min_431 = None
	        _assert_tensor_metadata_1291 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_287, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1291 = None
	        convert_element_type_859: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_287, torch.int8);  clamp_max_287 = None
	        view_2247: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_429, [sym_size_int, 1500, 1]);  clamp_min_429 = None
	        view_2248: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_858, [sym_size_int, 1500, 1]);  convert_element_type_858 = None
	        _assert_tensor_metadata_1292 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_859, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1292 = None
	        convert_element_type_860: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_859, torch.float32);  convert_element_type_859 = None
	        _assert_tensor_metadata_1293 = torch.ops.aten._assert_tensor_metadata.default(view_2248, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1293 = None
	        convert_element_type_861: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2248, torch.float32);  view_2248 = None
	        sub_6573: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_860, convert_element_type_861);  convert_element_type_860 = convert_element_type_861 = None
	        mul_13919: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6573, view_2247);  sub_6573 = view_2247 = None
	        _assert_tensor_metadata_1294 = torch.ops.aten._assert_tensor_metadata.default(mul_13919, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1294 = None
	        view_2250: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_23_fc2_parametrizations_weight_original0 = None
	        view_2251: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_23_fc2_parametrizations_weight_original1 = None
	        view_2252: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_23_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1295 = torch.ops.aten._assert_tensor_metadata.default(view_2250, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1295 = None
	        convert_element_type_862: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2250, torch.float32);  view_2250 = None
	        _assert_tensor_metadata_1296 = torch.ops.aten._assert_tensor_metadata.default(view_2252, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1296 = None
	        convert_element_type_863: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2252, torch.float32);  view_2252 = None
	        sub_6577: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_862, convert_element_type_863);  convert_element_type_862 = convert_element_type_863 = None
	        mul_13924: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6577, view_2251);  sub_6577 = view_2251 = None
	        view_2253: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13924, [1280, 5120]);  mul_13924 = None
	        _assert_tensor_metadata_1297 = torch.ops.aten._assert_tensor_metadata.default(view_2253, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1297 = None
	        mul_13929: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2254: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13919, [mul_13929, 5120]);  mul_13919 = mul_13929 = None
	        permute_240: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2253, [1, 0]);  view_2253 = None
	        addmm_119: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_23_fc2_bias, view_2254, permute_240);  model_audio_tower_layers_23_fc2_bias = view_2254 = permute_240 = None
	        view_2255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_119, [sym_size_int, 1500, 1280]);  addmm_119 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_22058: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_21760, view_2255);  add_21760 = view_2255 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_193: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_22058, memory_format = torch.contiguous_format)
	        var_mean_48 = torch.ops.aten.var_mean.correction(clone_193, [2], correction = 0, keepdim = True)
	        getitem_192: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_48[0]
	        getitem_193: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_48[1];  var_mean_48 = None
	        add_22063: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_192, 1e-05);  getitem_192 = None
	        rsqrt_48: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_22063);  add_22063 = None
	        sub_6583: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_193, getitem_193);  clone_193 = getitem_193 = None
	        mul_13940: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6583, rsqrt_48);  sub_6583 = rsqrt_48 = None
	        mul_13941: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13940, model_audio_tower_layers_24_self_attn_layer_norm_weight);  mul_13940 = model_audio_tower_layers_24_self_attn_layer_norm_weight = None
	        add_22064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13941, model_audio_tower_layers_24_self_attn_layer_norm_bias);  mul_13941 = model_audio_tower_layers_24_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22064, [2])
	        amax_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22064, [2])
	        full_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_144, full_288);  amin_144 = full_288 = None
	        full_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_144, full_289);  amax_144 = full_289 = None
	        sub_6594: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_144, minimum_144);  maximum_144 = None
	        div_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6594, 255.0);  sub_6594 = None
	        clamp_min_432: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_288, 1.1920928955078125e-07);  div_288 = None
	        div_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_144, clamp_min_432);  minimum_144 = None
	        round_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_289);  div_289 = None
	        sub_6600: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_289);  round_289 = None
	        clamp_min_433: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6600, -128);  sub_6600 = None
	        clamp_max_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_433, 127);  clamp_min_433 = None
	        _assert_tensor_metadata_1298 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_432, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1298 = None
	        _assert_tensor_metadata_1299 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_288, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1299 = None
	        convert_element_type_864: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_288, torch.int8);  clamp_max_288 = None
	        view_2258: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_432, [sym_size_int, 1500, 1])
	        view_2259: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_864, [sym_size_int, 1500, 1])
	        reciprocal_144: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2258);  view_2258 = None
	        mul_13989: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_144, 1.0);  reciprocal_144 = None
	        mul_13992: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22064, mul_13989);  mul_13989 = None
	        round_290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13992);  mul_13992 = None
	        add_22151: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_290, view_2259);  round_290 = view_2259 = None
	        clamp_min_434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22151, -128);  add_22151 = None
	        clamp_max_289: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_434, 127);  clamp_min_434 = None
	        _assert_tensor_metadata_1300 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_289, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1300 = None
	        convert_element_type_865: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_289, torch.int8);  clamp_max_289 = None
	        view_2262: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_432, [sym_size_int, 1500, 1]);  clamp_min_432 = None
	        view_2263: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_864, [sym_size_int, 1500, 1]);  convert_element_type_864 = None
	        _assert_tensor_metadata_1301 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_865, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1301 = None
	        convert_element_type_866: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_865, torch.float32);  convert_element_type_865 = None
	        _assert_tensor_metadata_1302 = torch.ops.aten._assert_tensor_metadata.default(view_2263, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1302 = None
	        convert_element_type_867: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2263, torch.float32);  view_2263 = None
	        sub_6620: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_866, convert_element_type_867);  convert_element_type_866 = convert_element_type_867 = None
	        mul_14014: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6620, view_2262);  sub_6620 = view_2262 = None
	        _assert_tensor_metadata_1303 = torch.ops.aten._assert_tensor_metadata.default(mul_14014, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1303 = None
	        view_2265: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2266: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2267: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1304 = torch.ops.aten._assert_tensor_metadata.default(view_2265, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1304 = None
	        convert_element_type_868: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2265, torch.float32);  view_2265 = None
	        _assert_tensor_metadata_1305 = torch.ops.aten._assert_tensor_metadata.default(view_2267, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1305 = None
	        convert_element_type_869: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2267, torch.float32);  view_2267 = None
	        sub_6624: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_868, convert_element_type_869);  convert_element_type_868 = convert_element_type_869 = None
	        mul_14019: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6624, view_2266);  sub_6624 = view_2266 = None
	        view_2268: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14019, [1280, 1280]);  mul_14019 = None
	        _assert_tensor_metadata_1306 = torch.ops.aten._assert_tensor_metadata.default(view_2268, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1306 = None
	        mul_14024: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2269: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14014, [mul_14024, 1280]);  mul_14014 = mul_14024 = None
	        permute_241: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2268, [1, 0]);  view_2268 = None
	        addmm_120: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_24_self_attn_q_proj_bias, view_2269, permute_241);  model_audio_tower_layers_24_self_attn_q_proj_bias = view_2269 = permute_241 = None
	        view_2270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_120, [sym_size_int, 1500, 1280]);  addmm_120 = None
	        mul_14031: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2270, 0.125);  view_2270 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2271: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14031, [sym_size_int, 1500, 20, 64]);  mul_14031 = None
	        permute_242: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2271, [0, 2, 1, 3]);  view_2271 = None
	        clone_194: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_242, memory_format = torch.contiguous_format);  permute_242 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22064, [2])
	        amax_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22064, [2])
	        full_290: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_145, full_290);  amin_145 = full_290 = None
	        full_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_145, full_291);  amax_145 = full_291 = None
	        sub_6639: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_145, minimum_145);  maximum_145 = None
	        div_290: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6639, 255.0);  sub_6639 = None
	        clamp_min_435: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_290, 1.1920928955078125e-07);  div_290 = None
	        div_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_145, clamp_min_435);  minimum_145 = None
	        round_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_291);  div_291 = None
	        sub_6645: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_291);  round_291 = None
	        clamp_min_436: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6645, -128);  sub_6645 = None
	        clamp_max_290: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_436, 127);  clamp_min_436 = None
	        _assert_tensor_metadata_1307 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_435, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1307 = None
	        _assert_tensor_metadata_1308 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_290, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1308 = None
	        convert_element_type_870: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_290, torch.int8);  clamp_max_290 = None
	        view_2274: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_435, [sym_size_int, 1500, 1])
	        view_2275: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_870, [sym_size_int, 1500, 1])
	        reciprocal_145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2274);  view_2274 = None
	        mul_14085: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_145, 1.0);  reciprocal_145 = None
	        mul_14088: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22064, mul_14085);  mul_14085 = None
	        round_292: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14088);  mul_14088 = None
	        add_22303: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_292, view_2275);  round_292 = view_2275 = None
	        clamp_min_437: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22303, -128);  add_22303 = None
	        clamp_max_291: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_437, 127);  clamp_min_437 = None
	        _assert_tensor_metadata_1309 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_291, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1309 = None
	        convert_element_type_871: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_291, torch.int8);  clamp_max_291 = None
	        view_2278: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_435, [sym_size_int, 1500, 1]);  clamp_min_435 = None
	        view_2279: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_870, [sym_size_int, 1500, 1]);  convert_element_type_870 = None
	        _assert_tensor_metadata_1310 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_871, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1310 = None
	        convert_element_type_872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_871, torch.float32);  convert_element_type_871 = None
	        _assert_tensor_metadata_1311 = torch.ops.aten._assert_tensor_metadata.default(view_2279, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1311 = None
	        convert_element_type_873: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2279, torch.float32);  view_2279 = None
	        sub_6665: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_872, convert_element_type_873);  convert_element_type_872 = convert_element_type_873 = None
	        mul_14110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6665, view_2278);  sub_6665 = view_2278 = None
	        _assert_tensor_metadata_1312 = torch.ops.aten._assert_tensor_metadata.default(mul_14110, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1312 = None
	        view_2281: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2282: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2283: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1313 = torch.ops.aten._assert_tensor_metadata.default(view_2281, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1313 = None
	        convert_element_type_874: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2281, torch.float32);  view_2281 = None
	        _assert_tensor_metadata_1314 = torch.ops.aten._assert_tensor_metadata.default(view_2283, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1314 = None
	        convert_element_type_875: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2283, torch.float32);  view_2283 = None
	        sub_6669: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_874, convert_element_type_875);  convert_element_type_874 = convert_element_type_875 = None
	        mul_14115: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6669, view_2282);  sub_6669 = view_2282 = None
	        view_2284: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14115, [1280, 1280]);  mul_14115 = None
	        _assert_tensor_metadata_1315 = torch.ops.aten._assert_tensor_metadata.default(view_2284, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1315 = None
	        permute_243: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2284, [1, 0]);  view_2284 = None
	        mul_14118: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2285: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14110, [mul_14118, 1280]);  mul_14110 = mul_14118 = None
	        mm_24: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2285, permute_243);  view_2285 = permute_243 = None
	        view_2286: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_24, [sym_size_int, 1500, 1280]);  mm_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2287: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2286, [sym_size_int, -1, 20, 64]);  view_2286 = None
	        permute_244: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2287, [0, 2, 1, 3]);  view_2287 = None
	        clone_195: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_244, memory_format = torch.contiguous_format);  permute_244 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22064, [2])
	        amax_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22064, [2])
	        full_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_146, full_292);  amin_146 = full_292 = None
	        full_293: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_146, full_293);  amax_146 = full_293 = None
	        sub_6683: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_146, minimum_146);  maximum_146 = None
	        div_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6683, 255.0);  sub_6683 = None
	        clamp_min_438: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_292, 1.1920928955078125e-07);  div_292 = None
	        div_293: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_146, clamp_min_438);  minimum_146 = None
	        round_293: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_293);  div_293 = None
	        sub_6689: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_293);  round_293 = None
	        clamp_min_439: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6689, -128);  sub_6689 = None
	        clamp_max_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_439, 127);  clamp_min_439 = None
	        _assert_tensor_metadata_1316 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_438, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1316 = None
	        _assert_tensor_metadata_1317 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_292, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1317 = None
	        convert_element_type_876: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_292, torch.int8);  clamp_max_292 = None
	        view_2290: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_438, [sym_size_int, 1500, 1])
	        view_2291: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_876, [sym_size_int, 1500, 1])
	        reciprocal_146: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2290);  view_2290 = None
	        mul_14184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_146, 1.0);  reciprocal_146 = None
	        mul_14187: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22064, mul_14184);  add_22064 = mul_14184 = None
	        round_294: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14187);  mul_14187 = None
	        add_22451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_294, view_2291);  round_294 = view_2291 = None
	        clamp_min_440: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22451, -128);  add_22451 = None
	        clamp_max_293: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_440, 127);  clamp_min_440 = None
	        _assert_tensor_metadata_1318 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_293, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1318 = None
	        convert_element_type_877: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_293, torch.int8);  clamp_max_293 = None
	        view_2294: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_438, [sym_size_int, 1500, 1]);  clamp_min_438 = None
	        view_2295: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_876, [sym_size_int, 1500, 1]);  convert_element_type_876 = None
	        _assert_tensor_metadata_1319 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_877, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1319 = None
	        convert_element_type_878: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_877, torch.float32);  convert_element_type_877 = None
	        _assert_tensor_metadata_1320 = torch.ops.aten._assert_tensor_metadata.default(view_2295, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1320 = None
	        convert_element_type_879: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2295, torch.float32);  view_2295 = None
	        sub_6709: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_878, convert_element_type_879);  convert_element_type_878 = convert_element_type_879 = None
	        mul_14209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6709, view_2294);  sub_6709 = view_2294 = None
	        _assert_tensor_metadata_1321 = torch.ops.aten._assert_tensor_metadata.default(mul_14209, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1321 = None
	        view_2297: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2298: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2299: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1322 = torch.ops.aten._assert_tensor_metadata.default(view_2297, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1322 = None
	        convert_element_type_880: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2297, torch.float32);  view_2297 = None
	        _assert_tensor_metadata_1323 = torch.ops.aten._assert_tensor_metadata.default(view_2299, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1323 = None
	        convert_element_type_881: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2299, torch.float32);  view_2299 = None
	        sub_6713: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_880, convert_element_type_881);  convert_element_type_880 = convert_element_type_881 = None
	        mul_14214: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6713, view_2298);  sub_6713 = view_2298 = None
	        view_2300: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14214, [1280, 1280]);  mul_14214 = None
	        _assert_tensor_metadata_1324 = torch.ops.aten._assert_tensor_metadata.default(view_2300, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1324 = None
	        mul_14219: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2301: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14209, [mul_14219, 1280]);  mul_14209 = mul_14219 = None
	        permute_245: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2300, [1, 0]);  view_2300 = None
	        addmm_121: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_24_self_attn_v_proj_bias, view_2301, permute_245);  model_audio_tower_layers_24_self_attn_v_proj_bias = view_2301 = permute_245 = None
	        view_2302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_121, [sym_size_int, 1500, 1280]);  addmm_121 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2303: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2302, [sym_size_int, -1, 20, 64]);  view_2302 = None
	        permute_246: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2303, [0, 2, 1, 3]);  view_2303 = None
	        clone_196: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_246, memory_format = torch.contiguous_format);  permute_246 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_24 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_194, clone_195, clone_196, None, False, scale = 1.0);  clone_194 = clone_195 = clone_196 = None
	        getitem_194: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_24[0];  _scaled_dot_product_efficient_attention_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_247: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_194, [0, 2, 1, 3]);  getitem_194 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2304: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_247, [sym_size_int, 1500, -1]);  permute_247 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2304, [2])
	        amax_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2304, [2])
	        full_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_147, full_294);  amin_147 = full_294 = None
	        full_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_147, full_295);  amax_147 = full_295 = None
	        sub_6731: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_147, minimum_147);  maximum_147 = None
	        div_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6731, 255.0);  sub_6731 = None
	        clamp_min_441: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_294, 1.1920928955078125e-07);  div_294 = None
	        div_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_147, clamp_min_441);  minimum_147 = None
	        round_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_295);  div_295 = None
	        sub_6737: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_295);  round_295 = None
	        clamp_min_442: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6737, -128);  sub_6737 = None
	        clamp_max_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_442, 127);  clamp_min_442 = None
	        _assert_tensor_metadata_1325 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_441, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1325 = None
	        _assert_tensor_metadata_1326 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_294, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1326 = None
	        convert_element_type_882: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_294, torch.int8);  clamp_max_294 = None
	        view_2307: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_441, [sym_size_int, 1500, 1])
	        view_2308: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_882, [sym_size_int, 1500, 1])
	        reciprocal_147: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2307);  view_2307 = None
	        mul_14289: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_147, 1.0);  reciprocal_147 = None
	        mul_14292: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2304, mul_14289);  view_2304 = mul_14289 = None
	        round_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14292);  mul_14292 = None
	        add_22615: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_296, view_2308);  round_296 = view_2308 = None
	        clamp_min_443: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22615, -128);  add_22615 = None
	        clamp_max_295: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_443, 127);  clamp_min_443 = None
	        _assert_tensor_metadata_1327 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_295, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1327 = None
	        convert_element_type_883: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_295, torch.int8);  clamp_max_295 = None
	        view_2311: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_441, [sym_size_int, 1500, 1]);  clamp_min_441 = None
	        view_2312: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_882, [sym_size_int, 1500, 1]);  convert_element_type_882 = None
	        _assert_tensor_metadata_1328 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_883, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1328 = None
	        convert_element_type_884: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_883, torch.float32);  convert_element_type_883 = None
	        _assert_tensor_metadata_1329 = torch.ops.aten._assert_tensor_metadata.default(view_2312, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1329 = None
	        convert_element_type_885: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2312, torch.float32);  view_2312 = None
	        sub_6757: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_884, convert_element_type_885);  convert_element_type_884 = convert_element_type_885 = None
	        mul_14314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6757, view_2311);  sub_6757 = view_2311 = None
	        _assert_tensor_metadata_1330 = torch.ops.aten._assert_tensor_metadata.default(mul_14314, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1330 = None
	        view_2314: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2315: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2316: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1331 = torch.ops.aten._assert_tensor_metadata.default(view_2314, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1331 = None
	        convert_element_type_886: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2314, torch.float32);  view_2314 = None
	        _assert_tensor_metadata_1332 = torch.ops.aten._assert_tensor_metadata.default(view_2316, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1332 = None
	        convert_element_type_887: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2316, torch.float32);  view_2316 = None
	        sub_6761: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_886, convert_element_type_887);  convert_element_type_886 = convert_element_type_887 = None
	        mul_14319: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6761, view_2315);  sub_6761 = view_2315 = None
	        view_2317: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14319, [1280, 1280]);  mul_14319 = None
	        _assert_tensor_metadata_1333 = torch.ops.aten._assert_tensor_metadata.default(view_2317, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1333 = None
	        mul_14324: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2318: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14314, [mul_14324, 1280]);  mul_14314 = mul_14324 = None
	        permute_248: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2317, [1, 0]);  view_2317 = None
	        addmm_122: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_24_self_attn_out_proj_bias, view_2318, permute_248);  model_audio_tower_layers_24_self_attn_out_proj_bias = view_2318 = permute_248 = None
	        view_2319: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_122, [sym_size_int, 1500, 1280]);  addmm_122 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_22678: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_22058, view_2319);  add_22058 = view_2319 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_198: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_22678, memory_format = torch.contiguous_format)
	        var_mean_49 = torch.ops.aten.var_mean.correction(clone_198, [2], correction = 0, keepdim = True)
	        getitem_198: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_49[0]
	        getitem_199: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_49[1];  var_mean_49 = None
	        add_22683: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_198, 1e-05);  getitem_198 = None
	        rsqrt_49: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_22683);  add_22683 = None
	        sub_6767: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_198, getitem_199);  clone_198 = getitem_199 = None
	        mul_14335: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6767, rsqrt_49);  sub_6767 = rsqrt_49 = None
	        mul_14336: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14335, model_audio_tower_layers_24_final_layer_norm_weight);  mul_14335 = model_audio_tower_layers_24_final_layer_norm_weight = None
	        add_22684: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_14336, model_audio_tower_layers_24_final_layer_norm_bias);  mul_14336 = model_audio_tower_layers_24_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22684, [2])
	        amax_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22684, [2])
	        full_296: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_148, full_296);  amin_148 = full_296 = None
	        full_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_148, full_297);  amax_148 = full_297 = None
	        sub_6778: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_148, minimum_148);  maximum_148 = None
	        div_296: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6778, 255.0);  sub_6778 = None
	        clamp_min_444: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_296, 1.1920928955078125e-07);  div_296 = None
	        div_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_148, clamp_min_444);  minimum_148 = None
	        round_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_297);  div_297 = None
	        sub_6784: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_297);  round_297 = None
	        clamp_min_445: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6784, -128);  sub_6784 = None
	        clamp_max_296: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_445, 127);  clamp_min_445 = None
	        _assert_tensor_metadata_1334 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_444, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1334 = None
	        _assert_tensor_metadata_1335 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_296, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1335 = None
	        convert_element_type_888: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_296, torch.int8);  clamp_max_296 = None
	        view_2322: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_444, [sym_size_int, 1500, 1])
	        view_2323: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_888, [sym_size_int, 1500, 1])
	        reciprocal_148: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2322);  view_2322 = None
	        mul_14384: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_148, 1.0);  reciprocal_148 = None
	        mul_14387: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22684, mul_14384);  add_22684 = mul_14384 = None
	        round_298: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14387);  mul_14387 = None
	        add_22771: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_298, view_2323);  round_298 = view_2323 = None
	        clamp_min_446: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22771, -128);  add_22771 = None
	        clamp_max_297: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_446, 127);  clamp_min_446 = None
	        _assert_tensor_metadata_1336 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_297, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1336 = None
	        convert_element_type_889: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_297, torch.int8);  clamp_max_297 = None
	        view_2326: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_444, [sym_size_int, 1500, 1]);  clamp_min_444 = None
	        view_2327: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_888, [sym_size_int, 1500, 1]);  convert_element_type_888 = None
	        _assert_tensor_metadata_1337 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_889, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1337 = None
	        convert_element_type_890: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_889, torch.float32);  convert_element_type_889 = None
	        _assert_tensor_metadata_1338 = torch.ops.aten._assert_tensor_metadata.default(view_2327, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1338 = None
	        convert_element_type_891: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2327, torch.float32);  view_2327 = None
	        sub_6804: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_890, convert_element_type_891);  convert_element_type_890 = convert_element_type_891 = None
	        mul_14409: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6804, view_2326);  sub_6804 = view_2326 = None
	        _assert_tensor_metadata_1339 = torch.ops.aten._assert_tensor_metadata.default(mul_14409, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1339 = None
	        view_2329: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_24_fc1_parametrizations_weight_original0 = None
	        view_2330: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_24_fc1_parametrizations_weight_original1 = None
	        view_2331: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_24_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1340 = torch.ops.aten._assert_tensor_metadata.default(view_2329, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1340 = None
	        convert_element_type_892: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2329, torch.float32);  view_2329 = None
	        _assert_tensor_metadata_1341 = torch.ops.aten._assert_tensor_metadata.default(view_2331, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1341 = None
	        convert_element_type_893: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2331, torch.float32);  view_2331 = None
	        sub_6808: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_892, convert_element_type_893);  convert_element_type_892 = convert_element_type_893 = None
	        mul_14414: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6808, view_2330);  sub_6808 = view_2330 = None
	        view_2332: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14414, [5120, 1280]);  mul_14414 = None
	        _assert_tensor_metadata_1342 = torch.ops.aten._assert_tensor_metadata.default(view_2332, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1342 = None
	        mul_14419: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2333: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14409, [mul_14419, 1280]);  mul_14409 = mul_14419 = None
	        permute_249: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2332, [1, 0]);  view_2332 = None
	        addmm_123: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_24_fc1_bias, view_2333, permute_249);  model_audio_tower_layers_24_fc1_bias = view_2333 = permute_249 = None
	        view_2334: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_123, [sym_size_int, 1500, 5120]);  addmm_123 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_14426: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2334, 0.5)
	        mul_14427: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2334, 0.7071067811865476);  view_2334 = None
	        erf_26: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_14427);  mul_14427 = None
	        add_22830: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_26, 1);  erf_26 = None
	        mul_14428: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14426, add_22830);  mul_14426 = add_22830 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_14428, [2])
	        amax_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_14428, [2])
	        full_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_149, full_298);  amin_149 = full_298 = None
	        full_299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_149, full_299);  amax_149 = full_299 = None
	        sub_6821: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_149, minimum_149);  maximum_149 = None
	        div_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6821, 255.0);  sub_6821 = None
	        clamp_min_447: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_298, 1.1920928955078125e-07);  div_298 = None
	        div_299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_149, clamp_min_447);  minimum_149 = None
	        round_299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_299);  div_299 = None
	        sub_6827: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_299);  round_299 = None
	        clamp_min_448: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6827, -128);  sub_6827 = None
	        clamp_max_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_448, 127);  clamp_min_448 = None
	        _assert_tensor_metadata_1343 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_447, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1343 = None
	        _assert_tensor_metadata_1344 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_298, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1344 = None
	        convert_element_type_894: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_298, torch.int8);  clamp_max_298 = None
	        view_2337: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_447, [sym_size_int, 1500, 1])
	        view_2338: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_894, [sym_size_int, 1500, 1])
	        reciprocal_149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2337);  view_2337 = None
	        mul_14474: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_149, 1.0);  reciprocal_149 = None
	        mul_14477: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14428, mul_14474);  mul_14428 = mul_14474 = None
	        round_300: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_14477);  mul_14477 = None
	        add_22913: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_300, view_2338);  round_300 = view_2338 = None
	        clamp_min_449: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22913, -128);  add_22913 = None
	        clamp_max_299: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_449, 127);  clamp_min_449 = None
	        _assert_tensor_metadata_1345 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_299, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1345 = None
	        convert_element_type_895: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_299, torch.int8);  clamp_max_299 = None
	        view_2341: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_447, [sym_size_int, 1500, 1]);  clamp_min_447 = None
	        view_2342: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_894, [sym_size_int, 1500, 1]);  convert_element_type_894 = None
	        _assert_tensor_metadata_1346 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_895, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1346 = None
	        convert_element_type_896: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_895, torch.float32);  convert_element_type_895 = None
	        _assert_tensor_metadata_1347 = torch.ops.aten._assert_tensor_metadata.default(view_2342, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1347 = None
	        convert_element_type_897: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2342, torch.float32);  view_2342 = None
	        sub_6847: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_896, convert_element_type_897);  convert_element_type_896 = convert_element_type_897 = None
	        mul_14499: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6847, view_2341);  sub_6847 = view_2341 = None
	        _assert_tensor_metadata_1348 = torch.ops.aten._assert_tensor_metadata.default(mul_14499, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1348 = None
	        view_2344: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_24_fc2_parametrizations_weight_original0 = None
	        view_2345: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_24_fc2_parametrizations_weight_original1 = None
	        view_2346: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_24_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1349 = torch.ops.aten._assert_tensor_metadata.default(view_2344, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1349 = None
	        convert_element_type_898: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2344, torch.float32);  view_2344 = None
	        _assert_tensor_metadata_1350 = torch.ops.aten._assert_tensor_metadata.default(view_2346, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1350 = None
	        convert_element_type_899: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2346, torch.float32);  view_2346 = None
	        sub_6851: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_898, convert_element_type_899);  convert_element_type_898 = convert_element_type_899 = None
	        mul_14504: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6851, view_2345);  sub_6851 = view_2345 = None
	        view_2347: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14504, [1280, 5120]);  mul_14504 = None
	        _assert_tensor_metadata_1351 = torch.ops.aten._assert_tensor_metadata.default(view_2347, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1351 = None
	        mul_14509: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2348: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14499, [mul_14509, 5120]);  mul_14499 = mul_14509 = None
	        permute_250: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2347, [1, 0]);  view_2347 = None
	        addmm_124: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_24_fc2_bias, view_2348, permute_250);  model_audio_tower_layers_24_fc2_bias = view_2348 = permute_250 = None
	        view_2349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_124, [sym_size_int, 1500, 1280]);  addmm_124 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_22976: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_22678, view_2349);  add_22678 = view_2349 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_201: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_22976, memory_format = torch.contiguous_format)
	        var_mean_50 = torch.ops.aten.var_mean.correction(clone_201, [2], correction = 0, keepdim = True)
	        getitem_200: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_50[0]
	        getitem_201: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_50[1];  var_mean_50 = None
	        add_22981: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_200, 1e-05);  getitem_200 = None
	        rsqrt_50: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_22981);  add_22981 = None
	        sub_6857: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_201, getitem_201);  clone_201 = getitem_201 = None
	        mul_14520: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6857, rsqrt_50);  sub_6857 = rsqrt_50 = None
	        mul_14521: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14520, model_audio_tower_layers_25_self_attn_layer_norm_weight);  mul_14520 = model_audio_tower_layers_25_self_attn_layer_norm_weight = None
	        add_22982: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_14521, model_audio_tower_layers_25_self_attn_layer_norm_bias);  mul_14521 = model_audio_tower_layers_25_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22982, [2])
	        amax_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22982, [2])
	        full_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_150, full_300);  amin_150 = full_300 = None
	        full_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_150, full_301);  amax_150 = full_301 = None
	        sub_6868: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_150, minimum_150);  maximum_150 = None
	        div_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6868, 255.0);  sub_6868 = None
	        clamp_min_450: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_300, 1.1920928955078125e-07);  div_300 = None
	        div_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_150, clamp_min_450);  minimum_150 = None
	        round_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_301);  div_301 = None
	        sub_6874: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_301);  round_301 = None
	        clamp_min_451: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6874, -128);  sub_6874 = None
	        clamp_max_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_451, 127);  clamp_min_451 = None
	        _assert_tensor_metadata_1352 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_450, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1352 = None
	        _assert_tensor_metadata_1353 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_300, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1353 = None
	        convert_element_type_900: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_300, torch.int8);  clamp_max_300 = None
	        view_2352: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_450, [sym_size_int, 1500, 1])
	        view_2353: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_900, [sym_size_int, 1500, 1])
	        reciprocal_150: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2352);  view_2352 = None
	        mul_14569: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_150, 1.0);  reciprocal_150 = None
	        mul_14572: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22982, mul_14569);  mul_14569 = None
	        round_302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14572);  mul_14572 = None
	        add_23069: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_302, view_2353);  round_302 = view_2353 = None
	        clamp_min_452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23069, -128);  add_23069 = None
	        clamp_max_301: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_452, 127);  clamp_min_452 = None
	        _assert_tensor_metadata_1354 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_301, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1354 = None
	        convert_element_type_901: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_301, torch.int8);  clamp_max_301 = None
	        view_2356: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_450, [sym_size_int, 1500, 1]);  clamp_min_450 = None
	        view_2357: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_900, [sym_size_int, 1500, 1]);  convert_element_type_900 = None
	        _assert_tensor_metadata_1355 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_901, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1355 = None
	        convert_element_type_902: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_901, torch.float32);  convert_element_type_901 = None
	        _assert_tensor_metadata_1356 = torch.ops.aten._assert_tensor_metadata.default(view_2357, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1356 = None
	        convert_element_type_903: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2357, torch.float32);  view_2357 = None
	        sub_6894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_902, convert_element_type_903);  convert_element_type_902 = convert_element_type_903 = None
	        mul_14594: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6894, view_2356);  sub_6894 = view_2356 = None
	        _assert_tensor_metadata_1357 = torch.ops.aten._assert_tensor_metadata.default(mul_14594, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1357 = None
	        view_2359: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2360: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2361: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1358 = torch.ops.aten._assert_tensor_metadata.default(view_2359, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1358 = None
	        convert_element_type_904: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2359, torch.float32);  view_2359 = None
	        _assert_tensor_metadata_1359 = torch.ops.aten._assert_tensor_metadata.default(view_2361, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1359 = None
	        convert_element_type_905: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2361, torch.float32);  view_2361 = None
	        sub_6898: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_904, convert_element_type_905);  convert_element_type_904 = convert_element_type_905 = None
	        mul_14599: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6898, view_2360);  sub_6898 = view_2360 = None
	        view_2362: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14599, [1280, 1280]);  mul_14599 = None
	        _assert_tensor_metadata_1360 = torch.ops.aten._assert_tensor_metadata.default(view_2362, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1360 = None
	        mul_14604: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2363: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14594, [mul_14604, 1280]);  mul_14594 = mul_14604 = None
	        permute_251: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2362, [1, 0]);  view_2362 = None
	        addmm_125: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_25_self_attn_q_proj_bias, view_2363, permute_251);  model_audio_tower_layers_25_self_attn_q_proj_bias = view_2363 = permute_251 = None
	        view_2364: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_125, [sym_size_int, 1500, 1280]);  addmm_125 = None
	        mul_14611: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2364, 0.125);  view_2364 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2365: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14611, [sym_size_int, 1500, 20, 64]);  mul_14611 = None
	        permute_252: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2365, [0, 2, 1, 3]);  view_2365 = None
	        clone_202: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_252, memory_format = torch.contiguous_format);  permute_252 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22982, [2])
	        amax_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22982, [2])
	        full_302: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_151, full_302);  amin_151 = full_302 = None
	        full_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_151, full_303);  amax_151 = full_303 = None
	        sub_6913: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_151, minimum_151);  maximum_151 = None
	        div_302: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6913, 255.0);  sub_6913 = None
	        clamp_min_453: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_302, 1.1920928955078125e-07);  div_302 = None
	        div_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_151, clamp_min_453);  minimum_151 = None
	        round_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_303);  div_303 = None
	        sub_6919: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_303);  round_303 = None
	        clamp_min_454: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6919, -128);  sub_6919 = None
	        clamp_max_302: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_454, 127);  clamp_min_454 = None
	        _assert_tensor_metadata_1361 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_453, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1361 = None
	        _assert_tensor_metadata_1362 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_302, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1362 = None
	        convert_element_type_906: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_302, torch.int8);  clamp_max_302 = None
	        view_2368: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_453, [sym_size_int, 1500, 1])
	        view_2369: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_906, [sym_size_int, 1500, 1])
	        reciprocal_151: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2368);  view_2368 = None
	        mul_14665: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_151, 1.0);  reciprocal_151 = None
	        mul_14668: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22982, mul_14665);  mul_14665 = None
	        round_304: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14668);  mul_14668 = None
	        add_23221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_304, view_2369);  round_304 = view_2369 = None
	        clamp_min_455: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23221, -128);  add_23221 = None
	        clamp_max_303: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_455, 127);  clamp_min_455 = None
	        _assert_tensor_metadata_1363 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_303, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1363 = None
	        convert_element_type_907: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_303, torch.int8);  clamp_max_303 = None
	        view_2372: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_453, [sym_size_int, 1500, 1]);  clamp_min_453 = None
	        view_2373: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_906, [sym_size_int, 1500, 1]);  convert_element_type_906 = None
	        _assert_tensor_metadata_1364 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_907, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1364 = None
	        convert_element_type_908: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_907, torch.float32);  convert_element_type_907 = None
	        _assert_tensor_metadata_1365 = torch.ops.aten._assert_tensor_metadata.default(view_2373, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1365 = None
	        convert_element_type_909: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2373, torch.float32);  view_2373 = None
	        sub_6939: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_908, convert_element_type_909);  convert_element_type_908 = convert_element_type_909 = None
	        mul_14690: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6939, view_2372);  sub_6939 = view_2372 = None
	        _assert_tensor_metadata_1366 = torch.ops.aten._assert_tensor_metadata.default(mul_14690, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1366 = None
	        view_2375: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2376: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2377: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1367 = torch.ops.aten._assert_tensor_metadata.default(view_2375, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1367 = None
	        convert_element_type_910: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2375, torch.float32);  view_2375 = None
	        _assert_tensor_metadata_1368 = torch.ops.aten._assert_tensor_metadata.default(view_2377, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1368 = None
	        convert_element_type_911: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2377, torch.float32);  view_2377 = None
	        sub_6943: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_910, convert_element_type_911);  convert_element_type_910 = convert_element_type_911 = None
	        mul_14695: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6943, view_2376);  sub_6943 = view_2376 = None
	        view_2378: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14695, [1280, 1280]);  mul_14695 = None
	        _assert_tensor_metadata_1369 = torch.ops.aten._assert_tensor_metadata.default(view_2378, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1369 = None
	        permute_253: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2378, [1, 0]);  view_2378 = None
	        mul_14698: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2379: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14690, [mul_14698, 1280]);  mul_14690 = mul_14698 = None
	        mm_25: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2379, permute_253);  view_2379 = permute_253 = None
	        view_2380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_25, [sym_size_int, 1500, 1280]);  mm_25 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2381: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2380, [sym_size_int, -1, 20, 64]);  view_2380 = None
	        permute_254: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2381, [0, 2, 1, 3]);  view_2381 = None
	        clone_203: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_254, memory_format = torch.contiguous_format);  permute_254 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22982, [2])
	        amax_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22982, [2])
	        full_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_152, full_304);  amin_152 = full_304 = None
	        full_305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_152, full_305);  amax_152 = full_305 = None
	        sub_6957: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_152, minimum_152);  maximum_152 = None
	        div_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6957, 255.0);  sub_6957 = None
	        clamp_min_456: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_304, 1.1920928955078125e-07);  div_304 = None
	        div_305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_152, clamp_min_456);  minimum_152 = None
	        round_305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_305);  div_305 = None
	        sub_6963: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_305);  round_305 = None
	        clamp_min_457: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6963, -128);  sub_6963 = None
	        clamp_max_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_457, 127);  clamp_min_457 = None
	        _assert_tensor_metadata_1370 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_456, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1370 = None
	        _assert_tensor_metadata_1371 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_304, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1371 = None
	        convert_element_type_912: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_304, torch.int8);  clamp_max_304 = None
	        view_2384: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_456, [sym_size_int, 1500, 1])
	        view_2385: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_912, [sym_size_int, 1500, 1])
	        reciprocal_152: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2384);  view_2384 = None
	        mul_14764: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_152, 1.0);  reciprocal_152 = None
	        mul_14767: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22982, mul_14764);  add_22982 = mul_14764 = None
	        round_306: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14767);  mul_14767 = None
	        add_23369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_306, view_2385);  round_306 = view_2385 = None
	        clamp_min_458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23369, -128);  add_23369 = None
	        clamp_max_305: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_458, 127);  clamp_min_458 = None
	        _assert_tensor_metadata_1372 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_305, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1372 = None
	        convert_element_type_913: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_305, torch.int8);  clamp_max_305 = None
	        view_2388: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_456, [sym_size_int, 1500, 1]);  clamp_min_456 = None
	        view_2389: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_912, [sym_size_int, 1500, 1]);  convert_element_type_912 = None
	        _assert_tensor_metadata_1373 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_913, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1373 = None
	        convert_element_type_914: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_913, torch.float32);  convert_element_type_913 = None
	        _assert_tensor_metadata_1374 = torch.ops.aten._assert_tensor_metadata.default(view_2389, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1374 = None
	        convert_element_type_915: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2389, torch.float32);  view_2389 = None
	        sub_6983: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_914, convert_element_type_915);  convert_element_type_914 = convert_element_type_915 = None
	        mul_14789: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6983, view_2388);  sub_6983 = view_2388 = None
	        _assert_tensor_metadata_1375 = torch.ops.aten._assert_tensor_metadata.default(mul_14789, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1375 = None
	        view_2391: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2392: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2393: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1376 = torch.ops.aten._assert_tensor_metadata.default(view_2391, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1376 = None
	        convert_element_type_916: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2391, torch.float32);  view_2391 = None
	        _assert_tensor_metadata_1377 = torch.ops.aten._assert_tensor_metadata.default(view_2393, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1377 = None
	        convert_element_type_917: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2393, torch.float32);  view_2393 = None
	        sub_6987: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_916, convert_element_type_917);  convert_element_type_916 = convert_element_type_917 = None
	        mul_14794: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6987, view_2392);  sub_6987 = view_2392 = None
	        view_2394: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14794, [1280, 1280]);  mul_14794 = None
	        _assert_tensor_metadata_1378 = torch.ops.aten._assert_tensor_metadata.default(view_2394, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1378 = None
	        mul_14799: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2395: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14789, [mul_14799, 1280]);  mul_14789 = mul_14799 = None
	        permute_255: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2394, [1, 0]);  view_2394 = None
	        addmm_126: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_25_self_attn_v_proj_bias, view_2395, permute_255);  model_audio_tower_layers_25_self_attn_v_proj_bias = view_2395 = permute_255 = None
	        view_2396: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_126, [sym_size_int, 1500, 1280]);  addmm_126 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2397: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2396, [sym_size_int, -1, 20, 64]);  view_2396 = None
	        permute_256: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2397, [0, 2, 1, 3]);  view_2397 = None
	        clone_204: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_256, memory_format = torch.contiguous_format);  permute_256 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_25 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_202, clone_203, clone_204, None, False, scale = 1.0);  clone_202 = clone_203 = clone_204 = None
	        getitem_202: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_25[0];  _scaled_dot_product_efficient_attention_25 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_257: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_202, [0, 2, 1, 3]);  getitem_202 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2398: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_257, [sym_size_int, 1500, -1]);  permute_257 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2398, [2])
	        amax_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2398, [2])
	        full_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_153, full_306);  amin_153 = full_306 = None
	        full_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_153, full_307);  amax_153 = full_307 = None
	        sub_7005: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_153, minimum_153);  maximum_153 = None
	        div_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7005, 255.0);  sub_7005 = None
	        clamp_min_459: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_306, 1.1920928955078125e-07);  div_306 = None
	        div_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_153, clamp_min_459);  minimum_153 = None
	        round_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_307);  div_307 = None
	        sub_7011: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_307);  round_307 = None
	        clamp_min_460: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7011, -128);  sub_7011 = None
	        clamp_max_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_460, 127);  clamp_min_460 = None
	        _assert_tensor_metadata_1379 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_459, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1379 = None
	        _assert_tensor_metadata_1380 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_306, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1380 = None
	        convert_element_type_918: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_306, torch.int8);  clamp_max_306 = None
	        view_2401: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_459, [sym_size_int, 1500, 1])
	        view_2402: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_918, [sym_size_int, 1500, 1])
	        reciprocal_153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2401);  view_2401 = None
	        mul_14869: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_153, 1.0);  reciprocal_153 = None
	        mul_14872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2398, mul_14869);  view_2398 = mul_14869 = None
	        round_308: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14872);  mul_14872 = None
	        add_23533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_308, view_2402);  round_308 = view_2402 = None
	        clamp_min_461: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23533, -128);  add_23533 = None
	        clamp_max_307: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_461, 127);  clamp_min_461 = None
	        _assert_tensor_metadata_1381 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_307, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1381 = None
	        convert_element_type_919: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_307, torch.int8);  clamp_max_307 = None
	        view_2405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_459, [sym_size_int, 1500, 1]);  clamp_min_459 = None
	        view_2406: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_918, [sym_size_int, 1500, 1]);  convert_element_type_918 = None
	        _assert_tensor_metadata_1382 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_919, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1382 = None
	        convert_element_type_920: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_919, torch.float32);  convert_element_type_919 = None
	        _assert_tensor_metadata_1383 = torch.ops.aten._assert_tensor_metadata.default(view_2406, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1383 = None
	        convert_element_type_921: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2406, torch.float32);  view_2406 = None
	        sub_7031: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_920, convert_element_type_921);  convert_element_type_920 = convert_element_type_921 = None
	        mul_14894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7031, view_2405);  sub_7031 = view_2405 = None
	        _assert_tensor_metadata_1384 = torch.ops.aten._assert_tensor_metadata.default(mul_14894, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1384 = None
	        view_2408: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2409: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2410: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1385 = torch.ops.aten._assert_tensor_metadata.default(view_2408, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1385 = None
	        convert_element_type_922: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2408, torch.float32);  view_2408 = None
	        _assert_tensor_metadata_1386 = torch.ops.aten._assert_tensor_metadata.default(view_2410, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1386 = None
	        convert_element_type_923: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2410, torch.float32);  view_2410 = None
	        sub_7035: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_922, convert_element_type_923);  convert_element_type_922 = convert_element_type_923 = None
	        mul_14899: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7035, view_2409);  sub_7035 = view_2409 = None
	        view_2411: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14899, [1280, 1280]);  mul_14899 = None
	        _assert_tensor_metadata_1387 = torch.ops.aten._assert_tensor_metadata.default(view_2411, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1387 = None
	        mul_14904: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2412: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14894, [mul_14904, 1280]);  mul_14894 = mul_14904 = None
	        permute_258: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2411, [1, 0]);  view_2411 = None
	        addmm_127: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_25_self_attn_out_proj_bias, view_2412, permute_258);  model_audio_tower_layers_25_self_attn_out_proj_bias = view_2412 = permute_258 = None
	        view_2413: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_127, [sym_size_int, 1500, 1280]);  addmm_127 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_23596: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_22976, view_2413);  add_22976 = view_2413 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_23596, memory_format = torch.contiguous_format)
	        var_mean_51 = torch.ops.aten.var_mean.correction(clone_206, [2], correction = 0, keepdim = True)
	        getitem_206: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_51[0]
	        getitem_207: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_51[1];  var_mean_51 = None
	        add_23601: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_206, 1e-05);  getitem_206 = None
	        rsqrt_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_23601);  add_23601 = None
	        sub_7041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_206, getitem_207);  clone_206 = getitem_207 = None
	        mul_14915: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7041, rsqrt_51);  sub_7041 = rsqrt_51 = None
	        mul_14916: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14915, model_audio_tower_layers_25_final_layer_norm_weight);  mul_14915 = model_audio_tower_layers_25_final_layer_norm_weight = None
	        add_23602: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_14916, model_audio_tower_layers_25_final_layer_norm_bias);  mul_14916 = model_audio_tower_layers_25_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_23602, [2])
	        amax_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_23602, [2])
	        full_308: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_154, full_308);  amin_154 = full_308 = None
	        full_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_154, full_309);  amax_154 = full_309 = None
	        sub_7052: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_154, minimum_154);  maximum_154 = None
	        div_308: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7052, 255.0);  sub_7052 = None
	        clamp_min_462: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_308, 1.1920928955078125e-07);  div_308 = None
	        div_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_154, clamp_min_462);  minimum_154 = None
	        round_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_309);  div_309 = None
	        sub_7058: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_309);  round_309 = None
	        clamp_min_463: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7058, -128);  sub_7058 = None
	        clamp_max_308: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_463, 127);  clamp_min_463 = None
	        _assert_tensor_metadata_1388 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_462, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1388 = None
	        _assert_tensor_metadata_1389 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_308, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1389 = None
	        convert_element_type_924: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_308, torch.int8);  clamp_max_308 = None
	        view_2416: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_462, [sym_size_int, 1500, 1])
	        view_2417: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_924, [sym_size_int, 1500, 1])
	        reciprocal_154: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2416);  view_2416 = None
	        mul_14964: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_154, 1.0);  reciprocal_154 = None
	        mul_14967: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_23602, mul_14964);  add_23602 = mul_14964 = None
	        round_310: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14967);  mul_14967 = None
	        add_23689: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_310, view_2417);  round_310 = view_2417 = None
	        clamp_min_464: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23689, -128);  add_23689 = None
	        clamp_max_309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_464, 127);  clamp_min_464 = None
	        _assert_tensor_metadata_1390 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_309, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1390 = None
	        convert_element_type_925: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_309, torch.int8);  clamp_max_309 = None
	        view_2420: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_462, [sym_size_int, 1500, 1]);  clamp_min_462 = None
	        view_2421: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_924, [sym_size_int, 1500, 1]);  convert_element_type_924 = None
	        _assert_tensor_metadata_1391 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_925, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1391 = None
	        convert_element_type_926: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_925, torch.float32);  convert_element_type_925 = None
	        _assert_tensor_metadata_1392 = torch.ops.aten._assert_tensor_metadata.default(view_2421, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1392 = None
	        convert_element_type_927: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2421, torch.float32);  view_2421 = None
	        sub_7078: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_926, convert_element_type_927);  convert_element_type_926 = convert_element_type_927 = None
	        mul_14989: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7078, view_2420);  sub_7078 = view_2420 = None
	        _assert_tensor_metadata_1393 = torch.ops.aten._assert_tensor_metadata.default(mul_14989, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1393 = None
	        view_2423: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_25_fc1_parametrizations_weight_original0 = None
	        view_2424: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_25_fc1_parametrizations_weight_original1 = None
	        view_2425: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_25_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1394 = torch.ops.aten._assert_tensor_metadata.default(view_2423, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1394 = None
	        convert_element_type_928: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2423, torch.float32);  view_2423 = None
	        _assert_tensor_metadata_1395 = torch.ops.aten._assert_tensor_metadata.default(view_2425, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1395 = None
	        convert_element_type_929: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2425, torch.float32);  view_2425 = None
	        sub_7082: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_928, convert_element_type_929);  convert_element_type_928 = convert_element_type_929 = None
	        mul_14994: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7082, view_2424);  sub_7082 = view_2424 = None
	        view_2426: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14994, [5120, 1280]);  mul_14994 = None
	        _assert_tensor_metadata_1396 = torch.ops.aten._assert_tensor_metadata.default(view_2426, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1396 = None
	        mul_14999: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2427: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14989, [mul_14999, 1280]);  mul_14989 = mul_14999 = None
	        permute_259: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2426, [1, 0]);  view_2426 = None
	        addmm_128: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_25_fc1_bias, view_2427, permute_259);  model_audio_tower_layers_25_fc1_bias = view_2427 = permute_259 = None
	        view_2428: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_128, [sym_size_int, 1500, 5120]);  addmm_128 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_15006: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2428, 0.5)
	        mul_15007: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2428, 0.7071067811865476);  view_2428 = None
	        erf_27: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_15007);  mul_15007 = None
	        add_23748: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_27, 1);  erf_27 = None
	        mul_15008: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15006, add_23748);  mul_15006 = add_23748 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_15008, [2])
	        amax_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_15008, [2])
	        full_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_155, full_310);  amin_155 = full_310 = None
	        full_311: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_155, full_311);  amax_155 = full_311 = None
	        sub_7095: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_155, minimum_155);  maximum_155 = None
	        div_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7095, 255.0);  sub_7095 = None
	        clamp_min_465: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_310, 1.1920928955078125e-07);  div_310 = None
	        div_311: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_155, clamp_min_465);  minimum_155 = None
	        round_311: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_311);  div_311 = None
	        sub_7101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_311);  round_311 = None
	        clamp_min_466: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7101, -128);  sub_7101 = None
	        clamp_max_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_466, 127);  clamp_min_466 = None
	        _assert_tensor_metadata_1397 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_465, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1397 = None
	        _assert_tensor_metadata_1398 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_310, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1398 = None
	        convert_element_type_930: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_310, torch.int8);  clamp_max_310 = None
	        view_2431: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_465, [sym_size_int, 1500, 1])
	        view_2432: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_930, [sym_size_int, 1500, 1])
	        reciprocal_155: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2431);  view_2431 = None
	        mul_15054: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_155, 1.0);  reciprocal_155 = None
	        mul_15057: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15008, mul_15054);  mul_15008 = mul_15054 = None
	        round_312: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_15057);  mul_15057 = None
	        add_23831: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_312, view_2432);  round_312 = view_2432 = None
	        clamp_min_467: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23831, -128);  add_23831 = None
	        clamp_max_311: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_467, 127);  clamp_min_467 = None
	        _assert_tensor_metadata_1399 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_311, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1399 = None
	        convert_element_type_931: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_311, torch.int8);  clamp_max_311 = None
	        view_2435: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_465, [sym_size_int, 1500, 1]);  clamp_min_465 = None
	        view_2436: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_930, [sym_size_int, 1500, 1]);  convert_element_type_930 = None
	        _assert_tensor_metadata_1400 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_931, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1400 = None
	        convert_element_type_932: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_931, torch.float32);  convert_element_type_931 = None
	        _assert_tensor_metadata_1401 = torch.ops.aten._assert_tensor_metadata.default(view_2436, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1401 = None
	        convert_element_type_933: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2436, torch.float32);  view_2436 = None
	        sub_7121: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_932, convert_element_type_933);  convert_element_type_932 = convert_element_type_933 = None
	        mul_15079: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7121, view_2435);  sub_7121 = view_2435 = None
	        _assert_tensor_metadata_1402 = torch.ops.aten._assert_tensor_metadata.default(mul_15079, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1402 = None
	        view_2438: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_25_fc2_parametrizations_weight_original0 = None
	        view_2439: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_25_fc2_parametrizations_weight_original1 = None
	        view_2440: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_25_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1403 = torch.ops.aten._assert_tensor_metadata.default(view_2438, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1403 = None
	        convert_element_type_934: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2438, torch.float32);  view_2438 = None
	        _assert_tensor_metadata_1404 = torch.ops.aten._assert_tensor_metadata.default(view_2440, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1404 = None
	        convert_element_type_935: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2440, torch.float32);  view_2440 = None
	        sub_7125: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_934, convert_element_type_935);  convert_element_type_934 = convert_element_type_935 = None
	        mul_15084: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7125, view_2439);  sub_7125 = view_2439 = None
	        view_2441: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15084, [1280, 5120]);  mul_15084 = None
	        _assert_tensor_metadata_1405 = torch.ops.aten._assert_tensor_metadata.default(view_2441, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1405 = None
	        mul_15089: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2442: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15079, [mul_15089, 5120]);  mul_15079 = mul_15089 = None
	        permute_260: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2441, [1, 0]);  view_2441 = None
	        addmm_129: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_25_fc2_bias, view_2442, permute_260);  model_audio_tower_layers_25_fc2_bias = view_2442 = permute_260 = None
	        view_2443: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_129, [sym_size_int, 1500, 1280]);  addmm_129 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_23894: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_23596, view_2443);  add_23596 = view_2443 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_23894, memory_format = torch.contiguous_format)
	        var_mean_52 = torch.ops.aten.var_mean.correction(clone_209, [2], correction = 0, keepdim = True)
	        getitem_208: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_52[0]
	        getitem_209: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_52[1];  var_mean_52 = None
	        add_23899: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_208, 1e-05);  getitem_208 = None
	        rsqrt_52: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_23899);  add_23899 = None
	        sub_7131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_209, getitem_209);  clone_209 = getitem_209 = None
	        mul_15100: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7131, rsqrt_52);  sub_7131 = rsqrt_52 = None
	        mul_15101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15100, model_audio_tower_layers_26_self_attn_layer_norm_weight);  mul_15100 = model_audio_tower_layers_26_self_attn_layer_norm_weight = None
	        add_23900: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_15101, model_audio_tower_layers_26_self_attn_layer_norm_bias);  mul_15101 = model_audio_tower_layers_26_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_23900, [2])
	        amax_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_23900, [2])
	        full_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_156, full_312);  amin_156 = full_312 = None
	        full_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_156, full_313);  amax_156 = full_313 = None
	        sub_7142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_156, minimum_156);  maximum_156 = None
	        div_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7142, 255.0);  sub_7142 = None
	        clamp_min_468: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_312, 1.1920928955078125e-07);  div_312 = None
	        div_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_156, clamp_min_468);  minimum_156 = None
	        round_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_313);  div_313 = None
	        sub_7148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_313);  round_313 = None
	        clamp_min_469: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7148, -128);  sub_7148 = None
	        clamp_max_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_469, 127);  clamp_min_469 = None
	        _assert_tensor_metadata_1406 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_468, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1406 = None
	        _assert_tensor_metadata_1407 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_312, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1407 = None
	        convert_element_type_936: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_312, torch.int8);  clamp_max_312 = None
	        view_2446: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_468, [sym_size_int, 1500, 1])
	        view_2447: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_936, [sym_size_int, 1500, 1])
	        reciprocal_156: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2446);  view_2446 = None
	        mul_15149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_156, 1.0);  reciprocal_156 = None
	        mul_15152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_23900, mul_15149);  mul_15149 = None
	        round_314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15152);  mul_15152 = None
	        add_23987: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_314, view_2447);  round_314 = view_2447 = None
	        clamp_min_470: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23987, -128);  add_23987 = None
	        clamp_max_313: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_470, 127);  clamp_min_470 = None
	        _assert_tensor_metadata_1408 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_313, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1408 = None
	        convert_element_type_937: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_313, torch.int8);  clamp_max_313 = None
	        view_2450: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_468, [sym_size_int, 1500, 1]);  clamp_min_468 = None
	        view_2451: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_936, [sym_size_int, 1500, 1]);  convert_element_type_936 = None
	        _assert_tensor_metadata_1409 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_937, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1409 = None
	        convert_element_type_938: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_937, torch.float32);  convert_element_type_937 = None
	        _assert_tensor_metadata_1410 = torch.ops.aten._assert_tensor_metadata.default(view_2451, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1410 = None
	        convert_element_type_939: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2451, torch.float32);  view_2451 = None
	        sub_7168: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_938, convert_element_type_939);  convert_element_type_938 = convert_element_type_939 = None
	        mul_15174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7168, view_2450);  sub_7168 = view_2450 = None
	        _assert_tensor_metadata_1411 = torch.ops.aten._assert_tensor_metadata.default(mul_15174, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1411 = None
	        view_2453: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2454: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2455: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1412 = torch.ops.aten._assert_tensor_metadata.default(view_2453, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1412 = None
	        convert_element_type_940: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2453, torch.float32);  view_2453 = None
	        _assert_tensor_metadata_1413 = torch.ops.aten._assert_tensor_metadata.default(view_2455, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1413 = None
	        convert_element_type_941: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2455, torch.float32);  view_2455 = None
	        sub_7172: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_940, convert_element_type_941);  convert_element_type_940 = convert_element_type_941 = None
	        mul_15179: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7172, view_2454);  sub_7172 = view_2454 = None
	        view_2456: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15179, [1280, 1280]);  mul_15179 = None
	        _assert_tensor_metadata_1414 = torch.ops.aten._assert_tensor_metadata.default(view_2456, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1414 = None
	        mul_15184: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2457: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15174, [mul_15184, 1280]);  mul_15174 = mul_15184 = None
	        permute_261: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2456, [1, 0]);  view_2456 = None
	        addmm_130: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_26_self_attn_q_proj_bias, view_2457, permute_261);  model_audio_tower_layers_26_self_attn_q_proj_bias = view_2457 = permute_261 = None
	        view_2458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_130, [sym_size_int, 1500, 1280]);  addmm_130 = None
	        mul_15191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2458, 0.125);  view_2458 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2459: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15191, [sym_size_int, 1500, 20, 64]);  mul_15191 = None
	        permute_262: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2459, [0, 2, 1, 3]);  view_2459 = None
	        clone_210: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_262, memory_format = torch.contiguous_format);  permute_262 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_23900, [2])
	        amax_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_23900, [2])
	        full_314: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_157, full_314);  amin_157 = full_314 = None
	        full_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_157, full_315);  amax_157 = full_315 = None
	        sub_7187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_157, minimum_157);  maximum_157 = None
	        div_314: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7187, 255.0);  sub_7187 = None
	        clamp_min_471: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_314, 1.1920928955078125e-07);  div_314 = None
	        div_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_157, clamp_min_471);  minimum_157 = None
	        round_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_315);  div_315 = None
	        sub_7193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_315);  round_315 = None
	        clamp_min_472: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7193, -128);  sub_7193 = None
	        clamp_max_314: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_472, 127);  clamp_min_472 = None
	        _assert_tensor_metadata_1415 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_471, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1415 = None
	        _assert_tensor_metadata_1416 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_314, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1416 = None
	        convert_element_type_942: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_314, torch.int8);  clamp_max_314 = None
	        view_2462: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_471, [sym_size_int, 1500, 1])
	        view_2463: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_942, [sym_size_int, 1500, 1])
	        reciprocal_157: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2462);  view_2462 = None
	        mul_15245: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_157, 1.0);  reciprocal_157 = None
	        mul_15248: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_23900, mul_15245);  mul_15245 = None
	        round_316: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15248);  mul_15248 = None
	        add_24139: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_316, view_2463);  round_316 = view_2463 = None
	        clamp_min_473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24139, -128);  add_24139 = None
	        clamp_max_315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_473, 127);  clamp_min_473 = None
	        _assert_tensor_metadata_1417 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_315, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1417 = None
	        convert_element_type_943: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_315, torch.int8);  clamp_max_315 = None
	        view_2466: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_471, [sym_size_int, 1500, 1]);  clamp_min_471 = None
	        view_2467: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_942, [sym_size_int, 1500, 1]);  convert_element_type_942 = None
	        _assert_tensor_metadata_1418 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_943, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1418 = None
	        convert_element_type_944: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_943, torch.float32);  convert_element_type_943 = None
	        _assert_tensor_metadata_1419 = torch.ops.aten._assert_tensor_metadata.default(view_2467, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1419 = None
	        convert_element_type_945: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2467, torch.float32);  view_2467 = None
	        sub_7213: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_944, convert_element_type_945);  convert_element_type_944 = convert_element_type_945 = None
	        mul_15270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7213, view_2466);  sub_7213 = view_2466 = None
	        _assert_tensor_metadata_1420 = torch.ops.aten._assert_tensor_metadata.default(mul_15270, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1420 = None
	        view_2469: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2470: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2471: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1421 = torch.ops.aten._assert_tensor_metadata.default(view_2469, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1421 = None
	        convert_element_type_946: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2469, torch.float32);  view_2469 = None
	        _assert_tensor_metadata_1422 = torch.ops.aten._assert_tensor_metadata.default(view_2471, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1422 = None
	        convert_element_type_947: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2471, torch.float32);  view_2471 = None
	        sub_7217: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_946, convert_element_type_947);  convert_element_type_946 = convert_element_type_947 = None
	        mul_15275: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7217, view_2470);  sub_7217 = view_2470 = None
	        view_2472: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15275, [1280, 1280]);  mul_15275 = None
	        _assert_tensor_metadata_1423 = torch.ops.aten._assert_tensor_metadata.default(view_2472, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1423 = None
	        permute_263: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2472, [1, 0]);  view_2472 = None
	        mul_15278: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2473: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15270, [mul_15278, 1280]);  mul_15270 = mul_15278 = None
	        mm_26: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2473, permute_263);  view_2473 = permute_263 = None
	        view_2474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_26, [sym_size_int, 1500, 1280]);  mm_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2475: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2474, [sym_size_int, -1, 20, 64]);  view_2474 = None
	        permute_264: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2475, [0, 2, 1, 3]);  view_2475 = None
	        clone_211: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_264, memory_format = torch.contiguous_format);  permute_264 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_23900, [2])
	        amax_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_23900, [2])
	        full_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_158, full_316);  amin_158 = full_316 = None
	        full_317: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_158, full_317);  amax_158 = full_317 = None
	        sub_7231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_158, minimum_158);  maximum_158 = None
	        div_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7231, 255.0);  sub_7231 = None
	        clamp_min_474: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_316, 1.1920928955078125e-07);  div_316 = None
	        div_317: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_158, clamp_min_474);  minimum_158 = None
	        round_317: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_317);  div_317 = None
	        sub_7237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_317);  round_317 = None
	        clamp_min_475: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7237, -128);  sub_7237 = None
	        clamp_max_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_475, 127);  clamp_min_475 = None
	        _assert_tensor_metadata_1424 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_474, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1424 = None
	        _assert_tensor_metadata_1425 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_316, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1425 = None
	        convert_element_type_948: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_316, torch.int8);  clamp_max_316 = None
	        view_2478: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_474, [sym_size_int, 1500, 1])
	        view_2479: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_948, [sym_size_int, 1500, 1])
	        reciprocal_158: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2478);  view_2478 = None
	        mul_15344: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_158, 1.0);  reciprocal_158 = None
	        mul_15347: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_23900, mul_15344);  add_23900 = mul_15344 = None
	        round_318: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15347);  mul_15347 = None
	        add_24287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_318, view_2479);  round_318 = view_2479 = None
	        clamp_min_476: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24287, -128);  add_24287 = None
	        clamp_max_317: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_476, 127);  clamp_min_476 = None
	        _assert_tensor_metadata_1426 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_317, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1426 = None
	        convert_element_type_949: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_317, torch.int8);  clamp_max_317 = None
	        view_2482: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_474, [sym_size_int, 1500, 1]);  clamp_min_474 = None
	        view_2483: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_948, [sym_size_int, 1500, 1]);  convert_element_type_948 = None
	        _assert_tensor_metadata_1427 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_949, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1427 = None
	        convert_element_type_950: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_949, torch.float32);  convert_element_type_949 = None
	        _assert_tensor_metadata_1428 = torch.ops.aten._assert_tensor_metadata.default(view_2483, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1428 = None
	        convert_element_type_951: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2483, torch.float32);  view_2483 = None
	        sub_7257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_950, convert_element_type_951);  convert_element_type_950 = convert_element_type_951 = None
	        mul_15369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7257, view_2482);  sub_7257 = view_2482 = None
	        _assert_tensor_metadata_1429 = torch.ops.aten._assert_tensor_metadata.default(mul_15369, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1429 = None
	        view_2485: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2486: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2487: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1430 = torch.ops.aten._assert_tensor_metadata.default(view_2485, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1430 = None
	        convert_element_type_952: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2485, torch.float32);  view_2485 = None
	        _assert_tensor_metadata_1431 = torch.ops.aten._assert_tensor_metadata.default(view_2487, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1431 = None
	        convert_element_type_953: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2487, torch.float32);  view_2487 = None
	        sub_7261: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_952, convert_element_type_953);  convert_element_type_952 = convert_element_type_953 = None
	        mul_15374: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7261, view_2486);  sub_7261 = view_2486 = None
	        view_2488: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15374, [1280, 1280]);  mul_15374 = None
	        _assert_tensor_metadata_1432 = torch.ops.aten._assert_tensor_metadata.default(view_2488, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1432 = None
	        mul_15379: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2489: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15369, [mul_15379, 1280]);  mul_15369 = mul_15379 = None
	        permute_265: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2488, [1, 0]);  view_2488 = None
	        addmm_131: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_26_self_attn_v_proj_bias, view_2489, permute_265);  model_audio_tower_layers_26_self_attn_v_proj_bias = view_2489 = permute_265 = None
	        view_2490: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_131, [sym_size_int, 1500, 1280]);  addmm_131 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2491: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2490, [sym_size_int, -1, 20, 64]);  view_2490 = None
	        permute_266: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2491, [0, 2, 1, 3]);  view_2491 = None
	        clone_212: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_266, memory_format = torch.contiguous_format);  permute_266 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_26 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_210, clone_211, clone_212, None, False, scale = 1.0);  clone_210 = clone_211 = clone_212 = None
	        getitem_210: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_26[0];  _scaled_dot_product_efficient_attention_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_267: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_210, [0, 2, 1, 3]);  getitem_210 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2492: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_267, [sym_size_int, 1500, -1]);  permute_267 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2492, [2])
	        amax_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2492, [2])
	        full_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_159, full_318);  amin_159 = full_318 = None
	        full_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_159, full_319);  amax_159 = full_319 = None
	        sub_7279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_159, minimum_159);  maximum_159 = None
	        div_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7279, 255.0);  sub_7279 = None
	        clamp_min_477: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_318, 1.1920928955078125e-07);  div_318 = None
	        div_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_159, clamp_min_477);  minimum_159 = None
	        round_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_319);  div_319 = None
	        sub_7285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_319);  round_319 = None
	        clamp_min_478: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7285, -128);  sub_7285 = None
	        clamp_max_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_478, 127);  clamp_min_478 = None
	        _assert_tensor_metadata_1433 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_477, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1433 = None
	        _assert_tensor_metadata_1434 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_318, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1434 = None
	        convert_element_type_954: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_318, torch.int8);  clamp_max_318 = None
	        view_2495: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_477, [sym_size_int, 1500, 1])
	        view_2496: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_954, [sym_size_int, 1500, 1])
	        reciprocal_159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2495);  view_2495 = None
	        mul_15449: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_159, 1.0);  reciprocal_159 = None
	        mul_15452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2492, mul_15449);  view_2492 = mul_15449 = None
	        round_320: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15452);  mul_15452 = None
	        add_24451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_320, view_2496);  round_320 = view_2496 = None
	        clamp_min_479: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24451, -128);  add_24451 = None
	        clamp_max_319: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_479, 127);  clamp_min_479 = None
	        _assert_tensor_metadata_1435 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_319, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1435 = None
	        convert_element_type_955: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_319, torch.int8);  clamp_max_319 = None
	        view_2499: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_477, [sym_size_int, 1500, 1]);  clamp_min_477 = None
	        view_2500: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_954, [sym_size_int, 1500, 1]);  convert_element_type_954 = None
	        _assert_tensor_metadata_1436 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_955, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1436 = None
	        convert_element_type_956: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_955, torch.float32);  convert_element_type_955 = None
	        _assert_tensor_metadata_1437 = torch.ops.aten._assert_tensor_metadata.default(view_2500, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1437 = None
	        convert_element_type_957: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2500, torch.float32);  view_2500 = None
	        sub_7305: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_956, convert_element_type_957);  convert_element_type_956 = convert_element_type_957 = None
	        mul_15474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7305, view_2499);  sub_7305 = view_2499 = None
	        _assert_tensor_metadata_1438 = torch.ops.aten._assert_tensor_metadata.default(mul_15474, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1438 = None
	        view_2502: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2503: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2504: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1439 = torch.ops.aten._assert_tensor_metadata.default(view_2502, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1439 = None
	        convert_element_type_958: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2502, torch.float32);  view_2502 = None
	        _assert_tensor_metadata_1440 = torch.ops.aten._assert_tensor_metadata.default(view_2504, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1440 = None
	        convert_element_type_959: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2504, torch.float32);  view_2504 = None
	        sub_7309: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_958, convert_element_type_959);  convert_element_type_958 = convert_element_type_959 = None
	        mul_15479: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7309, view_2503);  sub_7309 = view_2503 = None
	        view_2505: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15479, [1280, 1280]);  mul_15479 = None
	        _assert_tensor_metadata_1441 = torch.ops.aten._assert_tensor_metadata.default(view_2505, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1441 = None
	        mul_15484: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2506: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15474, [mul_15484, 1280]);  mul_15474 = mul_15484 = None
	        permute_268: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2505, [1, 0]);  view_2505 = None
	        addmm_132: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_26_self_attn_out_proj_bias, view_2506, permute_268);  model_audio_tower_layers_26_self_attn_out_proj_bias = view_2506 = permute_268 = None
	        view_2507: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_132, [sym_size_int, 1500, 1280]);  addmm_132 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_24514: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_23894, view_2507);  add_23894 = view_2507 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_24514, memory_format = torch.contiguous_format)
	        var_mean_53 = torch.ops.aten.var_mean.correction(clone_214, [2], correction = 0, keepdim = True)
	        getitem_214: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_53[0]
	        getitem_215: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_53[1];  var_mean_53 = None
	        add_24519: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_214, 1e-05);  getitem_214 = None
	        rsqrt_53: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_24519);  add_24519 = None
	        sub_7315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_214, getitem_215);  clone_214 = getitem_215 = None
	        mul_15495: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7315, rsqrt_53);  sub_7315 = rsqrt_53 = None
	        mul_15496: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15495, model_audio_tower_layers_26_final_layer_norm_weight);  mul_15495 = model_audio_tower_layers_26_final_layer_norm_weight = None
	        add_24520: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_15496, model_audio_tower_layers_26_final_layer_norm_bias);  mul_15496 = model_audio_tower_layers_26_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_24520, [2])
	        amax_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_24520, [2])
	        full_320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_160, full_320);  amin_160 = full_320 = None
	        full_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_160, full_321);  amax_160 = full_321 = None
	        sub_7326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_160, minimum_160);  maximum_160 = None
	        div_320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7326, 255.0);  sub_7326 = None
	        clamp_min_480: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_320, 1.1920928955078125e-07);  div_320 = None
	        div_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_160, clamp_min_480);  minimum_160 = None
	        round_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_321);  div_321 = None
	        sub_7332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_321);  round_321 = None
	        clamp_min_481: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7332, -128);  sub_7332 = None
	        clamp_max_320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_481, 127);  clamp_min_481 = None
	        _assert_tensor_metadata_1442 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_480, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1442 = None
	        _assert_tensor_metadata_1443 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_320, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1443 = None
	        convert_element_type_960: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_320, torch.int8);  clamp_max_320 = None
	        view_2510: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_480, [sym_size_int, 1500, 1])
	        view_2511: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_960, [sym_size_int, 1500, 1])
	        reciprocal_160: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2510);  view_2510 = None
	        mul_15544: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_160, 1.0);  reciprocal_160 = None
	        mul_15547: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_24520, mul_15544);  add_24520 = mul_15544 = None
	        round_322: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15547);  mul_15547 = None
	        add_24607: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_322, view_2511);  round_322 = view_2511 = None
	        clamp_min_482: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24607, -128);  add_24607 = None
	        clamp_max_321: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_482, 127);  clamp_min_482 = None
	        _assert_tensor_metadata_1444 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_321, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1444 = None
	        convert_element_type_961: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_321, torch.int8);  clamp_max_321 = None
	        view_2514: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_480, [sym_size_int, 1500, 1]);  clamp_min_480 = None
	        view_2515: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_960, [sym_size_int, 1500, 1]);  convert_element_type_960 = None
	        _assert_tensor_metadata_1445 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_961, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1445 = None
	        convert_element_type_962: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_961, torch.float32);  convert_element_type_961 = None
	        _assert_tensor_metadata_1446 = torch.ops.aten._assert_tensor_metadata.default(view_2515, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1446 = None
	        convert_element_type_963: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2515, torch.float32);  view_2515 = None
	        sub_7352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_962, convert_element_type_963);  convert_element_type_962 = convert_element_type_963 = None
	        mul_15569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7352, view_2514);  sub_7352 = view_2514 = None
	        _assert_tensor_metadata_1447 = torch.ops.aten._assert_tensor_metadata.default(mul_15569, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1447 = None
	        view_2517: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_26_fc1_parametrizations_weight_original0 = None
	        view_2518: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_26_fc1_parametrizations_weight_original1 = None
	        view_2519: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_26_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1448 = torch.ops.aten._assert_tensor_metadata.default(view_2517, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1448 = None
	        convert_element_type_964: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2517, torch.float32);  view_2517 = None
	        _assert_tensor_metadata_1449 = torch.ops.aten._assert_tensor_metadata.default(view_2519, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1449 = None
	        convert_element_type_965: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2519, torch.float32);  view_2519 = None
	        sub_7356: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_964, convert_element_type_965);  convert_element_type_964 = convert_element_type_965 = None
	        mul_15574: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7356, view_2518);  sub_7356 = view_2518 = None
	        view_2520: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15574, [5120, 1280]);  mul_15574 = None
	        _assert_tensor_metadata_1450 = torch.ops.aten._assert_tensor_metadata.default(view_2520, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1450 = None
	        mul_15579: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2521: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15569, [mul_15579, 1280]);  mul_15569 = mul_15579 = None
	        permute_269: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2520, [1, 0]);  view_2520 = None
	        addmm_133: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_26_fc1_bias, view_2521, permute_269);  model_audio_tower_layers_26_fc1_bias = view_2521 = permute_269 = None
	        view_2522: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_133, [sym_size_int, 1500, 5120]);  addmm_133 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_15586: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2522, 0.5)
	        mul_15587: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2522, 0.7071067811865476);  view_2522 = None
	        erf_28: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_15587);  mul_15587 = None
	        add_24666: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_28, 1);  erf_28 = None
	        mul_15588: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15586, add_24666);  mul_15586 = add_24666 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_15588, [2])
	        amax_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_15588, [2])
	        full_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_161, full_322);  amin_161 = full_322 = None
	        full_323: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_161, full_323);  amax_161 = full_323 = None
	        sub_7369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_161, minimum_161);  maximum_161 = None
	        div_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7369, 255.0);  sub_7369 = None
	        clamp_min_483: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_322, 1.1920928955078125e-07);  div_322 = None
	        div_323: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_161, clamp_min_483);  minimum_161 = None
	        round_323: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_323);  div_323 = None
	        sub_7375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_323);  round_323 = None
	        clamp_min_484: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7375, -128);  sub_7375 = None
	        clamp_max_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_484, 127);  clamp_min_484 = None
	        _assert_tensor_metadata_1451 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_483, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1451 = None
	        _assert_tensor_metadata_1452 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_322, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1452 = None
	        convert_element_type_966: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_322, torch.int8);  clamp_max_322 = None
	        view_2525: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_483, [sym_size_int, 1500, 1])
	        view_2526: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_966, [sym_size_int, 1500, 1])
	        reciprocal_161: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2525);  view_2525 = None
	        mul_15634: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_161, 1.0);  reciprocal_161 = None
	        mul_15637: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15588, mul_15634);  mul_15588 = mul_15634 = None
	        round_324: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_15637);  mul_15637 = None
	        add_24749: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_324, view_2526);  round_324 = view_2526 = None
	        clamp_min_485: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24749, -128);  add_24749 = None
	        clamp_max_323: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_485, 127);  clamp_min_485 = None
	        _assert_tensor_metadata_1453 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_323, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1453 = None
	        convert_element_type_967: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_323, torch.int8);  clamp_max_323 = None
	        view_2529: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_483, [sym_size_int, 1500, 1]);  clamp_min_483 = None
	        view_2530: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_966, [sym_size_int, 1500, 1]);  convert_element_type_966 = None
	        _assert_tensor_metadata_1454 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_967, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1454 = None
	        convert_element_type_968: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_967, torch.float32);  convert_element_type_967 = None
	        _assert_tensor_metadata_1455 = torch.ops.aten._assert_tensor_metadata.default(view_2530, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1455 = None
	        convert_element_type_969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2530, torch.float32);  view_2530 = None
	        sub_7395: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_968, convert_element_type_969);  convert_element_type_968 = convert_element_type_969 = None
	        mul_15659: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7395, view_2529);  sub_7395 = view_2529 = None
	        _assert_tensor_metadata_1456 = torch.ops.aten._assert_tensor_metadata.default(mul_15659, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1456 = None
	        view_2532: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_26_fc2_parametrizations_weight_original0 = None
	        view_2533: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_26_fc2_parametrizations_weight_original1 = None
	        view_2534: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_26_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1457 = torch.ops.aten._assert_tensor_metadata.default(view_2532, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1457 = None
	        convert_element_type_970: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2532, torch.float32);  view_2532 = None
	        _assert_tensor_metadata_1458 = torch.ops.aten._assert_tensor_metadata.default(view_2534, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1458 = None
	        convert_element_type_971: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2534, torch.float32);  view_2534 = None
	        sub_7399: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_970, convert_element_type_971);  convert_element_type_970 = convert_element_type_971 = None
	        mul_15664: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7399, view_2533);  sub_7399 = view_2533 = None
	        view_2535: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15664, [1280, 5120]);  mul_15664 = None
	        _assert_tensor_metadata_1459 = torch.ops.aten._assert_tensor_metadata.default(view_2535, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1459 = None
	        mul_15669: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2536: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15659, [mul_15669, 5120]);  mul_15659 = mul_15669 = None
	        permute_270: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2535, [1, 0]);  view_2535 = None
	        addmm_134: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_26_fc2_bias, view_2536, permute_270);  model_audio_tower_layers_26_fc2_bias = view_2536 = permute_270 = None
	        view_2537: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_134, [sym_size_int, 1500, 1280]);  addmm_134 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_24812: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_24514, view_2537);  add_24514 = view_2537 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_217: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_24812, memory_format = torch.contiguous_format)
	        var_mean_54 = torch.ops.aten.var_mean.correction(clone_217, [2], correction = 0, keepdim = True)
	        getitem_216: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_54[0]
	        getitem_217: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_54[1];  var_mean_54 = None
	        add_24817: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_216, 1e-05);  getitem_216 = None
	        rsqrt_54: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_24817);  add_24817 = None
	        sub_7405: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_217, getitem_217);  clone_217 = getitem_217 = None
	        mul_15680: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7405, rsqrt_54);  sub_7405 = rsqrt_54 = None
	        mul_15681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15680, model_audio_tower_layers_27_self_attn_layer_norm_weight);  mul_15680 = model_audio_tower_layers_27_self_attn_layer_norm_weight = None
	        add_24818: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_15681, model_audio_tower_layers_27_self_attn_layer_norm_bias);  mul_15681 = model_audio_tower_layers_27_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_24818, [2])
	        amax_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_24818, [2])
	        full_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_162, full_324);  amin_162 = full_324 = None
	        full_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_162, full_325);  amax_162 = full_325 = None
	        sub_7416: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_162, minimum_162);  maximum_162 = None
	        div_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7416, 255.0);  sub_7416 = None
	        clamp_min_486: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_324, 1.1920928955078125e-07);  div_324 = None
	        div_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_162, clamp_min_486);  minimum_162 = None
	        round_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_325);  div_325 = None
	        sub_7422: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_325);  round_325 = None
	        clamp_min_487: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7422, -128);  sub_7422 = None
	        clamp_max_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_487, 127);  clamp_min_487 = None
	        _assert_tensor_metadata_1460 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_486, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1460 = None
	        _assert_tensor_metadata_1461 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_324, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1461 = None
	        convert_element_type_972: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_324, torch.int8);  clamp_max_324 = None
	        view_2540: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_486, [sym_size_int, 1500, 1])
	        view_2541: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_972, [sym_size_int, 1500, 1])
	        reciprocal_162: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2540);  view_2540 = None
	        mul_15729: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_162, 1.0);  reciprocal_162 = None
	        mul_15732: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_24818, mul_15729);  mul_15729 = None
	        round_326: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15732);  mul_15732 = None
	        add_24905: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_326, view_2541);  round_326 = view_2541 = None
	        clamp_min_488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24905, -128);  add_24905 = None
	        clamp_max_325: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_488, 127);  clamp_min_488 = None
	        _assert_tensor_metadata_1462 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_325, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1462 = None
	        convert_element_type_973: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_325, torch.int8);  clamp_max_325 = None
	        view_2544: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_486, [sym_size_int, 1500, 1]);  clamp_min_486 = None
	        view_2545: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_972, [sym_size_int, 1500, 1]);  convert_element_type_972 = None
	        _assert_tensor_metadata_1463 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_973, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1463 = None
	        convert_element_type_974: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_973, torch.float32);  convert_element_type_973 = None
	        _assert_tensor_metadata_1464 = torch.ops.aten._assert_tensor_metadata.default(view_2545, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1464 = None
	        convert_element_type_975: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2545, torch.float32);  view_2545 = None
	        sub_7442: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_974, convert_element_type_975);  convert_element_type_974 = convert_element_type_975 = None
	        mul_15754: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7442, view_2544);  sub_7442 = view_2544 = None
	        _assert_tensor_metadata_1465 = torch.ops.aten._assert_tensor_metadata.default(mul_15754, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1465 = None
	        view_2547: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2548: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2549: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1466 = torch.ops.aten._assert_tensor_metadata.default(view_2547, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1466 = None
	        convert_element_type_976: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2547, torch.float32);  view_2547 = None
	        _assert_tensor_metadata_1467 = torch.ops.aten._assert_tensor_metadata.default(view_2549, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1467 = None
	        convert_element_type_977: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2549, torch.float32);  view_2549 = None
	        sub_7446: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_976, convert_element_type_977);  convert_element_type_976 = convert_element_type_977 = None
	        mul_15759: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7446, view_2548);  sub_7446 = view_2548 = None
	        view_2550: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15759, [1280, 1280]);  mul_15759 = None
	        _assert_tensor_metadata_1468 = torch.ops.aten._assert_tensor_metadata.default(view_2550, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1468 = None
	        mul_15764: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2551: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15754, [mul_15764, 1280]);  mul_15754 = mul_15764 = None
	        permute_271: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2550, [1, 0]);  view_2550 = None
	        addmm_135: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_27_self_attn_q_proj_bias, view_2551, permute_271);  model_audio_tower_layers_27_self_attn_q_proj_bias = view_2551 = permute_271 = None
	        view_2552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_135, [sym_size_int, 1500, 1280]);  addmm_135 = None
	        mul_15771: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2552, 0.125);  view_2552 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2553: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15771, [sym_size_int, 1500, 20, 64]);  mul_15771 = None
	        permute_272: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2553, [0, 2, 1, 3]);  view_2553 = None
	        clone_218: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_272, memory_format = torch.contiguous_format);  permute_272 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_24818, [2])
	        amax_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_24818, [2])
	        full_326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_163, full_326);  amin_163 = full_326 = None
	        full_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_163, full_327);  amax_163 = full_327 = None
	        sub_7461: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_163, minimum_163);  maximum_163 = None
	        div_326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7461, 255.0);  sub_7461 = None
	        clamp_min_489: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_326, 1.1920928955078125e-07);  div_326 = None
	        div_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_163, clamp_min_489);  minimum_163 = None
	        round_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_327);  div_327 = None
	        sub_7467: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_327);  round_327 = None
	        clamp_min_490: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7467, -128);  sub_7467 = None
	        clamp_max_326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_490, 127);  clamp_min_490 = None
	        _assert_tensor_metadata_1469 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_489, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1469 = None
	        _assert_tensor_metadata_1470 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_326, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1470 = None
	        convert_element_type_978: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_326, torch.int8);  clamp_max_326 = None
	        view_2556: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_489, [sym_size_int, 1500, 1])
	        view_2557: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_978, [sym_size_int, 1500, 1])
	        reciprocal_163: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2556);  view_2556 = None
	        mul_15825: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_163, 1.0);  reciprocal_163 = None
	        mul_15828: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_24818, mul_15825);  mul_15825 = None
	        round_328: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15828);  mul_15828 = None
	        add_25057: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_328, view_2557);  round_328 = view_2557 = None
	        clamp_min_491: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25057, -128);  add_25057 = None
	        clamp_max_327: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_491, 127);  clamp_min_491 = None
	        _assert_tensor_metadata_1471 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_327, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1471 = None
	        convert_element_type_979: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_327, torch.int8);  clamp_max_327 = None
	        view_2560: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_489, [sym_size_int, 1500, 1]);  clamp_min_489 = None
	        view_2561: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_978, [sym_size_int, 1500, 1]);  convert_element_type_978 = None
	        _assert_tensor_metadata_1472 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_979, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1472 = None
	        convert_element_type_980: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_979, torch.float32);  convert_element_type_979 = None
	        _assert_tensor_metadata_1473 = torch.ops.aten._assert_tensor_metadata.default(view_2561, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1473 = None
	        convert_element_type_981: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2561, torch.float32);  view_2561 = None
	        sub_7487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_980, convert_element_type_981);  convert_element_type_980 = convert_element_type_981 = None
	        mul_15850: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7487, view_2560);  sub_7487 = view_2560 = None
	        _assert_tensor_metadata_1474 = torch.ops.aten._assert_tensor_metadata.default(mul_15850, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1474 = None
	        view_2563: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2564: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2565: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1475 = torch.ops.aten._assert_tensor_metadata.default(view_2563, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1475 = None
	        convert_element_type_982: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2563, torch.float32);  view_2563 = None
	        _assert_tensor_metadata_1476 = torch.ops.aten._assert_tensor_metadata.default(view_2565, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1476 = None
	        convert_element_type_983: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2565, torch.float32);  view_2565 = None
	        sub_7491: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_982, convert_element_type_983);  convert_element_type_982 = convert_element_type_983 = None
	        mul_15855: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7491, view_2564);  sub_7491 = view_2564 = None
	        view_2566: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15855, [1280, 1280]);  mul_15855 = None
	        _assert_tensor_metadata_1477 = torch.ops.aten._assert_tensor_metadata.default(view_2566, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1477 = None
	        permute_273: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2566, [1, 0]);  view_2566 = None
	        mul_15858: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2567: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15850, [mul_15858, 1280]);  mul_15850 = mul_15858 = None
	        mm_27: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2567, permute_273);  view_2567 = permute_273 = None
	        view_2568: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_27, [sym_size_int, 1500, 1280]);  mm_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2569: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2568, [sym_size_int, -1, 20, 64]);  view_2568 = None
	        permute_274: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2569, [0, 2, 1, 3]);  view_2569 = None
	        clone_219: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_274, memory_format = torch.contiguous_format);  permute_274 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_24818, [2])
	        amax_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_24818, [2])
	        full_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_164, full_328);  amin_164 = full_328 = None
	        full_329: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_164, full_329);  amax_164 = full_329 = None
	        sub_7505: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_164, minimum_164);  maximum_164 = None
	        div_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7505, 255.0);  sub_7505 = None
	        clamp_min_492: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_328, 1.1920928955078125e-07);  div_328 = None
	        div_329: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_164, clamp_min_492);  minimum_164 = None
	        round_329: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_329);  div_329 = None
	        sub_7511: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_329);  round_329 = None
	        clamp_min_493: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7511, -128);  sub_7511 = None
	        clamp_max_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_493, 127);  clamp_min_493 = None
	        _assert_tensor_metadata_1478 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_492, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1478 = None
	        _assert_tensor_metadata_1479 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_328, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1479 = None
	        convert_element_type_984: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_328, torch.int8);  clamp_max_328 = None
	        view_2572: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_492, [sym_size_int, 1500, 1])
	        view_2573: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_984, [sym_size_int, 1500, 1])
	        reciprocal_164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2572);  view_2572 = None
	        mul_15924: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_164, 1.0);  reciprocal_164 = None
	        mul_15927: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_24818, mul_15924);  add_24818 = mul_15924 = None
	        round_330: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15927);  mul_15927 = None
	        add_25205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_330, view_2573);  round_330 = view_2573 = None
	        clamp_min_494: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25205, -128);  add_25205 = None
	        clamp_max_329: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_494, 127);  clamp_min_494 = None
	        _assert_tensor_metadata_1480 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_329, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1480 = None
	        convert_element_type_985: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_329, torch.int8);  clamp_max_329 = None
	        view_2576: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_492, [sym_size_int, 1500, 1]);  clamp_min_492 = None
	        view_2577: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_984, [sym_size_int, 1500, 1]);  convert_element_type_984 = None
	        _assert_tensor_metadata_1481 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_985, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1481 = None
	        convert_element_type_986: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_985, torch.float32);  convert_element_type_985 = None
	        _assert_tensor_metadata_1482 = torch.ops.aten._assert_tensor_metadata.default(view_2577, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1482 = None
	        convert_element_type_987: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2577, torch.float32);  view_2577 = None
	        sub_7531: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_986, convert_element_type_987);  convert_element_type_986 = convert_element_type_987 = None
	        mul_15949: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7531, view_2576);  sub_7531 = view_2576 = None
	        _assert_tensor_metadata_1483 = torch.ops.aten._assert_tensor_metadata.default(mul_15949, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1483 = None
	        view_2579: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2580: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2581: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1484 = torch.ops.aten._assert_tensor_metadata.default(view_2579, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1484 = None
	        convert_element_type_988: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2579, torch.float32);  view_2579 = None
	        _assert_tensor_metadata_1485 = torch.ops.aten._assert_tensor_metadata.default(view_2581, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1485 = None
	        convert_element_type_989: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2581, torch.float32);  view_2581 = None
	        sub_7535: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_988, convert_element_type_989);  convert_element_type_988 = convert_element_type_989 = None
	        mul_15954: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7535, view_2580);  sub_7535 = view_2580 = None
	        view_2582: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15954, [1280, 1280]);  mul_15954 = None
	        _assert_tensor_metadata_1486 = torch.ops.aten._assert_tensor_metadata.default(view_2582, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1486 = None
	        mul_15959: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2583: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15949, [mul_15959, 1280]);  mul_15949 = mul_15959 = None
	        permute_275: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2582, [1, 0]);  view_2582 = None
	        addmm_136: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_27_self_attn_v_proj_bias, view_2583, permute_275);  model_audio_tower_layers_27_self_attn_v_proj_bias = view_2583 = permute_275 = None
	        view_2584: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_136, [sym_size_int, 1500, 1280]);  addmm_136 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2585: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2584, [sym_size_int, -1, 20, 64]);  view_2584 = None
	        permute_276: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2585, [0, 2, 1, 3]);  view_2585 = None
	        clone_220: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_276, memory_format = torch.contiguous_format);  permute_276 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_27 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_218, clone_219, clone_220, None, False, scale = 1.0);  clone_218 = clone_219 = clone_220 = None
	        getitem_218: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_27[0];  _scaled_dot_product_efficient_attention_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_277: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_218, [0, 2, 1, 3]);  getitem_218 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2586: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_277, [sym_size_int, 1500, -1]);  permute_277 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2586, [2])
	        amax_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2586, [2])
	        full_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_165, full_330);  amin_165 = full_330 = None
	        full_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_165, full_331);  amax_165 = full_331 = None
	        sub_7553: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_165, minimum_165);  maximum_165 = None
	        div_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7553, 255.0);  sub_7553 = None
	        clamp_min_495: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_330, 1.1920928955078125e-07);  div_330 = None
	        div_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_165, clamp_min_495);  minimum_165 = None
	        round_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_331);  div_331 = None
	        sub_7559: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_331);  round_331 = None
	        clamp_min_496: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7559, -128);  sub_7559 = None
	        clamp_max_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_496, 127);  clamp_min_496 = None
	        _assert_tensor_metadata_1487 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_495, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1487 = None
	        _assert_tensor_metadata_1488 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_330, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1488 = None
	        convert_element_type_990: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_330, torch.int8);  clamp_max_330 = None
	        view_2589: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_495, [sym_size_int, 1500, 1])
	        view_2590: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_990, [sym_size_int, 1500, 1])
	        reciprocal_165: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2589);  view_2589 = None
	        mul_16029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_165, 1.0);  reciprocal_165 = None
	        mul_16032: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2586, mul_16029);  view_2586 = mul_16029 = None
	        round_332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16032);  mul_16032 = None
	        add_25369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_332, view_2590);  round_332 = view_2590 = None
	        clamp_min_497: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25369, -128);  add_25369 = None
	        clamp_max_331: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_497, 127);  clamp_min_497 = None
	        _assert_tensor_metadata_1489 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_331, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1489 = None
	        convert_element_type_991: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_331, torch.int8);  clamp_max_331 = None
	        view_2593: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_495, [sym_size_int, 1500, 1]);  clamp_min_495 = None
	        view_2594: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_990, [sym_size_int, 1500, 1]);  convert_element_type_990 = None
	        _assert_tensor_metadata_1490 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_991, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1490 = None
	        convert_element_type_992: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_991, torch.float32);  convert_element_type_991 = None
	        _assert_tensor_metadata_1491 = torch.ops.aten._assert_tensor_metadata.default(view_2594, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1491 = None
	        convert_element_type_993: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2594, torch.float32);  view_2594 = None
	        sub_7579: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_992, convert_element_type_993);  convert_element_type_992 = convert_element_type_993 = None
	        mul_16054: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7579, view_2593);  sub_7579 = view_2593 = None
	        _assert_tensor_metadata_1492 = torch.ops.aten._assert_tensor_metadata.default(mul_16054, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1492 = None
	        view_2596: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2597: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2598: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1493 = torch.ops.aten._assert_tensor_metadata.default(view_2596, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1493 = None
	        convert_element_type_994: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2596, torch.float32);  view_2596 = None
	        _assert_tensor_metadata_1494 = torch.ops.aten._assert_tensor_metadata.default(view_2598, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1494 = None
	        convert_element_type_995: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2598, torch.float32);  view_2598 = None
	        sub_7583: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_994, convert_element_type_995);  convert_element_type_994 = convert_element_type_995 = None
	        mul_16059: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7583, view_2597);  sub_7583 = view_2597 = None
	        view_2599: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16059, [1280, 1280]);  mul_16059 = None
	        _assert_tensor_metadata_1495 = torch.ops.aten._assert_tensor_metadata.default(view_2599, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1495 = None
	        mul_16064: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2600: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16054, [mul_16064, 1280]);  mul_16054 = mul_16064 = None
	        permute_278: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2599, [1, 0]);  view_2599 = None
	        addmm_137: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_27_self_attn_out_proj_bias, view_2600, permute_278);  model_audio_tower_layers_27_self_attn_out_proj_bias = view_2600 = permute_278 = None
	        view_2601: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_137, [sym_size_int, 1500, 1280]);  addmm_137 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_25432: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_24812, view_2601);  add_24812 = view_2601 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_222: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_25432, memory_format = torch.contiguous_format)
	        var_mean_55 = torch.ops.aten.var_mean.correction(clone_222, [2], correction = 0, keepdim = True)
	        getitem_222: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_55[0]
	        getitem_223: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_55[1];  var_mean_55 = None
	        add_25437: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_222, 1e-05);  getitem_222 = None
	        rsqrt_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_25437);  add_25437 = None
	        sub_7589: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_222, getitem_223);  clone_222 = getitem_223 = None
	        mul_16075: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7589, rsqrt_55);  sub_7589 = rsqrt_55 = None
	        mul_16076: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16075, model_audio_tower_layers_27_final_layer_norm_weight);  mul_16075 = model_audio_tower_layers_27_final_layer_norm_weight = None
	        add_25438: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16076, model_audio_tower_layers_27_final_layer_norm_bias);  mul_16076 = model_audio_tower_layers_27_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_25438, [2])
	        amax_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_25438, [2])
	        full_332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_166, full_332);  amin_166 = full_332 = None
	        full_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_166, full_333);  amax_166 = full_333 = None
	        sub_7600: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_166, minimum_166);  maximum_166 = None
	        div_332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7600, 255.0);  sub_7600 = None
	        clamp_min_498: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_332, 1.1920928955078125e-07);  div_332 = None
	        div_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_166, clamp_min_498);  minimum_166 = None
	        round_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_333);  div_333 = None
	        sub_7606: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_333);  round_333 = None
	        clamp_min_499: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7606, -128);  sub_7606 = None
	        clamp_max_332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_499, 127);  clamp_min_499 = None
	        _assert_tensor_metadata_1496 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_498, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1496 = None
	        _assert_tensor_metadata_1497 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_332, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1497 = None
	        convert_element_type_996: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_332, torch.int8);  clamp_max_332 = None
	        view_2604: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_498, [sym_size_int, 1500, 1])
	        view_2605: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_996, [sym_size_int, 1500, 1])
	        reciprocal_166: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2604);  view_2604 = None
	        mul_16124: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_166, 1.0);  reciprocal_166 = None
	        mul_16127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_25438, mul_16124);  add_25438 = mul_16124 = None
	        round_334: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16127);  mul_16127 = None
	        add_25525: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_334, view_2605);  round_334 = view_2605 = None
	        clamp_min_500: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25525, -128);  add_25525 = None
	        clamp_max_333: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_500, 127);  clamp_min_500 = None
	        _assert_tensor_metadata_1498 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_333, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1498 = None
	        convert_element_type_997: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_333, torch.int8);  clamp_max_333 = None
	        view_2608: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_498, [sym_size_int, 1500, 1]);  clamp_min_498 = None
	        view_2609: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_996, [sym_size_int, 1500, 1]);  convert_element_type_996 = None
	        _assert_tensor_metadata_1499 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_997, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1499 = None
	        convert_element_type_998: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_997, torch.float32);  convert_element_type_997 = None
	        _assert_tensor_metadata_1500 = torch.ops.aten._assert_tensor_metadata.default(view_2609, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1500 = None
	        convert_element_type_999: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2609, torch.float32);  view_2609 = None
	        sub_7626: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_998, convert_element_type_999);  convert_element_type_998 = convert_element_type_999 = None
	        mul_16149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7626, view_2608);  sub_7626 = view_2608 = None
	        _assert_tensor_metadata_1501 = torch.ops.aten._assert_tensor_metadata.default(mul_16149, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1501 = None
	        view_2611: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_27_fc1_parametrizations_weight_original0 = None
	        view_2612: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_27_fc1_parametrizations_weight_original1 = None
	        view_2613: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_27_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1502 = torch.ops.aten._assert_tensor_metadata.default(view_2611, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1502 = None
	        convert_element_type_1000: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2611, torch.float32);  view_2611 = None
	        _assert_tensor_metadata_1503 = torch.ops.aten._assert_tensor_metadata.default(view_2613, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1503 = None
	        convert_element_type_1001: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2613, torch.float32);  view_2613 = None
	        sub_7630: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1000, convert_element_type_1001);  convert_element_type_1000 = convert_element_type_1001 = None
	        mul_16154: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7630, view_2612);  sub_7630 = view_2612 = None
	        view_2614: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16154, [5120, 1280]);  mul_16154 = None
	        _assert_tensor_metadata_1504 = torch.ops.aten._assert_tensor_metadata.default(view_2614, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1504 = None
	        mul_16159: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2615: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16149, [mul_16159, 1280]);  mul_16149 = mul_16159 = None
	        permute_279: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2614, [1, 0]);  view_2614 = None
	        addmm_138: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_27_fc1_bias, view_2615, permute_279);  model_audio_tower_layers_27_fc1_bias = view_2615 = permute_279 = None
	        view_2616: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_138, [sym_size_int, 1500, 5120]);  addmm_138 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_16166: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2616, 0.5)
	        mul_16167: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2616, 0.7071067811865476);  view_2616 = None
	        erf_29: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_16167);  mul_16167 = None
	        add_25584: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_29, 1);  erf_29 = None
	        mul_16168: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16166, add_25584);  mul_16166 = add_25584 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_16168, [2])
	        amax_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_16168, [2])
	        full_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_167, full_334);  amin_167 = full_334 = None
	        full_335: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_167, full_335);  amax_167 = full_335 = None
	        sub_7643: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_167, minimum_167);  maximum_167 = None
	        div_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7643, 255.0);  sub_7643 = None
	        clamp_min_501: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_334, 1.1920928955078125e-07);  div_334 = None
	        div_335: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_167, clamp_min_501);  minimum_167 = None
	        round_335: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_335);  div_335 = None
	        sub_7649: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_335);  round_335 = None
	        clamp_min_502: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7649, -128);  sub_7649 = None
	        clamp_max_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_502, 127);  clamp_min_502 = None
	        _assert_tensor_metadata_1505 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_501, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1505 = None
	        _assert_tensor_metadata_1506 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_334, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1506 = None
	        convert_element_type_1002: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_334, torch.int8);  clamp_max_334 = None
	        view_2619: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_501, [sym_size_int, 1500, 1])
	        view_2620: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1002, [sym_size_int, 1500, 1])
	        reciprocal_167: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2619);  view_2619 = None
	        mul_16214: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_167, 1.0);  reciprocal_167 = None
	        mul_16217: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16168, mul_16214);  mul_16168 = mul_16214 = None
	        round_336: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_16217);  mul_16217 = None
	        add_25667: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_336, view_2620);  round_336 = view_2620 = None
	        clamp_min_503: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25667, -128);  add_25667 = None
	        clamp_max_335: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_503, 127);  clamp_min_503 = None
	        _assert_tensor_metadata_1507 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_335, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1507 = None
	        convert_element_type_1003: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_335, torch.int8);  clamp_max_335 = None
	        view_2623: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_501, [sym_size_int, 1500, 1]);  clamp_min_501 = None
	        view_2624: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1002, [sym_size_int, 1500, 1]);  convert_element_type_1002 = None
	        _assert_tensor_metadata_1508 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1003, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1508 = None
	        convert_element_type_1004: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1003, torch.float32);  convert_element_type_1003 = None
	        _assert_tensor_metadata_1509 = torch.ops.aten._assert_tensor_metadata.default(view_2624, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1509 = None
	        convert_element_type_1005: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2624, torch.float32);  view_2624 = None
	        sub_7669: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1004, convert_element_type_1005);  convert_element_type_1004 = convert_element_type_1005 = None
	        mul_16239: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7669, view_2623);  sub_7669 = view_2623 = None
	        _assert_tensor_metadata_1510 = torch.ops.aten._assert_tensor_metadata.default(mul_16239, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1510 = None
	        view_2626: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_27_fc2_parametrizations_weight_original0 = None
	        view_2627: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_27_fc2_parametrizations_weight_original1 = None
	        view_2628: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_27_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1511 = torch.ops.aten._assert_tensor_metadata.default(view_2626, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1511 = None
	        convert_element_type_1006: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2626, torch.float32);  view_2626 = None
	        _assert_tensor_metadata_1512 = torch.ops.aten._assert_tensor_metadata.default(view_2628, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1512 = None
	        convert_element_type_1007: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2628, torch.float32);  view_2628 = None
	        sub_7673: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1006, convert_element_type_1007);  convert_element_type_1006 = convert_element_type_1007 = None
	        mul_16244: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7673, view_2627);  sub_7673 = view_2627 = None
	        view_2629: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16244, [1280, 5120]);  mul_16244 = None
	        _assert_tensor_metadata_1513 = torch.ops.aten._assert_tensor_metadata.default(view_2629, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1513 = None
	        mul_16249: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2630: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16239, [mul_16249, 5120]);  mul_16239 = mul_16249 = None
	        permute_280: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2629, [1, 0]);  view_2629 = None
	        addmm_139: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_27_fc2_bias, view_2630, permute_280);  model_audio_tower_layers_27_fc2_bias = view_2630 = permute_280 = None
	        view_2631: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_139, [sym_size_int, 1500, 1280]);  addmm_139 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_25730: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_25432, view_2631);  add_25432 = view_2631 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_225: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_25730, memory_format = torch.contiguous_format)
	        var_mean_56 = torch.ops.aten.var_mean.correction(clone_225, [2], correction = 0, keepdim = True)
	        getitem_224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_56[0]
	        getitem_225: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_56[1];  var_mean_56 = None
	        add_25735: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_224, 1e-05);  getitem_224 = None
	        rsqrt_56: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_25735);  add_25735 = None
	        sub_7679: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_225, getitem_225);  clone_225 = getitem_225 = None
	        mul_16260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7679, rsqrt_56);  sub_7679 = rsqrt_56 = None
	        mul_16261: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16260, model_audio_tower_layers_28_self_attn_layer_norm_weight);  mul_16260 = model_audio_tower_layers_28_self_attn_layer_norm_weight = None
	        add_25736: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16261, model_audio_tower_layers_28_self_attn_layer_norm_bias);  mul_16261 = model_audio_tower_layers_28_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_25736, [2])
	        amax_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_25736, [2])
	        full_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_168, full_336);  amin_168 = full_336 = None
	        full_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_168, full_337);  amax_168 = full_337 = None
	        sub_7690: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_168, minimum_168);  maximum_168 = None
	        div_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7690, 255.0);  sub_7690 = None
	        clamp_min_504: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_336, 1.1920928955078125e-07);  div_336 = None
	        div_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_168, clamp_min_504);  minimum_168 = None
	        round_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_337);  div_337 = None
	        sub_7696: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_337);  round_337 = None
	        clamp_min_505: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7696, -128);  sub_7696 = None
	        clamp_max_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_505, 127);  clamp_min_505 = None
	        _assert_tensor_metadata_1514 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_504, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1514 = None
	        _assert_tensor_metadata_1515 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_336, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1515 = None
	        convert_element_type_1008: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_336, torch.int8);  clamp_max_336 = None
	        view_2634: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_504, [sym_size_int, 1500, 1])
	        view_2635: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1008, [sym_size_int, 1500, 1])
	        reciprocal_168: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2634);  view_2634 = None
	        mul_16309: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_168, 1.0);  reciprocal_168 = None
	        mul_16312: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_25736, mul_16309);  mul_16309 = None
	        round_338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16312);  mul_16312 = None
	        add_25823: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_338, view_2635);  round_338 = view_2635 = None
	        clamp_min_506: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25823, -128);  add_25823 = None
	        clamp_max_337: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_506, 127);  clamp_min_506 = None
	        _assert_tensor_metadata_1516 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_337, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1516 = None
	        convert_element_type_1009: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_337, torch.int8);  clamp_max_337 = None
	        view_2638: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_504, [sym_size_int, 1500, 1]);  clamp_min_504 = None
	        view_2639: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1008, [sym_size_int, 1500, 1]);  convert_element_type_1008 = None
	        _assert_tensor_metadata_1517 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1009, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1517 = None
	        convert_element_type_1010: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1009, torch.float32);  convert_element_type_1009 = None
	        _assert_tensor_metadata_1518 = torch.ops.aten._assert_tensor_metadata.default(view_2639, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1518 = None
	        convert_element_type_1011: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2639, torch.float32);  view_2639 = None
	        sub_7716: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1010, convert_element_type_1011);  convert_element_type_1010 = convert_element_type_1011 = None
	        mul_16334: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7716, view_2638);  sub_7716 = view_2638 = None
	        _assert_tensor_metadata_1519 = torch.ops.aten._assert_tensor_metadata.default(mul_16334, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1519 = None
	        view_2641: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2642: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2643: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1520 = torch.ops.aten._assert_tensor_metadata.default(view_2641, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1520 = None
	        convert_element_type_1012: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2641, torch.float32);  view_2641 = None
	        _assert_tensor_metadata_1521 = torch.ops.aten._assert_tensor_metadata.default(view_2643, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1521 = None
	        convert_element_type_1013: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2643, torch.float32);  view_2643 = None
	        sub_7720: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1012, convert_element_type_1013);  convert_element_type_1012 = convert_element_type_1013 = None
	        mul_16339: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7720, view_2642);  sub_7720 = view_2642 = None
	        view_2644: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16339, [1280, 1280]);  mul_16339 = None
	        _assert_tensor_metadata_1522 = torch.ops.aten._assert_tensor_metadata.default(view_2644, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1522 = None
	        mul_16344: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2645: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16334, [mul_16344, 1280]);  mul_16334 = mul_16344 = None
	        permute_281: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2644, [1, 0]);  view_2644 = None
	        addmm_140: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_28_self_attn_q_proj_bias, view_2645, permute_281);  model_audio_tower_layers_28_self_attn_q_proj_bias = view_2645 = permute_281 = None
	        view_2646: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_140, [sym_size_int, 1500, 1280]);  addmm_140 = None
	        mul_16351: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2646, 0.125);  view_2646 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2647: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16351, [sym_size_int, 1500, 20, 64]);  mul_16351 = None
	        permute_282: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2647, [0, 2, 1, 3]);  view_2647 = None
	        clone_226: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_282, memory_format = torch.contiguous_format);  permute_282 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_25736, [2])
	        amax_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_25736, [2])
	        full_338: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_169, full_338);  amin_169 = full_338 = None
	        full_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_169, full_339);  amax_169 = full_339 = None
	        sub_7735: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_169, minimum_169);  maximum_169 = None
	        div_338: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7735, 255.0);  sub_7735 = None
	        clamp_min_507: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_338, 1.1920928955078125e-07);  div_338 = None
	        div_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_169, clamp_min_507);  minimum_169 = None
	        round_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_339);  div_339 = None
	        sub_7741: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_339);  round_339 = None
	        clamp_min_508: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7741, -128);  sub_7741 = None
	        clamp_max_338: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_508, 127);  clamp_min_508 = None
	        _assert_tensor_metadata_1523 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_507, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1523 = None
	        _assert_tensor_metadata_1524 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_338, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1524 = None
	        convert_element_type_1014: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_338, torch.int8);  clamp_max_338 = None
	        view_2650: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_507, [sym_size_int, 1500, 1])
	        view_2651: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1014, [sym_size_int, 1500, 1])
	        reciprocal_169: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2650);  view_2650 = None
	        mul_16405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_169, 1.0);  reciprocal_169 = None
	        mul_16408: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_25736, mul_16405);  mul_16405 = None
	        round_340: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16408);  mul_16408 = None
	        add_25975: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_340, view_2651);  round_340 = view_2651 = None
	        clamp_min_509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25975, -128);  add_25975 = None
	        clamp_max_339: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_509, 127);  clamp_min_509 = None
	        _assert_tensor_metadata_1525 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_339, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1525 = None
	        convert_element_type_1015: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_339, torch.int8);  clamp_max_339 = None
	        view_2654: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_507, [sym_size_int, 1500, 1]);  clamp_min_507 = None
	        view_2655: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1014, [sym_size_int, 1500, 1]);  convert_element_type_1014 = None
	        _assert_tensor_metadata_1526 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1015, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1526 = None
	        convert_element_type_1016: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1015, torch.float32);  convert_element_type_1015 = None
	        _assert_tensor_metadata_1527 = torch.ops.aten._assert_tensor_metadata.default(view_2655, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1527 = None
	        convert_element_type_1017: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2655, torch.float32);  view_2655 = None
	        sub_7761: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1016, convert_element_type_1017);  convert_element_type_1016 = convert_element_type_1017 = None
	        mul_16430: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7761, view_2654);  sub_7761 = view_2654 = None
	        _assert_tensor_metadata_1528 = torch.ops.aten._assert_tensor_metadata.default(mul_16430, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1528 = None
	        view_2657: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2658: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2659: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1529 = torch.ops.aten._assert_tensor_metadata.default(view_2657, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1529 = None
	        convert_element_type_1018: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2657, torch.float32);  view_2657 = None
	        _assert_tensor_metadata_1530 = torch.ops.aten._assert_tensor_metadata.default(view_2659, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1530 = None
	        convert_element_type_1019: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2659, torch.float32);  view_2659 = None
	        sub_7765: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1018, convert_element_type_1019);  convert_element_type_1018 = convert_element_type_1019 = None
	        mul_16435: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7765, view_2658);  sub_7765 = view_2658 = None
	        view_2660: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16435, [1280, 1280]);  mul_16435 = None
	        _assert_tensor_metadata_1531 = torch.ops.aten._assert_tensor_metadata.default(view_2660, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1531 = None
	        permute_283: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2660, [1, 0]);  view_2660 = None
	        mul_16438: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2661: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16430, [mul_16438, 1280]);  mul_16430 = mul_16438 = None
	        mm_28: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2661, permute_283);  view_2661 = permute_283 = None
	        view_2662: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_28, [sym_size_int, 1500, 1280]);  mm_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2663: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2662, [sym_size_int, -1, 20, 64]);  view_2662 = None
	        permute_284: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2663, [0, 2, 1, 3]);  view_2663 = None
	        clone_227: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_284, memory_format = torch.contiguous_format);  permute_284 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_25736, [2])
	        amax_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_25736, [2])
	        full_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_170, full_340);  amin_170 = full_340 = None
	        full_341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_170, full_341);  amax_170 = full_341 = None
	        sub_7779: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_170, minimum_170);  maximum_170 = None
	        div_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7779, 255.0);  sub_7779 = None
	        clamp_min_510: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_340, 1.1920928955078125e-07);  div_340 = None
	        div_341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_170, clamp_min_510);  minimum_170 = None
	        round_341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_341);  div_341 = None
	        sub_7785: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_341);  round_341 = None
	        clamp_min_511: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7785, -128);  sub_7785 = None
	        clamp_max_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_511, 127);  clamp_min_511 = None
	        _assert_tensor_metadata_1532 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_510, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1532 = None
	        _assert_tensor_metadata_1533 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_340, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1533 = None
	        convert_element_type_1020: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_340, torch.int8);  clamp_max_340 = None
	        view_2666: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_510, [sym_size_int, 1500, 1])
	        view_2667: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1020, [sym_size_int, 1500, 1])
	        reciprocal_170: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2666);  view_2666 = None
	        mul_16504: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_170, 1.0);  reciprocal_170 = None
	        mul_16507: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_25736, mul_16504);  add_25736 = mul_16504 = None
	        round_342: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16507);  mul_16507 = None
	        add_26123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_342, view_2667);  round_342 = view_2667 = None
	        clamp_min_512: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26123, -128);  add_26123 = None
	        clamp_max_341: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_512, 127);  clamp_min_512 = None
	        _assert_tensor_metadata_1534 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_341, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1534 = None
	        convert_element_type_1021: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_341, torch.int8);  clamp_max_341 = None
	        view_2670: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_510, [sym_size_int, 1500, 1]);  clamp_min_510 = None
	        view_2671: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1020, [sym_size_int, 1500, 1]);  convert_element_type_1020 = None
	        _assert_tensor_metadata_1535 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1021, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1535 = None
	        convert_element_type_1022: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1021, torch.float32);  convert_element_type_1021 = None
	        _assert_tensor_metadata_1536 = torch.ops.aten._assert_tensor_metadata.default(view_2671, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1536 = None
	        convert_element_type_1023: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2671, torch.float32);  view_2671 = None
	        sub_7805: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1022, convert_element_type_1023);  convert_element_type_1022 = convert_element_type_1023 = None
	        mul_16529: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7805, view_2670);  sub_7805 = view_2670 = None
	        _assert_tensor_metadata_1537 = torch.ops.aten._assert_tensor_metadata.default(mul_16529, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1537 = None
	        view_2673: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2674: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2675: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1538 = torch.ops.aten._assert_tensor_metadata.default(view_2673, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1538 = None
	        convert_element_type_1024: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2673, torch.float32);  view_2673 = None
	        _assert_tensor_metadata_1539 = torch.ops.aten._assert_tensor_metadata.default(view_2675, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1539 = None
	        convert_element_type_1025: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2675, torch.float32);  view_2675 = None
	        sub_7809: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1024, convert_element_type_1025);  convert_element_type_1024 = convert_element_type_1025 = None
	        mul_16534: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7809, view_2674);  sub_7809 = view_2674 = None
	        view_2676: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16534, [1280, 1280]);  mul_16534 = None
	        _assert_tensor_metadata_1540 = torch.ops.aten._assert_tensor_metadata.default(view_2676, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1540 = None
	        mul_16539: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2677: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16529, [mul_16539, 1280]);  mul_16529 = mul_16539 = None
	        permute_285: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2676, [1, 0]);  view_2676 = None
	        addmm_141: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_28_self_attn_v_proj_bias, view_2677, permute_285);  model_audio_tower_layers_28_self_attn_v_proj_bias = view_2677 = permute_285 = None
	        view_2678: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_141, [sym_size_int, 1500, 1280]);  addmm_141 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2679: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2678, [sym_size_int, -1, 20, 64]);  view_2678 = None
	        permute_286: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2679, [0, 2, 1, 3]);  view_2679 = None
	        clone_228: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_286, memory_format = torch.contiguous_format);  permute_286 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_28 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_226, clone_227, clone_228, None, False, scale = 1.0);  clone_226 = clone_227 = clone_228 = None
	        getitem_226: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_28[0];  _scaled_dot_product_efficient_attention_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_287: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_226, [0, 2, 1, 3]);  getitem_226 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2680: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_287, [sym_size_int, 1500, -1]);  permute_287 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2680, [2])
	        amax_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2680, [2])
	        full_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_171, full_342);  amin_171 = full_342 = None
	        full_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_171, full_343);  amax_171 = full_343 = None
	        sub_7827: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_171, minimum_171);  maximum_171 = None
	        div_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7827, 255.0);  sub_7827 = None
	        clamp_min_513: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_342, 1.1920928955078125e-07);  div_342 = None
	        div_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_171, clamp_min_513);  minimum_171 = None
	        round_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_343);  div_343 = None
	        sub_7833: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_343);  round_343 = None
	        clamp_min_514: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7833, -128);  sub_7833 = None
	        clamp_max_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_514, 127);  clamp_min_514 = None
	        _assert_tensor_metadata_1541 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_513, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1541 = None
	        _assert_tensor_metadata_1542 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_342, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1542 = None
	        convert_element_type_1026: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_342, torch.int8);  clamp_max_342 = None
	        view_2683: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_513, [sym_size_int, 1500, 1])
	        view_2684: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1026, [sym_size_int, 1500, 1])
	        reciprocal_171: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2683);  view_2683 = None
	        mul_16609: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_171, 1.0);  reciprocal_171 = None
	        mul_16612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2680, mul_16609);  view_2680 = mul_16609 = None
	        round_344: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16612);  mul_16612 = None
	        add_26287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_344, view_2684);  round_344 = view_2684 = None
	        clamp_min_515: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26287, -128);  add_26287 = None
	        clamp_max_343: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_515, 127);  clamp_min_515 = None
	        _assert_tensor_metadata_1543 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_343, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1543 = None
	        convert_element_type_1027: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_343, torch.int8);  clamp_max_343 = None
	        view_2687: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_513, [sym_size_int, 1500, 1]);  clamp_min_513 = None
	        view_2688: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1026, [sym_size_int, 1500, 1]);  convert_element_type_1026 = None
	        _assert_tensor_metadata_1544 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1027, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1544 = None
	        convert_element_type_1028: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1027, torch.float32);  convert_element_type_1027 = None
	        _assert_tensor_metadata_1545 = torch.ops.aten._assert_tensor_metadata.default(view_2688, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1545 = None
	        convert_element_type_1029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2688, torch.float32);  view_2688 = None
	        sub_7853: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1028, convert_element_type_1029);  convert_element_type_1028 = convert_element_type_1029 = None
	        mul_16634: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7853, view_2687);  sub_7853 = view_2687 = None
	        _assert_tensor_metadata_1546 = torch.ops.aten._assert_tensor_metadata.default(mul_16634, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1546 = None
	        view_2690: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2691: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2692: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1547 = torch.ops.aten._assert_tensor_metadata.default(view_2690, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1547 = None
	        convert_element_type_1030: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2690, torch.float32);  view_2690 = None
	        _assert_tensor_metadata_1548 = torch.ops.aten._assert_tensor_metadata.default(view_2692, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1548 = None
	        convert_element_type_1031: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2692, torch.float32);  view_2692 = None
	        sub_7857: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1030, convert_element_type_1031);  convert_element_type_1030 = convert_element_type_1031 = None
	        mul_16639: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7857, view_2691);  sub_7857 = view_2691 = None
	        view_2693: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16639, [1280, 1280]);  mul_16639 = None
	        _assert_tensor_metadata_1549 = torch.ops.aten._assert_tensor_metadata.default(view_2693, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1549 = None
	        mul_16644: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2694: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16634, [mul_16644, 1280]);  mul_16634 = mul_16644 = None
	        permute_288: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2693, [1, 0]);  view_2693 = None
	        addmm_142: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_28_self_attn_out_proj_bias, view_2694, permute_288);  model_audio_tower_layers_28_self_attn_out_proj_bias = view_2694 = permute_288 = None
	        view_2695: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_142, [sym_size_int, 1500, 1280]);  addmm_142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_26350: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_25730, view_2695);  add_25730 = view_2695 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_26350, memory_format = torch.contiguous_format)
	        var_mean_57 = torch.ops.aten.var_mean.correction(clone_230, [2], correction = 0, keepdim = True)
	        getitem_230: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_57[0]
	        getitem_231: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_57[1];  var_mean_57 = None
	        add_26355: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_230, 1e-05);  getitem_230 = None
	        rsqrt_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_26355);  add_26355 = None
	        sub_7863: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_230, getitem_231);  clone_230 = getitem_231 = None
	        mul_16655: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7863, rsqrt_57);  sub_7863 = rsqrt_57 = None
	        mul_16656: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16655, model_audio_tower_layers_28_final_layer_norm_weight);  mul_16655 = model_audio_tower_layers_28_final_layer_norm_weight = None
	        add_26356: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16656, model_audio_tower_layers_28_final_layer_norm_bias);  mul_16656 = model_audio_tower_layers_28_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_26356, [2])
	        amax_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_26356, [2])
	        full_344: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_172, full_344);  amin_172 = full_344 = None
	        full_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_172, full_345);  amax_172 = full_345 = None
	        sub_7874: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_172, minimum_172);  maximum_172 = None
	        div_344: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7874, 255.0);  sub_7874 = None
	        clamp_min_516: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_344, 1.1920928955078125e-07);  div_344 = None
	        div_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_172, clamp_min_516);  minimum_172 = None
	        round_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_345);  div_345 = None
	        sub_7880: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_345);  round_345 = None
	        clamp_min_517: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7880, -128);  sub_7880 = None
	        clamp_max_344: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_517, 127);  clamp_min_517 = None
	        _assert_tensor_metadata_1550 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_516, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1550 = None
	        _assert_tensor_metadata_1551 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_344, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1551 = None
	        convert_element_type_1032: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_344, torch.int8);  clamp_max_344 = None
	        view_2698: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_516, [sym_size_int, 1500, 1])
	        view_2699: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1032, [sym_size_int, 1500, 1])
	        reciprocal_172: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2698);  view_2698 = None
	        mul_16704: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_172, 1.0);  reciprocal_172 = None
	        mul_16707: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_26356, mul_16704);  add_26356 = mul_16704 = None
	        round_346: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16707);  mul_16707 = None
	        add_26443: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_346, view_2699);  round_346 = view_2699 = None
	        clamp_min_518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26443, -128);  add_26443 = None
	        clamp_max_345: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_518, 127);  clamp_min_518 = None
	        _assert_tensor_metadata_1552 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_345, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1552 = None
	        convert_element_type_1033: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_345, torch.int8);  clamp_max_345 = None
	        view_2702: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_516, [sym_size_int, 1500, 1]);  clamp_min_516 = None
	        view_2703: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1032, [sym_size_int, 1500, 1]);  convert_element_type_1032 = None
	        _assert_tensor_metadata_1553 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1033, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1553 = None
	        convert_element_type_1034: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1033, torch.float32);  convert_element_type_1033 = None
	        _assert_tensor_metadata_1554 = torch.ops.aten._assert_tensor_metadata.default(view_2703, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1554 = None
	        convert_element_type_1035: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2703, torch.float32);  view_2703 = None
	        sub_7900: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1034, convert_element_type_1035);  convert_element_type_1034 = convert_element_type_1035 = None
	        mul_16729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7900, view_2702);  sub_7900 = view_2702 = None
	        _assert_tensor_metadata_1555 = torch.ops.aten._assert_tensor_metadata.default(mul_16729, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1555 = None
	        view_2705: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_28_fc1_parametrizations_weight_original0 = None
	        view_2706: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_28_fc1_parametrizations_weight_original1 = None
	        view_2707: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_28_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1556 = torch.ops.aten._assert_tensor_metadata.default(view_2705, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1556 = None
	        convert_element_type_1036: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2705, torch.float32);  view_2705 = None
	        _assert_tensor_metadata_1557 = torch.ops.aten._assert_tensor_metadata.default(view_2707, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1557 = None
	        convert_element_type_1037: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2707, torch.float32);  view_2707 = None
	        sub_7904: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1036, convert_element_type_1037);  convert_element_type_1036 = convert_element_type_1037 = None
	        mul_16734: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7904, view_2706);  sub_7904 = view_2706 = None
	        view_2708: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16734, [5120, 1280]);  mul_16734 = None
	        _assert_tensor_metadata_1558 = torch.ops.aten._assert_tensor_metadata.default(view_2708, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1558 = None
	        mul_16739: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2709: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16729, [mul_16739, 1280]);  mul_16729 = mul_16739 = None
	        permute_289: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2708, [1, 0]);  view_2708 = None
	        addmm_143: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_28_fc1_bias, view_2709, permute_289);  model_audio_tower_layers_28_fc1_bias = view_2709 = permute_289 = None
	        view_2710: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_143, [sym_size_int, 1500, 5120]);  addmm_143 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_16746: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2710, 0.5)
	        mul_16747: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2710, 0.7071067811865476);  view_2710 = None
	        erf_30: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_16747);  mul_16747 = None
	        add_26502: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_30, 1);  erf_30 = None
	        mul_16748: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16746, add_26502);  mul_16746 = add_26502 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_16748, [2])
	        amax_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_16748, [2])
	        full_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_173, full_346);  amin_173 = full_346 = None
	        full_347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_173, full_347);  amax_173 = full_347 = None
	        sub_7917: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_173, minimum_173);  maximum_173 = None
	        div_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7917, 255.0);  sub_7917 = None
	        clamp_min_519: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_346, 1.1920928955078125e-07);  div_346 = None
	        div_347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_173, clamp_min_519);  minimum_173 = None
	        round_347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_347);  div_347 = None
	        sub_7923: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_347);  round_347 = None
	        clamp_min_520: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7923, -128);  sub_7923 = None
	        clamp_max_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_520, 127);  clamp_min_520 = None
	        _assert_tensor_metadata_1559 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_519, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1559 = None
	        _assert_tensor_metadata_1560 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_346, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1560 = None
	        convert_element_type_1038: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_346, torch.int8);  clamp_max_346 = None
	        view_2713: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_519, [sym_size_int, 1500, 1])
	        view_2714: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1038, [sym_size_int, 1500, 1])
	        reciprocal_173: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2713);  view_2713 = None
	        mul_16794: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_173, 1.0);  reciprocal_173 = None
	        mul_16797: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16748, mul_16794);  mul_16748 = mul_16794 = None
	        round_348: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_16797);  mul_16797 = None
	        add_26585: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_348, view_2714);  round_348 = view_2714 = None
	        clamp_min_521: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26585, -128);  add_26585 = None
	        clamp_max_347: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_521, 127);  clamp_min_521 = None
	        _assert_tensor_metadata_1561 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_347, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1561 = None
	        convert_element_type_1039: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_347, torch.int8);  clamp_max_347 = None
	        view_2717: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_519, [sym_size_int, 1500, 1]);  clamp_min_519 = None
	        view_2718: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1038, [sym_size_int, 1500, 1]);  convert_element_type_1038 = None
	        _assert_tensor_metadata_1562 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1039, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1562 = None
	        convert_element_type_1040: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1039, torch.float32);  convert_element_type_1039 = None
	        _assert_tensor_metadata_1563 = torch.ops.aten._assert_tensor_metadata.default(view_2718, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1563 = None
	        convert_element_type_1041: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2718, torch.float32);  view_2718 = None
	        sub_7943: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1040, convert_element_type_1041);  convert_element_type_1040 = convert_element_type_1041 = None
	        mul_16819: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7943, view_2717);  sub_7943 = view_2717 = None
	        _assert_tensor_metadata_1564 = torch.ops.aten._assert_tensor_metadata.default(mul_16819, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1564 = None
	        view_2720: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_28_fc2_parametrizations_weight_original0 = None
	        view_2721: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_28_fc2_parametrizations_weight_original1 = None
	        view_2722: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_28_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1565 = torch.ops.aten._assert_tensor_metadata.default(view_2720, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1565 = None
	        convert_element_type_1042: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2720, torch.float32);  view_2720 = None
	        _assert_tensor_metadata_1566 = torch.ops.aten._assert_tensor_metadata.default(view_2722, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1566 = None
	        convert_element_type_1043: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2722, torch.float32);  view_2722 = None
	        sub_7947: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1042, convert_element_type_1043);  convert_element_type_1042 = convert_element_type_1043 = None
	        mul_16824: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7947, view_2721);  sub_7947 = view_2721 = None
	        view_2723: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16824, [1280, 5120]);  mul_16824 = None
	        _assert_tensor_metadata_1567 = torch.ops.aten._assert_tensor_metadata.default(view_2723, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1567 = None
	        mul_16829: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2724: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16819, [mul_16829, 5120]);  mul_16819 = mul_16829 = None
	        permute_290: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2723, [1, 0]);  view_2723 = None
	        addmm_144: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_28_fc2_bias, view_2724, permute_290);  model_audio_tower_layers_28_fc2_bias = view_2724 = permute_290 = None
	        view_2725: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_144, [sym_size_int, 1500, 1280]);  addmm_144 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_26648: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_26350, view_2725);  add_26350 = view_2725 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_233: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_26648, memory_format = torch.contiguous_format)
	        var_mean_58 = torch.ops.aten.var_mean.correction(clone_233, [2], correction = 0, keepdim = True)
	        getitem_232: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_58[0]
	        getitem_233: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_58[1];  var_mean_58 = None
	        add_26653: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_232, 1e-05);  getitem_232 = None
	        rsqrt_58: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_26653);  add_26653 = None
	        sub_7953: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_233, getitem_233);  clone_233 = getitem_233 = None
	        mul_16840: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7953, rsqrt_58);  sub_7953 = rsqrt_58 = None
	        mul_16841: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16840, model_audio_tower_layers_29_self_attn_layer_norm_weight);  mul_16840 = model_audio_tower_layers_29_self_attn_layer_norm_weight = None
	        add_26654: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16841, model_audio_tower_layers_29_self_attn_layer_norm_bias);  mul_16841 = model_audio_tower_layers_29_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_26654, [2])
	        amax_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_26654, [2])
	        full_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_174, full_348);  amin_174 = full_348 = None
	        full_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_174, full_349);  amax_174 = full_349 = None
	        sub_7964: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_174, minimum_174);  maximum_174 = None
	        div_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7964, 255.0);  sub_7964 = None
	        clamp_min_522: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_348, 1.1920928955078125e-07);  div_348 = None
	        div_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_174, clamp_min_522);  minimum_174 = None
	        round_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_349);  div_349 = None
	        sub_7970: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_349);  round_349 = None
	        clamp_min_523: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7970, -128);  sub_7970 = None
	        clamp_max_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_523, 127);  clamp_min_523 = None
	        _assert_tensor_metadata_1568 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_522, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1568 = None
	        _assert_tensor_metadata_1569 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_348, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1569 = None
	        convert_element_type_1044: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_348, torch.int8);  clamp_max_348 = None
	        view_2728: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_522, [sym_size_int, 1500, 1])
	        view_2729: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1044, [sym_size_int, 1500, 1])
	        reciprocal_174: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2728);  view_2728 = None
	        mul_16889: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_174, 1.0);  reciprocal_174 = None
	        mul_16892: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_26654, mul_16889);  mul_16889 = None
	        round_350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16892);  mul_16892 = None
	        add_26741: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_350, view_2729);  round_350 = view_2729 = None
	        clamp_min_524: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26741, -128);  add_26741 = None
	        clamp_max_349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_524, 127);  clamp_min_524 = None
	        _assert_tensor_metadata_1570 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_349, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1570 = None
	        convert_element_type_1045: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_349, torch.int8);  clamp_max_349 = None
	        view_2732: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_522, [sym_size_int, 1500, 1]);  clamp_min_522 = None
	        view_2733: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1044, [sym_size_int, 1500, 1]);  convert_element_type_1044 = None
	        _assert_tensor_metadata_1571 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1045, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1571 = None
	        convert_element_type_1046: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1045, torch.float32);  convert_element_type_1045 = None
	        _assert_tensor_metadata_1572 = torch.ops.aten._assert_tensor_metadata.default(view_2733, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1572 = None
	        convert_element_type_1047: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2733, torch.float32);  view_2733 = None
	        sub_7990: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1046, convert_element_type_1047);  convert_element_type_1046 = convert_element_type_1047 = None
	        mul_16914: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7990, view_2732);  sub_7990 = view_2732 = None
	        _assert_tensor_metadata_1573 = torch.ops.aten._assert_tensor_metadata.default(mul_16914, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1573 = None
	        view_2735: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2736: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2737: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1574 = torch.ops.aten._assert_tensor_metadata.default(view_2735, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1574 = None
	        convert_element_type_1048: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2735, torch.float32);  view_2735 = None
	        _assert_tensor_metadata_1575 = torch.ops.aten._assert_tensor_metadata.default(view_2737, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1575 = None
	        convert_element_type_1049: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2737, torch.float32);  view_2737 = None
	        sub_7994: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1048, convert_element_type_1049);  convert_element_type_1048 = convert_element_type_1049 = None
	        mul_16919: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7994, view_2736);  sub_7994 = view_2736 = None
	        view_2738: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16919, [1280, 1280]);  mul_16919 = None
	        _assert_tensor_metadata_1576 = torch.ops.aten._assert_tensor_metadata.default(view_2738, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1576 = None
	        mul_16924: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2739: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16914, [mul_16924, 1280]);  mul_16914 = mul_16924 = None
	        permute_291: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2738, [1, 0]);  view_2738 = None
	        addmm_145: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_29_self_attn_q_proj_bias, view_2739, permute_291);  model_audio_tower_layers_29_self_attn_q_proj_bias = view_2739 = permute_291 = None
	        view_2740: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_145, [sym_size_int, 1500, 1280]);  addmm_145 = None
	        mul_16931: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2740, 0.125);  view_2740 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2741: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16931, [sym_size_int, 1500, 20, 64]);  mul_16931 = None
	        permute_292: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2741, [0, 2, 1, 3]);  view_2741 = None
	        clone_234: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_292, memory_format = torch.contiguous_format);  permute_292 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_26654, [2])
	        amax_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_26654, [2])
	        full_350: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_175, full_350);  amin_175 = full_350 = None
	        full_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_175, full_351);  amax_175 = full_351 = None
	        sub_8009: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_175, minimum_175);  maximum_175 = None
	        div_350: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8009, 255.0);  sub_8009 = None
	        clamp_min_525: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_350, 1.1920928955078125e-07);  div_350 = None
	        div_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_175, clamp_min_525);  minimum_175 = None
	        round_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_351);  div_351 = None
	        sub_8015: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_351);  round_351 = None
	        clamp_min_526: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8015, -128);  sub_8015 = None
	        clamp_max_350: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_526, 127);  clamp_min_526 = None
	        _assert_tensor_metadata_1577 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_525, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1577 = None
	        _assert_tensor_metadata_1578 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_350, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1578 = None
	        convert_element_type_1050: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_350, torch.int8);  clamp_max_350 = None
	        view_2744: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_525, [sym_size_int, 1500, 1])
	        view_2745: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1050, [sym_size_int, 1500, 1])
	        reciprocal_175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2744);  view_2744 = None
	        mul_16985: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_175, 1.0);  reciprocal_175 = None
	        mul_16988: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_26654, mul_16985);  mul_16985 = None
	        round_352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16988);  mul_16988 = None
	        add_26893: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_352, view_2745);  round_352 = view_2745 = None
	        clamp_min_527: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26893, -128);  add_26893 = None
	        clamp_max_351: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_527, 127);  clamp_min_527 = None
	        _assert_tensor_metadata_1579 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_351, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1579 = None
	        convert_element_type_1051: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_351, torch.int8);  clamp_max_351 = None
	        view_2748: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_525, [sym_size_int, 1500, 1]);  clamp_min_525 = None
	        view_2749: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1050, [sym_size_int, 1500, 1]);  convert_element_type_1050 = None
	        _assert_tensor_metadata_1580 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1051, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1580 = None
	        convert_element_type_1052: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1051, torch.float32);  convert_element_type_1051 = None
	        _assert_tensor_metadata_1581 = torch.ops.aten._assert_tensor_metadata.default(view_2749, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1581 = None
	        convert_element_type_1053: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2749, torch.float32);  view_2749 = None
	        sub_8035: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1052, convert_element_type_1053);  convert_element_type_1052 = convert_element_type_1053 = None
	        mul_17010: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8035, view_2748);  sub_8035 = view_2748 = None
	        _assert_tensor_metadata_1582 = torch.ops.aten._assert_tensor_metadata.default(mul_17010, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1582 = None
	        view_2751: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2752: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2753: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1583 = torch.ops.aten._assert_tensor_metadata.default(view_2751, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1583 = None
	        convert_element_type_1054: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2751, torch.float32);  view_2751 = None
	        _assert_tensor_metadata_1584 = torch.ops.aten._assert_tensor_metadata.default(view_2753, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1584 = None
	        convert_element_type_1055: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2753, torch.float32);  view_2753 = None
	        sub_8039: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1054, convert_element_type_1055);  convert_element_type_1054 = convert_element_type_1055 = None
	        mul_17015: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8039, view_2752);  sub_8039 = view_2752 = None
	        view_2754: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17015, [1280, 1280]);  mul_17015 = None
	        _assert_tensor_metadata_1585 = torch.ops.aten._assert_tensor_metadata.default(view_2754, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1585 = None
	        permute_293: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2754, [1, 0]);  view_2754 = None
	        mul_17018: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2755: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17010, [mul_17018, 1280]);  mul_17010 = mul_17018 = None
	        mm_29: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2755, permute_293);  view_2755 = permute_293 = None
	        view_2756: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_29, [sym_size_int, 1500, 1280]);  mm_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2757: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2756, [sym_size_int, -1, 20, 64]);  view_2756 = None
	        permute_294: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2757, [0, 2, 1, 3]);  view_2757 = None
	        clone_235: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_294, memory_format = torch.contiguous_format);  permute_294 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_26654, [2])
	        amax_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_26654, [2])
	        full_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_176, full_352);  amin_176 = full_352 = None
	        full_353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_176, full_353);  amax_176 = full_353 = None
	        sub_8053: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_176, minimum_176);  maximum_176 = None
	        div_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8053, 255.0);  sub_8053 = None
	        clamp_min_528: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_352, 1.1920928955078125e-07);  div_352 = None
	        div_353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_176, clamp_min_528);  minimum_176 = None
	        round_353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_353);  div_353 = None
	        sub_8059: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_353);  round_353 = None
	        clamp_min_529: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8059, -128);  sub_8059 = None
	        clamp_max_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_529, 127);  clamp_min_529 = None
	        _assert_tensor_metadata_1586 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_528, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1586 = None
	        _assert_tensor_metadata_1587 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_352, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1587 = None
	        convert_element_type_1056: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_352, torch.int8);  clamp_max_352 = None
	        view_2760: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_528, [sym_size_int, 1500, 1])
	        view_2761: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1056, [sym_size_int, 1500, 1])
	        reciprocal_176: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2760);  view_2760 = None
	        mul_17084: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_176, 1.0);  reciprocal_176 = None
	        mul_17087: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_26654, mul_17084);  add_26654 = mul_17084 = None
	        round_354: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17087);  mul_17087 = None
	        add_27041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_354, view_2761);  round_354 = view_2761 = None
	        clamp_min_530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27041, -128);  add_27041 = None
	        clamp_max_353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_530, 127);  clamp_min_530 = None
	        _assert_tensor_metadata_1588 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_353, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1588 = None
	        convert_element_type_1057: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_353, torch.int8);  clamp_max_353 = None
	        view_2764: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_528, [sym_size_int, 1500, 1]);  clamp_min_528 = None
	        view_2765: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1056, [sym_size_int, 1500, 1]);  convert_element_type_1056 = None
	        _assert_tensor_metadata_1589 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1057, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1589 = None
	        convert_element_type_1058: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1057, torch.float32);  convert_element_type_1057 = None
	        _assert_tensor_metadata_1590 = torch.ops.aten._assert_tensor_metadata.default(view_2765, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1590 = None
	        convert_element_type_1059: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2765, torch.float32);  view_2765 = None
	        sub_8079: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1058, convert_element_type_1059);  convert_element_type_1058 = convert_element_type_1059 = None
	        mul_17109: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8079, view_2764);  sub_8079 = view_2764 = None
	        _assert_tensor_metadata_1591 = torch.ops.aten._assert_tensor_metadata.default(mul_17109, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1591 = None
	        view_2767: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2768: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2769: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1592 = torch.ops.aten._assert_tensor_metadata.default(view_2767, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1592 = None
	        convert_element_type_1060: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2767, torch.float32);  view_2767 = None
	        _assert_tensor_metadata_1593 = torch.ops.aten._assert_tensor_metadata.default(view_2769, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1593 = None
	        convert_element_type_1061: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2769, torch.float32);  view_2769 = None
	        sub_8083: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1060, convert_element_type_1061);  convert_element_type_1060 = convert_element_type_1061 = None
	        mul_17114: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8083, view_2768);  sub_8083 = view_2768 = None
	        view_2770: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17114, [1280, 1280]);  mul_17114 = None
	        _assert_tensor_metadata_1594 = torch.ops.aten._assert_tensor_metadata.default(view_2770, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1594 = None
	        mul_17119: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2771: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17109, [mul_17119, 1280]);  mul_17109 = mul_17119 = None
	        permute_295: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2770, [1, 0]);  view_2770 = None
	        addmm_146: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_29_self_attn_v_proj_bias, view_2771, permute_295);  model_audio_tower_layers_29_self_attn_v_proj_bias = view_2771 = permute_295 = None
	        view_2772: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_146, [sym_size_int, 1500, 1280]);  addmm_146 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2773: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2772, [sym_size_int, -1, 20, 64]);  view_2772 = None
	        permute_296: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2773, [0, 2, 1, 3]);  view_2773 = None
	        clone_236: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_296, memory_format = torch.contiguous_format);  permute_296 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_29 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_234, clone_235, clone_236, None, False, scale = 1.0);  clone_234 = clone_235 = clone_236 = None
	        getitem_234: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_29[0];  _scaled_dot_product_efficient_attention_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_297: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_234, [0, 2, 1, 3]);  getitem_234 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2774: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_297, [sym_size_int, 1500, -1]);  permute_297 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2774, [2])
	        amax_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2774, [2])
	        full_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_177, full_354);  amin_177 = full_354 = None
	        full_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_177, full_355);  amax_177 = full_355 = None
	        sub_8101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_177, minimum_177);  maximum_177 = None
	        div_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8101, 255.0);  sub_8101 = None
	        clamp_min_531: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_354, 1.1920928955078125e-07);  div_354 = None
	        div_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_177, clamp_min_531);  minimum_177 = None
	        round_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_355);  div_355 = None
	        sub_8107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_355);  round_355 = None
	        clamp_min_532: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8107, -128);  sub_8107 = None
	        clamp_max_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_532, 127);  clamp_min_532 = None
	        _assert_tensor_metadata_1595 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_531, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1595 = None
	        _assert_tensor_metadata_1596 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_354, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1596 = None
	        convert_element_type_1062: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_354, torch.int8);  clamp_max_354 = None
	        view_2777: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_531, [sym_size_int, 1500, 1])
	        view_2778: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1062, [sym_size_int, 1500, 1])
	        reciprocal_177: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2777);  view_2777 = None
	        mul_17189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_177, 1.0);  reciprocal_177 = None
	        mul_17192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2774, mul_17189);  view_2774 = mul_17189 = None
	        round_356: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17192);  mul_17192 = None
	        add_27205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_356, view_2778);  round_356 = view_2778 = None
	        clamp_min_533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27205, -128);  add_27205 = None
	        clamp_max_355: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_533, 127);  clamp_min_533 = None
	        _assert_tensor_metadata_1597 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_355, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1597 = None
	        convert_element_type_1063: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_355, torch.int8);  clamp_max_355 = None
	        view_2781: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_531, [sym_size_int, 1500, 1]);  clamp_min_531 = None
	        view_2782: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1062, [sym_size_int, 1500, 1]);  convert_element_type_1062 = None
	        _assert_tensor_metadata_1598 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1063, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1598 = None
	        convert_element_type_1064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1063, torch.float32);  convert_element_type_1063 = None
	        _assert_tensor_metadata_1599 = torch.ops.aten._assert_tensor_metadata.default(view_2782, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1599 = None
	        convert_element_type_1065: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2782, torch.float32);  view_2782 = None
	        sub_8127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1064, convert_element_type_1065);  convert_element_type_1064 = convert_element_type_1065 = None
	        mul_17214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8127, view_2781);  sub_8127 = view_2781 = None
	        _assert_tensor_metadata_1600 = torch.ops.aten._assert_tensor_metadata.default(mul_17214, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1600 = None
	        view_2784: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2785: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2786: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1601 = torch.ops.aten._assert_tensor_metadata.default(view_2784, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1601 = None
	        convert_element_type_1066: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2784, torch.float32);  view_2784 = None
	        _assert_tensor_metadata_1602 = torch.ops.aten._assert_tensor_metadata.default(view_2786, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1602 = None
	        convert_element_type_1067: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2786, torch.float32);  view_2786 = None
	        sub_8131: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1066, convert_element_type_1067);  convert_element_type_1066 = convert_element_type_1067 = None
	        mul_17219: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8131, view_2785);  sub_8131 = view_2785 = None
	        view_2787: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17219, [1280, 1280]);  mul_17219 = None
	        _assert_tensor_metadata_1603 = torch.ops.aten._assert_tensor_metadata.default(view_2787, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1603 = None
	        mul_17224: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2788: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17214, [mul_17224, 1280]);  mul_17214 = mul_17224 = None
	        permute_298: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2787, [1, 0]);  view_2787 = None
	        addmm_147: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_29_self_attn_out_proj_bias, view_2788, permute_298);  model_audio_tower_layers_29_self_attn_out_proj_bias = view_2788 = permute_298 = None
	        view_2789: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_147, [sym_size_int, 1500, 1280]);  addmm_147 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_27268: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_26648, view_2789);  add_26648 = view_2789 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_238: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_27268, memory_format = torch.contiguous_format)
	        var_mean_59 = torch.ops.aten.var_mean.correction(clone_238, [2], correction = 0, keepdim = True)
	        getitem_238: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_59[0]
	        getitem_239: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_59[1];  var_mean_59 = None
	        add_27273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_238, 1e-05);  getitem_238 = None
	        rsqrt_59: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_27273);  add_27273 = None
	        sub_8137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_238, getitem_239);  clone_238 = getitem_239 = None
	        mul_17235: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8137, rsqrt_59);  sub_8137 = rsqrt_59 = None
	        mul_17236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17235, model_audio_tower_layers_29_final_layer_norm_weight);  mul_17235 = model_audio_tower_layers_29_final_layer_norm_weight = None
	        add_27274: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_17236, model_audio_tower_layers_29_final_layer_norm_bias);  mul_17236 = model_audio_tower_layers_29_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_27274, [2])
	        amax_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_27274, [2])
	        full_356: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_178, full_356);  amin_178 = full_356 = None
	        full_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_178, full_357);  amax_178 = full_357 = None
	        sub_8148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_178, minimum_178);  maximum_178 = None
	        div_356: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8148, 255.0);  sub_8148 = None
	        clamp_min_534: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_356, 1.1920928955078125e-07);  div_356 = None
	        div_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_178, clamp_min_534);  minimum_178 = None
	        round_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_357);  div_357 = None
	        sub_8154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_357);  round_357 = None
	        clamp_min_535: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8154, -128);  sub_8154 = None
	        clamp_max_356: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_535, 127);  clamp_min_535 = None
	        _assert_tensor_metadata_1604 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_534, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1604 = None
	        _assert_tensor_metadata_1605 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_356, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1605 = None
	        convert_element_type_1068: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_356, torch.int8);  clamp_max_356 = None
	        view_2792: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_534, [sym_size_int, 1500, 1])
	        view_2793: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1068, [sym_size_int, 1500, 1])
	        reciprocal_178: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2792);  view_2792 = None
	        mul_17284: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_178, 1.0);  reciprocal_178 = None
	        mul_17287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_27274, mul_17284);  add_27274 = mul_17284 = None
	        round_358: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17287);  mul_17287 = None
	        add_27361: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_358, view_2793);  round_358 = view_2793 = None
	        clamp_min_536: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27361, -128);  add_27361 = None
	        clamp_max_357: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_536, 127);  clamp_min_536 = None
	        _assert_tensor_metadata_1606 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_357, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1606 = None
	        convert_element_type_1069: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_357, torch.int8);  clamp_max_357 = None
	        view_2796: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_534, [sym_size_int, 1500, 1]);  clamp_min_534 = None
	        view_2797: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1068, [sym_size_int, 1500, 1]);  convert_element_type_1068 = None
	        _assert_tensor_metadata_1607 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1069, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1607 = None
	        convert_element_type_1070: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1069, torch.float32);  convert_element_type_1069 = None
	        _assert_tensor_metadata_1608 = torch.ops.aten._assert_tensor_metadata.default(view_2797, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1608 = None
	        convert_element_type_1071: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2797, torch.float32);  view_2797 = None
	        sub_8174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1070, convert_element_type_1071);  convert_element_type_1070 = convert_element_type_1071 = None
	        mul_17309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8174, view_2796);  sub_8174 = view_2796 = None
	        _assert_tensor_metadata_1609 = torch.ops.aten._assert_tensor_metadata.default(mul_17309, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1609 = None
	        view_2799: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_29_fc1_parametrizations_weight_original0 = None
	        view_2800: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_29_fc1_parametrizations_weight_original1 = None
	        view_2801: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_29_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1610 = torch.ops.aten._assert_tensor_metadata.default(view_2799, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1610 = None
	        convert_element_type_1072: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2799, torch.float32);  view_2799 = None
	        _assert_tensor_metadata_1611 = torch.ops.aten._assert_tensor_metadata.default(view_2801, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1611 = None
	        convert_element_type_1073: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2801, torch.float32);  view_2801 = None
	        sub_8178: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1072, convert_element_type_1073);  convert_element_type_1072 = convert_element_type_1073 = None
	        mul_17314: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8178, view_2800);  sub_8178 = view_2800 = None
	        view_2802: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17314, [5120, 1280]);  mul_17314 = None
	        _assert_tensor_metadata_1612 = torch.ops.aten._assert_tensor_metadata.default(view_2802, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1612 = None
	        mul_17319: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2803: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17309, [mul_17319, 1280]);  mul_17309 = mul_17319 = None
	        permute_299: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2802, [1, 0]);  view_2802 = None
	        addmm_148: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_29_fc1_bias, view_2803, permute_299);  model_audio_tower_layers_29_fc1_bias = view_2803 = permute_299 = None
	        view_2804: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_148, [sym_size_int, 1500, 5120]);  addmm_148 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_17326: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2804, 0.5)
	        mul_17327: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2804, 0.7071067811865476);  view_2804 = None
	        erf_31: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_17327);  mul_17327 = None
	        add_27420: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_31, 1);  erf_31 = None
	        mul_17328: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17326, add_27420);  mul_17326 = add_27420 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_17328, [2])
	        amax_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_17328, [2])
	        full_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_179, full_358);  amin_179 = full_358 = None
	        full_359: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_179, full_359);  amax_179 = full_359 = None
	        sub_8191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_179, minimum_179);  maximum_179 = None
	        div_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8191, 255.0);  sub_8191 = None
	        clamp_min_537: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_358, 1.1920928955078125e-07);  div_358 = None
	        div_359: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_179, clamp_min_537);  minimum_179 = None
	        round_359: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_359);  div_359 = None
	        sub_8197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_359);  round_359 = None
	        clamp_min_538: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8197, -128);  sub_8197 = None
	        clamp_max_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_538, 127);  clamp_min_538 = None
	        _assert_tensor_metadata_1613 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_537, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1613 = None
	        _assert_tensor_metadata_1614 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_358, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1614 = None
	        convert_element_type_1074: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_358, torch.int8);  clamp_max_358 = None
	        view_2807: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_537, [sym_size_int, 1500, 1])
	        view_2808: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1074, [sym_size_int, 1500, 1])
	        reciprocal_179: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2807);  view_2807 = None
	        mul_17374: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_179, 1.0);  reciprocal_179 = None
	        mul_17377: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17328, mul_17374);  mul_17328 = mul_17374 = None
	        round_360: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_17377);  mul_17377 = None
	        add_27503: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_360, view_2808);  round_360 = view_2808 = None
	        clamp_min_539: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27503, -128);  add_27503 = None
	        clamp_max_359: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_539, 127);  clamp_min_539 = None
	        _assert_tensor_metadata_1615 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_359, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1615 = None
	        convert_element_type_1075: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_359, torch.int8);  clamp_max_359 = None
	        view_2811: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_537, [sym_size_int, 1500, 1]);  clamp_min_537 = None
	        view_2812: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1074, [sym_size_int, 1500, 1]);  convert_element_type_1074 = None
	        _assert_tensor_metadata_1616 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1075, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1616 = None
	        convert_element_type_1076: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1075, torch.float32);  convert_element_type_1075 = None
	        _assert_tensor_metadata_1617 = torch.ops.aten._assert_tensor_metadata.default(view_2812, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1617 = None
	        convert_element_type_1077: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2812, torch.float32);  view_2812 = None
	        sub_8217: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1076, convert_element_type_1077);  convert_element_type_1076 = convert_element_type_1077 = None
	        mul_17399: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8217, view_2811);  sub_8217 = view_2811 = None
	        _assert_tensor_metadata_1618 = torch.ops.aten._assert_tensor_metadata.default(mul_17399, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1618 = None
	        view_2814: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_29_fc2_parametrizations_weight_original0 = None
	        view_2815: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_29_fc2_parametrizations_weight_original1 = None
	        view_2816: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_29_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1619 = torch.ops.aten._assert_tensor_metadata.default(view_2814, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1619 = None
	        convert_element_type_1078: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2814, torch.float32);  view_2814 = None
	        _assert_tensor_metadata_1620 = torch.ops.aten._assert_tensor_metadata.default(view_2816, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1620 = None
	        convert_element_type_1079: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2816, torch.float32);  view_2816 = None
	        sub_8221: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1078, convert_element_type_1079);  convert_element_type_1078 = convert_element_type_1079 = None
	        mul_17404: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8221, view_2815);  sub_8221 = view_2815 = None
	        view_2817: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17404, [1280, 5120]);  mul_17404 = None
	        _assert_tensor_metadata_1621 = torch.ops.aten._assert_tensor_metadata.default(view_2817, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1621 = None
	        mul_17409: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2818: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17399, [mul_17409, 5120]);  mul_17399 = mul_17409 = None
	        permute_300: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2817, [1, 0]);  view_2817 = None
	        addmm_149: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_29_fc2_bias, view_2818, permute_300);  model_audio_tower_layers_29_fc2_bias = view_2818 = permute_300 = None
	        view_2819: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_149, [sym_size_int, 1500, 1280]);  addmm_149 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_27566: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_27268, view_2819);  add_27268 = view_2819 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_241: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_27566, memory_format = torch.contiguous_format)
	        var_mean_60 = torch.ops.aten.var_mean.correction(clone_241, [2], correction = 0, keepdim = True)
	        getitem_240: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_60[0]
	        getitem_241: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_60[1];  var_mean_60 = None
	        add_27571: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_240, 1e-05);  getitem_240 = None
	        rsqrt_60: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_27571);  add_27571 = None
	        sub_8227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_241, getitem_241);  clone_241 = getitem_241 = None
	        mul_17420: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8227, rsqrt_60);  sub_8227 = rsqrt_60 = None
	        mul_17421: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17420, model_audio_tower_layers_30_self_attn_layer_norm_weight);  mul_17420 = model_audio_tower_layers_30_self_attn_layer_norm_weight = None
	        add_27572: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_17421, model_audio_tower_layers_30_self_attn_layer_norm_bias);  mul_17421 = model_audio_tower_layers_30_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_27572, [2])
	        amax_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_27572, [2])
	        full_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_180, full_360);  amin_180 = full_360 = None
	        full_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_180, full_361);  amax_180 = full_361 = None
	        sub_8238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_180, minimum_180);  maximum_180 = None
	        div_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8238, 255.0);  sub_8238 = None
	        clamp_min_540: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_360, 1.1920928955078125e-07);  div_360 = None
	        div_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_180, clamp_min_540);  minimum_180 = None
	        round_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_361);  div_361 = None
	        sub_8244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_361);  round_361 = None
	        clamp_min_541: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8244, -128);  sub_8244 = None
	        clamp_max_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_541, 127);  clamp_min_541 = None
	        _assert_tensor_metadata_1622 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_540, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1622 = None
	        _assert_tensor_metadata_1623 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_360, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1623 = None
	        convert_element_type_1080: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_360, torch.int8);  clamp_max_360 = None
	        view_2822: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_540, [sym_size_int, 1500, 1])
	        view_2823: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1080, [sym_size_int, 1500, 1])
	        reciprocal_180: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2822);  view_2822 = None
	        mul_17469: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_180, 1.0);  reciprocal_180 = None
	        mul_17472: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_27572, mul_17469);  mul_17469 = None
	        round_362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17472);  mul_17472 = None
	        add_27659: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_362, view_2823);  round_362 = view_2823 = None
	        clamp_min_542: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27659, -128);  add_27659 = None
	        clamp_max_361: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_542, 127);  clamp_min_542 = None
	        _assert_tensor_metadata_1624 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_361, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1624 = None
	        convert_element_type_1081: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_361, torch.int8);  clamp_max_361 = None
	        view_2826: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_540, [sym_size_int, 1500, 1]);  clamp_min_540 = None
	        view_2827: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1080, [sym_size_int, 1500, 1]);  convert_element_type_1080 = None
	        _assert_tensor_metadata_1625 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1081, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1625 = None
	        convert_element_type_1082: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1081, torch.float32);  convert_element_type_1081 = None
	        _assert_tensor_metadata_1626 = torch.ops.aten._assert_tensor_metadata.default(view_2827, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1626 = None
	        convert_element_type_1083: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2827, torch.float32);  view_2827 = None
	        sub_8264: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1082, convert_element_type_1083);  convert_element_type_1082 = convert_element_type_1083 = None
	        mul_17494: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8264, view_2826);  sub_8264 = view_2826 = None
	        _assert_tensor_metadata_1627 = torch.ops.aten._assert_tensor_metadata.default(mul_17494, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1627 = None
	        view_2829: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2830: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2831: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1628 = torch.ops.aten._assert_tensor_metadata.default(view_2829, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1628 = None
	        convert_element_type_1084: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2829, torch.float32);  view_2829 = None
	        _assert_tensor_metadata_1629 = torch.ops.aten._assert_tensor_metadata.default(view_2831, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1629 = None
	        convert_element_type_1085: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2831, torch.float32);  view_2831 = None
	        sub_8268: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1084, convert_element_type_1085);  convert_element_type_1084 = convert_element_type_1085 = None
	        mul_17499: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8268, view_2830);  sub_8268 = view_2830 = None
	        view_2832: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17499, [1280, 1280]);  mul_17499 = None
	        _assert_tensor_metadata_1630 = torch.ops.aten._assert_tensor_metadata.default(view_2832, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1630 = None
	        mul_17504: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2833: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17494, [mul_17504, 1280]);  mul_17494 = mul_17504 = None
	        permute_301: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2832, [1, 0]);  view_2832 = None
	        addmm_150: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_30_self_attn_q_proj_bias, view_2833, permute_301);  model_audio_tower_layers_30_self_attn_q_proj_bias = view_2833 = permute_301 = None
	        view_2834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_150, [sym_size_int, 1500, 1280]);  addmm_150 = None
	        mul_17511: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2834, 0.125);  view_2834 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2835: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17511, [sym_size_int, 1500, 20, 64]);  mul_17511 = None
	        permute_302: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2835, [0, 2, 1, 3]);  view_2835 = None
	        clone_242: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_302, memory_format = torch.contiguous_format);  permute_302 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_27572, [2])
	        amax_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_27572, [2])
	        full_362: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_181, full_362);  amin_181 = full_362 = None
	        full_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_181, full_363);  amax_181 = full_363 = None
	        sub_8283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_181, minimum_181);  maximum_181 = None
	        div_362: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8283, 255.0);  sub_8283 = None
	        clamp_min_543: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_362, 1.1920928955078125e-07);  div_362 = None
	        div_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_181, clamp_min_543);  minimum_181 = None
	        round_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_363);  div_363 = None
	        sub_8289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_363);  round_363 = None
	        clamp_min_544: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8289, -128);  sub_8289 = None
	        clamp_max_362: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_544, 127);  clamp_min_544 = None
	        _assert_tensor_metadata_1631 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_543, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1631 = None
	        _assert_tensor_metadata_1632 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_362, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1632 = None
	        convert_element_type_1086: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_362, torch.int8);  clamp_max_362 = None
	        view_2838: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_543, [sym_size_int, 1500, 1])
	        view_2839: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1086, [sym_size_int, 1500, 1])
	        reciprocal_181: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2838);  view_2838 = None
	        mul_17565: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_181, 1.0);  reciprocal_181 = None
	        mul_17568: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_27572, mul_17565);  mul_17565 = None
	        round_364: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17568);  mul_17568 = None
	        add_27811: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_364, view_2839);  round_364 = view_2839 = None
	        clamp_min_545: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27811, -128);  add_27811 = None
	        clamp_max_363: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_545, 127);  clamp_min_545 = None
	        _assert_tensor_metadata_1633 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_363, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1633 = None
	        convert_element_type_1087: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_363, torch.int8);  clamp_max_363 = None
	        view_2842: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_543, [sym_size_int, 1500, 1]);  clamp_min_543 = None
	        view_2843: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1086, [sym_size_int, 1500, 1]);  convert_element_type_1086 = None
	        _assert_tensor_metadata_1634 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1087, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1634 = None
	        convert_element_type_1088: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1087, torch.float32);  convert_element_type_1087 = None
	        _assert_tensor_metadata_1635 = torch.ops.aten._assert_tensor_metadata.default(view_2843, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1635 = None
	        convert_element_type_1089: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2843, torch.float32);  view_2843 = None
	        sub_8309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1088, convert_element_type_1089);  convert_element_type_1088 = convert_element_type_1089 = None
	        mul_17590: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8309, view_2842);  sub_8309 = view_2842 = None
	        _assert_tensor_metadata_1636 = torch.ops.aten._assert_tensor_metadata.default(mul_17590, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1636 = None
	        view_2845: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2846: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2847: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1637 = torch.ops.aten._assert_tensor_metadata.default(view_2845, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1637 = None
	        convert_element_type_1090: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2845, torch.float32);  view_2845 = None
	        _assert_tensor_metadata_1638 = torch.ops.aten._assert_tensor_metadata.default(view_2847, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1638 = None
	        convert_element_type_1091: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2847, torch.float32);  view_2847 = None
	        sub_8313: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1090, convert_element_type_1091);  convert_element_type_1090 = convert_element_type_1091 = None
	        mul_17595: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8313, view_2846);  sub_8313 = view_2846 = None
	        view_2848: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17595, [1280, 1280]);  mul_17595 = None
	        _assert_tensor_metadata_1639 = torch.ops.aten._assert_tensor_metadata.default(view_2848, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1639 = None
	        permute_303: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2848, [1, 0]);  view_2848 = None
	        mul_17598: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2849: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17590, [mul_17598, 1280]);  mul_17590 = mul_17598 = None
	        mm_30: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2849, permute_303);  view_2849 = permute_303 = None
	        view_2850: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_30, [sym_size_int, 1500, 1280]);  mm_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2851: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2850, [sym_size_int, -1, 20, 64]);  view_2850 = None
	        permute_304: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2851, [0, 2, 1, 3]);  view_2851 = None
	        clone_243: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_304, memory_format = torch.contiguous_format);  permute_304 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_27572, [2])
	        amax_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_27572, [2])
	        full_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_182, full_364);  amin_182 = full_364 = None
	        full_365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_182, full_365);  amax_182 = full_365 = None
	        sub_8327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_182, minimum_182);  maximum_182 = None
	        div_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8327, 255.0);  sub_8327 = None
	        clamp_min_546: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_364, 1.1920928955078125e-07);  div_364 = None
	        div_365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_182, clamp_min_546);  minimum_182 = None
	        round_365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_365);  div_365 = None
	        sub_8333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_365);  round_365 = None
	        clamp_min_547: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8333, -128);  sub_8333 = None
	        clamp_max_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_547, 127);  clamp_min_547 = None
	        _assert_tensor_metadata_1640 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_546, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1640 = None
	        _assert_tensor_metadata_1641 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_364, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1641 = None
	        convert_element_type_1092: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_364, torch.int8);  clamp_max_364 = None
	        view_2854: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_546, [sym_size_int, 1500, 1])
	        view_2855: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1092, [sym_size_int, 1500, 1])
	        reciprocal_182: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2854);  view_2854 = None
	        mul_17664: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_182, 1.0);  reciprocal_182 = None
	        mul_17667: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_27572, mul_17664);  add_27572 = mul_17664 = None
	        round_366: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17667);  mul_17667 = None
	        add_27959: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_366, view_2855);  round_366 = view_2855 = None
	        clamp_min_548: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27959, -128);  add_27959 = None
	        clamp_max_365: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_548, 127);  clamp_min_548 = None
	        _assert_tensor_metadata_1642 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_365, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1642 = None
	        convert_element_type_1093: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_365, torch.int8);  clamp_max_365 = None
	        view_2858: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_546, [sym_size_int, 1500, 1]);  clamp_min_546 = None
	        view_2859: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1092, [sym_size_int, 1500, 1]);  convert_element_type_1092 = None
	        _assert_tensor_metadata_1643 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1093, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1643 = None
	        convert_element_type_1094: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1093, torch.float32);  convert_element_type_1093 = None
	        _assert_tensor_metadata_1644 = torch.ops.aten._assert_tensor_metadata.default(view_2859, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1644 = None
	        convert_element_type_1095: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2859, torch.float32);  view_2859 = None
	        sub_8353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1094, convert_element_type_1095);  convert_element_type_1094 = convert_element_type_1095 = None
	        mul_17689: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8353, view_2858);  sub_8353 = view_2858 = None
	        _assert_tensor_metadata_1645 = torch.ops.aten._assert_tensor_metadata.default(mul_17689, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1645 = None
	        view_2861: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2862: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2863: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1646 = torch.ops.aten._assert_tensor_metadata.default(view_2861, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1646 = None
	        convert_element_type_1096: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2861, torch.float32);  view_2861 = None
	        _assert_tensor_metadata_1647 = torch.ops.aten._assert_tensor_metadata.default(view_2863, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1647 = None
	        convert_element_type_1097: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2863, torch.float32);  view_2863 = None
	        sub_8357: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1096, convert_element_type_1097);  convert_element_type_1096 = convert_element_type_1097 = None
	        mul_17694: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8357, view_2862);  sub_8357 = view_2862 = None
	        view_2864: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17694, [1280, 1280]);  mul_17694 = None
	        _assert_tensor_metadata_1648 = torch.ops.aten._assert_tensor_metadata.default(view_2864, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1648 = None
	        mul_17699: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2865: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17689, [mul_17699, 1280]);  mul_17689 = mul_17699 = None
	        permute_305: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2864, [1, 0]);  view_2864 = None
	        addmm_151: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_30_self_attn_v_proj_bias, view_2865, permute_305);  model_audio_tower_layers_30_self_attn_v_proj_bias = view_2865 = permute_305 = None
	        view_2866: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_151, [sym_size_int, 1500, 1280]);  addmm_151 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2867: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2866, [sym_size_int, -1, 20, 64]);  view_2866 = None
	        permute_306: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2867, [0, 2, 1, 3]);  view_2867 = None
	        clone_244: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_306, memory_format = torch.contiguous_format);  permute_306 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_30 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_242, clone_243, clone_244, None, False, scale = 1.0);  clone_242 = clone_243 = clone_244 = None
	        getitem_242: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_30[0];  _scaled_dot_product_efficient_attention_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_307: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_242, [0, 2, 1, 3]);  getitem_242 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2868: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_307, [sym_size_int, 1500, -1]);  permute_307 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2868, [2])
	        amax_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2868, [2])
	        full_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_183, full_366);  amin_183 = full_366 = None
	        full_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_183, full_367);  amax_183 = full_367 = None
	        sub_8375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_183, minimum_183);  maximum_183 = None
	        div_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8375, 255.0);  sub_8375 = None
	        clamp_min_549: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_366, 1.1920928955078125e-07);  div_366 = None
	        div_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_183, clamp_min_549);  minimum_183 = None
	        round_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_367);  div_367 = None
	        sub_8381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_367);  round_367 = None
	        clamp_min_550: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8381, -128);  sub_8381 = None
	        clamp_max_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_550, 127);  clamp_min_550 = None
	        _assert_tensor_metadata_1649 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_549, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1649 = None
	        _assert_tensor_metadata_1650 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_366, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1650 = None
	        convert_element_type_1098: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_366, torch.int8);  clamp_max_366 = None
	        view_2871: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_549, [sym_size_int, 1500, 1])
	        view_2872: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1098, [sym_size_int, 1500, 1])
	        reciprocal_183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2871);  view_2871 = None
	        mul_17769: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_183, 1.0);  reciprocal_183 = None
	        mul_17772: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2868, mul_17769);  view_2868 = mul_17769 = None
	        round_368: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17772);  mul_17772 = None
	        add_28123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_368, view_2872);  round_368 = view_2872 = None
	        clamp_min_551: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28123, -128);  add_28123 = None
	        clamp_max_367: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_551, 127);  clamp_min_551 = None
	        _assert_tensor_metadata_1651 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_367, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1651 = None
	        convert_element_type_1099: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_367, torch.int8);  clamp_max_367 = None
	        view_2875: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_549, [sym_size_int, 1500, 1]);  clamp_min_549 = None
	        view_2876: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1098, [sym_size_int, 1500, 1]);  convert_element_type_1098 = None
	        _assert_tensor_metadata_1652 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1099, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1652 = None
	        convert_element_type_1100: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1099, torch.float32);  convert_element_type_1099 = None
	        _assert_tensor_metadata_1653 = torch.ops.aten._assert_tensor_metadata.default(view_2876, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1653 = None
	        convert_element_type_1101: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2876, torch.float32);  view_2876 = None
	        sub_8401: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1100, convert_element_type_1101);  convert_element_type_1100 = convert_element_type_1101 = None
	        mul_17794: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8401, view_2875);  sub_8401 = view_2875 = None
	        _assert_tensor_metadata_1654 = torch.ops.aten._assert_tensor_metadata.default(mul_17794, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1654 = None
	        view_2878: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2879: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2880: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1655 = torch.ops.aten._assert_tensor_metadata.default(view_2878, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1655 = None
	        convert_element_type_1102: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2878, torch.float32);  view_2878 = None
	        _assert_tensor_metadata_1656 = torch.ops.aten._assert_tensor_metadata.default(view_2880, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1656 = None
	        convert_element_type_1103: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2880, torch.float32);  view_2880 = None
	        sub_8405: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1102, convert_element_type_1103);  convert_element_type_1102 = convert_element_type_1103 = None
	        mul_17799: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8405, view_2879);  sub_8405 = view_2879 = None
	        view_2881: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17799, [1280, 1280]);  mul_17799 = None
	        _assert_tensor_metadata_1657 = torch.ops.aten._assert_tensor_metadata.default(view_2881, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1657 = None
	        mul_17804: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2882: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17794, [mul_17804, 1280]);  mul_17794 = mul_17804 = None
	        permute_308: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2881, [1, 0]);  view_2881 = None
	        addmm_152: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_30_self_attn_out_proj_bias, view_2882, permute_308);  model_audio_tower_layers_30_self_attn_out_proj_bias = view_2882 = permute_308 = None
	        view_2883: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_152, [sym_size_int, 1500, 1280]);  addmm_152 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_28186: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_27566, view_2883);  add_27566 = view_2883 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_246: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_28186, memory_format = torch.contiguous_format)
	        var_mean_61 = torch.ops.aten.var_mean.correction(clone_246, [2], correction = 0, keepdim = True)
	        getitem_246: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_61[0]
	        getitem_247: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_61[1];  var_mean_61 = None
	        add_28191: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_246, 1e-05);  getitem_246 = None
	        rsqrt_61: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_28191);  add_28191 = None
	        sub_8411: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_246, getitem_247);  clone_246 = getitem_247 = None
	        mul_17815: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8411, rsqrt_61);  sub_8411 = rsqrt_61 = None
	        mul_17816: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17815, model_audio_tower_layers_30_final_layer_norm_weight);  mul_17815 = model_audio_tower_layers_30_final_layer_norm_weight = None
	        add_28192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_17816, model_audio_tower_layers_30_final_layer_norm_bias);  mul_17816 = model_audio_tower_layers_30_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_28192, [2])
	        amax_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_28192, [2])
	        full_368: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_184, full_368);  amin_184 = full_368 = None
	        full_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_184, full_369);  amax_184 = full_369 = None
	        sub_8422: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_184, minimum_184);  maximum_184 = None
	        div_368: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8422, 255.0);  sub_8422 = None
	        clamp_min_552: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_368, 1.1920928955078125e-07);  div_368 = None
	        div_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_184, clamp_min_552);  minimum_184 = None
	        round_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_369);  div_369 = None
	        sub_8428: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_369);  round_369 = None
	        clamp_min_553: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8428, -128);  sub_8428 = None
	        clamp_max_368: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_553, 127);  clamp_min_553 = None
	        _assert_tensor_metadata_1658 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_552, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1658 = None
	        _assert_tensor_metadata_1659 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_368, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1659 = None
	        convert_element_type_1104: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_368, torch.int8);  clamp_max_368 = None
	        view_2886: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_552, [sym_size_int, 1500, 1])
	        view_2887: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1104, [sym_size_int, 1500, 1])
	        reciprocal_184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2886);  view_2886 = None
	        mul_17864: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_184, 1.0);  reciprocal_184 = None
	        mul_17867: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_28192, mul_17864);  add_28192 = mul_17864 = None
	        round_370: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17867);  mul_17867 = None
	        add_28279: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_370, view_2887);  round_370 = view_2887 = None
	        clamp_min_554: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28279, -128);  add_28279 = None
	        clamp_max_369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_554, 127);  clamp_min_554 = None
	        _assert_tensor_metadata_1660 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_369, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1660 = None
	        convert_element_type_1105: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_369, torch.int8);  clamp_max_369 = None
	        view_2890: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_552, [sym_size_int, 1500, 1]);  clamp_min_552 = None
	        view_2891: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1104, [sym_size_int, 1500, 1]);  convert_element_type_1104 = None
	        _assert_tensor_metadata_1661 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1105, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1661 = None
	        convert_element_type_1106: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1105, torch.float32);  convert_element_type_1105 = None
	        _assert_tensor_metadata_1662 = torch.ops.aten._assert_tensor_metadata.default(view_2891, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1662 = None
	        convert_element_type_1107: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2891, torch.float32);  view_2891 = None
	        sub_8448: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1106, convert_element_type_1107);  convert_element_type_1106 = convert_element_type_1107 = None
	        mul_17889: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8448, view_2890);  sub_8448 = view_2890 = None
	        _assert_tensor_metadata_1663 = torch.ops.aten._assert_tensor_metadata.default(mul_17889, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1663 = None
	        view_2893: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_30_fc1_parametrizations_weight_original0 = None
	        view_2894: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_30_fc1_parametrizations_weight_original1 = None
	        view_2895: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_30_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1664 = torch.ops.aten._assert_tensor_metadata.default(view_2893, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1664 = None
	        convert_element_type_1108: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2893, torch.float32);  view_2893 = None
	        _assert_tensor_metadata_1665 = torch.ops.aten._assert_tensor_metadata.default(view_2895, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1665 = None
	        convert_element_type_1109: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2895, torch.float32);  view_2895 = None
	        sub_8452: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1108, convert_element_type_1109);  convert_element_type_1108 = convert_element_type_1109 = None
	        mul_17894: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8452, view_2894);  sub_8452 = view_2894 = None
	        view_2896: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17894, [5120, 1280]);  mul_17894 = None
	        _assert_tensor_metadata_1666 = torch.ops.aten._assert_tensor_metadata.default(view_2896, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1666 = None
	        mul_17899: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2897: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17889, [mul_17899, 1280]);  mul_17889 = mul_17899 = None
	        permute_309: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2896, [1, 0]);  view_2896 = None
	        addmm_153: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_30_fc1_bias, view_2897, permute_309);  model_audio_tower_layers_30_fc1_bias = view_2897 = permute_309 = None
	        view_2898: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_153, [sym_size_int, 1500, 5120]);  addmm_153 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_17906: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2898, 0.5)
	        mul_17907: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2898, 0.7071067811865476);  view_2898 = None
	        erf_32: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_17907);  mul_17907 = None
	        add_28338: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_32, 1);  erf_32 = None
	        mul_17908: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17906, add_28338);  mul_17906 = add_28338 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_17908, [2])
	        amax_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_17908, [2])
	        full_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_185, full_370);  amin_185 = full_370 = None
	        full_371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_185, full_371);  amax_185 = full_371 = None
	        sub_8465: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_185, minimum_185);  maximum_185 = None
	        div_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8465, 255.0);  sub_8465 = None
	        clamp_min_555: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_370, 1.1920928955078125e-07);  div_370 = None
	        div_371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_185, clamp_min_555);  minimum_185 = None
	        round_371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_371);  div_371 = None
	        sub_8471: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_371);  round_371 = None
	        clamp_min_556: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8471, -128);  sub_8471 = None
	        clamp_max_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_556, 127);  clamp_min_556 = None
	        _assert_tensor_metadata_1667 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_555, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1667 = None
	        _assert_tensor_metadata_1668 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_370, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1668 = None
	        convert_element_type_1110: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_370, torch.int8);  clamp_max_370 = None
	        view_2901: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_555, [sym_size_int, 1500, 1])
	        view_2902: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1110, [sym_size_int, 1500, 1])
	        reciprocal_185: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2901);  view_2901 = None
	        mul_17954: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_185, 1.0);  reciprocal_185 = None
	        mul_17957: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17908, mul_17954);  mul_17908 = mul_17954 = None
	        round_372: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_17957);  mul_17957 = None
	        add_28421: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_372, view_2902);  round_372 = view_2902 = None
	        clamp_min_557: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28421, -128);  add_28421 = None
	        clamp_max_371: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_557, 127);  clamp_min_557 = None
	        _assert_tensor_metadata_1669 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_371, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1669 = None
	        convert_element_type_1111: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_371, torch.int8);  clamp_max_371 = None
	        view_2905: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_555, [sym_size_int, 1500, 1]);  clamp_min_555 = None
	        view_2906: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1110, [sym_size_int, 1500, 1]);  convert_element_type_1110 = None
	        _assert_tensor_metadata_1670 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1111, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1670 = None
	        convert_element_type_1112: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1111, torch.float32);  convert_element_type_1111 = None
	        _assert_tensor_metadata_1671 = torch.ops.aten._assert_tensor_metadata.default(view_2906, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1671 = None
	        convert_element_type_1113: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2906, torch.float32);  view_2906 = None
	        sub_8491: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1112, convert_element_type_1113);  convert_element_type_1112 = convert_element_type_1113 = None
	        mul_17979: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8491, view_2905);  sub_8491 = view_2905 = None
	        _assert_tensor_metadata_1672 = torch.ops.aten._assert_tensor_metadata.default(mul_17979, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1672 = None
	        view_2908: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_30_fc2_parametrizations_weight_original0 = None
	        view_2909: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_30_fc2_parametrizations_weight_original1 = None
	        view_2910: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_30_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1673 = torch.ops.aten._assert_tensor_metadata.default(view_2908, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1673 = None
	        convert_element_type_1114: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2908, torch.float32);  view_2908 = None
	        _assert_tensor_metadata_1674 = torch.ops.aten._assert_tensor_metadata.default(view_2910, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1674 = None
	        convert_element_type_1115: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2910, torch.float32);  view_2910 = None
	        sub_8495: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1114, convert_element_type_1115);  convert_element_type_1114 = convert_element_type_1115 = None
	        mul_17984: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8495, view_2909);  sub_8495 = view_2909 = None
	        view_2911: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17984, [1280, 5120]);  mul_17984 = None
	        _assert_tensor_metadata_1675 = torch.ops.aten._assert_tensor_metadata.default(view_2911, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1675 = None
	        mul_17989: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2912: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17979, [mul_17989, 5120]);  mul_17979 = mul_17989 = None
	        permute_310: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2911, [1, 0]);  view_2911 = None
	        addmm_154: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_30_fc2_bias, view_2912, permute_310);  model_audio_tower_layers_30_fc2_bias = view_2912 = permute_310 = None
	        view_2913: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_154, [sym_size_int, 1500, 1280]);  addmm_154 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_28484: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_28186, view_2913);  add_28186 = view_2913 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_28484, memory_format = torch.contiguous_format)
	        var_mean_62 = torch.ops.aten.var_mean.correction(clone_249, [2], correction = 0, keepdim = True)
	        getitem_248: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_62[0]
	        getitem_249: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_62[1];  var_mean_62 = None
	        add_28489: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_248, 1e-05);  getitem_248 = None
	        rsqrt_62: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_28489);  add_28489 = None
	        sub_8501: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_249, getitem_249);  clone_249 = getitem_249 = None
	        mul_18000: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8501, rsqrt_62);  sub_8501 = rsqrt_62 = None
	        mul_18001: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18000, model_audio_tower_layers_31_self_attn_layer_norm_weight);  mul_18000 = model_audio_tower_layers_31_self_attn_layer_norm_weight = None
	        add_28490: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_18001, model_audio_tower_layers_31_self_attn_layer_norm_bias);  mul_18001 = model_audio_tower_layers_31_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amin_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_28490, [2])
	        amax_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_28490, [2])
	        full_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_186, full_372);  amin_186 = full_372 = None
	        full_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_186, full_373);  amax_186 = full_373 = None
	        sub_8512: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_186, minimum_186);  maximum_186 = None
	        div_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8512, 255.0);  sub_8512 = None
	        clamp_min_558: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_372, 1.1920928955078125e-07);  div_372 = None
	        div_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_186, clamp_min_558);  minimum_186 = None
	        round_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_373);  div_373 = None
	        sub_8518: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_373);  round_373 = None
	        clamp_min_559: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8518, -128);  sub_8518 = None
	        clamp_max_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_559, 127);  clamp_min_559 = None
	        _assert_tensor_metadata_1676 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_558, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1676 = None
	        _assert_tensor_metadata_1677 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_372, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1677 = None
	        convert_element_type_1116: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_372, torch.int8);  clamp_max_372 = None
	        view_2916: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_558, [sym_size_int, 1500, 1])
	        view_2917: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1116, [sym_size_int, 1500, 1])
	        reciprocal_186: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2916);  view_2916 = None
	        mul_18049: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_186, 1.0);  reciprocal_186 = None
	        mul_18052: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_28490, mul_18049);  mul_18049 = None
	        round_374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18052);  mul_18052 = None
	        add_28577: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_374, view_2917);  round_374 = view_2917 = None
	        clamp_min_560: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28577, -128);  add_28577 = None
	        clamp_max_373: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_560, 127);  clamp_min_560 = None
	        _assert_tensor_metadata_1678 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_373, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1678 = None
	        convert_element_type_1117: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_373, torch.int8);  clamp_max_373 = None
	        view_2920: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_558, [sym_size_int, 1500, 1]);  clamp_min_558 = None
	        view_2921: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1116, [sym_size_int, 1500, 1]);  convert_element_type_1116 = None
	        _assert_tensor_metadata_1679 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1117, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1679 = None
	        convert_element_type_1118: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1117, torch.float32);  convert_element_type_1117 = None
	        _assert_tensor_metadata_1680 = torch.ops.aten._assert_tensor_metadata.default(view_2921, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1680 = None
	        convert_element_type_1119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2921, torch.float32);  view_2921 = None
	        sub_8538: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1118, convert_element_type_1119);  convert_element_type_1118 = convert_element_type_1119 = None
	        mul_18074: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8538, view_2920);  sub_8538 = view_2920 = None
	        _assert_tensor_metadata_1681 = torch.ops.aten._assert_tensor_metadata.default(mul_18074, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1681 = None
	        view_2923: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2924: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1 = None
	        view_2925: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1682 = torch.ops.aten._assert_tensor_metadata.default(view_2923, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1682 = None
	        convert_element_type_1120: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2923, torch.float32);  view_2923 = None
	        _assert_tensor_metadata_1683 = torch.ops.aten._assert_tensor_metadata.default(view_2925, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1683 = None
	        convert_element_type_1121: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2925, torch.float32);  view_2925 = None
	        sub_8542: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1120, convert_element_type_1121);  convert_element_type_1120 = convert_element_type_1121 = None
	        mul_18079: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8542, view_2924);  sub_8542 = view_2924 = None
	        view_2926: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18079, [1280, 1280]);  mul_18079 = None
	        _assert_tensor_metadata_1684 = torch.ops.aten._assert_tensor_metadata.default(view_2926, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1684 = None
	        mul_18084: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2927: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18074, [mul_18084, 1280]);  mul_18074 = mul_18084 = None
	        permute_311: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2926, [1, 0]);  view_2926 = None
	        addmm_155: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_31_self_attn_q_proj_bias, view_2927, permute_311);  model_audio_tower_layers_31_self_attn_q_proj_bias = view_2927 = permute_311 = None
	        view_2928: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_155, [sym_size_int, 1500, 1280]);  addmm_155 = None
	        mul_18091: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2928, 0.125);  view_2928 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2929: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18091, [sym_size_int, 1500, 20, 64]);  mul_18091 = None
	        permute_312: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2929, [0, 2, 1, 3]);  view_2929 = None
	        clone_250: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_312, memory_format = torch.contiguous_format);  permute_312 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amin_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_28490, [2])
	        amax_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_28490, [2])
	        full_374: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_187, full_374);  amin_187 = full_374 = None
	        full_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_187, full_375);  amax_187 = full_375 = None
	        sub_8557: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_187, minimum_187);  maximum_187 = None
	        div_374: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8557, 255.0);  sub_8557 = None
	        clamp_min_561: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_374, 1.1920928955078125e-07);  div_374 = None
	        div_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_187, clamp_min_561);  minimum_187 = None
	        round_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_375);  div_375 = None
	        sub_8563: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_375);  round_375 = None
	        clamp_min_562: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8563, -128);  sub_8563 = None
	        clamp_max_374: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_562, 127);  clamp_min_562 = None
	        _assert_tensor_metadata_1685 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_561, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1685 = None
	        _assert_tensor_metadata_1686 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_374, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1686 = None
	        convert_element_type_1122: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_374, torch.int8);  clamp_max_374 = None
	        view_2932: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_561, [sym_size_int, 1500, 1])
	        view_2933: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1122, [sym_size_int, 1500, 1])
	        reciprocal_187: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2932);  view_2932 = None
	        mul_18145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_187, 1.0);  reciprocal_187 = None
	        mul_18148: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_28490, mul_18145);  mul_18145 = None
	        round_376: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18148);  mul_18148 = None
	        add_28729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_376, view_2933);  round_376 = view_2933 = None
	        clamp_min_563: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28729, -128);  add_28729 = None
	        clamp_max_375: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_563, 127);  clamp_min_563 = None
	        _assert_tensor_metadata_1687 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_375, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1687 = None
	        convert_element_type_1123: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_375, torch.int8);  clamp_max_375 = None
	        view_2936: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_561, [sym_size_int, 1500, 1]);  clamp_min_561 = None
	        view_2937: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1122, [sym_size_int, 1500, 1]);  convert_element_type_1122 = None
	        _assert_tensor_metadata_1688 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1123, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1688 = None
	        convert_element_type_1124: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1123, torch.float32);  convert_element_type_1123 = None
	        _assert_tensor_metadata_1689 = torch.ops.aten._assert_tensor_metadata.default(view_2937, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1689 = None
	        convert_element_type_1125: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2937, torch.float32);  view_2937 = None
	        sub_8583: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1124, convert_element_type_1125);  convert_element_type_1124 = convert_element_type_1125 = None
	        mul_18170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8583, view_2936);  sub_8583 = view_2936 = None
	        _assert_tensor_metadata_1690 = torch.ops.aten._assert_tensor_metadata.default(mul_18170, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1690 = None
	        view_2939: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2940: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1 = None
	        view_2941: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1691 = torch.ops.aten._assert_tensor_metadata.default(view_2939, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1691 = None
	        convert_element_type_1126: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2939, torch.float32);  view_2939 = None
	        _assert_tensor_metadata_1692 = torch.ops.aten._assert_tensor_metadata.default(view_2941, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1692 = None
	        convert_element_type_1127: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2941, torch.float32);  view_2941 = None
	        sub_8587: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1126, convert_element_type_1127);  convert_element_type_1126 = convert_element_type_1127 = None
	        mul_18175: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8587, view_2940);  sub_8587 = view_2940 = None
	        view_2942: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18175, [1280, 1280]);  mul_18175 = None
	        _assert_tensor_metadata_1693 = torch.ops.aten._assert_tensor_metadata.default(view_2942, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1693 = None
	        permute_313: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2942, [1, 0]);  view_2942 = None
	        mul_18178: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2943: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18170, [mul_18178, 1280]);  mul_18170 = mul_18178 = None
	        mm_31: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2943, permute_313);  view_2943 = permute_313 = None
	        view_2944: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_31, [sym_size_int, 1500, 1280]);  mm_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2945: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2944, [sym_size_int, -1, 20, 64]);  view_2944 = None
	        permute_314: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2945, [0, 2, 1, 3]);  view_2945 = None
	        clone_251: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_314, memory_format = torch.contiguous_format);  permute_314 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amin_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_28490, [2])
	        amax_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_28490, [2])
	        full_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_188, full_376);  amin_188 = full_376 = None
	        full_377: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_188, full_377);  amax_188 = full_377 = None
	        sub_8601: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_188, minimum_188);  maximum_188 = None
	        div_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8601, 255.0);  sub_8601 = None
	        clamp_min_564: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_376, 1.1920928955078125e-07);  div_376 = None
	        div_377: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_188, clamp_min_564);  minimum_188 = None
	        round_377: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_377);  div_377 = None
	        sub_8607: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_377);  round_377 = None
	        clamp_min_565: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8607, -128);  sub_8607 = None
	        clamp_max_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_565, 127);  clamp_min_565 = None
	        _assert_tensor_metadata_1694 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_564, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1694 = None
	        _assert_tensor_metadata_1695 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_376, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1695 = None
	        convert_element_type_1128: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_376, torch.int8);  clamp_max_376 = None
	        view_2948: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_564, [sym_size_int, 1500, 1])
	        view_2949: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1128, [sym_size_int, 1500, 1])
	        reciprocal_188: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2948);  view_2948 = None
	        mul_18244: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_188, 1.0);  reciprocal_188 = None
	        mul_18247: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_28490, mul_18244);  add_28490 = mul_18244 = None
	        round_378: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18247);  mul_18247 = None
	        add_28877: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_378, view_2949);  round_378 = view_2949 = None
	        clamp_min_566: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28877, -128);  add_28877 = None
	        clamp_max_377: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_566, 127);  clamp_min_566 = None
	        _assert_tensor_metadata_1696 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_377, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1696 = None
	        convert_element_type_1129: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_377, torch.int8);  clamp_max_377 = None
	        view_2952: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_564, [sym_size_int, 1500, 1]);  clamp_min_564 = None
	        view_2953: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1128, [sym_size_int, 1500, 1]);  convert_element_type_1128 = None
	        _assert_tensor_metadata_1697 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1129, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1697 = None
	        convert_element_type_1130: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1129, torch.float32);  convert_element_type_1129 = None
	        _assert_tensor_metadata_1698 = torch.ops.aten._assert_tensor_metadata.default(view_2953, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1698 = None
	        convert_element_type_1131: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2953, torch.float32);  view_2953 = None
	        sub_8627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1130, convert_element_type_1131);  convert_element_type_1130 = convert_element_type_1131 = None
	        mul_18269: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8627, view_2952);  sub_8627 = view_2952 = None
	        _assert_tensor_metadata_1699 = torch.ops.aten._assert_tensor_metadata.default(mul_18269, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1699 = None
	        view_2955: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2956: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1 = None
	        view_2957: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1700 = torch.ops.aten._assert_tensor_metadata.default(view_2955, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1700 = None
	        convert_element_type_1132: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2955, torch.float32);  view_2955 = None
	        _assert_tensor_metadata_1701 = torch.ops.aten._assert_tensor_metadata.default(view_2957, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1701 = None
	        convert_element_type_1133: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2957, torch.float32);  view_2957 = None
	        sub_8631: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1132, convert_element_type_1133);  convert_element_type_1132 = convert_element_type_1133 = None
	        mul_18274: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8631, view_2956);  sub_8631 = view_2956 = None
	        view_2958: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18274, [1280, 1280]);  mul_18274 = None
	        _assert_tensor_metadata_1702 = torch.ops.aten._assert_tensor_metadata.default(view_2958, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1702 = None
	        mul_18279: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2959: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18269, [mul_18279, 1280]);  mul_18269 = mul_18279 = None
	        permute_315: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2958, [1, 0]);  view_2958 = None
	        addmm_156: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_31_self_attn_v_proj_bias, view_2959, permute_315);  model_audio_tower_layers_31_self_attn_v_proj_bias = view_2959 = permute_315 = None
	        view_2960: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_156, [sym_size_int, 1500, 1280]);  addmm_156 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2961: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2960, [sym_size_int, -1, 20, 64]);  view_2960 = None
	        permute_316: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2961, [0, 2, 1, 3]);  view_2961 = None
	        clone_252: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_316, memory_format = torch.contiguous_format);  permute_316 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_31 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_250, clone_251, clone_252, None, False, scale = 1.0);  clone_250 = clone_251 = clone_252 = None
	        getitem_250: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_31[0];  _scaled_dot_product_efficient_attention_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_317: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_250, [0, 2, 1, 3]);  getitem_250 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2962: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_317, [sym_size_int, 1500, -1]);  permute_317 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amin_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2962, [2])
	        amax_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2962, [2])
	        full_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_189, full_378);  amin_189 = full_378 = None
	        full_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_189, full_379);  amax_189 = full_379 = None
	        sub_8649: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_189, minimum_189);  maximum_189 = None
	        div_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8649, 255.0);  sub_8649 = None
	        clamp_min_567: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_378, 1.1920928955078125e-07);  div_378 = None
	        div_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_189, clamp_min_567);  minimum_189 = None
	        round_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_379);  div_379 = None
	        sub_8655: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_379);  round_379 = None
	        clamp_min_568: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8655, -128);  sub_8655 = None
	        clamp_max_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_568, 127);  clamp_min_568 = None
	        _assert_tensor_metadata_1703 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_567, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1703 = None
	        _assert_tensor_metadata_1704 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_378, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1704 = None
	        convert_element_type_1134: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_378, torch.int8);  clamp_max_378 = None
	        view_2965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_567, [sym_size_int, 1500, 1])
	        view_2966: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1134, [sym_size_int, 1500, 1])
	        reciprocal_189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2965);  view_2965 = None
	        mul_18349: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_189, 1.0);  reciprocal_189 = None
	        mul_18352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2962, mul_18349);  view_2962 = mul_18349 = None
	        round_380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18352);  mul_18352 = None
	        add_29041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_380, view_2966);  round_380 = view_2966 = None
	        clamp_min_569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29041, -128);  add_29041 = None
	        clamp_max_379: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_569, 127);  clamp_min_569 = None
	        _assert_tensor_metadata_1705 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_379, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1705 = None
	        convert_element_type_1135: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_379, torch.int8);  clamp_max_379 = None
	        view_2969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_567, [sym_size_int, 1500, 1]);  clamp_min_567 = None
	        view_2970: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1134, [sym_size_int, 1500, 1]);  convert_element_type_1134 = None
	        _assert_tensor_metadata_1706 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1135, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1706 = None
	        convert_element_type_1136: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1135, torch.float32);  convert_element_type_1135 = None
	        _assert_tensor_metadata_1707 = torch.ops.aten._assert_tensor_metadata.default(view_2970, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1707 = None
	        convert_element_type_1137: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2970, torch.float32);  view_2970 = None
	        sub_8675: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1136, convert_element_type_1137);  convert_element_type_1136 = convert_element_type_1137 = None
	        mul_18374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8675, view_2969);  sub_8675 = view_2969 = None
	        _assert_tensor_metadata_1708 = torch.ops.aten._assert_tensor_metadata.default(mul_18374, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1708 = None
	        view_2972: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2973: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1 = None
	        view_2974: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1709 = torch.ops.aten._assert_tensor_metadata.default(view_2972, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1709 = None
	        convert_element_type_1138: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2972, torch.float32);  view_2972 = None
	        _assert_tensor_metadata_1710 = torch.ops.aten._assert_tensor_metadata.default(view_2974, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1710 = None
	        convert_element_type_1139: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2974, torch.float32);  view_2974 = None
	        sub_8679: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1138, convert_element_type_1139);  convert_element_type_1138 = convert_element_type_1139 = None
	        mul_18379: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8679, view_2973);  sub_8679 = view_2973 = None
	        view_2975: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18379, [1280, 1280]);  mul_18379 = None
	        _assert_tensor_metadata_1711 = torch.ops.aten._assert_tensor_metadata.default(view_2975, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1711 = None
	        mul_18384: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2976: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18374, [mul_18384, 1280]);  mul_18374 = mul_18384 = None
	        permute_318: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2975, [1, 0]);  view_2975 = None
	        addmm_157: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_31_self_attn_out_proj_bias, view_2976, permute_318);  model_audio_tower_layers_31_self_attn_out_proj_bias = view_2976 = permute_318 = None
	        view_2977: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_157, [sym_size_int, 1500, 1280]);  addmm_157 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_29104: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_28484, view_2977);  add_28484 = view_2977 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_29104, memory_format = torch.contiguous_format)
	        var_mean_63 = torch.ops.aten.var_mean.correction(clone_254, [2], correction = 0, keepdim = True)
	        getitem_254: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_63[0]
	        getitem_255: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_63[1];  var_mean_63 = None
	        add_29109: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_254, 1e-05);  getitem_254 = None
	        rsqrt_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_29109);  add_29109 = None
	        sub_8685: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_254, getitem_255);  clone_254 = getitem_255 = None
	        mul_18395: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8685, rsqrt_63);  sub_8685 = rsqrt_63 = None
	        mul_18396: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18395, model_audio_tower_layers_31_final_layer_norm_weight);  mul_18395 = model_audio_tower_layers_31_final_layer_norm_weight = None
	        add_29110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_18396, model_audio_tower_layers_31_final_layer_norm_bias);  mul_18396 = model_audio_tower_layers_31_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amin_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_29110, [2])
	        amax_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_29110, [2])
	        full_380: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_190, full_380);  amin_190 = full_380 = None
	        full_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_190, full_381);  amax_190 = full_381 = None
	        sub_8696: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_190, minimum_190);  maximum_190 = None
	        div_380: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8696, 255.0);  sub_8696 = None
	        clamp_min_570: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_380, 1.1920928955078125e-07);  div_380 = None
	        div_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_190, clamp_min_570);  minimum_190 = None
	        round_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_381);  div_381 = None
	        sub_8702: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_381);  round_381 = None
	        clamp_min_571: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8702, -128);  sub_8702 = None
	        clamp_max_380: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_571, 127);  clamp_min_571 = None
	        _assert_tensor_metadata_1712 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_570, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1712 = None
	        _assert_tensor_metadata_1713 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_380, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1713 = None
	        convert_element_type_1140: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_380, torch.int8);  clamp_max_380 = None
	        view_2980: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_570, [sym_size_int, 1500, 1])
	        view_2981: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1140, [sym_size_int, 1500, 1])
	        reciprocal_190: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2980);  view_2980 = None
	        mul_18444: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_190, 1.0);  reciprocal_190 = None
	        mul_18447: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_29110, mul_18444);  add_29110 = mul_18444 = None
	        round_382: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18447);  mul_18447 = None
	        add_29197: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_382, view_2981);  round_382 = view_2981 = None
	        clamp_min_572: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29197, -128);  add_29197 = None
	        clamp_max_381: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_572, 127);  clamp_min_572 = None
	        _assert_tensor_metadata_1714 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_381, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1714 = None
	        convert_element_type_1141: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_381, torch.int8);  clamp_max_381 = None
	        view_2984: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_570, [sym_size_int, 1500, 1]);  clamp_min_570 = None
	        view_2985: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1140, [sym_size_int, 1500, 1]);  convert_element_type_1140 = None
	        _assert_tensor_metadata_1715 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1141, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1715 = None
	        convert_element_type_1142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1141, torch.float32);  convert_element_type_1141 = None
	        _assert_tensor_metadata_1716 = torch.ops.aten._assert_tensor_metadata.default(view_2985, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1716 = None
	        convert_element_type_1143: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2985, torch.float32);  view_2985 = None
	        sub_8722: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1142, convert_element_type_1143);  convert_element_type_1142 = convert_element_type_1143 = None
	        mul_18469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8722, view_2984);  sub_8722 = view_2984 = None
	        _assert_tensor_metadata_1717 = torch.ops.aten._assert_tensor_metadata.default(mul_18469, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1717 = None
	        view_2987: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_31_fc1_parametrizations_weight_original0 = None
	        view_2988: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_31_fc1_parametrizations_weight_original1 = None
	        view_2989: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_31_fc1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1718 = torch.ops.aten._assert_tensor_metadata.default(view_2987, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1718 = None
	        convert_element_type_1144: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2987, torch.float32);  view_2987 = None
	        _assert_tensor_metadata_1719 = torch.ops.aten._assert_tensor_metadata.default(view_2989, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1719 = None
	        convert_element_type_1145: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2989, torch.float32);  view_2989 = None
	        sub_8726: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1144, convert_element_type_1145);  convert_element_type_1144 = convert_element_type_1145 = None
	        mul_18474: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8726, view_2988);  sub_8726 = view_2988 = None
	        view_2990: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18474, [5120, 1280]);  mul_18474 = None
	        _assert_tensor_metadata_1720 = torch.ops.aten._assert_tensor_metadata.default(view_2990, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1720 = None
	        mul_18479: "Sym(1500*s6)" = sym_size_int * 1500
	        view_2991: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18469, [mul_18479, 1280]);  mul_18469 = mul_18479 = None
	        permute_319: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2990, [1, 0]);  view_2990 = None
	        addmm_158: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_31_fc1_bias, view_2991, permute_319);  model_audio_tower_layers_31_fc1_bias = view_2991 = permute_319 = None
	        view_2992: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_158, [sym_size_int, 1500, 5120]);  addmm_158 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_18486: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2992, 0.5)
	        mul_18487: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2992, 0.7071067811865476);  view_2992 = None
	        erf_33: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_18487);  mul_18487 = None
	        add_29256: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_33, 1);  erf_33 = None
	        mul_18488: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18486, add_29256);  mul_18486 = add_29256 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amin_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_18488, [2])
	        amax_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_18488, [2])
	        full_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_191, full_382);  amin_191 = full_382 = None
	        full_383: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_191, full_383);  amax_191 = full_383 = None
	        sub_8739: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_191, minimum_191);  maximum_191 = None
	        div_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8739, 255.0);  sub_8739 = None
	        clamp_min_573: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_382, 1.1920928955078125e-07);  div_382 = None
	        div_383: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_191, clamp_min_573);  minimum_191 = None
	        round_383: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_383);  div_383 = None
	        sub_8745: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_383);  round_383 = None
	        clamp_min_574: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8745, -128);  sub_8745 = None
	        clamp_max_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_574, 127);  clamp_min_574 = None
	        _assert_tensor_metadata_1721 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_573, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1721 = None
	        _assert_tensor_metadata_1722 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_382, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1722 = None
	        convert_element_type_1146: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_382, torch.int8);  clamp_max_382 = None
	        view_2995: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_573, [sym_size_int, 1500, 1])
	        view_2996: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1146, [sym_size_int, 1500, 1])
	        reciprocal_191: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2995);  view_2995 = None
	        mul_18534: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_191, 1.0);  reciprocal_191 = None
	        mul_18537: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18488, mul_18534);  mul_18488 = mul_18534 = None
	        round_384: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_18537);  mul_18537 = None
	        add_29339: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_384, view_2996);  round_384 = view_2996 = None
	        clamp_min_575: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29339, -128);  add_29339 = None
	        clamp_max_383: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_575, 127);  clamp_min_575 = None
	        _assert_tensor_metadata_1723 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_383, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1723 = None
	        convert_element_type_1147: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_383, torch.int8);  clamp_max_383 = None
	        view_2999: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_573, [sym_size_int, 1500, 1]);  clamp_min_573 = None
	        view_3000: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1146, [sym_size_int, 1500, 1]);  convert_element_type_1146 = None
	        _assert_tensor_metadata_1724 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1147, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1724 = None
	        convert_element_type_1148: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1147, torch.float32);  convert_element_type_1147 = None
	        _assert_tensor_metadata_1725 = torch.ops.aten._assert_tensor_metadata.default(view_3000, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1725 = None
	        convert_element_type_1149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3000, torch.float32);  view_3000 = None
	        sub_8765: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1148, convert_element_type_1149);  convert_element_type_1148 = convert_element_type_1149 = None
	        mul_18559: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8765, view_2999);  sub_8765 = view_2999 = None
	        _assert_tensor_metadata_1726 = torch.ops.aten._assert_tensor_metadata.default(mul_18559, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1726 = None
	        view_3002: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_31_fc2_parametrizations_weight_original0 = None
	        view_3003: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_31_fc2_parametrizations_weight_original1 = None
	        view_3004: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_31_fc2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1727 = torch.ops.aten._assert_tensor_metadata.default(view_3002, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1727 = None
	        convert_element_type_1150: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3002, torch.float32);  view_3002 = None
	        _assert_tensor_metadata_1728 = torch.ops.aten._assert_tensor_metadata.default(view_3004, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1728 = None
	        convert_element_type_1151: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3004, torch.float32);  view_3004 = None
	        sub_8769: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1150, convert_element_type_1151);  convert_element_type_1150 = convert_element_type_1151 = None
	        mul_18564: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8769, view_3003);  sub_8769 = view_3003 = None
	        view_3005: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18564, [1280, 5120]);  mul_18564 = None
	        _assert_tensor_metadata_1729 = torch.ops.aten._assert_tensor_metadata.default(view_3005, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1729 = None
	        mul_18569: "Sym(1500*s6)" = sym_size_int * 1500
	        view_3006: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18559, [mul_18569, 5120]);  mul_18559 = mul_18569 = None
	        permute_320: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_3005, [1, 0]);  view_3005 = None
	        addmm_159: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.addmm.default(model_audio_tower_layers_31_fc2_bias, view_3006, permute_320);  model_audio_tower_layers_31_fc2_bias = view_3006 = permute_320 = None
	        view_3007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(addmm_159, [sym_size_int, 1500, 1280]);  addmm_159 = sym_size_int = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_29402: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_29104, view_3007);  add_29104 = view_3007 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:365 in forward, code: hidden_states = self.layer_norm(hidden_states)
	        clone_257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_29402, memory_format = torch.contiguous_format);  add_29402 = None
	        var_mean_64 = torch.ops.aten.var_mean.correction(clone_257, [2], correction = 0, keepdim = True)
	        getitem_256: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_64[0]
	        getitem_257: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_64[1];  var_mean_64 = None
	        add_29407: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_256, 1e-05);  getitem_256 = None
	        rsqrt_64: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_29407);  add_29407 = None
	        sub_8775: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_257, getitem_257);  clone_257 = getitem_257 = None
	        mul_18580: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8775, rsqrt_64);  sub_8775 = rsqrt_64 = None
	        mul_18581: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18580, model_audio_tower_layer_norm_weight);  mul_18580 = model_audio_tower_layer_norm_weight = None
	        add_29408: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_18581, model_audio_tower_layer_norm_bias);  mul_18581 = model_audio_tower_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:451 in get_audio_embeds, code: audio_hidden_states = audio_hidden_states.reshape(-1, self.config.audio_config.intermediate_size)
	        view_3008: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_29408, [-1, 5120]);  add_29408 = None
	        sym_size_int_193: "Sym(375*s6)" = torch.ops.aten.sym_size.int(view_3008, 0)
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:389 in forward, code: hidden_states = self.linear_1(audio_features)
	        amin_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.amin.default(view_3008, [1])
	        amax_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.amax.default(view_3008, [1])
	        full_384: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.minimum.default(amin_192, full_384);  amin_192 = full_384 = None
	        full_385: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.maximum.default(amax_192, full_385);  amax_192 = full_385 = None
	        sub_8787: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_192, minimum_192);  maximum_192 = None
	        div_384: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(sub_8787, 255.0);  sub_8787 = None
	        clamp_min_576: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(div_384, 1.1920928955078125e-07);  div_384 = None
	        div_385: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(minimum_192, clamp_min_576);  minimum_192 = None
	        round_385: "f32[375*s6][1]cuda:0" = torch.ops.aten.round.default(div_385);  div_385 = None
	        sub_8793: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_385);  round_385 = None
	        clamp_min_577: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8793, -128);  sub_8793 = None
	        clamp_max_384: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_577, 127);  clamp_min_577 = None
	        _assert_tensor_metadata_1730 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_576, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1730 = None
	        _assert_tensor_metadata_1731 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_384, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1731 = None
	        convert_element_type_1152: "i8[375*s6][1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_384, torch.int8);  clamp_max_384 = None
	        view_3011: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_576, [sym_size_int_193, 1])
	        view_3012: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1152, [sym_size_int_193, 1])
	        reciprocal_192: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_3011);  view_3011 = None
	        mul_18613: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_192, 1.0);  reciprocal_192 = None
	        mul_18615: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_3008, mul_18613);  view_3008 = mul_18613 = None
	        round_386: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.round.default(mul_18615);  mul_18615 = None
	        add_29476: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_386, view_3012);  round_386 = view_3012 = None
	        clamp_min_578: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29476, -128);  add_29476 = None
	        clamp_max_385: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_578, 127);  clamp_min_578 = None
	        _assert_tensor_metadata_1732 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_385, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1732 = None
	        convert_element_type_1153: "i8[375*s6, 5120][5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_385, torch.int8);  clamp_max_385 = None
	        view_3015: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_576, [sym_size_int_193, 1]);  clamp_min_576 = None
	        view_3016: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1152, [sym_size_int_193, 1]);  convert_element_type_1152 = None
	        _assert_tensor_metadata_1733 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1153, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1733 = None
	        convert_element_type_1154: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1153, torch.float32);  convert_element_type_1153 = None
	        _assert_tensor_metadata_1734 = torch.ops.aten._assert_tensor_metadata.default(view_3016, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1734 = None
	        convert_element_type_1155: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3016, torch.float32);  view_3016 = None
	        sub_8813: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1154, convert_element_type_1155);  convert_element_type_1154 = convert_element_type_1155 = None
	        mul_18634: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8813, view_3015);  sub_8813 = view_3015 = None
	        _assert_tensor_metadata_1735 = torch.ops.aten._assert_tensor_metadata.default(mul_18634, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1735 = None
	        view_3018: "i8[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_multi_modal_projector_linear_1_parametrizations_weight_original0, [3072, 160, 32]);  model_multi_modal_projector_linear_1_parametrizations_weight_original0 = None
	        view_3019: "f32[3072, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_multi_modal_projector_linear_1_parametrizations_weight_original1, [3072, 160, 1]);  model_multi_modal_projector_linear_1_parametrizations_weight_original1 = None
	        view_3020: "i8[3072, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_multi_modal_projector_linear_1_parametrizations_weight_original2, [3072, 160, 1]);  model_multi_modal_projector_linear_1_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1736 = torch.ops.aten._assert_tensor_metadata.default(view_3018, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1736 = None
	        convert_element_type_1156: "f32[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3018, torch.float32);  view_3018 = None
	        _assert_tensor_metadata_1737 = torch.ops.aten._assert_tensor_metadata.default(view_3020, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1737 = None
	        convert_element_type_1157: "f32[3072, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3020, torch.float32);  view_3020 = None
	        sub_8817: "f32[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1156, convert_element_type_1157);  convert_element_type_1156 = convert_element_type_1157 = None
	        mul_18639: "f32[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8817, view_3019);  sub_8817 = view_3019 = None
	        view_3021: "f32[3072, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18639, [3072, 5120]);  mul_18639 = None
	        _assert_tensor_metadata_1738 = torch.ops.aten._assert_tensor_metadata.default(view_3021, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1738 = None
	        permute_321: "f32[5120, 3072][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_3021, [1, 0]);  view_3021 = None
	        mm_32: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mm.default(mul_18634, permute_321);  mul_18634 = permute_321 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_18642: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(mm_32, 0.5)
	        mul_18643: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(mm_32, 0.7071067811865476);  mm_32 = None
	        erf_34: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.erf.default(mul_18643);  mul_18643 = None
	        add_29516: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_34, 1);  erf_34 = None
	        mul_18644: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18642, add_29516);  mul_18642 = add_29516 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:391 in forward, code: hidden_states = self.linear_2(hidden_states)
	        amin_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.amin.default(mul_18644, [1])
	        amax_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.amax.default(mul_18644, [1])
	        full_386: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.minimum.default(amin_193, full_386);  amin_193 = full_386 = None
	        full_387: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.maximum.default(amax_193, full_387);  amax_193 = full_387 = None
	        sub_8827: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_193, minimum_193);  maximum_193 = None
	        div_386: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(sub_8827, 255.0);  sub_8827 = None
	        clamp_min_579: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(div_386, 1.1920928955078125e-07);  div_386 = None
	        div_387: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(minimum_193, clamp_min_579);  minimum_193 = None
	        round_387: "f32[375*s6][1]cuda:0" = torch.ops.aten.round.default(div_387);  div_387 = None
	        sub_8833: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_387);  round_387 = None
	        clamp_min_580: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8833, -128);  sub_8833 = None
	        clamp_max_386: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_580, 127);  clamp_min_580 = None
	        _assert_tensor_metadata_1739 = torch.ops.aten._assert_tensor_metadata.default(clamp_min_579, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1739 = None
	        _assert_tensor_metadata_1740 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_386, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1740 = None
	        convert_element_type_1158: "i8[375*s6][1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_386, torch.int8);  clamp_max_386 = None
	        view_3024: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_579, [sym_size_int_193, 1])
	        view_3025: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1158, [sym_size_int_193, 1])
	        reciprocal_193: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_3024);  view_3024 = None
	        mul_18666: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_193, 1.0);  reciprocal_193 = None
	        mul_18668: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18644, mul_18666);  mul_18644 = mul_18666 = None
	        round_388: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.round.default(mul_18668);  mul_18668 = None
	        add_29572: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.add.Tensor(round_388, view_3025);  round_388 = view_3025 = None
	        clamp_min_581: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29572, -128);  add_29572 = None
	        clamp_max_387: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_581, 127);  clamp_min_581 = None
	        _assert_tensor_metadata_1741 = torch.ops.aten._assert_tensor_metadata.default(clamp_max_387, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1741 = None
	        convert_element_type_1159: "i8[375*s6, 3072][3072, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_387, torch.int8);  clamp_max_387 = None
	        view_3028: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_579, [sym_size_int_193, 1]);  clamp_min_579 = None
	        view_3029: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1158, [sym_size_int_193, 1]);  convert_element_type_1158 = sym_size_int_193 = None
	        _assert_tensor_metadata_1742 = torch.ops.aten._assert_tensor_metadata.default(convert_element_type_1159, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1742 = None
	        convert_element_type_1160: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1159, torch.float32);  convert_element_type_1159 = None
	        _assert_tensor_metadata_1743 = torch.ops.aten._assert_tensor_metadata.default(view_3029, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1743 = None
	        convert_element_type_1161: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3029, torch.float32);  view_3029 = None
	        sub_8853: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1160, convert_element_type_1161);  convert_element_type_1160 = convert_element_type_1161 = None
	        mul_18687: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8853, view_3028);  sub_8853 = view_3028 = None
	        _assert_tensor_metadata_1744 = torch.ops.aten._assert_tensor_metadata.default(mul_18687, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1744 = None
	        view_3031: "i8[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_multi_modal_projector_linear_2_parametrizations_weight_original0, [3072, 96, 32]);  model_multi_modal_projector_linear_2_parametrizations_weight_original0 = None
	        view_3032: "f32[3072, 96, 1][96, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_multi_modal_projector_linear_2_parametrizations_weight_original1, [3072, 96, 1]);  model_multi_modal_projector_linear_2_parametrizations_weight_original1 = None
	        view_3033: "i8[3072, 96, 1][96, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_multi_modal_projector_linear_2_parametrizations_weight_original2, [3072, 96, 1]);  model_multi_modal_projector_linear_2_parametrizations_weight_original2 = None
	        _assert_tensor_metadata_1745 = torch.ops.aten._assert_tensor_metadata.default(view_3031, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1745 = None
	        convert_element_type_1162: "f32[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3031, torch.float32);  view_3031 = None
	        _assert_tensor_metadata_1746 = torch.ops.aten._assert_tensor_metadata.default(view_3033, None, None, torch.int8, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1746 = None
	        convert_element_type_1163: "f32[3072, 96, 1][96, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3033, torch.float32);  view_3033 = None
	        sub_8857: "f32[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1162, convert_element_type_1163);  convert_element_type_1162 = convert_element_type_1163 = None
	        mul_18692: "f32[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8857, view_3032);  sub_8857 = view_3032 = None
	        view_3034: "f32[3072, 3072][3072, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18692, [3072, 3072]);  mul_18692 = None
	        _assert_tensor_metadata_1747 = torch.ops.aten._assert_tensor_metadata.default(view_3034, None, None, torch.float32, device = device(type='cuda', index=0), layout = torch.strided);  _assert_tensor_metadata_1747 = None
	        permute_322: "f32[3072, 3072][1, 3072]cuda:0" = torch.ops.aten.permute.default(view_3034, [1, 0]);  view_3034 = None
	        mm_33: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mm.default(mul_18687, permute_322);  mul_18687 = permute_322 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py:83 in forward, code: return audio_embeds.unsqueeze(0)
	        unsqueeze: "f32[1, 375*s6, 3072][1152000*s6, 3072, 1]cuda:0" = torch.ops.aten.unsqueeze.default(mm_33, 0);  mm_33 = None
	        return (unsqueeze,)
	        
V0910 09:42:56.717000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "57cd37156f8f424e3dd875dbf33f1191"}
	{
	"name": "_recursive_post_grad_passes",
	"ts": 1757522576717500.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:04.550000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a296f7e66bc76b4adbd199f68252c68e"}
	{
	"name": "_recursive_post_grad_passes",
	"ts": 1757522584550598.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:05.101000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_inductor/compile_fx.py:1302] {"artifact": {"name": "after_post_grad_graph", "encoding": "string"}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2657, "name": "compile_fx", "filename": 19, "loc": "return inference_compiler(unlifted_gm, example_inputs_)"}, {"line": 1267, "name": "__call__", "filename": 33, "loc": "return self.compiler_fn(gm, example_inputs)"}, {"line": 2543, "name": "fw_compiler_base", "filename": 19, "loc": "return compile_fx_forward("}, {"line": 2260, "name": "compile_fx_forward", "filename": 19, "loc": "return inner_compile("}, {"line": 81, "name": "inner", "filename": 34, "loc": "return func(*args, **kwds)"}, {"line": 781, "name": "compile_fx_inner", "filename": 19, "loc": "return wrap_compiler_debug(_compile_fx_inner, compiler_name=\"inductor\")("}, {"line": 144, "name": "debug_wrapper", "filename": 35, "loc": "inner_compiled_fn = compiler_fn(gm, example_inputs)"}, {"line": 167, "name": "newFunction", "filename": 36, "loc": "return old_func(*args, **kwargs)"}, {"line": 962, "name": "_compile_fx_inner", "filename": 19, "loc": "mb_compiled_graph = fx_codegen_and_compile("}, {"line": 1694, "name": "fx_codegen_and_compile", "filename": 19, "loc": "return scheme.codegen_and_compile(gm, example_inputs, inputs_to_check, graph_kwargs)"}, {"line": 1302, "name": "codegen_and_compile", "filename": 19, "loc": "trace_structured("}], "has_payload": "af157a0bd6e22e19695c65b29a1ec67b"}
	class <lambda>(torch.nn.Module):
	    def forward(self):
	        arg877_1: "f32[s6, 128, 3000][384000, 3000, 1]cuda:0"; 
	    
	        arg877_1, = fx_pytree.tree_flatten_spec([], self._in_spec)
	        # No stacktrace found for following nodes
	        model_audio_tower_embed_positions_weight: "f32[1500, 1280][1280, 1]cuda:0" = self.model.audio_tower.embed_positions.weight
	        model_audio_tower_conv1_weight: "f32[1280, 128, 3][384, 3, 1]cuda:0" = self.model.audio_tower.conv1.weight
	        model_audio_tower_conv1_bias: "f32[1280][1]cuda:0" = self.model.audio_tower.conv1.bias
	        model_audio_tower_conv2_weight: "f32[1280, 1280, 3][3840, 3, 1]cuda:0" = self.model.audio_tower.conv2.weight
	        model_audio_tower_conv2_bias: "f32[1280][1]cuda:0" = self.model.audio_tower.conv2.bias
	        model_audio_tower_layers_0_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn_layer_norm.weight
	        model_audio_tower_layers_0_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn_layer_norm.bias
	        model_audio_tower_layers_0_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.bias
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.bias
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.bias
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_0_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").final_layer_norm.weight
	        model_audio_tower_layers_0_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").final_layer_norm.bias
	        model_audio_tower_layers_0_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.bias
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_0_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_0_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.bias
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_0_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "0").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn_layer_norm.weight
	        model_audio_tower_layers_1_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn_layer_norm.bias
	        model_audio_tower_layers_1_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.bias
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.bias
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.bias
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_1_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").final_layer_norm.weight
	        model_audio_tower_layers_1_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").final_layer_norm.bias
	        model_audio_tower_layers_1_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.bias
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_1_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_1_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.bias
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_1_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "1").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn_layer_norm.weight
	        model_audio_tower_layers_2_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn_layer_norm.bias
	        model_audio_tower_layers_2_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.bias
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.bias
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.bias
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_2_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").final_layer_norm.weight
	        model_audio_tower_layers_2_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").final_layer_norm.bias
	        model_audio_tower_layers_2_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.bias
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_2_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_2_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.bias
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_2_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "2").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn_layer_norm.weight
	        model_audio_tower_layers_3_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn_layer_norm.bias
	        model_audio_tower_layers_3_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.bias
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.bias
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.bias
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_3_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").final_layer_norm.weight
	        model_audio_tower_layers_3_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").final_layer_norm.bias
	        model_audio_tower_layers_3_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.bias
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_3_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_3_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.bias
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_3_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "3").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn_layer_norm.weight
	        model_audio_tower_layers_4_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn_layer_norm.bias
	        model_audio_tower_layers_4_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.bias
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.bias
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.bias
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_4_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").final_layer_norm.weight
	        model_audio_tower_layers_4_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").final_layer_norm.bias
	        model_audio_tower_layers_4_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.bias
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_4_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_4_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.bias
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_4_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "4").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn_layer_norm.weight
	        model_audio_tower_layers_5_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn_layer_norm.bias
	        model_audio_tower_layers_5_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.bias
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.bias
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.bias
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_5_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").final_layer_norm.weight
	        model_audio_tower_layers_5_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").final_layer_norm.bias
	        model_audio_tower_layers_5_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.bias
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_5_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_5_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.bias
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_5_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "5").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn_layer_norm.weight
	        model_audio_tower_layers_6_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn_layer_norm.bias
	        model_audio_tower_layers_6_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.bias
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.bias
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.bias
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_6_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").final_layer_norm.weight
	        model_audio_tower_layers_6_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").final_layer_norm.bias
	        model_audio_tower_layers_6_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.bias
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_6_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_6_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.bias
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_6_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "6").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn_layer_norm.weight
	        model_audio_tower_layers_7_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn_layer_norm.bias
	        model_audio_tower_layers_7_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.bias
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.bias
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.bias
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_7_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").final_layer_norm.weight
	        model_audio_tower_layers_7_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").final_layer_norm.bias
	        model_audio_tower_layers_7_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.bias
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_7_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_7_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.bias
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_7_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "7").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn_layer_norm.weight
	        model_audio_tower_layers_8_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn_layer_norm.bias
	        model_audio_tower_layers_8_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.bias
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.bias
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.bias
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_8_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").final_layer_norm.weight
	        model_audio_tower_layers_8_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").final_layer_norm.bias
	        model_audio_tower_layers_8_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.bias
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_8_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_8_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.bias
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_8_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "8").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn_layer_norm.weight
	        model_audio_tower_layers_9_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn_layer_norm.bias
	        model_audio_tower_layers_9_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.bias
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.bias
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.bias
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_9_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").final_layer_norm.weight
	        model_audio_tower_layers_9_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").final_layer_norm.bias
	        model_audio_tower_layers_9_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.bias
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_9_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_9_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.bias
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_9_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "9").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn_layer_norm.weight
	        model_audio_tower_layers_10_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn_layer_norm.bias
	        model_audio_tower_layers_10_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.bias
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.bias
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.bias
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_10_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").final_layer_norm.weight
	        model_audio_tower_layers_10_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").final_layer_norm.bias
	        model_audio_tower_layers_10_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.bias
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_10_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_10_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.bias
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_10_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "10").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn_layer_norm.weight
	        model_audio_tower_layers_11_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn_layer_norm.bias
	        model_audio_tower_layers_11_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.bias
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.bias
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.bias
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_11_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").final_layer_norm.weight
	        model_audio_tower_layers_11_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").final_layer_norm.bias
	        model_audio_tower_layers_11_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.bias
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_11_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_11_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.bias
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_11_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "11").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn_layer_norm.weight
	        model_audio_tower_layers_12_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn_layer_norm.bias
	        model_audio_tower_layers_12_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.bias
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.bias
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.bias
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_12_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").final_layer_norm.weight
	        model_audio_tower_layers_12_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").final_layer_norm.bias
	        model_audio_tower_layers_12_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.bias
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_12_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_12_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.bias
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_12_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "12").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn_layer_norm.weight
	        model_audio_tower_layers_13_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn_layer_norm.bias
	        model_audio_tower_layers_13_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.bias
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.bias
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.bias
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_13_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").final_layer_norm.weight
	        model_audio_tower_layers_13_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").final_layer_norm.bias
	        model_audio_tower_layers_13_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.bias
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_13_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_13_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.bias
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_13_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "13").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn_layer_norm.weight
	        model_audio_tower_layers_14_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn_layer_norm.bias
	        model_audio_tower_layers_14_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.bias
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.bias
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.bias
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_14_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").final_layer_norm.weight
	        model_audio_tower_layers_14_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").final_layer_norm.bias
	        model_audio_tower_layers_14_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.bias
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_14_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_14_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.bias
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_14_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "14").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn_layer_norm.weight
	        model_audio_tower_layers_15_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn_layer_norm.bias
	        model_audio_tower_layers_15_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.bias
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.bias
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.bias
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_15_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").final_layer_norm.weight
	        model_audio_tower_layers_15_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").final_layer_norm.bias
	        model_audio_tower_layers_15_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.bias
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_15_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_15_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.bias
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_15_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "15").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn_layer_norm.weight
	        model_audio_tower_layers_16_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn_layer_norm.bias
	        model_audio_tower_layers_16_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.bias
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.bias
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.bias
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_16_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").final_layer_norm.weight
	        model_audio_tower_layers_16_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").final_layer_norm.bias
	        model_audio_tower_layers_16_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.bias
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_16_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_16_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.bias
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_16_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "16").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn_layer_norm.weight
	        model_audio_tower_layers_17_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn_layer_norm.bias
	        model_audio_tower_layers_17_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.bias
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.bias
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.bias
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_17_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").final_layer_norm.weight
	        model_audio_tower_layers_17_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").final_layer_norm.bias
	        model_audio_tower_layers_17_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.bias
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_17_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_17_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.bias
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_17_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "17").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn_layer_norm.weight
	        model_audio_tower_layers_18_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn_layer_norm.bias
	        model_audio_tower_layers_18_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.bias
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.bias
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.bias
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_18_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").final_layer_norm.weight
	        model_audio_tower_layers_18_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").final_layer_norm.bias
	        model_audio_tower_layers_18_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.bias
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_18_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_18_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.bias
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_18_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "18").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn_layer_norm.weight
	        model_audio_tower_layers_19_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn_layer_norm.bias
	        model_audio_tower_layers_19_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.bias
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.bias
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.bias
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_19_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").final_layer_norm.weight
	        model_audio_tower_layers_19_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").final_layer_norm.bias
	        model_audio_tower_layers_19_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.bias
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_19_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_19_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.bias
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_19_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "19").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn_layer_norm.weight
	        model_audio_tower_layers_20_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn_layer_norm.bias
	        model_audio_tower_layers_20_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.bias
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.bias
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.bias
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_20_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").final_layer_norm.weight
	        model_audio_tower_layers_20_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").final_layer_norm.bias
	        model_audio_tower_layers_20_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.bias
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_20_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_20_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.bias
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_20_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "20").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn_layer_norm.weight
	        model_audio_tower_layers_21_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn_layer_norm.bias
	        model_audio_tower_layers_21_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.bias
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.bias
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.bias
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_21_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").final_layer_norm.weight
	        model_audio_tower_layers_21_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").final_layer_norm.bias
	        model_audio_tower_layers_21_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.bias
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_21_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_21_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.bias
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_21_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "21").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn_layer_norm.weight
	        model_audio_tower_layers_22_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn_layer_norm.bias
	        model_audio_tower_layers_22_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.bias
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.bias
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.bias
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_22_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").final_layer_norm.weight
	        model_audio_tower_layers_22_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").final_layer_norm.bias
	        model_audio_tower_layers_22_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.bias
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_22_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_22_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.bias
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_22_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "22").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn_layer_norm.weight
	        model_audio_tower_layers_23_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn_layer_norm.bias
	        model_audio_tower_layers_23_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.bias
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.bias
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.bias
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_23_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").final_layer_norm.weight
	        model_audio_tower_layers_23_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").final_layer_norm.bias
	        model_audio_tower_layers_23_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.bias
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_23_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_23_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.bias
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_23_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "23").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn_layer_norm.weight
	        model_audio_tower_layers_24_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn_layer_norm.bias
	        model_audio_tower_layers_24_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.bias
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.bias
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.bias
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_24_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").final_layer_norm.weight
	        model_audio_tower_layers_24_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").final_layer_norm.bias
	        model_audio_tower_layers_24_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.bias
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_24_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_24_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.bias
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_24_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "24").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn_layer_norm.weight
	        model_audio_tower_layers_25_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn_layer_norm.bias
	        model_audio_tower_layers_25_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.bias
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.bias
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.bias
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_25_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").final_layer_norm.weight
	        model_audio_tower_layers_25_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").final_layer_norm.bias
	        model_audio_tower_layers_25_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.bias
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_25_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_25_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.bias
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_25_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "25").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn_layer_norm.weight
	        model_audio_tower_layers_26_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn_layer_norm.bias
	        model_audio_tower_layers_26_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.bias
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.bias
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.bias
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_26_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").final_layer_norm.weight
	        model_audio_tower_layers_26_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").final_layer_norm.bias
	        model_audio_tower_layers_26_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.bias
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_26_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_26_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.bias
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_26_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "26").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn_layer_norm.weight
	        model_audio_tower_layers_27_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn_layer_norm.bias
	        model_audio_tower_layers_27_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.bias
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.bias
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.bias
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_27_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").final_layer_norm.weight
	        model_audio_tower_layers_27_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").final_layer_norm.bias
	        model_audio_tower_layers_27_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.bias
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_27_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_27_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.bias
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_27_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "27").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn_layer_norm.weight
	        model_audio_tower_layers_28_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn_layer_norm.bias
	        model_audio_tower_layers_28_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.bias
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.bias
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.bias
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_28_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").final_layer_norm.weight
	        model_audio_tower_layers_28_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").final_layer_norm.bias
	        model_audio_tower_layers_28_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.bias
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_28_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_28_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.bias
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_28_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "28").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn_layer_norm.weight
	        model_audio_tower_layers_29_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn_layer_norm.bias
	        model_audio_tower_layers_29_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.bias
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.bias
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.bias
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_29_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").final_layer_norm.weight
	        model_audio_tower_layers_29_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").final_layer_norm.bias
	        model_audio_tower_layers_29_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.bias
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_29_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_29_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.bias
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_29_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "29").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn_layer_norm.weight
	        model_audio_tower_layers_30_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn_layer_norm.bias
	        model_audio_tower_layers_30_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.bias
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.bias
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.bias
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_30_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").final_layer_norm.weight
	        model_audio_tower_layers_30_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").final_layer_norm.bias
	        model_audio_tower_layers_30_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.bias
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_30_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_30_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.bias
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_30_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "30").fc2.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn_layer_norm.weight
	        model_audio_tower_layers_31_self_attn_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn_layer_norm.bias
	        model_audio_tower_layers_31_self_attn_q_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.bias
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.q_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.k_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_v_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.bias
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.v_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_self_attn_out_proj_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.bias
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0: "i8[1280, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original0
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1: "f32[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original1
	        model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2: "i8[1280, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").self_attn.out_proj.parametrizations.weight.original2
	        model_audio_tower_layers_31_final_layer_norm_weight: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").final_layer_norm.weight
	        model_audio_tower_layers_31_final_layer_norm_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").final_layer_norm.bias
	        model_audio_tower_layers_31_fc1_bias: "f32[5120][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.bias
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original0: "i8[5120, 1280][1280, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original0
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original1: "f32[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original1
	        model_audio_tower_layers_31_fc1_parametrizations_weight_original2: "i8[5120, 40][40, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc1.parametrizations.weight.original2
	        model_audio_tower_layers_31_fc2_bias: "f32[1280][1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.bias
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original0: "i8[1280, 5120][5120, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original0
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original1: "f32[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original1
	        model_audio_tower_layers_31_fc2_parametrizations_weight_original2: "i8[1280, 160][160, 1]cuda:0" = getattr(self.model.audio_tower.layers, "31").fc2.parametrizations.weight.original2
	        model_audio_tower_layer_norm_weight: "f32[1280][1]cuda:0" = self.model.audio_tower.layer_norm.weight
	        model_audio_tower_layer_norm_bias: "f32[1280][1]cuda:0" = self.model.audio_tower.layer_norm.bias
	        model_multi_modal_projector_linear_1_parametrizations_weight_original0: "i8[3072, 5120][5120, 1]cuda:0" = self.model.multi_modal_projector.linear_1.parametrizations.weight.original0
	        model_multi_modal_projector_linear_1_parametrizations_weight_original1: "f32[3072, 160][160, 1]cuda:0" = self.model.multi_modal_projector.linear_1.parametrizations.weight.original1
	        model_multi_modal_projector_linear_1_parametrizations_weight_original2: "i8[3072, 160][160, 1]cuda:0" = self.model.multi_modal_projector.linear_1.parametrizations.weight.original2
	        model_multi_modal_projector_linear_2_parametrizations_weight_original0: "i8[3072, 3072][3072, 1]cuda:0" = self.model.multi_modal_projector.linear_2.parametrizations.weight.original0
	        model_multi_modal_projector_linear_2_parametrizations_weight_original1: "f32[3072, 96][96, 1]cuda:0" = self.model.multi_modal_projector.linear_2.parametrizations.weight.original1
	        model_multi_modal_projector_linear_2_parametrizations_weight_original2: "i8[3072, 96][96, 1]cuda:0" = self.model.multi_modal_projector.linear_2.parametrizations.weight.original2
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:349 in forward, code: inputs_embeds = nn.functional.gelu(self.conv1(input_features))
	        convolution: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.convolution.default(arg877_1, model_audio_tower_conv1_weight, model_audio_tower_conv1_bias, [1], [1], [1], False, [0], 1);  model_audio_tower_conv1_weight = model_audio_tower_conv1_bias = None
	        mul_2: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution, 0.5)
	        mul_3: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution, 0.7071067811865476);  convolution = None
	        erf: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.erf.default(mul_3);  mul_3 = None
	        add_4: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.add.Tensor(erf, 1);  erf = None
	        mul_4: "f32[s6, 1280, 3000][3840000, 3000, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2, add_4);  mul_2 = add_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:350 in forward, code: inputs_embeds = nn.functional.gelu(self.conv2(inputs_embeds))
	        convolution_1: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.convolution.default(mul_4, model_audio_tower_conv2_weight, model_audio_tower_conv2_bias, [2], [1], [1], False, [0], 1);  mul_4 = model_audio_tower_conv2_weight = model_audio_tower_conv2_bias = None
	        mul_9: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution_1, 0.5)
	        mul_10: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.mul.Tensor(convolution_1, 0.7071067811865476);  convolution_1 = None
	        erf_1: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.erf.default(mul_10);  mul_10 = None
	        add_13: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_1, 1);  erf_1 = None
	        mul_11: "f32[s6, 1280, 1500][1920000, 1500, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9, add_13);  mul_9 = add_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:351 in forward, code: inputs_embeds = inputs_embeds.permute(0, 2, 1)
	        permute: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.permute.default(mul_11, [0, 2, 1]);  mul_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:354 in forward, code: hidden_states = (inputs_embeds + embed_pos).to(inputs_embeds.dtype)
	        add_22: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(permute, model_audio_tower_embed_positions_weight);  permute = model_audio_tower_embed_positions_weight = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_22, memory_format = torch.contiguous_format)
	        var_mean = torch.ops.aten.var_mean.correction(clone_1, [2], correction = 0, keepdim = True)
	        getitem: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean[0]
	        getitem_1: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean[1];  var_mean = None
	        sub_7: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_1, getitem_1);  clone_1 = getitem_1 = None
	        add_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem, 1e-05);  getitem = None
	        rsqrt: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_31);  add_31 = None
	        mul_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7, rsqrt);  sub_7 = rsqrt = None
	        mul_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_20, model_audio_tower_layers_0_self_attn_layer_norm_weight);  mul_20 = model_audio_tower_layers_0_self_attn_layer_norm_weight = None
	        add_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_21, model_audio_tower_layers_0_self_attn_layer_norm_bias);  mul_21 = model_audio_tower_layers_0_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_32, [2])
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:349 in forward, code: inputs_embeds = nn.functional.gelu(self.conv1(input_features))
	        sym_size_int: "Sym(s6)" = torch.ops.aten.sym_size.int(arg877_1, 0);  arg877_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        full_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax, full_1);  amax = full_1 = None
	        amin: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_32, [2])
	        full: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin, full);  amin = full = None
	        sub_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum, minimum);  maximum = None
	        div: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_18, 255.0);  sub_18 = None
	        clamp_min: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div, 1.1920928955078125e-07);  div = None
	        div_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum, clamp_min);  minimum = None
	        round_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_1);  div_1 = None
	        sub_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_1);  round_1 = None
	        clamp_min_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_24, -128);  sub_24 = None
	        clamp_max: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_1, 127);  clamp_min_1 = None
	        view_2: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min, [sym_size_int, 1500, 1])
	        reciprocal: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2);  view_2 = None
	        mul_69: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal, 1.0);  reciprocal = None
	        mul_72: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_32, mul_69);  mul_69 = None
	        round_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_72);  mul_72 = None
	        convert_element_type: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max, torch.int8);  clamp_max = None
	        view_3: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type, [sym_size_int, 1500, 1])
	        add_119: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_2, view_3);  round_2 = view_3 = None
	        clamp_min_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_119, -128);  add_119 = None
	        clamp_max_1: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_2, 127);  clamp_min_2 = None
	        convert_element_type_1: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_1, torch.int8);  clamp_max_1 = None
	        view_7: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type, [sym_size_int, 1500, 1]);  convert_element_type = None
	        convert_element_type_2: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1, torch.float32);  convert_element_type_1 = None
	        convert_element_type_3: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_7, torch.float32);  view_7 = None
	        sub_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_2, convert_element_type_3);  convert_element_type_2 = convert_element_type_3 = None
	        view_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min, [sym_size_int, 1500, 1]);  clamp_min = None
	        mul_94: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_44, view_6);  sub_44 = view_6 = None
	        view_9: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_11: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_4: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_9, torch.float32);  view_9 = None
	        convert_element_type_5: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_11, torch.float32);  view_11 = None
	        sub_48: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_4, convert_element_type_5);  convert_element_type_4 = convert_element_type_5 = None
	        view_10: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_99: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_48, view_10);  sub_48 = view_10 = None
	        view_12: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_99, [1280, 1280]);  mul_99 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_32, [2])
	        full_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_1, full_3);  amax_1 = full_3 = None
	        amin_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_32, [2])
	        full_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_1: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_1, full_2);  amin_1 = full_2 = None
	        sub_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_1, minimum_1);  maximum_1 = None
	        div_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_63, 255.0);  sub_63 = None
	        clamp_min_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_2, 1.1920928955078125e-07);  div_2 = None
	        div_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_1, clamp_min_3);  minimum_1 = None
	        round_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_3);  div_3 = None
	        sub_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_3);  round_3 = None
	        clamp_min_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_69, -128);  sub_69 = None
	        clamp_max_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_4, 127);  clamp_min_4 = None
	        view_18: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_3, [sym_size_int, 1500, 1])
	        reciprocal_1: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_18);  view_18 = None
	        mul_165: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_1, 1.0);  reciprocal_1 = None
	        mul_168: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_32, mul_165);  mul_165 = None
	        round_4: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_168);  mul_168 = None
	        convert_element_type_6: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_2, torch.int8);  clamp_max_2 = None
	        view_19: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_6, [sym_size_int, 1500, 1])
	        add_271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_4, view_19);  round_4 = view_19 = None
	        clamp_min_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_271, -128);  add_271 = None
	        clamp_max_3: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_5, 127);  clamp_min_5 = None
	        convert_element_type_7: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_3, torch.int8);  clamp_max_3 = None
	        view_23: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_6, [sym_size_int, 1500, 1]);  convert_element_type_6 = None
	        convert_element_type_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_7, torch.float32);  convert_element_type_7 = None
	        convert_element_type_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_23, torch.float32);  view_23 = None
	        sub_89: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_8, convert_element_type_9);  convert_element_type_8 = convert_element_type_9 = None
	        view_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_3, [sym_size_int, 1500, 1]);  clamp_min_3 = None
	        mul_190: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_89, view_22);  sub_89 = view_22 = None
	        view_25: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_27: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_10: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_25, torch.float32);  view_25 = None
	        convert_element_type_11: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_27, torch.float32);  view_27 = None
	        sub_93: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_10, convert_element_type_11);  convert_element_type_10 = convert_element_type_11 = None
	        view_26: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_195: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_93, view_26);  sub_93 = view_26 = None
	        view_28: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_195, [1280, 1280]);  mul_195 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_32, [2])
	        full_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_2, full_5);  amax_2 = full_5 = None
	        amin_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_32, [2])
	        full_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_2: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_2, full_4);  amin_2 = full_4 = None
	        sub_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_2, minimum_2);  maximum_2 = None
	        div_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_107, 255.0);  sub_107 = None
	        clamp_min_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_4, 1.1920928955078125e-07);  div_4 = None
	        div_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_2, clamp_min_6);  minimum_2 = None
	        round_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_5);  div_5 = None
	        sub_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_5);  round_5 = None
	        clamp_min_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_113, -128);  sub_113 = None
	        clamp_max_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_7, 127);  clamp_min_7 = None
	        view_34: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_6, [sym_size_int, 1500, 1])
	        reciprocal_2: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_34);  view_34 = None
	        mul_264: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_2, 1.0);  reciprocal_2 = None
	        mul_267: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_32, mul_264);  add_32 = mul_264 = None
	        round_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_267);  mul_267 = None
	        convert_element_type_12: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_4, torch.int8);  clamp_max_4 = None
	        view_35: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_12, [sym_size_int, 1500, 1])
	        add_419: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_6, view_35);  round_6 = view_35 = None
	        clamp_min_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_419, -128);  add_419 = None
	        clamp_max_5: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_8, 127);  clamp_min_8 = None
	        convert_element_type_13: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_5, torch.int8);  clamp_max_5 = None
	        view_39: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_12, [sym_size_int, 1500, 1]);  convert_element_type_12 = None
	        convert_element_type_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_13, torch.float32);  convert_element_type_13 = None
	        convert_element_type_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_39, torch.float32);  view_39 = None
	        sub_133: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_14, convert_element_type_15);  convert_element_type_14 = convert_element_type_15 = None
	        view_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_6, [sym_size_int, 1500, 1]);  clamp_min_6 = None
	        mul_289: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_133, view_38);  sub_133 = view_38 = None
	        view_41: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_43: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_16: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_41, torch.float32);  view_41 = None
	        convert_element_type_17: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_43, torch.float32);  view_43 = None
	        sub_137: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_16, convert_element_type_17);  convert_element_type_16 = convert_element_type_17 = None
	        view_42: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_294: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_137, view_42);  sub_137 = view_42 = None
	        view_44: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_294, [1280, 1280]);  mul_294 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_104: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_13: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_94, [mul_104, 1280]);  mul_94 = mul_104 = None
	        permute_1: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_12, [1, 0]);  view_12 = None
	        mm_default_159: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_13, permute_1);  view_13 = permute_1 = None
	        add_tensor_159: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_159, model_audio_tower_layers_0_self_attn_q_proj_bias);  mm_default_159 = model_audio_tower_layers_0_self_attn_q_proj_bias = None
	        view_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_159, [sym_size_int, 1500, 1280]);  add_tensor_159 = None
	        mul_111: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_14, 0.125);  view_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_15: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_111, [sym_size_int, 1500, 20, 64]);  mul_111 = None
	        permute_2: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_15, [0, 2, 1, 3]);  view_15 = None
	        clone_2: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_2, memory_format = torch.contiguous_format);  permute_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_198: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_29: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_190, [mul_198, 1280]);  mul_190 = mul_198 = None
	        permute_3: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_28, [1, 0]);  view_28 = None
	        mm: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_29, permute_3);  view_29 = permute_3 = None
	        view_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm, [sym_size_int, 1500, 1280]);  mm = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_31: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_30, [sym_size_int, -1, 20, 64]);  view_30 = None
	        permute_4: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_31, [0, 2, 1, 3]);  view_31 = None
	        clone_3: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_4, memory_format = torch.contiguous_format);  permute_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_299: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_45: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_289, [mul_299, 1280]);  mul_289 = mul_299 = None
	        permute_5: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_44, [1, 0]);  view_44 = None
	        mm_default_158: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_45, permute_5);  view_45 = permute_5 = None
	        add_tensor_158: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_158, model_audio_tower_layers_0_self_attn_v_proj_bias);  mm_default_158 = model_audio_tower_layers_0_self_attn_v_proj_bias = None
	        view_46: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_158, [sym_size_int, 1500, 1280]);  add_tensor_158 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_47: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_46, [sym_size_int, -1, 20, 64]);  view_46 = None
	        permute_6: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_47, [0, 2, 1, 3]);  view_47 = None
	        clone_4: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_6, memory_format = torch.contiguous_format);  permute_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_2, clone_3, clone_4, None, False, scale = 1.0);  clone_2 = clone_3 = clone_4 = None
	        getitem_2: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention[0];  _scaled_dot_product_efficient_attention = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_7: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_2, [0, 2, 1, 3]);  getitem_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_48: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_7, [sym_size_int, 1500, -1]);  permute_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_48, [2])
	        full_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_3, full_7);  amax_3 = full_7 = None
	        amin_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_48, [2])
	        full_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_3: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_3, full_6);  amin_3 = full_6 = None
	        sub_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_3, minimum_3);  maximum_3 = None
	        div_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_155, 255.0);  sub_155 = None
	        clamp_min_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_6, 1.1920928955078125e-07);  div_6 = None
	        div_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_3, clamp_min_9);  minimum_3 = None
	        round_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_7);  div_7 = None
	        sub_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_7);  round_7 = None
	        clamp_min_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_161, -128);  sub_161 = None
	        clamp_max_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_10, 127);  clamp_min_10 = None
	        view_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_9, [sym_size_int, 1500, 1])
	        reciprocal_3: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_51);  view_51 = None
	        mul_369: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_3, 1.0);  reciprocal_3 = None
	        mul_372: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_48, mul_369);  view_48 = mul_369 = None
	        round_8: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_372);  mul_372 = None
	        convert_element_type_18: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_6, torch.int8);  clamp_max_6 = None
	        view_52: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_18, [sym_size_int, 1500, 1])
	        add_583: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_8, view_52);  round_8 = view_52 = None
	        clamp_min_11: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_583, -128);  add_583 = None
	        clamp_max_7: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_11, 127);  clamp_min_11 = None
	        convert_element_type_19: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_7, torch.int8);  clamp_max_7 = None
	        view_56: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_18, [sym_size_int, 1500, 1]);  convert_element_type_18 = None
	        convert_element_type_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_19, torch.float32);  convert_element_type_19 = None
	        convert_element_type_21: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_56, torch.float32);  view_56 = None
	        sub_181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_20, convert_element_type_21);  convert_element_type_20 = convert_element_type_21 = None
	        view_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_9, [sym_size_int, 1500, 1]);  clamp_min_9 = None
	        mul_394: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_181, view_55);  sub_181 = view_55 = None
	        view_58: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_60: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_22: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_58, torch.float32);  view_58 = None
	        convert_element_type_23: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_60, torch.float32);  view_60 = None
	        sub_185: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_22, convert_element_type_23);  convert_element_type_22 = convert_element_type_23 = None
	        view_59: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_399: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_185, view_59);  sub_185 = view_59 = None
	        view_61: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_399, [1280, 1280]);  mul_399 = None
	        mul_404: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_62: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_394, [mul_404, 1280]);  mul_394 = mul_404 = None
	        permute_8: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_61, [1, 0]);  view_61 = None
	        mm_default_157: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_62, permute_8);  view_62 = permute_8 = None
	        add_tensor_157: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_157, model_audio_tower_layers_0_self_attn_out_proj_bias);  mm_default_157 = model_audio_tower_layers_0_self_attn_out_proj_bias = None
	        view_63: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_157, [sym_size_int, 1500, 1280]);  add_tensor_157 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_646: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_22, view_63);  add_22 = view_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_6: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_646, memory_format = torch.contiguous_format)
	        var_mean_1 = torch.ops.aten.var_mean.correction(clone_6, [2], correction = 0, keepdim = True)
	        getitem_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_1[0]
	        getitem_7: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_1[1];  var_mean_1 = None
	        sub_191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_6, getitem_7);  clone_6 = getitem_7 = None
	        add_651: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_6, 1e-05);  getitem_6 = None
	        rsqrt_1: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_651);  add_651 = None
	        mul_415: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_191, rsqrt_1);  sub_191 = rsqrt_1 = None
	        mul_416: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_415, model_audio_tower_layers_0_final_layer_norm_weight);  mul_415 = model_audio_tower_layers_0_final_layer_norm_weight = None
	        add_652: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_416, model_audio_tower_layers_0_final_layer_norm_bias);  mul_416 = model_audio_tower_layers_0_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_652, [2])
	        full_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_4, full_9);  amax_4 = full_9 = None
	        amin_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_652, [2])
	        full_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_4: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_4, full_8);  amin_4 = full_8 = None
	        sub_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_4, minimum_4);  maximum_4 = None
	        div_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_202, 255.0);  sub_202 = None
	        clamp_min_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_8, 1.1920928955078125e-07);  div_8 = None
	        div_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_4, clamp_min_12);  minimum_4 = None
	        round_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_9);  div_9 = None
	        sub_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_9);  round_9 = None
	        clamp_min_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_208, -128);  sub_208 = None
	        clamp_max_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_13, 127);  clamp_min_13 = None
	        view_66: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_12, [sym_size_int, 1500, 1])
	        reciprocal_4: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_66);  view_66 = None
	        mul_464: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_4, 1.0);  reciprocal_4 = None
	        mul_467: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_652, mul_464);  add_652 = mul_464 = None
	        round_10: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_467);  mul_467 = None
	        convert_element_type_24: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_8, torch.int8);  clamp_max_8 = None
	        view_67: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_24, [sym_size_int, 1500, 1])
	        add_739: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_10, view_67);  round_10 = view_67 = None
	        clamp_min_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_739, -128);  add_739 = None
	        clamp_max_9: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_14, 127);  clamp_min_14 = None
	        convert_element_type_25: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_9, torch.int8);  clamp_max_9 = None
	        view_71: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_24, [sym_size_int, 1500, 1]);  convert_element_type_24 = None
	        convert_element_type_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_25, torch.float32);  convert_element_type_25 = None
	        convert_element_type_27: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_71, torch.float32);  view_71 = None
	        sub_228: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_26, convert_element_type_27);  convert_element_type_26 = convert_element_type_27 = None
	        view_70: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_12, [sym_size_int, 1500, 1]);  clamp_min_12 = None
	        mul_489: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_228, view_70);  sub_228 = view_70 = None
	        view_73: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_0_fc1_parametrizations_weight_original0 = None
	        view_75: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_0_fc1_parametrizations_weight_original2 = None
	        convert_element_type_28: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_73, torch.float32);  view_73 = None
	        convert_element_type_29: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_75, torch.float32);  view_75 = None
	        sub_232: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_28, convert_element_type_29);  convert_element_type_28 = convert_element_type_29 = None
	        view_74: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_0_fc1_parametrizations_weight_original1 = None
	        mul_494: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_232, view_74);  sub_232 = view_74 = None
	        view_76: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_494, [5120, 1280]);  mul_494 = None
	        mul_499: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_77: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_489, [mul_499, 1280]);  mul_489 = mul_499 = None
	        permute_9: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_76, [1, 0]);  view_76 = None
	        mm_default_156: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_77, permute_9);  view_77 = permute_9 = None
	        add_tensor_156: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_156, model_audio_tower_layers_0_fc1_bias);  mm_default_156 = model_audio_tower_layers_0_fc1_bias = None
	        view_78: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_156, [sym_size_int, 1500, 5120]);  add_tensor_156 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_506: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_78, 0.5)
	        mul_507: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_78, 0.7071067811865476);  view_78 = None
	        erf_2: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_507);  mul_507 = None
	        add_798: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_2, 1);  erf_2 = None
	        mul_508: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_506, add_798);  mul_506 = add_798 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_508, [2])
	        full_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_5, full_11);  amax_5 = full_11 = None
	        amin_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_508, [2])
	        full_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_5: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_5, full_10);  amin_5 = full_10 = None
	        sub_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_5, minimum_5);  maximum_5 = None
	        div_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_245, 255.0);  sub_245 = None
	        clamp_min_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_10, 1.1920928955078125e-07);  div_10 = None
	        div_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_5, clamp_min_15);  minimum_5 = None
	        round_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_11);  div_11 = None
	        sub_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_11);  round_11 = None
	        clamp_min_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_251, -128);  sub_251 = None
	        clamp_max_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_16, 127);  clamp_min_16 = None
	        view_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_15, [sym_size_int, 1500, 1])
	        reciprocal_5: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_81);  view_81 = None
	        mul_554: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_5, 1.0);  reciprocal_5 = None
	        mul_557: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_508, mul_554);  mul_508 = mul_554 = None
	        round_12: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_557);  mul_557 = None
	        convert_element_type_30: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_10, torch.int8);  clamp_max_10 = None
	        view_82: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_30, [sym_size_int, 1500, 1])
	        add_881: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_12, view_82);  round_12 = view_82 = None
	        clamp_min_17: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_881, -128);  add_881 = None
	        clamp_max_11: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_17, 127);  clamp_min_17 = None
	        convert_element_type_31: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_11, torch.int8);  clamp_max_11 = None
	        view_86: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_30, [sym_size_int, 1500, 1]);  convert_element_type_30 = None
	        convert_element_type_32: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_31, torch.float32);  convert_element_type_31 = None
	        convert_element_type_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_86, torch.float32);  view_86 = None
	        sub_271: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_32, convert_element_type_33);  convert_element_type_32 = convert_element_type_33 = None
	        view_85: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_15, [sym_size_int, 1500, 1]);  clamp_min_15 = None
	        mul_579: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_271, view_85);  sub_271 = view_85 = None
	        view_88: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_0_fc2_parametrizations_weight_original0 = None
	        view_90: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_0_fc2_parametrizations_weight_original2 = None
	        convert_element_type_34: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_88, torch.float32);  view_88 = None
	        convert_element_type_35: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_90, torch.float32);  view_90 = None
	        sub_275: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_34, convert_element_type_35);  convert_element_type_34 = convert_element_type_35 = None
	        view_89: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_0_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_0_fc2_parametrizations_weight_original1 = None
	        mul_584: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_275, view_89);  sub_275 = view_89 = None
	        view_91: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_584, [1280, 5120]);  mul_584 = None
	        mul_589: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_92: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_579, [mul_589, 5120]);  mul_579 = mul_589 = None
	        permute_10: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_91, [1, 0]);  view_91 = None
	        mm_default_155: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_92, permute_10);  view_92 = permute_10 = None
	        add_tensor_155: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_155, model_audio_tower_layers_0_fc2_bias);  mm_default_155 = model_audio_tower_layers_0_fc2_bias = None
	        view_93: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_155, [sym_size_int, 1500, 1280]);  add_tensor_155 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_944: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_646, view_93);  add_646 = view_93 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_9: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_944, memory_format = torch.contiguous_format)
	        var_mean_2 = torch.ops.aten.var_mean.correction(clone_9, [2], correction = 0, keepdim = True)
	        getitem_8: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_2[0]
	        getitem_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_2[1];  var_mean_2 = None
	        sub_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_9, getitem_9);  clone_9 = getitem_9 = None
	        add_949: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_8, 1e-05);  getitem_8 = None
	        rsqrt_2: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_949);  add_949 = None
	        mul_600: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_281, rsqrt_2);  sub_281 = rsqrt_2 = None
	        mul_601: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_600, model_audio_tower_layers_1_self_attn_layer_norm_weight);  mul_600 = model_audio_tower_layers_1_self_attn_layer_norm_weight = None
	        add_950: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_601, model_audio_tower_layers_1_self_attn_layer_norm_bias);  mul_601 = model_audio_tower_layers_1_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_950, [2])
	        full_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_6, full_13);  amax_6 = full_13 = None
	        amin_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_950, [2])
	        full_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_6: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_6, full_12);  amin_6 = full_12 = None
	        sub_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_6, minimum_6);  maximum_6 = None
	        div_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_292, 255.0);  sub_292 = None
	        clamp_min_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_12, 1.1920928955078125e-07);  div_12 = None
	        div_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_6, clamp_min_18);  minimum_6 = None
	        round_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_13);  div_13 = None
	        sub_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_13);  round_13 = None
	        clamp_min_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_298, -128);  sub_298 = None
	        clamp_max_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_19, 127);  clamp_min_19 = None
	        view_96: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_18, [sym_size_int, 1500, 1])
	        reciprocal_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_96);  view_96 = None
	        mul_649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_6, 1.0);  reciprocal_6 = None
	        mul_652: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_950, mul_649);  mul_649 = None
	        round_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_652);  mul_652 = None
	        convert_element_type_36: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_12, torch.int8);  clamp_max_12 = None
	        view_97: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_36, [sym_size_int, 1500, 1])
	        add_1037: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_14, view_97);  round_14 = view_97 = None
	        clamp_min_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1037, -128);  add_1037 = None
	        clamp_max_13: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_20, 127);  clamp_min_20 = None
	        convert_element_type_37: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_13, torch.int8);  clamp_max_13 = None
	        view_101: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_36, [sym_size_int, 1500, 1]);  convert_element_type_36 = None
	        convert_element_type_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_37, torch.float32);  convert_element_type_37 = None
	        convert_element_type_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_101, torch.float32);  view_101 = None
	        sub_318: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_38, convert_element_type_39);  convert_element_type_38 = convert_element_type_39 = None
	        view_100: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_18, [sym_size_int, 1500, 1]);  clamp_min_18 = None
	        mul_674: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_318, view_100);  sub_318 = view_100 = None
	        view_103: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_105: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_40: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_103, torch.float32);  view_103 = None
	        convert_element_type_41: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_105, torch.float32);  view_105 = None
	        sub_322: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_40, convert_element_type_41);  convert_element_type_40 = convert_element_type_41 = None
	        view_104: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_679: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_322, view_104);  sub_322 = view_104 = None
	        view_106: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_679, [1280, 1280]);  mul_679 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_950, [2])
	        full_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_7, full_15);  amax_7 = full_15 = None
	        amin_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_950, [2])
	        full_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_7: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_7, full_14);  amin_7 = full_14 = None
	        sub_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_7, minimum_7);  maximum_7 = None
	        div_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_337, 255.0);  sub_337 = None
	        clamp_min_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_14, 1.1920928955078125e-07);  div_14 = None
	        div_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_7, clamp_min_21);  minimum_7 = None
	        round_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_15);  div_15 = None
	        sub_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_15);  round_15 = None
	        clamp_min_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_343, -128);  sub_343 = None
	        clamp_max_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_22, 127);  clamp_min_22 = None
	        view_112: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_21, [sym_size_int, 1500, 1])
	        reciprocal_7: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_112);  view_112 = None
	        mul_745: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_7, 1.0);  reciprocal_7 = None
	        mul_748: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_950, mul_745);  mul_745 = None
	        round_16: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_748);  mul_748 = None
	        convert_element_type_42: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_14, torch.int8);  clamp_max_14 = None
	        view_113: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_42, [sym_size_int, 1500, 1])
	        add_1189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_16, view_113);  round_16 = view_113 = None
	        clamp_min_23: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1189, -128);  add_1189 = None
	        clamp_max_15: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_23, 127);  clamp_min_23 = None
	        convert_element_type_43: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_15, torch.int8);  clamp_max_15 = None
	        view_117: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_42, [sym_size_int, 1500, 1]);  convert_element_type_42 = None
	        convert_element_type_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_43, torch.float32);  convert_element_type_43 = None
	        convert_element_type_45: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_117, torch.float32);  view_117 = None
	        sub_363: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_44, convert_element_type_45);  convert_element_type_44 = convert_element_type_45 = None
	        view_116: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_21, [sym_size_int, 1500, 1]);  clamp_min_21 = None
	        mul_770: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_363, view_116);  sub_363 = view_116 = None
	        view_119: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_121: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_46: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_119, torch.float32);  view_119 = None
	        convert_element_type_47: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_121, torch.float32);  view_121 = None
	        sub_367: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_46, convert_element_type_47);  convert_element_type_46 = convert_element_type_47 = None
	        view_120: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_775: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_367, view_120);  sub_367 = view_120 = None
	        view_122: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_775, [1280, 1280]);  mul_775 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_950, [2])
	        full_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_8, full_17);  amax_8 = full_17 = None
	        amin_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_950, [2])
	        full_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_8: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_8, full_16);  amin_8 = full_16 = None
	        sub_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_8, minimum_8);  maximum_8 = None
	        div_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_381, 255.0);  sub_381 = None
	        clamp_min_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_16, 1.1920928955078125e-07);  div_16 = None
	        div_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_8, clamp_min_24);  minimum_8 = None
	        round_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_17);  div_17 = None
	        sub_387: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_17);  round_17 = None
	        clamp_min_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_387, -128);  sub_387 = None
	        clamp_max_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_25, 127);  clamp_min_25 = None
	        view_128: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_24, [sym_size_int, 1500, 1])
	        reciprocal_8: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_128);  view_128 = None
	        mul_844: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_8, 1.0);  reciprocal_8 = None
	        mul_847: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_950, mul_844);  add_950 = mul_844 = None
	        round_18: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_847);  mul_847 = None
	        convert_element_type_48: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_16, torch.int8);  clamp_max_16 = None
	        view_129: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_48, [sym_size_int, 1500, 1])
	        add_1337: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_18, view_129);  round_18 = view_129 = None
	        clamp_min_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1337, -128);  add_1337 = None
	        clamp_max_17: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_26, 127);  clamp_min_26 = None
	        convert_element_type_49: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_17, torch.int8);  clamp_max_17 = None
	        view_133: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_48, [sym_size_int, 1500, 1]);  convert_element_type_48 = None
	        convert_element_type_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_49, torch.float32);  convert_element_type_49 = None
	        convert_element_type_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_133, torch.float32);  view_133 = None
	        sub_407: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_50, convert_element_type_51);  convert_element_type_50 = convert_element_type_51 = None
	        view_132: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_24, [sym_size_int, 1500, 1]);  clamp_min_24 = None
	        mul_869: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_407, view_132);  sub_407 = view_132 = None
	        view_135: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_137: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_52: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_135, torch.float32);  view_135 = None
	        convert_element_type_53: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_137, torch.float32);  view_137 = None
	        sub_411: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_52, convert_element_type_53);  convert_element_type_52 = convert_element_type_53 = None
	        view_136: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_874: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_411, view_136);  sub_411 = view_136 = None
	        view_138: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_874, [1280, 1280]);  mul_874 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_684: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_107: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_674, [mul_684, 1280]);  mul_674 = mul_684 = None
	        permute_11: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_106, [1, 0]);  view_106 = None
	        mm_default_154: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_107, permute_11);  view_107 = permute_11 = None
	        add_tensor_154: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_154, model_audio_tower_layers_1_self_attn_q_proj_bias);  mm_default_154 = model_audio_tower_layers_1_self_attn_q_proj_bias = None
	        view_108: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_154, [sym_size_int, 1500, 1280]);  add_tensor_154 = None
	        mul_691: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_108, 0.125);  view_108 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_109: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_691, [sym_size_int, 1500, 20, 64]);  mul_691 = None
	        permute_12: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_109, [0, 2, 1, 3]);  view_109 = None
	        clone_10: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_12, memory_format = torch.contiguous_format);  permute_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_778: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_123: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_770, [mul_778, 1280]);  mul_770 = mul_778 = None
	        permute_13: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_122, [1, 0]);  view_122 = None
	        mm_1: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_123, permute_13);  view_123 = permute_13 = None
	        view_124: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_1, [sym_size_int, 1500, 1280]);  mm_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_125: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_124, [sym_size_int, -1, 20, 64]);  view_124 = None
	        permute_14: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_125, [0, 2, 1, 3]);  view_125 = None
	        clone_11: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_14, memory_format = torch.contiguous_format);  permute_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_879: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_139: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_869, [mul_879, 1280]);  mul_869 = mul_879 = None
	        permute_15: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_138, [1, 0]);  view_138 = None
	        mm_default_153: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_139, permute_15);  view_139 = permute_15 = None
	        add_tensor_153: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_153, model_audio_tower_layers_1_self_attn_v_proj_bias);  mm_default_153 = model_audio_tower_layers_1_self_attn_v_proj_bias = None
	        view_140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_153, [sym_size_int, 1500, 1280]);  add_tensor_153 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_141: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_140, [sym_size_int, -1, 20, 64]);  view_140 = None
	        permute_16: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_141, [0, 2, 1, 3]);  view_141 = None
	        clone_12: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_16, memory_format = torch.contiguous_format);  permute_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_1 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_10, clone_11, clone_12, None, False, scale = 1.0);  clone_10 = clone_11 = clone_12 = None
	        getitem_10: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_1[0];  _scaled_dot_product_efficient_attention_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_17: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_10, [0, 2, 1, 3]);  getitem_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_17, [sym_size_int, 1500, -1]);  permute_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_142, [2])
	        full_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_9, full_19);  amax_9 = full_19 = None
	        amin_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_142, [2])
	        full_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_9: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_9, full_18);  amin_9 = full_18 = None
	        sub_429: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_9, minimum_9);  maximum_9 = None
	        div_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_429, 255.0);  sub_429 = None
	        clamp_min_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_18, 1.1920928955078125e-07);  div_18 = None
	        div_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_9, clamp_min_27);  minimum_9 = None
	        round_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_19);  div_19 = None
	        sub_435: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_19);  round_19 = None
	        clamp_min_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_435, -128);  sub_435 = None
	        clamp_max_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_28, 127);  clamp_min_28 = None
	        view_145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_27, [sym_size_int, 1500, 1])
	        reciprocal_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_145);  view_145 = None
	        mul_949: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_9, 1.0);  reciprocal_9 = None
	        mul_952: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_142, mul_949);  view_142 = mul_949 = None
	        round_20: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_952);  mul_952 = None
	        convert_element_type_54: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_18, torch.int8);  clamp_max_18 = None
	        view_146: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_54, [sym_size_int, 1500, 1])
	        add_1501: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_20, view_146);  round_20 = view_146 = None
	        clamp_min_29: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1501, -128);  add_1501 = None
	        clamp_max_19: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_29, 127);  clamp_min_29 = None
	        convert_element_type_55: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_19, torch.int8);  clamp_max_19 = None
	        view_150: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_54, [sym_size_int, 1500, 1]);  convert_element_type_54 = None
	        convert_element_type_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_55, torch.float32);  convert_element_type_55 = None
	        convert_element_type_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_150, torch.float32);  view_150 = None
	        sub_455: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_56, convert_element_type_57);  convert_element_type_56 = convert_element_type_57 = None
	        view_149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_27, [sym_size_int, 1500, 1]);  clamp_min_27 = None
	        mul_974: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_455, view_149);  sub_455 = view_149 = None
	        view_152: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_154: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_58: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_152, torch.float32);  view_152 = None
	        convert_element_type_59: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_154, torch.float32);  view_154 = None
	        sub_459: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_58, convert_element_type_59);  convert_element_type_58 = convert_element_type_59 = None
	        view_153: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_979: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_459, view_153);  sub_459 = view_153 = None
	        view_155: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_979, [1280, 1280]);  mul_979 = None
	        mul_984: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_156: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_974, [mul_984, 1280]);  mul_974 = mul_984 = None
	        permute_18: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_155, [1, 0]);  view_155 = None
	        mm_default_152: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_156, permute_18);  view_156 = permute_18 = None
	        add_tensor_152: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_152, model_audio_tower_layers_1_self_attn_out_proj_bias);  mm_default_152 = model_audio_tower_layers_1_self_attn_out_proj_bias = None
	        view_157: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_152, [sym_size_int, 1500, 1280]);  add_tensor_152 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_1564: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_944, view_157);  add_944 = view_157 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_14: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_1564, memory_format = torch.contiguous_format)
	        var_mean_3 = torch.ops.aten.var_mean.correction(clone_14, [2], correction = 0, keepdim = True)
	        getitem_14: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_3[0]
	        getitem_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_3[1];  var_mean_3 = None
	        sub_465: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_14, getitem_15);  clone_14 = getitem_15 = None
	        add_1569: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_14, 1e-05);  getitem_14 = None
	        rsqrt_3: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_1569);  add_1569 = None
	        mul_995: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_465, rsqrt_3);  sub_465 = rsqrt_3 = None
	        mul_996: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_995, model_audio_tower_layers_1_final_layer_norm_weight);  mul_995 = model_audio_tower_layers_1_final_layer_norm_weight = None
	        add_1570: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_996, model_audio_tower_layers_1_final_layer_norm_bias);  mul_996 = model_audio_tower_layers_1_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_1570, [2])
	        full_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_10, full_21);  amax_10 = full_21 = None
	        amin_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_1570, [2])
	        full_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_10: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_10, full_20);  amin_10 = full_20 = None
	        sub_476: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_10, minimum_10);  maximum_10 = None
	        div_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_476, 255.0);  sub_476 = None
	        clamp_min_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_20, 1.1920928955078125e-07);  div_20 = None
	        div_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_10, clamp_min_30);  minimum_10 = None
	        round_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_21);  div_21 = None
	        sub_482: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_21);  round_21 = None
	        clamp_min_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_482, -128);  sub_482 = None
	        clamp_max_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_31, 127);  clamp_min_31 = None
	        view_160: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_30, [sym_size_int, 1500, 1])
	        reciprocal_10: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_160);  view_160 = None
	        mul_1044: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_10, 1.0);  reciprocal_10 = None
	        mul_1047: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_1570, mul_1044);  add_1570 = mul_1044 = None
	        round_22: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1047);  mul_1047 = None
	        convert_element_type_60: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_20, torch.int8);  clamp_max_20 = None
	        view_161: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_60, [sym_size_int, 1500, 1])
	        add_1657: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_22, view_161);  round_22 = view_161 = None
	        clamp_min_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1657, -128);  add_1657 = None
	        clamp_max_21: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_32, 127);  clamp_min_32 = None
	        convert_element_type_61: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_21, torch.int8);  clamp_max_21 = None
	        view_165: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_60, [sym_size_int, 1500, 1]);  convert_element_type_60 = None
	        convert_element_type_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_61, torch.float32);  convert_element_type_61 = None
	        convert_element_type_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_165, torch.float32);  view_165 = None
	        sub_502: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_62, convert_element_type_63);  convert_element_type_62 = convert_element_type_63 = None
	        view_164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_30, [sym_size_int, 1500, 1]);  clamp_min_30 = None
	        mul_1069: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_502, view_164);  sub_502 = view_164 = None
	        view_167: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_1_fc1_parametrizations_weight_original0 = None
	        view_169: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_1_fc1_parametrizations_weight_original2 = None
	        convert_element_type_64: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_167, torch.float32);  view_167 = None
	        convert_element_type_65: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_169, torch.float32);  view_169 = None
	        sub_506: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_64, convert_element_type_65);  convert_element_type_64 = convert_element_type_65 = None
	        view_168: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_1_fc1_parametrizations_weight_original1 = None
	        mul_1074: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_506, view_168);  sub_506 = view_168 = None
	        view_170: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1074, [5120, 1280]);  mul_1074 = None
	        mul_1079: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_171: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1069, [mul_1079, 1280]);  mul_1069 = mul_1079 = None
	        permute_19: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_170, [1, 0]);  view_170 = None
	        mm_default_151: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_171, permute_19);  view_171 = permute_19 = None
	        add_tensor_151: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_151, model_audio_tower_layers_1_fc1_bias);  mm_default_151 = model_audio_tower_layers_1_fc1_bias = None
	        view_172: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_151, [sym_size_int, 1500, 5120]);  add_tensor_151 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_1086: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_172, 0.5)
	        mul_1087: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_172, 0.7071067811865476);  view_172 = None
	        erf_3: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_1087);  mul_1087 = None
	        add_1716: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_3, 1);  erf_3 = None
	        mul_1088: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1086, add_1716);  mul_1086 = add_1716 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_1088, [2])
	        full_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_11, full_23);  amax_11 = full_23 = None
	        amin_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_1088, [2])
	        full_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_11: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_11, full_22);  amin_11 = full_22 = None
	        sub_519: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_11, minimum_11);  maximum_11 = None
	        div_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_519, 255.0);  sub_519 = None
	        clamp_min_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_22, 1.1920928955078125e-07);  div_22 = None
	        div_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_11, clamp_min_33);  minimum_11 = None
	        round_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_23);  div_23 = None
	        sub_525: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_23);  round_23 = None
	        clamp_min_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_525, -128);  sub_525 = None
	        clamp_max_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_34, 127);  clamp_min_34 = None
	        view_175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_33, [sym_size_int, 1500, 1])
	        reciprocal_11: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_175);  view_175 = None
	        mul_1134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_11, 1.0);  reciprocal_11 = None
	        mul_1137: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1088, mul_1134);  mul_1088 = mul_1134 = None
	        round_24: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_1137);  mul_1137 = None
	        convert_element_type_66: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_22, torch.int8);  clamp_max_22 = None
	        view_176: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_66, [sym_size_int, 1500, 1])
	        add_1799: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_24, view_176);  round_24 = view_176 = None
	        clamp_min_35: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1799, -128);  add_1799 = None
	        clamp_max_23: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_35, 127);  clamp_min_35 = None
	        convert_element_type_67: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_23, torch.int8);  clamp_max_23 = None
	        view_180: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_66, [sym_size_int, 1500, 1]);  convert_element_type_66 = None
	        convert_element_type_68: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_67, torch.float32);  convert_element_type_67 = None
	        convert_element_type_69: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_180, torch.float32);  view_180 = None
	        sub_545: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_68, convert_element_type_69);  convert_element_type_68 = convert_element_type_69 = None
	        view_179: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_33, [sym_size_int, 1500, 1]);  clamp_min_33 = None
	        mul_1159: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_545, view_179);  sub_545 = view_179 = None
	        view_182: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_1_fc2_parametrizations_weight_original0 = None
	        view_184: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_1_fc2_parametrizations_weight_original2 = None
	        convert_element_type_70: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_182, torch.float32);  view_182 = None
	        convert_element_type_71: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_184, torch.float32);  view_184 = None
	        sub_549: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_70, convert_element_type_71);  convert_element_type_70 = convert_element_type_71 = None
	        view_183: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_1_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_1_fc2_parametrizations_weight_original1 = None
	        mul_1164: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_549, view_183);  sub_549 = view_183 = None
	        view_185: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1164, [1280, 5120]);  mul_1164 = None
	        mul_1169: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_186: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1159, [mul_1169, 5120]);  mul_1159 = mul_1169 = None
	        permute_20: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_185, [1, 0]);  view_185 = None
	        mm_default_150: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_186, permute_20);  view_186 = permute_20 = None
	        add_tensor_150: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_150, model_audio_tower_layers_1_fc2_bias);  mm_default_150 = model_audio_tower_layers_1_fc2_bias = None
	        view_187: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_150, [sym_size_int, 1500, 1280]);  add_tensor_150 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_1862: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_1564, view_187);  add_1564 = view_187 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_17: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_1862, memory_format = torch.contiguous_format)
	        var_mean_4 = torch.ops.aten.var_mean.correction(clone_17, [2], correction = 0, keepdim = True)
	        getitem_16: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_4[0]
	        getitem_17: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_4[1];  var_mean_4 = None
	        sub_555: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_17, getitem_17);  clone_17 = getitem_17 = None
	        add_1867: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_16, 1e-05);  getitem_16 = None
	        rsqrt_4: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_1867);  add_1867 = None
	        mul_1180: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_555, rsqrt_4);  sub_555 = rsqrt_4 = None
	        mul_1181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1180, model_audio_tower_layers_2_self_attn_layer_norm_weight);  mul_1180 = model_audio_tower_layers_2_self_attn_layer_norm_weight = None
	        add_1868: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_1181, model_audio_tower_layers_2_self_attn_layer_norm_bias);  mul_1181 = model_audio_tower_layers_2_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_1868, [2])
	        full_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_12, full_25);  amax_12 = full_25 = None
	        amin_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_1868, [2])
	        full_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_12: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_12, full_24);  amin_12 = full_24 = None
	        sub_566: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_12, minimum_12);  maximum_12 = None
	        div_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_566, 255.0);  sub_566 = None
	        clamp_min_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_24, 1.1920928955078125e-07);  div_24 = None
	        div_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_12, clamp_min_36);  minimum_12 = None
	        round_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_25);  div_25 = None
	        sub_572: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_25);  round_25 = None
	        clamp_min_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_572, -128);  sub_572 = None
	        clamp_max_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_37, 127);  clamp_min_37 = None
	        view_190: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_36, [sym_size_int, 1500, 1])
	        reciprocal_12: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_190);  view_190 = None
	        mul_1229: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_12, 1.0);  reciprocal_12 = None
	        mul_1232: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_1868, mul_1229);  mul_1229 = None
	        round_26: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1232);  mul_1232 = None
	        convert_element_type_72: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_24, torch.int8);  clamp_max_24 = None
	        view_191: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_72, [sym_size_int, 1500, 1])
	        add_1955: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_26, view_191);  round_26 = view_191 = None
	        clamp_min_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_1955, -128);  add_1955 = None
	        clamp_max_25: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_38, 127);  clamp_min_38 = None
	        convert_element_type_73: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_25, torch.int8);  clamp_max_25 = None
	        view_195: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_72, [sym_size_int, 1500, 1]);  convert_element_type_72 = None
	        convert_element_type_74: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_73, torch.float32);  convert_element_type_73 = None
	        convert_element_type_75: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_195, torch.float32);  view_195 = None
	        sub_592: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_74, convert_element_type_75);  convert_element_type_74 = convert_element_type_75 = None
	        view_194: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_36, [sym_size_int, 1500, 1]);  clamp_min_36 = None
	        mul_1254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_592, view_194);  sub_592 = view_194 = None
	        view_197: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_199: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_76: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_197, torch.float32);  view_197 = None
	        convert_element_type_77: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_199, torch.float32);  view_199 = None
	        sub_596: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_76, convert_element_type_77);  convert_element_type_76 = convert_element_type_77 = None
	        view_198: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_1259: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_596, view_198);  sub_596 = view_198 = None
	        view_200: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1259, [1280, 1280]);  mul_1259 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_1868, [2])
	        full_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_13, full_27);  amax_13 = full_27 = None
	        amin_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_1868, [2])
	        full_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_13: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_13, full_26);  amin_13 = full_26 = None
	        sub_611: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_13, minimum_13);  maximum_13 = None
	        div_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_611, 255.0);  sub_611 = None
	        clamp_min_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_26, 1.1920928955078125e-07);  div_26 = None
	        div_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_13, clamp_min_39);  minimum_13 = None
	        round_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_27);  div_27 = None
	        sub_617: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_27);  round_27 = None
	        clamp_min_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_617, -128);  sub_617 = None
	        clamp_max_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_40, 127);  clamp_min_40 = None
	        view_206: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_39, [sym_size_int, 1500, 1])
	        reciprocal_13: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_206);  view_206 = None
	        mul_1325: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_13, 1.0);  reciprocal_13 = None
	        mul_1328: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_1868, mul_1325);  mul_1325 = None
	        round_28: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1328);  mul_1328 = None
	        convert_element_type_78: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_26, torch.int8);  clamp_max_26 = None
	        view_207: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_78, [sym_size_int, 1500, 1])
	        add_2107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_28, view_207);  round_28 = view_207 = None
	        clamp_min_41: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2107, -128);  add_2107 = None
	        clamp_max_27: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_41, 127);  clamp_min_41 = None
	        convert_element_type_79: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_27, torch.int8);  clamp_max_27 = None
	        view_211: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_78, [sym_size_int, 1500, 1]);  convert_element_type_78 = None
	        convert_element_type_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_79, torch.float32);  convert_element_type_79 = None
	        convert_element_type_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_211, torch.float32);  view_211 = None
	        sub_637: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_80, convert_element_type_81);  convert_element_type_80 = convert_element_type_81 = None
	        view_210: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_39, [sym_size_int, 1500, 1]);  clamp_min_39 = None
	        mul_1350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_637, view_210);  sub_637 = view_210 = None
	        view_213: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_215: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_82: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_213, torch.float32);  view_213 = None
	        convert_element_type_83: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_215, torch.float32);  view_215 = None
	        sub_641: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_82, convert_element_type_83);  convert_element_type_82 = convert_element_type_83 = None
	        view_214: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_1355: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_641, view_214);  sub_641 = view_214 = None
	        view_216: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1355, [1280, 1280]);  mul_1355 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_1868, [2])
	        full_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_14, full_29);  amax_14 = full_29 = None
	        amin_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_1868, [2])
	        full_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_14: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_14, full_28);  amin_14 = full_28 = None
	        sub_655: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_14, minimum_14);  maximum_14 = None
	        div_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_655, 255.0);  sub_655 = None
	        clamp_min_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_28, 1.1920928955078125e-07);  div_28 = None
	        div_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_14, clamp_min_42);  minimum_14 = None
	        round_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_29);  div_29 = None
	        sub_661: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_29);  round_29 = None
	        clamp_min_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_661, -128);  sub_661 = None
	        clamp_max_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_43, 127);  clamp_min_43 = None
	        view_222: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_42, [sym_size_int, 1500, 1])
	        reciprocal_14: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_222);  view_222 = None
	        mul_1424: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_14, 1.0);  reciprocal_14 = None
	        mul_1427: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_1868, mul_1424);  add_1868 = mul_1424 = None
	        round_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1427);  mul_1427 = None
	        convert_element_type_84: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_28, torch.int8);  clamp_max_28 = None
	        view_223: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_84, [sym_size_int, 1500, 1])
	        add_2255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_30, view_223);  round_30 = view_223 = None
	        clamp_min_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2255, -128);  add_2255 = None
	        clamp_max_29: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_44, 127);  clamp_min_44 = None
	        convert_element_type_85: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_29, torch.int8);  clamp_max_29 = None
	        view_227: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_84, [sym_size_int, 1500, 1]);  convert_element_type_84 = None
	        convert_element_type_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_85, torch.float32);  convert_element_type_85 = None
	        convert_element_type_87: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_227, torch.float32);  view_227 = None
	        sub_681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_86, convert_element_type_87);  convert_element_type_86 = convert_element_type_87 = None
	        view_226: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_42, [sym_size_int, 1500, 1]);  clamp_min_42 = None
	        mul_1449: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_681, view_226);  sub_681 = view_226 = None
	        view_229: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_231: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_88: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_229, torch.float32);  view_229 = None
	        convert_element_type_89: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_231, torch.float32);  view_231 = None
	        sub_685: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_88, convert_element_type_89);  convert_element_type_88 = convert_element_type_89 = None
	        view_230: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_1454: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_685, view_230);  sub_685 = view_230 = None
	        view_232: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1454, [1280, 1280]);  mul_1454 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_1264: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_201: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1254, [mul_1264, 1280]);  mul_1254 = mul_1264 = None
	        permute_21: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_200, [1, 0]);  view_200 = None
	        mm_default_149: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_201, permute_21);  view_201 = permute_21 = None
	        add_tensor_149: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_149, model_audio_tower_layers_2_self_attn_q_proj_bias);  mm_default_149 = model_audio_tower_layers_2_self_attn_q_proj_bias = None
	        view_202: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_149, [sym_size_int, 1500, 1280]);  add_tensor_149 = None
	        mul_1271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_202, 0.125);  view_202 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_203: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1271, [sym_size_int, 1500, 20, 64]);  mul_1271 = None
	        permute_22: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_203, [0, 2, 1, 3]);  view_203 = None
	        clone_18: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_22, memory_format = torch.contiguous_format);  permute_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_1358: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_217: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1350, [mul_1358, 1280]);  mul_1350 = mul_1358 = None
	        permute_23: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_216, [1, 0]);  view_216 = None
	        mm_2: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_217, permute_23);  view_217 = permute_23 = None
	        view_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_2, [sym_size_int, 1500, 1280]);  mm_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_219: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_218, [sym_size_int, -1, 20, 64]);  view_218 = None
	        permute_24: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_219, [0, 2, 1, 3]);  view_219 = None
	        clone_19: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_24, memory_format = torch.contiguous_format);  permute_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_1459: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_233: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1449, [mul_1459, 1280]);  mul_1449 = mul_1459 = None
	        permute_25: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_232, [1, 0]);  view_232 = None
	        mm_default_148: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_233, permute_25);  view_233 = permute_25 = None
	        add_tensor_148: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_148, model_audio_tower_layers_2_self_attn_v_proj_bias);  mm_default_148 = model_audio_tower_layers_2_self_attn_v_proj_bias = None
	        view_234: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_148, [sym_size_int, 1500, 1280]);  add_tensor_148 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_235: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_234, [sym_size_int, -1, 20, 64]);  view_234 = None
	        permute_26: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_235, [0, 2, 1, 3]);  view_235 = None
	        clone_20: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_26, memory_format = torch.contiguous_format);  permute_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_2 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_18, clone_19, clone_20, None, False, scale = 1.0);  clone_18 = clone_19 = clone_20 = None
	        getitem_18: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_2[0];  _scaled_dot_product_efficient_attention_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_27: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_18, [0, 2, 1, 3]);  getitem_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_27, [sym_size_int, 1500, -1]);  permute_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_236, [2])
	        full_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_15, full_31);  amax_15 = full_31 = None
	        amin_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_236, [2])
	        full_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_15: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_15, full_30);  amin_15 = full_30 = None
	        sub_703: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_15, minimum_15);  maximum_15 = None
	        div_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_703, 255.0);  sub_703 = None
	        clamp_min_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_30, 1.1920928955078125e-07);  div_30 = None
	        div_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_15, clamp_min_45);  minimum_15 = None
	        round_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_31);  div_31 = None
	        sub_709: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_31);  round_31 = None
	        clamp_min_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_709, -128);  sub_709 = None
	        clamp_max_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_46, 127);  clamp_min_46 = None
	        view_239: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_45, [sym_size_int, 1500, 1])
	        reciprocal_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_239);  view_239 = None
	        mul_1529: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_15, 1.0);  reciprocal_15 = None
	        mul_1532: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_236, mul_1529);  view_236 = mul_1529 = None
	        round_32: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1532);  mul_1532 = None
	        convert_element_type_90: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_30, torch.int8);  clamp_max_30 = None
	        view_240: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_90, [sym_size_int, 1500, 1])
	        add_2419: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_32, view_240);  round_32 = view_240 = None
	        clamp_min_47: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2419, -128);  add_2419 = None
	        clamp_max_31: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_47, 127);  clamp_min_47 = None
	        convert_element_type_91: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_31, torch.int8);  clamp_max_31 = None
	        view_244: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_90, [sym_size_int, 1500, 1]);  convert_element_type_90 = None
	        convert_element_type_92: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_91, torch.float32);  convert_element_type_91 = None
	        convert_element_type_93: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_244, torch.float32);  view_244 = None
	        sub_729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_92, convert_element_type_93);  convert_element_type_92 = convert_element_type_93 = None
	        view_243: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_45, [sym_size_int, 1500, 1]);  clamp_min_45 = None
	        mul_1554: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_729, view_243);  sub_729 = view_243 = None
	        view_246: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_248: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_94: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_246, torch.float32);  view_246 = None
	        convert_element_type_95: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_248, torch.float32);  view_248 = None
	        sub_733: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_94, convert_element_type_95);  convert_element_type_94 = convert_element_type_95 = None
	        view_247: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_1559: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_733, view_247);  sub_733 = view_247 = None
	        view_249: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1559, [1280, 1280]);  mul_1559 = None
	        mul_1564: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_250: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1554, [mul_1564, 1280]);  mul_1554 = mul_1564 = None
	        permute_28: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_249, [1, 0]);  view_249 = None
	        mm_default_147: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_250, permute_28);  view_250 = permute_28 = None
	        add_tensor_147: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_147, model_audio_tower_layers_2_self_attn_out_proj_bias);  mm_default_147 = model_audio_tower_layers_2_self_attn_out_proj_bias = None
	        view_251: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_147, [sym_size_int, 1500, 1280]);  add_tensor_147 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_2482: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_1862, view_251);  add_1862 = view_251 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_22: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_2482, memory_format = torch.contiguous_format)
	        var_mean_5 = torch.ops.aten.var_mean.correction(clone_22, [2], correction = 0, keepdim = True)
	        getitem_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_5[0]
	        getitem_23: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_5[1];  var_mean_5 = None
	        sub_739: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_22, getitem_23);  clone_22 = getitem_23 = None
	        add_2487: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_22, 1e-05);  getitem_22 = None
	        rsqrt_5: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_2487);  add_2487 = None
	        mul_1575: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_739, rsqrt_5);  sub_739 = rsqrt_5 = None
	        mul_1576: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1575, model_audio_tower_layers_2_final_layer_norm_weight);  mul_1575 = model_audio_tower_layers_2_final_layer_norm_weight = None
	        add_2488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_1576, model_audio_tower_layers_2_final_layer_norm_bias);  mul_1576 = model_audio_tower_layers_2_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_2488, [2])
	        full_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_16, full_33);  amax_16 = full_33 = None
	        amin_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_2488, [2])
	        full_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_16: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_16, full_32);  amin_16 = full_32 = None
	        sub_750: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_16, minimum_16);  maximum_16 = None
	        div_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_750, 255.0);  sub_750 = None
	        clamp_min_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_32, 1.1920928955078125e-07);  div_32 = None
	        div_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_16, clamp_min_48);  minimum_16 = None
	        round_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_33);  div_33 = None
	        sub_756: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_33);  round_33 = None
	        clamp_min_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_756, -128);  sub_756 = None
	        clamp_max_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_49, 127);  clamp_min_49 = None
	        view_254: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_48, [sym_size_int, 1500, 1])
	        reciprocal_16: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_254);  view_254 = None
	        mul_1624: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_16, 1.0);  reciprocal_16 = None
	        mul_1627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_2488, mul_1624);  add_2488 = mul_1624 = None
	        round_34: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1627);  mul_1627 = None
	        convert_element_type_96: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_32, torch.int8);  clamp_max_32 = None
	        view_255: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_96, [sym_size_int, 1500, 1])
	        add_2575: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_34, view_255);  round_34 = view_255 = None
	        clamp_min_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2575, -128);  add_2575 = None
	        clamp_max_33: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_50, 127);  clamp_min_50 = None
	        convert_element_type_97: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_33, torch.int8);  clamp_max_33 = None
	        view_259: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_96, [sym_size_int, 1500, 1]);  convert_element_type_96 = None
	        convert_element_type_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_97, torch.float32);  convert_element_type_97 = None
	        convert_element_type_99: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_259, torch.float32);  view_259 = None
	        sub_776: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_98, convert_element_type_99);  convert_element_type_98 = convert_element_type_99 = None
	        view_258: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_48, [sym_size_int, 1500, 1]);  clamp_min_48 = None
	        mul_1649: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_776, view_258);  sub_776 = view_258 = None
	        view_261: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_2_fc1_parametrizations_weight_original0 = None
	        view_263: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_2_fc1_parametrizations_weight_original2 = None
	        convert_element_type_100: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_261, torch.float32);  view_261 = None
	        convert_element_type_101: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_263, torch.float32);  view_263 = None
	        sub_780: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_100, convert_element_type_101);  convert_element_type_100 = convert_element_type_101 = None
	        view_262: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_2_fc1_parametrizations_weight_original1 = None
	        mul_1654: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_780, view_262);  sub_780 = view_262 = None
	        view_264: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1654, [5120, 1280]);  mul_1654 = None
	        mul_1659: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_265: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1649, [mul_1659, 1280]);  mul_1649 = mul_1659 = None
	        permute_29: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_264, [1, 0]);  view_264 = None
	        mm_default_146: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_265, permute_29);  view_265 = permute_29 = None
	        add_tensor_146: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_146, model_audio_tower_layers_2_fc1_bias);  mm_default_146 = model_audio_tower_layers_2_fc1_bias = None
	        view_266: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_146, [sym_size_int, 1500, 5120]);  add_tensor_146 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_1666: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_266, 0.5)
	        mul_1667: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_266, 0.7071067811865476);  view_266 = None
	        erf_4: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_1667);  mul_1667 = None
	        add_2634: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_4, 1);  erf_4 = None
	        mul_1668: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1666, add_2634);  mul_1666 = add_2634 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_1668, [2])
	        full_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_17, full_35);  amax_17 = full_35 = None
	        amin_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_1668, [2])
	        full_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_17: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_17, full_34);  amin_17 = full_34 = None
	        sub_793: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_17, minimum_17);  maximum_17 = None
	        div_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_793, 255.0);  sub_793 = None
	        clamp_min_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_34, 1.1920928955078125e-07);  div_34 = None
	        div_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_17, clamp_min_51);  minimum_17 = None
	        round_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_35);  div_35 = None
	        sub_799: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_35);  round_35 = None
	        clamp_min_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_799, -128);  sub_799 = None
	        clamp_max_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_52, 127);  clamp_min_52 = None
	        view_269: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_51, [sym_size_int, 1500, 1])
	        reciprocal_17: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_269);  view_269 = None
	        mul_1714: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_17, 1.0);  reciprocal_17 = None
	        mul_1717: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1668, mul_1714);  mul_1668 = mul_1714 = None
	        round_36: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_1717);  mul_1717 = None
	        convert_element_type_102: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_34, torch.int8);  clamp_max_34 = None
	        view_270: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_102, [sym_size_int, 1500, 1])
	        add_2717: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_36, view_270);  round_36 = view_270 = None
	        clamp_min_53: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2717, -128);  add_2717 = None
	        clamp_max_35: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_53, 127);  clamp_min_53 = None
	        convert_element_type_103: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_35, torch.int8);  clamp_max_35 = None
	        view_274: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_102, [sym_size_int, 1500, 1]);  convert_element_type_102 = None
	        convert_element_type_104: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_103, torch.float32);  convert_element_type_103 = None
	        convert_element_type_105: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_274, torch.float32);  view_274 = None
	        sub_819: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_104, convert_element_type_105);  convert_element_type_104 = convert_element_type_105 = None
	        view_273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_51, [sym_size_int, 1500, 1]);  clamp_min_51 = None
	        mul_1739: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_819, view_273);  sub_819 = view_273 = None
	        view_276: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_2_fc2_parametrizations_weight_original0 = None
	        view_278: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_2_fc2_parametrizations_weight_original2 = None
	        convert_element_type_106: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_276, torch.float32);  view_276 = None
	        convert_element_type_107: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_278, torch.float32);  view_278 = None
	        sub_823: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_106, convert_element_type_107);  convert_element_type_106 = convert_element_type_107 = None
	        view_277: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_2_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_2_fc2_parametrizations_weight_original1 = None
	        mul_1744: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_823, view_277);  sub_823 = view_277 = None
	        view_279: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1744, [1280, 5120]);  mul_1744 = None
	        mul_1749: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_280: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1739, [mul_1749, 5120]);  mul_1739 = mul_1749 = None
	        permute_30: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_279, [1, 0]);  view_279 = None
	        mm_default_145: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_280, permute_30);  view_280 = permute_30 = None
	        add_tensor_145: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_145, model_audio_tower_layers_2_fc2_bias);  mm_default_145 = model_audio_tower_layers_2_fc2_bias = None
	        view_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_145, [sym_size_int, 1500, 1280]);  add_tensor_145 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_2780: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_2482, view_281);  add_2482 = view_281 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_25: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_2780, memory_format = torch.contiguous_format)
	        var_mean_6 = torch.ops.aten.var_mean.correction(clone_25, [2], correction = 0, keepdim = True)
	        getitem_24: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_6[0]
	        getitem_25: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_6[1];  var_mean_6 = None
	        sub_829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_25, getitem_25);  clone_25 = getitem_25 = None
	        add_2785: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_24, 1e-05);  getitem_24 = None
	        rsqrt_6: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_2785);  add_2785 = None
	        mul_1760: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_829, rsqrt_6);  sub_829 = rsqrt_6 = None
	        mul_1761: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_1760, model_audio_tower_layers_3_self_attn_layer_norm_weight);  mul_1760 = model_audio_tower_layers_3_self_attn_layer_norm_weight = None
	        add_2786: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_1761, model_audio_tower_layers_3_self_attn_layer_norm_bias);  mul_1761 = model_audio_tower_layers_3_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_2786, [2])
	        full_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_18, full_37);  amax_18 = full_37 = None
	        amin_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_2786, [2])
	        full_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_18: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_18, full_36);  amin_18 = full_36 = None
	        sub_840: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_18, minimum_18);  maximum_18 = None
	        div_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_840, 255.0);  sub_840 = None
	        clamp_min_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_36, 1.1920928955078125e-07);  div_36 = None
	        div_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_18, clamp_min_54);  minimum_18 = None
	        round_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_37);  div_37 = None
	        sub_846: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_37);  round_37 = None
	        clamp_min_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_846, -128);  sub_846 = None
	        clamp_max_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_55, 127);  clamp_min_55 = None
	        view_284: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_54, [sym_size_int, 1500, 1])
	        reciprocal_18: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_284);  view_284 = None
	        mul_1809: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_18, 1.0);  reciprocal_18 = None
	        mul_1812: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_2786, mul_1809);  mul_1809 = None
	        round_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1812);  mul_1812 = None
	        convert_element_type_108: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_36, torch.int8);  clamp_max_36 = None
	        view_285: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_108, [sym_size_int, 1500, 1])
	        add_2873: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_38, view_285);  round_38 = view_285 = None
	        clamp_min_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_2873, -128);  add_2873 = None
	        clamp_max_37: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_56, 127);  clamp_min_56 = None
	        convert_element_type_109: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_37, torch.int8);  clamp_max_37 = None
	        view_289: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_108, [sym_size_int, 1500, 1]);  convert_element_type_108 = None
	        convert_element_type_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_109, torch.float32);  convert_element_type_109 = None
	        convert_element_type_111: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_289, torch.float32);  view_289 = None
	        sub_866: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_110, convert_element_type_111);  convert_element_type_110 = convert_element_type_111 = None
	        view_288: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_54, [sym_size_int, 1500, 1]);  clamp_min_54 = None
	        mul_1834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_866, view_288);  sub_866 = view_288 = None
	        view_291: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_293: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_112: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_291, torch.float32);  view_291 = None
	        convert_element_type_113: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_293, torch.float32);  view_293 = None
	        sub_870: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_112, convert_element_type_113);  convert_element_type_112 = convert_element_type_113 = None
	        view_292: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_1839: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_870, view_292);  sub_870 = view_292 = None
	        view_294: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1839, [1280, 1280]);  mul_1839 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_2786, [2])
	        full_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_19, full_39);  amax_19 = full_39 = None
	        amin_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_2786, [2])
	        full_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_19: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_19, full_38);  amin_19 = full_38 = None
	        sub_885: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_19, minimum_19);  maximum_19 = None
	        div_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_885, 255.0);  sub_885 = None
	        clamp_min_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_38, 1.1920928955078125e-07);  div_38 = None
	        div_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_19, clamp_min_57);  minimum_19 = None
	        round_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_39);  div_39 = None
	        sub_891: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_39);  round_39 = None
	        clamp_min_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_891, -128);  sub_891 = None
	        clamp_max_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_58, 127);  clamp_min_58 = None
	        view_300: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_57, [sym_size_int, 1500, 1])
	        reciprocal_19: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_300);  view_300 = None
	        mul_1905: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_19, 1.0);  reciprocal_19 = None
	        mul_1908: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_2786, mul_1905);  mul_1905 = None
	        round_40: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_1908);  mul_1908 = None
	        convert_element_type_114: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_38, torch.int8);  clamp_max_38 = None
	        view_301: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_114, [sym_size_int, 1500, 1])
	        add_3025: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_40, view_301);  round_40 = view_301 = None
	        clamp_min_59: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3025, -128);  add_3025 = None
	        clamp_max_39: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_59, 127);  clamp_min_59 = None
	        convert_element_type_115: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_39, torch.int8);  clamp_max_39 = None
	        view_305: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_114, [sym_size_int, 1500, 1]);  convert_element_type_114 = None
	        convert_element_type_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_115, torch.float32);  convert_element_type_115 = None
	        convert_element_type_117: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_305, torch.float32);  view_305 = None
	        sub_911: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_116, convert_element_type_117);  convert_element_type_116 = convert_element_type_117 = None
	        view_304: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_57, [sym_size_int, 1500, 1]);  clamp_min_57 = None
	        mul_1930: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_911, view_304);  sub_911 = view_304 = None
	        view_307: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_309: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_118: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_307, torch.float32);  view_307 = None
	        convert_element_type_119: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_309, torch.float32);  view_309 = None
	        sub_915: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_118, convert_element_type_119);  convert_element_type_118 = convert_element_type_119 = None
	        view_308: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_1935: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_915, view_308);  sub_915 = view_308 = None
	        view_310: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1935, [1280, 1280]);  mul_1935 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_2786, [2])
	        full_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_20, full_41);  amax_20 = full_41 = None
	        amin_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_2786, [2])
	        full_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_20: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_20, full_40);  amin_20 = full_40 = None
	        sub_929: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_20, minimum_20);  maximum_20 = None
	        div_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_929, 255.0);  sub_929 = None
	        clamp_min_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_40, 1.1920928955078125e-07);  div_40 = None
	        div_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_20, clamp_min_60);  minimum_20 = None
	        round_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_41);  div_41 = None
	        sub_935: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_41);  round_41 = None
	        clamp_min_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_935, -128);  sub_935 = None
	        clamp_max_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_61, 127);  clamp_min_61 = None
	        view_316: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_60, [sym_size_int, 1500, 1])
	        reciprocal_20: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_316);  view_316 = None
	        mul_2004: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_20, 1.0);  reciprocal_20 = None
	        mul_2007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_2786, mul_2004);  add_2786 = mul_2004 = None
	        round_42: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2007);  mul_2007 = None
	        convert_element_type_120: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_40, torch.int8);  clamp_max_40 = None
	        view_317: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_120, [sym_size_int, 1500, 1])
	        add_3173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_42, view_317);  round_42 = view_317 = None
	        clamp_min_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3173, -128);  add_3173 = None
	        clamp_max_41: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_62, 127);  clamp_min_62 = None
	        convert_element_type_121: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_41, torch.int8);  clamp_max_41 = None
	        view_321: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_120, [sym_size_int, 1500, 1]);  convert_element_type_120 = None
	        convert_element_type_122: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_121, torch.float32);  convert_element_type_121 = None
	        convert_element_type_123: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_321, torch.float32);  view_321 = None
	        sub_955: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_122, convert_element_type_123);  convert_element_type_122 = convert_element_type_123 = None
	        view_320: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_60, [sym_size_int, 1500, 1]);  clamp_min_60 = None
	        mul_2029: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_955, view_320);  sub_955 = view_320 = None
	        view_323: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_325: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_124: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_323, torch.float32);  view_323 = None
	        convert_element_type_125: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_325, torch.float32);  view_325 = None
	        sub_959: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_124, convert_element_type_125);  convert_element_type_124 = convert_element_type_125 = None
	        view_324: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_2034: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_959, view_324);  sub_959 = view_324 = None
	        view_326: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2034, [1280, 1280]);  mul_2034 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_1844: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_295: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1834, [mul_1844, 1280]);  mul_1834 = mul_1844 = None
	        permute_31: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_294, [1, 0]);  view_294 = None
	        mm_default_144: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_295, permute_31);  view_295 = permute_31 = None
	        add_tensor_144: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_144, model_audio_tower_layers_3_self_attn_q_proj_bias);  mm_default_144 = model_audio_tower_layers_3_self_attn_q_proj_bias = None
	        view_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_144, [sym_size_int, 1500, 1280]);  add_tensor_144 = None
	        mul_1851: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_296, 0.125);  view_296 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_297: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1851, [sym_size_int, 1500, 20, 64]);  mul_1851 = None
	        permute_32: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_297, [0, 2, 1, 3]);  view_297 = None
	        clone_26: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_32, memory_format = torch.contiguous_format);  permute_32 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_1938: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_311: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_1930, [mul_1938, 1280]);  mul_1930 = mul_1938 = None
	        permute_33: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_310, [1, 0]);  view_310 = None
	        mm_3: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_311, permute_33);  view_311 = permute_33 = None
	        view_312: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_3, [sym_size_int, 1500, 1280]);  mm_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_313: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_312, [sym_size_int, -1, 20, 64]);  view_312 = None
	        permute_34: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_313, [0, 2, 1, 3]);  view_313 = None
	        clone_27: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_34, memory_format = torch.contiguous_format);  permute_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_2039: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_327: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2029, [mul_2039, 1280]);  mul_2029 = mul_2039 = None
	        permute_35: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_326, [1, 0]);  view_326 = None
	        mm_default_143: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_327, permute_35);  view_327 = permute_35 = None
	        add_tensor_143: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_143, model_audio_tower_layers_3_self_attn_v_proj_bias);  mm_default_143 = model_audio_tower_layers_3_self_attn_v_proj_bias = None
	        view_328: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_143, [sym_size_int, 1500, 1280]);  add_tensor_143 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_329: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_328, [sym_size_int, -1, 20, 64]);  view_328 = None
	        permute_36: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_329, [0, 2, 1, 3]);  view_329 = None
	        clone_28: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_36, memory_format = torch.contiguous_format);  permute_36 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_3 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_26, clone_27, clone_28, None, False, scale = 1.0);  clone_26 = clone_27 = clone_28 = None
	        getitem_26: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_3[0];  _scaled_dot_product_efficient_attention_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_37: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_26, [0, 2, 1, 3]);  getitem_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_330: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_37, [sym_size_int, 1500, -1]);  permute_37 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_330, [2])
	        full_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_21, full_43);  amax_21 = full_43 = None
	        amin_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_330, [2])
	        full_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_21: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_21, full_42);  amin_21 = full_42 = None
	        sub_977: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_21, minimum_21);  maximum_21 = None
	        div_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_977, 255.0);  sub_977 = None
	        clamp_min_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_42, 1.1920928955078125e-07);  div_42 = None
	        div_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_21, clamp_min_63);  minimum_21 = None
	        round_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_43);  div_43 = None
	        sub_983: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_43);  round_43 = None
	        clamp_min_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_983, -128);  sub_983 = None
	        clamp_max_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_64, 127);  clamp_min_64 = None
	        view_333: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_63, [sym_size_int, 1500, 1])
	        reciprocal_21: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_333);  view_333 = None
	        mul_2109: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_21, 1.0);  reciprocal_21 = None
	        mul_2112: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_330, mul_2109);  view_330 = mul_2109 = None
	        round_44: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2112);  mul_2112 = None
	        convert_element_type_126: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_42, torch.int8);  clamp_max_42 = None
	        view_334: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_126, [sym_size_int, 1500, 1])
	        add_3337: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_44, view_334);  round_44 = view_334 = None
	        clamp_min_65: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3337, -128);  add_3337 = None
	        clamp_max_43: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_65, 127);  clamp_min_65 = None
	        convert_element_type_127: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_43, torch.int8);  clamp_max_43 = None
	        view_338: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_126, [sym_size_int, 1500, 1]);  convert_element_type_126 = None
	        convert_element_type_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_127, torch.float32);  convert_element_type_127 = None
	        convert_element_type_129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_338, torch.float32);  view_338 = None
	        sub_1003: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_128, convert_element_type_129);  convert_element_type_128 = convert_element_type_129 = None
	        view_337: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_63, [sym_size_int, 1500, 1]);  clamp_min_63 = None
	        mul_2134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1003, view_337);  sub_1003 = view_337 = None
	        view_340: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_342: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_130: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_340, torch.float32);  view_340 = None
	        convert_element_type_131: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_342, torch.float32);  view_342 = None
	        sub_1007: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_130, convert_element_type_131);  convert_element_type_130 = convert_element_type_131 = None
	        view_341: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_2139: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1007, view_341);  sub_1007 = view_341 = None
	        view_343: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2139, [1280, 1280]);  mul_2139 = None
	        mul_2144: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_344: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2134, [mul_2144, 1280]);  mul_2134 = mul_2144 = None
	        permute_38: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_343, [1, 0]);  view_343 = None
	        mm_default_142: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_344, permute_38);  view_344 = permute_38 = None
	        add_tensor_142: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_142, model_audio_tower_layers_3_self_attn_out_proj_bias);  mm_default_142 = model_audio_tower_layers_3_self_attn_out_proj_bias = None
	        view_345: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_142, [sym_size_int, 1500, 1280]);  add_tensor_142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_3400: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_2780, view_345);  add_2780 = view_345 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_30: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_3400, memory_format = torch.contiguous_format)
	        var_mean_7 = torch.ops.aten.var_mean.correction(clone_30, [2], correction = 0, keepdim = True)
	        getitem_30: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_7[0]
	        getitem_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_7[1];  var_mean_7 = None
	        sub_1013: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_30, getitem_31);  clone_30 = getitem_31 = None
	        add_3405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_30, 1e-05);  getitem_30 = None
	        rsqrt_7: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_3405);  add_3405 = None
	        mul_2155: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1013, rsqrt_7);  sub_1013 = rsqrt_7 = None
	        mul_2156: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2155, model_audio_tower_layers_3_final_layer_norm_weight);  mul_2155 = model_audio_tower_layers_3_final_layer_norm_weight = None
	        add_3406: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2156, model_audio_tower_layers_3_final_layer_norm_bias);  mul_2156 = model_audio_tower_layers_3_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_3406, [2])
	        full_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_22, full_45);  amax_22 = full_45 = None
	        amin_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_3406, [2])
	        full_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_22: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_22, full_44);  amin_22 = full_44 = None
	        sub_1024: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_22, minimum_22);  maximum_22 = None
	        div_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1024, 255.0);  sub_1024 = None
	        clamp_min_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_44, 1.1920928955078125e-07);  div_44 = None
	        div_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_22, clamp_min_66);  minimum_22 = None
	        round_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_45);  div_45 = None
	        sub_1030: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_45);  round_45 = None
	        clamp_min_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1030, -128);  sub_1030 = None
	        clamp_max_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_67, 127);  clamp_min_67 = None
	        view_348: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_66, [sym_size_int, 1500, 1])
	        reciprocal_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_348);  view_348 = None
	        mul_2204: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_22, 1.0);  reciprocal_22 = None
	        mul_2207: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_3406, mul_2204);  add_3406 = mul_2204 = None
	        round_46: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2207);  mul_2207 = None
	        convert_element_type_132: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_44, torch.int8);  clamp_max_44 = None
	        view_349: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_132, [sym_size_int, 1500, 1])
	        add_3493: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_46, view_349);  round_46 = view_349 = None
	        clamp_min_68: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3493, -128);  add_3493 = None
	        clamp_max_45: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_68, 127);  clamp_min_68 = None
	        convert_element_type_133: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_45, torch.int8);  clamp_max_45 = None
	        view_353: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_132, [sym_size_int, 1500, 1]);  convert_element_type_132 = None
	        convert_element_type_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_133, torch.float32);  convert_element_type_133 = None
	        convert_element_type_135: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_353, torch.float32);  view_353 = None
	        sub_1050: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_134, convert_element_type_135);  convert_element_type_134 = convert_element_type_135 = None
	        view_352: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_66, [sym_size_int, 1500, 1]);  clamp_min_66 = None
	        mul_2229: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1050, view_352);  sub_1050 = view_352 = None
	        view_355: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_3_fc1_parametrizations_weight_original0 = None
	        view_357: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_3_fc1_parametrizations_weight_original2 = None
	        convert_element_type_136: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_355, torch.float32);  view_355 = None
	        convert_element_type_137: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_357, torch.float32);  view_357 = None
	        sub_1054: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_136, convert_element_type_137);  convert_element_type_136 = convert_element_type_137 = None
	        view_356: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_3_fc1_parametrizations_weight_original1 = None
	        mul_2234: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1054, view_356);  sub_1054 = view_356 = None
	        view_358: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2234, [5120, 1280]);  mul_2234 = None
	        mul_2239: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_359: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2229, [mul_2239, 1280]);  mul_2229 = mul_2239 = None
	        permute_39: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_358, [1, 0]);  view_358 = None
	        mm_default_141: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_359, permute_39);  view_359 = permute_39 = None
	        add_tensor_141: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_141, model_audio_tower_layers_3_fc1_bias);  mm_default_141 = model_audio_tower_layers_3_fc1_bias = None
	        view_360: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_141, [sym_size_int, 1500, 5120]);  add_tensor_141 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_2246: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_360, 0.5)
	        mul_2247: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_360, 0.7071067811865476);  view_360 = None
	        erf_5: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_2247);  mul_2247 = None
	        add_3552: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_5, 1);  erf_5 = None
	        mul_2248: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2246, add_3552);  mul_2246 = add_3552 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_2248, [2])
	        full_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_23, full_47);  amax_23 = full_47 = None
	        amin_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_2248, [2])
	        full_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_23: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_23, full_46);  amin_23 = full_46 = None
	        sub_1067: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_23, minimum_23);  maximum_23 = None
	        div_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1067, 255.0);  sub_1067 = None
	        clamp_min_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_46, 1.1920928955078125e-07);  div_46 = None
	        div_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_23, clamp_min_69);  minimum_23 = None
	        round_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_47);  div_47 = None
	        sub_1073: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_47);  round_47 = None
	        clamp_min_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1073, -128);  sub_1073 = None
	        clamp_max_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_70, 127);  clamp_min_70 = None
	        view_363: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_69, [sym_size_int, 1500, 1])
	        reciprocal_23: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_363);  view_363 = None
	        mul_2294: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_23, 1.0);  reciprocal_23 = None
	        mul_2297: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2248, mul_2294);  mul_2248 = mul_2294 = None
	        round_48: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_2297);  mul_2297 = None
	        convert_element_type_138: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_46, torch.int8);  clamp_max_46 = None
	        view_364: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_138, [sym_size_int, 1500, 1])
	        add_3635: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_48, view_364);  round_48 = view_364 = None
	        clamp_min_71: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3635, -128);  add_3635 = None
	        clamp_max_47: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_71, 127);  clamp_min_71 = None
	        convert_element_type_139: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_47, torch.int8);  clamp_max_47 = None
	        view_368: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_138, [sym_size_int, 1500, 1]);  convert_element_type_138 = None
	        convert_element_type_140: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_139, torch.float32);  convert_element_type_139 = None
	        convert_element_type_141: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_368, torch.float32);  view_368 = None
	        sub_1093: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_140, convert_element_type_141);  convert_element_type_140 = convert_element_type_141 = None
	        view_367: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_69, [sym_size_int, 1500, 1]);  clamp_min_69 = None
	        mul_2319: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1093, view_367);  sub_1093 = view_367 = None
	        view_370: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_3_fc2_parametrizations_weight_original0 = None
	        view_372: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_3_fc2_parametrizations_weight_original2 = None
	        convert_element_type_142: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_370, torch.float32);  view_370 = None
	        convert_element_type_143: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_372, torch.float32);  view_372 = None
	        sub_1097: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_142, convert_element_type_143);  convert_element_type_142 = convert_element_type_143 = None
	        view_371: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_3_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_3_fc2_parametrizations_weight_original1 = None
	        mul_2324: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1097, view_371);  sub_1097 = view_371 = None
	        view_373: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2324, [1280, 5120]);  mul_2324 = None
	        mul_2329: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_374: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2319, [mul_2329, 5120]);  mul_2319 = mul_2329 = None
	        permute_40: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_373, [1, 0]);  view_373 = None
	        mm_default_140: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_374, permute_40);  view_374 = permute_40 = None
	        add_tensor_140: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_140, model_audio_tower_layers_3_fc2_bias);  mm_default_140 = model_audio_tower_layers_3_fc2_bias = None
	        view_375: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_140, [sym_size_int, 1500, 1280]);  add_tensor_140 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_3698: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_3400, view_375);  add_3400 = view_375 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_33: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_3698, memory_format = torch.contiguous_format)
	        var_mean_8 = torch.ops.aten.var_mean.correction(clone_33, [2], correction = 0, keepdim = True)
	        getitem_32: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_8[0]
	        getitem_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_8[1];  var_mean_8 = None
	        sub_1103: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_33, getitem_33);  clone_33 = getitem_33 = None
	        add_3703: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_32, 1e-05);  getitem_32 = None
	        rsqrt_8: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_3703);  add_3703 = None
	        mul_2340: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1103, rsqrt_8);  sub_1103 = rsqrt_8 = None
	        mul_2341: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2340, model_audio_tower_layers_4_self_attn_layer_norm_weight);  mul_2340 = model_audio_tower_layers_4_self_attn_layer_norm_weight = None
	        add_3704: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2341, model_audio_tower_layers_4_self_attn_layer_norm_bias);  mul_2341 = model_audio_tower_layers_4_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_3704, [2])
	        full_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_24, full_49);  amax_24 = full_49 = None
	        amin_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_3704, [2])
	        full_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_24: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_24, full_48);  amin_24 = full_48 = None
	        sub_1114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_24, minimum_24);  maximum_24 = None
	        div_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1114, 255.0);  sub_1114 = None
	        clamp_min_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_48, 1.1920928955078125e-07);  div_48 = None
	        div_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_24, clamp_min_72);  minimum_24 = None
	        round_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_49);  div_49 = None
	        sub_1120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_49);  round_49 = None
	        clamp_min_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1120, -128);  sub_1120 = None
	        clamp_max_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_73, 127);  clamp_min_73 = None
	        view_378: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_72, [sym_size_int, 1500, 1])
	        reciprocal_24: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_378);  view_378 = None
	        mul_2389: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_24, 1.0);  reciprocal_24 = None
	        mul_2392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_3704, mul_2389);  mul_2389 = None
	        round_50: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2392);  mul_2392 = None
	        convert_element_type_144: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_48, torch.int8);  clamp_max_48 = None
	        view_379: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_144, [sym_size_int, 1500, 1])
	        add_3791: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_50, view_379);  round_50 = view_379 = None
	        clamp_min_74: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3791, -128);  add_3791 = None
	        clamp_max_49: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_74, 127);  clamp_min_74 = None
	        convert_element_type_145: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_49, torch.int8);  clamp_max_49 = None
	        view_383: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_144, [sym_size_int, 1500, 1]);  convert_element_type_144 = None
	        convert_element_type_146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_145, torch.float32);  convert_element_type_145 = None
	        convert_element_type_147: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_383, torch.float32);  view_383 = None
	        sub_1140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_146, convert_element_type_147);  convert_element_type_146 = convert_element_type_147 = None
	        view_382: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_72, [sym_size_int, 1500, 1]);  clamp_min_72 = None
	        mul_2414: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1140, view_382);  sub_1140 = view_382 = None
	        view_385: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_387: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_148: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_385, torch.float32);  view_385 = None
	        convert_element_type_149: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_387, torch.float32);  view_387 = None
	        sub_1144: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_148, convert_element_type_149);  convert_element_type_148 = convert_element_type_149 = None
	        view_386: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_2419: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1144, view_386);  sub_1144 = view_386 = None
	        view_388: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2419, [1280, 1280]);  mul_2419 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_3704, [2])
	        full_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_25, full_51);  amax_25 = full_51 = None
	        amin_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_3704, [2])
	        full_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_25: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_25, full_50);  amin_25 = full_50 = None
	        sub_1159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_25, minimum_25);  maximum_25 = None
	        div_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1159, 255.0);  sub_1159 = None
	        clamp_min_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_50, 1.1920928955078125e-07);  div_50 = None
	        div_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_25, clamp_min_75);  minimum_25 = None
	        round_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_51);  div_51 = None
	        sub_1165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_51);  round_51 = None
	        clamp_min_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1165, -128);  sub_1165 = None
	        clamp_max_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_76, 127);  clamp_min_76 = None
	        view_394: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_75, [sym_size_int, 1500, 1])
	        reciprocal_25: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_394);  view_394 = None
	        mul_2485: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_25, 1.0);  reciprocal_25 = None
	        mul_2488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_3704, mul_2485);  mul_2485 = None
	        round_52: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2488);  mul_2488 = None
	        convert_element_type_150: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_50, torch.int8);  clamp_max_50 = None
	        view_395: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_150, [sym_size_int, 1500, 1])
	        add_3943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_52, view_395);  round_52 = view_395 = None
	        clamp_min_77: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_3943, -128);  add_3943 = None
	        clamp_max_51: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_77, 127);  clamp_min_77 = None
	        convert_element_type_151: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_51, torch.int8);  clamp_max_51 = None
	        view_399: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_150, [sym_size_int, 1500, 1]);  convert_element_type_150 = None
	        convert_element_type_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_151, torch.float32);  convert_element_type_151 = None
	        convert_element_type_153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_399, torch.float32);  view_399 = None
	        sub_1185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_152, convert_element_type_153);  convert_element_type_152 = convert_element_type_153 = None
	        view_398: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_75, [sym_size_int, 1500, 1]);  clamp_min_75 = None
	        mul_2510: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1185, view_398);  sub_1185 = view_398 = None
	        view_401: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_403: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_154: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_401, torch.float32);  view_401 = None
	        convert_element_type_155: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_403, torch.float32);  view_403 = None
	        sub_1189: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_154, convert_element_type_155);  convert_element_type_154 = convert_element_type_155 = None
	        view_402: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_2515: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1189, view_402);  sub_1189 = view_402 = None
	        view_404: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2515, [1280, 1280]);  mul_2515 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_3704, [2])
	        full_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_26, full_53);  amax_26 = full_53 = None
	        amin_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_3704, [2])
	        full_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_26: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_26, full_52);  amin_26 = full_52 = None
	        sub_1203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_26, minimum_26);  maximum_26 = None
	        div_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1203, 255.0);  sub_1203 = None
	        clamp_min_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_52, 1.1920928955078125e-07);  div_52 = None
	        div_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_26, clamp_min_78);  minimum_26 = None
	        round_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_53);  div_53 = None
	        sub_1209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_53);  round_53 = None
	        clamp_min_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1209, -128);  sub_1209 = None
	        clamp_max_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_79, 127);  clamp_min_79 = None
	        view_410: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_78, [sym_size_int, 1500, 1])
	        reciprocal_26: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_410);  view_410 = None
	        mul_2584: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_26, 1.0);  reciprocal_26 = None
	        mul_2587: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_3704, mul_2584);  add_3704 = mul_2584 = None
	        round_54: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2587);  mul_2587 = None
	        convert_element_type_156: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_52, torch.int8);  clamp_max_52 = None
	        view_411: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_156, [sym_size_int, 1500, 1])
	        add_4091: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_54, view_411);  round_54 = view_411 = None
	        clamp_min_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4091, -128);  add_4091 = None
	        clamp_max_53: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_80, 127);  clamp_min_80 = None
	        convert_element_type_157: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_53, torch.int8);  clamp_max_53 = None
	        view_415: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_156, [sym_size_int, 1500, 1]);  convert_element_type_156 = None
	        convert_element_type_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_157, torch.float32);  convert_element_type_157 = None
	        convert_element_type_159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_415, torch.float32);  view_415 = None
	        sub_1229: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_158, convert_element_type_159);  convert_element_type_158 = convert_element_type_159 = None
	        view_414: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_78, [sym_size_int, 1500, 1]);  clamp_min_78 = None
	        mul_2609: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1229, view_414);  sub_1229 = view_414 = None
	        view_417: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_419: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_160: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_417, torch.float32);  view_417 = None
	        convert_element_type_161: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_419, torch.float32);  view_419 = None
	        sub_1233: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_160, convert_element_type_161);  convert_element_type_160 = convert_element_type_161 = None
	        view_418: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_2614: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1233, view_418);  sub_1233 = view_418 = None
	        view_420: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2614, [1280, 1280]);  mul_2614 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_2424: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_389: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2414, [mul_2424, 1280]);  mul_2414 = mul_2424 = None
	        permute_41: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_388, [1, 0]);  view_388 = None
	        mm_default_139: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_389, permute_41);  view_389 = permute_41 = None
	        add_tensor_139: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_139, model_audio_tower_layers_4_self_attn_q_proj_bias);  mm_default_139 = model_audio_tower_layers_4_self_attn_q_proj_bias = None
	        view_390: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_139, [sym_size_int, 1500, 1280]);  add_tensor_139 = None
	        mul_2431: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_390, 0.125);  view_390 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_391: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2431, [sym_size_int, 1500, 20, 64]);  mul_2431 = None
	        permute_42: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_391, [0, 2, 1, 3]);  view_391 = None
	        clone_34: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_42, memory_format = torch.contiguous_format);  permute_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_2518: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_405: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2510, [mul_2518, 1280]);  mul_2510 = mul_2518 = None
	        permute_43: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_404, [1, 0]);  view_404 = None
	        mm_4: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_405, permute_43);  view_405 = permute_43 = None
	        view_406: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_4, [sym_size_int, 1500, 1280]);  mm_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_407: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_406, [sym_size_int, -1, 20, 64]);  view_406 = None
	        permute_44: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_407, [0, 2, 1, 3]);  view_407 = None
	        clone_35: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_44, memory_format = torch.contiguous_format);  permute_44 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_2619: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_421: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2609, [mul_2619, 1280]);  mul_2609 = mul_2619 = None
	        permute_45: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_420, [1, 0]);  view_420 = None
	        mm_default_138: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_421, permute_45);  view_421 = permute_45 = None
	        add_tensor_138: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_138, model_audio_tower_layers_4_self_attn_v_proj_bias);  mm_default_138 = model_audio_tower_layers_4_self_attn_v_proj_bias = None
	        view_422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_138, [sym_size_int, 1500, 1280]);  add_tensor_138 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_423: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_422, [sym_size_int, -1, 20, 64]);  view_422 = None
	        permute_46: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_423, [0, 2, 1, 3]);  view_423 = None
	        clone_36: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_46, memory_format = torch.contiguous_format);  permute_46 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_4 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_34, clone_35, clone_36, None, False, scale = 1.0);  clone_34 = clone_35 = clone_36 = None
	        getitem_34: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_4[0];  _scaled_dot_product_efficient_attention_4 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_47: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_34, [0, 2, 1, 3]);  getitem_34 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_424: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_47, [sym_size_int, 1500, -1]);  permute_47 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_424, [2])
	        full_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_27, full_55);  amax_27 = full_55 = None
	        amin_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_424, [2])
	        full_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_27: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_27, full_54);  amin_27 = full_54 = None
	        sub_1251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_27, minimum_27);  maximum_27 = None
	        div_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1251, 255.0);  sub_1251 = None
	        clamp_min_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_54, 1.1920928955078125e-07);  div_54 = None
	        div_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_27, clamp_min_81);  minimum_27 = None
	        round_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_55);  div_55 = None
	        sub_1257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_55);  round_55 = None
	        clamp_min_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1257, -128);  sub_1257 = None
	        clamp_max_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_82, 127);  clamp_min_82 = None
	        view_427: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_81, [sym_size_int, 1500, 1])
	        reciprocal_27: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_427);  view_427 = None
	        mul_2689: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_27, 1.0);  reciprocal_27 = None
	        mul_2692: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_424, mul_2689);  view_424 = mul_2689 = None
	        round_56: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2692);  mul_2692 = None
	        convert_element_type_162: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_54, torch.int8);  clamp_max_54 = None
	        view_428: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_162, [sym_size_int, 1500, 1])
	        add_4255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_56, view_428);  round_56 = view_428 = None
	        clamp_min_83: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4255, -128);  add_4255 = None
	        clamp_max_55: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_83, 127);  clamp_min_83 = None
	        convert_element_type_163: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_55, torch.int8);  clamp_max_55 = None
	        view_432: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_162, [sym_size_int, 1500, 1]);  convert_element_type_162 = None
	        convert_element_type_164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_163, torch.float32);  convert_element_type_163 = None
	        convert_element_type_165: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_432, torch.float32);  view_432 = None
	        sub_1277: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_164, convert_element_type_165);  convert_element_type_164 = convert_element_type_165 = None
	        view_431: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_81, [sym_size_int, 1500, 1]);  clamp_min_81 = None
	        mul_2714: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1277, view_431);  sub_1277 = view_431 = None
	        view_434: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_436: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_166: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_434, torch.float32);  view_434 = None
	        convert_element_type_167: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_436, torch.float32);  view_436 = None
	        sub_1281: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_166, convert_element_type_167);  convert_element_type_166 = convert_element_type_167 = None
	        view_435: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_2719: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1281, view_435);  sub_1281 = view_435 = None
	        view_437: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2719, [1280, 1280]);  mul_2719 = None
	        mul_2724: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_438: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2714, [mul_2724, 1280]);  mul_2714 = mul_2724 = None
	        permute_48: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_437, [1, 0]);  view_437 = None
	        mm_default_137: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_438, permute_48);  view_438 = permute_48 = None
	        add_tensor_137: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_137, model_audio_tower_layers_4_self_attn_out_proj_bias);  mm_default_137 = model_audio_tower_layers_4_self_attn_out_proj_bias = None
	        view_439: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_137, [sym_size_int, 1500, 1280]);  add_tensor_137 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_4318: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_3698, view_439);  add_3698 = view_439 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_38: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_4318, memory_format = torch.contiguous_format)
	        var_mean_9 = torch.ops.aten.var_mean.correction(clone_38, [2], correction = 0, keepdim = True)
	        getitem_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_9[0]
	        getitem_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_9[1];  var_mean_9 = None
	        sub_1287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_38, getitem_39);  clone_38 = getitem_39 = None
	        add_4323: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_38, 1e-05);  getitem_38 = None
	        rsqrt_9: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_4323);  add_4323 = None
	        mul_2735: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1287, rsqrt_9);  sub_1287 = rsqrt_9 = None
	        mul_2736: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2735, model_audio_tower_layers_4_final_layer_norm_weight);  mul_2735 = model_audio_tower_layers_4_final_layer_norm_weight = None
	        add_4324: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2736, model_audio_tower_layers_4_final_layer_norm_bias);  mul_2736 = model_audio_tower_layers_4_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_4324, [2])
	        full_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_28, full_57);  amax_28 = full_57 = None
	        amin_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_4324, [2])
	        full_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_28: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_28, full_56);  amin_28 = full_56 = None
	        sub_1298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_28, minimum_28);  maximum_28 = None
	        div_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1298, 255.0);  sub_1298 = None
	        clamp_min_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_56, 1.1920928955078125e-07);  div_56 = None
	        div_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_28, clamp_min_84);  minimum_28 = None
	        round_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_57);  div_57 = None
	        sub_1304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_57);  round_57 = None
	        clamp_min_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1304, -128);  sub_1304 = None
	        clamp_max_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_85, 127);  clamp_min_85 = None
	        view_442: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_84, [sym_size_int, 1500, 1])
	        reciprocal_28: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_442);  view_442 = None
	        mul_2784: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_28, 1.0);  reciprocal_28 = None
	        mul_2787: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_4324, mul_2784);  add_4324 = mul_2784 = None
	        round_58: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2787);  mul_2787 = None
	        convert_element_type_168: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_56, torch.int8);  clamp_max_56 = None
	        view_443: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_168, [sym_size_int, 1500, 1])
	        add_4411: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_58, view_443);  round_58 = view_443 = None
	        clamp_min_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4411, -128);  add_4411 = None
	        clamp_max_57: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_86, 127);  clamp_min_86 = None
	        convert_element_type_169: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_57, torch.int8);  clamp_max_57 = None
	        view_447: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_168, [sym_size_int, 1500, 1]);  convert_element_type_168 = None
	        convert_element_type_170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_169, torch.float32);  convert_element_type_169 = None
	        convert_element_type_171: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_447, torch.float32);  view_447 = None
	        sub_1324: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_170, convert_element_type_171);  convert_element_type_170 = convert_element_type_171 = None
	        view_446: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_84, [sym_size_int, 1500, 1]);  clamp_min_84 = None
	        mul_2809: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1324, view_446);  sub_1324 = view_446 = None
	        view_449: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_4_fc1_parametrizations_weight_original0 = None
	        view_451: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_4_fc1_parametrizations_weight_original2 = None
	        convert_element_type_172: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_449, torch.float32);  view_449 = None
	        convert_element_type_173: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_451, torch.float32);  view_451 = None
	        sub_1328: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_172, convert_element_type_173);  convert_element_type_172 = convert_element_type_173 = None
	        view_450: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_4_fc1_parametrizations_weight_original1 = None
	        mul_2814: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1328, view_450);  sub_1328 = view_450 = None
	        view_452: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2814, [5120, 1280]);  mul_2814 = None
	        mul_2819: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_453: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2809, [mul_2819, 1280]);  mul_2809 = mul_2819 = None
	        permute_49: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_452, [1, 0]);  view_452 = None
	        mm_default_136: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_453, permute_49);  view_453 = permute_49 = None
	        add_tensor_136: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_136, model_audio_tower_layers_4_fc1_bias);  mm_default_136 = model_audio_tower_layers_4_fc1_bias = None
	        view_454: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_136, [sym_size_int, 1500, 5120]);  add_tensor_136 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_2826: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_454, 0.5)
	        mul_2827: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_454, 0.7071067811865476);  view_454 = None
	        erf_6: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_2827);  mul_2827 = None
	        add_4470: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_6, 1);  erf_6 = None
	        mul_2828: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2826, add_4470);  mul_2826 = add_4470 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_2828, [2])
	        full_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_29, full_59);  amax_29 = full_59 = None
	        amin_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_2828, [2])
	        full_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_29: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_29, full_58);  amin_29 = full_58 = None
	        sub_1341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_29, minimum_29);  maximum_29 = None
	        div_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1341, 255.0);  sub_1341 = None
	        clamp_min_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_58, 1.1920928955078125e-07);  div_58 = None
	        div_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_29, clamp_min_87);  minimum_29 = None
	        round_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_59);  div_59 = None
	        sub_1347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_59);  round_59 = None
	        clamp_min_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1347, -128);  sub_1347 = None
	        clamp_max_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_88, 127);  clamp_min_88 = None
	        view_457: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_87, [sym_size_int, 1500, 1])
	        reciprocal_29: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_457);  view_457 = None
	        mul_2874: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_29, 1.0);  reciprocal_29 = None
	        mul_2877: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2828, mul_2874);  mul_2828 = mul_2874 = None
	        round_60: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_2877);  mul_2877 = None
	        convert_element_type_174: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_58, torch.int8);  clamp_max_58 = None
	        view_458: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_174, [sym_size_int, 1500, 1])
	        add_4553: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_60, view_458);  round_60 = view_458 = None
	        clamp_min_89: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4553, -128);  add_4553 = None
	        clamp_max_59: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_89, 127);  clamp_min_89 = None
	        convert_element_type_175: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_59, torch.int8);  clamp_max_59 = None
	        view_462: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_174, [sym_size_int, 1500, 1]);  convert_element_type_174 = None
	        convert_element_type_176: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_175, torch.float32);  convert_element_type_175 = None
	        convert_element_type_177: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_462, torch.float32);  view_462 = None
	        sub_1367: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_176, convert_element_type_177);  convert_element_type_176 = convert_element_type_177 = None
	        view_461: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_87, [sym_size_int, 1500, 1]);  clamp_min_87 = None
	        mul_2899: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1367, view_461);  sub_1367 = view_461 = None
	        view_464: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_4_fc2_parametrizations_weight_original0 = None
	        view_466: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_4_fc2_parametrizations_weight_original2 = None
	        convert_element_type_178: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_464, torch.float32);  view_464 = None
	        convert_element_type_179: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_466, torch.float32);  view_466 = None
	        sub_1371: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_178, convert_element_type_179);  convert_element_type_178 = convert_element_type_179 = None
	        view_465: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_4_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_4_fc2_parametrizations_weight_original1 = None
	        mul_2904: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1371, view_465);  sub_1371 = view_465 = None
	        view_467: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2904, [1280, 5120]);  mul_2904 = None
	        mul_2909: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_468: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2899, [mul_2909, 5120]);  mul_2899 = mul_2909 = None
	        permute_50: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_467, [1, 0]);  view_467 = None
	        mm_default_135: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_468, permute_50);  view_468 = permute_50 = None
	        add_tensor_135: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_135, model_audio_tower_layers_4_fc2_bias);  mm_default_135 = model_audio_tower_layers_4_fc2_bias = None
	        view_469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_135, [sym_size_int, 1500, 1280]);  add_tensor_135 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_4616: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_4318, view_469);  add_4318 = view_469 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_41: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_4616, memory_format = torch.contiguous_format)
	        var_mean_10 = torch.ops.aten.var_mean.correction(clone_41, [2], correction = 0, keepdim = True)
	        getitem_40: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_10[0]
	        getitem_41: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_10[1];  var_mean_10 = None
	        sub_1377: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_41, getitem_41);  clone_41 = getitem_41 = None
	        add_4621: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_40, 1e-05);  getitem_40 = None
	        rsqrt_10: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_4621);  add_4621 = None
	        mul_2920: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1377, rsqrt_10);  sub_1377 = rsqrt_10 = None
	        mul_2921: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_2920, model_audio_tower_layers_5_self_attn_layer_norm_weight);  mul_2920 = model_audio_tower_layers_5_self_attn_layer_norm_weight = None
	        add_4622: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_2921, model_audio_tower_layers_5_self_attn_layer_norm_bias);  mul_2921 = model_audio_tower_layers_5_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_4622, [2])
	        full_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_30, full_61);  amax_30 = full_61 = None
	        amin_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_4622, [2])
	        full_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_30: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_30, full_60);  amin_30 = full_60 = None
	        sub_1388: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_30, minimum_30);  maximum_30 = None
	        div_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1388, 255.0);  sub_1388 = None
	        clamp_min_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_60, 1.1920928955078125e-07);  div_60 = None
	        div_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_30, clamp_min_90);  minimum_30 = None
	        round_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_61);  div_61 = None
	        sub_1394: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_61);  round_61 = None
	        clamp_min_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1394, -128);  sub_1394 = None
	        clamp_max_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_91, 127);  clamp_min_91 = None
	        view_472: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_90, [sym_size_int, 1500, 1])
	        reciprocal_30: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_472);  view_472 = None
	        mul_2969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_30, 1.0);  reciprocal_30 = None
	        mul_2972: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_4622, mul_2969);  mul_2969 = None
	        round_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_2972);  mul_2972 = None
	        convert_element_type_180: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_60, torch.int8);  clamp_max_60 = None
	        view_473: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_180, [sym_size_int, 1500, 1])
	        add_4709: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_62, view_473);  round_62 = view_473 = None
	        clamp_min_92: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4709, -128);  add_4709 = None
	        clamp_max_61: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_92, 127);  clamp_min_92 = None
	        convert_element_type_181: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_61, torch.int8);  clamp_max_61 = None
	        view_477: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_180, [sym_size_int, 1500, 1]);  convert_element_type_180 = None
	        convert_element_type_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_181, torch.float32);  convert_element_type_181 = None
	        convert_element_type_183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_477, torch.float32);  view_477 = None
	        sub_1414: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_182, convert_element_type_183);  convert_element_type_182 = convert_element_type_183 = None
	        view_476: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_90, [sym_size_int, 1500, 1]);  clamp_min_90 = None
	        mul_2994: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1414, view_476);  sub_1414 = view_476 = None
	        view_479: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_481: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_184: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_479, torch.float32);  view_479 = None
	        convert_element_type_185: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_481, torch.float32);  view_481 = None
	        sub_1418: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_184, convert_element_type_185);  convert_element_type_184 = convert_element_type_185 = None
	        view_480: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_2999: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1418, view_480);  sub_1418 = view_480 = None
	        view_482: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2999, [1280, 1280]);  mul_2999 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_4622, [2])
	        full_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_31, full_63);  amax_31 = full_63 = None
	        amin_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_4622, [2])
	        full_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_31: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_31, full_62);  amin_31 = full_62 = None
	        sub_1433: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_31, minimum_31);  maximum_31 = None
	        div_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1433, 255.0);  sub_1433 = None
	        clamp_min_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_62, 1.1920928955078125e-07);  div_62 = None
	        div_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_31, clamp_min_93);  minimum_31 = None
	        round_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_63);  div_63 = None
	        sub_1439: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_63);  round_63 = None
	        clamp_min_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1439, -128);  sub_1439 = None
	        clamp_max_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_94, 127);  clamp_min_94 = None
	        view_488: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_93, [sym_size_int, 1500, 1])
	        reciprocal_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_488);  view_488 = None
	        mul_3065: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_31, 1.0);  reciprocal_31 = None
	        mul_3068: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_4622, mul_3065);  mul_3065 = None
	        round_64: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3068);  mul_3068 = None
	        convert_element_type_186: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_62, torch.int8);  clamp_max_62 = None
	        view_489: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_186, [sym_size_int, 1500, 1])
	        add_4861: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_64, view_489);  round_64 = view_489 = None
	        clamp_min_95: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_4861, -128);  add_4861 = None
	        clamp_max_63: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_95, 127);  clamp_min_95 = None
	        convert_element_type_187: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_63, torch.int8);  clamp_max_63 = None
	        view_493: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_186, [sym_size_int, 1500, 1]);  convert_element_type_186 = None
	        convert_element_type_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_187, torch.float32);  convert_element_type_187 = None
	        convert_element_type_189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_493, torch.float32);  view_493 = None
	        sub_1459: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_188, convert_element_type_189);  convert_element_type_188 = convert_element_type_189 = None
	        view_492: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_93, [sym_size_int, 1500, 1]);  clamp_min_93 = None
	        mul_3090: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1459, view_492);  sub_1459 = view_492 = None
	        view_495: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_497: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_190: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_495, torch.float32);  view_495 = None
	        convert_element_type_191: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_497, torch.float32);  view_497 = None
	        sub_1463: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_190, convert_element_type_191);  convert_element_type_190 = convert_element_type_191 = None
	        view_496: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_3095: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1463, view_496);  sub_1463 = view_496 = None
	        view_498: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3095, [1280, 1280]);  mul_3095 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_4622, [2])
	        full_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_32, full_65);  amax_32 = full_65 = None
	        amin_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_4622, [2])
	        full_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_32: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_32, full_64);  amin_32 = full_64 = None
	        sub_1477: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_32, minimum_32);  maximum_32 = None
	        div_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1477, 255.0);  sub_1477 = None
	        clamp_min_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_64, 1.1920928955078125e-07);  div_64 = None
	        div_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_32, clamp_min_96);  minimum_32 = None
	        round_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_65);  div_65 = None
	        sub_1483: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_65);  round_65 = None
	        clamp_min_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1483, -128);  sub_1483 = None
	        clamp_max_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_97, 127);  clamp_min_97 = None
	        view_504: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_96, [sym_size_int, 1500, 1])
	        reciprocal_32: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_504);  view_504 = None
	        mul_3164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_32, 1.0);  reciprocal_32 = None
	        mul_3167: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_4622, mul_3164);  add_4622 = mul_3164 = None
	        round_66: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3167);  mul_3167 = None
	        convert_element_type_192: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_64, torch.int8);  clamp_max_64 = None
	        view_505: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_192, [sym_size_int, 1500, 1])
	        add_5009: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_66, view_505);  round_66 = view_505 = None
	        clamp_min_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5009, -128);  add_5009 = None
	        clamp_max_65: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_98, 127);  clamp_min_98 = None
	        convert_element_type_193: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_65, torch.int8);  clamp_max_65 = None
	        view_509: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_192, [sym_size_int, 1500, 1]);  convert_element_type_192 = None
	        convert_element_type_194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_193, torch.float32);  convert_element_type_193 = None
	        convert_element_type_195: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_509, torch.float32);  view_509 = None
	        sub_1503: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_194, convert_element_type_195);  convert_element_type_194 = convert_element_type_195 = None
	        view_508: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_96, [sym_size_int, 1500, 1]);  clamp_min_96 = None
	        mul_3189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1503, view_508);  sub_1503 = view_508 = None
	        view_511: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_513: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_196: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_511, torch.float32);  view_511 = None
	        convert_element_type_197: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_513, torch.float32);  view_513 = None
	        sub_1507: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_196, convert_element_type_197);  convert_element_type_196 = convert_element_type_197 = None
	        view_512: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_3194: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1507, view_512);  sub_1507 = view_512 = None
	        view_514: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3194, [1280, 1280]);  mul_3194 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_3004: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_483: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_2994, [mul_3004, 1280]);  mul_2994 = mul_3004 = None
	        permute_51: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_482, [1, 0]);  view_482 = None
	        mm_default_134: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_483, permute_51);  view_483 = permute_51 = None
	        add_tensor_134: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_134, model_audio_tower_layers_5_self_attn_q_proj_bias);  mm_default_134 = model_audio_tower_layers_5_self_attn_q_proj_bias = None
	        view_484: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_134, [sym_size_int, 1500, 1280]);  add_tensor_134 = None
	        mul_3011: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_484, 0.125);  view_484 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_485: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3011, [sym_size_int, 1500, 20, 64]);  mul_3011 = None
	        permute_52: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_485, [0, 2, 1, 3]);  view_485 = None
	        clone_42: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_52, memory_format = torch.contiguous_format);  permute_52 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_3098: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_499: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3090, [mul_3098, 1280]);  mul_3090 = mul_3098 = None
	        permute_53: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_498, [1, 0]);  view_498 = None
	        mm_5: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_499, permute_53);  view_499 = permute_53 = None
	        view_500: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_5, [sym_size_int, 1500, 1280]);  mm_5 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_501: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_500, [sym_size_int, -1, 20, 64]);  view_500 = None
	        permute_54: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_501, [0, 2, 1, 3]);  view_501 = None
	        clone_43: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_54, memory_format = torch.contiguous_format);  permute_54 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_3199: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_515: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3189, [mul_3199, 1280]);  mul_3189 = mul_3199 = None
	        permute_55: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_514, [1, 0]);  view_514 = None
	        mm_default_133: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_515, permute_55);  view_515 = permute_55 = None
	        add_tensor_133: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_133, model_audio_tower_layers_5_self_attn_v_proj_bias);  mm_default_133 = model_audio_tower_layers_5_self_attn_v_proj_bias = None
	        view_516: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_133, [sym_size_int, 1500, 1280]);  add_tensor_133 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_517: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_516, [sym_size_int, -1, 20, 64]);  view_516 = None
	        permute_56: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_517, [0, 2, 1, 3]);  view_517 = None
	        clone_44: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_56, memory_format = torch.contiguous_format);  permute_56 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_5 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_42, clone_43, clone_44, None, False, scale = 1.0);  clone_42 = clone_43 = clone_44 = None
	        getitem_42: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_5[0];  _scaled_dot_product_efficient_attention_5 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_57: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_42, [0, 2, 1, 3]);  getitem_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_57, [sym_size_int, 1500, -1]);  permute_57 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_518, [2])
	        full_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_33, full_67);  amax_33 = full_67 = None
	        amin_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_518, [2])
	        full_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_33: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_33, full_66);  amin_33 = full_66 = None
	        sub_1525: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_33, minimum_33);  maximum_33 = None
	        div_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1525, 255.0);  sub_1525 = None
	        clamp_min_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_66, 1.1920928955078125e-07);  div_66 = None
	        div_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_33, clamp_min_99);  minimum_33 = None
	        round_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_67);  div_67 = None
	        sub_1531: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_67);  round_67 = None
	        clamp_min_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1531, -128);  sub_1531 = None
	        clamp_max_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_100, 127);  clamp_min_100 = None
	        view_521: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_99, [sym_size_int, 1500, 1])
	        reciprocal_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_521);  view_521 = None
	        mul_3269: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_33, 1.0);  reciprocal_33 = None
	        mul_3272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_518, mul_3269);  view_518 = mul_3269 = None
	        round_68: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3272);  mul_3272 = None
	        convert_element_type_198: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_66, torch.int8);  clamp_max_66 = None
	        view_522: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_198, [sym_size_int, 1500, 1])
	        add_5173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_68, view_522);  round_68 = view_522 = None
	        clamp_min_101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5173, -128);  add_5173 = None
	        clamp_max_67: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_101, 127);  clamp_min_101 = None
	        convert_element_type_199: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_67, torch.int8);  clamp_max_67 = None
	        view_526: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_198, [sym_size_int, 1500, 1]);  convert_element_type_198 = None
	        convert_element_type_200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_199, torch.float32);  convert_element_type_199 = None
	        convert_element_type_201: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_526, torch.float32);  view_526 = None
	        sub_1551: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_200, convert_element_type_201);  convert_element_type_200 = convert_element_type_201 = None
	        view_525: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_99, [sym_size_int, 1500, 1]);  clamp_min_99 = None
	        mul_3294: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1551, view_525);  sub_1551 = view_525 = None
	        view_528: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_530: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_202: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_528, torch.float32);  view_528 = None
	        convert_element_type_203: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_530, torch.float32);  view_530 = None
	        sub_1555: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_202, convert_element_type_203);  convert_element_type_202 = convert_element_type_203 = None
	        view_529: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_3299: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1555, view_529);  sub_1555 = view_529 = None
	        view_531: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3299, [1280, 1280]);  mul_3299 = None
	        mul_3304: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_532: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3294, [mul_3304, 1280]);  mul_3294 = mul_3304 = None
	        permute_58: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_531, [1, 0]);  view_531 = None
	        mm_default_132: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_532, permute_58);  view_532 = permute_58 = None
	        add_tensor_132: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_132, model_audio_tower_layers_5_self_attn_out_proj_bias);  mm_default_132 = model_audio_tower_layers_5_self_attn_out_proj_bias = None
	        view_533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_132, [sym_size_int, 1500, 1280]);  add_tensor_132 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_5236: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_4616, view_533);  add_4616 = view_533 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_46: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_5236, memory_format = torch.contiguous_format)
	        var_mean_11 = torch.ops.aten.var_mean.correction(clone_46, [2], correction = 0, keepdim = True)
	        getitem_46: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_11[0]
	        getitem_47: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_11[1];  var_mean_11 = None
	        sub_1561: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_46, getitem_47);  clone_46 = getitem_47 = None
	        add_5241: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_46, 1e-05);  getitem_46 = None
	        rsqrt_11: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_5241);  add_5241 = None
	        mul_3315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1561, rsqrt_11);  sub_1561 = rsqrt_11 = None
	        mul_3316: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3315, model_audio_tower_layers_5_final_layer_norm_weight);  mul_3315 = model_audio_tower_layers_5_final_layer_norm_weight = None
	        add_5242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_3316, model_audio_tower_layers_5_final_layer_norm_bias);  mul_3316 = model_audio_tower_layers_5_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_5242, [2])
	        full_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_34, full_69);  amax_34 = full_69 = None
	        amin_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_5242, [2])
	        full_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_34: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_34, full_68);  amin_34 = full_68 = None
	        sub_1572: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_34, minimum_34);  maximum_34 = None
	        div_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1572, 255.0);  sub_1572 = None
	        clamp_min_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_68, 1.1920928955078125e-07);  div_68 = None
	        div_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_34, clamp_min_102);  minimum_34 = None
	        round_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_69);  div_69 = None
	        sub_1578: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_69);  round_69 = None
	        clamp_min_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1578, -128);  sub_1578 = None
	        clamp_max_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_103, 127);  clamp_min_103 = None
	        view_536: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_102, [sym_size_int, 1500, 1])
	        reciprocal_34: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_536);  view_536 = None
	        mul_3364: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_34, 1.0);  reciprocal_34 = None
	        mul_3367: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_5242, mul_3364);  add_5242 = mul_3364 = None
	        round_70: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3367);  mul_3367 = None
	        convert_element_type_204: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_68, torch.int8);  clamp_max_68 = None
	        view_537: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_204, [sym_size_int, 1500, 1])
	        add_5329: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_70, view_537);  round_70 = view_537 = None
	        clamp_min_104: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5329, -128);  add_5329 = None
	        clamp_max_69: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_104, 127);  clamp_min_104 = None
	        convert_element_type_205: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_69, torch.int8);  clamp_max_69 = None
	        view_541: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_204, [sym_size_int, 1500, 1]);  convert_element_type_204 = None
	        convert_element_type_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_205, torch.float32);  convert_element_type_205 = None
	        convert_element_type_207: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_541, torch.float32);  view_541 = None
	        sub_1598: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_206, convert_element_type_207);  convert_element_type_206 = convert_element_type_207 = None
	        view_540: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_102, [sym_size_int, 1500, 1]);  clamp_min_102 = None
	        mul_3389: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1598, view_540);  sub_1598 = view_540 = None
	        view_543: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_5_fc1_parametrizations_weight_original0 = None
	        view_545: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_5_fc1_parametrizations_weight_original2 = None
	        convert_element_type_208: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_543, torch.float32);  view_543 = None
	        convert_element_type_209: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_545, torch.float32);  view_545 = None
	        sub_1602: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_208, convert_element_type_209);  convert_element_type_208 = convert_element_type_209 = None
	        view_544: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_5_fc1_parametrizations_weight_original1 = None
	        mul_3394: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1602, view_544);  sub_1602 = view_544 = None
	        view_546: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3394, [5120, 1280]);  mul_3394 = None
	        mul_3399: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_547: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3389, [mul_3399, 1280]);  mul_3389 = mul_3399 = None
	        permute_59: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_546, [1, 0]);  view_546 = None
	        mm_default_131: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_547, permute_59);  view_547 = permute_59 = None
	        add_tensor_131: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_131, model_audio_tower_layers_5_fc1_bias);  mm_default_131 = model_audio_tower_layers_5_fc1_bias = None
	        view_548: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_131, [sym_size_int, 1500, 5120]);  add_tensor_131 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_3406: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_548, 0.5)
	        mul_3407: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_548, 0.7071067811865476);  view_548 = None
	        erf_7: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_3407);  mul_3407 = None
	        add_5388: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_7, 1);  erf_7 = None
	        mul_3408: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3406, add_5388);  mul_3406 = add_5388 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_3408, [2])
	        full_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_35, full_71);  amax_35 = full_71 = None
	        amin_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_3408, [2])
	        full_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_35: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_35, full_70);  amin_35 = full_70 = None
	        sub_1615: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_35, minimum_35);  maximum_35 = None
	        div_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1615, 255.0);  sub_1615 = None
	        clamp_min_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_70, 1.1920928955078125e-07);  div_70 = None
	        div_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_35, clamp_min_105);  minimum_35 = None
	        round_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_71);  div_71 = None
	        sub_1621: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_71);  round_71 = None
	        clamp_min_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1621, -128);  sub_1621 = None
	        clamp_max_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_106, 127);  clamp_min_106 = None
	        view_551: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_105, [sym_size_int, 1500, 1])
	        reciprocal_35: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_551);  view_551 = None
	        mul_3454: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_35, 1.0);  reciprocal_35 = None
	        mul_3457: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3408, mul_3454);  mul_3408 = mul_3454 = None
	        round_72: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_3457);  mul_3457 = None
	        convert_element_type_210: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_70, torch.int8);  clamp_max_70 = None
	        view_552: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_210, [sym_size_int, 1500, 1])
	        add_5471: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_72, view_552);  round_72 = view_552 = None
	        clamp_min_107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5471, -128);  add_5471 = None
	        clamp_max_71: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_107, 127);  clamp_min_107 = None
	        convert_element_type_211: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_71, torch.int8);  clamp_max_71 = None
	        view_556: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_210, [sym_size_int, 1500, 1]);  convert_element_type_210 = None
	        convert_element_type_212: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_211, torch.float32);  convert_element_type_211 = None
	        convert_element_type_213: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_556, torch.float32);  view_556 = None
	        sub_1641: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_212, convert_element_type_213);  convert_element_type_212 = convert_element_type_213 = None
	        view_555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_105, [sym_size_int, 1500, 1]);  clamp_min_105 = None
	        mul_3479: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1641, view_555);  sub_1641 = view_555 = None
	        view_558: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_5_fc2_parametrizations_weight_original0 = None
	        view_560: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_5_fc2_parametrizations_weight_original2 = None
	        convert_element_type_214: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_558, torch.float32);  view_558 = None
	        convert_element_type_215: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_560, torch.float32);  view_560 = None
	        sub_1645: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_214, convert_element_type_215);  convert_element_type_214 = convert_element_type_215 = None
	        view_559: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_5_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_5_fc2_parametrizations_weight_original1 = None
	        mul_3484: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1645, view_559);  sub_1645 = view_559 = None
	        view_561: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3484, [1280, 5120]);  mul_3484 = None
	        mul_3489: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_562: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3479, [mul_3489, 5120]);  mul_3479 = mul_3489 = None
	        permute_60: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_561, [1, 0]);  view_561 = None
	        mm_default_130: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_562, permute_60);  view_562 = permute_60 = None
	        add_tensor_130: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_130, model_audio_tower_layers_5_fc2_bias);  mm_default_130 = model_audio_tower_layers_5_fc2_bias = None
	        view_563: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_130, [sym_size_int, 1500, 1280]);  add_tensor_130 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_5534: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_5236, view_563);  add_5236 = view_563 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_49: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_5534, memory_format = torch.contiguous_format)
	        var_mean_12 = torch.ops.aten.var_mean.correction(clone_49, [2], correction = 0, keepdim = True)
	        getitem_48: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_12[0]
	        getitem_49: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_12[1];  var_mean_12 = None
	        sub_1651: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_49, getitem_49);  clone_49 = getitem_49 = None
	        add_5539: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_48, 1e-05);  getitem_48 = None
	        rsqrt_12: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_5539);  add_5539 = None
	        mul_3500: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1651, rsqrt_12);  sub_1651 = rsqrt_12 = None
	        mul_3501: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3500, model_audio_tower_layers_6_self_attn_layer_norm_weight);  mul_3500 = model_audio_tower_layers_6_self_attn_layer_norm_weight = None
	        add_5540: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_3501, model_audio_tower_layers_6_self_attn_layer_norm_bias);  mul_3501 = model_audio_tower_layers_6_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_5540, [2])
	        full_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_36, full_73);  amax_36 = full_73 = None
	        amin_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_5540, [2])
	        full_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_36: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_36, full_72);  amin_36 = full_72 = None
	        sub_1662: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_36, minimum_36);  maximum_36 = None
	        div_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1662, 255.0);  sub_1662 = None
	        clamp_min_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_72, 1.1920928955078125e-07);  div_72 = None
	        div_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_36, clamp_min_108);  minimum_36 = None
	        round_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_73);  div_73 = None
	        sub_1668: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_73);  round_73 = None
	        clamp_min_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1668, -128);  sub_1668 = None
	        clamp_max_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_109, 127);  clamp_min_109 = None
	        view_566: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_108, [sym_size_int, 1500, 1])
	        reciprocal_36: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_566);  view_566 = None
	        mul_3549: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_36, 1.0);  reciprocal_36 = None
	        mul_3552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_5540, mul_3549);  mul_3549 = None
	        round_74: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3552);  mul_3552 = None
	        convert_element_type_216: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_72, torch.int8);  clamp_max_72 = None
	        view_567: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_216, [sym_size_int, 1500, 1])
	        add_5627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_74, view_567);  round_74 = view_567 = None
	        clamp_min_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5627, -128);  add_5627 = None
	        clamp_max_73: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_110, 127);  clamp_min_110 = None
	        convert_element_type_217: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_73, torch.int8);  clamp_max_73 = None
	        view_571: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_216, [sym_size_int, 1500, 1]);  convert_element_type_216 = None
	        convert_element_type_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_217, torch.float32);  convert_element_type_217 = None
	        convert_element_type_219: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_571, torch.float32);  view_571 = None
	        sub_1688: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_218, convert_element_type_219);  convert_element_type_218 = convert_element_type_219 = None
	        view_570: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_108, [sym_size_int, 1500, 1]);  clamp_min_108 = None
	        mul_3574: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1688, view_570);  sub_1688 = view_570 = None
	        view_573: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_575: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_220: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_573, torch.float32);  view_573 = None
	        convert_element_type_221: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_575, torch.float32);  view_575 = None
	        sub_1692: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_220, convert_element_type_221);  convert_element_type_220 = convert_element_type_221 = None
	        view_574: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_3579: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1692, view_574);  sub_1692 = view_574 = None
	        view_576: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3579, [1280, 1280]);  mul_3579 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_5540, [2])
	        full_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_37, full_75);  amax_37 = full_75 = None
	        amin_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_5540, [2])
	        full_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_37: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_37, full_74);  amin_37 = full_74 = None
	        sub_1707: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_37, minimum_37);  maximum_37 = None
	        div_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1707, 255.0);  sub_1707 = None
	        clamp_min_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_74, 1.1920928955078125e-07);  div_74 = None
	        div_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_37, clamp_min_111);  minimum_37 = None
	        round_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_75);  div_75 = None
	        sub_1713: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_75);  round_75 = None
	        clamp_min_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1713, -128);  sub_1713 = None
	        clamp_max_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_112, 127);  clamp_min_112 = None
	        view_582: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_111, [sym_size_int, 1500, 1])
	        reciprocal_37: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_582);  view_582 = None
	        mul_3645: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_37, 1.0);  reciprocal_37 = None
	        mul_3648: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_5540, mul_3645);  mul_3645 = None
	        round_76: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3648);  mul_3648 = None
	        convert_element_type_222: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_74, torch.int8);  clamp_max_74 = None
	        view_583: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_222, [sym_size_int, 1500, 1])
	        add_5779: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_76, view_583);  round_76 = view_583 = None
	        clamp_min_113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5779, -128);  add_5779 = None
	        clamp_max_75: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_113, 127);  clamp_min_113 = None
	        convert_element_type_223: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_75, torch.int8);  clamp_max_75 = None
	        view_587: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_222, [sym_size_int, 1500, 1]);  convert_element_type_222 = None
	        convert_element_type_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_223, torch.float32);  convert_element_type_223 = None
	        convert_element_type_225: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_587, torch.float32);  view_587 = None
	        sub_1733: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_224, convert_element_type_225);  convert_element_type_224 = convert_element_type_225 = None
	        view_586: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_111, [sym_size_int, 1500, 1]);  clamp_min_111 = None
	        mul_3670: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1733, view_586);  sub_1733 = view_586 = None
	        view_589: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_591: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_226: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_589, torch.float32);  view_589 = None
	        convert_element_type_227: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_591, torch.float32);  view_591 = None
	        sub_1737: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_226, convert_element_type_227);  convert_element_type_226 = convert_element_type_227 = None
	        view_590: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_3675: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1737, view_590);  sub_1737 = view_590 = None
	        view_592: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3675, [1280, 1280]);  mul_3675 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_5540, [2])
	        full_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_38, full_77);  amax_38 = full_77 = None
	        amin_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_5540, [2])
	        full_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_38: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_38, full_76);  amin_38 = full_76 = None
	        sub_1751: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_38, minimum_38);  maximum_38 = None
	        div_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1751, 255.0);  sub_1751 = None
	        clamp_min_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_76, 1.1920928955078125e-07);  div_76 = None
	        div_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_38, clamp_min_114);  minimum_38 = None
	        round_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_77);  div_77 = None
	        sub_1757: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_77);  round_77 = None
	        clamp_min_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1757, -128);  sub_1757 = None
	        clamp_max_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_115, 127);  clamp_min_115 = None
	        view_598: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_114, [sym_size_int, 1500, 1])
	        reciprocal_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_598);  view_598 = None
	        mul_3744: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_38, 1.0);  reciprocal_38 = None
	        mul_3747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_5540, mul_3744);  add_5540 = mul_3744 = None
	        round_78: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3747);  mul_3747 = None
	        convert_element_type_228: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_76, torch.int8);  clamp_max_76 = None
	        view_599: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_228, [sym_size_int, 1500, 1])
	        add_5927: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_78, view_599);  round_78 = view_599 = None
	        clamp_min_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_5927, -128);  add_5927 = None
	        clamp_max_77: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_116, 127);  clamp_min_116 = None
	        convert_element_type_229: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_77, torch.int8);  clamp_max_77 = None
	        view_603: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_228, [sym_size_int, 1500, 1]);  convert_element_type_228 = None
	        convert_element_type_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_229, torch.float32);  convert_element_type_229 = None
	        convert_element_type_231: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_603, torch.float32);  view_603 = None
	        sub_1777: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_230, convert_element_type_231);  convert_element_type_230 = convert_element_type_231 = None
	        view_602: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_114, [sym_size_int, 1500, 1]);  clamp_min_114 = None
	        mul_3769: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1777, view_602);  sub_1777 = view_602 = None
	        view_605: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_607: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_232: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_605, torch.float32);  view_605 = None
	        convert_element_type_233: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_607, torch.float32);  view_607 = None
	        sub_1781: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_232, convert_element_type_233);  convert_element_type_232 = convert_element_type_233 = None
	        view_606: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_3774: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1781, view_606);  sub_1781 = view_606 = None
	        view_608: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3774, [1280, 1280]);  mul_3774 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_3584: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_577: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3574, [mul_3584, 1280]);  mul_3574 = mul_3584 = None
	        permute_61: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_576, [1, 0]);  view_576 = None
	        mm_default_129: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_577, permute_61);  view_577 = permute_61 = None
	        add_tensor_129: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_129, model_audio_tower_layers_6_self_attn_q_proj_bias);  mm_default_129 = model_audio_tower_layers_6_self_attn_q_proj_bias = None
	        view_578: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_129, [sym_size_int, 1500, 1280]);  add_tensor_129 = None
	        mul_3591: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_578, 0.125);  view_578 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_579: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3591, [sym_size_int, 1500, 20, 64]);  mul_3591 = None
	        permute_62: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_579, [0, 2, 1, 3]);  view_579 = None
	        clone_50: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_62, memory_format = torch.contiguous_format);  permute_62 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_3678: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_593: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3670, [mul_3678, 1280]);  mul_3670 = mul_3678 = None
	        permute_63: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_592, [1, 0]);  view_592 = None
	        mm_6: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_593, permute_63);  view_593 = permute_63 = None
	        view_594: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_6, [sym_size_int, 1500, 1280]);  mm_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_595: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_594, [sym_size_int, -1, 20, 64]);  view_594 = None
	        permute_64: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_595, [0, 2, 1, 3]);  view_595 = None
	        clone_51: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_64, memory_format = torch.contiguous_format);  permute_64 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_3779: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_609: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3769, [mul_3779, 1280]);  mul_3769 = mul_3779 = None
	        permute_65: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_608, [1, 0]);  view_608 = None
	        mm_default_128: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_609, permute_65);  view_609 = permute_65 = None
	        add_tensor_128: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_128, model_audio_tower_layers_6_self_attn_v_proj_bias);  mm_default_128 = model_audio_tower_layers_6_self_attn_v_proj_bias = None
	        view_610: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_128, [sym_size_int, 1500, 1280]);  add_tensor_128 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_611: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_610, [sym_size_int, -1, 20, 64]);  view_610 = None
	        permute_66: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_611, [0, 2, 1, 3]);  view_611 = None
	        clone_52: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_66, memory_format = torch.contiguous_format);  permute_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_6 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_50, clone_51, clone_52, None, False, scale = 1.0);  clone_50 = clone_51 = clone_52 = None
	        getitem_50: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_6[0];  _scaled_dot_product_efficient_attention_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_67: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_50, [0, 2, 1, 3]);  getitem_50 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_67, [sym_size_int, 1500, -1]);  permute_67 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_612, [2])
	        full_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_39, full_79);  amax_39 = full_79 = None
	        amin_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_612, [2])
	        full_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_39: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_39, full_78);  amin_39 = full_78 = None
	        sub_1799: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_39, minimum_39);  maximum_39 = None
	        div_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1799, 255.0);  sub_1799 = None
	        clamp_min_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_78, 1.1920928955078125e-07);  div_78 = None
	        div_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_39, clamp_min_117);  minimum_39 = None
	        round_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_79);  div_79 = None
	        sub_1805: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_79);  round_79 = None
	        clamp_min_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1805, -128);  sub_1805 = None
	        clamp_max_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_118, 127);  clamp_min_118 = None
	        view_615: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_117, [sym_size_int, 1500, 1])
	        reciprocal_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_615);  view_615 = None
	        mul_3849: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_39, 1.0);  reciprocal_39 = None
	        mul_3852: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_612, mul_3849);  view_612 = mul_3849 = None
	        round_80: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3852);  mul_3852 = None
	        convert_element_type_234: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_78, torch.int8);  clamp_max_78 = None
	        view_616: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_234, [sym_size_int, 1500, 1])
	        add_6091: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_80, view_616);  round_80 = view_616 = None
	        clamp_min_119: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6091, -128);  add_6091 = None
	        clamp_max_79: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_119, 127);  clamp_min_119 = None
	        convert_element_type_235: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_79, torch.int8);  clamp_max_79 = None
	        view_620: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_234, [sym_size_int, 1500, 1]);  convert_element_type_234 = None
	        convert_element_type_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_235, torch.float32);  convert_element_type_235 = None
	        convert_element_type_237: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_620, torch.float32);  view_620 = None
	        sub_1825: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_236, convert_element_type_237);  convert_element_type_236 = convert_element_type_237 = None
	        view_619: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_117, [sym_size_int, 1500, 1]);  clamp_min_117 = None
	        mul_3874: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1825, view_619);  sub_1825 = view_619 = None
	        view_622: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_624: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_238: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_622, torch.float32);  view_622 = None
	        convert_element_type_239: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_624, torch.float32);  view_624 = None
	        sub_1829: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_238, convert_element_type_239);  convert_element_type_238 = convert_element_type_239 = None
	        view_623: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_3879: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1829, view_623);  sub_1829 = view_623 = None
	        view_625: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3879, [1280, 1280]);  mul_3879 = None
	        mul_3884: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_626: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3874, [mul_3884, 1280]);  mul_3874 = mul_3884 = None
	        permute_68: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_625, [1, 0]);  view_625 = None
	        mm_default_127: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_626, permute_68);  view_626 = permute_68 = None
	        add_tensor_127: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_127, model_audio_tower_layers_6_self_attn_out_proj_bias);  mm_default_127 = model_audio_tower_layers_6_self_attn_out_proj_bias = None
	        view_627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_127, [sym_size_int, 1500, 1280]);  add_tensor_127 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_6154: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_5534, view_627);  add_5534 = view_627 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_54: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_6154, memory_format = torch.contiguous_format)
	        var_mean_13 = torch.ops.aten.var_mean.correction(clone_54, [2], correction = 0, keepdim = True)
	        getitem_54: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_13[0]
	        getitem_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_13[1];  var_mean_13 = None
	        sub_1835: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_54, getitem_55);  clone_54 = getitem_55 = None
	        add_6159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_54, 1e-05);  getitem_54 = None
	        rsqrt_13: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_6159);  add_6159 = None
	        mul_3895: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1835, rsqrt_13);  sub_1835 = rsqrt_13 = None
	        mul_3896: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3895, model_audio_tower_layers_6_final_layer_norm_weight);  mul_3895 = model_audio_tower_layers_6_final_layer_norm_weight = None
	        add_6160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_3896, model_audio_tower_layers_6_final_layer_norm_bias);  mul_3896 = model_audio_tower_layers_6_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_6160, [2])
	        full_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_40, full_81);  amax_40 = full_81 = None
	        amin_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_6160, [2])
	        full_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_40: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_40, full_80);  amin_40 = full_80 = None
	        sub_1846: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_40, minimum_40);  maximum_40 = None
	        div_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1846, 255.0);  sub_1846 = None
	        clamp_min_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_80, 1.1920928955078125e-07);  div_80 = None
	        div_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_40, clamp_min_120);  minimum_40 = None
	        round_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_81);  div_81 = None
	        sub_1852: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_81);  round_81 = None
	        clamp_min_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1852, -128);  sub_1852 = None
	        clamp_max_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_121, 127);  clamp_min_121 = None
	        view_630: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_120, [sym_size_int, 1500, 1])
	        reciprocal_40: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_630);  view_630 = None
	        mul_3944: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_40, 1.0);  reciprocal_40 = None
	        mul_3947: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_6160, mul_3944);  add_6160 = mul_3944 = None
	        round_82: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_3947);  mul_3947 = None
	        convert_element_type_240: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_80, torch.int8);  clamp_max_80 = None
	        view_631: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_240, [sym_size_int, 1500, 1])
	        add_6247: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_82, view_631);  round_82 = view_631 = None
	        clamp_min_122: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6247, -128);  add_6247 = None
	        clamp_max_81: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_122, 127);  clamp_min_122 = None
	        convert_element_type_241: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_81, torch.int8);  clamp_max_81 = None
	        view_635: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_240, [sym_size_int, 1500, 1]);  convert_element_type_240 = None
	        convert_element_type_242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_241, torch.float32);  convert_element_type_241 = None
	        convert_element_type_243: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_635, torch.float32);  view_635 = None
	        sub_1872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_242, convert_element_type_243);  convert_element_type_242 = convert_element_type_243 = None
	        view_634: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_120, [sym_size_int, 1500, 1]);  clamp_min_120 = None
	        mul_3969: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1872, view_634);  sub_1872 = view_634 = None
	        view_637: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_6_fc1_parametrizations_weight_original0 = None
	        view_639: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_6_fc1_parametrizations_weight_original2 = None
	        convert_element_type_244: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_637, torch.float32);  view_637 = None
	        convert_element_type_245: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_639, torch.float32);  view_639 = None
	        sub_1876: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_244, convert_element_type_245);  convert_element_type_244 = convert_element_type_245 = None
	        view_638: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_6_fc1_parametrizations_weight_original1 = None
	        mul_3974: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1876, view_638);  sub_1876 = view_638 = None
	        view_640: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3974, [5120, 1280]);  mul_3974 = None
	        mul_3979: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_641: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_3969, [mul_3979, 1280]);  mul_3969 = mul_3979 = None
	        permute_69: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_640, [1, 0]);  view_640 = None
	        mm_default_126: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_641, permute_69);  view_641 = permute_69 = None
	        add_tensor_126: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_126, model_audio_tower_layers_6_fc1_bias);  mm_default_126 = model_audio_tower_layers_6_fc1_bias = None
	        view_642: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_126, [sym_size_int, 1500, 5120]);  add_tensor_126 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_3986: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_642, 0.5)
	        mul_3987: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_642, 0.7071067811865476);  view_642 = None
	        erf_8: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_3987);  mul_3987 = None
	        add_6306: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_8, 1);  erf_8 = None
	        mul_3988: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3986, add_6306);  mul_3986 = add_6306 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_3988, [2])
	        full_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_41, full_83);  amax_41 = full_83 = None
	        amin_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_3988, [2])
	        full_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_41: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_41, full_82);  amin_41 = full_82 = None
	        sub_1889: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_41, minimum_41);  maximum_41 = None
	        div_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1889, 255.0);  sub_1889 = None
	        clamp_min_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_82, 1.1920928955078125e-07);  div_82 = None
	        div_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_41, clamp_min_123);  minimum_41 = None
	        round_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_83);  div_83 = None
	        sub_1895: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_83);  round_83 = None
	        clamp_min_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1895, -128);  sub_1895 = None
	        clamp_max_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_124, 127);  clamp_min_124 = None
	        view_645: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_123, [sym_size_int, 1500, 1])
	        reciprocal_41: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_645);  view_645 = None
	        mul_4034: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_41, 1.0);  reciprocal_41 = None
	        mul_4037: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_3988, mul_4034);  mul_3988 = mul_4034 = None
	        round_84: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_4037);  mul_4037 = None
	        convert_element_type_246: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_82, torch.int8);  clamp_max_82 = None
	        view_646: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_246, [sym_size_int, 1500, 1])
	        add_6389: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_84, view_646);  round_84 = view_646 = None
	        clamp_min_125: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6389, -128);  add_6389 = None
	        clamp_max_83: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_125, 127);  clamp_min_125 = None
	        convert_element_type_247: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_83, torch.int8);  clamp_max_83 = None
	        view_650: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_246, [sym_size_int, 1500, 1]);  convert_element_type_246 = None
	        convert_element_type_248: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_247, torch.float32);  convert_element_type_247 = None
	        convert_element_type_249: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_650, torch.float32);  view_650 = None
	        sub_1915: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_248, convert_element_type_249);  convert_element_type_248 = convert_element_type_249 = None
	        view_649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_123, [sym_size_int, 1500, 1]);  clamp_min_123 = None
	        mul_4059: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1915, view_649);  sub_1915 = view_649 = None
	        view_652: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_6_fc2_parametrizations_weight_original0 = None
	        view_654: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_6_fc2_parametrizations_weight_original2 = None
	        convert_element_type_250: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_652, torch.float32);  view_652 = None
	        convert_element_type_251: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_654, torch.float32);  view_654 = None
	        sub_1919: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_250, convert_element_type_251);  convert_element_type_250 = convert_element_type_251 = None
	        view_653: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_6_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_6_fc2_parametrizations_weight_original1 = None
	        mul_4064: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1919, view_653);  sub_1919 = view_653 = None
	        view_655: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4064, [1280, 5120]);  mul_4064 = None
	        mul_4069: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_656: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4059, [mul_4069, 5120]);  mul_4059 = mul_4069 = None
	        permute_70: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_655, [1, 0]);  view_655 = None
	        mm_default_125: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_656, permute_70);  view_656 = permute_70 = None
	        add_tensor_125: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_125, model_audio_tower_layers_6_fc2_bias);  mm_default_125 = model_audio_tower_layers_6_fc2_bias = None
	        view_657: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_125, [sym_size_int, 1500, 1280]);  add_tensor_125 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_6452: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_6154, view_657);  add_6154 = view_657 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_57: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_6452, memory_format = torch.contiguous_format)
	        var_mean_14 = torch.ops.aten.var_mean.correction(clone_57, [2], correction = 0, keepdim = True)
	        getitem_56: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_14[0]
	        getitem_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_14[1];  var_mean_14 = None
	        sub_1925: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_57, getitem_57);  clone_57 = getitem_57 = None
	        add_6457: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_56, 1e-05);  getitem_56 = None
	        rsqrt_14: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_6457);  add_6457 = None
	        mul_4080: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1925, rsqrt_14);  sub_1925 = rsqrt_14 = None
	        mul_4081: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4080, model_audio_tower_layers_7_self_attn_layer_norm_weight);  mul_4080 = model_audio_tower_layers_7_self_attn_layer_norm_weight = None
	        add_6458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_4081, model_audio_tower_layers_7_self_attn_layer_norm_bias);  mul_4081 = model_audio_tower_layers_7_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_6458, [2])
	        full_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_42, full_85);  amax_42 = full_85 = None
	        amin_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_6458, [2])
	        full_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_42: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_42, full_84);  amin_42 = full_84 = None
	        sub_1936: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_42, minimum_42);  maximum_42 = None
	        div_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1936, 255.0);  sub_1936 = None
	        clamp_min_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_84, 1.1920928955078125e-07);  div_84 = None
	        div_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_42, clamp_min_126);  minimum_42 = None
	        round_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_85);  div_85 = None
	        sub_1942: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_85);  round_85 = None
	        clamp_min_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1942, -128);  sub_1942 = None
	        clamp_max_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_127, 127);  clamp_min_127 = None
	        view_660: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_126, [sym_size_int, 1500, 1])
	        reciprocal_42: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_660);  view_660 = None
	        mul_4129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_42, 1.0);  reciprocal_42 = None
	        mul_4132: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_6458, mul_4129);  mul_4129 = None
	        round_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4132);  mul_4132 = None
	        convert_element_type_252: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_84, torch.int8);  clamp_max_84 = None
	        view_661: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_252, [sym_size_int, 1500, 1])
	        add_6545: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_86, view_661);  round_86 = view_661 = None
	        clamp_min_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6545, -128);  add_6545 = None
	        clamp_max_85: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_128, 127);  clamp_min_128 = None
	        convert_element_type_253: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_85, torch.int8);  clamp_max_85 = None
	        view_665: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_252, [sym_size_int, 1500, 1]);  convert_element_type_252 = None
	        convert_element_type_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_253, torch.float32);  convert_element_type_253 = None
	        convert_element_type_255: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_665, torch.float32);  view_665 = None
	        sub_1962: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_254, convert_element_type_255);  convert_element_type_254 = convert_element_type_255 = None
	        view_664: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_126, [sym_size_int, 1500, 1]);  clamp_min_126 = None
	        mul_4154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1962, view_664);  sub_1962 = view_664 = None
	        view_667: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_669: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_256: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_667, torch.float32);  view_667 = None
	        convert_element_type_257: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_669, torch.float32);  view_669 = None
	        sub_1966: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_256, convert_element_type_257);  convert_element_type_256 = convert_element_type_257 = None
	        view_668: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_4159: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_1966, view_668);  sub_1966 = view_668 = None
	        view_670: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4159, [1280, 1280]);  mul_4159 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_6458, [2])
	        full_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_43, full_87);  amax_43 = full_87 = None
	        amin_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_6458, [2])
	        full_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_43: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_43, full_86);  amin_43 = full_86 = None
	        sub_1981: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_43, minimum_43);  maximum_43 = None
	        div_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_1981, 255.0);  sub_1981 = None
	        clamp_min_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_86, 1.1920928955078125e-07);  div_86 = None
	        div_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_43, clamp_min_129);  minimum_43 = None
	        round_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_87);  div_87 = None
	        sub_1987: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_87);  round_87 = None
	        clamp_min_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_1987, -128);  sub_1987 = None
	        clamp_max_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_130, 127);  clamp_min_130 = None
	        view_676: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_129, [sym_size_int, 1500, 1])
	        reciprocal_43: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_676);  view_676 = None
	        mul_4225: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_43, 1.0);  reciprocal_43 = None
	        mul_4228: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_6458, mul_4225);  mul_4225 = None
	        round_88: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4228);  mul_4228 = None
	        convert_element_type_258: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_86, torch.int8);  clamp_max_86 = None
	        view_677: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_258, [sym_size_int, 1500, 1])
	        add_6697: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_88, view_677);  round_88 = view_677 = None
	        clamp_min_131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6697, -128);  add_6697 = None
	        clamp_max_87: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_131, 127);  clamp_min_131 = None
	        convert_element_type_259: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_87, torch.int8);  clamp_max_87 = None
	        view_681: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_258, [sym_size_int, 1500, 1]);  convert_element_type_258 = None
	        convert_element_type_260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_259, torch.float32);  convert_element_type_259 = None
	        convert_element_type_261: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_681, torch.float32);  view_681 = None
	        sub_2007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_260, convert_element_type_261);  convert_element_type_260 = convert_element_type_261 = None
	        view_680: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_129, [sym_size_int, 1500, 1]);  clamp_min_129 = None
	        mul_4250: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2007, view_680);  sub_2007 = view_680 = None
	        view_683: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_685: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_262: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_683, torch.float32);  view_683 = None
	        convert_element_type_263: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_685, torch.float32);  view_685 = None
	        sub_2011: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_262, convert_element_type_263);  convert_element_type_262 = convert_element_type_263 = None
	        view_684: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_4255: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2011, view_684);  sub_2011 = view_684 = None
	        view_686: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4255, [1280, 1280]);  mul_4255 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_6458, [2])
	        full_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_44, full_89);  amax_44 = full_89 = None
	        amin_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_6458, [2])
	        full_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_44: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_44, full_88);  amin_44 = full_88 = None
	        sub_2025: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_44, minimum_44);  maximum_44 = None
	        div_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2025, 255.0);  sub_2025 = None
	        clamp_min_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_88, 1.1920928955078125e-07);  div_88 = None
	        div_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_44, clamp_min_132);  minimum_44 = None
	        round_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_89);  div_89 = None
	        sub_2031: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_89);  round_89 = None
	        clamp_min_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2031, -128);  sub_2031 = None
	        clamp_max_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_133, 127);  clamp_min_133 = None
	        view_692: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_132, [sym_size_int, 1500, 1])
	        reciprocal_44: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_692);  view_692 = None
	        mul_4324: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_44, 1.0);  reciprocal_44 = None
	        mul_4327: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_6458, mul_4324);  add_6458 = mul_4324 = None
	        round_90: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4327);  mul_4327 = None
	        convert_element_type_264: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_88, torch.int8);  clamp_max_88 = None
	        view_693: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_264, [sym_size_int, 1500, 1])
	        add_6845: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_90, view_693);  round_90 = view_693 = None
	        clamp_min_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_6845, -128);  add_6845 = None
	        clamp_max_89: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_134, 127);  clamp_min_134 = None
	        convert_element_type_265: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_89, torch.int8);  clamp_max_89 = None
	        view_697: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_264, [sym_size_int, 1500, 1]);  convert_element_type_264 = None
	        convert_element_type_266: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_265, torch.float32);  convert_element_type_265 = None
	        convert_element_type_267: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_697, torch.float32);  view_697 = None
	        sub_2051: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_266, convert_element_type_267);  convert_element_type_266 = convert_element_type_267 = None
	        view_696: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_132, [sym_size_int, 1500, 1]);  clamp_min_132 = None
	        mul_4349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2051, view_696);  sub_2051 = view_696 = None
	        view_699: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_701: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_268: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_699, torch.float32);  view_699 = None
	        convert_element_type_269: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_701, torch.float32);  view_701 = None
	        sub_2055: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_268, convert_element_type_269);  convert_element_type_268 = convert_element_type_269 = None
	        view_700: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_4354: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2055, view_700);  sub_2055 = view_700 = None
	        view_702: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4354, [1280, 1280]);  mul_4354 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_4164: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_671: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4154, [mul_4164, 1280]);  mul_4154 = mul_4164 = None
	        permute_71: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_670, [1, 0]);  view_670 = None
	        mm_default_124: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_671, permute_71);  view_671 = permute_71 = None
	        add_tensor_124: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_124, model_audio_tower_layers_7_self_attn_q_proj_bias);  mm_default_124 = model_audio_tower_layers_7_self_attn_q_proj_bias = None
	        view_672: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_124, [sym_size_int, 1500, 1280]);  add_tensor_124 = None
	        mul_4171: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_672, 0.125);  view_672 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_673: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4171, [sym_size_int, 1500, 20, 64]);  mul_4171 = None
	        permute_72: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_673, [0, 2, 1, 3]);  view_673 = None
	        clone_58: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_72, memory_format = torch.contiguous_format);  permute_72 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_4258: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_687: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4250, [mul_4258, 1280]);  mul_4250 = mul_4258 = None
	        permute_73: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_686, [1, 0]);  view_686 = None
	        mm_7: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_687, permute_73);  view_687 = permute_73 = None
	        view_688: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_7, [sym_size_int, 1500, 1280]);  mm_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_689: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_688, [sym_size_int, -1, 20, 64]);  view_688 = None
	        permute_74: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_689, [0, 2, 1, 3]);  view_689 = None
	        clone_59: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_74, memory_format = torch.contiguous_format);  permute_74 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_4359: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_703: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4349, [mul_4359, 1280]);  mul_4349 = mul_4359 = None
	        permute_75: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_702, [1, 0]);  view_702 = None
	        mm_default_123: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_703, permute_75);  view_703 = permute_75 = None
	        add_tensor_123: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_123, model_audio_tower_layers_7_self_attn_v_proj_bias);  mm_default_123 = model_audio_tower_layers_7_self_attn_v_proj_bias = None
	        view_704: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_123, [sym_size_int, 1500, 1280]);  add_tensor_123 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_705: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_704, [sym_size_int, -1, 20, 64]);  view_704 = None
	        permute_76: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_705, [0, 2, 1, 3]);  view_705 = None
	        clone_60: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_76, memory_format = torch.contiguous_format);  permute_76 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_7 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_58, clone_59, clone_60, None, False, scale = 1.0);  clone_58 = clone_59 = clone_60 = None
	        getitem_58: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_7[0];  _scaled_dot_product_efficient_attention_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_77: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_58, [0, 2, 1, 3]);  getitem_58 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_706: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_77, [sym_size_int, 1500, -1]);  permute_77 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_706, [2])
	        full_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_45, full_91);  amax_45 = full_91 = None
	        amin_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_706, [2])
	        full_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_45: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_45, full_90);  amin_45 = full_90 = None
	        sub_2073: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_45, minimum_45);  maximum_45 = None
	        div_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2073, 255.0);  sub_2073 = None
	        clamp_min_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_90, 1.1920928955078125e-07);  div_90 = None
	        div_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_45, clamp_min_135);  minimum_45 = None
	        round_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_91);  div_91 = None
	        sub_2079: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_91);  round_91 = None
	        clamp_min_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2079, -128);  sub_2079 = None
	        clamp_max_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_136, 127);  clamp_min_136 = None
	        view_709: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_135, [sym_size_int, 1500, 1])
	        reciprocal_45: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_709);  view_709 = None
	        mul_4429: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_45, 1.0);  reciprocal_45 = None
	        mul_4432: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_706, mul_4429);  view_706 = mul_4429 = None
	        round_92: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4432);  mul_4432 = None
	        convert_element_type_270: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_90, torch.int8);  clamp_max_90 = None
	        view_710: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_270, [sym_size_int, 1500, 1])
	        add_7009: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_92, view_710);  round_92 = view_710 = None
	        clamp_min_137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7009, -128);  add_7009 = None
	        clamp_max_91: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_137, 127);  clamp_min_137 = None
	        convert_element_type_271: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_91, torch.int8);  clamp_max_91 = None
	        view_714: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_270, [sym_size_int, 1500, 1]);  convert_element_type_270 = None
	        convert_element_type_272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_271, torch.float32);  convert_element_type_271 = None
	        convert_element_type_273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_714, torch.float32);  view_714 = None
	        sub_2099: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_272, convert_element_type_273);  convert_element_type_272 = convert_element_type_273 = None
	        view_713: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_135, [sym_size_int, 1500, 1]);  clamp_min_135 = None
	        mul_4454: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2099, view_713);  sub_2099 = view_713 = None
	        view_716: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_718: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_274: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_716, torch.float32);  view_716 = None
	        convert_element_type_275: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_718, torch.float32);  view_718 = None
	        sub_2103: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_274, convert_element_type_275);  convert_element_type_274 = convert_element_type_275 = None
	        view_717: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_4459: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2103, view_717);  sub_2103 = view_717 = None
	        view_719: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4459, [1280, 1280]);  mul_4459 = None
	        mul_4464: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_720: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4454, [mul_4464, 1280]);  mul_4454 = mul_4464 = None
	        permute_78: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_719, [1, 0]);  view_719 = None
	        mm_default_122: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_720, permute_78);  view_720 = permute_78 = None
	        add_tensor_122: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_122, model_audio_tower_layers_7_self_attn_out_proj_bias);  mm_default_122 = model_audio_tower_layers_7_self_attn_out_proj_bias = None
	        view_721: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_122, [sym_size_int, 1500, 1280]);  add_tensor_122 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_7072: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_6452, view_721);  add_6452 = view_721 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_62: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_7072, memory_format = torch.contiguous_format)
	        var_mean_15 = torch.ops.aten.var_mean.correction(clone_62, [2], correction = 0, keepdim = True)
	        getitem_62: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_15[0]
	        getitem_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_15[1];  var_mean_15 = None
	        sub_2109: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_62, getitem_63);  clone_62 = getitem_63 = None
	        add_7077: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_62, 1e-05);  getitem_62 = None
	        rsqrt_15: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_7077);  add_7077 = None
	        mul_4475: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2109, rsqrt_15);  sub_2109 = rsqrt_15 = None
	        mul_4476: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4475, model_audio_tower_layers_7_final_layer_norm_weight);  mul_4475 = model_audio_tower_layers_7_final_layer_norm_weight = None
	        add_7078: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_4476, model_audio_tower_layers_7_final_layer_norm_bias);  mul_4476 = model_audio_tower_layers_7_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_7078, [2])
	        full_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_46, full_93);  amax_46 = full_93 = None
	        amin_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_7078, [2])
	        full_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_46: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_46, full_92);  amin_46 = full_92 = None
	        sub_2120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_46, minimum_46);  maximum_46 = None
	        div_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2120, 255.0);  sub_2120 = None
	        clamp_min_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_92, 1.1920928955078125e-07);  div_92 = None
	        div_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_46, clamp_min_138);  minimum_46 = None
	        round_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_93);  div_93 = None
	        sub_2126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_93);  round_93 = None
	        clamp_min_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2126, -128);  sub_2126 = None
	        clamp_max_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_139, 127);  clamp_min_139 = None
	        view_724: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_138, [sym_size_int, 1500, 1])
	        reciprocal_46: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_724);  view_724 = None
	        mul_4524: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_46, 1.0);  reciprocal_46 = None
	        mul_4527: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_7078, mul_4524);  add_7078 = mul_4524 = None
	        round_94: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4527);  mul_4527 = None
	        convert_element_type_276: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_92, torch.int8);  clamp_max_92 = None
	        view_725: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_276, [sym_size_int, 1500, 1])
	        add_7165: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_94, view_725);  round_94 = view_725 = None
	        clamp_min_140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7165, -128);  add_7165 = None
	        clamp_max_93: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_140, 127);  clamp_min_140 = None
	        convert_element_type_277: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_93, torch.int8);  clamp_max_93 = None
	        view_729: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_276, [sym_size_int, 1500, 1]);  convert_element_type_276 = None
	        convert_element_type_278: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_277, torch.float32);  convert_element_type_277 = None
	        convert_element_type_279: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_729, torch.float32);  view_729 = None
	        sub_2146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_278, convert_element_type_279);  convert_element_type_278 = convert_element_type_279 = None
	        view_728: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_138, [sym_size_int, 1500, 1]);  clamp_min_138 = None
	        mul_4549: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2146, view_728);  sub_2146 = view_728 = None
	        view_731: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_7_fc1_parametrizations_weight_original0 = None
	        view_733: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_7_fc1_parametrizations_weight_original2 = None
	        convert_element_type_280: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_731, torch.float32);  view_731 = None
	        convert_element_type_281: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_733, torch.float32);  view_733 = None
	        sub_2150: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_280, convert_element_type_281);  convert_element_type_280 = convert_element_type_281 = None
	        view_732: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_7_fc1_parametrizations_weight_original1 = None
	        mul_4554: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2150, view_732);  sub_2150 = view_732 = None
	        view_734: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4554, [5120, 1280]);  mul_4554 = None
	        mul_4559: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_735: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4549, [mul_4559, 1280]);  mul_4549 = mul_4559 = None
	        permute_79: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_734, [1, 0]);  view_734 = None
	        mm_default_121: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_735, permute_79);  view_735 = permute_79 = None
	        add_tensor_121: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_121, model_audio_tower_layers_7_fc1_bias);  mm_default_121 = model_audio_tower_layers_7_fc1_bias = None
	        view_736: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_121, [sym_size_int, 1500, 5120]);  add_tensor_121 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_4566: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_736, 0.5)
	        mul_4567: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_736, 0.7071067811865476);  view_736 = None
	        erf_9: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_4567);  mul_4567 = None
	        add_7224: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_9, 1);  erf_9 = None
	        mul_4568: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4566, add_7224);  mul_4566 = add_7224 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_4568, [2])
	        full_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_47, full_95);  amax_47 = full_95 = None
	        amin_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_4568, [2])
	        full_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_47: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_47, full_94);  amin_47 = full_94 = None
	        sub_2163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_47, minimum_47);  maximum_47 = None
	        div_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2163, 255.0);  sub_2163 = None
	        clamp_min_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_94, 1.1920928955078125e-07);  div_94 = None
	        div_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_47, clamp_min_141);  minimum_47 = None
	        round_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_95);  div_95 = None
	        sub_2169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_95);  round_95 = None
	        clamp_min_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2169, -128);  sub_2169 = None
	        clamp_max_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_142, 127);  clamp_min_142 = None
	        view_739: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_141, [sym_size_int, 1500, 1])
	        reciprocal_47: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_739);  view_739 = None
	        mul_4614: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_47, 1.0);  reciprocal_47 = None
	        mul_4617: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4568, mul_4614);  mul_4568 = mul_4614 = None
	        round_96: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_4617);  mul_4617 = None
	        convert_element_type_282: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_94, torch.int8);  clamp_max_94 = None
	        view_740: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_282, [sym_size_int, 1500, 1])
	        add_7307: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_96, view_740);  round_96 = view_740 = None
	        clamp_min_143: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7307, -128);  add_7307 = None
	        clamp_max_95: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_143, 127);  clamp_min_143 = None
	        convert_element_type_283: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_95, torch.int8);  clamp_max_95 = None
	        view_744: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_282, [sym_size_int, 1500, 1]);  convert_element_type_282 = None
	        convert_element_type_284: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_283, torch.float32);  convert_element_type_283 = None
	        convert_element_type_285: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_744, torch.float32);  view_744 = None
	        sub_2189: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_284, convert_element_type_285);  convert_element_type_284 = convert_element_type_285 = None
	        view_743: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_141, [sym_size_int, 1500, 1]);  clamp_min_141 = None
	        mul_4639: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2189, view_743);  sub_2189 = view_743 = None
	        view_746: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_7_fc2_parametrizations_weight_original0 = None
	        view_748: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_7_fc2_parametrizations_weight_original2 = None
	        convert_element_type_286: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_746, torch.float32);  view_746 = None
	        convert_element_type_287: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_748, torch.float32);  view_748 = None
	        sub_2193: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_286, convert_element_type_287);  convert_element_type_286 = convert_element_type_287 = None
	        view_747: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_7_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_7_fc2_parametrizations_weight_original1 = None
	        mul_4644: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2193, view_747);  sub_2193 = view_747 = None
	        view_749: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4644, [1280, 5120]);  mul_4644 = None
	        mul_4649: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_750: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4639, [mul_4649, 5120]);  mul_4639 = mul_4649 = None
	        permute_80: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_749, [1, 0]);  view_749 = None
	        mm_default_120: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_750, permute_80);  view_750 = permute_80 = None
	        add_tensor_120: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_120, model_audio_tower_layers_7_fc2_bias);  mm_default_120 = model_audio_tower_layers_7_fc2_bias = None
	        view_751: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_120, [sym_size_int, 1500, 1280]);  add_tensor_120 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_7370: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_7072, view_751);  add_7072 = view_751 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_65: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_7370, memory_format = torch.contiguous_format)
	        var_mean_16 = torch.ops.aten.var_mean.correction(clone_65, [2], correction = 0, keepdim = True)
	        getitem_64: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_16[0]
	        getitem_65: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_16[1];  var_mean_16 = None
	        sub_2199: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_65, getitem_65);  clone_65 = getitem_65 = None
	        add_7375: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_64, 1e-05);  getitem_64 = None
	        rsqrt_16: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_7375);  add_7375 = None
	        mul_4660: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2199, rsqrt_16);  sub_2199 = rsqrt_16 = None
	        mul_4661: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_4660, model_audio_tower_layers_8_self_attn_layer_norm_weight);  mul_4660 = model_audio_tower_layers_8_self_attn_layer_norm_weight = None
	        add_7376: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_4661, model_audio_tower_layers_8_self_attn_layer_norm_bias);  mul_4661 = model_audio_tower_layers_8_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_7376, [2])
	        full_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_48, full_97);  amax_48 = full_97 = None
	        amin_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_7376, [2])
	        full_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_48: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_48, full_96);  amin_48 = full_96 = None
	        sub_2210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_48, minimum_48);  maximum_48 = None
	        div_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2210, 255.0);  sub_2210 = None
	        clamp_min_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_96, 1.1920928955078125e-07);  div_96 = None
	        div_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_48, clamp_min_144);  minimum_48 = None
	        round_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_97);  div_97 = None
	        sub_2216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_97);  round_97 = None
	        clamp_min_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2216, -128);  sub_2216 = None
	        clamp_max_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_145, 127);  clamp_min_145 = None
	        view_754: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_144, [sym_size_int, 1500, 1])
	        reciprocal_48: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_754);  view_754 = None
	        mul_4709: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_48, 1.0);  reciprocal_48 = None
	        mul_4712: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_7376, mul_4709);  mul_4709 = None
	        round_98: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4712);  mul_4712 = None
	        convert_element_type_288: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_96, torch.int8);  clamp_max_96 = None
	        view_755: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_288, [sym_size_int, 1500, 1])
	        add_7463: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_98, view_755);  round_98 = view_755 = None
	        clamp_min_146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7463, -128);  add_7463 = None
	        clamp_max_97: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_146, 127);  clamp_min_146 = None
	        convert_element_type_289: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_97, torch.int8);  clamp_max_97 = None
	        view_759: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_288, [sym_size_int, 1500, 1]);  convert_element_type_288 = None
	        convert_element_type_290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_289, torch.float32);  convert_element_type_289 = None
	        convert_element_type_291: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_759, torch.float32);  view_759 = None
	        sub_2236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_290, convert_element_type_291);  convert_element_type_290 = convert_element_type_291 = None
	        view_758: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_144, [sym_size_int, 1500, 1]);  clamp_min_144 = None
	        mul_4734: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2236, view_758);  sub_2236 = view_758 = None
	        view_761: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_763: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_292: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_761, torch.float32);  view_761 = None
	        convert_element_type_293: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_763, torch.float32);  view_763 = None
	        sub_2240: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_292, convert_element_type_293);  convert_element_type_292 = convert_element_type_293 = None
	        view_762: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_4739: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2240, view_762);  sub_2240 = view_762 = None
	        view_764: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4739, [1280, 1280]);  mul_4739 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_7376, [2])
	        full_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_49, full_99);  amax_49 = full_99 = None
	        amin_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_7376, [2])
	        full_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_49: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_49, full_98);  amin_49 = full_98 = None
	        sub_2255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_49, minimum_49);  maximum_49 = None
	        div_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2255, 255.0);  sub_2255 = None
	        clamp_min_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_98, 1.1920928955078125e-07);  div_98 = None
	        div_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_49, clamp_min_147);  minimum_49 = None
	        round_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_99);  div_99 = None
	        sub_2261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_99);  round_99 = None
	        clamp_min_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2261, -128);  sub_2261 = None
	        clamp_max_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_148, 127);  clamp_min_148 = None
	        view_770: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_147, [sym_size_int, 1500, 1])
	        reciprocal_49: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_770);  view_770 = None
	        mul_4805: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_49, 1.0);  reciprocal_49 = None
	        mul_4808: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_7376, mul_4805);  mul_4805 = None
	        round_100: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4808);  mul_4808 = None
	        convert_element_type_294: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_98, torch.int8);  clamp_max_98 = None
	        view_771: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_294, [sym_size_int, 1500, 1])
	        add_7615: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_100, view_771);  round_100 = view_771 = None
	        clamp_min_149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7615, -128);  add_7615 = None
	        clamp_max_99: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_149, 127);  clamp_min_149 = None
	        convert_element_type_295: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_99, torch.int8);  clamp_max_99 = None
	        view_775: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_294, [sym_size_int, 1500, 1]);  convert_element_type_294 = None
	        convert_element_type_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_295, torch.float32);  convert_element_type_295 = None
	        convert_element_type_297: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_775, torch.float32);  view_775 = None
	        sub_2281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_296, convert_element_type_297);  convert_element_type_296 = convert_element_type_297 = None
	        view_774: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_147, [sym_size_int, 1500, 1]);  clamp_min_147 = None
	        mul_4830: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2281, view_774);  sub_2281 = view_774 = None
	        view_777: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_779: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_298: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_777, torch.float32);  view_777 = None
	        convert_element_type_299: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_779, torch.float32);  view_779 = None
	        sub_2285: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_298, convert_element_type_299);  convert_element_type_298 = convert_element_type_299 = None
	        view_778: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_4835: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2285, view_778);  sub_2285 = view_778 = None
	        view_780: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4835, [1280, 1280]);  mul_4835 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_7376, [2])
	        full_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_50, full_101);  amax_50 = full_101 = None
	        amin_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_7376, [2])
	        full_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_50: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_50, full_100);  amin_50 = full_100 = None
	        sub_2299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_50, minimum_50);  maximum_50 = None
	        div_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2299, 255.0);  sub_2299 = None
	        clamp_min_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_100, 1.1920928955078125e-07);  div_100 = None
	        div_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_50, clamp_min_150);  minimum_50 = None
	        round_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_101);  div_101 = None
	        sub_2305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_101);  round_101 = None
	        clamp_min_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2305, -128);  sub_2305 = None
	        clamp_max_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_151, 127);  clamp_min_151 = None
	        view_786: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_150, [sym_size_int, 1500, 1])
	        reciprocal_50: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_786);  view_786 = None
	        mul_4904: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_50, 1.0);  reciprocal_50 = None
	        mul_4907: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_7376, mul_4904);  add_7376 = mul_4904 = None
	        round_102: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_4907);  mul_4907 = None
	        convert_element_type_300: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_100, torch.int8);  clamp_max_100 = None
	        view_787: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_300, [sym_size_int, 1500, 1])
	        add_7763: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_102, view_787);  round_102 = view_787 = None
	        clamp_min_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7763, -128);  add_7763 = None
	        clamp_max_101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_152, 127);  clamp_min_152 = None
	        convert_element_type_301: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_101, torch.int8);  clamp_max_101 = None
	        view_791: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_300, [sym_size_int, 1500, 1]);  convert_element_type_300 = None
	        convert_element_type_302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_301, torch.float32);  convert_element_type_301 = None
	        convert_element_type_303: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_791, torch.float32);  view_791 = None
	        sub_2325: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_302, convert_element_type_303);  convert_element_type_302 = convert_element_type_303 = None
	        view_790: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_150, [sym_size_int, 1500, 1]);  clamp_min_150 = None
	        mul_4929: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2325, view_790);  sub_2325 = view_790 = None
	        view_793: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_795: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_304: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_793, torch.float32);  view_793 = None
	        convert_element_type_305: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_795, torch.float32);  view_795 = None
	        sub_2329: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_304, convert_element_type_305);  convert_element_type_304 = convert_element_type_305 = None
	        view_794: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_4934: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2329, view_794);  sub_2329 = view_794 = None
	        view_796: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4934, [1280, 1280]);  mul_4934 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_4744: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_765: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4734, [mul_4744, 1280]);  mul_4734 = mul_4744 = None
	        permute_81: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_764, [1, 0]);  view_764 = None
	        mm_default_119: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_765, permute_81);  view_765 = permute_81 = None
	        add_tensor_119: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_119, model_audio_tower_layers_8_self_attn_q_proj_bias);  mm_default_119 = model_audio_tower_layers_8_self_attn_q_proj_bias = None
	        view_766: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_119, [sym_size_int, 1500, 1280]);  add_tensor_119 = None
	        mul_4751: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_766, 0.125);  view_766 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_767: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4751, [sym_size_int, 1500, 20, 64]);  mul_4751 = None
	        permute_82: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_767, [0, 2, 1, 3]);  view_767 = None
	        clone_66: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_82, memory_format = torch.contiguous_format);  permute_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_4838: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_781: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4830, [mul_4838, 1280]);  mul_4830 = mul_4838 = None
	        permute_83: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_780, [1, 0]);  view_780 = None
	        mm_8: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_781, permute_83);  view_781 = permute_83 = None
	        view_782: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_8, [sym_size_int, 1500, 1280]);  mm_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_783: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_782, [sym_size_int, -1, 20, 64]);  view_782 = None
	        permute_84: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_783, [0, 2, 1, 3]);  view_783 = None
	        clone_67: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_84, memory_format = torch.contiguous_format);  permute_84 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_4939: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_797: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_4929, [mul_4939, 1280]);  mul_4929 = mul_4939 = None
	        permute_85: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_796, [1, 0]);  view_796 = None
	        mm_default_118: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_797, permute_85);  view_797 = permute_85 = None
	        add_tensor_118: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_118, model_audio_tower_layers_8_self_attn_v_proj_bias);  mm_default_118 = model_audio_tower_layers_8_self_attn_v_proj_bias = None
	        view_798: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_118, [sym_size_int, 1500, 1280]);  add_tensor_118 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_799: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_798, [sym_size_int, -1, 20, 64]);  view_798 = None
	        permute_86: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_799, [0, 2, 1, 3]);  view_799 = None
	        clone_68: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_86, memory_format = torch.contiguous_format);  permute_86 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_8 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_66, clone_67, clone_68, None, False, scale = 1.0);  clone_66 = clone_67 = clone_68 = None
	        getitem_66: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_8[0];  _scaled_dot_product_efficient_attention_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_87: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_66, [0, 2, 1, 3]);  getitem_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_800: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_87, [sym_size_int, 1500, -1]);  permute_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_800, [2])
	        full_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_51, full_103);  amax_51 = full_103 = None
	        amin_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_800, [2])
	        full_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_51: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_51, full_102);  amin_51 = full_102 = None
	        sub_2347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_51, minimum_51);  maximum_51 = None
	        div_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2347, 255.0);  sub_2347 = None
	        clamp_min_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_102, 1.1920928955078125e-07);  div_102 = None
	        div_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_51, clamp_min_153);  minimum_51 = None
	        round_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_103);  div_103 = None
	        sub_2353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_103);  round_103 = None
	        clamp_min_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2353, -128);  sub_2353 = None
	        clamp_max_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_154, 127);  clamp_min_154 = None
	        view_803: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_153, [sym_size_int, 1500, 1])
	        reciprocal_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_803);  view_803 = None
	        mul_5009: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_51, 1.0);  reciprocal_51 = None
	        mul_5012: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_800, mul_5009);  view_800 = mul_5009 = None
	        round_104: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5012);  mul_5012 = None
	        convert_element_type_306: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_102, torch.int8);  clamp_max_102 = None
	        view_804: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_306, [sym_size_int, 1500, 1])
	        add_7927: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_104, view_804);  round_104 = view_804 = None
	        clamp_min_155: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_7927, -128);  add_7927 = None
	        clamp_max_103: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_155, 127);  clamp_min_155 = None
	        convert_element_type_307: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_103, torch.int8);  clamp_max_103 = None
	        view_808: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_306, [sym_size_int, 1500, 1]);  convert_element_type_306 = None
	        convert_element_type_308: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_307, torch.float32);  convert_element_type_307 = None
	        convert_element_type_309: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_808, torch.float32);  view_808 = None
	        sub_2373: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_308, convert_element_type_309);  convert_element_type_308 = convert_element_type_309 = None
	        view_807: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_153, [sym_size_int, 1500, 1]);  clamp_min_153 = None
	        mul_5034: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2373, view_807);  sub_2373 = view_807 = None
	        view_810: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_812: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_310: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_810, torch.float32);  view_810 = None
	        convert_element_type_311: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_812, torch.float32);  view_812 = None
	        sub_2377: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_310, convert_element_type_311);  convert_element_type_310 = convert_element_type_311 = None
	        view_811: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_5039: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2377, view_811);  sub_2377 = view_811 = None
	        view_813: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5039, [1280, 1280]);  mul_5039 = None
	        mul_5044: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_814: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5034, [mul_5044, 1280]);  mul_5034 = mul_5044 = None
	        permute_88: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_813, [1, 0]);  view_813 = None
	        mm_default_117: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_814, permute_88);  view_814 = permute_88 = None
	        add_tensor_117: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_117, model_audio_tower_layers_8_self_attn_out_proj_bias);  mm_default_117 = model_audio_tower_layers_8_self_attn_out_proj_bias = None
	        view_815: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_117, [sym_size_int, 1500, 1280]);  add_tensor_117 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_7990: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_7370, view_815);  add_7370 = view_815 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_70: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_7990, memory_format = torch.contiguous_format)
	        var_mean_17 = torch.ops.aten.var_mean.correction(clone_70, [2], correction = 0, keepdim = True)
	        getitem_70: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_17[0]
	        getitem_71: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_17[1];  var_mean_17 = None
	        sub_2383: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_70, getitem_71);  clone_70 = getitem_71 = None
	        add_7995: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_70, 1e-05);  getitem_70 = None
	        rsqrt_17: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_7995);  add_7995 = None
	        mul_5055: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2383, rsqrt_17);  sub_2383 = rsqrt_17 = None
	        mul_5056: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5055, model_audio_tower_layers_8_final_layer_norm_weight);  mul_5055 = model_audio_tower_layers_8_final_layer_norm_weight = None
	        add_7996: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5056, model_audio_tower_layers_8_final_layer_norm_bias);  mul_5056 = model_audio_tower_layers_8_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_7996, [2])
	        full_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_52, full_105);  amax_52 = full_105 = None
	        amin_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_7996, [2])
	        full_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_52: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_52, full_104);  amin_52 = full_104 = None
	        sub_2394: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_52, minimum_52);  maximum_52 = None
	        div_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2394, 255.0);  sub_2394 = None
	        clamp_min_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_104, 1.1920928955078125e-07);  div_104 = None
	        div_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_52, clamp_min_156);  minimum_52 = None
	        round_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_105);  div_105 = None
	        sub_2400: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_105);  round_105 = None
	        clamp_min_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2400, -128);  sub_2400 = None
	        clamp_max_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_157, 127);  clamp_min_157 = None
	        view_818: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_156, [sym_size_int, 1500, 1])
	        reciprocal_52: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_818);  view_818 = None
	        mul_5104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_52, 1.0);  reciprocal_52 = None
	        mul_5107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_7996, mul_5104);  add_7996 = mul_5104 = None
	        round_106: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5107);  mul_5107 = None
	        convert_element_type_312: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_104, torch.int8);  clamp_max_104 = None
	        view_819: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_312, [sym_size_int, 1500, 1])
	        add_8083: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_106, view_819);  round_106 = view_819 = None
	        clamp_min_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8083, -128);  add_8083 = None
	        clamp_max_105: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_158, 127);  clamp_min_158 = None
	        convert_element_type_313: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_105, torch.int8);  clamp_max_105 = None
	        view_823: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_312, [sym_size_int, 1500, 1]);  convert_element_type_312 = None
	        convert_element_type_314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_313, torch.float32);  convert_element_type_313 = None
	        convert_element_type_315: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_823, torch.float32);  view_823 = None
	        sub_2420: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_314, convert_element_type_315);  convert_element_type_314 = convert_element_type_315 = None
	        view_822: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_156, [sym_size_int, 1500, 1]);  clamp_min_156 = None
	        mul_5129: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2420, view_822);  sub_2420 = view_822 = None
	        view_825: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_8_fc1_parametrizations_weight_original0 = None
	        view_827: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_8_fc1_parametrizations_weight_original2 = None
	        convert_element_type_316: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_825, torch.float32);  view_825 = None
	        convert_element_type_317: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_827, torch.float32);  view_827 = None
	        sub_2424: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_316, convert_element_type_317);  convert_element_type_316 = convert_element_type_317 = None
	        view_826: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_8_fc1_parametrizations_weight_original1 = None
	        mul_5134: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2424, view_826);  sub_2424 = view_826 = None
	        view_828: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5134, [5120, 1280]);  mul_5134 = None
	        mul_5139: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_829: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5129, [mul_5139, 1280]);  mul_5129 = mul_5139 = None
	        permute_89: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_828, [1, 0]);  view_828 = None
	        mm_default_116: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_829, permute_89);  view_829 = permute_89 = None
	        add_tensor_116: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_116, model_audio_tower_layers_8_fc1_bias);  mm_default_116 = model_audio_tower_layers_8_fc1_bias = None
	        view_830: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_116, [sym_size_int, 1500, 5120]);  add_tensor_116 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_5146: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_830, 0.5)
	        mul_5147: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_830, 0.7071067811865476);  view_830 = None
	        erf_10: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_5147);  mul_5147 = None
	        add_8142: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_10, 1);  erf_10 = None
	        mul_5148: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5146, add_8142);  mul_5146 = add_8142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_5148, [2])
	        full_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_53, full_107);  amax_53 = full_107 = None
	        amin_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_5148, [2])
	        full_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_53: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_53, full_106);  amin_53 = full_106 = None
	        sub_2437: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_53, minimum_53);  maximum_53 = None
	        div_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2437, 255.0);  sub_2437 = None
	        clamp_min_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_106, 1.1920928955078125e-07);  div_106 = None
	        div_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_53, clamp_min_159);  minimum_53 = None
	        round_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_107);  div_107 = None
	        sub_2443: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_107);  round_107 = None
	        clamp_min_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2443, -128);  sub_2443 = None
	        clamp_max_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_160, 127);  clamp_min_160 = None
	        view_833: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_159, [sym_size_int, 1500, 1])
	        reciprocal_53: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_833);  view_833 = None
	        mul_5194: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_53, 1.0);  reciprocal_53 = None
	        mul_5197: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5148, mul_5194);  mul_5148 = mul_5194 = None
	        round_108: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_5197);  mul_5197 = None
	        convert_element_type_318: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_106, torch.int8);  clamp_max_106 = None
	        view_834: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_318, [sym_size_int, 1500, 1])
	        add_8225: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_108, view_834);  round_108 = view_834 = None
	        clamp_min_161: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8225, -128);  add_8225 = None
	        clamp_max_107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_161, 127);  clamp_min_161 = None
	        convert_element_type_319: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_107, torch.int8);  clamp_max_107 = None
	        view_838: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_318, [sym_size_int, 1500, 1]);  convert_element_type_318 = None
	        convert_element_type_320: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_319, torch.float32);  convert_element_type_319 = None
	        convert_element_type_321: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_838, torch.float32);  view_838 = None
	        sub_2463: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_320, convert_element_type_321);  convert_element_type_320 = convert_element_type_321 = None
	        view_837: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_159, [sym_size_int, 1500, 1]);  clamp_min_159 = None
	        mul_5219: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2463, view_837);  sub_2463 = view_837 = None
	        view_840: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_8_fc2_parametrizations_weight_original0 = None
	        view_842: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_8_fc2_parametrizations_weight_original2 = None
	        convert_element_type_322: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_840, torch.float32);  view_840 = None
	        convert_element_type_323: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_842, torch.float32);  view_842 = None
	        sub_2467: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_322, convert_element_type_323);  convert_element_type_322 = convert_element_type_323 = None
	        view_841: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_8_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_8_fc2_parametrizations_weight_original1 = None
	        mul_5224: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2467, view_841);  sub_2467 = view_841 = None
	        view_843: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5224, [1280, 5120]);  mul_5224 = None
	        mul_5229: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_844: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5219, [mul_5229, 5120]);  mul_5219 = mul_5229 = None
	        permute_90: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_843, [1, 0]);  view_843 = None
	        mm_default_115: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_844, permute_90);  view_844 = permute_90 = None
	        add_tensor_115: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_115, model_audio_tower_layers_8_fc2_bias);  mm_default_115 = model_audio_tower_layers_8_fc2_bias = None
	        view_845: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_115, [sym_size_int, 1500, 1280]);  add_tensor_115 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_8288: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_7990, view_845);  add_7990 = view_845 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_73: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_8288, memory_format = torch.contiguous_format)
	        var_mean_18 = torch.ops.aten.var_mean.correction(clone_73, [2], correction = 0, keepdim = True)
	        getitem_72: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_18[0]
	        getitem_73: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_18[1];  var_mean_18 = None
	        sub_2473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_73, getitem_73);  clone_73 = getitem_73 = None
	        add_8293: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_72, 1e-05);  getitem_72 = None
	        rsqrt_18: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_8293);  add_8293 = None
	        mul_5240: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2473, rsqrt_18);  sub_2473 = rsqrt_18 = None
	        mul_5241: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5240, model_audio_tower_layers_9_self_attn_layer_norm_weight);  mul_5240 = model_audio_tower_layers_9_self_attn_layer_norm_weight = None
	        add_8294: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5241, model_audio_tower_layers_9_self_attn_layer_norm_bias);  mul_5241 = model_audio_tower_layers_9_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_8294, [2])
	        full_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_54, full_109);  amax_54 = full_109 = None
	        amin_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_8294, [2])
	        full_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_54: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_54, full_108);  amin_54 = full_108 = None
	        sub_2484: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_54, minimum_54);  maximum_54 = None
	        div_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2484, 255.0);  sub_2484 = None
	        clamp_min_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_108, 1.1920928955078125e-07);  div_108 = None
	        div_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_54, clamp_min_162);  minimum_54 = None
	        round_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_109);  div_109 = None
	        sub_2490: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_109);  round_109 = None
	        clamp_min_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2490, -128);  sub_2490 = None
	        clamp_max_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_163, 127);  clamp_min_163 = None
	        view_848: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_162, [sym_size_int, 1500, 1])
	        reciprocal_54: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_848);  view_848 = None
	        mul_5289: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_54, 1.0);  reciprocal_54 = None
	        mul_5292: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_8294, mul_5289);  mul_5289 = None
	        round_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5292);  mul_5292 = None
	        convert_element_type_324: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_108, torch.int8);  clamp_max_108 = None
	        view_849: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_324, [sym_size_int, 1500, 1])
	        add_8381: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_110, view_849);  round_110 = view_849 = None
	        clamp_min_164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8381, -128);  add_8381 = None
	        clamp_max_109: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_164, 127);  clamp_min_164 = None
	        convert_element_type_325: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_109, torch.int8);  clamp_max_109 = None
	        view_853: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_324, [sym_size_int, 1500, 1]);  convert_element_type_324 = None
	        convert_element_type_326: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_325, torch.float32);  convert_element_type_325 = None
	        convert_element_type_327: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_853, torch.float32);  view_853 = None
	        sub_2510: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_326, convert_element_type_327);  convert_element_type_326 = convert_element_type_327 = None
	        view_852: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_162, [sym_size_int, 1500, 1]);  clamp_min_162 = None
	        mul_5314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2510, view_852);  sub_2510 = view_852 = None
	        view_855: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_857: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_328: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_855, torch.float32);  view_855 = None
	        convert_element_type_329: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_857, torch.float32);  view_857 = None
	        sub_2514: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_328, convert_element_type_329);  convert_element_type_328 = convert_element_type_329 = None
	        view_856: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_5319: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2514, view_856);  sub_2514 = view_856 = None
	        view_858: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5319, [1280, 1280]);  mul_5319 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_8294, [2])
	        full_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_55, full_111);  amax_55 = full_111 = None
	        amin_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_8294, [2])
	        full_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_55: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_55, full_110);  amin_55 = full_110 = None
	        sub_2529: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_55, minimum_55);  maximum_55 = None
	        div_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2529, 255.0);  sub_2529 = None
	        clamp_min_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_110, 1.1920928955078125e-07);  div_110 = None
	        div_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_55, clamp_min_165);  minimum_55 = None
	        round_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_111);  div_111 = None
	        sub_2535: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_111);  round_111 = None
	        clamp_min_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2535, -128);  sub_2535 = None
	        clamp_max_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_166, 127);  clamp_min_166 = None
	        view_864: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_165, [sym_size_int, 1500, 1])
	        reciprocal_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_864);  view_864 = None
	        mul_5385: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_55, 1.0);  reciprocal_55 = None
	        mul_5388: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_8294, mul_5385);  mul_5385 = None
	        round_112: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5388);  mul_5388 = None
	        convert_element_type_330: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_110, torch.int8);  clamp_max_110 = None
	        view_865: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_330, [sym_size_int, 1500, 1])
	        add_8533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_112, view_865);  round_112 = view_865 = None
	        clamp_min_167: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8533, -128);  add_8533 = None
	        clamp_max_111: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_167, 127);  clamp_min_167 = None
	        convert_element_type_331: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_111, torch.int8);  clamp_max_111 = None
	        view_869: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_330, [sym_size_int, 1500, 1]);  convert_element_type_330 = None
	        convert_element_type_332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_331, torch.float32);  convert_element_type_331 = None
	        convert_element_type_333: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_869, torch.float32);  view_869 = None
	        sub_2555: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_332, convert_element_type_333);  convert_element_type_332 = convert_element_type_333 = None
	        view_868: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_165, [sym_size_int, 1500, 1]);  clamp_min_165 = None
	        mul_5410: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2555, view_868);  sub_2555 = view_868 = None
	        view_871: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_873: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_334: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_871, torch.float32);  view_871 = None
	        convert_element_type_335: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_873, torch.float32);  view_873 = None
	        sub_2559: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_334, convert_element_type_335);  convert_element_type_334 = convert_element_type_335 = None
	        view_872: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_5415: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2559, view_872);  sub_2559 = view_872 = None
	        view_874: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5415, [1280, 1280]);  mul_5415 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_8294, [2])
	        full_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_56, full_113);  amax_56 = full_113 = None
	        amin_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_8294, [2])
	        full_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_56: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_56, full_112);  amin_56 = full_112 = None
	        sub_2573: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_56, minimum_56);  maximum_56 = None
	        div_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2573, 255.0);  sub_2573 = None
	        clamp_min_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_112, 1.1920928955078125e-07);  div_112 = None
	        div_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_56, clamp_min_168);  minimum_56 = None
	        round_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_113);  div_113 = None
	        sub_2579: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_113);  round_113 = None
	        clamp_min_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2579, -128);  sub_2579 = None
	        clamp_max_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_169, 127);  clamp_min_169 = None
	        view_880: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_168, [sym_size_int, 1500, 1])
	        reciprocal_56: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_880);  view_880 = None
	        mul_5484: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_56, 1.0);  reciprocal_56 = None
	        mul_5487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_8294, mul_5484);  add_8294 = mul_5484 = None
	        round_114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5487);  mul_5487 = None
	        convert_element_type_336: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_112, torch.int8);  clamp_max_112 = None
	        view_881: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_336, [sym_size_int, 1500, 1])
	        add_8681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_114, view_881);  round_114 = view_881 = None
	        clamp_min_170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8681, -128);  add_8681 = None
	        clamp_max_113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_170, 127);  clamp_min_170 = None
	        convert_element_type_337: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_113, torch.int8);  clamp_max_113 = None
	        view_885: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_336, [sym_size_int, 1500, 1]);  convert_element_type_336 = None
	        convert_element_type_338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_337, torch.float32);  convert_element_type_337 = None
	        convert_element_type_339: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_885, torch.float32);  view_885 = None
	        sub_2599: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_338, convert_element_type_339);  convert_element_type_338 = convert_element_type_339 = None
	        view_884: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_168, [sym_size_int, 1500, 1]);  clamp_min_168 = None
	        mul_5509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2599, view_884);  sub_2599 = view_884 = None
	        view_887: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_889: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_340: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_887, torch.float32);  view_887 = None
	        convert_element_type_341: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_889, torch.float32);  view_889 = None
	        sub_2603: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_340, convert_element_type_341);  convert_element_type_340 = convert_element_type_341 = None
	        view_888: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_5514: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2603, view_888);  sub_2603 = view_888 = None
	        view_890: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5514, [1280, 1280]);  mul_5514 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_5324: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_859: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5314, [mul_5324, 1280]);  mul_5314 = mul_5324 = None
	        permute_91: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_858, [1, 0]);  view_858 = None
	        mm_default_114: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_859, permute_91);  view_859 = permute_91 = None
	        add_tensor_114: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_114, model_audio_tower_layers_9_self_attn_q_proj_bias);  mm_default_114 = model_audio_tower_layers_9_self_attn_q_proj_bias = None
	        view_860: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_114, [sym_size_int, 1500, 1280]);  add_tensor_114 = None
	        mul_5331: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_860, 0.125);  view_860 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_861: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5331, [sym_size_int, 1500, 20, 64]);  mul_5331 = None
	        permute_92: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_861, [0, 2, 1, 3]);  view_861 = None
	        clone_74: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_92, memory_format = torch.contiguous_format);  permute_92 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_5418: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_875: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5410, [mul_5418, 1280]);  mul_5410 = mul_5418 = None
	        permute_93: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_874, [1, 0]);  view_874 = None
	        mm_9: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_875, permute_93);  view_875 = permute_93 = None
	        view_876: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_9, [sym_size_int, 1500, 1280]);  mm_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_877: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_876, [sym_size_int, -1, 20, 64]);  view_876 = None
	        permute_94: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_877, [0, 2, 1, 3]);  view_877 = None
	        clone_75: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_94, memory_format = torch.contiguous_format);  permute_94 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_5519: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_891: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5509, [mul_5519, 1280]);  mul_5509 = mul_5519 = None
	        permute_95: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_890, [1, 0]);  view_890 = None
	        mm_default_113: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_891, permute_95);  view_891 = permute_95 = None
	        add_tensor_113: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_113, model_audio_tower_layers_9_self_attn_v_proj_bias);  mm_default_113 = model_audio_tower_layers_9_self_attn_v_proj_bias = None
	        view_892: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_113, [sym_size_int, 1500, 1280]);  add_tensor_113 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_893: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_892, [sym_size_int, -1, 20, 64]);  view_892 = None
	        permute_96: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_893, [0, 2, 1, 3]);  view_893 = None
	        clone_76: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_96, memory_format = torch.contiguous_format);  permute_96 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_9 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_74, clone_75, clone_76, None, False, scale = 1.0);  clone_74 = clone_75 = clone_76 = None
	        getitem_74: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_9[0];  _scaled_dot_product_efficient_attention_9 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_97: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_74, [0, 2, 1, 3]);  getitem_74 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_97, [sym_size_int, 1500, -1]);  permute_97 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_894, [2])
	        full_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_57, full_115);  amax_57 = full_115 = None
	        amin_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_894, [2])
	        full_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_57: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_57, full_114);  amin_57 = full_114 = None
	        sub_2621: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_57, minimum_57);  maximum_57 = None
	        div_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2621, 255.0);  sub_2621 = None
	        clamp_min_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_114, 1.1920928955078125e-07);  div_114 = None
	        div_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_57, clamp_min_171);  minimum_57 = None
	        round_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_115);  div_115 = None
	        sub_2627: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_115);  round_115 = None
	        clamp_min_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2627, -128);  sub_2627 = None
	        clamp_max_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_172, 127);  clamp_min_172 = None
	        view_897: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_171, [sym_size_int, 1500, 1])
	        reciprocal_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_897);  view_897 = None
	        mul_5589: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_57, 1.0);  reciprocal_57 = None
	        mul_5592: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_894, mul_5589);  view_894 = mul_5589 = None
	        round_116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5592);  mul_5592 = None
	        convert_element_type_342: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_114, torch.int8);  clamp_max_114 = None
	        view_898: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_342, [sym_size_int, 1500, 1])
	        add_8845: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_116, view_898);  round_116 = view_898 = None
	        clamp_min_173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_8845, -128);  add_8845 = None
	        clamp_max_115: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_173, 127);  clamp_min_173 = None
	        convert_element_type_343: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_115, torch.int8);  clamp_max_115 = None
	        view_902: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_342, [sym_size_int, 1500, 1]);  convert_element_type_342 = None
	        convert_element_type_344: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_343, torch.float32);  convert_element_type_343 = None
	        convert_element_type_345: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_902, torch.float32);  view_902 = None
	        sub_2647: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_344, convert_element_type_345);  convert_element_type_344 = convert_element_type_345 = None
	        view_901: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_171, [sym_size_int, 1500, 1]);  clamp_min_171 = None
	        mul_5614: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2647, view_901);  sub_2647 = view_901 = None
	        view_904: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_906: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_346: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_904, torch.float32);  view_904 = None
	        convert_element_type_347: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_906, torch.float32);  view_906 = None
	        sub_2651: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_346, convert_element_type_347);  convert_element_type_346 = convert_element_type_347 = None
	        view_905: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_5619: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2651, view_905);  sub_2651 = view_905 = None
	        view_907: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5619, [1280, 1280]);  mul_5619 = None
	        mul_5624: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_908: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5614, [mul_5624, 1280]);  mul_5614 = mul_5624 = None
	        permute_98: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_907, [1, 0]);  view_907 = None
	        mm_default_112: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_908, permute_98);  view_908 = permute_98 = None
	        add_tensor_112: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_112, model_audio_tower_layers_9_self_attn_out_proj_bias);  mm_default_112 = model_audio_tower_layers_9_self_attn_out_proj_bias = None
	        view_909: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_112, [sym_size_int, 1500, 1280]);  add_tensor_112 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_8908: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_8288, view_909);  add_8288 = view_909 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_78: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_8908, memory_format = torch.contiguous_format)
	        var_mean_19 = torch.ops.aten.var_mean.correction(clone_78, [2], correction = 0, keepdim = True)
	        getitem_78: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_19[0]
	        getitem_79: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_19[1];  var_mean_19 = None
	        sub_2657: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_78, getitem_79);  clone_78 = getitem_79 = None
	        add_8913: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_78, 1e-05);  getitem_78 = None
	        rsqrt_19: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_8913);  add_8913 = None
	        mul_5635: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2657, rsqrt_19);  sub_2657 = rsqrt_19 = None
	        mul_5636: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5635, model_audio_tower_layers_9_final_layer_norm_weight);  mul_5635 = model_audio_tower_layers_9_final_layer_norm_weight = None
	        add_8914: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5636, model_audio_tower_layers_9_final_layer_norm_bias);  mul_5636 = model_audio_tower_layers_9_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_8914, [2])
	        full_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_58, full_117);  amax_58 = full_117 = None
	        amin_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_8914, [2])
	        full_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_58: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_58, full_116);  amin_58 = full_116 = None
	        sub_2668: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_58, minimum_58);  maximum_58 = None
	        div_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2668, 255.0);  sub_2668 = None
	        clamp_min_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_116, 1.1920928955078125e-07);  div_116 = None
	        div_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_58, clamp_min_174);  minimum_58 = None
	        round_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_117);  div_117 = None
	        sub_2674: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_117);  round_117 = None
	        clamp_min_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2674, -128);  sub_2674 = None
	        clamp_max_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_175, 127);  clamp_min_175 = None
	        view_912: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_174, [sym_size_int, 1500, 1])
	        reciprocal_58: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_912);  view_912 = None
	        mul_5684: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_58, 1.0);  reciprocal_58 = None
	        mul_5687: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_8914, mul_5684);  add_8914 = mul_5684 = None
	        round_118: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5687);  mul_5687 = None
	        convert_element_type_348: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_116, torch.int8);  clamp_max_116 = None
	        view_913: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_348, [sym_size_int, 1500, 1])
	        add_9001: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_118, view_913);  round_118 = view_913 = None
	        clamp_min_176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9001, -128);  add_9001 = None
	        clamp_max_117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_176, 127);  clamp_min_176 = None
	        convert_element_type_349: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_117, torch.int8);  clamp_max_117 = None
	        view_917: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_348, [sym_size_int, 1500, 1]);  convert_element_type_348 = None
	        convert_element_type_350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_349, torch.float32);  convert_element_type_349 = None
	        convert_element_type_351: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_917, torch.float32);  view_917 = None
	        sub_2694: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_350, convert_element_type_351);  convert_element_type_350 = convert_element_type_351 = None
	        view_916: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_174, [sym_size_int, 1500, 1]);  clamp_min_174 = None
	        mul_5709: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2694, view_916);  sub_2694 = view_916 = None
	        view_919: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_9_fc1_parametrizations_weight_original0 = None
	        view_921: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_9_fc1_parametrizations_weight_original2 = None
	        convert_element_type_352: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_919, torch.float32);  view_919 = None
	        convert_element_type_353: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_921, torch.float32);  view_921 = None
	        sub_2698: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_352, convert_element_type_353);  convert_element_type_352 = convert_element_type_353 = None
	        view_920: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_9_fc1_parametrizations_weight_original1 = None
	        mul_5714: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2698, view_920);  sub_2698 = view_920 = None
	        view_922: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5714, [5120, 1280]);  mul_5714 = None
	        mul_5719: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_923: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5709, [mul_5719, 1280]);  mul_5709 = mul_5719 = None
	        permute_99: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_922, [1, 0]);  view_922 = None
	        mm_default_111: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_923, permute_99);  view_923 = permute_99 = None
	        add_tensor_111: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_111, model_audio_tower_layers_9_fc1_bias);  mm_default_111 = model_audio_tower_layers_9_fc1_bias = None
	        view_924: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_111, [sym_size_int, 1500, 5120]);  add_tensor_111 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_5726: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_924, 0.5)
	        mul_5727: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_924, 0.7071067811865476);  view_924 = None
	        erf_11: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_5727);  mul_5727 = None
	        add_9060: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_11, 1);  erf_11 = None
	        mul_5728: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5726, add_9060);  mul_5726 = add_9060 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_5728, [2])
	        full_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_59, full_119);  amax_59 = full_119 = None
	        amin_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_5728, [2])
	        full_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_59: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_59, full_118);  amin_59 = full_118 = None
	        sub_2711: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_59, minimum_59);  maximum_59 = None
	        div_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2711, 255.0);  sub_2711 = None
	        clamp_min_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_118, 1.1920928955078125e-07);  div_118 = None
	        div_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_59, clamp_min_177);  minimum_59 = None
	        round_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_119);  div_119 = None
	        sub_2717: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_119);  round_119 = None
	        clamp_min_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2717, -128);  sub_2717 = None
	        clamp_max_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_178, 127);  clamp_min_178 = None
	        view_927: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_177, [sym_size_int, 1500, 1])
	        reciprocal_59: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_927);  view_927 = None
	        mul_5774: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_59, 1.0);  reciprocal_59 = None
	        mul_5777: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5728, mul_5774);  mul_5728 = mul_5774 = None
	        round_120: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_5777);  mul_5777 = None
	        convert_element_type_354: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_118, torch.int8);  clamp_max_118 = None
	        view_928: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_354, [sym_size_int, 1500, 1])
	        add_9143: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_120, view_928);  round_120 = view_928 = None
	        clamp_min_179: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9143, -128);  add_9143 = None
	        clamp_max_119: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_179, 127);  clamp_min_179 = None
	        convert_element_type_355: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_119, torch.int8);  clamp_max_119 = None
	        view_932: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_354, [sym_size_int, 1500, 1]);  convert_element_type_354 = None
	        convert_element_type_356: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_355, torch.float32);  convert_element_type_355 = None
	        convert_element_type_357: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_932, torch.float32);  view_932 = None
	        sub_2737: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_356, convert_element_type_357);  convert_element_type_356 = convert_element_type_357 = None
	        view_931: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_177, [sym_size_int, 1500, 1]);  clamp_min_177 = None
	        mul_5799: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2737, view_931);  sub_2737 = view_931 = None
	        view_934: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_9_fc2_parametrizations_weight_original0 = None
	        view_936: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_9_fc2_parametrizations_weight_original2 = None
	        convert_element_type_358: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_934, torch.float32);  view_934 = None
	        convert_element_type_359: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_936, torch.float32);  view_936 = None
	        sub_2741: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_358, convert_element_type_359);  convert_element_type_358 = convert_element_type_359 = None
	        view_935: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_9_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_9_fc2_parametrizations_weight_original1 = None
	        mul_5804: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2741, view_935);  sub_2741 = view_935 = None
	        view_937: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5804, [1280, 5120]);  mul_5804 = None
	        mul_5809: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_938: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5799, [mul_5809, 5120]);  mul_5799 = mul_5809 = None
	        permute_100: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_937, [1, 0]);  view_937 = None
	        mm_default_110: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_938, permute_100);  view_938 = permute_100 = None
	        add_tensor_110: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_110, model_audio_tower_layers_9_fc2_bias);  mm_default_110 = model_audio_tower_layers_9_fc2_bias = None
	        view_939: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_110, [sym_size_int, 1500, 1280]);  add_tensor_110 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_9206: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_8908, view_939);  add_8908 = view_939 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_81: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_9206, memory_format = torch.contiguous_format)
	        var_mean_20 = torch.ops.aten.var_mean.correction(clone_81, [2], correction = 0, keepdim = True)
	        getitem_80: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_20[0]
	        getitem_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_20[1];  var_mean_20 = None
	        sub_2747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_81, getitem_81);  clone_81 = getitem_81 = None
	        add_9211: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_80, 1e-05);  getitem_80 = None
	        rsqrt_20: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_9211);  add_9211 = None
	        mul_5820: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2747, rsqrt_20);  sub_2747 = rsqrt_20 = None
	        mul_5821: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_5820, model_audio_tower_layers_10_self_attn_layer_norm_weight);  mul_5820 = model_audio_tower_layers_10_self_attn_layer_norm_weight = None
	        add_9212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_5821, model_audio_tower_layers_10_self_attn_layer_norm_bias);  mul_5821 = model_audio_tower_layers_10_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_9212, [2])
	        full_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_60, full_121);  amax_60 = full_121 = None
	        amin_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_9212, [2])
	        full_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_60: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_60, full_120);  amin_60 = full_120 = None
	        sub_2758: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_60, minimum_60);  maximum_60 = None
	        div_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2758, 255.0);  sub_2758 = None
	        clamp_min_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_120, 1.1920928955078125e-07);  div_120 = None
	        div_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_60, clamp_min_180);  minimum_60 = None
	        round_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_121);  div_121 = None
	        sub_2764: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_121);  round_121 = None
	        clamp_min_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2764, -128);  sub_2764 = None
	        clamp_max_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_181, 127);  clamp_min_181 = None
	        view_942: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_180, [sym_size_int, 1500, 1])
	        reciprocal_60: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_942);  view_942 = None
	        mul_5869: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_60, 1.0);  reciprocal_60 = None
	        mul_5872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_9212, mul_5869);  mul_5869 = None
	        round_122: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5872);  mul_5872 = None
	        convert_element_type_360: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_120, torch.int8);  clamp_max_120 = None
	        view_943: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_360, [sym_size_int, 1500, 1])
	        add_9299: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_122, view_943);  round_122 = view_943 = None
	        clamp_min_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9299, -128);  add_9299 = None
	        clamp_max_121: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_182, 127);  clamp_min_182 = None
	        convert_element_type_361: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_121, torch.int8);  clamp_max_121 = None
	        view_947: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_360, [sym_size_int, 1500, 1]);  convert_element_type_360 = None
	        convert_element_type_362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_361, torch.float32);  convert_element_type_361 = None
	        convert_element_type_363: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_947, torch.float32);  view_947 = None
	        sub_2784: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_362, convert_element_type_363);  convert_element_type_362 = convert_element_type_363 = None
	        view_946: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_180, [sym_size_int, 1500, 1]);  clamp_min_180 = None
	        mul_5894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2784, view_946);  sub_2784 = view_946 = None
	        view_949: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_951: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_364: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_949, torch.float32);  view_949 = None
	        convert_element_type_365: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_951, torch.float32);  view_951 = None
	        sub_2788: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_364, convert_element_type_365);  convert_element_type_364 = convert_element_type_365 = None
	        view_950: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_5899: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2788, view_950);  sub_2788 = view_950 = None
	        view_952: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5899, [1280, 1280]);  mul_5899 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_9212, [2])
	        full_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_61, full_123);  amax_61 = full_123 = None
	        amin_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_9212, [2])
	        full_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_61: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_61, full_122);  amin_61 = full_122 = None
	        sub_2803: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_61, minimum_61);  maximum_61 = None
	        div_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2803, 255.0);  sub_2803 = None
	        clamp_min_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_122, 1.1920928955078125e-07);  div_122 = None
	        div_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_61, clamp_min_183);  minimum_61 = None
	        round_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_123);  div_123 = None
	        sub_2809: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_123);  round_123 = None
	        clamp_min_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2809, -128);  sub_2809 = None
	        clamp_max_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_184, 127);  clamp_min_184 = None
	        view_958: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_183, [sym_size_int, 1500, 1])
	        reciprocal_61: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_958);  view_958 = None
	        mul_5965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_61, 1.0);  reciprocal_61 = None
	        mul_5968: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_9212, mul_5965);  mul_5965 = None
	        round_124: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_5968);  mul_5968 = None
	        convert_element_type_366: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_122, torch.int8);  clamp_max_122 = None
	        view_959: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_366, [sym_size_int, 1500, 1])
	        add_9451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_124, view_959);  round_124 = view_959 = None
	        clamp_min_185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9451, -128);  add_9451 = None
	        clamp_max_123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_185, 127);  clamp_min_185 = None
	        convert_element_type_367: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_123, torch.int8);  clamp_max_123 = None
	        view_963: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_366, [sym_size_int, 1500, 1]);  convert_element_type_366 = None
	        convert_element_type_368: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_367, torch.float32);  convert_element_type_367 = None
	        convert_element_type_369: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_963, torch.float32);  view_963 = None
	        sub_2829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_368, convert_element_type_369);  convert_element_type_368 = convert_element_type_369 = None
	        view_962: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_183, [sym_size_int, 1500, 1]);  clamp_min_183 = None
	        mul_5990: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2829, view_962);  sub_2829 = view_962 = None
	        view_965: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_967: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_370: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_965, torch.float32);  view_965 = None
	        convert_element_type_371: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_967, torch.float32);  view_967 = None
	        sub_2833: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_370, convert_element_type_371);  convert_element_type_370 = convert_element_type_371 = None
	        view_966: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_5995: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2833, view_966);  sub_2833 = view_966 = None
	        view_968: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5995, [1280, 1280]);  mul_5995 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_9212, [2])
	        full_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_62, full_125);  amax_62 = full_125 = None
	        amin_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_9212, [2])
	        full_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_62: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_62, full_124);  amin_62 = full_124 = None
	        sub_2847: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_62, minimum_62);  maximum_62 = None
	        div_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2847, 255.0);  sub_2847 = None
	        clamp_min_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_124, 1.1920928955078125e-07);  div_124 = None
	        div_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_62, clamp_min_186);  minimum_62 = None
	        round_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_125);  div_125 = None
	        sub_2853: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_125);  round_125 = None
	        clamp_min_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2853, -128);  sub_2853 = None
	        clamp_max_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_187, 127);  clamp_min_187 = None
	        view_974: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_186, [sym_size_int, 1500, 1])
	        reciprocal_62: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_974);  view_974 = None
	        mul_6064: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_62, 1.0);  reciprocal_62 = None
	        mul_6067: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_9212, mul_6064);  add_9212 = mul_6064 = None
	        round_126: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6067);  mul_6067 = None
	        convert_element_type_372: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_124, torch.int8);  clamp_max_124 = None
	        view_975: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_372, [sym_size_int, 1500, 1])
	        add_9599: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_126, view_975);  round_126 = view_975 = None
	        clamp_min_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9599, -128);  add_9599 = None
	        clamp_max_125: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_188, 127);  clamp_min_188 = None
	        convert_element_type_373: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_125, torch.int8);  clamp_max_125 = None
	        view_979: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_372, [sym_size_int, 1500, 1]);  convert_element_type_372 = None
	        convert_element_type_374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_373, torch.float32);  convert_element_type_373 = None
	        convert_element_type_375: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_979, torch.float32);  view_979 = None
	        sub_2873: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_374, convert_element_type_375);  convert_element_type_374 = convert_element_type_375 = None
	        view_978: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_186, [sym_size_int, 1500, 1]);  clamp_min_186 = None
	        mul_6089: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2873, view_978);  sub_2873 = view_978 = None
	        view_981: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_983: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_376: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_981, torch.float32);  view_981 = None
	        convert_element_type_377: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_983, torch.float32);  view_983 = None
	        sub_2877: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_376, convert_element_type_377);  convert_element_type_376 = convert_element_type_377 = None
	        view_982: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_6094: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2877, view_982);  sub_2877 = view_982 = None
	        view_984: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6094, [1280, 1280]);  mul_6094 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_5904: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_953: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5894, [mul_5904, 1280]);  mul_5894 = mul_5904 = None
	        permute_101: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_952, [1, 0]);  view_952 = None
	        mm_default_109: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_953, permute_101);  view_953 = permute_101 = None
	        add_tensor_109: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_109, model_audio_tower_layers_10_self_attn_q_proj_bias);  mm_default_109 = model_audio_tower_layers_10_self_attn_q_proj_bias = None
	        view_954: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_109, [sym_size_int, 1500, 1280]);  add_tensor_109 = None
	        mul_5911: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_954, 0.125);  view_954 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_955: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5911, [sym_size_int, 1500, 20, 64]);  mul_5911 = None
	        permute_102: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_955, [0, 2, 1, 3]);  view_955 = None
	        clone_82: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_102, memory_format = torch.contiguous_format);  permute_102 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_5998: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_969: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_5990, [mul_5998, 1280]);  mul_5990 = mul_5998 = None
	        permute_103: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_968, [1, 0]);  view_968 = None
	        mm_10: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_969, permute_103);  view_969 = permute_103 = None
	        view_970: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_10, [sym_size_int, 1500, 1280]);  mm_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_971: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_970, [sym_size_int, -1, 20, 64]);  view_970 = None
	        permute_104: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_971, [0, 2, 1, 3]);  view_971 = None
	        clone_83: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_104, memory_format = torch.contiguous_format);  permute_104 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_6099: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_985: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6089, [mul_6099, 1280]);  mul_6089 = mul_6099 = None
	        permute_105: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_984, [1, 0]);  view_984 = None
	        mm_default_108: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_985, permute_105);  view_985 = permute_105 = None
	        add_tensor_108: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_108, model_audio_tower_layers_10_self_attn_v_proj_bias);  mm_default_108 = model_audio_tower_layers_10_self_attn_v_proj_bias = None
	        view_986: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_108, [sym_size_int, 1500, 1280]);  add_tensor_108 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_987: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_986, [sym_size_int, -1, 20, 64]);  view_986 = None
	        permute_106: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_987, [0, 2, 1, 3]);  view_987 = None
	        clone_84: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_106, memory_format = torch.contiguous_format);  permute_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_10 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_82, clone_83, clone_84, None, False, scale = 1.0);  clone_82 = clone_83 = clone_84 = None
	        getitem_82: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_10[0];  _scaled_dot_product_efficient_attention_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_107: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_82, [0, 2, 1, 3]);  getitem_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_988: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_107, [sym_size_int, 1500, -1]);  permute_107 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_988, [2])
	        full_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_63, full_127);  amax_63 = full_127 = None
	        amin_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_988, [2])
	        full_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_63: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_63, full_126);  amin_63 = full_126 = None
	        sub_2895: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_63, minimum_63);  maximum_63 = None
	        div_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2895, 255.0);  sub_2895 = None
	        clamp_min_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_126, 1.1920928955078125e-07);  div_126 = None
	        div_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_63, clamp_min_189);  minimum_63 = None
	        round_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_127);  div_127 = None
	        sub_2901: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_127);  round_127 = None
	        clamp_min_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2901, -128);  sub_2901 = None
	        clamp_max_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_190, 127);  clamp_min_190 = None
	        view_991: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_189, [sym_size_int, 1500, 1])
	        reciprocal_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_991);  view_991 = None
	        mul_6169: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_63, 1.0);  reciprocal_63 = None
	        mul_6172: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_988, mul_6169);  view_988 = mul_6169 = None
	        round_128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6172);  mul_6172 = None
	        convert_element_type_378: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_126, torch.int8);  clamp_max_126 = None
	        view_992: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_378, [sym_size_int, 1500, 1])
	        add_9763: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_128, view_992);  round_128 = view_992 = None
	        clamp_min_191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9763, -128);  add_9763 = None
	        clamp_max_127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_191, 127);  clamp_min_191 = None
	        convert_element_type_379: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_127, torch.int8);  clamp_max_127 = None
	        view_996: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_378, [sym_size_int, 1500, 1]);  convert_element_type_378 = None
	        convert_element_type_380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_379, torch.float32);  convert_element_type_379 = None
	        convert_element_type_381: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_996, torch.float32);  view_996 = None
	        sub_2921: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_380, convert_element_type_381);  convert_element_type_380 = convert_element_type_381 = None
	        view_995: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_189, [sym_size_int, 1500, 1]);  clamp_min_189 = None
	        mul_6194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2921, view_995);  sub_2921 = view_995 = None
	        view_998: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1000: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_382: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_998, torch.float32);  view_998 = None
	        convert_element_type_383: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1000, torch.float32);  view_1000 = None
	        sub_2925: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_382, convert_element_type_383);  convert_element_type_382 = convert_element_type_383 = None
	        view_999: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_6199: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2925, view_999);  sub_2925 = view_999 = None
	        view_1001: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6199, [1280, 1280]);  mul_6199 = None
	        mul_6204: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1002: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6194, [mul_6204, 1280]);  mul_6194 = mul_6204 = None
	        permute_108: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1001, [1, 0]);  view_1001 = None
	        mm_default_107: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1002, permute_108);  view_1002 = permute_108 = None
	        add_tensor_107: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_107, model_audio_tower_layers_10_self_attn_out_proj_bias);  mm_default_107 = model_audio_tower_layers_10_self_attn_out_proj_bias = None
	        view_1003: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_107, [sym_size_int, 1500, 1280]);  add_tensor_107 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_9826: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_9206, view_1003);  add_9206 = view_1003 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_86: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_9826, memory_format = torch.contiguous_format)
	        var_mean_21 = torch.ops.aten.var_mean.correction(clone_86, [2], correction = 0, keepdim = True)
	        getitem_86: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_21[0]
	        getitem_87: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_21[1];  var_mean_21 = None
	        sub_2931: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_86, getitem_87);  clone_86 = getitem_87 = None
	        add_9831: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_86, 1e-05);  getitem_86 = None
	        rsqrt_21: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_9831);  add_9831 = None
	        mul_6215: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2931, rsqrt_21);  sub_2931 = rsqrt_21 = None
	        mul_6216: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6215, model_audio_tower_layers_10_final_layer_norm_weight);  mul_6215 = model_audio_tower_layers_10_final_layer_norm_weight = None
	        add_9832: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6216, model_audio_tower_layers_10_final_layer_norm_bias);  mul_6216 = model_audio_tower_layers_10_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_9832, [2])
	        full_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_64, full_129);  amax_64 = full_129 = None
	        amin_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_9832, [2])
	        full_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_64: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_64, full_128);  amin_64 = full_128 = None
	        sub_2942: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_64, minimum_64);  maximum_64 = None
	        div_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2942, 255.0);  sub_2942 = None
	        clamp_min_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_128, 1.1920928955078125e-07);  div_128 = None
	        div_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_64, clamp_min_192);  minimum_64 = None
	        round_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_129);  div_129 = None
	        sub_2948: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_129);  round_129 = None
	        clamp_min_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2948, -128);  sub_2948 = None
	        clamp_max_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_193, 127);  clamp_min_193 = None
	        view_1006: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_192, [sym_size_int, 1500, 1])
	        reciprocal_64: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1006);  view_1006 = None
	        mul_6264: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_64, 1.0);  reciprocal_64 = None
	        mul_6267: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_9832, mul_6264);  add_9832 = mul_6264 = None
	        round_130: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6267);  mul_6267 = None
	        convert_element_type_384: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_128, torch.int8);  clamp_max_128 = None
	        view_1007: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_384, [sym_size_int, 1500, 1])
	        add_9919: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_130, view_1007);  round_130 = view_1007 = None
	        clamp_min_194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_9919, -128);  add_9919 = None
	        clamp_max_129: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_194, 127);  clamp_min_194 = None
	        convert_element_type_385: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_129, torch.int8);  clamp_max_129 = None
	        view_1011: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_384, [sym_size_int, 1500, 1]);  convert_element_type_384 = None
	        convert_element_type_386: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_385, torch.float32);  convert_element_type_385 = None
	        convert_element_type_387: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1011, torch.float32);  view_1011 = None
	        sub_2968: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_386, convert_element_type_387);  convert_element_type_386 = convert_element_type_387 = None
	        view_1010: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_192, [sym_size_int, 1500, 1]);  clamp_min_192 = None
	        mul_6289: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2968, view_1010);  sub_2968 = view_1010 = None
	        view_1013: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_10_fc1_parametrizations_weight_original0 = None
	        view_1015: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_10_fc1_parametrizations_weight_original2 = None
	        convert_element_type_388: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1013, torch.float32);  view_1013 = None
	        convert_element_type_389: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1015, torch.float32);  view_1015 = None
	        sub_2972: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_388, convert_element_type_389);  convert_element_type_388 = convert_element_type_389 = None
	        view_1014: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_10_fc1_parametrizations_weight_original1 = None
	        mul_6294: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_2972, view_1014);  sub_2972 = view_1014 = None
	        view_1016: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6294, [5120, 1280]);  mul_6294 = None
	        mul_6299: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1017: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6289, [mul_6299, 1280]);  mul_6289 = mul_6299 = None
	        permute_109: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1016, [1, 0]);  view_1016 = None
	        mm_default_106: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_1017, permute_109);  view_1017 = permute_109 = None
	        add_tensor_106: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_106, model_audio_tower_layers_10_fc1_bias);  mm_default_106 = model_audio_tower_layers_10_fc1_bias = None
	        view_1018: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_106, [sym_size_int, 1500, 5120]);  add_tensor_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_6306: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1018, 0.5)
	        mul_6307: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1018, 0.7071067811865476);  view_1018 = None
	        erf_12: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_6307);  mul_6307 = None
	        add_9978: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_12, 1);  erf_12 = None
	        mul_6308: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6306, add_9978);  mul_6306 = add_9978 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_6308, [2])
	        full_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_65, full_131);  amax_65 = full_131 = None
	        amin_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_6308, [2])
	        full_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_65: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_65, full_130);  amin_65 = full_130 = None
	        sub_2985: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_65, minimum_65);  maximum_65 = None
	        div_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_2985, 255.0);  sub_2985 = None
	        clamp_min_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_130, 1.1920928955078125e-07);  div_130 = None
	        div_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_65, clamp_min_195);  minimum_65 = None
	        round_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_131);  div_131 = None
	        sub_2991: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_131);  round_131 = None
	        clamp_min_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_2991, -128);  sub_2991 = None
	        clamp_max_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_196, 127);  clamp_min_196 = None
	        view_1021: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_195, [sym_size_int, 1500, 1])
	        reciprocal_65: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1021);  view_1021 = None
	        mul_6354: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_65, 1.0);  reciprocal_65 = None
	        mul_6357: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6308, mul_6354);  mul_6308 = mul_6354 = None
	        round_132: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_6357);  mul_6357 = None
	        convert_element_type_390: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_130, torch.int8);  clamp_max_130 = None
	        view_1022: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_390, [sym_size_int, 1500, 1])
	        add_10061: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_132, view_1022);  round_132 = view_1022 = None
	        clamp_min_197: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10061, -128);  add_10061 = None
	        clamp_max_131: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_197, 127);  clamp_min_197 = None
	        convert_element_type_391: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_131, torch.int8);  clamp_max_131 = None
	        view_1026: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_390, [sym_size_int, 1500, 1]);  convert_element_type_390 = None
	        convert_element_type_392: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_391, torch.float32);  convert_element_type_391 = None
	        convert_element_type_393: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1026, torch.float32);  view_1026 = None
	        sub_3011: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_392, convert_element_type_393);  convert_element_type_392 = convert_element_type_393 = None
	        view_1025: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_195, [sym_size_int, 1500, 1]);  clamp_min_195 = None
	        mul_6379: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3011, view_1025);  sub_3011 = view_1025 = None
	        view_1028: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_10_fc2_parametrizations_weight_original0 = None
	        view_1030: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_10_fc2_parametrizations_weight_original2 = None
	        convert_element_type_394: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1028, torch.float32);  view_1028 = None
	        convert_element_type_395: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1030, torch.float32);  view_1030 = None
	        sub_3015: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_394, convert_element_type_395);  convert_element_type_394 = convert_element_type_395 = None
	        view_1029: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_10_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_10_fc2_parametrizations_weight_original1 = None
	        mul_6384: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3015, view_1029);  sub_3015 = view_1029 = None
	        view_1031: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6384, [1280, 5120]);  mul_6384 = None
	        mul_6389: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1032: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6379, [mul_6389, 5120]);  mul_6379 = mul_6389 = None
	        permute_110: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1031, [1, 0]);  view_1031 = None
	        mm_default_105: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1032, permute_110);  view_1032 = permute_110 = None
	        add_tensor_105: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_105, model_audio_tower_layers_10_fc2_bias);  mm_default_105 = model_audio_tower_layers_10_fc2_bias = None
	        view_1033: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_105, [sym_size_int, 1500, 1280]);  add_tensor_105 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_10124: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_9826, view_1033);  add_9826 = view_1033 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_89: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_10124, memory_format = torch.contiguous_format)
	        var_mean_22 = torch.ops.aten.var_mean.correction(clone_89, [2], correction = 0, keepdim = True)
	        getitem_88: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_22[0]
	        getitem_89: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_22[1];  var_mean_22 = None
	        sub_3021: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_89, getitem_89);  clone_89 = getitem_89 = None
	        add_10129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_88, 1e-05);  getitem_88 = None
	        rsqrt_22: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_10129);  add_10129 = None
	        mul_6400: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3021, rsqrt_22);  sub_3021 = rsqrt_22 = None
	        mul_6401: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6400, model_audio_tower_layers_11_self_attn_layer_norm_weight);  mul_6400 = model_audio_tower_layers_11_self_attn_layer_norm_weight = None
	        add_10130: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6401, model_audio_tower_layers_11_self_attn_layer_norm_bias);  mul_6401 = model_audio_tower_layers_11_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_10130, [2])
	        full_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_66, full_133);  amax_66 = full_133 = None
	        amin_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_10130, [2])
	        full_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_66: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_66, full_132);  amin_66 = full_132 = None
	        sub_3032: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_66, minimum_66);  maximum_66 = None
	        div_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3032, 255.0);  sub_3032 = None
	        clamp_min_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_132, 1.1920928955078125e-07);  div_132 = None
	        div_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_66, clamp_min_198);  minimum_66 = None
	        round_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_133);  div_133 = None
	        sub_3038: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_133);  round_133 = None
	        clamp_min_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3038, -128);  sub_3038 = None
	        clamp_max_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_199, 127);  clamp_min_199 = None
	        view_1036: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_198, [sym_size_int, 1500, 1])
	        reciprocal_66: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1036);  view_1036 = None
	        mul_6449: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_66, 1.0);  reciprocal_66 = None
	        mul_6452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_10130, mul_6449);  mul_6449 = None
	        round_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6452);  mul_6452 = None
	        convert_element_type_396: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_132, torch.int8);  clamp_max_132 = None
	        view_1037: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_396, [sym_size_int, 1500, 1])
	        add_10217: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_134, view_1037);  round_134 = view_1037 = None
	        clamp_min_200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10217, -128);  add_10217 = None
	        clamp_max_133: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_200, 127);  clamp_min_200 = None
	        convert_element_type_397: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_133, torch.int8);  clamp_max_133 = None
	        view_1041: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_396, [sym_size_int, 1500, 1]);  convert_element_type_396 = None
	        convert_element_type_398: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_397, torch.float32);  convert_element_type_397 = None
	        convert_element_type_399: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1041, torch.float32);  view_1041 = None
	        sub_3058: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_398, convert_element_type_399);  convert_element_type_398 = convert_element_type_399 = None
	        view_1040: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_198, [sym_size_int, 1500, 1]);  clamp_min_198 = None
	        mul_6474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3058, view_1040);  sub_3058 = view_1040 = None
	        view_1043: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1045: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_400: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1043, torch.float32);  view_1043 = None
	        convert_element_type_401: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1045, torch.float32);  view_1045 = None
	        sub_3062: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_400, convert_element_type_401);  convert_element_type_400 = convert_element_type_401 = None
	        view_1044: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_6479: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3062, view_1044);  sub_3062 = view_1044 = None
	        view_1046: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6479, [1280, 1280]);  mul_6479 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_10130, [2])
	        full_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_67, full_135);  amax_67 = full_135 = None
	        amin_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_10130, [2])
	        full_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_67: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_67, full_134);  amin_67 = full_134 = None
	        sub_3077: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_67, minimum_67);  maximum_67 = None
	        div_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3077, 255.0);  sub_3077 = None
	        clamp_min_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_134, 1.1920928955078125e-07);  div_134 = None
	        div_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_67, clamp_min_201);  minimum_67 = None
	        round_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_135);  div_135 = None
	        sub_3083: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_135);  round_135 = None
	        clamp_min_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3083, -128);  sub_3083 = None
	        clamp_max_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_202, 127);  clamp_min_202 = None
	        view_1052: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_201, [sym_size_int, 1500, 1])
	        reciprocal_67: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1052);  view_1052 = None
	        mul_6545: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_67, 1.0);  reciprocal_67 = None
	        mul_6548: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_10130, mul_6545);  mul_6545 = None
	        round_136: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6548);  mul_6548 = None
	        convert_element_type_402: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_134, torch.int8);  clamp_max_134 = None
	        view_1053: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_402, [sym_size_int, 1500, 1])
	        add_10369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_136, view_1053);  round_136 = view_1053 = None
	        clamp_min_203: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10369, -128);  add_10369 = None
	        clamp_max_135: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_203, 127);  clamp_min_203 = None
	        convert_element_type_403: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_135, torch.int8);  clamp_max_135 = None
	        view_1057: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_402, [sym_size_int, 1500, 1]);  convert_element_type_402 = None
	        convert_element_type_404: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_403, torch.float32);  convert_element_type_403 = None
	        convert_element_type_405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1057, torch.float32);  view_1057 = None
	        sub_3103: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_404, convert_element_type_405);  convert_element_type_404 = convert_element_type_405 = None
	        view_1056: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_201, [sym_size_int, 1500, 1]);  clamp_min_201 = None
	        mul_6570: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3103, view_1056);  sub_3103 = view_1056 = None
	        view_1059: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1061: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_406: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1059, torch.float32);  view_1059 = None
	        convert_element_type_407: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1061, torch.float32);  view_1061 = None
	        sub_3107: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_406, convert_element_type_407);  convert_element_type_406 = convert_element_type_407 = None
	        view_1060: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_6575: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3107, view_1060);  sub_3107 = view_1060 = None
	        view_1062: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6575, [1280, 1280]);  mul_6575 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_10130, [2])
	        full_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_68, full_137);  amax_68 = full_137 = None
	        amin_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_10130, [2])
	        full_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_68: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_68, full_136);  amin_68 = full_136 = None
	        sub_3121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_68, minimum_68);  maximum_68 = None
	        div_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3121, 255.0);  sub_3121 = None
	        clamp_min_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_136, 1.1920928955078125e-07);  div_136 = None
	        div_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_68, clamp_min_204);  minimum_68 = None
	        round_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_137);  div_137 = None
	        sub_3127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_137);  round_137 = None
	        clamp_min_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3127, -128);  sub_3127 = None
	        clamp_max_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_205, 127);  clamp_min_205 = None
	        view_1068: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_204, [sym_size_int, 1500, 1])
	        reciprocal_68: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1068);  view_1068 = None
	        mul_6644: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_68, 1.0);  reciprocal_68 = None
	        mul_6647: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_10130, mul_6644);  add_10130 = mul_6644 = None
	        round_138: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6647);  mul_6647 = None
	        convert_element_type_408: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_136, torch.int8);  clamp_max_136 = None
	        view_1069: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_408, [sym_size_int, 1500, 1])
	        add_10517: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_138, view_1069);  round_138 = view_1069 = None
	        clamp_min_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10517, -128);  add_10517 = None
	        clamp_max_137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_206, 127);  clamp_min_206 = None
	        convert_element_type_409: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_137, torch.int8);  clamp_max_137 = None
	        view_1073: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_408, [sym_size_int, 1500, 1]);  convert_element_type_408 = None
	        convert_element_type_410: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_409, torch.float32);  convert_element_type_409 = None
	        convert_element_type_411: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1073, torch.float32);  view_1073 = None
	        sub_3147: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_410, convert_element_type_411);  convert_element_type_410 = convert_element_type_411 = None
	        view_1072: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_204, [sym_size_int, 1500, 1]);  clamp_min_204 = None
	        mul_6669: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3147, view_1072);  sub_3147 = view_1072 = None
	        view_1075: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1077: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_412: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1075, torch.float32);  view_1075 = None
	        convert_element_type_413: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1077, torch.float32);  view_1077 = None
	        sub_3151: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_412, convert_element_type_413);  convert_element_type_412 = convert_element_type_413 = None
	        view_1076: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_6674: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3151, view_1076);  sub_3151 = view_1076 = None
	        view_1078: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6674, [1280, 1280]);  mul_6674 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_6484: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1047: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6474, [mul_6484, 1280]);  mul_6474 = mul_6484 = None
	        permute_111: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1046, [1, 0]);  view_1046 = None
	        mm_default_104: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1047, permute_111);  view_1047 = permute_111 = None
	        add_tensor_104: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_104, model_audio_tower_layers_11_self_attn_q_proj_bias);  mm_default_104 = model_audio_tower_layers_11_self_attn_q_proj_bias = None
	        view_1048: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_104, [sym_size_int, 1500, 1280]);  add_tensor_104 = None
	        mul_6491: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1048, 0.125);  view_1048 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1049: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6491, [sym_size_int, 1500, 20, 64]);  mul_6491 = None
	        permute_112: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1049, [0, 2, 1, 3]);  view_1049 = None
	        clone_90: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_112, memory_format = torch.contiguous_format);  permute_112 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_6578: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1063: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6570, [mul_6578, 1280]);  mul_6570 = mul_6578 = None
	        permute_113: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1062, [1, 0]);  view_1062 = None
	        mm_11: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1063, permute_113);  view_1063 = permute_113 = None
	        view_1064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_11, [sym_size_int, 1500, 1280]);  mm_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1065: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1064, [sym_size_int, -1, 20, 64]);  view_1064 = None
	        permute_114: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1065, [0, 2, 1, 3]);  view_1065 = None
	        clone_91: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_114, memory_format = torch.contiguous_format);  permute_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_6679: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1079: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6669, [mul_6679, 1280]);  mul_6669 = mul_6679 = None
	        permute_115: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1078, [1, 0]);  view_1078 = None
	        mm_default_103: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1079, permute_115);  view_1079 = permute_115 = None
	        add_tensor_103: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_103, model_audio_tower_layers_11_self_attn_v_proj_bias);  mm_default_103 = model_audio_tower_layers_11_self_attn_v_proj_bias = None
	        view_1080: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_103, [sym_size_int, 1500, 1280]);  add_tensor_103 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1081: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1080, [sym_size_int, -1, 20, 64]);  view_1080 = None
	        permute_116: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1081, [0, 2, 1, 3]);  view_1081 = None
	        clone_92: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_116, memory_format = torch.contiguous_format);  permute_116 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_11 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_90, clone_91, clone_92, None, False, scale = 1.0);  clone_90 = clone_91 = clone_92 = None
	        getitem_90: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_11[0];  _scaled_dot_product_efficient_attention_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_117: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_90, [0, 2, 1, 3]);  getitem_90 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1082: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_117, [sym_size_int, 1500, -1]);  permute_117 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1082, [2])
	        full_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_69, full_139);  amax_69 = full_139 = None
	        amin_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1082, [2])
	        full_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_69: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_69, full_138);  amin_69 = full_138 = None
	        sub_3169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_69, minimum_69);  maximum_69 = None
	        div_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3169, 255.0);  sub_3169 = None
	        clamp_min_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_138, 1.1920928955078125e-07);  div_138 = None
	        div_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_69, clamp_min_207);  minimum_69 = None
	        round_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_139);  div_139 = None
	        sub_3175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_139);  round_139 = None
	        clamp_min_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3175, -128);  sub_3175 = None
	        clamp_max_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_208, 127);  clamp_min_208 = None
	        view_1085: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_207, [sym_size_int, 1500, 1])
	        reciprocal_69: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1085);  view_1085 = None
	        mul_6749: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_69, 1.0);  reciprocal_69 = None
	        mul_6752: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1082, mul_6749);  view_1082 = mul_6749 = None
	        round_140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6752);  mul_6752 = None
	        convert_element_type_414: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_138, torch.int8);  clamp_max_138 = None
	        view_1086: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_414, [sym_size_int, 1500, 1])
	        add_10681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_140, view_1086);  round_140 = view_1086 = None
	        clamp_min_209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10681, -128);  add_10681 = None
	        clamp_max_139: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_209, 127);  clamp_min_209 = None
	        convert_element_type_415: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_139, torch.int8);  clamp_max_139 = None
	        view_1090: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_414, [sym_size_int, 1500, 1]);  convert_element_type_414 = None
	        convert_element_type_416: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_415, torch.float32);  convert_element_type_415 = None
	        convert_element_type_417: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1090, torch.float32);  view_1090 = None
	        sub_3195: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_416, convert_element_type_417);  convert_element_type_416 = convert_element_type_417 = None
	        view_1089: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_207, [sym_size_int, 1500, 1]);  clamp_min_207 = None
	        mul_6774: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3195, view_1089);  sub_3195 = view_1089 = None
	        view_1092: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1094: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_418: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1092, torch.float32);  view_1092 = None
	        convert_element_type_419: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1094, torch.float32);  view_1094 = None
	        sub_3199: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_418, convert_element_type_419);  convert_element_type_418 = convert_element_type_419 = None
	        view_1093: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_6779: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3199, view_1093);  sub_3199 = view_1093 = None
	        view_1095: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6779, [1280, 1280]);  mul_6779 = None
	        mul_6784: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1096: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6774, [mul_6784, 1280]);  mul_6774 = mul_6784 = None
	        permute_118: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1095, [1, 0]);  view_1095 = None
	        mm_default_102: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1096, permute_118);  view_1096 = permute_118 = None
	        add_tensor_102: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_102, model_audio_tower_layers_11_self_attn_out_proj_bias);  mm_default_102 = model_audio_tower_layers_11_self_attn_out_proj_bias = None
	        view_1097: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_102, [sym_size_int, 1500, 1280]);  add_tensor_102 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_10744: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_10124, view_1097);  add_10124 = view_1097 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_94: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_10744, memory_format = torch.contiguous_format)
	        var_mean_23 = torch.ops.aten.var_mean.correction(clone_94, [2], correction = 0, keepdim = True)
	        getitem_94: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_23[0]
	        getitem_95: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_23[1];  var_mean_23 = None
	        sub_3205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_94, getitem_95);  clone_94 = getitem_95 = None
	        add_10749: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_94, 1e-05);  getitem_94 = None
	        rsqrt_23: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_10749);  add_10749 = None
	        mul_6795: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3205, rsqrt_23);  sub_3205 = rsqrt_23 = None
	        mul_6796: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6795, model_audio_tower_layers_11_final_layer_norm_weight);  mul_6795 = model_audio_tower_layers_11_final_layer_norm_weight = None
	        add_10750: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6796, model_audio_tower_layers_11_final_layer_norm_bias);  mul_6796 = model_audio_tower_layers_11_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_10750, [2])
	        full_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_70, full_141);  amax_70 = full_141 = None
	        amin_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_10750, [2])
	        full_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_70: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_70, full_140);  amin_70 = full_140 = None
	        sub_3216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_70, minimum_70);  maximum_70 = None
	        div_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3216, 255.0);  sub_3216 = None
	        clamp_min_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_140, 1.1920928955078125e-07);  div_140 = None
	        div_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_70, clamp_min_210);  minimum_70 = None
	        round_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_141);  div_141 = None
	        sub_3222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_141);  round_141 = None
	        clamp_min_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3222, -128);  sub_3222 = None
	        clamp_max_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_211, 127);  clamp_min_211 = None
	        view_1100: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_210, [sym_size_int, 1500, 1])
	        reciprocal_70: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1100);  view_1100 = None
	        mul_6844: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_70, 1.0);  reciprocal_70 = None
	        mul_6847: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_10750, mul_6844);  add_10750 = mul_6844 = None
	        round_142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_6847);  mul_6847 = None
	        convert_element_type_420: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_140, torch.int8);  clamp_max_140 = None
	        view_1101: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_420, [sym_size_int, 1500, 1])
	        add_10837: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_142, view_1101);  round_142 = view_1101 = None
	        clamp_min_212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10837, -128);  add_10837 = None
	        clamp_max_141: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_212, 127);  clamp_min_212 = None
	        convert_element_type_421: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_141, torch.int8);  clamp_max_141 = None
	        view_1105: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_420, [sym_size_int, 1500, 1]);  convert_element_type_420 = None
	        convert_element_type_422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_421, torch.float32);  convert_element_type_421 = None
	        convert_element_type_423: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1105, torch.float32);  view_1105 = None
	        sub_3242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_422, convert_element_type_423);  convert_element_type_422 = convert_element_type_423 = None
	        view_1104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_210, [sym_size_int, 1500, 1]);  clamp_min_210 = None
	        mul_6869: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3242, view_1104);  sub_3242 = view_1104 = None
	        view_1107: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_11_fc1_parametrizations_weight_original0 = None
	        view_1109: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_11_fc1_parametrizations_weight_original2 = None
	        convert_element_type_424: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1107, torch.float32);  view_1107 = None
	        convert_element_type_425: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1109, torch.float32);  view_1109 = None
	        sub_3246: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_424, convert_element_type_425);  convert_element_type_424 = convert_element_type_425 = None
	        view_1108: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_11_fc1_parametrizations_weight_original1 = None
	        mul_6874: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3246, view_1108);  sub_3246 = view_1108 = None
	        view_1110: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6874, [5120, 1280]);  mul_6874 = None
	        mul_6879: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1111: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6869, [mul_6879, 1280]);  mul_6869 = mul_6879 = None
	        permute_119: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1110, [1, 0]);  view_1110 = None
	        mm_default_101: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_1111, permute_119);  view_1111 = permute_119 = None
	        add_tensor_101: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_101, model_audio_tower_layers_11_fc1_bias);  mm_default_101 = model_audio_tower_layers_11_fc1_bias = None
	        view_1112: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_101, [sym_size_int, 1500, 5120]);  add_tensor_101 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_6886: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1112, 0.5)
	        mul_6887: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1112, 0.7071067811865476);  view_1112 = None
	        erf_13: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_6887);  mul_6887 = None
	        add_10896: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_13, 1);  erf_13 = None
	        mul_6888: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6886, add_10896);  mul_6886 = add_10896 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_6888, [2])
	        full_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_71, full_143);  amax_71 = full_143 = None
	        amin_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_6888, [2])
	        full_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_71: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_71, full_142);  amin_71 = full_142 = None
	        sub_3259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_71, minimum_71);  maximum_71 = None
	        div_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3259, 255.0);  sub_3259 = None
	        clamp_min_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_142, 1.1920928955078125e-07);  div_142 = None
	        div_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_71, clamp_min_213);  minimum_71 = None
	        round_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_143);  div_143 = None
	        sub_3265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_143);  round_143 = None
	        clamp_min_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3265, -128);  sub_3265 = None
	        clamp_max_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_214, 127);  clamp_min_214 = None
	        view_1115: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_213, [sym_size_int, 1500, 1])
	        reciprocal_71: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1115);  view_1115 = None
	        mul_6934: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_71, 1.0);  reciprocal_71 = None
	        mul_6937: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6888, mul_6934);  mul_6888 = mul_6934 = None
	        round_144: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_6937);  mul_6937 = None
	        convert_element_type_426: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_142, torch.int8);  clamp_max_142 = None
	        view_1116: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_426, [sym_size_int, 1500, 1])
	        add_10979: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_144, view_1116);  round_144 = view_1116 = None
	        clamp_min_215: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_10979, -128);  add_10979 = None
	        clamp_max_143: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_215, 127);  clamp_min_215 = None
	        convert_element_type_427: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_143, torch.int8);  clamp_max_143 = None
	        view_1120: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_426, [sym_size_int, 1500, 1]);  convert_element_type_426 = None
	        convert_element_type_428: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_427, torch.float32);  convert_element_type_427 = None
	        convert_element_type_429: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1120, torch.float32);  view_1120 = None
	        sub_3285: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_428, convert_element_type_429);  convert_element_type_428 = convert_element_type_429 = None
	        view_1119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_213, [sym_size_int, 1500, 1]);  clamp_min_213 = None
	        mul_6959: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3285, view_1119);  sub_3285 = view_1119 = None
	        view_1122: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_11_fc2_parametrizations_weight_original0 = None
	        view_1124: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_11_fc2_parametrizations_weight_original2 = None
	        convert_element_type_430: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1122, torch.float32);  view_1122 = None
	        convert_element_type_431: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1124, torch.float32);  view_1124 = None
	        sub_3289: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_430, convert_element_type_431);  convert_element_type_430 = convert_element_type_431 = None
	        view_1123: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_11_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_11_fc2_parametrizations_weight_original1 = None
	        mul_6964: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3289, view_1123);  sub_3289 = view_1123 = None
	        view_1125: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6964, [1280, 5120]);  mul_6964 = None
	        mul_6969: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1126: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_6959, [mul_6969, 5120]);  mul_6959 = mul_6969 = None
	        permute_120: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1125, [1, 0]);  view_1125 = None
	        mm_default_100: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1126, permute_120);  view_1126 = permute_120 = None
	        add_tensor_100: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_100, model_audio_tower_layers_11_fc2_bias);  mm_default_100 = model_audio_tower_layers_11_fc2_bias = None
	        view_1127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_100, [sym_size_int, 1500, 1280]);  add_tensor_100 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_11042: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_10744, view_1127);  add_10744 = view_1127 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_97: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_11042, memory_format = torch.contiguous_format)
	        var_mean_24 = torch.ops.aten.var_mean.correction(clone_97, [2], correction = 0, keepdim = True)
	        getitem_96: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_24[0]
	        getitem_97: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_24[1];  var_mean_24 = None
	        sub_3295: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_97, getitem_97);  clone_97 = getitem_97 = None
	        add_11047: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_96, 1e-05);  getitem_96 = None
	        rsqrt_24: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_11047);  add_11047 = None
	        mul_6980: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3295, rsqrt_24);  sub_3295 = rsqrt_24 = None
	        mul_6981: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_6980, model_audio_tower_layers_12_self_attn_layer_norm_weight);  mul_6980 = model_audio_tower_layers_12_self_attn_layer_norm_weight = None
	        add_11048: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_6981, model_audio_tower_layers_12_self_attn_layer_norm_bias);  mul_6981 = model_audio_tower_layers_12_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11048, [2])
	        full_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_72, full_145);  amax_72 = full_145 = None
	        amin_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11048, [2])
	        full_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_72: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_72, full_144);  amin_72 = full_144 = None
	        sub_3306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_72, minimum_72);  maximum_72 = None
	        div_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3306, 255.0);  sub_3306 = None
	        clamp_min_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_144, 1.1920928955078125e-07);  div_144 = None
	        div_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_72, clamp_min_216);  minimum_72 = None
	        round_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_145);  div_145 = None
	        sub_3312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_145);  round_145 = None
	        clamp_min_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3312, -128);  sub_3312 = None
	        clamp_max_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_217, 127);  clamp_min_217 = None
	        view_1130: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_216, [sym_size_int, 1500, 1])
	        reciprocal_72: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1130);  view_1130 = None
	        mul_7029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_72, 1.0);  reciprocal_72 = None
	        mul_7032: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11048, mul_7029);  mul_7029 = None
	        round_146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7032);  mul_7032 = None
	        convert_element_type_432: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_144, torch.int8);  clamp_max_144 = None
	        view_1131: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_432, [sym_size_int, 1500, 1])
	        add_11135: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_146, view_1131);  round_146 = view_1131 = None
	        clamp_min_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11135, -128);  add_11135 = None
	        clamp_max_145: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_218, 127);  clamp_min_218 = None
	        convert_element_type_433: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_145, torch.int8);  clamp_max_145 = None
	        view_1135: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_432, [sym_size_int, 1500, 1]);  convert_element_type_432 = None
	        convert_element_type_434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_433, torch.float32);  convert_element_type_433 = None
	        convert_element_type_435: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1135, torch.float32);  view_1135 = None
	        sub_3332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_434, convert_element_type_435);  convert_element_type_434 = convert_element_type_435 = None
	        view_1134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_216, [sym_size_int, 1500, 1]);  clamp_min_216 = None
	        mul_7054: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3332, view_1134);  sub_3332 = view_1134 = None
	        view_1137: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1139: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_436: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1137, torch.float32);  view_1137 = None
	        convert_element_type_437: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1139, torch.float32);  view_1139 = None
	        sub_3336: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_436, convert_element_type_437);  convert_element_type_436 = convert_element_type_437 = None
	        view_1138: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_7059: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3336, view_1138);  sub_3336 = view_1138 = None
	        view_1140: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7059, [1280, 1280]);  mul_7059 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11048, [2])
	        full_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_73, full_147);  amax_73 = full_147 = None
	        amin_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11048, [2])
	        full_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_73: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_73, full_146);  amin_73 = full_146 = None
	        sub_3351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_73, minimum_73);  maximum_73 = None
	        div_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3351, 255.0);  sub_3351 = None
	        clamp_min_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_146, 1.1920928955078125e-07);  div_146 = None
	        div_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_73, clamp_min_219);  minimum_73 = None
	        round_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_147);  div_147 = None
	        sub_3357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_147);  round_147 = None
	        clamp_min_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3357, -128);  sub_3357 = None
	        clamp_max_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_220, 127);  clamp_min_220 = None
	        view_1146: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_219, [sym_size_int, 1500, 1])
	        reciprocal_73: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1146);  view_1146 = None
	        mul_7125: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_73, 1.0);  reciprocal_73 = None
	        mul_7128: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11048, mul_7125);  mul_7125 = None
	        round_148: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7128);  mul_7128 = None
	        convert_element_type_438: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_146, torch.int8);  clamp_max_146 = None
	        view_1147: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_438, [sym_size_int, 1500, 1])
	        add_11287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_148, view_1147);  round_148 = view_1147 = None
	        clamp_min_221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11287, -128);  add_11287 = None
	        clamp_max_147: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_221, 127);  clamp_min_221 = None
	        convert_element_type_439: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_147, torch.int8);  clamp_max_147 = None
	        view_1151: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_438, [sym_size_int, 1500, 1]);  convert_element_type_438 = None
	        convert_element_type_440: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_439, torch.float32);  convert_element_type_439 = None
	        convert_element_type_441: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1151, torch.float32);  view_1151 = None
	        sub_3377: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_440, convert_element_type_441);  convert_element_type_440 = convert_element_type_441 = None
	        view_1150: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_219, [sym_size_int, 1500, 1]);  clamp_min_219 = None
	        mul_7150: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3377, view_1150);  sub_3377 = view_1150 = None
	        view_1153: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1155: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_442: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1153, torch.float32);  view_1153 = None
	        convert_element_type_443: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1155, torch.float32);  view_1155 = None
	        sub_3381: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_442, convert_element_type_443);  convert_element_type_442 = convert_element_type_443 = None
	        view_1154: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_7155: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3381, view_1154);  sub_3381 = view_1154 = None
	        view_1156: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7155, [1280, 1280]);  mul_7155 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11048, [2])
	        full_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_74, full_149);  amax_74 = full_149 = None
	        amin_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11048, [2])
	        full_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_74: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_74, full_148);  amin_74 = full_148 = None
	        sub_3395: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_74, minimum_74);  maximum_74 = None
	        div_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3395, 255.0);  sub_3395 = None
	        clamp_min_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_148, 1.1920928955078125e-07);  div_148 = None
	        div_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_74, clamp_min_222);  minimum_74 = None
	        round_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_149);  div_149 = None
	        sub_3401: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_149);  round_149 = None
	        clamp_min_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3401, -128);  sub_3401 = None
	        clamp_max_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_223, 127);  clamp_min_223 = None
	        view_1162: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_222, [sym_size_int, 1500, 1])
	        reciprocal_74: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1162);  view_1162 = None
	        mul_7224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_74, 1.0);  reciprocal_74 = None
	        mul_7227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11048, mul_7224);  add_11048 = mul_7224 = None
	        round_150: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7227);  mul_7227 = None
	        convert_element_type_444: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_148, torch.int8);  clamp_max_148 = None
	        view_1163: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_444, [sym_size_int, 1500, 1])
	        add_11435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_150, view_1163);  round_150 = view_1163 = None
	        clamp_min_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11435, -128);  add_11435 = None
	        clamp_max_149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_224, 127);  clamp_min_224 = None
	        convert_element_type_445: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_149, torch.int8);  clamp_max_149 = None
	        view_1167: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_444, [sym_size_int, 1500, 1]);  convert_element_type_444 = None
	        convert_element_type_446: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_445, torch.float32);  convert_element_type_445 = None
	        convert_element_type_447: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1167, torch.float32);  view_1167 = None
	        sub_3421: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_446, convert_element_type_447);  convert_element_type_446 = convert_element_type_447 = None
	        view_1166: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_222, [sym_size_int, 1500, 1]);  clamp_min_222 = None
	        mul_7249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3421, view_1166);  sub_3421 = view_1166 = None
	        view_1169: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1171: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_448: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1169, torch.float32);  view_1169 = None
	        convert_element_type_449: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1171, torch.float32);  view_1171 = None
	        sub_3425: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_448, convert_element_type_449);  convert_element_type_448 = convert_element_type_449 = None
	        view_1170: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_7254: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3425, view_1170);  sub_3425 = view_1170 = None
	        view_1172: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7254, [1280, 1280]);  mul_7254 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_7064: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1141: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7054, [mul_7064, 1280]);  mul_7054 = mul_7064 = None
	        permute_121: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1140, [1, 0]);  view_1140 = None
	        mm_default_99: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1141, permute_121);  view_1141 = permute_121 = None
	        add_tensor_99: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_99, model_audio_tower_layers_12_self_attn_q_proj_bias);  mm_default_99 = model_audio_tower_layers_12_self_attn_q_proj_bias = None
	        view_1142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_99, [sym_size_int, 1500, 1280]);  add_tensor_99 = None
	        mul_7071: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1142, 0.125);  view_1142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1143: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7071, [sym_size_int, 1500, 20, 64]);  mul_7071 = None
	        permute_122: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1143, [0, 2, 1, 3]);  view_1143 = None
	        clone_98: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_122, memory_format = torch.contiguous_format);  permute_122 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_7158: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1157: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7150, [mul_7158, 1280]);  mul_7150 = mul_7158 = None
	        permute_123: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1156, [1, 0]);  view_1156 = None
	        mm_12: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1157, permute_123);  view_1157 = permute_123 = None
	        view_1158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_12, [sym_size_int, 1500, 1280]);  mm_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1159: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1158, [sym_size_int, -1, 20, 64]);  view_1158 = None
	        permute_124: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1159, [0, 2, 1, 3]);  view_1159 = None
	        clone_99: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_124, memory_format = torch.contiguous_format);  permute_124 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_7259: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1173: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7249, [mul_7259, 1280]);  mul_7249 = mul_7259 = None
	        permute_125: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1172, [1, 0]);  view_1172 = None
	        mm_default_98: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1173, permute_125);  view_1173 = permute_125 = None
	        add_tensor_98: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_98, model_audio_tower_layers_12_self_attn_v_proj_bias);  mm_default_98 = model_audio_tower_layers_12_self_attn_v_proj_bias = None
	        view_1174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_98, [sym_size_int, 1500, 1280]);  add_tensor_98 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1175: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1174, [sym_size_int, -1, 20, 64]);  view_1174 = None
	        permute_126: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1175, [0, 2, 1, 3]);  view_1175 = None
	        clone_100: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_126, memory_format = torch.contiguous_format);  permute_126 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_12 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_98, clone_99, clone_100, None, False, scale = 1.0);  clone_98 = clone_99 = clone_100 = None
	        getitem_98: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_12[0];  _scaled_dot_product_efficient_attention_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_127: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_98, [0, 2, 1, 3]);  getitem_98 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_127, [sym_size_int, 1500, -1]);  permute_127 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1176, [2])
	        full_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_75, full_151);  amax_75 = full_151 = None
	        amin_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1176, [2])
	        full_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_75: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_75, full_150);  amin_75 = full_150 = None
	        sub_3443: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_75, minimum_75);  maximum_75 = None
	        div_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3443, 255.0);  sub_3443 = None
	        clamp_min_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_150, 1.1920928955078125e-07);  div_150 = None
	        div_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_75, clamp_min_225);  minimum_75 = None
	        round_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_151);  div_151 = None
	        sub_3449: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_151);  round_151 = None
	        clamp_min_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3449, -128);  sub_3449 = None
	        clamp_max_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_226, 127);  clamp_min_226 = None
	        view_1179: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_225, [sym_size_int, 1500, 1])
	        reciprocal_75: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1179);  view_1179 = None
	        mul_7329: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_75, 1.0);  reciprocal_75 = None
	        mul_7332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1176, mul_7329);  view_1176 = mul_7329 = None
	        round_152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7332);  mul_7332 = None
	        convert_element_type_450: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_150, torch.int8);  clamp_max_150 = None
	        view_1180: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_450, [sym_size_int, 1500, 1])
	        add_11599: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_152, view_1180);  round_152 = view_1180 = None
	        clamp_min_227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11599, -128);  add_11599 = None
	        clamp_max_151: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_227, 127);  clamp_min_227 = None
	        convert_element_type_451: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_151, torch.int8);  clamp_max_151 = None
	        view_1184: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_450, [sym_size_int, 1500, 1]);  convert_element_type_450 = None
	        convert_element_type_452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_451, torch.float32);  convert_element_type_451 = None
	        convert_element_type_453: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1184, torch.float32);  view_1184 = None
	        sub_3469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_452, convert_element_type_453);  convert_element_type_452 = convert_element_type_453 = None
	        view_1183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_225, [sym_size_int, 1500, 1]);  clamp_min_225 = None
	        mul_7354: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3469, view_1183);  sub_3469 = view_1183 = None
	        view_1186: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1188: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_454: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1186, torch.float32);  view_1186 = None
	        convert_element_type_455: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1188, torch.float32);  view_1188 = None
	        sub_3473: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_454, convert_element_type_455);  convert_element_type_454 = convert_element_type_455 = None
	        view_1187: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_7359: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3473, view_1187);  sub_3473 = view_1187 = None
	        view_1189: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7359, [1280, 1280]);  mul_7359 = None
	        mul_7364: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1190: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7354, [mul_7364, 1280]);  mul_7354 = mul_7364 = None
	        permute_128: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1189, [1, 0]);  view_1189 = None
	        mm_default_97: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1190, permute_128);  view_1190 = permute_128 = None
	        add_tensor_97: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_97, model_audio_tower_layers_12_self_attn_out_proj_bias);  mm_default_97 = model_audio_tower_layers_12_self_attn_out_proj_bias = None
	        view_1191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_97, [sym_size_int, 1500, 1280]);  add_tensor_97 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_11662: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_11042, view_1191);  add_11042 = view_1191 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_102: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_11662, memory_format = torch.contiguous_format)
	        var_mean_25 = torch.ops.aten.var_mean.correction(clone_102, [2], correction = 0, keepdim = True)
	        getitem_102: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_25[0]
	        getitem_103: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_25[1];  var_mean_25 = None
	        sub_3479: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_102, getitem_103);  clone_102 = getitem_103 = None
	        add_11667: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_102, 1e-05);  getitem_102 = None
	        rsqrt_25: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_11667);  add_11667 = None
	        mul_7375: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3479, rsqrt_25);  sub_3479 = rsqrt_25 = None
	        mul_7376: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7375, model_audio_tower_layers_12_final_layer_norm_weight);  mul_7375 = model_audio_tower_layers_12_final_layer_norm_weight = None
	        add_11668: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_7376, model_audio_tower_layers_12_final_layer_norm_bias);  mul_7376 = model_audio_tower_layers_12_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11668, [2])
	        full_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_76, full_153);  amax_76 = full_153 = None
	        amin_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11668, [2])
	        full_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_76: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_76, full_152);  amin_76 = full_152 = None
	        sub_3490: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_76, minimum_76);  maximum_76 = None
	        div_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3490, 255.0);  sub_3490 = None
	        clamp_min_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_152, 1.1920928955078125e-07);  div_152 = None
	        div_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_76, clamp_min_228);  minimum_76 = None
	        round_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_153);  div_153 = None
	        sub_3496: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_153);  round_153 = None
	        clamp_min_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3496, -128);  sub_3496 = None
	        clamp_max_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_229, 127);  clamp_min_229 = None
	        view_1194: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_228, [sym_size_int, 1500, 1])
	        reciprocal_76: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1194);  view_1194 = None
	        mul_7424: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_76, 1.0);  reciprocal_76 = None
	        mul_7427: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11668, mul_7424);  add_11668 = mul_7424 = None
	        round_154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7427);  mul_7427 = None
	        convert_element_type_456: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_152, torch.int8);  clamp_max_152 = None
	        view_1195: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_456, [sym_size_int, 1500, 1])
	        add_11755: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_154, view_1195);  round_154 = view_1195 = None
	        clamp_min_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11755, -128);  add_11755 = None
	        clamp_max_153: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_230, 127);  clamp_min_230 = None
	        convert_element_type_457: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_153, torch.int8);  clamp_max_153 = None
	        view_1199: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_456, [sym_size_int, 1500, 1]);  convert_element_type_456 = None
	        convert_element_type_458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_457, torch.float32);  convert_element_type_457 = None
	        convert_element_type_459: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1199, torch.float32);  view_1199 = None
	        sub_3516: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_458, convert_element_type_459);  convert_element_type_458 = convert_element_type_459 = None
	        view_1198: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_228, [sym_size_int, 1500, 1]);  clamp_min_228 = None
	        mul_7449: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3516, view_1198);  sub_3516 = view_1198 = None
	        view_1201: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_12_fc1_parametrizations_weight_original0 = None
	        view_1203: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_12_fc1_parametrizations_weight_original2 = None
	        convert_element_type_460: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1201, torch.float32);  view_1201 = None
	        convert_element_type_461: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1203, torch.float32);  view_1203 = None
	        sub_3520: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_460, convert_element_type_461);  convert_element_type_460 = convert_element_type_461 = None
	        view_1202: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_12_fc1_parametrizations_weight_original1 = None
	        mul_7454: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3520, view_1202);  sub_3520 = view_1202 = None
	        view_1204: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7454, [5120, 1280]);  mul_7454 = None
	        mul_7459: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1205: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7449, [mul_7459, 1280]);  mul_7449 = mul_7459 = None
	        permute_129: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1204, [1, 0]);  view_1204 = None
	        mm_default_96: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_1205, permute_129);  view_1205 = permute_129 = None
	        add_tensor_96: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_96, model_audio_tower_layers_12_fc1_bias);  mm_default_96 = model_audio_tower_layers_12_fc1_bias = None
	        view_1206: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_96, [sym_size_int, 1500, 5120]);  add_tensor_96 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_7466: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1206, 0.5)
	        mul_7467: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1206, 0.7071067811865476);  view_1206 = None
	        erf_14: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_7467);  mul_7467 = None
	        add_11814: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_14, 1);  erf_14 = None
	        mul_7468: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7466, add_11814);  mul_7466 = add_11814 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_7468, [2])
	        full_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_77, full_155);  amax_77 = full_155 = None
	        amin_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_7468, [2])
	        full_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_77: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_77, full_154);  amin_77 = full_154 = None
	        sub_3533: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_77, minimum_77);  maximum_77 = None
	        div_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3533, 255.0);  sub_3533 = None
	        clamp_min_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_154, 1.1920928955078125e-07);  div_154 = None
	        div_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_77, clamp_min_231);  minimum_77 = None
	        round_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_155);  div_155 = None
	        sub_3539: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_155);  round_155 = None
	        clamp_min_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3539, -128);  sub_3539 = None
	        clamp_max_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_232, 127);  clamp_min_232 = None
	        view_1209: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_231, [sym_size_int, 1500, 1])
	        reciprocal_77: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1209);  view_1209 = None
	        mul_7514: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_77, 1.0);  reciprocal_77 = None
	        mul_7517: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7468, mul_7514);  mul_7468 = mul_7514 = None
	        round_156: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_7517);  mul_7517 = None
	        convert_element_type_462: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_154, torch.int8);  clamp_max_154 = None
	        view_1210: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_462, [sym_size_int, 1500, 1])
	        add_11897: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_156, view_1210);  round_156 = view_1210 = None
	        clamp_min_233: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_11897, -128);  add_11897 = None
	        clamp_max_155: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_233, 127);  clamp_min_233 = None
	        convert_element_type_463: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_155, torch.int8);  clamp_max_155 = None
	        view_1214: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_462, [sym_size_int, 1500, 1]);  convert_element_type_462 = None
	        convert_element_type_464: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_463, torch.float32);  convert_element_type_463 = None
	        convert_element_type_465: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1214, torch.float32);  view_1214 = None
	        sub_3559: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_464, convert_element_type_465);  convert_element_type_464 = convert_element_type_465 = None
	        view_1213: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_231, [sym_size_int, 1500, 1]);  clamp_min_231 = None
	        mul_7539: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3559, view_1213);  sub_3559 = view_1213 = None
	        view_1216: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_12_fc2_parametrizations_weight_original0 = None
	        view_1218: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_12_fc2_parametrizations_weight_original2 = None
	        convert_element_type_466: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1216, torch.float32);  view_1216 = None
	        convert_element_type_467: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1218, torch.float32);  view_1218 = None
	        sub_3563: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_466, convert_element_type_467);  convert_element_type_466 = convert_element_type_467 = None
	        view_1217: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_12_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_12_fc2_parametrizations_weight_original1 = None
	        mul_7544: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3563, view_1217);  sub_3563 = view_1217 = None
	        view_1219: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7544, [1280, 5120]);  mul_7544 = None
	        mul_7549: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1220: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7539, [mul_7549, 5120]);  mul_7539 = mul_7549 = None
	        permute_130: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1219, [1, 0]);  view_1219 = None
	        mm_default_95: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1220, permute_130);  view_1220 = permute_130 = None
	        add_tensor_95: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_95, model_audio_tower_layers_12_fc2_bias);  mm_default_95 = model_audio_tower_layers_12_fc2_bias = None
	        view_1221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_95, [sym_size_int, 1500, 1280]);  add_tensor_95 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_11960: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_11662, view_1221);  add_11662 = view_1221 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_105: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_11960, memory_format = torch.contiguous_format)
	        var_mean_26 = torch.ops.aten.var_mean.correction(clone_105, [2], correction = 0, keepdim = True)
	        getitem_104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_26[0]
	        getitem_105: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_26[1];  var_mean_26 = None
	        sub_3569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_105, getitem_105);  clone_105 = getitem_105 = None
	        add_11965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_104, 1e-05);  getitem_104 = None
	        rsqrt_26: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_11965);  add_11965 = None
	        mul_7560: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3569, rsqrt_26);  sub_3569 = rsqrt_26 = None
	        mul_7561: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7560, model_audio_tower_layers_13_self_attn_layer_norm_weight);  mul_7560 = model_audio_tower_layers_13_self_attn_layer_norm_weight = None
	        add_11966: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_7561, model_audio_tower_layers_13_self_attn_layer_norm_bias);  mul_7561 = model_audio_tower_layers_13_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11966, [2])
	        full_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_78, full_157);  amax_78 = full_157 = None
	        amin_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11966, [2])
	        full_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_78: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_78, full_156);  amin_78 = full_156 = None
	        sub_3580: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_78, minimum_78);  maximum_78 = None
	        div_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3580, 255.0);  sub_3580 = None
	        clamp_min_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_156, 1.1920928955078125e-07);  div_156 = None
	        div_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_78, clamp_min_234);  minimum_78 = None
	        round_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_157);  div_157 = None
	        sub_3586: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_157);  round_157 = None
	        clamp_min_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3586, -128);  sub_3586 = None
	        clamp_max_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_235, 127);  clamp_min_235 = None
	        view_1224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_234, [sym_size_int, 1500, 1])
	        reciprocal_78: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1224);  view_1224 = None
	        mul_7609: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_78, 1.0);  reciprocal_78 = None
	        mul_7612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11966, mul_7609);  mul_7609 = None
	        round_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7612);  mul_7612 = None
	        convert_element_type_468: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_156, torch.int8);  clamp_max_156 = None
	        view_1225: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_468, [sym_size_int, 1500, 1])
	        add_12053: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_158, view_1225);  round_158 = view_1225 = None
	        clamp_min_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12053, -128);  add_12053 = None
	        clamp_max_157: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_236, 127);  clamp_min_236 = None
	        convert_element_type_469: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_157, torch.int8);  clamp_max_157 = None
	        view_1229: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_468, [sym_size_int, 1500, 1]);  convert_element_type_468 = None
	        convert_element_type_470: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_469, torch.float32);  convert_element_type_469 = None
	        convert_element_type_471: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1229, torch.float32);  view_1229 = None
	        sub_3606: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_470, convert_element_type_471);  convert_element_type_470 = convert_element_type_471 = None
	        view_1228: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_234, [sym_size_int, 1500, 1]);  clamp_min_234 = None
	        mul_7634: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3606, view_1228);  sub_3606 = view_1228 = None
	        view_1231: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1233: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_472: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1231, torch.float32);  view_1231 = None
	        convert_element_type_473: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1233, torch.float32);  view_1233 = None
	        sub_3610: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_472, convert_element_type_473);  convert_element_type_472 = convert_element_type_473 = None
	        view_1232: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_7639: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3610, view_1232);  sub_3610 = view_1232 = None
	        view_1234: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7639, [1280, 1280]);  mul_7639 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11966, [2])
	        full_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_79, full_159);  amax_79 = full_159 = None
	        amin_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11966, [2])
	        full_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_79: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_79, full_158);  amin_79 = full_158 = None
	        sub_3625: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_79, minimum_79);  maximum_79 = None
	        div_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3625, 255.0);  sub_3625 = None
	        clamp_min_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_158, 1.1920928955078125e-07);  div_158 = None
	        div_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_79, clamp_min_237);  minimum_79 = None
	        round_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_159);  div_159 = None
	        sub_3631: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_159);  round_159 = None
	        clamp_min_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3631, -128);  sub_3631 = None
	        clamp_max_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_238, 127);  clamp_min_238 = None
	        view_1240: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_237, [sym_size_int, 1500, 1])
	        reciprocal_79: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1240);  view_1240 = None
	        mul_7705: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_79, 1.0);  reciprocal_79 = None
	        mul_7708: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11966, mul_7705);  mul_7705 = None
	        round_160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7708);  mul_7708 = None
	        convert_element_type_474: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_158, torch.int8);  clamp_max_158 = None
	        view_1241: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_474, [sym_size_int, 1500, 1])
	        add_12205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_160, view_1241);  round_160 = view_1241 = None
	        clamp_min_239: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12205, -128);  add_12205 = None
	        clamp_max_159: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_239, 127);  clamp_min_239 = None
	        convert_element_type_475: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_159, torch.int8);  clamp_max_159 = None
	        view_1245: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_474, [sym_size_int, 1500, 1]);  convert_element_type_474 = None
	        convert_element_type_476: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_475, torch.float32);  convert_element_type_475 = None
	        convert_element_type_477: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1245, torch.float32);  view_1245 = None
	        sub_3651: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_476, convert_element_type_477);  convert_element_type_476 = convert_element_type_477 = None
	        view_1244: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_237, [sym_size_int, 1500, 1]);  clamp_min_237 = None
	        mul_7730: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3651, view_1244);  sub_3651 = view_1244 = None
	        view_1247: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1249: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_478: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1247, torch.float32);  view_1247 = None
	        convert_element_type_479: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1249, torch.float32);  view_1249 = None
	        sub_3655: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_478, convert_element_type_479);  convert_element_type_478 = convert_element_type_479 = None
	        view_1248: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_7735: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3655, view_1248);  sub_3655 = view_1248 = None
	        view_1250: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7735, [1280, 1280]);  mul_7735 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_11966, [2])
	        full_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_80, full_161);  amax_80 = full_161 = None
	        amin_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_11966, [2])
	        full_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_80: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_80, full_160);  amin_80 = full_160 = None
	        sub_3669: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_80, minimum_80);  maximum_80 = None
	        div_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3669, 255.0);  sub_3669 = None
	        clamp_min_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_160, 1.1920928955078125e-07);  div_160 = None
	        div_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_80, clamp_min_240);  minimum_80 = None
	        round_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_161);  div_161 = None
	        sub_3675: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_161);  round_161 = None
	        clamp_min_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3675, -128);  sub_3675 = None
	        clamp_max_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_241, 127);  clamp_min_241 = None
	        view_1256: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_240, [sym_size_int, 1500, 1])
	        reciprocal_80: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1256);  view_1256 = None
	        mul_7804: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_80, 1.0);  reciprocal_80 = None
	        mul_7807: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_11966, mul_7804);  add_11966 = mul_7804 = None
	        round_162: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7807);  mul_7807 = None
	        convert_element_type_480: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_160, torch.int8);  clamp_max_160 = None
	        view_1257: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_480, [sym_size_int, 1500, 1])
	        add_12353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_162, view_1257);  round_162 = view_1257 = None
	        clamp_min_242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12353, -128);  add_12353 = None
	        clamp_max_161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_242, 127);  clamp_min_242 = None
	        convert_element_type_481: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_161, torch.int8);  clamp_max_161 = None
	        view_1261: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_480, [sym_size_int, 1500, 1]);  convert_element_type_480 = None
	        convert_element_type_482: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_481, torch.float32);  convert_element_type_481 = None
	        convert_element_type_483: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1261, torch.float32);  view_1261 = None
	        sub_3695: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_482, convert_element_type_483);  convert_element_type_482 = convert_element_type_483 = None
	        view_1260: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_240, [sym_size_int, 1500, 1]);  clamp_min_240 = None
	        mul_7829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3695, view_1260);  sub_3695 = view_1260 = None
	        view_1263: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1265: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_484: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1263, torch.float32);  view_1263 = None
	        convert_element_type_485: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1265, torch.float32);  view_1265 = None
	        sub_3699: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_484, convert_element_type_485);  convert_element_type_484 = convert_element_type_485 = None
	        view_1264: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_7834: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3699, view_1264);  sub_3699 = view_1264 = None
	        view_1266: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7834, [1280, 1280]);  mul_7834 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_7644: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1235: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7634, [mul_7644, 1280]);  mul_7634 = mul_7644 = None
	        permute_131: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1234, [1, 0]);  view_1234 = None
	        mm_default_94: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1235, permute_131);  view_1235 = permute_131 = None
	        add_tensor_94: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_94, model_audio_tower_layers_13_self_attn_q_proj_bias);  mm_default_94 = model_audio_tower_layers_13_self_attn_q_proj_bias = None
	        view_1236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_94, [sym_size_int, 1500, 1280]);  add_tensor_94 = None
	        mul_7651: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1236, 0.125);  view_1236 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1237: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7651, [sym_size_int, 1500, 20, 64]);  mul_7651 = None
	        permute_132: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1237, [0, 2, 1, 3]);  view_1237 = None
	        clone_106: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_132, memory_format = torch.contiguous_format);  permute_132 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_7738: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1251: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7730, [mul_7738, 1280]);  mul_7730 = mul_7738 = None
	        permute_133: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1250, [1, 0]);  view_1250 = None
	        mm_13: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1251, permute_133);  view_1251 = permute_133 = None
	        view_1252: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_13, [sym_size_int, 1500, 1280]);  mm_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1253: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1252, [sym_size_int, -1, 20, 64]);  view_1252 = None
	        permute_134: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1253, [0, 2, 1, 3]);  view_1253 = None
	        clone_107: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_134, memory_format = torch.contiguous_format);  permute_134 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_7839: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1267: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7829, [mul_7839, 1280]);  mul_7829 = mul_7839 = None
	        permute_135: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1266, [1, 0]);  view_1266 = None
	        mm_default_93: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1267, permute_135);  view_1267 = permute_135 = None
	        add_tensor_93: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_93, model_audio_tower_layers_13_self_attn_v_proj_bias);  mm_default_93 = model_audio_tower_layers_13_self_attn_v_proj_bias = None
	        view_1268: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_93, [sym_size_int, 1500, 1280]);  add_tensor_93 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1269: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1268, [sym_size_int, -1, 20, 64]);  view_1268 = None
	        permute_136: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1269, [0, 2, 1, 3]);  view_1269 = None
	        clone_108: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_136, memory_format = torch.contiguous_format);  permute_136 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_13 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_106, clone_107, clone_108, None, False, scale = 1.0);  clone_106 = clone_107 = clone_108 = None
	        getitem_106: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_13[0];  _scaled_dot_product_efficient_attention_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_137: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_106, [0, 2, 1, 3]);  getitem_106 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_137, [sym_size_int, 1500, -1]);  permute_137 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1270, [2])
	        full_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_81, full_163);  amax_81 = full_163 = None
	        amin_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1270, [2])
	        full_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_81: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_81, full_162);  amin_81 = full_162 = None
	        sub_3717: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_81, minimum_81);  maximum_81 = None
	        div_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3717, 255.0);  sub_3717 = None
	        clamp_min_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_162, 1.1920928955078125e-07);  div_162 = None
	        div_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_81, clamp_min_243);  minimum_81 = None
	        round_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_163);  div_163 = None
	        sub_3723: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_163);  round_163 = None
	        clamp_min_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3723, -128);  sub_3723 = None
	        clamp_max_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_244, 127);  clamp_min_244 = None
	        view_1273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_243, [sym_size_int, 1500, 1])
	        reciprocal_81: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1273);  view_1273 = None
	        mul_7909: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_81, 1.0);  reciprocal_81 = None
	        mul_7912: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1270, mul_7909);  view_1270 = mul_7909 = None
	        round_164: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_7912);  mul_7912 = None
	        convert_element_type_486: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_162, torch.int8);  clamp_max_162 = None
	        view_1274: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_486, [sym_size_int, 1500, 1])
	        add_12517: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_164, view_1274);  round_164 = view_1274 = None
	        clamp_min_245: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12517, -128);  add_12517 = None
	        clamp_max_163: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_245, 127);  clamp_min_245 = None
	        convert_element_type_487: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_163, torch.int8);  clamp_max_163 = None
	        view_1278: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_486, [sym_size_int, 1500, 1]);  convert_element_type_486 = None
	        convert_element_type_488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_487, torch.float32);  convert_element_type_487 = None
	        convert_element_type_489: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1278, torch.float32);  view_1278 = None
	        sub_3743: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_488, convert_element_type_489);  convert_element_type_488 = convert_element_type_489 = None
	        view_1277: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_243, [sym_size_int, 1500, 1]);  clamp_min_243 = None
	        mul_7934: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3743, view_1277);  sub_3743 = view_1277 = None
	        view_1280: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1282: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_490: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1280, torch.float32);  view_1280 = None
	        convert_element_type_491: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1282, torch.float32);  view_1282 = None
	        sub_3747: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_490, convert_element_type_491);  convert_element_type_490 = convert_element_type_491 = None
	        view_1281: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_7939: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3747, view_1281);  sub_3747 = view_1281 = None
	        view_1283: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7939, [1280, 1280]);  mul_7939 = None
	        mul_7944: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1284: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_7934, [mul_7944, 1280]);  mul_7934 = mul_7944 = None
	        permute_138: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1283, [1, 0]);  view_1283 = None
	        mm_default_92: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1284, permute_138);  view_1284 = permute_138 = None
	        add_tensor_92: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_92, model_audio_tower_layers_13_self_attn_out_proj_bias);  mm_default_92 = model_audio_tower_layers_13_self_attn_out_proj_bias = None
	        view_1285: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_92, [sym_size_int, 1500, 1280]);  add_tensor_92 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_12580: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_11960, view_1285);  add_11960 = view_1285 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_12580, memory_format = torch.contiguous_format)
	        var_mean_27 = torch.ops.aten.var_mean.correction(clone_110, [2], correction = 0, keepdim = True)
	        getitem_110: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_27[0]
	        getitem_111: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_27[1];  var_mean_27 = None
	        sub_3753: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_110, getitem_111);  clone_110 = getitem_111 = None
	        add_12585: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_110, 1e-05);  getitem_110 = None
	        rsqrt_27: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_12585);  add_12585 = None
	        mul_7955: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3753, rsqrt_27);  sub_3753 = rsqrt_27 = None
	        mul_7956: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_7955, model_audio_tower_layers_13_final_layer_norm_weight);  mul_7955 = model_audio_tower_layers_13_final_layer_norm_weight = None
	        add_12586: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_7956, model_audio_tower_layers_13_final_layer_norm_bias);  mul_7956 = model_audio_tower_layers_13_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_12586, [2])
	        full_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_82, full_165);  amax_82 = full_165 = None
	        amin_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_12586, [2])
	        full_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_82: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_82, full_164);  amin_82 = full_164 = None
	        sub_3764: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_82, minimum_82);  maximum_82 = None
	        div_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3764, 255.0);  sub_3764 = None
	        clamp_min_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_164, 1.1920928955078125e-07);  div_164 = None
	        div_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_82, clamp_min_246);  minimum_82 = None
	        round_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_165);  div_165 = None
	        sub_3770: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_165);  round_165 = None
	        clamp_min_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3770, -128);  sub_3770 = None
	        clamp_max_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_247, 127);  clamp_min_247 = None
	        view_1288: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_246, [sym_size_int, 1500, 1])
	        reciprocal_82: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1288);  view_1288 = None
	        mul_8004: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_82, 1.0);  reciprocal_82 = None
	        mul_8007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_12586, mul_8004);  add_12586 = mul_8004 = None
	        round_166: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8007);  mul_8007 = None
	        convert_element_type_492: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_164, torch.int8);  clamp_max_164 = None
	        view_1289: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_492, [sym_size_int, 1500, 1])
	        add_12673: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_166, view_1289);  round_166 = view_1289 = None
	        clamp_min_248: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12673, -128);  add_12673 = None
	        clamp_max_165: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_248, 127);  clamp_min_248 = None
	        convert_element_type_493: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_165, torch.int8);  clamp_max_165 = None
	        view_1293: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_492, [sym_size_int, 1500, 1]);  convert_element_type_492 = None
	        convert_element_type_494: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_493, torch.float32);  convert_element_type_493 = None
	        convert_element_type_495: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1293, torch.float32);  view_1293 = None
	        sub_3790: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_494, convert_element_type_495);  convert_element_type_494 = convert_element_type_495 = None
	        view_1292: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_246, [sym_size_int, 1500, 1]);  clamp_min_246 = None
	        mul_8029: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3790, view_1292);  sub_3790 = view_1292 = None
	        view_1295: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_13_fc1_parametrizations_weight_original0 = None
	        view_1297: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_13_fc1_parametrizations_weight_original2 = None
	        convert_element_type_496: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1295, torch.float32);  view_1295 = None
	        convert_element_type_497: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1297, torch.float32);  view_1297 = None
	        sub_3794: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_496, convert_element_type_497);  convert_element_type_496 = convert_element_type_497 = None
	        view_1296: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_13_fc1_parametrizations_weight_original1 = None
	        mul_8034: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3794, view_1296);  sub_3794 = view_1296 = None
	        view_1298: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8034, [5120, 1280]);  mul_8034 = None
	        mul_8039: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1299: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8029, [mul_8039, 1280]);  mul_8029 = mul_8039 = None
	        permute_139: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1298, [1, 0]);  view_1298 = None
	        mm_default_91: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_1299, permute_139);  view_1299 = permute_139 = None
	        add_tensor_91: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_91, model_audio_tower_layers_13_fc1_bias);  mm_default_91 = model_audio_tower_layers_13_fc1_bias = None
	        view_1300: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_91, [sym_size_int, 1500, 5120]);  add_tensor_91 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_8046: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1300, 0.5)
	        mul_8047: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1300, 0.7071067811865476);  view_1300 = None
	        erf_15: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_8047);  mul_8047 = None
	        add_12732: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_15, 1);  erf_15 = None
	        mul_8048: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8046, add_12732);  mul_8046 = add_12732 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_8048, [2])
	        full_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_83, full_167);  amax_83 = full_167 = None
	        amin_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_8048, [2])
	        full_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_83: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_83, full_166);  amin_83 = full_166 = None
	        sub_3807: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_83, minimum_83);  maximum_83 = None
	        div_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3807, 255.0);  sub_3807 = None
	        clamp_min_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_166, 1.1920928955078125e-07);  div_166 = None
	        div_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_83, clamp_min_249);  minimum_83 = None
	        round_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_167);  div_167 = None
	        sub_3813: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_167);  round_167 = None
	        clamp_min_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3813, -128);  sub_3813 = None
	        clamp_max_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_250, 127);  clamp_min_250 = None
	        view_1303: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_249, [sym_size_int, 1500, 1])
	        reciprocal_83: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1303);  view_1303 = None
	        mul_8094: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_83, 1.0);  reciprocal_83 = None
	        mul_8097: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8048, mul_8094);  mul_8048 = mul_8094 = None
	        round_168: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_8097);  mul_8097 = None
	        convert_element_type_498: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_166, torch.int8);  clamp_max_166 = None
	        view_1304: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_498, [sym_size_int, 1500, 1])
	        add_12815: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_168, view_1304);  round_168 = view_1304 = None
	        clamp_min_251: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12815, -128);  add_12815 = None
	        clamp_max_167: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_251, 127);  clamp_min_251 = None
	        convert_element_type_499: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_167, torch.int8);  clamp_max_167 = None
	        view_1308: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_498, [sym_size_int, 1500, 1]);  convert_element_type_498 = None
	        convert_element_type_500: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_499, torch.float32);  convert_element_type_499 = None
	        convert_element_type_501: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1308, torch.float32);  view_1308 = None
	        sub_3833: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_500, convert_element_type_501);  convert_element_type_500 = convert_element_type_501 = None
	        view_1307: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_249, [sym_size_int, 1500, 1]);  clamp_min_249 = None
	        mul_8119: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3833, view_1307);  sub_3833 = view_1307 = None
	        view_1310: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_13_fc2_parametrizations_weight_original0 = None
	        view_1312: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_13_fc2_parametrizations_weight_original2 = None
	        convert_element_type_502: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1310, torch.float32);  view_1310 = None
	        convert_element_type_503: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1312, torch.float32);  view_1312 = None
	        sub_3837: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_502, convert_element_type_503);  convert_element_type_502 = convert_element_type_503 = None
	        view_1311: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_13_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_13_fc2_parametrizations_weight_original1 = None
	        mul_8124: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3837, view_1311);  sub_3837 = view_1311 = None
	        view_1313: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8124, [1280, 5120]);  mul_8124 = None
	        mul_8129: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1314: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8119, [mul_8129, 5120]);  mul_8119 = mul_8129 = None
	        permute_140: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1313, [1, 0]);  view_1313 = None
	        mm_default_90: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1314, permute_140);  view_1314 = permute_140 = None
	        add_tensor_90: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_90, model_audio_tower_layers_13_fc2_bias);  mm_default_90 = model_audio_tower_layers_13_fc2_bias = None
	        view_1315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_90, [sym_size_int, 1500, 1280]);  add_tensor_90 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_12878: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_12580, view_1315);  add_12580 = view_1315 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_12878, memory_format = torch.contiguous_format)
	        var_mean_28 = torch.ops.aten.var_mean.correction(clone_113, [2], correction = 0, keepdim = True)
	        getitem_112: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_28[0]
	        getitem_113: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_28[1];  var_mean_28 = None
	        sub_3843: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_113, getitem_113);  clone_113 = getitem_113 = None
	        add_12883: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_112, 1e-05);  getitem_112 = None
	        rsqrt_28: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_12883);  add_12883 = None
	        mul_8140: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3843, rsqrt_28);  sub_3843 = rsqrt_28 = None
	        mul_8141: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8140, model_audio_tower_layers_14_self_attn_layer_norm_weight);  mul_8140 = model_audio_tower_layers_14_self_attn_layer_norm_weight = None
	        add_12884: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_8141, model_audio_tower_layers_14_self_attn_layer_norm_bias);  mul_8141 = model_audio_tower_layers_14_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_12884, [2])
	        full_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_84, full_169);  amax_84 = full_169 = None
	        amin_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_12884, [2])
	        full_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_84: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_84, full_168);  amin_84 = full_168 = None
	        sub_3854: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_84, minimum_84);  maximum_84 = None
	        div_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3854, 255.0);  sub_3854 = None
	        clamp_min_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_168, 1.1920928955078125e-07);  div_168 = None
	        div_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_84, clamp_min_252);  minimum_84 = None
	        round_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_169);  div_169 = None
	        sub_3860: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_169);  round_169 = None
	        clamp_min_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3860, -128);  sub_3860 = None
	        clamp_max_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_253, 127);  clamp_min_253 = None
	        view_1318: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_252, [sym_size_int, 1500, 1])
	        reciprocal_84: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1318);  view_1318 = None
	        mul_8189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_84, 1.0);  reciprocal_84 = None
	        mul_8192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_12884, mul_8189);  mul_8189 = None
	        round_170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8192);  mul_8192 = None
	        convert_element_type_504: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_168, torch.int8);  clamp_max_168 = None
	        view_1319: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_504, [sym_size_int, 1500, 1])
	        add_12971: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_170, view_1319);  round_170 = view_1319 = None
	        clamp_min_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_12971, -128);  add_12971 = None
	        clamp_max_169: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_254, 127);  clamp_min_254 = None
	        convert_element_type_505: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_169, torch.int8);  clamp_max_169 = None
	        view_1323: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_504, [sym_size_int, 1500, 1]);  convert_element_type_504 = None
	        convert_element_type_506: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_505, torch.float32);  convert_element_type_505 = None
	        convert_element_type_507: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1323, torch.float32);  view_1323 = None
	        sub_3880: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_506, convert_element_type_507);  convert_element_type_506 = convert_element_type_507 = None
	        view_1322: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_252, [sym_size_int, 1500, 1]);  clamp_min_252 = None
	        mul_8214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3880, view_1322);  sub_3880 = view_1322 = None
	        view_1325: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1327: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_508: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1325, torch.float32);  view_1325 = None
	        convert_element_type_509: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1327, torch.float32);  view_1327 = None
	        sub_3884: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_508, convert_element_type_509);  convert_element_type_508 = convert_element_type_509 = None
	        view_1326: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_8219: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3884, view_1326);  sub_3884 = view_1326 = None
	        view_1328: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8219, [1280, 1280]);  mul_8219 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_12884, [2])
	        full_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_85, full_171);  amax_85 = full_171 = None
	        amin_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_12884, [2])
	        full_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_85: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_85, full_170);  amin_85 = full_170 = None
	        sub_3899: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_85, minimum_85);  maximum_85 = None
	        div_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3899, 255.0);  sub_3899 = None
	        clamp_min_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_170, 1.1920928955078125e-07);  div_170 = None
	        div_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_85, clamp_min_255);  minimum_85 = None
	        round_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_171);  div_171 = None
	        sub_3905: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_171);  round_171 = None
	        clamp_min_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3905, -128);  sub_3905 = None
	        clamp_max_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_256, 127);  clamp_min_256 = None
	        view_1334: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_255, [sym_size_int, 1500, 1])
	        reciprocal_85: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1334);  view_1334 = None
	        mul_8285: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_85, 1.0);  reciprocal_85 = None
	        mul_8288: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_12884, mul_8285);  mul_8285 = None
	        round_172: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8288);  mul_8288 = None
	        convert_element_type_510: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_170, torch.int8);  clamp_max_170 = None
	        view_1335: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_510, [sym_size_int, 1500, 1])
	        add_13123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_172, view_1335);  round_172 = view_1335 = None
	        clamp_min_257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13123, -128);  add_13123 = None
	        clamp_max_171: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_257, 127);  clamp_min_257 = None
	        convert_element_type_511: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_171, torch.int8);  clamp_max_171 = None
	        view_1339: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_510, [sym_size_int, 1500, 1]);  convert_element_type_510 = None
	        convert_element_type_512: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_511, torch.float32);  convert_element_type_511 = None
	        convert_element_type_513: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1339, torch.float32);  view_1339 = None
	        sub_3925: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_512, convert_element_type_513);  convert_element_type_512 = convert_element_type_513 = None
	        view_1338: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_255, [sym_size_int, 1500, 1]);  clamp_min_255 = None
	        mul_8310: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3925, view_1338);  sub_3925 = view_1338 = None
	        view_1341: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1343: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_514: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1341, torch.float32);  view_1341 = None
	        convert_element_type_515: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1343, torch.float32);  view_1343 = None
	        sub_3929: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_514, convert_element_type_515);  convert_element_type_514 = convert_element_type_515 = None
	        view_1342: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_8315: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3929, view_1342);  sub_3929 = view_1342 = None
	        view_1344: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8315, [1280, 1280]);  mul_8315 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_12884, [2])
	        full_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_86, full_173);  amax_86 = full_173 = None
	        amin_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_12884, [2])
	        full_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_86: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_86, full_172);  amin_86 = full_172 = None
	        sub_3943: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_86, minimum_86);  maximum_86 = None
	        div_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3943, 255.0);  sub_3943 = None
	        clamp_min_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_172, 1.1920928955078125e-07);  div_172 = None
	        div_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_86, clamp_min_258);  minimum_86 = None
	        round_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_173);  div_173 = None
	        sub_3949: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_173);  round_173 = None
	        clamp_min_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3949, -128);  sub_3949 = None
	        clamp_max_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_259, 127);  clamp_min_259 = None
	        view_1350: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_258, [sym_size_int, 1500, 1])
	        reciprocal_86: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1350);  view_1350 = None
	        mul_8384: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_86, 1.0);  reciprocal_86 = None
	        mul_8387: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_12884, mul_8384);  add_12884 = mul_8384 = None
	        round_174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8387);  mul_8387 = None
	        convert_element_type_516: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_172, torch.int8);  clamp_max_172 = None
	        view_1351: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_516, [sym_size_int, 1500, 1])
	        add_13271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_174, view_1351);  round_174 = view_1351 = None
	        clamp_min_260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13271, -128);  add_13271 = None
	        clamp_max_173: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_260, 127);  clamp_min_260 = None
	        convert_element_type_517: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_173, torch.int8);  clamp_max_173 = None
	        view_1355: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_516, [sym_size_int, 1500, 1]);  convert_element_type_516 = None
	        convert_element_type_518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_517, torch.float32);  convert_element_type_517 = None
	        convert_element_type_519: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1355, torch.float32);  view_1355 = None
	        sub_3969: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_518, convert_element_type_519);  convert_element_type_518 = convert_element_type_519 = None
	        view_1354: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_258, [sym_size_int, 1500, 1]);  clamp_min_258 = None
	        mul_8409: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3969, view_1354);  sub_3969 = view_1354 = None
	        view_1357: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1359: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_520: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1357, torch.float32);  view_1357 = None
	        convert_element_type_521: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1359, torch.float32);  view_1359 = None
	        sub_3973: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_520, convert_element_type_521);  convert_element_type_520 = convert_element_type_521 = None
	        view_1358: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_8414: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_3973, view_1358);  sub_3973 = view_1358 = None
	        view_1360: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8414, [1280, 1280]);  mul_8414 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_8224: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1329: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8214, [mul_8224, 1280]);  mul_8214 = mul_8224 = None
	        permute_141: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1328, [1, 0]);  view_1328 = None
	        mm_default_89: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1329, permute_141);  view_1329 = permute_141 = None
	        add_tensor_89: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_89, model_audio_tower_layers_14_self_attn_q_proj_bias);  mm_default_89 = model_audio_tower_layers_14_self_attn_q_proj_bias = None
	        view_1330: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_89, [sym_size_int, 1500, 1280]);  add_tensor_89 = None
	        mul_8231: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1330, 0.125);  view_1330 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1331: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8231, [sym_size_int, 1500, 20, 64]);  mul_8231 = None
	        permute_142: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1331, [0, 2, 1, 3]);  view_1331 = None
	        clone_114: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_142, memory_format = torch.contiguous_format);  permute_142 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_8318: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1345: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8310, [mul_8318, 1280]);  mul_8310 = mul_8318 = None
	        permute_143: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1344, [1, 0]);  view_1344 = None
	        mm_14: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1345, permute_143);  view_1345 = permute_143 = None
	        view_1346: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_14, [sym_size_int, 1500, 1280]);  mm_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1347: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1346, [sym_size_int, -1, 20, 64]);  view_1346 = None
	        permute_144: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1347, [0, 2, 1, 3]);  view_1347 = None
	        clone_115: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_144, memory_format = torch.contiguous_format);  permute_144 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_8419: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1361: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8409, [mul_8419, 1280]);  mul_8409 = mul_8419 = None
	        permute_145: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1360, [1, 0]);  view_1360 = None
	        mm_default_88: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1361, permute_145);  view_1361 = permute_145 = None
	        add_tensor_88: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_88, model_audio_tower_layers_14_self_attn_v_proj_bias);  mm_default_88 = model_audio_tower_layers_14_self_attn_v_proj_bias = None
	        view_1362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_88, [sym_size_int, 1500, 1280]);  add_tensor_88 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1363: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1362, [sym_size_int, -1, 20, 64]);  view_1362 = None
	        permute_146: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1363, [0, 2, 1, 3]);  view_1363 = None
	        clone_116: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_146, memory_format = torch.contiguous_format);  permute_146 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_14 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_114, clone_115, clone_116, None, False, scale = 1.0);  clone_114 = clone_115 = clone_116 = None
	        getitem_114: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_14[0];  _scaled_dot_product_efficient_attention_14 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_147: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_114, [0, 2, 1, 3]);  getitem_114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1364: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_147, [sym_size_int, 1500, -1]);  permute_147 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1364, [2])
	        full_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_87, full_175);  amax_87 = full_175 = None
	        amin_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1364, [2])
	        full_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_87: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_87, full_174);  amin_87 = full_174 = None
	        sub_3991: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_87, minimum_87);  maximum_87 = None
	        div_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_3991, 255.0);  sub_3991 = None
	        clamp_min_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_174, 1.1920928955078125e-07);  div_174 = None
	        div_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_87, clamp_min_261);  minimum_87 = None
	        round_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_175);  div_175 = None
	        sub_3997: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_175);  round_175 = None
	        clamp_min_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_3997, -128);  sub_3997 = None
	        clamp_max_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_262, 127);  clamp_min_262 = None
	        view_1367: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_261, [sym_size_int, 1500, 1])
	        reciprocal_87: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1367);  view_1367 = None
	        mul_8489: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_87, 1.0);  reciprocal_87 = None
	        mul_8492: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1364, mul_8489);  view_1364 = mul_8489 = None
	        round_176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8492);  mul_8492 = None
	        convert_element_type_522: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_174, torch.int8);  clamp_max_174 = None
	        view_1368: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_522, [sym_size_int, 1500, 1])
	        add_13435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_176, view_1368);  round_176 = view_1368 = None
	        clamp_min_263: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13435, -128);  add_13435 = None
	        clamp_max_175: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_263, 127);  clamp_min_263 = None
	        convert_element_type_523: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_175, torch.int8);  clamp_max_175 = None
	        view_1372: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_522, [sym_size_int, 1500, 1]);  convert_element_type_522 = None
	        convert_element_type_524: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_523, torch.float32);  convert_element_type_523 = None
	        convert_element_type_525: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1372, torch.float32);  view_1372 = None
	        sub_4017: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_524, convert_element_type_525);  convert_element_type_524 = convert_element_type_525 = None
	        view_1371: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_261, [sym_size_int, 1500, 1]);  clamp_min_261 = None
	        mul_8514: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4017, view_1371);  sub_4017 = view_1371 = None
	        view_1374: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1376: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_526: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1374, torch.float32);  view_1374 = None
	        convert_element_type_527: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1376, torch.float32);  view_1376 = None
	        sub_4021: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_526, convert_element_type_527);  convert_element_type_526 = convert_element_type_527 = None
	        view_1375: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_8519: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4021, view_1375);  sub_4021 = view_1375 = None
	        view_1377: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8519, [1280, 1280]);  mul_8519 = None
	        mul_8524: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1378: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8514, [mul_8524, 1280]);  mul_8514 = mul_8524 = None
	        permute_148: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1377, [1, 0]);  view_1377 = None
	        mm_default_87: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1378, permute_148);  view_1378 = permute_148 = None
	        add_tensor_87: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_87, model_audio_tower_layers_14_self_attn_out_proj_bias);  mm_default_87 = model_audio_tower_layers_14_self_attn_out_proj_bias = None
	        view_1379: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_87, [sym_size_int, 1500, 1280]);  add_tensor_87 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_13498: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_12878, view_1379);  add_12878 = view_1379 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_118: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_13498, memory_format = torch.contiguous_format)
	        var_mean_29 = torch.ops.aten.var_mean.correction(clone_118, [2], correction = 0, keepdim = True)
	        getitem_118: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_29[0]
	        getitem_119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_29[1];  var_mean_29 = None
	        sub_4027: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_118, getitem_119);  clone_118 = getitem_119 = None
	        add_13503: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_118, 1e-05);  getitem_118 = None
	        rsqrt_29: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_13503);  add_13503 = None
	        mul_8535: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4027, rsqrt_29);  sub_4027 = rsqrt_29 = None
	        mul_8536: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8535, model_audio_tower_layers_14_final_layer_norm_weight);  mul_8535 = model_audio_tower_layers_14_final_layer_norm_weight = None
	        add_13504: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_8536, model_audio_tower_layers_14_final_layer_norm_bias);  mul_8536 = model_audio_tower_layers_14_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_13504, [2])
	        full_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_88, full_177);  amax_88 = full_177 = None
	        amin_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_13504, [2])
	        full_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_88: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_88, full_176);  amin_88 = full_176 = None
	        sub_4038: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_88, minimum_88);  maximum_88 = None
	        div_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4038, 255.0);  sub_4038 = None
	        clamp_min_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_176, 1.1920928955078125e-07);  div_176 = None
	        div_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_88, clamp_min_264);  minimum_88 = None
	        round_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_177);  div_177 = None
	        sub_4044: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_177);  round_177 = None
	        clamp_min_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4044, -128);  sub_4044 = None
	        clamp_max_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_265, 127);  clamp_min_265 = None
	        view_1382: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_264, [sym_size_int, 1500, 1])
	        reciprocal_88: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1382);  view_1382 = None
	        mul_8584: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_88, 1.0);  reciprocal_88 = None
	        mul_8587: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_13504, mul_8584);  add_13504 = mul_8584 = None
	        round_178: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8587);  mul_8587 = None
	        convert_element_type_528: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_176, torch.int8);  clamp_max_176 = None
	        view_1383: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_528, [sym_size_int, 1500, 1])
	        add_13591: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_178, view_1383);  round_178 = view_1383 = None
	        clamp_min_266: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13591, -128);  add_13591 = None
	        clamp_max_177: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_266, 127);  clamp_min_266 = None
	        convert_element_type_529: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_177, torch.int8);  clamp_max_177 = None
	        view_1387: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_528, [sym_size_int, 1500, 1]);  convert_element_type_528 = None
	        convert_element_type_530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_529, torch.float32);  convert_element_type_529 = None
	        convert_element_type_531: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1387, torch.float32);  view_1387 = None
	        sub_4064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_530, convert_element_type_531);  convert_element_type_530 = convert_element_type_531 = None
	        view_1386: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_264, [sym_size_int, 1500, 1]);  clamp_min_264 = None
	        mul_8609: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4064, view_1386);  sub_4064 = view_1386 = None
	        view_1389: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_14_fc1_parametrizations_weight_original0 = None
	        view_1391: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_14_fc1_parametrizations_weight_original2 = None
	        convert_element_type_532: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1389, torch.float32);  view_1389 = None
	        convert_element_type_533: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1391, torch.float32);  view_1391 = None
	        sub_4068: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_532, convert_element_type_533);  convert_element_type_532 = convert_element_type_533 = None
	        view_1390: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_14_fc1_parametrizations_weight_original1 = None
	        mul_8614: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4068, view_1390);  sub_4068 = view_1390 = None
	        view_1392: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8614, [5120, 1280]);  mul_8614 = None
	        mul_8619: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1393: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8609, [mul_8619, 1280]);  mul_8609 = mul_8619 = None
	        permute_149: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1392, [1, 0]);  view_1392 = None
	        mm_default_86: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_1393, permute_149);  view_1393 = permute_149 = None
	        add_tensor_86: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_86, model_audio_tower_layers_14_fc1_bias);  mm_default_86 = model_audio_tower_layers_14_fc1_bias = None
	        view_1394: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_86, [sym_size_int, 1500, 5120]);  add_tensor_86 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_8626: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1394, 0.5)
	        mul_8627: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1394, 0.7071067811865476);  view_1394 = None
	        erf_16: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_8627);  mul_8627 = None
	        add_13650: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_16, 1);  erf_16 = None
	        mul_8628: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8626, add_13650);  mul_8626 = add_13650 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_8628, [2])
	        full_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_89, full_179);  amax_89 = full_179 = None
	        amin_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_8628, [2])
	        full_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_89: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_89, full_178);  amin_89 = full_178 = None
	        sub_4081: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_89, minimum_89);  maximum_89 = None
	        div_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4081, 255.0);  sub_4081 = None
	        clamp_min_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_178, 1.1920928955078125e-07);  div_178 = None
	        div_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_89, clamp_min_267);  minimum_89 = None
	        round_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_179);  div_179 = None
	        sub_4087: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_179);  round_179 = None
	        clamp_min_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4087, -128);  sub_4087 = None
	        clamp_max_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_268, 127);  clamp_min_268 = None
	        view_1397: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_267, [sym_size_int, 1500, 1])
	        reciprocal_89: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1397);  view_1397 = None
	        mul_8674: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_89, 1.0);  reciprocal_89 = None
	        mul_8677: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8628, mul_8674);  mul_8628 = mul_8674 = None
	        round_180: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_8677);  mul_8677 = None
	        convert_element_type_534: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_178, torch.int8);  clamp_max_178 = None
	        view_1398: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_534, [sym_size_int, 1500, 1])
	        add_13733: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_180, view_1398);  round_180 = view_1398 = None
	        clamp_min_269: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13733, -128);  add_13733 = None
	        clamp_max_179: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_269, 127);  clamp_min_269 = None
	        convert_element_type_535: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_179, torch.int8);  clamp_max_179 = None
	        view_1402: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_534, [sym_size_int, 1500, 1]);  convert_element_type_534 = None
	        convert_element_type_536: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_535, torch.float32);  convert_element_type_535 = None
	        convert_element_type_537: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1402, torch.float32);  view_1402 = None
	        sub_4107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_536, convert_element_type_537);  convert_element_type_536 = convert_element_type_537 = None
	        view_1401: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_267, [sym_size_int, 1500, 1]);  clamp_min_267 = None
	        mul_8699: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4107, view_1401);  sub_4107 = view_1401 = None
	        view_1404: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_14_fc2_parametrizations_weight_original0 = None
	        view_1406: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_14_fc2_parametrizations_weight_original2 = None
	        convert_element_type_538: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1404, torch.float32);  view_1404 = None
	        convert_element_type_539: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1406, torch.float32);  view_1406 = None
	        sub_4111: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_538, convert_element_type_539);  convert_element_type_538 = convert_element_type_539 = None
	        view_1405: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_14_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_14_fc2_parametrizations_weight_original1 = None
	        mul_8704: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4111, view_1405);  sub_4111 = view_1405 = None
	        view_1407: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8704, [1280, 5120]);  mul_8704 = None
	        mul_8709: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1408: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8699, [mul_8709, 5120]);  mul_8699 = mul_8709 = None
	        permute_150: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1407, [1, 0]);  view_1407 = None
	        mm_default_85: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1408, permute_150);  view_1408 = permute_150 = None
	        add_tensor_85: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_85, model_audio_tower_layers_14_fc2_bias);  mm_default_85 = model_audio_tower_layers_14_fc2_bias = None
	        view_1409: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_85, [sym_size_int, 1500, 1280]);  add_tensor_85 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_13796: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_13498, view_1409);  add_13498 = view_1409 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_121: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_13796, memory_format = torch.contiguous_format)
	        var_mean_30 = torch.ops.aten.var_mean.correction(clone_121, [2], correction = 0, keepdim = True)
	        getitem_120: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_30[0]
	        getitem_121: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_30[1];  var_mean_30 = None
	        sub_4117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_121, getitem_121);  clone_121 = getitem_121 = None
	        add_13801: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_120, 1e-05);  getitem_120 = None
	        rsqrt_30: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_13801);  add_13801 = None
	        mul_8720: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4117, rsqrt_30);  sub_4117 = rsqrt_30 = None
	        mul_8721: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_8720, model_audio_tower_layers_15_self_attn_layer_norm_weight);  mul_8720 = model_audio_tower_layers_15_self_attn_layer_norm_weight = None
	        add_13802: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_8721, model_audio_tower_layers_15_self_attn_layer_norm_bias);  mul_8721 = model_audio_tower_layers_15_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_13802, [2])
	        full_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_90, full_181);  amax_90 = full_181 = None
	        amin_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_13802, [2])
	        full_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_90: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_90, full_180);  amin_90 = full_180 = None
	        sub_4128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_90, minimum_90);  maximum_90 = None
	        div_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4128, 255.0);  sub_4128 = None
	        clamp_min_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_180, 1.1920928955078125e-07);  div_180 = None
	        div_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_90, clamp_min_270);  minimum_90 = None
	        round_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_181);  div_181 = None
	        sub_4134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_181);  round_181 = None
	        clamp_min_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4134, -128);  sub_4134 = None
	        clamp_max_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_271, 127);  clamp_min_271 = None
	        view_1412: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_270, [sym_size_int, 1500, 1])
	        reciprocal_90: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1412);  view_1412 = None
	        mul_8769: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_90, 1.0);  reciprocal_90 = None
	        mul_8772: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_13802, mul_8769);  mul_8769 = None
	        round_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8772);  mul_8772 = None
	        convert_element_type_540: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_180, torch.int8);  clamp_max_180 = None
	        view_1413: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_540, [sym_size_int, 1500, 1])
	        add_13889: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_182, view_1413);  round_182 = view_1413 = None
	        clamp_min_272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_13889, -128);  add_13889 = None
	        clamp_max_181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_272, 127);  clamp_min_272 = None
	        convert_element_type_541: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_181, torch.int8);  clamp_max_181 = None
	        view_1417: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_540, [sym_size_int, 1500, 1]);  convert_element_type_540 = None
	        convert_element_type_542: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_541, torch.float32);  convert_element_type_541 = None
	        convert_element_type_543: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1417, torch.float32);  view_1417 = None
	        sub_4154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_542, convert_element_type_543);  convert_element_type_542 = convert_element_type_543 = None
	        view_1416: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_270, [sym_size_int, 1500, 1]);  clamp_min_270 = None
	        mul_8794: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4154, view_1416);  sub_4154 = view_1416 = None
	        view_1419: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1421: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_544: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1419, torch.float32);  view_1419 = None
	        convert_element_type_545: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1421, torch.float32);  view_1421 = None
	        sub_4158: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_544, convert_element_type_545);  convert_element_type_544 = convert_element_type_545 = None
	        view_1420: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_8799: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4158, view_1420);  sub_4158 = view_1420 = None
	        view_1422: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8799, [1280, 1280]);  mul_8799 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_13802, [2])
	        full_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_91, full_183);  amax_91 = full_183 = None
	        amin_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_13802, [2])
	        full_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_91: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_91, full_182);  amin_91 = full_182 = None
	        sub_4173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_91, minimum_91);  maximum_91 = None
	        div_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4173, 255.0);  sub_4173 = None
	        clamp_min_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_182, 1.1920928955078125e-07);  div_182 = None
	        div_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_91, clamp_min_273);  minimum_91 = None
	        round_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_183);  div_183 = None
	        sub_4179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_183);  round_183 = None
	        clamp_min_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4179, -128);  sub_4179 = None
	        clamp_max_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_274, 127);  clamp_min_274 = None
	        view_1428: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_273, [sym_size_int, 1500, 1])
	        reciprocal_91: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1428);  view_1428 = None
	        mul_8865: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_91, 1.0);  reciprocal_91 = None
	        mul_8868: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_13802, mul_8865);  mul_8865 = None
	        round_184: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8868);  mul_8868 = None
	        convert_element_type_546: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_182, torch.int8);  clamp_max_182 = None
	        view_1429: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_546, [sym_size_int, 1500, 1])
	        add_14041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_184, view_1429);  round_184 = view_1429 = None
	        clamp_min_275: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14041, -128);  add_14041 = None
	        clamp_max_183: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_275, 127);  clamp_min_275 = None
	        convert_element_type_547: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_183, torch.int8);  clamp_max_183 = None
	        view_1433: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_546, [sym_size_int, 1500, 1]);  convert_element_type_546 = None
	        convert_element_type_548: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_547, torch.float32);  convert_element_type_547 = None
	        convert_element_type_549: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1433, torch.float32);  view_1433 = None
	        sub_4199: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_548, convert_element_type_549);  convert_element_type_548 = convert_element_type_549 = None
	        view_1432: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_273, [sym_size_int, 1500, 1]);  clamp_min_273 = None
	        mul_8890: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4199, view_1432);  sub_4199 = view_1432 = None
	        view_1435: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1437: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_550: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1435, torch.float32);  view_1435 = None
	        convert_element_type_551: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1437, torch.float32);  view_1437 = None
	        sub_4203: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_550, convert_element_type_551);  convert_element_type_550 = convert_element_type_551 = None
	        view_1436: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_8895: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4203, view_1436);  sub_4203 = view_1436 = None
	        view_1438: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8895, [1280, 1280]);  mul_8895 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_13802, [2])
	        full_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_92, full_185);  amax_92 = full_185 = None
	        amin_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_13802, [2])
	        full_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_92: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_92, full_184);  amin_92 = full_184 = None
	        sub_4217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_92, minimum_92);  maximum_92 = None
	        div_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4217, 255.0);  sub_4217 = None
	        clamp_min_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_184, 1.1920928955078125e-07);  div_184 = None
	        div_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_92, clamp_min_276);  minimum_92 = None
	        round_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_185);  div_185 = None
	        sub_4223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_185);  round_185 = None
	        clamp_min_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4223, -128);  sub_4223 = None
	        clamp_max_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_277, 127);  clamp_min_277 = None
	        view_1444: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_276, [sym_size_int, 1500, 1])
	        reciprocal_92: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1444);  view_1444 = None
	        mul_8964: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_92, 1.0);  reciprocal_92 = None
	        mul_8967: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_13802, mul_8964);  add_13802 = mul_8964 = None
	        round_186: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_8967);  mul_8967 = None
	        convert_element_type_552: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_184, torch.int8);  clamp_max_184 = None
	        view_1445: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_552, [sym_size_int, 1500, 1])
	        add_14189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_186, view_1445);  round_186 = view_1445 = None
	        clamp_min_278: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14189, -128);  add_14189 = None
	        clamp_max_185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_278, 127);  clamp_min_278 = None
	        convert_element_type_553: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_185, torch.int8);  clamp_max_185 = None
	        view_1449: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_552, [sym_size_int, 1500, 1]);  convert_element_type_552 = None
	        convert_element_type_554: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_553, torch.float32);  convert_element_type_553 = None
	        convert_element_type_555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1449, torch.float32);  view_1449 = None
	        sub_4243: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_554, convert_element_type_555);  convert_element_type_554 = convert_element_type_555 = None
	        view_1448: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_276, [sym_size_int, 1500, 1]);  clamp_min_276 = None
	        mul_8989: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4243, view_1448);  sub_4243 = view_1448 = None
	        view_1451: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1453: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_556: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1451, torch.float32);  view_1451 = None
	        convert_element_type_557: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1453, torch.float32);  view_1453 = None
	        sub_4247: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_556, convert_element_type_557);  convert_element_type_556 = convert_element_type_557 = None
	        view_1452: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_8994: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4247, view_1452);  sub_4247 = view_1452 = None
	        view_1454: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8994, [1280, 1280]);  mul_8994 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_8804: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1423: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8794, [mul_8804, 1280]);  mul_8794 = mul_8804 = None
	        permute_151: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1422, [1, 0]);  view_1422 = None
	        mm_default_84: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1423, permute_151);  view_1423 = permute_151 = None
	        add_tensor_84: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_84, model_audio_tower_layers_15_self_attn_q_proj_bias);  mm_default_84 = model_audio_tower_layers_15_self_attn_q_proj_bias = None
	        view_1424: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_84, [sym_size_int, 1500, 1280]);  add_tensor_84 = None
	        mul_8811: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1424, 0.125);  view_1424 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1425: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8811, [sym_size_int, 1500, 20, 64]);  mul_8811 = None
	        permute_152: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1425, [0, 2, 1, 3]);  view_1425 = None
	        clone_122: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_152, memory_format = torch.contiguous_format);  permute_152 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_8898: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1439: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8890, [mul_8898, 1280]);  mul_8890 = mul_8898 = None
	        permute_153: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1438, [1, 0]);  view_1438 = None
	        mm_15: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1439, permute_153);  view_1439 = permute_153 = None
	        view_1440: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_15, [sym_size_int, 1500, 1280]);  mm_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1441: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1440, [sym_size_int, -1, 20, 64]);  view_1440 = None
	        permute_154: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1441, [0, 2, 1, 3]);  view_1441 = None
	        clone_123: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_154, memory_format = torch.contiguous_format);  permute_154 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_8999: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1455: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_8989, [mul_8999, 1280]);  mul_8989 = mul_8999 = None
	        permute_155: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1454, [1, 0]);  view_1454 = None
	        mm_default_83: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1455, permute_155);  view_1455 = permute_155 = None
	        add_tensor_83: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_83, model_audio_tower_layers_15_self_attn_v_proj_bias);  mm_default_83 = model_audio_tower_layers_15_self_attn_v_proj_bias = None
	        view_1456: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_83, [sym_size_int, 1500, 1280]);  add_tensor_83 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1457: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1456, [sym_size_int, -1, 20, 64]);  view_1456 = None
	        permute_156: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1457, [0, 2, 1, 3]);  view_1457 = None
	        clone_124: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_156, memory_format = torch.contiguous_format);  permute_156 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_15 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_122, clone_123, clone_124, None, False, scale = 1.0);  clone_122 = clone_123 = clone_124 = None
	        getitem_122: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_15[0];  _scaled_dot_product_efficient_attention_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_157: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_122, [0, 2, 1, 3]);  getitem_122 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_157, [sym_size_int, 1500, -1]);  permute_157 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1458, [2])
	        full_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_93, full_187);  amax_93 = full_187 = None
	        amin_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1458, [2])
	        full_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_93: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_93, full_186);  amin_93 = full_186 = None
	        sub_4265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_93, minimum_93);  maximum_93 = None
	        div_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4265, 255.0);  sub_4265 = None
	        clamp_min_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_186, 1.1920928955078125e-07);  div_186 = None
	        div_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_93, clamp_min_279);  minimum_93 = None
	        round_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_187);  div_187 = None
	        sub_4271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_187);  round_187 = None
	        clamp_min_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4271, -128);  sub_4271 = None
	        clamp_max_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_280, 127);  clamp_min_280 = None
	        view_1461: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_279, [sym_size_int, 1500, 1])
	        reciprocal_93: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1461);  view_1461 = None
	        mul_9069: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_93, 1.0);  reciprocal_93 = None
	        mul_9072: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1458, mul_9069);  view_1458 = mul_9069 = None
	        round_188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9072);  mul_9072 = None
	        convert_element_type_558: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_186, torch.int8);  clamp_max_186 = None
	        view_1462: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_558, [sym_size_int, 1500, 1])
	        add_14353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_188, view_1462);  round_188 = view_1462 = None
	        clamp_min_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14353, -128);  add_14353 = None
	        clamp_max_187: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_281, 127);  clamp_min_281 = None
	        convert_element_type_559: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_187, torch.int8);  clamp_max_187 = None
	        view_1466: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_558, [sym_size_int, 1500, 1]);  convert_element_type_558 = None
	        convert_element_type_560: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_559, torch.float32);  convert_element_type_559 = None
	        convert_element_type_561: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1466, torch.float32);  view_1466 = None
	        sub_4291: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_560, convert_element_type_561);  convert_element_type_560 = convert_element_type_561 = None
	        view_1465: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_279, [sym_size_int, 1500, 1]);  clamp_min_279 = None
	        mul_9094: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4291, view_1465);  sub_4291 = view_1465 = None
	        view_1468: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1470: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_562: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1468, torch.float32);  view_1468 = None
	        convert_element_type_563: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1470, torch.float32);  view_1470 = None
	        sub_4295: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_562, convert_element_type_563);  convert_element_type_562 = convert_element_type_563 = None
	        view_1469: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_9099: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4295, view_1469);  sub_4295 = view_1469 = None
	        view_1471: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9099, [1280, 1280]);  mul_9099 = None
	        mul_9104: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1472: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9094, [mul_9104, 1280]);  mul_9094 = mul_9104 = None
	        permute_158: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1471, [1, 0]);  view_1471 = None
	        mm_default_82: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1472, permute_158);  view_1472 = permute_158 = None
	        add_tensor_82: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_82, model_audio_tower_layers_15_self_attn_out_proj_bias);  mm_default_82 = model_audio_tower_layers_15_self_attn_out_proj_bias = None
	        view_1473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_82, [sym_size_int, 1500, 1280]);  add_tensor_82 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_14416: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_13796, view_1473);  add_13796 = view_1473 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_126: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_14416, memory_format = torch.contiguous_format)
	        var_mean_31 = torch.ops.aten.var_mean.correction(clone_126, [2], correction = 0, keepdim = True)
	        getitem_126: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_31[0]
	        getitem_127: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_31[1];  var_mean_31 = None
	        sub_4301: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_126, getitem_127);  clone_126 = getitem_127 = None
	        add_14421: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_126, 1e-05);  getitem_126 = None
	        rsqrt_31: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_14421);  add_14421 = None
	        mul_9115: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4301, rsqrt_31);  sub_4301 = rsqrt_31 = None
	        mul_9116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9115, model_audio_tower_layers_15_final_layer_norm_weight);  mul_9115 = model_audio_tower_layers_15_final_layer_norm_weight = None
	        add_14422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9116, model_audio_tower_layers_15_final_layer_norm_bias);  mul_9116 = model_audio_tower_layers_15_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_14422, [2])
	        full_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_94, full_189);  amax_94 = full_189 = None
	        amin_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_14422, [2])
	        full_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_94: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_94, full_188);  amin_94 = full_188 = None
	        sub_4312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_94, minimum_94);  maximum_94 = None
	        div_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4312, 255.0);  sub_4312 = None
	        clamp_min_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_188, 1.1920928955078125e-07);  div_188 = None
	        div_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_94, clamp_min_282);  minimum_94 = None
	        round_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_189);  div_189 = None
	        sub_4318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_189);  round_189 = None
	        clamp_min_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4318, -128);  sub_4318 = None
	        clamp_max_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_283, 127);  clamp_min_283 = None
	        view_1476: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_282, [sym_size_int, 1500, 1])
	        reciprocal_94: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1476);  view_1476 = None
	        mul_9164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_94, 1.0);  reciprocal_94 = None
	        mul_9167: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_14422, mul_9164);  add_14422 = mul_9164 = None
	        round_190: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9167);  mul_9167 = None
	        convert_element_type_564: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_188, torch.int8);  clamp_max_188 = None
	        view_1477: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_564, [sym_size_int, 1500, 1])
	        add_14509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_190, view_1477);  round_190 = view_1477 = None
	        clamp_min_284: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14509, -128);  add_14509 = None
	        clamp_max_189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_284, 127);  clamp_min_284 = None
	        convert_element_type_565: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_189, torch.int8);  clamp_max_189 = None
	        view_1481: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_564, [sym_size_int, 1500, 1]);  convert_element_type_564 = None
	        convert_element_type_566: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_565, torch.float32);  convert_element_type_565 = None
	        convert_element_type_567: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1481, torch.float32);  view_1481 = None
	        sub_4338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_566, convert_element_type_567);  convert_element_type_566 = convert_element_type_567 = None
	        view_1480: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_282, [sym_size_int, 1500, 1]);  clamp_min_282 = None
	        mul_9189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4338, view_1480);  sub_4338 = view_1480 = None
	        view_1483: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_15_fc1_parametrizations_weight_original0 = None
	        view_1485: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_15_fc1_parametrizations_weight_original2 = None
	        convert_element_type_568: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1483, torch.float32);  view_1483 = None
	        convert_element_type_569: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1485, torch.float32);  view_1485 = None
	        sub_4342: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_568, convert_element_type_569);  convert_element_type_568 = convert_element_type_569 = None
	        view_1484: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_15_fc1_parametrizations_weight_original1 = None
	        mul_9194: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4342, view_1484);  sub_4342 = view_1484 = None
	        view_1486: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9194, [5120, 1280]);  mul_9194 = None
	        mul_9199: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1487: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9189, [mul_9199, 1280]);  mul_9189 = mul_9199 = None
	        permute_159: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1486, [1, 0]);  view_1486 = None
	        mm_default_81: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_1487, permute_159);  view_1487 = permute_159 = None
	        add_tensor_81: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_81, model_audio_tower_layers_15_fc1_bias);  mm_default_81 = model_audio_tower_layers_15_fc1_bias = None
	        view_1488: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_81, [sym_size_int, 1500, 5120]);  add_tensor_81 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_9206: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1488, 0.5)
	        mul_9207: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1488, 0.7071067811865476);  view_1488 = None
	        erf_17: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_9207);  mul_9207 = None
	        add_14568: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_17, 1);  erf_17 = None
	        mul_9208: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9206, add_14568);  mul_9206 = add_14568 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_9208, [2])
	        full_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_95, full_191);  amax_95 = full_191 = None
	        amin_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_9208, [2])
	        full_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_95: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_95, full_190);  amin_95 = full_190 = None
	        sub_4355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_95, minimum_95);  maximum_95 = None
	        div_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4355, 255.0);  sub_4355 = None
	        clamp_min_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_190, 1.1920928955078125e-07);  div_190 = None
	        div_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_95, clamp_min_285);  minimum_95 = None
	        round_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_191);  div_191 = None
	        sub_4361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_191);  round_191 = None
	        clamp_min_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4361, -128);  sub_4361 = None
	        clamp_max_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_286, 127);  clamp_min_286 = None
	        view_1491: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_285, [sym_size_int, 1500, 1])
	        reciprocal_95: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1491);  view_1491 = None
	        mul_9254: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_95, 1.0);  reciprocal_95 = None
	        mul_9257: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9208, mul_9254);  mul_9208 = mul_9254 = None
	        round_192: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_9257);  mul_9257 = None
	        convert_element_type_570: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_190, torch.int8);  clamp_max_190 = None
	        view_1492: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_570, [sym_size_int, 1500, 1])
	        add_14651: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_192, view_1492);  round_192 = view_1492 = None
	        clamp_min_287: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14651, -128);  add_14651 = None
	        clamp_max_191: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_287, 127);  clamp_min_287 = None
	        convert_element_type_571: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_191, torch.int8);  clamp_max_191 = None
	        view_1496: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_570, [sym_size_int, 1500, 1]);  convert_element_type_570 = None
	        convert_element_type_572: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_571, torch.float32);  convert_element_type_571 = None
	        convert_element_type_573: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1496, torch.float32);  view_1496 = None
	        sub_4381: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_572, convert_element_type_573);  convert_element_type_572 = convert_element_type_573 = None
	        view_1495: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_285, [sym_size_int, 1500, 1]);  clamp_min_285 = None
	        mul_9279: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4381, view_1495);  sub_4381 = view_1495 = None
	        view_1498: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_15_fc2_parametrizations_weight_original0 = None
	        view_1500: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_15_fc2_parametrizations_weight_original2 = None
	        convert_element_type_574: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1498, torch.float32);  view_1498 = None
	        convert_element_type_575: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1500, torch.float32);  view_1500 = None
	        sub_4385: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_574, convert_element_type_575);  convert_element_type_574 = convert_element_type_575 = None
	        view_1499: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_15_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_15_fc2_parametrizations_weight_original1 = None
	        mul_9284: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4385, view_1499);  sub_4385 = view_1499 = None
	        view_1501: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9284, [1280, 5120]);  mul_9284 = None
	        mul_9289: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1502: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9279, [mul_9289, 5120]);  mul_9279 = mul_9289 = None
	        permute_160: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1501, [1, 0]);  view_1501 = None
	        mm_default_80: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1502, permute_160);  view_1502 = permute_160 = None
	        add_tensor_80: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_80, model_audio_tower_layers_15_fc2_bias);  mm_default_80 = model_audio_tower_layers_15_fc2_bias = None
	        view_1503: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_80, [sym_size_int, 1500, 1280]);  add_tensor_80 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_14714: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_14416, view_1503);  add_14416 = view_1503 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_129: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_14714, memory_format = torch.contiguous_format)
	        var_mean_32 = torch.ops.aten.var_mean.correction(clone_129, [2], correction = 0, keepdim = True)
	        getitem_128: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_32[0]
	        getitem_129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_32[1];  var_mean_32 = None
	        sub_4391: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_129, getitem_129);  clone_129 = getitem_129 = None
	        add_14719: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_128, 1e-05);  getitem_128 = None
	        rsqrt_32: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_14719);  add_14719 = None
	        mul_9300: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4391, rsqrt_32);  sub_4391 = rsqrt_32 = None
	        mul_9301: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9300, model_audio_tower_layers_16_self_attn_layer_norm_weight);  mul_9300 = model_audio_tower_layers_16_self_attn_layer_norm_weight = None
	        add_14720: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9301, model_audio_tower_layers_16_self_attn_layer_norm_bias);  mul_9301 = model_audio_tower_layers_16_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_14720, [2])
	        full_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_96, full_193);  amax_96 = full_193 = None
	        amin_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_14720, [2])
	        full_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_96: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_96, full_192);  amin_96 = full_192 = None
	        sub_4402: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_96, minimum_96);  maximum_96 = None
	        div_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4402, 255.0);  sub_4402 = None
	        clamp_min_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_192, 1.1920928955078125e-07);  div_192 = None
	        div_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_96, clamp_min_288);  minimum_96 = None
	        round_193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_193);  div_193 = None
	        sub_4408: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_193);  round_193 = None
	        clamp_min_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4408, -128);  sub_4408 = None
	        clamp_max_192: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_289, 127);  clamp_min_289 = None
	        view_1506: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_288, [sym_size_int, 1500, 1])
	        reciprocal_96: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1506);  view_1506 = None
	        mul_9349: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_96, 1.0);  reciprocal_96 = None
	        mul_9352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_14720, mul_9349);  mul_9349 = None
	        round_194: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9352);  mul_9352 = None
	        convert_element_type_576: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_192, torch.int8);  clamp_max_192 = None
	        view_1507: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_576, [sym_size_int, 1500, 1])
	        add_14807: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_194, view_1507);  round_194 = view_1507 = None
	        clamp_min_290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14807, -128);  add_14807 = None
	        clamp_max_193: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_290, 127);  clamp_min_290 = None
	        convert_element_type_577: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_193, torch.int8);  clamp_max_193 = None
	        view_1511: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_576, [sym_size_int, 1500, 1]);  convert_element_type_576 = None
	        convert_element_type_578: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_577, torch.float32);  convert_element_type_577 = None
	        convert_element_type_579: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1511, torch.float32);  view_1511 = None
	        sub_4428: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_578, convert_element_type_579);  convert_element_type_578 = convert_element_type_579 = None
	        view_1510: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_288, [sym_size_int, 1500, 1]);  clamp_min_288 = None
	        mul_9374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4428, view_1510);  sub_4428 = view_1510 = None
	        view_1513: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1515: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_580: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1513, torch.float32);  view_1513 = None
	        convert_element_type_581: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1515, torch.float32);  view_1515 = None
	        sub_4432: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_580, convert_element_type_581);  convert_element_type_580 = convert_element_type_581 = None
	        view_1514: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_9379: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4432, view_1514);  sub_4432 = view_1514 = None
	        view_1516: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9379, [1280, 1280]);  mul_9379 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_14720, [2])
	        full_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_97, full_195);  amax_97 = full_195 = None
	        amin_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_14720, [2])
	        full_194: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_97: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_97, full_194);  amin_97 = full_194 = None
	        sub_4447: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_97, minimum_97);  maximum_97 = None
	        div_194: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4447, 255.0);  sub_4447 = None
	        clamp_min_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_194, 1.1920928955078125e-07);  div_194 = None
	        div_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_97, clamp_min_291);  minimum_97 = None
	        round_195: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_195);  div_195 = None
	        sub_4453: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_195);  round_195 = None
	        clamp_min_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4453, -128);  sub_4453 = None
	        clamp_max_194: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_292, 127);  clamp_min_292 = None
	        view_1522: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_291, [sym_size_int, 1500, 1])
	        reciprocal_97: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1522);  view_1522 = None
	        mul_9445: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_97, 1.0);  reciprocal_97 = None
	        mul_9448: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_14720, mul_9445);  mul_9445 = None
	        round_196: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9448);  mul_9448 = None
	        convert_element_type_582: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_194, torch.int8);  clamp_max_194 = None
	        view_1523: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_582, [sym_size_int, 1500, 1])
	        add_14959: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_196, view_1523);  round_196 = view_1523 = None
	        clamp_min_293: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_14959, -128);  add_14959 = None
	        clamp_max_195: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_293, 127);  clamp_min_293 = None
	        convert_element_type_583: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_195, torch.int8);  clamp_max_195 = None
	        view_1527: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_582, [sym_size_int, 1500, 1]);  convert_element_type_582 = None
	        convert_element_type_584: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_583, torch.float32);  convert_element_type_583 = None
	        convert_element_type_585: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1527, torch.float32);  view_1527 = None
	        sub_4473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_584, convert_element_type_585);  convert_element_type_584 = convert_element_type_585 = None
	        view_1526: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_291, [sym_size_int, 1500, 1]);  clamp_min_291 = None
	        mul_9470: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4473, view_1526);  sub_4473 = view_1526 = None
	        view_1529: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1531: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_586: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1529, torch.float32);  view_1529 = None
	        convert_element_type_587: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1531, torch.float32);  view_1531 = None
	        sub_4477: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_586, convert_element_type_587);  convert_element_type_586 = convert_element_type_587 = None
	        view_1530: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_9475: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4477, view_1530);  sub_4477 = view_1530 = None
	        view_1532: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9475, [1280, 1280]);  mul_9475 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_14720, [2])
	        full_197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_98, full_197);  amax_98 = full_197 = None
	        amin_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_14720, [2])
	        full_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_98: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_98, full_196);  amin_98 = full_196 = None
	        sub_4491: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_98, minimum_98);  maximum_98 = None
	        div_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4491, 255.0);  sub_4491 = None
	        clamp_min_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_196, 1.1920928955078125e-07);  div_196 = None
	        div_197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_98, clamp_min_294);  minimum_98 = None
	        round_197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_197);  div_197 = None
	        sub_4497: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_197);  round_197 = None
	        clamp_min_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4497, -128);  sub_4497 = None
	        clamp_max_196: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_295, 127);  clamp_min_295 = None
	        view_1538: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_294, [sym_size_int, 1500, 1])
	        reciprocal_98: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1538);  view_1538 = None
	        mul_9544: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_98, 1.0);  reciprocal_98 = None
	        mul_9547: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_14720, mul_9544);  add_14720 = mul_9544 = None
	        round_198: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9547);  mul_9547 = None
	        convert_element_type_588: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_196, torch.int8);  clamp_max_196 = None
	        view_1539: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_588, [sym_size_int, 1500, 1])
	        add_15107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_198, view_1539);  round_198 = view_1539 = None
	        clamp_min_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15107, -128);  add_15107 = None
	        clamp_max_197: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_296, 127);  clamp_min_296 = None
	        convert_element_type_589: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_197, torch.int8);  clamp_max_197 = None
	        view_1543: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_588, [sym_size_int, 1500, 1]);  convert_element_type_588 = None
	        convert_element_type_590: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_589, torch.float32);  convert_element_type_589 = None
	        convert_element_type_591: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1543, torch.float32);  view_1543 = None
	        sub_4517: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_590, convert_element_type_591);  convert_element_type_590 = convert_element_type_591 = None
	        view_1542: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_294, [sym_size_int, 1500, 1]);  clamp_min_294 = None
	        mul_9569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4517, view_1542);  sub_4517 = view_1542 = None
	        view_1545: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1547: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_592: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1545, torch.float32);  view_1545 = None
	        convert_element_type_593: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1547, torch.float32);  view_1547 = None
	        sub_4521: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_592, convert_element_type_593);  convert_element_type_592 = convert_element_type_593 = None
	        view_1546: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_9574: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4521, view_1546);  sub_4521 = view_1546 = None
	        view_1548: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9574, [1280, 1280]);  mul_9574 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_9384: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1517: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9374, [mul_9384, 1280]);  mul_9374 = mul_9384 = None
	        permute_161: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1516, [1, 0]);  view_1516 = None
	        mm_default_79: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1517, permute_161);  view_1517 = permute_161 = None
	        add_tensor_79: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_79, model_audio_tower_layers_16_self_attn_q_proj_bias);  mm_default_79 = model_audio_tower_layers_16_self_attn_q_proj_bias = None
	        view_1518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_79, [sym_size_int, 1500, 1280]);  add_tensor_79 = None
	        mul_9391: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1518, 0.125);  view_1518 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1519: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9391, [sym_size_int, 1500, 20, 64]);  mul_9391 = None
	        permute_162: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1519, [0, 2, 1, 3]);  view_1519 = None
	        clone_130: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_162, memory_format = torch.contiguous_format);  permute_162 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_9478: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1533: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9470, [mul_9478, 1280]);  mul_9470 = mul_9478 = None
	        permute_163: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1532, [1, 0]);  view_1532 = None
	        mm_16: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1533, permute_163);  view_1533 = permute_163 = None
	        view_1534: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_16, [sym_size_int, 1500, 1280]);  mm_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1535: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1534, [sym_size_int, -1, 20, 64]);  view_1534 = None
	        permute_164: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1535, [0, 2, 1, 3]);  view_1535 = None
	        clone_131: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_164, memory_format = torch.contiguous_format);  permute_164 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_9579: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1549: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9569, [mul_9579, 1280]);  mul_9569 = mul_9579 = None
	        permute_165: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1548, [1, 0]);  view_1548 = None
	        mm_default_78: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1549, permute_165);  view_1549 = permute_165 = None
	        add_tensor_78: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_78, model_audio_tower_layers_16_self_attn_v_proj_bias);  mm_default_78 = model_audio_tower_layers_16_self_attn_v_proj_bias = None
	        view_1550: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_78, [sym_size_int, 1500, 1280]);  add_tensor_78 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1551: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1550, [sym_size_int, -1, 20, 64]);  view_1550 = None
	        permute_166: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1551, [0, 2, 1, 3]);  view_1551 = None
	        clone_132: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_166, memory_format = torch.contiguous_format);  permute_166 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_16 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_130, clone_131, clone_132, None, False, scale = 1.0);  clone_130 = clone_131 = clone_132 = None
	        getitem_130: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_16[0];  _scaled_dot_product_efficient_attention_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_167: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_130, [0, 2, 1, 3]);  getitem_130 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_167, [sym_size_int, 1500, -1]);  permute_167 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1552, [2])
	        full_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_99, full_199);  amax_99 = full_199 = None
	        amin_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1552, [2])
	        full_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_99: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_99, full_198);  amin_99 = full_198 = None
	        sub_4539: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_99, minimum_99);  maximum_99 = None
	        div_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4539, 255.0);  sub_4539 = None
	        clamp_min_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_198, 1.1920928955078125e-07);  div_198 = None
	        div_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_99, clamp_min_297);  minimum_99 = None
	        round_199: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_199);  div_199 = None
	        sub_4545: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_199);  round_199 = None
	        clamp_min_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4545, -128);  sub_4545 = None
	        clamp_max_198: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_298, 127);  clamp_min_298 = None
	        view_1555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_297, [sym_size_int, 1500, 1])
	        reciprocal_99: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1555);  view_1555 = None
	        mul_9649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_99, 1.0);  reciprocal_99 = None
	        mul_9652: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1552, mul_9649);  view_1552 = mul_9649 = None
	        round_200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9652);  mul_9652 = None
	        convert_element_type_594: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_198, torch.int8);  clamp_max_198 = None
	        view_1556: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_594, [sym_size_int, 1500, 1])
	        add_15271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_200, view_1556);  round_200 = view_1556 = None
	        clamp_min_299: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15271, -128);  add_15271 = None
	        clamp_max_199: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_299, 127);  clamp_min_299 = None
	        convert_element_type_595: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_199, torch.int8);  clamp_max_199 = None
	        view_1560: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_594, [sym_size_int, 1500, 1]);  convert_element_type_594 = None
	        convert_element_type_596: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_595, torch.float32);  convert_element_type_595 = None
	        convert_element_type_597: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1560, torch.float32);  view_1560 = None
	        sub_4565: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_596, convert_element_type_597);  convert_element_type_596 = convert_element_type_597 = None
	        view_1559: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_297, [sym_size_int, 1500, 1]);  clamp_min_297 = None
	        mul_9674: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4565, view_1559);  sub_4565 = view_1559 = None
	        view_1562: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1564: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_598: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1562, torch.float32);  view_1562 = None
	        convert_element_type_599: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1564, torch.float32);  view_1564 = None
	        sub_4569: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_598, convert_element_type_599);  convert_element_type_598 = convert_element_type_599 = None
	        view_1563: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_9679: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4569, view_1563);  sub_4569 = view_1563 = None
	        view_1565: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9679, [1280, 1280]);  mul_9679 = None
	        mul_9684: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1566: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9674, [mul_9684, 1280]);  mul_9674 = mul_9684 = None
	        permute_168: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1565, [1, 0]);  view_1565 = None
	        mm_default_77: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1566, permute_168);  view_1566 = permute_168 = None
	        add_tensor_77: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_77, model_audio_tower_layers_16_self_attn_out_proj_bias);  mm_default_77 = model_audio_tower_layers_16_self_attn_out_proj_bias = None
	        view_1567: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_77, [sym_size_int, 1500, 1280]);  add_tensor_77 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_15334: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_14714, view_1567);  add_14714 = view_1567 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_134: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_15334, memory_format = torch.contiguous_format)
	        var_mean_33 = torch.ops.aten.var_mean.correction(clone_134, [2], correction = 0, keepdim = True)
	        getitem_134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_33[0]
	        getitem_135: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_33[1];  var_mean_33 = None
	        sub_4575: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_134, getitem_135);  clone_134 = getitem_135 = None
	        add_15339: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_134, 1e-05);  getitem_134 = None
	        rsqrt_33: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_15339);  add_15339 = None
	        mul_9695: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4575, rsqrt_33);  sub_4575 = rsqrt_33 = None
	        mul_9696: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9695, model_audio_tower_layers_16_final_layer_norm_weight);  mul_9695 = model_audio_tower_layers_16_final_layer_norm_weight = None
	        add_15340: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9696, model_audio_tower_layers_16_final_layer_norm_bias);  mul_9696 = model_audio_tower_layers_16_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_15340, [2])
	        full_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_100, full_201);  amax_100 = full_201 = None
	        amin_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_15340, [2])
	        full_200: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_100: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_100, full_200);  amin_100 = full_200 = None
	        sub_4586: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_100, minimum_100);  maximum_100 = None
	        div_200: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4586, 255.0);  sub_4586 = None
	        clamp_min_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_200, 1.1920928955078125e-07);  div_200 = None
	        div_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_100, clamp_min_300);  minimum_100 = None
	        round_201: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_201);  div_201 = None
	        sub_4592: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_201);  round_201 = None
	        clamp_min_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4592, -128);  sub_4592 = None
	        clamp_max_200: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_301, 127);  clamp_min_301 = None
	        view_1570: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_300, [sym_size_int, 1500, 1])
	        reciprocal_100: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1570);  view_1570 = None
	        mul_9744: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_100, 1.0);  reciprocal_100 = None
	        mul_9747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_15340, mul_9744);  add_15340 = mul_9744 = None
	        round_202: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9747);  mul_9747 = None
	        convert_element_type_600: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_200, torch.int8);  clamp_max_200 = None
	        view_1571: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_600, [sym_size_int, 1500, 1])
	        add_15427: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_202, view_1571);  round_202 = view_1571 = None
	        clamp_min_302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15427, -128);  add_15427 = None
	        clamp_max_201: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_302, 127);  clamp_min_302 = None
	        convert_element_type_601: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_201, torch.int8);  clamp_max_201 = None
	        view_1575: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_600, [sym_size_int, 1500, 1]);  convert_element_type_600 = None
	        convert_element_type_602: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_601, torch.float32);  convert_element_type_601 = None
	        convert_element_type_603: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1575, torch.float32);  view_1575 = None
	        sub_4612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_602, convert_element_type_603);  convert_element_type_602 = convert_element_type_603 = None
	        view_1574: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_300, [sym_size_int, 1500, 1]);  clamp_min_300 = None
	        mul_9769: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4612, view_1574);  sub_4612 = view_1574 = None
	        view_1577: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_16_fc1_parametrizations_weight_original0 = None
	        view_1579: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_16_fc1_parametrizations_weight_original2 = None
	        convert_element_type_604: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1577, torch.float32);  view_1577 = None
	        convert_element_type_605: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1579, torch.float32);  view_1579 = None
	        sub_4616: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_604, convert_element_type_605);  convert_element_type_604 = convert_element_type_605 = None
	        view_1578: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_16_fc1_parametrizations_weight_original1 = None
	        mul_9774: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4616, view_1578);  sub_4616 = view_1578 = None
	        view_1580: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9774, [5120, 1280]);  mul_9774 = None
	        mul_9779: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1581: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9769, [mul_9779, 1280]);  mul_9769 = mul_9779 = None
	        permute_169: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1580, [1, 0]);  view_1580 = None
	        mm_default_76: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_1581, permute_169);  view_1581 = permute_169 = None
	        add_tensor_76: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_76, model_audio_tower_layers_16_fc1_bias);  mm_default_76 = model_audio_tower_layers_16_fc1_bias = None
	        view_1582: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_76, [sym_size_int, 1500, 5120]);  add_tensor_76 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_9786: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1582, 0.5)
	        mul_9787: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1582, 0.7071067811865476);  view_1582 = None
	        erf_18: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_9787);  mul_9787 = None
	        add_15486: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_18, 1);  erf_18 = None
	        mul_9788: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9786, add_15486);  mul_9786 = add_15486 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_9788, [2])
	        full_203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_101, full_203);  amax_101 = full_203 = None
	        amin_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_9788, [2])
	        full_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_101, full_202);  amin_101 = full_202 = None
	        sub_4629: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_101, minimum_101);  maximum_101 = None
	        div_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4629, 255.0);  sub_4629 = None
	        clamp_min_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_202, 1.1920928955078125e-07);  div_202 = None
	        div_203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_101, clamp_min_303);  minimum_101 = None
	        round_203: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_203);  div_203 = None
	        sub_4635: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_203);  round_203 = None
	        clamp_min_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4635, -128);  sub_4635 = None
	        clamp_max_202: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_304, 127);  clamp_min_304 = None
	        view_1585: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_303, [sym_size_int, 1500, 1])
	        reciprocal_101: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1585);  view_1585 = None
	        mul_9834: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_101, 1.0);  reciprocal_101 = None
	        mul_9837: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9788, mul_9834);  mul_9788 = mul_9834 = None
	        round_204: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_9837);  mul_9837 = None
	        convert_element_type_606: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_202, torch.int8);  clamp_max_202 = None
	        view_1586: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_606, [sym_size_int, 1500, 1])
	        add_15569: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_204, view_1586);  round_204 = view_1586 = None
	        clamp_min_305: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15569, -128);  add_15569 = None
	        clamp_max_203: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_305, 127);  clamp_min_305 = None
	        convert_element_type_607: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_203, torch.int8);  clamp_max_203 = None
	        view_1590: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_606, [sym_size_int, 1500, 1]);  convert_element_type_606 = None
	        convert_element_type_608: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_607, torch.float32);  convert_element_type_607 = None
	        convert_element_type_609: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1590, torch.float32);  view_1590 = None
	        sub_4655: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_608, convert_element_type_609);  convert_element_type_608 = convert_element_type_609 = None
	        view_1589: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_303, [sym_size_int, 1500, 1]);  clamp_min_303 = None
	        mul_9859: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4655, view_1589);  sub_4655 = view_1589 = None
	        view_1592: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_16_fc2_parametrizations_weight_original0 = None
	        view_1594: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_16_fc2_parametrizations_weight_original2 = None
	        convert_element_type_610: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1592, torch.float32);  view_1592 = None
	        convert_element_type_611: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1594, torch.float32);  view_1594 = None
	        sub_4659: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_610, convert_element_type_611);  convert_element_type_610 = convert_element_type_611 = None
	        view_1593: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_16_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_16_fc2_parametrizations_weight_original1 = None
	        mul_9864: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4659, view_1593);  sub_4659 = view_1593 = None
	        view_1595: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9864, [1280, 5120]);  mul_9864 = None
	        mul_9869: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1596: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9859, [mul_9869, 5120]);  mul_9859 = mul_9869 = None
	        permute_170: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1595, [1, 0]);  view_1595 = None
	        mm_default_75: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1596, permute_170);  view_1596 = permute_170 = None
	        add_tensor_75: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_75, model_audio_tower_layers_16_fc2_bias);  mm_default_75 = model_audio_tower_layers_16_fc2_bias = None
	        view_1597: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_75, [sym_size_int, 1500, 1280]);  add_tensor_75 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_15632: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_15334, view_1597);  add_15334 = view_1597 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_15632, memory_format = torch.contiguous_format)
	        var_mean_34 = torch.ops.aten.var_mean.correction(clone_137, [2], correction = 0, keepdim = True)
	        getitem_136: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_34[0]
	        getitem_137: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_34[1];  var_mean_34 = None
	        sub_4665: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_137, getitem_137);  clone_137 = getitem_137 = None
	        add_15637: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_136, 1e-05);  getitem_136 = None
	        rsqrt_34: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_15637);  add_15637 = None
	        mul_9880: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4665, rsqrt_34);  sub_4665 = rsqrt_34 = None
	        mul_9881: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_9880, model_audio_tower_layers_17_self_attn_layer_norm_weight);  mul_9880 = model_audio_tower_layers_17_self_attn_layer_norm_weight = None
	        add_15638: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_9881, model_audio_tower_layers_17_self_attn_layer_norm_bias);  mul_9881 = model_audio_tower_layers_17_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_15638, [2])
	        full_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_102, full_205);  amax_102 = full_205 = None
	        amin_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_15638, [2])
	        full_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_102: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_102, full_204);  amin_102 = full_204 = None
	        sub_4676: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_102, minimum_102);  maximum_102 = None
	        div_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4676, 255.0);  sub_4676 = None
	        clamp_min_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_204, 1.1920928955078125e-07);  div_204 = None
	        div_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_102, clamp_min_306);  minimum_102 = None
	        round_205: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_205);  div_205 = None
	        sub_4682: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_205);  round_205 = None
	        clamp_min_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4682, -128);  sub_4682 = None
	        clamp_max_204: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_307, 127);  clamp_min_307 = None
	        view_1600: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_306, [sym_size_int, 1500, 1])
	        reciprocal_102: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1600);  view_1600 = None
	        mul_9929: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_102, 1.0);  reciprocal_102 = None
	        mul_9932: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_15638, mul_9929);  mul_9929 = None
	        round_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_9932);  mul_9932 = None
	        convert_element_type_612: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_204, torch.int8);  clamp_max_204 = None
	        view_1601: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_612, [sym_size_int, 1500, 1])
	        add_15725: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_206, view_1601);  round_206 = view_1601 = None
	        clamp_min_308: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15725, -128);  add_15725 = None
	        clamp_max_205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_308, 127);  clamp_min_308 = None
	        convert_element_type_613: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_205, torch.int8);  clamp_max_205 = None
	        view_1605: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_612, [sym_size_int, 1500, 1]);  convert_element_type_612 = None
	        convert_element_type_614: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_613, torch.float32);  convert_element_type_613 = None
	        convert_element_type_615: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1605, torch.float32);  view_1605 = None
	        sub_4702: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_614, convert_element_type_615);  convert_element_type_614 = convert_element_type_615 = None
	        view_1604: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_306, [sym_size_int, 1500, 1]);  clamp_min_306 = None
	        mul_9954: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4702, view_1604);  sub_4702 = view_1604 = None
	        view_1607: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1609: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_616: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1607, torch.float32);  view_1607 = None
	        convert_element_type_617: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1609, torch.float32);  view_1609 = None
	        sub_4706: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_616, convert_element_type_617);  convert_element_type_616 = convert_element_type_617 = None
	        view_1608: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_9959: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4706, view_1608);  sub_4706 = view_1608 = None
	        view_1610: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9959, [1280, 1280]);  mul_9959 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_15638, [2])
	        full_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_103, full_207);  amax_103 = full_207 = None
	        amin_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_15638, [2])
	        full_206: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_103: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_103, full_206);  amin_103 = full_206 = None
	        sub_4721: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_103, minimum_103);  maximum_103 = None
	        div_206: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4721, 255.0);  sub_4721 = None
	        clamp_min_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_206, 1.1920928955078125e-07);  div_206 = None
	        div_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_103, clamp_min_309);  minimum_103 = None
	        round_207: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_207);  div_207 = None
	        sub_4727: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_207);  round_207 = None
	        clamp_min_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4727, -128);  sub_4727 = None
	        clamp_max_206: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_310, 127);  clamp_min_310 = None
	        view_1616: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_309, [sym_size_int, 1500, 1])
	        reciprocal_103: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1616);  view_1616 = None
	        mul_10025: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_103, 1.0);  reciprocal_103 = None
	        mul_10028: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_15638, mul_10025);  mul_10025 = None
	        round_208: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10028);  mul_10028 = None
	        convert_element_type_618: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_206, torch.int8);  clamp_max_206 = None
	        view_1617: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_618, [sym_size_int, 1500, 1])
	        add_15877: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_208, view_1617);  round_208 = view_1617 = None
	        clamp_min_311: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_15877, -128);  add_15877 = None
	        clamp_max_207: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_311, 127);  clamp_min_311 = None
	        convert_element_type_619: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_207, torch.int8);  clamp_max_207 = None
	        view_1621: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_618, [sym_size_int, 1500, 1]);  convert_element_type_618 = None
	        convert_element_type_620: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_619, torch.float32);  convert_element_type_619 = None
	        convert_element_type_621: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1621, torch.float32);  view_1621 = None
	        sub_4747: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_620, convert_element_type_621);  convert_element_type_620 = convert_element_type_621 = None
	        view_1620: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_309, [sym_size_int, 1500, 1]);  clamp_min_309 = None
	        mul_10050: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4747, view_1620);  sub_4747 = view_1620 = None
	        view_1623: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1625: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_622: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1623, torch.float32);  view_1623 = None
	        convert_element_type_623: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1625, torch.float32);  view_1625 = None
	        sub_4751: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_622, convert_element_type_623);  convert_element_type_622 = convert_element_type_623 = None
	        view_1624: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_10055: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4751, view_1624);  sub_4751 = view_1624 = None
	        view_1626: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10055, [1280, 1280]);  mul_10055 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_15638, [2])
	        full_209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_104, full_209);  amax_104 = full_209 = None
	        amin_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_15638, [2])
	        full_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_104: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_104, full_208);  amin_104 = full_208 = None
	        sub_4765: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_104, minimum_104);  maximum_104 = None
	        div_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4765, 255.0);  sub_4765 = None
	        clamp_min_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_208, 1.1920928955078125e-07);  div_208 = None
	        div_209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_104, clamp_min_312);  minimum_104 = None
	        round_209: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_209);  div_209 = None
	        sub_4771: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_209);  round_209 = None
	        clamp_min_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4771, -128);  sub_4771 = None
	        clamp_max_208: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_313, 127);  clamp_min_313 = None
	        view_1632: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_312, [sym_size_int, 1500, 1])
	        reciprocal_104: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1632);  view_1632 = None
	        mul_10124: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_104, 1.0);  reciprocal_104 = None
	        mul_10127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_15638, mul_10124);  add_15638 = mul_10124 = None
	        round_210: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10127);  mul_10127 = None
	        convert_element_type_624: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_208, torch.int8);  clamp_max_208 = None
	        view_1633: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_624, [sym_size_int, 1500, 1])
	        add_16025: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_210, view_1633);  round_210 = view_1633 = None
	        clamp_min_314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16025, -128);  add_16025 = None
	        clamp_max_209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_314, 127);  clamp_min_314 = None
	        convert_element_type_625: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_209, torch.int8);  clamp_max_209 = None
	        view_1637: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_624, [sym_size_int, 1500, 1]);  convert_element_type_624 = None
	        convert_element_type_626: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_625, torch.float32);  convert_element_type_625 = None
	        convert_element_type_627: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1637, torch.float32);  view_1637 = None
	        sub_4791: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_626, convert_element_type_627);  convert_element_type_626 = convert_element_type_627 = None
	        view_1636: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_312, [sym_size_int, 1500, 1]);  clamp_min_312 = None
	        mul_10149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4791, view_1636);  sub_4791 = view_1636 = None
	        view_1639: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1641: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_628: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1639, torch.float32);  view_1639 = None
	        convert_element_type_629: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1641, torch.float32);  view_1641 = None
	        sub_4795: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_628, convert_element_type_629);  convert_element_type_628 = convert_element_type_629 = None
	        view_1640: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_10154: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4795, view_1640);  sub_4795 = view_1640 = None
	        view_1642: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10154, [1280, 1280]);  mul_10154 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_9964: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1611: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9954, [mul_9964, 1280]);  mul_9954 = mul_9964 = None
	        permute_171: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1610, [1, 0]);  view_1610 = None
	        mm_default_74: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1611, permute_171);  view_1611 = permute_171 = None
	        add_tensor_74: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_74, model_audio_tower_layers_17_self_attn_q_proj_bias);  mm_default_74 = model_audio_tower_layers_17_self_attn_q_proj_bias = None
	        view_1612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_74, [sym_size_int, 1500, 1280]);  add_tensor_74 = None
	        mul_9971: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1612, 0.125);  view_1612 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1613: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_9971, [sym_size_int, 1500, 20, 64]);  mul_9971 = None
	        permute_172: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1613, [0, 2, 1, 3]);  view_1613 = None
	        clone_138: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_172, memory_format = torch.contiguous_format);  permute_172 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_10058: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1627: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10050, [mul_10058, 1280]);  mul_10050 = mul_10058 = None
	        permute_173: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1626, [1, 0]);  view_1626 = None
	        mm_17: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1627, permute_173);  view_1627 = permute_173 = None
	        view_1628: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_17, [sym_size_int, 1500, 1280]);  mm_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1629: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1628, [sym_size_int, -1, 20, 64]);  view_1628 = None
	        permute_174: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1629, [0, 2, 1, 3]);  view_1629 = None
	        clone_139: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_174, memory_format = torch.contiguous_format);  permute_174 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_10159: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1643: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10149, [mul_10159, 1280]);  mul_10149 = mul_10159 = None
	        permute_175: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1642, [1, 0]);  view_1642 = None
	        mm_default_73: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1643, permute_175);  view_1643 = permute_175 = None
	        add_tensor_73: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_73, model_audio_tower_layers_17_self_attn_v_proj_bias);  mm_default_73 = model_audio_tower_layers_17_self_attn_v_proj_bias = None
	        view_1644: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_73, [sym_size_int, 1500, 1280]);  add_tensor_73 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1645: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1644, [sym_size_int, -1, 20, 64]);  view_1644 = None
	        permute_176: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1645, [0, 2, 1, 3]);  view_1645 = None
	        clone_140: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_176, memory_format = torch.contiguous_format);  permute_176 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_17 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_138, clone_139, clone_140, None, False, scale = 1.0);  clone_138 = clone_139 = clone_140 = None
	        getitem_138: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_17[0];  _scaled_dot_product_efficient_attention_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_177: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_138, [0, 2, 1, 3]);  getitem_138 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1646: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_177, [sym_size_int, 1500, -1]);  permute_177 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1646, [2])
	        full_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_105, full_211);  amax_105 = full_211 = None
	        amin_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1646, [2])
	        full_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_105: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_105, full_210);  amin_105 = full_210 = None
	        sub_4813: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_105, minimum_105);  maximum_105 = None
	        div_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4813, 255.0);  sub_4813 = None
	        clamp_min_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_210, 1.1920928955078125e-07);  div_210 = None
	        div_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_105, clamp_min_315);  minimum_105 = None
	        round_211: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_211);  div_211 = None
	        sub_4819: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_211);  round_211 = None
	        clamp_min_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4819, -128);  sub_4819 = None
	        clamp_max_210: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_316, 127);  clamp_min_316 = None
	        view_1649: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_315, [sym_size_int, 1500, 1])
	        reciprocal_105: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1649);  view_1649 = None
	        mul_10229: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_105, 1.0);  reciprocal_105 = None
	        mul_10232: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1646, mul_10229);  view_1646 = mul_10229 = None
	        round_212: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10232);  mul_10232 = None
	        convert_element_type_630: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_210, torch.int8);  clamp_max_210 = None
	        view_1650: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_630, [sym_size_int, 1500, 1])
	        add_16189: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_212, view_1650);  round_212 = view_1650 = None
	        clamp_min_317: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16189, -128);  add_16189 = None
	        clamp_max_211: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_317, 127);  clamp_min_317 = None
	        convert_element_type_631: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_211, torch.int8);  clamp_max_211 = None
	        view_1654: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_630, [sym_size_int, 1500, 1]);  convert_element_type_630 = None
	        convert_element_type_632: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_631, torch.float32);  convert_element_type_631 = None
	        convert_element_type_633: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1654, torch.float32);  view_1654 = None
	        sub_4839: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_632, convert_element_type_633);  convert_element_type_632 = convert_element_type_633 = None
	        view_1653: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_315, [sym_size_int, 1500, 1]);  clamp_min_315 = None
	        mul_10254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4839, view_1653);  sub_4839 = view_1653 = None
	        view_1656: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1658: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_634: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1656, torch.float32);  view_1656 = None
	        convert_element_type_635: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1658, torch.float32);  view_1658 = None
	        sub_4843: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_634, convert_element_type_635);  convert_element_type_634 = convert_element_type_635 = None
	        view_1657: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_10259: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4843, view_1657);  sub_4843 = view_1657 = None
	        view_1659: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10259, [1280, 1280]);  mul_10259 = None
	        mul_10264: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1660: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10254, [mul_10264, 1280]);  mul_10254 = mul_10264 = None
	        permute_178: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1659, [1, 0]);  view_1659 = None
	        mm_default_72: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1660, permute_178);  view_1660 = permute_178 = None
	        add_tensor_72: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_72, model_audio_tower_layers_17_self_attn_out_proj_bias);  mm_default_72 = model_audio_tower_layers_17_self_attn_out_proj_bias = None
	        view_1661: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_72, [sym_size_int, 1500, 1280]);  add_tensor_72 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_16252: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_15632, view_1661);  add_15632 = view_1661 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_16252, memory_format = torch.contiguous_format)
	        var_mean_35 = torch.ops.aten.var_mean.correction(clone_142, [2], correction = 0, keepdim = True)
	        getitem_142: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_35[0]
	        getitem_143: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_35[1];  var_mean_35 = None
	        sub_4849: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_142, getitem_143);  clone_142 = getitem_143 = None
	        add_16257: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_142, 1e-05);  getitem_142 = None
	        rsqrt_35: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_16257);  add_16257 = None
	        mul_10275: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4849, rsqrt_35);  sub_4849 = rsqrt_35 = None
	        mul_10276: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10275, model_audio_tower_layers_17_final_layer_norm_weight);  mul_10275 = model_audio_tower_layers_17_final_layer_norm_weight = None
	        add_16258: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_10276, model_audio_tower_layers_17_final_layer_norm_bias);  mul_10276 = model_audio_tower_layers_17_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_16258, [2])
	        full_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_106, full_213);  amax_106 = full_213 = None
	        amin_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_16258, [2])
	        full_212: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_106: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_106, full_212);  amin_106 = full_212 = None
	        sub_4860: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_106, minimum_106);  maximum_106 = None
	        div_212: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4860, 255.0);  sub_4860 = None
	        clamp_min_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_212, 1.1920928955078125e-07);  div_212 = None
	        div_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_106, clamp_min_318);  minimum_106 = None
	        round_213: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_213);  div_213 = None
	        sub_4866: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_213);  round_213 = None
	        clamp_min_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4866, -128);  sub_4866 = None
	        clamp_max_212: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_319, 127);  clamp_min_319 = None
	        view_1664: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_318, [sym_size_int, 1500, 1])
	        reciprocal_106: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1664);  view_1664 = None
	        mul_10324: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_106, 1.0);  reciprocal_106 = None
	        mul_10327: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_16258, mul_10324);  add_16258 = mul_10324 = None
	        round_214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10327);  mul_10327 = None
	        convert_element_type_636: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_212, torch.int8);  clamp_max_212 = None
	        view_1665: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_636, [sym_size_int, 1500, 1])
	        add_16345: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_214, view_1665);  round_214 = view_1665 = None
	        clamp_min_320: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16345, -128);  add_16345 = None
	        clamp_max_213: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_320, 127);  clamp_min_320 = None
	        convert_element_type_637: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_213, torch.int8);  clamp_max_213 = None
	        view_1669: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_636, [sym_size_int, 1500, 1]);  convert_element_type_636 = None
	        convert_element_type_638: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_637, torch.float32);  convert_element_type_637 = None
	        convert_element_type_639: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1669, torch.float32);  view_1669 = None
	        sub_4886: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_638, convert_element_type_639);  convert_element_type_638 = convert_element_type_639 = None
	        view_1668: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_318, [sym_size_int, 1500, 1]);  clamp_min_318 = None
	        mul_10349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4886, view_1668);  sub_4886 = view_1668 = None
	        view_1671: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_17_fc1_parametrizations_weight_original0 = None
	        view_1673: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_17_fc1_parametrizations_weight_original2 = None
	        convert_element_type_640: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1671, torch.float32);  view_1671 = None
	        convert_element_type_641: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1673, torch.float32);  view_1673 = None
	        sub_4890: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_640, convert_element_type_641);  convert_element_type_640 = convert_element_type_641 = None
	        view_1672: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_17_fc1_parametrizations_weight_original1 = None
	        mul_10354: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4890, view_1672);  sub_4890 = view_1672 = None
	        view_1674: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10354, [5120, 1280]);  mul_10354 = None
	        mul_10359: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1675: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10349, [mul_10359, 1280]);  mul_10349 = mul_10359 = None
	        permute_179: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1674, [1, 0]);  view_1674 = None
	        mm_default_71: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_1675, permute_179);  view_1675 = permute_179 = None
	        add_tensor_71: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_71, model_audio_tower_layers_17_fc1_bias);  mm_default_71 = model_audio_tower_layers_17_fc1_bias = None
	        view_1676: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_71, [sym_size_int, 1500, 5120]);  add_tensor_71 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_10366: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1676, 0.5)
	        mul_10367: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1676, 0.7071067811865476);  view_1676 = None
	        erf_19: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_10367);  mul_10367 = None
	        add_16404: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_19, 1);  erf_19 = None
	        mul_10368: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10366, add_16404);  mul_10366 = add_16404 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_10368, [2])
	        full_215: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_107, full_215);  amax_107 = full_215 = None
	        amin_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_10368, [2])
	        full_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_107, full_214);  amin_107 = full_214 = None
	        sub_4903: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_107, minimum_107);  maximum_107 = None
	        div_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4903, 255.0);  sub_4903 = None
	        clamp_min_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_214, 1.1920928955078125e-07);  div_214 = None
	        div_215: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_107, clamp_min_321);  minimum_107 = None
	        round_215: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_215);  div_215 = None
	        sub_4909: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_215);  round_215 = None
	        clamp_min_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4909, -128);  sub_4909 = None
	        clamp_max_214: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_322, 127);  clamp_min_322 = None
	        view_1679: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_321, [sym_size_int, 1500, 1])
	        reciprocal_107: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1679);  view_1679 = None
	        mul_10414: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_107, 1.0);  reciprocal_107 = None
	        mul_10417: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10368, mul_10414);  mul_10368 = mul_10414 = None
	        round_216: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_10417);  mul_10417 = None
	        convert_element_type_642: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_214, torch.int8);  clamp_max_214 = None
	        view_1680: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_642, [sym_size_int, 1500, 1])
	        add_16487: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_216, view_1680);  round_216 = view_1680 = None
	        clamp_min_323: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16487, -128);  add_16487 = None
	        clamp_max_215: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_323, 127);  clamp_min_323 = None
	        convert_element_type_643: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_215, torch.int8);  clamp_max_215 = None
	        view_1684: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_642, [sym_size_int, 1500, 1]);  convert_element_type_642 = None
	        convert_element_type_644: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_643, torch.float32);  convert_element_type_643 = None
	        convert_element_type_645: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1684, torch.float32);  view_1684 = None
	        sub_4929: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_644, convert_element_type_645);  convert_element_type_644 = convert_element_type_645 = None
	        view_1683: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_321, [sym_size_int, 1500, 1]);  clamp_min_321 = None
	        mul_10439: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4929, view_1683);  sub_4929 = view_1683 = None
	        view_1686: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_17_fc2_parametrizations_weight_original0 = None
	        view_1688: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_17_fc2_parametrizations_weight_original2 = None
	        convert_element_type_646: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1686, torch.float32);  view_1686 = None
	        convert_element_type_647: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1688, torch.float32);  view_1688 = None
	        sub_4933: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_646, convert_element_type_647);  convert_element_type_646 = convert_element_type_647 = None
	        view_1687: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_17_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_17_fc2_parametrizations_weight_original1 = None
	        mul_10444: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4933, view_1687);  sub_4933 = view_1687 = None
	        view_1689: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10444, [1280, 5120]);  mul_10444 = None
	        mul_10449: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1690: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10439, [mul_10449, 5120]);  mul_10439 = mul_10449 = None
	        permute_180: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1689, [1, 0]);  view_1689 = None
	        mm_default_70: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1690, permute_180);  view_1690 = permute_180 = None
	        add_tensor_70: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_70, model_audio_tower_layers_17_fc2_bias);  mm_default_70 = model_audio_tower_layers_17_fc2_bias = None
	        view_1691: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_70, [sym_size_int, 1500, 1280]);  add_tensor_70 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_16550: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_16252, view_1691);  add_16252 = view_1691 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_145: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_16550, memory_format = torch.contiguous_format)
	        var_mean_36 = torch.ops.aten.var_mean.correction(clone_145, [2], correction = 0, keepdim = True)
	        getitem_144: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_36[0]
	        getitem_145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_36[1];  var_mean_36 = None
	        sub_4939: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_145, getitem_145);  clone_145 = getitem_145 = None
	        add_16555: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_144, 1e-05);  getitem_144 = None
	        rsqrt_36: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_16555);  add_16555 = None
	        mul_10460: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4939, rsqrt_36);  sub_4939 = rsqrt_36 = None
	        mul_10461: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10460, model_audio_tower_layers_18_self_attn_layer_norm_weight);  mul_10460 = model_audio_tower_layers_18_self_attn_layer_norm_weight = None
	        add_16556: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_10461, model_audio_tower_layers_18_self_attn_layer_norm_bias);  mul_10461 = model_audio_tower_layers_18_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_16556, [2])
	        full_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_108, full_217);  amax_108 = full_217 = None
	        amin_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_16556, [2])
	        full_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_108: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_108, full_216);  amin_108 = full_216 = None
	        sub_4950: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_108, minimum_108);  maximum_108 = None
	        div_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4950, 255.0);  sub_4950 = None
	        clamp_min_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_216, 1.1920928955078125e-07);  div_216 = None
	        div_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_108, clamp_min_324);  minimum_108 = None
	        round_217: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_217);  div_217 = None
	        sub_4956: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_217);  round_217 = None
	        clamp_min_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_4956, -128);  sub_4956 = None
	        clamp_max_216: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_325, 127);  clamp_min_325 = None
	        view_1694: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_324, [sym_size_int, 1500, 1])
	        reciprocal_108: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1694);  view_1694 = None
	        mul_10509: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_108, 1.0);  reciprocal_108 = None
	        mul_10512: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_16556, mul_10509);  mul_10509 = None
	        round_218: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10512);  mul_10512 = None
	        convert_element_type_648: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_216, torch.int8);  clamp_max_216 = None
	        view_1695: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_648, [sym_size_int, 1500, 1])
	        add_16643: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_218, view_1695);  round_218 = view_1695 = None
	        clamp_min_326: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16643, -128);  add_16643 = None
	        clamp_max_217: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_326, 127);  clamp_min_326 = None
	        convert_element_type_649: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_217, torch.int8);  clamp_max_217 = None
	        view_1699: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_648, [sym_size_int, 1500, 1]);  convert_element_type_648 = None
	        convert_element_type_650: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_649, torch.float32);  convert_element_type_649 = None
	        convert_element_type_651: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1699, torch.float32);  view_1699 = None
	        sub_4976: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_650, convert_element_type_651);  convert_element_type_650 = convert_element_type_651 = None
	        view_1698: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_324, [sym_size_int, 1500, 1]);  clamp_min_324 = None
	        mul_10534: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4976, view_1698);  sub_4976 = view_1698 = None
	        view_1701: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1703: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_652: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1701, torch.float32);  view_1701 = None
	        convert_element_type_653: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1703, torch.float32);  view_1703 = None
	        sub_4980: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_652, convert_element_type_653);  convert_element_type_652 = convert_element_type_653 = None
	        view_1702: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_10539: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_4980, view_1702);  sub_4980 = view_1702 = None
	        view_1704: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10539, [1280, 1280]);  mul_10539 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_16556, [2])
	        full_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_109, full_219);  amax_109 = full_219 = None
	        amin_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_16556, [2])
	        full_218: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_109: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_109, full_218);  amin_109 = full_218 = None
	        sub_4995: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_109, minimum_109);  maximum_109 = None
	        div_218: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_4995, 255.0);  sub_4995 = None
	        clamp_min_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_218, 1.1920928955078125e-07);  div_218 = None
	        div_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_109, clamp_min_327);  minimum_109 = None
	        round_219: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_219);  div_219 = None
	        sub_5001: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_219);  round_219 = None
	        clamp_min_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5001, -128);  sub_5001 = None
	        clamp_max_218: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_328, 127);  clamp_min_328 = None
	        view_1710: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_327, [sym_size_int, 1500, 1])
	        reciprocal_109: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1710);  view_1710 = None
	        mul_10605: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_109, 1.0);  reciprocal_109 = None
	        mul_10608: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_16556, mul_10605);  mul_10605 = None
	        round_220: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10608);  mul_10608 = None
	        convert_element_type_654: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_218, torch.int8);  clamp_max_218 = None
	        view_1711: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_654, [sym_size_int, 1500, 1])
	        add_16795: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_220, view_1711);  round_220 = view_1711 = None
	        clamp_min_329: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16795, -128);  add_16795 = None
	        clamp_max_219: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_329, 127);  clamp_min_329 = None
	        convert_element_type_655: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_219, torch.int8);  clamp_max_219 = None
	        view_1715: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_654, [sym_size_int, 1500, 1]);  convert_element_type_654 = None
	        convert_element_type_656: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_655, torch.float32);  convert_element_type_655 = None
	        convert_element_type_657: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1715, torch.float32);  view_1715 = None
	        sub_5021: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_656, convert_element_type_657);  convert_element_type_656 = convert_element_type_657 = None
	        view_1714: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_327, [sym_size_int, 1500, 1]);  clamp_min_327 = None
	        mul_10630: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5021, view_1714);  sub_5021 = view_1714 = None
	        view_1717: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1719: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_658: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1717, torch.float32);  view_1717 = None
	        convert_element_type_659: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1719, torch.float32);  view_1719 = None
	        sub_5025: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_658, convert_element_type_659);  convert_element_type_658 = convert_element_type_659 = None
	        view_1718: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_10635: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5025, view_1718);  sub_5025 = view_1718 = None
	        view_1720: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10635, [1280, 1280]);  mul_10635 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_16556, [2])
	        full_221: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_110, full_221);  amax_110 = full_221 = None
	        amin_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_16556, [2])
	        full_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_110: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_110, full_220);  amin_110 = full_220 = None
	        sub_5039: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_110, minimum_110);  maximum_110 = None
	        div_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5039, 255.0);  sub_5039 = None
	        clamp_min_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_220, 1.1920928955078125e-07);  div_220 = None
	        div_221: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_110, clamp_min_330);  minimum_110 = None
	        round_221: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_221);  div_221 = None
	        sub_5045: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_221);  round_221 = None
	        clamp_min_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5045, -128);  sub_5045 = None
	        clamp_max_220: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_331, 127);  clamp_min_331 = None
	        view_1726: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_330, [sym_size_int, 1500, 1])
	        reciprocal_110: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1726);  view_1726 = None
	        mul_10704: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_110, 1.0);  reciprocal_110 = None
	        mul_10707: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_16556, mul_10704);  add_16556 = mul_10704 = None
	        round_222: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10707);  mul_10707 = None
	        convert_element_type_660: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_220, torch.int8);  clamp_max_220 = None
	        view_1727: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_660, [sym_size_int, 1500, 1])
	        add_16943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_222, view_1727);  round_222 = view_1727 = None
	        clamp_min_332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_16943, -128);  add_16943 = None
	        clamp_max_221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_332, 127);  clamp_min_332 = None
	        convert_element_type_661: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_221, torch.int8);  clamp_max_221 = None
	        view_1731: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_660, [sym_size_int, 1500, 1]);  convert_element_type_660 = None
	        convert_element_type_662: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_661, torch.float32);  convert_element_type_661 = None
	        convert_element_type_663: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1731, torch.float32);  view_1731 = None
	        sub_5065: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_662, convert_element_type_663);  convert_element_type_662 = convert_element_type_663 = None
	        view_1730: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_330, [sym_size_int, 1500, 1]);  clamp_min_330 = None
	        mul_10729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5065, view_1730);  sub_5065 = view_1730 = None
	        view_1733: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1735: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_664: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1733, torch.float32);  view_1733 = None
	        convert_element_type_665: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1735, torch.float32);  view_1735 = None
	        sub_5069: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_664, convert_element_type_665);  convert_element_type_664 = convert_element_type_665 = None
	        view_1734: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_10734: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5069, view_1734);  sub_5069 = view_1734 = None
	        view_1736: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10734, [1280, 1280]);  mul_10734 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_10544: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1705: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10534, [mul_10544, 1280]);  mul_10534 = mul_10544 = None
	        permute_181: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1704, [1, 0]);  view_1704 = None
	        mm_default_69: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1705, permute_181);  view_1705 = permute_181 = None
	        add_tensor_69: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_69, model_audio_tower_layers_18_self_attn_q_proj_bias);  mm_default_69 = model_audio_tower_layers_18_self_attn_q_proj_bias = None
	        view_1706: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_69, [sym_size_int, 1500, 1280]);  add_tensor_69 = None
	        mul_10551: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1706, 0.125);  view_1706 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1707: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10551, [sym_size_int, 1500, 20, 64]);  mul_10551 = None
	        permute_182: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1707, [0, 2, 1, 3]);  view_1707 = None
	        clone_146: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_182, memory_format = torch.contiguous_format);  permute_182 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_10638: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1721: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10630, [mul_10638, 1280]);  mul_10630 = mul_10638 = None
	        permute_183: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1720, [1, 0]);  view_1720 = None
	        mm_18: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1721, permute_183);  view_1721 = permute_183 = None
	        view_1722: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_18, [sym_size_int, 1500, 1280]);  mm_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1723: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1722, [sym_size_int, -1, 20, 64]);  view_1722 = None
	        permute_184: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1723, [0, 2, 1, 3]);  view_1723 = None
	        clone_147: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_184, memory_format = torch.contiguous_format);  permute_184 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_10739: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1737: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10729, [mul_10739, 1280]);  mul_10729 = mul_10739 = None
	        permute_185: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1736, [1, 0]);  view_1736 = None
	        mm_default_68: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1737, permute_185);  view_1737 = permute_185 = None
	        add_tensor_68: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_68, model_audio_tower_layers_18_self_attn_v_proj_bias);  mm_default_68 = model_audio_tower_layers_18_self_attn_v_proj_bias = None
	        view_1738: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_68, [sym_size_int, 1500, 1280]);  add_tensor_68 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1739: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1738, [sym_size_int, -1, 20, 64]);  view_1738 = None
	        permute_186: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1739, [0, 2, 1, 3]);  view_1739 = None
	        clone_148: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_186, memory_format = torch.contiguous_format);  permute_186 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_18 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_146, clone_147, clone_148, None, False, scale = 1.0);  clone_146 = clone_147 = clone_148 = None
	        getitem_146: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_18[0];  _scaled_dot_product_efficient_attention_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_187: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_146, [0, 2, 1, 3]);  getitem_146 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1740: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_187, [sym_size_int, 1500, -1]);  permute_187 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1740, [2])
	        full_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_111, full_223);  amax_111 = full_223 = None
	        amin_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1740, [2])
	        full_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_111: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_111, full_222);  amin_111 = full_222 = None
	        sub_5087: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_111, minimum_111);  maximum_111 = None
	        div_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5087, 255.0);  sub_5087 = None
	        clamp_min_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_222, 1.1920928955078125e-07);  div_222 = None
	        div_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_111, clamp_min_333);  minimum_111 = None
	        round_223: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_223);  div_223 = None
	        sub_5093: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_223);  round_223 = None
	        clamp_min_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5093, -128);  sub_5093 = None
	        clamp_max_222: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_334, 127);  clamp_min_334 = None
	        view_1743: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_333, [sym_size_int, 1500, 1])
	        reciprocal_111: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1743);  view_1743 = None
	        mul_10809: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_111, 1.0);  reciprocal_111 = None
	        mul_10812: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1740, mul_10809);  view_1740 = mul_10809 = None
	        round_224: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10812);  mul_10812 = None
	        convert_element_type_666: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_222, torch.int8);  clamp_max_222 = None
	        view_1744: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_666, [sym_size_int, 1500, 1])
	        add_17107: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_224, view_1744);  round_224 = view_1744 = None
	        clamp_min_335: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17107, -128);  add_17107 = None
	        clamp_max_223: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_335, 127);  clamp_min_335 = None
	        convert_element_type_667: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_223, torch.int8);  clamp_max_223 = None
	        view_1748: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_666, [sym_size_int, 1500, 1]);  convert_element_type_666 = None
	        convert_element_type_668: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_667, torch.float32);  convert_element_type_667 = None
	        convert_element_type_669: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1748, torch.float32);  view_1748 = None
	        sub_5113: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_668, convert_element_type_669);  convert_element_type_668 = convert_element_type_669 = None
	        view_1747: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_333, [sym_size_int, 1500, 1]);  clamp_min_333 = None
	        mul_10834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5113, view_1747);  sub_5113 = view_1747 = None
	        view_1750: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1752: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_670: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1750, torch.float32);  view_1750 = None
	        convert_element_type_671: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1752, torch.float32);  view_1752 = None
	        sub_5117: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_670, convert_element_type_671);  convert_element_type_670 = convert_element_type_671 = None
	        view_1751: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_10839: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5117, view_1751);  sub_5117 = view_1751 = None
	        view_1753: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10839, [1280, 1280]);  mul_10839 = None
	        mul_10844: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1754: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10834, [mul_10844, 1280]);  mul_10834 = mul_10844 = None
	        permute_188: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1753, [1, 0]);  view_1753 = None
	        mm_default_67: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1754, permute_188);  view_1754 = permute_188 = None
	        add_tensor_67: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_67, model_audio_tower_layers_18_self_attn_out_proj_bias);  mm_default_67 = model_audio_tower_layers_18_self_attn_out_proj_bias = None
	        view_1755: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_67, [sym_size_int, 1500, 1280]);  add_tensor_67 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_17170: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_16550, view_1755);  add_16550 = view_1755 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_150: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_17170, memory_format = torch.contiguous_format)
	        var_mean_37 = torch.ops.aten.var_mean.correction(clone_150, [2], correction = 0, keepdim = True)
	        getitem_150: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_37[0]
	        getitem_151: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_37[1];  var_mean_37 = None
	        sub_5123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_150, getitem_151);  clone_150 = getitem_151 = None
	        add_17175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_150, 1e-05);  getitem_150 = None
	        rsqrt_37: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_17175);  add_17175 = None
	        mul_10855: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5123, rsqrt_37);  sub_5123 = rsqrt_37 = None
	        mul_10856: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10855, model_audio_tower_layers_18_final_layer_norm_weight);  mul_10855 = model_audio_tower_layers_18_final_layer_norm_weight = None
	        add_17176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_10856, model_audio_tower_layers_18_final_layer_norm_bias);  mul_10856 = model_audio_tower_layers_18_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_17176, [2])
	        full_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_112, full_225);  amax_112 = full_225 = None
	        amin_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_17176, [2])
	        full_224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_112: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_112, full_224);  amin_112 = full_224 = None
	        sub_5134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_112, minimum_112);  maximum_112 = None
	        div_224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5134, 255.0);  sub_5134 = None
	        clamp_min_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_224, 1.1920928955078125e-07);  div_224 = None
	        div_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_112, clamp_min_336);  minimum_112 = None
	        round_225: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_225);  div_225 = None
	        sub_5140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_225);  round_225 = None
	        clamp_min_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5140, -128);  sub_5140 = None
	        clamp_max_224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_337, 127);  clamp_min_337 = None
	        view_1758: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_336, [sym_size_int, 1500, 1])
	        reciprocal_112: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1758);  view_1758 = None
	        mul_10904: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_112, 1.0);  reciprocal_112 = None
	        mul_10907: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_17176, mul_10904);  add_17176 = mul_10904 = None
	        round_226: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_10907);  mul_10907 = None
	        convert_element_type_672: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_224, torch.int8);  clamp_max_224 = None
	        view_1759: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_672, [sym_size_int, 1500, 1])
	        add_17263: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_226, view_1759);  round_226 = view_1759 = None
	        clamp_min_338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17263, -128);  add_17263 = None
	        clamp_max_225: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_338, 127);  clamp_min_338 = None
	        convert_element_type_673: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_225, torch.int8);  clamp_max_225 = None
	        view_1763: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_672, [sym_size_int, 1500, 1]);  convert_element_type_672 = None
	        convert_element_type_674: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_673, torch.float32);  convert_element_type_673 = None
	        convert_element_type_675: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1763, torch.float32);  view_1763 = None
	        sub_5160: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_674, convert_element_type_675);  convert_element_type_674 = convert_element_type_675 = None
	        view_1762: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_336, [sym_size_int, 1500, 1]);  clamp_min_336 = None
	        mul_10929: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5160, view_1762);  sub_5160 = view_1762 = None
	        view_1765: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_18_fc1_parametrizations_weight_original0 = None
	        view_1767: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_18_fc1_parametrizations_weight_original2 = None
	        convert_element_type_676: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1765, torch.float32);  view_1765 = None
	        convert_element_type_677: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1767, torch.float32);  view_1767 = None
	        sub_5164: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_676, convert_element_type_677);  convert_element_type_676 = convert_element_type_677 = None
	        view_1766: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_18_fc1_parametrizations_weight_original1 = None
	        mul_10934: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5164, view_1766);  sub_5164 = view_1766 = None
	        view_1768: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10934, [5120, 1280]);  mul_10934 = None
	        mul_10939: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1769: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_10929, [mul_10939, 1280]);  mul_10929 = mul_10939 = None
	        permute_189: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1768, [1, 0]);  view_1768 = None
	        mm_default_66: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_1769, permute_189);  view_1769 = permute_189 = None
	        add_tensor_66: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_66, model_audio_tower_layers_18_fc1_bias);  mm_default_66 = model_audio_tower_layers_18_fc1_bias = None
	        view_1770: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_66, [sym_size_int, 1500, 5120]);  add_tensor_66 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_10946: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1770, 0.5)
	        mul_10947: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1770, 0.7071067811865476);  view_1770 = None
	        erf_20: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_10947);  mul_10947 = None
	        add_17322: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_20, 1);  erf_20 = None
	        mul_10948: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10946, add_17322);  mul_10946 = add_17322 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_10948, [2])
	        full_227: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_113, full_227);  amax_113 = full_227 = None
	        amin_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_10948, [2])
	        full_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_113: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_113, full_226);  amin_113 = full_226 = None
	        sub_5177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_113, minimum_113);  maximum_113 = None
	        div_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5177, 255.0);  sub_5177 = None
	        clamp_min_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_226, 1.1920928955078125e-07);  div_226 = None
	        div_227: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_113, clamp_min_339);  minimum_113 = None
	        round_227: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_227);  div_227 = None
	        sub_5183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_227);  round_227 = None
	        clamp_min_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5183, -128);  sub_5183 = None
	        clamp_max_226: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_340, 127);  clamp_min_340 = None
	        view_1773: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_339, [sym_size_int, 1500, 1])
	        reciprocal_113: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1773);  view_1773 = None
	        mul_10994: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_113, 1.0);  reciprocal_113 = None
	        mul_10997: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_10948, mul_10994);  mul_10948 = mul_10994 = None
	        round_228: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_10997);  mul_10997 = None
	        convert_element_type_678: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_226, torch.int8);  clamp_max_226 = None
	        view_1774: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_678, [sym_size_int, 1500, 1])
	        add_17405: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_228, view_1774);  round_228 = view_1774 = None
	        clamp_min_341: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17405, -128);  add_17405 = None
	        clamp_max_227: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_341, 127);  clamp_min_341 = None
	        convert_element_type_679: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_227, torch.int8);  clamp_max_227 = None
	        view_1778: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_678, [sym_size_int, 1500, 1]);  convert_element_type_678 = None
	        convert_element_type_680: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_679, torch.float32);  convert_element_type_679 = None
	        convert_element_type_681: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1778, torch.float32);  view_1778 = None
	        sub_5203: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_680, convert_element_type_681);  convert_element_type_680 = convert_element_type_681 = None
	        view_1777: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_339, [sym_size_int, 1500, 1]);  clamp_min_339 = None
	        mul_11019: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5203, view_1777);  sub_5203 = view_1777 = None
	        view_1780: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_18_fc2_parametrizations_weight_original0 = None
	        view_1782: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_18_fc2_parametrizations_weight_original2 = None
	        convert_element_type_682: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1780, torch.float32);  view_1780 = None
	        convert_element_type_683: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1782, torch.float32);  view_1782 = None
	        sub_5207: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_682, convert_element_type_683);  convert_element_type_682 = convert_element_type_683 = None
	        view_1781: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_18_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_18_fc2_parametrizations_weight_original1 = None
	        mul_11024: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5207, view_1781);  sub_5207 = view_1781 = None
	        view_1783: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11024, [1280, 5120]);  mul_11024 = None
	        mul_11029: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1784: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11019, [mul_11029, 5120]);  mul_11019 = mul_11029 = None
	        permute_190: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1783, [1, 0]);  view_1783 = None
	        mm_default_65: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1784, permute_190);  view_1784 = permute_190 = None
	        add_tensor_65: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_65, model_audio_tower_layers_18_fc2_bias);  mm_default_65 = model_audio_tower_layers_18_fc2_bias = None
	        view_1785: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_65, [sym_size_int, 1500, 1280]);  add_tensor_65 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_17468: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_17170, view_1785);  add_17170 = view_1785 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_153: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_17468, memory_format = torch.contiguous_format)
	        var_mean_38 = torch.ops.aten.var_mean.correction(clone_153, [2], correction = 0, keepdim = True)
	        getitem_152: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_38[0]
	        getitem_153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_38[1];  var_mean_38 = None
	        sub_5213: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_153, getitem_153);  clone_153 = getitem_153 = None
	        add_17473: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_152, 1e-05);  getitem_152 = None
	        rsqrt_38: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_17473);  add_17473 = None
	        mul_11040: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5213, rsqrt_38);  sub_5213 = rsqrt_38 = None
	        mul_11041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11040, model_audio_tower_layers_19_self_attn_layer_norm_weight);  mul_11040 = model_audio_tower_layers_19_self_attn_layer_norm_weight = None
	        add_17474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_11041, model_audio_tower_layers_19_self_attn_layer_norm_bias);  mul_11041 = model_audio_tower_layers_19_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_17474, [2])
	        full_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_114, full_229);  amax_114 = full_229 = None
	        amin_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_17474, [2])
	        full_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_114: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_114, full_228);  amin_114 = full_228 = None
	        sub_5224: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_114, minimum_114);  maximum_114 = None
	        div_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5224, 255.0);  sub_5224 = None
	        clamp_min_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_228, 1.1920928955078125e-07);  div_228 = None
	        div_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_114, clamp_min_342);  minimum_114 = None
	        round_229: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_229);  div_229 = None
	        sub_5230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_229);  round_229 = None
	        clamp_min_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5230, -128);  sub_5230 = None
	        clamp_max_228: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_343, 127);  clamp_min_343 = None
	        view_1788: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_342, [sym_size_int, 1500, 1])
	        reciprocal_114: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1788);  view_1788 = None
	        mul_11089: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_114, 1.0);  reciprocal_114 = None
	        mul_11092: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_17474, mul_11089);  mul_11089 = None
	        round_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11092);  mul_11092 = None
	        convert_element_type_684: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_228, torch.int8);  clamp_max_228 = None
	        view_1789: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_684, [sym_size_int, 1500, 1])
	        add_17561: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_230, view_1789);  round_230 = view_1789 = None
	        clamp_min_344: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17561, -128);  add_17561 = None
	        clamp_max_229: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_344, 127);  clamp_min_344 = None
	        convert_element_type_685: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_229, torch.int8);  clamp_max_229 = None
	        view_1793: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_684, [sym_size_int, 1500, 1]);  convert_element_type_684 = None
	        convert_element_type_686: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_685, torch.float32);  convert_element_type_685 = None
	        convert_element_type_687: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1793, torch.float32);  view_1793 = None
	        sub_5250: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_686, convert_element_type_687);  convert_element_type_686 = convert_element_type_687 = None
	        view_1792: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_342, [sym_size_int, 1500, 1]);  clamp_min_342 = None
	        mul_11114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5250, view_1792);  sub_5250 = view_1792 = None
	        view_1795: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1797: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_688: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1795, torch.float32);  view_1795 = None
	        convert_element_type_689: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1797, torch.float32);  view_1797 = None
	        sub_5254: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_688, convert_element_type_689);  convert_element_type_688 = convert_element_type_689 = None
	        view_1796: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_11119: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5254, view_1796);  sub_5254 = view_1796 = None
	        view_1798: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11119, [1280, 1280]);  mul_11119 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_17474, [2])
	        full_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_115, full_231);  amax_115 = full_231 = None
	        amin_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_17474, [2])
	        full_230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_115: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_115, full_230);  amin_115 = full_230 = None
	        sub_5269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_115, minimum_115);  maximum_115 = None
	        div_230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5269, 255.0);  sub_5269 = None
	        clamp_min_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_230, 1.1920928955078125e-07);  div_230 = None
	        div_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_115, clamp_min_345);  minimum_115 = None
	        round_231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_231);  div_231 = None
	        sub_5275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_231);  round_231 = None
	        clamp_min_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5275, -128);  sub_5275 = None
	        clamp_max_230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_346, 127);  clamp_min_346 = None
	        view_1804: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_345, [sym_size_int, 1500, 1])
	        reciprocal_115: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1804);  view_1804 = None
	        mul_11185: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_115, 1.0);  reciprocal_115 = None
	        mul_11188: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_17474, mul_11185);  mul_11185 = None
	        round_232: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11188);  mul_11188 = None
	        convert_element_type_690: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_230, torch.int8);  clamp_max_230 = None
	        view_1805: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_690, [sym_size_int, 1500, 1])
	        add_17713: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_232, view_1805);  round_232 = view_1805 = None
	        clamp_min_347: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17713, -128);  add_17713 = None
	        clamp_max_231: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_347, 127);  clamp_min_347 = None
	        convert_element_type_691: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_231, torch.int8);  clamp_max_231 = None
	        view_1809: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_690, [sym_size_int, 1500, 1]);  convert_element_type_690 = None
	        convert_element_type_692: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_691, torch.float32);  convert_element_type_691 = None
	        convert_element_type_693: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1809, torch.float32);  view_1809 = None
	        sub_5295: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_692, convert_element_type_693);  convert_element_type_692 = convert_element_type_693 = None
	        view_1808: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_345, [sym_size_int, 1500, 1]);  clamp_min_345 = None
	        mul_11210: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5295, view_1808);  sub_5295 = view_1808 = None
	        view_1811: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1813: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_694: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1811, torch.float32);  view_1811 = None
	        convert_element_type_695: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1813, torch.float32);  view_1813 = None
	        sub_5299: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_694, convert_element_type_695);  convert_element_type_694 = convert_element_type_695 = None
	        view_1812: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_11215: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5299, view_1812);  sub_5299 = view_1812 = None
	        view_1814: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11215, [1280, 1280]);  mul_11215 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_17474, [2])
	        full_233: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_116, full_233);  amax_116 = full_233 = None
	        amin_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_17474, [2])
	        full_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_116: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_116, full_232);  amin_116 = full_232 = None
	        sub_5313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_116, minimum_116);  maximum_116 = None
	        div_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5313, 255.0);  sub_5313 = None
	        clamp_min_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_232, 1.1920928955078125e-07);  div_232 = None
	        div_233: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_116, clamp_min_348);  minimum_116 = None
	        round_233: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_233);  div_233 = None
	        sub_5319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_233);  round_233 = None
	        clamp_min_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5319, -128);  sub_5319 = None
	        clamp_max_232: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_349, 127);  clamp_min_349 = None
	        view_1820: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_348, [sym_size_int, 1500, 1])
	        reciprocal_116: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1820);  view_1820 = None
	        mul_11284: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_116, 1.0);  reciprocal_116 = None
	        mul_11287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_17474, mul_11284);  add_17474 = mul_11284 = None
	        round_234: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11287);  mul_11287 = None
	        convert_element_type_696: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_232, torch.int8);  clamp_max_232 = None
	        view_1821: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_696, [sym_size_int, 1500, 1])
	        add_17861: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_234, view_1821);  round_234 = view_1821 = None
	        clamp_min_350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_17861, -128);  add_17861 = None
	        clamp_max_233: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_350, 127);  clamp_min_350 = None
	        convert_element_type_697: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_233, torch.int8);  clamp_max_233 = None
	        view_1825: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_696, [sym_size_int, 1500, 1]);  convert_element_type_696 = None
	        convert_element_type_698: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_697, torch.float32);  convert_element_type_697 = None
	        convert_element_type_699: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1825, torch.float32);  view_1825 = None
	        sub_5339: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_698, convert_element_type_699);  convert_element_type_698 = convert_element_type_699 = None
	        view_1824: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_348, [sym_size_int, 1500, 1]);  clamp_min_348 = None
	        mul_11309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5339, view_1824);  sub_5339 = view_1824 = None
	        view_1827: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1829: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_700: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1827, torch.float32);  view_1827 = None
	        convert_element_type_701: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1829, torch.float32);  view_1829 = None
	        sub_5343: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_700, convert_element_type_701);  convert_element_type_700 = convert_element_type_701 = None
	        view_1828: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_11314: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5343, view_1828);  sub_5343 = view_1828 = None
	        view_1830: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11314, [1280, 1280]);  mul_11314 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_11124: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1799: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11114, [mul_11124, 1280]);  mul_11114 = mul_11124 = None
	        permute_191: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1798, [1, 0]);  view_1798 = None
	        mm_default_64: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1799, permute_191);  view_1799 = permute_191 = None
	        add_tensor_64: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_64, model_audio_tower_layers_19_self_attn_q_proj_bias);  mm_default_64 = model_audio_tower_layers_19_self_attn_q_proj_bias = None
	        view_1800: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_64, [sym_size_int, 1500, 1280]);  add_tensor_64 = None
	        mul_11131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1800, 0.125);  view_1800 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1801: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11131, [sym_size_int, 1500, 20, 64]);  mul_11131 = None
	        permute_192: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1801, [0, 2, 1, 3]);  view_1801 = None
	        clone_154: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_192, memory_format = torch.contiguous_format);  permute_192 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_11218: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1815: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11210, [mul_11218, 1280]);  mul_11210 = mul_11218 = None
	        permute_193: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1814, [1, 0]);  view_1814 = None
	        mm_19: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1815, permute_193);  view_1815 = permute_193 = None
	        view_1816: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_19, [sym_size_int, 1500, 1280]);  mm_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1817: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1816, [sym_size_int, -1, 20, 64]);  view_1816 = None
	        permute_194: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1817, [0, 2, 1, 3]);  view_1817 = None
	        clone_155: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_194, memory_format = torch.contiguous_format);  permute_194 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_11319: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1831: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11309, [mul_11319, 1280]);  mul_11309 = mul_11319 = None
	        permute_195: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1830, [1, 0]);  view_1830 = None
	        mm_default_63: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1831, permute_195);  view_1831 = permute_195 = None
	        add_tensor_63: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_63, model_audio_tower_layers_19_self_attn_v_proj_bias);  mm_default_63 = model_audio_tower_layers_19_self_attn_v_proj_bias = None
	        view_1832: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_63, [sym_size_int, 1500, 1280]);  add_tensor_63 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1833: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1832, [sym_size_int, -1, 20, 64]);  view_1832 = None
	        permute_196: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1833, [0, 2, 1, 3]);  view_1833 = None
	        clone_156: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_196, memory_format = torch.contiguous_format);  permute_196 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_19 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_154, clone_155, clone_156, None, False, scale = 1.0);  clone_154 = clone_155 = clone_156 = None
	        getitem_154: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_19[0];  _scaled_dot_product_efficient_attention_19 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_197: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_154, [0, 2, 1, 3]);  getitem_154 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_197, [sym_size_int, 1500, -1]);  permute_197 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1834, [2])
	        full_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_117, full_235);  amax_117 = full_235 = None
	        amin_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1834, [2])
	        full_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_117: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_117, full_234);  amin_117 = full_234 = None
	        sub_5361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_117, minimum_117);  maximum_117 = None
	        div_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5361, 255.0);  sub_5361 = None
	        clamp_min_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_234, 1.1920928955078125e-07);  div_234 = None
	        div_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_117, clamp_min_351);  minimum_117 = None
	        round_235: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_235);  div_235 = None
	        sub_5367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_235);  round_235 = None
	        clamp_min_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5367, -128);  sub_5367 = None
	        clamp_max_234: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_352, 127);  clamp_min_352 = None
	        view_1837: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_351, [sym_size_int, 1500, 1])
	        reciprocal_117: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1837);  view_1837 = None
	        mul_11389: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_117, 1.0);  reciprocal_117 = None
	        mul_11392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1834, mul_11389);  view_1834 = mul_11389 = None
	        round_236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11392);  mul_11392 = None
	        convert_element_type_702: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_234, torch.int8);  clamp_max_234 = None
	        view_1838: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_702, [sym_size_int, 1500, 1])
	        add_18025: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_236, view_1838);  round_236 = view_1838 = None
	        clamp_min_353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18025, -128);  add_18025 = None
	        clamp_max_235: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_353, 127);  clamp_min_353 = None
	        convert_element_type_703: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_235, torch.int8);  clamp_max_235 = None
	        view_1842: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_702, [sym_size_int, 1500, 1]);  convert_element_type_702 = None
	        convert_element_type_704: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_703, torch.float32);  convert_element_type_703 = None
	        convert_element_type_705: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1842, torch.float32);  view_1842 = None
	        sub_5387: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_704, convert_element_type_705);  convert_element_type_704 = convert_element_type_705 = None
	        view_1841: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_351, [sym_size_int, 1500, 1]);  clamp_min_351 = None
	        mul_11414: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5387, view_1841);  sub_5387 = view_1841 = None
	        view_1844: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1846: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_706: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1844, torch.float32);  view_1844 = None
	        convert_element_type_707: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1846, torch.float32);  view_1846 = None
	        sub_5391: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_706, convert_element_type_707);  convert_element_type_706 = convert_element_type_707 = None
	        view_1845: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_11419: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5391, view_1845);  sub_5391 = view_1845 = None
	        view_1847: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11419, [1280, 1280]);  mul_11419 = None
	        mul_11424: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1848: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11414, [mul_11424, 1280]);  mul_11414 = mul_11424 = None
	        permute_198: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1847, [1, 0]);  view_1847 = None
	        mm_default_62: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1848, permute_198);  view_1848 = permute_198 = None
	        add_tensor_62: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_62, model_audio_tower_layers_19_self_attn_out_proj_bias);  mm_default_62 = model_audio_tower_layers_19_self_attn_out_proj_bias = None
	        view_1849: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_62, [sym_size_int, 1500, 1280]);  add_tensor_62 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_18088: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_17468, view_1849);  add_17468 = view_1849 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_158: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_18088, memory_format = torch.contiguous_format)
	        var_mean_39 = torch.ops.aten.var_mean.correction(clone_158, [2], correction = 0, keepdim = True)
	        getitem_158: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_39[0]
	        getitem_159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_39[1];  var_mean_39 = None
	        sub_5397: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_158, getitem_159);  clone_158 = getitem_159 = None
	        add_18093: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_158, 1e-05);  getitem_158 = None
	        rsqrt_39: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_18093);  add_18093 = None
	        mul_11435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5397, rsqrt_39);  sub_5397 = rsqrt_39 = None
	        mul_11436: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11435, model_audio_tower_layers_19_final_layer_norm_weight);  mul_11435 = model_audio_tower_layers_19_final_layer_norm_weight = None
	        add_18094: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_11436, model_audio_tower_layers_19_final_layer_norm_bias);  mul_11436 = model_audio_tower_layers_19_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_18094, [2])
	        full_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_118, full_237);  amax_118 = full_237 = None
	        amin_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_18094, [2])
	        full_236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_118: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_118, full_236);  amin_118 = full_236 = None
	        sub_5408: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_118, minimum_118);  maximum_118 = None
	        div_236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5408, 255.0);  sub_5408 = None
	        clamp_min_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_236, 1.1920928955078125e-07);  div_236 = None
	        div_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_118, clamp_min_354);  minimum_118 = None
	        round_237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_237);  div_237 = None
	        sub_5414: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_237);  round_237 = None
	        clamp_min_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5414, -128);  sub_5414 = None
	        clamp_max_236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_355, 127);  clamp_min_355 = None
	        view_1852: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_354, [sym_size_int, 1500, 1])
	        reciprocal_118: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1852);  view_1852 = None
	        mul_11484: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_118, 1.0);  reciprocal_118 = None
	        mul_11487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_18094, mul_11484);  add_18094 = mul_11484 = None
	        round_238: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11487);  mul_11487 = None
	        convert_element_type_708: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_236, torch.int8);  clamp_max_236 = None
	        view_1853: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_708, [sym_size_int, 1500, 1])
	        add_18181: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_238, view_1853);  round_238 = view_1853 = None
	        clamp_min_356: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18181, -128);  add_18181 = None
	        clamp_max_237: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_356, 127);  clamp_min_356 = None
	        convert_element_type_709: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_237, torch.int8);  clamp_max_237 = None
	        view_1857: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_708, [sym_size_int, 1500, 1]);  convert_element_type_708 = None
	        convert_element_type_710: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_709, torch.float32);  convert_element_type_709 = None
	        convert_element_type_711: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1857, torch.float32);  view_1857 = None
	        sub_5434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_710, convert_element_type_711);  convert_element_type_710 = convert_element_type_711 = None
	        view_1856: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_354, [sym_size_int, 1500, 1]);  clamp_min_354 = None
	        mul_11509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5434, view_1856);  sub_5434 = view_1856 = None
	        view_1859: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_19_fc1_parametrizations_weight_original0 = None
	        view_1861: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_19_fc1_parametrizations_weight_original2 = None
	        convert_element_type_712: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1859, torch.float32);  view_1859 = None
	        convert_element_type_713: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1861, torch.float32);  view_1861 = None
	        sub_5438: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_712, convert_element_type_713);  convert_element_type_712 = convert_element_type_713 = None
	        view_1860: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_19_fc1_parametrizations_weight_original1 = None
	        mul_11514: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5438, view_1860);  sub_5438 = view_1860 = None
	        view_1862: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11514, [5120, 1280]);  mul_11514 = None
	        mul_11519: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1863: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11509, [mul_11519, 1280]);  mul_11509 = mul_11519 = None
	        permute_199: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1862, [1, 0]);  view_1862 = None
	        mm_default_61: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_1863, permute_199);  view_1863 = permute_199 = None
	        add_tensor_61: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_61, model_audio_tower_layers_19_fc1_bias);  mm_default_61 = model_audio_tower_layers_19_fc1_bias = None
	        view_1864: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_61, [sym_size_int, 1500, 5120]);  add_tensor_61 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_11526: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1864, 0.5)
	        mul_11527: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1864, 0.7071067811865476);  view_1864 = None
	        erf_21: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_11527);  mul_11527 = None
	        add_18240: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_21, 1);  erf_21 = None
	        mul_11528: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11526, add_18240);  mul_11526 = add_18240 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_11528, [2])
	        full_239: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_119, full_239);  amax_119 = full_239 = None
	        amin_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_11528, [2])
	        full_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_119: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_119, full_238);  amin_119 = full_238 = None
	        sub_5451: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_119, minimum_119);  maximum_119 = None
	        div_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5451, 255.0);  sub_5451 = None
	        clamp_min_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_238, 1.1920928955078125e-07);  div_238 = None
	        div_239: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_119, clamp_min_357);  minimum_119 = None
	        round_239: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_239);  div_239 = None
	        sub_5457: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_239);  round_239 = None
	        clamp_min_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5457, -128);  sub_5457 = None
	        clamp_max_238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_358, 127);  clamp_min_358 = None
	        view_1867: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_357, [sym_size_int, 1500, 1])
	        reciprocal_119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1867);  view_1867 = None
	        mul_11574: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_119, 1.0);  reciprocal_119 = None
	        mul_11577: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11528, mul_11574);  mul_11528 = mul_11574 = None
	        round_240: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_11577);  mul_11577 = None
	        convert_element_type_714: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_238, torch.int8);  clamp_max_238 = None
	        view_1868: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_714, [sym_size_int, 1500, 1])
	        add_18323: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_240, view_1868);  round_240 = view_1868 = None
	        clamp_min_359: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18323, -128);  add_18323 = None
	        clamp_max_239: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_359, 127);  clamp_min_359 = None
	        convert_element_type_715: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_239, torch.int8);  clamp_max_239 = None
	        view_1872: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_714, [sym_size_int, 1500, 1]);  convert_element_type_714 = None
	        convert_element_type_716: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_715, torch.float32);  convert_element_type_715 = None
	        convert_element_type_717: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1872, torch.float32);  view_1872 = None
	        sub_5477: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_716, convert_element_type_717);  convert_element_type_716 = convert_element_type_717 = None
	        view_1871: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_357, [sym_size_int, 1500, 1]);  clamp_min_357 = None
	        mul_11599: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5477, view_1871);  sub_5477 = view_1871 = None
	        view_1874: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_19_fc2_parametrizations_weight_original0 = None
	        view_1876: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_19_fc2_parametrizations_weight_original2 = None
	        convert_element_type_718: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1874, torch.float32);  view_1874 = None
	        convert_element_type_719: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1876, torch.float32);  view_1876 = None
	        sub_5481: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_718, convert_element_type_719);  convert_element_type_718 = convert_element_type_719 = None
	        view_1875: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_19_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_19_fc2_parametrizations_weight_original1 = None
	        mul_11604: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5481, view_1875);  sub_5481 = view_1875 = None
	        view_1877: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11604, [1280, 5120]);  mul_11604 = None
	        mul_11609: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1878: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11599, [mul_11609, 5120]);  mul_11599 = mul_11609 = None
	        permute_200: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1877, [1, 0]);  view_1877 = None
	        mm_default_60: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1878, permute_200);  view_1878 = permute_200 = None
	        add_tensor_60: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_60, model_audio_tower_layers_19_fc2_bias);  mm_default_60 = model_audio_tower_layers_19_fc2_bias = None
	        view_1879: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_60, [sym_size_int, 1500, 1280]);  add_tensor_60 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_18386: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_18088, view_1879);  add_18088 = view_1879 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_18386, memory_format = torch.contiguous_format)
	        var_mean_40 = torch.ops.aten.var_mean.correction(clone_161, [2], correction = 0, keepdim = True)
	        getitem_160: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_40[0]
	        getitem_161: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_40[1];  var_mean_40 = None
	        sub_5487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_161, getitem_161);  clone_161 = getitem_161 = None
	        add_18391: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_160, 1e-05);  getitem_160 = None
	        rsqrt_40: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_18391);  add_18391 = None
	        mul_11620: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5487, rsqrt_40);  sub_5487 = rsqrt_40 = None
	        mul_11621: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_11620, model_audio_tower_layers_20_self_attn_layer_norm_weight);  mul_11620 = model_audio_tower_layers_20_self_attn_layer_norm_weight = None
	        add_18392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_11621, model_audio_tower_layers_20_self_attn_layer_norm_bias);  mul_11621 = model_audio_tower_layers_20_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_18392, [2])
	        full_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_120, full_241);  amax_120 = full_241 = None
	        amin_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_18392, [2])
	        full_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_120: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_120, full_240);  amin_120 = full_240 = None
	        sub_5498: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_120, minimum_120);  maximum_120 = None
	        div_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5498, 255.0);  sub_5498 = None
	        clamp_min_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_240, 1.1920928955078125e-07);  div_240 = None
	        div_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_120, clamp_min_360);  minimum_120 = None
	        round_241: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_241);  div_241 = None
	        sub_5504: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_241);  round_241 = None
	        clamp_min_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5504, -128);  sub_5504 = None
	        clamp_max_240: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_361, 127);  clamp_min_361 = None
	        view_1882: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_360, [sym_size_int, 1500, 1])
	        reciprocal_120: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1882);  view_1882 = None
	        mul_11669: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_120, 1.0);  reciprocal_120 = None
	        mul_11672: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_18392, mul_11669);  mul_11669 = None
	        round_242: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11672);  mul_11672 = None
	        convert_element_type_720: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_240, torch.int8);  clamp_max_240 = None
	        view_1883: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_720, [sym_size_int, 1500, 1])
	        add_18479: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_242, view_1883);  round_242 = view_1883 = None
	        clamp_min_362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18479, -128);  add_18479 = None
	        clamp_max_241: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_362, 127);  clamp_min_362 = None
	        convert_element_type_721: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_241, torch.int8);  clamp_max_241 = None
	        view_1887: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_720, [sym_size_int, 1500, 1]);  convert_element_type_720 = None
	        convert_element_type_722: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_721, torch.float32);  convert_element_type_721 = None
	        convert_element_type_723: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1887, torch.float32);  view_1887 = None
	        sub_5524: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_722, convert_element_type_723);  convert_element_type_722 = convert_element_type_723 = None
	        view_1886: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_360, [sym_size_int, 1500, 1]);  clamp_min_360 = None
	        mul_11694: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5524, view_1886);  sub_5524 = view_1886 = None
	        view_1889: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1891: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_724: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1889, torch.float32);  view_1889 = None
	        convert_element_type_725: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1891, torch.float32);  view_1891 = None
	        sub_5528: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_724, convert_element_type_725);  convert_element_type_724 = convert_element_type_725 = None
	        view_1890: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_11699: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5528, view_1890);  sub_5528 = view_1890 = None
	        view_1892: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11699, [1280, 1280]);  mul_11699 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_18392, [2])
	        full_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_121, full_243);  amax_121 = full_243 = None
	        amin_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_18392, [2])
	        full_242: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_121: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_121, full_242);  amin_121 = full_242 = None
	        sub_5543: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_121, minimum_121);  maximum_121 = None
	        div_242: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5543, 255.0);  sub_5543 = None
	        clamp_min_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_242, 1.1920928955078125e-07);  div_242 = None
	        div_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_121, clamp_min_363);  minimum_121 = None
	        round_243: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_243);  div_243 = None
	        sub_5549: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_243);  round_243 = None
	        clamp_min_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5549, -128);  sub_5549 = None
	        clamp_max_242: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_364, 127);  clamp_min_364 = None
	        view_1898: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_363, [sym_size_int, 1500, 1])
	        reciprocal_121: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1898);  view_1898 = None
	        mul_11765: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_121, 1.0);  reciprocal_121 = None
	        mul_11768: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_18392, mul_11765);  mul_11765 = None
	        round_244: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11768);  mul_11768 = None
	        convert_element_type_726: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_242, torch.int8);  clamp_max_242 = None
	        view_1899: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_726, [sym_size_int, 1500, 1])
	        add_18631: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_244, view_1899);  round_244 = view_1899 = None
	        clamp_min_365: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18631, -128);  add_18631 = None
	        clamp_max_243: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_365, 127);  clamp_min_365 = None
	        convert_element_type_727: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_243, torch.int8);  clamp_max_243 = None
	        view_1903: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_726, [sym_size_int, 1500, 1]);  convert_element_type_726 = None
	        convert_element_type_728: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_727, torch.float32);  convert_element_type_727 = None
	        convert_element_type_729: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1903, torch.float32);  view_1903 = None
	        sub_5569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_728, convert_element_type_729);  convert_element_type_728 = convert_element_type_729 = None
	        view_1902: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_363, [sym_size_int, 1500, 1]);  clamp_min_363 = None
	        mul_11790: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5569, view_1902);  sub_5569 = view_1902 = None
	        view_1905: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_1907: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_730: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1905, torch.float32);  view_1905 = None
	        convert_element_type_731: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1907, torch.float32);  view_1907 = None
	        sub_5573: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_730, convert_element_type_731);  convert_element_type_730 = convert_element_type_731 = None
	        view_1906: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_11795: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5573, view_1906);  sub_5573 = view_1906 = None
	        view_1908: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11795, [1280, 1280]);  mul_11795 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_18392, [2])
	        full_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_122, full_245);  amax_122 = full_245 = None
	        amin_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_18392, [2])
	        full_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_122: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_122, full_244);  amin_122 = full_244 = None
	        sub_5587: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_122, minimum_122);  maximum_122 = None
	        div_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5587, 255.0);  sub_5587 = None
	        clamp_min_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_244, 1.1920928955078125e-07);  div_244 = None
	        div_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_122, clamp_min_366);  minimum_122 = None
	        round_245: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_245);  div_245 = None
	        sub_5593: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_245);  round_245 = None
	        clamp_min_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5593, -128);  sub_5593 = None
	        clamp_max_244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_367, 127);  clamp_min_367 = None
	        view_1914: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_366, [sym_size_int, 1500, 1])
	        reciprocal_122: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1914);  view_1914 = None
	        mul_11864: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_122, 1.0);  reciprocal_122 = None
	        mul_11867: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_18392, mul_11864);  add_18392 = mul_11864 = None
	        round_246: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11867);  mul_11867 = None
	        convert_element_type_732: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_244, torch.int8);  clamp_max_244 = None
	        view_1915: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_732, [sym_size_int, 1500, 1])
	        add_18779: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_246, view_1915);  round_246 = view_1915 = None
	        clamp_min_368: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18779, -128);  add_18779 = None
	        clamp_max_245: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_368, 127);  clamp_min_368 = None
	        convert_element_type_733: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_245, torch.int8);  clamp_max_245 = None
	        view_1919: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_732, [sym_size_int, 1500, 1]);  convert_element_type_732 = None
	        convert_element_type_734: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_733, torch.float32);  convert_element_type_733 = None
	        convert_element_type_735: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1919, torch.float32);  view_1919 = None
	        sub_5613: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_734, convert_element_type_735);  convert_element_type_734 = convert_element_type_735 = None
	        view_1918: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_366, [sym_size_int, 1500, 1]);  clamp_min_366 = None
	        mul_11889: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5613, view_1918);  sub_5613 = view_1918 = None
	        view_1921: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_1923: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_736: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1921, torch.float32);  view_1921 = None
	        convert_element_type_737: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1923, torch.float32);  view_1923 = None
	        sub_5617: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_736, convert_element_type_737);  convert_element_type_736 = convert_element_type_737 = None
	        view_1922: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_11894: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5617, view_1922);  sub_5617 = view_1922 = None
	        view_1924: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11894, [1280, 1280]);  mul_11894 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_11704: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1893: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11694, [mul_11704, 1280]);  mul_11694 = mul_11704 = None
	        permute_201: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1892, [1, 0]);  view_1892 = None
	        mm_default_59: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1893, permute_201);  view_1893 = permute_201 = None
	        add_tensor_59: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_59, model_audio_tower_layers_20_self_attn_q_proj_bias);  mm_default_59 = model_audio_tower_layers_20_self_attn_q_proj_bias = None
	        view_1894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_59, [sym_size_int, 1500, 1280]);  add_tensor_59 = None
	        mul_11711: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1894, 0.125);  view_1894 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1895: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11711, [sym_size_int, 1500, 20, 64]);  mul_11711 = None
	        permute_202: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1895, [0, 2, 1, 3]);  view_1895 = None
	        clone_162: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_202, memory_format = torch.contiguous_format);  permute_202 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_11798: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1909: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11790, [mul_11798, 1280]);  mul_11790 = mul_11798 = None
	        permute_203: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1908, [1, 0]);  view_1908 = None
	        mm_20: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1909, permute_203);  view_1909 = permute_203 = None
	        view_1910: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_20, [sym_size_int, 1500, 1280]);  mm_20 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1911: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1910, [sym_size_int, -1, 20, 64]);  view_1910 = None
	        permute_204: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1911, [0, 2, 1, 3]);  view_1911 = None
	        clone_163: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_204, memory_format = torch.contiguous_format);  permute_204 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_11899: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1925: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11889, [mul_11899, 1280]);  mul_11889 = mul_11899 = None
	        permute_205: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1924, [1, 0]);  view_1924 = None
	        mm_default_58: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1925, permute_205);  view_1925 = permute_205 = None
	        add_tensor_58: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_58, model_audio_tower_layers_20_self_attn_v_proj_bias);  mm_default_58 = model_audio_tower_layers_20_self_attn_v_proj_bias = None
	        view_1926: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_58, [sym_size_int, 1500, 1280]);  add_tensor_58 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1927: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_1926, [sym_size_int, -1, 20, 64]);  view_1926 = None
	        permute_206: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1927, [0, 2, 1, 3]);  view_1927 = None
	        clone_164: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_206, memory_format = torch.contiguous_format);  permute_206 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_20 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_162, clone_163, clone_164, None, False, scale = 1.0);  clone_162 = clone_163 = clone_164 = None
	        getitem_162: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_20[0];  _scaled_dot_product_efficient_attention_20 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_207: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_162, [0, 2, 1, 3]);  getitem_162 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_1928: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_207, [sym_size_int, 1500, -1]);  permute_207 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_1928, [2])
	        full_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_123, full_247);  amax_123 = full_247 = None
	        amin_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_1928, [2])
	        full_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_123: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_123, full_246);  amin_123 = full_246 = None
	        sub_5635: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_123, minimum_123);  maximum_123 = None
	        div_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5635, 255.0);  sub_5635 = None
	        clamp_min_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_246, 1.1920928955078125e-07);  div_246 = None
	        div_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_123, clamp_min_369);  minimum_123 = None
	        round_247: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_247);  div_247 = None
	        sub_5641: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_247);  round_247 = None
	        clamp_min_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5641, -128);  sub_5641 = None
	        clamp_max_246: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_370, 127);  clamp_min_370 = None
	        view_1931: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_369, [sym_size_int, 1500, 1])
	        reciprocal_123: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1931);  view_1931 = None
	        mul_11969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_123, 1.0);  reciprocal_123 = None
	        mul_11972: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1928, mul_11969);  view_1928 = mul_11969 = None
	        round_248: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_11972);  mul_11972 = None
	        convert_element_type_738: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_246, torch.int8);  clamp_max_246 = None
	        view_1932: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_738, [sym_size_int, 1500, 1])
	        add_18943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_248, view_1932);  round_248 = view_1932 = None
	        clamp_min_371: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_18943, -128);  add_18943 = None
	        clamp_max_247: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_371, 127);  clamp_min_371 = None
	        convert_element_type_739: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_247, torch.int8);  clamp_max_247 = None
	        view_1936: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_738, [sym_size_int, 1500, 1]);  convert_element_type_738 = None
	        convert_element_type_740: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_739, torch.float32);  convert_element_type_739 = None
	        convert_element_type_741: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1936, torch.float32);  view_1936 = None
	        sub_5661: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_740, convert_element_type_741);  convert_element_type_740 = convert_element_type_741 = None
	        view_1935: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_369, [sym_size_int, 1500, 1]);  clamp_min_369 = None
	        mul_11994: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5661, view_1935);  sub_5661 = view_1935 = None
	        view_1938: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_1940: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_742: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1938, torch.float32);  view_1938 = None
	        convert_element_type_743: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1940, torch.float32);  view_1940 = None
	        sub_5665: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_742, convert_element_type_743);  convert_element_type_742 = convert_element_type_743 = None
	        view_1939: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_11999: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5665, view_1939);  sub_5665 = view_1939 = None
	        view_1941: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11999, [1280, 1280]);  mul_11999 = None
	        mul_12004: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1942: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_11994, [mul_12004, 1280]);  mul_11994 = mul_12004 = None
	        permute_208: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1941, [1, 0]);  view_1941 = None
	        mm_default_57: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1942, permute_208);  view_1942 = permute_208 = None
	        add_tensor_57: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_57, model_audio_tower_layers_20_self_attn_out_proj_bias);  mm_default_57 = model_audio_tower_layers_20_self_attn_out_proj_bias = None
	        view_1943: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_57, [sym_size_int, 1500, 1280]);  add_tensor_57 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_19006: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_18386, view_1943);  add_18386 = view_1943 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_166: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_19006, memory_format = torch.contiguous_format)
	        var_mean_41 = torch.ops.aten.var_mean.correction(clone_166, [2], correction = 0, keepdim = True)
	        getitem_166: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_41[0]
	        getitem_167: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_41[1];  var_mean_41 = None
	        sub_5671: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_166, getitem_167);  clone_166 = getitem_167 = None
	        add_19011: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_166, 1e-05);  getitem_166 = None
	        rsqrt_41: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_19011);  add_19011 = None
	        mul_12015: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5671, rsqrt_41);  sub_5671 = rsqrt_41 = None
	        mul_12016: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12015, model_audio_tower_layers_20_final_layer_norm_weight);  mul_12015 = model_audio_tower_layers_20_final_layer_norm_weight = None
	        add_19012: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12016, model_audio_tower_layers_20_final_layer_norm_bias);  mul_12016 = model_audio_tower_layers_20_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_19012, [2])
	        full_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_124, full_249);  amax_124 = full_249 = None
	        amin_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_19012, [2])
	        full_248: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_124: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_124, full_248);  amin_124 = full_248 = None
	        sub_5682: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_124, minimum_124);  maximum_124 = None
	        div_248: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5682, 255.0);  sub_5682 = None
	        clamp_min_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_248, 1.1920928955078125e-07);  div_248 = None
	        div_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_124, clamp_min_372);  minimum_124 = None
	        round_249: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_249);  div_249 = None
	        sub_5688: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_249);  round_249 = None
	        clamp_min_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5688, -128);  sub_5688 = None
	        clamp_max_248: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_373, 127);  clamp_min_373 = None
	        view_1946: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_372, [sym_size_int, 1500, 1])
	        reciprocal_124: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1946);  view_1946 = None
	        mul_12064: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_124, 1.0);  reciprocal_124 = None
	        mul_12067: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_19012, mul_12064);  add_19012 = mul_12064 = None
	        round_250: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12067);  mul_12067 = None
	        convert_element_type_744: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_248, torch.int8);  clamp_max_248 = None
	        view_1947: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_744, [sym_size_int, 1500, 1])
	        add_19099: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_250, view_1947);  round_250 = view_1947 = None
	        clamp_min_374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19099, -128);  add_19099 = None
	        clamp_max_249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_374, 127);  clamp_min_374 = None
	        convert_element_type_745: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_249, torch.int8);  clamp_max_249 = None
	        view_1951: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_744, [sym_size_int, 1500, 1]);  convert_element_type_744 = None
	        convert_element_type_746: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_745, torch.float32);  convert_element_type_745 = None
	        convert_element_type_747: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1951, torch.float32);  view_1951 = None
	        sub_5708: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_746, convert_element_type_747);  convert_element_type_746 = convert_element_type_747 = None
	        view_1950: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_372, [sym_size_int, 1500, 1]);  clamp_min_372 = None
	        mul_12089: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5708, view_1950);  sub_5708 = view_1950 = None
	        view_1953: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_20_fc1_parametrizations_weight_original0 = None
	        view_1955: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_20_fc1_parametrizations_weight_original2 = None
	        convert_element_type_748: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1953, torch.float32);  view_1953 = None
	        convert_element_type_749: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1955, torch.float32);  view_1955 = None
	        sub_5712: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_748, convert_element_type_749);  convert_element_type_748 = convert_element_type_749 = None
	        view_1954: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_20_fc1_parametrizations_weight_original1 = None
	        mul_12094: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5712, view_1954);  sub_5712 = view_1954 = None
	        view_1956: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12094, [5120, 1280]);  mul_12094 = None
	        mul_12099: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1957: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12089, [mul_12099, 1280]);  mul_12089 = mul_12099 = None
	        permute_209: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1956, [1, 0]);  view_1956 = None
	        mm_default_56: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_1957, permute_209);  view_1957 = permute_209 = None
	        add_tensor_56: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_56, model_audio_tower_layers_20_fc1_bias);  mm_default_56 = model_audio_tower_layers_20_fc1_bias = None
	        view_1958: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_56, [sym_size_int, 1500, 5120]);  add_tensor_56 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_12106: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1958, 0.5)
	        mul_12107: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1958, 0.7071067811865476);  view_1958 = None
	        erf_22: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_12107);  mul_12107 = None
	        add_19158: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_22, 1);  erf_22 = None
	        mul_12108: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12106, add_19158);  mul_12106 = add_19158 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_12108, [2])
	        full_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_125, full_251);  amax_125 = full_251 = None
	        amin_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_12108, [2])
	        full_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_125: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_125, full_250);  amin_125 = full_250 = None
	        sub_5725: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_125, minimum_125);  maximum_125 = None
	        div_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5725, 255.0);  sub_5725 = None
	        clamp_min_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_250, 1.1920928955078125e-07);  div_250 = None
	        div_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_125, clamp_min_375);  minimum_125 = None
	        round_251: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_251);  div_251 = None
	        sub_5731: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_251);  round_251 = None
	        clamp_min_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5731, -128);  sub_5731 = None
	        clamp_max_250: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_376, 127);  clamp_min_376 = None
	        view_1961: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_375, [sym_size_int, 1500, 1])
	        reciprocal_125: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1961);  view_1961 = None
	        mul_12154: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_125, 1.0);  reciprocal_125 = None
	        mul_12157: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12108, mul_12154);  mul_12108 = mul_12154 = None
	        round_252: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_12157);  mul_12157 = None
	        convert_element_type_750: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_250, torch.int8);  clamp_max_250 = None
	        view_1962: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_750, [sym_size_int, 1500, 1])
	        add_19241: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_252, view_1962);  round_252 = view_1962 = None
	        clamp_min_377: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19241, -128);  add_19241 = None
	        clamp_max_251: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_377, 127);  clamp_min_377 = None
	        convert_element_type_751: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_251, torch.int8);  clamp_max_251 = None
	        view_1966: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_750, [sym_size_int, 1500, 1]);  convert_element_type_750 = None
	        convert_element_type_752: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_751, torch.float32);  convert_element_type_751 = None
	        convert_element_type_753: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1966, torch.float32);  view_1966 = None
	        sub_5751: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_752, convert_element_type_753);  convert_element_type_752 = convert_element_type_753 = None
	        view_1965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_375, [sym_size_int, 1500, 1]);  clamp_min_375 = None
	        mul_12179: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5751, view_1965);  sub_5751 = view_1965 = None
	        view_1968: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_20_fc2_parametrizations_weight_original0 = None
	        view_1970: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_20_fc2_parametrizations_weight_original2 = None
	        convert_element_type_754: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1968, torch.float32);  view_1968 = None
	        convert_element_type_755: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1970, torch.float32);  view_1970 = None
	        sub_5755: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_754, convert_element_type_755);  convert_element_type_754 = convert_element_type_755 = None
	        view_1969: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_20_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_20_fc2_parametrizations_weight_original1 = None
	        mul_12184: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5755, view_1969);  sub_5755 = view_1969 = None
	        view_1971: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12184, [1280, 5120]);  mul_12184 = None
	        mul_12189: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1972: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12179, [mul_12189, 5120]);  mul_12179 = mul_12189 = None
	        permute_210: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_1971, [1, 0]);  view_1971 = None
	        mm_default_55: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1972, permute_210);  view_1972 = permute_210 = None
	        add_tensor_55: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_55, model_audio_tower_layers_20_fc2_bias);  mm_default_55 = model_audio_tower_layers_20_fc2_bias = None
	        view_1973: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_55, [sym_size_int, 1500, 1280]);  add_tensor_55 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_19304: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_19006, view_1973);  add_19006 = view_1973 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_169: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_19304, memory_format = torch.contiguous_format)
	        var_mean_42 = torch.ops.aten.var_mean.correction(clone_169, [2], correction = 0, keepdim = True)
	        getitem_168: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_42[0]
	        getitem_169: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_42[1];  var_mean_42 = None
	        sub_5761: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_169, getitem_169);  clone_169 = getitem_169 = None
	        add_19309: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_168, 1e-05);  getitem_168 = None
	        rsqrt_42: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_19309);  add_19309 = None
	        mul_12200: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5761, rsqrt_42);  sub_5761 = rsqrt_42 = None
	        mul_12201: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12200, model_audio_tower_layers_21_self_attn_layer_norm_weight);  mul_12200 = model_audio_tower_layers_21_self_attn_layer_norm_weight = None
	        add_19310: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12201, model_audio_tower_layers_21_self_attn_layer_norm_bias);  mul_12201 = model_audio_tower_layers_21_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_19310, [2])
	        full_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_126, full_253);  amax_126 = full_253 = None
	        amin_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_19310, [2])
	        full_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_126: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_126, full_252);  amin_126 = full_252 = None
	        sub_5772: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_126, minimum_126);  maximum_126 = None
	        div_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5772, 255.0);  sub_5772 = None
	        clamp_min_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_252, 1.1920928955078125e-07);  div_252 = None
	        div_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_126, clamp_min_378);  minimum_126 = None
	        round_253: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_253);  div_253 = None
	        sub_5778: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_253);  round_253 = None
	        clamp_min_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5778, -128);  sub_5778 = None
	        clamp_max_252: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_379, 127);  clamp_min_379 = None
	        view_1976: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_378, [sym_size_int, 1500, 1])
	        reciprocal_126: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1976);  view_1976 = None
	        mul_12249: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_126, 1.0);  reciprocal_126 = None
	        mul_12252: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_19310, mul_12249);  mul_12249 = None
	        round_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12252);  mul_12252 = None
	        convert_element_type_756: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_252, torch.int8);  clamp_max_252 = None
	        view_1977: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_756, [sym_size_int, 1500, 1])
	        add_19397: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_254, view_1977);  round_254 = view_1977 = None
	        clamp_min_380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19397, -128);  add_19397 = None
	        clamp_max_253: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_380, 127);  clamp_min_380 = None
	        convert_element_type_757: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_253, torch.int8);  clamp_max_253 = None
	        view_1981: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_756, [sym_size_int, 1500, 1]);  convert_element_type_756 = None
	        convert_element_type_758: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_757, torch.float32);  convert_element_type_757 = None
	        convert_element_type_759: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1981, torch.float32);  view_1981 = None
	        sub_5798: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_758, convert_element_type_759);  convert_element_type_758 = convert_element_type_759 = None
	        view_1980: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_378, [sym_size_int, 1500, 1]);  clamp_min_378 = None
	        mul_12274: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5798, view_1980);  sub_5798 = view_1980 = None
	        view_1983: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_1985: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_760: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1983, torch.float32);  view_1983 = None
	        convert_element_type_761: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1985, torch.float32);  view_1985 = None
	        sub_5802: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_760, convert_element_type_761);  convert_element_type_760 = convert_element_type_761 = None
	        view_1984: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_12279: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5802, view_1984);  sub_5802 = view_1984 = None
	        view_1986: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12279, [1280, 1280]);  mul_12279 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_19310, [2])
	        full_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_127, full_255);  amax_127 = full_255 = None
	        amin_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_19310, [2])
	        full_254: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_127: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_127, full_254);  amin_127 = full_254 = None
	        sub_5817: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_127, minimum_127);  maximum_127 = None
	        div_254: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5817, 255.0);  sub_5817 = None
	        clamp_min_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_254, 1.1920928955078125e-07);  div_254 = None
	        div_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_127, clamp_min_381);  minimum_127 = None
	        round_255: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_255);  div_255 = None
	        sub_5823: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_255);  round_255 = None
	        clamp_min_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5823, -128);  sub_5823 = None
	        clamp_max_254: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_382, 127);  clamp_min_382 = None
	        view_1992: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_381, [sym_size_int, 1500, 1])
	        reciprocal_127: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_1992);  view_1992 = None
	        mul_12345: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_127, 1.0);  reciprocal_127 = None
	        mul_12348: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_19310, mul_12345);  mul_12345 = None
	        round_256: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12348);  mul_12348 = None
	        convert_element_type_762: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_254, torch.int8);  clamp_max_254 = None
	        view_1993: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_762, [sym_size_int, 1500, 1])
	        add_19549: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_256, view_1993);  round_256 = view_1993 = None
	        clamp_min_383: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19549, -128);  add_19549 = None
	        clamp_max_255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_383, 127);  clamp_min_383 = None
	        convert_element_type_763: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_255, torch.int8);  clamp_max_255 = None
	        view_1997: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_762, [sym_size_int, 1500, 1]);  convert_element_type_762 = None
	        convert_element_type_764: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_763, torch.float32);  convert_element_type_763 = None
	        convert_element_type_765: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1997, torch.float32);  view_1997 = None
	        sub_5843: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_764, convert_element_type_765);  convert_element_type_764 = convert_element_type_765 = None
	        view_1996: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_381, [sym_size_int, 1500, 1]);  clamp_min_381 = None
	        mul_12370: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5843, view_1996);  sub_5843 = view_1996 = None
	        view_1999: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2001: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_766: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_1999, torch.float32);  view_1999 = None
	        convert_element_type_767: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2001, torch.float32);  view_2001 = None
	        sub_5847: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_766, convert_element_type_767);  convert_element_type_766 = convert_element_type_767 = None
	        view_2000: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_12375: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5847, view_2000);  sub_5847 = view_2000 = None
	        view_2002: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12375, [1280, 1280]);  mul_12375 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_19310, [2])
	        full_257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_128, full_257);  amax_128 = full_257 = None
	        amin_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_19310, [2])
	        full_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_128: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_128, full_256);  amin_128 = full_256 = None
	        sub_5861: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_128, minimum_128);  maximum_128 = None
	        div_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5861, 255.0);  sub_5861 = None
	        clamp_min_384: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_256, 1.1920928955078125e-07);  div_256 = None
	        div_257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_128, clamp_min_384);  minimum_128 = None
	        round_257: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_257);  div_257 = None
	        sub_5867: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_257);  round_257 = None
	        clamp_min_385: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5867, -128);  sub_5867 = None
	        clamp_max_256: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_385, 127);  clamp_min_385 = None
	        view_2008: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_384, [sym_size_int, 1500, 1])
	        reciprocal_128: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2008);  view_2008 = None
	        mul_12444: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_128, 1.0);  reciprocal_128 = None
	        mul_12447: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_19310, mul_12444);  add_19310 = mul_12444 = None
	        round_258: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12447);  mul_12447 = None
	        convert_element_type_768: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_256, torch.int8);  clamp_max_256 = None
	        view_2009: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_768, [sym_size_int, 1500, 1])
	        add_19697: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_258, view_2009);  round_258 = view_2009 = None
	        clamp_min_386: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19697, -128);  add_19697 = None
	        clamp_max_257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_386, 127);  clamp_min_386 = None
	        convert_element_type_769: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_257, torch.int8);  clamp_max_257 = None
	        view_2013: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_768, [sym_size_int, 1500, 1]);  convert_element_type_768 = None
	        convert_element_type_770: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_769, torch.float32);  convert_element_type_769 = None
	        convert_element_type_771: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2013, torch.float32);  view_2013 = None
	        sub_5887: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_770, convert_element_type_771);  convert_element_type_770 = convert_element_type_771 = None
	        view_2012: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_384, [sym_size_int, 1500, 1]);  clamp_min_384 = None
	        mul_12469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5887, view_2012);  sub_5887 = view_2012 = None
	        view_2015: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2017: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_772: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2015, torch.float32);  view_2015 = None
	        convert_element_type_773: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2017, torch.float32);  view_2017 = None
	        sub_5891: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_772, convert_element_type_773);  convert_element_type_772 = convert_element_type_773 = None
	        view_2016: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_12474: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5891, view_2016);  sub_5891 = view_2016 = None
	        view_2018: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12474, [1280, 1280]);  mul_12474 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_12284: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_1987: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12274, [mul_12284, 1280]);  mul_12274 = mul_12284 = None
	        permute_211: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_1986, [1, 0]);  view_1986 = None
	        mm_default_54: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_1987, permute_211);  view_1987 = permute_211 = None
	        add_tensor_54: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_54, model_audio_tower_layers_21_self_attn_q_proj_bias);  mm_default_54 = model_audio_tower_layers_21_self_attn_q_proj_bias = None
	        view_1988: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_54, [sym_size_int, 1500, 1280]);  add_tensor_54 = None
	        mul_12291: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_1988, 0.125);  view_1988 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_1989: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12291, [sym_size_int, 1500, 20, 64]);  mul_12291 = None
	        permute_212: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_1989, [0, 2, 1, 3]);  view_1989 = None
	        clone_170: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_212, memory_format = torch.contiguous_format);  permute_212 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_12378: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2003: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12370, [mul_12378, 1280]);  mul_12370 = mul_12378 = None
	        permute_213: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2002, [1, 0]);  view_2002 = None
	        mm_21: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2003, permute_213);  view_2003 = permute_213 = None
	        view_2004: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_21, [sym_size_int, 1500, 1280]);  mm_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2005: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2004, [sym_size_int, -1, 20, 64]);  view_2004 = None
	        permute_214: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2005, [0, 2, 1, 3]);  view_2005 = None
	        clone_171: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_214, memory_format = torch.contiguous_format);  permute_214 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_12479: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2019: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12469, [mul_12479, 1280]);  mul_12469 = mul_12479 = None
	        permute_215: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2018, [1, 0]);  view_2018 = None
	        mm_default_53: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2019, permute_215);  view_2019 = permute_215 = None
	        add_tensor_53: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_53, model_audio_tower_layers_21_self_attn_v_proj_bias);  mm_default_53 = model_audio_tower_layers_21_self_attn_v_proj_bias = None
	        view_2020: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_53, [sym_size_int, 1500, 1280]);  add_tensor_53 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2021: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2020, [sym_size_int, -1, 20, 64]);  view_2020 = None
	        permute_216: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2021, [0, 2, 1, 3]);  view_2021 = None
	        clone_172: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_216, memory_format = torch.contiguous_format);  permute_216 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_21 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_170, clone_171, clone_172, None, False, scale = 1.0);  clone_170 = clone_171 = clone_172 = None
	        getitem_170: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_21[0];  _scaled_dot_product_efficient_attention_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_217: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_170, [0, 2, 1, 3]);  getitem_170 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2022: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_217, [sym_size_int, 1500, -1]);  permute_217 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2022, [2])
	        full_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_129, full_259);  amax_129 = full_259 = None
	        amin_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2022, [2])
	        full_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_129: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_129, full_258);  amin_129 = full_258 = None
	        sub_5909: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_129, minimum_129);  maximum_129 = None
	        div_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5909, 255.0);  sub_5909 = None
	        clamp_min_387: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_258, 1.1920928955078125e-07);  div_258 = None
	        div_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_129, clamp_min_387);  minimum_129 = None
	        round_259: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_259);  div_259 = None
	        sub_5915: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_259);  round_259 = None
	        clamp_min_388: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5915, -128);  sub_5915 = None
	        clamp_max_258: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_388, 127);  clamp_min_388 = None
	        view_2025: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_387, [sym_size_int, 1500, 1])
	        reciprocal_129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2025);  view_2025 = None
	        mul_12549: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_129, 1.0);  reciprocal_129 = None
	        mul_12552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2022, mul_12549);  view_2022 = mul_12549 = None
	        round_260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12552);  mul_12552 = None
	        convert_element_type_774: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_258, torch.int8);  clamp_max_258 = None
	        view_2026: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_774, [sym_size_int, 1500, 1])
	        add_19861: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_260, view_2026);  round_260 = view_2026 = None
	        clamp_min_389: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_19861, -128);  add_19861 = None
	        clamp_max_259: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_389, 127);  clamp_min_389 = None
	        convert_element_type_775: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_259, torch.int8);  clamp_max_259 = None
	        view_2030: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_774, [sym_size_int, 1500, 1]);  convert_element_type_774 = None
	        convert_element_type_776: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_775, torch.float32);  convert_element_type_775 = None
	        convert_element_type_777: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2030, torch.float32);  view_2030 = None
	        sub_5935: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_776, convert_element_type_777);  convert_element_type_776 = convert_element_type_777 = None
	        view_2029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_387, [sym_size_int, 1500, 1]);  clamp_min_387 = None
	        mul_12574: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5935, view_2029);  sub_5935 = view_2029 = None
	        view_2032: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2034: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_778: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2032, torch.float32);  view_2032 = None
	        convert_element_type_779: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2034, torch.float32);  view_2034 = None
	        sub_5939: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_778, convert_element_type_779);  convert_element_type_778 = convert_element_type_779 = None
	        view_2033: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_12579: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5939, view_2033);  sub_5939 = view_2033 = None
	        view_2035: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12579, [1280, 1280]);  mul_12579 = None
	        mul_12584: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2036: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12574, [mul_12584, 1280]);  mul_12574 = mul_12584 = None
	        permute_218: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2035, [1, 0]);  view_2035 = None
	        mm_default_52: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2036, permute_218);  view_2036 = permute_218 = None
	        add_tensor_52: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_52, model_audio_tower_layers_21_self_attn_out_proj_bias);  mm_default_52 = model_audio_tower_layers_21_self_attn_out_proj_bias = None
	        view_2037: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_52, [sym_size_int, 1500, 1280]);  add_tensor_52 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_19924: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_19304, view_2037);  add_19304 = view_2037 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_19924, memory_format = torch.contiguous_format)
	        var_mean_43 = torch.ops.aten.var_mean.correction(clone_174, [2], correction = 0, keepdim = True)
	        getitem_174: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_43[0]
	        getitem_175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_43[1];  var_mean_43 = None
	        sub_5945: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_174, getitem_175);  clone_174 = getitem_175 = None
	        add_19929: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_174, 1e-05);  getitem_174 = None
	        rsqrt_43: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_19929);  add_19929 = None
	        mul_12595: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5945, rsqrt_43);  sub_5945 = rsqrt_43 = None
	        mul_12596: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12595, model_audio_tower_layers_21_final_layer_norm_weight);  mul_12595 = model_audio_tower_layers_21_final_layer_norm_weight = None
	        add_19930: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12596, model_audio_tower_layers_21_final_layer_norm_bias);  mul_12596 = model_audio_tower_layers_21_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_19930, [2])
	        full_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_130, full_261);  amax_130 = full_261 = None
	        amin_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_19930, [2])
	        full_260: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_130: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_130, full_260);  amin_130 = full_260 = None
	        sub_5956: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_130, minimum_130);  maximum_130 = None
	        div_260: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5956, 255.0);  sub_5956 = None
	        clamp_min_390: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_260, 1.1920928955078125e-07);  div_260 = None
	        div_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_130, clamp_min_390);  minimum_130 = None
	        round_261: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_261);  div_261 = None
	        sub_5962: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_261);  round_261 = None
	        clamp_min_391: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_5962, -128);  sub_5962 = None
	        clamp_max_260: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_391, 127);  clamp_min_391 = None
	        view_2040: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_390, [sym_size_int, 1500, 1])
	        reciprocal_130: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2040);  view_2040 = None
	        mul_12644: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_130, 1.0);  reciprocal_130 = None
	        mul_12647: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_19930, mul_12644);  add_19930 = mul_12644 = None
	        round_262: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12647);  mul_12647 = None
	        convert_element_type_780: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_260, torch.int8);  clamp_max_260 = None
	        view_2041: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_780, [sym_size_int, 1500, 1])
	        add_20017: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_262, view_2041);  round_262 = view_2041 = None
	        clamp_min_392: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20017, -128);  add_20017 = None
	        clamp_max_261: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_392, 127);  clamp_min_392 = None
	        convert_element_type_781: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_261, torch.int8);  clamp_max_261 = None
	        view_2045: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_780, [sym_size_int, 1500, 1]);  convert_element_type_780 = None
	        convert_element_type_782: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_781, torch.float32);  convert_element_type_781 = None
	        convert_element_type_783: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2045, torch.float32);  view_2045 = None
	        sub_5982: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_782, convert_element_type_783);  convert_element_type_782 = convert_element_type_783 = None
	        view_2044: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_390, [sym_size_int, 1500, 1]);  clamp_min_390 = None
	        mul_12669: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5982, view_2044);  sub_5982 = view_2044 = None
	        view_2047: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_21_fc1_parametrizations_weight_original0 = None
	        view_2049: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_21_fc1_parametrizations_weight_original2 = None
	        convert_element_type_784: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2047, torch.float32);  view_2047 = None
	        convert_element_type_785: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2049, torch.float32);  view_2049 = None
	        sub_5986: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_784, convert_element_type_785);  convert_element_type_784 = convert_element_type_785 = None
	        view_2048: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_21_fc1_parametrizations_weight_original1 = None
	        mul_12674: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_5986, view_2048);  sub_5986 = view_2048 = None
	        view_2050: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12674, [5120, 1280]);  mul_12674 = None
	        mul_12679: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2051: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12669, [mul_12679, 1280]);  mul_12669 = mul_12679 = None
	        permute_219: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2050, [1, 0]);  view_2050 = None
	        mm_default_51: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_2051, permute_219);  view_2051 = permute_219 = None
	        add_tensor_51: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_51, model_audio_tower_layers_21_fc1_bias);  mm_default_51 = model_audio_tower_layers_21_fc1_bias = None
	        view_2052: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_51, [sym_size_int, 1500, 5120]);  add_tensor_51 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_12686: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2052, 0.5)
	        mul_12687: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2052, 0.7071067811865476);  view_2052 = None
	        erf_23: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_12687);  mul_12687 = None
	        add_20076: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_23, 1);  erf_23 = None
	        mul_12688: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12686, add_20076);  mul_12686 = add_20076 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_12688, [2])
	        full_263: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_131, full_263);  amax_131 = full_263 = None
	        amin_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_12688, [2])
	        full_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_131: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_131, full_262);  amin_131 = full_262 = None
	        sub_5999: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_131, minimum_131);  maximum_131 = None
	        div_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_5999, 255.0);  sub_5999 = None
	        clamp_min_393: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_262, 1.1920928955078125e-07);  div_262 = None
	        div_263: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_131, clamp_min_393);  minimum_131 = None
	        round_263: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_263);  div_263 = None
	        sub_6005: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_263);  round_263 = None
	        clamp_min_394: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6005, -128);  sub_6005 = None
	        clamp_max_262: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_394, 127);  clamp_min_394 = None
	        view_2055: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_393, [sym_size_int, 1500, 1])
	        reciprocal_131: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2055);  view_2055 = None
	        mul_12734: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_131, 1.0);  reciprocal_131 = None
	        mul_12737: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12688, mul_12734);  mul_12688 = mul_12734 = None
	        round_264: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_12737);  mul_12737 = None
	        convert_element_type_786: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_262, torch.int8);  clamp_max_262 = None
	        view_2056: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_786, [sym_size_int, 1500, 1])
	        add_20159: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_264, view_2056);  round_264 = view_2056 = None
	        clamp_min_395: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20159, -128);  add_20159 = None
	        clamp_max_263: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_395, 127);  clamp_min_395 = None
	        convert_element_type_787: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_263, torch.int8);  clamp_max_263 = None
	        view_2060: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_786, [sym_size_int, 1500, 1]);  convert_element_type_786 = None
	        convert_element_type_788: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_787, torch.float32);  convert_element_type_787 = None
	        convert_element_type_789: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2060, torch.float32);  view_2060 = None
	        sub_6025: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_788, convert_element_type_789);  convert_element_type_788 = convert_element_type_789 = None
	        view_2059: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_393, [sym_size_int, 1500, 1]);  clamp_min_393 = None
	        mul_12759: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6025, view_2059);  sub_6025 = view_2059 = None
	        view_2062: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_21_fc2_parametrizations_weight_original0 = None
	        view_2064: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_21_fc2_parametrizations_weight_original2 = None
	        convert_element_type_790: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2062, torch.float32);  view_2062 = None
	        convert_element_type_791: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2064, torch.float32);  view_2064 = None
	        sub_6029: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_790, convert_element_type_791);  convert_element_type_790 = convert_element_type_791 = None
	        view_2063: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_21_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_21_fc2_parametrizations_weight_original1 = None
	        mul_12764: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6029, view_2063);  sub_6029 = view_2063 = None
	        view_2065: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12764, [1280, 5120]);  mul_12764 = None
	        mul_12769: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2066: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12759, [mul_12769, 5120]);  mul_12759 = mul_12769 = None
	        permute_220: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2065, [1, 0]);  view_2065 = None
	        mm_default_50: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2066, permute_220);  view_2066 = permute_220 = None
	        add_tensor_50: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_50, model_audio_tower_layers_21_fc2_bias);  mm_default_50 = model_audio_tower_layers_21_fc2_bias = None
	        view_2067: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_50, [sym_size_int, 1500, 1280]);  add_tensor_50 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_20222: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_19924, view_2067);  add_19924 = view_2067 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_177: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_20222, memory_format = torch.contiguous_format)
	        var_mean_44 = torch.ops.aten.var_mean.correction(clone_177, [2], correction = 0, keepdim = True)
	        getitem_176: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_44[0]
	        getitem_177: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_44[1];  var_mean_44 = None
	        sub_6035: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_177, getitem_177);  clone_177 = getitem_177 = None
	        add_20227: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_176, 1e-05);  getitem_176 = None
	        rsqrt_44: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_20227);  add_20227 = None
	        mul_12780: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6035, rsqrt_44);  sub_6035 = rsqrt_44 = None
	        mul_12781: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_12780, model_audio_tower_layers_22_self_attn_layer_norm_weight);  mul_12780 = model_audio_tower_layers_22_self_attn_layer_norm_weight = None
	        add_20228: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_12781, model_audio_tower_layers_22_self_attn_layer_norm_bias);  mul_12781 = model_audio_tower_layers_22_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_20228, [2])
	        full_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_132, full_265);  amax_132 = full_265 = None
	        amin_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_20228, [2])
	        full_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_132: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_132, full_264);  amin_132 = full_264 = None
	        sub_6046: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_132, minimum_132);  maximum_132 = None
	        div_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6046, 255.0);  sub_6046 = None
	        clamp_min_396: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_264, 1.1920928955078125e-07);  div_264 = None
	        div_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_132, clamp_min_396);  minimum_132 = None
	        round_265: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_265);  div_265 = None
	        sub_6052: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_265);  round_265 = None
	        clamp_min_397: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6052, -128);  sub_6052 = None
	        clamp_max_264: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_397, 127);  clamp_min_397 = None
	        view_2070: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_396, [sym_size_int, 1500, 1])
	        reciprocal_132: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2070);  view_2070 = None
	        mul_12829: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_132, 1.0);  reciprocal_132 = None
	        mul_12832: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_20228, mul_12829);  mul_12829 = None
	        round_266: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12832);  mul_12832 = None
	        convert_element_type_792: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_264, torch.int8);  clamp_max_264 = None
	        view_2071: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_792, [sym_size_int, 1500, 1])
	        add_20315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_266, view_2071);  round_266 = view_2071 = None
	        clamp_min_398: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20315, -128);  add_20315 = None
	        clamp_max_265: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_398, 127);  clamp_min_398 = None
	        convert_element_type_793: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_265, torch.int8);  clamp_max_265 = None
	        view_2075: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_792, [sym_size_int, 1500, 1]);  convert_element_type_792 = None
	        convert_element_type_794: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_793, torch.float32);  convert_element_type_793 = None
	        convert_element_type_795: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2075, torch.float32);  view_2075 = None
	        sub_6072: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_794, convert_element_type_795);  convert_element_type_794 = convert_element_type_795 = None
	        view_2074: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_396, [sym_size_int, 1500, 1]);  clamp_min_396 = None
	        mul_12854: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6072, view_2074);  sub_6072 = view_2074 = None
	        view_2077: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2079: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_796: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2077, torch.float32);  view_2077 = None
	        convert_element_type_797: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2079, torch.float32);  view_2079 = None
	        sub_6076: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_796, convert_element_type_797);  convert_element_type_796 = convert_element_type_797 = None
	        view_2078: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_12859: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6076, view_2078);  sub_6076 = view_2078 = None
	        view_2080: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12859, [1280, 1280]);  mul_12859 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_20228, [2])
	        full_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_133, full_267);  amax_133 = full_267 = None
	        amin_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_20228, [2])
	        full_266: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_133: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_133, full_266);  amin_133 = full_266 = None
	        sub_6091: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_133, minimum_133);  maximum_133 = None
	        div_266: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6091, 255.0);  sub_6091 = None
	        clamp_min_399: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_266, 1.1920928955078125e-07);  div_266 = None
	        div_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_133, clamp_min_399);  minimum_133 = None
	        round_267: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_267);  div_267 = None
	        sub_6097: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_267);  round_267 = None
	        clamp_min_400: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6097, -128);  sub_6097 = None
	        clamp_max_266: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_400, 127);  clamp_min_400 = None
	        view_2086: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_399, [sym_size_int, 1500, 1])
	        reciprocal_133: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2086);  view_2086 = None
	        mul_12925: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_133, 1.0);  reciprocal_133 = None
	        mul_12928: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_20228, mul_12925);  mul_12925 = None
	        round_268: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_12928);  mul_12928 = None
	        convert_element_type_798: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_266, torch.int8);  clamp_max_266 = None
	        view_2087: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_798, [sym_size_int, 1500, 1])
	        add_20467: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_268, view_2087);  round_268 = view_2087 = None
	        clamp_min_401: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20467, -128);  add_20467 = None
	        clamp_max_267: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_401, 127);  clamp_min_401 = None
	        convert_element_type_799: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_267, torch.int8);  clamp_max_267 = None
	        view_2091: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_798, [sym_size_int, 1500, 1]);  convert_element_type_798 = None
	        convert_element_type_800: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_799, torch.float32);  convert_element_type_799 = None
	        convert_element_type_801: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2091, torch.float32);  view_2091 = None
	        sub_6117: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_800, convert_element_type_801);  convert_element_type_800 = convert_element_type_801 = None
	        view_2090: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_399, [sym_size_int, 1500, 1]);  clamp_min_399 = None
	        mul_12950: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6117, view_2090);  sub_6117 = view_2090 = None
	        view_2093: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2095: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_802: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2093, torch.float32);  view_2093 = None
	        convert_element_type_803: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2095, torch.float32);  view_2095 = None
	        sub_6121: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_802, convert_element_type_803);  convert_element_type_802 = convert_element_type_803 = None
	        view_2094: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_12955: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6121, view_2094);  sub_6121 = view_2094 = None
	        view_2096: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12955, [1280, 1280]);  mul_12955 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_20228, [2])
	        full_269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_134, full_269);  amax_134 = full_269 = None
	        amin_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_20228, [2])
	        full_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_134: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_134, full_268);  amin_134 = full_268 = None
	        sub_6135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_134, minimum_134);  maximum_134 = None
	        div_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6135, 255.0);  sub_6135 = None
	        clamp_min_402: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_268, 1.1920928955078125e-07);  div_268 = None
	        div_269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_134, clamp_min_402);  minimum_134 = None
	        round_269: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_269);  div_269 = None
	        sub_6141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_269);  round_269 = None
	        clamp_min_403: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6141, -128);  sub_6141 = None
	        clamp_max_268: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_403, 127);  clamp_min_403 = None
	        view_2102: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_402, [sym_size_int, 1500, 1])
	        reciprocal_134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2102);  view_2102 = None
	        mul_13024: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_134, 1.0);  reciprocal_134 = None
	        mul_13027: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_20228, mul_13024);  add_20228 = mul_13024 = None
	        round_270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13027);  mul_13027 = None
	        convert_element_type_804: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_268, torch.int8);  clamp_max_268 = None
	        view_2103: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_804, [sym_size_int, 1500, 1])
	        add_20615: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_270, view_2103);  round_270 = view_2103 = None
	        clamp_min_404: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20615, -128);  add_20615 = None
	        clamp_max_269: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_404, 127);  clamp_min_404 = None
	        convert_element_type_805: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_269, torch.int8);  clamp_max_269 = None
	        view_2107: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_804, [sym_size_int, 1500, 1]);  convert_element_type_804 = None
	        convert_element_type_806: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_805, torch.float32);  convert_element_type_805 = None
	        convert_element_type_807: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2107, torch.float32);  view_2107 = None
	        sub_6161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_806, convert_element_type_807);  convert_element_type_806 = convert_element_type_807 = None
	        view_2106: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_402, [sym_size_int, 1500, 1]);  clamp_min_402 = None
	        mul_13049: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6161, view_2106);  sub_6161 = view_2106 = None
	        view_2109: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2111: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_808: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2109, torch.float32);  view_2109 = None
	        convert_element_type_809: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2111, torch.float32);  view_2111 = None
	        sub_6165: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_808, convert_element_type_809);  convert_element_type_808 = convert_element_type_809 = None
	        view_2110: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_13054: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6165, view_2110);  sub_6165 = view_2110 = None
	        view_2112: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13054, [1280, 1280]);  mul_13054 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_12864: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2081: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12854, [mul_12864, 1280]);  mul_12854 = mul_12864 = None
	        permute_221: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2080, [1, 0]);  view_2080 = None
	        mm_default_49: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2081, permute_221);  view_2081 = permute_221 = None
	        add_tensor_49: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_49, model_audio_tower_layers_22_self_attn_q_proj_bias);  mm_default_49 = model_audio_tower_layers_22_self_attn_q_proj_bias = None
	        view_2082: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_49, [sym_size_int, 1500, 1280]);  add_tensor_49 = None
	        mul_12871: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2082, 0.125);  view_2082 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2083: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12871, [sym_size_int, 1500, 20, 64]);  mul_12871 = None
	        permute_222: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2083, [0, 2, 1, 3]);  view_2083 = None
	        clone_178: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_222, memory_format = torch.contiguous_format);  permute_222 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_12958: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2097: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_12950, [mul_12958, 1280]);  mul_12950 = mul_12958 = None
	        permute_223: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2096, [1, 0]);  view_2096 = None
	        mm_22: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2097, permute_223);  view_2097 = permute_223 = None
	        view_2098: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_22, [sym_size_int, 1500, 1280]);  mm_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2099: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2098, [sym_size_int, -1, 20, 64]);  view_2098 = None
	        permute_224: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2099, [0, 2, 1, 3]);  view_2099 = None
	        clone_179: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_224, memory_format = torch.contiguous_format);  permute_224 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_13059: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2113: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13049, [mul_13059, 1280]);  mul_13049 = mul_13059 = None
	        permute_225: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2112, [1, 0]);  view_2112 = None
	        mm_default_48: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2113, permute_225);  view_2113 = permute_225 = None
	        add_tensor_48: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_48, model_audio_tower_layers_22_self_attn_v_proj_bias);  mm_default_48 = model_audio_tower_layers_22_self_attn_v_proj_bias = None
	        view_2114: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_48, [sym_size_int, 1500, 1280]);  add_tensor_48 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2115: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2114, [sym_size_int, -1, 20, 64]);  view_2114 = None
	        permute_226: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2115, [0, 2, 1, 3]);  view_2115 = None
	        clone_180: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_226, memory_format = torch.contiguous_format);  permute_226 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_22 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_178, clone_179, clone_180, None, False, scale = 1.0);  clone_178 = clone_179 = clone_180 = None
	        getitem_178: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_22[0];  _scaled_dot_product_efficient_attention_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_227: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_178, [0, 2, 1, 3]);  getitem_178 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2116: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_227, [sym_size_int, 1500, -1]);  permute_227 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2116, [2])
	        full_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_135, full_271);  amax_135 = full_271 = None
	        amin_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2116, [2])
	        full_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_135: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_135, full_270);  amin_135 = full_270 = None
	        sub_6183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_135, minimum_135);  maximum_135 = None
	        div_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6183, 255.0);  sub_6183 = None
	        clamp_min_405: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_270, 1.1920928955078125e-07);  div_270 = None
	        div_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_135, clamp_min_405);  minimum_135 = None
	        round_271: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_271);  div_271 = None
	        sub_6189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_271);  round_271 = None
	        clamp_min_406: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6189, -128);  sub_6189 = None
	        clamp_max_270: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_406, 127);  clamp_min_406 = None
	        view_2119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_405, [sym_size_int, 1500, 1])
	        reciprocal_135: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2119);  view_2119 = None
	        mul_13129: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_135, 1.0);  reciprocal_135 = None
	        mul_13132: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2116, mul_13129);  view_2116 = mul_13129 = None
	        round_272: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13132);  mul_13132 = None
	        convert_element_type_810: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_270, torch.int8);  clamp_max_270 = None
	        view_2120: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_810, [sym_size_int, 1500, 1])
	        add_20779: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_272, view_2120);  round_272 = view_2120 = None
	        clamp_min_407: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20779, -128);  add_20779 = None
	        clamp_max_271: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_407, 127);  clamp_min_407 = None
	        convert_element_type_811: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_271, torch.int8);  clamp_max_271 = None
	        view_2124: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_810, [sym_size_int, 1500, 1]);  convert_element_type_810 = None
	        convert_element_type_812: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_811, torch.float32);  convert_element_type_811 = None
	        convert_element_type_813: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2124, torch.float32);  view_2124 = None
	        sub_6209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_812, convert_element_type_813);  convert_element_type_812 = convert_element_type_813 = None
	        view_2123: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_405, [sym_size_int, 1500, 1]);  clamp_min_405 = None
	        mul_13154: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6209, view_2123);  sub_6209 = view_2123 = None
	        view_2126: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2128: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_814: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2126, torch.float32);  view_2126 = None
	        convert_element_type_815: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2128, torch.float32);  view_2128 = None
	        sub_6213: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_814, convert_element_type_815);  convert_element_type_814 = convert_element_type_815 = None
	        view_2127: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_13159: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6213, view_2127);  sub_6213 = view_2127 = None
	        view_2129: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13159, [1280, 1280]);  mul_13159 = None
	        mul_13164: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2130: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13154, [mul_13164, 1280]);  mul_13154 = mul_13164 = None
	        permute_228: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2129, [1, 0]);  view_2129 = None
	        mm_default_47: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2130, permute_228);  view_2130 = permute_228 = None
	        add_tensor_47: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_47, model_audio_tower_layers_22_self_attn_out_proj_bias);  mm_default_47 = model_audio_tower_layers_22_self_attn_out_proj_bias = None
	        view_2131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_47, [sym_size_int, 1500, 1280]);  add_tensor_47 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_20842: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_20222, view_2131);  add_20222 = view_2131 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_182: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_20842, memory_format = torch.contiguous_format)
	        var_mean_45 = torch.ops.aten.var_mean.correction(clone_182, [2], correction = 0, keepdim = True)
	        getitem_182: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_45[0]
	        getitem_183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_45[1];  var_mean_45 = None
	        sub_6219: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_182, getitem_183);  clone_182 = getitem_183 = None
	        add_20847: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_182, 1e-05);  getitem_182 = None
	        rsqrt_45: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_20847);  add_20847 = None
	        mul_13175: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6219, rsqrt_45);  sub_6219 = rsqrt_45 = None
	        mul_13176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13175, model_audio_tower_layers_22_final_layer_norm_weight);  mul_13175 = model_audio_tower_layers_22_final_layer_norm_weight = None
	        add_20848: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13176, model_audio_tower_layers_22_final_layer_norm_bias);  mul_13176 = model_audio_tower_layers_22_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_20848, [2])
	        full_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_136, full_273);  amax_136 = full_273 = None
	        amin_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_20848, [2])
	        full_272: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_136: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_136, full_272);  amin_136 = full_272 = None
	        sub_6230: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_136, minimum_136);  maximum_136 = None
	        div_272: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6230, 255.0);  sub_6230 = None
	        clamp_min_408: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_272, 1.1920928955078125e-07);  div_272 = None
	        div_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_136, clamp_min_408);  minimum_136 = None
	        round_273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_273);  div_273 = None
	        sub_6236: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_273);  round_273 = None
	        clamp_min_409: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6236, -128);  sub_6236 = None
	        clamp_max_272: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_409, 127);  clamp_min_409 = None
	        view_2134: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_408, [sym_size_int, 1500, 1])
	        reciprocal_136: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2134);  view_2134 = None
	        mul_13224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_136, 1.0);  reciprocal_136 = None
	        mul_13227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_20848, mul_13224);  add_20848 = mul_13224 = None
	        round_274: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13227);  mul_13227 = None
	        convert_element_type_816: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_272, torch.int8);  clamp_max_272 = None
	        view_2135: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_816, [sym_size_int, 1500, 1])
	        add_20935: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_274, view_2135);  round_274 = view_2135 = None
	        clamp_min_410: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_20935, -128);  add_20935 = None
	        clamp_max_273: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_410, 127);  clamp_min_410 = None
	        convert_element_type_817: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_273, torch.int8);  clamp_max_273 = None
	        view_2139: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_816, [sym_size_int, 1500, 1]);  convert_element_type_816 = None
	        convert_element_type_818: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_817, torch.float32);  convert_element_type_817 = None
	        convert_element_type_819: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2139, torch.float32);  view_2139 = None
	        sub_6256: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_818, convert_element_type_819);  convert_element_type_818 = convert_element_type_819 = None
	        view_2138: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_408, [sym_size_int, 1500, 1]);  clamp_min_408 = None
	        mul_13249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6256, view_2138);  sub_6256 = view_2138 = None
	        view_2141: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_22_fc1_parametrizations_weight_original0 = None
	        view_2143: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_22_fc1_parametrizations_weight_original2 = None
	        convert_element_type_820: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2141, torch.float32);  view_2141 = None
	        convert_element_type_821: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2143, torch.float32);  view_2143 = None
	        sub_6260: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_820, convert_element_type_821);  convert_element_type_820 = convert_element_type_821 = None
	        view_2142: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_22_fc1_parametrizations_weight_original1 = None
	        mul_13254: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6260, view_2142);  sub_6260 = view_2142 = None
	        view_2144: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13254, [5120, 1280]);  mul_13254 = None
	        mul_13259: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2145: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13249, [mul_13259, 1280]);  mul_13249 = mul_13259 = None
	        permute_229: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2144, [1, 0]);  view_2144 = None
	        mm_default_46: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_2145, permute_229);  view_2145 = permute_229 = None
	        add_tensor_46: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_46, model_audio_tower_layers_22_fc1_bias);  mm_default_46 = model_audio_tower_layers_22_fc1_bias = None
	        view_2146: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_46, [sym_size_int, 1500, 5120]);  add_tensor_46 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_13266: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2146, 0.5)
	        mul_13267: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2146, 0.7071067811865476);  view_2146 = None
	        erf_24: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_13267);  mul_13267 = None
	        add_20994: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_24, 1);  erf_24 = None
	        mul_13268: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13266, add_20994);  mul_13266 = add_20994 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_13268, [2])
	        full_275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_137, full_275);  amax_137 = full_275 = None
	        amin_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_13268, [2])
	        full_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_137: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_137, full_274);  amin_137 = full_274 = None
	        sub_6273: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_137, minimum_137);  maximum_137 = None
	        div_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6273, 255.0);  sub_6273 = None
	        clamp_min_411: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_274, 1.1920928955078125e-07);  div_274 = None
	        div_275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_137, clamp_min_411);  minimum_137 = None
	        round_275: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_275);  div_275 = None
	        sub_6279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_275);  round_275 = None
	        clamp_min_412: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6279, -128);  sub_6279 = None
	        clamp_max_274: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_412, 127);  clamp_min_412 = None
	        view_2149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_411, [sym_size_int, 1500, 1])
	        reciprocal_137: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2149);  view_2149 = None
	        mul_13314: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_137, 1.0);  reciprocal_137 = None
	        mul_13317: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13268, mul_13314);  mul_13268 = mul_13314 = None
	        round_276: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_13317);  mul_13317 = None
	        convert_element_type_822: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_274, torch.int8);  clamp_max_274 = None
	        view_2150: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_822, [sym_size_int, 1500, 1])
	        add_21077: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_276, view_2150);  round_276 = view_2150 = None
	        clamp_min_413: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21077, -128);  add_21077 = None
	        clamp_max_275: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_413, 127);  clamp_min_413 = None
	        convert_element_type_823: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_275, torch.int8);  clamp_max_275 = None
	        view_2154: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_822, [sym_size_int, 1500, 1]);  convert_element_type_822 = None
	        convert_element_type_824: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_823, torch.float32);  convert_element_type_823 = None
	        convert_element_type_825: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2154, torch.float32);  view_2154 = None
	        sub_6299: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_824, convert_element_type_825);  convert_element_type_824 = convert_element_type_825 = None
	        view_2153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_411, [sym_size_int, 1500, 1]);  clamp_min_411 = None
	        mul_13339: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6299, view_2153);  sub_6299 = view_2153 = None
	        view_2156: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_22_fc2_parametrizations_weight_original0 = None
	        view_2158: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_22_fc2_parametrizations_weight_original2 = None
	        convert_element_type_826: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2156, torch.float32);  view_2156 = None
	        convert_element_type_827: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2158, torch.float32);  view_2158 = None
	        sub_6303: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_826, convert_element_type_827);  convert_element_type_826 = convert_element_type_827 = None
	        view_2157: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_22_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_22_fc2_parametrizations_weight_original1 = None
	        mul_13344: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6303, view_2157);  sub_6303 = view_2157 = None
	        view_2159: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13344, [1280, 5120]);  mul_13344 = None
	        mul_13349: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2160: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13339, [mul_13349, 5120]);  mul_13339 = mul_13349 = None
	        permute_230: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2159, [1, 0]);  view_2159 = None
	        mm_default_45: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2160, permute_230);  view_2160 = permute_230 = None
	        add_tensor_45: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_45, model_audio_tower_layers_22_fc2_bias);  mm_default_45 = model_audio_tower_layers_22_fc2_bias = None
	        view_2161: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_45, [sym_size_int, 1500, 1280]);  add_tensor_45 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_21140: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_20842, view_2161);  add_20842 = view_2161 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_185: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_21140, memory_format = torch.contiguous_format)
	        var_mean_46 = torch.ops.aten.var_mean.correction(clone_185, [2], correction = 0, keepdim = True)
	        getitem_184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_46[0]
	        getitem_185: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_46[1];  var_mean_46 = None
	        sub_6309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_185, getitem_185);  clone_185 = getitem_185 = None
	        add_21145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_184, 1e-05);  getitem_184 = None
	        rsqrt_46: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_21145);  add_21145 = None
	        mul_13360: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6309, rsqrt_46);  sub_6309 = rsqrt_46 = None
	        mul_13361: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13360, model_audio_tower_layers_23_self_attn_layer_norm_weight);  mul_13360 = model_audio_tower_layers_23_self_attn_layer_norm_weight = None
	        add_21146: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13361, model_audio_tower_layers_23_self_attn_layer_norm_bias);  mul_13361 = model_audio_tower_layers_23_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_21146, [2])
	        full_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_138, full_277);  amax_138 = full_277 = None
	        amin_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_21146, [2])
	        full_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_138: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_138, full_276);  amin_138 = full_276 = None
	        sub_6320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_138, minimum_138);  maximum_138 = None
	        div_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6320, 255.0);  sub_6320 = None
	        clamp_min_414: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_276, 1.1920928955078125e-07);  div_276 = None
	        div_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_138, clamp_min_414);  minimum_138 = None
	        round_277: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_277);  div_277 = None
	        sub_6326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_277);  round_277 = None
	        clamp_min_415: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6326, -128);  sub_6326 = None
	        clamp_max_276: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_415, 127);  clamp_min_415 = None
	        view_2164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_414, [sym_size_int, 1500, 1])
	        reciprocal_138: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2164);  view_2164 = None
	        mul_13409: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_138, 1.0);  reciprocal_138 = None
	        mul_13412: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_21146, mul_13409);  mul_13409 = None
	        round_278: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13412);  mul_13412 = None
	        convert_element_type_828: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_276, torch.int8);  clamp_max_276 = None
	        view_2165: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_828, [sym_size_int, 1500, 1])
	        add_21233: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_278, view_2165);  round_278 = view_2165 = None
	        clamp_min_416: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21233, -128);  add_21233 = None
	        clamp_max_277: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_416, 127);  clamp_min_416 = None
	        convert_element_type_829: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_277, torch.int8);  clamp_max_277 = None
	        view_2169: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_828, [sym_size_int, 1500, 1]);  convert_element_type_828 = None
	        convert_element_type_830: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_829, torch.float32);  convert_element_type_829 = None
	        convert_element_type_831: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2169, torch.float32);  view_2169 = None
	        sub_6346: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_830, convert_element_type_831);  convert_element_type_830 = convert_element_type_831 = None
	        view_2168: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_414, [sym_size_int, 1500, 1]);  clamp_min_414 = None
	        mul_13434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6346, view_2168);  sub_6346 = view_2168 = None
	        view_2171: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2173: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_832: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2171, torch.float32);  view_2171 = None
	        convert_element_type_833: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2173, torch.float32);  view_2173 = None
	        sub_6350: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_832, convert_element_type_833);  convert_element_type_832 = convert_element_type_833 = None
	        view_2172: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_13439: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6350, view_2172);  sub_6350 = view_2172 = None
	        view_2174: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13439, [1280, 1280]);  mul_13439 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_21146, [2])
	        full_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_139, full_279);  amax_139 = full_279 = None
	        amin_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_21146, [2])
	        full_278: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_139: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_139, full_278);  amin_139 = full_278 = None
	        sub_6365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_139, minimum_139);  maximum_139 = None
	        div_278: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6365, 255.0);  sub_6365 = None
	        clamp_min_417: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_278, 1.1920928955078125e-07);  div_278 = None
	        div_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_139, clamp_min_417);  minimum_139 = None
	        round_279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_279);  div_279 = None
	        sub_6371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_279);  round_279 = None
	        clamp_min_418: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6371, -128);  sub_6371 = None
	        clamp_max_278: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_418, 127);  clamp_min_418 = None
	        view_2180: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_417, [sym_size_int, 1500, 1])
	        reciprocal_139: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2180);  view_2180 = None
	        mul_13505: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_139, 1.0);  reciprocal_139 = None
	        mul_13508: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_21146, mul_13505);  mul_13505 = None
	        round_280: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13508);  mul_13508 = None
	        convert_element_type_834: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_278, torch.int8);  clamp_max_278 = None
	        view_2181: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_834, [sym_size_int, 1500, 1])
	        add_21385: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_280, view_2181);  round_280 = view_2181 = None
	        clamp_min_419: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21385, -128);  add_21385 = None
	        clamp_max_279: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_419, 127);  clamp_min_419 = None
	        convert_element_type_835: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_279, torch.int8);  clamp_max_279 = None
	        view_2185: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_834, [sym_size_int, 1500, 1]);  convert_element_type_834 = None
	        convert_element_type_836: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_835, torch.float32);  convert_element_type_835 = None
	        convert_element_type_837: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2185, torch.float32);  view_2185 = None
	        sub_6391: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_836, convert_element_type_837);  convert_element_type_836 = convert_element_type_837 = None
	        view_2184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_417, [sym_size_int, 1500, 1]);  clamp_min_417 = None
	        mul_13530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6391, view_2184);  sub_6391 = view_2184 = None
	        view_2187: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2189: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_838: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2187, torch.float32);  view_2187 = None
	        convert_element_type_839: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2189, torch.float32);  view_2189 = None
	        sub_6395: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_838, convert_element_type_839);  convert_element_type_838 = convert_element_type_839 = None
	        view_2188: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_13535: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6395, view_2188);  sub_6395 = view_2188 = None
	        view_2190: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13535, [1280, 1280]);  mul_13535 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_21146, [2])
	        full_281: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_140, full_281);  amax_140 = full_281 = None
	        amin_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_21146, [2])
	        full_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_140: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_140, full_280);  amin_140 = full_280 = None
	        sub_6409: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_140, minimum_140);  maximum_140 = None
	        div_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6409, 255.0);  sub_6409 = None
	        clamp_min_420: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_280, 1.1920928955078125e-07);  div_280 = None
	        div_281: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_140, clamp_min_420);  minimum_140 = None
	        round_281: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_281);  div_281 = None
	        sub_6415: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_281);  round_281 = None
	        clamp_min_421: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6415, -128);  sub_6415 = None
	        clamp_max_280: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_421, 127);  clamp_min_421 = None
	        view_2196: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_420, [sym_size_int, 1500, 1])
	        reciprocal_140: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2196);  view_2196 = None
	        mul_13604: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_140, 1.0);  reciprocal_140 = None
	        mul_13607: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_21146, mul_13604);  add_21146 = mul_13604 = None
	        round_282: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13607);  mul_13607 = None
	        convert_element_type_840: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_280, torch.int8);  clamp_max_280 = None
	        view_2197: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_840, [sym_size_int, 1500, 1])
	        add_21533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_282, view_2197);  round_282 = view_2197 = None
	        clamp_min_422: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21533, -128);  add_21533 = None
	        clamp_max_281: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_422, 127);  clamp_min_422 = None
	        convert_element_type_841: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_281, torch.int8);  clamp_max_281 = None
	        view_2201: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_840, [sym_size_int, 1500, 1]);  convert_element_type_840 = None
	        convert_element_type_842: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_841, torch.float32);  convert_element_type_841 = None
	        convert_element_type_843: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2201, torch.float32);  view_2201 = None
	        sub_6435: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_842, convert_element_type_843);  convert_element_type_842 = convert_element_type_843 = None
	        view_2200: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_420, [sym_size_int, 1500, 1]);  clamp_min_420 = None
	        mul_13629: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6435, view_2200);  sub_6435 = view_2200 = None
	        view_2203: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2205: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_844: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2203, torch.float32);  view_2203 = None
	        convert_element_type_845: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2205, torch.float32);  view_2205 = None
	        sub_6439: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_844, convert_element_type_845);  convert_element_type_844 = convert_element_type_845 = None
	        view_2204: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_13634: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6439, view_2204);  sub_6439 = view_2204 = None
	        view_2206: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13634, [1280, 1280]);  mul_13634 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_13444: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2175: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13434, [mul_13444, 1280]);  mul_13434 = mul_13444 = None
	        permute_231: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2174, [1, 0]);  view_2174 = None
	        mm_default_44: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2175, permute_231);  view_2175 = permute_231 = None
	        add_tensor_44: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_44, model_audio_tower_layers_23_self_attn_q_proj_bias);  mm_default_44 = model_audio_tower_layers_23_self_attn_q_proj_bias = None
	        view_2176: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_44, [sym_size_int, 1500, 1280]);  add_tensor_44 = None
	        mul_13451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2176, 0.125);  view_2176 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2177: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13451, [sym_size_int, 1500, 20, 64]);  mul_13451 = None
	        permute_232: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2177, [0, 2, 1, 3]);  view_2177 = None
	        clone_186: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_232, memory_format = torch.contiguous_format);  permute_232 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_13538: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2191: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13530, [mul_13538, 1280]);  mul_13530 = mul_13538 = None
	        permute_233: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2190, [1, 0]);  view_2190 = None
	        mm_23: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2191, permute_233);  view_2191 = permute_233 = None
	        view_2192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_23, [sym_size_int, 1500, 1280]);  mm_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2193: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2192, [sym_size_int, -1, 20, 64]);  view_2192 = None
	        permute_234: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2193, [0, 2, 1, 3]);  view_2193 = None
	        clone_187: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_234, memory_format = torch.contiguous_format);  permute_234 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_13639: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2207: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13629, [mul_13639, 1280]);  mul_13629 = mul_13639 = None
	        permute_235: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2206, [1, 0]);  view_2206 = None
	        mm_default_43: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2207, permute_235);  view_2207 = permute_235 = None
	        add_tensor_43: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_43, model_audio_tower_layers_23_self_attn_v_proj_bias);  mm_default_43 = model_audio_tower_layers_23_self_attn_v_proj_bias = None
	        view_2208: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_43, [sym_size_int, 1500, 1280]);  add_tensor_43 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2209: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2208, [sym_size_int, -1, 20, 64]);  view_2208 = None
	        permute_236: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2209, [0, 2, 1, 3]);  view_2209 = None
	        clone_188: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_236, memory_format = torch.contiguous_format);  permute_236 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_23 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_186, clone_187, clone_188, None, False, scale = 1.0);  clone_186 = clone_187 = clone_188 = None
	        getitem_186: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_23[0];  _scaled_dot_product_efficient_attention_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_237: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_186, [0, 2, 1, 3]);  getitem_186 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2210: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_237, [sym_size_int, 1500, -1]);  permute_237 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2210, [2])
	        full_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_141, full_283);  amax_141 = full_283 = None
	        amin_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2210, [2])
	        full_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_141: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_141, full_282);  amin_141 = full_282 = None
	        sub_6457: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_141, minimum_141);  maximum_141 = None
	        div_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6457, 255.0);  sub_6457 = None
	        clamp_min_423: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_282, 1.1920928955078125e-07);  div_282 = None
	        div_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_141, clamp_min_423);  minimum_141 = None
	        round_283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_283);  div_283 = None
	        sub_6463: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_283);  round_283 = None
	        clamp_min_424: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6463, -128);  sub_6463 = None
	        clamp_max_282: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_424, 127);  clamp_min_424 = None
	        view_2213: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_423, [sym_size_int, 1500, 1])
	        reciprocal_141: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2213);  view_2213 = None
	        mul_13709: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_141, 1.0);  reciprocal_141 = None
	        mul_13712: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2210, mul_13709);  view_2210 = mul_13709 = None
	        round_284: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13712);  mul_13712 = None
	        convert_element_type_846: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_282, torch.int8);  clamp_max_282 = None
	        view_2214: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_846, [sym_size_int, 1500, 1])
	        add_21697: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_284, view_2214);  round_284 = view_2214 = None
	        clamp_min_425: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21697, -128);  add_21697 = None
	        clamp_max_283: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_425, 127);  clamp_min_425 = None
	        convert_element_type_847: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_283, torch.int8);  clamp_max_283 = None
	        view_2218: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_846, [sym_size_int, 1500, 1]);  convert_element_type_846 = None
	        convert_element_type_848: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_847, torch.float32);  convert_element_type_847 = None
	        convert_element_type_849: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2218, torch.float32);  view_2218 = None
	        sub_6483: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_848, convert_element_type_849);  convert_element_type_848 = convert_element_type_849 = None
	        view_2217: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_423, [sym_size_int, 1500, 1]);  clamp_min_423 = None
	        mul_13734: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6483, view_2217);  sub_6483 = view_2217 = None
	        view_2220: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2222: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_850: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2220, torch.float32);  view_2220 = None
	        convert_element_type_851: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2222, torch.float32);  view_2222 = None
	        sub_6487: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_850, convert_element_type_851);  convert_element_type_850 = convert_element_type_851 = None
	        view_2221: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_13739: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6487, view_2221);  sub_6487 = view_2221 = None
	        view_2223: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13739, [1280, 1280]);  mul_13739 = None
	        mul_13744: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2224: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13734, [mul_13744, 1280]);  mul_13734 = mul_13744 = None
	        permute_238: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2223, [1, 0]);  view_2223 = None
	        mm_default_42: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2224, permute_238);  view_2224 = permute_238 = None
	        add_tensor_42: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_42, model_audio_tower_layers_23_self_attn_out_proj_bias);  mm_default_42 = model_audio_tower_layers_23_self_attn_out_proj_bias = None
	        view_2225: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_42, [sym_size_int, 1500, 1280]);  add_tensor_42 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_21760: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_21140, view_2225);  add_21140 = view_2225 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_190: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_21760, memory_format = torch.contiguous_format)
	        var_mean_47 = torch.ops.aten.var_mean.correction(clone_190, [2], correction = 0, keepdim = True)
	        getitem_190: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_47[0]
	        getitem_191: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_47[1];  var_mean_47 = None
	        sub_6493: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_190, getitem_191);  clone_190 = getitem_191 = None
	        add_21765: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_190, 1e-05);  getitem_190 = None
	        rsqrt_47: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_21765);  add_21765 = None
	        mul_13755: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6493, rsqrt_47);  sub_6493 = rsqrt_47 = None
	        mul_13756: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13755, model_audio_tower_layers_23_final_layer_norm_weight);  mul_13755 = model_audio_tower_layers_23_final_layer_norm_weight = None
	        add_21766: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13756, model_audio_tower_layers_23_final_layer_norm_bias);  mul_13756 = model_audio_tower_layers_23_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_21766, [2])
	        full_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_142, full_285);  amax_142 = full_285 = None
	        amin_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_21766, [2])
	        full_284: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_142, full_284);  amin_142 = full_284 = None
	        sub_6504: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_142, minimum_142);  maximum_142 = None
	        div_284: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6504, 255.0);  sub_6504 = None
	        clamp_min_426: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_284, 1.1920928955078125e-07);  div_284 = None
	        div_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_142, clamp_min_426);  minimum_142 = None
	        round_285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_285);  div_285 = None
	        sub_6510: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_285);  round_285 = None
	        clamp_min_427: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6510, -128);  sub_6510 = None
	        clamp_max_284: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_427, 127);  clamp_min_427 = None
	        view_2228: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_426, [sym_size_int, 1500, 1])
	        reciprocal_142: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2228);  view_2228 = None
	        mul_13804: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_142, 1.0);  reciprocal_142 = None
	        mul_13807: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_21766, mul_13804);  add_21766 = mul_13804 = None
	        round_286: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13807);  mul_13807 = None
	        convert_element_type_852: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_284, torch.int8);  clamp_max_284 = None
	        view_2229: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_852, [sym_size_int, 1500, 1])
	        add_21853: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_286, view_2229);  round_286 = view_2229 = None
	        clamp_min_428: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21853, -128);  add_21853 = None
	        clamp_max_285: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_428, 127);  clamp_min_428 = None
	        convert_element_type_853: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_285, torch.int8);  clamp_max_285 = None
	        view_2233: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_852, [sym_size_int, 1500, 1]);  convert_element_type_852 = None
	        convert_element_type_854: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_853, torch.float32);  convert_element_type_853 = None
	        convert_element_type_855: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2233, torch.float32);  view_2233 = None
	        sub_6530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_854, convert_element_type_855);  convert_element_type_854 = convert_element_type_855 = None
	        view_2232: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_426, [sym_size_int, 1500, 1]);  clamp_min_426 = None
	        mul_13829: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6530, view_2232);  sub_6530 = view_2232 = None
	        view_2235: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_23_fc1_parametrizations_weight_original0 = None
	        view_2237: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_23_fc1_parametrizations_weight_original2 = None
	        convert_element_type_856: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2235, torch.float32);  view_2235 = None
	        convert_element_type_857: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2237, torch.float32);  view_2237 = None
	        sub_6534: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_856, convert_element_type_857);  convert_element_type_856 = convert_element_type_857 = None
	        view_2236: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_23_fc1_parametrizations_weight_original1 = None
	        mul_13834: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6534, view_2236);  sub_6534 = view_2236 = None
	        view_2238: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13834, [5120, 1280]);  mul_13834 = None
	        mul_13839: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2239: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13829, [mul_13839, 1280]);  mul_13829 = mul_13839 = None
	        permute_239: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2238, [1, 0]);  view_2238 = None
	        mm_default_41: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_2239, permute_239);  view_2239 = permute_239 = None
	        add_tensor_41: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_41, model_audio_tower_layers_23_fc1_bias);  mm_default_41 = model_audio_tower_layers_23_fc1_bias = None
	        view_2240: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_41, [sym_size_int, 1500, 5120]);  add_tensor_41 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_13846: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2240, 0.5)
	        mul_13847: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2240, 0.7071067811865476);  view_2240 = None
	        erf_25: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_13847);  mul_13847 = None
	        add_21912: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_25, 1);  erf_25 = None
	        mul_13848: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13846, add_21912);  mul_13846 = add_21912 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_13848, [2])
	        full_287: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_143, full_287);  amax_143 = full_287 = None
	        amin_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_13848, [2])
	        full_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_143: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_143, full_286);  amin_143 = full_286 = None
	        sub_6547: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_143, minimum_143);  maximum_143 = None
	        div_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6547, 255.0);  sub_6547 = None
	        clamp_min_429: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_286, 1.1920928955078125e-07);  div_286 = None
	        div_287: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_143, clamp_min_429);  minimum_143 = None
	        round_287: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_287);  div_287 = None
	        sub_6553: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_287);  round_287 = None
	        clamp_min_430: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6553, -128);  sub_6553 = None
	        clamp_max_286: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_430, 127);  clamp_min_430 = None
	        view_2243: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_429, [sym_size_int, 1500, 1])
	        reciprocal_143: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2243);  view_2243 = None
	        mul_13894: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_143, 1.0);  reciprocal_143 = None
	        mul_13897: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13848, mul_13894);  mul_13848 = mul_13894 = None
	        round_288: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_13897);  mul_13897 = None
	        convert_element_type_858: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_286, torch.int8);  clamp_max_286 = None
	        view_2244: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_858, [sym_size_int, 1500, 1])
	        add_21995: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_288, view_2244);  round_288 = view_2244 = None
	        clamp_min_431: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_21995, -128);  add_21995 = None
	        clamp_max_287: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_431, 127);  clamp_min_431 = None
	        convert_element_type_859: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_287, torch.int8);  clamp_max_287 = None
	        view_2248: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_858, [sym_size_int, 1500, 1]);  convert_element_type_858 = None
	        convert_element_type_860: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_859, torch.float32);  convert_element_type_859 = None
	        convert_element_type_861: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2248, torch.float32);  view_2248 = None
	        sub_6573: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_860, convert_element_type_861);  convert_element_type_860 = convert_element_type_861 = None
	        view_2247: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_429, [sym_size_int, 1500, 1]);  clamp_min_429 = None
	        mul_13919: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6573, view_2247);  sub_6573 = view_2247 = None
	        view_2250: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_23_fc2_parametrizations_weight_original0 = None
	        view_2252: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_23_fc2_parametrizations_weight_original2 = None
	        convert_element_type_862: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2250, torch.float32);  view_2250 = None
	        convert_element_type_863: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2252, torch.float32);  view_2252 = None
	        sub_6577: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_862, convert_element_type_863);  convert_element_type_862 = convert_element_type_863 = None
	        view_2251: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_23_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_23_fc2_parametrizations_weight_original1 = None
	        mul_13924: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6577, view_2251);  sub_6577 = view_2251 = None
	        view_2253: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13924, [1280, 5120]);  mul_13924 = None
	        mul_13929: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2254: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_13919, [mul_13929, 5120]);  mul_13919 = mul_13929 = None
	        permute_240: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2253, [1, 0]);  view_2253 = None
	        mm_default_40: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2254, permute_240);  view_2254 = permute_240 = None
	        add_tensor_40: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_40, model_audio_tower_layers_23_fc2_bias);  mm_default_40 = model_audio_tower_layers_23_fc2_bias = None
	        view_2255: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_40, [sym_size_int, 1500, 1280]);  add_tensor_40 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_22058: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_21760, view_2255);  add_21760 = view_2255 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_193: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_22058, memory_format = torch.contiguous_format)
	        var_mean_48 = torch.ops.aten.var_mean.correction(clone_193, [2], correction = 0, keepdim = True)
	        getitem_192: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_48[0]
	        getitem_193: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_48[1];  var_mean_48 = None
	        sub_6583: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_193, getitem_193);  clone_193 = getitem_193 = None
	        add_22063: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_192, 1e-05);  getitem_192 = None
	        rsqrt_48: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_22063);  add_22063 = None
	        mul_13940: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6583, rsqrt_48);  sub_6583 = rsqrt_48 = None
	        mul_13941: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_13940, model_audio_tower_layers_24_self_attn_layer_norm_weight);  mul_13940 = model_audio_tower_layers_24_self_attn_layer_norm_weight = None
	        add_22064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_13941, model_audio_tower_layers_24_self_attn_layer_norm_bias);  mul_13941 = model_audio_tower_layers_24_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22064, [2])
	        full_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_144, full_289);  amax_144 = full_289 = None
	        amin_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22064, [2])
	        full_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_144: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_144, full_288);  amin_144 = full_288 = None
	        sub_6594: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_144, minimum_144);  maximum_144 = None
	        div_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6594, 255.0);  sub_6594 = None
	        clamp_min_432: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_288, 1.1920928955078125e-07);  div_288 = None
	        div_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_144, clamp_min_432);  minimum_144 = None
	        round_289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_289);  div_289 = None
	        sub_6600: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_289);  round_289 = None
	        clamp_min_433: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6600, -128);  sub_6600 = None
	        clamp_max_288: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_433, 127);  clamp_min_433 = None
	        view_2258: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_432, [sym_size_int, 1500, 1])
	        reciprocal_144: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2258);  view_2258 = None
	        mul_13989: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_144, 1.0);  reciprocal_144 = None
	        mul_13992: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22064, mul_13989);  mul_13989 = None
	        round_290: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_13992);  mul_13992 = None
	        convert_element_type_864: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_288, torch.int8);  clamp_max_288 = None
	        view_2259: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_864, [sym_size_int, 1500, 1])
	        add_22151: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_290, view_2259);  round_290 = view_2259 = None
	        clamp_min_434: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22151, -128);  add_22151 = None
	        clamp_max_289: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_434, 127);  clamp_min_434 = None
	        convert_element_type_865: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_289, torch.int8);  clamp_max_289 = None
	        view_2263: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_864, [sym_size_int, 1500, 1]);  convert_element_type_864 = None
	        convert_element_type_866: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_865, torch.float32);  convert_element_type_865 = None
	        convert_element_type_867: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2263, torch.float32);  view_2263 = None
	        sub_6620: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_866, convert_element_type_867);  convert_element_type_866 = convert_element_type_867 = None
	        view_2262: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_432, [sym_size_int, 1500, 1]);  clamp_min_432 = None
	        mul_14014: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6620, view_2262);  sub_6620 = view_2262 = None
	        view_2265: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2267: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_868: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2265, torch.float32);  view_2265 = None
	        convert_element_type_869: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2267, torch.float32);  view_2267 = None
	        sub_6624: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_868, convert_element_type_869);  convert_element_type_868 = convert_element_type_869 = None
	        view_2266: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_14019: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6624, view_2266);  sub_6624 = view_2266 = None
	        view_2268: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14019, [1280, 1280]);  mul_14019 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22064, [2])
	        full_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_145, full_291);  amax_145 = full_291 = None
	        amin_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22064, [2])
	        full_290: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_145: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_145, full_290);  amin_145 = full_290 = None
	        sub_6639: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_145, minimum_145);  maximum_145 = None
	        div_290: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6639, 255.0);  sub_6639 = None
	        clamp_min_435: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_290, 1.1920928955078125e-07);  div_290 = None
	        div_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_145, clamp_min_435);  minimum_145 = None
	        round_291: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_291);  div_291 = None
	        sub_6645: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_291);  round_291 = None
	        clamp_min_436: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6645, -128);  sub_6645 = None
	        clamp_max_290: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_436, 127);  clamp_min_436 = None
	        view_2274: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_435, [sym_size_int, 1500, 1])
	        reciprocal_145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2274);  view_2274 = None
	        mul_14085: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_145, 1.0);  reciprocal_145 = None
	        mul_14088: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22064, mul_14085);  mul_14085 = None
	        round_292: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14088);  mul_14088 = None
	        convert_element_type_870: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_290, torch.int8);  clamp_max_290 = None
	        view_2275: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_870, [sym_size_int, 1500, 1])
	        add_22303: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_292, view_2275);  round_292 = view_2275 = None
	        clamp_min_437: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22303, -128);  add_22303 = None
	        clamp_max_291: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_437, 127);  clamp_min_437 = None
	        convert_element_type_871: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_291, torch.int8);  clamp_max_291 = None
	        view_2279: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_870, [sym_size_int, 1500, 1]);  convert_element_type_870 = None
	        convert_element_type_872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_871, torch.float32);  convert_element_type_871 = None
	        convert_element_type_873: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2279, torch.float32);  view_2279 = None
	        sub_6665: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_872, convert_element_type_873);  convert_element_type_872 = convert_element_type_873 = None
	        view_2278: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_435, [sym_size_int, 1500, 1]);  clamp_min_435 = None
	        mul_14110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6665, view_2278);  sub_6665 = view_2278 = None
	        view_2281: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2283: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_874: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2281, torch.float32);  view_2281 = None
	        convert_element_type_875: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2283, torch.float32);  view_2283 = None
	        sub_6669: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_874, convert_element_type_875);  convert_element_type_874 = convert_element_type_875 = None
	        view_2282: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_14115: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6669, view_2282);  sub_6669 = view_2282 = None
	        view_2284: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14115, [1280, 1280]);  mul_14115 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22064, [2])
	        full_293: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_146, full_293);  amax_146 = full_293 = None
	        amin_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22064, [2])
	        full_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_146: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_146, full_292);  amin_146 = full_292 = None
	        sub_6683: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_146, minimum_146);  maximum_146 = None
	        div_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6683, 255.0);  sub_6683 = None
	        clamp_min_438: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_292, 1.1920928955078125e-07);  div_292 = None
	        div_293: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_146, clamp_min_438);  minimum_146 = None
	        round_293: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_293);  div_293 = None
	        sub_6689: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_293);  round_293 = None
	        clamp_min_439: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6689, -128);  sub_6689 = None
	        clamp_max_292: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_439, 127);  clamp_min_439 = None
	        view_2290: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_438, [sym_size_int, 1500, 1])
	        reciprocal_146: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2290);  view_2290 = None
	        mul_14184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_146, 1.0);  reciprocal_146 = None
	        mul_14187: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22064, mul_14184);  add_22064 = mul_14184 = None
	        round_294: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14187);  mul_14187 = None
	        convert_element_type_876: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_292, torch.int8);  clamp_max_292 = None
	        view_2291: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_876, [sym_size_int, 1500, 1])
	        add_22451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_294, view_2291);  round_294 = view_2291 = None
	        clamp_min_440: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22451, -128);  add_22451 = None
	        clamp_max_293: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_440, 127);  clamp_min_440 = None
	        convert_element_type_877: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_293, torch.int8);  clamp_max_293 = None
	        view_2295: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_876, [sym_size_int, 1500, 1]);  convert_element_type_876 = None
	        convert_element_type_878: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_877, torch.float32);  convert_element_type_877 = None
	        convert_element_type_879: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2295, torch.float32);  view_2295 = None
	        sub_6709: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_878, convert_element_type_879);  convert_element_type_878 = convert_element_type_879 = None
	        view_2294: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_438, [sym_size_int, 1500, 1]);  clamp_min_438 = None
	        mul_14209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6709, view_2294);  sub_6709 = view_2294 = None
	        view_2297: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2299: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_880: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2297, torch.float32);  view_2297 = None
	        convert_element_type_881: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2299, torch.float32);  view_2299 = None
	        sub_6713: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_880, convert_element_type_881);  convert_element_type_880 = convert_element_type_881 = None
	        view_2298: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_14214: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6713, view_2298);  sub_6713 = view_2298 = None
	        view_2300: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14214, [1280, 1280]);  mul_14214 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_14024: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2269: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14014, [mul_14024, 1280]);  mul_14014 = mul_14024 = None
	        permute_241: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2268, [1, 0]);  view_2268 = None
	        mm_default_39: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2269, permute_241);  view_2269 = permute_241 = None
	        add_tensor_39: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_39, model_audio_tower_layers_24_self_attn_q_proj_bias);  mm_default_39 = model_audio_tower_layers_24_self_attn_q_proj_bias = None
	        view_2270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_39, [sym_size_int, 1500, 1280]);  add_tensor_39 = None
	        mul_14031: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2270, 0.125);  view_2270 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2271: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14031, [sym_size_int, 1500, 20, 64]);  mul_14031 = None
	        permute_242: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2271, [0, 2, 1, 3]);  view_2271 = None
	        clone_194: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_242, memory_format = torch.contiguous_format);  permute_242 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_14118: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2285: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14110, [mul_14118, 1280]);  mul_14110 = mul_14118 = None
	        permute_243: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2284, [1, 0]);  view_2284 = None
	        mm_24: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2285, permute_243);  view_2285 = permute_243 = None
	        view_2286: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_24, [sym_size_int, 1500, 1280]);  mm_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2287: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2286, [sym_size_int, -1, 20, 64]);  view_2286 = None
	        permute_244: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2287, [0, 2, 1, 3]);  view_2287 = None
	        clone_195: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_244, memory_format = torch.contiguous_format);  permute_244 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_14219: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2301: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14209, [mul_14219, 1280]);  mul_14209 = mul_14219 = None
	        permute_245: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2300, [1, 0]);  view_2300 = None
	        mm_default_38: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2301, permute_245);  view_2301 = permute_245 = None
	        add_tensor_38: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_38, model_audio_tower_layers_24_self_attn_v_proj_bias);  mm_default_38 = model_audio_tower_layers_24_self_attn_v_proj_bias = None
	        view_2302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_38, [sym_size_int, 1500, 1280]);  add_tensor_38 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2303: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2302, [sym_size_int, -1, 20, 64]);  view_2302 = None
	        permute_246: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2303, [0, 2, 1, 3]);  view_2303 = None
	        clone_196: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_246, memory_format = torch.contiguous_format);  permute_246 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_24 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_194, clone_195, clone_196, None, False, scale = 1.0);  clone_194 = clone_195 = clone_196 = None
	        getitem_194: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_24[0];  _scaled_dot_product_efficient_attention_24 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_247: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_194, [0, 2, 1, 3]);  getitem_194 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2304: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_247, [sym_size_int, 1500, -1]);  permute_247 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2304, [2])
	        full_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_147, full_295);  amax_147 = full_295 = None
	        amin_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2304, [2])
	        full_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_147: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_147, full_294);  amin_147 = full_294 = None
	        sub_6731: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_147, minimum_147);  maximum_147 = None
	        div_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6731, 255.0);  sub_6731 = None
	        clamp_min_441: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_294, 1.1920928955078125e-07);  div_294 = None
	        div_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_147, clamp_min_441);  minimum_147 = None
	        round_295: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_295);  div_295 = None
	        sub_6737: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_295);  round_295 = None
	        clamp_min_442: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6737, -128);  sub_6737 = None
	        clamp_max_294: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_442, 127);  clamp_min_442 = None
	        view_2307: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_441, [sym_size_int, 1500, 1])
	        reciprocal_147: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2307);  view_2307 = None
	        mul_14289: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_147, 1.0);  reciprocal_147 = None
	        mul_14292: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2304, mul_14289);  view_2304 = mul_14289 = None
	        round_296: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14292);  mul_14292 = None
	        convert_element_type_882: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_294, torch.int8);  clamp_max_294 = None
	        view_2308: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_882, [sym_size_int, 1500, 1])
	        add_22615: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_296, view_2308);  round_296 = view_2308 = None
	        clamp_min_443: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22615, -128);  add_22615 = None
	        clamp_max_295: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_443, 127);  clamp_min_443 = None
	        convert_element_type_883: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_295, torch.int8);  clamp_max_295 = None
	        view_2312: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_882, [sym_size_int, 1500, 1]);  convert_element_type_882 = None
	        convert_element_type_884: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_883, torch.float32);  convert_element_type_883 = None
	        convert_element_type_885: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2312, torch.float32);  view_2312 = None
	        sub_6757: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_884, convert_element_type_885);  convert_element_type_884 = convert_element_type_885 = None
	        view_2311: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_441, [sym_size_int, 1500, 1]);  clamp_min_441 = None
	        mul_14314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6757, view_2311);  sub_6757 = view_2311 = None
	        view_2314: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2316: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_886: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2314, torch.float32);  view_2314 = None
	        convert_element_type_887: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2316, torch.float32);  view_2316 = None
	        sub_6761: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_886, convert_element_type_887);  convert_element_type_886 = convert_element_type_887 = None
	        view_2315: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_14319: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6761, view_2315);  sub_6761 = view_2315 = None
	        view_2317: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14319, [1280, 1280]);  mul_14319 = None
	        mul_14324: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2318: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14314, [mul_14324, 1280]);  mul_14314 = mul_14324 = None
	        permute_248: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2317, [1, 0]);  view_2317 = None
	        mm_default_37: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2318, permute_248);  view_2318 = permute_248 = None
	        add_tensor_37: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_37, model_audio_tower_layers_24_self_attn_out_proj_bias);  mm_default_37 = model_audio_tower_layers_24_self_attn_out_proj_bias = None
	        view_2319: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_37, [sym_size_int, 1500, 1280]);  add_tensor_37 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_22678: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_22058, view_2319);  add_22058 = view_2319 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_198: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_22678, memory_format = torch.contiguous_format)
	        var_mean_49 = torch.ops.aten.var_mean.correction(clone_198, [2], correction = 0, keepdim = True)
	        getitem_198: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_49[0]
	        getitem_199: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_49[1];  var_mean_49 = None
	        sub_6767: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_198, getitem_199);  clone_198 = getitem_199 = None
	        add_22683: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_198, 1e-05);  getitem_198 = None
	        rsqrt_49: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_22683);  add_22683 = None
	        mul_14335: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6767, rsqrt_49);  sub_6767 = rsqrt_49 = None
	        mul_14336: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14335, model_audio_tower_layers_24_final_layer_norm_weight);  mul_14335 = model_audio_tower_layers_24_final_layer_norm_weight = None
	        add_22684: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_14336, model_audio_tower_layers_24_final_layer_norm_bias);  mul_14336 = model_audio_tower_layers_24_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22684, [2])
	        full_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_148, full_297);  amax_148 = full_297 = None
	        amin_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22684, [2])
	        full_296: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_148, full_296);  amin_148 = full_296 = None
	        sub_6778: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_148, minimum_148);  maximum_148 = None
	        div_296: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6778, 255.0);  sub_6778 = None
	        clamp_min_444: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_296, 1.1920928955078125e-07);  div_296 = None
	        div_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_148, clamp_min_444);  minimum_148 = None
	        round_297: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_297);  div_297 = None
	        sub_6784: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_297);  round_297 = None
	        clamp_min_445: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6784, -128);  sub_6784 = None
	        clamp_max_296: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_445, 127);  clamp_min_445 = None
	        view_2322: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_444, [sym_size_int, 1500, 1])
	        reciprocal_148: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2322);  view_2322 = None
	        mul_14384: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_148, 1.0);  reciprocal_148 = None
	        mul_14387: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22684, mul_14384);  add_22684 = mul_14384 = None
	        round_298: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14387);  mul_14387 = None
	        convert_element_type_888: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_296, torch.int8);  clamp_max_296 = None
	        view_2323: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_888, [sym_size_int, 1500, 1])
	        add_22771: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_298, view_2323);  round_298 = view_2323 = None
	        clamp_min_446: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22771, -128);  add_22771 = None
	        clamp_max_297: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_446, 127);  clamp_min_446 = None
	        convert_element_type_889: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_297, torch.int8);  clamp_max_297 = None
	        view_2327: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_888, [sym_size_int, 1500, 1]);  convert_element_type_888 = None
	        convert_element_type_890: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_889, torch.float32);  convert_element_type_889 = None
	        convert_element_type_891: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2327, torch.float32);  view_2327 = None
	        sub_6804: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_890, convert_element_type_891);  convert_element_type_890 = convert_element_type_891 = None
	        view_2326: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_444, [sym_size_int, 1500, 1]);  clamp_min_444 = None
	        mul_14409: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6804, view_2326);  sub_6804 = view_2326 = None
	        view_2329: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_24_fc1_parametrizations_weight_original0 = None
	        view_2331: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_24_fc1_parametrizations_weight_original2 = None
	        convert_element_type_892: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2329, torch.float32);  view_2329 = None
	        convert_element_type_893: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2331, torch.float32);  view_2331 = None
	        sub_6808: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_892, convert_element_type_893);  convert_element_type_892 = convert_element_type_893 = None
	        view_2330: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_24_fc1_parametrizations_weight_original1 = None
	        mul_14414: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6808, view_2330);  sub_6808 = view_2330 = None
	        view_2332: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14414, [5120, 1280]);  mul_14414 = None
	        mul_14419: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2333: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14409, [mul_14419, 1280]);  mul_14409 = mul_14419 = None
	        permute_249: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2332, [1, 0]);  view_2332 = None
	        mm_default_36: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_2333, permute_249);  view_2333 = permute_249 = None
	        add_tensor_36: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_36, model_audio_tower_layers_24_fc1_bias);  mm_default_36 = model_audio_tower_layers_24_fc1_bias = None
	        view_2334: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_36, [sym_size_int, 1500, 5120]);  add_tensor_36 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_14426: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2334, 0.5)
	        mul_14427: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2334, 0.7071067811865476);  view_2334 = None
	        erf_26: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_14427);  mul_14427 = None
	        add_22830: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_26, 1);  erf_26 = None
	        mul_14428: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14426, add_22830);  mul_14426 = add_22830 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_14428, [2])
	        full_299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_149, full_299);  amax_149 = full_299 = None
	        amin_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_14428, [2])
	        full_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_149: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_149, full_298);  amin_149 = full_298 = None
	        sub_6821: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_149, minimum_149);  maximum_149 = None
	        div_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6821, 255.0);  sub_6821 = None
	        clamp_min_447: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_298, 1.1920928955078125e-07);  div_298 = None
	        div_299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_149, clamp_min_447);  minimum_149 = None
	        round_299: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_299);  div_299 = None
	        sub_6827: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_299);  round_299 = None
	        clamp_min_448: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6827, -128);  sub_6827 = None
	        clamp_max_298: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_448, 127);  clamp_min_448 = None
	        view_2337: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_447, [sym_size_int, 1500, 1])
	        reciprocal_149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2337);  view_2337 = None
	        mul_14474: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_149, 1.0);  reciprocal_149 = None
	        mul_14477: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14428, mul_14474);  mul_14428 = mul_14474 = None
	        round_300: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_14477);  mul_14477 = None
	        convert_element_type_894: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_298, torch.int8);  clamp_max_298 = None
	        view_2338: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_894, [sym_size_int, 1500, 1])
	        add_22913: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_300, view_2338);  round_300 = view_2338 = None
	        clamp_min_449: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_22913, -128);  add_22913 = None
	        clamp_max_299: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_449, 127);  clamp_min_449 = None
	        convert_element_type_895: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_299, torch.int8);  clamp_max_299 = None
	        view_2342: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_894, [sym_size_int, 1500, 1]);  convert_element_type_894 = None
	        convert_element_type_896: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_895, torch.float32);  convert_element_type_895 = None
	        convert_element_type_897: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2342, torch.float32);  view_2342 = None
	        sub_6847: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_896, convert_element_type_897);  convert_element_type_896 = convert_element_type_897 = None
	        view_2341: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_447, [sym_size_int, 1500, 1]);  clamp_min_447 = None
	        mul_14499: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6847, view_2341);  sub_6847 = view_2341 = None
	        view_2344: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_24_fc2_parametrizations_weight_original0 = None
	        view_2346: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_24_fc2_parametrizations_weight_original2 = None
	        convert_element_type_898: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2344, torch.float32);  view_2344 = None
	        convert_element_type_899: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2346, torch.float32);  view_2346 = None
	        sub_6851: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_898, convert_element_type_899);  convert_element_type_898 = convert_element_type_899 = None
	        view_2345: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_24_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_24_fc2_parametrizations_weight_original1 = None
	        mul_14504: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6851, view_2345);  sub_6851 = view_2345 = None
	        view_2347: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14504, [1280, 5120]);  mul_14504 = None
	        mul_14509: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2348: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14499, [mul_14509, 5120]);  mul_14499 = mul_14509 = None
	        permute_250: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2347, [1, 0]);  view_2347 = None
	        mm_default_35: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2348, permute_250);  view_2348 = permute_250 = None
	        add_tensor_35: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_35, model_audio_tower_layers_24_fc2_bias);  mm_default_35 = model_audio_tower_layers_24_fc2_bias = None
	        view_2349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_35, [sym_size_int, 1500, 1280]);  add_tensor_35 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_22976: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_22678, view_2349);  add_22678 = view_2349 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_201: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_22976, memory_format = torch.contiguous_format)
	        var_mean_50 = torch.ops.aten.var_mean.correction(clone_201, [2], correction = 0, keepdim = True)
	        getitem_200: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_50[0]
	        getitem_201: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_50[1];  var_mean_50 = None
	        sub_6857: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_201, getitem_201);  clone_201 = getitem_201 = None
	        add_22981: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_200, 1e-05);  getitem_200 = None
	        rsqrt_50: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_22981);  add_22981 = None
	        mul_14520: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6857, rsqrt_50);  sub_6857 = rsqrt_50 = None
	        mul_14521: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14520, model_audio_tower_layers_25_self_attn_layer_norm_weight);  mul_14520 = model_audio_tower_layers_25_self_attn_layer_norm_weight = None
	        add_22982: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_14521, model_audio_tower_layers_25_self_attn_layer_norm_bias);  mul_14521 = model_audio_tower_layers_25_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22982, [2])
	        full_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_150, full_301);  amax_150 = full_301 = None
	        amin_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22982, [2])
	        full_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_150: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_150, full_300);  amin_150 = full_300 = None
	        sub_6868: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_150, minimum_150);  maximum_150 = None
	        div_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6868, 255.0);  sub_6868 = None
	        clamp_min_450: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_300, 1.1920928955078125e-07);  div_300 = None
	        div_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_150, clamp_min_450);  minimum_150 = None
	        round_301: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_301);  div_301 = None
	        sub_6874: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_301);  round_301 = None
	        clamp_min_451: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6874, -128);  sub_6874 = None
	        clamp_max_300: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_451, 127);  clamp_min_451 = None
	        view_2352: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_450, [sym_size_int, 1500, 1])
	        reciprocal_150: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2352);  view_2352 = None
	        mul_14569: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_150, 1.0);  reciprocal_150 = None
	        mul_14572: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22982, mul_14569);  mul_14569 = None
	        round_302: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14572);  mul_14572 = None
	        convert_element_type_900: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_300, torch.int8);  clamp_max_300 = None
	        view_2353: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_900, [sym_size_int, 1500, 1])
	        add_23069: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_302, view_2353);  round_302 = view_2353 = None
	        clamp_min_452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23069, -128);  add_23069 = None
	        clamp_max_301: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_452, 127);  clamp_min_452 = None
	        convert_element_type_901: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_301, torch.int8);  clamp_max_301 = None
	        view_2357: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_900, [sym_size_int, 1500, 1]);  convert_element_type_900 = None
	        convert_element_type_902: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_901, torch.float32);  convert_element_type_901 = None
	        convert_element_type_903: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2357, torch.float32);  view_2357 = None
	        sub_6894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_902, convert_element_type_903);  convert_element_type_902 = convert_element_type_903 = None
	        view_2356: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_450, [sym_size_int, 1500, 1]);  clamp_min_450 = None
	        mul_14594: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6894, view_2356);  sub_6894 = view_2356 = None
	        view_2359: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2361: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_904: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2359, torch.float32);  view_2359 = None
	        convert_element_type_905: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2361, torch.float32);  view_2361 = None
	        sub_6898: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_904, convert_element_type_905);  convert_element_type_904 = convert_element_type_905 = None
	        view_2360: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_14599: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6898, view_2360);  sub_6898 = view_2360 = None
	        view_2362: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14599, [1280, 1280]);  mul_14599 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22982, [2])
	        full_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_151, full_303);  amax_151 = full_303 = None
	        amin_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22982, [2])
	        full_302: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_151: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_151, full_302);  amin_151 = full_302 = None
	        sub_6913: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_151, minimum_151);  maximum_151 = None
	        div_302: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6913, 255.0);  sub_6913 = None
	        clamp_min_453: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_302, 1.1920928955078125e-07);  div_302 = None
	        div_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_151, clamp_min_453);  minimum_151 = None
	        round_303: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_303);  div_303 = None
	        sub_6919: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_303);  round_303 = None
	        clamp_min_454: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6919, -128);  sub_6919 = None
	        clamp_max_302: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_454, 127);  clamp_min_454 = None
	        view_2368: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_453, [sym_size_int, 1500, 1])
	        reciprocal_151: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2368);  view_2368 = None
	        mul_14665: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_151, 1.0);  reciprocal_151 = None
	        mul_14668: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22982, mul_14665);  mul_14665 = None
	        round_304: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14668);  mul_14668 = None
	        convert_element_type_906: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_302, torch.int8);  clamp_max_302 = None
	        view_2369: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_906, [sym_size_int, 1500, 1])
	        add_23221: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_304, view_2369);  round_304 = view_2369 = None
	        clamp_min_455: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23221, -128);  add_23221 = None
	        clamp_max_303: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_455, 127);  clamp_min_455 = None
	        convert_element_type_907: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_303, torch.int8);  clamp_max_303 = None
	        view_2373: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_906, [sym_size_int, 1500, 1]);  convert_element_type_906 = None
	        convert_element_type_908: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_907, torch.float32);  convert_element_type_907 = None
	        convert_element_type_909: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2373, torch.float32);  view_2373 = None
	        sub_6939: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_908, convert_element_type_909);  convert_element_type_908 = convert_element_type_909 = None
	        view_2372: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_453, [sym_size_int, 1500, 1]);  clamp_min_453 = None
	        mul_14690: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6939, view_2372);  sub_6939 = view_2372 = None
	        view_2375: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2377: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_910: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2375, torch.float32);  view_2375 = None
	        convert_element_type_911: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2377, torch.float32);  view_2377 = None
	        sub_6943: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_910, convert_element_type_911);  convert_element_type_910 = convert_element_type_911 = None
	        view_2376: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_14695: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6943, view_2376);  sub_6943 = view_2376 = None
	        view_2378: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14695, [1280, 1280]);  mul_14695 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_22982, [2])
	        full_305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_152, full_305);  amax_152 = full_305 = None
	        amin_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_22982, [2])
	        full_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_152: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_152, full_304);  amin_152 = full_304 = None
	        sub_6957: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_152, minimum_152);  maximum_152 = None
	        div_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_6957, 255.0);  sub_6957 = None
	        clamp_min_456: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_304, 1.1920928955078125e-07);  div_304 = None
	        div_305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_152, clamp_min_456);  minimum_152 = None
	        round_305: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_305);  div_305 = None
	        sub_6963: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_305);  round_305 = None
	        clamp_min_457: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_6963, -128);  sub_6963 = None
	        clamp_max_304: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_457, 127);  clamp_min_457 = None
	        view_2384: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_456, [sym_size_int, 1500, 1])
	        reciprocal_152: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2384);  view_2384 = None
	        mul_14764: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_152, 1.0);  reciprocal_152 = None
	        mul_14767: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_22982, mul_14764);  add_22982 = mul_14764 = None
	        round_306: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14767);  mul_14767 = None
	        convert_element_type_912: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_304, torch.int8);  clamp_max_304 = None
	        view_2385: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_912, [sym_size_int, 1500, 1])
	        add_23369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_306, view_2385);  round_306 = view_2385 = None
	        clamp_min_458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23369, -128);  add_23369 = None
	        clamp_max_305: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_458, 127);  clamp_min_458 = None
	        convert_element_type_913: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_305, torch.int8);  clamp_max_305 = None
	        view_2389: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_912, [sym_size_int, 1500, 1]);  convert_element_type_912 = None
	        convert_element_type_914: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_913, torch.float32);  convert_element_type_913 = None
	        convert_element_type_915: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2389, torch.float32);  view_2389 = None
	        sub_6983: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_914, convert_element_type_915);  convert_element_type_914 = convert_element_type_915 = None
	        view_2388: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_456, [sym_size_int, 1500, 1]);  clamp_min_456 = None
	        mul_14789: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6983, view_2388);  sub_6983 = view_2388 = None
	        view_2391: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2393: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_916: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2391, torch.float32);  view_2391 = None
	        convert_element_type_917: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2393, torch.float32);  view_2393 = None
	        sub_6987: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_916, convert_element_type_917);  convert_element_type_916 = convert_element_type_917 = None
	        view_2392: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_14794: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_6987, view_2392);  sub_6987 = view_2392 = None
	        view_2394: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14794, [1280, 1280]);  mul_14794 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_14604: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2363: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14594, [mul_14604, 1280]);  mul_14594 = mul_14604 = None
	        permute_251: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2362, [1, 0]);  view_2362 = None
	        mm_default_34: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2363, permute_251);  view_2363 = permute_251 = None
	        add_tensor_34: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_34, model_audio_tower_layers_25_self_attn_q_proj_bias);  mm_default_34 = model_audio_tower_layers_25_self_attn_q_proj_bias = None
	        view_2364: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_34, [sym_size_int, 1500, 1280]);  add_tensor_34 = None
	        mul_14611: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2364, 0.125);  view_2364 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2365: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14611, [sym_size_int, 1500, 20, 64]);  mul_14611 = None
	        permute_252: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2365, [0, 2, 1, 3]);  view_2365 = None
	        clone_202: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_252, memory_format = torch.contiguous_format);  permute_252 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_14698: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2379: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14690, [mul_14698, 1280]);  mul_14690 = mul_14698 = None
	        permute_253: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2378, [1, 0]);  view_2378 = None
	        mm_25: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2379, permute_253);  view_2379 = permute_253 = None
	        view_2380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_25, [sym_size_int, 1500, 1280]);  mm_25 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2381: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2380, [sym_size_int, -1, 20, 64]);  view_2380 = None
	        permute_254: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2381, [0, 2, 1, 3]);  view_2381 = None
	        clone_203: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_254, memory_format = torch.contiguous_format);  permute_254 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_14799: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2395: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14789, [mul_14799, 1280]);  mul_14789 = mul_14799 = None
	        permute_255: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2394, [1, 0]);  view_2394 = None
	        mm_default_33: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2395, permute_255);  view_2395 = permute_255 = None
	        add_tensor_33: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_33, model_audio_tower_layers_25_self_attn_v_proj_bias);  mm_default_33 = model_audio_tower_layers_25_self_attn_v_proj_bias = None
	        view_2396: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_33, [sym_size_int, 1500, 1280]);  add_tensor_33 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2397: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2396, [sym_size_int, -1, 20, 64]);  view_2396 = None
	        permute_256: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2397, [0, 2, 1, 3]);  view_2397 = None
	        clone_204: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_256, memory_format = torch.contiguous_format);  permute_256 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_25 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_202, clone_203, clone_204, None, False, scale = 1.0);  clone_202 = clone_203 = clone_204 = None
	        getitem_202: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_25[0];  _scaled_dot_product_efficient_attention_25 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_257: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_202, [0, 2, 1, 3]);  getitem_202 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2398: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_257, [sym_size_int, 1500, -1]);  permute_257 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2398, [2])
	        full_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_153, full_307);  amax_153 = full_307 = None
	        amin_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2398, [2])
	        full_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_153: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_153, full_306);  amin_153 = full_306 = None
	        sub_7005: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_153, minimum_153);  maximum_153 = None
	        div_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7005, 255.0);  sub_7005 = None
	        clamp_min_459: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_306, 1.1920928955078125e-07);  div_306 = None
	        div_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_153, clamp_min_459);  minimum_153 = None
	        round_307: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_307);  div_307 = None
	        sub_7011: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_307);  round_307 = None
	        clamp_min_460: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7011, -128);  sub_7011 = None
	        clamp_max_306: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_460, 127);  clamp_min_460 = None
	        view_2401: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_459, [sym_size_int, 1500, 1])
	        reciprocal_153: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2401);  view_2401 = None
	        mul_14869: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_153, 1.0);  reciprocal_153 = None
	        mul_14872: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2398, mul_14869);  view_2398 = mul_14869 = None
	        round_308: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14872);  mul_14872 = None
	        convert_element_type_918: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_306, torch.int8);  clamp_max_306 = None
	        view_2402: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_918, [sym_size_int, 1500, 1])
	        add_23533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_308, view_2402);  round_308 = view_2402 = None
	        clamp_min_461: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23533, -128);  add_23533 = None
	        clamp_max_307: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_461, 127);  clamp_min_461 = None
	        convert_element_type_919: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_307, torch.int8);  clamp_max_307 = None
	        view_2406: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_918, [sym_size_int, 1500, 1]);  convert_element_type_918 = None
	        convert_element_type_920: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_919, torch.float32);  convert_element_type_919 = None
	        convert_element_type_921: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2406, torch.float32);  view_2406 = None
	        sub_7031: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_920, convert_element_type_921);  convert_element_type_920 = convert_element_type_921 = None
	        view_2405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_459, [sym_size_int, 1500, 1]);  clamp_min_459 = None
	        mul_14894: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7031, view_2405);  sub_7031 = view_2405 = None
	        view_2408: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2410: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_922: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2408, torch.float32);  view_2408 = None
	        convert_element_type_923: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2410, torch.float32);  view_2410 = None
	        sub_7035: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_922, convert_element_type_923);  convert_element_type_922 = convert_element_type_923 = None
	        view_2409: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_14899: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7035, view_2409);  sub_7035 = view_2409 = None
	        view_2411: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14899, [1280, 1280]);  mul_14899 = None
	        mul_14904: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2412: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14894, [mul_14904, 1280]);  mul_14894 = mul_14904 = None
	        permute_258: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2411, [1, 0]);  view_2411 = None
	        mm_default_32: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2412, permute_258);  view_2412 = permute_258 = None
	        add_tensor_32: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_32, model_audio_tower_layers_25_self_attn_out_proj_bias);  mm_default_32 = model_audio_tower_layers_25_self_attn_out_proj_bias = None
	        view_2413: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_32, [sym_size_int, 1500, 1280]);  add_tensor_32 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_23596: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_22976, view_2413);  add_22976 = view_2413 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_206: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_23596, memory_format = torch.contiguous_format)
	        var_mean_51 = torch.ops.aten.var_mean.correction(clone_206, [2], correction = 0, keepdim = True)
	        getitem_206: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_51[0]
	        getitem_207: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_51[1];  var_mean_51 = None
	        sub_7041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_206, getitem_207);  clone_206 = getitem_207 = None
	        add_23601: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_206, 1e-05);  getitem_206 = None
	        rsqrt_51: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_23601);  add_23601 = None
	        mul_14915: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7041, rsqrt_51);  sub_7041 = rsqrt_51 = None
	        mul_14916: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_14915, model_audio_tower_layers_25_final_layer_norm_weight);  mul_14915 = model_audio_tower_layers_25_final_layer_norm_weight = None
	        add_23602: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_14916, model_audio_tower_layers_25_final_layer_norm_bias);  mul_14916 = model_audio_tower_layers_25_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_23602, [2])
	        full_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_154, full_309);  amax_154 = full_309 = None
	        amin_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_23602, [2])
	        full_308: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_154, full_308);  amin_154 = full_308 = None
	        sub_7052: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_154, minimum_154);  maximum_154 = None
	        div_308: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7052, 255.0);  sub_7052 = None
	        clamp_min_462: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_308, 1.1920928955078125e-07);  div_308 = None
	        div_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_154, clamp_min_462);  minimum_154 = None
	        round_309: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_309);  div_309 = None
	        sub_7058: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_309);  round_309 = None
	        clamp_min_463: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7058, -128);  sub_7058 = None
	        clamp_max_308: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_463, 127);  clamp_min_463 = None
	        view_2416: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_462, [sym_size_int, 1500, 1])
	        reciprocal_154: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2416);  view_2416 = None
	        mul_14964: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_154, 1.0);  reciprocal_154 = None
	        mul_14967: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_23602, mul_14964);  add_23602 = mul_14964 = None
	        round_310: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_14967);  mul_14967 = None
	        convert_element_type_924: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_308, torch.int8);  clamp_max_308 = None
	        view_2417: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_924, [sym_size_int, 1500, 1])
	        add_23689: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_310, view_2417);  round_310 = view_2417 = None
	        clamp_min_464: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23689, -128);  add_23689 = None
	        clamp_max_309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_464, 127);  clamp_min_464 = None
	        convert_element_type_925: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_309, torch.int8);  clamp_max_309 = None
	        view_2421: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_924, [sym_size_int, 1500, 1]);  convert_element_type_924 = None
	        convert_element_type_926: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_925, torch.float32);  convert_element_type_925 = None
	        convert_element_type_927: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2421, torch.float32);  view_2421 = None
	        sub_7078: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_926, convert_element_type_927);  convert_element_type_926 = convert_element_type_927 = None
	        view_2420: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_462, [sym_size_int, 1500, 1]);  clamp_min_462 = None
	        mul_14989: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7078, view_2420);  sub_7078 = view_2420 = None
	        view_2423: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_25_fc1_parametrizations_weight_original0 = None
	        view_2425: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_25_fc1_parametrizations_weight_original2 = None
	        convert_element_type_928: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2423, torch.float32);  view_2423 = None
	        convert_element_type_929: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2425, torch.float32);  view_2425 = None
	        sub_7082: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_928, convert_element_type_929);  convert_element_type_928 = convert_element_type_929 = None
	        view_2424: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_25_fc1_parametrizations_weight_original1 = None
	        mul_14994: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7082, view_2424);  sub_7082 = view_2424 = None
	        view_2426: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14994, [5120, 1280]);  mul_14994 = None
	        mul_14999: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2427: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_14989, [mul_14999, 1280]);  mul_14989 = mul_14999 = None
	        permute_259: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2426, [1, 0]);  view_2426 = None
	        mm_default_31: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_2427, permute_259);  view_2427 = permute_259 = None
	        add_tensor_31: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_31, model_audio_tower_layers_25_fc1_bias);  mm_default_31 = model_audio_tower_layers_25_fc1_bias = None
	        view_2428: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_31, [sym_size_int, 1500, 5120]);  add_tensor_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_15006: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2428, 0.5)
	        mul_15007: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2428, 0.7071067811865476);  view_2428 = None
	        erf_27: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_15007);  mul_15007 = None
	        add_23748: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_27, 1);  erf_27 = None
	        mul_15008: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15006, add_23748);  mul_15006 = add_23748 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_15008, [2])
	        full_311: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_155, full_311);  amax_155 = full_311 = None
	        amin_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_15008, [2])
	        full_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_155: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_155, full_310);  amin_155 = full_310 = None
	        sub_7095: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_155, minimum_155);  maximum_155 = None
	        div_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7095, 255.0);  sub_7095 = None
	        clamp_min_465: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_310, 1.1920928955078125e-07);  div_310 = None
	        div_311: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_155, clamp_min_465);  minimum_155 = None
	        round_311: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_311);  div_311 = None
	        sub_7101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_311);  round_311 = None
	        clamp_min_466: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7101, -128);  sub_7101 = None
	        clamp_max_310: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_466, 127);  clamp_min_466 = None
	        view_2431: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_465, [sym_size_int, 1500, 1])
	        reciprocal_155: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2431);  view_2431 = None
	        mul_15054: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_155, 1.0);  reciprocal_155 = None
	        mul_15057: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15008, mul_15054);  mul_15008 = mul_15054 = None
	        round_312: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_15057);  mul_15057 = None
	        convert_element_type_930: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_310, torch.int8);  clamp_max_310 = None
	        view_2432: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_930, [sym_size_int, 1500, 1])
	        add_23831: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_312, view_2432);  round_312 = view_2432 = None
	        clamp_min_467: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23831, -128);  add_23831 = None
	        clamp_max_311: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_467, 127);  clamp_min_467 = None
	        convert_element_type_931: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_311, torch.int8);  clamp_max_311 = None
	        view_2436: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_930, [sym_size_int, 1500, 1]);  convert_element_type_930 = None
	        convert_element_type_932: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_931, torch.float32);  convert_element_type_931 = None
	        convert_element_type_933: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2436, torch.float32);  view_2436 = None
	        sub_7121: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_932, convert_element_type_933);  convert_element_type_932 = convert_element_type_933 = None
	        view_2435: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_465, [sym_size_int, 1500, 1]);  clamp_min_465 = None
	        mul_15079: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7121, view_2435);  sub_7121 = view_2435 = None
	        view_2438: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_25_fc2_parametrizations_weight_original0 = None
	        view_2440: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_25_fc2_parametrizations_weight_original2 = None
	        convert_element_type_934: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2438, torch.float32);  view_2438 = None
	        convert_element_type_935: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2440, torch.float32);  view_2440 = None
	        sub_7125: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_934, convert_element_type_935);  convert_element_type_934 = convert_element_type_935 = None
	        view_2439: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_25_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_25_fc2_parametrizations_weight_original1 = None
	        mul_15084: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7125, view_2439);  sub_7125 = view_2439 = None
	        view_2441: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15084, [1280, 5120]);  mul_15084 = None
	        mul_15089: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2442: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15079, [mul_15089, 5120]);  mul_15079 = mul_15089 = None
	        permute_260: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2441, [1, 0]);  view_2441 = None
	        mm_default_30: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2442, permute_260);  view_2442 = permute_260 = None
	        add_tensor_30: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_30, model_audio_tower_layers_25_fc2_bias);  mm_default_30 = model_audio_tower_layers_25_fc2_bias = None
	        view_2443: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_30, [sym_size_int, 1500, 1280]);  add_tensor_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_23894: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_23596, view_2443);  add_23596 = view_2443 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_209: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_23894, memory_format = torch.contiguous_format)
	        var_mean_52 = torch.ops.aten.var_mean.correction(clone_209, [2], correction = 0, keepdim = True)
	        getitem_208: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_52[0]
	        getitem_209: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_52[1];  var_mean_52 = None
	        sub_7131: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_209, getitem_209);  clone_209 = getitem_209 = None
	        add_23899: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_208, 1e-05);  getitem_208 = None
	        rsqrt_52: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_23899);  add_23899 = None
	        mul_15100: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7131, rsqrt_52);  sub_7131 = rsqrt_52 = None
	        mul_15101: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15100, model_audio_tower_layers_26_self_attn_layer_norm_weight);  mul_15100 = model_audio_tower_layers_26_self_attn_layer_norm_weight = None
	        add_23900: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_15101, model_audio_tower_layers_26_self_attn_layer_norm_bias);  mul_15101 = model_audio_tower_layers_26_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_23900, [2])
	        full_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_156, full_313);  amax_156 = full_313 = None
	        amin_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_23900, [2])
	        full_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_156: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_156, full_312);  amin_156 = full_312 = None
	        sub_7142: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_156, minimum_156);  maximum_156 = None
	        div_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7142, 255.0);  sub_7142 = None
	        clamp_min_468: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_312, 1.1920928955078125e-07);  div_312 = None
	        div_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_156, clamp_min_468);  minimum_156 = None
	        round_313: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_313);  div_313 = None
	        sub_7148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_313);  round_313 = None
	        clamp_min_469: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7148, -128);  sub_7148 = None
	        clamp_max_312: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_469, 127);  clamp_min_469 = None
	        view_2446: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_468, [sym_size_int, 1500, 1])
	        reciprocal_156: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2446);  view_2446 = None
	        mul_15149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_156, 1.0);  reciprocal_156 = None
	        mul_15152: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_23900, mul_15149);  mul_15149 = None
	        round_314: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15152);  mul_15152 = None
	        convert_element_type_936: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_312, torch.int8);  clamp_max_312 = None
	        view_2447: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_936, [sym_size_int, 1500, 1])
	        add_23987: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_314, view_2447);  round_314 = view_2447 = None
	        clamp_min_470: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_23987, -128);  add_23987 = None
	        clamp_max_313: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_470, 127);  clamp_min_470 = None
	        convert_element_type_937: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_313, torch.int8);  clamp_max_313 = None
	        view_2451: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_936, [sym_size_int, 1500, 1]);  convert_element_type_936 = None
	        convert_element_type_938: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_937, torch.float32);  convert_element_type_937 = None
	        convert_element_type_939: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2451, torch.float32);  view_2451 = None
	        sub_7168: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_938, convert_element_type_939);  convert_element_type_938 = convert_element_type_939 = None
	        view_2450: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_468, [sym_size_int, 1500, 1]);  clamp_min_468 = None
	        mul_15174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7168, view_2450);  sub_7168 = view_2450 = None
	        view_2453: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2455: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_940: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2453, torch.float32);  view_2453 = None
	        convert_element_type_941: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2455, torch.float32);  view_2455 = None
	        sub_7172: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_940, convert_element_type_941);  convert_element_type_940 = convert_element_type_941 = None
	        view_2454: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_15179: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7172, view_2454);  sub_7172 = view_2454 = None
	        view_2456: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15179, [1280, 1280]);  mul_15179 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_23900, [2])
	        full_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_157, full_315);  amax_157 = full_315 = None
	        amin_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_23900, [2])
	        full_314: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_157: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_157, full_314);  amin_157 = full_314 = None
	        sub_7187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_157, minimum_157);  maximum_157 = None
	        div_314: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7187, 255.0);  sub_7187 = None
	        clamp_min_471: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_314, 1.1920928955078125e-07);  div_314 = None
	        div_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_157, clamp_min_471);  minimum_157 = None
	        round_315: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_315);  div_315 = None
	        sub_7193: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_315);  round_315 = None
	        clamp_min_472: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7193, -128);  sub_7193 = None
	        clamp_max_314: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_472, 127);  clamp_min_472 = None
	        view_2462: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_471, [sym_size_int, 1500, 1])
	        reciprocal_157: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2462);  view_2462 = None
	        mul_15245: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_157, 1.0);  reciprocal_157 = None
	        mul_15248: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_23900, mul_15245);  mul_15245 = None
	        round_316: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15248);  mul_15248 = None
	        convert_element_type_942: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_314, torch.int8);  clamp_max_314 = None
	        view_2463: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_942, [sym_size_int, 1500, 1])
	        add_24139: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_316, view_2463);  round_316 = view_2463 = None
	        clamp_min_473: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24139, -128);  add_24139 = None
	        clamp_max_315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_473, 127);  clamp_min_473 = None
	        convert_element_type_943: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_315, torch.int8);  clamp_max_315 = None
	        view_2467: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_942, [sym_size_int, 1500, 1]);  convert_element_type_942 = None
	        convert_element_type_944: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_943, torch.float32);  convert_element_type_943 = None
	        convert_element_type_945: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2467, torch.float32);  view_2467 = None
	        sub_7213: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_944, convert_element_type_945);  convert_element_type_944 = convert_element_type_945 = None
	        view_2466: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_471, [sym_size_int, 1500, 1]);  clamp_min_471 = None
	        mul_15270: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7213, view_2466);  sub_7213 = view_2466 = None
	        view_2469: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2471: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_946: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2469, torch.float32);  view_2469 = None
	        convert_element_type_947: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2471, torch.float32);  view_2471 = None
	        sub_7217: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_946, convert_element_type_947);  convert_element_type_946 = convert_element_type_947 = None
	        view_2470: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_15275: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7217, view_2470);  sub_7217 = view_2470 = None
	        view_2472: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15275, [1280, 1280]);  mul_15275 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_23900, [2])
	        full_317: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_158, full_317);  amax_158 = full_317 = None
	        amin_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_23900, [2])
	        full_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_158: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_158, full_316);  amin_158 = full_316 = None
	        sub_7231: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_158, minimum_158);  maximum_158 = None
	        div_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7231, 255.0);  sub_7231 = None
	        clamp_min_474: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_316, 1.1920928955078125e-07);  div_316 = None
	        div_317: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_158, clamp_min_474);  minimum_158 = None
	        round_317: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_317);  div_317 = None
	        sub_7237: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_317);  round_317 = None
	        clamp_min_475: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7237, -128);  sub_7237 = None
	        clamp_max_316: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_475, 127);  clamp_min_475 = None
	        view_2478: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_474, [sym_size_int, 1500, 1])
	        reciprocal_158: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2478);  view_2478 = None
	        mul_15344: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_158, 1.0);  reciprocal_158 = None
	        mul_15347: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_23900, mul_15344);  add_23900 = mul_15344 = None
	        round_318: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15347);  mul_15347 = None
	        convert_element_type_948: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_316, torch.int8);  clamp_max_316 = None
	        view_2479: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_948, [sym_size_int, 1500, 1])
	        add_24287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_318, view_2479);  round_318 = view_2479 = None
	        clamp_min_476: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24287, -128);  add_24287 = None
	        clamp_max_317: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_476, 127);  clamp_min_476 = None
	        convert_element_type_949: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_317, torch.int8);  clamp_max_317 = None
	        view_2483: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_948, [sym_size_int, 1500, 1]);  convert_element_type_948 = None
	        convert_element_type_950: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_949, torch.float32);  convert_element_type_949 = None
	        convert_element_type_951: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2483, torch.float32);  view_2483 = None
	        sub_7257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_950, convert_element_type_951);  convert_element_type_950 = convert_element_type_951 = None
	        view_2482: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_474, [sym_size_int, 1500, 1]);  clamp_min_474 = None
	        mul_15369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7257, view_2482);  sub_7257 = view_2482 = None
	        view_2485: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2487: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_952: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2485, torch.float32);  view_2485 = None
	        convert_element_type_953: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2487, torch.float32);  view_2487 = None
	        sub_7261: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_952, convert_element_type_953);  convert_element_type_952 = convert_element_type_953 = None
	        view_2486: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_15374: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7261, view_2486);  sub_7261 = view_2486 = None
	        view_2488: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15374, [1280, 1280]);  mul_15374 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_15184: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2457: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15174, [mul_15184, 1280]);  mul_15174 = mul_15184 = None
	        permute_261: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2456, [1, 0]);  view_2456 = None
	        mm_default_29: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2457, permute_261);  view_2457 = permute_261 = None
	        add_tensor_29: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_29, model_audio_tower_layers_26_self_attn_q_proj_bias);  mm_default_29 = model_audio_tower_layers_26_self_attn_q_proj_bias = None
	        view_2458: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_29, [sym_size_int, 1500, 1280]);  add_tensor_29 = None
	        mul_15191: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2458, 0.125);  view_2458 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2459: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15191, [sym_size_int, 1500, 20, 64]);  mul_15191 = None
	        permute_262: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2459, [0, 2, 1, 3]);  view_2459 = None
	        clone_210: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_262, memory_format = torch.contiguous_format);  permute_262 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_15278: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2473: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15270, [mul_15278, 1280]);  mul_15270 = mul_15278 = None
	        permute_263: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2472, [1, 0]);  view_2472 = None
	        mm_26: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2473, permute_263);  view_2473 = permute_263 = None
	        view_2474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_26, [sym_size_int, 1500, 1280]);  mm_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2475: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2474, [sym_size_int, -1, 20, 64]);  view_2474 = None
	        permute_264: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2475, [0, 2, 1, 3]);  view_2475 = None
	        clone_211: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_264, memory_format = torch.contiguous_format);  permute_264 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_15379: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2489: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15369, [mul_15379, 1280]);  mul_15369 = mul_15379 = None
	        permute_265: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2488, [1, 0]);  view_2488 = None
	        mm_default_28: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2489, permute_265);  view_2489 = permute_265 = None
	        add_tensor_28: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_28, model_audio_tower_layers_26_self_attn_v_proj_bias);  mm_default_28 = model_audio_tower_layers_26_self_attn_v_proj_bias = None
	        view_2490: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_28, [sym_size_int, 1500, 1280]);  add_tensor_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2491: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2490, [sym_size_int, -1, 20, 64]);  view_2490 = None
	        permute_266: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2491, [0, 2, 1, 3]);  view_2491 = None
	        clone_212: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_266, memory_format = torch.contiguous_format);  permute_266 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_26 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_210, clone_211, clone_212, None, False, scale = 1.0);  clone_210 = clone_211 = clone_212 = None
	        getitem_210: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_26[0];  _scaled_dot_product_efficient_attention_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_267: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_210, [0, 2, 1, 3]);  getitem_210 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2492: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_267, [sym_size_int, 1500, -1]);  permute_267 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2492, [2])
	        full_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_159, full_319);  amax_159 = full_319 = None
	        amin_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2492, [2])
	        full_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_159: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_159, full_318);  amin_159 = full_318 = None
	        sub_7279: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_159, minimum_159);  maximum_159 = None
	        div_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7279, 255.0);  sub_7279 = None
	        clamp_min_477: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_318, 1.1920928955078125e-07);  div_318 = None
	        div_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_159, clamp_min_477);  minimum_159 = None
	        round_319: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_319);  div_319 = None
	        sub_7285: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_319);  round_319 = None
	        clamp_min_478: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7285, -128);  sub_7285 = None
	        clamp_max_318: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_478, 127);  clamp_min_478 = None
	        view_2495: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_477, [sym_size_int, 1500, 1])
	        reciprocal_159: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2495);  view_2495 = None
	        mul_15449: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_159, 1.0);  reciprocal_159 = None
	        mul_15452: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2492, mul_15449);  view_2492 = mul_15449 = None
	        round_320: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15452);  mul_15452 = None
	        convert_element_type_954: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_318, torch.int8);  clamp_max_318 = None
	        view_2496: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_954, [sym_size_int, 1500, 1])
	        add_24451: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_320, view_2496);  round_320 = view_2496 = None
	        clamp_min_479: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24451, -128);  add_24451 = None
	        clamp_max_319: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_479, 127);  clamp_min_479 = None
	        convert_element_type_955: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_319, torch.int8);  clamp_max_319 = None
	        view_2500: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_954, [sym_size_int, 1500, 1]);  convert_element_type_954 = None
	        convert_element_type_956: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_955, torch.float32);  convert_element_type_955 = None
	        convert_element_type_957: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2500, torch.float32);  view_2500 = None
	        sub_7305: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_956, convert_element_type_957);  convert_element_type_956 = convert_element_type_957 = None
	        view_2499: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_477, [sym_size_int, 1500, 1]);  clamp_min_477 = None
	        mul_15474: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7305, view_2499);  sub_7305 = view_2499 = None
	        view_2502: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2504: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_958: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2502, torch.float32);  view_2502 = None
	        convert_element_type_959: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2504, torch.float32);  view_2504 = None
	        sub_7309: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_958, convert_element_type_959);  convert_element_type_958 = convert_element_type_959 = None
	        view_2503: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_15479: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7309, view_2503);  sub_7309 = view_2503 = None
	        view_2505: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15479, [1280, 1280]);  mul_15479 = None
	        mul_15484: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2506: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15474, [mul_15484, 1280]);  mul_15474 = mul_15484 = None
	        permute_268: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2505, [1, 0]);  view_2505 = None
	        mm_default_27: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2506, permute_268);  view_2506 = permute_268 = None
	        add_tensor_27: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_27, model_audio_tower_layers_26_self_attn_out_proj_bias);  mm_default_27 = model_audio_tower_layers_26_self_attn_out_proj_bias = None
	        view_2507: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_27, [sym_size_int, 1500, 1280]);  add_tensor_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_24514: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_23894, view_2507);  add_23894 = view_2507 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_24514, memory_format = torch.contiguous_format)
	        var_mean_53 = torch.ops.aten.var_mean.correction(clone_214, [2], correction = 0, keepdim = True)
	        getitem_214: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_53[0]
	        getitem_215: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_53[1];  var_mean_53 = None
	        sub_7315: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_214, getitem_215);  clone_214 = getitem_215 = None
	        add_24519: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_214, 1e-05);  getitem_214 = None
	        rsqrt_53: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_24519);  add_24519 = None
	        mul_15495: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7315, rsqrt_53);  sub_7315 = rsqrt_53 = None
	        mul_15496: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15495, model_audio_tower_layers_26_final_layer_norm_weight);  mul_15495 = model_audio_tower_layers_26_final_layer_norm_weight = None
	        add_24520: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_15496, model_audio_tower_layers_26_final_layer_norm_bias);  mul_15496 = model_audio_tower_layers_26_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_24520, [2])
	        full_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_160, full_321);  amax_160 = full_321 = None
	        amin_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_24520, [2])
	        full_320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_160: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_160, full_320);  amin_160 = full_320 = None
	        sub_7326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_160, minimum_160);  maximum_160 = None
	        div_320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7326, 255.0);  sub_7326 = None
	        clamp_min_480: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_320, 1.1920928955078125e-07);  div_320 = None
	        div_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_160, clamp_min_480);  minimum_160 = None
	        round_321: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_321);  div_321 = None
	        sub_7332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_321);  round_321 = None
	        clamp_min_481: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7332, -128);  sub_7332 = None
	        clamp_max_320: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_481, 127);  clamp_min_481 = None
	        view_2510: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_480, [sym_size_int, 1500, 1])
	        reciprocal_160: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2510);  view_2510 = None
	        mul_15544: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_160, 1.0);  reciprocal_160 = None
	        mul_15547: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_24520, mul_15544);  add_24520 = mul_15544 = None
	        round_322: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15547);  mul_15547 = None
	        convert_element_type_960: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_320, torch.int8);  clamp_max_320 = None
	        view_2511: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_960, [sym_size_int, 1500, 1])
	        add_24607: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_322, view_2511);  round_322 = view_2511 = None
	        clamp_min_482: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24607, -128);  add_24607 = None
	        clamp_max_321: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_482, 127);  clamp_min_482 = None
	        convert_element_type_961: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_321, torch.int8);  clamp_max_321 = None
	        view_2515: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_960, [sym_size_int, 1500, 1]);  convert_element_type_960 = None
	        convert_element_type_962: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_961, torch.float32);  convert_element_type_961 = None
	        convert_element_type_963: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2515, torch.float32);  view_2515 = None
	        sub_7352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_962, convert_element_type_963);  convert_element_type_962 = convert_element_type_963 = None
	        view_2514: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_480, [sym_size_int, 1500, 1]);  clamp_min_480 = None
	        mul_15569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7352, view_2514);  sub_7352 = view_2514 = None
	        view_2517: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_26_fc1_parametrizations_weight_original0 = None
	        view_2519: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_26_fc1_parametrizations_weight_original2 = None
	        convert_element_type_964: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2517, torch.float32);  view_2517 = None
	        convert_element_type_965: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2519, torch.float32);  view_2519 = None
	        sub_7356: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_964, convert_element_type_965);  convert_element_type_964 = convert_element_type_965 = None
	        view_2518: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_26_fc1_parametrizations_weight_original1 = None
	        mul_15574: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7356, view_2518);  sub_7356 = view_2518 = None
	        view_2520: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15574, [5120, 1280]);  mul_15574 = None
	        mul_15579: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2521: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15569, [mul_15579, 1280]);  mul_15569 = mul_15579 = None
	        permute_269: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2520, [1, 0]);  view_2520 = None
	        mm_default_26: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_2521, permute_269);  view_2521 = permute_269 = None
	        add_tensor_26: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_26, model_audio_tower_layers_26_fc1_bias);  mm_default_26 = model_audio_tower_layers_26_fc1_bias = None
	        view_2522: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_26, [sym_size_int, 1500, 5120]);  add_tensor_26 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_15586: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2522, 0.5)
	        mul_15587: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2522, 0.7071067811865476);  view_2522 = None
	        erf_28: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_15587);  mul_15587 = None
	        add_24666: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_28, 1);  erf_28 = None
	        mul_15588: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15586, add_24666);  mul_15586 = add_24666 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_15588, [2])
	        full_323: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_161, full_323);  amax_161 = full_323 = None
	        amin_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_15588, [2])
	        full_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_161: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_161, full_322);  amin_161 = full_322 = None
	        sub_7369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_161, minimum_161);  maximum_161 = None
	        div_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7369, 255.0);  sub_7369 = None
	        clamp_min_483: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_322, 1.1920928955078125e-07);  div_322 = None
	        div_323: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_161, clamp_min_483);  minimum_161 = None
	        round_323: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_323);  div_323 = None
	        sub_7375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_323);  round_323 = None
	        clamp_min_484: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7375, -128);  sub_7375 = None
	        clamp_max_322: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_484, 127);  clamp_min_484 = None
	        view_2525: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_483, [sym_size_int, 1500, 1])
	        reciprocal_161: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2525);  view_2525 = None
	        mul_15634: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_161, 1.0);  reciprocal_161 = None
	        mul_15637: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15588, mul_15634);  mul_15588 = mul_15634 = None
	        round_324: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_15637);  mul_15637 = None
	        convert_element_type_966: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_322, torch.int8);  clamp_max_322 = None
	        view_2526: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_966, [sym_size_int, 1500, 1])
	        add_24749: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_324, view_2526);  round_324 = view_2526 = None
	        clamp_min_485: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24749, -128);  add_24749 = None
	        clamp_max_323: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_485, 127);  clamp_min_485 = None
	        convert_element_type_967: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_323, torch.int8);  clamp_max_323 = None
	        view_2530: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_966, [sym_size_int, 1500, 1]);  convert_element_type_966 = None
	        convert_element_type_968: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_967, torch.float32);  convert_element_type_967 = None
	        convert_element_type_969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2530, torch.float32);  view_2530 = None
	        sub_7395: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_968, convert_element_type_969);  convert_element_type_968 = convert_element_type_969 = None
	        view_2529: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_483, [sym_size_int, 1500, 1]);  clamp_min_483 = None
	        mul_15659: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7395, view_2529);  sub_7395 = view_2529 = None
	        view_2532: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_26_fc2_parametrizations_weight_original0 = None
	        view_2534: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_26_fc2_parametrizations_weight_original2 = None
	        convert_element_type_970: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2532, torch.float32);  view_2532 = None
	        convert_element_type_971: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2534, torch.float32);  view_2534 = None
	        sub_7399: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_970, convert_element_type_971);  convert_element_type_970 = convert_element_type_971 = None
	        view_2533: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_26_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_26_fc2_parametrizations_weight_original1 = None
	        mul_15664: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7399, view_2533);  sub_7399 = view_2533 = None
	        view_2535: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15664, [1280, 5120]);  mul_15664 = None
	        mul_15669: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2536: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15659, [mul_15669, 5120]);  mul_15659 = mul_15669 = None
	        permute_270: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2535, [1, 0]);  view_2535 = None
	        mm_default_25: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2536, permute_270);  view_2536 = permute_270 = None
	        add_tensor_25: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_25, model_audio_tower_layers_26_fc2_bias);  mm_default_25 = model_audio_tower_layers_26_fc2_bias = None
	        view_2537: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_25, [sym_size_int, 1500, 1280]);  add_tensor_25 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_24812: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_24514, view_2537);  add_24514 = view_2537 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_217: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_24812, memory_format = torch.contiguous_format)
	        var_mean_54 = torch.ops.aten.var_mean.correction(clone_217, [2], correction = 0, keepdim = True)
	        getitem_216: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_54[0]
	        getitem_217: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_54[1];  var_mean_54 = None
	        sub_7405: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_217, getitem_217);  clone_217 = getitem_217 = None
	        add_24817: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_216, 1e-05);  getitem_216 = None
	        rsqrt_54: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_24817);  add_24817 = None
	        mul_15680: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7405, rsqrt_54);  sub_7405 = rsqrt_54 = None
	        mul_15681: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_15680, model_audio_tower_layers_27_self_attn_layer_norm_weight);  mul_15680 = model_audio_tower_layers_27_self_attn_layer_norm_weight = None
	        add_24818: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_15681, model_audio_tower_layers_27_self_attn_layer_norm_bias);  mul_15681 = model_audio_tower_layers_27_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_24818, [2])
	        full_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_162, full_325);  amax_162 = full_325 = None
	        amin_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_24818, [2])
	        full_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_162: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_162, full_324);  amin_162 = full_324 = None
	        sub_7416: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_162, minimum_162);  maximum_162 = None
	        div_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7416, 255.0);  sub_7416 = None
	        clamp_min_486: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_324, 1.1920928955078125e-07);  div_324 = None
	        div_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_162, clamp_min_486);  minimum_162 = None
	        round_325: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_325);  div_325 = None
	        sub_7422: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_325);  round_325 = None
	        clamp_min_487: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7422, -128);  sub_7422 = None
	        clamp_max_324: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_487, 127);  clamp_min_487 = None
	        view_2540: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_486, [sym_size_int, 1500, 1])
	        reciprocal_162: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2540);  view_2540 = None
	        mul_15729: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_162, 1.0);  reciprocal_162 = None
	        mul_15732: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_24818, mul_15729);  mul_15729 = None
	        round_326: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15732);  mul_15732 = None
	        convert_element_type_972: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_324, torch.int8);  clamp_max_324 = None
	        view_2541: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_972, [sym_size_int, 1500, 1])
	        add_24905: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_326, view_2541);  round_326 = view_2541 = None
	        clamp_min_488: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_24905, -128);  add_24905 = None
	        clamp_max_325: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_488, 127);  clamp_min_488 = None
	        convert_element_type_973: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_325, torch.int8);  clamp_max_325 = None
	        view_2545: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_972, [sym_size_int, 1500, 1]);  convert_element_type_972 = None
	        convert_element_type_974: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_973, torch.float32);  convert_element_type_973 = None
	        convert_element_type_975: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2545, torch.float32);  view_2545 = None
	        sub_7442: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_974, convert_element_type_975);  convert_element_type_974 = convert_element_type_975 = None
	        view_2544: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_486, [sym_size_int, 1500, 1]);  clamp_min_486 = None
	        mul_15754: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7442, view_2544);  sub_7442 = view_2544 = None
	        view_2547: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2549: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_976: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2547, torch.float32);  view_2547 = None
	        convert_element_type_977: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2549, torch.float32);  view_2549 = None
	        sub_7446: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_976, convert_element_type_977);  convert_element_type_976 = convert_element_type_977 = None
	        view_2548: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_15759: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7446, view_2548);  sub_7446 = view_2548 = None
	        view_2550: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15759, [1280, 1280]);  mul_15759 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_24818, [2])
	        full_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_163, full_327);  amax_163 = full_327 = None
	        amin_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_24818, [2])
	        full_326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_163: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_163, full_326);  amin_163 = full_326 = None
	        sub_7461: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_163, minimum_163);  maximum_163 = None
	        div_326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7461, 255.0);  sub_7461 = None
	        clamp_min_489: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_326, 1.1920928955078125e-07);  div_326 = None
	        div_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_163, clamp_min_489);  minimum_163 = None
	        round_327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_327);  div_327 = None
	        sub_7467: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_327);  round_327 = None
	        clamp_min_490: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7467, -128);  sub_7467 = None
	        clamp_max_326: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_490, 127);  clamp_min_490 = None
	        view_2556: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_489, [sym_size_int, 1500, 1])
	        reciprocal_163: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2556);  view_2556 = None
	        mul_15825: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_163, 1.0);  reciprocal_163 = None
	        mul_15828: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_24818, mul_15825);  mul_15825 = None
	        round_328: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15828);  mul_15828 = None
	        convert_element_type_978: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_326, torch.int8);  clamp_max_326 = None
	        view_2557: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_978, [sym_size_int, 1500, 1])
	        add_25057: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_328, view_2557);  round_328 = view_2557 = None
	        clamp_min_491: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25057, -128);  add_25057 = None
	        clamp_max_327: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_491, 127);  clamp_min_491 = None
	        convert_element_type_979: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_327, torch.int8);  clamp_max_327 = None
	        view_2561: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_978, [sym_size_int, 1500, 1]);  convert_element_type_978 = None
	        convert_element_type_980: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_979, torch.float32);  convert_element_type_979 = None
	        convert_element_type_981: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2561, torch.float32);  view_2561 = None
	        sub_7487: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_980, convert_element_type_981);  convert_element_type_980 = convert_element_type_981 = None
	        view_2560: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_489, [sym_size_int, 1500, 1]);  clamp_min_489 = None
	        mul_15850: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7487, view_2560);  sub_7487 = view_2560 = None
	        view_2563: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2565: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_982: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2563, torch.float32);  view_2563 = None
	        convert_element_type_983: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2565, torch.float32);  view_2565 = None
	        sub_7491: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_982, convert_element_type_983);  convert_element_type_982 = convert_element_type_983 = None
	        view_2564: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_15855: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7491, view_2564);  sub_7491 = view_2564 = None
	        view_2566: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15855, [1280, 1280]);  mul_15855 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_24818, [2])
	        full_329: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_164, full_329);  amax_164 = full_329 = None
	        amin_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_24818, [2])
	        full_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_164: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_164, full_328);  amin_164 = full_328 = None
	        sub_7505: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_164, minimum_164);  maximum_164 = None
	        div_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7505, 255.0);  sub_7505 = None
	        clamp_min_492: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_328, 1.1920928955078125e-07);  div_328 = None
	        div_329: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_164, clamp_min_492);  minimum_164 = None
	        round_329: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_329);  div_329 = None
	        sub_7511: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_329);  round_329 = None
	        clamp_min_493: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7511, -128);  sub_7511 = None
	        clamp_max_328: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_493, 127);  clamp_min_493 = None
	        view_2572: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_492, [sym_size_int, 1500, 1])
	        reciprocal_164: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2572);  view_2572 = None
	        mul_15924: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_164, 1.0);  reciprocal_164 = None
	        mul_15927: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_24818, mul_15924);  add_24818 = mul_15924 = None
	        round_330: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_15927);  mul_15927 = None
	        convert_element_type_984: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_328, torch.int8);  clamp_max_328 = None
	        view_2573: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_984, [sym_size_int, 1500, 1])
	        add_25205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_330, view_2573);  round_330 = view_2573 = None
	        clamp_min_494: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25205, -128);  add_25205 = None
	        clamp_max_329: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_494, 127);  clamp_min_494 = None
	        convert_element_type_985: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_329, torch.int8);  clamp_max_329 = None
	        view_2577: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_984, [sym_size_int, 1500, 1]);  convert_element_type_984 = None
	        convert_element_type_986: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_985, torch.float32);  convert_element_type_985 = None
	        convert_element_type_987: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2577, torch.float32);  view_2577 = None
	        sub_7531: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_986, convert_element_type_987);  convert_element_type_986 = convert_element_type_987 = None
	        view_2576: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_492, [sym_size_int, 1500, 1]);  clamp_min_492 = None
	        mul_15949: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7531, view_2576);  sub_7531 = view_2576 = None
	        view_2579: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2581: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_988: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2579, torch.float32);  view_2579 = None
	        convert_element_type_989: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2581, torch.float32);  view_2581 = None
	        sub_7535: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_988, convert_element_type_989);  convert_element_type_988 = convert_element_type_989 = None
	        view_2580: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_15954: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7535, view_2580);  sub_7535 = view_2580 = None
	        view_2582: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15954, [1280, 1280]);  mul_15954 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_15764: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2551: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15754, [mul_15764, 1280]);  mul_15754 = mul_15764 = None
	        permute_271: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2550, [1, 0]);  view_2550 = None
	        mm_default_24: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2551, permute_271);  view_2551 = permute_271 = None
	        add_tensor_24: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_24, model_audio_tower_layers_27_self_attn_q_proj_bias);  mm_default_24 = model_audio_tower_layers_27_self_attn_q_proj_bias = None
	        view_2552: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_24, [sym_size_int, 1500, 1280]);  add_tensor_24 = None
	        mul_15771: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2552, 0.125);  view_2552 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2553: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15771, [sym_size_int, 1500, 20, 64]);  mul_15771 = None
	        permute_272: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2553, [0, 2, 1, 3]);  view_2553 = None
	        clone_218: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_272, memory_format = torch.contiguous_format);  permute_272 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_15858: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2567: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15850, [mul_15858, 1280]);  mul_15850 = mul_15858 = None
	        permute_273: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2566, [1, 0]);  view_2566 = None
	        mm_27: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2567, permute_273);  view_2567 = permute_273 = None
	        view_2568: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_27, [sym_size_int, 1500, 1280]);  mm_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2569: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2568, [sym_size_int, -1, 20, 64]);  view_2568 = None
	        permute_274: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2569, [0, 2, 1, 3]);  view_2569 = None
	        clone_219: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_274, memory_format = torch.contiguous_format);  permute_274 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_15959: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2583: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_15949, [mul_15959, 1280]);  mul_15949 = mul_15959 = None
	        permute_275: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2582, [1, 0]);  view_2582 = None
	        mm_default_23: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2583, permute_275);  view_2583 = permute_275 = None
	        add_tensor_23: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_23, model_audio_tower_layers_27_self_attn_v_proj_bias);  mm_default_23 = model_audio_tower_layers_27_self_attn_v_proj_bias = None
	        view_2584: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_23, [sym_size_int, 1500, 1280]);  add_tensor_23 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2585: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2584, [sym_size_int, -1, 20, 64]);  view_2584 = None
	        permute_276: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2585, [0, 2, 1, 3]);  view_2585 = None
	        clone_220: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_276, memory_format = torch.contiguous_format);  permute_276 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_27 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_218, clone_219, clone_220, None, False, scale = 1.0);  clone_218 = clone_219 = clone_220 = None
	        getitem_218: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_27[0];  _scaled_dot_product_efficient_attention_27 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_277: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_218, [0, 2, 1, 3]);  getitem_218 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2586: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_277, [sym_size_int, 1500, -1]);  permute_277 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2586, [2])
	        full_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_165, full_331);  amax_165 = full_331 = None
	        amin_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2586, [2])
	        full_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_165: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_165, full_330);  amin_165 = full_330 = None
	        sub_7553: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_165, minimum_165);  maximum_165 = None
	        div_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7553, 255.0);  sub_7553 = None
	        clamp_min_495: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_330, 1.1920928955078125e-07);  div_330 = None
	        div_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_165, clamp_min_495);  minimum_165 = None
	        round_331: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_331);  div_331 = None
	        sub_7559: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_331);  round_331 = None
	        clamp_min_496: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7559, -128);  sub_7559 = None
	        clamp_max_330: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_496, 127);  clamp_min_496 = None
	        view_2589: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_495, [sym_size_int, 1500, 1])
	        reciprocal_165: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2589);  view_2589 = None
	        mul_16029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_165, 1.0);  reciprocal_165 = None
	        mul_16032: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2586, mul_16029);  view_2586 = mul_16029 = None
	        round_332: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16032);  mul_16032 = None
	        convert_element_type_990: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_330, torch.int8);  clamp_max_330 = None
	        view_2590: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_990, [sym_size_int, 1500, 1])
	        add_25369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_332, view_2590);  round_332 = view_2590 = None
	        clamp_min_497: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25369, -128);  add_25369 = None
	        clamp_max_331: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_497, 127);  clamp_min_497 = None
	        convert_element_type_991: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_331, torch.int8);  clamp_max_331 = None
	        view_2594: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_990, [sym_size_int, 1500, 1]);  convert_element_type_990 = None
	        convert_element_type_992: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_991, torch.float32);  convert_element_type_991 = None
	        convert_element_type_993: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2594, torch.float32);  view_2594 = None
	        sub_7579: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_992, convert_element_type_993);  convert_element_type_992 = convert_element_type_993 = None
	        view_2593: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_495, [sym_size_int, 1500, 1]);  clamp_min_495 = None
	        mul_16054: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7579, view_2593);  sub_7579 = view_2593 = None
	        view_2596: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2598: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_994: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2596, torch.float32);  view_2596 = None
	        convert_element_type_995: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2598, torch.float32);  view_2598 = None
	        sub_7583: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_994, convert_element_type_995);  convert_element_type_994 = convert_element_type_995 = None
	        view_2597: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_16059: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7583, view_2597);  sub_7583 = view_2597 = None
	        view_2599: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16059, [1280, 1280]);  mul_16059 = None
	        mul_16064: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2600: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16054, [mul_16064, 1280]);  mul_16054 = mul_16064 = None
	        permute_278: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2599, [1, 0]);  view_2599 = None
	        mm_default_22: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2600, permute_278);  view_2600 = permute_278 = None
	        add_tensor_22: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_22, model_audio_tower_layers_27_self_attn_out_proj_bias);  mm_default_22 = model_audio_tower_layers_27_self_attn_out_proj_bias = None
	        view_2601: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_22, [sym_size_int, 1500, 1280]);  add_tensor_22 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_25432: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_24812, view_2601);  add_24812 = view_2601 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_222: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_25432, memory_format = torch.contiguous_format)
	        var_mean_55 = torch.ops.aten.var_mean.correction(clone_222, [2], correction = 0, keepdim = True)
	        getitem_222: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_55[0]
	        getitem_223: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_55[1];  var_mean_55 = None
	        sub_7589: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_222, getitem_223);  clone_222 = getitem_223 = None
	        add_25437: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_222, 1e-05);  getitem_222 = None
	        rsqrt_55: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_25437);  add_25437 = None
	        mul_16075: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7589, rsqrt_55);  sub_7589 = rsqrt_55 = None
	        mul_16076: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16075, model_audio_tower_layers_27_final_layer_norm_weight);  mul_16075 = model_audio_tower_layers_27_final_layer_norm_weight = None
	        add_25438: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16076, model_audio_tower_layers_27_final_layer_norm_bias);  mul_16076 = model_audio_tower_layers_27_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_25438, [2])
	        full_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_166, full_333);  amax_166 = full_333 = None
	        amin_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_25438, [2])
	        full_332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_166: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_166, full_332);  amin_166 = full_332 = None
	        sub_7600: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_166, minimum_166);  maximum_166 = None
	        div_332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7600, 255.0);  sub_7600 = None
	        clamp_min_498: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_332, 1.1920928955078125e-07);  div_332 = None
	        div_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_166, clamp_min_498);  minimum_166 = None
	        round_333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_333);  div_333 = None
	        sub_7606: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_333);  round_333 = None
	        clamp_min_499: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7606, -128);  sub_7606 = None
	        clamp_max_332: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_499, 127);  clamp_min_499 = None
	        view_2604: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_498, [sym_size_int, 1500, 1])
	        reciprocal_166: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2604);  view_2604 = None
	        mul_16124: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_166, 1.0);  reciprocal_166 = None
	        mul_16127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_25438, mul_16124);  add_25438 = mul_16124 = None
	        round_334: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16127);  mul_16127 = None
	        convert_element_type_996: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_332, torch.int8);  clamp_max_332 = None
	        view_2605: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_996, [sym_size_int, 1500, 1])
	        add_25525: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_334, view_2605);  round_334 = view_2605 = None
	        clamp_min_500: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25525, -128);  add_25525 = None
	        clamp_max_333: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_500, 127);  clamp_min_500 = None
	        convert_element_type_997: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_333, torch.int8);  clamp_max_333 = None
	        view_2609: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_996, [sym_size_int, 1500, 1]);  convert_element_type_996 = None
	        convert_element_type_998: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_997, torch.float32);  convert_element_type_997 = None
	        convert_element_type_999: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2609, torch.float32);  view_2609 = None
	        sub_7626: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_998, convert_element_type_999);  convert_element_type_998 = convert_element_type_999 = None
	        view_2608: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_498, [sym_size_int, 1500, 1]);  clamp_min_498 = None
	        mul_16149: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7626, view_2608);  sub_7626 = view_2608 = None
	        view_2611: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_27_fc1_parametrizations_weight_original0 = None
	        view_2613: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_27_fc1_parametrizations_weight_original2 = None
	        convert_element_type_1000: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2611, torch.float32);  view_2611 = None
	        convert_element_type_1001: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2613, torch.float32);  view_2613 = None
	        sub_7630: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1000, convert_element_type_1001);  convert_element_type_1000 = convert_element_type_1001 = None
	        view_2612: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_27_fc1_parametrizations_weight_original1 = None
	        mul_16154: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7630, view_2612);  sub_7630 = view_2612 = None
	        view_2614: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16154, [5120, 1280]);  mul_16154 = None
	        mul_16159: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2615: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16149, [mul_16159, 1280]);  mul_16149 = mul_16159 = None
	        permute_279: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2614, [1, 0]);  view_2614 = None
	        mm_default_21: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_2615, permute_279);  view_2615 = permute_279 = None
	        add_tensor_21: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_21, model_audio_tower_layers_27_fc1_bias);  mm_default_21 = model_audio_tower_layers_27_fc1_bias = None
	        view_2616: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_21, [sym_size_int, 1500, 5120]);  add_tensor_21 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_16166: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2616, 0.5)
	        mul_16167: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2616, 0.7071067811865476);  view_2616 = None
	        erf_29: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_16167);  mul_16167 = None
	        add_25584: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_29, 1);  erf_29 = None
	        mul_16168: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16166, add_25584);  mul_16166 = add_25584 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_16168, [2])
	        full_335: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_167, full_335);  amax_167 = full_335 = None
	        amin_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_16168, [2])
	        full_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_167: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_167, full_334);  amin_167 = full_334 = None
	        sub_7643: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_167, minimum_167);  maximum_167 = None
	        div_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7643, 255.0);  sub_7643 = None
	        clamp_min_501: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_334, 1.1920928955078125e-07);  div_334 = None
	        div_335: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_167, clamp_min_501);  minimum_167 = None
	        round_335: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_335);  div_335 = None
	        sub_7649: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_335);  round_335 = None
	        clamp_min_502: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7649, -128);  sub_7649 = None
	        clamp_max_334: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_502, 127);  clamp_min_502 = None
	        view_2619: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_501, [sym_size_int, 1500, 1])
	        reciprocal_167: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2619);  view_2619 = None
	        mul_16214: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_167, 1.0);  reciprocal_167 = None
	        mul_16217: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16168, mul_16214);  mul_16168 = mul_16214 = None
	        round_336: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_16217);  mul_16217 = None
	        convert_element_type_1002: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_334, torch.int8);  clamp_max_334 = None
	        view_2620: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1002, [sym_size_int, 1500, 1])
	        add_25667: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_336, view_2620);  round_336 = view_2620 = None
	        clamp_min_503: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25667, -128);  add_25667 = None
	        clamp_max_335: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_503, 127);  clamp_min_503 = None
	        convert_element_type_1003: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_335, torch.int8);  clamp_max_335 = None
	        view_2624: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1002, [sym_size_int, 1500, 1]);  convert_element_type_1002 = None
	        convert_element_type_1004: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1003, torch.float32);  convert_element_type_1003 = None
	        convert_element_type_1005: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2624, torch.float32);  view_2624 = None
	        sub_7669: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1004, convert_element_type_1005);  convert_element_type_1004 = convert_element_type_1005 = None
	        view_2623: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_501, [sym_size_int, 1500, 1]);  clamp_min_501 = None
	        mul_16239: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7669, view_2623);  sub_7669 = view_2623 = None
	        view_2626: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_27_fc2_parametrizations_weight_original0 = None
	        view_2628: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_27_fc2_parametrizations_weight_original2 = None
	        convert_element_type_1006: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2626, torch.float32);  view_2626 = None
	        convert_element_type_1007: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2628, torch.float32);  view_2628 = None
	        sub_7673: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1006, convert_element_type_1007);  convert_element_type_1006 = convert_element_type_1007 = None
	        view_2627: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_27_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_27_fc2_parametrizations_weight_original1 = None
	        mul_16244: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7673, view_2627);  sub_7673 = view_2627 = None
	        view_2629: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16244, [1280, 5120]);  mul_16244 = None
	        mul_16249: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2630: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16239, [mul_16249, 5120]);  mul_16239 = mul_16249 = None
	        permute_280: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2629, [1, 0]);  view_2629 = None
	        mm_default_20: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2630, permute_280);  view_2630 = permute_280 = None
	        add_tensor_20: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_20, model_audio_tower_layers_27_fc2_bias);  mm_default_20 = model_audio_tower_layers_27_fc2_bias = None
	        view_2631: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_20, [sym_size_int, 1500, 1280]);  add_tensor_20 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_25730: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_25432, view_2631);  add_25432 = view_2631 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_225: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_25730, memory_format = torch.contiguous_format)
	        var_mean_56 = torch.ops.aten.var_mean.correction(clone_225, [2], correction = 0, keepdim = True)
	        getitem_224: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_56[0]
	        getitem_225: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_56[1];  var_mean_56 = None
	        sub_7679: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_225, getitem_225);  clone_225 = getitem_225 = None
	        add_25735: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_224, 1e-05);  getitem_224 = None
	        rsqrt_56: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_25735);  add_25735 = None
	        mul_16260: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7679, rsqrt_56);  sub_7679 = rsqrt_56 = None
	        mul_16261: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16260, model_audio_tower_layers_28_self_attn_layer_norm_weight);  mul_16260 = model_audio_tower_layers_28_self_attn_layer_norm_weight = None
	        add_25736: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16261, model_audio_tower_layers_28_self_attn_layer_norm_bias);  mul_16261 = model_audio_tower_layers_28_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_25736, [2])
	        full_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_168, full_337);  amax_168 = full_337 = None
	        amin_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_25736, [2])
	        full_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_168: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_168, full_336);  amin_168 = full_336 = None
	        sub_7690: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_168, minimum_168);  maximum_168 = None
	        div_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7690, 255.0);  sub_7690 = None
	        clamp_min_504: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_336, 1.1920928955078125e-07);  div_336 = None
	        div_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_168, clamp_min_504);  minimum_168 = None
	        round_337: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_337);  div_337 = None
	        sub_7696: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_337);  round_337 = None
	        clamp_min_505: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7696, -128);  sub_7696 = None
	        clamp_max_336: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_505, 127);  clamp_min_505 = None
	        view_2634: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_504, [sym_size_int, 1500, 1])
	        reciprocal_168: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2634);  view_2634 = None
	        mul_16309: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_168, 1.0);  reciprocal_168 = None
	        mul_16312: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_25736, mul_16309);  mul_16309 = None
	        round_338: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16312);  mul_16312 = None
	        convert_element_type_1008: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_336, torch.int8);  clamp_max_336 = None
	        view_2635: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1008, [sym_size_int, 1500, 1])
	        add_25823: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_338, view_2635);  round_338 = view_2635 = None
	        clamp_min_506: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25823, -128);  add_25823 = None
	        clamp_max_337: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_506, 127);  clamp_min_506 = None
	        convert_element_type_1009: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_337, torch.int8);  clamp_max_337 = None
	        view_2639: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1008, [sym_size_int, 1500, 1]);  convert_element_type_1008 = None
	        convert_element_type_1010: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1009, torch.float32);  convert_element_type_1009 = None
	        convert_element_type_1011: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2639, torch.float32);  view_2639 = None
	        sub_7716: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1010, convert_element_type_1011);  convert_element_type_1010 = convert_element_type_1011 = None
	        view_2638: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_504, [sym_size_int, 1500, 1]);  clamp_min_504 = None
	        mul_16334: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7716, view_2638);  sub_7716 = view_2638 = None
	        view_2641: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2643: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_1012: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2641, torch.float32);  view_2641 = None
	        convert_element_type_1013: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2643, torch.float32);  view_2643 = None
	        sub_7720: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1012, convert_element_type_1013);  convert_element_type_1012 = convert_element_type_1013 = None
	        view_2642: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_16339: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7720, view_2642);  sub_7720 = view_2642 = None
	        view_2644: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16339, [1280, 1280]);  mul_16339 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_25736, [2])
	        full_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_169, full_339);  amax_169 = full_339 = None
	        amin_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_25736, [2])
	        full_338: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_169: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_169, full_338);  amin_169 = full_338 = None
	        sub_7735: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_169, minimum_169);  maximum_169 = None
	        div_338: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7735, 255.0);  sub_7735 = None
	        clamp_min_507: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_338, 1.1920928955078125e-07);  div_338 = None
	        div_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_169, clamp_min_507);  minimum_169 = None
	        round_339: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_339);  div_339 = None
	        sub_7741: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_339);  round_339 = None
	        clamp_min_508: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7741, -128);  sub_7741 = None
	        clamp_max_338: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_508, 127);  clamp_min_508 = None
	        view_2650: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_507, [sym_size_int, 1500, 1])
	        reciprocal_169: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2650);  view_2650 = None
	        mul_16405: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_169, 1.0);  reciprocal_169 = None
	        mul_16408: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_25736, mul_16405);  mul_16405 = None
	        round_340: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16408);  mul_16408 = None
	        convert_element_type_1014: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_338, torch.int8);  clamp_max_338 = None
	        view_2651: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1014, [sym_size_int, 1500, 1])
	        add_25975: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_340, view_2651);  round_340 = view_2651 = None
	        clamp_min_509: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_25975, -128);  add_25975 = None
	        clamp_max_339: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_509, 127);  clamp_min_509 = None
	        convert_element_type_1015: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_339, torch.int8);  clamp_max_339 = None
	        view_2655: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1014, [sym_size_int, 1500, 1]);  convert_element_type_1014 = None
	        convert_element_type_1016: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1015, torch.float32);  convert_element_type_1015 = None
	        convert_element_type_1017: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2655, torch.float32);  view_2655 = None
	        sub_7761: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1016, convert_element_type_1017);  convert_element_type_1016 = convert_element_type_1017 = None
	        view_2654: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_507, [sym_size_int, 1500, 1]);  clamp_min_507 = None
	        mul_16430: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7761, view_2654);  sub_7761 = view_2654 = None
	        view_2657: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2659: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_1018: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2657, torch.float32);  view_2657 = None
	        convert_element_type_1019: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2659, torch.float32);  view_2659 = None
	        sub_7765: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1018, convert_element_type_1019);  convert_element_type_1018 = convert_element_type_1019 = None
	        view_2658: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_16435: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7765, view_2658);  sub_7765 = view_2658 = None
	        view_2660: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16435, [1280, 1280]);  mul_16435 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_25736, [2])
	        full_341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_170, full_341);  amax_170 = full_341 = None
	        amin_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_25736, [2])
	        full_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_170: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_170, full_340);  amin_170 = full_340 = None
	        sub_7779: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_170, minimum_170);  maximum_170 = None
	        div_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7779, 255.0);  sub_7779 = None
	        clamp_min_510: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_340, 1.1920928955078125e-07);  div_340 = None
	        div_341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_170, clamp_min_510);  minimum_170 = None
	        round_341: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_341);  div_341 = None
	        sub_7785: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_341);  round_341 = None
	        clamp_min_511: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7785, -128);  sub_7785 = None
	        clamp_max_340: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_511, 127);  clamp_min_511 = None
	        view_2666: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_510, [sym_size_int, 1500, 1])
	        reciprocal_170: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2666);  view_2666 = None
	        mul_16504: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_170, 1.0);  reciprocal_170 = None
	        mul_16507: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_25736, mul_16504);  add_25736 = mul_16504 = None
	        round_342: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16507);  mul_16507 = None
	        convert_element_type_1020: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_340, torch.int8);  clamp_max_340 = None
	        view_2667: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1020, [sym_size_int, 1500, 1])
	        add_26123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_342, view_2667);  round_342 = view_2667 = None
	        clamp_min_512: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26123, -128);  add_26123 = None
	        clamp_max_341: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_512, 127);  clamp_min_512 = None
	        convert_element_type_1021: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_341, torch.int8);  clamp_max_341 = None
	        view_2671: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1020, [sym_size_int, 1500, 1]);  convert_element_type_1020 = None
	        convert_element_type_1022: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1021, torch.float32);  convert_element_type_1021 = None
	        convert_element_type_1023: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2671, torch.float32);  view_2671 = None
	        sub_7805: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1022, convert_element_type_1023);  convert_element_type_1022 = convert_element_type_1023 = None
	        view_2670: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_510, [sym_size_int, 1500, 1]);  clamp_min_510 = None
	        mul_16529: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7805, view_2670);  sub_7805 = view_2670 = None
	        view_2673: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2675: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_1024: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2673, torch.float32);  view_2673 = None
	        convert_element_type_1025: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2675, torch.float32);  view_2675 = None
	        sub_7809: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1024, convert_element_type_1025);  convert_element_type_1024 = convert_element_type_1025 = None
	        view_2674: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_16534: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7809, view_2674);  sub_7809 = view_2674 = None
	        view_2676: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16534, [1280, 1280]);  mul_16534 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_16344: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2645: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16334, [mul_16344, 1280]);  mul_16334 = mul_16344 = None
	        permute_281: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2644, [1, 0]);  view_2644 = None
	        mm_default_19: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2645, permute_281);  view_2645 = permute_281 = None
	        add_tensor_19: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_19, model_audio_tower_layers_28_self_attn_q_proj_bias);  mm_default_19 = model_audio_tower_layers_28_self_attn_q_proj_bias = None
	        view_2646: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_19, [sym_size_int, 1500, 1280]);  add_tensor_19 = None
	        mul_16351: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2646, 0.125);  view_2646 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2647: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16351, [sym_size_int, 1500, 20, 64]);  mul_16351 = None
	        permute_282: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2647, [0, 2, 1, 3]);  view_2647 = None
	        clone_226: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_282, memory_format = torch.contiguous_format);  permute_282 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_16438: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2661: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16430, [mul_16438, 1280]);  mul_16430 = mul_16438 = None
	        permute_283: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2660, [1, 0]);  view_2660 = None
	        mm_28: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2661, permute_283);  view_2661 = permute_283 = None
	        view_2662: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_28, [sym_size_int, 1500, 1280]);  mm_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2663: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2662, [sym_size_int, -1, 20, 64]);  view_2662 = None
	        permute_284: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2663, [0, 2, 1, 3]);  view_2663 = None
	        clone_227: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_284, memory_format = torch.contiguous_format);  permute_284 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_16539: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2677: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16529, [mul_16539, 1280]);  mul_16529 = mul_16539 = None
	        permute_285: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2676, [1, 0]);  view_2676 = None
	        mm_default_18: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2677, permute_285);  view_2677 = permute_285 = None
	        add_tensor_18: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_18, model_audio_tower_layers_28_self_attn_v_proj_bias);  mm_default_18 = model_audio_tower_layers_28_self_attn_v_proj_bias = None
	        view_2678: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_18, [sym_size_int, 1500, 1280]);  add_tensor_18 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2679: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2678, [sym_size_int, -1, 20, 64]);  view_2678 = None
	        permute_286: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2679, [0, 2, 1, 3]);  view_2679 = None
	        clone_228: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_286, memory_format = torch.contiguous_format);  permute_286 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_28 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_226, clone_227, clone_228, None, False, scale = 1.0);  clone_226 = clone_227 = clone_228 = None
	        getitem_226: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_28[0];  _scaled_dot_product_efficient_attention_28 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_287: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_226, [0, 2, 1, 3]);  getitem_226 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2680: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_287, [sym_size_int, 1500, -1]);  permute_287 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2680, [2])
	        full_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_171, full_343);  amax_171 = full_343 = None
	        amin_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2680, [2])
	        full_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_171: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_171, full_342);  amin_171 = full_342 = None
	        sub_7827: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_171, minimum_171);  maximum_171 = None
	        div_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7827, 255.0);  sub_7827 = None
	        clamp_min_513: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_342, 1.1920928955078125e-07);  div_342 = None
	        div_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_171, clamp_min_513);  minimum_171 = None
	        round_343: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_343);  div_343 = None
	        sub_7833: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_343);  round_343 = None
	        clamp_min_514: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7833, -128);  sub_7833 = None
	        clamp_max_342: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_514, 127);  clamp_min_514 = None
	        view_2683: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_513, [sym_size_int, 1500, 1])
	        reciprocal_171: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2683);  view_2683 = None
	        mul_16609: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_171, 1.0);  reciprocal_171 = None
	        mul_16612: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2680, mul_16609);  view_2680 = mul_16609 = None
	        round_344: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16612);  mul_16612 = None
	        convert_element_type_1026: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_342, torch.int8);  clamp_max_342 = None
	        view_2684: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1026, [sym_size_int, 1500, 1])
	        add_26287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_344, view_2684);  round_344 = view_2684 = None
	        clamp_min_515: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26287, -128);  add_26287 = None
	        clamp_max_343: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_515, 127);  clamp_min_515 = None
	        convert_element_type_1027: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_343, torch.int8);  clamp_max_343 = None
	        view_2688: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1026, [sym_size_int, 1500, 1]);  convert_element_type_1026 = None
	        convert_element_type_1028: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1027, torch.float32);  convert_element_type_1027 = None
	        convert_element_type_1029: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2688, torch.float32);  view_2688 = None
	        sub_7853: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1028, convert_element_type_1029);  convert_element_type_1028 = convert_element_type_1029 = None
	        view_2687: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_513, [sym_size_int, 1500, 1]);  clamp_min_513 = None
	        mul_16634: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7853, view_2687);  sub_7853 = view_2687 = None
	        view_2690: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2692: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_1030: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2690, torch.float32);  view_2690 = None
	        convert_element_type_1031: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2692, torch.float32);  view_2692 = None
	        sub_7857: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1030, convert_element_type_1031);  convert_element_type_1030 = convert_element_type_1031 = None
	        view_2691: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_16639: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7857, view_2691);  sub_7857 = view_2691 = None
	        view_2693: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16639, [1280, 1280]);  mul_16639 = None
	        mul_16644: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2694: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16634, [mul_16644, 1280]);  mul_16634 = mul_16644 = None
	        permute_288: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2693, [1, 0]);  view_2693 = None
	        mm_default_17: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2694, permute_288);  view_2694 = permute_288 = None
	        add_tensor_17: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_17, model_audio_tower_layers_28_self_attn_out_proj_bias);  mm_default_17 = model_audio_tower_layers_28_self_attn_out_proj_bias = None
	        view_2695: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_17, [sym_size_int, 1500, 1280]);  add_tensor_17 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_26350: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_25730, view_2695);  add_25730 = view_2695 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_230: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_26350, memory_format = torch.contiguous_format)
	        var_mean_57 = torch.ops.aten.var_mean.correction(clone_230, [2], correction = 0, keepdim = True)
	        getitem_230: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_57[0]
	        getitem_231: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_57[1];  var_mean_57 = None
	        sub_7863: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_230, getitem_231);  clone_230 = getitem_231 = None
	        add_26355: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_230, 1e-05);  getitem_230 = None
	        rsqrt_57: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_26355);  add_26355 = None
	        mul_16655: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7863, rsqrt_57);  sub_7863 = rsqrt_57 = None
	        mul_16656: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16655, model_audio_tower_layers_28_final_layer_norm_weight);  mul_16655 = model_audio_tower_layers_28_final_layer_norm_weight = None
	        add_26356: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16656, model_audio_tower_layers_28_final_layer_norm_bias);  mul_16656 = model_audio_tower_layers_28_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_26356, [2])
	        full_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_172, full_345);  amax_172 = full_345 = None
	        amin_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_26356, [2])
	        full_344: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_172: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_172, full_344);  amin_172 = full_344 = None
	        sub_7874: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_172, minimum_172);  maximum_172 = None
	        div_344: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7874, 255.0);  sub_7874 = None
	        clamp_min_516: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_344, 1.1920928955078125e-07);  div_344 = None
	        div_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_172, clamp_min_516);  minimum_172 = None
	        round_345: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_345);  div_345 = None
	        sub_7880: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_345);  round_345 = None
	        clamp_min_517: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7880, -128);  sub_7880 = None
	        clamp_max_344: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_517, 127);  clamp_min_517 = None
	        view_2698: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_516, [sym_size_int, 1500, 1])
	        reciprocal_172: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2698);  view_2698 = None
	        mul_16704: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_172, 1.0);  reciprocal_172 = None
	        mul_16707: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_26356, mul_16704);  add_26356 = mul_16704 = None
	        round_346: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16707);  mul_16707 = None
	        convert_element_type_1032: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_344, torch.int8);  clamp_max_344 = None
	        view_2699: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1032, [sym_size_int, 1500, 1])
	        add_26443: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_346, view_2699);  round_346 = view_2699 = None
	        clamp_min_518: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26443, -128);  add_26443 = None
	        clamp_max_345: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_518, 127);  clamp_min_518 = None
	        convert_element_type_1033: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_345, torch.int8);  clamp_max_345 = None
	        view_2703: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1032, [sym_size_int, 1500, 1]);  convert_element_type_1032 = None
	        convert_element_type_1034: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1033, torch.float32);  convert_element_type_1033 = None
	        convert_element_type_1035: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2703, torch.float32);  view_2703 = None
	        sub_7900: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1034, convert_element_type_1035);  convert_element_type_1034 = convert_element_type_1035 = None
	        view_2702: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_516, [sym_size_int, 1500, 1]);  clamp_min_516 = None
	        mul_16729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7900, view_2702);  sub_7900 = view_2702 = None
	        view_2705: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_28_fc1_parametrizations_weight_original0 = None
	        view_2707: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_28_fc1_parametrizations_weight_original2 = None
	        convert_element_type_1036: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2705, torch.float32);  view_2705 = None
	        convert_element_type_1037: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2707, torch.float32);  view_2707 = None
	        sub_7904: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1036, convert_element_type_1037);  convert_element_type_1036 = convert_element_type_1037 = None
	        view_2706: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_28_fc1_parametrizations_weight_original1 = None
	        mul_16734: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7904, view_2706);  sub_7904 = view_2706 = None
	        view_2708: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16734, [5120, 1280]);  mul_16734 = None
	        mul_16739: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2709: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16729, [mul_16739, 1280]);  mul_16729 = mul_16739 = None
	        permute_289: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2708, [1, 0]);  view_2708 = None
	        mm_default_16: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_2709, permute_289);  view_2709 = permute_289 = None
	        add_tensor_16: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_16, model_audio_tower_layers_28_fc1_bias);  mm_default_16 = model_audio_tower_layers_28_fc1_bias = None
	        view_2710: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_16, [sym_size_int, 1500, 5120]);  add_tensor_16 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_16746: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2710, 0.5)
	        mul_16747: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2710, 0.7071067811865476);  view_2710 = None
	        erf_30: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_16747);  mul_16747 = None
	        add_26502: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_30, 1);  erf_30 = None
	        mul_16748: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16746, add_26502);  mul_16746 = add_26502 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_16748, [2])
	        full_347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_173, full_347);  amax_173 = full_347 = None
	        amin_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_16748, [2])
	        full_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_173: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_173, full_346);  amin_173 = full_346 = None
	        sub_7917: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_173, minimum_173);  maximum_173 = None
	        div_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7917, 255.0);  sub_7917 = None
	        clamp_min_519: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_346, 1.1920928955078125e-07);  div_346 = None
	        div_347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_173, clamp_min_519);  minimum_173 = None
	        round_347: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_347);  div_347 = None
	        sub_7923: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_347);  round_347 = None
	        clamp_min_520: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7923, -128);  sub_7923 = None
	        clamp_max_346: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_520, 127);  clamp_min_520 = None
	        view_2713: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_519, [sym_size_int, 1500, 1])
	        reciprocal_173: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2713);  view_2713 = None
	        mul_16794: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_173, 1.0);  reciprocal_173 = None
	        mul_16797: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16748, mul_16794);  mul_16748 = mul_16794 = None
	        round_348: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_16797);  mul_16797 = None
	        convert_element_type_1038: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_346, torch.int8);  clamp_max_346 = None
	        view_2714: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1038, [sym_size_int, 1500, 1])
	        add_26585: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_348, view_2714);  round_348 = view_2714 = None
	        clamp_min_521: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26585, -128);  add_26585 = None
	        clamp_max_347: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_521, 127);  clamp_min_521 = None
	        convert_element_type_1039: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_347, torch.int8);  clamp_max_347 = None
	        view_2718: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1038, [sym_size_int, 1500, 1]);  convert_element_type_1038 = None
	        convert_element_type_1040: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1039, torch.float32);  convert_element_type_1039 = None
	        convert_element_type_1041: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2718, torch.float32);  view_2718 = None
	        sub_7943: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1040, convert_element_type_1041);  convert_element_type_1040 = convert_element_type_1041 = None
	        view_2717: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_519, [sym_size_int, 1500, 1]);  clamp_min_519 = None
	        mul_16819: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7943, view_2717);  sub_7943 = view_2717 = None
	        view_2720: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_28_fc2_parametrizations_weight_original0 = None
	        view_2722: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_28_fc2_parametrizations_weight_original2 = None
	        convert_element_type_1042: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2720, torch.float32);  view_2720 = None
	        convert_element_type_1043: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2722, torch.float32);  view_2722 = None
	        sub_7947: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1042, convert_element_type_1043);  convert_element_type_1042 = convert_element_type_1043 = None
	        view_2721: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_28_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_28_fc2_parametrizations_weight_original1 = None
	        mul_16824: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7947, view_2721);  sub_7947 = view_2721 = None
	        view_2723: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16824, [1280, 5120]);  mul_16824 = None
	        mul_16829: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2724: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16819, [mul_16829, 5120]);  mul_16819 = mul_16829 = None
	        permute_290: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2723, [1, 0]);  view_2723 = None
	        mm_default_15: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2724, permute_290);  view_2724 = permute_290 = None
	        add_tensor_15: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_15, model_audio_tower_layers_28_fc2_bias);  mm_default_15 = model_audio_tower_layers_28_fc2_bias = None
	        view_2725: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_15, [sym_size_int, 1500, 1280]);  add_tensor_15 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_26648: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_26350, view_2725);  add_26350 = view_2725 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_233: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_26648, memory_format = torch.contiguous_format)
	        var_mean_58 = torch.ops.aten.var_mean.correction(clone_233, [2], correction = 0, keepdim = True)
	        getitem_232: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_58[0]
	        getitem_233: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_58[1];  var_mean_58 = None
	        sub_7953: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_233, getitem_233);  clone_233 = getitem_233 = None
	        add_26653: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_232, 1e-05);  getitem_232 = None
	        rsqrt_58: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_26653);  add_26653 = None
	        mul_16840: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7953, rsqrt_58);  sub_7953 = rsqrt_58 = None
	        mul_16841: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_16840, model_audio_tower_layers_29_self_attn_layer_norm_weight);  mul_16840 = model_audio_tower_layers_29_self_attn_layer_norm_weight = None
	        add_26654: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_16841, model_audio_tower_layers_29_self_attn_layer_norm_bias);  mul_16841 = model_audio_tower_layers_29_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_26654, [2])
	        full_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_174, full_349);  amax_174 = full_349 = None
	        amin_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_26654, [2])
	        full_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_174: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_174, full_348);  amin_174 = full_348 = None
	        sub_7964: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_174, minimum_174);  maximum_174 = None
	        div_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_7964, 255.0);  sub_7964 = None
	        clamp_min_522: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_348, 1.1920928955078125e-07);  div_348 = None
	        div_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_174, clamp_min_522);  minimum_174 = None
	        round_349: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_349);  div_349 = None
	        sub_7970: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_349);  round_349 = None
	        clamp_min_523: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_7970, -128);  sub_7970 = None
	        clamp_max_348: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_523, 127);  clamp_min_523 = None
	        view_2728: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_522, [sym_size_int, 1500, 1])
	        reciprocal_174: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2728);  view_2728 = None
	        mul_16889: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_174, 1.0);  reciprocal_174 = None
	        mul_16892: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_26654, mul_16889);  mul_16889 = None
	        round_350: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16892);  mul_16892 = None
	        convert_element_type_1044: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_348, torch.int8);  clamp_max_348 = None
	        view_2729: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1044, [sym_size_int, 1500, 1])
	        add_26741: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_350, view_2729);  round_350 = view_2729 = None
	        clamp_min_524: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26741, -128);  add_26741 = None
	        clamp_max_349: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_524, 127);  clamp_min_524 = None
	        convert_element_type_1045: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_349, torch.int8);  clamp_max_349 = None
	        view_2733: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1044, [sym_size_int, 1500, 1]);  convert_element_type_1044 = None
	        convert_element_type_1046: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1045, torch.float32);  convert_element_type_1045 = None
	        convert_element_type_1047: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2733, torch.float32);  view_2733 = None
	        sub_7990: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1046, convert_element_type_1047);  convert_element_type_1046 = convert_element_type_1047 = None
	        view_2732: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_522, [sym_size_int, 1500, 1]);  clamp_min_522 = None
	        mul_16914: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7990, view_2732);  sub_7990 = view_2732 = None
	        view_2735: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2737: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_1048: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2735, torch.float32);  view_2735 = None
	        convert_element_type_1049: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2737, torch.float32);  view_2737 = None
	        sub_7994: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1048, convert_element_type_1049);  convert_element_type_1048 = convert_element_type_1049 = None
	        view_2736: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_16919: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_7994, view_2736);  sub_7994 = view_2736 = None
	        view_2738: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16919, [1280, 1280]);  mul_16919 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_26654, [2])
	        full_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_175, full_351);  amax_175 = full_351 = None
	        amin_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_26654, [2])
	        full_350: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_175: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_175, full_350);  amin_175 = full_350 = None
	        sub_8009: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_175, minimum_175);  maximum_175 = None
	        div_350: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8009, 255.0);  sub_8009 = None
	        clamp_min_525: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_350, 1.1920928955078125e-07);  div_350 = None
	        div_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_175, clamp_min_525);  minimum_175 = None
	        round_351: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_351);  div_351 = None
	        sub_8015: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_351);  round_351 = None
	        clamp_min_526: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8015, -128);  sub_8015 = None
	        clamp_max_350: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_526, 127);  clamp_min_526 = None
	        view_2744: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_525, [sym_size_int, 1500, 1])
	        reciprocal_175: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2744);  view_2744 = None
	        mul_16985: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_175, 1.0);  reciprocal_175 = None
	        mul_16988: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_26654, mul_16985);  mul_16985 = None
	        round_352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_16988);  mul_16988 = None
	        convert_element_type_1050: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_350, torch.int8);  clamp_max_350 = None
	        view_2745: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1050, [sym_size_int, 1500, 1])
	        add_26893: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_352, view_2745);  round_352 = view_2745 = None
	        clamp_min_527: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_26893, -128);  add_26893 = None
	        clamp_max_351: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_527, 127);  clamp_min_527 = None
	        convert_element_type_1051: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_351, torch.int8);  clamp_max_351 = None
	        view_2749: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1050, [sym_size_int, 1500, 1]);  convert_element_type_1050 = None
	        convert_element_type_1052: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1051, torch.float32);  convert_element_type_1051 = None
	        convert_element_type_1053: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2749, torch.float32);  view_2749 = None
	        sub_8035: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1052, convert_element_type_1053);  convert_element_type_1052 = convert_element_type_1053 = None
	        view_2748: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_525, [sym_size_int, 1500, 1]);  clamp_min_525 = None
	        mul_17010: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8035, view_2748);  sub_8035 = view_2748 = None
	        view_2751: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2753: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_1054: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2751, torch.float32);  view_2751 = None
	        convert_element_type_1055: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2753, torch.float32);  view_2753 = None
	        sub_8039: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1054, convert_element_type_1055);  convert_element_type_1054 = convert_element_type_1055 = None
	        view_2752: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_17015: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8039, view_2752);  sub_8039 = view_2752 = None
	        view_2754: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17015, [1280, 1280]);  mul_17015 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_26654, [2])
	        full_353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_176, full_353);  amax_176 = full_353 = None
	        amin_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_26654, [2])
	        full_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_176: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_176, full_352);  amin_176 = full_352 = None
	        sub_8053: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_176, minimum_176);  maximum_176 = None
	        div_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8053, 255.0);  sub_8053 = None
	        clamp_min_528: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_352, 1.1920928955078125e-07);  div_352 = None
	        div_353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_176, clamp_min_528);  minimum_176 = None
	        round_353: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_353);  div_353 = None
	        sub_8059: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_353);  round_353 = None
	        clamp_min_529: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8059, -128);  sub_8059 = None
	        clamp_max_352: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_529, 127);  clamp_min_529 = None
	        view_2760: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_528, [sym_size_int, 1500, 1])
	        reciprocal_176: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2760);  view_2760 = None
	        mul_17084: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_176, 1.0);  reciprocal_176 = None
	        mul_17087: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_26654, mul_17084);  add_26654 = mul_17084 = None
	        round_354: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17087);  mul_17087 = None
	        convert_element_type_1056: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_352, torch.int8);  clamp_max_352 = None
	        view_2761: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1056, [sym_size_int, 1500, 1])
	        add_27041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_354, view_2761);  round_354 = view_2761 = None
	        clamp_min_530: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27041, -128);  add_27041 = None
	        clamp_max_353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_530, 127);  clamp_min_530 = None
	        convert_element_type_1057: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_353, torch.int8);  clamp_max_353 = None
	        view_2765: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1056, [sym_size_int, 1500, 1]);  convert_element_type_1056 = None
	        convert_element_type_1058: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1057, torch.float32);  convert_element_type_1057 = None
	        convert_element_type_1059: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2765, torch.float32);  view_2765 = None
	        sub_8079: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1058, convert_element_type_1059);  convert_element_type_1058 = convert_element_type_1059 = None
	        view_2764: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_528, [sym_size_int, 1500, 1]);  clamp_min_528 = None
	        mul_17109: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8079, view_2764);  sub_8079 = view_2764 = None
	        view_2767: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2769: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_1060: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2767, torch.float32);  view_2767 = None
	        convert_element_type_1061: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2769, torch.float32);  view_2769 = None
	        sub_8083: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1060, convert_element_type_1061);  convert_element_type_1060 = convert_element_type_1061 = None
	        view_2768: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_17114: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8083, view_2768);  sub_8083 = view_2768 = None
	        view_2770: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17114, [1280, 1280]);  mul_17114 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_16924: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2739: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16914, [mul_16924, 1280]);  mul_16914 = mul_16924 = None
	        permute_291: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2738, [1, 0]);  view_2738 = None
	        mm_default_14: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2739, permute_291);  view_2739 = permute_291 = None
	        add_tensor_14: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_14, model_audio_tower_layers_29_self_attn_q_proj_bias);  mm_default_14 = model_audio_tower_layers_29_self_attn_q_proj_bias = None
	        view_2740: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_14, [sym_size_int, 1500, 1280]);  add_tensor_14 = None
	        mul_16931: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2740, 0.125);  view_2740 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2741: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_16931, [sym_size_int, 1500, 20, 64]);  mul_16931 = None
	        permute_292: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2741, [0, 2, 1, 3]);  view_2741 = None
	        clone_234: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_292, memory_format = torch.contiguous_format);  permute_292 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_17018: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2755: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17010, [mul_17018, 1280]);  mul_17010 = mul_17018 = None
	        permute_293: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2754, [1, 0]);  view_2754 = None
	        mm_29: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2755, permute_293);  view_2755 = permute_293 = None
	        view_2756: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_29, [sym_size_int, 1500, 1280]);  mm_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2757: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2756, [sym_size_int, -1, 20, 64]);  view_2756 = None
	        permute_294: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2757, [0, 2, 1, 3]);  view_2757 = None
	        clone_235: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_294, memory_format = torch.contiguous_format);  permute_294 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_17119: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2771: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17109, [mul_17119, 1280]);  mul_17109 = mul_17119 = None
	        permute_295: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2770, [1, 0]);  view_2770 = None
	        mm_default_13: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2771, permute_295);  view_2771 = permute_295 = None
	        add_tensor_13: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_13, model_audio_tower_layers_29_self_attn_v_proj_bias);  mm_default_13 = model_audio_tower_layers_29_self_attn_v_proj_bias = None
	        view_2772: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_13, [sym_size_int, 1500, 1280]);  add_tensor_13 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2773: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2772, [sym_size_int, -1, 20, 64]);  view_2772 = None
	        permute_296: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2773, [0, 2, 1, 3]);  view_2773 = None
	        clone_236: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_296, memory_format = torch.contiguous_format);  permute_296 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_29 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_234, clone_235, clone_236, None, False, scale = 1.0);  clone_234 = clone_235 = clone_236 = None
	        getitem_234: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_29[0];  _scaled_dot_product_efficient_attention_29 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_297: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_234, [0, 2, 1, 3]);  getitem_234 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2774: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_297, [sym_size_int, 1500, -1]);  permute_297 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2774, [2])
	        full_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_177, full_355);  amax_177 = full_355 = None
	        amin_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2774, [2])
	        full_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_177: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_177, full_354);  amin_177 = full_354 = None
	        sub_8101: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_177, minimum_177);  maximum_177 = None
	        div_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8101, 255.0);  sub_8101 = None
	        clamp_min_531: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_354, 1.1920928955078125e-07);  div_354 = None
	        div_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_177, clamp_min_531);  minimum_177 = None
	        round_355: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_355);  div_355 = None
	        sub_8107: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_355);  round_355 = None
	        clamp_min_532: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8107, -128);  sub_8107 = None
	        clamp_max_354: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_532, 127);  clamp_min_532 = None
	        view_2777: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_531, [sym_size_int, 1500, 1])
	        reciprocal_177: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2777);  view_2777 = None
	        mul_17189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_177, 1.0);  reciprocal_177 = None
	        mul_17192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2774, mul_17189);  view_2774 = mul_17189 = None
	        round_356: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17192);  mul_17192 = None
	        convert_element_type_1062: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_354, torch.int8);  clamp_max_354 = None
	        view_2778: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1062, [sym_size_int, 1500, 1])
	        add_27205: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_356, view_2778);  round_356 = view_2778 = None
	        clamp_min_533: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27205, -128);  add_27205 = None
	        clamp_max_355: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_533, 127);  clamp_min_533 = None
	        convert_element_type_1063: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_355, torch.int8);  clamp_max_355 = None
	        view_2782: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1062, [sym_size_int, 1500, 1]);  convert_element_type_1062 = None
	        convert_element_type_1064: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1063, torch.float32);  convert_element_type_1063 = None
	        convert_element_type_1065: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2782, torch.float32);  view_2782 = None
	        sub_8127: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1064, convert_element_type_1065);  convert_element_type_1064 = convert_element_type_1065 = None
	        view_2781: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_531, [sym_size_int, 1500, 1]);  clamp_min_531 = None
	        mul_17214: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8127, view_2781);  sub_8127 = view_2781 = None
	        view_2784: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2786: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_1066: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2784, torch.float32);  view_2784 = None
	        convert_element_type_1067: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2786, torch.float32);  view_2786 = None
	        sub_8131: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1066, convert_element_type_1067);  convert_element_type_1066 = convert_element_type_1067 = None
	        view_2785: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_17219: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8131, view_2785);  sub_8131 = view_2785 = None
	        view_2787: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17219, [1280, 1280]);  mul_17219 = None
	        mul_17224: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2788: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17214, [mul_17224, 1280]);  mul_17214 = mul_17224 = None
	        permute_298: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2787, [1, 0]);  view_2787 = None
	        mm_default_12: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2788, permute_298);  view_2788 = permute_298 = None
	        add_tensor_12: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_12, model_audio_tower_layers_29_self_attn_out_proj_bias);  mm_default_12 = model_audio_tower_layers_29_self_attn_out_proj_bias = None
	        view_2789: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_12, [sym_size_int, 1500, 1280]);  add_tensor_12 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_27268: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_26648, view_2789);  add_26648 = view_2789 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_238: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_27268, memory_format = torch.contiguous_format)
	        var_mean_59 = torch.ops.aten.var_mean.correction(clone_238, [2], correction = 0, keepdim = True)
	        getitem_238: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_59[0]
	        getitem_239: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_59[1];  var_mean_59 = None
	        sub_8137: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_238, getitem_239);  clone_238 = getitem_239 = None
	        add_27273: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_238, 1e-05);  getitem_238 = None
	        rsqrt_59: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_27273);  add_27273 = None
	        mul_17235: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8137, rsqrt_59);  sub_8137 = rsqrt_59 = None
	        mul_17236: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17235, model_audio_tower_layers_29_final_layer_norm_weight);  mul_17235 = model_audio_tower_layers_29_final_layer_norm_weight = None
	        add_27274: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_17236, model_audio_tower_layers_29_final_layer_norm_bias);  mul_17236 = model_audio_tower_layers_29_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_27274, [2])
	        full_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_178, full_357);  amax_178 = full_357 = None
	        amin_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_27274, [2])
	        full_356: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_178: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_178, full_356);  amin_178 = full_356 = None
	        sub_8148: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_178, minimum_178);  maximum_178 = None
	        div_356: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8148, 255.0);  sub_8148 = None
	        clamp_min_534: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_356, 1.1920928955078125e-07);  div_356 = None
	        div_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_178, clamp_min_534);  minimum_178 = None
	        round_357: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_357);  div_357 = None
	        sub_8154: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_357);  round_357 = None
	        clamp_min_535: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8154, -128);  sub_8154 = None
	        clamp_max_356: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_535, 127);  clamp_min_535 = None
	        view_2792: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_534, [sym_size_int, 1500, 1])
	        reciprocal_178: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2792);  view_2792 = None
	        mul_17284: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_178, 1.0);  reciprocal_178 = None
	        mul_17287: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_27274, mul_17284);  add_27274 = mul_17284 = None
	        round_358: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17287);  mul_17287 = None
	        convert_element_type_1068: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_356, torch.int8);  clamp_max_356 = None
	        view_2793: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1068, [sym_size_int, 1500, 1])
	        add_27361: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_358, view_2793);  round_358 = view_2793 = None
	        clamp_min_536: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27361, -128);  add_27361 = None
	        clamp_max_357: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_536, 127);  clamp_min_536 = None
	        convert_element_type_1069: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_357, torch.int8);  clamp_max_357 = None
	        view_2797: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1068, [sym_size_int, 1500, 1]);  convert_element_type_1068 = None
	        convert_element_type_1070: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1069, torch.float32);  convert_element_type_1069 = None
	        convert_element_type_1071: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2797, torch.float32);  view_2797 = None
	        sub_8174: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1070, convert_element_type_1071);  convert_element_type_1070 = convert_element_type_1071 = None
	        view_2796: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_534, [sym_size_int, 1500, 1]);  clamp_min_534 = None
	        mul_17309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8174, view_2796);  sub_8174 = view_2796 = None
	        view_2799: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_29_fc1_parametrizations_weight_original0 = None
	        view_2801: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_29_fc1_parametrizations_weight_original2 = None
	        convert_element_type_1072: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2799, torch.float32);  view_2799 = None
	        convert_element_type_1073: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2801, torch.float32);  view_2801 = None
	        sub_8178: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1072, convert_element_type_1073);  convert_element_type_1072 = convert_element_type_1073 = None
	        view_2800: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_29_fc1_parametrizations_weight_original1 = None
	        mul_17314: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8178, view_2800);  sub_8178 = view_2800 = None
	        view_2802: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17314, [5120, 1280]);  mul_17314 = None
	        mul_17319: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2803: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17309, [mul_17319, 1280]);  mul_17309 = mul_17319 = None
	        permute_299: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2802, [1, 0]);  view_2802 = None
	        mm_default_11: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_2803, permute_299);  view_2803 = permute_299 = None
	        add_tensor_11: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_11, model_audio_tower_layers_29_fc1_bias);  mm_default_11 = model_audio_tower_layers_29_fc1_bias = None
	        view_2804: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_11, [sym_size_int, 1500, 5120]);  add_tensor_11 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_17326: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2804, 0.5)
	        mul_17327: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2804, 0.7071067811865476);  view_2804 = None
	        erf_31: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_17327);  mul_17327 = None
	        add_27420: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_31, 1);  erf_31 = None
	        mul_17328: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17326, add_27420);  mul_17326 = add_27420 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_17328, [2])
	        full_359: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_179, full_359);  amax_179 = full_359 = None
	        amin_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_17328, [2])
	        full_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_179: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_179, full_358);  amin_179 = full_358 = None
	        sub_8191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_179, minimum_179);  maximum_179 = None
	        div_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8191, 255.0);  sub_8191 = None
	        clamp_min_537: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_358, 1.1920928955078125e-07);  div_358 = None
	        div_359: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_179, clamp_min_537);  minimum_179 = None
	        round_359: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_359);  div_359 = None
	        sub_8197: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_359);  round_359 = None
	        clamp_min_538: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8197, -128);  sub_8197 = None
	        clamp_max_358: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_538, 127);  clamp_min_538 = None
	        view_2807: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_537, [sym_size_int, 1500, 1])
	        reciprocal_179: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2807);  view_2807 = None
	        mul_17374: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_179, 1.0);  reciprocal_179 = None
	        mul_17377: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17328, mul_17374);  mul_17328 = mul_17374 = None
	        round_360: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_17377);  mul_17377 = None
	        convert_element_type_1074: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_358, torch.int8);  clamp_max_358 = None
	        view_2808: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1074, [sym_size_int, 1500, 1])
	        add_27503: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_360, view_2808);  round_360 = view_2808 = None
	        clamp_min_539: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27503, -128);  add_27503 = None
	        clamp_max_359: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_539, 127);  clamp_min_539 = None
	        convert_element_type_1075: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_359, torch.int8);  clamp_max_359 = None
	        view_2812: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1074, [sym_size_int, 1500, 1]);  convert_element_type_1074 = None
	        convert_element_type_1076: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1075, torch.float32);  convert_element_type_1075 = None
	        convert_element_type_1077: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2812, torch.float32);  view_2812 = None
	        sub_8217: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1076, convert_element_type_1077);  convert_element_type_1076 = convert_element_type_1077 = None
	        view_2811: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_537, [sym_size_int, 1500, 1]);  clamp_min_537 = None
	        mul_17399: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8217, view_2811);  sub_8217 = view_2811 = None
	        view_2814: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_29_fc2_parametrizations_weight_original0 = None
	        view_2816: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_29_fc2_parametrizations_weight_original2 = None
	        convert_element_type_1078: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2814, torch.float32);  view_2814 = None
	        convert_element_type_1079: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2816, torch.float32);  view_2816 = None
	        sub_8221: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1078, convert_element_type_1079);  convert_element_type_1078 = convert_element_type_1079 = None
	        view_2815: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_29_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_29_fc2_parametrizations_weight_original1 = None
	        mul_17404: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8221, view_2815);  sub_8221 = view_2815 = None
	        view_2817: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17404, [1280, 5120]);  mul_17404 = None
	        mul_17409: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2818: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17399, [mul_17409, 5120]);  mul_17399 = mul_17409 = None
	        permute_300: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2817, [1, 0]);  view_2817 = None
	        mm_default_10: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2818, permute_300);  view_2818 = permute_300 = None
	        add_tensor_10: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_10, model_audio_tower_layers_29_fc2_bias);  mm_default_10 = model_audio_tower_layers_29_fc2_bias = None
	        view_2819: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_10, [sym_size_int, 1500, 1280]);  add_tensor_10 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_27566: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_27268, view_2819);  add_27268 = view_2819 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_241: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_27566, memory_format = torch.contiguous_format)
	        var_mean_60 = torch.ops.aten.var_mean.correction(clone_241, [2], correction = 0, keepdim = True)
	        getitem_240: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_60[0]
	        getitem_241: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_60[1];  var_mean_60 = None
	        sub_8227: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_241, getitem_241);  clone_241 = getitem_241 = None
	        add_27571: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_240, 1e-05);  getitem_240 = None
	        rsqrt_60: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_27571);  add_27571 = None
	        mul_17420: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8227, rsqrt_60);  sub_8227 = rsqrt_60 = None
	        mul_17421: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17420, model_audio_tower_layers_30_self_attn_layer_norm_weight);  mul_17420 = model_audio_tower_layers_30_self_attn_layer_norm_weight = None
	        add_27572: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_17421, model_audio_tower_layers_30_self_attn_layer_norm_bias);  mul_17421 = model_audio_tower_layers_30_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_27572, [2])
	        full_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_180, full_361);  amax_180 = full_361 = None
	        amin_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_27572, [2])
	        full_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_180: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_180, full_360);  amin_180 = full_360 = None
	        sub_8238: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_180, minimum_180);  maximum_180 = None
	        div_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8238, 255.0);  sub_8238 = None
	        clamp_min_540: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_360, 1.1920928955078125e-07);  div_360 = None
	        div_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_180, clamp_min_540);  minimum_180 = None
	        round_361: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_361);  div_361 = None
	        sub_8244: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_361);  round_361 = None
	        clamp_min_541: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8244, -128);  sub_8244 = None
	        clamp_max_360: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_541, 127);  clamp_min_541 = None
	        view_2822: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_540, [sym_size_int, 1500, 1])
	        reciprocal_180: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2822);  view_2822 = None
	        mul_17469: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_180, 1.0);  reciprocal_180 = None
	        mul_17472: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_27572, mul_17469);  mul_17469 = None
	        round_362: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17472);  mul_17472 = None
	        convert_element_type_1080: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_360, torch.int8);  clamp_max_360 = None
	        view_2823: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1080, [sym_size_int, 1500, 1])
	        add_27659: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_362, view_2823);  round_362 = view_2823 = None
	        clamp_min_542: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27659, -128);  add_27659 = None
	        clamp_max_361: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_542, 127);  clamp_min_542 = None
	        convert_element_type_1081: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_361, torch.int8);  clamp_max_361 = None
	        view_2827: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1080, [sym_size_int, 1500, 1]);  convert_element_type_1080 = None
	        convert_element_type_1082: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1081, torch.float32);  convert_element_type_1081 = None
	        convert_element_type_1083: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2827, torch.float32);  view_2827 = None
	        sub_8264: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1082, convert_element_type_1083);  convert_element_type_1082 = convert_element_type_1083 = None
	        view_2826: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_540, [sym_size_int, 1500, 1]);  clamp_min_540 = None
	        mul_17494: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8264, view_2826);  sub_8264 = view_2826 = None
	        view_2829: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2831: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_1084: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2829, torch.float32);  view_2829 = None
	        convert_element_type_1085: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2831, torch.float32);  view_2831 = None
	        sub_8268: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1084, convert_element_type_1085);  convert_element_type_1084 = convert_element_type_1085 = None
	        view_2830: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_17499: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8268, view_2830);  sub_8268 = view_2830 = None
	        view_2832: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17499, [1280, 1280]);  mul_17499 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_27572, [2])
	        full_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_181, full_363);  amax_181 = full_363 = None
	        amin_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_27572, [2])
	        full_362: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_181: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_181, full_362);  amin_181 = full_362 = None
	        sub_8283: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_181, minimum_181);  maximum_181 = None
	        div_362: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8283, 255.0);  sub_8283 = None
	        clamp_min_543: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_362, 1.1920928955078125e-07);  div_362 = None
	        div_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_181, clamp_min_543);  minimum_181 = None
	        round_363: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_363);  div_363 = None
	        sub_8289: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_363);  round_363 = None
	        clamp_min_544: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8289, -128);  sub_8289 = None
	        clamp_max_362: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_544, 127);  clamp_min_544 = None
	        view_2838: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_543, [sym_size_int, 1500, 1])
	        reciprocal_181: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2838);  view_2838 = None
	        mul_17565: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_181, 1.0);  reciprocal_181 = None
	        mul_17568: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_27572, mul_17565);  mul_17565 = None
	        round_364: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17568);  mul_17568 = None
	        convert_element_type_1086: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_362, torch.int8);  clamp_max_362 = None
	        view_2839: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1086, [sym_size_int, 1500, 1])
	        add_27811: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_364, view_2839);  round_364 = view_2839 = None
	        clamp_min_545: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27811, -128);  add_27811 = None
	        clamp_max_363: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_545, 127);  clamp_min_545 = None
	        convert_element_type_1087: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_363, torch.int8);  clamp_max_363 = None
	        view_2843: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1086, [sym_size_int, 1500, 1]);  convert_element_type_1086 = None
	        convert_element_type_1088: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1087, torch.float32);  convert_element_type_1087 = None
	        convert_element_type_1089: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2843, torch.float32);  view_2843 = None
	        sub_8309: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1088, convert_element_type_1089);  convert_element_type_1088 = convert_element_type_1089 = None
	        view_2842: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_543, [sym_size_int, 1500, 1]);  clamp_min_543 = None
	        mul_17590: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8309, view_2842);  sub_8309 = view_2842 = None
	        view_2845: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2847: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_1090: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2845, torch.float32);  view_2845 = None
	        convert_element_type_1091: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2847, torch.float32);  view_2847 = None
	        sub_8313: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1090, convert_element_type_1091);  convert_element_type_1090 = convert_element_type_1091 = None
	        view_2846: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_17595: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8313, view_2846);  sub_8313 = view_2846 = None
	        view_2848: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17595, [1280, 1280]);  mul_17595 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_27572, [2])
	        full_365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_182, full_365);  amax_182 = full_365 = None
	        amin_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_27572, [2])
	        full_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_182: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_182, full_364);  amin_182 = full_364 = None
	        sub_8327: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_182, minimum_182);  maximum_182 = None
	        div_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8327, 255.0);  sub_8327 = None
	        clamp_min_546: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_364, 1.1920928955078125e-07);  div_364 = None
	        div_365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_182, clamp_min_546);  minimum_182 = None
	        round_365: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_365);  div_365 = None
	        sub_8333: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_365);  round_365 = None
	        clamp_min_547: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8333, -128);  sub_8333 = None
	        clamp_max_364: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_547, 127);  clamp_min_547 = None
	        view_2854: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_546, [sym_size_int, 1500, 1])
	        reciprocal_182: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2854);  view_2854 = None
	        mul_17664: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_182, 1.0);  reciprocal_182 = None
	        mul_17667: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_27572, mul_17664);  add_27572 = mul_17664 = None
	        round_366: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17667);  mul_17667 = None
	        convert_element_type_1092: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_364, torch.int8);  clamp_max_364 = None
	        view_2855: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1092, [sym_size_int, 1500, 1])
	        add_27959: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_366, view_2855);  round_366 = view_2855 = None
	        clamp_min_548: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_27959, -128);  add_27959 = None
	        clamp_max_365: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_548, 127);  clamp_min_548 = None
	        convert_element_type_1093: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_365, torch.int8);  clamp_max_365 = None
	        view_2859: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1092, [sym_size_int, 1500, 1]);  convert_element_type_1092 = None
	        convert_element_type_1094: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1093, torch.float32);  convert_element_type_1093 = None
	        convert_element_type_1095: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2859, torch.float32);  view_2859 = None
	        sub_8353: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1094, convert_element_type_1095);  convert_element_type_1094 = convert_element_type_1095 = None
	        view_2858: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_546, [sym_size_int, 1500, 1]);  clamp_min_546 = None
	        mul_17689: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8353, view_2858);  sub_8353 = view_2858 = None
	        view_2861: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2863: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_1096: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2861, torch.float32);  view_2861 = None
	        convert_element_type_1097: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2863, torch.float32);  view_2863 = None
	        sub_8357: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1096, convert_element_type_1097);  convert_element_type_1096 = convert_element_type_1097 = None
	        view_2862: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_17694: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8357, view_2862);  sub_8357 = view_2862 = None
	        view_2864: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17694, [1280, 1280]);  mul_17694 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_17504: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2833: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17494, [mul_17504, 1280]);  mul_17494 = mul_17504 = None
	        permute_301: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2832, [1, 0]);  view_2832 = None
	        mm_default_9: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2833, permute_301);  view_2833 = permute_301 = None
	        add_tensor_9: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_9, model_audio_tower_layers_30_self_attn_q_proj_bias);  mm_default_9 = model_audio_tower_layers_30_self_attn_q_proj_bias = None
	        view_2834: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_9, [sym_size_int, 1500, 1280]);  add_tensor_9 = None
	        mul_17511: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2834, 0.125);  view_2834 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2835: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17511, [sym_size_int, 1500, 20, 64]);  mul_17511 = None
	        permute_302: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2835, [0, 2, 1, 3]);  view_2835 = None
	        clone_242: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_302, memory_format = torch.contiguous_format);  permute_302 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_17598: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2849: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17590, [mul_17598, 1280]);  mul_17590 = mul_17598 = None
	        permute_303: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2848, [1, 0]);  view_2848 = None
	        mm_30: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2849, permute_303);  view_2849 = permute_303 = None
	        view_2850: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_30, [sym_size_int, 1500, 1280]);  mm_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2851: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2850, [sym_size_int, -1, 20, 64]);  view_2850 = None
	        permute_304: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2851, [0, 2, 1, 3]);  view_2851 = None
	        clone_243: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_304, memory_format = torch.contiguous_format);  permute_304 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_17699: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2865: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17689, [mul_17699, 1280]);  mul_17689 = mul_17699 = None
	        permute_305: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2864, [1, 0]);  view_2864 = None
	        mm_default_8: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2865, permute_305);  view_2865 = permute_305 = None
	        add_tensor_8: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_8, model_audio_tower_layers_30_self_attn_v_proj_bias);  mm_default_8 = model_audio_tower_layers_30_self_attn_v_proj_bias = None
	        view_2866: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_8, [sym_size_int, 1500, 1280]);  add_tensor_8 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2867: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2866, [sym_size_int, -1, 20, 64]);  view_2866 = None
	        permute_306: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2867, [0, 2, 1, 3]);  view_2867 = None
	        clone_244: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_306, memory_format = torch.contiguous_format);  permute_306 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_30 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_242, clone_243, clone_244, None, False, scale = 1.0);  clone_242 = clone_243 = clone_244 = None
	        getitem_242: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_30[0];  _scaled_dot_product_efficient_attention_30 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_307: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_242, [0, 2, 1, 3]);  getitem_242 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2868: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_307, [sym_size_int, 1500, -1]);  permute_307 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2868, [2])
	        full_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_183, full_367);  amax_183 = full_367 = None
	        amin_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2868, [2])
	        full_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_183: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_183, full_366);  amin_183 = full_366 = None
	        sub_8375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_183, minimum_183);  maximum_183 = None
	        div_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8375, 255.0);  sub_8375 = None
	        clamp_min_549: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_366, 1.1920928955078125e-07);  div_366 = None
	        div_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_183, clamp_min_549);  minimum_183 = None
	        round_367: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_367);  div_367 = None
	        sub_8381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_367);  round_367 = None
	        clamp_min_550: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8381, -128);  sub_8381 = None
	        clamp_max_366: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_550, 127);  clamp_min_550 = None
	        view_2871: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_549, [sym_size_int, 1500, 1])
	        reciprocal_183: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2871);  view_2871 = None
	        mul_17769: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_183, 1.0);  reciprocal_183 = None
	        mul_17772: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2868, mul_17769);  view_2868 = mul_17769 = None
	        round_368: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17772);  mul_17772 = None
	        convert_element_type_1098: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_366, torch.int8);  clamp_max_366 = None
	        view_2872: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1098, [sym_size_int, 1500, 1])
	        add_28123: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_368, view_2872);  round_368 = view_2872 = None
	        clamp_min_551: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28123, -128);  add_28123 = None
	        clamp_max_367: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_551, 127);  clamp_min_551 = None
	        convert_element_type_1099: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_367, torch.int8);  clamp_max_367 = None
	        view_2876: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1098, [sym_size_int, 1500, 1]);  convert_element_type_1098 = None
	        convert_element_type_1100: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1099, torch.float32);  convert_element_type_1099 = None
	        convert_element_type_1101: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2876, torch.float32);  view_2876 = None
	        sub_8401: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1100, convert_element_type_1101);  convert_element_type_1100 = convert_element_type_1101 = None
	        view_2875: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_549, [sym_size_int, 1500, 1]);  clamp_min_549 = None
	        mul_17794: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8401, view_2875);  sub_8401 = view_2875 = None
	        view_2878: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2880: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_1102: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2878, torch.float32);  view_2878 = None
	        convert_element_type_1103: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2880, torch.float32);  view_2880 = None
	        sub_8405: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1102, convert_element_type_1103);  convert_element_type_1102 = convert_element_type_1103 = None
	        view_2879: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_17799: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8405, view_2879);  sub_8405 = view_2879 = None
	        view_2881: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17799, [1280, 1280]);  mul_17799 = None
	        mul_17804: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2882: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17794, [mul_17804, 1280]);  mul_17794 = mul_17804 = None
	        permute_308: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2881, [1, 0]);  view_2881 = None
	        mm_default_7: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2882, permute_308);  view_2882 = permute_308 = None
	        add_tensor_7: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_7, model_audio_tower_layers_30_self_attn_out_proj_bias);  mm_default_7 = model_audio_tower_layers_30_self_attn_out_proj_bias = None
	        view_2883: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_7, [sym_size_int, 1500, 1280]);  add_tensor_7 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_28186: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_27566, view_2883);  add_27566 = view_2883 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_246: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_28186, memory_format = torch.contiguous_format)
	        var_mean_61 = torch.ops.aten.var_mean.correction(clone_246, [2], correction = 0, keepdim = True)
	        getitem_246: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_61[0]
	        getitem_247: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_61[1];  var_mean_61 = None
	        sub_8411: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_246, getitem_247);  clone_246 = getitem_247 = None
	        add_28191: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_246, 1e-05);  getitem_246 = None
	        rsqrt_61: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_28191);  add_28191 = None
	        mul_17815: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8411, rsqrt_61);  sub_8411 = rsqrt_61 = None
	        mul_17816: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17815, model_audio_tower_layers_30_final_layer_norm_weight);  mul_17815 = model_audio_tower_layers_30_final_layer_norm_weight = None
	        add_28192: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_17816, model_audio_tower_layers_30_final_layer_norm_bias);  mul_17816 = model_audio_tower_layers_30_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_28192, [2])
	        full_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_184, full_369);  amax_184 = full_369 = None
	        amin_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_28192, [2])
	        full_368: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_184: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_184, full_368);  amin_184 = full_368 = None
	        sub_8422: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_184, minimum_184);  maximum_184 = None
	        div_368: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8422, 255.0);  sub_8422 = None
	        clamp_min_552: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_368, 1.1920928955078125e-07);  div_368 = None
	        div_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_184, clamp_min_552);  minimum_184 = None
	        round_369: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_369);  div_369 = None
	        sub_8428: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_369);  round_369 = None
	        clamp_min_553: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8428, -128);  sub_8428 = None
	        clamp_max_368: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_553, 127);  clamp_min_553 = None
	        view_2886: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_552, [sym_size_int, 1500, 1])
	        reciprocal_184: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2886);  view_2886 = None
	        mul_17864: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_184, 1.0);  reciprocal_184 = None
	        mul_17867: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_28192, mul_17864);  add_28192 = mul_17864 = None
	        round_370: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_17867);  mul_17867 = None
	        convert_element_type_1104: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_368, torch.int8);  clamp_max_368 = None
	        view_2887: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1104, [sym_size_int, 1500, 1])
	        add_28279: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_370, view_2887);  round_370 = view_2887 = None
	        clamp_min_554: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28279, -128);  add_28279 = None
	        clamp_max_369: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_554, 127);  clamp_min_554 = None
	        convert_element_type_1105: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_369, torch.int8);  clamp_max_369 = None
	        view_2891: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1104, [sym_size_int, 1500, 1]);  convert_element_type_1104 = None
	        convert_element_type_1106: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1105, torch.float32);  convert_element_type_1105 = None
	        convert_element_type_1107: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2891, torch.float32);  view_2891 = None
	        sub_8448: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1106, convert_element_type_1107);  convert_element_type_1106 = convert_element_type_1107 = None
	        view_2890: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_552, [sym_size_int, 1500, 1]);  clamp_min_552 = None
	        mul_17889: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8448, view_2890);  sub_8448 = view_2890 = None
	        view_2893: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_30_fc1_parametrizations_weight_original0 = None
	        view_2895: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_30_fc1_parametrizations_weight_original2 = None
	        convert_element_type_1108: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2893, torch.float32);  view_2893 = None
	        convert_element_type_1109: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2895, torch.float32);  view_2895 = None
	        sub_8452: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1108, convert_element_type_1109);  convert_element_type_1108 = convert_element_type_1109 = None
	        view_2894: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_30_fc1_parametrizations_weight_original1 = None
	        mul_17894: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8452, view_2894);  sub_8452 = view_2894 = None
	        view_2896: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17894, [5120, 1280]);  mul_17894 = None
	        mul_17899: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2897: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17889, [mul_17899, 1280]);  mul_17889 = mul_17899 = None
	        permute_309: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2896, [1, 0]);  view_2896 = None
	        mm_default_6: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_2897, permute_309);  view_2897 = permute_309 = None
	        add_tensor_6: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_6, model_audio_tower_layers_30_fc1_bias);  mm_default_6 = model_audio_tower_layers_30_fc1_bias = None
	        view_2898: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_6, [sym_size_int, 1500, 5120]);  add_tensor_6 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_17906: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2898, 0.5)
	        mul_17907: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2898, 0.7071067811865476);  view_2898 = None
	        erf_32: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_17907);  mul_17907 = None
	        add_28338: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_32, 1);  erf_32 = None
	        mul_17908: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17906, add_28338);  mul_17906 = add_28338 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_17908, [2])
	        full_371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_185, full_371);  amax_185 = full_371 = None
	        amin_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_17908, [2])
	        full_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_185: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_185, full_370);  amin_185 = full_370 = None
	        sub_8465: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_185, minimum_185);  maximum_185 = None
	        div_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8465, 255.0);  sub_8465 = None
	        clamp_min_555: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_370, 1.1920928955078125e-07);  div_370 = None
	        div_371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_185, clamp_min_555);  minimum_185 = None
	        round_371: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_371);  div_371 = None
	        sub_8471: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_371);  round_371 = None
	        clamp_min_556: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8471, -128);  sub_8471 = None
	        clamp_max_370: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_556, 127);  clamp_min_556 = None
	        view_2901: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_555, [sym_size_int, 1500, 1])
	        reciprocal_185: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2901);  view_2901 = None
	        mul_17954: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_185, 1.0);  reciprocal_185 = None
	        mul_17957: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_17908, mul_17954);  mul_17908 = mul_17954 = None
	        round_372: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_17957);  mul_17957 = None
	        convert_element_type_1110: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_370, torch.int8);  clamp_max_370 = None
	        view_2902: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1110, [sym_size_int, 1500, 1])
	        add_28421: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_372, view_2902);  round_372 = view_2902 = None
	        clamp_min_557: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28421, -128);  add_28421 = None
	        clamp_max_371: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_557, 127);  clamp_min_557 = None
	        convert_element_type_1111: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_371, torch.int8);  clamp_max_371 = None
	        view_2906: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1110, [sym_size_int, 1500, 1]);  convert_element_type_1110 = None
	        convert_element_type_1112: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1111, torch.float32);  convert_element_type_1111 = None
	        convert_element_type_1113: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2906, torch.float32);  view_2906 = None
	        sub_8491: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1112, convert_element_type_1113);  convert_element_type_1112 = convert_element_type_1113 = None
	        view_2905: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_555, [sym_size_int, 1500, 1]);  clamp_min_555 = None
	        mul_17979: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8491, view_2905);  sub_8491 = view_2905 = None
	        view_2908: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_30_fc2_parametrizations_weight_original0 = None
	        view_2910: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_30_fc2_parametrizations_weight_original2 = None
	        convert_element_type_1114: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2908, torch.float32);  view_2908 = None
	        convert_element_type_1115: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2910, torch.float32);  view_2910 = None
	        sub_8495: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1114, convert_element_type_1115);  convert_element_type_1114 = convert_element_type_1115 = None
	        view_2909: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_30_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_30_fc2_parametrizations_weight_original1 = None
	        mul_17984: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8495, view_2909);  sub_8495 = view_2909 = None
	        view_2911: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17984, [1280, 5120]);  mul_17984 = None
	        mul_17989: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2912: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_17979, [mul_17989, 5120]);  mul_17979 = mul_17989 = None
	        permute_310: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_2911, [1, 0]);  view_2911 = None
	        mm_default_5: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2912, permute_310);  view_2912 = permute_310 = None
	        add_tensor_5: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_5, model_audio_tower_layers_30_fc2_bias);  mm_default_5 = model_audio_tower_layers_30_fc2_bias = None
	        view_2913: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_5, [sym_size_int, 1500, 1280]);  add_tensor_5 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_28484: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_28186, view_2913);  add_28186 = view_2913 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:202 in forward, code: hidden_states = self.self_attn_layer_norm(hidden_states)
	        clone_249: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_28484, memory_format = torch.contiguous_format)
	        var_mean_62 = torch.ops.aten.var_mean.correction(clone_249, [2], correction = 0, keepdim = True)
	        getitem_248: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_62[0]
	        getitem_249: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_62[1];  var_mean_62 = None
	        sub_8501: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_249, getitem_249);  clone_249 = getitem_249 = None
	        add_28489: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_248, 1e-05);  getitem_248 = None
	        rsqrt_62: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_28489);  add_28489 = None
	        mul_18000: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8501, rsqrt_62);  sub_8501 = rsqrt_62 = None
	        mul_18001: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18000, model_audio_tower_layers_31_self_attn_layer_norm_weight);  mul_18000 = model_audio_tower_layers_31_self_attn_layer_norm_weight = None
	        add_28490: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_18001, model_audio_tower_layers_31_self_attn_layer_norm_bias);  mul_18001 = model_audio_tower_layers_31_self_attn_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        amax_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_28490, [2])
	        full_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_186, full_373);  amax_186 = full_373 = None
	        amin_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_28490, [2])
	        full_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_186: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_186, full_372);  amin_186 = full_372 = None
	        sub_8512: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_186, minimum_186);  maximum_186 = None
	        div_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8512, 255.0);  sub_8512 = None
	        clamp_min_558: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_372, 1.1920928955078125e-07);  div_372 = None
	        div_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_186, clamp_min_558);  minimum_186 = None
	        round_373: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_373);  div_373 = None
	        sub_8518: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_373);  round_373 = None
	        clamp_min_559: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8518, -128);  sub_8518 = None
	        clamp_max_372: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_559, 127);  clamp_min_559 = None
	        view_2916: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_558, [sym_size_int, 1500, 1])
	        reciprocal_186: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2916);  view_2916 = None
	        mul_18049: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_186, 1.0);  reciprocal_186 = None
	        mul_18052: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_28490, mul_18049);  mul_18049 = None
	        round_374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18052);  mul_18052 = None
	        convert_element_type_1116: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_372, torch.int8);  clamp_max_372 = None
	        view_2917: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1116, [sym_size_int, 1500, 1])
	        add_28577: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_374, view_2917);  round_374 = view_2917 = None
	        clamp_min_560: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28577, -128);  add_28577 = None
	        clamp_max_373: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_560, 127);  clamp_min_560 = None
	        convert_element_type_1117: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_373, torch.int8);  clamp_max_373 = None
	        view_2921: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1116, [sym_size_int, 1500, 1]);  convert_element_type_1116 = None
	        convert_element_type_1118: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1117, torch.float32);  convert_element_type_1117 = None
	        convert_element_type_1119: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2921, torch.float32);  view_2921 = None
	        sub_8538: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1118, convert_element_type_1119);  convert_element_type_1118 = convert_element_type_1119 = None
	        view_2920: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_558, [sym_size_int, 1500, 1]);  clamp_min_558 = None
	        mul_18074: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8538, view_2920);  sub_8538 = view_2920 = None
	        view_2923: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0 = None
	        view_2925: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2 = None
	        convert_element_type_1120: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2923, torch.float32);  view_2923 = None
	        convert_element_type_1121: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2925, torch.float32);  view_2925 = None
	        sub_8542: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1120, convert_element_type_1121);  convert_element_type_1120 = convert_element_type_1121 = None
	        view_2924: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1 = None
	        mul_18079: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8542, view_2924);  sub_8542 = view_2924 = None
	        view_2926: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18079, [1280, 1280]);  mul_18079 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        amax_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_28490, [2])
	        full_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_187, full_375);  amax_187 = full_375 = None
	        amin_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_28490, [2])
	        full_374: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_187: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_187, full_374);  amin_187 = full_374 = None
	        sub_8557: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_187, minimum_187);  maximum_187 = None
	        div_374: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8557, 255.0);  sub_8557 = None
	        clamp_min_561: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_374, 1.1920928955078125e-07);  div_374 = None
	        div_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_187, clamp_min_561);  minimum_187 = None
	        round_375: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_375);  div_375 = None
	        sub_8563: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_375);  round_375 = None
	        clamp_min_562: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8563, -128);  sub_8563 = None
	        clamp_max_374: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_562, 127);  clamp_min_562 = None
	        view_2932: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_561, [sym_size_int, 1500, 1])
	        reciprocal_187: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2932);  view_2932 = None
	        mul_18145: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_187, 1.0);  reciprocal_187 = None
	        mul_18148: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_28490, mul_18145);  mul_18145 = None
	        round_376: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18148);  mul_18148 = None
	        convert_element_type_1122: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_374, torch.int8);  clamp_max_374 = None
	        view_2933: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1122, [sym_size_int, 1500, 1])
	        add_28729: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_376, view_2933);  round_376 = view_2933 = None
	        clamp_min_563: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28729, -128);  add_28729 = None
	        clamp_max_375: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_563, 127);  clamp_min_563 = None
	        convert_element_type_1123: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_375, torch.int8);  clamp_max_375 = None
	        view_2937: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1122, [sym_size_int, 1500, 1]);  convert_element_type_1122 = None
	        convert_element_type_1124: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1123, torch.float32);  convert_element_type_1123 = None
	        convert_element_type_1125: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2937, torch.float32);  view_2937 = None
	        sub_8583: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1124, convert_element_type_1125);  convert_element_type_1124 = convert_element_type_1125 = None
	        view_2936: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_561, [sym_size_int, 1500, 1]);  clamp_min_561 = None
	        mul_18170: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8583, view_2936);  sub_8583 = view_2936 = None
	        view_2939: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0 = None
	        view_2941: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2 = None
	        convert_element_type_1126: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2939, torch.float32);  view_2939 = None
	        convert_element_type_1127: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2941, torch.float32);  view_2941 = None
	        sub_8587: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1126, convert_element_type_1127);  convert_element_type_1126 = convert_element_type_1127 = None
	        view_2940: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1 = None
	        mul_18175: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8587, view_2940);  sub_8587 = view_2940 = None
	        view_2942: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18175, [1280, 1280]);  mul_18175 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        amax_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_28490, [2])
	        full_377: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_188, full_377);  amax_188 = full_377 = None
	        amin_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_28490, [2])
	        full_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_188: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_188, full_376);  amin_188 = full_376 = None
	        sub_8601: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_188, minimum_188);  maximum_188 = None
	        div_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8601, 255.0);  sub_8601 = None
	        clamp_min_564: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_376, 1.1920928955078125e-07);  div_376 = None
	        div_377: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_188, clamp_min_564);  minimum_188 = None
	        round_377: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_377);  div_377 = None
	        sub_8607: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_377);  round_377 = None
	        clamp_min_565: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8607, -128);  sub_8607 = None
	        clamp_max_376: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_565, 127);  clamp_min_565 = None
	        view_2948: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_564, [sym_size_int, 1500, 1])
	        reciprocal_188: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2948);  view_2948 = None
	        mul_18244: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_188, 1.0);  reciprocal_188 = None
	        mul_18247: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_28490, mul_18244);  add_28490 = mul_18244 = None
	        round_378: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18247);  mul_18247 = None
	        convert_element_type_1128: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_376, torch.int8);  clamp_max_376 = None
	        view_2949: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1128, [sym_size_int, 1500, 1])
	        add_28877: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_378, view_2949);  round_378 = view_2949 = None
	        clamp_min_566: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_28877, -128);  add_28877 = None
	        clamp_max_377: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_566, 127);  clamp_min_566 = None
	        convert_element_type_1129: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_377, torch.int8);  clamp_max_377 = None
	        view_2953: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1128, [sym_size_int, 1500, 1]);  convert_element_type_1128 = None
	        convert_element_type_1130: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1129, torch.float32);  convert_element_type_1129 = None
	        convert_element_type_1131: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2953, torch.float32);  view_2953 = None
	        sub_8627: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1130, convert_element_type_1131);  convert_element_type_1130 = convert_element_type_1131 = None
	        view_2952: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_564, [sym_size_int, 1500, 1]);  clamp_min_564 = None
	        mul_18269: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8627, view_2952);  sub_8627 = view_2952 = None
	        view_2955: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0 = None
	        view_2957: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2 = None
	        convert_element_type_1132: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2955, torch.float32);  view_2955 = None
	        convert_element_type_1133: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2957, torch.float32);  view_2957 = None
	        sub_8631: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1132, convert_element_type_1133);  convert_element_type_1132 = convert_element_type_1133 = None
	        view_2956: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1 = None
	        mul_18274: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8631, view_2956);  sub_8631 = view_2956 = None
	        view_2958: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18274, [1280, 1280]);  mul_18274 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:137 in forward, code: query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)
	        mul_18084: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2927: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18074, [mul_18084, 1280]);  mul_18074 = mul_18084 = None
	        permute_311: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2926, [1, 0]);  view_2926 = None
	        mm_default_4: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2927, permute_311);  view_2927 = permute_311 = None
	        add_tensor_4: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_4, model_audio_tower_layers_31_self_attn_q_proj_bias);  mm_default_4 = model_audio_tower_layers_31_self_attn_q_proj_bias = None
	        view_2928: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_4, [sym_size_int, 1500, 1280]);  add_tensor_4 = None
	        mul_18091: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2928, 0.125);  view_2928 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2929: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18091, [sym_size_int, 1500, 20, 64]);  mul_18091 = None
	        permute_312: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2929, [0, 2, 1, 3]);  view_2929 = None
	        clone_250: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_312, memory_format = torch.contiguous_format);  permute_312 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:138 in forward, code: key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
	        mul_18178: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2943: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18170, [mul_18178, 1280]);  mul_18170 = mul_18178 = None
	        permute_313: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2942, [1, 0]);  view_2942 = None
	        mm_31: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2943, permute_313);  view_2943 = permute_313 = None
	        view_2944: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(mm_31, [sym_size_int, 1500, 1280]);  mm_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2945: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2944, [sym_size_int, -1, 20, 64]);  view_2944 = None
	        permute_314: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2945, [0, 2, 1, 3]);  view_2945 = None
	        clone_251: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_314, memory_format = torch.contiguous_format);  permute_314 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:139 in forward, code: value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
	        mul_18279: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2959: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18269, [mul_18279, 1280]);  mul_18269 = mul_18279 = None
	        permute_315: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2958, [1, 0]);  view_2958 = None
	        mm_default_3: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2959, permute_315);  view_2959 = permute_315 = None
	        add_tensor_3: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_3, model_audio_tower_layers_31_self_attn_v_proj_bias);  mm_default_3 = model_audio_tower_layers_31_self_attn_v_proj_bias = None
	        view_2960: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_3, [sym_size_int, 1500, 1280]);  add_tensor_3 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:118 in _shape, code: return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
	        view_2961: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.reshape.default(view_2960, [sym_size_int, -1, 20, 64]);  view_2960 = None
	        permute_316: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = torch.ops.aten.permute.default(view_2961, [0, 2, 1, 3]);  view_2961 = None
	        clone_252: "f32[s6, 20, 1500, 64][1920000, 96000, 64, 1]cuda:0" = torch.ops.aten.clone.default(permute_316, memory_format = torch.contiguous_format);  permute_316 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:89 in sdpa_attention_forward, code: attn_output = torch.nn.functional.scaled_dot_product_attention(
	        _scaled_dot_product_efficient_attention_31 = torch.ops.aten._scaled_dot_product_efficient_attention.default(clone_250, clone_251, clone_252, None, False, scale = 1.0);  clone_250 = clone_251 = clone_252 = None
	        getitem_250: "f32[s6, 20, 1500, 64][1920000, 64, 1280, 1]cuda:0" = _scaled_dot_product_efficient_attention_31[0];  _scaled_dot_product_efficient_attention_31 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py:99 in sdpa_attention_forward, code: attn_output = attn_output.transpose(1, 2).contiguous()
	        permute_317: "f32[s6, 1500, 20, 64][1920000, 1280, 64, 1]cuda:0" = torch.ops.aten.permute.default(getitem_250, [0, 2, 1, 3]);  getitem_250 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:158 in forward, code: attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()
	        view_2962: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(permute_317, [sym_size_int, 1500, -1]);  permute_317 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:159 in forward, code: attn_output = self.out_proj(attn_output)
	        amax_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(view_2962, [2])
	        full_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_189, full_379);  amax_189 = full_379 = None
	        amin_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(view_2962, [2])
	        full_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_189: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_189, full_378);  amin_189 = full_378 = None
	        sub_8649: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_189, minimum_189);  maximum_189 = None
	        div_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8649, 255.0);  sub_8649 = None
	        clamp_min_567: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_378, 1.1920928955078125e-07);  div_378 = None
	        div_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_189, clamp_min_567);  minimum_189 = None
	        round_379: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_379);  div_379 = None
	        sub_8655: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_379);  round_379 = None
	        clamp_min_568: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8655, -128);  sub_8655 = None
	        clamp_max_378: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_568, 127);  clamp_min_568 = None
	        view_2965: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_567, [sym_size_int, 1500, 1])
	        reciprocal_189: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2965);  view_2965 = None
	        mul_18349: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_189, 1.0);  reciprocal_189 = None
	        mul_18352: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2962, mul_18349);  view_2962 = mul_18349 = None
	        round_380: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18352);  mul_18352 = None
	        convert_element_type_1134: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_378, torch.int8);  clamp_max_378 = None
	        view_2966: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1134, [sym_size_int, 1500, 1])
	        add_29041: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_380, view_2966);  round_380 = view_2966 = None
	        clamp_min_569: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29041, -128);  add_29041 = None
	        clamp_max_379: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_569, 127);  clamp_min_569 = None
	        convert_element_type_1135: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_379, torch.int8);  clamp_max_379 = None
	        view_2970: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1134, [sym_size_int, 1500, 1]);  convert_element_type_1134 = None
	        convert_element_type_1136: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1135, torch.float32);  convert_element_type_1135 = None
	        convert_element_type_1137: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2970, torch.float32);  view_2970 = None
	        sub_8675: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1136, convert_element_type_1137);  convert_element_type_1136 = convert_element_type_1137 = None
	        view_2969: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_567, [sym_size_int, 1500, 1]);  clamp_min_567 = None
	        mul_18374: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8675, view_2969);  sub_8675 = view_2969 = None
	        view_2972: "i8[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0, [1280, 40, 32]);  model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0 = None
	        view_2974: "i8[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2 = None
	        convert_element_type_1138: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2972, torch.float32);  view_2972 = None
	        convert_element_type_1139: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2974, torch.float32);  view_2974 = None
	        sub_8679: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1138, convert_element_type_1139);  convert_element_type_1138 = convert_element_type_1139 = None
	        view_2973: "f32[1280, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1, [1280, 40, 1]);  model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1 = None
	        mul_18379: "f32[1280, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8679, view_2973);  sub_8679 = view_2973 = None
	        view_2975: "f32[1280, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18379, [1280, 1280]);  mul_18379 = None
	        mul_18384: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2976: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18374, [mul_18384, 1280]);  mul_18374 = mul_18384 = None
	        permute_318: "f32[1280, 1280][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2975, [1, 0]);  view_2975 = None
	        mm_default_2: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_2976, permute_318);  view_2976 = permute_318 = None
	        add_tensor_2: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_2, model_audio_tower_layers_31_self_attn_out_proj_bias);  mm_default_2 = model_audio_tower_layers_31_self_attn_out_proj_bias = None
	        view_2977: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_2, [sym_size_int, 1500, 1280]);  add_tensor_2 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:210 in forward, code: hidden_states = residual + hidden_states
	        add_29104: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_28484, view_2977);  add_28484 = view_2977 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:213 in forward, code: hidden_states = self.final_layer_norm(hidden_states)
	        clone_254: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_29104, memory_format = torch.contiguous_format)
	        var_mean_63 = torch.ops.aten.var_mean.correction(clone_254, [2], correction = 0, keepdim = True)
	        getitem_254: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_63[0]
	        getitem_255: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_63[1];  var_mean_63 = None
	        sub_8685: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_254, getitem_255);  clone_254 = getitem_255 = None
	        add_29109: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_254, 1e-05);  getitem_254 = None
	        rsqrt_63: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_29109);  add_29109 = None
	        mul_18395: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8685, rsqrt_63);  sub_8685 = rsqrt_63 = None
	        mul_18396: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18395, model_audio_tower_layers_31_final_layer_norm_weight);  mul_18395 = model_audio_tower_layers_31_final_layer_norm_weight = None
	        add_29110: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_18396, model_audio_tower_layers_31_final_layer_norm_bias);  mul_18396 = model_audio_tower_layers_31_final_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:214 in forward, code: hidden_states = self.activation_fn(self.fc1(hidden_states))
	        amax_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(add_29110, [2])
	        full_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_190, full_381);  amax_190 = full_381 = None
	        amin_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(add_29110, [2])
	        full_380: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_190: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_190, full_380);  amin_190 = full_380 = None
	        sub_8696: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_190, minimum_190);  maximum_190 = None
	        div_380: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8696, 255.0);  sub_8696 = None
	        clamp_min_570: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_380, 1.1920928955078125e-07);  div_380 = None
	        div_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_190, clamp_min_570);  minimum_190 = None
	        round_381: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_381);  div_381 = None
	        sub_8702: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_381);  round_381 = None
	        clamp_min_571: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8702, -128);  sub_8702 = None
	        clamp_max_380: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_571, 127);  clamp_min_571 = None
	        view_2980: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_570, [sym_size_int, 1500, 1])
	        reciprocal_190: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2980);  view_2980 = None
	        mul_18444: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_190, 1.0);  reciprocal_190 = None
	        mul_18447: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(add_29110, mul_18444);  add_29110 = mul_18444 = None
	        round_382: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.round.default(mul_18447);  mul_18447 = None
	        convert_element_type_1140: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_380, torch.int8);  clamp_max_380 = None
	        view_2981: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1140, [sym_size_int, 1500, 1])
	        add_29197: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(round_382, view_2981);  round_382 = view_2981 = None
	        clamp_min_572: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29197, -128);  add_29197 = None
	        clamp_max_381: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_572, 127);  clamp_min_572 = None
	        convert_element_type_1141: "i8[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_381, torch.int8);  clamp_max_381 = None
	        view_2985: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1140, [sym_size_int, 1500, 1]);  convert_element_type_1140 = None
	        convert_element_type_1142: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1141, torch.float32);  convert_element_type_1141 = None
	        convert_element_type_1143: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2985, torch.float32);  view_2985 = None
	        sub_8722: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1142, convert_element_type_1143);  convert_element_type_1142 = convert_element_type_1143 = None
	        view_2984: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_570, [sym_size_int, 1500, 1]);  clamp_min_570 = None
	        mul_18469: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8722, view_2984);  sub_8722 = view_2984 = None
	        view_2987: "i8[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_fc1_parametrizations_weight_original0, [5120, 40, 32]);  model_audio_tower_layers_31_fc1_parametrizations_weight_original0 = None
	        view_2989: "i8[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_fc1_parametrizations_weight_original2, [5120, 40, 1]);  model_audio_tower_layers_31_fc1_parametrizations_weight_original2 = None
	        convert_element_type_1144: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2987, torch.float32);  view_2987 = None
	        convert_element_type_1145: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_2989, torch.float32);  view_2989 = None
	        sub_8726: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1144, convert_element_type_1145);  convert_element_type_1144 = convert_element_type_1145 = None
	        view_2988: "f32[5120, 40, 1][40, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_fc1_parametrizations_weight_original1, [5120, 40, 1]);  model_audio_tower_layers_31_fc1_parametrizations_weight_original1 = None
	        mul_18474: "f32[5120, 40, 32][1280, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8726, view_2988);  sub_8726 = view_2988 = None
	        view_2990: "f32[5120, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18474, [5120, 1280]);  mul_18474 = None
	        mul_18479: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_2991: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18469, [mul_18479, 1280]);  mul_18469 = mul_18479 = None
	        permute_319: "f32[1280, 5120][1, 1280]cuda:0" = torch.ops.aten.permute.default(view_2990, [1, 0]);  view_2990 = None
	        mm_default_1: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mm.default(view_2991, permute_319);  view_2991 = permute_319 = None
	        add_tensor_1: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default_1, model_audio_tower_layers_31_fc1_bias);  mm_default_1 = model_audio_tower_layers_31_fc1_bias = None
	        view_2992: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor_1, [sym_size_int, 1500, 5120]);  add_tensor_1 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_18486: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2992, 0.5)
	        mul_18487: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_2992, 0.7071067811865476);  view_2992 = None
	        erf_33: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.erf.default(mul_18487);  mul_18487 = None
	        add_29256: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_33, 1);  erf_33 = None
	        mul_18488: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18486, add_29256);  mul_18486 = add_29256 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:216 in forward, code: hidden_states = self.fc2(hidden_states)
	        amax_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amax.default(mul_18488, [2])
	        full_383: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.maximum.default(amax_191, full_383);  amax_191 = full_383 = None
	        amin_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.amin.default(mul_18488, [2])
	        full_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.full.default([sym_size_int, 1500], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_191: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.minimum.default(amin_191, full_382);  amin_191 = full_382 = None
	        sub_8739: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_191, minimum_191);  maximum_191 = None
	        div_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(sub_8739, 255.0);  sub_8739 = None
	        clamp_min_573: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(div_382, 1.1920928955078125e-07);  div_382 = None
	        div_383: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.div.Tensor(minimum_191, clamp_min_573);  minimum_191 = None
	        round_383: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.round.default(div_383);  div_383 = None
	        sub_8745: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_383);  round_383 = None
	        clamp_min_574: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8745, -128);  sub_8745 = None
	        clamp_max_382: "f32[s6, 1500][1500, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_574, 127);  clamp_min_574 = None
	        view_2995: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_573, [sym_size_int, 1500, 1])
	        reciprocal_191: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_2995);  view_2995 = None
	        mul_18534: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_191, 1.0);  reciprocal_191 = None
	        mul_18537: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18488, mul_18534);  mul_18488 = mul_18534 = None
	        round_384: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.round.default(mul_18537);  mul_18537 = None
	        convert_element_type_1146: "i8[s6, 1500][1500, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_382, torch.int8);  clamp_max_382 = None
	        view_2996: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1146, [sym_size_int, 1500, 1])
	        add_29339: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_384, view_2996);  round_384 = view_2996 = None
	        clamp_min_575: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29339, -128);  add_29339 = None
	        clamp_max_383: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_575, 127);  clamp_min_575 = None
	        convert_element_type_1147: "i8[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_383, torch.int8);  clamp_max_383 = None
	        view_3000: "i8[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1146, [sym_size_int, 1500, 1]);  convert_element_type_1146 = None
	        convert_element_type_1148: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1147, torch.float32);  convert_element_type_1147 = None
	        convert_element_type_1149: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3000, torch.float32);  view_3000 = None
	        sub_8765: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1148, convert_element_type_1149);  convert_element_type_1148 = convert_element_type_1149 = None
	        view_2999: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_573, [sym_size_int, 1500, 1]);  clamp_min_573 = None
	        mul_18559: "f32[s6, 1500, 5120][7680000, 5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8765, view_2999);  sub_8765 = view_2999 = None
	        view_3002: "i8[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_fc2_parametrizations_weight_original0, [1280, 160, 32]);  model_audio_tower_layers_31_fc2_parametrizations_weight_original0 = None
	        view_3004: "i8[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_fc2_parametrizations_weight_original2, [1280, 160, 1]);  model_audio_tower_layers_31_fc2_parametrizations_weight_original2 = None
	        convert_element_type_1150: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3002, torch.float32);  view_3002 = None
	        convert_element_type_1151: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3004, torch.float32);  view_3004 = None
	        sub_8769: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1150, convert_element_type_1151);  convert_element_type_1150 = convert_element_type_1151 = None
	        view_3003: "f32[1280, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_audio_tower_layers_31_fc2_parametrizations_weight_original1, [1280, 160, 1]);  model_audio_tower_layers_31_fc2_parametrizations_weight_original1 = None
	        mul_18564: "f32[1280, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8769, view_3003);  sub_8769 = view_3003 = None
	        view_3005: "f32[1280, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18564, [1280, 5120]);  mul_18564 = None
	        mul_18569: "Sym(1500 * s6)" = sym_size_int * 1500
	        view_3006: "f32[1500*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18559, [mul_18569, 5120]);  mul_18559 = mul_18569 = None
	        permute_320: "f32[5120, 1280][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_3005, [1, 0]);  view_3005 = None
	        mm_default: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.mm.default(view_3006, permute_320);  view_3006 = permute_320 = None
	        add_tensor: "f32[1500*s6, 1280][1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mm_default, model_audio_tower_layers_31_fc2_bias);  mm_default = model_audio_tower_layers_31_fc2_bias = None
	        view_3007: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.reshape.default(add_tensor, [sym_size_int, 1500, 1280]);  add_tensor = sym_size_int = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:218 in forward, code: hidden_states = residual + hidden_states
	        add_29402: "f32[s6, 1500, 1280][1920000, 1, 1500]cuda:0" = torch.ops.aten.add.Tensor(add_29104, view_3007);  add_29104 = view_3007 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:365 in forward, code: hidden_states = self.layer_norm(hidden_states)
	        clone_257: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.clone.default(add_29402, memory_format = torch.contiguous_format);  add_29402 = None
	        var_mean_64 = torch.ops.aten.var_mean.correction(clone_257, [2], correction = 0, keepdim = True)
	        getitem_256: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_64[0]
	        getitem_257: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = var_mean_64[1];  var_mean_64 = None
	        sub_8775: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.sub.Tensor(clone_257, getitem_257);  clone_257 = getitem_257 = None
	        add_29407: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.add.Tensor(getitem_256, 1e-05);  getitem_256 = None
	        rsqrt_64: "f32[s6, 1500, 1][1500, 1, 1]cuda:0" = torch.ops.aten.rsqrt.default(add_29407);  add_29407 = None
	        mul_18580: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8775, rsqrt_64);  sub_8775 = rsqrt_64 = None
	        mul_18581: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18580, model_audio_tower_layer_norm_weight);  mul_18580 = model_audio_tower_layer_norm_weight = None
	        add_29408: "f32[s6, 1500, 1280][1920000, 1280, 1]cuda:0" = torch.ops.aten.add.Tensor(mul_18581, model_audio_tower_layer_norm_bias);  mul_18581 = model_audio_tower_layer_norm_bias = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:451 in get_audio_embeds, code: audio_hidden_states = audio_hidden_states.reshape(-1, self.config.audio_config.intermediate_size)
	        view_3008: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(add_29408, [-1, 5120]);  add_29408 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:389 in forward, code: hidden_states = self.linear_1(audio_features)
	        amax_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.amax.default(view_3008, [1])
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:451 in get_audio_embeds, code: audio_hidden_states = audio_hidden_states.reshape(-1, self.config.audio_config.intermediate_size)
	        sym_size_int_193: "Sym(375 * s6)" = torch.ops.aten.sym_size.int(view_3008, 0)
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:389 in forward, code: hidden_states = self.linear_1(audio_features)
	        full_385: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.maximum.default(amax_192, full_385);  amax_192 = full_385 = None
	        amin_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.amin.default(view_3008, [1])
	        full_384: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_192: "f32[375*s6][1]cuda:0" = torch.ops.aten.minimum.default(amin_192, full_384);  amin_192 = full_384 = None
	        sub_8787: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_192, minimum_192);  maximum_192 = None
	        div_384: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(sub_8787, 255.0);  sub_8787 = None
	        clamp_min_576: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(div_384, 1.1920928955078125e-07);  div_384 = None
	        div_385: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(minimum_192, clamp_min_576);  minimum_192 = None
	        round_385: "f32[375*s6][1]cuda:0" = torch.ops.aten.round.default(div_385);  div_385 = None
	        sub_8793: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_385);  round_385 = None
	        clamp_min_577: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8793, -128);  sub_8793 = None
	        clamp_max_384: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_577, 127);  clamp_min_577 = None
	        view_3011: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_576, [sym_size_int_193, 1])
	        reciprocal_192: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_3011);  view_3011 = None
	        mul_18613: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_192, 1.0);  reciprocal_192 = None
	        mul_18615: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(view_3008, mul_18613);  view_3008 = mul_18613 = None
	        round_386: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.round.default(mul_18615);  mul_18615 = None
	        convert_element_type_1152: "i8[375*s6][1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_384, torch.int8);  clamp_max_384 = None
	        view_3012: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1152, [sym_size_int_193, 1])
	        add_29476: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.add.Tensor(round_386, view_3012);  round_386 = view_3012 = None
	        clamp_min_578: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29476, -128);  add_29476 = None
	        clamp_max_385: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_578, 127);  clamp_min_578 = None
	        convert_element_type_1153: "i8[375*s6, 5120][5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_385, torch.int8);  clamp_max_385 = None
	        view_3016: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1152, [sym_size_int_193, 1]);  convert_element_type_1152 = None
	        convert_element_type_1154: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1153, torch.float32);  convert_element_type_1153 = None
	        convert_element_type_1155: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3016, torch.float32);  view_3016 = None
	        sub_8813: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1154, convert_element_type_1155);  convert_element_type_1154 = convert_element_type_1155 = None
	        view_3015: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_576, [sym_size_int_193, 1]);  clamp_min_576 = None
	        mul_18634: "f32[375*s6, 5120][5120, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8813, view_3015);  sub_8813 = view_3015 = None
	        view_3018: "i8[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_multi_modal_projector_linear_1_parametrizations_weight_original0, [3072, 160, 32]);  model_multi_modal_projector_linear_1_parametrizations_weight_original0 = None
	        view_3020: "i8[3072, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_multi_modal_projector_linear_1_parametrizations_weight_original2, [3072, 160, 1]);  model_multi_modal_projector_linear_1_parametrizations_weight_original2 = None
	        convert_element_type_1156: "f32[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3018, torch.float32);  view_3018 = None
	        convert_element_type_1157: "f32[3072, 160, 1][160, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3020, torch.float32);  view_3020 = None
	        sub_8817: "f32[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1156, convert_element_type_1157);  convert_element_type_1156 = convert_element_type_1157 = None
	        view_3019: "f32[3072, 160, 1][160, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_multi_modal_projector_linear_1_parametrizations_weight_original1, [3072, 160, 1]);  model_multi_modal_projector_linear_1_parametrizations_weight_original1 = None
	        mul_18639: "f32[3072, 160, 32][5120, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8817, view_3019);  sub_8817 = view_3019 = None
	        view_3021: "f32[3072, 5120][5120, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18639, [3072, 5120]);  mul_18639 = None
	        permute_321: "f32[5120, 3072][1, 5120]cuda:0" = torch.ops.aten.permute.default(view_3021, [1, 0]);  view_3021 = None
	        mm_32: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mm.default(mul_18634, permute_321);  mul_18634 = permute_321 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py:69 in forward, code: return self.act(input)
	        mul_18642: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(mm_32, 0.5)
	        mul_18643: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(mm_32, 0.7071067811865476);  mm_32 = None
	        erf_34: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.erf.default(mul_18643);  mul_18643 = None
	        add_29516: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.add.Tensor(erf_34, 1);  erf_34 = None
	        mul_18644: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18642, add_29516);  mul_18642 = add_29516 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py:391 in forward, code: hidden_states = self.linear_2(hidden_states)
	        amax_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.amax.default(mul_18644, [1])
	        full_387: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        maximum_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.maximum.default(amax_193, full_387);  amax_193 = full_387 = None
	        amin_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.amin.default(mul_18644, [1])
	        full_386: "f32[375*s6][1]cuda:0" = torch.ops.aten.full.default([sym_size_int_193], 0, dtype = torch.float32, layout = torch.strided, device = device(type='cuda', index=0), pin_memory = False)
	        minimum_193: "f32[375*s6][1]cuda:0" = torch.ops.aten.minimum.default(amin_193, full_386);  amin_193 = full_386 = None
	        sub_8827: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(maximum_193, minimum_193);  maximum_193 = None
	        div_386: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(sub_8827, 255.0);  sub_8827 = None
	        clamp_min_579: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(div_386, 1.1920928955078125e-07);  div_386 = None
	        div_387: "f32[375*s6][1]cuda:0" = torch.ops.aten.div.Tensor(minimum_193, clamp_min_579);  minimum_193 = None
	        round_387: "f32[375*s6][1]cuda:0" = torch.ops.aten.round.default(div_387);  div_387 = None
	        sub_8833: "f32[375*s6][1]cuda:0" = torch.ops.aten.sub.Tensor(-128, round_387);  round_387 = None
	        clamp_min_580: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_min.default(sub_8833, -128);  sub_8833 = None
	        clamp_max_386: "f32[375*s6][1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_580, 127);  clamp_min_580 = None
	        view_3024: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_579, [sym_size_int_193, 1])
	        reciprocal_193: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reciprocal.default(view_3024);  view_3024 = None
	        mul_18666: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.mul.Tensor(reciprocal_193, 1.0);  reciprocal_193 = None
	        mul_18668: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(mul_18644, mul_18666);  mul_18644 = mul_18666 = None
	        round_388: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.round.default(mul_18668);  mul_18668 = None
	        convert_element_type_1158: "i8[375*s6][1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_386, torch.int8);  clamp_max_386 = None
	        view_3025: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1158, [sym_size_int_193, 1])
	        add_29572: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.add.Tensor(round_388, view_3025);  round_388 = view_3025 = None
	        clamp_min_581: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.clamp_min.default(add_29572, -128);  add_29572 = None
	        clamp_max_387: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.clamp_max.default(clamp_min_581, 127);  clamp_min_581 = None
	        convert_element_type_1159: "i8[375*s6, 3072][3072, 1]cuda:0" = torch.ops.prims.convert_element_type.default(clamp_max_387, torch.int8);  clamp_max_387 = None
	        view_3029: "i8[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reshape.default(convert_element_type_1158, [sym_size_int_193, 1]);  convert_element_type_1158 = None
	        convert_element_type_1160: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.prims.convert_element_type.default(convert_element_type_1159, torch.float32);  convert_element_type_1159 = None
	        convert_element_type_1161: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3029, torch.float32);  view_3029 = None
	        sub_8853: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1160, convert_element_type_1161);  convert_element_type_1160 = convert_element_type_1161 = None
	        view_3028: "f32[375*s6, 1][1, 1]cuda:0" = torch.ops.aten.reshape.default(clamp_min_579, [sym_size_int_193, 1]);  clamp_min_579 = sym_size_int_193 = None
	        mul_18687: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8853, view_3028);  sub_8853 = view_3028 = None
	        view_3031: "i8[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.aten.reshape.default(model_multi_modal_projector_linear_2_parametrizations_weight_original0, [3072, 96, 32]);  model_multi_modal_projector_linear_2_parametrizations_weight_original0 = None
	        view_3033: "i8[3072, 96, 1][96, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_multi_modal_projector_linear_2_parametrizations_weight_original2, [3072, 96, 1]);  model_multi_modal_projector_linear_2_parametrizations_weight_original2 = None
	        convert_element_type_1162: "f32[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3031, torch.float32);  view_3031 = None
	        convert_element_type_1163: "f32[3072, 96, 1][96, 1, 1]cuda:0" = torch.ops.prims.convert_element_type.default(view_3033, torch.float32);  view_3033 = None
	        sub_8857: "f32[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.aten.sub.Tensor(convert_element_type_1162, convert_element_type_1163);  convert_element_type_1162 = convert_element_type_1163 = None
	        view_3032: "f32[3072, 96, 1][96, 1, 1]cuda:0" = torch.ops.aten.reshape.default(model_multi_modal_projector_linear_2_parametrizations_weight_original1, [3072, 96, 1]);  model_multi_modal_projector_linear_2_parametrizations_weight_original1 = None
	        mul_18692: "f32[3072, 96, 32][3072, 32, 1]cuda:0" = torch.ops.aten.mul.Tensor(sub_8857, view_3032);  sub_8857 = view_3032 = None
	        view_3034: "f32[3072, 3072][3072, 1]cuda:0" = torch.ops.aten.reshape.default(mul_18692, [3072, 3072]);  mul_18692 = None
	        permute_322: "f32[3072, 3072][1, 3072]cuda:0" = torch.ops.aten.permute.default(view_3034, [1, 0]);  view_3034 = None
	        mm_33: "f32[375*s6, 3072][3072, 1]cuda:0" = torch.ops.aten.mm.default(mul_18687, permute_322);  mul_18687 = permute_322 = None
	        
	         # File: /home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py:83 in forward, code: return audio_embeds.unsqueeze(0)
	        unsqueeze: "f32[1, 375*s6, 3072][1152000*s6, 3072, 1]cuda:0" = torch.ops.aten.unsqueeze.default(mm_33, 0);  mm_33 = None
	        return (unsqueeze,)
	        
V0910 09:43:05.443000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c1b9e92d41aa11b13aa60a6f5b4b2183"}
	{
	"name": "GraphLowering.run",
	"ts": 1757522585443591.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:06.068000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_inductor/graph.py", 37]}
V0910 09:43:06.069000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_inductor/lowering.py", 38]}
V0910 09:43:06.070000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_inductor/kernel/conv.py", 39]}
V0910 09:43:06.071000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/fx/experimental/symbolic_shapes.py:7190] {"guard_added_fast": {"expr": "Ne(s6, 1)", "user_stack": [], "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2657, "name": "compile_fx", "filename": 19, "loc": "return inference_compiler(unlifted_gm, example_inputs_)"}, {"line": 1267, "name": "__call__", "filename": 33, "loc": "return self.compiler_fn(gm, example_inputs)"}, {"line": 2543, "name": "fw_compiler_base", "filename": 19, "loc": "return compile_fx_forward("}, {"line": 2260, "name": "compile_fx_forward", "filename": 19, "loc": "return inner_compile("}, {"line": 81, "name": "inner", "filename": 34, "loc": "return func(*args, **kwds)"}, {"line": 781, "name": "compile_fx_inner", "filename": 19, "loc": "return wrap_compiler_debug(_compile_fx_inner, compiler_name=\"inductor\")("}, {"line": 144, "name": "debug_wrapper", "filename": 35, "loc": "inner_compiled_fn = compiler_fn(gm, example_inputs)"}, {"line": 167, "name": "newFunction", "filename": 36, "loc": "return old_func(*args, **kwargs)"}, {"line": 962, "name": "_compile_fx_inner", "filename": 19, "loc": "mb_compiled_graph = fx_codegen_and_compile("}, {"line": 1694, "name": "fx_codegen_and_compile", "filename": 19, "loc": "return scheme.codegen_and_compile(gm, example_inputs, inputs_to_check, graph_kwargs)"}, {"line": 1419, "name": "codegen_and_compile", "filename": 19, "loc": "graph.run(*example_inputs)"}, {"line": 937, "name": "run", "filename": 37, "loc": "return super().run(*args)"}, {"line": 174, "name": "run", "filename": 24, "loc": "self.env[node] = self.run_node(node)"}, {"line": 1624, "name": "run_node", "filename": 37, "loc": "result = super().run_node(n)"}, {"line": 256, "name": "run_node", "filename": 24, "loc": "return getattr(self, n.op)(n.target, args, kwargs)"}, {"line": 1279, "name": "call_function", "filename": 37, "loc": "out = lowerings[target](*args, **kwargs)  # type: ignore[index]"}, {"line": 458, "name": "wrapped", "filename": 38, "loc": "out = decomp_fn(*args, **kwargs)"}, {"line": 520, "name": "convolution", "filename": 39, "loc": "return L[aten.add]("}, {"line": 451, "name": "wrapped", "filename": 38, "loc": "args, kwargs = transform_args("}, {"line": 364, "name": "transform_args", "filename": 38, "loc": "broadcasted = broadcast_tensors("}, {"line": 458, "name": "wrapped", "filename": 38, "loc": "out = decomp_fn(*args, **kwargs)"}, {"line": 945, "name": "broadcast_tensors", "filename": 38, "loc": "target: list[sympy.Expr] = functools.reduce("}, {"line": 500, "name": "broadcast_symbolic_shapes", "filename": 38, "loc": "if V.graph.sizevars.shape_env.evaluate_expr("}, {"line": 7339, "name": "evaluate_expr", "filename": 14, "loc": "return self._inner_evaluate_expr("}, {"line": 272, "name": "wrapper", "filename": 15, "loc": "return retlog(fn(*args, **kwargs))"}, {"line": 7362, "name": "_inner_evaluate_expr", "filename": 14, "loc": "return self._evaluate_expr("}, {"line": 7615, "name": "_evaluate_expr", "filename": 14, "loc": "self._log_guard(\"eval\", g, forcing_spec=forcing_spec)"}, {"line": 7190, "name": "_log_guard", "filename": 14, "loc": "trace_structured("}, {"line": 1346, "name": "trace_structured", "filename": 16, "loc": "record[name] = metadata_fn()"}]}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2657, "name": "compile_fx", "filename": 19, "loc": "return inference_compiler(unlifted_gm, example_inputs_)"}, {"line": 1267, "name": "__call__", "filename": 33, "loc": "return self.compiler_fn(gm, example_inputs)"}, {"line": 2543, "name": "fw_compiler_base", "filename": 19, "loc": "return compile_fx_forward("}, {"line": 2260, "name": "compile_fx_forward", "filename": 19, "loc": "return inner_compile("}, {"line": 81, "name": "inner", "filename": 34, "loc": "return func(*args, **kwds)"}, {"line": 781, "name": "compile_fx_inner", "filename": 19, "loc": "return wrap_compiler_debug(_compile_fx_inner, compiler_name=\"inductor\")("}, {"line": 144, "name": "debug_wrapper", "filename": 35, "loc": "inner_compiled_fn = compiler_fn(gm, example_inputs)"}, {"line": 167, "name": "newFunction", "filename": 36, "loc": "return old_func(*args, **kwargs)"}, {"line": 962, "name": "_compile_fx_inner", "filename": 19, "loc": "mb_compiled_graph = fx_codegen_and_compile("}, {"line": 1694, "name": "fx_codegen_and_compile", "filename": 19, "loc": "return scheme.codegen_and_compile(gm, example_inputs, inputs_to_check, graph_kwargs)"}, {"line": 1419, "name": "codegen_and_compile", "filename": 19, "loc": "graph.run(*example_inputs)"}, {"line": 937, "name": "run", "filename": 37, "loc": "return super().run(*args)"}, {"line": 174, "name": "run", "filename": 24, "loc": "self.env[node] = self.run_node(node)"}, {"line": 1624, "name": "run_node", "filename": 37, "loc": "result = super().run_node(n)"}, {"line": 256, "name": "run_node", "filename": 24, "loc": "return getattr(self, n.op)(n.target, args, kwargs)"}, {"line": 1279, "name": "call_function", "filename": 37, "loc": "out = lowerings[target](*args, **kwargs)  # type: ignore[index]"}, {"line": 458, "name": "wrapped", "filename": 38, "loc": "out = decomp_fn(*args, **kwargs)"}, {"line": 520, "name": "convolution", "filename": 39, "loc": "return L[aten.add]("}, {"line": 451, "name": "wrapped", "filename": 38, "loc": "args, kwargs = transform_args("}, {"line": 364, "name": "transform_args", "filename": 38, "loc": "broadcasted = broadcast_tensors("}, {"line": 458, "name": "wrapped", "filename": 38, "loc": "out = decomp_fn(*args, **kwargs)"}, {"line": 945, "name": "broadcast_tensors", "filename": 38, "loc": "target: list[sympy.Expr] = functools.reduce("}, {"line": 500, "name": "broadcast_symbolic_shapes", "filename": 38, "loc": "if V.graph.sizevars.shape_env.evaluate_expr("}, {"line": 7339, "name": "evaluate_expr", "filename": 14, "loc": "return self._inner_evaluate_expr("}, {"line": 272, "name": "wrapper", "filename": 15, "loc": "return retlog(fn(*args, **kwargs))"}, {"line": 7362, "name": "_inner_evaluate_expr", "filename": 14, "loc": "return self._evaluate_expr("}, {"line": 7615, "name": "_evaluate_expr", "filename": 14, "loc": "self._log_guard(\"eval\", g, forcing_spec=forcing_spec)"}, {"line": 7190, "name": "_log_guard", "filename": 14, "loc": "trace_structured("}]}
V0910 09:43:25.124000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "425fad9707f21d727f6a5a8f32063116"}
	{
	"name": "GraphLowering.run",
	"ts": 1757522605124520.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:25.126000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "80820b3d447c757d2dcbbfaf20455858"}
	{
	"name": "GraphLowering.compile_to_fn",
	"ts": 1757522605126593.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:25.128000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e703fd491179207e915d43afb860db8f"}
	{
	"name": "GraphLowering.codegen",
	"ts": 1757522605128036.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:25.136000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9aafbbca7bcd6a65dfc519bdefc095f5"}
	{
	"name": "Scheduler.__init__",
	"ts": 1757522605135935.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:29.577000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1ab8515d5b97f9737531c5f6003db81d"}
	{
	"name": "Scheduler.fused_nodes",
	"ts": 1757522609577492.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:36.748000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "76d8b8323625152cede9f24bdcad4f14"}
	{
	"name": "Scheduler.fused_nodes",
	"ts": 1757522616748750.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:36.885000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4997ec45877452a5eda084c4a98093f1"}
	{
	"name": "Scheduler.__init__",
	"ts": 1757522616885755.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:36.887000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e51e63b0897a763e4f5fddf079c6b698"}
	{
	"name": "Scheduler.codegen",
	"ts": 1757522616887618.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.233000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "47f36f30b66388a2a90c82ff72f7e60e"}
	{
	"name": "Scheduler.codegen",
	"ts": 1757522632233074.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.234000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7417ca946d497bcb68b72fbdad3ba5ea"}
	{
	"name": "CppWrapperGpu.generate",
	"ts": 1757522632234775.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.236000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "15785a61d4ae7ec86028af5188057803"}
	{
	"name": "CppWrapperCpu.generate",
	"ts": 1757522632236127.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.239000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "af30dd446980eb18ec630db7c2942a9d"}
	{
	"name": "PythonWrapperCodegen.generate",
	"ts": 1757522632239841.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.605000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "da0375fc8cd83caf6bbbd5347f49654f"}
	{
	"name": "async_compile.wait",
	"ts": 1757522632605647.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.608000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ee3ec5dc52ae33bfdd693d920562400b"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632607970.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.609000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b227a4e0db2ecaa9d01f68695ac5a5eb"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632609506.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.615000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0d4f13cf7850a30f408972382dc125c8"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632615814.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.617000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c4eab68f66873e3066774e29241e7d43"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632617182.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.619000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4ecfab8556cb30d89cae57790c993723"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632619581.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.621000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4b3f1259abbfef02f40a72ba73f5b713"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632620942.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.623000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0964dcb24587606b68f4d3f689ba87cb"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632623250.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.624000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "34ee847fb92cf6803d53e7a762b1a392"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632624608.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.629000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d6af479a65a401652f6d06402d7fe044"}
	{
	"name": "reload_kernel_in_parent",
	"ts": 1757522632629922.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.637000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6acb0d1f1d39e5655356928bf422f8e5"}
	{
	"name": "reload_kernel_in_parent",
	"ts": 1757522632637593.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.656000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "14eacf98f65f2acf564f6e671a818cbf"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632656144.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.657000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a6c63931cb09efcb0dfa47a86b9f8d63"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632657618.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.662000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5dbff2c5c15dfaa4166f6b3c6c34dc87"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632662221.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.663000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "86314f6a80e346e6ac11d71c61347a80"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632663572.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.665000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "83741077cbf1e62d5c6cb2cec6a33b77"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632665670.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.667000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c81a7b10f9198431c547de9cdc057cb9"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632667386.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.669000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "107bad3fd3e279ac02253094abd6684d"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632669511.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.670000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1605e9d41646d5411fc28740a4ad8d44"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632670847.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.672000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1e9c8ca69f437caad932c1e1b6dcbbd1"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632672938.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.674000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4949c401bc6cf4920a6eb5f605741234"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632674273.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.676000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6d64a87a898bdd4b8ad94a78f25e0e88"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632676269.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.677000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5be826be11491ae823325a6324e48ab8"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632677600.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.679000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7ba2758f0550fa334fe14ea2b7f6e275"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632679887.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.681000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "aea1f91198fe88ccd634d2e05316f3a6"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632681231.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.686000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "91657b953393547ef01d9da40d77dd26"}
	{
	"name": "reload_kernel_in_parent",
	"ts": 1757522632686351.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.694000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d486724e31838e07367a672b28b43f1a"}
	{
	"name": "reload_kernel_in_parent",
	"ts": 1757522632694049.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.707000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "de421cb39bd323b22adda36caf8366bf"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632707680.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.709000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f8bd0b434fae81eaeb57e4aa8d111fda"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632709134.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.713000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ee945e176b9bb794a9860e11b9bf0b76"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632713588.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.715000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "04462a6d8c014c1bd075007022228c4d"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632714955.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.717000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "64497aa1fc17bfe2a44f60a2c8404ac9"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632716984.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.718000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ed27fc8080af0cfa184dc2cd4aa61a04"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632718314.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.720000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3aab6c9f21999ac6f2566aa29e55c37e"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632720552.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.721000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "53d59ef3d46a9c41eee457cf115d2a19"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632721925.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.724000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "65eebce8a8d64c8593d7f2a3d6b1b465"}
	{
	"name": "reload_kernel_in_parent",
	"ts": 1757522632724059.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.730000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "af3b95d1b4d317a66232c7972f606e96"}
	{
	"name": "reload_kernel_in_parent",
	"ts": 1757522632730640.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.739000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0afd5c09063c537f6771e3fe657dfba5"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632738985.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.740000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d654b6c4e25a8f06b493c8fc212601d6"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632740355.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.742000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b9b1f79c4fb68fdcec336f6a97d22818"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632742785.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.744000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b68a0ac20b431e4bb96d081369f06fe4"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632744131.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.746000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "16e936515eb7c1b49cf92deba3b737e0"}
	{
	"name": "reload_kernel_in_parent",
	"ts": 1757522632746133.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.752000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4b8b3acde4b11bdbececaa19501a99eb"}
	{
	"name": "reload_kernel_in_parent",
	"ts": 1757522632752430.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.760000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a4631a3613f094812012327d6259de94"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632760567.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.762000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3f3637ec87c049de895bd4f5379630d3"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632762008.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.764000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "44eed4516cb29584eb875b9b6219dd26"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632764556.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.765000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "055facf8fa0e7d10d05beb053915c5ae"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632765892.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.768000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "04ca4198f102b49d243b3ddecaae8dda"}
	{
	"name": "reload_kernel_in_parent",
	"ts": 1757522632768086.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.773000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e1e64c8851fcd075a738df608a10a295"}
	{
	"name": "reload_kernel_in_parent",
	"ts": 1757522632773062.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.779000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5b2a9dc94e6e07129ea31b9762269e1f"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632779246.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.780000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6aeb7d05a582f5b68cfc96e5a66b1cc0"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632780565.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.783000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "782fa4fcaef8d6fbff90203dcc5b2919"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632782943.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.784000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "69c320a3dec57ff6f3761789dbd07171"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632784323.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.786000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8c821b723583365f3574481329e4a118"}
	{
	"name": "reload_kernel_in_parent",
	"ts": 1757522632786467.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.791000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e55407d88930dd91affe5b2e34076a01"}
	{
	"name": "reload_kernel_in_parent",
	"ts": 1757522632791318.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.797000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c4d926041e71240be6245ef09266e6a1"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632797517.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:52.798000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c0c17836a1fa03090a35205e2c369cce"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522632798856.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.348000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "de4378395320ffaa9f809528b9fb0b46"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522633348410.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.350000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d5f0be96ab285b3bba22df355dd1d41d"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522633350353.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.352000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d72fc265c6bc71503515ec2801d53cd1"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522633352722.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.354000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b8f579461027979cb860b78b20c1287f"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522633354011.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.433000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b0724804e58b483eb09ca8249fae248c"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522633433090.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.435000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ae1fca7b328345ef5d419664c546fe9e"}
	{
	"name": "CachingAutotuner.synchronize",
	"ts": 1757522633435127.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.437000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "701c852760fed36d2034c0ab3da34d77"}
	{
	"name": "async_compile.wait",
	"ts": 1757522633437462.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.442000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_inductor/codegen/cpp_wrapper_gpu.py", 40]}
V0910 09:43:53.443000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_inductor/codegen/cpp_wrapper_cpu.py", 41]}
V0910 09:43:53.443000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_inductor/codegen/wrapper.py", 42]}
V0910 09:43:53.444000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["<string>", 43]}
V0910 09:43:53.445000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_inductor/async_compile.py", 44]}
V0910 09:43:53.446000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_inductor/async_compile.py:121] {"artifact": {"name": "triton_kernel_info", "encoding": "json"}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2657, "name": "compile_fx", "filename": 19, "loc": "return inference_compiler(unlifted_gm, example_inputs_)"}, {"line": 1267, "name": "__call__", "filename": 33, "loc": "return self.compiler_fn(gm, example_inputs)"}, {"line": 2543, "name": "fw_compiler_base", "filename": 19, "loc": "return compile_fx_forward("}, {"line": 2260, "name": "compile_fx_forward", "filename": 19, "loc": "return inner_compile("}, {"line": 81, "name": "inner", "filename": 34, "loc": "return func(*args, **kwds)"}, {"line": 781, "name": "compile_fx_inner", "filename": 19, "loc": "return wrap_compiler_debug(_compile_fx_inner, compiler_name=\"inductor\")("}, {"line": 144, "name": "debug_wrapper", "filename": 35, "loc": "inner_compiled_fn = compiler_fn(gm, example_inputs)"}, {"line": 167, "name": "newFunction", "filename": 36, "loc": "return old_func(*args, **kwargs)"}, {"line": 962, "name": "_compile_fx_inner", "filename": 19, "loc": "mb_compiled_graph = fx_codegen_and_compile("}, {"line": 1694, "name": "fx_codegen_and_compile", "filename": 19, "loc": "return scheme.codegen_and_compile(gm, example_inputs, inputs_to_check, graph_kwargs)"}, {"line": 1463, "name": "codegen_and_compile", "filename": 19, "loc": "wrapper_code, kernel_code = graph.codegen_with_cpp_wrapper()"}, {"line": 2220, "name": "codegen_with_cpp_wrapper", "filename": 37, "loc": "return self.codegen()"}, {"line": 2271, "name": "codegen", "filename": 37, "loc": "result = self.wrapper_code.generate(self.is_inference)"}, {"line": 355, "name": "generate", "filename": 40, "loc": "return super().generate(is_inference)"}, {"line": 1005, "name": "generate", "filename": 41, "loc": "return super().generate(is_inference)"}, {"line": 1551, "name": "generate", "filename": 42, "loc": "return self._generate(is_inference)"}, {"line": 1614, "name": "_generate", "filename": 42, "loc": "self.generate_and_run_autotune_block()"}, {"line": 1692, "name": "generate_and_run_autotune_block", "filename": 42, "loc": "exec(tuning_code, scope)"}, {"line": 1371, "name": "<module>", "filename": 43, "loc": ""}, {"line": 631, "name": "wait", "filename": 44, "loc": "_compile_end()"}, {"line": 121, "name": "_compile_end", "filename": 44, "loc": "torch._logging.trace_structured("}], "has_payload": "a27d508924b6be7c0178d9db569fc53a"}
	{"triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2": {"autotune_cache_state": "only 1 config", "only_config": [["XBLOCK", 128], ["num_warps", 8], ["num_stages", 1]], "compile_time_us": 63806}, "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5": {"best_config": [["XBLOCK", 1024], ["num_warps", 4], ["num_stages", 1]], "autotune_cache_state": "hit", "compile_time_us": 1183541}, "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6": {"best_config": [["XBLOCK", 1024], ["num_warps", 4], ["num_stages", 1]], "autotune_cache_state": "hit", "compile_time_us": 1171123}, "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7": {"best_config": [["XBLOCK", 1024], ["num_warps", 4], ["num_stages", 1]], "autotune_cache_state": "hit", "compile_time_us": 1308640}, "triton_poi_fused__to_copy_mul_sub_view_10": {"best_config": [["XBLOCK", 1024], ["num_warps", 4], ["num_stages", 1]], "autotune_cache_state": "hit", "compile_time_us": 1185446}, "triton_poi_fused__to_copy_mul_sub_view_16": {"best_config": [["XBLOCK", 512], ["num_warps", 8], ["num_stages", 1]], "autotune_cache_state": "hit", "compile_time_us": 1184074}, "triton_poi_fused__to_copy_mul_sub_view_18": {"best_config": [["XBLOCK", 512], ["num_warps", 8], ["num_stages", 1]], "autotune_cache_state": "hit", "compile_time_us": 1186781}, "triton_poi_fused__to_copy_mul_sub_view_4": {"best_config": [["XBLOCK", 1024], ["num_warps", 4], ["num_stages", 1]], "autotune_cache_state": "hit", "compile_time_us": 1196146}, "triton_poi_fused_convolution_gelu_0": {"best_config": [["XBLOCK", 512], ["num_warps", 8], ["num_stages", 1]], "autotune_cache_state": "hit", "compile_time_us": 1254263}, "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9": {"autotune_cache_state": "miss", "num_configs": 5, "compile_time_us": 1230700}, "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11": {"autotune_cache_state": "only 1 config", "only_config": [["R0_BLOCK", 2048], ["XBLOCK", 1], ["num_warps", 16], ["num_stages", 1]], "compile_time_us": 70518}, "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12": {"autotune_cache_state": "only 1 config", "only_config": [["R0_BLOCK", 1024], ["XBLOCK", 1], ["num_warps", 8], ["num_stages", 1]], "compile_time_us": 75710}, "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13": {"autotune_cache_state": "only 1 config", "only_config": [["R0_BLOCK", 1024], ["XBLOCK", 1], ["num_warps", 8], ["num_stages", 1]], "compile_time_us": 75641}, "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15": {"autotune_cache_state": "only 1 config", "only_config": [["R0_BLOCK", 2048], ["XBLOCK", 1], ["num_warps", 16], ["num_stages", 1]], "compile_time_us": 66948}, "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3": {"autotune_cache_state": "miss", "num_configs": 5, "compile_time_us": 1227301}, "triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17": {"autotune_cache_state": "only 1 config", "only_config": [["R0_BLOCK", 2048], ["XBLOCK", 1], ["num_warps", 16], ["num_stages", 1]], "compile_time_us": 67450}, "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8": {"autotune_cache_state": "only 1 config", "only_config": [["R0_BLOCK", 2048], ["XBLOCK", 1], ["num_warps", 16], ["num_stages", 1]], "compile_time_us": 67031}, "triton_red_fused_add_addmm_native_layer_norm_view_14": {"autotune_cache_state": "only 1 config", "only_config": [["R0_BLOCK", 1024], ["XBLOCK", 1], ["num_warps", 8], ["num_stages", 1]], "compile_time_us": 94186}, "triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1": {"autotune_cache_state": "only 1 config", "only_config": [["R0_BLOCK", 8], ["XBLOCK", 64], ["num_warps", 4], ["num_stages", 1]], "compile_time_us": 68299}}
V0910 09:43:53.455000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6466e503f9028354970094d443f0550e"}
	{
	"name": "CachingAutotuner.benchmark_all_configs",
	"ts": 1757522633455029.2,
	"args": {
	"kernel_name": "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3",
	"is_backward": false,
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.510000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "96dc46c2086346187d1db42cf8895367"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522633509968.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.613000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "bd2d003d6bf7941ff0e49322e5d83837"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522633613093.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.615000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4303c770e5ab753d62cc2fc12f8c619b"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522633615490.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.646000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "cbf0f419e8a89fd8f9a3f3da3863c2a8"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522633645915.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.648000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7bbcdb764f6cfca7f75617a004d38a4a"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522633648203.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.709000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f6835c6883d45bafec856d4f09f557fb"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522633708991.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.711000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6a2daea12a313dee0823f01e488688a8"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522633711393.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.771000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "cc353a66339c6d01f4c7153bf5eafdc0"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522633771565.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.773000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "219f9df0230344c35bb9825eeba252b8"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522633773762.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.840000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d14ab0cfdde12dba7e6c0ef4876cab50"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522633839911.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.841000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0cbbe70df7c95c25308c4d75c6192acb"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522633841857.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.901000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6f34d718e1179ef8a1badedf65940db5"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522633901405.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.903000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "25d23f193cdfe439516418c48643c95b"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522633903045.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.962000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "17dd8aa2c3a353304f9bb5c288687b0a"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522633962672.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.964000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "5c3f3a043c2493881ae08e2d5ac9b9ca"}
	{
	"name": "CachingAutotuner.benchmark_all_configs",
	"ts": 1757522633964023.0,
	"args": {
	"kernel_name": "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3",
	"is_backward": false,
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.984000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "495e40cf62e907051632e247b44f3655"}
	{
	"name": "CachingAutotuner.benchmark_all_configs",
	"ts": 1757522633984491.0,
	"args": {
	"kernel_name": "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9",
	"is_backward": false,
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:53.986000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c788b646c5ef70f27af053a225c1207f"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522633986215.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.045000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d727c81c721e38c0145f2af0e56a6c28"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634045469.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.047000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c00168b81fe7ac919786a39d3e084f47"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634047343.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.112000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0d670c5a0711457296433bc06f933e2d"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634112486.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.114000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d05c5069bddc79b155c8cbc20c310c25"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634114662.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.174000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "e38752ad3c7a0bbd78332578d733fcd9"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634174426.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.176000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "11a48c9258fb6f2e96946ae4ec6c118a"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634176266.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.237000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "ea1e0a21f69ac836422f0f8bdec99101"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634237457.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.239000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2a7ac1ad3469da7aae3ca0476e48f6b7"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634239270.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.305000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "c267501ad20ebcc40f60b581fa58290f"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634305640.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.307000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6116daa7eeb11d4a95baa0c5be72b5af"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634307549.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.366000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1923f719f7e30b2507b2b8ab651bd36e"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634366768.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.369000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "7ccde22e1d950e96920aea2ab819bb13"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634368951.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.427000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "de009807ba3e52b5d87384d80ab9d0ef"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634427390.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.429000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "cf6a33ece62c5421a74fb47082770689"}
	{
	"name": "CachingAutotuner.benchmark_all_configs",
	"ts": 1757522634429265.8,
	"args": {
	"kernel_name": "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9",
	"is_backward": false,
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.444000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d3f440959a046fc467ffd06728e4a2e1"}
	{
	"name": "CachingAutotuner.benchmark_all_configs",
	"ts": 1757522634444465.2,
	"args": {
	"kernel_name": "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12",
	"is_backward": false,
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.446000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d9c503a3fe39c1e0b7348e268d38eca0"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634446365.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.492000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "492b43a7dc7c654c9131712d77af4412"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634492555.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.494000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2c76f5cbc74a2188dd1b6fcc54e0823b"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634494412.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.555000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "653641c0ca0aa59d8ca0fa94aa6f9534"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634555597.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.557000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "766773edf630f783477328c407ef2261"}
	{
	"name": "CachingAutotuner.benchmark_all_configs",
	"ts": 1757522634557021.8,
	"args": {
	"kernel_name": "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12",
	"is_backward": false,
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.562000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1d76acb886e51397c03307ff392fd7f7"}
	{
	"name": "CachingAutotuner.benchmark_all_configs",
	"ts": 1757522634562866.0,
	"args": {
	"kernel_name": "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13",
	"is_backward": false,
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.564000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6c7d35c4872dcec9c1cc1e9c6479ae28"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634564597.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.623000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f3aaee6766ee0999e6ad08c4c51cd2f5"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634623146.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.625000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "41a57363b9a44e2d5c2dd3f0fbed010a"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634624971.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.683000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2eaed128c4a2bd1b5bb2c789649dc9fb"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634682912.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.684000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "aa0d1a59e19b36046a4c89924b00d89d"}
	{
	"name": "CachingAutotuner.benchmark_all_configs",
	"ts": 1757522634684448.2,
	"args": {
	"kernel_name": "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13",
	"is_backward": false,
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.690000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a9f292d49ba6a05c14e3ddbba58f6295"}
	{
	"name": "CachingAutotuner.benchmark_all_configs",
	"ts": 1757522634690228.5,
	"args": {
	"kernel_name": "triton_red_fused_add_addmm_native_layer_norm_view_14",
	"is_backward": false,
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.691000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3ea45eaecb1a5966ebc45b214c5395c1"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634691897.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.746000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "79e8a873beceb083ea8e0c4afe0d545c"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634746698.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.748000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3ff78aeb02ba114d9da8ca37738f6389"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634748592.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.805000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "b97af94e6c337b3692c25a53a86e9a2f"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634805253.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.806000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1e090f8d6ee10af4ee380bfdc17667bf"}
	{
	"name": "CachingAutotuner.benchmark_all_configs",
	"ts": 1757522634806774.0,
	"args": {
	"kernel_name": "triton_red_fused_add_addmm_native_layer_norm_view_14",
	"is_backward": false,
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.811000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9d9be8190df31fa7467c6540f1e97bad"}
	{
	"name": "CachingAutotuner.benchmark_all_configs",
	"ts": 1757522634811586.8,
	"args": {
	"kernel_name": "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15",
	"is_backward": false,
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.813000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "203f33f5880eea0042961978a390eecb"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634813490.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.867000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9d24728003c307048511a8ee2921b440"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634867687.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.869000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a3862230437ac9bcbc1696057d8df7dc"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634869511.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.923000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "96fa24530ffc77fe563fb08c8a9cdebd"}
	{
	"name": "TritonBenchmarker.benchmark_gpu",
	"ts": 1757522634923621.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:54.925000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6556ea7b558f4c39e68d05bac5fa14e9"}
	{
	"name": "CachingAutotuner.benchmark_all_configs",
	"ts": 1757522634924940.2,
	"args": {
	"kernel_name": "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15",
	"is_backward": false,
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:55.961000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "16580813693ef5c283da19d2b0a7f883"}
	{
	"name": "PythonWrapperCodegen.generate",
	"ts": 1757522635960931.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:55.962000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "61f8c24364589f5f4f5f6d3bc9d8b717"}
	{
	"name": "CppWrapperCpu.generate",
	"ts": 1757522635962783.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:55.966000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "642652c3cdf0225190e5833ac2050251"}
	{
	"name": "CppWrapperGpu.generate",
	"ts": 1757522635966606.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:55.970000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "a2463f7f687b02e8e54ce4a4318e8432"}
	{
	"name": "GraphLowering.codegen",
	"ts": 1757522635970069.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:55.973000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "761a298a1f7c939f0518a714690d08a7"}
	{
	"name": "AotCodeCompiler.compile",
	"ts": 1757522635973645.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:43:56.003000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_inductor/codecache.py", 45]}
V0910 09:43:56.007000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_inductor/codecache.py:1813] {"graph_dump": {"name": "inductor_aot_wrapper_code", "type": "cpp", "filename": "/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/c352umhij7bqg7p3tqfyegchuzaz5aagiv7cyzfgjxn3nlod5dvn.wrapper.cpp"}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2657, "name": "compile_fx", "filename": 19, "loc": "return inference_compiler(unlifted_gm, example_inputs_)"}, {"line": 1267, "name": "__call__", "filename": 33, "loc": "return self.compiler_fn(gm, example_inputs)"}, {"line": 2543, "name": "fw_compiler_base", "filename": 19, "loc": "return compile_fx_forward("}, {"line": 2260, "name": "compile_fx_forward", "filename": 19, "loc": "return inner_compile("}, {"line": 81, "name": "inner", "filename": 34, "loc": "return func(*args, **kwds)"}, {"line": 781, "name": "compile_fx_inner", "filename": 19, "loc": "return wrap_compiler_debug(_compile_fx_inner, compiler_name=\"inductor\")("}, {"line": 144, "name": "debug_wrapper", "filename": 35, "loc": "inner_compiled_fn = compiler_fn(gm, example_inputs)"}, {"line": 167, "name": "newFunction", "filename": 36, "loc": "return old_func(*args, **kwargs)"}, {"line": 962, "name": "_compile_fx_inner", "filename": 19, "loc": "mb_compiled_graph = fx_codegen_and_compile("}, {"line": 1694, "name": "fx_codegen_and_compile", "filename": 19, "loc": "return scheme.codegen_and_compile(gm, example_inputs, inputs_to_check, graph_kwargs)"}, {"line": 1486, "name": "codegen_and_compile", "filename": 19, "loc": "compiled_fn = AotCodeCompiler.compile("}, {"line": 1813, "name": "compile", "filename": 45, "loc": "trace_structured("}], "has_payload": "1863203d08b4dfd93b2686176021e7ee"}
	
	#include <torch/csrc/inductor/aoti_include/cuda.h>
	// Definition of AOTI runtime interface functions
	
	#include <torch/csrc/inductor/aoti_runtime/interface.h>
	#include <torch/csrc/inductor/aoti_runtime/model_container.h>
	
	#include <iostream>
	#include <vector>
	
	#define CONVERT_EXCEPTION_TO_ERROR_CODE(...)      \
	  try {                                           \
	    __VA_ARGS__                                   \
	  } catch (const std::exception& e) {             \
	    std::cerr << "Error: " << e.what() << '\n';   \
	    return AOTI_RUNTIME_FAILURE;                  \
	  } catch (...) {                                 \
	    std::cerr << "Unknown exception occurred.\n"; \
	    return AOTI_RUNTIME_FAILURE;                  \
	  }                                               \
	  return AOTI_RUNTIME_SUCCESS;
	
	#define AOTI_VECTOR_SIZE_CHECK(actual_size, expected_size, name)  \
	  do {                                                            \
	    AOTI_RUNTIME_CHECK(                                           \
	        actual_size == expected_size,                             \
	        "expected " + std::string(name) + " vector size to be " + \
	            std::to_string(expected_size) + ", but got " +        \
	            std::to_string(actual_size));                         \
	  } while (0)
	
	// AOTInductor uses at::addmm_out, which doesn't supports
	// arguments that requires gradient. For this reason, we
	// enforce no_grad context for run APIs.
	//
	// A RAII, thread local (!) guard that enables or disables grad mode upon
	// construction, and sets it back to the original value upon destruction.
	struct AOTINoGradGuard {
	  AOTINoGradGuard() {
	    aoti_torch_grad_mode_set_enabled(false);
	  }
	  AOTINoGradGuard(const AOTINoGradGuard&) = delete;
	  AOTINoGradGuard(AOTINoGradGuard&&) noexcept = delete;
	  ~AOTINoGradGuard() {
	    aoti_torch_grad_mode_set_enabled(prev_mode);
	  }
	  AOTINoGradGuard& operator=(const AOTINoGradGuard&) = delete;
	  AOTINoGradGuard& operator=(AOTINoGradGuard&&) noexcept = delete;
	  bool prev_mode{aoti_torch_grad_mode_is_enabled()};
	};
	
	extern "C" {
	
	AOTIRuntimeError AOTInductorModelContainerCreate(
	    AOTInductorModelContainerHandle* container_handle,
	    size_t num_models,
	    bool is_cpu,
	    const char* cubin_dir) {
	      return AOTInductorModelContainerCreateWithDevice(
	        container_handle,
	        num_models,
	        is_cpu ? "cpu" : "cuda",
	        cubin_dir);
	}
	
	AOTIRuntimeError AOTInductorModelContainerCreateWithDevice(
	    AOTInductorModelContainerHandle* container_handle,
	    size_t num_models,
	    const char* device_str,
	    const char* cubin_dir) {
	  if (num_models == 0) {
	    std::cerr << "Error: num_models must be positive, but got 0\n";
	    return AOTI_RUNTIME_FAILURE;
	  }
	  CONVERT_EXCEPTION_TO_ERROR_CODE({
	    std::optional<std::string> cubin_dir_opt;
	    if (cubin_dir != nullptr) {
	      cubin_dir_opt.emplace(cubin_dir);
	    }
	    auto* container = new torch::aot_inductor::AOTInductorModelContainer(
	        num_models, std::string(device_str), cubin_dir_opt);
	    *container_handle =
	        reinterpret_cast<AOTInductorModelContainerHandle>(container);
	  })
	}
	
	AOTIRuntimeError AOTInductorModelContainerDelete(
	    AOTInductorModelContainerHandle container_handle) {
	  CONVERT_EXCEPTION_TO_ERROR_CODE({
	    auto* container =
	        reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(
	            container_handle);
	    delete container;
	  });
	}
	
	AOTIRuntimeError AOTInductorModelContainerRun(
	    AOTInductorModelContainerHandle container_handle,
	    AtenTensorHandle* input_handles, // array of input AtenTensorHandle; handles
	                                     // are stolen; the array itself is borrowed
	    size_t num_inputs,
	    AtenTensorHandle*
	        output_handles, // array for writing output AtenTensorHandle; handles
	                        // will be stolen by the caller; the array itself is
	                        // borrowed
	    size_t num_outputs,
	    AOTInductorStreamHandle stream_handle,
	    AOTIProxyExecutorHandle proxy_executor_handle) {
	  auto* container =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(
	          container_handle);
	  AOTI_VECTOR_SIZE_CHECK(num_inputs, container->num_inputs(), "inputs");
	  AOTI_VECTOR_SIZE_CHECK(num_outputs, container->num_outputs(), "outputs");
	
	  auto stream =
	      reinterpret_cast<torch::aot_inductor::DeviceStreamType>(stream_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE({
	    AOTINoGradGuard guard;
	    container->run(
	        input_handles, output_handles, stream, proxy_executor_handle);
	  })
	}
	
	AOTIRuntimeError AOTInductorModelContainerRunSingleThreaded(
	    AOTInductorModelContainerHandle container_handle,
	    AtenTensorHandle* input_handles, // array of input AtenTensorHandle; handles
	                                     // are stolen; the array itself is borrowed
	    size_t num_inputs,
	    AtenTensorHandle*
	        output_handles, // array for writing output AtenTensorHandle; handles
	                        // will be stolen by the caller; the array itself is
	                        // borrowed
	    size_t num_outputs,
	    AOTInductorStreamHandle stream_handle,
	    AOTIProxyExecutorHandle proxy_executor_handle) {
	  auto* container =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(
	          container_handle);
	  AOTI_VECTOR_SIZE_CHECK(num_inputs, container->num_inputs(), "inputs");
	  AOTI_VECTOR_SIZE_CHECK(num_outputs, container->num_outputs(), "outputs");
	
	  auto stream =
	      reinterpret_cast<torch::aot_inductor::DeviceStreamType>(stream_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE({
	    AOTINoGradGuard guard;
	    container->run_single_threaded(
	        input_handles, output_handles, stream, proxy_executor_handle);
	  })
	}
	
	AOTIRuntimeError AOTInductorModelContainerGetNumConstants(
	    AOTInductorModelContainerHandle container_handle,
	    size_t* num_constants) {
	  auto* container =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(
	          container_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE(
	    { *num_constants = container->num_constants(); })
	}
	
	AOTIRuntimeError AOTInductorModelContainerGetConstantName(
	    AOTInductorModelContainerHandle container_handle,
	    size_t idx,
	    const char** name) {
	  auto* container =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(
	          container_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE(
	    { *name = container->constant_name(idx); })
	}
	
	AOTIRuntimeError AOTInductorModelContainerGetConstantOriginalFQN(
	    AOTInductorModelContainerHandle container_handle,
	    size_t idx,
	    const char** original_fqn) {
	  auto* container =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(
	          container_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE(
	    { *original_fqn = container->constant_original_fqn(idx); })
	}
	
	AOTIRuntimeError AOTInductorModelContainerGetConstantFromFolded(
	    AOTInductorModelContainerHandle container_handle,
	    size_t idx,
	    bool* from_folded) {
	  auto* container =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(container_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE({ *from_folded = container->constant_from_folded(idx); })
	}
	
	AOTIRuntimeError AOTInductorModelContainerGetConstantType(
	    AOTInductorModelContainerHandle container_handle,
	    size_t idx,
	    int32_t* type) {
	  auto* container =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(container_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE({ *type = container->constant_type(idx); })
	}
	
	AOTIRuntimeError AOTInductorModelContainerGetConstantDtype(
	    AOTInductorModelContainerHandle container_handle,
	    size_t idx,
	    int32_t* dtype) {
	  auto* container =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(
	          container_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE(
	    { *dtype = container->constant_dtype(idx); })
	}
	
	AOTIRuntimeError AOTInductorModelContainerGetConstantDataSize(
	  AOTInductorModelContainerHandle container_handle,
	  size_t idx,
	  size_t* data_size) {
	  auto* container =
	    reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(
	        container_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE(
	    { *data_size = container->constant_data_size(idx); })
	}
	
	AOTIRuntimeError AOTInductorModelContainerExtractConstantsMap(
	    AOTInductorModelContainerHandle container_handle,
	    AOTInductorConstantMapHandle constant_map_handle,
	    bool use_inactive) {
	  auto* container =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(
	          container_handle);
	  auto constants_map = reinterpret_cast<std::unordered_map<std::string, AtenTensorHandle>*>(constant_map_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE(
	    { const auto ret = container->extract_constants_map(use_inactive);
	      for (const auto& pair: ret) {
	        constants_map->emplace(pair.first, pair.second);
	      }
	    })
	}
	
	AOTIRuntimeError AOTInductorModelContainerUpdateUserManagedConstantBuffer(
	    AOTInductorModelContainerHandle container_handle,
	    AOTInductorConstantMapHandle constant_map_handle,
	    bool use_inactive,
	    bool validate_full_update) {
	  auto* container =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(
	          container_handle);
	  auto input_map = reinterpret_cast<std::unordered_map<std::string, AtenTensorHandle>*>(constant_map_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE({
	    container->update_constant_buffer(
	        *input_map, use_inactive, validate_full_update, /* user_managed = */ true);
	  })
	}
	
	AOTIRuntimeError AOTInductorModelContainerUpdateConstantBuffer(
	    AOTInductorModelContainerHandle container_handle,
	    AOTInductorConstantMapHandle constant_map_handle,
	    bool use_inactive,
	    bool validate_full_update) {
	  auto* container =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(
	          container_handle);
	  auto input_map = reinterpret_cast<std::unordered_map<std::string, AtenTensorHandle>*>(constant_map_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE({
	    container->update_constant_buffer(
	        *input_map, use_inactive, validate_full_update);
	  })
	}
	
	AOTIRuntimeError AOTInductorModelContainerUpdateInactiveConstantBuffer(
	    AOTInductorModelContainerHandle container_handle,
	    AOTInductorConstantMapHandle constant_map_handle) {
	  return AOTInductorModelContainerUpdateConstantBuffer(container_handle,
	          constant_map_handle,
	          /*use_inactive*/ true,
	          /*validate_full_update*/ true);
	}
	
	AOTIRuntimeError AOTInductorModelContainerFreeInactiveConstantBuffer(
	    AOTInductorModelContainerHandle container_handle) {
	  auto* container =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(
	          container_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE({
	    container->free_inactive_constant_buffer();
	  })
	}
	
	AOTIRuntimeError AOTInductorModelContainerRunConstantFolding(
	    AOTInductorModelContainerHandle container_handle,
	    bool use_inactive,
	    AOTInductorStreamHandle stream_handle,
	    AOTIProxyExecutorHandle proxy_executor_handle) {
	  auto* container =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(
	          container_handle);
	  auto stream =
	      reinterpret_cast<torch::aot_inductor::DeviceStreamType>(stream_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE({
	    AOTINoGradGuard guard;
	    container->run_const_fold(use_inactive, stream, proxy_executor_handle);
	  })
	}
	
	AOTIRuntimeError AOTInductorModelContainerSwapConstantBuffer(
	    AOTInductorModelContainerHandle container_handle) {
	  auto* container =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(
	          container_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE({
	    container->swap_constant_buffer();
	  })
	}
	
	AOTIRuntimeError AOTInductorModelContainerGetNumInputs(
	    AOTInductorModelContainerHandle container_handle,
	    size_t* ret_num_inputs) {
	  auto* container =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(
	          container_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE(
	      { *ret_num_inputs = container->num_inputs(); })
	}
	
	AOTIRuntimeError AOTInductorModelContainerGetInputName(
	    AOTInductorModelContainerHandle container_handle,
	    size_t input_idx,
	    const char** ret_input_names) {
	  auto* container =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(
	          container_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE(
	      { *ret_input_names = container->input_name(input_idx); })
	}
	
	AOTIRuntimeError AOTInductorModelContainerGetNumOutputs(
	    AOTInductorModelContainerHandle container_handle,
	    size_t* ret_num_outputs) {
	  auto* container =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(
	          container_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE(
	      { *ret_num_outputs = container->num_outputs(); })
	}
	
	AOTIRuntimeError AOTInductorModelContainerGetOutputName(
	    AOTInductorModelContainerHandle container_handle,
	    size_t output_idx,
	    const char** ret_output_names) {
	  auto* container =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(
	          container_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE(
	      { *ret_output_names = container->output_name(output_idx); })
	}
	
	AOTIRuntimeError AOTInductorModelContainerGetCallSpec(
	    AOTInductorModelContainerHandle container_handle,
	    const char** in_spec,
	    const char** out_spec) {
	  auto* container =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModelContainer*>(
	          container_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE({
	    *in_spec = container->get_in_spec();
	    *out_spec = container->get_out_spec();
	  })
	}
	
	AOTIRuntimeError AOTInductorModelCreate(
	    AOTInductorModelHandle* model_handle,
	    AOTInductorConstantMapHandle constant_map_handle){
	    CONVERT_EXCEPTION_TO_ERROR_CODE({
	      auto constant_map = std::make_shared<torch::aot_inductor::ConstantMap>();
	      auto constant_array = std::make_shared<std::vector<torch::aot_inductor::ConstantHandle>>();
	      auto input_map = reinterpret_cast<std::unordered_map<std::string, AtenTensorHandle>*>(constant_map_handle);
	
	      auto model = new torch::aot_inductor::AOTInductorModel(
	          constant_map,
	          constant_array,
	          "cpu", // device_str is hardcoded, as AOTInductorModelCreate is only use for CPU models
	          ""
	      );
	
	      if (input_map) {
	        for (auto const& kv : *input_map) {
	          constant_map->emplace(kv.first, kv.second);
	        }
	      } else {
	        model->load_constants();
	      }
	
	      *model_handle = reinterpret_cast<AOTInductorModelHandle>(model);
	    })}
	
	AOTIRuntimeError AOTInductorModelRun(
	    AOTInductorModelHandle model_handle,
	    AtenTensorHandle* input_handles,
	    AtenTensorHandle* output_handles) {
	  auto model =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModel*>(model_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE({
	    AOTINoGradGuard guard;
	    model->run_impl(
	        input_handles,
	        output_handles,
	        (torch::aot_inductor::DeviceStreamType) nullptr,
	        nullptr);
	  })
	}
	
	AOTIRuntimeError AOTInductorModelDelete(AOTInductorModelHandle model_handle){
	    CONVERT_EXCEPTION_TO_ERROR_CODE({
	      auto model = reinterpret_cast<torch::aot_inductor::AOTInductorModel*>(
	          model_handle);
	      delete model;
	    })}
	
	AOTIRuntimeError AOTInductorModelGetNumOutputs(
	    AOTInductorModelHandle model_handle,
	    size_t* ret_num_outputs) {
	  CONVERT_EXCEPTION_TO_ERROR_CODE({
	      auto model = reinterpret_cast<torch::aot_inductor::AOTInductorModel*>(model_handle);
	      *ret_num_outputs = model->num_outputs();
	  })
	}
	
	AOTIRuntimeError AOTInductorModelUpdateConstantsMap(
	    AOTInductorModelHandle model_handle,
	    AOTInductorConstantMapHandle constant_map_handle) {
	  auto model =
	      reinterpret_cast<torch::aot_inductor::AOTInductorModel*>(model_handle);
	  CONVERT_EXCEPTION_TO_ERROR_CODE({
	    auto constant_map = std::make_shared<torch::aot_inductor::ConstantMap>();
	    auto input_map =
	        reinterpret_cast<std::unordered_map<std::string, AtenTensorHandle>*>(
	            constant_map_handle);
	
	    for (auto const& kv : *input_map) {
	      constant_map->emplace(kv.first, kv.second);
	    }
	    model->update_constants_map(std::move(constant_map));
	  })
	}
	
	} // extern "C"
	
	
	#define CUDA_DRIVER_CHECK(EXPR)                    \
	do {                                               \
	    CUresult code = EXPR;                          \
	    const char *msg;                               \
	    CUresult code_get_error = cuGetErrorString(code, &msg); \
	    if (code_get_error != CUDA_SUCCESS) {          \
	        throw std::runtime_error(                  \
	            std::string("CUDA driver error: ") +   \
	            std::string("invalid error code!"));   \
	    }                                              \
	    if (code != CUDA_SUCCESS) {                    \
	        throw std::runtime_error(                  \
	            std::string("CUDA driver error: ") +   \
	            std::string(msg));                     \
	    }                                              \
	} while (0);
	
	static inline CUfunction loadKernel(
	        std::string filePath,
	        const std::string &funcName,
	        uint32_t sharedMemBytes,
	        const std::optional<std::string> &cubinDir = std::nullopt) {
	    if (cubinDir) {
	        std::filesystem::path p1{*cubinDir};
	        std::filesystem::path p2{filePath};
	        filePath = (p1 / p2.filename()).string();
	    }
	
	    CUmodule mod;
	    CUfunction func;
	    CUDA_DRIVER_CHECK(cuModuleLoad(&mod, filePath.c_str()));
	    CUDA_DRIVER_CHECK(cuModuleGetFunction(&func, mod, funcName.c_str()));
	    if (sharedMemBytes > 0) {
	        CUDA_DRIVER_CHECK(cuFuncSetAttribute(
	            func,
	            CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES,
	            sharedMemBytes
	        ))
	    }
	    return func;
	}
	
	static inline CUfunction loadKernel(const void* start, const std::string &funcName, uint32_t sharedMemBytes) {
	    CUmodule mod;
	    CUfunction func;
	    CUDA_DRIVER_CHECK(cuModuleLoadData(&mod, start));
	    CUDA_DRIVER_CHECK(cuModuleGetFunction(&func, mod, funcName.c_str()));
	    if (sharedMemBytes > 0) {
	        CUDA_DRIVER_CHECK(cuFuncSetAttribute(
	            func,
	            CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES,
	            sharedMemBytes
	        ))
	    }
	    return func;
	}
	
	static inline void launchKernel(
	        CUfunction func,
	        uint32_t gridX,
	        uint32_t gridY,
	        uint32_t gridZ,
	        uint32_t numWarps,
	        uint32_t sharedMemBytes,
	        void* args[],
	        cudaStream_t stream) {
	    CUDA_DRIVER_CHECK(cuLaunchKernel(
	        func, gridX, gridY, gridZ, 32*numWarps, 1, 1, sharedMemBytes, stream, args, nullptr
	    ));
	}
	CACHE_TORCH_DTYPE(float32);
	CACHE_TORCH_DTYPE(int8);
	CACHE_TORCH_DEVICE(cuda);
	CACHE_TORCH_LAYOUT(strided);
	namespace torch::aot_inductor {
	namespace {
	class AOTInductorModelKernels : public AOTInductorModelKernelsBase {
	  public:
	    CUfunction triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2{nullptr};
	    CUfunction triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5{nullptr};
	    CUfunction triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6{nullptr};
	    CUfunction triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7{nullptr};
	    CUfunction triton_poi_fused__to_copy_mul_sub_view_10{nullptr};
	    CUfunction triton_poi_fused__to_copy_mul_sub_view_16{nullptr};
	    CUfunction triton_poi_fused__to_copy_mul_sub_view_18{nullptr};
	    CUfunction triton_poi_fused__to_copy_mul_sub_view_4{nullptr};
	    CUfunction triton_poi_fused_convolution_gelu_0{nullptr};
	    CUfunction triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9{nullptr};
	    CUfunction triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11{nullptr};
	    CUfunction triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12{nullptr};
	    CUfunction triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13{nullptr};
	    CUfunction triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15{nullptr};
	    CUfunction triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3{nullptr};
	    CUfunction triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17{nullptr};
	    CUfunction triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8{nullptr};
	    CUfunction triton_red_fused_add_addmm_native_layer_norm_view_14{nullptr};
	    CUfunction triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1{nullptr};
	};
	}  // namespace
	
	
	
	AOTInductorModel::AOTInductorModel(std::shared_ptr<ConstantMap> constants_map,
	                                   std::shared_ptr<std::vector<ConstantHandle>> constants_array,
	                                   const std::string& device_str,
	                                   std::optional<std::string> cubin_dir)
	    : AOTInductorModelBase(1,
	                           1,
	                           877,
	                           device_str,
	                           std::move(cubin_dir),
	                           true) {
	    inputs_info_[0].name = "arg877_1";
	    constants_info_[0].name = "model_audio_tower_embed_positions_weight";
	    constants_info_[0].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[0].offset = 0;
	    constants_info_[0].data_size = 7680000;
	    constants_info_[0].from_folded = false;
	    constants_info_[0].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[0].shape = {1500, 1280};
	    constants_info_[0].stride = {1280, 1};
	    constants_info_[0].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[0].original_fqn = "model.audio_tower.embed_positions.weight";
	    constants_info_[1].name = "model_audio_tower_conv1_weight";
	    constants_info_[1].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[1].offset = 0;
	    constants_info_[1].data_size = 1966080;
	    constants_info_[1].from_folded = false;
	    constants_info_[1].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[1].shape = {1280, 128, 3};
	    constants_info_[1].stride = {384, 3, 1};
	    constants_info_[1].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[1].original_fqn = "model.audio_tower.conv1.weight";
	    constants_info_[2].name = "model_audio_tower_conv1_bias";
	    constants_info_[2].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[2].offset = 0;
	    constants_info_[2].data_size = 5120;
	    constants_info_[2].from_folded = false;
	    constants_info_[2].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[2].shape = {1280};
	    constants_info_[2].stride = {1};
	    constants_info_[2].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[2].original_fqn = "model.audio_tower.conv1.bias";
	    constants_info_[3].name = "model_audio_tower_conv2_weight";
	    constants_info_[3].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[3].offset = 0;
	    constants_info_[3].data_size = 19660800;
	    constants_info_[3].from_folded = false;
	    constants_info_[3].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[3].shape = {1280, 1280, 3};
	    constants_info_[3].stride = {3840, 3, 1};
	    constants_info_[3].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[3].original_fqn = "model.audio_tower.conv2.weight";
	    constants_info_[4].name = "model_audio_tower_conv2_bias";
	    constants_info_[4].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[4].offset = 0;
	    constants_info_[4].data_size = 5120;
	    constants_info_[4].from_folded = false;
	    constants_info_[4].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[4].shape = {1280};
	    constants_info_[4].stride = {1};
	    constants_info_[4].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[4].original_fqn = "model.audio_tower.conv2.bias";
	    constants_info_[5].name = "model_audio_tower_layers_0_self_attn_layer_norm_weight";
	    constants_info_[5].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[5].offset = 0;
	    constants_info_[5].data_size = 5120;
	    constants_info_[5].from_folded = false;
	    constants_info_[5].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[5].shape = {1280};
	    constants_info_[5].stride = {1};
	    constants_info_[5].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[5].original_fqn = "model.audio_tower.layers.0.self_attn_layer_norm.weight";
	    constants_info_[6].name = "model_audio_tower_layers_0_self_attn_layer_norm_bias";
	    constants_info_[6].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[6].offset = 0;
	    constants_info_[6].data_size = 5120;
	    constants_info_[6].from_folded = false;
	    constants_info_[6].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[6].shape = {1280};
	    constants_info_[6].stride = {1};
	    constants_info_[6].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[6].original_fqn = "model.audio_tower.layers.0.self_attn_layer_norm.bias";
	    constants_info_[7].name = "model_audio_tower_layers_0_self_attn_q_proj_bias";
	    constants_info_[7].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[7].offset = 0;
	    constants_info_[7].data_size = 5120;
	    constants_info_[7].from_folded = false;
	    constants_info_[7].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[7].shape = {1280};
	    constants_info_[7].stride = {1};
	    constants_info_[7].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[7].original_fqn = "model.audio_tower.layers.0.self_attn.q_proj.bias";
	    constants_info_[8].name = "model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[8].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[8].offset = 0;
	    constants_info_[8].data_size = 1638400;
	    constants_info_[8].from_folded = false;
	    constants_info_[8].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[8].shape = {1280, 1280};
	    constants_info_[8].stride = {1280, 1};
	    constants_info_[8].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[8].original_fqn = "model.audio_tower.layers.0.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[9].name = "model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[9].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[9].offset = 0;
	    constants_info_[9].data_size = 204800;
	    constants_info_[9].from_folded = false;
	    constants_info_[9].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[9].shape = {1280, 40};
	    constants_info_[9].stride = {40, 1};
	    constants_info_[9].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[9].original_fqn = "model.audio_tower.layers.0.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[10].name = "model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[10].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[10].offset = 0;
	    constants_info_[10].data_size = 51200;
	    constants_info_[10].from_folded = false;
	    constants_info_[10].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[10].shape = {1280, 40};
	    constants_info_[10].stride = {40, 1};
	    constants_info_[10].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[10].original_fqn = "model.audio_tower.layers.0.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[11].name = "model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[11].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[11].offset = 0;
	    constants_info_[11].data_size = 1638400;
	    constants_info_[11].from_folded = false;
	    constants_info_[11].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[11].shape = {1280, 1280};
	    constants_info_[11].stride = {1280, 1};
	    constants_info_[11].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[11].original_fqn = "model.audio_tower.layers.0.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[12].name = "model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[12].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[12].offset = 0;
	    constants_info_[12].data_size = 204800;
	    constants_info_[12].from_folded = false;
	    constants_info_[12].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[12].shape = {1280, 40};
	    constants_info_[12].stride = {40, 1};
	    constants_info_[12].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[12].original_fqn = "model.audio_tower.layers.0.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[13].name = "model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[13].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[13].offset = 0;
	    constants_info_[13].data_size = 51200;
	    constants_info_[13].from_folded = false;
	    constants_info_[13].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[13].shape = {1280, 40};
	    constants_info_[13].stride = {40, 1};
	    constants_info_[13].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[13].original_fqn = "model.audio_tower.layers.0.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[14].name = "model_audio_tower_layers_0_self_attn_v_proj_bias";
	    constants_info_[14].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[14].offset = 0;
	    constants_info_[14].data_size = 5120;
	    constants_info_[14].from_folded = false;
	    constants_info_[14].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[14].shape = {1280};
	    constants_info_[14].stride = {1};
	    constants_info_[14].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[14].original_fqn = "model.audio_tower.layers.0.self_attn.v_proj.bias";
	    constants_info_[15].name = "model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[15].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[15].offset = 0;
	    constants_info_[15].data_size = 1638400;
	    constants_info_[15].from_folded = false;
	    constants_info_[15].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[15].shape = {1280, 1280};
	    constants_info_[15].stride = {1280, 1};
	    constants_info_[15].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[15].original_fqn = "model.audio_tower.layers.0.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[16].name = "model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[16].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[16].offset = 0;
	    constants_info_[16].data_size = 204800;
	    constants_info_[16].from_folded = false;
	    constants_info_[16].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[16].shape = {1280, 40};
	    constants_info_[16].stride = {40, 1};
	    constants_info_[16].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[16].original_fqn = "model.audio_tower.layers.0.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[17].name = "model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[17].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[17].offset = 0;
	    constants_info_[17].data_size = 51200;
	    constants_info_[17].from_folded = false;
	    constants_info_[17].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[17].shape = {1280, 40};
	    constants_info_[17].stride = {40, 1};
	    constants_info_[17].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[17].original_fqn = "model.audio_tower.layers.0.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[18].name = "model_audio_tower_layers_0_self_attn_out_proj_bias";
	    constants_info_[18].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[18].offset = 0;
	    constants_info_[18].data_size = 5120;
	    constants_info_[18].from_folded = false;
	    constants_info_[18].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[18].shape = {1280};
	    constants_info_[18].stride = {1};
	    constants_info_[18].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[18].original_fqn = "model.audio_tower.layers.0.self_attn.out_proj.bias";
	    constants_info_[19].name = "model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[19].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[19].offset = 0;
	    constants_info_[19].data_size = 1638400;
	    constants_info_[19].from_folded = false;
	    constants_info_[19].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[19].shape = {1280, 1280};
	    constants_info_[19].stride = {1280, 1};
	    constants_info_[19].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[19].original_fqn = "model.audio_tower.layers.0.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[20].name = "model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[20].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[20].offset = 0;
	    constants_info_[20].data_size = 204800;
	    constants_info_[20].from_folded = false;
	    constants_info_[20].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[20].shape = {1280, 40};
	    constants_info_[20].stride = {40, 1};
	    constants_info_[20].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[20].original_fqn = "model.audio_tower.layers.0.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[21].name = "model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[21].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[21].offset = 0;
	    constants_info_[21].data_size = 51200;
	    constants_info_[21].from_folded = false;
	    constants_info_[21].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[21].shape = {1280, 40};
	    constants_info_[21].stride = {40, 1};
	    constants_info_[21].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[21].original_fqn = "model.audio_tower.layers.0.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[22].name = "model_audio_tower_layers_0_final_layer_norm_weight";
	    constants_info_[22].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[22].offset = 0;
	    constants_info_[22].data_size = 5120;
	    constants_info_[22].from_folded = false;
	    constants_info_[22].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[22].shape = {1280};
	    constants_info_[22].stride = {1};
	    constants_info_[22].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[22].original_fqn = "model.audio_tower.layers.0.final_layer_norm.weight";
	    constants_info_[23].name = "model_audio_tower_layers_0_final_layer_norm_bias";
	    constants_info_[23].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[23].offset = 0;
	    constants_info_[23].data_size = 5120;
	    constants_info_[23].from_folded = false;
	    constants_info_[23].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[23].shape = {1280};
	    constants_info_[23].stride = {1};
	    constants_info_[23].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[23].original_fqn = "model.audio_tower.layers.0.final_layer_norm.bias";
	    constants_info_[24].name = "model_audio_tower_layers_0_fc1_bias";
	    constants_info_[24].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[24].offset = 0;
	    constants_info_[24].data_size = 20480;
	    constants_info_[24].from_folded = false;
	    constants_info_[24].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[24].shape = {5120};
	    constants_info_[24].stride = {1};
	    constants_info_[24].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[24].original_fqn = "model.audio_tower.layers.0.fc1.bias";
	    constants_info_[25].name = "model_audio_tower_layers_0_fc1_parametrizations_weight_original0";
	    constants_info_[25].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[25].offset = 0;
	    constants_info_[25].data_size = 6553600;
	    constants_info_[25].from_folded = false;
	    constants_info_[25].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[25].shape = {5120, 1280};
	    constants_info_[25].stride = {1280, 1};
	    constants_info_[25].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[25].original_fqn = "model.audio_tower.layers.0.fc1.parametrizations.weight.original0";
	    constants_info_[26].name = "model_audio_tower_layers_0_fc1_parametrizations_weight_original1";
	    constants_info_[26].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[26].offset = 0;
	    constants_info_[26].data_size = 819200;
	    constants_info_[26].from_folded = false;
	    constants_info_[26].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[26].shape = {5120, 40};
	    constants_info_[26].stride = {40, 1};
	    constants_info_[26].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[26].original_fqn = "model.audio_tower.layers.0.fc1.parametrizations.weight.original1";
	    constants_info_[27].name = "model_audio_tower_layers_0_fc1_parametrizations_weight_original2";
	    constants_info_[27].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[27].offset = 0;
	    constants_info_[27].data_size = 204800;
	    constants_info_[27].from_folded = false;
	    constants_info_[27].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[27].shape = {5120, 40};
	    constants_info_[27].stride = {40, 1};
	    constants_info_[27].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[27].original_fqn = "model.audio_tower.layers.0.fc1.parametrizations.weight.original2";
	    constants_info_[28].name = "model_audio_tower_layers_0_fc2_bias";
	    constants_info_[28].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[28].offset = 0;
	    constants_info_[28].data_size = 5120;
	    constants_info_[28].from_folded = false;
	    constants_info_[28].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[28].shape = {1280};
	    constants_info_[28].stride = {1};
	    constants_info_[28].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[28].original_fqn = "model.audio_tower.layers.0.fc2.bias";
	    constants_info_[29].name = "model_audio_tower_layers_0_fc2_parametrizations_weight_original0";
	    constants_info_[29].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[29].offset = 0;
	    constants_info_[29].data_size = 6553600;
	    constants_info_[29].from_folded = false;
	    constants_info_[29].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[29].shape = {1280, 5120};
	    constants_info_[29].stride = {5120, 1};
	    constants_info_[29].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[29].original_fqn = "model.audio_tower.layers.0.fc2.parametrizations.weight.original0";
	    constants_info_[30].name = "model_audio_tower_layers_0_fc2_parametrizations_weight_original1";
	    constants_info_[30].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[30].offset = 0;
	    constants_info_[30].data_size = 819200;
	    constants_info_[30].from_folded = false;
	    constants_info_[30].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[30].shape = {1280, 160};
	    constants_info_[30].stride = {160, 1};
	    constants_info_[30].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[30].original_fqn = "model.audio_tower.layers.0.fc2.parametrizations.weight.original1";
	    constants_info_[31].name = "model_audio_tower_layers_0_fc2_parametrizations_weight_original2";
	    constants_info_[31].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[31].offset = 0;
	    constants_info_[31].data_size = 204800;
	    constants_info_[31].from_folded = false;
	    constants_info_[31].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[31].shape = {1280, 160};
	    constants_info_[31].stride = {160, 1};
	    constants_info_[31].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[31].original_fqn = "model.audio_tower.layers.0.fc2.parametrizations.weight.original2";
	    constants_info_[32].name = "model_audio_tower_layers_1_self_attn_layer_norm_weight";
	    constants_info_[32].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[32].offset = 0;
	    constants_info_[32].data_size = 5120;
	    constants_info_[32].from_folded = false;
	    constants_info_[32].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[32].shape = {1280};
	    constants_info_[32].stride = {1};
	    constants_info_[32].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[32].original_fqn = "model.audio_tower.layers.1.self_attn_layer_norm.weight";
	    constants_info_[33].name = "model_audio_tower_layers_1_self_attn_layer_norm_bias";
	    constants_info_[33].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[33].offset = 0;
	    constants_info_[33].data_size = 5120;
	    constants_info_[33].from_folded = false;
	    constants_info_[33].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[33].shape = {1280};
	    constants_info_[33].stride = {1};
	    constants_info_[33].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[33].original_fqn = "model.audio_tower.layers.1.self_attn_layer_norm.bias";
	    constants_info_[34].name = "model_audio_tower_layers_1_self_attn_q_proj_bias";
	    constants_info_[34].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[34].offset = 0;
	    constants_info_[34].data_size = 5120;
	    constants_info_[34].from_folded = false;
	    constants_info_[34].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[34].shape = {1280};
	    constants_info_[34].stride = {1};
	    constants_info_[34].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[34].original_fqn = "model.audio_tower.layers.1.self_attn.q_proj.bias";
	    constants_info_[35].name = "model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[35].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[35].offset = 0;
	    constants_info_[35].data_size = 1638400;
	    constants_info_[35].from_folded = false;
	    constants_info_[35].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[35].shape = {1280, 1280};
	    constants_info_[35].stride = {1280, 1};
	    constants_info_[35].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[35].original_fqn = "model.audio_tower.layers.1.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[36].name = "model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[36].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[36].offset = 0;
	    constants_info_[36].data_size = 204800;
	    constants_info_[36].from_folded = false;
	    constants_info_[36].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[36].shape = {1280, 40};
	    constants_info_[36].stride = {40, 1};
	    constants_info_[36].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[36].original_fqn = "model.audio_tower.layers.1.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[37].name = "model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[37].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[37].offset = 0;
	    constants_info_[37].data_size = 51200;
	    constants_info_[37].from_folded = false;
	    constants_info_[37].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[37].shape = {1280, 40};
	    constants_info_[37].stride = {40, 1};
	    constants_info_[37].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[37].original_fqn = "model.audio_tower.layers.1.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[38].name = "model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[38].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[38].offset = 0;
	    constants_info_[38].data_size = 1638400;
	    constants_info_[38].from_folded = false;
	    constants_info_[38].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[38].shape = {1280, 1280};
	    constants_info_[38].stride = {1280, 1};
	    constants_info_[38].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[38].original_fqn = "model.audio_tower.layers.1.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[39].name = "model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[39].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[39].offset = 0;
	    constants_info_[39].data_size = 204800;
	    constants_info_[39].from_folded = false;
	    constants_info_[39].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[39].shape = {1280, 40};
	    constants_info_[39].stride = {40, 1};
	    constants_info_[39].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[39].original_fqn = "model.audio_tower.layers.1.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[40].name = "model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[40].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[40].offset = 0;
	    constants_info_[40].data_size = 51200;
	    constants_info_[40].from_folded = false;
	    constants_info_[40].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[40].shape = {1280, 40};
	    constants_info_[40].stride = {40, 1};
	    constants_info_[40].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[40].original_fqn = "model.audio_tower.layers.1.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[41].name = "model_audio_tower_layers_1_self_attn_v_proj_bias";
	    constants_info_[41].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[41].offset = 0;
	    constants_info_[41].data_size = 5120;
	    constants_info_[41].from_folded = false;
	    constants_info_[41].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[41].shape = {1280};
	    constants_info_[41].stride = {1};
	    constants_info_[41].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[41].original_fqn = "model.audio_tower.layers.1.self_attn.v_proj.bias";
	    constants_info_[42].name = "model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[42].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[42].offset = 0;
	    constants_info_[42].data_size = 1638400;
	    constants_info_[42].from_folded = false;
	    constants_info_[42].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[42].shape = {1280, 1280};
	    constants_info_[42].stride = {1280, 1};
	    constants_info_[42].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[42].original_fqn = "model.audio_tower.layers.1.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[43].name = "model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[43].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[43].offset = 0;
	    constants_info_[43].data_size = 204800;
	    constants_info_[43].from_folded = false;
	    constants_info_[43].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[43].shape = {1280, 40};
	    constants_info_[43].stride = {40, 1};
	    constants_info_[43].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[43].original_fqn = "model.audio_tower.layers.1.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[44].name = "model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[44].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[44].offset = 0;
	    constants_info_[44].data_size = 51200;
	    constants_info_[44].from_folded = false;
	    constants_info_[44].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[44].shape = {1280, 40};
	    constants_info_[44].stride = {40, 1};
	    constants_info_[44].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[44].original_fqn = "model.audio_tower.layers.1.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[45].name = "model_audio_tower_layers_1_self_attn_out_proj_bias";
	    constants_info_[45].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[45].offset = 0;
	    constants_info_[45].data_size = 5120;
	    constants_info_[45].from_folded = false;
	    constants_info_[45].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[45].shape = {1280};
	    constants_info_[45].stride = {1};
	    constants_info_[45].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[45].original_fqn = "model.audio_tower.layers.1.self_attn.out_proj.bias";
	    constants_info_[46].name = "model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[46].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[46].offset = 0;
	    constants_info_[46].data_size = 1638400;
	    constants_info_[46].from_folded = false;
	    constants_info_[46].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[46].shape = {1280, 1280};
	    constants_info_[46].stride = {1280, 1};
	    constants_info_[46].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[46].original_fqn = "model.audio_tower.layers.1.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[47].name = "model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[47].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[47].offset = 0;
	    constants_info_[47].data_size = 204800;
	    constants_info_[47].from_folded = false;
	    constants_info_[47].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[47].shape = {1280, 40};
	    constants_info_[47].stride = {40, 1};
	    constants_info_[47].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[47].original_fqn = "model.audio_tower.layers.1.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[48].name = "model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[48].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[48].offset = 0;
	    constants_info_[48].data_size = 51200;
	    constants_info_[48].from_folded = false;
	    constants_info_[48].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[48].shape = {1280, 40};
	    constants_info_[48].stride = {40, 1};
	    constants_info_[48].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[48].original_fqn = "model.audio_tower.layers.1.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[49].name = "model_audio_tower_layers_1_final_layer_norm_weight";
	    constants_info_[49].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[49].offset = 0;
	    constants_info_[49].data_size = 5120;
	    constants_info_[49].from_folded = false;
	    constants_info_[49].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[49].shape = {1280};
	    constants_info_[49].stride = {1};
	    constants_info_[49].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[49].original_fqn = "model.audio_tower.layers.1.final_layer_norm.weight";
	    constants_info_[50].name = "model_audio_tower_layers_1_final_layer_norm_bias";
	    constants_info_[50].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[50].offset = 0;
	    constants_info_[50].data_size = 5120;
	    constants_info_[50].from_folded = false;
	    constants_info_[50].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[50].shape = {1280};
	    constants_info_[50].stride = {1};
	    constants_info_[50].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[50].original_fqn = "model.audio_tower.layers.1.final_layer_norm.bias";
	    constants_info_[51].name = "model_audio_tower_layers_1_fc1_bias";
	    constants_info_[51].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[51].offset = 0;
	    constants_info_[51].data_size = 20480;
	    constants_info_[51].from_folded = false;
	    constants_info_[51].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[51].shape = {5120};
	    constants_info_[51].stride = {1};
	    constants_info_[51].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[51].original_fqn = "model.audio_tower.layers.1.fc1.bias";
	    constants_info_[52].name = "model_audio_tower_layers_1_fc1_parametrizations_weight_original0";
	    constants_info_[52].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[52].offset = 0;
	    constants_info_[52].data_size = 6553600;
	    constants_info_[52].from_folded = false;
	    constants_info_[52].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[52].shape = {5120, 1280};
	    constants_info_[52].stride = {1280, 1};
	    constants_info_[52].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[52].original_fqn = "model.audio_tower.layers.1.fc1.parametrizations.weight.original0";
	    constants_info_[53].name = "model_audio_tower_layers_1_fc1_parametrizations_weight_original1";
	    constants_info_[53].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[53].offset = 0;
	    constants_info_[53].data_size = 819200;
	    constants_info_[53].from_folded = false;
	    constants_info_[53].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[53].shape = {5120, 40};
	    constants_info_[53].stride = {40, 1};
	    constants_info_[53].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[53].original_fqn = "model.audio_tower.layers.1.fc1.parametrizations.weight.original1";
	    constants_info_[54].name = "model_audio_tower_layers_1_fc1_parametrizations_weight_original2";
	    constants_info_[54].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[54].offset = 0;
	    constants_info_[54].data_size = 204800;
	    constants_info_[54].from_folded = false;
	    constants_info_[54].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[54].shape = {5120, 40};
	    constants_info_[54].stride = {40, 1};
	    constants_info_[54].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[54].original_fqn = "model.audio_tower.layers.1.fc1.parametrizations.weight.original2";
	    constants_info_[55].name = "model_audio_tower_layers_1_fc2_bias";
	    constants_info_[55].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[55].offset = 0;
	    constants_info_[55].data_size = 5120;
	    constants_info_[55].from_folded = false;
	    constants_info_[55].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[55].shape = {1280};
	    constants_info_[55].stride = {1};
	    constants_info_[55].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[55].original_fqn = "model.audio_tower.layers.1.fc2.bias";
	    constants_info_[56].name = "model_audio_tower_layers_1_fc2_parametrizations_weight_original0";
	    constants_info_[56].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[56].offset = 0;
	    constants_info_[56].data_size = 6553600;
	    constants_info_[56].from_folded = false;
	    constants_info_[56].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[56].shape = {1280, 5120};
	    constants_info_[56].stride = {5120, 1};
	    constants_info_[56].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[56].original_fqn = "model.audio_tower.layers.1.fc2.parametrizations.weight.original0";
	    constants_info_[57].name = "model_audio_tower_layers_1_fc2_parametrizations_weight_original1";
	    constants_info_[57].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[57].offset = 0;
	    constants_info_[57].data_size = 819200;
	    constants_info_[57].from_folded = false;
	    constants_info_[57].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[57].shape = {1280, 160};
	    constants_info_[57].stride = {160, 1};
	    constants_info_[57].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[57].original_fqn = "model.audio_tower.layers.1.fc2.parametrizations.weight.original1";
	    constants_info_[58].name = "model_audio_tower_layers_1_fc2_parametrizations_weight_original2";
	    constants_info_[58].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[58].offset = 0;
	    constants_info_[58].data_size = 204800;
	    constants_info_[58].from_folded = false;
	    constants_info_[58].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[58].shape = {1280, 160};
	    constants_info_[58].stride = {160, 1};
	    constants_info_[58].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[58].original_fqn = "model.audio_tower.layers.1.fc2.parametrizations.weight.original2";
	    constants_info_[59].name = "model_audio_tower_layers_2_self_attn_layer_norm_weight";
	    constants_info_[59].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[59].offset = 0;
	    constants_info_[59].data_size = 5120;
	    constants_info_[59].from_folded = false;
	    constants_info_[59].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[59].shape = {1280};
	    constants_info_[59].stride = {1};
	    constants_info_[59].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[59].original_fqn = "model.audio_tower.layers.2.self_attn_layer_norm.weight";
	    constants_info_[60].name = "model_audio_tower_layers_2_self_attn_layer_norm_bias";
	    constants_info_[60].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[60].offset = 0;
	    constants_info_[60].data_size = 5120;
	    constants_info_[60].from_folded = false;
	    constants_info_[60].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[60].shape = {1280};
	    constants_info_[60].stride = {1};
	    constants_info_[60].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[60].original_fqn = "model.audio_tower.layers.2.self_attn_layer_norm.bias";
	    constants_info_[61].name = "model_audio_tower_layers_2_self_attn_q_proj_bias";
	    constants_info_[61].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[61].offset = 0;
	    constants_info_[61].data_size = 5120;
	    constants_info_[61].from_folded = false;
	    constants_info_[61].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[61].shape = {1280};
	    constants_info_[61].stride = {1};
	    constants_info_[61].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[61].original_fqn = "model.audio_tower.layers.2.self_attn.q_proj.bias";
	    constants_info_[62].name = "model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[62].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[62].offset = 0;
	    constants_info_[62].data_size = 1638400;
	    constants_info_[62].from_folded = false;
	    constants_info_[62].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[62].shape = {1280, 1280};
	    constants_info_[62].stride = {1280, 1};
	    constants_info_[62].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[62].original_fqn = "model.audio_tower.layers.2.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[63].name = "model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[63].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[63].offset = 0;
	    constants_info_[63].data_size = 204800;
	    constants_info_[63].from_folded = false;
	    constants_info_[63].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[63].shape = {1280, 40};
	    constants_info_[63].stride = {40, 1};
	    constants_info_[63].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[63].original_fqn = "model.audio_tower.layers.2.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[64].name = "model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[64].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[64].offset = 0;
	    constants_info_[64].data_size = 51200;
	    constants_info_[64].from_folded = false;
	    constants_info_[64].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[64].shape = {1280, 40};
	    constants_info_[64].stride = {40, 1};
	    constants_info_[64].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[64].original_fqn = "model.audio_tower.layers.2.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[65].name = "model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[65].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[65].offset = 0;
	    constants_info_[65].data_size = 1638400;
	    constants_info_[65].from_folded = false;
	    constants_info_[65].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[65].shape = {1280, 1280};
	    constants_info_[65].stride = {1280, 1};
	    constants_info_[65].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[65].original_fqn = "model.audio_tower.layers.2.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[66].name = "model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[66].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[66].offset = 0;
	    constants_info_[66].data_size = 204800;
	    constants_info_[66].from_folded = false;
	    constants_info_[66].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[66].shape = {1280, 40};
	    constants_info_[66].stride = {40, 1};
	    constants_info_[66].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[66].original_fqn = "model.audio_tower.layers.2.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[67].name = "model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[67].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[67].offset = 0;
	    constants_info_[67].data_size = 51200;
	    constants_info_[67].from_folded = false;
	    constants_info_[67].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[67].shape = {1280, 40};
	    constants_info_[67].stride = {40, 1};
	    constants_info_[67].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[67].original_fqn = "model.audio_tower.layers.2.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[68].name = "model_audio_tower_layers_2_self_attn_v_proj_bias";
	    constants_info_[68].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[68].offset = 0;
	    constants_info_[68].data_size = 5120;
	    constants_info_[68].from_folded = false;
	    constants_info_[68].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[68].shape = {1280};
	    constants_info_[68].stride = {1};
	    constants_info_[68].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[68].original_fqn = "model.audio_tower.layers.2.self_attn.v_proj.bias";
	    constants_info_[69].name = "model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[69].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[69].offset = 0;
	    constants_info_[69].data_size = 1638400;
	    constants_info_[69].from_folded = false;
	    constants_info_[69].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[69].shape = {1280, 1280};
	    constants_info_[69].stride = {1280, 1};
	    constants_info_[69].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[69].original_fqn = "model.audio_tower.layers.2.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[70].name = "model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[70].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[70].offset = 0;
	    constants_info_[70].data_size = 204800;
	    constants_info_[70].from_folded = false;
	    constants_info_[70].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[70].shape = {1280, 40};
	    constants_info_[70].stride = {40, 1};
	    constants_info_[70].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[70].original_fqn = "model.audio_tower.layers.2.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[71].name = "model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[71].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[71].offset = 0;
	    constants_info_[71].data_size = 51200;
	    constants_info_[71].from_folded = false;
	    constants_info_[71].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[71].shape = {1280, 40};
	    constants_info_[71].stride = {40, 1};
	    constants_info_[71].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[71].original_fqn = "model.audio_tower.layers.2.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[72].name = "model_audio_tower_layers_2_self_attn_out_proj_bias";
	    constants_info_[72].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[72].offset = 0;
	    constants_info_[72].data_size = 5120;
	    constants_info_[72].from_folded = false;
	    constants_info_[72].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[72].shape = {1280};
	    constants_info_[72].stride = {1};
	    constants_info_[72].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[72].original_fqn = "model.audio_tower.layers.2.self_attn.out_proj.bias";
	    constants_info_[73].name = "model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[73].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[73].offset = 0;
	    constants_info_[73].data_size = 1638400;
	    constants_info_[73].from_folded = false;
	    constants_info_[73].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[73].shape = {1280, 1280};
	    constants_info_[73].stride = {1280, 1};
	    constants_info_[73].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[73].original_fqn = "model.audio_tower.layers.2.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[74].name = "model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[74].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[74].offset = 0;
	    constants_info_[74].data_size = 204800;
	    constants_info_[74].from_folded = false;
	    constants_info_[74].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[74].shape = {1280, 40};
	    constants_info_[74].stride = {40, 1};
	    constants_info_[74].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[74].original_fqn = "model.audio_tower.layers.2.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[75].name = "model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[75].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[75].offset = 0;
	    constants_info_[75].data_size = 51200;
	    constants_info_[75].from_folded = false;
	    constants_info_[75].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[75].shape = {1280, 40};
	    constants_info_[75].stride = {40, 1};
	    constants_info_[75].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[75].original_fqn = "model.audio_tower.layers.2.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[76].name = "model_audio_tower_layers_2_final_layer_norm_weight";
	    constants_info_[76].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[76].offset = 0;
	    constants_info_[76].data_size = 5120;
	    constants_info_[76].from_folded = false;
	    constants_info_[76].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[76].shape = {1280};
	    constants_info_[76].stride = {1};
	    constants_info_[76].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[76].original_fqn = "model.audio_tower.layers.2.final_layer_norm.weight";
	    constants_info_[77].name = "model_audio_tower_layers_2_final_layer_norm_bias";
	    constants_info_[77].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[77].offset = 0;
	    constants_info_[77].data_size = 5120;
	    constants_info_[77].from_folded = false;
	    constants_info_[77].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[77].shape = {1280};
	    constants_info_[77].stride = {1};
	    constants_info_[77].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[77].original_fqn = "model.audio_tower.layers.2.final_layer_norm.bias";
	    constants_info_[78].name = "model_audio_tower_layers_2_fc1_bias";
	    constants_info_[78].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[78].offset = 0;
	    constants_info_[78].data_size = 20480;
	    constants_info_[78].from_folded = false;
	    constants_info_[78].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[78].shape = {5120};
	    constants_info_[78].stride = {1};
	    constants_info_[78].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[78].original_fqn = "model.audio_tower.layers.2.fc1.bias";
	    constants_info_[79].name = "model_audio_tower_layers_2_fc1_parametrizations_weight_original0";
	    constants_info_[79].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[79].offset = 0;
	    constants_info_[79].data_size = 6553600;
	    constants_info_[79].from_folded = false;
	    constants_info_[79].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[79].shape = {5120, 1280};
	    constants_info_[79].stride = {1280, 1};
	    constants_info_[79].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[79].original_fqn = "model.audio_tower.layers.2.fc1.parametrizations.weight.original0";
	    constants_info_[80].name = "model_audio_tower_layers_2_fc1_parametrizations_weight_original1";
	    constants_info_[80].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[80].offset = 0;
	    constants_info_[80].data_size = 819200;
	    constants_info_[80].from_folded = false;
	    constants_info_[80].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[80].shape = {5120, 40};
	    constants_info_[80].stride = {40, 1};
	    constants_info_[80].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[80].original_fqn = "model.audio_tower.layers.2.fc1.parametrizations.weight.original1";
	    constants_info_[81].name = "model_audio_tower_layers_2_fc1_parametrizations_weight_original2";
	    constants_info_[81].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[81].offset = 0;
	    constants_info_[81].data_size = 204800;
	    constants_info_[81].from_folded = false;
	    constants_info_[81].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[81].shape = {5120, 40};
	    constants_info_[81].stride = {40, 1};
	    constants_info_[81].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[81].original_fqn = "model.audio_tower.layers.2.fc1.parametrizations.weight.original2";
	    constants_info_[82].name = "model_audio_tower_layers_2_fc2_bias";
	    constants_info_[82].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[82].offset = 0;
	    constants_info_[82].data_size = 5120;
	    constants_info_[82].from_folded = false;
	    constants_info_[82].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[82].shape = {1280};
	    constants_info_[82].stride = {1};
	    constants_info_[82].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[82].original_fqn = "model.audio_tower.layers.2.fc2.bias";
	    constants_info_[83].name = "model_audio_tower_layers_2_fc2_parametrizations_weight_original0";
	    constants_info_[83].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[83].offset = 0;
	    constants_info_[83].data_size = 6553600;
	    constants_info_[83].from_folded = false;
	    constants_info_[83].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[83].shape = {1280, 5120};
	    constants_info_[83].stride = {5120, 1};
	    constants_info_[83].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[83].original_fqn = "model.audio_tower.layers.2.fc2.parametrizations.weight.original0";
	    constants_info_[84].name = "model_audio_tower_layers_2_fc2_parametrizations_weight_original1";
	    constants_info_[84].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[84].offset = 0;
	    constants_info_[84].data_size = 819200;
	    constants_info_[84].from_folded = false;
	    constants_info_[84].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[84].shape = {1280, 160};
	    constants_info_[84].stride = {160, 1};
	    constants_info_[84].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[84].original_fqn = "model.audio_tower.layers.2.fc2.parametrizations.weight.original1";
	    constants_info_[85].name = "model_audio_tower_layers_2_fc2_parametrizations_weight_original2";
	    constants_info_[85].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[85].offset = 0;
	    constants_info_[85].data_size = 204800;
	    constants_info_[85].from_folded = false;
	    constants_info_[85].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[85].shape = {1280, 160};
	    constants_info_[85].stride = {160, 1};
	    constants_info_[85].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[85].original_fqn = "model.audio_tower.layers.2.fc2.parametrizations.weight.original2";
	    constants_info_[86].name = "model_audio_tower_layers_3_self_attn_layer_norm_weight";
	    constants_info_[86].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[86].offset = 0;
	    constants_info_[86].data_size = 5120;
	    constants_info_[86].from_folded = false;
	    constants_info_[86].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[86].shape = {1280};
	    constants_info_[86].stride = {1};
	    constants_info_[86].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[86].original_fqn = "model.audio_tower.layers.3.self_attn_layer_norm.weight";
	    constants_info_[87].name = "model_audio_tower_layers_3_self_attn_layer_norm_bias";
	    constants_info_[87].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[87].offset = 0;
	    constants_info_[87].data_size = 5120;
	    constants_info_[87].from_folded = false;
	    constants_info_[87].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[87].shape = {1280};
	    constants_info_[87].stride = {1};
	    constants_info_[87].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[87].original_fqn = "model.audio_tower.layers.3.self_attn_layer_norm.bias";
	    constants_info_[88].name = "model_audio_tower_layers_3_self_attn_q_proj_bias";
	    constants_info_[88].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[88].offset = 0;
	    constants_info_[88].data_size = 5120;
	    constants_info_[88].from_folded = false;
	    constants_info_[88].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[88].shape = {1280};
	    constants_info_[88].stride = {1};
	    constants_info_[88].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[88].original_fqn = "model.audio_tower.layers.3.self_attn.q_proj.bias";
	    constants_info_[89].name = "model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[89].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[89].offset = 0;
	    constants_info_[89].data_size = 1638400;
	    constants_info_[89].from_folded = false;
	    constants_info_[89].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[89].shape = {1280, 1280};
	    constants_info_[89].stride = {1280, 1};
	    constants_info_[89].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[89].original_fqn = "model.audio_tower.layers.3.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[90].name = "model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[90].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[90].offset = 0;
	    constants_info_[90].data_size = 204800;
	    constants_info_[90].from_folded = false;
	    constants_info_[90].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[90].shape = {1280, 40};
	    constants_info_[90].stride = {40, 1};
	    constants_info_[90].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[90].original_fqn = "model.audio_tower.layers.3.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[91].name = "model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[91].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[91].offset = 0;
	    constants_info_[91].data_size = 51200;
	    constants_info_[91].from_folded = false;
	    constants_info_[91].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[91].shape = {1280, 40};
	    constants_info_[91].stride = {40, 1};
	    constants_info_[91].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[91].original_fqn = "model.audio_tower.layers.3.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[92].name = "model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[92].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[92].offset = 0;
	    constants_info_[92].data_size = 1638400;
	    constants_info_[92].from_folded = false;
	    constants_info_[92].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[92].shape = {1280, 1280};
	    constants_info_[92].stride = {1280, 1};
	    constants_info_[92].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[92].original_fqn = "model.audio_tower.layers.3.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[93].name = "model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[93].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[93].offset = 0;
	    constants_info_[93].data_size = 204800;
	    constants_info_[93].from_folded = false;
	    constants_info_[93].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[93].shape = {1280, 40};
	    constants_info_[93].stride = {40, 1};
	    constants_info_[93].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[93].original_fqn = "model.audio_tower.layers.3.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[94].name = "model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[94].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[94].offset = 0;
	    constants_info_[94].data_size = 51200;
	    constants_info_[94].from_folded = false;
	    constants_info_[94].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[94].shape = {1280, 40};
	    constants_info_[94].stride = {40, 1};
	    constants_info_[94].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[94].original_fqn = "model.audio_tower.layers.3.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[95].name = "model_audio_tower_layers_3_self_attn_v_proj_bias";
	    constants_info_[95].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[95].offset = 0;
	    constants_info_[95].data_size = 5120;
	    constants_info_[95].from_folded = false;
	    constants_info_[95].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[95].shape = {1280};
	    constants_info_[95].stride = {1};
	    constants_info_[95].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[95].original_fqn = "model.audio_tower.layers.3.self_attn.v_proj.bias";
	    constants_info_[96].name = "model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[96].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[96].offset = 0;
	    constants_info_[96].data_size = 1638400;
	    constants_info_[96].from_folded = false;
	    constants_info_[96].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[96].shape = {1280, 1280};
	    constants_info_[96].stride = {1280, 1};
	    constants_info_[96].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[96].original_fqn = "model.audio_tower.layers.3.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[97].name = "model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[97].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[97].offset = 0;
	    constants_info_[97].data_size = 204800;
	    constants_info_[97].from_folded = false;
	    constants_info_[97].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[97].shape = {1280, 40};
	    constants_info_[97].stride = {40, 1};
	    constants_info_[97].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[97].original_fqn = "model.audio_tower.layers.3.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[98].name = "model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[98].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[98].offset = 0;
	    constants_info_[98].data_size = 51200;
	    constants_info_[98].from_folded = false;
	    constants_info_[98].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[98].shape = {1280, 40};
	    constants_info_[98].stride = {40, 1};
	    constants_info_[98].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[98].original_fqn = "model.audio_tower.layers.3.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[99].name = "model_audio_tower_layers_3_self_attn_out_proj_bias";
	    constants_info_[99].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[99].offset = 0;
	    constants_info_[99].data_size = 5120;
	    constants_info_[99].from_folded = false;
	    constants_info_[99].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[99].shape = {1280};
	    constants_info_[99].stride = {1};
	    constants_info_[99].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[99].original_fqn = "model.audio_tower.layers.3.self_attn.out_proj.bias";
	    constants_info_[100].name = "model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[100].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[100].offset = 0;
	    constants_info_[100].data_size = 1638400;
	    constants_info_[100].from_folded = false;
	    constants_info_[100].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[100].shape = {1280, 1280};
	    constants_info_[100].stride = {1280, 1};
	    constants_info_[100].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[100].original_fqn = "model.audio_tower.layers.3.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[101].name = "model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[101].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[101].offset = 0;
	    constants_info_[101].data_size = 204800;
	    constants_info_[101].from_folded = false;
	    constants_info_[101].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[101].shape = {1280, 40};
	    constants_info_[101].stride = {40, 1};
	    constants_info_[101].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[101].original_fqn = "model.audio_tower.layers.3.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[102].name = "model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[102].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[102].offset = 0;
	    constants_info_[102].data_size = 51200;
	    constants_info_[102].from_folded = false;
	    constants_info_[102].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[102].shape = {1280, 40};
	    constants_info_[102].stride = {40, 1};
	    constants_info_[102].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[102].original_fqn = "model.audio_tower.layers.3.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[103].name = "model_audio_tower_layers_3_final_layer_norm_weight";
	    constants_info_[103].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[103].offset = 0;
	    constants_info_[103].data_size = 5120;
	    constants_info_[103].from_folded = false;
	    constants_info_[103].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[103].shape = {1280};
	    constants_info_[103].stride = {1};
	    constants_info_[103].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[103].original_fqn = "model.audio_tower.layers.3.final_layer_norm.weight";
	    constants_info_[104].name = "model_audio_tower_layers_3_final_layer_norm_bias";
	    constants_info_[104].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[104].offset = 0;
	    constants_info_[104].data_size = 5120;
	    constants_info_[104].from_folded = false;
	    constants_info_[104].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[104].shape = {1280};
	    constants_info_[104].stride = {1};
	    constants_info_[104].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[104].original_fqn = "model.audio_tower.layers.3.final_layer_norm.bias";
	    constants_info_[105].name = "model_audio_tower_layers_3_fc1_bias";
	    constants_info_[105].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[105].offset = 0;
	    constants_info_[105].data_size = 20480;
	    constants_info_[105].from_folded = false;
	    constants_info_[105].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[105].shape = {5120};
	    constants_info_[105].stride = {1};
	    constants_info_[105].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[105].original_fqn = "model.audio_tower.layers.3.fc1.bias";
	    constants_info_[106].name = "model_audio_tower_layers_3_fc1_parametrizations_weight_original0";
	    constants_info_[106].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[106].offset = 0;
	    constants_info_[106].data_size = 6553600;
	    constants_info_[106].from_folded = false;
	    constants_info_[106].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[106].shape = {5120, 1280};
	    constants_info_[106].stride = {1280, 1};
	    constants_info_[106].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[106].original_fqn = "model.audio_tower.layers.3.fc1.parametrizations.weight.original0";
	    constants_info_[107].name = "model_audio_tower_layers_3_fc1_parametrizations_weight_original1";
	    constants_info_[107].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[107].offset = 0;
	    constants_info_[107].data_size = 819200;
	    constants_info_[107].from_folded = false;
	    constants_info_[107].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[107].shape = {5120, 40};
	    constants_info_[107].stride = {40, 1};
	    constants_info_[107].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[107].original_fqn = "model.audio_tower.layers.3.fc1.parametrizations.weight.original1";
	    constants_info_[108].name = "model_audio_tower_layers_3_fc1_parametrizations_weight_original2";
	    constants_info_[108].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[108].offset = 0;
	    constants_info_[108].data_size = 204800;
	    constants_info_[108].from_folded = false;
	    constants_info_[108].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[108].shape = {5120, 40};
	    constants_info_[108].stride = {40, 1};
	    constants_info_[108].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[108].original_fqn = "model.audio_tower.layers.3.fc1.parametrizations.weight.original2";
	    constants_info_[109].name = "model_audio_tower_layers_3_fc2_bias";
	    constants_info_[109].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[109].offset = 0;
	    constants_info_[109].data_size = 5120;
	    constants_info_[109].from_folded = false;
	    constants_info_[109].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[109].shape = {1280};
	    constants_info_[109].stride = {1};
	    constants_info_[109].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[109].original_fqn = "model.audio_tower.layers.3.fc2.bias";
	    constants_info_[110].name = "model_audio_tower_layers_3_fc2_parametrizations_weight_original0";
	    constants_info_[110].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[110].offset = 0;
	    constants_info_[110].data_size = 6553600;
	    constants_info_[110].from_folded = false;
	    constants_info_[110].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[110].shape = {1280, 5120};
	    constants_info_[110].stride = {5120, 1};
	    constants_info_[110].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[110].original_fqn = "model.audio_tower.layers.3.fc2.parametrizations.weight.original0";
	    constants_info_[111].name = "model_audio_tower_layers_3_fc2_parametrizations_weight_original1";
	    constants_info_[111].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[111].offset = 0;
	    constants_info_[111].data_size = 819200;
	    constants_info_[111].from_folded = false;
	    constants_info_[111].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[111].shape = {1280, 160};
	    constants_info_[111].stride = {160, 1};
	    constants_info_[111].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[111].original_fqn = "model.audio_tower.layers.3.fc2.parametrizations.weight.original1";
	    constants_info_[112].name = "model_audio_tower_layers_3_fc2_parametrizations_weight_original2";
	    constants_info_[112].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[112].offset = 0;
	    constants_info_[112].data_size = 204800;
	    constants_info_[112].from_folded = false;
	    constants_info_[112].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[112].shape = {1280, 160};
	    constants_info_[112].stride = {160, 1};
	    constants_info_[112].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[112].original_fqn = "model.audio_tower.layers.3.fc2.parametrizations.weight.original2";
	    constants_info_[113].name = "model_audio_tower_layers_4_self_attn_layer_norm_weight";
	    constants_info_[113].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[113].offset = 0;
	    constants_info_[113].data_size = 5120;
	    constants_info_[113].from_folded = false;
	    constants_info_[113].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[113].shape = {1280};
	    constants_info_[113].stride = {1};
	    constants_info_[113].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[113].original_fqn = "model.audio_tower.layers.4.self_attn_layer_norm.weight";
	    constants_info_[114].name = "model_audio_tower_layers_4_self_attn_layer_norm_bias";
	    constants_info_[114].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[114].offset = 0;
	    constants_info_[114].data_size = 5120;
	    constants_info_[114].from_folded = false;
	    constants_info_[114].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[114].shape = {1280};
	    constants_info_[114].stride = {1};
	    constants_info_[114].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[114].original_fqn = "model.audio_tower.layers.4.self_attn_layer_norm.bias";
	    constants_info_[115].name = "model_audio_tower_layers_4_self_attn_q_proj_bias";
	    constants_info_[115].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[115].offset = 0;
	    constants_info_[115].data_size = 5120;
	    constants_info_[115].from_folded = false;
	    constants_info_[115].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[115].shape = {1280};
	    constants_info_[115].stride = {1};
	    constants_info_[115].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[115].original_fqn = "model.audio_tower.layers.4.self_attn.q_proj.bias";
	    constants_info_[116].name = "model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[116].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[116].offset = 0;
	    constants_info_[116].data_size = 1638400;
	    constants_info_[116].from_folded = false;
	    constants_info_[116].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[116].shape = {1280, 1280};
	    constants_info_[116].stride = {1280, 1};
	    constants_info_[116].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[116].original_fqn = "model.audio_tower.layers.4.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[117].name = "model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[117].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[117].offset = 0;
	    constants_info_[117].data_size = 204800;
	    constants_info_[117].from_folded = false;
	    constants_info_[117].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[117].shape = {1280, 40};
	    constants_info_[117].stride = {40, 1};
	    constants_info_[117].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[117].original_fqn = "model.audio_tower.layers.4.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[118].name = "model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[118].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[118].offset = 0;
	    constants_info_[118].data_size = 51200;
	    constants_info_[118].from_folded = false;
	    constants_info_[118].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[118].shape = {1280, 40};
	    constants_info_[118].stride = {40, 1};
	    constants_info_[118].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[118].original_fqn = "model.audio_tower.layers.4.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[119].name = "model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[119].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[119].offset = 0;
	    constants_info_[119].data_size = 1638400;
	    constants_info_[119].from_folded = false;
	    constants_info_[119].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[119].shape = {1280, 1280};
	    constants_info_[119].stride = {1280, 1};
	    constants_info_[119].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[119].original_fqn = "model.audio_tower.layers.4.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[120].name = "model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[120].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[120].offset = 0;
	    constants_info_[120].data_size = 204800;
	    constants_info_[120].from_folded = false;
	    constants_info_[120].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[120].shape = {1280, 40};
	    constants_info_[120].stride = {40, 1};
	    constants_info_[120].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[120].original_fqn = "model.audio_tower.layers.4.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[121].name = "model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[121].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[121].offset = 0;
	    constants_info_[121].data_size = 51200;
	    constants_info_[121].from_folded = false;
	    constants_info_[121].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[121].shape = {1280, 40};
	    constants_info_[121].stride = {40, 1};
	    constants_info_[121].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[121].original_fqn = "model.audio_tower.layers.4.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[122].name = "model_audio_tower_layers_4_self_attn_v_proj_bias";
	    constants_info_[122].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[122].offset = 0;
	    constants_info_[122].data_size = 5120;
	    constants_info_[122].from_folded = false;
	    constants_info_[122].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[122].shape = {1280};
	    constants_info_[122].stride = {1};
	    constants_info_[122].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[122].original_fqn = "model.audio_tower.layers.4.self_attn.v_proj.bias";
	    constants_info_[123].name = "model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[123].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[123].offset = 0;
	    constants_info_[123].data_size = 1638400;
	    constants_info_[123].from_folded = false;
	    constants_info_[123].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[123].shape = {1280, 1280};
	    constants_info_[123].stride = {1280, 1};
	    constants_info_[123].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[123].original_fqn = "model.audio_tower.layers.4.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[124].name = "model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[124].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[124].offset = 0;
	    constants_info_[124].data_size = 204800;
	    constants_info_[124].from_folded = false;
	    constants_info_[124].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[124].shape = {1280, 40};
	    constants_info_[124].stride = {40, 1};
	    constants_info_[124].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[124].original_fqn = "model.audio_tower.layers.4.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[125].name = "model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[125].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[125].offset = 0;
	    constants_info_[125].data_size = 51200;
	    constants_info_[125].from_folded = false;
	    constants_info_[125].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[125].shape = {1280, 40};
	    constants_info_[125].stride = {40, 1};
	    constants_info_[125].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[125].original_fqn = "model.audio_tower.layers.4.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[126].name = "model_audio_tower_layers_4_self_attn_out_proj_bias";
	    constants_info_[126].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[126].offset = 0;
	    constants_info_[126].data_size = 5120;
	    constants_info_[126].from_folded = false;
	    constants_info_[126].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[126].shape = {1280};
	    constants_info_[126].stride = {1};
	    constants_info_[126].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[126].original_fqn = "model.audio_tower.layers.4.self_attn.out_proj.bias";
	    constants_info_[127].name = "model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[127].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[127].offset = 0;
	    constants_info_[127].data_size = 1638400;
	    constants_info_[127].from_folded = false;
	    constants_info_[127].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[127].shape = {1280, 1280};
	    constants_info_[127].stride = {1280, 1};
	    constants_info_[127].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[127].original_fqn = "model.audio_tower.layers.4.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[128].name = "model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[128].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[128].offset = 0;
	    constants_info_[128].data_size = 204800;
	    constants_info_[128].from_folded = false;
	    constants_info_[128].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[128].shape = {1280, 40};
	    constants_info_[128].stride = {40, 1};
	    constants_info_[128].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[128].original_fqn = "model.audio_tower.layers.4.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[129].name = "model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[129].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[129].offset = 0;
	    constants_info_[129].data_size = 51200;
	    constants_info_[129].from_folded = false;
	    constants_info_[129].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[129].shape = {1280, 40};
	    constants_info_[129].stride = {40, 1};
	    constants_info_[129].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[129].original_fqn = "model.audio_tower.layers.4.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[130].name = "model_audio_tower_layers_4_final_layer_norm_weight";
	    constants_info_[130].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[130].offset = 0;
	    constants_info_[130].data_size = 5120;
	    constants_info_[130].from_folded = false;
	    constants_info_[130].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[130].shape = {1280};
	    constants_info_[130].stride = {1};
	    constants_info_[130].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[130].original_fqn = "model.audio_tower.layers.4.final_layer_norm.weight";
	    constants_info_[131].name = "model_audio_tower_layers_4_final_layer_norm_bias";
	    constants_info_[131].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[131].offset = 0;
	    constants_info_[131].data_size = 5120;
	    constants_info_[131].from_folded = false;
	    constants_info_[131].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[131].shape = {1280};
	    constants_info_[131].stride = {1};
	    constants_info_[131].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[131].original_fqn = "model.audio_tower.layers.4.final_layer_norm.bias";
	    constants_info_[132].name = "model_audio_tower_layers_4_fc1_bias";
	    constants_info_[132].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[132].offset = 0;
	    constants_info_[132].data_size = 20480;
	    constants_info_[132].from_folded = false;
	    constants_info_[132].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[132].shape = {5120};
	    constants_info_[132].stride = {1};
	    constants_info_[132].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[132].original_fqn = "model.audio_tower.layers.4.fc1.bias";
	    constants_info_[133].name = "model_audio_tower_layers_4_fc1_parametrizations_weight_original0";
	    constants_info_[133].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[133].offset = 0;
	    constants_info_[133].data_size = 6553600;
	    constants_info_[133].from_folded = false;
	    constants_info_[133].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[133].shape = {5120, 1280};
	    constants_info_[133].stride = {1280, 1};
	    constants_info_[133].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[133].original_fqn = "model.audio_tower.layers.4.fc1.parametrizations.weight.original0";
	    constants_info_[134].name = "model_audio_tower_layers_4_fc1_parametrizations_weight_original1";
	    constants_info_[134].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[134].offset = 0;
	    constants_info_[134].data_size = 819200;
	    constants_info_[134].from_folded = false;
	    constants_info_[134].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[134].shape = {5120, 40};
	    constants_info_[134].stride = {40, 1};
	    constants_info_[134].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[134].original_fqn = "model.audio_tower.layers.4.fc1.parametrizations.weight.original1";
	    constants_info_[135].name = "model_audio_tower_layers_4_fc1_parametrizations_weight_original2";
	    constants_info_[135].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[135].offset = 0;
	    constants_info_[135].data_size = 204800;
	    constants_info_[135].from_folded = false;
	    constants_info_[135].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[135].shape = {5120, 40};
	    constants_info_[135].stride = {40, 1};
	    constants_info_[135].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[135].original_fqn = "model.audio_tower.layers.4.fc1.parametrizations.weight.original2";
	    constants_info_[136].name = "model_audio_tower_layers_4_fc2_bias";
	    constants_info_[136].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[136].offset = 0;
	    constants_info_[136].data_size = 5120;
	    constants_info_[136].from_folded = false;
	    constants_info_[136].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[136].shape = {1280};
	    constants_info_[136].stride = {1};
	    constants_info_[136].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[136].original_fqn = "model.audio_tower.layers.4.fc2.bias";
	    constants_info_[137].name = "model_audio_tower_layers_4_fc2_parametrizations_weight_original0";
	    constants_info_[137].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[137].offset = 0;
	    constants_info_[137].data_size = 6553600;
	    constants_info_[137].from_folded = false;
	    constants_info_[137].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[137].shape = {1280, 5120};
	    constants_info_[137].stride = {5120, 1};
	    constants_info_[137].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[137].original_fqn = "model.audio_tower.layers.4.fc2.parametrizations.weight.original0";
	    constants_info_[138].name = "model_audio_tower_layers_4_fc2_parametrizations_weight_original1";
	    constants_info_[138].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[138].offset = 0;
	    constants_info_[138].data_size = 819200;
	    constants_info_[138].from_folded = false;
	    constants_info_[138].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[138].shape = {1280, 160};
	    constants_info_[138].stride = {160, 1};
	    constants_info_[138].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[138].original_fqn = "model.audio_tower.layers.4.fc2.parametrizations.weight.original1";
	    constants_info_[139].name = "model_audio_tower_layers_4_fc2_parametrizations_weight_original2";
	    constants_info_[139].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[139].offset = 0;
	    constants_info_[139].data_size = 204800;
	    constants_info_[139].from_folded = false;
	    constants_info_[139].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[139].shape = {1280, 160};
	    constants_info_[139].stride = {160, 1};
	    constants_info_[139].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[139].original_fqn = "model.audio_tower.layers.4.fc2.parametrizations.weight.original2";
	    constants_info_[140].name = "model_audio_tower_layers_5_self_attn_layer_norm_weight";
	    constants_info_[140].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[140].offset = 0;
	    constants_info_[140].data_size = 5120;
	    constants_info_[140].from_folded = false;
	    constants_info_[140].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[140].shape = {1280};
	    constants_info_[140].stride = {1};
	    constants_info_[140].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[140].original_fqn = "model.audio_tower.layers.5.self_attn_layer_norm.weight";
	    constants_info_[141].name = "model_audio_tower_layers_5_self_attn_layer_norm_bias";
	    constants_info_[141].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[141].offset = 0;
	    constants_info_[141].data_size = 5120;
	    constants_info_[141].from_folded = false;
	    constants_info_[141].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[141].shape = {1280};
	    constants_info_[141].stride = {1};
	    constants_info_[141].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[141].original_fqn = "model.audio_tower.layers.5.self_attn_layer_norm.bias";
	    constants_info_[142].name = "model_audio_tower_layers_5_self_attn_q_proj_bias";
	    constants_info_[142].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[142].offset = 0;
	    constants_info_[142].data_size = 5120;
	    constants_info_[142].from_folded = false;
	    constants_info_[142].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[142].shape = {1280};
	    constants_info_[142].stride = {1};
	    constants_info_[142].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[142].original_fqn = "model.audio_tower.layers.5.self_attn.q_proj.bias";
	    constants_info_[143].name = "model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[143].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[143].offset = 0;
	    constants_info_[143].data_size = 1638400;
	    constants_info_[143].from_folded = false;
	    constants_info_[143].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[143].shape = {1280, 1280};
	    constants_info_[143].stride = {1280, 1};
	    constants_info_[143].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[143].original_fqn = "model.audio_tower.layers.5.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[144].name = "model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[144].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[144].offset = 0;
	    constants_info_[144].data_size = 204800;
	    constants_info_[144].from_folded = false;
	    constants_info_[144].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[144].shape = {1280, 40};
	    constants_info_[144].stride = {40, 1};
	    constants_info_[144].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[144].original_fqn = "model.audio_tower.layers.5.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[145].name = "model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[145].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[145].offset = 0;
	    constants_info_[145].data_size = 51200;
	    constants_info_[145].from_folded = false;
	    constants_info_[145].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[145].shape = {1280, 40};
	    constants_info_[145].stride = {40, 1};
	    constants_info_[145].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[145].original_fqn = "model.audio_tower.layers.5.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[146].name = "model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[146].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[146].offset = 0;
	    constants_info_[146].data_size = 1638400;
	    constants_info_[146].from_folded = false;
	    constants_info_[146].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[146].shape = {1280, 1280};
	    constants_info_[146].stride = {1280, 1};
	    constants_info_[146].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[146].original_fqn = "model.audio_tower.layers.5.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[147].name = "model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[147].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[147].offset = 0;
	    constants_info_[147].data_size = 204800;
	    constants_info_[147].from_folded = false;
	    constants_info_[147].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[147].shape = {1280, 40};
	    constants_info_[147].stride = {40, 1};
	    constants_info_[147].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[147].original_fqn = "model.audio_tower.layers.5.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[148].name = "model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[148].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[148].offset = 0;
	    constants_info_[148].data_size = 51200;
	    constants_info_[148].from_folded = false;
	    constants_info_[148].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[148].shape = {1280, 40};
	    constants_info_[148].stride = {40, 1};
	    constants_info_[148].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[148].original_fqn = "model.audio_tower.layers.5.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[149].name = "model_audio_tower_layers_5_self_attn_v_proj_bias";
	    constants_info_[149].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[149].offset = 0;
	    constants_info_[149].data_size = 5120;
	    constants_info_[149].from_folded = false;
	    constants_info_[149].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[149].shape = {1280};
	    constants_info_[149].stride = {1};
	    constants_info_[149].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[149].original_fqn = "model.audio_tower.layers.5.self_attn.v_proj.bias";
	    constants_info_[150].name = "model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[150].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[150].offset = 0;
	    constants_info_[150].data_size = 1638400;
	    constants_info_[150].from_folded = false;
	    constants_info_[150].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[150].shape = {1280, 1280};
	    constants_info_[150].stride = {1280, 1};
	    constants_info_[150].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[150].original_fqn = "model.audio_tower.layers.5.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[151].name = "model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[151].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[151].offset = 0;
	    constants_info_[151].data_size = 204800;
	    constants_info_[151].from_folded = false;
	    constants_info_[151].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[151].shape = {1280, 40};
	    constants_info_[151].stride = {40, 1};
	    constants_info_[151].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[151].original_fqn = "model.audio_tower.layers.5.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[152].name = "model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[152].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[152].offset = 0;
	    constants_info_[152].data_size = 51200;
	    constants_info_[152].from_folded = false;
	    constants_info_[152].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[152].shape = {1280, 40};
	    constants_info_[152].stride = {40, 1};
	    constants_info_[152].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[152].original_fqn = "model.audio_tower.layers.5.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[153].name = "model_audio_tower_layers_5_self_attn_out_proj_bias";
	    constants_info_[153].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[153].offset = 0;
	    constants_info_[153].data_size = 5120;
	    constants_info_[153].from_folded = false;
	    constants_info_[153].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[153].shape = {1280};
	    constants_info_[153].stride = {1};
	    constants_info_[153].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[153].original_fqn = "model.audio_tower.layers.5.self_attn.out_proj.bias";
	    constants_info_[154].name = "model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[154].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[154].offset = 0;
	    constants_info_[154].data_size = 1638400;
	    constants_info_[154].from_folded = false;
	    constants_info_[154].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[154].shape = {1280, 1280};
	    constants_info_[154].stride = {1280, 1};
	    constants_info_[154].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[154].original_fqn = "model.audio_tower.layers.5.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[155].name = "model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[155].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[155].offset = 0;
	    constants_info_[155].data_size = 204800;
	    constants_info_[155].from_folded = false;
	    constants_info_[155].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[155].shape = {1280, 40};
	    constants_info_[155].stride = {40, 1};
	    constants_info_[155].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[155].original_fqn = "model.audio_tower.layers.5.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[156].name = "model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[156].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[156].offset = 0;
	    constants_info_[156].data_size = 51200;
	    constants_info_[156].from_folded = false;
	    constants_info_[156].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[156].shape = {1280, 40};
	    constants_info_[156].stride = {40, 1};
	    constants_info_[156].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[156].original_fqn = "model.audio_tower.layers.5.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[157].name = "model_audio_tower_layers_5_final_layer_norm_weight";
	    constants_info_[157].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[157].offset = 0;
	    constants_info_[157].data_size = 5120;
	    constants_info_[157].from_folded = false;
	    constants_info_[157].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[157].shape = {1280};
	    constants_info_[157].stride = {1};
	    constants_info_[157].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[157].original_fqn = "model.audio_tower.layers.5.final_layer_norm.weight";
	    constants_info_[158].name = "model_audio_tower_layers_5_final_layer_norm_bias";
	    constants_info_[158].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[158].offset = 0;
	    constants_info_[158].data_size = 5120;
	    constants_info_[158].from_folded = false;
	    constants_info_[158].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[158].shape = {1280};
	    constants_info_[158].stride = {1};
	    constants_info_[158].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[158].original_fqn = "model.audio_tower.layers.5.final_layer_norm.bias";
	    constants_info_[159].name = "model_audio_tower_layers_5_fc1_bias";
	    constants_info_[159].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[159].offset = 0;
	    constants_info_[159].data_size = 20480;
	    constants_info_[159].from_folded = false;
	    constants_info_[159].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[159].shape = {5120};
	    constants_info_[159].stride = {1};
	    constants_info_[159].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[159].original_fqn = "model.audio_tower.layers.5.fc1.bias";
	    constants_info_[160].name = "model_audio_tower_layers_5_fc1_parametrizations_weight_original0";
	    constants_info_[160].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[160].offset = 0;
	    constants_info_[160].data_size = 6553600;
	    constants_info_[160].from_folded = false;
	    constants_info_[160].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[160].shape = {5120, 1280};
	    constants_info_[160].stride = {1280, 1};
	    constants_info_[160].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[160].original_fqn = "model.audio_tower.layers.5.fc1.parametrizations.weight.original0";
	    constants_info_[161].name = "model_audio_tower_layers_5_fc1_parametrizations_weight_original1";
	    constants_info_[161].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[161].offset = 0;
	    constants_info_[161].data_size = 819200;
	    constants_info_[161].from_folded = false;
	    constants_info_[161].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[161].shape = {5120, 40};
	    constants_info_[161].stride = {40, 1};
	    constants_info_[161].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[161].original_fqn = "model.audio_tower.layers.5.fc1.parametrizations.weight.original1";
	    constants_info_[162].name = "model_audio_tower_layers_5_fc1_parametrizations_weight_original2";
	    constants_info_[162].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[162].offset = 0;
	    constants_info_[162].data_size = 204800;
	    constants_info_[162].from_folded = false;
	    constants_info_[162].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[162].shape = {5120, 40};
	    constants_info_[162].stride = {40, 1};
	    constants_info_[162].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[162].original_fqn = "model.audio_tower.layers.5.fc1.parametrizations.weight.original2";
	    constants_info_[163].name = "model_audio_tower_layers_5_fc2_bias";
	    constants_info_[163].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[163].offset = 0;
	    constants_info_[163].data_size = 5120;
	    constants_info_[163].from_folded = false;
	    constants_info_[163].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[163].shape = {1280};
	    constants_info_[163].stride = {1};
	    constants_info_[163].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[163].original_fqn = "model.audio_tower.layers.5.fc2.bias";
	    constants_info_[164].name = "model_audio_tower_layers_5_fc2_parametrizations_weight_original0";
	    constants_info_[164].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[164].offset = 0;
	    constants_info_[164].data_size = 6553600;
	    constants_info_[164].from_folded = false;
	    constants_info_[164].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[164].shape = {1280, 5120};
	    constants_info_[164].stride = {5120, 1};
	    constants_info_[164].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[164].original_fqn = "model.audio_tower.layers.5.fc2.parametrizations.weight.original0";
	    constants_info_[165].name = "model_audio_tower_layers_5_fc2_parametrizations_weight_original1";
	    constants_info_[165].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[165].offset = 0;
	    constants_info_[165].data_size = 819200;
	    constants_info_[165].from_folded = false;
	    constants_info_[165].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[165].shape = {1280, 160};
	    constants_info_[165].stride = {160, 1};
	    constants_info_[165].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[165].original_fqn = "model.audio_tower.layers.5.fc2.parametrizations.weight.original1";
	    constants_info_[166].name = "model_audio_tower_layers_5_fc2_parametrizations_weight_original2";
	    constants_info_[166].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[166].offset = 0;
	    constants_info_[166].data_size = 204800;
	    constants_info_[166].from_folded = false;
	    constants_info_[166].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[166].shape = {1280, 160};
	    constants_info_[166].stride = {160, 1};
	    constants_info_[166].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[166].original_fqn = "model.audio_tower.layers.5.fc2.parametrizations.weight.original2";
	    constants_info_[167].name = "model_audio_tower_layers_6_self_attn_layer_norm_weight";
	    constants_info_[167].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[167].offset = 0;
	    constants_info_[167].data_size = 5120;
	    constants_info_[167].from_folded = false;
	    constants_info_[167].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[167].shape = {1280};
	    constants_info_[167].stride = {1};
	    constants_info_[167].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[167].original_fqn = "model.audio_tower.layers.6.self_attn_layer_norm.weight";
	    constants_info_[168].name = "model_audio_tower_layers_6_self_attn_layer_norm_bias";
	    constants_info_[168].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[168].offset = 0;
	    constants_info_[168].data_size = 5120;
	    constants_info_[168].from_folded = false;
	    constants_info_[168].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[168].shape = {1280};
	    constants_info_[168].stride = {1};
	    constants_info_[168].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[168].original_fqn = "model.audio_tower.layers.6.self_attn_layer_norm.bias";
	    constants_info_[169].name = "model_audio_tower_layers_6_self_attn_q_proj_bias";
	    constants_info_[169].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[169].offset = 0;
	    constants_info_[169].data_size = 5120;
	    constants_info_[169].from_folded = false;
	    constants_info_[169].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[169].shape = {1280};
	    constants_info_[169].stride = {1};
	    constants_info_[169].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[169].original_fqn = "model.audio_tower.layers.6.self_attn.q_proj.bias";
	    constants_info_[170].name = "model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[170].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[170].offset = 0;
	    constants_info_[170].data_size = 1638400;
	    constants_info_[170].from_folded = false;
	    constants_info_[170].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[170].shape = {1280, 1280};
	    constants_info_[170].stride = {1280, 1};
	    constants_info_[170].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[170].original_fqn = "model.audio_tower.layers.6.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[171].name = "model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[171].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[171].offset = 0;
	    constants_info_[171].data_size = 204800;
	    constants_info_[171].from_folded = false;
	    constants_info_[171].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[171].shape = {1280, 40};
	    constants_info_[171].stride = {40, 1};
	    constants_info_[171].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[171].original_fqn = "model.audio_tower.layers.6.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[172].name = "model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[172].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[172].offset = 0;
	    constants_info_[172].data_size = 51200;
	    constants_info_[172].from_folded = false;
	    constants_info_[172].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[172].shape = {1280, 40};
	    constants_info_[172].stride = {40, 1};
	    constants_info_[172].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[172].original_fqn = "model.audio_tower.layers.6.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[173].name = "model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[173].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[173].offset = 0;
	    constants_info_[173].data_size = 1638400;
	    constants_info_[173].from_folded = false;
	    constants_info_[173].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[173].shape = {1280, 1280};
	    constants_info_[173].stride = {1280, 1};
	    constants_info_[173].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[173].original_fqn = "model.audio_tower.layers.6.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[174].name = "model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[174].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[174].offset = 0;
	    constants_info_[174].data_size = 204800;
	    constants_info_[174].from_folded = false;
	    constants_info_[174].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[174].shape = {1280, 40};
	    constants_info_[174].stride = {40, 1};
	    constants_info_[174].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[174].original_fqn = "model.audio_tower.layers.6.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[175].name = "model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[175].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[175].offset = 0;
	    constants_info_[175].data_size = 51200;
	    constants_info_[175].from_folded = false;
	    constants_info_[175].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[175].shape = {1280, 40};
	    constants_info_[175].stride = {40, 1};
	    constants_info_[175].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[175].original_fqn = "model.audio_tower.layers.6.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[176].name = "model_audio_tower_layers_6_self_attn_v_proj_bias";
	    constants_info_[176].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[176].offset = 0;
	    constants_info_[176].data_size = 5120;
	    constants_info_[176].from_folded = false;
	    constants_info_[176].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[176].shape = {1280};
	    constants_info_[176].stride = {1};
	    constants_info_[176].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[176].original_fqn = "model.audio_tower.layers.6.self_attn.v_proj.bias";
	    constants_info_[177].name = "model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[177].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[177].offset = 0;
	    constants_info_[177].data_size = 1638400;
	    constants_info_[177].from_folded = false;
	    constants_info_[177].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[177].shape = {1280, 1280};
	    constants_info_[177].stride = {1280, 1};
	    constants_info_[177].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[177].original_fqn = "model.audio_tower.layers.6.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[178].name = "model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[178].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[178].offset = 0;
	    constants_info_[178].data_size = 204800;
	    constants_info_[178].from_folded = false;
	    constants_info_[178].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[178].shape = {1280, 40};
	    constants_info_[178].stride = {40, 1};
	    constants_info_[178].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[178].original_fqn = "model.audio_tower.layers.6.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[179].name = "model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[179].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[179].offset = 0;
	    constants_info_[179].data_size = 51200;
	    constants_info_[179].from_folded = false;
	    constants_info_[179].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[179].shape = {1280, 40};
	    constants_info_[179].stride = {40, 1};
	    constants_info_[179].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[179].original_fqn = "model.audio_tower.layers.6.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[180].name = "model_audio_tower_layers_6_self_attn_out_proj_bias";
	    constants_info_[180].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[180].offset = 0;
	    constants_info_[180].data_size = 5120;
	    constants_info_[180].from_folded = false;
	    constants_info_[180].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[180].shape = {1280};
	    constants_info_[180].stride = {1};
	    constants_info_[180].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[180].original_fqn = "model.audio_tower.layers.6.self_attn.out_proj.bias";
	    constants_info_[181].name = "model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[181].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[181].offset = 0;
	    constants_info_[181].data_size = 1638400;
	    constants_info_[181].from_folded = false;
	    constants_info_[181].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[181].shape = {1280, 1280};
	    constants_info_[181].stride = {1280, 1};
	    constants_info_[181].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[181].original_fqn = "model.audio_tower.layers.6.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[182].name = "model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[182].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[182].offset = 0;
	    constants_info_[182].data_size = 204800;
	    constants_info_[182].from_folded = false;
	    constants_info_[182].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[182].shape = {1280, 40};
	    constants_info_[182].stride = {40, 1};
	    constants_info_[182].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[182].original_fqn = "model.audio_tower.layers.6.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[183].name = "model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[183].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[183].offset = 0;
	    constants_info_[183].data_size = 51200;
	    constants_info_[183].from_folded = false;
	    constants_info_[183].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[183].shape = {1280, 40};
	    constants_info_[183].stride = {40, 1};
	    constants_info_[183].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[183].original_fqn = "model.audio_tower.layers.6.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[184].name = "model_audio_tower_layers_6_final_layer_norm_weight";
	    constants_info_[184].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[184].offset = 0;
	    constants_info_[184].data_size = 5120;
	    constants_info_[184].from_folded = false;
	    constants_info_[184].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[184].shape = {1280};
	    constants_info_[184].stride = {1};
	    constants_info_[184].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[184].original_fqn = "model.audio_tower.layers.6.final_layer_norm.weight";
	    constants_info_[185].name = "model_audio_tower_layers_6_final_layer_norm_bias";
	    constants_info_[185].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[185].offset = 0;
	    constants_info_[185].data_size = 5120;
	    constants_info_[185].from_folded = false;
	    constants_info_[185].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[185].shape = {1280};
	    constants_info_[185].stride = {1};
	    constants_info_[185].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[185].original_fqn = "model.audio_tower.layers.6.final_layer_norm.bias";
	    constants_info_[186].name = "model_audio_tower_layers_6_fc1_bias";
	    constants_info_[186].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[186].offset = 0;
	    constants_info_[186].data_size = 20480;
	    constants_info_[186].from_folded = false;
	    constants_info_[186].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[186].shape = {5120};
	    constants_info_[186].stride = {1};
	    constants_info_[186].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[186].original_fqn = "model.audio_tower.layers.6.fc1.bias";
	    constants_info_[187].name = "model_audio_tower_layers_6_fc1_parametrizations_weight_original0";
	    constants_info_[187].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[187].offset = 0;
	    constants_info_[187].data_size = 6553600;
	    constants_info_[187].from_folded = false;
	    constants_info_[187].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[187].shape = {5120, 1280};
	    constants_info_[187].stride = {1280, 1};
	    constants_info_[187].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[187].original_fqn = "model.audio_tower.layers.6.fc1.parametrizations.weight.original0";
	    constants_info_[188].name = "model_audio_tower_layers_6_fc1_parametrizations_weight_original1";
	    constants_info_[188].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[188].offset = 0;
	    constants_info_[188].data_size = 819200;
	    constants_info_[188].from_folded = false;
	    constants_info_[188].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[188].shape = {5120, 40};
	    constants_info_[188].stride = {40, 1};
	    constants_info_[188].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[188].original_fqn = "model.audio_tower.layers.6.fc1.parametrizations.weight.original1";
	    constants_info_[189].name = "model_audio_tower_layers_6_fc1_parametrizations_weight_original2";
	    constants_info_[189].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[189].offset = 0;
	    constants_info_[189].data_size = 204800;
	    constants_info_[189].from_folded = false;
	    constants_info_[189].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[189].shape = {5120, 40};
	    constants_info_[189].stride = {40, 1};
	    constants_info_[189].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[189].original_fqn = "model.audio_tower.layers.6.fc1.parametrizations.weight.original2";
	    constants_info_[190].name = "model_audio_tower_layers_6_fc2_bias";
	    constants_info_[190].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[190].offset = 0;
	    constants_info_[190].data_size = 5120;
	    constants_info_[190].from_folded = false;
	    constants_info_[190].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[190].shape = {1280};
	    constants_info_[190].stride = {1};
	    constants_info_[190].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[190].original_fqn = "model.audio_tower.layers.6.fc2.bias";
	    constants_info_[191].name = "model_audio_tower_layers_6_fc2_parametrizations_weight_original0";
	    constants_info_[191].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[191].offset = 0;
	    constants_info_[191].data_size = 6553600;
	    constants_info_[191].from_folded = false;
	    constants_info_[191].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[191].shape = {1280, 5120};
	    constants_info_[191].stride = {5120, 1};
	    constants_info_[191].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[191].original_fqn = "model.audio_tower.layers.6.fc2.parametrizations.weight.original0";
	    constants_info_[192].name = "model_audio_tower_layers_6_fc2_parametrizations_weight_original1";
	    constants_info_[192].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[192].offset = 0;
	    constants_info_[192].data_size = 819200;
	    constants_info_[192].from_folded = false;
	    constants_info_[192].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[192].shape = {1280, 160};
	    constants_info_[192].stride = {160, 1};
	    constants_info_[192].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[192].original_fqn = "model.audio_tower.layers.6.fc2.parametrizations.weight.original1";
	    constants_info_[193].name = "model_audio_tower_layers_6_fc2_parametrizations_weight_original2";
	    constants_info_[193].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[193].offset = 0;
	    constants_info_[193].data_size = 204800;
	    constants_info_[193].from_folded = false;
	    constants_info_[193].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[193].shape = {1280, 160};
	    constants_info_[193].stride = {160, 1};
	    constants_info_[193].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[193].original_fqn = "model.audio_tower.layers.6.fc2.parametrizations.weight.original2";
	    constants_info_[194].name = "model_audio_tower_layers_7_self_attn_layer_norm_weight";
	    constants_info_[194].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[194].offset = 0;
	    constants_info_[194].data_size = 5120;
	    constants_info_[194].from_folded = false;
	    constants_info_[194].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[194].shape = {1280};
	    constants_info_[194].stride = {1};
	    constants_info_[194].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[194].original_fqn = "model.audio_tower.layers.7.self_attn_layer_norm.weight";
	    constants_info_[195].name = "model_audio_tower_layers_7_self_attn_layer_norm_bias";
	    constants_info_[195].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[195].offset = 0;
	    constants_info_[195].data_size = 5120;
	    constants_info_[195].from_folded = false;
	    constants_info_[195].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[195].shape = {1280};
	    constants_info_[195].stride = {1};
	    constants_info_[195].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[195].original_fqn = "model.audio_tower.layers.7.self_attn_layer_norm.bias";
	    constants_info_[196].name = "model_audio_tower_layers_7_self_attn_q_proj_bias";
	    constants_info_[196].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[196].offset = 0;
	    constants_info_[196].data_size = 5120;
	    constants_info_[196].from_folded = false;
	    constants_info_[196].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[196].shape = {1280};
	    constants_info_[196].stride = {1};
	    constants_info_[196].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[196].original_fqn = "model.audio_tower.layers.7.self_attn.q_proj.bias";
	    constants_info_[197].name = "model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[197].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[197].offset = 0;
	    constants_info_[197].data_size = 1638400;
	    constants_info_[197].from_folded = false;
	    constants_info_[197].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[197].shape = {1280, 1280};
	    constants_info_[197].stride = {1280, 1};
	    constants_info_[197].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[197].original_fqn = "model.audio_tower.layers.7.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[198].name = "model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[198].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[198].offset = 0;
	    constants_info_[198].data_size = 204800;
	    constants_info_[198].from_folded = false;
	    constants_info_[198].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[198].shape = {1280, 40};
	    constants_info_[198].stride = {40, 1};
	    constants_info_[198].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[198].original_fqn = "model.audio_tower.layers.7.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[199].name = "model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[199].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[199].offset = 0;
	    constants_info_[199].data_size = 51200;
	    constants_info_[199].from_folded = false;
	    constants_info_[199].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[199].shape = {1280, 40};
	    constants_info_[199].stride = {40, 1};
	    constants_info_[199].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[199].original_fqn = "model.audio_tower.layers.7.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[200].name = "model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[200].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[200].offset = 0;
	    constants_info_[200].data_size = 1638400;
	    constants_info_[200].from_folded = false;
	    constants_info_[200].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[200].shape = {1280, 1280};
	    constants_info_[200].stride = {1280, 1};
	    constants_info_[200].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[200].original_fqn = "model.audio_tower.layers.7.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[201].name = "model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[201].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[201].offset = 0;
	    constants_info_[201].data_size = 204800;
	    constants_info_[201].from_folded = false;
	    constants_info_[201].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[201].shape = {1280, 40};
	    constants_info_[201].stride = {40, 1};
	    constants_info_[201].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[201].original_fqn = "model.audio_tower.layers.7.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[202].name = "model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[202].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[202].offset = 0;
	    constants_info_[202].data_size = 51200;
	    constants_info_[202].from_folded = false;
	    constants_info_[202].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[202].shape = {1280, 40};
	    constants_info_[202].stride = {40, 1};
	    constants_info_[202].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[202].original_fqn = "model.audio_tower.layers.7.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[203].name = "model_audio_tower_layers_7_self_attn_v_proj_bias";
	    constants_info_[203].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[203].offset = 0;
	    constants_info_[203].data_size = 5120;
	    constants_info_[203].from_folded = false;
	    constants_info_[203].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[203].shape = {1280};
	    constants_info_[203].stride = {1};
	    constants_info_[203].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[203].original_fqn = "model.audio_tower.layers.7.self_attn.v_proj.bias";
	    constants_info_[204].name = "model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[204].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[204].offset = 0;
	    constants_info_[204].data_size = 1638400;
	    constants_info_[204].from_folded = false;
	    constants_info_[204].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[204].shape = {1280, 1280};
	    constants_info_[204].stride = {1280, 1};
	    constants_info_[204].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[204].original_fqn = "model.audio_tower.layers.7.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[205].name = "model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[205].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[205].offset = 0;
	    constants_info_[205].data_size = 204800;
	    constants_info_[205].from_folded = false;
	    constants_info_[205].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[205].shape = {1280, 40};
	    constants_info_[205].stride = {40, 1};
	    constants_info_[205].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[205].original_fqn = "model.audio_tower.layers.7.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[206].name = "model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[206].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[206].offset = 0;
	    constants_info_[206].data_size = 51200;
	    constants_info_[206].from_folded = false;
	    constants_info_[206].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[206].shape = {1280, 40};
	    constants_info_[206].stride = {40, 1};
	    constants_info_[206].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[206].original_fqn = "model.audio_tower.layers.7.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[207].name = "model_audio_tower_layers_7_self_attn_out_proj_bias";
	    constants_info_[207].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[207].offset = 0;
	    constants_info_[207].data_size = 5120;
	    constants_info_[207].from_folded = false;
	    constants_info_[207].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[207].shape = {1280};
	    constants_info_[207].stride = {1};
	    constants_info_[207].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[207].original_fqn = "model.audio_tower.layers.7.self_attn.out_proj.bias";
	    constants_info_[208].name = "model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[208].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[208].offset = 0;
	    constants_info_[208].data_size = 1638400;
	    constants_info_[208].from_folded = false;
	    constants_info_[208].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[208].shape = {1280, 1280};
	    constants_info_[208].stride = {1280, 1};
	    constants_info_[208].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[208].original_fqn = "model.audio_tower.layers.7.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[209].name = "model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[209].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[209].offset = 0;
	    constants_info_[209].data_size = 204800;
	    constants_info_[209].from_folded = false;
	    constants_info_[209].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[209].shape = {1280, 40};
	    constants_info_[209].stride = {40, 1};
	    constants_info_[209].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[209].original_fqn = "model.audio_tower.layers.7.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[210].name = "model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[210].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[210].offset = 0;
	    constants_info_[210].data_size = 51200;
	    constants_info_[210].from_folded = false;
	    constants_info_[210].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[210].shape = {1280, 40};
	    constants_info_[210].stride = {40, 1};
	    constants_info_[210].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[210].original_fqn = "model.audio_tower.layers.7.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[211].name = "model_audio_tower_layers_7_final_layer_norm_weight";
	    constants_info_[211].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[211].offset = 0;
	    constants_info_[211].data_size = 5120;
	    constants_info_[211].from_folded = false;
	    constants_info_[211].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[211].shape = {1280};
	    constants_info_[211].stride = {1};
	    constants_info_[211].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[211].original_fqn = "model.audio_tower.layers.7.final_layer_norm.weight";
	    constants_info_[212].name = "model_audio_tower_layers_7_final_layer_norm_bias";
	    constants_info_[212].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[212].offset = 0;
	    constants_info_[212].data_size = 5120;
	    constants_info_[212].from_folded = false;
	    constants_info_[212].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[212].shape = {1280};
	    constants_info_[212].stride = {1};
	    constants_info_[212].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[212].original_fqn = "model.audio_tower.layers.7.final_layer_norm.bias";
	    constants_info_[213].name = "model_audio_tower_layers_7_fc1_bias";
	    constants_info_[213].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[213].offset = 0;
	    constants_info_[213].data_size = 20480;
	    constants_info_[213].from_folded = false;
	    constants_info_[213].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[213].shape = {5120};
	    constants_info_[213].stride = {1};
	    constants_info_[213].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[213].original_fqn = "model.audio_tower.layers.7.fc1.bias";
	    constants_info_[214].name = "model_audio_tower_layers_7_fc1_parametrizations_weight_original0";
	    constants_info_[214].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[214].offset = 0;
	    constants_info_[214].data_size = 6553600;
	    constants_info_[214].from_folded = false;
	    constants_info_[214].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[214].shape = {5120, 1280};
	    constants_info_[214].stride = {1280, 1};
	    constants_info_[214].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[214].original_fqn = "model.audio_tower.layers.7.fc1.parametrizations.weight.original0";
	    constants_info_[215].name = "model_audio_tower_layers_7_fc1_parametrizations_weight_original1";
	    constants_info_[215].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[215].offset = 0;
	    constants_info_[215].data_size = 819200;
	    constants_info_[215].from_folded = false;
	    constants_info_[215].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[215].shape = {5120, 40};
	    constants_info_[215].stride = {40, 1};
	    constants_info_[215].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[215].original_fqn = "model.audio_tower.layers.7.fc1.parametrizations.weight.original1";
	    constants_info_[216].name = "model_audio_tower_layers_7_fc1_parametrizations_weight_original2";
	    constants_info_[216].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[216].offset = 0;
	    constants_info_[216].data_size = 204800;
	    constants_info_[216].from_folded = false;
	    constants_info_[216].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[216].shape = {5120, 40};
	    constants_info_[216].stride = {40, 1};
	    constants_info_[216].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[216].original_fqn = "model.audio_tower.layers.7.fc1.parametrizations.weight.original2";
	    constants_info_[217].name = "model_audio_tower_layers_7_fc2_bias";
	    constants_info_[217].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[217].offset = 0;
	    constants_info_[217].data_size = 5120;
	    constants_info_[217].from_folded = false;
	    constants_info_[217].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[217].shape = {1280};
	    constants_info_[217].stride = {1};
	    constants_info_[217].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[217].original_fqn = "model.audio_tower.layers.7.fc2.bias";
	    constants_info_[218].name = "model_audio_tower_layers_7_fc2_parametrizations_weight_original0";
	    constants_info_[218].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[218].offset = 0;
	    constants_info_[218].data_size = 6553600;
	    constants_info_[218].from_folded = false;
	    constants_info_[218].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[218].shape = {1280, 5120};
	    constants_info_[218].stride = {5120, 1};
	    constants_info_[218].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[218].original_fqn = "model.audio_tower.layers.7.fc2.parametrizations.weight.original0";
	    constants_info_[219].name = "model_audio_tower_layers_7_fc2_parametrizations_weight_original1";
	    constants_info_[219].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[219].offset = 0;
	    constants_info_[219].data_size = 819200;
	    constants_info_[219].from_folded = false;
	    constants_info_[219].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[219].shape = {1280, 160};
	    constants_info_[219].stride = {160, 1};
	    constants_info_[219].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[219].original_fqn = "model.audio_tower.layers.7.fc2.parametrizations.weight.original1";
	    constants_info_[220].name = "model_audio_tower_layers_7_fc2_parametrizations_weight_original2";
	    constants_info_[220].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[220].offset = 0;
	    constants_info_[220].data_size = 204800;
	    constants_info_[220].from_folded = false;
	    constants_info_[220].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[220].shape = {1280, 160};
	    constants_info_[220].stride = {160, 1};
	    constants_info_[220].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[220].original_fqn = "model.audio_tower.layers.7.fc2.parametrizations.weight.original2";
	    constants_info_[221].name = "model_audio_tower_layers_8_self_attn_layer_norm_weight";
	    constants_info_[221].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[221].offset = 0;
	    constants_info_[221].data_size = 5120;
	    constants_info_[221].from_folded = false;
	    constants_info_[221].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[221].shape = {1280};
	    constants_info_[221].stride = {1};
	    constants_info_[221].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[221].original_fqn = "model.audio_tower.layers.8.self_attn_layer_norm.weight";
	    constants_info_[222].name = "model_audio_tower_layers_8_self_attn_layer_norm_bias";
	    constants_info_[222].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[222].offset = 0;
	    constants_info_[222].data_size = 5120;
	    constants_info_[222].from_folded = false;
	    constants_info_[222].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[222].shape = {1280};
	    constants_info_[222].stride = {1};
	    constants_info_[222].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[222].original_fqn = "model.audio_tower.layers.8.self_attn_layer_norm.bias";
	    constants_info_[223].name = "model_audio_tower_layers_8_self_attn_q_proj_bias";
	    constants_info_[223].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[223].offset = 0;
	    constants_info_[223].data_size = 5120;
	    constants_info_[223].from_folded = false;
	    constants_info_[223].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[223].shape = {1280};
	    constants_info_[223].stride = {1};
	    constants_info_[223].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[223].original_fqn = "model.audio_tower.layers.8.self_attn.q_proj.bias";
	    constants_info_[224].name = "model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[224].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[224].offset = 0;
	    constants_info_[224].data_size = 1638400;
	    constants_info_[224].from_folded = false;
	    constants_info_[224].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[224].shape = {1280, 1280};
	    constants_info_[224].stride = {1280, 1};
	    constants_info_[224].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[224].original_fqn = "model.audio_tower.layers.8.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[225].name = "model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[225].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[225].offset = 0;
	    constants_info_[225].data_size = 204800;
	    constants_info_[225].from_folded = false;
	    constants_info_[225].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[225].shape = {1280, 40};
	    constants_info_[225].stride = {40, 1};
	    constants_info_[225].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[225].original_fqn = "model.audio_tower.layers.8.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[226].name = "model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[226].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[226].offset = 0;
	    constants_info_[226].data_size = 51200;
	    constants_info_[226].from_folded = false;
	    constants_info_[226].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[226].shape = {1280, 40};
	    constants_info_[226].stride = {40, 1};
	    constants_info_[226].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[226].original_fqn = "model.audio_tower.layers.8.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[227].name = "model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[227].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[227].offset = 0;
	    constants_info_[227].data_size = 1638400;
	    constants_info_[227].from_folded = false;
	    constants_info_[227].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[227].shape = {1280, 1280};
	    constants_info_[227].stride = {1280, 1};
	    constants_info_[227].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[227].original_fqn = "model.audio_tower.layers.8.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[228].name = "model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[228].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[228].offset = 0;
	    constants_info_[228].data_size = 204800;
	    constants_info_[228].from_folded = false;
	    constants_info_[228].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[228].shape = {1280, 40};
	    constants_info_[228].stride = {40, 1};
	    constants_info_[228].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[228].original_fqn = "model.audio_tower.layers.8.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[229].name = "model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[229].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[229].offset = 0;
	    constants_info_[229].data_size = 51200;
	    constants_info_[229].from_folded = false;
	    constants_info_[229].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[229].shape = {1280, 40};
	    constants_info_[229].stride = {40, 1};
	    constants_info_[229].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[229].original_fqn = "model.audio_tower.layers.8.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[230].name = "model_audio_tower_layers_8_self_attn_v_proj_bias";
	    constants_info_[230].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[230].offset = 0;
	    constants_info_[230].data_size = 5120;
	    constants_info_[230].from_folded = false;
	    constants_info_[230].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[230].shape = {1280};
	    constants_info_[230].stride = {1};
	    constants_info_[230].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[230].original_fqn = "model.audio_tower.layers.8.self_attn.v_proj.bias";
	    constants_info_[231].name = "model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[231].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[231].offset = 0;
	    constants_info_[231].data_size = 1638400;
	    constants_info_[231].from_folded = false;
	    constants_info_[231].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[231].shape = {1280, 1280};
	    constants_info_[231].stride = {1280, 1};
	    constants_info_[231].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[231].original_fqn = "model.audio_tower.layers.8.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[232].name = "model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[232].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[232].offset = 0;
	    constants_info_[232].data_size = 204800;
	    constants_info_[232].from_folded = false;
	    constants_info_[232].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[232].shape = {1280, 40};
	    constants_info_[232].stride = {40, 1};
	    constants_info_[232].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[232].original_fqn = "model.audio_tower.layers.8.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[233].name = "model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[233].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[233].offset = 0;
	    constants_info_[233].data_size = 51200;
	    constants_info_[233].from_folded = false;
	    constants_info_[233].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[233].shape = {1280, 40};
	    constants_info_[233].stride = {40, 1};
	    constants_info_[233].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[233].original_fqn = "model.audio_tower.layers.8.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[234].name = "model_audio_tower_layers_8_self_attn_out_proj_bias";
	    constants_info_[234].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[234].offset = 0;
	    constants_info_[234].data_size = 5120;
	    constants_info_[234].from_folded = false;
	    constants_info_[234].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[234].shape = {1280};
	    constants_info_[234].stride = {1};
	    constants_info_[234].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[234].original_fqn = "model.audio_tower.layers.8.self_attn.out_proj.bias";
	    constants_info_[235].name = "model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[235].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[235].offset = 0;
	    constants_info_[235].data_size = 1638400;
	    constants_info_[235].from_folded = false;
	    constants_info_[235].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[235].shape = {1280, 1280};
	    constants_info_[235].stride = {1280, 1};
	    constants_info_[235].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[235].original_fqn = "model.audio_tower.layers.8.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[236].name = "model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[236].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[236].offset = 0;
	    constants_info_[236].data_size = 204800;
	    constants_info_[236].from_folded = false;
	    constants_info_[236].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[236].shape = {1280, 40};
	    constants_info_[236].stride = {40, 1};
	    constants_info_[236].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[236].original_fqn = "model.audio_tower.layers.8.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[237].name = "model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[237].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[237].offset = 0;
	    constants_info_[237].data_size = 51200;
	    constants_info_[237].from_folded = false;
	    constants_info_[237].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[237].shape = {1280, 40};
	    constants_info_[237].stride = {40, 1};
	    constants_info_[237].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[237].original_fqn = "model.audio_tower.layers.8.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[238].name = "model_audio_tower_layers_8_final_layer_norm_weight";
	    constants_info_[238].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[238].offset = 0;
	    constants_info_[238].data_size = 5120;
	    constants_info_[238].from_folded = false;
	    constants_info_[238].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[238].shape = {1280};
	    constants_info_[238].stride = {1};
	    constants_info_[238].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[238].original_fqn = "model.audio_tower.layers.8.final_layer_norm.weight";
	    constants_info_[239].name = "model_audio_tower_layers_8_final_layer_norm_bias";
	    constants_info_[239].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[239].offset = 0;
	    constants_info_[239].data_size = 5120;
	    constants_info_[239].from_folded = false;
	    constants_info_[239].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[239].shape = {1280};
	    constants_info_[239].stride = {1};
	    constants_info_[239].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[239].original_fqn = "model.audio_tower.layers.8.final_layer_norm.bias";
	    constants_info_[240].name = "model_audio_tower_layers_8_fc1_bias";
	    constants_info_[240].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[240].offset = 0;
	    constants_info_[240].data_size = 20480;
	    constants_info_[240].from_folded = false;
	    constants_info_[240].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[240].shape = {5120};
	    constants_info_[240].stride = {1};
	    constants_info_[240].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[240].original_fqn = "model.audio_tower.layers.8.fc1.bias";
	    constants_info_[241].name = "model_audio_tower_layers_8_fc1_parametrizations_weight_original0";
	    constants_info_[241].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[241].offset = 0;
	    constants_info_[241].data_size = 6553600;
	    constants_info_[241].from_folded = false;
	    constants_info_[241].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[241].shape = {5120, 1280};
	    constants_info_[241].stride = {1280, 1};
	    constants_info_[241].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[241].original_fqn = "model.audio_tower.layers.8.fc1.parametrizations.weight.original0";
	    constants_info_[242].name = "model_audio_tower_layers_8_fc1_parametrizations_weight_original1";
	    constants_info_[242].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[242].offset = 0;
	    constants_info_[242].data_size = 819200;
	    constants_info_[242].from_folded = false;
	    constants_info_[242].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[242].shape = {5120, 40};
	    constants_info_[242].stride = {40, 1};
	    constants_info_[242].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[242].original_fqn = "model.audio_tower.layers.8.fc1.parametrizations.weight.original1";
	    constants_info_[243].name = "model_audio_tower_layers_8_fc1_parametrizations_weight_original2";
	    constants_info_[243].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[243].offset = 0;
	    constants_info_[243].data_size = 204800;
	    constants_info_[243].from_folded = false;
	    constants_info_[243].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[243].shape = {5120, 40};
	    constants_info_[243].stride = {40, 1};
	    constants_info_[243].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[243].original_fqn = "model.audio_tower.layers.8.fc1.parametrizations.weight.original2";
	    constants_info_[244].name = "model_audio_tower_layers_8_fc2_bias";
	    constants_info_[244].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[244].offset = 0;
	    constants_info_[244].data_size = 5120;
	    constants_info_[244].from_folded = false;
	    constants_info_[244].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[244].shape = {1280};
	    constants_info_[244].stride = {1};
	    constants_info_[244].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[244].original_fqn = "model.audio_tower.layers.8.fc2.bias";
	    constants_info_[245].name = "model_audio_tower_layers_8_fc2_parametrizations_weight_original0";
	    constants_info_[245].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[245].offset = 0;
	    constants_info_[245].data_size = 6553600;
	    constants_info_[245].from_folded = false;
	    constants_info_[245].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[245].shape = {1280, 5120};
	    constants_info_[245].stride = {5120, 1};
	    constants_info_[245].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[245].original_fqn = "model.audio_tower.layers.8.fc2.parametrizations.weight.original0";
	    constants_info_[246].name = "model_audio_tower_layers_8_fc2_parametrizations_weight_original1";
	    constants_info_[246].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[246].offset = 0;
	    constants_info_[246].data_size = 819200;
	    constants_info_[246].from_folded = false;
	    constants_info_[246].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[246].shape = {1280, 160};
	    constants_info_[246].stride = {160, 1};
	    constants_info_[246].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[246].original_fqn = "model.audio_tower.layers.8.fc2.parametrizations.weight.original1";
	    constants_info_[247].name = "model_audio_tower_layers_8_fc2_parametrizations_weight_original2";
	    constants_info_[247].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[247].offset = 0;
	    constants_info_[247].data_size = 204800;
	    constants_info_[247].from_folded = false;
	    constants_info_[247].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[247].shape = {1280, 160};
	    constants_info_[247].stride = {160, 1};
	    constants_info_[247].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[247].original_fqn = "model.audio_tower.layers.8.fc2.parametrizations.weight.original2";
	    constants_info_[248].name = "model_audio_tower_layers_9_self_attn_layer_norm_weight";
	    constants_info_[248].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[248].offset = 0;
	    constants_info_[248].data_size = 5120;
	    constants_info_[248].from_folded = false;
	    constants_info_[248].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[248].shape = {1280};
	    constants_info_[248].stride = {1};
	    constants_info_[248].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[248].original_fqn = "model.audio_tower.layers.9.self_attn_layer_norm.weight";
	    constants_info_[249].name = "model_audio_tower_layers_9_self_attn_layer_norm_bias";
	    constants_info_[249].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[249].offset = 0;
	    constants_info_[249].data_size = 5120;
	    constants_info_[249].from_folded = false;
	    constants_info_[249].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[249].shape = {1280};
	    constants_info_[249].stride = {1};
	    constants_info_[249].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[249].original_fqn = "model.audio_tower.layers.9.self_attn_layer_norm.bias";
	    constants_info_[250].name = "model_audio_tower_layers_9_self_attn_q_proj_bias";
	    constants_info_[250].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[250].offset = 0;
	    constants_info_[250].data_size = 5120;
	    constants_info_[250].from_folded = false;
	    constants_info_[250].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[250].shape = {1280};
	    constants_info_[250].stride = {1};
	    constants_info_[250].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[250].original_fqn = "model.audio_tower.layers.9.self_attn.q_proj.bias";
	    constants_info_[251].name = "model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[251].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[251].offset = 0;
	    constants_info_[251].data_size = 1638400;
	    constants_info_[251].from_folded = false;
	    constants_info_[251].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[251].shape = {1280, 1280};
	    constants_info_[251].stride = {1280, 1};
	    constants_info_[251].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[251].original_fqn = "model.audio_tower.layers.9.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[252].name = "model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[252].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[252].offset = 0;
	    constants_info_[252].data_size = 204800;
	    constants_info_[252].from_folded = false;
	    constants_info_[252].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[252].shape = {1280, 40};
	    constants_info_[252].stride = {40, 1};
	    constants_info_[252].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[252].original_fqn = "model.audio_tower.layers.9.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[253].name = "model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[253].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[253].offset = 0;
	    constants_info_[253].data_size = 51200;
	    constants_info_[253].from_folded = false;
	    constants_info_[253].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[253].shape = {1280, 40};
	    constants_info_[253].stride = {40, 1};
	    constants_info_[253].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[253].original_fqn = "model.audio_tower.layers.9.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[254].name = "model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[254].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[254].offset = 0;
	    constants_info_[254].data_size = 1638400;
	    constants_info_[254].from_folded = false;
	    constants_info_[254].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[254].shape = {1280, 1280};
	    constants_info_[254].stride = {1280, 1};
	    constants_info_[254].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[254].original_fqn = "model.audio_tower.layers.9.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[255].name = "model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[255].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[255].offset = 0;
	    constants_info_[255].data_size = 204800;
	    constants_info_[255].from_folded = false;
	    constants_info_[255].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[255].shape = {1280, 40};
	    constants_info_[255].stride = {40, 1};
	    constants_info_[255].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[255].original_fqn = "model.audio_tower.layers.9.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[256].name = "model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[256].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[256].offset = 0;
	    constants_info_[256].data_size = 51200;
	    constants_info_[256].from_folded = false;
	    constants_info_[256].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[256].shape = {1280, 40};
	    constants_info_[256].stride = {40, 1};
	    constants_info_[256].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[256].original_fqn = "model.audio_tower.layers.9.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[257].name = "model_audio_tower_layers_9_self_attn_v_proj_bias";
	    constants_info_[257].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[257].offset = 0;
	    constants_info_[257].data_size = 5120;
	    constants_info_[257].from_folded = false;
	    constants_info_[257].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[257].shape = {1280};
	    constants_info_[257].stride = {1};
	    constants_info_[257].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[257].original_fqn = "model.audio_tower.layers.9.self_attn.v_proj.bias";
	    constants_info_[258].name = "model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[258].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[258].offset = 0;
	    constants_info_[258].data_size = 1638400;
	    constants_info_[258].from_folded = false;
	    constants_info_[258].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[258].shape = {1280, 1280};
	    constants_info_[258].stride = {1280, 1};
	    constants_info_[258].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[258].original_fqn = "model.audio_tower.layers.9.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[259].name = "model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[259].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[259].offset = 0;
	    constants_info_[259].data_size = 204800;
	    constants_info_[259].from_folded = false;
	    constants_info_[259].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[259].shape = {1280, 40};
	    constants_info_[259].stride = {40, 1};
	    constants_info_[259].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[259].original_fqn = "model.audio_tower.layers.9.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[260].name = "model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[260].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[260].offset = 0;
	    constants_info_[260].data_size = 51200;
	    constants_info_[260].from_folded = false;
	    constants_info_[260].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[260].shape = {1280, 40};
	    constants_info_[260].stride = {40, 1};
	    constants_info_[260].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[260].original_fqn = "model.audio_tower.layers.9.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[261].name = "model_audio_tower_layers_9_self_attn_out_proj_bias";
	    constants_info_[261].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[261].offset = 0;
	    constants_info_[261].data_size = 5120;
	    constants_info_[261].from_folded = false;
	    constants_info_[261].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[261].shape = {1280};
	    constants_info_[261].stride = {1};
	    constants_info_[261].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[261].original_fqn = "model.audio_tower.layers.9.self_attn.out_proj.bias";
	    constants_info_[262].name = "model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[262].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[262].offset = 0;
	    constants_info_[262].data_size = 1638400;
	    constants_info_[262].from_folded = false;
	    constants_info_[262].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[262].shape = {1280, 1280};
	    constants_info_[262].stride = {1280, 1};
	    constants_info_[262].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[262].original_fqn = "model.audio_tower.layers.9.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[263].name = "model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[263].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[263].offset = 0;
	    constants_info_[263].data_size = 204800;
	    constants_info_[263].from_folded = false;
	    constants_info_[263].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[263].shape = {1280, 40};
	    constants_info_[263].stride = {40, 1};
	    constants_info_[263].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[263].original_fqn = "model.audio_tower.layers.9.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[264].name = "model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[264].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[264].offset = 0;
	    constants_info_[264].data_size = 51200;
	    constants_info_[264].from_folded = false;
	    constants_info_[264].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[264].shape = {1280, 40};
	    constants_info_[264].stride = {40, 1};
	    constants_info_[264].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[264].original_fqn = "model.audio_tower.layers.9.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[265].name = "model_audio_tower_layers_9_final_layer_norm_weight";
	    constants_info_[265].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[265].offset = 0;
	    constants_info_[265].data_size = 5120;
	    constants_info_[265].from_folded = false;
	    constants_info_[265].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[265].shape = {1280};
	    constants_info_[265].stride = {1};
	    constants_info_[265].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[265].original_fqn = "model.audio_tower.layers.9.final_layer_norm.weight";
	    constants_info_[266].name = "model_audio_tower_layers_9_final_layer_norm_bias";
	    constants_info_[266].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[266].offset = 0;
	    constants_info_[266].data_size = 5120;
	    constants_info_[266].from_folded = false;
	    constants_info_[266].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[266].shape = {1280};
	    constants_info_[266].stride = {1};
	    constants_info_[266].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[266].original_fqn = "model.audio_tower.layers.9.final_layer_norm.bias";
	    constants_info_[267].name = "model_audio_tower_layers_9_fc1_bias";
	    constants_info_[267].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[267].offset = 0;
	    constants_info_[267].data_size = 20480;
	    constants_info_[267].from_folded = false;
	    constants_info_[267].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[267].shape = {5120};
	    constants_info_[267].stride = {1};
	    constants_info_[267].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[267].original_fqn = "model.audio_tower.layers.9.fc1.bias";
	    constants_info_[268].name = "model_audio_tower_layers_9_fc1_parametrizations_weight_original0";
	    constants_info_[268].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[268].offset = 0;
	    constants_info_[268].data_size = 6553600;
	    constants_info_[268].from_folded = false;
	    constants_info_[268].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[268].shape = {5120, 1280};
	    constants_info_[268].stride = {1280, 1};
	    constants_info_[268].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[268].original_fqn = "model.audio_tower.layers.9.fc1.parametrizations.weight.original0";
	    constants_info_[269].name = "model_audio_tower_layers_9_fc1_parametrizations_weight_original1";
	    constants_info_[269].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[269].offset = 0;
	    constants_info_[269].data_size = 819200;
	    constants_info_[269].from_folded = false;
	    constants_info_[269].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[269].shape = {5120, 40};
	    constants_info_[269].stride = {40, 1};
	    constants_info_[269].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[269].original_fqn = "model.audio_tower.layers.9.fc1.parametrizations.weight.original1";
	    constants_info_[270].name = "model_audio_tower_layers_9_fc1_parametrizations_weight_original2";
	    constants_info_[270].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[270].offset = 0;
	    constants_info_[270].data_size = 204800;
	    constants_info_[270].from_folded = false;
	    constants_info_[270].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[270].shape = {5120, 40};
	    constants_info_[270].stride = {40, 1};
	    constants_info_[270].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[270].original_fqn = "model.audio_tower.layers.9.fc1.parametrizations.weight.original2";
	    constants_info_[271].name = "model_audio_tower_layers_9_fc2_bias";
	    constants_info_[271].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[271].offset = 0;
	    constants_info_[271].data_size = 5120;
	    constants_info_[271].from_folded = false;
	    constants_info_[271].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[271].shape = {1280};
	    constants_info_[271].stride = {1};
	    constants_info_[271].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[271].original_fqn = "model.audio_tower.layers.9.fc2.bias";
	    constants_info_[272].name = "model_audio_tower_layers_9_fc2_parametrizations_weight_original0";
	    constants_info_[272].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[272].offset = 0;
	    constants_info_[272].data_size = 6553600;
	    constants_info_[272].from_folded = false;
	    constants_info_[272].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[272].shape = {1280, 5120};
	    constants_info_[272].stride = {5120, 1};
	    constants_info_[272].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[272].original_fqn = "model.audio_tower.layers.9.fc2.parametrizations.weight.original0";
	    constants_info_[273].name = "model_audio_tower_layers_9_fc2_parametrizations_weight_original1";
	    constants_info_[273].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[273].offset = 0;
	    constants_info_[273].data_size = 819200;
	    constants_info_[273].from_folded = false;
	    constants_info_[273].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[273].shape = {1280, 160};
	    constants_info_[273].stride = {160, 1};
	    constants_info_[273].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[273].original_fqn = "model.audio_tower.layers.9.fc2.parametrizations.weight.original1";
	    constants_info_[274].name = "model_audio_tower_layers_9_fc2_parametrizations_weight_original2";
	    constants_info_[274].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[274].offset = 0;
	    constants_info_[274].data_size = 204800;
	    constants_info_[274].from_folded = false;
	    constants_info_[274].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[274].shape = {1280, 160};
	    constants_info_[274].stride = {160, 1};
	    constants_info_[274].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[274].original_fqn = "model.audio_tower.layers.9.fc2.parametrizations.weight.original2";
	    constants_info_[275].name = "model_audio_tower_layers_10_self_attn_layer_norm_weight";
	    constants_info_[275].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[275].offset = 0;
	    constants_info_[275].data_size = 5120;
	    constants_info_[275].from_folded = false;
	    constants_info_[275].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[275].shape = {1280};
	    constants_info_[275].stride = {1};
	    constants_info_[275].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[275].original_fqn = "model.audio_tower.layers.10.self_attn_layer_norm.weight";
	    constants_info_[276].name = "model_audio_tower_layers_10_self_attn_layer_norm_bias";
	    constants_info_[276].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[276].offset = 0;
	    constants_info_[276].data_size = 5120;
	    constants_info_[276].from_folded = false;
	    constants_info_[276].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[276].shape = {1280};
	    constants_info_[276].stride = {1};
	    constants_info_[276].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[276].original_fqn = "model.audio_tower.layers.10.self_attn_layer_norm.bias";
	    constants_info_[277].name = "model_audio_tower_layers_10_self_attn_q_proj_bias";
	    constants_info_[277].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[277].offset = 0;
	    constants_info_[277].data_size = 5120;
	    constants_info_[277].from_folded = false;
	    constants_info_[277].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[277].shape = {1280};
	    constants_info_[277].stride = {1};
	    constants_info_[277].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[277].original_fqn = "model.audio_tower.layers.10.self_attn.q_proj.bias";
	    constants_info_[278].name = "model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[278].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[278].offset = 0;
	    constants_info_[278].data_size = 1638400;
	    constants_info_[278].from_folded = false;
	    constants_info_[278].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[278].shape = {1280, 1280};
	    constants_info_[278].stride = {1280, 1};
	    constants_info_[278].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[278].original_fqn = "model.audio_tower.layers.10.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[279].name = "model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[279].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[279].offset = 0;
	    constants_info_[279].data_size = 204800;
	    constants_info_[279].from_folded = false;
	    constants_info_[279].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[279].shape = {1280, 40};
	    constants_info_[279].stride = {40, 1};
	    constants_info_[279].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[279].original_fqn = "model.audio_tower.layers.10.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[280].name = "model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[280].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[280].offset = 0;
	    constants_info_[280].data_size = 51200;
	    constants_info_[280].from_folded = false;
	    constants_info_[280].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[280].shape = {1280, 40};
	    constants_info_[280].stride = {40, 1};
	    constants_info_[280].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[280].original_fqn = "model.audio_tower.layers.10.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[281].name = "model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[281].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[281].offset = 0;
	    constants_info_[281].data_size = 1638400;
	    constants_info_[281].from_folded = false;
	    constants_info_[281].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[281].shape = {1280, 1280};
	    constants_info_[281].stride = {1280, 1};
	    constants_info_[281].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[281].original_fqn = "model.audio_tower.layers.10.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[282].name = "model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[282].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[282].offset = 0;
	    constants_info_[282].data_size = 204800;
	    constants_info_[282].from_folded = false;
	    constants_info_[282].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[282].shape = {1280, 40};
	    constants_info_[282].stride = {40, 1};
	    constants_info_[282].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[282].original_fqn = "model.audio_tower.layers.10.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[283].name = "model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[283].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[283].offset = 0;
	    constants_info_[283].data_size = 51200;
	    constants_info_[283].from_folded = false;
	    constants_info_[283].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[283].shape = {1280, 40};
	    constants_info_[283].stride = {40, 1};
	    constants_info_[283].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[283].original_fqn = "model.audio_tower.layers.10.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[284].name = "model_audio_tower_layers_10_self_attn_v_proj_bias";
	    constants_info_[284].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[284].offset = 0;
	    constants_info_[284].data_size = 5120;
	    constants_info_[284].from_folded = false;
	    constants_info_[284].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[284].shape = {1280};
	    constants_info_[284].stride = {1};
	    constants_info_[284].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[284].original_fqn = "model.audio_tower.layers.10.self_attn.v_proj.bias";
	    constants_info_[285].name = "model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[285].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[285].offset = 0;
	    constants_info_[285].data_size = 1638400;
	    constants_info_[285].from_folded = false;
	    constants_info_[285].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[285].shape = {1280, 1280};
	    constants_info_[285].stride = {1280, 1};
	    constants_info_[285].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[285].original_fqn = "model.audio_tower.layers.10.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[286].name = "model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[286].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[286].offset = 0;
	    constants_info_[286].data_size = 204800;
	    constants_info_[286].from_folded = false;
	    constants_info_[286].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[286].shape = {1280, 40};
	    constants_info_[286].stride = {40, 1};
	    constants_info_[286].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[286].original_fqn = "model.audio_tower.layers.10.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[287].name = "model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[287].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[287].offset = 0;
	    constants_info_[287].data_size = 51200;
	    constants_info_[287].from_folded = false;
	    constants_info_[287].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[287].shape = {1280, 40};
	    constants_info_[287].stride = {40, 1};
	    constants_info_[287].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[287].original_fqn = "model.audio_tower.layers.10.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[288].name = "model_audio_tower_layers_10_self_attn_out_proj_bias";
	    constants_info_[288].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[288].offset = 0;
	    constants_info_[288].data_size = 5120;
	    constants_info_[288].from_folded = false;
	    constants_info_[288].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[288].shape = {1280};
	    constants_info_[288].stride = {1};
	    constants_info_[288].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[288].original_fqn = "model.audio_tower.layers.10.self_attn.out_proj.bias";
	    constants_info_[289].name = "model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[289].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[289].offset = 0;
	    constants_info_[289].data_size = 1638400;
	    constants_info_[289].from_folded = false;
	    constants_info_[289].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[289].shape = {1280, 1280};
	    constants_info_[289].stride = {1280, 1};
	    constants_info_[289].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[289].original_fqn = "model.audio_tower.layers.10.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[290].name = "model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[290].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[290].offset = 0;
	    constants_info_[290].data_size = 204800;
	    constants_info_[290].from_folded = false;
	    constants_info_[290].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[290].shape = {1280, 40};
	    constants_info_[290].stride = {40, 1};
	    constants_info_[290].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[290].original_fqn = "model.audio_tower.layers.10.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[291].name = "model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[291].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[291].offset = 0;
	    constants_info_[291].data_size = 51200;
	    constants_info_[291].from_folded = false;
	    constants_info_[291].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[291].shape = {1280, 40};
	    constants_info_[291].stride = {40, 1};
	    constants_info_[291].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[291].original_fqn = "model.audio_tower.layers.10.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[292].name = "model_audio_tower_layers_10_final_layer_norm_weight";
	    constants_info_[292].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[292].offset = 0;
	    constants_info_[292].data_size = 5120;
	    constants_info_[292].from_folded = false;
	    constants_info_[292].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[292].shape = {1280};
	    constants_info_[292].stride = {1};
	    constants_info_[292].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[292].original_fqn = "model.audio_tower.layers.10.final_layer_norm.weight";
	    constants_info_[293].name = "model_audio_tower_layers_10_final_layer_norm_bias";
	    constants_info_[293].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[293].offset = 0;
	    constants_info_[293].data_size = 5120;
	    constants_info_[293].from_folded = false;
	    constants_info_[293].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[293].shape = {1280};
	    constants_info_[293].stride = {1};
	    constants_info_[293].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[293].original_fqn = "model.audio_tower.layers.10.final_layer_norm.bias";
	    constants_info_[294].name = "model_audio_tower_layers_10_fc1_bias";
	    constants_info_[294].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[294].offset = 0;
	    constants_info_[294].data_size = 20480;
	    constants_info_[294].from_folded = false;
	    constants_info_[294].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[294].shape = {5120};
	    constants_info_[294].stride = {1};
	    constants_info_[294].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[294].original_fqn = "model.audio_tower.layers.10.fc1.bias";
	    constants_info_[295].name = "model_audio_tower_layers_10_fc1_parametrizations_weight_original0";
	    constants_info_[295].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[295].offset = 0;
	    constants_info_[295].data_size = 6553600;
	    constants_info_[295].from_folded = false;
	    constants_info_[295].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[295].shape = {5120, 1280};
	    constants_info_[295].stride = {1280, 1};
	    constants_info_[295].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[295].original_fqn = "model.audio_tower.layers.10.fc1.parametrizations.weight.original0";
	    constants_info_[296].name = "model_audio_tower_layers_10_fc1_parametrizations_weight_original1";
	    constants_info_[296].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[296].offset = 0;
	    constants_info_[296].data_size = 819200;
	    constants_info_[296].from_folded = false;
	    constants_info_[296].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[296].shape = {5120, 40};
	    constants_info_[296].stride = {40, 1};
	    constants_info_[296].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[296].original_fqn = "model.audio_tower.layers.10.fc1.parametrizations.weight.original1";
	    constants_info_[297].name = "model_audio_tower_layers_10_fc1_parametrizations_weight_original2";
	    constants_info_[297].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[297].offset = 0;
	    constants_info_[297].data_size = 204800;
	    constants_info_[297].from_folded = false;
	    constants_info_[297].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[297].shape = {5120, 40};
	    constants_info_[297].stride = {40, 1};
	    constants_info_[297].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[297].original_fqn = "model.audio_tower.layers.10.fc1.parametrizations.weight.original2";
	    constants_info_[298].name = "model_audio_tower_layers_10_fc2_bias";
	    constants_info_[298].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[298].offset = 0;
	    constants_info_[298].data_size = 5120;
	    constants_info_[298].from_folded = false;
	    constants_info_[298].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[298].shape = {1280};
	    constants_info_[298].stride = {1};
	    constants_info_[298].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[298].original_fqn = "model.audio_tower.layers.10.fc2.bias";
	    constants_info_[299].name = "model_audio_tower_layers_10_fc2_parametrizations_weight_original0";
	    constants_info_[299].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[299].offset = 0;
	    constants_info_[299].data_size = 6553600;
	    constants_info_[299].from_folded = false;
	    constants_info_[299].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[299].shape = {1280, 5120};
	    constants_info_[299].stride = {5120, 1};
	    constants_info_[299].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[299].original_fqn = "model.audio_tower.layers.10.fc2.parametrizations.weight.original0";
	    constants_info_[300].name = "model_audio_tower_layers_10_fc2_parametrizations_weight_original1";
	    constants_info_[300].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[300].offset = 0;
	    constants_info_[300].data_size = 819200;
	    constants_info_[300].from_folded = false;
	    constants_info_[300].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[300].shape = {1280, 160};
	    constants_info_[300].stride = {160, 1};
	    constants_info_[300].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[300].original_fqn = "model.audio_tower.layers.10.fc2.parametrizations.weight.original1";
	    constants_info_[301].name = "model_audio_tower_layers_10_fc2_parametrizations_weight_original2";
	    constants_info_[301].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[301].offset = 0;
	    constants_info_[301].data_size = 204800;
	    constants_info_[301].from_folded = false;
	    constants_info_[301].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[301].shape = {1280, 160};
	    constants_info_[301].stride = {160, 1};
	    constants_info_[301].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[301].original_fqn = "model.audio_tower.layers.10.fc2.parametrizations.weight.original2";
	    constants_info_[302].name = "model_audio_tower_layers_11_self_attn_layer_norm_weight";
	    constants_info_[302].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[302].offset = 0;
	    constants_info_[302].data_size = 5120;
	    constants_info_[302].from_folded = false;
	    constants_info_[302].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[302].shape = {1280};
	    constants_info_[302].stride = {1};
	    constants_info_[302].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[302].original_fqn = "model.audio_tower.layers.11.self_attn_layer_norm.weight";
	    constants_info_[303].name = "model_audio_tower_layers_11_self_attn_layer_norm_bias";
	    constants_info_[303].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[303].offset = 0;
	    constants_info_[303].data_size = 5120;
	    constants_info_[303].from_folded = false;
	    constants_info_[303].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[303].shape = {1280};
	    constants_info_[303].stride = {1};
	    constants_info_[303].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[303].original_fqn = "model.audio_tower.layers.11.self_attn_layer_norm.bias";
	    constants_info_[304].name = "model_audio_tower_layers_11_self_attn_q_proj_bias";
	    constants_info_[304].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[304].offset = 0;
	    constants_info_[304].data_size = 5120;
	    constants_info_[304].from_folded = false;
	    constants_info_[304].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[304].shape = {1280};
	    constants_info_[304].stride = {1};
	    constants_info_[304].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[304].original_fqn = "model.audio_tower.layers.11.self_attn.q_proj.bias";
	    constants_info_[305].name = "model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[305].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[305].offset = 0;
	    constants_info_[305].data_size = 1638400;
	    constants_info_[305].from_folded = false;
	    constants_info_[305].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[305].shape = {1280, 1280};
	    constants_info_[305].stride = {1280, 1};
	    constants_info_[305].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[305].original_fqn = "model.audio_tower.layers.11.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[306].name = "model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[306].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[306].offset = 0;
	    constants_info_[306].data_size = 204800;
	    constants_info_[306].from_folded = false;
	    constants_info_[306].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[306].shape = {1280, 40};
	    constants_info_[306].stride = {40, 1};
	    constants_info_[306].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[306].original_fqn = "model.audio_tower.layers.11.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[307].name = "model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[307].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[307].offset = 0;
	    constants_info_[307].data_size = 51200;
	    constants_info_[307].from_folded = false;
	    constants_info_[307].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[307].shape = {1280, 40};
	    constants_info_[307].stride = {40, 1};
	    constants_info_[307].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[307].original_fqn = "model.audio_tower.layers.11.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[308].name = "model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[308].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[308].offset = 0;
	    constants_info_[308].data_size = 1638400;
	    constants_info_[308].from_folded = false;
	    constants_info_[308].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[308].shape = {1280, 1280};
	    constants_info_[308].stride = {1280, 1};
	    constants_info_[308].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[308].original_fqn = "model.audio_tower.layers.11.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[309].name = "model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[309].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[309].offset = 0;
	    constants_info_[309].data_size = 204800;
	    constants_info_[309].from_folded = false;
	    constants_info_[309].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[309].shape = {1280, 40};
	    constants_info_[309].stride = {40, 1};
	    constants_info_[309].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[309].original_fqn = "model.audio_tower.layers.11.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[310].name = "model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[310].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[310].offset = 0;
	    constants_info_[310].data_size = 51200;
	    constants_info_[310].from_folded = false;
	    constants_info_[310].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[310].shape = {1280, 40};
	    constants_info_[310].stride = {40, 1};
	    constants_info_[310].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[310].original_fqn = "model.audio_tower.layers.11.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[311].name = "model_audio_tower_layers_11_self_attn_v_proj_bias";
	    constants_info_[311].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[311].offset = 0;
	    constants_info_[311].data_size = 5120;
	    constants_info_[311].from_folded = false;
	    constants_info_[311].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[311].shape = {1280};
	    constants_info_[311].stride = {1};
	    constants_info_[311].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[311].original_fqn = "model.audio_tower.layers.11.self_attn.v_proj.bias";
	    constants_info_[312].name = "model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[312].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[312].offset = 0;
	    constants_info_[312].data_size = 1638400;
	    constants_info_[312].from_folded = false;
	    constants_info_[312].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[312].shape = {1280, 1280};
	    constants_info_[312].stride = {1280, 1};
	    constants_info_[312].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[312].original_fqn = "model.audio_tower.layers.11.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[313].name = "model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[313].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[313].offset = 0;
	    constants_info_[313].data_size = 204800;
	    constants_info_[313].from_folded = false;
	    constants_info_[313].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[313].shape = {1280, 40};
	    constants_info_[313].stride = {40, 1};
	    constants_info_[313].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[313].original_fqn = "model.audio_tower.layers.11.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[314].name = "model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[314].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[314].offset = 0;
	    constants_info_[314].data_size = 51200;
	    constants_info_[314].from_folded = false;
	    constants_info_[314].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[314].shape = {1280, 40};
	    constants_info_[314].stride = {40, 1};
	    constants_info_[314].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[314].original_fqn = "model.audio_tower.layers.11.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[315].name = "model_audio_tower_layers_11_self_attn_out_proj_bias";
	    constants_info_[315].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[315].offset = 0;
	    constants_info_[315].data_size = 5120;
	    constants_info_[315].from_folded = false;
	    constants_info_[315].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[315].shape = {1280};
	    constants_info_[315].stride = {1};
	    constants_info_[315].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[315].original_fqn = "model.audio_tower.layers.11.self_attn.out_proj.bias";
	    constants_info_[316].name = "model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[316].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[316].offset = 0;
	    constants_info_[316].data_size = 1638400;
	    constants_info_[316].from_folded = false;
	    constants_info_[316].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[316].shape = {1280, 1280};
	    constants_info_[316].stride = {1280, 1};
	    constants_info_[316].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[316].original_fqn = "model.audio_tower.layers.11.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[317].name = "model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[317].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[317].offset = 0;
	    constants_info_[317].data_size = 204800;
	    constants_info_[317].from_folded = false;
	    constants_info_[317].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[317].shape = {1280, 40};
	    constants_info_[317].stride = {40, 1};
	    constants_info_[317].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[317].original_fqn = "model.audio_tower.layers.11.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[318].name = "model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[318].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[318].offset = 0;
	    constants_info_[318].data_size = 51200;
	    constants_info_[318].from_folded = false;
	    constants_info_[318].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[318].shape = {1280, 40};
	    constants_info_[318].stride = {40, 1};
	    constants_info_[318].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[318].original_fqn = "model.audio_tower.layers.11.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[319].name = "model_audio_tower_layers_11_final_layer_norm_weight";
	    constants_info_[319].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[319].offset = 0;
	    constants_info_[319].data_size = 5120;
	    constants_info_[319].from_folded = false;
	    constants_info_[319].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[319].shape = {1280};
	    constants_info_[319].stride = {1};
	    constants_info_[319].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[319].original_fqn = "model.audio_tower.layers.11.final_layer_norm.weight";
	    constants_info_[320].name = "model_audio_tower_layers_11_final_layer_norm_bias";
	    constants_info_[320].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[320].offset = 0;
	    constants_info_[320].data_size = 5120;
	    constants_info_[320].from_folded = false;
	    constants_info_[320].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[320].shape = {1280};
	    constants_info_[320].stride = {1};
	    constants_info_[320].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[320].original_fqn = "model.audio_tower.layers.11.final_layer_norm.bias";
	    constants_info_[321].name = "model_audio_tower_layers_11_fc1_bias";
	    constants_info_[321].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[321].offset = 0;
	    constants_info_[321].data_size = 20480;
	    constants_info_[321].from_folded = false;
	    constants_info_[321].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[321].shape = {5120};
	    constants_info_[321].stride = {1};
	    constants_info_[321].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[321].original_fqn = "model.audio_tower.layers.11.fc1.bias";
	    constants_info_[322].name = "model_audio_tower_layers_11_fc1_parametrizations_weight_original0";
	    constants_info_[322].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[322].offset = 0;
	    constants_info_[322].data_size = 6553600;
	    constants_info_[322].from_folded = false;
	    constants_info_[322].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[322].shape = {5120, 1280};
	    constants_info_[322].stride = {1280, 1};
	    constants_info_[322].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[322].original_fqn = "model.audio_tower.layers.11.fc1.parametrizations.weight.original0";
	    constants_info_[323].name = "model_audio_tower_layers_11_fc1_parametrizations_weight_original1";
	    constants_info_[323].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[323].offset = 0;
	    constants_info_[323].data_size = 819200;
	    constants_info_[323].from_folded = false;
	    constants_info_[323].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[323].shape = {5120, 40};
	    constants_info_[323].stride = {40, 1};
	    constants_info_[323].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[323].original_fqn = "model.audio_tower.layers.11.fc1.parametrizations.weight.original1";
	    constants_info_[324].name = "model_audio_tower_layers_11_fc1_parametrizations_weight_original2";
	    constants_info_[324].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[324].offset = 0;
	    constants_info_[324].data_size = 204800;
	    constants_info_[324].from_folded = false;
	    constants_info_[324].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[324].shape = {5120, 40};
	    constants_info_[324].stride = {40, 1};
	    constants_info_[324].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[324].original_fqn = "model.audio_tower.layers.11.fc1.parametrizations.weight.original2";
	    constants_info_[325].name = "model_audio_tower_layers_11_fc2_bias";
	    constants_info_[325].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[325].offset = 0;
	    constants_info_[325].data_size = 5120;
	    constants_info_[325].from_folded = false;
	    constants_info_[325].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[325].shape = {1280};
	    constants_info_[325].stride = {1};
	    constants_info_[325].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[325].original_fqn = "model.audio_tower.layers.11.fc2.bias";
	    constants_info_[326].name = "model_audio_tower_layers_11_fc2_parametrizations_weight_original0";
	    constants_info_[326].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[326].offset = 0;
	    constants_info_[326].data_size = 6553600;
	    constants_info_[326].from_folded = false;
	    constants_info_[326].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[326].shape = {1280, 5120};
	    constants_info_[326].stride = {5120, 1};
	    constants_info_[326].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[326].original_fqn = "model.audio_tower.layers.11.fc2.parametrizations.weight.original0";
	    constants_info_[327].name = "model_audio_tower_layers_11_fc2_parametrizations_weight_original1";
	    constants_info_[327].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[327].offset = 0;
	    constants_info_[327].data_size = 819200;
	    constants_info_[327].from_folded = false;
	    constants_info_[327].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[327].shape = {1280, 160};
	    constants_info_[327].stride = {160, 1};
	    constants_info_[327].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[327].original_fqn = "model.audio_tower.layers.11.fc2.parametrizations.weight.original1";
	    constants_info_[328].name = "model_audio_tower_layers_11_fc2_parametrizations_weight_original2";
	    constants_info_[328].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[328].offset = 0;
	    constants_info_[328].data_size = 204800;
	    constants_info_[328].from_folded = false;
	    constants_info_[328].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[328].shape = {1280, 160};
	    constants_info_[328].stride = {160, 1};
	    constants_info_[328].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[328].original_fqn = "model.audio_tower.layers.11.fc2.parametrizations.weight.original2";
	    constants_info_[329].name = "model_audio_tower_layers_12_self_attn_layer_norm_weight";
	    constants_info_[329].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[329].offset = 0;
	    constants_info_[329].data_size = 5120;
	    constants_info_[329].from_folded = false;
	    constants_info_[329].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[329].shape = {1280};
	    constants_info_[329].stride = {1};
	    constants_info_[329].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[329].original_fqn = "model.audio_tower.layers.12.self_attn_layer_norm.weight";
	    constants_info_[330].name = "model_audio_tower_layers_12_self_attn_layer_norm_bias";
	    constants_info_[330].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[330].offset = 0;
	    constants_info_[330].data_size = 5120;
	    constants_info_[330].from_folded = false;
	    constants_info_[330].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[330].shape = {1280};
	    constants_info_[330].stride = {1};
	    constants_info_[330].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[330].original_fqn = "model.audio_tower.layers.12.self_attn_layer_norm.bias";
	    constants_info_[331].name = "model_audio_tower_layers_12_self_attn_q_proj_bias";
	    constants_info_[331].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[331].offset = 0;
	    constants_info_[331].data_size = 5120;
	    constants_info_[331].from_folded = false;
	    constants_info_[331].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[331].shape = {1280};
	    constants_info_[331].stride = {1};
	    constants_info_[331].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[331].original_fqn = "model.audio_tower.layers.12.self_attn.q_proj.bias";
	    constants_info_[332].name = "model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[332].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[332].offset = 0;
	    constants_info_[332].data_size = 1638400;
	    constants_info_[332].from_folded = false;
	    constants_info_[332].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[332].shape = {1280, 1280};
	    constants_info_[332].stride = {1280, 1};
	    constants_info_[332].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[332].original_fqn = "model.audio_tower.layers.12.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[333].name = "model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[333].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[333].offset = 0;
	    constants_info_[333].data_size = 204800;
	    constants_info_[333].from_folded = false;
	    constants_info_[333].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[333].shape = {1280, 40};
	    constants_info_[333].stride = {40, 1};
	    constants_info_[333].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[333].original_fqn = "model.audio_tower.layers.12.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[334].name = "model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[334].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[334].offset = 0;
	    constants_info_[334].data_size = 51200;
	    constants_info_[334].from_folded = false;
	    constants_info_[334].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[334].shape = {1280, 40};
	    constants_info_[334].stride = {40, 1};
	    constants_info_[334].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[334].original_fqn = "model.audio_tower.layers.12.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[335].name = "model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[335].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[335].offset = 0;
	    constants_info_[335].data_size = 1638400;
	    constants_info_[335].from_folded = false;
	    constants_info_[335].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[335].shape = {1280, 1280};
	    constants_info_[335].stride = {1280, 1};
	    constants_info_[335].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[335].original_fqn = "model.audio_tower.layers.12.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[336].name = "model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[336].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[336].offset = 0;
	    constants_info_[336].data_size = 204800;
	    constants_info_[336].from_folded = false;
	    constants_info_[336].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[336].shape = {1280, 40};
	    constants_info_[336].stride = {40, 1};
	    constants_info_[336].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[336].original_fqn = "model.audio_tower.layers.12.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[337].name = "model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[337].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[337].offset = 0;
	    constants_info_[337].data_size = 51200;
	    constants_info_[337].from_folded = false;
	    constants_info_[337].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[337].shape = {1280, 40};
	    constants_info_[337].stride = {40, 1};
	    constants_info_[337].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[337].original_fqn = "model.audio_tower.layers.12.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[338].name = "model_audio_tower_layers_12_self_attn_v_proj_bias";
	    constants_info_[338].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[338].offset = 0;
	    constants_info_[338].data_size = 5120;
	    constants_info_[338].from_folded = false;
	    constants_info_[338].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[338].shape = {1280};
	    constants_info_[338].stride = {1};
	    constants_info_[338].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[338].original_fqn = "model.audio_tower.layers.12.self_attn.v_proj.bias";
	    constants_info_[339].name = "model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[339].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[339].offset = 0;
	    constants_info_[339].data_size = 1638400;
	    constants_info_[339].from_folded = false;
	    constants_info_[339].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[339].shape = {1280, 1280};
	    constants_info_[339].stride = {1280, 1};
	    constants_info_[339].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[339].original_fqn = "model.audio_tower.layers.12.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[340].name = "model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[340].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[340].offset = 0;
	    constants_info_[340].data_size = 204800;
	    constants_info_[340].from_folded = false;
	    constants_info_[340].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[340].shape = {1280, 40};
	    constants_info_[340].stride = {40, 1};
	    constants_info_[340].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[340].original_fqn = "model.audio_tower.layers.12.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[341].name = "model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[341].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[341].offset = 0;
	    constants_info_[341].data_size = 51200;
	    constants_info_[341].from_folded = false;
	    constants_info_[341].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[341].shape = {1280, 40};
	    constants_info_[341].stride = {40, 1};
	    constants_info_[341].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[341].original_fqn = "model.audio_tower.layers.12.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[342].name = "model_audio_tower_layers_12_self_attn_out_proj_bias";
	    constants_info_[342].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[342].offset = 0;
	    constants_info_[342].data_size = 5120;
	    constants_info_[342].from_folded = false;
	    constants_info_[342].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[342].shape = {1280};
	    constants_info_[342].stride = {1};
	    constants_info_[342].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[342].original_fqn = "model.audio_tower.layers.12.self_attn.out_proj.bias";
	    constants_info_[343].name = "model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[343].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[343].offset = 0;
	    constants_info_[343].data_size = 1638400;
	    constants_info_[343].from_folded = false;
	    constants_info_[343].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[343].shape = {1280, 1280};
	    constants_info_[343].stride = {1280, 1};
	    constants_info_[343].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[343].original_fqn = "model.audio_tower.layers.12.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[344].name = "model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[344].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[344].offset = 0;
	    constants_info_[344].data_size = 204800;
	    constants_info_[344].from_folded = false;
	    constants_info_[344].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[344].shape = {1280, 40};
	    constants_info_[344].stride = {40, 1};
	    constants_info_[344].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[344].original_fqn = "model.audio_tower.layers.12.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[345].name = "model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[345].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[345].offset = 0;
	    constants_info_[345].data_size = 51200;
	    constants_info_[345].from_folded = false;
	    constants_info_[345].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[345].shape = {1280, 40};
	    constants_info_[345].stride = {40, 1};
	    constants_info_[345].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[345].original_fqn = "model.audio_tower.layers.12.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[346].name = "model_audio_tower_layers_12_final_layer_norm_weight";
	    constants_info_[346].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[346].offset = 0;
	    constants_info_[346].data_size = 5120;
	    constants_info_[346].from_folded = false;
	    constants_info_[346].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[346].shape = {1280};
	    constants_info_[346].stride = {1};
	    constants_info_[346].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[346].original_fqn = "model.audio_tower.layers.12.final_layer_norm.weight";
	    constants_info_[347].name = "model_audio_tower_layers_12_final_layer_norm_bias";
	    constants_info_[347].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[347].offset = 0;
	    constants_info_[347].data_size = 5120;
	    constants_info_[347].from_folded = false;
	    constants_info_[347].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[347].shape = {1280};
	    constants_info_[347].stride = {1};
	    constants_info_[347].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[347].original_fqn = "model.audio_tower.layers.12.final_layer_norm.bias";
	    constants_info_[348].name = "model_audio_tower_layers_12_fc1_bias";
	    constants_info_[348].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[348].offset = 0;
	    constants_info_[348].data_size = 20480;
	    constants_info_[348].from_folded = false;
	    constants_info_[348].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[348].shape = {5120};
	    constants_info_[348].stride = {1};
	    constants_info_[348].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[348].original_fqn = "model.audio_tower.layers.12.fc1.bias";
	    constants_info_[349].name = "model_audio_tower_layers_12_fc1_parametrizations_weight_original0";
	    constants_info_[349].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[349].offset = 0;
	    constants_info_[349].data_size = 6553600;
	    constants_info_[349].from_folded = false;
	    constants_info_[349].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[349].shape = {5120, 1280};
	    constants_info_[349].stride = {1280, 1};
	    constants_info_[349].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[349].original_fqn = "model.audio_tower.layers.12.fc1.parametrizations.weight.original0";
	    constants_info_[350].name = "model_audio_tower_layers_12_fc1_parametrizations_weight_original1";
	    constants_info_[350].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[350].offset = 0;
	    constants_info_[350].data_size = 819200;
	    constants_info_[350].from_folded = false;
	    constants_info_[350].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[350].shape = {5120, 40};
	    constants_info_[350].stride = {40, 1};
	    constants_info_[350].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[350].original_fqn = "model.audio_tower.layers.12.fc1.parametrizations.weight.original1";
	    constants_info_[351].name = "model_audio_tower_layers_12_fc1_parametrizations_weight_original2";
	    constants_info_[351].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[351].offset = 0;
	    constants_info_[351].data_size = 204800;
	    constants_info_[351].from_folded = false;
	    constants_info_[351].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[351].shape = {5120, 40};
	    constants_info_[351].stride = {40, 1};
	    constants_info_[351].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[351].original_fqn = "model.audio_tower.layers.12.fc1.parametrizations.weight.original2";
	    constants_info_[352].name = "model_audio_tower_layers_12_fc2_bias";
	    constants_info_[352].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[352].offset = 0;
	    constants_info_[352].data_size = 5120;
	    constants_info_[352].from_folded = false;
	    constants_info_[352].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[352].shape = {1280};
	    constants_info_[352].stride = {1};
	    constants_info_[352].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[352].original_fqn = "model.audio_tower.layers.12.fc2.bias";
	    constants_info_[353].name = "model_audio_tower_layers_12_fc2_parametrizations_weight_original0";
	    constants_info_[353].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[353].offset = 0;
	    constants_info_[353].data_size = 6553600;
	    constants_info_[353].from_folded = false;
	    constants_info_[353].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[353].shape = {1280, 5120};
	    constants_info_[353].stride = {5120, 1};
	    constants_info_[353].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[353].original_fqn = "model.audio_tower.layers.12.fc2.parametrizations.weight.original0";
	    constants_info_[354].name = "model_audio_tower_layers_12_fc2_parametrizations_weight_original1";
	    constants_info_[354].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[354].offset = 0;
	    constants_info_[354].data_size = 819200;
	    constants_info_[354].from_folded = false;
	    constants_info_[354].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[354].shape = {1280, 160};
	    constants_info_[354].stride = {160, 1};
	    constants_info_[354].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[354].original_fqn = "model.audio_tower.layers.12.fc2.parametrizations.weight.original1";
	    constants_info_[355].name = "model_audio_tower_layers_12_fc2_parametrizations_weight_original2";
	    constants_info_[355].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[355].offset = 0;
	    constants_info_[355].data_size = 204800;
	    constants_info_[355].from_folded = false;
	    constants_info_[355].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[355].shape = {1280, 160};
	    constants_info_[355].stride = {160, 1};
	    constants_info_[355].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[355].original_fqn = "model.audio_tower.layers.12.fc2.parametrizations.weight.original2";
	    constants_info_[356].name = "model_audio_tower_layers_13_self_attn_layer_norm_weight";
	    constants_info_[356].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[356].offset = 0;
	    constants_info_[356].data_size = 5120;
	    constants_info_[356].from_folded = false;
	    constants_info_[356].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[356].shape = {1280};
	    constants_info_[356].stride = {1};
	    constants_info_[356].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[356].original_fqn = "model.audio_tower.layers.13.self_attn_layer_norm.weight";
	    constants_info_[357].name = "model_audio_tower_layers_13_self_attn_layer_norm_bias";
	    constants_info_[357].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[357].offset = 0;
	    constants_info_[357].data_size = 5120;
	    constants_info_[357].from_folded = false;
	    constants_info_[357].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[357].shape = {1280};
	    constants_info_[357].stride = {1};
	    constants_info_[357].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[357].original_fqn = "model.audio_tower.layers.13.self_attn_layer_norm.bias";
	    constants_info_[358].name = "model_audio_tower_layers_13_self_attn_q_proj_bias";
	    constants_info_[358].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[358].offset = 0;
	    constants_info_[358].data_size = 5120;
	    constants_info_[358].from_folded = false;
	    constants_info_[358].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[358].shape = {1280};
	    constants_info_[358].stride = {1};
	    constants_info_[358].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[358].original_fqn = "model.audio_tower.layers.13.self_attn.q_proj.bias";
	    constants_info_[359].name = "model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[359].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[359].offset = 0;
	    constants_info_[359].data_size = 1638400;
	    constants_info_[359].from_folded = false;
	    constants_info_[359].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[359].shape = {1280, 1280};
	    constants_info_[359].stride = {1280, 1};
	    constants_info_[359].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[359].original_fqn = "model.audio_tower.layers.13.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[360].name = "model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[360].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[360].offset = 0;
	    constants_info_[360].data_size = 204800;
	    constants_info_[360].from_folded = false;
	    constants_info_[360].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[360].shape = {1280, 40};
	    constants_info_[360].stride = {40, 1};
	    constants_info_[360].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[360].original_fqn = "model.audio_tower.layers.13.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[361].name = "model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[361].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[361].offset = 0;
	    constants_info_[361].data_size = 51200;
	    constants_info_[361].from_folded = false;
	    constants_info_[361].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[361].shape = {1280, 40};
	    constants_info_[361].stride = {40, 1};
	    constants_info_[361].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[361].original_fqn = "model.audio_tower.layers.13.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[362].name = "model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[362].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[362].offset = 0;
	    constants_info_[362].data_size = 1638400;
	    constants_info_[362].from_folded = false;
	    constants_info_[362].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[362].shape = {1280, 1280};
	    constants_info_[362].stride = {1280, 1};
	    constants_info_[362].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[362].original_fqn = "model.audio_tower.layers.13.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[363].name = "model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[363].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[363].offset = 0;
	    constants_info_[363].data_size = 204800;
	    constants_info_[363].from_folded = false;
	    constants_info_[363].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[363].shape = {1280, 40};
	    constants_info_[363].stride = {40, 1};
	    constants_info_[363].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[363].original_fqn = "model.audio_tower.layers.13.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[364].name = "model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[364].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[364].offset = 0;
	    constants_info_[364].data_size = 51200;
	    constants_info_[364].from_folded = false;
	    constants_info_[364].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[364].shape = {1280, 40};
	    constants_info_[364].stride = {40, 1};
	    constants_info_[364].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[364].original_fqn = "model.audio_tower.layers.13.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[365].name = "model_audio_tower_layers_13_self_attn_v_proj_bias";
	    constants_info_[365].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[365].offset = 0;
	    constants_info_[365].data_size = 5120;
	    constants_info_[365].from_folded = false;
	    constants_info_[365].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[365].shape = {1280};
	    constants_info_[365].stride = {1};
	    constants_info_[365].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[365].original_fqn = "model.audio_tower.layers.13.self_attn.v_proj.bias";
	    constants_info_[366].name = "model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[366].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[366].offset = 0;
	    constants_info_[366].data_size = 1638400;
	    constants_info_[366].from_folded = false;
	    constants_info_[366].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[366].shape = {1280, 1280};
	    constants_info_[366].stride = {1280, 1};
	    constants_info_[366].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[366].original_fqn = "model.audio_tower.layers.13.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[367].name = "model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[367].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[367].offset = 0;
	    constants_info_[367].data_size = 204800;
	    constants_info_[367].from_folded = false;
	    constants_info_[367].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[367].shape = {1280, 40};
	    constants_info_[367].stride = {40, 1};
	    constants_info_[367].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[367].original_fqn = "model.audio_tower.layers.13.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[368].name = "model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[368].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[368].offset = 0;
	    constants_info_[368].data_size = 51200;
	    constants_info_[368].from_folded = false;
	    constants_info_[368].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[368].shape = {1280, 40};
	    constants_info_[368].stride = {40, 1};
	    constants_info_[368].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[368].original_fqn = "model.audio_tower.layers.13.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[369].name = "model_audio_tower_layers_13_self_attn_out_proj_bias";
	    constants_info_[369].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[369].offset = 0;
	    constants_info_[369].data_size = 5120;
	    constants_info_[369].from_folded = false;
	    constants_info_[369].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[369].shape = {1280};
	    constants_info_[369].stride = {1};
	    constants_info_[369].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[369].original_fqn = "model.audio_tower.layers.13.self_attn.out_proj.bias";
	    constants_info_[370].name = "model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[370].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[370].offset = 0;
	    constants_info_[370].data_size = 1638400;
	    constants_info_[370].from_folded = false;
	    constants_info_[370].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[370].shape = {1280, 1280};
	    constants_info_[370].stride = {1280, 1};
	    constants_info_[370].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[370].original_fqn = "model.audio_tower.layers.13.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[371].name = "model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[371].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[371].offset = 0;
	    constants_info_[371].data_size = 204800;
	    constants_info_[371].from_folded = false;
	    constants_info_[371].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[371].shape = {1280, 40};
	    constants_info_[371].stride = {40, 1};
	    constants_info_[371].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[371].original_fqn = "model.audio_tower.layers.13.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[372].name = "model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[372].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[372].offset = 0;
	    constants_info_[372].data_size = 51200;
	    constants_info_[372].from_folded = false;
	    constants_info_[372].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[372].shape = {1280, 40};
	    constants_info_[372].stride = {40, 1};
	    constants_info_[372].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[372].original_fqn = "model.audio_tower.layers.13.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[373].name = "model_audio_tower_layers_13_final_layer_norm_weight";
	    constants_info_[373].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[373].offset = 0;
	    constants_info_[373].data_size = 5120;
	    constants_info_[373].from_folded = false;
	    constants_info_[373].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[373].shape = {1280};
	    constants_info_[373].stride = {1};
	    constants_info_[373].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[373].original_fqn = "model.audio_tower.layers.13.final_layer_norm.weight";
	    constants_info_[374].name = "model_audio_tower_layers_13_final_layer_norm_bias";
	    constants_info_[374].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[374].offset = 0;
	    constants_info_[374].data_size = 5120;
	    constants_info_[374].from_folded = false;
	    constants_info_[374].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[374].shape = {1280};
	    constants_info_[374].stride = {1};
	    constants_info_[374].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[374].original_fqn = "model.audio_tower.layers.13.final_layer_norm.bias";
	    constants_info_[375].name = "model_audio_tower_layers_13_fc1_bias";
	    constants_info_[375].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[375].offset = 0;
	    constants_info_[375].data_size = 20480;
	    constants_info_[375].from_folded = false;
	    constants_info_[375].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[375].shape = {5120};
	    constants_info_[375].stride = {1};
	    constants_info_[375].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[375].original_fqn = "model.audio_tower.layers.13.fc1.bias";
	    constants_info_[376].name = "model_audio_tower_layers_13_fc1_parametrizations_weight_original0";
	    constants_info_[376].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[376].offset = 0;
	    constants_info_[376].data_size = 6553600;
	    constants_info_[376].from_folded = false;
	    constants_info_[376].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[376].shape = {5120, 1280};
	    constants_info_[376].stride = {1280, 1};
	    constants_info_[376].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[376].original_fqn = "model.audio_tower.layers.13.fc1.parametrizations.weight.original0";
	    constants_info_[377].name = "model_audio_tower_layers_13_fc1_parametrizations_weight_original1";
	    constants_info_[377].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[377].offset = 0;
	    constants_info_[377].data_size = 819200;
	    constants_info_[377].from_folded = false;
	    constants_info_[377].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[377].shape = {5120, 40};
	    constants_info_[377].stride = {40, 1};
	    constants_info_[377].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[377].original_fqn = "model.audio_tower.layers.13.fc1.parametrizations.weight.original1";
	    constants_info_[378].name = "model_audio_tower_layers_13_fc1_parametrizations_weight_original2";
	    constants_info_[378].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[378].offset = 0;
	    constants_info_[378].data_size = 204800;
	    constants_info_[378].from_folded = false;
	    constants_info_[378].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[378].shape = {5120, 40};
	    constants_info_[378].stride = {40, 1};
	    constants_info_[378].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[378].original_fqn = "model.audio_tower.layers.13.fc1.parametrizations.weight.original2";
	    constants_info_[379].name = "model_audio_tower_layers_13_fc2_bias";
	    constants_info_[379].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[379].offset = 0;
	    constants_info_[379].data_size = 5120;
	    constants_info_[379].from_folded = false;
	    constants_info_[379].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[379].shape = {1280};
	    constants_info_[379].stride = {1};
	    constants_info_[379].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[379].original_fqn = "model.audio_tower.layers.13.fc2.bias";
	    constants_info_[380].name = "model_audio_tower_layers_13_fc2_parametrizations_weight_original0";
	    constants_info_[380].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[380].offset = 0;
	    constants_info_[380].data_size = 6553600;
	    constants_info_[380].from_folded = false;
	    constants_info_[380].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[380].shape = {1280, 5120};
	    constants_info_[380].stride = {5120, 1};
	    constants_info_[380].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[380].original_fqn = "model.audio_tower.layers.13.fc2.parametrizations.weight.original0";
	    constants_info_[381].name = "model_audio_tower_layers_13_fc2_parametrizations_weight_original1";
	    constants_info_[381].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[381].offset = 0;
	    constants_info_[381].data_size = 819200;
	    constants_info_[381].from_folded = false;
	    constants_info_[381].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[381].shape = {1280, 160};
	    constants_info_[381].stride = {160, 1};
	    constants_info_[381].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[381].original_fqn = "model.audio_tower.layers.13.fc2.parametrizations.weight.original1";
	    constants_info_[382].name = "model_audio_tower_layers_13_fc2_parametrizations_weight_original2";
	    constants_info_[382].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[382].offset = 0;
	    constants_info_[382].data_size = 204800;
	    constants_info_[382].from_folded = false;
	    constants_info_[382].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[382].shape = {1280, 160};
	    constants_info_[382].stride = {160, 1};
	    constants_info_[382].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[382].original_fqn = "model.audio_tower.layers.13.fc2.parametrizations.weight.original2";
	    constants_info_[383].name = "model_audio_tower_layers_14_self_attn_layer_norm_weight";
	    constants_info_[383].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[383].offset = 0;
	    constants_info_[383].data_size = 5120;
	    constants_info_[383].from_folded = false;
	    constants_info_[383].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[383].shape = {1280};
	    constants_info_[383].stride = {1};
	    constants_info_[383].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[383].original_fqn = "model.audio_tower.layers.14.self_attn_layer_norm.weight";
	    constants_info_[384].name = "model_audio_tower_layers_14_self_attn_layer_norm_bias";
	    constants_info_[384].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[384].offset = 0;
	    constants_info_[384].data_size = 5120;
	    constants_info_[384].from_folded = false;
	    constants_info_[384].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[384].shape = {1280};
	    constants_info_[384].stride = {1};
	    constants_info_[384].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[384].original_fqn = "model.audio_tower.layers.14.self_attn_layer_norm.bias";
	    constants_info_[385].name = "model_audio_tower_layers_14_self_attn_q_proj_bias";
	    constants_info_[385].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[385].offset = 0;
	    constants_info_[385].data_size = 5120;
	    constants_info_[385].from_folded = false;
	    constants_info_[385].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[385].shape = {1280};
	    constants_info_[385].stride = {1};
	    constants_info_[385].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[385].original_fqn = "model.audio_tower.layers.14.self_attn.q_proj.bias";
	    constants_info_[386].name = "model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[386].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[386].offset = 0;
	    constants_info_[386].data_size = 1638400;
	    constants_info_[386].from_folded = false;
	    constants_info_[386].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[386].shape = {1280, 1280};
	    constants_info_[386].stride = {1280, 1};
	    constants_info_[386].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[386].original_fqn = "model.audio_tower.layers.14.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[387].name = "model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[387].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[387].offset = 0;
	    constants_info_[387].data_size = 204800;
	    constants_info_[387].from_folded = false;
	    constants_info_[387].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[387].shape = {1280, 40};
	    constants_info_[387].stride = {40, 1};
	    constants_info_[387].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[387].original_fqn = "model.audio_tower.layers.14.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[388].name = "model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[388].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[388].offset = 0;
	    constants_info_[388].data_size = 51200;
	    constants_info_[388].from_folded = false;
	    constants_info_[388].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[388].shape = {1280, 40};
	    constants_info_[388].stride = {40, 1};
	    constants_info_[388].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[388].original_fqn = "model.audio_tower.layers.14.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[389].name = "model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[389].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[389].offset = 0;
	    constants_info_[389].data_size = 1638400;
	    constants_info_[389].from_folded = false;
	    constants_info_[389].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[389].shape = {1280, 1280};
	    constants_info_[389].stride = {1280, 1};
	    constants_info_[389].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[389].original_fqn = "model.audio_tower.layers.14.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[390].name = "model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[390].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[390].offset = 0;
	    constants_info_[390].data_size = 204800;
	    constants_info_[390].from_folded = false;
	    constants_info_[390].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[390].shape = {1280, 40};
	    constants_info_[390].stride = {40, 1};
	    constants_info_[390].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[390].original_fqn = "model.audio_tower.layers.14.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[391].name = "model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[391].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[391].offset = 0;
	    constants_info_[391].data_size = 51200;
	    constants_info_[391].from_folded = false;
	    constants_info_[391].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[391].shape = {1280, 40};
	    constants_info_[391].stride = {40, 1};
	    constants_info_[391].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[391].original_fqn = "model.audio_tower.layers.14.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[392].name = "model_audio_tower_layers_14_self_attn_v_proj_bias";
	    constants_info_[392].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[392].offset = 0;
	    constants_info_[392].data_size = 5120;
	    constants_info_[392].from_folded = false;
	    constants_info_[392].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[392].shape = {1280};
	    constants_info_[392].stride = {1};
	    constants_info_[392].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[392].original_fqn = "model.audio_tower.layers.14.self_attn.v_proj.bias";
	    constants_info_[393].name = "model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[393].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[393].offset = 0;
	    constants_info_[393].data_size = 1638400;
	    constants_info_[393].from_folded = false;
	    constants_info_[393].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[393].shape = {1280, 1280};
	    constants_info_[393].stride = {1280, 1};
	    constants_info_[393].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[393].original_fqn = "model.audio_tower.layers.14.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[394].name = "model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[394].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[394].offset = 0;
	    constants_info_[394].data_size = 204800;
	    constants_info_[394].from_folded = false;
	    constants_info_[394].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[394].shape = {1280, 40};
	    constants_info_[394].stride = {40, 1};
	    constants_info_[394].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[394].original_fqn = "model.audio_tower.layers.14.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[395].name = "model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[395].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[395].offset = 0;
	    constants_info_[395].data_size = 51200;
	    constants_info_[395].from_folded = false;
	    constants_info_[395].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[395].shape = {1280, 40};
	    constants_info_[395].stride = {40, 1};
	    constants_info_[395].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[395].original_fqn = "model.audio_tower.layers.14.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[396].name = "model_audio_tower_layers_14_self_attn_out_proj_bias";
	    constants_info_[396].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[396].offset = 0;
	    constants_info_[396].data_size = 5120;
	    constants_info_[396].from_folded = false;
	    constants_info_[396].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[396].shape = {1280};
	    constants_info_[396].stride = {1};
	    constants_info_[396].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[396].original_fqn = "model.audio_tower.layers.14.self_attn.out_proj.bias";
	    constants_info_[397].name = "model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[397].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[397].offset = 0;
	    constants_info_[397].data_size = 1638400;
	    constants_info_[397].from_folded = false;
	    constants_info_[397].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[397].shape = {1280, 1280};
	    constants_info_[397].stride = {1280, 1};
	    constants_info_[397].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[397].original_fqn = "model.audio_tower.layers.14.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[398].name = "model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[398].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[398].offset = 0;
	    constants_info_[398].data_size = 204800;
	    constants_info_[398].from_folded = false;
	    constants_info_[398].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[398].shape = {1280, 40};
	    constants_info_[398].stride = {40, 1};
	    constants_info_[398].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[398].original_fqn = "model.audio_tower.layers.14.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[399].name = "model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[399].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[399].offset = 0;
	    constants_info_[399].data_size = 51200;
	    constants_info_[399].from_folded = false;
	    constants_info_[399].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[399].shape = {1280, 40};
	    constants_info_[399].stride = {40, 1};
	    constants_info_[399].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[399].original_fqn = "model.audio_tower.layers.14.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[400].name = "model_audio_tower_layers_14_final_layer_norm_weight";
	    constants_info_[400].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[400].offset = 0;
	    constants_info_[400].data_size = 5120;
	    constants_info_[400].from_folded = false;
	    constants_info_[400].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[400].shape = {1280};
	    constants_info_[400].stride = {1};
	    constants_info_[400].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[400].original_fqn = "model.audio_tower.layers.14.final_layer_norm.weight";
	    constants_info_[401].name = "model_audio_tower_layers_14_final_layer_norm_bias";
	    constants_info_[401].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[401].offset = 0;
	    constants_info_[401].data_size = 5120;
	    constants_info_[401].from_folded = false;
	    constants_info_[401].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[401].shape = {1280};
	    constants_info_[401].stride = {1};
	    constants_info_[401].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[401].original_fqn = "model.audio_tower.layers.14.final_layer_norm.bias";
	    constants_info_[402].name = "model_audio_tower_layers_14_fc1_bias";
	    constants_info_[402].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[402].offset = 0;
	    constants_info_[402].data_size = 20480;
	    constants_info_[402].from_folded = false;
	    constants_info_[402].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[402].shape = {5120};
	    constants_info_[402].stride = {1};
	    constants_info_[402].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[402].original_fqn = "model.audio_tower.layers.14.fc1.bias";
	    constants_info_[403].name = "model_audio_tower_layers_14_fc1_parametrizations_weight_original0";
	    constants_info_[403].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[403].offset = 0;
	    constants_info_[403].data_size = 6553600;
	    constants_info_[403].from_folded = false;
	    constants_info_[403].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[403].shape = {5120, 1280};
	    constants_info_[403].stride = {1280, 1};
	    constants_info_[403].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[403].original_fqn = "model.audio_tower.layers.14.fc1.parametrizations.weight.original0";
	    constants_info_[404].name = "model_audio_tower_layers_14_fc1_parametrizations_weight_original1";
	    constants_info_[404].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[404].offset = 0;
	    constants_info_[404].data_size = 819200;
	    constants_info_[404].from_folded = false;
	    constants_info_[404].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[404].shape = {5120, 40};
	    constants_info_[404].stride = {40, 1};
	    constants_info_[404].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[404].original_fqn = "model.audio_tower.layers.14.fc1.parametrizations.weight.original1";
	    constants_info_[405].name = "model_audio_tower_layers_14_fc1_parametrizations_weight_original2";
	    constants_info_[405].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[405].offset = 0;
	    constants_info_[405].data_size = 204800;
	    constants_info_[405].from_folded = false;
	    constants_info_[405].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[405].shape = {5120, 40};
	    constants_info_[405].stride = {40, 1};
	    constants_info_[405].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[405].original_fqn = "model.audio_tower.layers.14.fc1.parametrizations.weight.original2";
	    constants_info_[406].name = "model_audio_tower_layers_14_fc2_bias";
	    constants_info_[406].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[406].offset = 0;
	    constants_info_[406].data_size = 5120;
	    constants_info_[406].from_folded = false;
	    constants_info_[406].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[406].shape = {1280};
	    constants_info_[406].stride = {1};
	    constants_info_[406].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[406].original_fqn = "model.audio_tower.layers.14.fc2.bias";
	    constants_info_[407].name = "model_audio_tower_layers_14_fc2_parametrizations_weight_original0";
	    constants_info_[407].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[407].offset = 0;
	    constants_info_[407].data_size = 6553600;
	    constants_info_[407].from_folded = false;
	    constants_info_[407].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[407].shape = {1280, 5120};
	    constants_info_[407].stride = {5120, 1};
	    constants_info_[407].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[407].original_fqn = "model.audio_tower.layers.14.fc2.parametrizations.weight.original0";
	    constants_info_[408].name = "model_audio_tower_layers_14_fc2_parametrizations_weight_original1";
	    constants_info_[408].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[408].offset = 0;
	    constants_info_[408].data_size = 819200;
	    constants_info_[408].from_folded = false;
	    constants_info_[408].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[408].shape = {1280, 160};
	    constants_info_[408].stride = {160, 1};
	    constants_info_[408].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[408].original_fqn = "model.audio_tower.layers.14.fc2.parametrizations.weight.original1";
	    constants_info_[409].name = "model_audio_tower_layers_14_fc2_parametrizations_weight_original2";
	    constants_info_[409].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[409].offset = 0;
	    constants_info_[409].data_size = 204800;
	    constants_info_[409].from_folded = false;
	    constants_info_[409].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[409].shape = {1280, 160};
	    constants_info_[409].stride = {160, 1};
	    constants_info_[409].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[409].original_fqn = "model.audio_tower.layers.14.fc2.parametrizations.weight.original2";
	    constants_info_[410].name = "model_audio_tower_layers_15_self_attn_layer_norm_weight";
	    constants_info_[410].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[410].offset = 0;
	    constants_info_[410].data_size = 5120;
	    constants_info_[410].from_folded = false;
	    constants_info_[410].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[410].shape = {1280};
	    constants_info_[410].stride = {1};
	    constants_info_[410].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[410].original_fqn = "model.audio_tower.layers.15.self_attn_layer_norm.weight";
	    constants_info_[411].name = "model_audio_tower_layers_15_self_attn_layer_norm_bias";
	    constants_info_[411].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[411].offset = 0;
	    constants_info_[411].data_size = 5120;
	    constants_info_[411].from_folded = false;
	    constants_info_[411].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[411].shape = {1280};
	    constants_info_[411].stride = {1};
	    constants_info_[411].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[411].original_fqn = "model.audio_tower.layers.15.self_attn_layer_norm.bias";
	    constants_info_[412].name = "model_audio_tower_layers_15_self_attn_q_proj_bias";
	    constants_info_[412].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[412].offset = 0;
	    constants_info_[412].data_size = 5120;
	    constants_info_[412].from_folded = false;
	    constants_info_[412].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[412].shape = {1280};
	    constants_info_[412].stride = {1};
	    constants_info_[412].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[412].original_fqn = "model.audio_tower.layers.15.self_attn.q_proj.bias";
	    constants_info_[413].name = "model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[413].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[413].offset = 0;
	    constants_info_[413].data_size = 1638400;
	    constants_info_[413].from_folded = false;
	    constants_info_[413].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[413].shape = {1280, 1280};
	    constants_info_[413].stride = {1280, 1};
	    constants_info_[413].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[413].original_fqn = "model.audio_tower.layers.15.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[414].name = "model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[414].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[414].offset = 0;
	    constants_info_[414].data_size = 204800;
	    constants_info_[414].from_folded = false;
	    constants_info_[414].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[414].shape = {1280, 40};
	    constants_info_[414].stride = {40, 1};
	    constants_info_[414].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[414].original_fqn = "model.audio_tower.layers.15.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[415].name = "model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[415].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[415].offset = 0;
	    constants_info_[415].data_size = 51200;
	    constants_info_[415].from_folded = false;
	    constants_info_[415].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[415].shape = {1280, 40};
	    constants_info_[415].stride = {40, 1};
	    constants_info_[415].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[415].original_fqn = "model.audio_tower.layers.15.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[416].name = "model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[416].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[416].offset = 0;
	    constants_info_[416].data_size = 1638400;
	    constants_info_[416].from_folded = false;
	    constants_info_[416].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[416].shape = {1280, 1280};
	    constants_info_[416].stride = {1280, 1};
	    constants_info_[416].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[416].original_fqn = "model.audio_tower.layers.15.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[417].name = "model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[417].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[417].offset = 0;
	    constants_info_[417].data_size = 204800;
	    constants_info_[417].from_folded = false;
	    constants_info_[417].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[417].shape = {1280, 40};
	    constants_info_[417].stride = {40, 1};
	    constants_info_[417].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[417].original_fqn = "model.audio_tower.layers.15.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[418].name = "model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[418].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[418].offset = 0;
	    constants_info_[418].data_size = 51200;
	    constants_info_[418].from_folded = false;
	    constants_info_[418].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[418].shape = {1280, 40};
	    constants_info_[418].stride = {40, 1};
	    constants_info_[418].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[418].original_fqn = "model.audio_tower.layers.15.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[419].name = "model_audio_tower_layers_15_self_attn_v_proj_bias";
	    constants_info_[419].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[419].offset = 0;
	    constants_info_[419].data_size = 5120;
	    constants_info_[419].from_folded = false;
	    constants_info_[419].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[419].shape = {1280};
	    constants_info_[419].stride = {1};
	    constants_info_[419].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[419].original_fqn = "model.audio_tower.layers.15.self_attn.v_proj.bias";
	    constants_info_[420].name = "model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[420].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[420].offset = 0;
	    constants_info_[420].data_size = 1638400;
	    constants_info_[420].from_folded = false;
	    constants_info_[420].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[420].shape = {1280, 1280};
	    constants_info_[420].stride = {1280, 1};
	    constants_info_[420].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[420].original_fqn = "model.audio_tower.layers.15.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[421].name = "model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[421].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[421].offset = 0;
	    constants_info_[421].data_size = 204800;
	    constants_info_[421].from_folded = false;
	    constants_info_[421].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[421].shape = {1280, 40};
	    constants_info_[421].stride = {40, 1};
	    constants_info_[421].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[421].original_fqn = "model.audio_tower.layers.15.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[422].name = "model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[422].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[422].offset = 0;
	    constants_info_[422].data_size = 51200;
	    constants_info_[422].from_folded = false;
	    constants_info_[422].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[422].shape = {1280, 40};
	    constants_info_[422].stride = {40, 1};
	    constants_info_[422].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[422].original_fqn = "model.audio_tower.layers.15.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[423].name = "model_audio_tower_layers_15_self_attn_out_proj_bias";
	    constants_info_[423].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[423].offset = 0;
	    constants_info_[423].data_size = 5120;
	    constants_info_[423].from_folded = false;
	    constants_info_[423].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[423].shape = {1280};
	    constants_info_[423].stride = {1};
	    constants_info_[423].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[423].original_fqn = "model.audio_tower.layers.15.self_attn.out_proj.bias";
	    constants_info_[424].name = "model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[424].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[424].offset = 0;
	    constants_info_[424].data_size = 1638400;
	    constants_info_[424].from_folded = false;
	    constants_info_[424].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[424].shape = {1280, 1280};
	    constants_info_[424].stride = {1280, 1};
	    constants_info_[424].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[424].original_fqn = "model.audio_tower.layers.15.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[425].name = "model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[425].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[425].offset = 0;
	    constants_info_[425].data_size = 204800;
	    constants_info_[425].from_folded = false;
	    constants_info_[425].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[425].shape = {1280, 40};
	    constants_info_[425].stride = {40, 1};
	    constants_info_[425].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[425].original_fqn = "model.audio_tower.layers.15.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[426].name = "model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[426].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[426].offset = 0;
	    constants_info_[426].data_size = 51200;
	    constants_info_[426].from_folded = false;
	    constants_info_[426].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[426].shape = {1280, 40};
	    constants_info_[426].stride = {40, 1};
	    constants_info_[426].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[426].original_fqn = "model.audio_tower.layers.15.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[427].name = "model_audio_tower_layers_15_final_layer_norm_weight";
	    constants_info_[427].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[427].offset = 0;
	    constants_info_[427].data_size = 5120;
	    constants_info_[427].from_folded = false;
	    constants_info_[427].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[427].shape = {1280};
	    constants_info_[427].stride = {1};
	    constants_info_[427].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[427].original_fqn = "model.audio_tower.layers.15.final_layer_norm.weight";
	    constants_info_[428].name = "model_audio_tower_layers_15_final_layer_norm_bias";
	    constants_info_[428].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[428].offset = 0;
	    constants_info_[428].data_size = 5120;
	    constants_info_[428].from_folded = false;
	    constants_info_[428].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[428].shape = {1280};
	    constants_info_[428].stride = {1};
	    constants_info_[428].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[428].original_fqn = "model.audio_tower.layers.15.final_layer_norm.bias";
	    constants_info_[429].name = "model_audio_tower_layers_15_fc1_bias";
	    constants_info_[429].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[429].offset = 0;
	    constants_info_[429].data_size = 20480;
	    constants_info_[429].from_folded = false;
	    constants_info_[429].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[429].shape = {5120};
	    constants_info_[429].stride = {1};
	    constants_info_[429].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[429].original_fqn = "model.audio_tower.layers.15.fc1.bias";
	    constants_info_[430].name = "model_audio_tower_layers_15_fc1_parametrizations_weight_original0";
	    constants_info_[430].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[430].offset = 0;
	    constants_info_[430].data_size = 6553600;
	    constants_info_[430].from_folded = false;
	    constants_info_[430].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[430].shape = {5120, 1280};
	    constants_info_[430].stride = {1280, 1};
	    constants_info_[430].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[430].original_fqn = "model.audio_tower.layers.15.fc1.parametrizations.weight.original0";
	    constants_info_[431].name = "model_audio_tower_layers_15_fc1_parametrizations_weight_original1";
	    constants_info_[431].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[431].offset = 0;
	    constants_info_[431].data_size = 819200;
	    constants_info_[431].from_folded = false;
	    constants_info_[431].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[431].shape = {5120, 40};
	    constants_info_[431].stride = {40, 1};
	    constants_info_[431].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[431].original_fqn = "model.audio_tower.layers.15.fc1.parametrizations.weight.original1";
	    constants_info_[432].name = "model_audio_tower_layers_15_fc1_parametrizations_weight_original2";
	    constants_info_[432].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[432].offset = 0;
	    constants_info_[432].data_size = 204800;
	    constants_info_[432].from_folded = false;
	    constants_info_[432].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[432].shape = {5120, 40};
	    constants_info_[432].stride = {40, 1};
	    constants_info_[432].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[432].original_fqn = "model.audio_tower.layers.15.fc1.parametrizations.weight.original2";
	    constants_info_[433].name = "model_audio_tower_layers_15_fc2_bias";
	    constants_info_[433].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[433].offset = 0;
	    constants_info_[433].data_size = 5120;
	    constants_info_[433].from_folded = false;
	    constants_info_[433].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[433].shape = {1280};
	    constants_info_[433].stride = {1};
	    constants_info_[433].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[433].original_fqn = "model.audio_tower.layers.15.fc2.bias";
	    constants_info_[434].name = "model_audio_tower_layers_15_fc2_parametrizations_weight_original0";
	    constants_info_[434].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[434].offset = 0;
	    constants_info_[434].data_size = 6553600;
	    constants_info_[434].from_folded = false;
	    constants_info_[434].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[434].shape = {1280, 5120};
	    constants_info_[434].stride = {5120, 1};
	    constants_info_[434].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[434].original_fqn = "model.audio_tower.layers.15.fc2.parametrizations.weight.original0";
	    constants_info_[435].name = "model_audio_tower_layers_15_fc2_parametrizations_weight_original1";
	    constants_info_[435].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[435].offset = 0;
	    constants_info_[435].data_size = 819200;
	    constants_info_[435].from_folded = false;
	    constants_info_[435].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[435].shape = {1280, 160};
	    constants_info_[435].stride = {160, 1};
	    constants_info_[435].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[435].original_fqn = "model.audio_tower.layers.15.fc2.parametrizations.weight.original1";
	    constants_info_[436].name = "model_audio_tower_layers_15_fc2_parametrizations_weight_original2";
	    constants_info_[436].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[436].offset = 0;
	    constants_info_[436].data_size = 204800;
	    constants_info_[436].from_folded = false;
	    constants_info_[436].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[436].shape = {1280, 160};
	    constants_info_[436].stride = {160, 1};
	    constants_info_[436].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[436].original_fqn = "model.audio_tower.layers.15.fc2.parametrizations.weight.original2";
	    constants_info_[437].name = "model_audio_tower_layers_16_self_attn_layer_norm_weight";
	    constants_info_[437].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[437].offset = 0;
	    constants_info_[437].data_size = 5120;
	    constants_info_[437].from_folded = false;
	    constants_info_[437].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[437].shape = {1280};
	    constants_info_[437].stride = {1};
	    constants_info_[437].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[437].original_fqn = "model.audio_tower.layers.16.self_attn_layer_norm.weight";
	    constants_info_[438].name = "model_audio_tower_layers_16_self_attn_layer_norm_bias";
	    constants_info_[438].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[438].offset = 0;
	    constants_info_[438].data_size = 5120;
	    constants_info_[438].from_folded = false;
	    constants_info_[438].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[438].shape = {1280};
	    constants_info_[438].stride = {1};
	    constants_info_[438].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[438].original_fqn = "model.audio_tower.layers.16.self_attn_layer_norm.bias";
	    constants_info_[439].name = "model_audio_tower_layers_16_self_attn_q_proj_bias";
	    constants_info_[439].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[439].offset = 0;
	    constants_info_[439].data_size = 5120;
	    constants_info_[439].from_folded = false;
	    constants_info_[439].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[439].shape = {1280};
	    constants_info_[439].stride = {1};
	    constants_info_[439].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[439].original_fqn = "model.audio_tower.layers.16.self_attn.q_proj.bias";
	    constants_info_[440].name = "model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[440].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[440].offset = 0;
	    constants_info_[440].data_size = 1638400;
	    constants_info_[440].from_folded = false;
	    constants_info_[440].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[440].shape = {1280, 1280};
	    constants_info_[440].stride = {1280, 1};
	    constants_info_[440].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[440].original_fqn = "model.audio_tower.layers.16.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[441].name = "model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[441].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[441].offset = 0;
	    constants_info_[441].data_size = 204800;
	    constants_info_[441].from_folded = false;
	    constants_info_[441].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[441].shape = {1280, 40};
	    constants_info_[441].stride = {40, 1};
	    constants_info_[441].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[441].original_fqn = "model.audio_tower.layers.16.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[442].name = "model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[442].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[442].offset = 0;
	    constants_info_[442].data_size = 51200;
	    constants_info_[442].from_folded = false;
	    constants_info_[442].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[442].shape = {1280, 40};
	    constants_info_[442].stride = {40, 1};
	    constants_info_[442].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[442].original_fqn = "model.audio_tower.layers.16.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[443].name = "model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[443].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[443].offset = 0;
	    constants_info_[443].data_size = 1638400;
	    constants_info_[443].from_folded = false;
	    constants_info_[443].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[443].shape = {1280, 1280};
	    constants_info_[443].stride = {1280, 1};
	    constants_info_[443].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[443].original_fqn = "model.audio_tower.layers.16.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[444].name = "model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[444].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[444].offset = 0;
	    constants_info_[444].data_size = 204800;
	    constants_info_[444].from_folded = false;
	    constants_info_[444].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[444].shape = {1280, 40};
	    constants_info_[444].stride = {40, 1};
	    constants_info_[444].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[444].original_fqn = "model.audio_tower.layers.16.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[445].name = "model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[445].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[445].offset = 0;
	    constants_info_[445].data_size = 51200;
	    constants_info_[445].from_folded = false;
	    constants_info_[445].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[445].shape = {1280, 40};
	    constants_info_[445].stride = {40, 1};
	    constants_info_[445].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[445].original_fqn = "model.audio_tower.layers.16.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[446].name = "model_audio_tower_layers_16_self_attn_v_proj_bias";
	    constants_info_[446].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[446].offset = 0;
	    constants_info_[446].data_size = 5120;
	    constants_info_[446].from_folded = false;
	    constants_info_[446].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[446].shape = {1280};
	    constants_info_[446].stride = {1};
	    constants_info_[446].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[446].original_fqn = "model.audio_tower.layers.16.self_attn.v_proj.bias";
	    constants_info_[447].name = "model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[447].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[447].offset = 0;
	    constants_info_[447].data_size = 1638400;
	    constants_info_[447].from_folded = false;
	    constants_info_[447].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[447].shape = {1280, 1280};
	    constants_info_[447].stride = {1280, 1};
	    constants_info_[447].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[447].original_fqn = "model.audio_tower.layers.16.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[448].name = "model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[448].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[448].offset = 0;
	    constants_info_[448].data_size = 204800;
	    constants_info_[448].from_folded = false;
	    constants_info_[448].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[448].shape = {1280, 40};
	    constants_info_[448].stride = {40, 1};
	    constants_info_[448].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[448].original_fqn = "model.audio_tower.layers.16.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[449].name = "model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[449].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[449].offset = 0;
	    constants_info_[449].data_size = 51200;
	    constants_info_[449].from_folded = false;
	    constants_info_[449].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[449].shape = {1280, 40};
	    constants_info_[449].stride = {40, 1};
	    constants_info_[449].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[449].original_fqn = "model.audio_tower.layers.16.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[450].name = "model_audio_tower_layers_16_self_attn_out_proj_bias";
	    constants_info_[450].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[450].offset = 0;
	    constants_info_[450].data_size = 5120;
	    constants_info_[450].from_folded = false;
	    constants_info_[450].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[450].shape = {1280};
	    constants_info_[450].stride = {1};
	    constants_info_[450].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[450].original_fqn = "model.audio_tower.layers.16.self_attn.out_proj.bias";
	    constants_info_[451].name = "model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[451].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[451].offset = 0;
	    constants_info_[451].data_size = 1638400;
	    constants_info_[451].from_folded = false;
	    constants_info_[451].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[451].shape = {1280, 1280};
	    constants_info_[451].stride = {1280, 1};
	    constants_info_[451].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[451].original_fqn = "model.audio_tower.layers.16.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[452].name = "model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[452].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[452].offset = 0;
	    constants_info_[452].data_size = 204800;
	    constants_info_[452].from_folded = false;
	    constants_info_[452].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[452].shape = {1280, 40};
	    constants_info_[452].stride = {40, 1};
	    constants_info_[452].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[452].original_fqn = "model.audio_tower.layers.16.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[453].name = "model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[453].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[453].offset = 0;
	    constants_info_[453].data_size = 51200;
	    constants_info_[453].from_folded = false;
	    constants_info_[453].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[453].shape = {1280, 40};
	    constants_info_[453].stride = {40, 1};
	    constants_info_[453].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[453].original_fqn = "model.audio_tower.layers.16.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[454].name = "model_audio_tower_layers_16_final_layer_norm_weight";
	    constants_info_[454].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[454].offset = 0;
	    constants_info_[454].data_size = 5120;
	    constants_info_[454].from_folded = false;
	    constants_info_[454].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[454].shape = {1280};
	    constants_info_[454].stride = {1};
	    constants_info_[454].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[454].original_fqn = "model.audio_tower.layers.16.final_layer_norm.weight";
	    constants_info_[455].name = "model_audio_tower_layers_16_final_layer_norm_bias";
	    constants_info_[455].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[455].offset = 0;
	    constants_info_[455].data_size = 5120;
	    constants_info_[455].from_folded = false;
	    constants_info_[455].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[455].shape = {1280};
	    constants_info_[455].stride = {1};
	    constants_info_[455].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[455].original_fqn = "model.audio_tower.layers.16.final_layer_norm.bias";
	    constants_info_[456].name = "model_audio_tower_layers_16_fc1_bias";
	    constants_info_[456].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[456].offset = 0;
	    constants_info_[456].data_size = 20480;
	    constants_info_[456].from_folded = false;
	    constants_info_[456].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[456].shape = {5120};
	    constants_info_[456].stride = {1};
	    constants_info_[456].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[456].original_fqn = "model.audio_tower.layers.16.fc1.bias";
	    constants_info_[457].name = "model_audio_tower_layers_16_fc1_parametrizations_weight_original0";
	    constants_info_[457].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[457].offset = 0;
	    constants_info_[457].data_size = 6553600;
	    constants_info_[457].from_folded = false;
	    constants_info_[457].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[457].shape = {5120, 1280};
	    constants_info_[457].stride = {1280, 1};
	    constants_info_[457].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[457].original_fqn = "model.audio_tower.layers.16.fc1.parametrizations.weight.original0";
	    constants_info_[458].name = "model_audio_tower_layers_16_fc1_parametrizations_weight_original1";
	    constants_info_[458].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[458].offset = 0;
	    constants_info_[458].data_size = 819200;
	    constants_info_[458].from_folded = false;
	    constants_info_[458].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[458].shape = {5120, 40};
	    constants_info_[458].stride = {40, 1};
	    constants_info_[458].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[458].original_fqn = "model.audio_tower.layers.16.fc1.parametrizations.weight.original1";
	    constants_info_[459].name = "model_audio_tower_layers_16_fc1_parametrizations_weight_original2";
	    constants_info_[459].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[459].offset = 0;
	    constants_info_[459].data_size = 204800;
	    constants_info_[459].from_folded = false;
	    constants_info_[459].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[459].shape = {5120, 40};
	    constants_info_[459].stride = {40, 1};
	    constants_info_[459].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[459].original_fqn = "model.audio_tower.layers.16.fc1.parametrizations.weight.original2";
	    constants_info_[460].name = "model_audio_tower_layers_16_fc2_bias";
	    constants_info_[460].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[460].offset = 0;
	    constants_info_[460].data_size = 5120;
	    constants_info_[460].from_folded = false;
	    constants_info_[460].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[460].shape = {1280};
	    constants_info_[460].stride = {1};
	    constants_info_[460].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[460].original_fqn = "model.audio_tower.layers.16.fc2.bias";
	    constants_info_[461].name = "model_audio_tower_layers_16_fc2_parametrizations_weight_original0";
	    constants_info_[461].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[461].offset = 0;
	    constants_info_[461].data_size = 6553600;
	    constants_info_[461].from_folded = false;
	    constants_info_[461].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[461].shape = {1280, 5120};
	    constants_info_[461].stride = {5120, 1};
	    constants_info_[461].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[461].original_fqn = "model.audio_tower.layers.16.fc2.parametrizations.weight.original0";
	    constants_info_[462].name = "model_audio_tower_layers_16_fc2_parametrizations_weight_original1";
	    constants_info_[462].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[462].offset = 0;
	    constants_info_[462].data_size = 819200;
	    constants_info_[462].from_folded = false;
	    constants_info_[462].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[462].shape = {1280, 160};
	    constants_info_[462].stride = {160, 1};
	    constants_info_[462].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[462].original_fqn = "model.audio_tower.layers.16.fc2.parametrizations.weight.original1";
	    constants_info_[463].name = "model_audio_tower_layers_16_fc2_parametrizations_weight_original2";
	    constants_info_[463].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[463].offset = 0;
	    constants_info_[463].data_size = 204800;
	    constants_info_[463].from_folded = false;
	    constants_info_[463].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[463].shape = {1280, 160};
	    constants_info_[463].stride = {160, 1};
	    constants_info_[463].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[463].original_fqn = "model.audio_tower.layers.16.fc2.parametrizations.weight.original2";
	    constants_info_[464].name = "model_audio_tower_layers_17_self_attn_layer_norm_weight";
	    constants_info_[464].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[464].offset = 0;
	    constants_info_[464].data_size = 5120;
	    constants_info_[464].from_folded = false;
	    constants_info_[464].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[464].shape = {1280};
	    constants_info_[464].stride = {1};
	    constants_info_[464].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[464].original_fqn = "model.audio_tower.layers.17.self_attn_layer_norm.weight";
	    constants_info_[465].name = "model_audio_tower_layers_17_self_attn_layer_norm_bias";
	    constants_info_[465].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[465].offset = 0;
	    constants_info_[465].data_size = 5120;
	    constants_info_[465].from_folded = false;
	    constants_info_[465].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[465].shape = {1280};
	    constants_info_[465].stride = {1};
	    constants_info_[465].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[465].original_fqn = "model.audio_tower.layers.17.self_attn_layer_norm.bias";
	    constants_info_[466].name = "model_audio_tower_layers_17_self_attn_q_proj_bias";
	    constants_info_[466].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[466].offset = 0;
	    constants_info_[466].data_size = 5120;
	    constants_info_[466].from_folded = false;
	    constants_info_[466].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[466].shape = {1280};
	    constants_info_[466].stride = {1};
	    constants_info_[466].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[466].original_fqn = "model.audio_tower.layers.17.self_attn.q_proj.bias";
	    constants_info_[467].name = "model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[467].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[467].offset = 0;
	    constants_info_[467].data_size = 1638400;
	    constants_info_[467].from_folded = false;
	    constants_info_[467].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[467].shape = {1280, 1280};
	    constants_info_[467].stride = {1280, 1};
	    constants_info_[467].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[467].original_fqn = "model.audio_tower.layers.17.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[468].name = "model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[468].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[468].offset = 0;
	    constants_info_[468].data_size = 204800;
	    constants_info_[468].from_folded = false;
	    constants_info_[468].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[468].shape = {1280, 40};
	    constants_info_[468].stride = {40, 1};
	    constants_info_[468].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[468].original_fqn = "model.audio_tower.layers.17.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[469].name = "model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[469].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[469].offset = 0;
	    constants_info_[469].data_size = 51200;
	    constants_info_[469].from_folded = false;
	    constants_info_[469].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[469].shape = {1280, 40};
	    constants_info_[469].stride = {40, 1};
	    constants_info_[469].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[469].original_fqn = "model.audio_tower.layers.17.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[470].name = "model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[470].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[470].offset = 0;
	    constants_info_[470].data_size = 1638400;
	    constants_info_[470].from_folded = false;
	    constants_info_[470].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[470].shape = {1280, 1280};
	    constants_info_[470].stride = {1280, 1};
	    constants_info_[470].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[470].original_fqn = "model.audio_tower.layers.17.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[471].name = "model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[471].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[471].offset = 0;
	    constants_info_[471].data_size = 204800;
	    constants_info_[471].from_folded = false;
	    constants_info_[471].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[471].shape = {1280, 40};
	    constants_info_[471].stride = {40, 1};
	    constants_info_[471].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[471].original_fqn = "model.audio_tower.layers.17.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[472].name = "model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[472].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[472].offset = 0;
	    constants_info_[472].data_size = 51200;
	    constants_info_[472].from_folded = false;
	    constants_info_[472].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[472].shape = {1280, 40};
	    constants_info_[472].stride = {40, 1};
	    constants_info_[472].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[472].original_fqn = "model.audio_tower.layers.17.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[473].name = "model_audio_tower_layers_17_self_attn_v_proj_bias";
	    constants_info_[473].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[473].offset = 0;
	    constants_info_[473].data_size = 5120;
	    constants_info_[473].from_folded = false;
	    constants_info_[473].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[473].shape = {1280};
	    constants_info_[473].stride = {1};
	    constants_info_[473].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[473].original_fqn = "model.audio_tower.layers.17.self_attn.v_proj.bias";
	    constants_info_[474].name = "model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[474].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[474].offset = 0;
	    constants_info_[474].data_size = 1638400;
	    constants_info_[474].from_folded = false;
	    constants_info_[474].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[474].shape = {1280, 1280};
	    constants_info_[474].stride = {1280, 1};
	    constants_info_[474].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[474].original_fqn = "model.audio_tower.layers.17.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[475].name = "model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[475].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[475].offset = 0;
	    constants_info_[475].data_size = 204800;
	    constants_info_[475].from_folded = false;
	    constants_info_[475].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[475].shape = {1280, 40};
	    constants_info_[475].stride = {40, 1};
	    constants_info_[475].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[475].original_fqn = "model.audio_tower.layers.17.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[476].name = "model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[476].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[476].offset = 0;
	    constants_info_[476].data_size = 51200;
	    constants_info_[476].from_folded = false;
	    constants_info_[476].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[476].shape = {1280, 40};
	    constants_info_[476].stride = {40, 1};
	    constants_info_[476].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[476].original_fqn = "model.audio_tower.layers.17.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[477].name = "model_audio_tower_layers_17_self_attn_out_proj_bias";
	    constants_info_[477].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[477].offset = 0;
	    constants_info_[477].data_size = 5120;
	    constants_info_[477].from_folded = false;
	    constants_info_[477].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[477].shape = {1280};
	    constants_info_[477].stride = {1};
	    constants_info_[477].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[477].original_fqn = "model.audio_tower.layers.17.self_attn.out_proj.bias";
	    constants_info_[478].name = "model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[478].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[478].offset = 0;
	    constants_info_[478].data_size = 1638400;
	    constants_info_[478].from_folded = false;
	    constants_info_[478].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[478].shape = {1280, 1280};
	    constants_info_[478].stride = {1280, 1};
	    constants_info_[478].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[478].original_fqn = "model.audio_tower.layers.17.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[479].name = "model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[479].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[479].offset = 0;
	    constants_info_[479].data_size = 204800;
	    constants_info_[479].from_folded = false;
	    constants_info_[479].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[479].shape = {1280, 40};
	    constants_info_[479].stride = {40, 1};
	    constants_info_[479].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[479].original_fqn = "model.audio_tower.layers.17.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[480].name = "model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[480].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[480].offset = 0;
	    constants_info_[480].data_size = 51200;
	    constants_info_[480].from_folded = false;
	    constants_info_[480].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[480].shape = {1280, 40};
	    constants_info_[480].stride = {40, 1};
	    constants_info_[480].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[480].original_fqn = "model.audio_tower.layers.17.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[481].name = "model_audio_tower_layers_17_final_layer_norm_weight";
	    constants_info_[481].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[481].offset = 0;
	    constants_info_[481].data_size = 5120;
	    constants_info_[481].from_folded = false;
	    constants_info_[481].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[481].shape = {1280};
	    constants_info_[481].stride = {1};
	    constants_info_[481].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[481].original_fqn = "model.audio_tower.layers.17.final_layer_norm.weight";
	    constants_info_[482].name = "model_audio_tower_layers_17_final_layer_norm_bias";
	    constants_info_[482].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[482].offset = 0;
	    constants_info_[482].data_size = 5120;
	    constants_info_[482].from_folded = false;
	    constants_info_[482].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[482].shape = {1280};
	    constants_info_[482].stride = {1};
	    constants_info_[482].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[482].original_fqn = "model.audio_tower.layers.17.final_layer_norm.bias";
	    constants_info_[483].name = "model_audio_tower_layers_17_fc1_bias";
	    constants_info_[483].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[483].offset = 0;
	    constants_info_[483].data_size = 20480;
	    constants_info_[483].from_folded = false;
	    constants_info_[483].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[483].shape = {5120};
	    constants_info_[483].stride = {1};
	    constants_info_[483].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[483].original_fqn = "model.audio_tower.layers.17.fc1.bias";
	    constants_info_[484].name = "model_audio_tower_layers_17_fc1_parametrizations_weight_original0";
	    constants_info_[484].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[484].offset = 0;
	    constants_info_[484].data_size = 6553600;
	    constants_info_[484].from_folded = false;
	    constants_info_[484].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[484].shape = {5120, 1280};
	    constants_info_[484].stride = {1280, 1};
	    constants_info_[484].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[484].original_fqn = "model.audio_tower.layers.17.fc1.parametrizations.weight.original0";
	    constants_info_[485].name = "model_audio_tower_layers_17_fc1_parametrizations_weight_original1";
	    constants_info_[485].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[485].offset = 0;
	    constants_info_[485].data_size = 819200;
	    constants_info_[485].from_folded = false;
	    constants_info_[485].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[485].shape = {5120, 40};
	    constants_info_[485].stride = {40, 1};
	    constants_info_[485].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[485].original_fqn = "model.audio_tower.layers.17.fc1.parametrizations.weight.original1";
	    constants_info_[486].name = "model_audio_tower_layers_17_fc1_parametrizations_weight_original2";
	    constants_info_[486].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[486].offset = 0;
	    constants_info_[486].data_size = 204800;
	    constants_info_[486].from_folded = false;
	    constants_info_[486].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[486].shape = {5120, 40};
	    constants_info_[486].stride = {40, 1};
	    constants_info_[486].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[486].original_fqn = "model.audio_tower.layers.17.fc1.parametrizations.weight.original2";
	    constants_info_[487].name = "model_audio_tower_layers_17_fc2_bias";
	    constants_info_[487].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[487].offset = 0;
	    constants_info_[487].data_size = 5120;
	    constants_info_[487].from_folded = false;
	    constants_info_[487].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[487].shape = {1280};
	    constants_info_[487].stride = {1};
	    constants_info_[487].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[487].original_fqn = "model.audio_tower.layers.17.fc2.bias";
	    constants_info_[488].name = "model_audio_tower_layers_17_fc2_parametrizations_weight_original0";
	    constants_info_[488].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[488].offset = 0;
	    constants_info_[488].data_size = 6553600;
	    constants_info_[488].from_folded = false;
	    constants_info_[488].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[488].shape = {1280, 5120};
	    constants_info_[488].stride = {5120, 1};
	    constants_info_[488].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[488].original_fqn = "model.audio_tower.layers.17.fc2.parametrizations.weight.original0";
	    constants_info_[489].name = "model_audio_tower_layers_17_fc2_parametrizations_weight_original1";
	    constants_info_[489].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[489].offset = 0;
	    constants_info_[489].data_size = 819200;
	    constants_info_[489].from_folded = false;
	    constants_info_[489].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[489].shape = {1280, 160};
	    constants_info_[489].stride = {160, 1};
	    constants_info_[489].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[489].original_fqn = "model.audio_tower.layers.17.fc2.parametrizations.weight.original1";
	    constants_info_[490].name = "model_audio_tower_layers_17_fc2_parametrizations_weight_original2";
	    constants_info_[490].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[490].offset = 0;
	    constants_info_[490].data_size = 204800;
	    constants_info_[490].from_folded = false;
	    constants_info_[490].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[490].shape = {1280, 160};
	    constants_info_[490].stride = {160, 1};
	    constants_info_[490].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[490].original_fqn = "model.audio_tower.layers.17.fc2.parametrizations.weight.original2";
	    constants_info_[491].name = "model_audio_tower_layers_18_self_attn_layer_norm_weight";
	    constants_info_[491].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[491].offset = 0;
	    constants_info_[491].data_size = 5120;
	    constants_info_[491].from_folded = false;
	    constants_info_[491].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[491].shape = {1280};
	    constants_info_[491].stride = {1};
	    constants_info_[491].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[491].original_fqn = "model.audio_tower.layers.18.self_attn_layer_norm.weight";
	    constants_info_[492].name = "model_audio_tower_layers_18_self_attn_layer_norm_bias";
	    constants_info_[492].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[492].offset = 0;
	    constants_info_[492].data_size = 5120;
	    constants_info_[492].from_folded = false;
	    constants_info_[492].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[492].shape = {1280};
	    constants_info_[492].stride = {1};
	    constants_info_[492].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[492].original_fqn = "model.audio_tower.layers.18.self_attn_layer_norm.bias";
	    constants_info_[493].name = "model_audio_tower_layers_18_self_attn_q_proj_bias";
	    constants_info_[493].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[493].offset = 0;
	    constants_info_[493].data_size = 5120;
	    constants_info_[493].from_folded = false;
	    constants_info_[493].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[493].shape = {1280};
	    constants_info_[493].stride = {1};
	    constants_info_[493].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[493].original_fqn = "model.audio_tower.layers.18.self_attn.q_proj.bias";
	    constants_info_[494].name = "model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[494].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[494].offset = 0;
	    constants_info_[494].data_size = 1638400;
	    constants_info_[494].from_folded = false;
	    constants_info_[494].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[494].shape = {1280, 1280};
	    constants_info_[494].stride = {1280, 1};
	    constants_info_[494].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[494].original_fqn = "model.audio_tower.layers.18.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[495].name = "model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[495].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[495].offset = 0;
	    constants_info_[495].data_size = 204800;
	    constants_info_[495].from_folded = false;
	    constants_info_[495].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[495].shape = {1280, 40};
	    constants_info_[495].stride = {40, 1};
	    constants_info_[495].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[495].original_fqn = "model.audio_tower.layers.18.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[496].name = "model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[496].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[496].offset = 0;
	    constants_info_[496].data_size = 51200;
	    constants_info_[496].from_folded = false;
	    constants_info_[496].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[496].shape = {1280, 40};
	    constants_info_[496].stride = {40, 1};
	    constants_info_[496].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[496].original_fqn = "model.audio_tower.layers.18.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[497].name = "model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[497].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[497].offset = 0;
	    constants_info_[497].data_size = 1638400;
	    constants_info_[497].from_folded = false;
	    constants_info_[497].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[497].shape = {1280, 1280};
	    constants_info_[497].stride = {1280, 1};
	    constants_info_[497].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[497].original_fqn = "model.audio_tower.layers.18.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[498].name = "model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[498].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[498].offset = 0;
	    constants_info_[498].data_size = 204800;
	    constants_info_[498].from_folded = false;
	    constants_info_[498].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[498].shape = {1280, 40};
	    constants_info_[498].stride = {40, 1};
	    constants_info_[498].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[498].original_fqn = "model.audio_tower.layers.18.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[499].name = "model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[499].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[499].offset = 0;
	    constants_info_[499].data_size = 51200;
	    constants_info_[499].from_folded = false;
	    constants_info_[499].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[499].shape = {1280, 40};
	    constants_info_[499].stride = {40, 1};
	    constants_info_[499].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[499].original_fqn = "model.audio_tower.layers.18.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[500].name = "model_audio_tower_layers_18_self_attn_v_proj_bias";
	    constants_info_[500].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[500].offset = 0;
	    constants_info_[500].data_size = 5120;
	    constants_info_[500].from_folded = false;
	    constants_info_[500].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[500].shape = {1280};
	    constants_info_[500].stride = {1};
	    constants_info_[500].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[500].original_fqn = "model.audio_tower.layers.18.self_attn.v_proj.bias";
	    constants_info_[501].name = "model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[501].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[501].offset = 0;
	    constants_info_[501].data_size = 1638400;
	    constants_info_[501].from_folded = false;
	    constants_info_[501].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[501].shape = {1280, 1280};
	    constants_info_[501].stride = {1280, 1};
	    constants_info_[501].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[501].original_fqn = "model.audio_tower.layers.18.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[502].name = "model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[502].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[502].offset = 0;
	    constants_info_[502].data_size = 204800;
	    constants_info_[502].from_folded = false;
	    constants_info_[502].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[502].shape = {1280, 40};
	    constants_info_[502].stride = {40, 1};
	    constants_info_[502].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[502].original_fqn = "model.audio_tower.layers.18.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[503].name = "model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[503].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[503].offset = 0;
	    constants_info_[503].data_size = 51200;
	    constants_info_[503].from_folded = false;
	    constants_info_[503].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[503].shape = {1280, 40};
	    constants_info_[503].stride = {40, 1};
	    constants_info_[503].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[503].original_fqn = "model.audio_tower.layers.18.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[504].name = "model_audio_tower_layers_18_self_attn_out_proj_bias";
	    constants_info_[504].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[504].offset = 0;
	    constants_info_[504].data_size = 5120;
	    constants_info_[504].from_folded = false;
	    constants_info_[504].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[504].shape = {1280};
	    constants_info_[504].stride = {1};
	    constants_info_[504].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[504].original_fqn = "model.audio_tower.layers.18.self_attn.out_proj.bias";
	    constants_info_[505].name = "model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[505].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[505].offset = 0;
	    constants_info_[505].data_size = 1638400;
	    constants_info_[505].from_folded = false;
	    constants_info_[505].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[505].shape = {1280, 1280};
	    constants_info_[505].stride = {1280, 1};
	    constants_info_[505].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[505].original_fqn = "model.audio_tower.layers.18.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[506].name = "model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[506].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[506].offset = 0;
	    constants_info_[506].data_size = 204800;
	    constants_info_[506].from_folded = false;
	    constants_info_[506].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[506].shape = {1280, 40};
	    constants_info_[506].stride = {40, 1};
	    constants_info_[506].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[506].original_fqn = "model.audio_tower.layers.18.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[507].name = "model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[507].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[507].offset = 0;
	    constants_info_[507].data_size = 51200;
	    constants_info_[507].from_folded = false;
	    constants_info_[507].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[507].shape = {1280, 40};
	    constants_info_[507].stride = {40, 1};
	    constants_info_[507].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[507].original_fqn = "model.audio_tower.layers.18.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[508].name = "model_audio_tower_layers_18_final_layer_norm_weight";
	    constants_info_[508].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[508].offset = 0;
	    constants_info_[508].data_size = 5120;
	    constants_info_[508].from_folded = false;
	    constants_info_[508].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[508].shape = {1280};
	    constants_info_[508].stride = {1};
	    constants_info_[508].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[508].original_fqn = "model.audio_tower.layers.18.final_layer_norm.weight";
	    constants_info_[509].name = "model_audio_tower_layers_18_final_layer_norm_bias";
	    constants_info_[509].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[509].offset = 0;
	    constants_info_[509].data_size = 5120;
	    constants_info_[509].from_folded = false;
	    constants_info_[509].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[509].shape = {1280};
	    constants_info_[509].stride = {1};
	    constants_info_[509].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[509].original_fqn = "model.audio_tower.layers.18.final_layer_norm.bias";
	    constants_info_[510].name = "model_audio_tower_layers_18_fc1_bias";
	    constants_info_[510].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[510].offset = 0;
	    constants_info_[510].data_size = 20480;
	    constants_info_[510].from_folded = false;
	    constants_info_[510].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[510].shape = {5120};
	    constants_info_[510].stride = {1};
	    constants_info_[510].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[510].original_fqn = "model.audio_tower.layers.18.fc1.bias";
	    constants_info_[511].name = "model_audio_tower_layers_18_fc1_parametrizations_weight_original0";
	    constants_info_[511].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[511].offset = 0;
	    constants_info_[511].data_size = 6553600;
	    constants_info_[511].from_folded = false;
	    constants_info_[511].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[511].shape = {5120, 1280};
	    constants_info_[511].stride = {1280, 1};
	    constants_info_[511].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[511].original_fqn = "model.audio_tower.layers.18.fc1.parametrizations.weight.original0";
	    constants_info_[512].name = "model_audio_tower_layers_18_fc1_parametrizations_weight_original1";
	    constants_info_[512].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[512].offset = 0;
	    constants_info_[512].data_size = 819200;
	    constants_info_[512].from_folded = false;
	    constants_info_[512].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[512].shape = {5120, 40};
	    constants_info_[512].stride = {40, 1};
	    constants_info_[512].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[512].original_fqn = "model.audio_tower.layers.18.fc1.parametrizations.weight.original1";
	    constants_info_[513].name = "model_audio_tower_layers_18_fc1_parametrizations_weight_original2";
	    constants_info_[513].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[513].offset = 0;
	    constants_info_[513].data_size = 204800;
	    constants_info_[513].from_folded = false;
	    constants_info_[513].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[513].shape = {5120, 40};
	    constants_info_[513].stride = {40, 1};
	    constants_info_[513].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[513].original_fqn = "model.audio_tower.layers.18.fc1.parametrizations.weight.original2";
	    constants_info_[514].name = "model_audio_tower_layers_18_fc2_bias";
	    constants_info_[514].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[514].offset = 0;
	    constants_info_[514].data_size = 5120;
	    constants_info_[514].from_folded = false;
	    constants_info_[514].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[514].shape = {1280};
	    constants_info_[514].stride = {1};
	    constants_info_[514].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[514].original_fqn = "model.audio_tower.layers.18.fc2.bias";
	    constants_info_[515].name = "model_audio_tower_layers_18_fc2_parametrizations_weight_original0";
	    constants_info_[515].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[515].offset = 0;
	    constants_info_[515].data_size = 6553600;
	    constants_info_[515].from_folded = false;
	    constants_info_[515].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[515].shape = {1280, 5120};
	    constants_info_[515].stride = {5120, 1};
	    constants_info_[515].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[515].original_fqn = "model.audio_tower.layers.18.fc2.parametrizations.weight.original0";
	    constants_info_[516].name = "model_audio_tower_layers_18_fc2_parametrizations_weight_original1";
	    constants_info_[516].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[516].offset = 0;
	    constants_info_[516].data_size = 819200;
	    constants_info_[516].from_folded = false;
	    constants_info_[516].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[516].shape = {1280, 160};
	    constants_info_[516].stride = {160, 1};
	    constants_info_[516].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[516].original_fqn = "model.audio_tower.layers.18.fc2.parametrizations.weight.original1";
	    constants_info_[517].name = "model_audio_tower_layers_18_fc2_parametrizations_weight_original2";
	    constants_info_[517].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[517].offset = 0;
	    constants_info_[517].data_size = 204800;
	    constants_info_[517].from_folded = false;
	    constants_info_[517].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[517].shape = {1280, 160};
	    constants_info_[517].stride = {160, 1};
	    constants_info_[517].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[517].original_fqn = "model.audio_tower.layers.18.fc2.parametrizations.weight.original2";
	    constants_info_[518].name = "model_audio_tower_layers_19_self_attn_layer_norm_weight";
	    constants_info_[518].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[518].offset = 0;
	    constants_info_[518].data_size = 5120;
	    constants_info_[518].from_folded = false;
	    constants_info_[518].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[518].shape = {1280};
	    constants_info_[518].stride = {1};
	    constants_info_[518].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[518].original_fqn = "model.audio_tower.layers.19.self_attn_layer_norm.weight";
	    constants_info_[519].name = "model_audio_tower_layers_19_self_attn_layer_norm_bias";
	    constants_info_[519].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[519].offset = 0;
	    constants_info_[519].data_size = 5120;
	    constants_info_[519].from_folded = false;
	    constants_info_[519].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[519].shape = {1280};
	    constants_info_[519].stride = {1};
	    constants_info_[519].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[519].original_fqn = "model.audio_tower.layers.19.self_attn_layer_norm.bias";
	    constants_info_[520].name = "model_audio_tower_layers_19_self_attn_q_proj_bias";
	    constants_info_[520].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[520].offset = 0;
	    constants_info_[520].data_size = 5120;
	    constants_info_[520].from_folded = false;
	    constants_info_[520].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[520].shape = {1280};
	    constants_info_[520].stride = {1};
	    constants_info_[520].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[520].original_fqn = "model.audio_tower.layers.19.self_attn.q_proj.bias";
	    constants_info_[521].name = "model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[521].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[521].offset = 0;
	    constants_info_[521].data_size = 1638400;
	    constants_info_[521].from_folded = false;
	    constants_info_[521].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[521].shape = {1280, 1280};
	    constants_info_[521].stride = {1280, 1};
	    constants_info_[521].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[521].original_fqn = "model.audio_tower.layers.19.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[522].name = "model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[522].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[522].offset = 0;
	    constants_info_[522].data_size = 204800;
	    constants_info_[522].from_folded = false;
	    constants_info_[522].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[522].shape = {1280, 40};
	    constants_info_[522].stride = {40, 1};
	    constants_info_[522].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[522].original_fqn = "model.audio_tower.layers.19.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[523].name = "model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[523].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[523].offset = 0;
	    constants_info_[523].data_size = 51200;
	    constants_info_[523].from_folded = false;
	    constants_info_[523].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[523].shape = {1280, 40};
	    constants_info_[523].stride = {40, 1};
	    constants_info_[523].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[523].original_fqn = "model.audio_tower.layers.19.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[524].name = "model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[524].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[524].offset = 0;
	    constants_info_[524].data_size = 1638400;
	    constants_info_[524].from_folded = false;
	    constants_info_[524].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[524].shape = {1280, 1280};
	    constants_info_[524].stride = {1280, 1};
	    constants_info_[524].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[524].original_fqn = "model.audio_tower.layers.19.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[525].name = "model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[525].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[525].offset = 0;
	    constants_info_[525].data_size = 204800;
	    constants_info_[525].from_folded = false;
	    constants_info_[525].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[525].shape = {1280, 40};
	    constants_info_[525].stride = {40, 1};
	    constants_info_[525].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[525].original_fqn = "model.audio_tower.layers.19.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[526].name = "model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[526].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[526].offset = 0;
	    constants_info_[526].data_size = 51200;
	    constants_info_[526].from_folded = false;
	    constants_info_[526].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[526].shape = {1280, 40};
	    constants_info_[526].stride = {40, 1};
	    constants_info_[526].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[526].original_fqn = "model.audio_tower.layers.19.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[527].name = "model_audio_tower_layers_19_self_attn_v_proj_bias";
	    constants_info_[527].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[527].offset = 0;
	    constants_info_[527].data_size = 5120;
	    constants_info_[527].from_folded = false;
	    constants_info_[527].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[527].shape = {1280};
	    constants_info_[527].stride = {1};
	    constants_info_[527].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[527].original_fqn = "model.audio_tower.layers.19.self_attn.v_proj.bias";
	    constants_info_[528].name = "model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[528].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[528].offset = 0;
	    constants_info_[528].data_size = 1638400;
	    constants_info_[528].from_folded = false;
	    constants_info_[528].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[528].shape = {1280, 1280};
	    constants_info_[528].stride = {1280, 1};
	    constants_info_[528].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[528].original_fqn = "model.audio_tower.layers.19.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[529].name = "model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[529].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[529].offset = 0;
	    constants_info_[529].data_size = 204800;
	    constants_info_[529].from_folded = false;
	    constants_info_[529].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[529].shape = {1280, 40};
	    constants_info_[529].stride = {40, 1};
	    constants_info_[529].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[529].original_fqn = "model.audio_tower.layers.19.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[530].name = "model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[530].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[530].offset = 0;
	    constants_info_[530].data_size = 51200;
	    constants_info_[530].from_folded = false;
	    constants_info_[530].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[530].shape = {1280, 40};
	    constants_info_[530].stride = {40, 1};
	    constants_info_[530].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[530].original_fqn = "model.audio_tower.layers.19.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[531].name = "model_audio_tower_layers_19_self_attn_out_proj_bias";
	    constants_info_[531].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[531].offset = 0;
	    constants_info_[531].data_size = 5120;
	    constants_info_[531].from_folded = false;
	    constants_info_[531].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[531].shape = {1280};
	    constants_info_[531].stride = {1};
	    constants_info_[531].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[531].original_fqn = "model.audio_tower.layers.19.self_attn.out_proj.bias";
	    constants_info_[532].name = "model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[532].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[532].offset = 0;
	    constants_info_[532].data_size = 1638400;
	    constants_info_[532].from_folded = false;
	    constants_info_[532].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[532].shape = {1280, 1280};
	    constants_info_[532].stride = {1280, 1};
	    constants_info_[532].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[532].original_fqn = "model.audio_tower.layers.19.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[533].name = "model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[533].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[533].offset = 0;
	    constants_info_[533].data_size = 204800;
	    constants_info_[533].from_folded = false;
	    constants_info_[533].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[533].shape = {1280, 40};
	    constants_info_[533].stride = {40, 1};
	    constants_info_[533].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[533].original_fqn = "model.audio_tower.layers.19.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[534].name = "model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[534].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[534].offset = 0;
	    constants_info_[534].data_size = 51200;
	    constants_info_[534].from_folded = false;
	    constants_info_[534].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[534].shape = {1280, 40};
	    constants_info_[534].stride = {40, 1};
	    constants_info_[534].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[534].original_fqn = "model.audio_tower.layers.19.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[535].name = "model_audio_tower_layers_19_final_layer_norm_weight";
	    constants_info_[535].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[535].offset = 0;
	    constants_info_[535].data_size = 5120;
	    constants_info_[535].from_folded = false;
	    constants_info_[535].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[535].shape = {1280};
	    constants_info_[535].stride = {1};
	    constants_info_[535].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[535].original_fqn = "model.audio_tower.layers.19.final_layer_norm.weight";
	    constants_info_[536].name = "model_audio_tower_layers_19_final_layer_norm_bias";
	    constants_info_[536].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[536].offset = 0;
	    constants_info_[536].data_size = 5120;
	    constants_info_[536].from_folded = false;
	    constants_info_[536].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[536].shape = {1280};
	    constants_info_[536].stride = {1};
	    constants_info_[536].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[536].original_fqn = "model.audio_tower.layers.19.final_layer_norm.bias";
	    constants_info_[537].name = "model_audio_tower_layers_19_fc1_bias";
	    constants_info_[537].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[537].offset = 0;
	    constants_info_[537].data_size = 20480;
	    constants_info_[537].from_folded = false;
	    constants_info_[537].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[537].shape = {5120};
	    constants_info_[537].stride = {1};
	    constants_info_[537].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[537].original_fqn = "model.audio_tower.layers.19.fc1.bias";
	    constants_info_[538].name = "model_audio_tower_layers_19_fc1_parametrizations_weight_original0";
	    constants_info_[538].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[538].offset = 0;
	    constants_info_[538].data_size = 6553600;
	    constants_info_[538].from_folded = false;
	    constants_info_[538].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[538].shape = {5120, 1280};
	    constants_info_[538].stride = {1280, 1};
	    constants_info_[538].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[538].original_fqn = "model.audio_tower.layers.19.fc1.parametrizations.weight.original0";
	    constants_info_[539].name = "model_audio_tower_layers_19_fc1_parametrizations_weight_original1";
	    constants_info_[539].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[539].offset = 0;
	    constants_info_[539].data_size = 819200;
	    constants_info_[539].from_folded = false;
	    constants_info_[539].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[539].shape = {5120, 40};
	    constants_info_[539].stride = {40, 1};
	    constants_info_[539].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[539].original_fqn = "model.audio_tower.layers.19.fc1.parametrizations.weight.original1";
	    constants_info_[540].name = "model_audio_tower_layers_19_fc1_parametrizations_weight_original2";
	    constants_info_[540].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[540].offset = 0;
	    constants_info_[540].data_size = 204800;
	    constants_info_[540].from_folded = false;
	    constants_info_[540].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[540].shape = {5120, 40};
	    constants_info_[540].stride = {40, 1};
	    constants_info_[540].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[540].original_fqn = "model.audio_tower.layers.19.fc1.parametrizations.weight.original2";
	    constants_info_[541].name = "model_audio_tower_layers_19_fc2_bias";
	    constants_info_[541].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[541].offset = 0;
	    constants_info_[541].data_size = 5120;
	    constants_info_[541].from_folded = false;
	    constants_info_[541].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[541].shape = {1280};
	    constants_info_[541].stride = {1};
	    constants_info_[541].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[541].original_fqn = "model.audio_tower.layers.19.fc2.bias";
	    constants_info_[542].name = "model_audio_tower_layers_19_fc2_parametrizations_weight_original0";
	    constants_info_[542].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[542].offset = 0;
	    constants_info_[542].data_size = 6553600;
	    constants_info_[542].from_folded = false;
	    constants_info_[542].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[542].shape = {1280, 5120};
	    constants_info_[542].stride = {5120, 1};
	    constants_info_[542].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[542].original_fqn = "model.audio_tower.layers.19.fc2.parametrizations.weight.original0";
	    constants_info_[543].name = "model_audio_tower_layers_19_fc2_parametrizations_weight_original1";
	    constants_info_[543].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[543].offset = 0;
	    constants_info_[543].data_size = 819200;
	    constants_info_[543].from_folded = false;
	    constants_info_[543].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[543].shape = {1280, 160};
	    constants_info_[543].stride = {160, 1};
	    constants_info_[543].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[543].original_fqn = "model.audio_tower.layers.19.fc2.parametrizations.weight.original1";
	    constants_info_[544].name = "model_audio_tower_layers_19_fc2_parametrizations_weight_original2";
	    constants_info_[544].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[544].offset = 0;
	    constants_info_[544].data_size = 204800;
	    constants_info_[544].from_folded = false;
	    constants_info_[544].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[544].shape = {1280, 160};
	    constants_info_[544].stride = {160, 1};
	    constants_info_[544].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[544].original_fqn = "model.audio_tower.layers.19.fc2.parametrizations.weight.original2";
	    constants_info_[545].name = "model_audio_tower_layers_20_self_attn_layer_norm_weight";
	    constants_info_[545].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[545].offset = 0;
	    constants_info_[545].data_size = 5120;
	    constants_info_[545].from_folded = false;
	    constants_info_[545].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[545].shape = {1280};
	    constants_info_[545].stride = {1};
	    constants_info_[545].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[545].original_fqn = "model.audio_tower.layers.20.self_attn_layer_norm.weight";
	    constants_info_[546].name = "model_audio_tower_layers_20_self_attn_layer_norm_bias";
	    constants_info_[546].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[546].offset = 0;
	    constants_info_[546].data_size = 5120;
	    constants_info_[546].from_folded = false;
	    constants_info_[546].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[546].shape = {1280};
	    constants_info_[546].stride = {1};
	    constants_info_[546].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[546].original_fqn = "model.audio_tower.layers.20.self_attn_layer_norm.bias";
	    constants_info_[547].name = "model_audio_tower_layers_20_self_attn_q_proj_bias";
	    constants_info_[547].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[547].offset = 0;
	    constants_info_[547].data_size = 5120;
	    constants_info_[547].from_folded = false;
	    constants_info_[547].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[547].shape = {1280};
	    constants_info_[547].stride = {1};
	    constants_info_[547].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[547].original_fqn = "model.audio_tower.layers.20.self_attn.q_proj.bias";
	    constants_info_[548].name = "model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[548].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[548].offset = 0;
	    constants_info_[548].data_size = 1638400;
	    constants_info_[548].from_folded = false;
	    constants_info_[548].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[548].shape = {1280, 1280};
	    constants_info_[548].stride = {1280, 1};
	    constants_info_[548].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[548].original_fqn = "model.audio_tower.layers.20.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[549].name = "model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[549].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[549].offset = 0;
	    constants_info_[549].data_size = 204800;
	    constants_info_[549].from_folded = false;
	    constants_info_[549].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[549].shape = {1280, 40};
	    constants_info_[549].stride = {40, 1};
	    constants_info_[549].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[549].original_fqn = "model.audio_tower.layers.20.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[550].name = "model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[550].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[550].offset = 0;
	    constants_info_[550].data_size = 51200;
	    constants_info_[550].from_folded = false;
	    constants_info_[550].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[550].shape = {1280, 40};
	    constants_info_[550].stride = {40, 1};
	    constants_info_[550].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[550].original_fqn = "model.audio_tower.layers.20.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[551].name = "model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[551].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[551].offset = 0;
	    constants_info_[551].data_size = 1638400;
	    constants_info_[551].from_folded = false;
	    constants_info_[551].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[551].shape = {1280, 1280};
	    constants_info_[551].stride = {1280, 1};
	    constants_info_[551].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[551].original_fqn = "model.audio_tower.layers.20.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[552].name = "model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[552].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[552].offset = 0;
	    constants_info_[552].data_size = 204800;
	    constants_info_[552].from_folded = false;
	    constants_info_[552].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[552].shape = {1280, 40};
	    constants_info_[552].stride = {40, 1};
	    constants_info_[552].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[552].original_fqn = "model.audio_tower.layers.20.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[553].name = "model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[553].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[553].offset = 0;
	    constants_info_[553].data_size = 51200;
	    constants_info_[553].from_folded = false;
	    constants_info_[553].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[553].shape = {1280, 40};
	    constants_info_[553].stride = {40, 1};
	    constants_info_[553].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[553].original_fqn = "model.audio_tower.layers.20.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[554].name = "model_audio_tower_layers_20_self_attn_v_proj_bias";
	    constants_info_[554].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[554].offset = 0;
	    constants_info_[554].data_size = 5120;
	    constants_info_[554].from_folded = false;
	    constants_info_[554].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[554].shape = {1280};
	    constants_info_[554].stride = {1};
	    constants_info_[554].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[554].original_fqn = "model.audio_tower.layers.20.self_attn.v_proj.bias";
	    constants_info_[555].name = "model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[555].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[555].offset = 0;
	    constants_info_[555].data_size = 1638400;
	    constants_info_[555].from_folded = false;
	    constants_info_[555].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[555].shape = {1280, 1280};
	    constants_info_[555].stride = {1280, 1};
	    constants_info_[555].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[555].original_fqn = "model.audio_tower.layers.20.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[556].name = "model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[556].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[556].offset = 0;
	    constants_info_[556].data_size = 204800;
	    constants_info_[556].from_folded = false;
	    constants_info_[556].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[556].shape = {1280, 40};
	    constants_info_[556].stride = {40, 1};
	    constants_info_[556].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[556].original_fqn = "model.audio_tower.layers.20.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[557].name = "model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[557].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[557].offset = 0;
	    constants_info_[557].data_size = 51200;
	    constants_info_[557].from_folded = false;
	    constants_info_[557].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[557].shape = {1280, 40};
	    constants_info_[557].stride = {40, 1};
	    constants_info_[557].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[557].original_fqn = "model.audio_tower.layers.20.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[558].name = "model_audio_tower_layers_20_self_attn_out_proj_bias";
	    constants_info_[558].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[558].offset = 0;
	    constants_info_[558].data_size = 5120;
	    constants_info_[558].from_folded = false;
	    constants_info_[558].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[558].shape = {1280};
	    constants_info_[558].stride = {1};
	    constants_info_[558].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[558].original_fqn = "model.audio_tower.layers.20.self_attn.out_proj.bias";
	    constants_info_[559].name = "model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[559].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[559].offset = 0;
	    constants_info_[559].data_size = 1638400;
	    constants_info_[559].from_folded = false;
	    constants_info_[559].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[559].shape = {1280, 1280};
	    constants_info_[559].stride = {1280, 1};
	    constants_info_[559].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[559].original_fqn = "model.audio_tower.layers.20.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[560].name = "model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[560].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[560].offset = 0;
	    constants_info_[560].data_size = 204800;
	    constants_info_[560].from_folded = false;
	    constants_info_[560].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[560].shape = {1280, 40};
	    constants_info_[560].stride = {40, 1};
	    constants_info_[560].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[560].original_fqn = "model.audio_tower.layers.20.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[561].name = "model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[561].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[561].offset = 0;
	    constants_info_[561].data_size = 51200;
	    constants_info_[561].from_folded = false;
	    constants_info_[561].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[561].shape = {1280, 40};
	    constants_info_[561].stride = {40, 1};
	    constants_info_[561].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[561].original_fqn = "model.audio_tower.layers.20.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[562].name = "model_audio_tower_layers_20_final_layer_norm_weight";
	    constants_info_[562].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[562].offset = 0;
	    constants_info_[562].data_size = 5120;
	    constants_info_[562].from_folded = false;
	    constants_info_[562].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[562].shape = {1280};
	    constants_info_[562].stride = {1};
	    constants_info_[562].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[562].original_fqn = "model.audio_tower.layers.20.final_layer_norm.weight";
	    constants_info_[563].name = "model_audio_tower_layers_20_final_layer_norm_bias";
	    constants_info_[563].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[563].offset = 0;
	    constants_info_[563].data_size = 5120;
	    constants_info_[563].from_folded = false;
	    constants_info_[563].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[563].shape = {1280};
	    constants_info_[563].stride = {1};
	    constants_info_[563].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[563].original_fqn = "model.audio_tower.layers.20.final_layer_norm.bias";
	    constants_info_[564].name = "model_audio_tower_layers_20_fc1_bias";
	    constants_info_[564].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[564].offset = 0;
	    constants_info_[564].data_size = 20480;
	    constants_info_[564].from_folded = false;
	    constants_info_[564].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[564].shape = {5120};
	    constants_info_[564].stride = {1};
	    constants_info_[564].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[564].original_fqn = "model.audio_tower.layers.20.fc1.bias";
	    constants_info_[565].name = "model_audio_tower_layers_20_fc1_parametrizations_weight_original0";
	    constants_info_[565].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[565].offset = 0;
	    constants_info_[565].data_size = 6553600;
	    constants_info_[565].from_folded = false;
	    constants_info_[565].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[565].shape = {5120, 1280};
	    constants_info_[565].stride = {1280, 1};
	    constants_info_[565].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[565].original_fqn = "model.audio_tower.layers.20.fc1.parametrizations.weight.original0";
	    constants_info_[566].name = "model_audio_tower_layers_20_fc1_parametrizations_weight_original1";
	    constants_info_[566].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[566].offset = 0;
	    constants_info_[566].data_size = 819200;
	    constants_info_[566].from_folded = false;
	    constants_info_[566].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[566].shape = {5120, 40};
	    constants_info_[566].stride = {40, 1};
	    constants_info_[566].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[566].original_fqn = "model.audio_tower.layers.20.fc1.parametrizations.weight.original1";
	    constants_info_[567].name = "model_audio_tower_layers_20_fc1_parametrizations_weight_original2";
	    constants_info_[567].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[567].offset = 0;
	    constants_info_[567].data_size = 204800;
	    constants_info_[567].from_folded = false;
	    constants_info_[567].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[567].shape = {5120, 40};
	    constants_info_[567].stride = {40, 1};
	    constants_info_[567].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[567].original_fqn = "model.audio_tower.layers.20.fc1.parametrizations.weight.original2";
	    constants_info_[568].name = "model_audio_tower_layers_20_fc2_bias";
	    constants_info_[568].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[568].offset = 0;
	    constants_info_[568].data_size = 5120;
	    constants_info_[568].from_folded = false;
	    constants_info_[568].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[568].shape = {1280};
	    constants_info_[568].stride = {1};
	    constants_info_[568].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[568].original_fqn = "model.audio_tower.layers.20.fc2.bias";
	    constants_info_[569].name = "model_audio_tower_layers_20_fc2_parametrizations_weight_original0";
	    constants_info_[569].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[569].offset = 0;
	    constants_info_[569].data_size = 6553600;
	    constants_info_[569].from_folded = false;
	    constants_info_[569].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[569].shape = {1280, 5120};
	    constants_info_[569].stride = {5120, 1};
	    constants_info_[569].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[569].original_fqn = "model.audio_tower.layers.20.fc2.parametrizations.weight.original0";
	    constants_info_[570].name = "model_audio_tower_layers_20_fc2_parametrizations_weight_original1";
	    constants_info_[570].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[570].offset = 0;
	    constants_info_[570].data_size = 819200;
	    constants_info_[570].from_folded = false;
	    constants_info_[570].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[570].shape = {1280, 160};
	    constants_info_[570].stride = {160, 1};
	    constants_info_[570].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[570].original_fqn = "model.audio_tower.layers.20.fc2.parametrizations.weight.original1";
	    constants_info_[571].name = "model_audio_tower_layers_20_fc2_parametrizations_weight_original2";
	    constants_info_[571].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[571].offset = 0;
	    constants_info_[571].data_size = 204800;
	    constants_info_[571].from_folded = false;
	    constants_info_[571].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[571].shape = {1280, 160};
	    constants_info_[571].stride = {160, 1};
	    constants_info_[571].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[571].original_fqn = "model.audio_tower.layers.20.fc2.parametrizations.weight.original2";
	    constants_info_[572].name = "model_audio_tower_layers_21_self_attn_layer_norm_weight";
	    constants_info_[572].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[572].offset = 0;
	    constants_info_[572].data_size = 5120;
	    constants_info_[572].from_folded = false;
	    constants_info_[572].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[572].shape = {1280};
	    constants_info_[572].stride = {1};
	    constants_info_[572].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[572].original_fqn = "model.audio_tower.layers.21.self_attn_layer_norm.weight";
	    constants_info_[573].name = "model_audio_tower_layers_21_self_attn_layer_norm_bias";
	    constants_info_[573].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[573].offset = 0;
	    constants_info_[573].data_size = 5120;
	    constants_info_[573].from_folded = false;
	    constants_info_[573].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[573].shape = {1280};
	    constants_info_[573].stride = {1};
	    constants_info_[573].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[573].original_fqn = "model.audio_tower.layers.21.self_attn_layer_norm.bias";
	    constants_info_[574].name = "model_audio_tower_layers_21_self_attn_q_proj_bias";
	    constants_info_[574].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[574].offset = 0;
	    constants_info_[574].data_size = 5120;
	    constants_info_[574].from_folded = false;
	    constants_info_[574].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[574].shape = {1280};
	    constants_info_[574].stride = {1};
	    constants_info_[574].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[574].original_fqn = "model.audio_tower.layers.21.self_attn.q_proj.bias";
	    constants_info_[575].name = "model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[575].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[575].offset = 0;
	    constants_info_[575].data_size = 1638400;
	    constants_info_[575].from_folded = false;
	    constants_info_[575].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[575].shape = {1280, 1280};
	    constants_info_[575].stride = {1280, 1};
	    constants_info_[575].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[575].original_fqn = "model.audio_tower.layers.21.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[576].name = "model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[576].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[576].offset = 0;
	    constants_info_[576].data_size = 204800;
	    constants_info_[576].from_folded = false;
	    constants_info_[576].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[576].shape = {1280, 40};
	    constants_info_[576].stride = {40, 1};
	    constants_info_[576].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[576].original_fqn = "model.audio_tower.layers.21.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[577].name = "model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[577].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[577].offset = 0;
	    constants_info_[577].data_size = 51200;
	    constants_info_[577].from_folded = false;
	    constants_info_[577].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[577].shape = {1280, 40};
	    constants_info_[577].stride = {40, 1};
	    constants_info_[577].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[577].original_fqn = "model.audio_tower.layers.21.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[578].name = "model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[578].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[578].offset = 0;
	    constants_info_[578].data_size = 1638400;
	    constants_info_[578].from_folded = false;
	    constants_info_[578].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[578].shape = {1280, 1280};
	    constants_info_[578].stride = {1280, 1};
	    constants_info_[578].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[578].original_fqn = "model.audio_tower.layers.21.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[579].name = "model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[579].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[579].offset = 0;
	    constants_info_[579].data_size = 204800;
	    constants_info_[579].from_folded = false;
	    constants_info_[579].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[579].shape = {1280, 40};
	    constants_info_[579].stride = {40, 1};
	    constants_info_[579].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[579].original_fqn = "model.audio_tower.layers.21.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[580].name = "model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[580].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[580].offset = 0;
	    constants_info_[580].data_size = 51200;
	    constants_info_[580].from_folded = false;
	    constants_info_[580].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[580].shape = {1280, 40};
	    constants_info_[580].stride = {40, 1};
	    constants_info_[580].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[580].original_fqn = "model.audio_tower.layers.21.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[581].name = "model_audio_tower_layers_21_self_attn_v_proj_bias";
	    constants_info_[581].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[581].offset = 0;
	    constants_info_[581].data_size = 5120;
	    constants_info_[581].from_folded = false;
	    constants_info_[581].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[581].shape = {1280};
	    constants_info_[581].stride = {1};
	    constants_info_[581].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[581].original_fqn = "model.audio_tower.layers.21.self_attn.v_proj.bias";
	    constants_info_[582].name = "model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[582].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[582].offset = 0;
	    constants_info_[582].data_size = 1638400;
	    constants_info_[582].from_folded = false;
	    constants_info_[582].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[582].shape = {1280, 1280};
	    constants_info_[582].stride = {1280, 1};
	    constants_info_[582].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[582].original_fqn = "model.audio_tower.layers.21.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[583].name = "model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[583].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[583].offset = 0;
	    constants_info_[583].data_size = 204800;
	    constants_info_[583].from_folded = false;
	    constants_info_[583].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[583].shape = {1280, 40};
	    constants_info_[583].stride = {40, 1};
	    constants_info_[583].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[583].original_fqn = "model.audio_tower.layers.21.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[584].name = "model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[584].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[584].offset = 0;
	    constants_info_[584].data_size = 51200;
	    constants_info_[584].from_folded = false;
	    constants_info_[584].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[584].shape = {1280, 40};
	    constants_info_[584].stride = {40, 1};
	    constants_info_[584].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[584].original_fqn = "model.audio_tower.layers.21.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[585].name = "model_audio_tower_layers_21_self_attn_out_proj_bias";
	    constants_info_[585].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[585].offset = 0;
	    constants_info_[585].data_size = 5120;
	    constants_info_[585].from_folded = false;
	    constants_info_[585].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[585].shape = {1280};
	    constants_info_[585].stride = {1};
	    constants_info_[585].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[585].original_fqn = "model.audio_tower.layers.21.self_attn.out_proj.bias";
	    constants_info_[586].name = "model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[586].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[586].offset = 0;
	    constants_info_[586].data_size = 1638400;
	    constants_info_[586].from_folded = false;
	    constants_info_[586].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[586].shape = {1280, 1280};
	    constants_info_[586].stride = {1280, 1};
	    constants_info_[586].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[586].original_fqn = "model.audio_tower.layers.21.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[587].name = "model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[587].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[587].offset = 0;
	    constants_info_[587].data_size = 204800;
	    constants_info_[587].from_folded = false;
	    constants_info_[587].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[587].shape = {1280, 40};
	    constants_info_[587].stride = {40, 1};
	    constants_info_[587].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[587].original_fqn = "model.audio_tower.layers.21.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[588].name = "model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[588].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[588].offset = 0;
	    constants_info_[588].data_size = 51200;
	    constants_info_[588].from_folded = false;
	    constants_info_[588].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[588].shape = {1280, 40};
	    constants_info_[588].stride = {40, 1};
	    constants_info_[588].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[588].original_fqn = "model.audio_tower.layers.21.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[589].name = "model_audio_tower_layers_21_final_layer_norm_weight";
	    constants_info_[589].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[589].offset = 0;
	    constants_info_[589].data_size = 5120;
	    constants_info_[589].from_folded = false;
	    constants_info_[589].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[589].shape = {1280};
	    constants_info_[589].stride = {1};
	    constants_info_[589].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[589].original_fqn = "model.audio_tower.layers.21.final_layer_norm.weight";
	    constants_info_[590].name = "model_audio_tower_layers_21_final_layer_norm_bias";
	    constants_info_[590].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[590].offset = 0;
	    constants_info_[590].data_size = 5120;
	    constants_info_[590].from_folded = false;
	    constants_info_[590].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[590].shape = {1280};
	    constants_info_[590].stride = {1};
	    constants_info_[590].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[590].original_fqn = "model.audio_tower.layers.21.final_layer_norm.bias";
	    constants_info_[591].name = "model_audio_tower_layers_21_fc1_bias";
	    constants_info_[591].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[591].offset = 0;
	    constants_info_[591].data_size = 20480;
	    constants_info_[591].from_folded = false;
	    constants_info_[591].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[591].shape = {5120};
	    constants_info_[591].stride = {1};
	    constants_info_[591].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[591].original_fqn = "model.audio_tower.layers.21.fc1.bias";
	    constants_info_[592].name = "model_audio_tower_layers_21_fc1_parametrizations_weight_original0";
	    constants_info_[592].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[592].offset = 0;
	    constants_info_[592].data_size = 6553600;
	    constants_info_[592].from_folded = false;
	    constants_info_[592].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[592].shape = {5120, 1280};
	    constants_info_[592].stride = {1280, 1};
	    constants_info_[592].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[592].original_fqn = "model.audio_tower.layers.21.fc1.parametrizations.weight.original0";
	    constants_info_[593].name = "model_audio_tower_layers_21_fc1_parametrizations_weight_original1";
	    constants_info_[593].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[593].offset = 0;
	    constants_info_[593].data_size = 819200;
	    constants_info_[593].from_folded = false;
	    constants_info_[593].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[593].shape = {5120, 40};
	    constants_info_[593].stride = {40, 1};
	    constants_info_[593].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[593].original_fqn = "model.audio_tower.layers.21.fc1.parametrizations.weight.original1";
	    constants_info_[594].name = "model_audio_tower_layers_21_fc1_parametrizations_weight_original2";
	    constants_info_[594].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[594].offset = 0;
	    constants_info_[594].data_size = 204800;
	    constants_info_[594].from_folded = false;
	    constants_info_[594].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[594].shape = {5120, 40};
	    constants_info_[594].stride = {40, 1};
	    constants_info_[594].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[594].original_fqn = "model.audio_tower.layers.21.fc1.parametrizations.weight.original2";
	    constants_info_[595].name = "model_audio_tower_layers_21_fc2_bias";
	    constants_info_[595].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[595].offset = 0;
	    constants_info_[595].data_size = 5120;
	    constants_info_[595].from_folded = false;
	    constants_info_[595].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[595].shape = {1280};
	    constants_info_[595].stride = {1};
	    constants_info_[595].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[595].original_fqn = "model.audio_tower.layers.21.fc2.bias";
	    constants_info_[596].name = "model_audio_tower_layers_21_fc2_parametrizations_weight_original0";
	    constants_info_[596].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[596].offset = 0;
	    constants_info_[596].data_size = 6553600;
	    constants_info_[596].from_folded = false;
	    constants_info_[596].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[596].shape = {1280, 5120};
	    constants_info_[596].stride = {5120, 1};
	    constants_info_[596].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[596].original_fqn = "model.audio_tower.layers.21.fc2.parametrizations.weight.original0";
	    constants_info_[597].name = "model_audio_tower_layers_21_fc2_parametrizations_weight_original1";
	    constants_info_[597].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[597].offset = 0;
	    constants_info_[597].data_size = 819200;
	    constants_info_[597].from_folded = false;
	    constants_info_[597].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[597].shape = {1280, 160};
	    constants_info_[597].stride = {160, 1};
	    constants_info_[597].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[597].original_fqn = "model.audio_tower.layers.21.fc2.parametrizations.weight.original1";
	    constants_info_[598].name = "model_audio_tower_layers_21_fc2_parametrizations_weight_original2";
	    constants_info_[598].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[598].offset = 0;
	    constants_info_[598].data_size = 204800;
	    constants_info_[598].from_folded = false;
	    constants_info_[598].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[598].shape = {1280, 160};
	    constants_info_[598].stride = {160, 1};
	    constants_info_[598].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[598].original_fqn = "model.audio_tower.layers.21.fc2.parametrizations.weight.original2";
	    constants_info_[599].name = "model_audio_tower_layers_22_self_attn_layer_norm_weight";
	    constants_info_[599].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[599].offset = 0;
	    constants_info_[599].data_size = 5120;
	    constants_info_[599].from_folded = false;
	    constants_info_[599].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[599].shape = {1280};
	    constants_info_[599].stride = {1};
	    constants_info_[599].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[599].original_fqn = "model.audio_tower.layers.22.self_attn_layer_norm.weight";
	    constants_info_[600].name = "model_audio_tower_layers_22_self_attn_layer_norm_bias";
	    constants_info_[600].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[600].offset = 0;
	    constants_info_[600].data_size = 5120;
	    constants_info_[600].from_folded = false;
	    constants_info_[600].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[600].shape = {1280};
	    constants_info_[600].stride = {1};
	    constants_info_[600].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[600].original_fqn = "model.audio_tower.layers.22.self_attn_layer_norm.bias";
	    constants_info_[601].name = "model_audio_tower_layers_22_self_attn_q_proj_bias";
	    constants_info_[601].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[601].offset = 0;
	    constants_info_[601].data_size = 5120;
	    constants_info_[601].from_folded = false;
	    constants_info_[601].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[601].shape = {1280};
	    constants_info_[601].stride = {1};
	    constants_info_[601].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[601].original_fqn = "model.audio_tower.layers.22.self_attn.q_proj.bias";
	    constants_info_[602].name = "model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[602].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[602].offset = 0;
	    constants_info_[602].data_size = 1638400;
	    constants_info_[602].from_folded = false;
	    constants_info_[602].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[602].shape = {1280, 1280};
	    constants_info_[602].stride = {1280, 1};
	    constants_info_[602].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[602].original_fqn = "model.audio_tower.layers.22.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[603].name = "model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[603].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[603].offset = 0;
	    constants_info_[603].data_size = 204800;
	    constants_info_[603].from_folded = false;
	    constants_info_[603].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[603].shape = {1280, 40};
	    constants_info_[603].stride = {40, 1};
	    constants_info_[603].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[603].original_fqn = "model.audio_tower.layers.22.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[604].name = "model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[604].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[604].offset = 0;
	    constants_info_[604].data_size = 51200;
	    constants_info_[604].from_folded = false;
	    constants_info_[604].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[604].shape = {1280, 40};
	    constants_info_[604].stride = {40, 1};
	    constants_info_[604].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[604].original_fqn = "model.audio_tower.layers.22.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[605].name = "model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[605].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[605].offset = 0;
	    constants_info_[605].data_size = 1638400;
	    constants_info_[605].from_folded = false;
	    constants_info_[605].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[605].shape = {1280, 1280};
	    constants_info_[605].stride = {1280, 1};
	    constants_info_[605].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[605].original_fqn = "model.audio_tower.layers.22.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[606].name = "model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[606].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[606].offset = 0;
	    constants_info_[606].data_size = 204800;
	    constants_info_[606].from_folded = false;
	    constants_info_[606].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[606].shape = {1280, 40};
	    constants_info_[606].stride = {40, 1};
	    constants_info_[606].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[606].original_fqn = "model.audio_tower.layers.22.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[607].name = "model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[607].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[607].offset = 0;
	    constants_info_[607].data_size = 51200;
	    constants_info_[607].from_folded = false;
	    constants_info_[607].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[607].shape = {1280, 40};
	    constants_info_[607].stride = {40, 1};
	    constants_info_[607].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[607].original_fqn = "model.audio_tower.layers.22.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[608].name = "model_audio_tower_layers_22_self_attn_v_proj_bias";
	    constants_info_[608].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[608].offset = 0;
	    constants_info_[608].data_size = 5120;
	    constants_info_[608].from_folded = false;
	    constants_info_[608].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[608].shape = {1280};
	    constants_info_[608].stride = {1};
	    constants_info_[608].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[608].original_fqn = "model.audio_tower.layers.22.self_attn.v_proj.bias";
	    constants_info_[609].name = "model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[609].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[609].offset = 0;
	    constants_info_[609].data_size = 1638400;
	    constants_info_[609].from_folded = false;
	    constants_info_[609].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[609].shape = {1280, 1280};
	    constants_info_[609].stride = {1280, 1};
	    constants_info_[609].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[609].original_fqn = "model.audio_tower.layers.22.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[610].name = "model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[610].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[610].offset = 0;
	    constants_info_[610].data_size = 204800;
	    constants_info_[610].from_folded = false;
	    constants_info_[610].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[610].shape = {1280, 40};
	    constants_info_[610].stride = {40, 1};
	    constants_info_[610].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[610].original_fqn = "model.audio_tower.layers.22.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[611].name = "model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[611].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[611].offset = 0;
	    constants_info_[611].data_size = 51200;
	    constants_info_[611].from_folded = false;
	    constants_info_[611].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[611].shape = {1280, 40};
	    constants_info_[611].stride = {40, 1};
	    constants_info_[611].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[611].original_fqn = "model.audio_tower.layers.22.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[612].name = "model_audio_tower_layers_22_self_attn_out_proj_bias";
	    constants_info_[612].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[612].offset = 0;
	    constants_info_[612].data_size = 5120;
	    constants_info_[612].from_folded = false;
	    constants_info_[612].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[612].shape = {1280};
	    constants_info_[612].stride = {1};
	    constants_info_[612].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[612].original_fqn = "model.audio_tower.layers.22.self_attn.out_proj.bias";
	    constants_info_[613].name = "model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[613].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[613].offset = 0;
	    constants_info_[613].data_size = 1638400;
	    constants_info_[613].from_folded = false;
	    constants_info_[613].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[613].shape = {1280, 1280};
	    constants_info_[613].stride = {1280, 1};
	    constants_info_[613].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[613].original_fqn = "model.audio_tower.layers.22.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[614].name = "model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[614].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[614].offset = 0;
	    constants_info_[614].data_size = 204800;
	    constants_info_[614].from_folded = false;
	    constants_info_[614].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[614].shape = {1280, 40};
	    constants_info_[614].stride = {40, 1};
	    constants_info_[614].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[614].original_fqn = "model.audio_tower.layers.22.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[615].name = "model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[615].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[615].offset = 0;
	    constants_info_[615].data_size = 51200;
	    constants_info_[615].from_folded = false;
	    constants_info_[615].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[615].shape = {1280, 40};
	    constants_info_[615].stride = {40, 1};
	    constants_info_[615].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[615].original_fqn = "model.audio_tower.layers.22.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[616].name = "model_audio_tower_layers_22_final_layer_norm_weight";
	    constants_info_[616].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[616].offset = 0;
	    constants_info_[616].data_size = 5120;
	    constants_info_[616].from_folded = false;
	    constants_info_[616].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[616].shape = {1280};
	    constants_info_[616].stride = {1};
	    constants_info_[616].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[616].original_fqn = "model.audio_tower.layers.22.final_layer_norm.weight";
	    constants_info_[617].name = "model_audio_tower_layers_22_final_layer_norm_bias";
	    constants_info_[617].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[617].offset = 0;
	    constants_info_[617].data_size = 5120;
	    constants_info_[617].from_folded = false;
	    constants_info_[617].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[617].shape = {1280};
	    constants_info_[617].stride = {1};
	    constants_info_[617].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[617].original_fqn = "model.audio_tower.layers.22.final_layer_norm.bias";
	    constants_info_[618].name = "model_audio_tower_layers_22_fc1_bias";
	    constants_info_[618].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[618].offset = 0;
	    constants_info_[618].data_size = 20480;
	    constants_info_[618].from_folded = false;
	    constants_info_[618].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[618].shape = {5120};
	    constants_info_[618].stride = {1};
	    constants_info_[618].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[618].original_fqn = "model.audio_tower.layers.22.fc1.bias";
	    constants_info_[619].name = "model_audio_tower_layers_22_fc1_parametrizations_weight_original0";
	    constants_info_[619].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[619].offset = 0;
	    constants_info_[619].data_size = 6553600;
	    constants_info_[619].from_folded = false;
	    constants_info_[619].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[619].shape = {5120, 1280};
	    constants_info_[619].stride = {1280, 1};
	    constants_info_[619].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[619].original_fqn = "model.audio_tower.layers.22.fc1.parametrizations.weight.original0";
	    constants_info_[620].name = "model_audio_tower_layers_22_fc1_parametrizations_weight_original1";
	    constants_info_[620].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[620].offset = 0;
	    constants_info_[620].data_size = 819200;
	    constants_info_[620].from_folded = false;
	    constants_info_[620].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[620].shape = {5120, 40};
	    constants_info_[620].stride = {40, 1};
	    constants_info_[620].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[620].original_fqn = "model.audio_tower.layers.22.fc1.parametrizations.weight.original1";
	    constants_info_[621].name = "model_audio_tower_layers_22_fc1_parametrizations_weight_original2";
	    constants_info_[621].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[621].offset = 0;
	    constants_info_[621].data_size = 204800;
	    constants_info_[621].from_folded = false;
	    constants_info_[621].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[621].shape = {5120, 40};
	    constants_info_[621].stride = {40, 1};
	    constants_info_[621].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[621].original_fqn = "model.audio_tower.layers.22.fc1.parametrizations.weight.original2";
	    constants_info_[622].name = "model_audio_tower_layers_22_fc2_bias";
	    constants_info_[622].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[622].offset = 0;
	    constants_info_[622].data_size = 5120;
	    constants_info_[622].from_folded = false;
	    constants_info_[622].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[622].shape = {1280};
	    constants_info_[622].stride = {1};
	    constants_info_[622].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[622].original_fqn = "model.audio_tower.layers.22.fc2.bias";
	    constants_info_[623].name = "model_audio_tower_layers_22_fc2_parametrizations_weight_original0";
	    constants_info_[623].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[623].offset = 0;
	    constants_info_[623].data_size = 6553600;
	    constants_info_[623].from_folded = false;
	    constants_info_[623].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[623].shape = {1280, 5120};
	    constants_info_[623].stride = {5120, 1};
	    constants_info_[623].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[623].original_fqn = "model.audio_tower.layers.22.fc2.parametrizations.weight.original0";
	    constants_info_[624].name = "model_audio_tower_layers_22_fc2_parametrizations_weight_original1";
	    constants_info_[624].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[624].offset = 0;
	    constants_info_[624].data_size = 819200;
	    constants_info_[624].from_folded = false;
	    constants_info_[624].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[624].shape = {1280, 160};
	    constants_info_[624].stride = {160, 1};
	    constants_info_[624].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[624].original_fqn = "model.audio_tower.layers.22.fc2.parametrizations.weight.original1";
	    constants_info_[625].name = "model_audio_tower_layers_22_fc2_parametrizations_weight_original2";
	    constants_info_[625].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[625].offset = 0;
	    constants_info_[625].data_size = 204800;
	    constants_info_[625].from_folded = false;
	    constants_info_[625].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[625].shape = {1280, 160};
	    constants_info_[625].stride = {160, 1};
	    constants_info_[625].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[625].original_fqn = "model.audio_tower.layers.22.fc2.parametrizations.weight.original2";
	    constants_info_[626].name = "model_audio_tower_layers_23_self_attn_layer_norm_weight";
	    constants_info_[626].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[626].offset = 0;
	    constants_info_[626].data_size = 5120;
	    constants_info_[626].from_folded = false;
	    constants_info_[626].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[626].shape = {1280};
	    constants_info_[626].stride = {1};
	    constants_info_[626].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[626].original_fqn = "model.audio_tower.layers.23.self_attn_layer_norm.weight";
	    constants_info_[627].name = "model_audio_tower_layers_23_self_attn_layer_norm_bias";
	    constants_info_[627].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[627].offset = 0;
	    constants_info_[627].data_size = 5120;
	    constants_info_[627].from_folded = false;
	    constants_info_[627].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[627].shape = {1280};
	    constants_info_[627].stride = {1};
	    constants_info_[627].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[627].original_fqn = "model.audio_tower.layers.23.self_attn_layer_norm.bias";
	    constants_info_[628].name = "model_audio_tower_layers_23_self_attn_q_proj_bias";
	    constants_info_[628].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[628].offset = 0;
	    constants_info_[628].data_size = 5120;
	    constants_info_[628].from_folded = false;
	    constants_info_[628].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[628].shape = {1280};
	    constants_info_[628].stride = {1};
	    constants_info_[628].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[628].original_fqn = "model.audio_tower.layers.23.self_attn.q_proj.bias";
	    constants_info_[629].name = "model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[629].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[629].offset = 0;
	    constants_info_[629].data_size = 1638400;
	    constants_info_[629].from_folded = false;
	    constants_info_[629].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[629].shape = {1280, 1280};
	    constants_info_[629].stride = {1280, 1};
	    constants_info_[629].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[629].original_fqn = "model.audio_tower.layers.23.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[630].name = "model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[630].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[630].offset = 0;
	    constants_info_[630].data_size = 204800;
	    constants_info_[630].from_folded = false;
	    constants_info_[630].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[630].shape = {1280, 40};
	    constants_info_[630].stride = {40, 1};
	    constants_info_[630].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[630].original_fqn = "model.audio_tower.layers.23.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[631].name = "model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[631].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[631].offset = 0;
	    constants_info_[631].data_size = 51200;
	    constants_info_[631].from_folded = false;
	    constants_info_[631].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[631].shape = {1280, 40};
	    constants_info_[631].stride = {40, 1};
	    constants_info_[631].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[631].original_fqn = "model.audio_tower.layers.23.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[632].name = "model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[632].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[632].offset = 0;
	    constants_info_[632].data_size = 1638400;
	    constants_info_[632].from_folded = false;
	    constants_info_[632].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[632].shape = {1280, 1280};
	    constants_info_[632].stride = {1280, 1};
	    constants_info_[632].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[632].original_fqn = "model.audio_tower.layers.23.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[633].name = "model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[633].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[633].offset = 0;
	    constants_info_[633].data_size = 204800;
	    constants_info_[633].from_folded = false;
	    constants_info_[633].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[633].shape = {1280, 40};
	    constants_info_[633].stride = {40, 1};
	    constants_info_[633].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[633].original_fqn = "model.audio_tower.layers.23.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[634].name = "model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[634].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[634].offset = 0;
	    constants_info_[634].data_size = 51200;
	    constants_info_[634].from_folded = false;
	    constants_info_[634].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[634].shape = {1280, 40};
	    constants_info_[634].stride = {40, 1};
	    constants_info_[634].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[634].original_fqn = "model.audio_tower.layers.23.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[635].name = "model_audio_tower_layers_23_self_attn_v_proj_bias";
	    constants_info_[635].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[635].offset = 0;
	    constants_info_[635].data_size = 5120;
	    constants_info_[635].from_folded = false;
	    constants_info_[635].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[635].shape = {1280};
	    constants_info_[635].stride = {1};
	    constants_info_[635].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[635].original_fqn = "model.audio_tower.layers.23.self_attn.v_proj.bias";
	    constants_info_[636].name = "model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[636].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[636].offset = 0;
	    constants_info_[636].data_size = 1638400;
	    constants_info_[636].from_folded = false;
	    constants_info_[636].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[636].shape = {1280, 1280};
	    constants_info_[636].stride = {1280, 1};
	    constants_info_[636].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[636].original_fqn = "model.audio_tower.layers.23.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[637].name = "model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[637].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[637].offset = 0;
	    constants_info_[637].data_size = 204800;
	    constants_info_[637].from_folded = false;
	    constants_info_[637].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[637].shape = {1280, 40};
	    constants_info_[637].stride = {40, 1};
	    constants_info_[637].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[637].original_fqn = "model.audio_tower.layers.23.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[638].name = "model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[638].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[638].offset = 0;
	    constants_info_[638].data_size = 51200;
	    constants_info_[638].from_folded = false;
	    constants_info_[638].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[638].shape = {1280, 40};
	    constants_info_[638].stride = {40, 1};
	    constants_info_[638].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[638].original_fqn = "model.audio_tower.layers.23.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[639].name = "model_audio_tower_layers_23_self_attn_out_proj_bias";
	    constants_info_[639].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[639].offset = 0;
	    constants_info_[639].data_size = 5120;
	    constants_info_[639].from_folded = false;
	    constants_info_[639].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[639].shape = {1280};
	    constants_info_[639].stride = {1};
	    constants_info_[639].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[639].original_fqn = "model.audio_tower.layers.23.self_attn.out_proj.bias";
	    constants_info_[640].name = "model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[640].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[640].offset = 0;
	    constants_info_[640].data_size = 1638400;
	    constants_info_[640].from_folded = false;
	    constants_info_[640].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[640].shape = {1280, 1280};
	    constants_info_[640].stride = {1280, 1};
	    constants_info_[640].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[640].original_fqn = "model.audio_tower.layers.23.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[641].name = "model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[641].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[641].offset = 0;
	    constants_info_[641].data_size = 204800;
	    constants_info_[641].from_folded = false;
	    constants_info_[641].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[641].shape = {1280, 40};
	    constants_info_[641].stride = {40, 1};
	    constants_info_[641].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[641].original_fqn = "model.audio_tower.layers.23.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[642].name = "model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[642].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[642].offset = 0;
	    constants_info_[642].data_size = 51200;
	    constants_info_[642].from_folded = false;
	    constants_info_[642].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[642].shape = {1280, 40};
	    constants_info_[642].stride = {40, 1};
	    constants_info_[642].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[642].original_fqn = "model.audio_tower.layers.23.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[643].name = "model_audio_tower_layers_23_final_layer_norm_weight";
	    constants_info_[643].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[643].offset = 0;
	    constants_info_[643].data_size = 5120;
	    constants_info_[643].from_folded = false;
	    constants_info_[643].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[643].shape = {1280};
	    constants_info_[643].stride = {1};
	    constants_info_[643].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[643].original_fqn = "model.audio_tower.layers.23.final_layer_norm.weight";
	    constants_info_[644].name = "model_audio_tower_layers_23_final_layer_norm_bias";
	    constants_info_[644].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[644].offset = 0;
	    constants_info_[644].data_size = 5120;
	    constants_info_[644].from_folded = false;
	    constants_info_[644].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[644].shape = {1280};
	    constants_info_[644].stride = {1};
	    constants_info_[644].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[644].original_fqn = "model.audio_tower.layers.23.final_layer_norm.bias";
	    constants_info_[645].name = "model_audio_tower_layers_23_fc1_bias";
	    constants_info_[645].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[645].offset = 0;
	    constants_info_[645].data_size = 20480;
	    constants_info_[645].from_folded = false;
	    constants_info_[645].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[645].shape = {5120};
	    constants_info_[645].stride = {1};
	    constants_info_[645].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[645].original_fqn = "model.audio_tower.layers.23.fc1.bias";
	    constants_info_[646].name = "model_audio_tower_layers_23_fc1_parametrizations_weight_original0";
	    constants_info_[646].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[646].offset = 0;
	    constants_info_[646].data_size = 6553600;
	    constants_info_[646].from_folded = false;
	    constants_info_[646].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[646].shape = {5120, 1280};
	    constants_info_[646].stride = {1280, 1};
	    constants_info_[646].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[646].original_fqn = "model.audio_tower.layers.23.fc1.parametrizations.weight.original0";
	    constants_info_[647].name = "model_audio_tower_layers_23_fc1_parametrizations_weight_original1";
	    constants_info_[647].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[647].offset = 0;
	    constants_info_[647].data_size = 819200;
	    constants_info_[647].from_folded = false;
	    constants_info_[647].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[647].shape = {5120, 40};
	    constants_info_[647].stride = {40, 1};
	    constants_info_[647].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[647].original_fqn = "model.audio_tower.layers.23.fc1.parametrizations.weight.original1";
	    constants_info_[648].name = "model_audio_tower_layers_23_fc1_parametrizations_weight_original2";
	    constants_info_[648].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[648].offset = 0;
	    constants_info_[648].data_size = 204800;
	    constants_info_[648].from_folded = false;
	    constants_info_[648].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[648].shape = {5120, 40};
	    constants_info_[648].stride = {40, 1};
	    constants_info_[648].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[648].original_fqn = "model.audio_tower.layers.23.fc1.parametrizations.weight.original2";
	    constants_info_[649].name = "model_audio_tower_layers_23_fc2_bias";
	    constants_info_[649].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[649].offset = 0;
	    constants_info_[649].data_size = 5120;
	    constants_info_[649].from_folded = false;
	    constants_info_[649].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[649].shape = {1280};
	    constants_info_[649].stride = {1};
	    constants_info_[649].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[649].original_fqn = "model.audio_tower.layers.23.fc2.bias";
	    constants_info_[650].name = "model_audio_tower_layers_23_fc2_parametrizations_weight_original0";
	    constants_info_[650].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[650].offset = 0;
	    constants_info_[650].data_size = 6553600;
	    constants_info_[650].from_folded = false;
	    constants_info_[650].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[650].shape = {1280, 5120};
	    constants_info_[650].stride = {5120, 1};
	    constants_info_[650].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[650].original_fqn = "model.audio_tower.layers.23.fc2.parametrizations.weight.original0";
	    constants_info_[651].name = "model_audio_tower_layers_23_fc2_parametrizations_weight_original1";
	    constants_info_[651].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[651].offset = 0;
	    constants_info_[651].data_size = 819200;
	    constants_info_[651].from_folded = false;
	    constants_info_[651].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[651].shape = {1280, 160};
	    constants_info_[651].stride = {160, 1};
	    constants_info_[651].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[651].original_fqn = "model.audio_tower.layers.23.fc2.parametrizations.weight.original1";
	    constants_info_[652].name = "model_audio_tower_layers_23_fc2_parametrizations_weight_original2";
	    constants_info_[652].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[652].offset = 0;
	    constants_info_[652].data_size = 204800;
	    constants_info_[652].from_folded = false;
	    constants_info_[652].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[652].shape = {1280, 160};
	    constants_info_[652].stride = {160, 1};
	    constants_info_[652].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[652].original_fqn = "model.audio_tower.layers.23.fc2.parametrizations.weight.original2";
	    constants_info_[653].name = "model_audio_tower_layers_24_self_attn_layer_norm_weight";
	    constants_info_[653].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[653].offset = 0;
	    constants_info_[653].data_size = 5120;
	    constants_info_[653].from_folded = false;
	    constants_info_[653].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[653].shape = {1280};
	    constants_info_[653].stride = {1};
	    constants_info_[653].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[653].original_fqn = "model.audio_tower.layers.24.self_attn_layer_norm.weight";
	    constants_info_[654].name = "model_audio_tower_layers_24_self_attn_layer_norm_bias";
	    constants_info_[654].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[654].offset = 0;
	    constants_info_[654].data_size = 5120;
	    constants_info_[654].from_folded = false;
	    constants_info_[654].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[654].shape = {1280};
	    constants_info_[654].stride = {1};
	    constants_info_[654].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[654].original_fqn = "model.audio_tower.layers.24.self_attn_layer_norm.bias";
	    constants_info_[655].name = "model_audio_tower_layers_24_self_attn_q_proj_bias";
	    constants_info_[655].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[655].offset = 0;
	    constants_info_[655].data_size = 5120;
	    constants_info_[655].from_folded = false;
	    constants_info_[655].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[655].shape = {1280};
	    constants_info_[655].stride = {1};
	    constants_info_[655].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[655].original_fqn = "model.audio_tower.layers.24.self_attn.q_proj.bias";
	    constants_info_[656].name = "model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[656].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[656].offset = 0;
	    constants_info_[656].data_size = 1638400;
	    constants_info_[656].from_folded = false;
	    constants_info_[656].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[656].shape = {1280, 1280};
	    constants_info_[656].stride = {1280, 1};
	    constants_info_[656].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[656].original_fqn = "model.audio_tower.layers.24.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[657].name = "model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[657].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[657].offset = 0;
	    constants_info_[657].data_size = 204800;
	    constants_info_[657].from_folded = false;
	    constants_info_[657].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[657].shape = {1280, 40};
	    constants_info_[657].stride = {40, 1};
	    constants_info_[657].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[657].original_fqn = "model.audio_tower.layers.24.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[658].name = "model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[658].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[658].offset = 0;
	    constants_info_[658].data_size = 51200;
	    constants_info_[658].from_folded = false;
	    constants_info_[658].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[658].shape = {1280, 40};
	    constants_info_[658].stride = {40, 1};
	    constants_info_[658].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[658].original_fqn = "model.audio_tower.layers.24.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[659].name = "model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[659].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[659].offset = 0;
	    constants_info_[659].data_size = 1638400;
	    constants_info_[659].from_folded = false;
	    constants_info_[659].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[659].shape = {1280, 1280};
	    constants_info_[659].stride = {1280, 1};
	    constants_info_[659].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[659].original_fqn = "model.audio_tower.layers.24.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[660].name = "model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[660].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[660].offset = 0;
	    constants_info_[660].data_size = 204800;
	    constants_info_[660].from_folded = false;
	    constants_info_[660].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[660].shape = {1280, 40};
	    constants_info_[660].stride = {40, 1};
	    constants_info_[660].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[660].original_fqn = "model.audio_tower.layers.24.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[661].name = "model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[661].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[661].offset = 0;
	    constants_info_[661].data_size = 51200;
	    constants_info_[661].from_folded = false;
	    constants_info_[661].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[661].shape = {1280, 40};
	    constants_info_[661].stride = {40, 1};
	    constants_info_[661].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[661].original_fqn = "model.audio_tower.layers.24.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[662].name = "model_audio_tower_layers_24_self_attn_v_proj_bias";
	    constants_info_[662].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[662].offset = 0;
	    constants_info_[662].data_size = 5120;
	    constants_info_[662].from_folded = false;
	    constants_info_[662].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[662].shape = {1280};
	    constants_info_[662].stride = {1};
	    constants_info_[662].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[662].original_fqn = "model.audio_tower.layers.24.self_attn.v_proj.bias";
	    constants_info_[663].name = "model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[663].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[663].offset = 0;
	    constants_info_[663].data_size = 1638400;
	    constants_info_[663].from_folded = false;
	    constants_info_[663].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[663].shape = {1280, 1280};
	    constants_info_[663].stride = {1280, 1};
	    constants_info_[663].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[663].original_fqn = "model.audio_tower.layers.24.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[664].name = "model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[664].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[664].offset = 0;
	    constants_info_[664].data_size = 204800;
	    constants_info_[664].from_folded = false;
	    constants_info_[664].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[664].shape = {1280, 40};
	    constants_info_[664].stride = {40, 1};
	    constants_info_[664].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[664].original_fqn = "model.audio_tower.layers.24.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[665].name = "model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[665].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[665].offset = 0;
	    constants_info_[665].data_size = 51200;
	    constants_info_[665].from_folded = false;
	    constants_info_[665].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[665].shape = {1280, 40};
	    constants_info_[665].stride = {40, 1};
	    constants_info_[665].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[665].original_fqn = "model.audio_tower.layers.24.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[666].name = "model_audio_tower_layers_24_self_attn_out_proj_bias";
	    constants_info_[666].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[666].offset = 0;
	    constants_info_[666].data_size = 5120;
	    constants_info_[666].from_folded = false;
	    constants_info_[666].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[666].shape = {1280};
	    constants_info_[666].stride = {1};
	    constants_info_[666].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[666].original_fqn = "model.audio_tower.layers.24.self_attn.out_proj.bias";
	    constants_info_[667].name = "model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[667].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[667].offset = 0;
	    constants_info_[667].data_size = 1638400;
	    constants_info_[667].from_folded = false;
	    constants_info_[667].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[667].shape = {1280, 1280};
	    constants_info_[667].stride = {1280, 1};
	    constants_info_[667].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[667].original_fqn = "model.audio_tower.layers.24.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[668].name = "model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[668].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[668].offset = 0;
	    constants_info_[668].data_size = 204800;
	    constants_info_[668].from_folded = false;
	    constants_info_[668].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[668].shape = {1280, 40};
	    constants_info_[668].stride = {40, 1};
	    constants_info_[668].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[668].original_fqn = "model.audio_tower.layers.24.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[669].name = "model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[669].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[669].offset = 0;
	    constants_info_[669].data_size = 51200;
	    constants_info_[669].from_folded = false;
	    constants_info_[669].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[669].shape = {1280, 40};
	    constants_info_[669].stride = {40, 1};
	    constants_info_[669].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[669].original_fqn = "model.audio_tower.layers.24.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[670].name = "model_audio_tower_layers_24_final_layer_norm_weight";
	    constants_info_[670].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[670].offset = 0;
	    constants_info_[670].data_size = 5120;
	    constants_info_[670].from_folded = false;
	    constants_info_[670].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[670].shape = {1280};
	    constants_info_[670].stride = {1};
	    constants_info_[670].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[670].original_fqn = "model.audio_tower.layers.24.final_layer_norm.weight";
	    constants_info_[671].name = "model_audio_tower_layers_24_final_layer_norm_bias";
	    constants_info_[671].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[671].offset = 0;
	    constants_info_[671].data_size = 5120;
	    constants_info_[671].from_folded = false;
	    constants_info_[671].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[671].shape = {1280};
	    constants_info_[671].stride = {1};
	    constants_info_[671].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[671].original_fqn = "model.audio_tower.layers.24.final_layer_norm.bias";
	    constants_info_[672].name = "model_audio_tower_layers_24_fc1_bias";
	    constants_info_[672].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[672].offset = 0;
	    constants_info_[672].data_size = 20480;
	    constants_info_[672].from_folded = false;
	    constants_info_[672].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[672].shape = {5120};
	    constants_info_[672].stride = {1};
	    constants_info_[672].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[672].original_fqn = "model.audio_tower.layers.24.fc1.bias";
	    constants_info_[673].name = "model_audio_tower_layers_24_fc1_parametrizations_weight_original0";
	    constants_info_[673].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[673].offset = 0;
	    constants_info_[673].data_size = 6553600;
	    constants_info_[673].from_folded = false;
	    constants_info_[673].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[673].shape = {5120, 1280};
	    constants_info_[673].stride = {1280, 1};
	    constants_info_[673].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[673].original_fqn = "model.audio_tower.layers.24.fc1.parametrizations.weight.original0";
	    constants_info_[674].name = "model_audio_tower_layers_24_fc1_parametrizations_weight_original1";
	    constants_info_[674].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[674].offset = 0;
	    constants_info_[674].data_size = 819200;
	    constants_info_[674].from_folded = false;
	    constants_info_[674].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[674].shape = {5120, 40};
	    constants_info_[674].stride = {40, 1};
	    constants_info_[674].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[674].original_fqn = "model.audio_tower.layers.24.fc1.parametrizations.weight.original1";
	    constants_info_[675].name = "model_audio_tower_layers_24_fc1_parametrizations_weight_original2";
	    constants_info_[675].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[675].offset = 0;
	    constants_info_[675].data_size = 204800;
	    constants_info_[675].from_folded = false;
	    constants_info_[675].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[675].shape = {5120, 40};
	    constants_info_[675].stride = {40, 1};
	    constants_info_[675].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[675].original_fqn = "model.audio_tower.layers.24.fc1.parametrizations.weight.original2";
	    constants_info_[676].name = "model_audio_tower_layers_24_fc2_bias";
	    constants_info_[676].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[676].offset = 0;
	    constants_info_[676].data_size = 5120;
	    constants_info_[676].from_folded = false;
	    constants_info_[676].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[676].shape = {1280};
	    constants_info_[676].stride = {1};
	    constants_info_[676].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[676].original_fqn = "model.audio_tower.layers.24.fc2.bias";
	    constants_info_[677].name = "model_audio_tower_layers_24_fc2_parametrizations_weight_original0";
	    constants_info_[677].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[677].offset = 0;
	    constants_info_[677].data_size = 6553600;
	    constants_info_[677].from_folded = false;
	    constants_info_[677].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[677].shape = {1280, 5120};
	    constants_info_[677].stride = {5120, 1};
	    constants_info_[677].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[677].original_fqn = "model.audio_tower.layers.24.fc2.parametrizations.weight.original0";
	    constants_info_[678].name = "model_audio_tower_layers_24_fc2_parametrizations_weight_original1";
	    constants_info_[678].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[678].offset = 0;
	    constants_info_[678].data_size = 819200;
	    constants_info_[678].from_folded = false;
	    constants_info_[678].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[678].shape = {1280, 160};
	    constants_info_[678].stride = {160, 1};
	    constants_info_[678].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[678].original_fqn = "model.audio_tower.layers.24.fc2.parametrizations.weight.original1";
	    constants_info_[679].name = "model_audio_tower_layers_24_fc2_parametrizations_weight_original2";
	    constants_info_[679].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[679].offset = 0;
	    constants_info_[679].data_size = 204800;
	    constants_info_[679].from_folded = false;
	    constants_info_[679].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[679].shape = {1280, 160};
	    constants_info_[679].stride = {160, 1};
	    constants_info_[679].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[679].original_fqn = "model.audio_tower.layers.24.fc2.parametrizations.weight.original2";
	    constants_info_[680].name = "model_audio_tower_layers_25_self_attn_layer_norm_weight";
	    constants_info_[680].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[680].offset = 0;
	    constants_info_[680].data_size = 5120;
	    constants_info_[680].from_folded = false;
	    constants_info_[680].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[680].shape = {1280};
	    constants_info_[680].stride = {1};
	    constants_info_[680].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[680].original_fqn = "model.audio_tower.layers.25.self_attn_layer_norm.weight";
	    constants_info_[681].name = "model_audio_tower_layers_25_self_attn_layer_norm_bias";
	    constants_info_[681].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[681].offset = 0;
	    constants_info_[681].data_size = 5120;
	    constants_info_[681].from_folded = false;
	    constants_info_[681].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[681].shape = {1280};
	    constants_info_[681].stride = {1};
	    constants_info_[681].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[681].original_fqn = "model.audio_tower.layers.25.self_attn_layer_norm.bias";
	    constants_info_[682].name = "model_audio_tower_layers_25_self_attn_q_proj_bias";
	    constants_info_[682].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[682].offset = 0;
	    constants_info_[682].data_size = 5120;
	    constants_info_[682].from_folded = false;
	    constants_info_[682].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[682].shape = {1280};
	    constants_info_[682].stride = {1};
	    constants_info_[682].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[682].original_fqn = "model.audio_tower.layers.25.self_attn.q_proj.bias";
	    constants_info_[683].name = "model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[683].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[683].offset = 0;
	    constants_info_[683].data_size = 1638400;
	    constants_info_[683].from_folded = false;
	    constants_info_[683].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[683].shape = {1280, 1280};
	    constants_info_[683].stride = {1280, 1};
	    constants_info_[683].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[683].original_fqn = "model.audio_tower.layers.25.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[684].name = "model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[684].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[684].offset = 0;
	    constants_info_[684].data_size = 204800;
	    constants_info_[684].from_folded = false;
	    constants_info_[684].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[684].shape = {1280, 40};
	    constants_info_[684].stride = {40, 1};
	    constants_info_[684].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[684].original_fqn = "model.audio_tower.layers.25.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[685].name = "model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[685].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[685].offset = 0;
	    constants_info_[685].data_size = 51200;
	    constants_info_[685].from_folded = false;
	    constants_info_[685].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[685].shape = {1280, 40};
	    constants_info_[685].stride = {40, 1};
	    constants_info_[685].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[685].original_fqn = "model.audio_tower.layers.25.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[686].name = "model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[686].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[686].offset = 0;
	    constants_info_[686].data_size = 1638400;
	    constants_info_[686].from_folded = false;
	    constants_info_[686].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[686].shape = {1280, 1280};
	    constants_info_[686].stride = {1280, 1};
	    constants_info_[686].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[686].original_fqn = "model.audio_tower.layers.25.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[687].name = "model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[687].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[687].offset = 0;
	    constants_info_[687].data_size = 204800;
	    constants_info_[687].from_folded = false;
	    constants_info_[687].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[687].shape = {1280, 40};
	    constants_info_[687].stride = {40, 1};
	    constants_info_[687].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[687].original_fqn = "model.audio_tower.layers.25.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[688].name = "model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[688].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[688].offset = 0;
	    constants_info_[688].data_size = 51200;
	    constants_info_[688].from_folded = false;
	    constants_info_[688].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[688].shape = {1280, 40};
	    constants_info_[688].stride = {40, 1};
	    constants_info_[688].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[688].original_fqn = "model.audio_tower.layers.25.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[689].name = "model_audio_tower_layers_25_self_attn_v_proj_bias";
	    constants_info_[689].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[689].offset = 0;
	    constants_info_[689].data_size = 5120;
	    constants_info_[689].from_folded = false;
	    constants_info_[689].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[689].shape = {1280};
	    constants_info_[689].stride = {1};
	    constants_info_[689].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[689].original_fqn = "model.audio_tower.layers.25.self_attn.v_proj.bias";
	    constants_info_[690].name = "model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[690].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[690].offset = 0;
	    constants_info_[690].data_size = 1638400;
	    constants_info_[690].from_folded = false;
	    constants_info_[690].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[690].shape = {1280, 1280};
	    constants_info_[690].stride = {1280, 1};
	    constants_info_[690].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[690].original_fqn = "model.audio_tower.layers.25.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[691].name = "model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[691].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[691].offset = 0;
	    constants_info_[691].data_size = 204800;
	    constants_info_[691].from_folded = false;
	    constants_info_[691].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[691].shape = {1280, 40};
	    constants_info_[691].stride = {40, 1};
	    constants_info_[691].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[691].original_fqn = "model.audio_tower.layers.25.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[692].name = "model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[692].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[692].offset = 0;
	    constants_info_[692].data_size = 51200;
	    constants_info_[692].from_folded = false;
	    constants_info_[692].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[692].shape = {1280, 40};
	    constants_info_[692].stride = {40, 1};
	    constants_info_[692].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[692].original_fqn = "model.audio_tower.layers.25.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[693].name = "model_audio_tower_layers_25_self_attn_out_proj_bias";
	    constants_info_[693].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[693].offset = 0;
	    constants_info_[693].data_size = 5120;
	    constants_info_[693].from_folded = false;
	    constants_info_[693].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[693].shape = {1280};
	    constants_info_[693].stride = {1};
	    constants_info_[693].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[693].original_fqn = "model.audio_tower.layers.25.self_attn.out_proj.bias";
	    constants_info_[694].name = "model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[694].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[694].offset = 0;
	    constants_info_[694].data_size = 1638400;
	    constants_info_[694].from_folded = false;
	    constants_info_[694].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[694].shape = {1280, 1280};
	    constants_info_[694].stride = {1280, 1};
	    constants_info_[694].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[694].original_fqn = "model.audio_tower.layers.25.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[695].name = "model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[695].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[695].offset = 0;
	    constants_info_[695].data_size = 204800;
	    constants_info_[695].from_folded = false;
	    constants_info_[695].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[695].shape = {1280, 40};
	    constants_info_[695].stride = {40, 1};
	    constants_info_[695].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[695].original_fqn = "model.audio_tower.layers.25.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[696].name = "model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[696].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[696].offset = 0;
	    constants_info_[696].data_size = 51200;
	    constants_info_[696].from_folded = false;
	    constants_info_[696].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[696].shape = {1280, 40};
	    constants_info_[696].stride = {40, 1};
	    constants_info_[696].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[696].original_fqn = "model.audio_tower.layers.25.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[697].name = "model_audio_tower_layers_25_final_layer_norm_weight";
	    constants_info_[697].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[697].offset = 0;
	    constants_info_[697].data_size = 5120;
	    constants_info_[697].from_folded = false;
	    constants_info_[697].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[697].shape = {1280};
	    constants_info_[697].stride = {1};
	    constants_info_[697].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[697].original_fqn = "model.audio_tower.layers.25.final_layer_norm.weight";
	    constants_info_[698].name = "model_audio_tower_layers_25_final_layer_norm_bias";
	    constants_info_[698].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[698].offset = 0;
	    constants_info_[698].data_size = 5120;
	    constants_info_[698].from_folded = false;
	    constants_info_[698].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[698].shape = {1280};
	    constants_info_[698].stride = {1};
	    constants_info_[698].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[698].original_fqn = "model.audio_tower.layers.25.final_layer_norm.bias";
	    constants_info_[699].name = "model_audio_tower_layers_25_fc1_bias";
	    constants_info_[699].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[699].offset = 0;
	    constants_info_[699].data_size = 20480;
	    constants_info_[699].from_folded = false;
	    constants_info_[699].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[699].shape = {5120};
	    constants_info_[699].stride = {1};
	    constants_info_[699].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[699].original_fqn = "model.audio_tower.layers.25.fc1.bias";
	    constants_info_[700].name = "model_audio_tower_layers_25_fc1_parametrizations_weight_original0";
	    constants_info_[700].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[700].offset = 0;
	    constants_info_[700].data_size = 6553600;
	    constants_info_[700].from_folded = false;
	    constants_info_[700].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[700].shape = {5120, 1280};
	    constants_info_[700].stride = {1280, 1};
	    constants_info_[700].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[700].original_fqn = "model.audio_tower.layers.25.fc1.parametrizations.weight.original0";
	    constants_info_[701].name = "model_audio_tower_layers_25_fc1_parametrizations_weight_original1";
	    constants_info_[701].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[701].offset = 0;
	    constants_info_[701].data_size = 819200;
	    constants_info_[701].from_folded = false;
	    constants_info_[701].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[701].shape = {5120, 40};
	    constants_info_[701].stride = {40, 1};
	    constants_info_[701].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[701].original_fqn = "model.audio_tower.layers.25.fc1.parametrizations.weight.original1";
	    constants_info_[702].name = "model_audio_tower_layers_25_fc1_parametrizations_weight_original2";
	    constants_info_[702].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[702].offset = 0;
	    constants_info_[702].data_size = 204800;
	    constants_info_[702].from_folded = false;
	    constants_info_[702].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[702].shape = {5120, 40};
	    constants_info_[702].stride = {40, 1};
	    constants_info_[702].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[702].original_fqn = "model.audio_tower.layers.25.fc1.parametrizations.weight.original2";
	    constants_info_[703].name = "model_audio_tower_layers_25_fc2_bias";
	    constants_info_[703].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[703].offset = 0;
	    constants_info_[703].data_size = 5120;
	    constants_info_[703].from_folded = false;
	    constants_info_[703].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[703].shape = {1280};
	    constants_info_[703].stride = {1};
	    constants_info_[703].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[703].original_fqn = "model.audio_tower.layers.25.fc2.bias";
	    constants_info_[704].name = "model_audio_tower_layers_25_fc2_parametrizations_weight_original0";
	    constants_info_[704].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[704].offset = 0;
	    constants_info_[704].data_size = 6553600;
	    constants_info_[704].from_folded = false;
	    constants_info_[704].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[704].shape = {1280, 5120};
	    constants_info_[704].stride = {5120, 1};
	    constants_info_[704].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[704].original_fqn = "model.audio_tower.layers.25.fc2.parametrizations.weight.original0";
	    constants_info_[705].name = "model_audio_tower_layers_25_fc2_parametrizations_weight_original1";
	    constants_info_[705].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[705].offset = 0;
	    constants_info_[705].data_size = 819200;
	    constants_info_[705].from_folded = false;
	    constants_info_[705].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[705].shape = {1280, 160};
	    constants_info_[705].stride = {160, 1};
	    constants_info_[705].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[705].original_fqn = "model.audio_tower.layers.25.fc2.parametrizations.weight.original1";
	    constants_info_[706].name = "model_audio_tower_layers_25_fc2_parametrizations_weight_original2";
	    constants_info_[706].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[706].offset = 0;
	    constants_info_[706].data_size = 204800;
	    constants_info_[706].from_folded = false;
	    constants_info_[706].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[706].shape = {1280, 160};
	    constants_info_[706].stride = {160, 1};
	    constants_info_[706].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[706].original_fqn = "model.audio_tower.layers.25.fc2.parametrizations.weight.original2";
	    constants_info_[707].name = "model_audio_tower_layers_26_self_attn_layer_norm_weight";
	    constants_info_[707].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[707].offset = 0;
	    constants_info_[707].data_size = 5120;
	    constants_info_[707].from_folded = false;
	    constants_info_[707].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[707].shape = {1280};
	    constants_info_[707].stride = {1};
	    constants_info_[707].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[707].original_fqn = "model.audio_tower.layers.26.self_attn_layer_norm.weight";
	    constants_info_[708].name = "model_audio_tower_layers_26_self_attn_layer_norm_bias";
	    constants_info_[708].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[708].offset = 0;
	    constants_info_[708].data_size = 5120;
	    constants_info_[708].from_folded = false;
	    constants_info_[708].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[708].shape = {1280};
	    constants_info_[708].stride = {1};
	    constants_info_[708].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[708].original_fqn = "model.audio_tower.layers.26.self_attn_layer_norm.bias";
	    constants_info_[709].name = "model_audio_tower_layers_26_self_attn_q_proj_bias";
	    constants_info_[709].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[709].offset = 0;
	    constants_info_[709].data_size = 5120;
	    constants_info_[709].from_folded = false;
	    constants_info_[709].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[709].shape = {1280};
	    constants_info_[709].stride = {1};
	    constants_info_[709].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[709].original_fqn = "model.audio_tower.layers.26.self_attn.q_proj.bias";
	    constants_info_[710].name = "model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[710].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[710].offset = 0;
	    constants_info_[710].data_size = 1638400;
	    constants_info_[710].from_folded = false;
	    constants_info_[710].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[710].shape = {1280, 1280};
	    constants_info_[710].stride = {1280, 1};
	    constants_info_[710].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[710].original_fqn = "model.audio_tower.layers.26.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[711].name = "model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[711].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[711].offset = 0;
	    constants_info_[711].data_size = 204800;
	    constants_info_[711].from_folded = false;
	    constants_info_[711].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[711].shape = {1280, 40};
	    constants_info_[711].stride = {40, 1};
	    constants_info_[711].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[711].original_fqn = "model.audio_tower.layers.26.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[712].name = "model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[712].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[712].offset = 0;
	    constants_info_[712].data_size = 51200;
	    constants_info_[712].from_folded = false;
	    constants_info_[712].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[712].shape = {1280, 40};
	    constants_info_[712].stride = {40, 1};
	    constants_info_[712].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[712].original_fqn = "model.audio_tower.layers.26.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[713].name = "model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[713].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[713].offset = 0;
	    constants_info_[713].data_size = 1638400;
	    constants_info_[713].from_folded = false;
	    constants_info_[713].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[713].shape = {1280, 1280};
	    constants_info_[713].stride = {1280, 1};
	    constants_info_[713].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[713].original_fqn = "model.audio_tower.layers.26.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[714].name = "model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[714].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[714].offset = 0;
	    constants_info_[714].data_size = 204800;
	    constants_info_[714].from_folded = false;
	    constants_info_[714].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[714].shape = {1280, 40};
	    constants_info_[714].stride = {40, 1};
	    constants_info_[714].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[714].original_fqn = "model.audio_tower.layers.26.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[715].name = "model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[715].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[715].offset = 0;
	    constants_info_[715].data_size = 51200;
	    constants_info_[715].from_folded = false;
	    constants_info_[715].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[715].shape = {1280, 40};
	    constants_info_[715].stride = {40, 1};
	    constants_info_[715].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[715].original_fqn = "model.audio_tower.layers.26.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[716].name = "model_audio_tower_layers_26_self_attn_v_proj_bias";
	    constants_info_[716].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[716].offset = 0;
	    constants_info_[716].data_size = 5120;
	    constants_info_[716].from_folded = false;
	    constants_info_[716].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[716].shape = {1280};
	    constants_info_[716].stride = {1};
	    constants_info_[716].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[716].original_fqn = "model.audio_tower.layers.26.self_attn.v_proj.bias";
	    constants_info_[717].name = "model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[717].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[717].offset = 0;
	    constants_info_[717].data_size = 1638400;
	    constants_info_[717].from_folded = false;
	    constants_info_[717].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[717].shape = {1280, 1280};
	    constants_info_[717].stride = {1280, 1};
	    constants_info_[717].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[717].original_fqn = "model.audio_tower.layers.26.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[718].name = "model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[718].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[718].offset = 0;
	    constants_info_[718].data_size = 204800;
	    constants_info_[718].from_folded = false;
	    constants_info_[718].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[718].shape = {1280, 40};
	    constants_info_[718].stride = {40, 1};
	    constants_info_[718].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[718].original_fqn = "model.audio_tower.layers.26.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[719].name = "model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[719].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[719].offset = 0;
	    constants_info_[719].data_size = 51200;
	    constants_info_[719].from_folded = false;
	    constants_info_[719].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[719].shape = {1280, 40};
	    constants_info_[719].stride = {40, 1};
	    constants_info_[719].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[719].original_fqn = "model.audio_tower.layers.26.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[720].name = "model_audio_tower_layers_26_self_attn_out_proj_bias";
	    constants_info_[720].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[720].offset = 0;
	    constants_info_[720].data_size = 5120;
	    constants_info_[720].from_folded = false;
	    constants_info_[720].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[720].shape = {1280};
	    constants_info_[720].stride = {1};
	    constants_info_[720].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[720].original_fqn = "model.audio_tower.layers.26.self_attn.out_proj.bias";
	    constants_info_[721].name = "model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[721].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[721].offset = 0;
	    constants_info_[721].data_size = 1638400;
	    constants_info_[721].from_folded = false;
	    constants_info_[721].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[721].shape = {1280, 1280};
	    constants_info_[721].stride = {1280, 1};
	    constants_info_[721].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[721].original_fqn = "model.audio_tower.layers.26.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[722].name = "model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[722].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[722].offset = 0;
	    constants_info_[722].data_size = 204800;
	    constants_info_[722].from_folded = false;
	    constants_info_[722].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[722].shape = {1280, 40};
	    constants_info_[722].stride = {40, 1};
	    constants_info_[722].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[722].original_fqn = "model.audio_tower.layers.26.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[723].name = "model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[723].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[723].offset = 0;
	    constants_info_[723].data_size = 51200;
	    constants_info_[723].from_folded = false;
	    constants_info_[723].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[723].shape = {1280, 40};
	    constants_info_[723].stride = {40, 1};
	    constants_info_[723].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[723].original_fqn = "model.audio_tower.layers.26.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[724].name = "model_audio_tower_layers_26_final_layer_norm_weight";
	    constants_info_[724].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[724].offset = 0;
	    constants_info_[724].data_size = 5120;
	    constants_info_[724].from_folded = false;
	    constants_info_[724].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[724].shape = {1280};
	    constants_info_[724].stride = {1};
	    constants_info_[724].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[724].original_fqn = "model.audio_tower.layers.26.final_layer_norm.weight";
	    constants_info_[725].name = "model_audio_tower_layers_26_final_layer_norm_bias";
	    constants_info_[725].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[725].offset = 0;
	    constants_info_[725].data_size = 5120;
	    constants_info_[725].from_folded = false;
	    constants_info_[725].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[725].shape = {1280};
	    constants_info_[725].stride = {1};
	    constants_info_[725].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[725].original_fqn = "model.audio_tower.layers.26.final_layer_norm.bias";
	    constants_info_[726].name = "model_audio_tower_layers_26_fc1_bias";
	    constants_info_[726].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[726].offset = 0;
	    constants_info_[726].data_size = 20480;
	    constants_info_[726].from_folded = false;
	    constants_info_[726].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[726].shape = {5120};
	    constants_info_[726].stride = {1};
	    constants_info_[726].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[726].original_fqn = "model.audio_tower.layers.26.fc1.bias";
	    constants_info_[727].name = "model_audio_tower_layers_26_fc1_parametrizations_weight_original0";
	    constants_info_[727].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[727].offset = 0;
	    constants_info_[727].data_size = 6553600;
	    constants_info_[727].from_folded = false;
	    constants_info_[727].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[727].shape = {5120, 1280};
	    constants_info_[727].stride = {1280, 1};
	    constants_info_[727].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[727].original_fqn = "model.audio_tower.layers.26.fc1.parametrizations.weight.original0";
	    constants_info_[728].name = "model_audio_tower_layers_26_fc1_parametrizations_weight_original1";
	    constants_info_[728].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[728].offset = 0;
	    constants_info_[728].data_size = 819200;
	    constants_info_[728].from_folded = false;
	    constants_info_[728].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[728].shape = {5120, 40};
	    constants_info_[728].stride = {40, 1};
	    constants_info_[728].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[728].original_fqn = "model.audio_tower.layers.26.fc1.parametrizations.weight.original1";
	    constants_info_[729].name = "model_audio_tower_layers_26_fc1_parametrizations_weight_original2";
	    constants_info_[729].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[729].offset = 0;
	    constants_info_[729].data_size = 204800;
	    constants_info_[729].from_folded = false;
	    constants_info_[729].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[729].shape = {5120, 40};
	    constants_info_[729].stride = {40, 1};
	    constants_info_[729].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[729].original_fqn = "model.audio_tower.layers.26.fc1.parametrizations.weight.original2";
	    constants_info_[730].name = "model_audio_tower_layers_26_fc2_bias";
	    constants_info_[730].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[730].offset = 0;
	    constants_info_[730].data_size = 5120;
	    constants_info_[730].from_folded = false;
	    constants_info_[730].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[730].shape = {1280};
	    constants_info_[730].stride = {1};
	    constants_info_[730].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[730].original_fqn = "model.audio_tower.layers.26.fc2.bias";
	    constants_info_[731].name = "model_audio_tower_layers_26_fc2_parametrizations_weight_original0";
	    constants_info_[731].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[731].offset = 0;
	    constants_info_[731].data_size = 6553600;
	    constants_info_[731].from_folded = false;
	    constants_info_[731].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[731].shape = {1280, 5120};
	    constants_info_[731].stride = {5120, 1};
	    constants_info_[731].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[731].original_fqn = "model.audio_tower.layers.26.fc2.parametrizations.weight.original0";
	    constants_info_[732].name = "model_audio_tower_layers_26_fc2_parametrizations_weight_original1";
	    constants_info_[732].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[732].offset = 0;
	    constants_info_[732].data_size = 819200;
	    constants_info_[732].from_folded = false;
	    constants_info_[732].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[732].shape = {1280, 160};
	    constants_info_[732].stride = {160, 1};
	    constants_info_[732].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[732].original_fqn = "model.audio_tower.layers.26.fc2.parametrizations.weight.original1";
	    constants_info_[733].name = "model_audio_tower_layers_26_fc2_parametrizations_weight_original2";
	    constants_info_[733].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[733].offset = 0;
	    constants_info_[733].data_size = 204800;
	    constants_info_[733].from_folded = false;
	    constants_info_[733].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[733].shape = {1280, 160};
	    constants_info_[733].stride = {160, 1};
	    constants_info_[733].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[733].original_fqn = "model.audio_tower.layers.26.fc2.parametrizations.weight.original2";
	    constants_info_[734].name = "model_audio_tower_layers_27_self_attn_layer_norm_weight";
	    constants_info_[734].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[734].offset = 0;
	    constants_info_[734].data_size = 5120;
	    constants_info_[734].from_folded = false;
	    constants_info_[734].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[734].shape = {1280};
	    constants_info_[734].stride = {1};
	    constants_info_[734].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[734].original_fqn = "model.audio_tower.layers.27.self_attn_layer_norm.weight";
	    constants_info_[735].name = "model_audio_tower_layers_27_self_attn_layer_norm_bias";
	    constants_info_[735].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[735].offset = 0;
	    constants_info_[735].data_size = 5120;
	    constants_info_[735].from_folded = false;
	    constants_info_[735].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[735].shape = {1280};
	    constants_info_[735].stride = {1};
	    constants_info_[735].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[735].original_fqn = "model.audio_tower.layers.27.self_attn_layer_norm.bias";
	    constants_info_[736].name = "model_audio_tower_layers_27_self_attn_q_proj_bias";
	    constants_info_[736].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[736].offset = 0;
	    constants_info_[736].data_size = 5120;
	    constants_info_[736].from_folded = false;
	    constants_info_[736].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[736].shape = {1280};
	    constants_info_[736].stride = {1};
	    constants_info_[736].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[736].original_fqn = "model.audio_tower.layers.27.self_attn.q_proj.bias";
	    constants_info_[737].name = "model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[737].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[737].offset = 0;
	    constants_info_[737].data_size = 1638400;
	    constants_info_[737].from_folded = false;
	    constants_info_[737].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[737].shape = {1280, 1280};
	    constants_info_[737].stride = {1280, 1};
	    constants_info_[737].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[737].original_fqn = "model.audio_tower.layers.27.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[738].name = "model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[738].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[738].offset = 0;
	    constants_info_[738].data_size = 204800;
	    constants_info_[738].from_folded = false;
	    constants_info_[738].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[738].shape = {1280, 40};
	    constants_info_[738].stride = {40, 1};
	    constants_info_[738].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[738].original_fqn = "model.audio_tower.layers.27.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[739].name = "model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[739].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[739].offset = 0;
	    constants_info_[739].data_size = 51200;
	    constants_info_[739].from_folded = false;
	    constants_info_[739].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[739].shape = {1280, 40};
	    constants_info_[739].stride = {40, 1};
	    constants_info_[739].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[739].original_fqn = "model.audio_tower.layers.27.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[740].name = "model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[740].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[740].offset = 0;
	    constants_info_[740].data_size = 1638400;
	    constants_info_[740].from_folded = false;
	    constants_info_[740].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[740].shape = {1280, 1280};
	    constants_info_[740].stride = {1280, 1};
	    constants_info_[740].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[740].original_fqn = "model.audio_tower.layers.27.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[741].name = "model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[741].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[741].offset = 0;
	    constants_info_[741].data_size = 204800;
	    constants_info_[741].from_folded = false;
	    constants_info_[741].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[741].shape = {1280, 40};
	    constants_info_[741].stride = {40, 1};
	    constants_info_[741].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[741].original_fqn = "model.audio_tower.layers.27.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[742].name = "model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[742].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[742].offset = 0;
	    constants_info_[742].data_size = 51200;
	    constants_info_[742].from_folded = false;
	    constants_info_[742].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[742].shape = {1280, 40};
	    constants_info_[742].stride = {40, 1};
	    constants_info_[742].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[742].original_fqn = "model.audio_tower.layers.27.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[743].name = "model_audio_tower_layers_27_self_attn_v_proj_bias";
	    constants_info_[743].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[743].offset = 0;
	    constants_info_[743].data_size = 5120;
	    constants_info_[743].from_folded = false;
	    constants_info_[743].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[743].shape = {1280};
	    constants_info_[743].stride = {1};
	    constants_info_[743].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[743].original_fqn = "model.audio_tower.layers.27.self_attn.v_proj.bias";
	    constants_info_[744].name = "model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[744].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[744].offset = 0;
	    constants_info_[744].data_size = 1638400;
	    constants_info_[744].from_folded = false;
	    constants_info_[744].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[744].shape = {1280, 1280};
	    constants_info_[744].stride = {1280, 1};
	    constants_info_[744].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[744].original_fqn = "model.audio_tower.layers.27.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[745].name = "model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[745].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[745].offset = 0;
	    constants_info_[745].data_size = 204800;
	    constants_info_[745].from_folded = false;
	    constants_info_[745].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[745].shape = {1280, 40};
	    constants_info_[745].stride = {40, 1};
	    constants_info_[745].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[745].original_fqn = "model.audio_tower.layers.27.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[746].name = "model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[746].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[746].offset = 0;
	    constants_info_[746].data_size = 51200;
	    constants_info_[746].from_folded = false;
	    constants_info_[746].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[746].shape = {1280, 40};
	    constants_info_[746].stride = {40, 1};
	    constants_info_[746].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[746].original_fqn = "model.audio_tower.layers.27.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[747].name = "model_audio_tower_layers_27_self_attn_out_proj_bias";
	    constants_info_[747].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[747].offset = 0;
	    constants_info_[747].data_size = 5120;
	    constants_info_[747].from_folded = false;
	    constants_info_[747].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[747].shape = {1280};
	    constants_info_[747].stride = {1};
	    constants_info_[747].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[747].original_fqn = "model.audio_tower.layers.27.self_attn.out_proj.bias";
	    constants_info_[748].name = "model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[748].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[748].offset = 0;
	    constants_info_[748].data_size = 1638400;
	    constants_info_[748].from_folded = false;
	    constants_info_[748].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[748].shape = {1280, 1280};
	    constants_info_[748].stride = {1280, 1};
	    constants_info_[748].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[748].original_fqn = "model.audio_tower.layers.27.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[749].name = "model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[749].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[749].offset = 0;
	    constants_info_[749].data_size = 204800;
	    constants_info_[749].from_folded = false;
	    constants_info_[749].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[749].shape = {1280, 40};
	    constants_info_[749].stride = {40, 1};
	    constants_info_[749].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[749].original_fqn = "model.audio_tower.layers.27.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[750].name = "model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[750].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[750].offset = 0;
	    constants_info_[750].data_size = 51200;
	    constants_info_[750].from_folded = false;
	    constants_info_[750].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[750].shape = {1280, 40};
	    constants_info_[750].stride = {40, 1};
	    constants_info_[750].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[750].original_fqn = "model.audio_tower.layers.27.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[751].name = "model_audio_tower_layers_27_final_layer_norm_weight";
	    constants_info_[751].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[751].offset = 0;
	    constants_info_[751].data_size = 5120;
	    constants_info_[751].from_folded = false;
	    constants_info_[751].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[751].shape = {1280};
	    constants_info_[751].stride = {1};
	    constants_info_[751].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[751].original_fqn = "model.audio_tower.layers.27.final_layer_norm.weight";
	    constants_info_[752].name = "model_audio_tower_layers_27_final_layer_norm_bias";
	    constants_info_[752].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[752].offset = 0;
	    constants_info_[752].data_size = 5120;
	    constants_info_[752].from_folded = false;
	    constants_info_[752].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[752].shape = {1280};
	    constants_info_[752].stride = {1};
	    constants_info_[752].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[752].original_fqn = "model.audio_tower.layers.27.final_layer_norm.bias";
	    constants_info_[753].name = "model_audio_tower_layers_27_fc1_bias";
	    constants_info_[753].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[753].offset = 0;
	    constants_info_[753].data_size = 20480;
	    constants_info_[753].from_folded = false;
	    constants_info_[753].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[753].shape = {5120};
	    constants_info_[753].stride = {1};
	    constants_info_[753].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[753].original_fqn = "model.audio_tower.layers.27.fc1.bias";
	    constants_info_[754].name = "model_audio_tower_layers_27_fc1_parametrizations_weight_original0";
	    constants_info_[754].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[754].offset = 0;
	    constants_info_[754].data_size = 6553600;
	    constants_info_[754].from_folded = false;
	    constants_info_[754].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[754].shape = {5120, 1280};
	    constants_info_[754].stride = {1280, 1};
	    constants_info_[754].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[754].original_fqn = "model.audio_tower.layers.27.fc1.parametrizations.weight.original0";
	    constants_info_[755].name = "model_audio_tower_layers_27_fc1_parametrizations_weight_original1";
	    constants_info_[755].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[755].offset = 0;
	    constants_info_[755].data_size = 819200;
	    constants_info_[755].from_folded = false;
	    constants_info_[755].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[755].shape = {5120, 40};
	    constants_info_[755].stride = {40, 1};
	    constants_info_[755].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[755].original_fqn = "model.audio_tower.layers.27.fc1.parametrizations.weight.original1";
	    constants_info_[756].name = "model_audio_tower_layers_27_fc1_parametrizations_weight_original2";
	    constants_info_[756].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[756].offset = 0;
	    constants_info_[756].data_size = 204800;
	    constants_info_[756].from_folded = false;
	    constants_info_[756].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[756].shape = {5120, 40};
	    constants_info_[756].stride = {40, 1};
	    constants_info_[756].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[756].original_fqn = "model.audio_tower.layers.27.fc1.parametrizations.weight.original2";
	    constants_info_[757].name = "model_audio_tower_layers_27_fc2_bias";
	    constants_info_[757].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[757].offset = 0;
	    constants_info_[757].data_size = 5120;
	    constants_info_[757].from_folded = false;
	    constants_info_[757].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[757].shape = {1280};
	    constants_info_[757].stride = {1};
	    constants_info_[757].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[757].original_fqn = "model.audio_tower.layers.27.fc2.bias";
	    constants_info_[758].name = "model_audio_tower_layers_27_fc2_parametrizations_weight_original0";
	    constants_info_[758].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[758].offset = 0;
	    constants_info_[758].data_size = 6553600;
	    constants_info_[758].from_folded = false;
	    constants_info_[758].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[758].shape = {1280, 5120};
	    constants_info_[758].stride = {5120, 1};
	    constants_info_[758].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[758].original_fqn = "model.audio_tower.layers.27.fc2.parametrizations.weight.original0";
	    constants_info_[759].name = "model_audio_tower_layers_27_fc2_parametrizations_weight_original1";
	    constants_info_[759].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[759].offset = 0;
	    constants_info_[759].data_size = 819200;
	    constants_info_[759].from_folded = false;
	    constants_info_[759].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[759].shape = {1280, 160};
	    constants_info_[759].stride = {160, 1};
	    constants_info_[759].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[759].original_fqn = "model.audio_tower.layers.27.fc2.parametrizations.weight.original1";
	    constants_info_[760].name = "model_audio_tower_layers_27_fc2_parametrizations_weight_original2";
	    constants_info_[760].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[760].offset = 0;
	    constants_info_[760].data_size = 204800;
	    constants_info_[760].from_folded = false;
	    constants_info_[760].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[760].shape = {1280, 160};
	    constants_info_[760].stride = {160, 1};
	    constants_info_[760].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[760].original_fqn = "model.audio_tower.layers.27.fc2.parametrizations.weight.original2";
	    constants_info_[761].name = "model_audio_tower_layers_28_self_attn_layer_norm_weight";
	    constants_info_[761].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[761].offset = 0;
	    constants_info_[761].data_size = 5120;
	    constants_info_[761].from_folded = false;
	    constants_info_[761].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[761].shape = {1280};
	    constants_info_[761].stride = {1};
	    constants_info_[761].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[761].original_fqn = "model.audio_tower.layers.28.self_attn_layer_norm.weight";
	    constants_info_[762].name = "model_audio_tower_layers_28_self_attn_layer_norm_bias";
	    constants_info_[762].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[762].offset = 0;
	    constants_info_[762].data_size = 5120;
	    constants_info_[762].from_folded = false;
	    constants_info_[762].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[762].shape = {1280};
	    constants_info_[762].stride = {1};
	    constants_info_[762].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[762].original_fqn = "model.audio_tower.layers.28.self_attn_layer_norm.bias";
	    constants_info_[763].name = "model_audio_tower_layers_28_self_attn_q_proj_bias";
	    constants_info_[763].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[763].offset = 0;
	    constants_info_[763].data_size = 5120;
	    constants_info_[763].from_folded = false;
	    constants_info_[763].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[763].shape = {1280};
	    constants_info_[763].stride = {1};
	    constants_info_[763].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[763].original_fqn = "model.audio_tower.layers.28.self_attn.q_proj.bias";
	    constants_info_[764].name = "model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[764].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[764].offset = 0;
	    constants_info_[764].data_size = 1638400;
	    constants_info_[764].from_folded = false;
	    constants_info_[764].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[764].shape = {1280, 1280};
	    constants_info_[764].stride = {1280, 1};
	    constants_info_[764].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[764].original_fqn = "model.audio_tower.layers.28.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[765].name = "model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[765].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[765].offset = 0;
	    constants_info_[765].data_size = 204800;
	    constants_info_[765].from_folded = false;
	    constants_info_[765].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[765].shape = {1280, 40};
	    constants_info_[765].stride = {40, 1};
	    constants_info_[765].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[765].original_fqn = "model.audio_tower.layers.28.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[766].name = "model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[766].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[766].offset = 0;
	    constants_info_[766].data_size = 51200;
	    constants_info_[766].from_folded = false;
	    constants_info_[766].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[766].shape = {1280, 40};
	    constants_info_[766].stride = {40, 1};
	    constants_info_[766].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[766].original_fqn = "model.audio_tower.layers.28.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[767].name = "model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[767].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[767].offset = 0;
	    constants_info_[767].data_size = 1638400;
	    constants_info_[767].from_folded = false;
	    constants_info_[767].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[767].shape = {1280, 1280};
	    constants_info_[767].stride = {1280, 1};
	    constants_info_[767].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[767].original_fqn = "model.audio_tower.layers.28.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[768].name = "model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[768].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[768].offset = 0;
	    constants_info_[768].data_size = 204800;
	    constants_info_[768].from_folded = false;
	    constants_info_[768].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[768].shape = {1280, 40};
	    constants_info_[768].stride = {40, 1};
	    constants_info_[768].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[768].original_fqn = "model.audio_tower.layers.28.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[769].name = "model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[769].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[769].offset = 0;
	    constants_info_[769].data_size = 51200;
	    constants_info_[769].from_folded = false;
	    constants_info_[769].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[769].shape = {1280, 40};
	    constants_info_[769].stride = {40, 1};
	    constants_info_[769].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[769].original_fqn = "model.audio_tower.layers.28.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[770].name = "model_audio_tower_layers_28_self_attn_v_proj_bias";
	    constants_info_[770].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[770].offset = 0;
	    constants_info_[770].data_size = 5120;
	    constants_info_[770].from_folded = false;
	    constants_info_[770].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[770].shape = {1280};
	    constants_info_[770].stride = {1};
	    constants_info_[770].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[770].original_fqn = "model.audio_tower.layers.28.self_attn.v_proj.bias";
	    constants_info_[771].name = "model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[771].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[771].offset = 0;
	    constants_info_[771].data_size = 1638400;
	    constants_info_[771].from_folded = false;
	    constants_info_[771].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[771].shape = {1280, 1280};
	    constants_info_[771].stride = {1280, 1};
	    constants_info_[771].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[771].original_fqn = "model.audio_tower.layers.28.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[772].name = "model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[772].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[772].offset = 0;
	    constants_info_[772].data_size = 204800;
	    constants_info_[772].from_folded = false;
	    constants_info_[772].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[772].shape = {1280, 40};
	    constants_info_[772].stride = {40, 1};
	    constants_info_[772].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[772].original_fqn = "model.audio_tower.layers.28.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[773].name = "model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[773].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[773].offset = 0;
	    constants_info_[773].data_size = 51200;
	    constants_info_[773].from_folded = false;
	    constants_info_[773].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[773].shape = {1280, 40};
	    constants_info_[773].stride = {40, 1};
	    constants_info_[773].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[773].original_fqn = "model.audio_tower.layers.28.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[774].name = "model_audio_tower_layers_28_self_attn_out_proj_bias";
	    constants_info_[774].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[774].offset = 0;
	    constants_info_[774].data_size = 5120;
	    constants_info_[774].from_folded = false;
	    constants_info_[774].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[774].shape = {1280};
	    constants_info_[774].stride = {1};
	    constants_info_[774].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[774].original_fqn = "model.audio_tower.layers.28.self_attn.out_proj.bias";
	    constants_info_[775].name = "model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[775].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[775].offset = 0;
	    constants_info_[775].data_size = 1638400;
	    constants_info_[775].from_folded = false;
	    constants_info_[775].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[775].shape = {1280, 1280};
	    constants_info_[775].stride = {1280, 1};
	    constants_info_[775].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[775].original_fqn = "model.audio_tower.layers.28.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[776].name = "model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[776].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[776].offset = 0;
	    constants_info_[776].data_size = 204800;
	    constants_info_[776].from_folded = false;
	    constants_info_[776].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[776].shape = {1280, 40};
	    constants_info_[776].stride = {40, 1};
	    constants_info_[776].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[776].original_fqn = "model.audio_tower.layers.28.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[777].name = "model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[777].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[777].offset = 0;
	    constants_info_[777].data_size = 51200;
	    constants_info_[777].from_folded = false;
	    constants_info_[777].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[777].shape = {1280, 40};
	    constants_info_[777].stride = {40, 1};
	    constants_info_[777].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[777].original_fqn = "model.audio_tower.layers.28.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[778].name = "model_audio_tower_layers_28_final_layer_norm_weight";
	    constants_info_[778].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[778].offset = 0;
	    constants_info_[778].data_size = 5120;
	    constants_info_[778].from_folded = false;
	    constants_info_[778].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[778].shape = {1280};
	    constants_info_[778].stride = {1};
	    constants_info_[778].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[778].original_fqn = "model.audio_tower.layers.28.final_layer_norm.weight";
	    constants_info_[779].name = "model_audio_tower_layers_28_final_layer_norm_bias";
	    constants_info_[779].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[779].offset = 0;
	    constants_info_[779].data_size = 5120;
	    constants_info_[779].from_folded = false;
	    constants_info_[779].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[779].shape = {1280};
	    constants_info_[779].stride = {1};
	    constants_info_[779].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[779].original_fqn = "model.audio_tower.layers.28.final_layer_norm.bias";
	    constants_info_[780].name = "model_audio_tower_layers_28_fc1_bias";
	    constants_info_[780].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[780].offset = 0;
	    constants_info_[780].data_size = 20480;
	    constants_info_[780].from_folded = false;
	    constants_info_[780].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[780].shape = {5120};
	    constants_info_[780].stride = {1};
	    constants_info_[780].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[780].original_fqn = "model.audio_tower.layers.28.fc1.bias";
	    constants_info_[781].name = "model_audio_tower_layers_28_fc1_parametrizations_weight_original0";
	    constants_info_[781].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[781].offset = 0;
	    constants_info_[781].data_size = 6553600;
	    constants_info_[781].from_folded = false;
	    constants_info_[781].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[781].shape = {5120, 1280};
	    constants_info_[781].stride = {1280, 1};
	    constants_info_[781].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[781].original_fqn = "model.audio_tower.layers.28.fc1.parametrizations.weight.original0";
	    constants_info_[782].name = "model_audio_tower_layers_28_fc1_parametrizations_weight_original1";
	    constants_info_[782].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[782].offset = 0;
	    constants_info_[782].data_size = 819200;
	    constants_info_[782].from_folded = false;
	    constants_info_[782].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[782].shape = {5120, 40};
	    constants_info_[782].stride = {40, 1};
	    constants_info_[782].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[782].original_fqn = "model.audio_tower.layers.28.fc1.parametrizations.weight.original1";
	    constants_info_[783].name = "model_audio_tower_layers_28_fc1_parametrizations_weight_original2";
	    constants_info_[783].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[783].offset = 0;
	    constants_info_[783].data_size = 204800;
	    constants_info_[783].from_folded = false;
	    constants_info_[783].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[783].shape = {5120, 40};
	    constants_info_[783].stride = {40, 1};
	    constants_info_[783].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[783].original_fqn = "model.audio_tower.layers.28.fc1.parametrizations.weight.original2";
	    constants_info_[784].name = "model_audio_tower_layers_28_fc2_bias";
	    constants_info_[784].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[784].offset = 0;
	    constants_info_[784].data_size = 5120;
	    constants_info_[784].from_folded = false;
	    constants_info_[784].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[784].shape = {1280};
	    constants_info_[784].stride = {1};
	    constants_info_[784].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[784].original_fqn = "model.audio_tower.layers.28.fc2.bias";
	    constants_info_[785].name = "model_audio_tower_layers_28_fc2_parametrizations_weight_original0";
	    constants_info_[785].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[785].offset = 0;
	    constants_info_[785].data_size = 6553600;
	    constants_info_[785].from_folded = false;
	    constants_info_[785].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[785].shape = {1280, 5120};
	    constants_info_[785].stride = {5120, 1};
	    constants_info_[785].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[785].original_fqn = "model.audio_tower.layers.28.fc2.parametrizations.weight.original0";
	    constants_info_[786].name = "model_audio_tower_layers_28_fc2_parametrizations_weight_original1";
	    constants_info_[786].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[786].offset = 0;
	    constants_info_[786].data_size = 819200;
	    constants_info_[786].from_folded = false;
	    constants_info_[786].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[786].shape = {1280, 160};
	    constants_info_[786].stride = {160, 1};
	    constants_info_[786].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[786].original_fqn = "model.audio_tower.layers.28.fc2.parametrizations.weight.original1";
	    constants_info_[787].name = "model_audio_tower_layers_28_fc2_parametrizations_weight_original2";
	    constants_info_[787].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[787].offset = 0;
	    constants_info_[787].data_size = 204800;
	    constants_info_[787].from_folded = false;
	    constants_info_[787].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[787].shape = {1280, 160};
	    constants_info_[787].stride = {160, 1};
	    constants_info_[787].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[787].original_fqn = "model.audio_tower.layers.28.fc2.parametrizations.weight.original2";
	    constants_info_[788].name = "model_audio_tower_layers_29_self_attn_layer_norm_weight";
	    constants_info_[788].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[788].offset = 0;
	    constants_info_[788].data_size = 5120;
	    constants_info_[788].from_folded = false;
	    constants_info_[788].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[788].shape = {1280};
	    constants_info_[788].stride = {1};
	    constants_info_[788].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[788].original_fqn = "model.audio_tower.layers.29.self_attn_layer_norm.weight";
	    constants_info_[789].name = "model_audio_tower_layers_29_self_attn_layer_norm_bias";
	    constants_info_[789].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[789].offset = 0;
	    constants_info_[789].data_size = 5120;
	    constants_info_[789].from_folded = false;
	    constants_info_[789].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[789].shape = {1280};
	    constants_info_[789].stride = {1};
	    constants_info_[789].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[789].original_fqn = "model.audio_tower.layers.29.self_attn_layer_norm.bias";
	    constants_info_[790].name = "model_audio_tower_layers_29_self_attn_q_proj_bias";
	    constants_info_[790].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[790].offset = 0;
	    constants_info_[790].data_size = 5120;
	    constants_info_[790].from_folded = false;
	    constants_info_[790].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[790].shape = {1280};
	    constants_info_[790].stride = {1};
	    constants_info_[790].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[790].original_fqn = "model.audio_tower.layers.29.self_attn.q_proj.bias";
	    constants_info_[791].name = "model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[791].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[791].offset = 0;
	    constants_info_[791].data_size = 1638400;
	    constants_info_[791].from_folded = false;
	    constants_info_[791].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[791].shape = {1280, 1280};
	    constants_info_[791].stride = {1280, 1};
	    constants_info_[791].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[791].original_fqn = "model.audio_tower.layers.29.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[792].name = "model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[792].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[792].offset = 0;
	    constants_info_[792].data_size = 204800;
	    constants_info_[792].from_folded = false;
	    constants_info_[792].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[792].shape = {1280, 40};
	    constants_info_[792].stride = {40, 1};
	    constants_info_[792].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[792].original_fqn = "model.audio_tower.layers.29.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[793].name = "model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[793].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[793].offset = 0;
	    constants_info_[793].data_size = 51200;
	    constants_info_[793].from_folded = false;
	    constants_info_[793].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[793].shape = {1280, 40};
	    constants_info_[793].stride = {40, 1};
	    constants_info_[793].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[793].original_fqn = "model.audio_tower.layers.29.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[794].name = "model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[794].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[794].offset = 0;
	    constants_info_[794].data_size = 1638400;
	    constants_info_[794].from_folded = false;
	    constants_info_[794].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[794].shape = {1280, 1280};
	    constants_info_[794].stride = {1280, 1};
	    constants_info_[794].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[794].original_fqn = "model.audio_tower.layers.29.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[795].name = "model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[795].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[795].offset = 0;
	    constants_info_[795].data_size = 204800;
	    constants_info_[795].from_folded = false;
	    constants_info_[795].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[795].shape = {1280, 40};
	    constants_info_[795].stride = {40, 1};
	    constants_info_[795].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[795].original_fqn = "model.audio_tower.layers.29.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[796].name = "model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[796].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[796].offset = 0;
	    constants_info_[796].data_size = 51200;
	    constants_info_[796].from_folded = false;
	    constants_info_[796].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[796].shape = {1280, 40};
	    constants_info_[796].stride = {40, 1};
	    constants_info_[796].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[796].original_fqn = "model.audio_tower.layers.29.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[797].name = "model_audio_tower_layers_29_self_attn_v_proj_bias";
	    constants_info_[797].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[797].offset = 0;
	    constants_info_[797].data_size = 5120;
	    constants_info_[797].from_folded = false;
	    constants_info_[797].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[797].shape = {1280};
	    constants_info_[797].stride = {1};
	    constants_info_[797].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[797].original_fqn = "model.audio_tower.layers.29.self_attn.v_proj.bias";
	    constants_info_[798].name = "model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[798].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[798].offset = 0;
	    constants_info_[798].data_size = 1638400;
	    constants_info_[798].from_folded = false;
	    constants_info_[798].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[798].shape = {1280, 1280};
	    constants_info_[798].stride = {1280, 1};
	    constants_info_[798].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[798].original_fqn = "model.audio_tower.layers.29.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[799].name = "model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[799].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[799].offset = 0;
	    constants_info_[799].data_size = 204800;
	    constants_info_[799].from_folded = false;
	    constants_info_[799].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[799].shape = {1280, 40};
	    constants_info_[799].stride = {40, 1};
	    constants_info_[799].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[799].original_fqn = "model.audio_tower.layers.29.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[800].name = "model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[800].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[800].offset = 0;
	    constants_info_[800].data_size = 51200;
	    constants_info_[800].from_folded = false;
	    constants_info_[800].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[800].shape = {1280, 40};
	    constants_info_[800].stride = {40, 1};
	    constants_info_[800].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[800].original_fqn = "model.audio_tower.layers.29.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[801].name = "model_audio_tower_layers_29_self_attn_out_proj_bias";
	    constants_info_[801].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[801].offset = 0;
	    constants_info_[801].data_size = 5120;
	    constants_info_[801].from_folded = false;
	    constants_info_[801].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[801].shape = {1280};
	    constants_info_[801].stride = {1};
	    constants_info_[801].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[801].original_fqn = "model.audio_tower.layers.29.self_attn.out_proj.bias";
	    constants_info_[802].name = "model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[802].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[802].offset = 0;
	    constants_info_[802].data_size = 1638400;
	    constants_info_[802].from_folded = false;
	    constants_info_[802].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[802].shape = {1280, 1280};
	    constants_info_[802].stride = {1280, 1};
	    constants_info_[802].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[802].original_fqn = "model.audio_tower.layers.29.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[803].name = "model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[803].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[803].offset = 0;
	    constants_info_[803].data_size = 204800;
	    constants_info_[803].from_folded = false;
	    constants_info_[803].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[803].shape = {1280, 40};
	    constants_info_[803].stride = {40, 1};
	    constants_info_[803].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[803].original_fqn = "model.audio_tower.layers.29.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[804].name = "model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[804].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[804].offset = 0;
	    constants_info_[804].data_size = 51200;
	    constants_info_[804].from_folded = false;
	    constants_info_[804].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[804].shape = {1280, 40};
	    constants_info_[804].stride = {40, 1};
	    constants_info_[804].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[804].original_fqn = "model.audio_tower.layers.29.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[805].name = "model_audio_tower_layers_29_final_layer_norm_weight";
	    constants_info_[805].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[805].offset = 0;
	    constants_info_[805].data_size = 5120;
	    constants_info_[805].from_folded = false;
	    constants_info_[805].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[805].shape = {1280};
	    constants_info_[805].stride = {1};
	    constants_info_[805].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[805].original_fqn = "model.audio_tower.layers.29.final_layer_norm.weight";
	    constants_info_[806].name = "model_audio_tower_layers_29_final_layer_norm_bias";
	    constants_info_[806].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[806].offset = 0;
	    constants_info_[806].data_size = 5120;
	    constants_info_[806].from_folded = false;
	    constants_info_[806].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[806].shape = {1280};
	    constants_info_[806].stride = {1};
	    constants_info_[806].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[806].original_fqn = "model.audio_tower.layers.29.final_layer_norm.bias";
	    constants_info_[807].name = "model_audio_tower_layers_29_fc1_bias";
	    constants_info_[807].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[807].offset = 0;
	    constants_info_[807].data_size = 20480;
	    constants_info_[807].from_folded = false;
	    constants_info_[807].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[807].shape = {5120};
	    constants_info_[807].stride = {1};
	    constants_info_[807].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[807].original_fqn = "model.audio_tower.layers.29.fc1.bias";
	    constants_info_[808].name = "model_audio_tower_layers_29_fc1_parametrizations_weight_original0";
	    constants_info_[808].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[808].offset = 0;
	    constants_info_[808].data_size = 6553600;
	    constants_info_[808].from_folded = false;
	    constants_info_[808].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[808].shape = {5120, 1280};
	    constants_info_[808].stride = {1280, 1};
	    constants_info_[808].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[808].original_fqn = "model.audio_tower.layers.29.fc1.parametrizations.weight.original0";
	    constants_info_[809].name = "model_audio_tower_layers_29_fc1_parametrizations_weight_original1";
	    constants_info_[809].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[809].offset = 0;
	    constants_info_[809].data_size = 819200;
	    constants_info_[809].from_folded = false;
	    constants_info_[809].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[809].shape = {5120, 40};
	    constants_info_[809].stride = {40, 1};
	    constants_info_[809].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[809].original_fqn = "model.audio_tower.layers.29.fc1.parametrizations.weight.original1";
	    constants_info_[810].name = "model_audio_tower_layers_29_fc1_parametrizations_weight_original2";
	    constants_info_[810].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[810].offset = 0;
	    constants_info_[810].data_size = 204800;
	    constants_info_[810].from_folded = false;
	    constants_info_[810].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[810].shape = {5120, 40};
	    constants_info_[810].stride = {40, 1};
	    constants_info_[810].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[810].original_fqn = "model.audio_tower.layers.29.fc1.parametrizations.weight.original2";
	    constants_info_[811].name = "model_audio_tower_layers_29_fc2_bias";
	    constants_info_[811].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[811].offset = 0;
	    constants_info_[811].data_size = 5120;
	    constants_info_[811].from_folded = false;
	    constants_info_[811].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[811].shape = {1280};
	    constants_info_[811].stride = {1};
	    constants_info_[811].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[811].original_fqn = "model.audio_tower.layers.29.fc2.bias";
	    constants_info_[812].name = "model_audio_tower_layers_29_fc2_parametrizations_weight_original0";
	    constants_info_[812].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[812].offset = 0;
	    constants_info_[812].data_size = 6553600;
	    constants_info_[812].from_folded = false;
	    constants_info_[812].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[812].shape = {1280, 5120};
	    constants_info_[812].stride = {5120, 1};
	    constants_info_[812].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[812].original_fqn = "model.audio_tower.layers.29.fc2.parametrizations.weight.original0";
	    constants_info_[813].name = "model_audio_tower_layers_29_fc2_parametrizations_weight_original1";
	    constants_info_[813].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[813].offset = 0;
	    constants_info_[813].data_size = 819200;
	    constants_info_[813].from_folded = false;
	    constants_info_[813].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[813].shape = {1280, 160};
	    constants_info_[813].stride = {160, 1};
	    constants_info_[813].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[813].original_fqn = "model.audio_tower.layers.29.fc2.parametrizations.weight.original1";
	    constants_info_[814].name = "model_audio_tower_layers_29_fc2_parametrizations_weight_original2";
	    constants_info_[814].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[814].offset = 0;
	    constants_info_[814].data_size = 204800;
	    constants_info_[814].from_folded = false;
	    constants_info_[814].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[814].shape = {1280, 160};
	    constants_info_[814].stride = {160, 1};
	    constants_info_[814].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[814].original_fqn = "model.audio_tower.layers.29.fc2.parametrizations.weight.original2";
	    constants_info_[815].name = "model_audio_tower_layers_30_self_attn_layer_norm_weight";
	    constants_info_[815].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[815].offset = 0;
	    constants_info_[815].data_size = 5120;
	    constants_info_[815].from_folded = false;
	    constants_info_[815].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[815].shape = {1280};
	    constants_info_[815].stride = {1};
	    constants_info_[815].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[815].original_fqn = "model.audio_tower.layers.30.self_attn_layer_norm.weight";
	    constants_info_[816].name = "model_audio_tower_layers_30_self_attn_layer_norm_bias";
	    constants_info_[816].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[816].offset = 0;
	    constants_info_[816].data_size = 5120;
	    constants_info_[816].from_folded = false;
	    constants_info_[816].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[816].shape = {1280};
	    constants_info_[816].stride = {1};
	    constants_info_[816].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[816].original_fqn = "model.audio_tower.layers.30.self_attn_layer_norm.bias";
	    constants_info_[817].name = "model_audio_tower_layers_30_self_attn_q_proj_bias";
	    constants_info_[817].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[817].offset = 0;
	    constants_info_[817].data_size = 5120;
	    constants_info_[817].from_folded = false;
	    constants_info_[817].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[817].shape = {1280};
	    constants_info_[817].stride = {1};
	    constants_info_[817].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[817].original_fqn = "model.audio_tower.layers.30.self_attn.q_proj.bias";
	    constants_info_[818].name = "model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[818].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[818].offset = 0;
	    constants_info_[818].data_size = 1638400;
	    constants_info_[818].from_folded = false;
	    constants_info_[818].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[818].shape = {1280, 1280};
	    constants_info_[818].stride = {1280, 1};
	    constants_info_[818].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[818].original_fqn = "model.audio_tower.layers.30.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[819].name = "model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[819].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[819].offset = 0;
	    constants_info_[819].data_size = 204800;
	    constants_info_[819].from_folded = false;
	    constants_info_[819].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[819].shape = {1280, 40};
	    constants_info_[819].stride = {40, 1};
	    constants_info_[819].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[819].original_fqn = "model.audio_tower.layers.30.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[820].name = "model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[820].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[820].offset = 0;
	    constants_info_[820].data_size = 51200;
	    constants_info_[820].from_folded = false;
	    constants_info_[820].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[820].shape = {1280, 40};
	    constants_info_[820].stride = {40, 1};
	    constants_info_[820].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[820].original_fqn = "model.audio_tower.layers.30.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[821].name = "model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[821].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[821].offset = 0;
	    constants_info_[821].data_size = 1638400;
	    constants_info_[821].from_folded = false;
	    constants_info_[821].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[821].shape = {1280, 1280};
	    constants_info_[821].stride = {1280, 1};
	    constants_info_[821].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[821].original_fqn = "model.audio_tower.layers.30.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[822].name = "model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[822].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[822].offset = 0;
	    constants_info_[822].data_size = 204800;
	    constants_info_[822].from_folded = false;
	    constants_info_[822].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[822].shape = {1280, 40};
	    constants_info_[822].stride = {40, 1};
	    constants_info_[822].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[822].original_fqn = "model.audio_tower.layers.30.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[823].name = "model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[823].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[823].offset = 0;
	    constants_info_[823].data_size = 51200;
	    constants_info_[823].from_folded = false;
	    constants_info_[823].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[823].shape = {1280, 40};
	    constants_info_[823].stride = {40, 1};
	    constants_info_[823].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[823].original_fqn = "model.audio_tower.layers.30.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[824].name = "model_audio_tower_layers_30_self_attn_v_proj_bias";
	    constants_info_[824].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[824].offset = 0;
	    constants_info_[824].data_size = 5120;
	    constants_info_[824].from_folded = false;
	    constants_info_[824].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[824].shape = {1280};
	    constants_info_[824].stride = {1};
	    constants_info_[824].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[824].original_fqn = "model.audio_tower.layers.30.self_attn.v_proj.bias";
	    constants_info_[825].name = "model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[825].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[825].offset = 0;
	    constants_info_[825].data_size = 1638400;
	    constants_info_[825].from_folded = false;
	    constants_info_[825].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[825].shape = {1280, 1280};
	    constants_info_[825].stride = {1280, 1};
	    constants_info_[825].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[825].original_fqn = "model.audio_tower.layers.30.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[826].name = "model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[826].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[826].offset = 0;
	    constants_info_[826].data_size = 204800;
	    constants_info_[826].from_folded = false;
	    constants_info_[826].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[826].shape = {1280, 40};
	    constants_info_[826].stride = {40, 1};
	    constants_info_[826].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[826].original_fqn = "model.audio_tower.layers.30.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[827].name = "model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[827].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[827].offset = 0;
	    constants_info_[827].data_size = 51200;
	    constants_info_[827].from_folded = false;
	    constants_info_[827].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[827].shape = {1280, 40};
	    constants_info_[827].stride = {40, 1};
	    constants_info_[827].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[827].original_fqn = "model.audio_tower.layers.30.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[828].name = "model_audio_tower_layers_30_self_attn_out_proj_bias";
	    constants_info_[828].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[828].offset = 0;
	    constants_info_[828].data_size = 5120;
	    constants_info_[828].from_folded = false;
	    constants_info_[828].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[828].shape = {1280};
	    constants_info_[828].stride = {1};
	    constants_info_[828].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[828].original_fqn = "model.audio_tower.layers.30.self_attn.out_proj.bias";
	    constants_info_[829].name = "model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[829].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[829].offset = 0;
	    constants_info_[829].data_size = 1638400;
	    constants_info_[829].from_folded = false;
	    constants_info_[829].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[829].shape = {1280, 1280};
	    constants_info_[829].stride = {1280, 1};
	    constants_info_[829].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[829].original_fqn = "model.audio_tower.layers.30.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[830].name = "model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[830].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[830].offset = 0;
	    constants_info_[830].data_size = 204800;
	    constants_info_[830].from_folded = false;
	    constants_info_[830].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[830].shape = {1280, 40};
	    constants_info_[830].stride = {40, 1};
	    constants_info_[830].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[830].original_fqn = "model.audio_tower.layers.30.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[831].name = "model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[831].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[831].offset = 0;
	    constants_info_[831].data_size = 51200;
	    constants_info_[831].from_folded = false;
	    constants_info_[831].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[831].shape = {1280, 40};
	    constants_info_[831].stride = {40, 1};
	    constants_info_[831].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[831].original_fqn = "model.audio_tower.layers.30.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[832].name = "model_audio_tower_layers_30_final_layer_norm_weight";
	    constants_info_[832].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[832].offset = 0;
	    constants_info_[832].data_size = 5120;
	    constants_info_[832].from_folded = false;
	    constants_info_[832].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[832].shape = {1280};
	    constants_info_[832].stride = {1};
	    constants_info_[832].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[832].original_fqn = "model.audio_tower.layers.30.final_layer_norm.weight";
	    constants_info_[833].name = "model_audio_tower_layers_30_final_layer_norm_bias";
	    constants_info_[833].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[833].offset = 0;
	    constants_info_[833].data_size = 5120;
	    constants_info_[833].from_folded = false;
	    constants_info_[833].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[833].shape = {1280};
	    constants_info_[833].stride = {1};
	    constants_info_[833].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[833].original_fqn = "model.audio_tower.layers.30.final_layer_norm.bias";
	    constants_info_[834].name = "model_audio_tower_layers_30_fc1_bias";
	    constants_info_[834].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[834].offset = 0;
	    constants_info_[834].data_size = 20480;
	    constants_info_[834].from_folded = false;
	    constants_info_[834].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[834].shape = {5120};
	    constants_info_[834].stride = {1};
	    constants_info_[834].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[834].original_fqn = "model.audio_tower.layers.30.fc1.bias";
	    constants_info_[835].name = "model_audio_tower_layers_30_fc1_parametrizations_weight_original0";
	    constants_info_[835].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[835].offset = 0;
	    constants_info_[835].data_size = 6553600;
	    constants_info_[835].from_folded = false;
	    constants_info_[835].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[835].shape = {5120, 1280};
	    constants_info_[835].stride = {1280, 1};
	    constants_info_[835].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[835].original_fqn = "model.audio_tower.layers.30.fc1.parametrizations.weight.original0";
	    constants_info_[836].name = "model_audio_tower_layers_30_fc1_parametrizations_weight_original1";
	    constants_info_[836].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[836].offset = 0;
	    constants_info_[836].data_size = 819200;
	    constants_info_[836].from_folded = false;
	    constants_info_[836].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[836].shape = {5120, 40};
	    constants_info_[836].stride = {40, 1};
	    constants_info_[836].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[836].original_fqn = "model.audio_tower.layers.30.fc1.parametrizations.weight.original1";
	    constants_info_[837].name = "model_audio_tower_layers_30_fc1_parametrizations_weight_original2";
	    constants_info_[837].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[837].offset = 0;
	    constants_info_[837].data_size = 204800;
	    constants_info_[837].from_folded = false;
	    constants_info_[837].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[837].shape = {5120, 40};
	    constants_info_[837].stride = {40, 1};
	    constants_info_[837].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[837].original_fqn = "model.audio_tower.layers.30.fc1.parametrizations.weight.original2";
	    constants_info_[838].name = "model_audio_tower_layers_30_fc2_bias";
	    constants_info_[838].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[838].offset = 0;
	    constants_info_[838].data_size = 5120;
	    constants_info_[838].from_folded = false;
	    constants_info_[838].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[838].shape = {1280};
	    constants_info_[838].stride = {1};
	    constants_info_[838].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[838].original_fqn = "model.audio_tower.layers.30.fc2.bias";
	    constants_info_[839].name = "model_audio_tower_layers_30_fc2_parametrizations_weight_original0";
	    constants_info_[839].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[839].offset = 0;
	    constants_info_[839].data_size = 6553600;
	    constants_info_[839].from_folded = false;
	    constants_info_[839].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[839].shape = {1280, 5120};
	    constants_info_[839].stride = {5120, 1};
	    constants_info_[839].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[839].original_fqn = "model.audio_tower.layers.30.fc2.parametrizations.weight.original0";
	    constants_info_[840].name = "model_audio_tower_layers_30_fc2_parametrizations_weight_original1";
	    constants_info_[840].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[840].offset = 0;
	    constants_info_[840].data_size = 819200;
	    constants_info_[840].from_folded = false;
	    constants_info_[840].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[840].shape = {1280, 160};
	    constants_info_[840].stride = {160, 1};
	    constants_info_[840].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[840].original_fqn = "model.audio_tower.layers.30.fc2.parametrizations.weight.original1";
	    constants_info_[841].name = "model_audio_tower_layers_30_fc2_parametrizations_weight_original2";
	    constants_info_[841].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[841].offset = 0;
	    constants_info_[841].data_size = 204800;
	    constants_info_[841].from_folded = false;
	    constants_info_[841].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[841].shape = {1280, 160};
	    constants_info_[841].stride = {160, 1};
	    constants_info_[841].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[841].original_fqn = "model.audio_tower.layers.30.fc2.parametrizations.weight.original2";
	    constants_info_[842].name = "model_audio_tower_layers_31_self_attn_layer_norm_weight";
	    constants_info_[842].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[842].offset = 0;
	    constants_info_[842].data_size = 5120;
	    constants_info_[842].from_folded = false;
	    constants_info_[842].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[842].shape = {1280};
	    constants_info_[842].stride = {1};
	    constants_info_[842].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[842].original_fqn = "model.audio_tower.layers.31.self_attn_layer_norm.weight";
	    constants_info_[843].name = "model_audio_tower_layers_31_self_attn_layer_norm_bias";
	    constants_info_[843].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[843].offset = 0;
	    constants_info_[843].data_size = 5120;
	    constants_info_[843].from_folded = false;
	    constants_info_[843].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[843].shape = {1280};
	    constants_info_[843].stride = {1};
	    constants_info_[843].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[843].original_fqn = "model.audio_tower.layers.31.self_attn_layer_norm.bias";
	    constants_info_[844].name = "model_audio_tower_layers_31_self_attn_q_proj_bias";
	    constants_info_[844].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[844].offset = 0;
	    constants_info_[844].data_size = 5120;
	    constants_info_[844].from_folded = false;
	    constants_info_[844].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[844].shape = {1280};
	    constants_info_[844].stride = {1};
	    constants_info_[844].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[844].original_fqn = "model.audio_tower.layers.31.self_attn.q_proj.bias";
	    constants_info_[845].name = "model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0";
	    constants_info_[845].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[845].offset = 0;
	    constants_info_[845].data_size = 1638400;
	    constants_info_[845].from_folded = false;
	    constants_info_[845].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[845].shape = {1280, 1280};
	    constants_info_[845].stride = {1280, 1};
	    constants_info_[845].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[845].original_fqn = "model.audio_tower.layers.31.self_attn.q_proj.parametrizations.weight.original0";
	    constants_info_[846].name = "model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1";
	    constants_info_[846].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[846].offset = 0;
	    constants_info_[846].data_size = 204800;
	    constants_info_[846].from_folded = false;
	    constants_info_[846].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[846].shape = {1280, 40};
	    constants_info_[846].stride = {40, 1};
	    constants_info_[846].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[846].original_fqn = "model.audio_tower.layers.31.self_attn.q_proj.parametrizations.weight.original1";
	    constants_info_[847].name = "model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2";
	    constants_info_[847].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[847].offset = 0;
	    constants_info_[847].data_size = 51200;
	    constants_info_[847].from_folded = false;
	    constants_info_[847].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[847].shape = {1280, 40};
	    constants_info_[847].stride = {40, 1};
	    constants_info_[847].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[847].original_fqn = "model.audio_tower.layers.31.self_attn.q_proj.parametrizations.weight.original2";
	    constants_info_[848].name = "model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0";
	    constants_info_[848].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[848].offset = 0;
	    constants_info_[848].data_size = 1638400;
	    constants_info_[848].from_folded = false;
	    constants_info_[848].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[848].shape = {1280, 1280};
	    constants_info_[848].stride = {1280, 1};
	    constants_info_[848].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[848].original_fqn = "model.audio_tower.layers.31.self_attn.k_proj.parametrizations.weight.original0";
	    constants_info_[849].name = "model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1";
	    constants_info_[849].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[849].offset = 0;
	    constants_info_[849].data_size = 204800;
	    constants_info_[849].from_folded = false;
	    constants_info_[849].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[849].shape = {1280, 40};
	    constants_info_[849].stride = {40, 1};
	    constants_info_[849].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[849].original_fqn = "model.audio_tower.layers.31.self_attn.k_proj.parametrizations.weight.original1";
	    constants_info_[850].name = "model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2";
	    constants_info_[850].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[850].offset = 0;
	    constants_info_[850].data_size = 51200;
	    constants_info_[850].from_folded = false;
	    constants_info_[850].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[850].shape = {1280, 40};
	    constants_info_[850].stride = {40, 1};
	    constants_info_[850].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[850].original_fqn = "model.audio_tower.layers.31.self_attn.k_proj.parametrizations.weight.original2";
	    constants_info_[851].name = "model_audio_tower_layers_31_self_attn_v_proj_bias";
	    constants_info_[851].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[851].offset = 0;
	    constants_info_[851].data_size = 5120;
	    constants_info_[851].from_folded = false;
	    constants_info_[851].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[851].shape = {1280};
	    constants_info_[851].stride = {1};
	    constants_info_[851].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[851].original_fqn = "model.audio_tower.layers.31.self_attn.v_proj.bias";
	    constants_info_[852].name = "model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0";
	    constants_info_[852].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[852].offset = 0;
	    constants_info_[852].data_size = 1638400;
	    constants_info_[852].from_folded = false;
	    constants_info_[852].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[852].shape = {1280, 1280};
	    constants_info_[852].stride = {1280, 1};
	    constants_info_[852].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[852].original_fqn = "model.audio_tower.layers.31.self_attn.v_proj.parametrizations.weight.original0";
	    constants_info_[853].name = "model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1";
	    constants_info_[853].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[853].offset = 0;
	    constants_info_[853].data_size = 204800;
	    constants_info_[853].from_folded = false;
	    constants_info_[853].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[853].shape = {1280, 40};
	    constants_info_[853].stride = {40, 1};
	    constants_info_[853].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[853].original_fqn = "model.audio_tower.layers.31.self_attn.v_proj.parametrizations.weight.original1";
	    constants_info_[854].name = "model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2";
	    constants_info_[854].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[854].offset = 0;
	    constants_info_[854].data_size = 51200;
	    constants_info_[854].from_folded = false;
	    constants_info_[854].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[854].shape = {1280, 40};
	    constants_info_[854].stride = {40, 1};
	    constants_info_[854].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[854].original_fqn = "model.audio_tower.layers.31.self_attn.v_proj.parametrizations.weight.original2";
	    constants_info_[855].name = "model_audio_tower_layers_31_self_attn_out_proj_bias";
	    constants_info_[855].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[855].offset = 0;
	    constants_info_[855].data_size = 5120;
	    constants_info_[855].from_folded = false;
	    constants_info_[855].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[855].shape = {1280};
	    constants_info_[855].stride = {1};
	    constants_info_[855].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[855].original_fqn = "model.audio_tower.layers.31.self_attn.out_proj.bias";
	    constants_info_[856].name = "model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0";
	    constants_info_[856].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[856].offset = 0;
	    constants_info_[856].data_size = 1638400;
	    constants_info_[856].from_folded = false;
	    constants_info_[856].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[856].shape = {1280, 1280};
	    constants_info_[856].stride = {1280, 1};
	    constants_info_[856].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[856].original_fqn = "model.audio_tower.layers.31.self_attn.out_proj.parametrizations.weight.original0";
	    constants_info_[857].name = "model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1";
	    constants_info_[857].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[857].offset = 0;
	    constants_info_[857].data_size = 204800;
	    constants_info_[857].from_folded = false;
	    constants_info_[857].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[857].shape = {1280, 40};
	    constants_info_[857].stride = {40, 1};
	    constants_info_[857].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[857].original_fqn = "model.audio_tower.layers.31.self_attn.out_proj.parametrizations.weight.original1";
	    constants_info_[858].name = "model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2";
	    constants_info_[858].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[858].offset = 0;
	    constants_info_[858].data_size = 51200;
	    constants_info_[858].from_folded = false;
	    constants_info_[858].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[858].shape = {1280, 40};
	    constants_info_[858].stride = {40, 1};
	    constants_info_[858].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[858].original_fqn = "model.audio_tower.layers.31.self_attn.out_proj.parametrizations.weight.original2";
	    constants_info_[859].name = "model_audio_tower_layers_31_final_layer_norm_weight";
	    constants_info_[859].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[859].offset = 0;
	    constants_info_[859].data_size = 5120;
	    constants_info_[859].from_folded = false;
	    constants_info_[859].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[859].shape = {1280};
	    constants_info_[859].stride = {1};
	    constants_info_[859].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[859].original_fqn = "model.audio_tower.layers.31.final_layer_norm.weight";
	    constants_info_[860].name = "model_audio_tower_layers_31_final_layer_norm_bias";
	    constants_info_[860].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[860].offset = 0;
	    constants_info_[860].data_size = 5120;
	    constants_info_[860].from_folded = false;
	    constants_info_[860].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[860].shape = {1280};
	    constants_info_[860].stride = {1};
	    constants_info_[860].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[860].original_fqn = "model.audio_tower.layers.31.final_layer_norm.bias";
	    constants_info_[861].name = "model_audio_tower_layers_31_fc1_bias";
	    constants_info_[861].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[861].offset = 0;
	    constants_info_[861].data_size = 20480;
	    constants_info_[861].from_folded = false;
	    constants_info_[861].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[861].shape = {5120};
	    constants_info_[861].stride = {1};
	    constants_info_[861].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[861].original_fqn = "model.audio_tower.layers.31.fc1.bias";
	    constants_info_[862].name = "model_audio_tower_layers_31_fc1_parametrizations_weight_original0";
	    constants_info_[862].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[862].offset = 0;
	    constants_info_[862].data_size = 6553600;
	    constants_info_[862].from_folded = false;
	    constants_info_[862].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[862].shape = {5120, 1280};
	    constants_info_[862].stride = {1280, 1};
	    constants_info_[862].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[862].original_fqn = "model.audio_tower.layers.31.fc1.parametrizations.weight.original0";
	    constants_info_[863].name = "model_audio_tower_layers_31_fc1_parametrizations_weight_original1";
	    constants_info_[863].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[863].offset = 0;
	    constants_info_[863].data_size = 819200;
	    constants_info_[863].from_folded = false;
	    constants_info_[863].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[863].shape = {5120, 40};
	    constants_info_[863].stride = {40, 1};
	    constants_info_[863].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[863].original_fqn = "model.audio_tower.layers.31.fc1.parametrizations.weight.original1";
	    constants_info_[864].name = "model_audio_tower_layers_31_fc1_parametrizations_weight_original2";
	    constants_info_[864].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[864].offset = 0;
	    constants_info_[864].data_size = 204800;
	    constants_info_[864].from_folded = false;
	    constants_info_[864].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[864].shape = {5120, 40};
	    constants_info_[864].stride = {40, 1};
	    constants_info_[864].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[864].original_fqn = "model.audio_tower.layers.31.fc1.parametrizations.weight.original2";
	    constants_info_[865].name = "model_audio_tower_layers_31_fc2_bias";
	    constants_info_[865].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[865].offset = 0;
	    constants_info_[865].data_size = 5120;
	    constants_info_[865].from_folded = false;
	    constants_info_[865].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[865].shape = {1280};
	    constants_info_[865].stride = {1};
	    constants_info_[865].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[865].original_fqn = "model.audio_tower.layers.31.fc2.bias";
	    constants_info_[866].name = "model_audio_tower_layers_31_fc2_parametrizations_weight_original0";
	    constants_info_[866].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[866].offset = 0;
	    constants_info_[866].data_size = 6553600;
	    constants_info_[866].from_folded = false;
	    constants_info_[866].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[866].shape = {1280, 5120};
	    constants_info_[866].stride = {5120, 1};
	    constants_info_[866].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[866].original_fqn = "model.audio_tower.layers.31.fc2.parametrizations.weight.original0";
	    constants_info_[867].name = "model_audio_tower_layers_31_fc2_parametrizations_weight_original1";
	    constants_info_[867].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[867].offset = 0;
	    constants_info_[867].data_size = 819200;
	    constants_info_[867].from_folded = false;
	    constants_info_[867].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[867].shape = {1280, 160};
	    constants_info_[867].stride = {160, 1};
	    constants_info_[867].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[867].original_fqn = "model.audio_tower.layers.31.fc2.parametrizations.weight.original1";
	    constants_info_[868].name = "model_audio_tower_layers_31_fc2_parametrizations_weight_original2";
	    constants_info_[868].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[868].offset = 0;
	    constants_info_[868].data_size = 204800;
	    constants_info_[868].from_folded = false;
	    constants_info_[868].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[868].shape = {1280, 160};
	    constants_info_[868].stride = {160, 1};
	    constants_info_[868].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[868].original_fqn = "model.audio_tower.layers.31.fc2.parametrizations.weight.original2";
	    constants_info_[869].name = "model_audio_tower_layer_norm_weight";
	    constants_info_[869].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[869].offset = 0;
	    constants_info_[869].data_size = 5120;
	    constants_info_[869].from_folded = false;
	    constants_info_[869].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[869].shape = {1280};
	    constants_info_[869].stride = {1};
	    constants_info_[869].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[869].original_fqn = "model.audio_tower.layer_norm.weight";
	    constants_info_[870].name = "model_audio_tower_layer_norm_bias";
	    constants_info_[870].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[870].offset = 0;
	    constants_info_[870].data_size = 5120;
	    constants_info_[870].from_folded = false;
	    constants_info_[870].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[870].shape = {1280};
	    constants_info_[870].stride = {1};
	    constants_info_[870].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[870].original_fqn = "model.audio_tower.layer_norm.bias";
	    constants_info_[871].name = "model_multi_modal_projector_linear_1_parametrizations_weight_original0";
	    constants_info_[871].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[871].offset = 0;
	    constants_info_[871].data_size = 15728640;
	    constants_info_[871].from_folded = false;
	    constants_info_[871].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[871].shape = {3072, 5120};
	    constants_info_[871].stride = {5120, 1};
	    constants_info_[871].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[871].original_fqn = "model.multi_modal_projector.linear_1.parametrizations.weight.original0";
	    constants_info_[872].name = "model_multi_modal_projector_linear_1_parametrizations_weight_original1";
	    constants_info_[872].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[872].offset = 0;
	    constants_info_[872].data_size = 1966080;
	    constants_info_[872].from_folded = false;
	    constants_info_[872].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[872].shape = {3072, 160};
	    constants_info_[872].stride = {160, 1};
	    constants_info_[872].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[872].original_fqn = "model.multi_modal_projector.linear_1.parametrizations.weight.original1";
	    constants_info_[873].name = "model_multi_modal_projector_linear_1_parametrizations_weight_original2";
	    constants_info_[873].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[873].offset = 0;
	    constants_info_[873].data_size = 491520;
	    constants_info_[873].from_folded = false;
	    constants_info_[873].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[873].shape = {3072, 160};
	    constants_info_[873].stride = {160, 1};
	    constants_info_[873].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[873].original_fqn = "model.multi_modal_projector.linear_1.parametrizations.weight.original2";
	    constants_info_[874].name = "model_multi_modal_projector_linear_2_parametrizations_weight_original0";
	    constants_info_[874].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[874].offset = 0;
	    constants_info_[874].data_size = 9437184;
	    constants_info_[874].from_folded = false;
	    constants_info_[874].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[874].shape = {3072, 3072};
	    constants_info_[874].stride = {3072, 1};
	    constants_info_[874].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[874].original_fqn = "model.multi_modal_projector.linear_2.parametrizations.weight.original0";
	    constants_info_[875].name = "model_multi_modal_projector_linear_2_parametrizations_weight_original1";
	    constants_info_[875].dtype = static_cast<int32_t>(cached_torch_dtype_float32);
	    constants_info_[875].offset = 0;
	    constants_info_[875].data_size = 1179648;
	    constants_info_[875].from_folded = false;
	    constants_info_[875].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[875].shape = {3072, 96};
	    constants_info_[875].stride = {96, 1};
	    constants_info_[875].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[875].original_fqn = "model.multi_modal_projector.linear_2.parametrizations.weight.original1";
	    constants_info_[876].name = "model_multi_modal_projector_linear_2_parametrizations_weight_original2";
	    constants_info_[876].dtype = static_cast<int32_t>(cached_torch_dtype_int8);
	    constants_info_[876].offset = 0;
	    constants_info_[876].data_size = 294912;
	    constants_info_[876].from_folded = false;
	    constants_info_[876].type = static_cast<int32_t>(torch::aot_inductor::ConstantType::Parameter);
	    constants_info_[876].shape = {3072, 96};
	    constants_info_[876].stride = {96, 1};
	    constants_info_[876].layout = static_cast<int32_t>(cached_torch_layout_strided);
	    constants_info_[876].original_fqn = "model.multi_modal_projector.linear_2.parametrizations.weight.original2";
	    update_constants_map(std::move(constants_map));
	    update_constants_array(std::move(constants_array));
	    in_spec_ = R"([1, {"type": "builtins.tuple", "context": "null", "children_spec": [{"type": "builtins.tuple", "context": "null", "children_spec": []}, {"type": "builtins.dict", "context": "[\"input_features\"]", "children_spec": [{"type": null, "context": null, "children_spec": []}]}]}])";
	    out_spec_ = R"([1, {"type": null, "context": null, "children_spec": []}])";
	    outputs_info_[0].name = "output0";
	    this->kernels_ = std::make_unique<AOTInductorModelKernels>();
	}
	
	std::unordered_map<std::string, AtenTensorHandle> AOTInductorModel::const_run_impl(
	    DeviceStreamType stream,
	    AOTIProxyExecutorHandle proxy_executor,
	    bool initialization
	) {
	
	    if (!initialization) {
	        std::cerr << "[WARNING] Calling constant_folding in model, but compiled with config: "
	                  << "aot_inductor.use_runtime_constant_folding=False\n";
	    }
	    return {};
	}
	} // namespace torch::aot_inductor
	using namespace torch::aot_inductor;
	
	template <typename in_out_ptr0_type_, typename in_ptr0_type_, typename kernels_type_>
	static inline void call_triton_poi_fused_convolution_gelu_0(
	    const in_out_ptr0_type_& in_out_ptr0,
	    const in_ptr0_type_& in_ptr0,
	    int64_t xnumel,
	    int32_t device_idx_,
	    cudaStream_t stream_,
	    kernels_type_& kernels_,
	    const std::optional<std::string>& cubin_dir_ = std::nullopt
	){
	    /*
	    async_compile.triton('triton_poi_fused_convolution_gelu_0', '''
	    import triton
	    import triton.language as tl
	
	    from torch._inductor.runtime import triton_helpers, triton_heuristics
	    from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
	    from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, DeviceProperties
	    triton_helpers.set_driver_to_gpu()
	
	    @triton_heuristics.pointwise(
	        size_hints={'x': 16777216}, 
	        filename=__file__,
	        triton_meta={'signature': {'in_out_ptr0': '*fp32', 'in_ptr0': '*fp32', 'xnumel': 'i32', 'XBLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=108, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, warp_size=32), 'constants': {}, 'configs': [{(0,): [['tt.divisibility', 16]], (1,): [['tt.divisibility', 16]], (2,): [['tt.divisibility', 16]]}]},
	        inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_poi_fused_convolution_gelu_0', 'mutated_arg_names': ['in_out_ptr0'], 'optimize_mem': True, 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': '07F8534F6C2EC4E11ACA26C8C9D43565780E009807A566E459118140DB4BC4AB', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': True, 'is_fbcode': True},
	        min_elem_per_thread=0
	    )
	    @triton.jit
	    def triton_poi_fused_convolution_gelu_0(in_out_ptr0, in_ptr0, xnumel, XBLOCK : tl.constexpr):
	        xoffset = tl.program_id(0) * XBLOCK
	        xindex = xoffset + tl.arange(0, XBLOCK)[:]
	        xmask = xindex < xnumel
	        x3 = xindex
	        x1 = ((xindex // 3000) % 1280)
	        tmp0 = tl.load(in_out_ptr0 + (x3), xmask)
	        tmp1 = tl.load(in_ptr0 + (x1), xmask, eviction_policy='evict_last')
	        tmp2 = tmp0 + tmp1
	        tmp3 = 0.5
	        tmp4 = tmp2 * tmp3
	        tmp5 = 0.7071067811865476
	        tmp6 = tmp2 * tmp5
	        tmp7 = libdevice.erf(tmp6)
	        tmp8 = 1.0
	        tmp9 = tmp7 + tmp8
	        tmp10 = tmp4 * tmp9
	        tl.store(in_out_ptr0 + (x3), tmp10, xmask)
	    ''', device_str='cuda')
	    */
	    uint32_t grid_0 = ((xnumel + (512 - 1)) / (512));
	    uint32_t grid_1 = 1;
	    uint32_t grid_2 = 1;
	    if (grid_0 == 0 || grid_1 == 0 || grid_2 == 0) return;
	    if (kernels_.triton_poi_fused_convolution_gelu_0 == nullptr) {
	        kernels_.triton_poi_fused_convolution_gelu_0 = loadKernel("/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/ckjk6vsfguqn7iv75aqlmo7t4mm3sfnt27wkl3n2s5iotvpxkpbi.cubin", "triton_poi_fused_convolution_gelu_0", 0, cubin_dir_); 
	    }
	    CUdeviceptr var_33 = reinterpret_cast<CUdeviceptr>(in_out_ptr0.data_ptr());
	    CUdeviceptr var_34 = reinterpret_cast<CUdeviceptr>(in_ptr0.data_ptr());
	    int32_t var_35 = xnumel;
	    CUdeviceptr global_scratch_scratch_36 = 0;
	    void* kernel_args_[] = {&var_33, &var_34, &var_35, &global_scratch_scratch_36};
	    launchKernel(kernels_.triton_poi_fused_convolution_gelu_0, grid_0, grid_1, grid_2, 8, 0, kernel_args_, stream_);
	}
	
	template <typename in_ptr0_type_, typename in_ptr1_type_, typename in_ptr2_type_, typename out_ptr0_type_, typename out_ptr1_type_, typename out_ptr2_type_, typename kernels_type_>
	static inline void call_triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1(
	    const in_ptr0_type_& in_ptr0,
	    const in_ptr1_type_& in_ptr1,
	    const in_ptr2_type_& in_ptr2,
	    const out_ptr0_type_& out_ptr0,
	    const out_ptr1_type_& out_ptr1,
	    const out_ptr2_type_& out_ptr2,
	    int64_t xnumel,
	    int64_t r0_numel,
	    int32_t device_idx_,
	    cudaStream_t stream_,
	    kernels_type_& kernels_,
	    const std::optional<std::string>& cubin_dir_ = std::nullopt
	){
	    /*
	    async_compile.triton('triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1', '''
	    import triton
	    import triton.language as tl
	
	    from torch._inductor.runtime import triton_helpers, triton_heuristics
	    from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
	    from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, DeviceProperties
	    triton_helpers.set_driver_to_gpu()
	
	    @triton_heuristics.reduction(
	        size_hints={'x': 65536, 'r0_': 128},
	        reduction_hint=ReductionHint.OUTER,
	        filename=__file__,
	        triton_meta={'signature': {'in_ptr0': '*fp32', 'in_ptr1': '*fp32', 'in_ptr2': '*fp32', 'out_ptr0': '*fp32', 'out_ptr1': '*fp32', 'out_ptr2': '*fp32', 'xnumel': 'i32', 'r0_numel': 'i32', 'XBLOCK': 'constexpr', 'R0_BLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=108, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, warp_size=32), 'constants': {}, 'configs': [{(0,): [['tt.divisibility', 16]], (1,): [['tt.divisibility', 16]], (2,): [['tt.divisibility', 16]], (3,): [['tt.divisibility', 16]], (4,): [['tt.divisibility', 16]], (5,): [['tt.divisibility', 16]], (7,): [['tt.divisibility', 16]]}]},
	        inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1', 'mutated_arg_names': [], 'optimize_mem': True, 'no_x_dim': False, 'num_load': 3, 'num_reduction': 3, 'backend_hash': '07F8534F6C2EC4E11ACA26C8C9D43565780E009807A566E459118140DB4BC4AB', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': True, 'is_fbcode': True}
	    )
	    @triton.jit
	    def triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1(in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1, out_ptr2, xnumel, r0_numel, XBLOCK : tl.constexpr, R0_BLOCK : tl.constexpr):
	        r0_numel = 128
	        rnumel = r0_numel
	        RBLOCK: tl.constexpr = R0_BLOCK
	        xoffset = tl.program_id(0) * XBLOCK
	        xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
	        xmask = xindex < xnumel
	        r0_base = tl.arange(0, R0_BLOCK)[None, :]
	        rbase = r0_base
	        x0 = (xindex % 10)
	        x1 = ((xindex // 10) % 1500)
	        x2 = xindex // 15000
	        x5 = (xindex % 15000)
	        tmp14_mean = tl.zeros([XBLOCK, R0_BLOCK], tl.float32)
	        tmp14_m2 = tl.zeros([XBLOCK, R0_BLOCK], tl.float32)
	        tmp14_weight = tl.zeros([XBLOCK, R0_BLOCK], tl.float32)
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_3 = r0_index
	            tmp0 = tl.load(in_ptr0 + (x1 + 1500*r0_3 + 192000*x0 + 1920000*x2), r0_mask & xmask, eviction_policy='evict_last', other=0.0)
	            tmp1 = tl.load(in_ptr1 + (r0_3 + 128*x0), r0_mask & xmask, eviction_policy='evict_last', other=0.0)
	            tmp11 = tl.load(in_ptr2 + (r0_3 + 128*x5), r0_mask & xmask, eviction_policy='evict_last', other=0.0)
	            tmp2 = tmp0 + tmp1
	            tmp3 = 0.5
	            tmp4 = tmp2 * tmp3
	            tmp5 = 0.7071067811865476
	            tmp6 = tmp2 * tmp5
	            tmp7 = libdevice.erf(tmp6)
	            tmp8 = 1.0
	            tmp9 = tmp7 + tmp8
	            tmp10 = tmp4 * tmp9
	            tmp12 = tmp10 + tmp11
	            tmp13 = tl.broadcast_to(tmp12, [XBLOCK, R0_BLOCK])
	            tmp14_mean_next, tmp14_m2_next, tmp14_weight_next = triton_helpers.welford_reduce(
	                tmp13, tmp14_mean, tmp14_m2, tmp14_weight, roffset == 0
	            )
	            tmp14_mean = tl.where(r0_mask & xmask, tmp14_mean_next, tmp14_mean)
	            tmp14_m2 = tl.where(r0_mask & xmask, tmp14_m2_next, tmp14_m2)
	            tmp14_weight = tl.where(r0_mask & xmask, tmp14_weight_next, tmp14_weight)
	        tmp15, tmp16, tmp17 = triton_helpers.welford(tmp14_mean, tmp14_m2, tmp14_weight, 1)
	        tmp14 = tmp15[:, None]
	        tmp18 = tmp16[:, None]
	        tmp19 = tmp17[:, None]
	        tl.store(out_ptr0 + (x5 + 15008*x2), tmp14, xmask)
	        tl.store(out_ptr1 + (x5 + 15008*x2), tmp18, xmask)
	        tl.store(out_ptr2 + (x5 + 15008*x2), tmp19, xmask)
	    ''', device_str='cuda')
	    */
	    uint32_t grid_0 = ((xnumel + (64 - 1)) / (64));
	    uint32_t grid_1 = 1;
	    uint32_t grid_2 = 1;
	    if (grid_0 == 0 || grid_1 == 0 || grid_2 == 0) return;
	    if (kernels_.triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1 == nullptr) {
	        kernels_.triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1 = loadKernel("/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/cxhbdmw2k45mgnneulfggbubywar4ri67rbszuqga7dwe5r34vyt.cubin", "triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1", 3072, cubin_dir_); 
	    }
	    CUdeviceptr var_37 = reinterpret_cast<CUdeviceptr>(in_ptr0.data_ptr());
	    CUdeviceptr var_38 = reinterpret_cast<CUdeviceptr>(in_ptr1.data_ptr());
	    CUdeviceptr var_39 = reinterpret_cast<CUdeviceptr>(in_ptr2.data_ptr());
	    CUdeviceptr var_40 = reinterpret_cast<CUdeviceptr>(out_ptr0.data_ptr());
	    CUdeviceptr var_41 = reinterpret_cast<CUdeviceptr>(out_ptr1.data_ptr());
	    CUdeviceptr var_42 = reinterpret_cast<CUdeviceptr>(out_ptr2.data_ptr());
	    int32_t var_43 = xnumel;
	    int var_44 = r0_numel;
	    CUdeviceptr global_scratch_scratch_45 = 0;
	    void* kernel_args_[] = {&var_37, &var_38, &var_39, &var_40, &var_41, &var_42, &var_43, &var_44, &global_scratch_scratch_45};
	    launchKernel(kernels_.triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1, grid_0, grid_1, grid_2, 4, 3072, kernel_args_, stream_);
	}
	
	template <typename in_ptr0_type_, typename in_ptr1_type_, typename in_ptr2_type_, typename out_ptr0_type_, typename out_ptr1_type_, typename kernels_type_>
	static inline void call_triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2(
	    const in_ptr0_type_& in_ptr0,
	    const in_ptr1_type_& in_ptr1,
	    const in_ptr2_type_& in_ptr2,
	    const out_ptr0_type_& out_ptr0,
	    const out_ptr1_type_& out_ptr1,
	    int64_t xnumel,
	    int64_t r0_numel,
	    int32_t device_idx_,
	    cudaStream_t stream_,
	    kernels_type_& kernels_,
	    const std::optional<std::string>& cubin_dir_ = std::nullopt
	){
	    /*
	    async_compile.triton('triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2', '''
	    import triton
	    import triton.language as tl
	
	    from torch._inductor.runtime import triton_helpers, triton_heuristics
	    from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
	    from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, DeviceProperties
	    triton_helpers.set_driver_to_gpu()
	
	    @triton_heuristics.persistent_reduction(
	        size_hints={'x': 8192, 'r0_': 16},
	        reduction_hint=ReductionHint.OUTER,
	        filename=__file__,
	        triton_meta={'signature': {'in_ptr0': '*fp32', 'in_ptr1': '*fp32', 'in_ptr2': '*fp32', 'out_ptr0': '*fp32', 'out_ptr1': '*fp32', 'xnumel': 'i32', 'r0_numel': 'i32', 'XBLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=108, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, warp_size=32), 'constants': {}, 'configs': [{(0,): [['tt.divisibility', 16]], (1,): [['tt.divisibility', 16]], (2,): [['tt.divisibility', 16]], (3,): [['tt.divisibility', 16]], (4,): [['tt.divisibility', 16]]}]},
	        inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2', 'mutated_arg_names': [], 'optimize_mem': True, 'no_x_dim': None, 'num_load': 3, 'num_reduction': 2, 'backend_hash': '07F8534F6C2EC4E11ACA26C8C9D43565780E009807A566E459118140DB4BC4AB', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': True, 'is_fbcode': True}
	    )
	    @triton.jit
	    def triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2(in_ptr0, in_ptr1, in_ptr2, out_ptr0, out_ptr1, xnumel, r0_numel, XBLOCK : tl.constexpr):
	        r0_numel = 10
	        R0_BLOCK: tl.constexpr = 16
	        rnumel = r0_numel
	        RBLOCK: tl.constexpr = R0_BLOCK
	        xoffset = tl.program_id(0) * XBLOCK
	        xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
	        xmask = xindex < xnumel
	        r0_index = tl.arange(0, R0_BLOCK)[None, :]
	        r0_offset = 0
	        r0_mask = r0_index < r0_numel
	        roffset = r0_offset
	        rindex = r0_index
	        r0_2 = r0_index
	        x0 = (xindex % 1500)
	        x1 = xindex // 1500
	        tmp0 = tl.load(in_ptr0 + (r0_2 + 10*x0 + 15008*x1), r0_mask & xmask, other=0.0)
	        tmp1 = tl.load(in_ptr1 + (r0_2 + 10*x0 + 15008*x1), r0_mask & xmask, other=0.0)
	        tmp2 = tl.load(in_ptr2 + (r0_2 + 10*x0 + 15008*x1), r0_mask & xmask, other=0.0)
	        tmp3 = tl.broadcast_to(tmp0, [XBLOCK, R0_BLOCK])
	        tmp4 = tl.broadcast_to(tmp1, [XBLOCK, R0_BLOCK])
	        tmp5 = tl.broadcast_to(tmp2, [XBLOCK, R0_BLOCK])
	        tmp7 = tl.where(r0_mask & xmask, tmp3, 0)
	        tmp8 = tl.where(r0_mask & xmask, tmp4, 0)
	        tmp9 = tl.where(r0_mask & xmask, tmp5, 0)
	        tmp10, tmp11, tmp12 = triton_helpers.welford(tmp7, tmp8, tmp9, 1)
	        tmp13 = tmp10[:, None]
	        tmp14 = tmp11[:, None]
	        tmp15 = tmp12[:, None]
	        tl.store(out_ptr0 + (x0 + 1504*x1), tmp13, xmask)
	        tl.store(out_ptr1 + (x0 + 1504*x1), tmp14, xmask)
	    ''', device_str='cuda')
	    */
	    uint32_t grid_0 = ((xnumel + (128 - 1)) / (128));
	    uint32_t grid_1 = 1;
	    uint32_t grid_2 = 1;
	    if (grid_0 == 0 || grid_1 == 0 || grid_2 == 0) return;
	    if (kernels_.triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2 == nullptr) {
	        kernels_.triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2 = loadKernel("/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/cwhmurfngnqlco3ikplruep3nvkomvvm7zxiyxcpnonrortsjaub.cubin", "triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2", 512, cubin_dir_); 
	    }
	    CUdeviceptr var_46 = reinterpret_cast<CUdeviceptr>(in_ptr0.data_ptr());
	    CUdeviceptr var_47 = reinterpret_cast<CUdeviceptr>(in_ptr1.data_ptr());
	    CUdeviceptr var_48 = reinterpret_cast<CUdeviceptr>(in_ptr2.data_ptr());
	    CUdeviceptr var_49 = reinterpret_cast<CUdeviceptr>(out_ptr0.data_ptr());
	    CUdeviceptr var_50 = reinterpret_cast<CUdeviceptr>(out_ptr1.data_ptr());
	    int32_t var_51 = xnumel;
	    int var_52 = r0_numel;
	    CUdeviceptr global_scratch_scratch_53 = 0;
	    void* kernel_args_[] = {&var_46, &var_47, &var_48, &var_49, &var_50, &var_51, &var_52, &global_scratch_scratch_53};
	    launchKernel(kernels_.triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2, grid_0, grid_1, grid_2, 8, 512, kernel_args_, stream_);
	}
	
	template <typename in_out_ptr0_type_, typename in_out_ptr1_type_, typename in_out_ptr2_type_, typename in_ptr0_type_, typename in_ptr1_type_, typename in_ptr2_type_, typename in_ptr3_type_, typename in_ptr4_type_, typename in_ptr5_type_, typename in_ptr6_type_, typename out_ptr0_type_, typename kernels_type_>
	static inline void call_triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3(
	    const in_out_ptr0_type_& in_out_ptr0,
	    const in_out_ptr1_type_& in_out_ptr1,
	    const in_out_ptr2_type_& in_out_ptr2,
	    const in_ptr0_type_& in_ptr0,
	    const in_ptr1_type_& in_ptr1,
	    const in_ptr2_type_& in_ptr2,
	    const in_ptr3_type_& in_ptr3,
	    const in_ptr4_type_& in_ptr4,
	    const in_ptr5_type_& in_ptr5,
	    const in_ptr6_type_& in_ptr6,
	    const out_ptr0_type_& out_ptr0,
	    int64_t xnumel,
	    int64_t r0_numel,
	    int32_t device_idx_,
	    cudaStream_t stream_,
	    kernels_type_& kernels_,
	    const std::optional<std::string>& cubin_dir_ = std::nullopt
	){
	    /*
	    async_compile.triton('triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3', '''
	    import triton
	    import triton.language as tl
	
	    from torch._inductor.runtime import triton_helpers, triton_heuristics
	    from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
	    from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, DeviceProperties
	    triton_helpers.set_driver_to_gpu()
	
	    @triton_heuristics.reduction(
	        size_hints={'x': 8192, 'r0_': 2048},
	        reduction_hint=ReductionHint.DEFAULT,
	        filename=__file__,
	        triton_meta={'signature': {'in_out_ptr0': '*fp32', 'in_out_ptr1': '*fp32', 'in_out_ptr2': '*fp32', 'in_ptr0': '*fp32', 'in_ptr1': '*fp32', 'in_ptr2': '*fp32', 'in_ptr3': '*fp32', 'in_ptr4': '*fp32', 'in_ptr5': '*fp32', 'in_ptr6': '*fp32', 'out_ptr0': '*fp32', 'xnumel': 'i32', 'r0_numel': 'i32', 'XBLOCK': 'constexpr', 'R0_BLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=108, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, warp_size=32), 'constants': {}, 'configs': [{(0,): [['tt.divisibility', 16]], (1,): [['tt.divisibility', 16]], (2,): [['tt.divisibility', 16]], (3,): [['tt.divisibility', 16]], (4,): [['tt.divisibility', 16]], (5,): [['tt.divisibility', 16]], (6,): [['tt.divisibility', 16]], (7,): [['tt.divisibility', 16]], (8,): [['tt.divisibility', 16]], (9,): [['tt.divisibility', 16]], (10,): [['tt.divisibility', 16]], (12,): [['tt.divisibility', 16]]}]},
	        inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3', 'mutated_arg_names': ['in_out_ptr0', 'in_out_ptr1', 'in_out_ptr2'], 'optimize_mem': True, 'no_x_dim': False, 'num_load': 8, 'num_reduction': 6, 'backend_hash': '07F8534F6C2EC4E11ACA26C8C9D43565780E009807A566E459118140DB4BC4AB', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': True, 'is_fbcode': True}
	    )
	    @triton.jit
	    def triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3(in_out_ptr0, in_out_ptr1, in_out_ptr2, in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, in_ptr6, out_ptr0, xnumel, r0_numel, XBLOCK : tl.constexpr, R0_BLOCK : tl.constexpr):
	        r0_numel = 1280
	        rnumel = r0_numel
	        RBLOCK: tl.constexpr = R0_BLOCK
	        xoffset = tl.program_id(0) * XBLOCK
	        xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
	        xmask = xindex < xnumel
	        r0_base = tl.arange(0, R0_BLOCK)[None, :]
	        rbase = r0_base
	        x0 = (xindex % 1500)
	        x1 = xindex // 1500
	        tmp13 = tl.load(in_ptr3 + (x0 + 1504*x1), xmask, eviction_policy='evict_last')
	        tmp15 = tl.load(in_ptr4 + (x0 + 1504*x1), xmask, eviction_policy='evict_last')
	        x3 = xindex
	        _tmp27 = tl.full([XBLOCK, R0_BLOCK], float("-inf"), tl.float32)
	        _tmp29 = tl.full([XBLOCK, R0_BLOCK], float("inf"), tl.float32)
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_2 = r0_index
	            tmp0 = tl.load(in_ptr0 + (x0 + 1500*r0_2 + 1920000*x1), r0_mask & xmask, eviction_policy='evict_first', other=0.0)
	            tmp1 = tl.load(in_ptr1 + (r0_2), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp11 = tl.load(in_ptr2 + (r0_2 + 1280*x0), r0_mask & xmask, eviction_policy='evict_last', other=0.0)
	            tmp22 = tl.load(in_ptr5 + (r0_2), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp24 = tl.load(in_ptr6 + (r0_2), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp2 = tmp0 + tmp1
	            tmp3 = 0.5
	            tmp4 = tmp2 * tmp3
	            tmp5 = 0.7071067811865476
	            tmp6 = tmp2 * tmp5
	            tmp7 = libdevice.erf(tmp6)
	            tmp8 = 1.0
	            tmp9 = tmp7 + tmp8
	            tmp10 = tmp4 * tmp9
	            tmp12 = tmp10 + tmp11
	            tmp14 = tmp12 - tmp13
	            tmp16 = 1280.0
	            tmp17 = (tmp15 / tmp16)
	            tmp18 = 1e-05
	            tmp19 = tmp17 + tmp18
	            tmp20 = libdevice.rsqrt(tmp19)
	            tmp21 = tmp14 * tmp20
	            tmp23 = tmp21 * tmp22
	            tmp25 = tmp23 + tmp24
	            tmp26 = tl.broadcast_to(tmp25, [XBLOCK, R0_BLOCK])
	            tmp28 = triton_helpers.maximum(_tmp27, tmp26)
	            _tmp27 = tl.where(r0_mask & xmask, tmp28, _tmp27)
	            tmp30 = triton_helpers.minimum(_tmp29, tmp26)
	            _tmp29 = tl.where(r0_mask & xmask, tmp30, _tmp29)
	            tl.store(out_ptr0 + (r0_2 + 1280*x3), tmp25, r0_mask & xmask)
	        tmp27 = triton_helpers.max2(_tmp27, 1)[:, None]
	        tmp29 = triton_helpers.min2(_tmp29, 1)[:, None]
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_2 = r0_index
	            tmp31 = tl.load(out_ptr0 + (r0_2 + 1280*x3), r0_mask & xmask, eviction_policy='evict_first', other=0.0)
	            tmp32 = 0.0
	            tmp33 = triton_helpers.maximum(tmp27, tmp32)
	            tmp34 = triton_helpers.minimum(tmp29, tmp32)
	            tmp35 = tmp33 - tmp34
	            tmp36 = 0.00392156862745098
	            tmp37 = tmp35 * tmp36
	            tmp38 = 1.1920928955078125e-07
	            tmp39 = triton_helpers.maximum(tmp37, tmp38)
	            tmp40 = tl.full([1, 1], 1, tl.int32)
	            tmp41 = (tmp40 / tmp39)
	            tmp42 = 1.0
	            tmp43 = tmp41 * tmp42
	            tmp44 = tmp31 * tmp43
	            tmp45 = libdevice.nearbyint(tmp44)
	            tmp46 = (tmp34 / tmp39)
	            tmp47 = libdevice.nearbyint(tmp46)
	            tmp48 = -128.0
	            tmp49 = tmp48 - tmp47
	            tmp50 = triton_helpers.maximum(tmp49, tmp48)
	            tmp51 = 127.0
	            tmp52 = triton_helpers.minimum(tmp50, tmp51)
	            tmp53 = tmp52.to(tl.int8)
	            tmp54 = tmp53.to(tl.float32)
	            tmp55 = tmp45 + tmp54
	            tmp56 = triton_helpers.maximum(tmp55, tmp48)
	            tmp57 = triton_helpers.minimum(tmp56, tmp51)
	            tmp58 = tmp57.to(tl.int8)
	            tmp59 = tmp58.to(tl.float32)
	            tmp60 = tmp59 - tmp54
	            tmp61 = tmp60 * tmp39
	            tl.store(in_out_ptr0 + (r0_2 + 1280*x3), tmp61, r0_mask & xmask)
	            tl.store(in_out_ptr1 + (r0_2 + 1280*x3), tmp61, r0_mask & xmask)
	            tl.store(in_out_ptr2 + (r0_2 + 1280*x3), tmp61, r0_mask & xmask)
	    ''', device_str='cuda')
	    */
	    uint32_t grid_0 = ((xnumel + (8 - 1)) / (8));
	    uint32_t grid_1 = 1;
	    uint32_t grid_2 = 1;
	    if (grid_0 == 0 || grid_1 == 0 || grid_2 == 0) return;
	    if (kernels_.triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3 == nullptr) {
	        kernels_.triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3 = loadKernel("/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/cv4sqdr7hiwd2dv7yrnjjdue67n7hdzozkx3gmkn24mxolejam6h.cubin", "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3", 8320, cubin_dir_); 
	    }
	    CUdeviceptr var_54 = reinterpret_cast<CUdeviceptr>(in_out_ptr0.data_ptr());
	    CUdeviceptr var_55 = reinterpret_cast<CUdeviceptr>(in_out_ptr1.data_ptr());
	    CUdeviceptr var_56 = reinterpret_cast<CUdeviceptr>(in_out_ptr2.data_ptr());
	    CUdeviceptr var_57 = reinterpret_cast<CUdeviceptr>(in_ptr0.data_ptr());
	    CUdeviceptr var_58 = reinterpret_cast<CUdeviceptr>(in_ptr1.data_ptr());
	    CUdeviceptr var_59 = reinterpret_cast<CUdeviceptr>(in_ptr2.data_ptr());
	    CUdeviceptr var_60 = reinterpret_cast<CUdeviceptr>(in_ptr3.data_ptr());
	    CUdeviceptr var_61 = reinterpret_cast<CUdeviceptr>(in_ptr4.data_ptr());
	    CUdeviceptr var_62 = reinterpret_cast<CUdeviceptr>(in_ptr5.data_ptr());
	    CUdeviceptr var_63 = reinterpret_cast<CUdeviceptr>(in_ptr6.data_ptr());
	    CUdeviceptr var_64 = reinterpret_cast<CUdeviceptr>(out_ptr0.data_ptr());
	    int32_t var_65 = xnumel;
	    int var_66 = r0_numel;
	    CUdeviceptr global_scratch_scratch_67 = 0;
	    void* kernel_args_[] = {&var_54, &var_55, &var_56, &var_57, &var_58, &var_59, &var_60, &var_61, &var_62, &var_63, &var_64, &var_65, &var_66, &global_scratch_scratch_67};
	    launchKernel(kernels_.triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3, grid_0, grid_1, grid_2, 16, 8320, kernel_args_, stream_);
	}
	
	template <typename in_ptr0_type_, typename in_ptr1_type_, typename in_ptr2_type_, typename out_ptr0_type_, typename kernels_type_>
	static inline void call_triton_poi_fused__to_copy_mul_sub_view_4(
	    const in_ptr0_type_& in_ptr0,
	    const in_ptr1_type_& in_ptr1,
	    const in_ptr2_type_& in_ptr2,
	    const out_ptr0_type_& out_ptr0,
	    int64_t xnumel,
	    int32_t device_idx_,
	    cudaStream_t stream_,
	    kernels_type_& kernels_,
	    const std::optional<std::string>& cubin_dir_ = std::nullopt
	){
	    /*
	    async_compile.triton('triton_poi_fused__to_copy_mul_sub_view_4', '''
	    import triton
	    import triton.language as tl
	
	    from torch._inductor.runtime import triton_helpers, triton_heuristics
	    from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
	    from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, DeviceProperties
	    triton_helpers.set_driver_to_gpu()
	
	    @triton_heuristics.pointwise(
	        size_hints={'x': 2097152}, 
	        filename=__file__,
	        triton_meta={'signature': {'in_ptr0': '*i8', 'in_ptr1': '*i8', 'in_ptr2': '*fp32', 'out_ptr0': '*fp32', 'xnumel': 'i32', 'XBLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=108, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, warp_size=32), 'constants': {}, 'configs': [{(0,): [['tt.divisibility', 16]], (1,): [['tt.divisibility', 16]], (2,): [['tt.divisibility', 16]], (3,): [['tt.divisibility', 16]], (4,): [['tt.divisibility', 16]]}]},
	        inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__to_copy_mul_sub_view_4', 'mutated_arg_names': [], 'optimize_mem': True, 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': '07F8534F6C2EC4E11ACA26C8C9D43565780E009807A566E459118140DB4BC4AB', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': True, 'is_fbcode': True},
	        min_elem_per_thread=0
	    )
	    @triton.jit
	    def triton_poi_fused__to_copy_mul_sub_view_4(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
	        xnumel = 1638400
	        xoffset = tl.program_id(0) * XBLOCK
	        xindex = xoffset + tl.arange(0, XBLOCK)[:]
	        xmask = tl.full([XBLOCK], True, tl.int1)
	        x2 = xindex
	        x1 = xindex // 32
	        tmp0 = tl.load(in_ptr0 + (x2), None)
	        tmp2 = tl.load(in_ptr1 + (x1), None, eviction_policy='evict_last')
	        tmp5 = tl.load(in_ptr2 + (x1), None, eviction_policy='evict_last')
	        tmp1 = tmp0.to(tl.float32)
	        tmp3 = tmp2.to(tl.float32)
	        tmp4 = tmp1 - tmp3
	        tmp6 = tmp4 * tmp5
	        tl.store(out_ptr0 + (x2), tmp6, None)
	    ''', device_str='cuda')
	    */
	    uint32_t grid_0 = ((xnumel + (1024 - 1)) / (1024));
	    uint32_t grid_1 = 1;
	    uint32_t grid_2 = 1;
	    if (grid_0 == 0 || grid_1 == 0 || grid_2 == 0) return;
	    if (kernels_.triton_poi_fused__to_copy_mul_sub_view_4 == nullptr) {
	        kernels_.triton_poi_fused__to_copy_mul_sub_view_4 = loadKernel("/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/crim3y3fcxlcgeal7h3kushdcjjlnfg3wnabnbsj7ava2ml7xsy6.cubin", "triton_poi_fused__to_copy_mul_sub_view_4", 4096, cubin_dir_); 
	    }
	    CUdeviceptr var_68 = reinterpret_cast<CUdeviceptr>(in_ptr0.data_ptr());
	    CUdeviceptr var_69 = reinterpret_cast<CUdeviceptr>(in_ptr1.data_ptr());
	    CUdeviceptr var_70 = reinterpret_cast<CUdeviceptr>(in_ptr2.data_ptr());
	    CUdeviceptr var_71 = reinterpret_cast<CUdeviceptr>(out_ptr0.data_ptr());
	    int var_72 = xnumel;
	    CUdeviceptr global_scratch_scratch_73 = 0;
	    void* kernel_args_[] = {&var_68, &var_69, &var_70, &var_71, &var_72, &global_scratch_scratch_73};
	    launchKernel(kernels_.triton_poi_fused__to_copy_mul_sub_view_4, grid_0, grid_1, grid_2, 4, 4096, kernel_args_, stream_);
	}
	
	template <typename in_ptr0_type_, typename in_ptr1_type_, typename out_ptr0_type_, typename kernels_type_>
	static inline void call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(
	    const in_ptr0_type_& in_ptr0,
	    const in_ptr1_type_& in_ptr1,
	    const out_ptr0_type_& out_ptr0,
	    int64_t xnumel,
	    int32_t device_idx_,
	    cudaStream_t stream_,
	    kernels_type_& kernels_,
	    const std::optional<std::string>& cubin_dir_ = std::nullopt
	){
	    /*
	    async_compile.triton('triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5', '''
	    import triton
	    import triton.language as tl
	
	    from torch._inductor.runtime import triton_helpers, triton_heuristics
	    from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
	    from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, DeviceProperties
	    triton_helpers.set_driver_to_gpu()
	
	    @triton_heuristics.pointwise(
	        size_hints={'x': 8388608}, 
	        filename=__file__,
	        triton_meta={'signature': {'in_ptr0': '*fp32', 'in_ptr1': '*fp32', 'out_ptr0': '*fp32', 'xnumel': 'i32', 'XBLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=108, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, warp_size=32), 'constants': {}, 'configs': [{(0,): [['tt.divisibility', 16]], (1,): [['tt.divisibility', 16]], (2,): [['tt.divisibility', 16]], (3,): [['tt.divisibility', 16]]}]},
	        inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5', 'mutated_arg_names': [], 'optimize_mem': True, 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': '07F8534F6C2EC4E11ACA26C8C9D43565780E009807A566E459118140DB4BC4AB', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': True, 'is_fbcode': True},
	        min_elem_per_thread=0
	    )
	    @triton.jit
	    def triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
	        xoffset = tl.program_id(0) * XBLOCK
	        xindex = xoffset + tl.arange(0, XBLOCK)[:]
	        xmask = xindex < xnumel
	        x0 = (xindex % 64)
	        x1 = ((xindex // 64) % 1500)
	        x2 = ((xindex // 96000) % 20)
	        x3 = xindex // 1920000
	        x4 = xindex
	        tmp0 = tl.load(in_ptr0 + (x0 + 64*x2 + 1280*x1 + 1920000*x3), xmask)
	        tmp1 = tl.load(in_ptr1 + (x0 + 64*x2), xmask, eviction_policy='evict_last')
	        tmp2 = tmp0 + tmp1
	        tmp3 = 0.125
	        tmp4 = tmp2 * tmp3
	        tl.store(out_ptr0 + (x4), tmp4, xmask)
	    ''', device_str='cuda')
	    */
	    uint32_t grid_0 = ((xnumel + (1024 - 1)) / (1024));
	    uint32_t grid_1 = 1;
	    uint32_t grid_2 = 1;
	    if (grid_0 == 0 || grid_1 == 0 || grid_2 == 0) return;
	    if (kernels_.triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5 == nullptr) {
	        kernels_.triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5 = loadKernel("/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/c6qcqn6v3pzqwjmsichbvqvhjzg5wm7q3qjcewq3sl76lcn7te7r.cubin", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5", 0, cubin_dir_); 
	    }
	    CUdeviceptr var_74 = reinterpret_cast<CUdeviceptr>(in_ptr0.data_ptr());
	    CUdeviceptr var_75 = reinterpret_cast<CUdeviceptr>(in_ptr1.data_ptr());
	    CUdeviceptr var_76 = reinterpret_cast<CUdeviceptr>(out_ptr0.data_ptr());
	    int32_t var_77 = xnumel;
	    CUdeviceptr global_scratch_scratch_78 = 0;
	    void* kernel_args_[] = {&var_74, &var_75, &var_76, &var_77, &global_scratch_scratch_78};
	    launchKernel(kernels_.triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5, grid_0, grid_1, grid_2, 4, 0, kernel_args_, stream_);
	}
	
	template <typename in_ptr0_type_, typename out_ptr0_type_, typename kernels_type_>
	static inline void call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(
	    const in_ptr0_type_& in_ptr0,
	    const out_ptr0_type_& out_ptr0,
	    int64_t xnumel,
	    int32_t device_idx_,
	    cudaStream_t stream_,
	    kernels_type_& kernels_,
	    const std::optional<std::string>& cubin_dir_ = std::nullopt
	){
	    /*
	    async_compile.triton('triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6', '''
	    import triton
	    import triton.language as tl
	
	    from torch._inductor.runtime import triton_helpers, triton_heuristics
	    from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
	    from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, DeviceProperties
	    triton_helpers.set_driver_to_gpu()
	
	    @triton_heuristics.pointwise(
	        size_hints={'x': 8388608}, 
	        filename=__file__,
	        triton_meta={'signature': {'in_ptr0': '*fp32', 'out_ptr0': '*fp32', 'xnumel': 'i32', 'XBLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=108, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, warp_size=32), 'constants': {}, 'configs': [{(0,): [['tt.divisibility', 16]], (1,): [['tt.divisibility', 16]], (2,): [['tt.divisibility', 16]]}]},
	        inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6', 'mutated_arg_names': [], 'optimize_mem': True, 'no_x_dim': False, 'num_load': 1, 'num_reduction': 0, 'backend_hash': '07F8534F6C2EC4E11ACA26C8C9D43565780E009807A566E459118140DB4BC4AB', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': True, 'is_fbcode': True},
	        min_elem_per_thread=0
	    )
	    @triton.jit
	    def triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(in_ptr0, out_ptr0, xnumel, XBLOCK : tl.constexpr):
	        xoffset = tl.program_id(0) * XBLOCK
	        xindex = xoffset + tl.arange(0, XBLOCK)[:]
	        xmask = xindex < xnumel
	        x0 = (xindex % 64)
	        x1 = ((xindex // 64) % 1500)
	        x2 = ((xindex // 96000) % 20)
	        x3 = xindex // 1920000
	        x4 = xindex
	        tmp0 = tl.load(in_ptr0 + (x0 + 64*x2 + 1280*x1 + 1920000*x3), xmask)
	        tl.store(out_ptr0 + (x4), tmp0, xmask)
	    ''', device_str='cuda')
	    */
	    uint32_t grid_0 = ((xnumel + (1024 - 1)) / (1024));
	    uint32_t grid_1 = 1;
	    uint32_t grid_2 = 1;
	    if (grid_0 == 0 || grid_1 == 0 || grid_2 == 0) return;
	    if (kernels_.triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6 == nullptr) {
	        kernels_.triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6 = loadKernel("/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/cqyi3egykrlrcagy25yclakg6xxi3veqbeg6u2amrvkflonjkkff.cubin", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6", 0, cubin_dir_); 
	    }
	    CUdeviceptr var_79 = reinterpret_cast<CUdeviceptr>(in_ptr0.data_ptr());
	    CUdeviceptr var_80 = reinterpret_cast<CUdeviceptr>(out_ptr0.data_ptr());
	    int32_t var_81 = xnumel;
	    CUdeviceptr global_scratch_scratch_82 = 0;
	    void* kernel_args_[] = {&var_79, &var_80, &var_81, &global_scratch_scratch_82};
	    launchKernel(kernels_.triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6, grid_0, grid_1, grid_2, 4, 0, kernel_args_, stream_);
	}
	
	template <typename in_ptr0_type_, typename in_ptr1_type_, typename out_ptr0_type_, typename kernels_type_>
	static inline void call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(
	    const in_ptr0_type_& in_ptr0,
	    const in_ptr1_type_& in_ptr1,
	    const out_ptr0_type_& out_ptr0,
	    int64_t xnumel,
	    int32_t device_idx_,
	    cudaStream_t stream_,
	    kernels_type_& kernels_,
	    const std::optional<std::string>& cubin_dir_ = std::nullopt
	){
	    /*
	    async_compile.triton('triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7', '''
	    import triton
	    import triton.language as tl
	
	    from torch._inductor.runtime import triton_helpers, triton_heuristics
	    from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
	    from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, DeviceProperties
	    triton_helpers.set_driver_to_gpu()
	
	    @triton_heuristics.pointwise(
	        size_hints={'x': 8388608}, 
	        filename=__file__,
	        triton_meta={'signature': {'in_ptr0': '*fp32', 'in_ptr1': '*fp32', 'out_ptr0': '*fp32', 'xnumel': 'i32', 'XBLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=108, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, warp_size=32), 'constants': {}, 'configs': [{(0,): [['tt.divisibility', 16]], (1,): [['tt.divisibility', 16]], (2,): [['tt.divisibility', 16]], (3,): [['tt.divisibility', 16]]}]},
	        inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7', 'mutated_arg_names': [], 'optimize_mem': True, 'no_x_dim': False, 'num_load': 2, 'num_reduction': 0, 'backend_hash': '07F8534F6C2EC4E11ACA26C8C9D43565780E009807A566E459118140DB4BC4AB', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': True, 'is_fbcode': True},
	        min_elem_per_thread=0
	    )
	    @triton.jit
	    def triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
	        xoffset = tl.program_id(0) * XBLOCK
	        xindex = xoffset + tl.arange(0, XBLOCK)[:]
	        xmask = xindex < xnumel
	        x0 = (xindex % 64)
	        x1 = ((xindex // 64) % 1500)
	        x2 = ((xindex // 96000) % 20)
	        x3 = xindex // 1920000
	        x4 = xindex
	        tmp0 = tl.load(in_ptr0 + (x0 + 64*x2 + 1280*x1 + 1920000*x3), xmask)
	        tmp1 = tl.load(in_ptr1 + (x0 + 64*x2), xmask, eviction_policy='evict_last')
	        tmp2 = tmp0 + tmp1
	        tl.store(out_ptr0 + (x4), tmp2, xmask)
	    ''', device_str='cuda')
	    */
	    uint32_t grid_0 = ((xnumel + (1024 - 1)) / (1024));
	    uint32_t grid_1 = 1;
	    uint32_t grid_2 = 1;
	    if (grid_0 == 0 || grid_1 == 0 || grid_2 == 0) return;
	    if (kernels_.triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7 == nullptr) {
	        kernels_.triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7 = loadKernel("/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/cxgj4wqj4kc6vetuem44gmn47ujbpicq7fjl7enf6l7372n5gx64.cubin", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7", 0, cubin_dir_); 
	    }
	    CUdeviceptr var_83 = reinterpret_cast<CUdeviceptr>(in_ptr0.data_ptr());
	    CUdeviceptr var_84 = reinterpret_cast<CUdeviceptr>(in_ptr1.data_ptr());
	    CUdeviceptr var_85 = reinterpret_cast<CUdeviceptr>(out_ptr0.data_ptr());
	    int32_t var_86 = xnumel;
	    CUdeviceptr global_scratch_scratch_87 = 0;
	    void* kernel_args_[] = {&var_83, &var_84, &var_85, &var_86, &global_scratch_scratch_87};
	    launchKernel(kernels_.triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7, grid_0, grid_1, grid_2, 4, 0, kernel_args_, stream_);
	}
	
	template <typename in_out_ptr0_type_, typename kernels_type_>
	static inline void call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(
	    const in_out_ptr0_type_& in_out_ptr0,
	    int64_t xnumel,
	    int64_t r0_numel,
	    int32_t device_idx_,
	    cudaStream_t stream_,
	    kernels_type_& kernels_,
	    const std::optional<std::string>& cubin_dir_ = std::nullopt
	){
	    /*
	    async_compile.triton('triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8', '''
	    import triton
	    import triton.language as tl
	
	    from torch._inductor.runtime import triton_helpers, triton_heuristics
	    from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
	    from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, DeviceProperties
	    triton_helpers.set_driver_to_gpu()
	
	    @triton_heuristics.reduction(
	        size_hints={'x': 8192, 'r0_': 2048},
	        reduction_hint=ReductionHint.INNER,
	        filename=__file__,
	        triton_meta={'signature': {'in_out_ptr0': '*fp32', 'xnumel': 'i32', 'r0_numel': 'i32', 'XBLOCK': 'constexpr', 'R0_BLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=108, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, warp_size=32), 'constants': {}, 'configs': [{(0,): [['tt.divisibility', 16]], (2,): [['tt.divisibility', 16]]}]},
	        inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8', 'mutated_arg_names': ['in_out_ptr0'], 'optimize_mem': True, 'no_x_dim': False, 'num_load': 2, 'num_reduction': 2, 'backend_hash': '07F8534F6C2EC4E11ACA26C8C9D43565780E009807A566E459118140DB4BC4AB', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': True, 'is_fbcode': True}
	    )
	    @triton.jit
	    def triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(in_out_ptr0, xnumel, r0_numel, XBLOCK : tl.constexpr, R0_BLOCK : tl.constexpr):
	        r0_numel = 1280
	        rnumel = r0_numel
	        RBLOCK: tl.constexpr = R0_BLOCK
	        xoffset = tl.program_id(0) * XBLOCK
	        xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
	        xmask = xindex < xnumel
	        r0_base = tl.arange(0, R0_BLOCK)[None, :]
	        rbase = r0_base
	        x3 = xindex
	        _tmp2 = tl.full([XBLOCK, R0_BLOCK], float("-inf"), tl.float32)
	        x0 = (xindex % 1500)
	        x1 = xindex // 1500
	        _tmp4 = tl.full([XBLOCK, R0_BLOCK], float("inf"), tl.float32)
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_2 = r0_index
	            tmp0 = tl.load(in_out_ptr0 + (r0_2 + 1280*x3), r0_mask & xmask, eviction_policy='evict_last', other=0.0)
	            tmp1 = tl.broadcast_to(tmp0, [XBLOCK, R0_BLOCK])
	            tmp3 = triton_helpers.maximum(_tmp2, tmp1)
	            _tmp2 = tl.where(r0_mask & xmask, tmp3, _tmp2)
	            tmp5 = triton_helpers.minimum(_tmp4, tmp1)
	            _tmp4 = tl.where(r0_mask & xmask, tmp5, _tmp4)
	        tmp2 = triton_helpers.max2(_tmp2, 1)[:, None]
	        tmp4 = triton_helpers.min2(_tmp4, 1)[:, None]
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_2 = r0_index
	            tmp6 = tl.load(in_out_ptr0 + (r0_2 + 1280*x3), r0_mask & xmask, eviction_policy='evict_first', other=0.0)
	            tmp7 = 0.0
	            tmp8 = triton_helpers.maximum(tmp2, tmp7)
	            tmp9 = triton_helpers.minimum(tmp4, tmp7)
	            tmp10 = tmp8 - tmp9
	            tmp11 = 0.00392156862745098
	            tmp12 = tmp10 * tmp11
	            tmp13 = 1.1920928955078125e-07
	            tmp14 = triton_helpers.maximum(tmp12, tmp13)
	            tmp15 = tl.full([1, 1], 1, tl.int32)
	            tmp16 = (tmp15 / tmp14)
	            tmp17 = 1.0
	            tmp18 = tmp16 * tmp17
	            tmp19 = tmp6 * tmp18
	            tmp20 = libdevice.nearbyint(tmp19)
	            tmp21 = (tmp9 / tmp14)
	            tmp22 = libdevice.nearbyint(tmp21)
	            tmp23 = -128.0
	            tmp24 = tmp23 - tmp22
	            tmp25 = triton_helpers.maximum(tmp24, tmp23)
	            tmp26 = 127.0
	            tmp27 = triton_helpers.minimum(tmp25, tmp26)
	            tmp28 = tmp27.to(tl.int8)
	            tmp29 = tmp28.to(tl.float32)
	            tmp30 = tmp20 + tmp29
	            tmp31 = triton_helpers.maximum(tmp30, tmp23)
	            tmp32 = triton_helpers.minimum(tmp31, tmp26)
	            tmp33 = tmp32.to(tl.int8)
	            tmp34 = tmp33.to(tl.float32)
	            tmp35 = tmp34 - tmp29
	            tmp36 = tmp35 * tmp14
	            tl.store(in_out_ptr0 + (r0_2 + 1280*x3), tmp36, r0_mask & xmask)
	    ''', device_str='cuda')
	    */
	    uint32_t grid_0 = xnumel;
	    uint32_t grid_1 = 1;
	    uint32_t grid_2 = 1;
	    if (grid_0 == 0 || grid_1 == 0 || grid_2 == 0) return;
	    if (kernels_.triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8 == nullptr) {
	        kernels_.triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8 = loadKernel("/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/cwczgsigukwadvpw37ddr6344oosgalv5bknevxg7j6cyusvpy3e.cubin", "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8", 64, cubin_dir_); 
	    }
	    CUdeviceptr var_88 = reinterpret_cast<CUdeviceptr>(in_out_ptr0.data_ptr());
	    int32_t var_89 = xnumel;
	    int var_90 = r0_numel;
	    CUdeviceptr global_scratch_scratch_91 = 0;
	    void* kernel_args_[] = {&var_88, &var_89, &var_90, &global_scratch_scratch_91};
	    launchKernel(kernels_.triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8, grid_0, grid_1, grid_2, 16, 64, kernel_args_, stream_);
	}
	
	template <typename in_out_ptr0_type_, typename in_out_ptr1_type_, typename in_ptr0_type_, typename in_ptr1_type_, typename in_ptr2_type_, typename in_ptr3_type_, typename in_ptr4_type_, typename in_ptr5_type_, typename kernels_type_>
	static inline void call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9(
	    const in_out_ptr0_type_& in_out_ptr0,
	    const in_out_ptr1_type_& in_out_ptr1,
	    const in_ptr0_type_& in_ptr0,
	    const in_ptr1_type_& in_ptr1,
	    const in_ptr2_type_& in_ptr2,
	    const in_ptr3_type_& in_ptr3,
	    const in_ptr4_type_& in_ptr4,
	    const in_ptr5_type_& in_ptr5,
	    int64_t xnumel,
	    int64_t r0_numel,
	    int32_t device_idx_,
	    cudaStream_t stream_,
	    kernels_type_& kernels_,
	    const std::optional<std::string>& cubin_dir_ = std::nullopt
	){
	    /*
	    async_compile.triton('triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9', '''
	    import triton
	    import triton.language as tl
	
	    from torch._inductor.runtime import triton_helpers, triton_heuristics
	    from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
	    from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, DeviceProperties
	    triton_helpers.set_driver_to_gpu()
	
	    @triton_heuristics.reduction(
	        size_hints={'x': 8192, 'r0_': 2048},
	        reduction_hint=ReductionHint.DEFAULT,
	        filename=__file__,
	        triton_meta={'signature': {'in_out_ptr0': '*fp32', 'in_out_ptr1': '*fp32', 'in_ptr0': '*fp32', 'in_ptr1': '*fp32', 'in_ptr2': '*fp32', 'in_ptr3': '*fp32', 'in_ptr4': '*fp32', 'in_ptr5': '*fp32', 'xnumel': 'i32', 'r0_numel': 'i32', 'XBLOCK': 'constexpr', 'R0_BLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=108, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, warp_size=32), 'constants': {}, 'configs': [{(0,): [['tt.divisibility', 16]], (1,): [['tt.divisibility', 16]], (2,): [['tt.divisibility', 16]], (3,): [['tt.divisibility', 16]], (4,): [['tt.divisibility', 16]], (5,): [['tt.divisibility', 16]], (6,): [['tt.divisibility', 16]], (7,): [['tt.divisibility', 16]], (9,): [['tt.divisibility', 16]]}]},
	        inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9', 'mutated_arg_names': ['in_out_ptr0', 'in_out_ptr1'], 'optimize_mem': True, 'no_x_dim': False, 'num_load': 9, 'num_reduction': 4, 'backend_hash': '07F8534F6C2EC4E11ACA26C8C9D43565780E009807A566E459118140DB4BC4AB', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': True, 'is_fbcode': True}
	    )
	    @triton.jit
	    def triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9(in_out_ptr0, in_out_ptr1, in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, xnumel, r0_numel, XBLOCK : tl.constexpr, R0_BLOCK : tl.constexpr):
	        r0_numel = 1280
	        rnumel = r0_numel
	        RBLOCK: tl.constexpr = R0_BLOCK
	        xoffset = tl.program_id(0) * XBLOCK
	        xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
	        xmask = xindex < xnumel
	        r0_base = tl.arange(0, R0_BLOCK)[None, :]
	        rbase = r0_base
	        x0 = (xindex % 1500)
	        x1 = xindex // 1500
	        x3 = xindex
	        tmp18_mean = tl.zeros([XBLOCK, R0_BLOCK], tl.float32)
	        tmp18_m2 = tl.zeros([XBLOCK, R0_BLOCK], tl.float32)
	        tmp18_weight = tl.zeros([XBLOCK, R0_BLOCK], tl.float32)
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_2 = r0_index
	            tmp0 = tl.load(in_ptr0 + (x0 + 1500*r0_2 + 1920000*x1), r0_mask & xmask, eviction_policy='evict_first', other=0.0)
	            tmp1 = tl.load(in_ptr1 + (r0_2), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp11 = tl.load(in_ptr2 + (r0_2 + 1280*x0), r0_mask & xmask, eviction_policy='evict_last', other=0.0)
	            tmp13 = tl.load(in_out_ptr0 + (r0_2 + 1280*x3), r0_mask & xmask, eviction_policy='evict_first', other=0.0)
	            tmp14 = tl.load(in_ptr3 + (r0_2), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp2 = tmp0 + tmp1
	            tmp3 = 0.5
	            tmp4 = tmp2 * tmp3
	            tmp5 = 0.7071067811865476
	            tmp6 = tmp2 * tmp5
	            tmp7 = libdevice.erf(tmp6)
	            tmp8 = 1.0
	            tmp9 = tmp7 + tmp8
	            tmp10 = tmp4 * tmp9
	            tmp12 = tmp10 + tmp11
	            tmp15 = tmp13 + tmp14
	            tmp16 = tmp12 + tmp15
	            tmp17 = tl.broadcast_to(tmp16, [XBLOCK, R0_BLOCK])
	            tmp18_mean_next, tmp18_m2_next, tmp18_weight_next = triton_helpers.welford_reduce(
	                tmp17, tmp18_mean, tmp18_m2, tmp18_weight, roffset == 0
	            )
	            tmp18_mean = tl.where(r0_mask & xmask, tmp18_mean_next, tmp18_mean)
	            tmp18_m2 = tl.where(r0_mask & xmask, tmp18_m2_next, tmp18_m2)
	            tmp18_weight = tl.where(r0_mask & xmask, tmp18_weight_next, tmp18_weight)
	            tl.store(in_out_ptr0 + (r0_2 + 1280*x3), tmp16, r0_mask & xmask)
	        tmp19, tmp20, tmp21 = triton_helpers.welford(tmp18_mean, tmp18_m2, tmp18_weight, 1)
	        tmp18 = tmp19[:, None]
	        tmp22 = tmp20[:, None]
	        tmp23 = tmp21[:, None]
	        _tmp37 = tl.full([XBLOCK, R0_BLOCK], float("-inf"), tl.float32)
	        _tmp39 = tl.full([XBLOCK, R0_BLOCK], float("inf"), tl.float32)
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_2 = r0_index
	            tmp24 = tl.load(in_out_ptr0 + (r0_2 + 1280*x3), r0_mask & xmask, eviction_policy='evict_first', other=0.0)
	            tmp32 = tl.load(in_ptr4 + (r0_2), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp34 = tl.load(in_ptr5 + (r0_2), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp25 = tmp24 - tmp18
	            tmp26 = 1280.0
	            tmp27 = (tmp22 / tmp26)
	            tmp28 = 1e-05
	            tmp29 = tmp27 + tmp28
	            tmp30 = libdevice.rsqrt(tmp29)
	            tmp31 = tmp25 * tmp30
	            tmp33 = tmp31 * tmp32
	            tmp35 = tmp33 + tmp34
	            tmp36 = tl.broadcast_to(tmp35, [XBLOCK, R0_BLOCK])
	            tmp38 = triton_helpers.maximum(_tmp37, tmp36)
	            _tmp37 = tl.where(r0_mask & xmask, tmp38, _tmp37)
	            tmp40 = triton_helpers.minimum(_tmp39, tmp36)
	            _tmp39 = tl.where(r0_mask & xmask, tmp40, _tmp39)
	            tl.store(in_out_ptr1 + (r0_2 + 1280*x3), tmp35, r0_mask & xmask)
	        tmp37 = triton_helpers.max2(_tmp37, 1)[:, None]
	        tmp39 = triton_helpers.min2(_tmp39, 1)[:, None]
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_2 = r0_index
	            tmp41 = tl.load(in_out_ptr1 + (r0_2 + 1280*x3), r0_mask & xmask, eviction_policy='evict_first', other=0.0)
	            tmp42 = 0.0
	            tmp43 = triton_helpers.maximum(tmp37, tmp42)
	            tmp44 = triton_helpers.minimum(tmp39, tmp42)
	            tmp45 = tmp43 - tmp44
	            tmp46 = 0.00392156862745098
	            tmp47 = tmp45 * tmp46
	            tmp48 = 1.1920928955078125e-07
	            tmp49 = triton_helpers.maximum(tmp47, tmp48)
	            tmp50 = tl.full([1, 1], 1, tl.int32)
	            tmp51 = (tmp50 / tmp49)
	            tmp52 = 1.0
	            tmp53 = tmp51 * tmp52
	            tmp54 = tmp41 * tmp53
	            tmp55 = libdevice.nearbyint(tmp54)
	            tmp56 = (tmp44 / tmp49)
	            tmp57 = libdevice.nearbyint(tmp56)
	            tmp58 = -128.0
	            tmp59 = tmp58 - tmp57
	            tmp60 = triton_helpers.maximum(tmp59, tmp58)
	            tmp61 = 127.0
	            tmp62 = triton_helpers.minimum(tmp60, tmp61)
	            tmp63 = tmp62.to(tl.int8)
	            tmp64 = tmp63.to(tl.float32)
	            tmp65 = tmp55 + tmp64
	            tmp66 = triton_helpers.maximum(tmp65, tmp58)
	            tmp67 = triton_helpers.minimum(tmp66, tmp61)
	            tmp68 = tmp67.to(tl.int8)
	            tmp69 = tmp68.to(tl.float32)
	            tmp70 = tmp69 - tmp64
	            tmp71 = tmp70 * tmp49
	            tl.store(in_out_ptr1 + (r0_2 + 1280*x3), tmp71, r0_mask & xmask)
	    ''', device_str='cuda')
	    */
	    uint32_t grid_0 = ((xnumel + (8 - 1)) / (8));
	    uint32_t grid_1 = 1;
	    uint32_t grid_2 = 1;
	    if (grid_0 == 0 || grid_1 == 0 || grid_2 == 0) return;
	    if (kernels_.triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9 == nullptr) {
	        kernels_.triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9 = loadKernel("/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/chezqj33jk55dyvgieb7j7fgrgvxnlsm2oneefitclz4asun2pel.cubin", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9", 8320, cubin_dir_); 
	    }
	    CUdeviceptr var_92 = reinterpret_cast<CUdeviceptr>(in_out_ptr0.data_ptr());
	    CUdeviceptr var_93 = reinterpret_cast<CUdeviceptr>(in_out_ptr1.data_ptr());
	    CUdeviceptr var_94 = reinterpret_cast<CUdeviceptr>(in_ptr0.data_ptr());
	    CUdeviceptr var_95 = reinterpret_cast<CUdeviceptr>(in_ptr1.data_ptr());
	    CUdeviceptr var_96 = reinterpret_cast<CUdeviceptr>(in_ptr2.data_ptr());
	    CUdeviceptr var_97 = reinterpret_cast<CUdeviceptr>(in_ptr3.data_ptr());
	    CUdeviceptr var_98 = reinterpret_cast<CUdeviceptr>(in_ptr4.data_ptr());
	    CUdeviceptr var_99 = reinterpret_cast<CUdeviceptr>(in_ptr5.data_ptr());
	    int32_t var_100 = xnumel;
	    int var_101 = r0_numel;
	    CUdeviceptr global_scratch_scratch_102 = 0;
	    void* kernel_args_[] = {&var_92, &var_93, &var_94, &var_95, &var_96, &var_97, &var_98, &var_99, &var_100, &var_101, &global_scratch_scratch_102};
	    launchKernel(kernels_.triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9, grid_0, grid_1, grid_2, 16, 8320, kernel_args_, stream_);
	}
	
	template <typename in_ptr0_type_, typename in_ptr1_type_, typename in_ptr2_type_, typename out_ptr0_type_, typename kernels_type_>
	static inline void call_triton_poi_fused__to_copy_mul_sub_view_10(
	    const in_ptr0_type_& in_ptr0,
	    const in_ptr1_type_& in_ptr1,
	    const in_ptr2_type_& in_ptr2,
	    const out_ptr0_type_& out_ptr0,
	    int64_t xnumel,
	    int32_t device_idx_,
	    cudaStream_t stream_,
	    kernels_type_& kernels_,
	    const std::optional<std::string>& cubin_dir_ = std::nullopt
	){
	    /*
	    async_compile.triton('triton_poi_fused__to_copy_mul_sub_view_10', '''
	    import triton
	    import triton.language as tl
	
	    from torch._inductor.runtime import triton_helpers, triton_heuristics
	    from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
	    from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, DeviceProperties
	    triton_helpers.set_driver_to_gpu()
	
	    @triton_heuristics.pointwise(
	        size_hints={'x': 8388608}, 
	        filename=__file__,
	        triton_meta={'signature': {'in_ptr0': '*i8', 'in_ptr1': '*i8', 'in_ptr2': '*fp32', 'out_ptr0': '*fp32', 'xnumel': 'i32', 'XBLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=108, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, warp_size=32), 'constants': {}, 'configs': [{(0,): [['tt.divisibility', 16]], (1,): [['tt.divisibility', 16]], (2,): [['tt.divisibility', 16]], (3,): [['tt.divisibility', 16]], (4,): [['tt.divisibility', 16]]}]},
	        inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__to_copy_mul_sub_view_10', 'mutated_arg_names': [], 'optimize_mem': True, 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': '07F8534F6C2EC4E11ACA26C8C9D43565780E009807A566E459118140DB4BC4AB', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': True, 'is_fbcode': True},
	        min_elem_per_thread=0
	    )
	    @triton.jit
	    def triton_poi_fused__to_copy_mul_sub_view_10(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
	        xnumel = 6553600
	        xoffset = tl.program_id(0) * XBLOCK
	        xindex = xoffset + tl.arange(0, XBLOCK)[:]
	        xmask = tl.full([XBLOCK], True, tl.int1)
	        x2 = xindex
	        x1 = xindex // 32
	        tmp0 = tl.load(in_ptr0 + (x2), None)
	        tmp2 = tl.load(in_ptr1 + (x1), None, eviction_policy='evict_last')
	        tmp5 = tl.load(in_ptr2 + (x1), None, eviction_policy='evict_last')
	        tmp1 = tmp0.to(tl.float32)
	        tmp3 = tmp2.to(tl.float32)
	        tmp4 = tmp1 - tmp3
	        tmp6 = tmp4 * tmp5
	        tl.store(out_ptr0 + (x2), tmp6, None)
	    ''', device_str='cuda')
	    */
	    uint32_t grid_0 = ((xnumel + (1024 - 1)) / (1024));
	    uint32_t grid_1 = 1;
	    uint32_t grid_2 = 1;
	    if (grid_0 == 0 || grid_1 == 0 || grid_2 == 0) return;
	    if (kernels_.triton_poi_fused__to_copy_mul_sub_view_10 == nullptr) {
	        kernels_.triton_poi_fused__to_copy_mul_sub_view_10 = loadKernel("/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/cg2nd3euryckc65lul3e7h7paqhe3qg763kswlxjj2oehnaarv25.cubin", "triton_poi_fused__to_copy_mul_sub_view_10", 4096, cubin_dir_); 
	    }
	    CUdeviceptr var_103 = reinterpret_cast<CUdeviceptr>(in_ptr0.data_ptr());
	    CUdeviceptr var_104 = reinterpret_cast<CUdeviceptr>(in_ptr1.data_ptr());
	    CUdeviceptr var_105 = reinterpret_cast<CUdeviceptr>(in_ptr2.data_ptr());
	    CUdeviceptr var_106 = reinterpret_cast<CUdeviceptr>(out_ptr0.data_ptr());
	    int var_107 = xnumel;
	    CUdeviceptr global_scratch_scratch_108 = 0;
	    void* kernel_args_[] = {&var_103, &var_104, &var_105, &var_106, &var_107, &global_scratch_scratch_108};
	    launchKernel(kernels_.triton_poi_fused__to_copy_mul_sub_view_10, grid_0, grid_1, grid_2, 4, 4096, kernel_args_, stream_);
	}
	
	template <typename in_out_ptr0_type_, typename in_ptr0_type_, typename kernels_type_>
	static inline void call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(
	    const in_out_ptr0_type_& in_out_ptr0,
	    const in_ptr0_type_& in_ptr0,
	    int64_t xnumel,
	    int64_t r0_numel,
	    int32_t device_idx_,
	    cudaStream_t stream_,
	    kernels_type_& kernels_,
	    const std::optional<std::string>& cubin_dir_ = std::nullopt
	){
	    /*
	    async_compile.triton('triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11', '''
	    import triton
	    import triton.language as tl
	
	    from torch._inductor.runtime import triton_helpers, triton_heuristics
	    from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
	    from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, DeviceProperties
	    triton_helpers.set_driver_to_gpu()
	
	    @triton_heuristics.reduction(
	        size_hints={'x': 8192, 'r0_': 8192},
	        reduction_hint=ReductionHint.INNER,
	        filename=__file__,
	        triton_meta={'signature': {'in_out_ptr0': '*fp32', 'in_ptr0': '*fp32', 'xnumel': 'i32', 'r0_numel': 'i32', 'XBLOCK': 'constexpr', 'R0_BLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=108, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, warp_size=32), 'constants': {}, 'configs': [{(0,): [['tt.divisibility', 16]], (1,): [['tt.divisibility', 16]], (3,): [['tt.divisibility', 16]]}]},
	        inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11', 'mutated_arg_names': ['in_out_ptr0'], 'optimize_mem': True, 'no_x_dim': False, 'num_load': 4, 'num_reduction': 2, 'backend_hash': '07F8534F6C2EC4E11ACA26C8C9D43565780E009807A566E459118140DB4BC4AB', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': True, 'is_fbcode': True}
	    )
	    @triton.jit
	    def triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(in_out_ptr0, in_ptr0, xnumel, r0_numel, XBLOCK : tl.constexpr, R0_BLOCK : tl.constexpr):
	        r0_numel = 5120
	        rnumel = r0_numel
	        RBLOCK: tl.constexpr = R0_BLOCK
	        xoffset = tl.program_id(0) * XBLOCK
	        xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
	        xmask = xindex < xnumel
	        r0_base = tl.arange(0, R0_BLOCK)[None, :]
	        rbase = r0_base
	        x3 = xindex
	        _tmp12 = tl.full([XBLOCK, R0_BLOCK], float("-inf"), tl.float32)
	        x0 = (xindex % 1500)
	        x1 = xindex // 1500
	        _tmp14 = tl.full([XBLOCK, R0_BLOCK], float("inf"), tl.float32)
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_2 = r0_index
	            tmp0 = tl.load(in_out_ptr0 + (r0_2 + 5120*x3), r0_mask & xmask, eviction_policy='evict_last', other=0.0)
	            tmp1 = tl.load(in_ptr0 + (r0_2), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp2 = tmp0 + tmp1
	            tmp3 = 0.5
	            tmp4 = tmp2 * tmp3
	            tmp5 = 0.7071067811865476
	            tmp6 = tmp2 * tmp5
	            tmp7 = libdevice.erf(tmp6)
	            tmp8 = 1.0
	            tmp9 = tmp7 + tmp8
	            tmp10 = tmp4 * tmp9
	            tmp11 = tl.broadcast_to(tmp10, [XBLOCK, R0_BLOCK])
	            tmp13 = triton_helpers.maximum(_tmp12, tmp11)
	            _tmp12 = tl.where(r0_mask & xmask, tmp13, _tmp12)
	            tmp15 = triton_helpers.minimum(_tmp14, tmp11)
	            _tmp14 = tl.where(r0_mask & xmask, tmp15, _tmp14)
	        tmp12 = triton_helpers.max2(_tmp12, 1)[:, None]
	        tmp14 = triton_helpers.min2(_tmp14, 1)[:, None]
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_2 = r0_index
	            tmp16 = tl.load(in_out_ptr0 + (r0_2 + 5120*x3), r0_mask & xmask, eviction_policy='evict_first', other=0.0)
	            tmp17 = tl.load(in_ptr0 + (r0_2), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp18 = tmp16 + tmp17
	            tmp19 = 0.5
	            tmp20 = tmp18 * tmp19
	            tmp21 = 0.7071067811865476
	            tmp22 = tmp18 * tmp21
	            tmp23 = libdevice.erf(tmp22)
	            tmp24 = 1.0
	            tmp25 = tmp23 + tmp24
	            tmp26 = tmp20 * tmp25
	            tmp27 = 0.0
	            tmp28 = triton_helpers.maximum(tmp12, tmp27)
	            tmp29 = triton_helpers.minimum(tmp14, tmp27)
	            tmp30 = tmp28 - tmp29
	            tmp31 = 0.00392156862745098
	            tmp32 = tmp30 * tmp31
	            tmp33 = 1.1920928955078125e-07
	            tmp34 = triton_helpers.maximum(tmp32, tmp33)
	            tmp35 = tl.full([1, 1], 1, tl.int32)
	            tmp36 = (tmp35 / tmp34)
	            tmp37 = tmp36 * tmp24
	            tmp38 = tmp26 * tmp37
	            tmp39 = libdevice.nearbyint(tmp38)
	            tmp40 = (tmp29 / tmp34)
	            tmp41 = libdevice.nearbyint(tmp40)
	            tmp42 = -128.0
	            tmp43 = tmp42 - tmp41
	            tmp44 = triton_helpers.maximum(tmp43, tmp42)
	            tmp45 = 127.0
	            tmp46 = triton_helpers.minimum(tmp44, tmp45)
	            tmp47 = tmp46.to(tl.int8)
	            tmp48 = tmp47.to(tl.float32)
	            tmp49 = tmp39 + tmp48
	            tmp50 = triton_helpers.maximum(tmp49, tmp42)
	            tmp51 = triton_helpers.minimum(tmp50, tmp45)
	            tmp52 = tmp51.to(tl.int8)
	            tmp53 = tmp52.to(tl.float32)
	            tmp54 = tmp53 - tmp48
	            tmp55 = tmp54 * tmp34
	            tl.store(in_out_ptr0 + (r0_2 + 5120*x3), tmp55, r0_mask & xmask)
	    ''', device_str='cuda')
	    */
	    uint32_t grid_0 = xnumel;
	    uint32_t grid_1 = 1;
	    uint32_t grid_2 = 1;
	    if (grid_0 == 0 || grid_1 == 0 || grid_2 == 0) return;
	    if (kernels_.triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11 == nullptr) {
	        kernels_.triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11 = loadKernel("/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/cavazqc3gkdeptxgbzrvvuqymx6g547trodmyi56gbkskogtfqtu.cubin", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11", 64, cubin_dir_); 
	    }
	    CUdeviceptr var_109 = reinterpret_cast<CUdeviceptr>(in_out_ptr0.data_ptr());
	    CUdeviceptr var_110 = reinterpret_cast<CUdeviceptr>(in_ptr0.data_ptr());
	    int32_t var_111 = xnumel;
	    int var_112 = r0_numel;
	    CUdeviceptr global_scratch_scratch_113 = 0;
	    void* kernel_args_[] = {&var_109, &var_110, &var_111, &var_112, &global_scratch_scratch_113};
	    launchKernel(kernels_.triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11, grid_0, grid_1, grid_2, 16, 64, kernel_args_, stream_);
	}
	
	template <typename in_out_ptr0_type_, typename in_out_ptr1_type_, typename in_out_ptr2_type_, typename in_ptr0_type_, typename in_ptr1_type_, typename in_ptr2_type_, typename in_ptr3_type_, typename in_ptr4_type_, typename out_ptr2_type_, typename kernels_type_>
	static inline void call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(
	    const in_out_ptr0_type_& in_out_ptr0,
	    const in_out_ptr1_type_& in_out_ptr1,
	    const in_out_ptr2_type_& in_out_ptr2,
	    const in_ptr0_type_& in_ptr0,
	    const in_ptr1_type_& in_ptr1,
	    const in_ptr2_type_& in_ptr2,
	    const in_ptr3_type_& in_ptr3,
	    const in_ptr4_type_& in_ptr4,
	    const out_ptr2_type_& out_ptr2,
	    int64_t xnumel,
	    int64_t r0_numel,
	    int32_t device_idx_,
	    cudaStream_t stream_,
	    kernels_type_& kernels_,
	    const std::optional<std::string>& cubin_dir_ = std::nullopt
	){
	    /*
	    async_compile.triton('triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12', '''
	    import triton
	    import triton.language as tl
	
	    from torch._inductor.runtime import triton_helpers, triton_heuristics
	    from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
	    from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, DeviceProperties
	    triton_helpers.set_driver_to_gpu()
	
	    @triton_heuristics.reduction(
	        size_hints={'x': 8192, 'r0_': 2048},
	        reduction_hint=ReductionHint.INNER,
	        filename=__file__,
	        triton_meta={'signature': {'in_out_ptr0': '*fp32', 'in_out_ptr1': '*fp32', 'in_out_ptr2': '*fp32', 'in_ptr0': '*fp32', 'in_ptr1': '*fp32', 'in_ptr2': '*fp32', 'in_ptr3': '*fp32', 'in_ptr4': '*fp32', 'out_ptr2': '*fp32', 'xnumel': 'i32', 'r0_numel': 'i32', 'XBLOCK': 'constexpr', 'R0_BLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=108, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, warp_size=32), 'constants': {}, 'configs': [{(0,): [['tt.divisibility', 16]], (1,): [['tt.divisibility', 16]], (2,): [['tt.divisibility', 16]], (3,): [['tt.divisibility', 16]], (4,): [['tt.divisibility', 16]], (5,): [['tt.divisibility', 16]], (6,): [['tt.divisibility', 16]], (7,): [['tt.divisibility', 16]], (8,): [['tt.divisibility', 16]], (10,): [['tt.divisibility', 16]]}]},
	        inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12', 'mutated_arg_names': ['in_out_ptr0', 'in_out_ptr1', 'in_out_ptr2'], 'optimize_mem': True, 'no_x_dim': False, 'num_load': 9, 'num_reduction': 8, 'backend_hash': '07F8534F6C2EC4E11ACA26C8C9D43565780E009807A566E459118140DB4BC4AB', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': True, 'is_fbcode': True}
	    )
	    @triton.jit
	    def triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(in_out_ptr0, in_out_ptr1, in_out_ptr2, in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, out_ptr2, xnumel, r0_numel, XBLOCK : tl.constexpr, R0_BLOCK : tl.constexpr):
	        r0_numel = 1280
	        rnumel = r0_numel
	        RBLOCK: tl.constexpr = R0_BLOCK
	        xoffset = tl.program_id(0) * XBLOCK
	        xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
	        xmask = xindex < xnumel
	        r0_base = tl.arange(0, R0_BLOCK)[None, :]
	        rbase = r0_base
	        x3 = xindex
	        tmp6_mean = tl.zeros([XBLOCK, R0_BLOCK], tl.float32)
	        tmp6_m2 = tl.zeros([XBLOCK, R0_BLOCK], tl.float32)
	        tmp6_weight = tl.zeros([XBLOCK, R0_BLOCK], tl.float32)
	        x0 = (xindex % 1500)
	        x1 = xindex // 1500
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_2 = r0_index
	            tmp0 = tl.load(in_ptr0 + (r0_2 + 1280*x3), r0_mask & xmask, eviction_policy='evict_last', other=0.0)
	            tmp1 = tl.load(in_ptr1 + (r0_2 + 1280*x3), r0_mask & xmask, eviction_policy='evict_last', other=0.0)
	            tmp2 = tl.load(in_ptr2 + (r0_2), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp3 = tmp1 + tmp2
	            tmp4 = tmp0 + tmp3
	            tmp5 = tl.broadcast_to(tmp4, [XBLOCK, R0_BLOCK])
	            tmp6_mean_next, tmp6_m2_next, tmp6_weight_next = triton_helpers.welford_reduce(
	                tmp5, tmp6_mean, tmp6_m2, tmp6_weight, roffset == 0
	            )
	            tmp6_mean = tl.where(r0_mask & xmask, tmp6_mean_next, tmp6_mean)
	            tmp6_m2 = tl.where(r0_mask & xmask, tmp6_m2_next, tmp6_m2)
	            tmp6_weight = tl.where(r0_mask & xmask, tmp6_weight_next, tmp6_weight)
	        tmp7, tmp8, tmp9 = triton_helpers.welford(tmp6_mean, tmp6_m2, tmp6_weight, 1)
	        tmp6 = tmp7[:, None]
	        tmp10 = tmp8[:, None]
	        tmp11 = tmp9[:, None]
	        _tmp29 = tl.full([XBLOCK, R0_BLOCK], float("-inf"), tl.float32)
	        _tmp31 = tl.full([XBLOCK, R0_BLOCK], float("inf"), tl.float32)
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_2 = r0_index
	            tmp12 = tl.load(in_ptr0 + (r0_2 + 1280*x3), r0_mask & xmask, eviction_policy='evict_first', other=0.0)
	            tmp13 = tl.load(in_ptr1 + (r0_2 + 1280*x3), r0_mask & xmask, eviction_policy='evict_first', other=0.0)
	            tmp14 = tl.load(in_ptr2 + (r0_2), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp24 = tl.load(in_ptr3 + (r0_2), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp26 = tl.load(in_ptr4 + (r0_2), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp15 = tmp13 + tmp14
	            tmp16 = tmp12 + tmp15
	            tmp17 = tmp16 - tmp6
	            tmp18 = 1280.0
	            tmp19 = (tmp10 / tmp18)
	            tmp20 = 1e-05
	            tmp21 = tmp19 + tmp20
	            tmp22 = libdevice.rsqrt(tmp21)
	            tmp23 = tmp17 * tmp22
	            tmp25 = tmp23 * tmp24
	            tmp27 = tmp25 + tmp26
	            tmp28 = tl.broadcast_to(tmp27, [XBLOCK, R0_BLOCK])
	            tmp30 = triton_helpers.maximum(_tmp29, tmp28)
	            _tmp29 = tl.where(r0_mask & xmask, tmp30, _tmp29)
	            tmp32 = triton_helpers.minimum(_tmp31, tmp28)
	            _tmp31 = tl.where(r0_mask & xmask, tmp32, _tmp31)
	            tl.store(out_ptr2 + (r0_2 + 1280*x3), tmp27, r0_mask & xmask)
	        tmp29 = triton_helpers.max2(_tmp29, 1)[:, None]
	        tmp31 = triton_helpers.min2(_tmp31, 1)[:, None]
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_2 = r0_index
	            tmp33 = tl.load(out_ptr2 + (r0_2 + 1280*x3), r0_mask & xmask, eviction_policy='evict_first', other=0.0)
	            tmp34 = 0.0
	            tmp35 = triton_helpers.maximum(tmp29, tmp34)
	            tmp36 = triton_helpers.minimum(tmp31, tmp34)
	            tmp37 = tmp35 - tmp36
	            tmp38 = 0.00392156862745098
	            tmp39 = tmp37 * tmp38
	            tmp40 = 1.1920928955078125e-07
	            tmp41 = triton_helpers.maximum(tmp39, tmp40)
	            tmp42 = tl.full([1, 1], 1, tl.int32)
	            tmp43 = (tmp42 / tmp41)
	            tmp44 = 1.0
	            tmp45 = tmp43 * tmp44
	            tmp46 = tmp33 * tmp45
	            tmp47 = libdevice.nearbyint(tmp46)
	            tmp48 = (tmp36 / tmp41)
	            tmp49 = libdevice.nearbyint(tmp48)
	            tmp50 = -128.0
	            tmp51 = tmp50 - tmp49
	            tmp52 = triton_helpers.maximum(tmp51, tmp50)
	            tmp53 = 127.0
	            tmp54 = triton_helpers.minimum(tmp52, tmp53)
	            tmp55 = tmp54.to(tl.int8)
	            tmp56 = tmp55.to(tl.float32)
	            tmp57 = tmp47 + tmp56
	            tmp58 = triton_helpers.maximum(tmp57, tmp50)
	            tmp59 = triton_helpers.minimum(tmp58, tmp53)
	            tmp60 = tmp59.to(tl.int8)
	            tmp61 = tmp60.to(tl.float32)
	            tmp62 = tmp61 - tmp56
	            tmp63 = tmp62 * tmp41
	            tl.store(in_out_ptr0 + (r0_2 + 1280*x3), tmp63, r0_mask & xmask)
	            tl.store(in_out_ptr1 + (r0_2 + 1280*x3), tmp63, r0_mask & xmask)
	            tl.store(in_out_ptr2 + (r0_2 + 1280*x3), tmp63, r0_mask & xmask)
	    ''', device_str='cuda')
	    */
	    uint32_t grid_0 = xnumel;
	    uint32_t grid_1 = 1;
	    uint32_t grid_2 = 1;
	    if (grid_0 == 0 || grid_1 == 0 || grid_2 == 0) return;
	    if (kernels_.triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12 == nullptr) {
	        kernels_.triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12 = loadKernel("/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/c5l6ujih7wrghszprumdzl3qefy6u44hwgsfg4b2llgas6cg43mk.cubin", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12", 96, cubin_dir_); 
	    }
	    CUdeviceptr var_114 = reinterpret_cast<CUdeviceptr>(in_out_ptr0.data_ptr());
	    CUdeviceptr var_115 = reinterpret_cast<CUdeviceptr>(in_out_ptr1.data_ptr());
	    CUdeviceptr var_116 = reinterpret_cast<CUdeviceptr>(in_out_ptr2.data_ptr());
	    CUdeviceptr var_117 = reinterpret_cast<CUdeviceptr>(in_ptr0.data_ptr());
	    CUdeviceptr var_118 = reinterpret_cast<CUdeviceptr>(in_ptr1.data_ptr());
	    CUdeviceptr var_119 = reinterpret_cast<CUdeviceptr>(in_ptr2.data_ptr());
	    CUdeviceptr var_120 = reinterpret_cast<CUdeviceptr>(in_ptr3.data_ptr());
	    CUdeviceptr var_121 = reinterpret_cast<CUdeviceptr>(in_ptr4.data_ptr());
	    CUdeviceptr var_122 = reinterpret_cast<CUdeviceptr>(out_ptr2.data_ptr());
	    int32_t var_123 = xnumel;
	    int var_124 = r0_numel;
	    CUdeviceptr global_scratch_scratch_125 = 0;
	    void* kernel_args_[] = {&var_114, &var_115, &var_116, &var_117, &var_118, &var_119, &var_120, &var_121, &var_122, &var_123, &var_124, &global_scratch_scratch_125};
	    launchKernel(kernels_.triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12, grid_0, grid_1, grid_2, 8, 96, kernel_args_, stream_);
	}
	
	template <typename in_out_ptr0_type_, typename in_out_ptr1_type_, typename in_ptr0_type_, typename in_ptr1_type_, typename in_ptr2_type_, typename in_ptr3_type_, typename in_ptr4_type_, typename in_ptr5_type_, typename kernels_type_>
	static inline void call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(
	    const in_out_ptr0_type_& in_out_ptr0,
	    const in_out_ptr1_type_& in_out_ptr1,
	    const in_ptr0_type_& in_ptr0,
	    const in_ptr1_type_& in_ptr1,
	    const in_ptr2_type_& in_ptr2,
	    const in_ptr3_type_& in_ptr3,
	    const in_ptr4_type_& in_ptr4,
	    const in_ptr5_type_& in_ptr5,
	    int64_t xnumel,
	    int64_t r0_numel,
	    int32_t device_idx_,
	    cudaStream_t stream_,
	    kernels_type_& kernels_,
	    const std::optional<std::string>& cubin_dir_ = std::nullopt
	){
	    /*
	    async_compile.triton('triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13', '''
	    import triton
	    import triton.language as tl
	
	    from torch._inductor.runtime import triton_helpers, triton_heuristics
	    from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
	    from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, DeviceProperties
	    triton_helpers.set_driver_to_gpu()
	
	    @triton_heuristics.reduction(
	        size_hints={'x': 8192, 'r0_': 2048},
	        reduction_hint=ReductionHint.INNER,
	        filename=__file__,
	        triton_meta={'signature': {'in_out_ptr0': '*fp32', 'in_out_ptr1': '*fp32', 'in_ptr0': '*fp32', 'in_ptr1': '*fp32', 'in_ptr2': '*fp32', 'in_ptr3': '*fp32', 'in_ptr4': '*fp32', 'in_ptr5': '*fp32', 'xnumel': 'i32', 'r0_numel': 'i32', 'XBLOCK': 'constexpr', 'R0_BLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=108, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, warp_size=32), 'constants': {}, 'configs': [{(0,): [['tt.divisibility', 16]], (1,): [['tt.divisibility', 16]], (2,): [['tt.divisibility', 16]], (3,): [['tt.divisibility', 16]], (4,): [['tt.divisibility', 16]], (5,): [['tt.divisibility', 16]], (6,): [['tt.divisibility', 16]], (7,): [['tt.divisibility', 16]], (9,): [['tt.divisibility', 16]]}]},
	        inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13', 'mutated_arg_names': ['in_out_ptr0', 'in_out_ptr1'], 'optimize_mem': True, 'no_x_dim': False, 'num_load': 9, 'num_reduction': 4, 'backend_hash': '07F8534F6C2EC4E11ACA26C8C9D43565780E009807A566E459118140DB4BC4AB', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': True, 'is_fbcode': True}
	    )
	    @triton.jit
	    def triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(in_out_ptr0, in_out_ptr1, in_ptr0, in_ptr1, in_ptr2, in_ptr3, in_ptr4, in_ptr5, xnumel, r0_numel, XBLOCK : tl.constexpr, R0_BLOCK : tl.constexpr):
	        r0_numel = 1280
	        rnumel = r0_numel
	        RBLOCK: tl.constexpr = R0_BLOCK
	        xoffset = tl.program_id(0) * XBLOCK
	        xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
	        xmask = xindex < xnumel
	        r0_base = tl.arange(0, R0_BLOCK)[None, :]
	        rbase = r0_base
	        x0 = xindex
	        tmp10_mean = tl.zeros([XBLOCK, R0_BLOCK], tl.float32)
	        tmp10_m2 = tl.zeros([XBLOCK, R0_BLOCK], tl.float32)
	        tmp10_weight = tl.zeros([XBLOCK, R0_BLOCK], tl.float32)
	        x2 = (xindex % 1500)
	        x3 = xindex // 1500
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_1 = r0_index
	            tmp0 = tl.load(in_out_ptr0 + (r0_1 + 1280*x0), r0_mask & xmask, eviction_policy='evict_first', other=0.0)
	            tmp1 = tl.load(in_ptr0 + (r0_1 + 1280*x0), r0_mask & xmask, eviction_policy='evict_first', other=0.0)
	            tmp2 = tl.load(in_ptr1 + (r0_1), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp5 = tl.load(in_ptr2 + (r0_1 + 1280*x0), r0_mask & xmask, eviction_policy='evict_first', other=0.0)
	            tmp6 = tl.load(in_ptr3 + (r0_1), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp3 = tmp1 + tmp2
	            tmp4 = tmp0 + tmp3
	            tmp7 = tmp5 + tmp6
	            tmp8 = tmp4 + tmp7
	            tmp9 = tl.broadcast_to(tmp8, [XBLOCK, R0_BLOCK])
	            tmp10_mean_next, tmp10_m2_next, tmp10_weight_next = triton_helpers.welford_reduce(
	                tmp9, tmp10_mean, tmp10_m2, tmp10_weight, roffset == 0
	            )
	            tmp10_mean = tl.where(r0_mask & xmask, tmp10_mean_next, tmp10_mean)
	            tmp10_m2 = tl.where(r0_mask & xmask, tmp10_m2_next, tmp10_m2)
	            tmp10_weight = tl.where(r0_mask & xmask, tmp10_weight_next, tmp10_weight)
	            tl.store(in_out_ptr0 + (r0_1 + 1280*x0), tmp8, r0_mask & xmask)
	        tmp11, tmp12, tmp13 = triton_helpers.welford(tmp10_mean, tmp10_m2, tmp10_weight, 1)
	        tmp10 = tmp11[:, None]
	        tmp14 = tmp12[:, None]
	        tmp15 = tmp13[:, None]
	        _tmp29 = tl.full([XBLOCK, R0_BLOCK], float("-inf"), tl.float32)
	        _tmp31 = tl.full([XBLOCK, R0_BLOCK], float("inf"), tl.float32)
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_1 = r0_index
	            tmp16 = tl.load(in_out_ptr0 + (r0_1 + 1280*x0), r0_mask & xmask, eviction_policy='evict_first', other=0.0)
	            tmp24 = tl.load(in_ptr4 + (r0_1), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp26 = tl.load(in_ptr5 + (r0_1), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp17 = tmp16 - tmp10
	            tmp18 = 1280.0
	            tmp19 = (tmp14 / tmp18)
	            tmp20 = 1e-05
	            tmp21 = tmp19 + tmp20
	            tmp22 = libdevice.rsqrt(tmp21)
	            tmp23 = tmp17 * tmp22
	            tmp25 = tmp23 * tmp24
	            tmp27 = tmp25 + tmp26
	            tmp28 = tl.broadcast_to(tmp27, [XBLOCK, R0_BLOCK])
	            tmp30 = triton_helpers.maximum(_tmp29, tmp28)
	            _tmp29 = tl.where(r0_mask & xmask, tmp30, _tmp29)
	            tmp32 = triton_helpers.minimum(_tmp31, tmp28)
	            _tmp31 = tl.where(r0_mask & xmask, tmp32, _tmp31)
	            tl.store(in_out_ptr1 + (r0_1 + 1280*x0), tmp27, r0_mask & xmask)
	        tmp29 = triton_helpers.max2(_tmp29, 1)[:, None]
	        tmp31 = triton_helpers.min2(_tmp31, 1)[:, None]
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_1 = r0_index
	            tmp33 = tl.load(in_out_ptr1 + (r0_1 + 1280*x0), r0_mask & xmask, eviction_policy='evict_first', other=0.0)
	            tmp34 = 0.0
	            tmp35 = triton_helpers.maximum(tmp29, tmp34)
	            tmp36 = triton_helpers.minimum(tmp31, tmp34)
	            tmp37 = tmp35 - tmp36
	            tmp38 = 0.00392156862745098
	            tmp39 = tmp37 * tmp38
	            tmp40 = 1.1920928955078125e-07
	            tmp41 = triton_helpers.maximum(tmp39, tmp40)
	            tmp42 = tl.full([1, 1], 1, tl.int32)
	            tmp43 = (tmp42 / tmp41)
	            tmp44 = 1.0
	            tmp45 = tmp43 * tmp44
	            tmp46 = tmp33 * tmp45
	            tmp47 = libdevice.nearbyint(tmp46)
	            tmp48 = (tmp36 / tmp41)
	            tmp49 = libdevice.nearbyint(tmp48)
	            tmp50 = -128.0
	            tmp51 = tmp50 - tmp49
	            tmp52 = triton_helpers.maximum(tmp51, tmp50)
	            tmp53 = 127.0
	            tmp54 = triton_helpers.minimum(tmp52, tmp53)
	            tmp55 = tmp54.to(tl.int8)
	            tmp56 = tmp55.to(tl.float32)
	            tmp57 = tmp47 + tmp56
	            tmp58 = triton_helpers.maximum(tmp57, tmp50)
	            tmp59 = triton_helpers.minimum(tmp58, tmp53)
	            tmp60 = tmp59.to(tl.int8)
	            tmp61 = tmp60.to(tl.float32)
	            tmp62 = tmp61 - tmp56
	            tmp63 = tmp62 * tmp41
	            tl.store(in_out_ptr1 + (r0_1 + 1280*x0), tmp63, r0_mask & xmask)
	    ''', device_str='cuda')
	    */
	    uint32_t grid_0 = xnumel;
	    uint32_t grid_1 = 1;
	    uint32_t grid_2 = 1;
	    if (grid_0 == 0 || grid_1 == 0 || grid_2 == 0) return;
	    if (kernels_.triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13 == nullptr) {
	        kernels_.triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13 = loadKernel("/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/cdqp6wtsnwd37woqtjjzog3teor7ainio63eiod4lzf7ma6q4dko.cubin", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13", 96, cubin_dir_); 
	    }
	    CUdeviceptr var_126 = reinterpret_cast<CUdeviceptr>(in_out_ptr0.data_ptr());
	    CUdeviceptr var_127 = reinterpret_cast<CUdeviceptr>(in_out_ptr1.data_ptr());
	    CUdeviceptr var_128 = reinterpret_cast<CUdeviceptr>(in_ptr0.data_ptr());
	    CUdeviceptr var_129 = reinterpret_cast<CUdeviceptr>(in_ptr1.data_ptr());
	    CUdeviceptr var_130 = reinterpret_cast<CUdeviceptr>(in_ptr2.data_ptr());
	    CUdeviceptr var_131 = reinterpret_cast<CUdeviceptr>(in_ptr3.data_ptr());
	    CUdeviceptr var_132 = reinterpret_cast<CUdeviceptr>(in_ptr4.data_ptr());
	    CUdeviceptr var_133 = reinterpret_cast<CUdeviceptr>(in_ptr5.data_ptr());
	    int32_t var_134 = xnumel;
	    int var_135 = r0_numel;
	    CUdeviceptr global_scratch_scratch_136 = 0;
	    void* kernel_args_[] = {&var_126, &var_127, &var_128, &var_129, &var_130, &var_131, &var_132, &var_133, &var_134, &var_135, &global_scratch_scratch_136};
	    launchKernel(kernels_.triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13, grid_0, grid_1, grid_2, 8, 96, kernel_args_, stream_);
	}
	
	template <typename in_out_ptr0_type_, typename in_ptr0_type_, typename in_ptr1_type_, typename in_ptr2_type_, typename in_ptr3_type_, typename kernels_type_>
	static inline void call_triton_red_fused_add_addmm_native_layer_norm_view_14(
	    const in_out_ptr0_type_& in_out_ptr0,
	    const in_ptr0_type_& in_ptr0,
	    const in_ptr1_type_& in_ptr1,
	    const in_ptr2_type_& in_ptr2,
	    const in_ptr3_type_& in_ptr3,
	    int64_t xnumel,
	    int64_t r0_numel,
	    int32_t device_idx_,
	    cudaStream_t stream_,
	    kernels_type_& kernels_,
	    const std::optional<std::string>& cubin_dir_ = std::nullopt
	){
	    /*
	    async_compile.triton('triton_red_fused_add_addmm_native_layer_norm_view_14', '''
	    import triton
	    import triton.language as tl
	
	    from torch._inductor.runtime import triton_helpers, triton_heuristics
	    from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
	    from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, DeviceProperties
	    triton_helpers.set_driver_to_gpu()
	
	    @triton_heuristics.reduction(
	        size_hints={'x': 8192, 'r0_': 2048},
	        reduction_hint=ReductionHint.INNER,
	        filename=__file__,
	        triton_meta={'signature': {'in_out_ptr0': '*fp32', 'in_ptr0': '*fp32', 'in_ptr1': '*fp32', 'in_ptr2': '*fp32', 'in_ptr3': '*fp32', 'xnumel': 'i32', 'r0_numel': 'i32', 'XBLOCK': 'constexpr', 'R0_BLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=108, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, warp_size=32), 'constants': {}, 'configs': [{(0,): [['tt.divisibility', 16]], (1,): [['tt.divisibility', 16]], (2,): [['tt.divisibility', 16]], (3,): [['tt.divisibility', 16]], (4,): [['tt.divisibility', 16]], (6,): [['tt.divisibility', 16]]}]},
	        inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_red_fused_add_addmm_native_layer_norm_view_14', 'mutated_arg_names': ['in_out_ptr0'], 'optimize_mem': True, 'no_x_dim': False, 'num_load': 8, 'num_reduction': 2, 'backend_hash': '07F8534F6C2EC4E11ACA26C8C9D43565780E009807A566E459118140DB4BC4AB', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': True, 'is_fbcode': True}
	    )
	    @triton.jit
	    def triton_red_fused_add_addmm_native_layer_norm_view_14(in_out_ptr0, in_ptr0, in_ptr1, in_ptr2, in_ptr3, xnumel, r0_numel, XBLOCK : tl.constexpr, R0_BLOCK : tl.constexpr):
	        r0_numel = 1280
	        rnumel = r0_numel
	        RBLOCK: tl.constexpr = R0_BLOCK
	        xoffset = tl.program_id(0) * XBLOCK
	        xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
	        xmask = xindex < xnumel
	        r0_base = tl.arange(0, R0_BLOCK)[None, :]
	        rbase = r0_base
	        x3 = xindex
	        tmp6_mean = tl.zeros([XBLOCK, R0_BLOCK], tl.float32)
	        tmp6_m2 = tl.zeros([XBLOCK, R0_BLOCK], tl.float32)
	        tmp6_weight = tl.zeros([XBLOCK, R0_BLOCK], tl.float32)
	        x0 = (xindex % 1500)
	        x1 = xindex // 1500
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_2 = r0_index
	            tmp0 = tl.load(in_out_ptr0 + (r0_2 + 1280*x3), r0_mask & xmask, eviction_policy='evict_last', other=0.0)
	            tmp1 = tl.load(in_ptr0 + (r0_2 + 1280*x3), r0_mask & xmask, eviction_policy='evict_last', other=0.0)
	            tmp2 = tl.load(in_ptr1 + (r0_2), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp3 = tmp1 + tmp2
	            tmp4 = tmp0 + tmp3
	            tmp5 = tl.broadcast_to(tmp4, [XBLOCK, R0_BLOCK])
	            tmp6_mean_next, tmp6_m2_next, tmp6_weight_next = triton_helpers.welford_reduce(
	                tmp5, tmp6_mean, tmp6_m2, tmp6_weight, roffset == 0
	            )
	            tmp6_mean = tl.where(r0_mask & xmask, tmp6_mean_next, tmp6_mean)
	            tmp6_m2 = tl.where(r0_mask & xmask, tmp6_m2_next, tmp6_m2)
	            tmp6_weight = tl.where(r0_mask & xmask, tmp6_weight_next, tmp6_weight)
	        tmp7, tmp8, tmp9 = triton_helpers.welford(tmp6_mean, tmp6_m2, tmp6_weight, 1)
	        tmp6 = tmp7[:, None]
	        tmp10 = tmp8[:, None]
	        tmp11 = tmp9[:, None]
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_2 = r0_index
	            tmp12 = tl.load(in_out_ptr0 + (r0_2 + 1280*x3), r0_mask & xmask, eviction_policy='evict_first', other=0.0)
	            tmp13 = tl.load(in_ptr0 + (r0_2 + 1280*x3), r0_mask & xmask, eviction_policy='evict_first', other=0.0)
	            tmp14 = tl.load(in_ptr1 + (r0_2), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp24 = tl.load(in_ptr2 + (r0_2), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp26 = tl.load(in_ptr3 + (r0_2), r0_mask, eviction_policy='evict_last', other=0.0)
	            tmp15 = tmp13 + tmp14
	            tmp16 = tmp12 + tmp15
	            tmp17 = tmp16 - tmp6
	            tmp18 = 1280.0
	            tmp19 = (tmp10 / tmp18)
	            tmp20 = 1e-05
	            tmp21 = tmp19 + tmp20
	            tmp22 = libdevice.rsqrt(tmp21)
	            tmp23 = tmp17 * tmp22
	            tmp25 = tmp23 * tmp24
	            tmp27 = tmp25 + tmp26
	            tl.store(in_out_ptr0 + (r0_2 + 1280*x3), tmp27, r0_mask & xmask)
	    ''', device_str='cuda')
	    */
	    uint32_t grid_0 = xnumel;
	    uint32_t grid_1 = 1;
	    uint32_t grid_2 = 1;
	    if (grid_0 == 0 || grid_1 == 0 || grid_2 == 0) return;
	    if (kernels_.triton_red_fused_add_addmm_native_layer_norm_view_14 == nullptr) {
	        kernels_.triton_red_fused_add_addmm_native_layer_norm_view_14 = loadKernel("/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/cgxis7wdf4yyzrqr4e752jpq6eunqvdkbfrcog4gic2lpowbctwe.cubin", "triton_red_fused_add_addmm_native_layer_norm_view_14", 96, cubin_dir_); 
	    }
	    CUdeviceptr var_137 = reinterpret_cast<CUdeviceptr>(in_out_ptr0.data_ptr());
	    CUdeviceptr var_138 = reinterpret_cast<CUdeviceptr>(in_ptr0.data_ptr());
	    CUdeviceptr var_139 = reinterpret_cast<CUdeviceptr>(in_ptr1.data_ptr());
	    CUdeviceptr var_140 = reinterpret_cast<CUdeviceptr>(in_ptr2.data_ptr());
	    CUdeviceptr var_141 = reinterpret_cast<CUdeviceptr>(in_ptr3.data_ptr());
	    int32_t var_142 = xnumel;
	    int var_143 = r0_numel;
	    CUdeviceptr global_scratch_scratch_144 = 0;
	    void* kernel_args_[] = {&var_137, &var_138, &var_139, &var_140, &var_141, &var_142, &var_143, &global_scratch_scratch_144};
	    launchKernel(kernels_.triton_red_fused_add_addmm_native_layer_norm_view_14, grid_0, grid_1, grid_2, 8, 96, kernel_args_, stream_);
	}
	
	template <typename in_out_ptr0_type_, typename in_ptr0_type_, typename kernels_type_>
	static inline void call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15(
	    const in_out_ptr0_type_& in_out_ptr0,
	    const in_ptr0_type_& in_ptr0,
	    int64_t ks0,
	    int64_t xnumel,
	    int64_t r0_numel,
	    int32_t device_idx_,
	    cudaStream_t stream_,
	    kernels_type_& kernels_,
	    const std::optional<std::string>& cubin_dir_ = std::nullopt
	){
	    /*
	    async_compile.triton('triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15', '''
	    import triton
	    import triton.language as tl
	
	    from torch._inductor.runtime import triton_helpers, triton_heuristics
	    from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
	    from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, DeviceProperties
	    triton_helpers.set_driver_to_gpu()
	
	    @triton_heuristics.reduction(
	        size_hints={'x': 2048, 'r0_': 8192},
	        reduction_hint=ReductionHint.INNER,
	        filename=__file__,
	        triton_meta={'signature': {'in_out_ptr0': '*fp32', 'in_ptr0': '*fp32', 'ks0': 'i64', 'xnumel': 'i32', 'r0_numel': 'i32', 'XBLOCK': 'constexpr', 'R0_BLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=108, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, warp_size=32), 'constants': {}, 'configs': [{(0,): [['tt.divisibility', 16]], (1,): [['tt.divisibility', 16]], (4,): [['tt.divisibility', 16]]}]},
	        inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15', 'mutated_arg_names': ['in_out_ptr0'], 'optimize_mem': True, 'no_x_dim': False, 'num_load': 2, 'num_reduction': 2, 'backend_hash': '07F8534F6C2EC4E11ACA26C8C9D43565780E009807A566E459118140DB4BC4AB', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': True, 'is_fbcode': True}
	    )
	    @triton.jit
	    def triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15(in_out_ptr0, in_ptr0, ks0, xnumel, r0_numel, XBLOCK : tl.constexpr, R0_BLOCK : tl.constexpr):
	        r0_numel = 5120
	        rnumel = r0_numel
	        RBLOCK: tl.constexpr = R0_BLOCK
	        xoffset = tl.program_id(0) * XBLOCK
	        xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
	        xmask = xindex < xnumel
	        r0_base = tl.arange(0, R0_BLOCK)[None, :]
	        rbase = r0_base
	        x0 = xindex
	        _tmp2 = tl.full([XBLOCK, R0_BLOCK], float("-inf"), tl.float32)
	        _tmp4 = tl.full([XBLOCK, R0_BLOCK], float("inf"), tl.float32)
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_1 = r0_index
	            tmp0 = tl.load(in_ptr0 + (1280*((((r0_1 + 5120*x0) // 1280) % (1500*ks0))) + ((r0_1 % 1280))), r0_mask & xmask, eviction_policy='evict_last', other=0.0)
	            tmp1 = tl.broadcast_to(tmp0, [XBLOCK, R0_BLOCK])
	            tmp3 = triton_helpers.maximum(_tmp2, tmp1)
	            _tmp2 = tl.where(r0_mask & xmask, tmp3, _tmp2)
	            tmp5 = triton_helpers.minimum(_tmp4, tmp1)
	            _tmp4 = tl.where(r0_mask & xmask, tmp5, _tmp4)
	        tmp2 = triton_helpers.max2(_tmp2, 1)[:, None]
	        tmp4 = triton_helpers.min2(_tmp4, 1)[:, None]
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_1 = r0_index
	            tmp6 = tl.load(in_ptr0 + (1280*((((r0_1 + 5120*x0) // 1280) % (1500*ks0))) + ((r0_1 % 1280))), r0_mask & xmask, eviction_policy='evict_last', other=0.0)
	            tmp7 = 0.0
	            tmp8 = triton_helpers.maximum(tmp2, tmp7)
	            tmp9 = triton_helpers.minimum(tmp4, tmp7)
	            tmp10 = tmp8 - tmp9
	            tmp11 = 0.00392156862745098
	            tmp12 = tmp10 * tmp11
	            tmp13 = 1.1920928955078125e-07
	            tmp14 = triton_helpers.maximum(tmp12, tmp13)
	            tmp15 = tl.full([1, 1], 1, tl.int32)
	            tmp16 = (tmp15 / tmp14)
	            tmp17 = 1.0
	            tmp18 = tmp16 * tmp17
	            tmp19 = tmp6 * tmp18
	            tmp20 = libdevice.nearbyint(tmp19)
	            tmp21 = (tmp9 / tmp14)
	            tmp22 = libdevice.nearbyint(tmp21)
	            tmp23 = -128.0
	            tmp24 = tmp23 - tmp22
	            tmp25 = triton_helpers.maximum(tmp24, tmp23)
	            tmp26 = 127.0
	            tmp27 = triton_helpers.minimum(tmp25, tmp26)
	            tmp28 = tmp27.to(tl.int8)
	            tmp29 = tmp28.to(tl.float32)
	            tmp30 = tmp20 + tmp29
	            tmp31 = triton_helpers.maximum(tmp30, tmp23)
	            tmp32 = triton_helpers.minimum(tmp31, tmp26)
	            tmp33 = tmp32.to(tl.int8)
	            tmp34 = tmp33.to(tl.float32)
	            tmp35 = tmp34 - tmp29
	            tmp36 = tmp35 * tmp14
	            tl.store(in_out_ptr0 + (r0_1 + 5120*x0), tmp36, r0_mask & xmask)
	    ''', device_str='cuda')
	    */
	    uint32_t grid_0 = xnumel;
	    uint32_t grid_1 = 1;
	    uint32_t grid_2 = 1;
	    if (grid_0 == 0 || grid_1 == 0 || grid_2 == 0) return;
	    if (kernels_.triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15 == nullptr) {
	        kernels_.triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15 = loadKernel("/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/ciw56hes7a7b3bb5i6smv2wwwbq7knrhj2dz4lxaqg57bzvsp3tl.cubin", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15", 64, cubin_dir_); 
	    }
	    CUdeviceptr var_145 = reinterpret_cast<CUdeviceptr>(in_out_ptr0.data_ptr());
	    CUdeviceptr var_146 = reinterpret_cast<CUdeviceptr>(in_ptr0.data_ptr());
	    int64_t var_147 = ks0;
	    int32_t var_148 = xnumel;
	    int var_149 = r0_numel;
	    CUdeviceptr global_scratch_scratch_150 = 0;
	    void* kernel_args_[] = {&var_145, &var_146, &var_147, &var_148, &var_149, &global_scratch_scratch_150};
	    launchKernel(kernels_.triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15, grid_0, grid_1, grid_2, 16, 64, kernel_args_, stream_);
	}
	
	template <typename in_ptr0_type_, typename in_ptr1_type_, typename in_ptr2_type_, typename out_ptr0_type_, typename kernels_type_>
	static inline void call_triton_poi_fused__to_copy_mul_sub_view_16(
	    const in_ptr0_type_& in_ptr0,
	    const in_ptr1_type_& in_ptr1,
	    const in_ptr2_type_& in_ptr2,
	    const out_ptr0_type_& out_ptr0,
	    int64_t xnumel,
	    int32_t device_idx_,
	    cudaStream_t stream_,
	    kernels_type_& kernels_,
	    const std::optional<std::string>& cubin_dir_ = std::nullopt
	){
	    /*
	    async_compile.triton('triton_poi_fused__to_copy_mul_sub_view_16', '''
	    import triton
	    import triton.language as tl
	
	    from torch._inductor.runtime import triton_helpers, triton_heuristics
	    from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
	    from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, DeviceProperties
	    triton_helpers.set_driver_to_gpu()
	
	    @triton_heuristics.pointwise(
	        size_hints={'x': 16777216}, 
	        filename=__file__,
	        triton_meta={'signature': {'in_ptr0': '*i8', 'in_ptr1': '*i8', 'in_ptr2': '*fp32', 'out_ptr0': '*fp32', 'xnumel': 'i32', 'XBLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=108, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, warp_size=32), 'constants': {}, 'configs': [{(0,): [['tt.divisibility', 16]], (1,): [['tt.divisibility', 16]], (2,): [['tt.divisibility', 16]], (3,): [['tt.divisibility', 16]], (4,): [['tt.divisibility', 16]]}]},
	        inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__to_copy_mul_sub_view_16', 'mutated_arg_names': [], 'optimize_mem': True, 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': '07F8534F6C2EC4E11ACA26C8C9D43565780E009807A566E459118140DB4BC4AB', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': True, 'is_fbcode': True},
	        min_elem_per_thread=0
	    )
	    @triton.jit
	    def triton_poi_fused__to_copy_mul_sub_view_16(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
	        xnumel = 15728640
	        xoffset = tl.program_id(0) * XBLOCK
	        xindex = xoffset + tl.arange(0, XBLOCK)[:]
	        xmask = tl.full([XBLOCK], True, tl.int1)
	        x2 = xindex
	        x1 = xindex // 32
	        tmp0 = tl.load(in_ptr0 + (x2), None)
	        tmp2 = tl.load(in_ptr1 + (x1), None, eviction_policy='evict_last')
	        tmp5 = tl.load(in_ptr2 + (x1), None, eviction_policy='evict_last')
	        tmp1 = tmp0.to(tl.float32)
	        tmp3 = tmp2.to(tl.float32)
	        tmp4 = tmp1 - tmp3
	        tmp6 = tmp4 * tmp5
	        tl.store(out_ptr0 + (x2), tmp6, None)
	    ''', device_str='cuda')
	    */
	    uint32_t grid_0 = ((xnumel + (512 - 1)) / (512));
	    uint32_t grid_1 = 1;
	    uint32_t grid_2 = 1;
	    if (grid_0 == 0 || grid_1 == 0 || grid_2 == 0) return;
	    if (kernels_.triton_poi_fused__to_copy_mul_sub_view_16 == nullptr) {
	        kernels_.triton_poi_fused__to_copy_mul_sub_view_16 = loadKernel("/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/cguijlrbjoft62e5ek56tidqfnayhhkbumqe2ik5pq2sli6mrf4k.cubin", "triton_poi_fused__to_copy_mul_sub_view_16", 0, cubin_dir_); 
	    }
	    CUdeviceptr var_151 = reinterpret_cast<CUdeviceptr>(in_ptr0.data_ptr());
	    CUdeviceptr var_152 = reinterpret_cast<CUdeviceptr>(in_ptr1.data_ptr());
	    CUdeviceptr var_153 = reinterpret_cast<CUdeviceptr>(in_ptr2.data_ptr());
	    CUdeviceptr var_154 = reinterpret_cast<CUdeviceptr>(out_ptr0.data_ptr());
	    int var_155 = xnumel;
	    CUdeviceptr global_scratch_scratch_156 = 0;
	    void* kernel_args_[] = {&var_151, &var_152, &var_153, &var_154, &var_155, &global_scratch_scratch_156};
	    launchKernel(kernels_.triton_poi_fused__to_copy_mul_sub_view_16, grid_0, grid_1, grid_2, 8, 0, kernel_args_, stream_);
	}
	
	template <typename in_out_ptr0_type_, typename kernels_type_>
	static inline void call_triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17(
	    const in_out_ptr0_type_& in_out_ptr0,
	    int64_t xnumel,
	    int64_t r0_numel,
	    int32_t device_idx_,
	    cudaStream_t stream_,
	    kernels_type_& kernels_,
	    const std::optional<std::string>& cubin_dir_ = std::nullopt
	){
	    /*
	    async_compile.triton('triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17', '''
	    import triton
	    import triton.language as tl
	
	    from torch._inductor.runtime import triton_helpers, triton_heuristics
	    from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
	    from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, DeviceProperties
	    triton_helpers.set_driver_to_gpu()
	
	    @triton_heuristics.reduction(
	        size_hints={'x': 2048, 'r0_': 4096},
	        reduction_hint=ReductionHint.INNER,
	        filename=__file__,
	        triton_meta={'signature': {'in_out_ptr0': '*fp32', 'xnumel': 'i32', 'r0_numel': 'i32', 'XBLOCK': 'constexpr', 'R0_BLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=108, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, warp_size=32), 'constants': {}, 'configs': [{(0,): [['tt.divisibility', 16]], (2,): [['tt.divisibility', 16]]}]},
	        inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17', 'mutated_arg_names': ['in_out_ptr0'], 'optimize_mem': True, 'no_x_dim': False, 'num_load': 2, 'num_reduction': 2, 'backend_hash': '07F8534F6C2EC4E11ACA26C8C9D43565780E009807A566E459118140DB4BC4AB', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': True, 'is_fbcode': True}
	    )
	    @triton.jit
	    def triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17(in_out_ptr0, xnumel, r0_numel, XBLOCK : tl.constexpr, R0_BLOCK : tl.constexpr):
	        r0_numel = 3072
	        rnumel = r0_numel
	        RBLOCK: tl.constexpr = R0_BLOCK
	        xoffset = tl.program_id(0) * XBLOCK
	        xindex = xoffset + tl.arange(0, XBLOCK)[:, None]
	        xmask = xindex < xnumel
	        r0_base = tl.arange(0, R0_BLOCK)[None, :]
	        rbase = r0_base
	        x0 = xindex
	        _tmp10 = tl.full([XBLOCK, R0_BLOCK], float("-inf"), tl.float32)
	        _tmp12 = tl.full([XBLOCK, R0_BLOCK], float("inf"), tl.float32)
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_1 = r0_index
	            tmp0 = tl.load(in_out_ptr0 + (r0_1 + 3072*x0), r0_mask & xmask, eviction_policy='evict_last', other=0.0)
	            tmp1 = 0.5
	            tmp2 = tmp0 * tmp1
	            tmp3 = 0.7071067811865476
	            tmp4 = tmp0 * tmp3
	            tmp5 = libdevice.erf(tmp4)
	            tmp6 = 1.0
	            tmp7 = tmp5 + tmp6
	            tmp8 = tmp2 * tmp7
	            tmp9 = tl.broadcast_to(tmp8, [XBLOCK, R0_BLOCK])
	            tmp11 = triton_helpers.maximum(_tmp10, tmp9)
	            _tmp10 = tl.where(r0_mask & xmask, tmp11, _tmp10)
	            tmp13 = triton_helpers.minimum(_tmp12, tmp9)
	            _tmp12 = tl.where(r0_mask & xmask, tmp13, _tmp12)
	        tmp10 = triton_helpers.max2(_tmp10, 1)[:, None]
	        tmp12 = triton_helpers.min2(_tmp12, 1)[:, None]
	        for r0_offset in range(0, r0_numel, R0_BLOCK):
	            r0_index = r0_offset + r0_base
	            r0_mask = r0_index < r0_numel
	            roffset = r0_offset
	            rindex = r0_index
	            r0_1 = r0_index
	            tmp14 = tl.load(in_out_ptr0 + (r0_1 + 3072*x0), r0_mask & xmask, eviction_policy='evict_first', other=0.0)
	            tmp15 = 0.5
	            tmp16 = tmp14 * tmp15
	            tmp17 = 0.7071067811865476
	            tmp18 = tmp14 * tmp17
	            tmp19 = libdevice.erf(tmp18)
	            tmp20 = 1.0
	            tmp21 = tmp19 + tmp20
	            tmp22 = tmp16 * tmp21
	            tmp23 = 0.0
	            tmp24 = triton_helpers.maximum(tmp10, tmp23)
	            tmp25 = triton_helpers.minimum(tmp12, tmp23)
	            tmp26 = tmp24 - tmp25
	            tmp27 = 0.00392156862745098
	            tmp28 = tmp26 * tmp27
	            tmp29 = 1.1920928955078125e-07
	            tmp30 = triton_helpers.maximum(tmp28, tmp29)
	            tmp31 = tl.full([1, 1], 1, tl.int32)
	            tmp32 = (tmp31 / tmp30)
	            tmp33 = tmp32 * tmp20
	            tmp34 = tmp22 * tmp33
	            tmp35 = libdevice.nearbyint(tmp34)
	            tmp36 = (tmp25 / tmp30)
	            tmp37 = libdevice.nearbyint(tmp36)
	            tmp38 = -128.0
	            tmp39 = tmp38 - tmp37
	            tmp40 = triton_helpers.maximum(tmp39, tmp38)
	            tmp41 = 127.0
	            tmp42 = triton_helpers.minimum(tmp40, tmp41)
	            tmp43 = tmp42.to(tl.int8)
	            tmp44 = tmp43.to(tl.float32)
	            tmp45 = tmp35 + tmp44
	            tmp46 = triton_helpers.maximum(tmp45, tmp38)
	            tmp47 = triton_helpers.minimum(tmp46, tmp41)
	            tmp48 = tmp47.to(tl.int8)
	            tmp49 = tmp48.to(tl.float32)
	            tmp50 = tmp49 - tmp44
	            tmp51 = tmp50 * tmp30
	            tl.store(in_out_ptr0 + (r0_1 + 3072*x0), tmp51, r0_mask & xmask)
	    ''', device_str='cuda')
	    */
	    uint32_t grid_0 = xnumel;
	    uint32_t grid_1 = 1;
	    uint32_t grid_2 = 1;
	    if (grid_0 == 0 || grid_1 == 0 || grid_2 == 0) return;
	    if (kernels_.triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17 == nullptr) {
	        kernels_.triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17 = loadKernel("/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/c6vgxoezwdhozlv6bvoixfo6ngv75iwimwrz65jmzb5se3xtjzyt.cubin", "triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17", 64, cubin_dir_); 
	    }
	    CUdeviceptr var_157 = reinterpret_cast<CUdeviceptr>(in_out_ptr0.data_ptr());
	    int32_t var_158 = xnumel;
	    int var_159 = r0_numel;
	    CUdeviceptr global_scratch_scratch_160 = 0;
	    void* kernel_args_[] = {&var_157, &var_158, &var_159, &global_scratch_scratch_160};
	    launchKernel(kernels_.triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17, grid_0, grid_1, grid_2, 16, 64, kernel_args_, stream_);
	}
	
	template <typename in_ptr0_type_, typename in_ptr1_type_, typename in_ptr2_type_, typename out_ptr0_type_, typename kernels_type_>
	static inline void call_triton_poi_fused__to_copy_mul_sub_view_18(
	    const in_ptr0_type_& in_ptr0,
	    const in_ptr1_type_& in_ptr1,
	    const in_ptr2_type_& in_ptr2,
	    const out_ptr0_type_& out_ptr0,
	    int64_t xnumel,
	    int32_t device_idx_,
	    cudaStream_t stream_,
	    kernels_type_& kernels_,
	    const std::optional<std::string>& cubin_dir_ = std::nullopt
	){
	    /*
	    async_compile.triton('triton_poi_fused__to_copy_mul_sub_view_18', '''
	    import triton
	    import triton.language as tl
	
	    from torch._inductor.runtime import triton_helpers, triton_heuristics
	    from torch._inductor.runtime.triton_helpers import libdevice, math as tl_math
	    from torch._inductor.runtime.hints import AutotuneHint, ReductionHint, TileHint, DeviceProperties
	    triton_helpers.set_driver_to_gpu()
	
	    @triton_heuristics.pointwise(
	        size_hints={'x': 16777216}, 
	        filename=__file__,
	        triton_meta={'signature': {'in_ptr0': '*i8', 'in_ptr1': '*i8', 'in_ptr2': '*fp32', 'out_ptr0': '*fp32', 'xnumel': 'i32', 'XBLOCK': 'constexpr'}, 'device': DeviceProperties(type='cuda', index=0, multi_processor_count=108, cc=80, major=8, regs_per_multiprocessor=65536, max_threads_per_multi_processor=2048, warp_size=32), 'constants': {}, 'configs': [{(0,): [['tt.divisibility', 16]], (1,): [['tt.divisibility', 16]], (2,): [['tt.divisibility', 16]], (3,): [['tt.divisibility', 16]], (4,): [['tt.divisibility', 16]]}]},
	        inductor_meta={'grid_type': 'Grid1D', 'autotune_hints': set(), 'kernel_name': 'triton_poi_fused__to_copy_mul_sub_view_18', 'mutated_arg_names': [], 'optimize_mem': True, 'no_x_dim': False, 'num_load': 3, 'num_reduction': 0, 'backend_hash': '07F8534F6C2EC4E11ACA26C8C9D43565780E009807A566E459118140DB4BC4AB', 'are_deterministic_algorithms_enabled': False, 'assert_indirect_indexing': True, 'autotune_local_cache': True, 'autotune_pointwise': True, 'autotune_remote_cache': None, 'force_disable_caches': False, 'dynamic_scale_rblock': True, 'max_autotune': False, 'max_autotune_pointwise': False, 'min_split_scan_rblock': 256, 'spill_threshold': 16, 'store_cubin': True, 'is_fbcode': True},
	        min_elem_per_thread=0
	    )
	    @triton.jit
	    def triton_poi_fused__to_copy_mul_sub_view_18(in_ptr0, in_ptr1, in_ptr2, out_ptr0, xnumel, XBLOCK : tl.constexpr):
	        xnumel = 9437184
	        xoffset = tl.program_id(0) * XBLOCK
	        xindex = xoffset + tl.arange(0, XBLOCK)[:]
	        xmask = tl.full([XBLOCK], True, tl.int1)
	        x2 = xindex
	        x1 = xindex // 32
	        tmp0 = tl.load(in_ptr0 + (x2), None)
	        tmp2 = tl.load(in_ptr1 + (x1), None, eviction_policy='evict_last')
	        tmp5 = tl.load(in_ptr2 + (x1), None, eviction_policy='evict_last')
	        tmp1 = tmp0.to(tl.float32)
	        tmp3 = tmp2.to(tl.float32)
	        tmp4 = tmp1 - tmp3
	        tmp6 = tmp4 * tmp5
	        tl.store(out_ptr0 + (x2), tmp6, None)
	    ''', device_str='cuda')
	    */
	    uint32_t grid_0 = ((xnumel + (512 - 1)) / (512));
	    uint32_t grid_1 = 1;
	    uint32_t grid_2 = 1;
	    if (grid_0 == 0 || grid_1 == 0 || grid_2 == 0) return;
	    if (kernels_.triton_poi_fused__to_copy_mul_sub_view_18 == nullptr) {
	        kernels_.triton_poi_fused__to_copy_mul_sub_view_18 = loadKernel("/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/c7ckgvbpng7kv6gh4cbbwg4i27lw2xucdlsyjtyuevoqshliflc3.cubin", "triton_poi_fused__to_copy_mul_sub_view_18", 0, cubin_dir_); 
	    }
	    CUdeviceptr var_161 = reinterpret_cast<CUdeviceptr>(in_ptr0.data_ptr());
	    CUdeviceptr var_162 = reinterpret_cast<CUdeviceptr>(in_ptr1.data_ptr());
	    CUdeviceptr var_163 = reinterpret_cast<CUdeviceptr>(in_ptr2.data_ptr());
	    CUdeviceptr var_164 = reinterpret_cast<CUdeviceptr>(out_ptr0.data_ptr());
	    int var_165 = xnumel;
	    CUdeviceptr global_scratch_scratch_166 = 0;
	    void* kernel_args_[] = {&var_161, &var_162, &var_163, &var_164, &var_165, &global_scratch_scratch_166};
	    launchKernel(kernels_.triton_poi_fused__to_copy_mul_sub_view_18, grid_0, grid_1, grid_2, 8, 0, kernel_args_, stream_);
	}
	
	namespace torch::aot_inductor {
	
	void AOTInductorModel::_const_run_impl(
	    std::vector<AtenTensorHandle>& output_handles,
	    DeviceStreamType stream,
	    AOTIProxyExecutorHandle proxy_executor
	) {}
	
	AOTI_NOINLINE static void check_input_0(
	    AtenTensorHandle* input_handles
	) {
	    ConstantHandle arg877_1 = ConstantHandle(input_handles[0]);
	    int32_t arg877_1_dtype;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_get_dtype(arg877_1, &arg877_1_dtype));
	
	    int32_t arg877_1_expected_dtype = aoti_torch_dtype_float32();
	    if (arg877_1_expected_dtype != arg877_1_dtype) {
	        std::stringstream ss;
	        ss << "input_handles[0]: unmatched dtype, "
	           << "expected: " << arg877_1_expected_dtype << "(at::kFloat), "
	           << "but got: " << arg877_1_dtype << "\n";
	        throw std::runtime_error(ss.str());
	    }
	    auto arg877_1_size = arg877_1.sizes();
	
	    if (arg877_1_size[0] < 1) {
	        std::stringstream ss;
	        ss << "input_handles[0]: dim value is too small at 0, "
	           << "expected it to be >= 1, " << "but got: "
	           << arg877_1_size[0] << "\n";
	        throw std::runtime_error(ss.str());
	    }
	
	    if (arg877_1_size[0] > 10) {
	        std::stringstream ss;
	        ss << "input_handles[0]: dim value is too large at 0, "
	           << "expected to be <= 10, " << "but got: "
	           << arg877_1_size[0] << "\n";
	        throw std::runtime_error(ss.str());
	    }
	
	    if (128 != arg877_1_size[1]) {
	        std::stringstream ss;
	        ss << "input_handles[0]: unmatched dim value at 1, "
	           << "expected: 128, " << "but got: " << arg877_1_size[1]
	           << "\n";
	        throw std::runtime_error(ss.str());
	    }
	
	    if (3000 != arg877_1_size[2]) {
	        std::stringstream ss;
	        ss << "input_handles[0]: unmatched dim value at 2, "
	           << "expected: 3000, " << "but got: " << arg877_1_size[2]
	           << "\n";
	        throw std::runtime_error(ss.str());
	    }
	    auto arg877_1_stride = arg877_1.strides();
	
	    if (384000 != arg877_1_stride[0]) {
	        std::stringstream ss;
	        ss << "input_handles[0]: unmatched stride value at 0, "
	           << "expected: 384000, " << "but got: " << arg877_1_stride[0]
	           << "\n";
	        throw std::runtime_error(ss.str());
	    }
	
	    if (3000 != arg877_1_stride[1]) {
	        std::stringstream ss;
	        ss << "input_handles[0]: unmatched stride value at 1, "
	           << "expected: 3000, " << "but got: " << arg877_1_stride[1]
	           << "\n";
	        throw std::runtime_error(ss.str());
	    }
	
	    if (1 != arg877_1_stride[2]) {
	        std::stringstream ss;
	        ss << "input_handles[0]: unmatched stride value at 2, "
	           << "expected: 1, " << "but got: " << arg877_1_stride[2]
	           << "\n";
	        throw std::runtime_error(ss.str());
	    }
	    int32_t arg877_1_device_type;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_get_device_type(arg877_1, &arg877_1_device_type));
	
	    int32_t arg877_1_expected_device_type = 1;
	    if (arg877_1_expected_device_type != arg877_1_device_type) {
	        std::stringstream ss;
	        ss << "input_handles[0]: unmatched device type, "
	        << "expected: " << arg877_1_expected_device_type << "1(cuda), "
	        << "but got: " << arg877_1_device_type << "\n";
	        throw std::runtime_error(ss.str());
	    }
	}
	
	static bool _check_aoti_runtime_check_inputs_env() {
	    const static char* env_var_value = getenv("AOTI_RUNTIME_CHECK_INPUTS");
	    const static bool result = env_var_value != nullptr && env_var_value[0] != '0';
	    return result;
	}
	
	AOTI_NOINLINE static void __check_inputs_outputs(
	    AtenTensorHandle* input_handles,
	    AtenTensorHandle* output_handles) {
	    if (!_check_aoti_runtime_check_inputs_env()){
	        return;
	    }
	    check_input_0(input_handles);
	}
	
	void AOTInductorModel::run_impl(
	    AtenTensorHandle*
	        input_handles, // array of input AtenTensorHandle; handles
	                        // are stolen; the array itself is borrowed
	    AtenTensorHandle*
	        output_handles, // array for writing output AtenTensorHandle; handles
	                        // will be stolen by the caller; the array itself is
	                        // borrowed
	    DeviceStreamType stream,
	    AOTIProxyExecutorHandle proxy_executor
	) {
	    __check_inputs_outputs(input_handles, output_handles);
	
	    auto inputs = steal_from_raw_handles_to_raii_handles(input_handles, 1);
	    auto arg877_1 = std::move(inputs[0]);
	    [[maybe_unused]] auto& model_audio_tower_embed_positions_weight = constants_->at(0);
	    [[maybe_unused]] auto& model_audio_tower_conv1_weight = constants_->at(1);
	    [[maybe_unused]] auto& model_audio_tower_conv1_bias = constants_->at(2);
	    [[maybe_unused]] auto& model_audio_tower_conv2_weight = constants_->at(3);
	    [[maybe_unused]] auto& model_audio_tower_conv2_bias = constants_->at(4);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_self_attn_layer_norm_weight = constants_->at(5);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_self_attn_layer_norm_bias = constants_->at(6);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_self_attn_q_proj_bias = constants_->at(7);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(8);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(9);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(10);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(11);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(12);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(13);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_self_attn_v_proj_bias = constants_->at(14);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(15);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(16);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(17);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_self_attn_out_proj_bias = constants_->at(18);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(19);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(20);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(21);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_final_layer_norm_weight = constants_->at(22);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_final_layer_norm_bias = constants_->at(23);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_fc1_bias = constants_->at(24);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_fc1_parametrizations_weight_original0 = constants_->at(25);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_fc1_parametrizations_weight_original1 = constants_->at(26);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_fc1_parametrizations_weight_original2 = constants_->at(27);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_fc2_bias = constants_->at(28);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_fc2_parametrizations_weight_original0 = constants_->at(29);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_fc2_parametrizations_weight_original1 = constants_->at(30);
	    [[maybe_unused]] auto& model_audio_tower_layers_0_fc2_parametrizations_weight_original2 = constants_->at(31);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_self_attn_layer_norm_weight = constants_->at(32);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_self_attn_layer_norm_bias = constants_->at(33);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_self_attn_q_proj_bias = constants_->at(34);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(35);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(36);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(37);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(38);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(39);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(40);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_self_attn_v_proj_bias = constants_->at(41);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(42);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(43);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(44);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_self_attn_out_proj_bias = constants_->at(45);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(46);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(47);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(48);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_final_layer_norm_weight = constants_->at(49);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_final_layer_norm_bias = constants_->at(50);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_fc1_bias = constants_->at(51);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_fc1_parametrizations_weight_original0 = constants_->at(52);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_fc1_parametrizations_weight_original1 = constants_->at(53);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_fc1_parametrizations_weight_original2 = constants_->at(54);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_fc2_bias = constants_->at(55);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_fc2_parametrizations_weight_original0 = constants_->at(56);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_fc2_parametrizations_weight_original1 = constants_->at(57);
	    [[maybe_unused]] auto& model_audio_tower_layers_1_fc2_parametrizations_weight_original2 = constants_->at(58);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_self_attn_layer_norm_weight = constants_->at(59);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_self_attn_layer_norm_bias = constants_->at(60);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_self_attn_q_proj_bias = constants_->at(61);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(62);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(63);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(64);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(65);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(66);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(67);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_self_attn_v_proj_bias = constants_->at(68);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(69);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(70);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(71);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_self_attn_out_proj_bias = constants_->at(72);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(73);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(74);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(75);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_final_layer_norm_weight = constants_->at(76);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_final_layer_norm_bias = constants_->at(77);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_fc1_bias = constants_->at(78);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_fc1_parametrizations_weight_original0 = constants_->at(79);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_fc1_parametrizations_weight_original1 = constants_->at(80);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_fc1_parametrizations_weight_original2 = constants_->at(81);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_fc2_bias = constants_->at(82);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_fc2_parametrizations_weight_original0 = constants_->at(83);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_fc2_parametrizations_weight_original1 = constants_->at(84);
	    [[maybe_unused]] auto& model_audio_tower_layers_2_fc2_parametrizations_weight_original2 = constants_->at(85);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_self_attn_layer_norm_weight = constants_->at(86);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_self_attn_layer_norm_bias = constants_->at(87);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_self_attn_q_proj_bias = constants_->at(88);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(89);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(90);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(91);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(92);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(93);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(94);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_self_attn_v_proj_bias = constants_->at(95);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(96);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(97);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(98);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_self_attn_out_proj_bias = constants_->at(99);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(100);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(101);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(102);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_final_layer_norm_weight = constants_->at(103);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_final_layer_norm_bias = constants_->at(104);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_fc1_bias = constants_->at(105);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_fc1_parametrizations_weight_original0 = constants_->at(106);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_fc1_parametrizations_weight_original1 = constants_->at(107);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_fc1_parametrizations_weight_original2 = constants_->at(108);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_fc2_bias = constants_->at(109);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_fc2_parametrizations_weight_original0 = constants_->at(110);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_fc2_parametrizations_weight_original1 = constants_->at(111);
	    [[maybe_unused]] auto& model_audio_tower_layers_3_fc2_parametrizations_weight_original2 = constants_->at(112);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_self_attn_layer_norm_weight = constants_->at(113);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_self_attn_layer_norm_bias = constants_->at(114);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_self_attn_q_proj_bias = constants_->at(115);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(116);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(117);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(118);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(119);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(120);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(121);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_self_attn_v_proj_bias = constants_->at(122);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(123);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(124);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(125);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_self_attn_out_proj_bias = constants_->at(126);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(127);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(128);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(129);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_final_layer_norm_weight = constants_->at(130);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_final_layer_norm_bias = constants_->at(131);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_fc1_bias = constants_->at(132);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_fc1_parametrizations_weight_original0 = constants_->at(133);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_fc1_parametrizations_weight_original1 = constants_->at(134);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_fc1_parametrizations_weight_original2 = constants_->at(135);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_fc2_bias = constants_->at(136);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_fc2_parametrizations_weight_original0 = constants_->at(137);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_fc2_parametrizations_weight_original1 = constants_->at(138);
	    [[maybe_unused]] auto& model_audio_tower_layers_4_fc2_parametrizations_weight_original2 = constants_->at(139);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_self_attn_layer_norm_weight = constants_->at(140);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_self_attn_layer_norm_bias = constants_->at(141);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_self_attn_q_proj_bias = constants_->at(142);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(143);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(144);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(145);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(146);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(147);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(148);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_self_attn_v_proj_bias = constants_->at(149);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(150);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(151);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(152);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_self_attn_out_proj_bias = constants_->at(153);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(154);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(155);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(156);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_final_layer_norm_weight = constants_->at(157);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_final_layer_norm_bias = constants_->at(158);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_fc1_bias = constants_->at(159);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_fc1_parametrizations_weight_original0 = constants_->at(160);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_fc1_parametrizations_weight_original1 = constants_->at(161);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_fc1_parametrizations_weight_original2 = constants_->at(162);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_fc2_bias = constants_->at(163);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_fc2_parametrizations_weight_original0 = constants_->at(164);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_fc2_parametrizations_weight_original1 = constants_->at(165);
	    [[maybe_unused]] auto& model_audio_tower_layers_5_fc2_parametrizations_weight_original2 = constants_->at(166);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_self_attn_layer_norm_weight = constants_->at(167);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_self_attn_layer_norm_bias = constants_->at(168);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_self_attn_q_proj_bias = constants_->at(169);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(170);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(171);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(172);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(173);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(174);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(175);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_self_attn_v_proj_bias = constants_->at(176);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(177);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(178);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(179);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_self_attn_out_proj_bias = constants_->at(180);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(181);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(182);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(183);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_final_layer_norm_weight = constants_->at(184);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_final_layer_norm_bias = constants_->at(185);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_fc1_bias = constants_->at(186);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_fc1_parametrizations_weight_original0 = constants_->at(187);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_fc1_parametrizations_weight_original1 = constants_->at(188);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_fc1_parametrizations_weight_original2 = constants_->at(189);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_fc2_bias = constants_->at(190);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_fc2_parametrizations_weight_original0 = constants_->at(191);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_fc2_parametrizations_weight_original1 = constants_->at(192);
	    [[maybe_unused]] auto& model_audio_tower_layers_6_fc2_parametrizations_weight_original2 = constants_->at(193);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_self_attn_layer_norm_weight = constants_->at(194);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_self_attn_layer_norm_bias = constants_->at(195);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_self_attn_q_proj_bias = constants_->at(196);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(197);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(198);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(199);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(200);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(201);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(202);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_self_attn_v_proj_bias = constants_->at(203);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(204);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(205);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(206);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_self_attn_out_proj_bias = constants_->at(207);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(208);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(209);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(210);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_final_layer_norm_weight = constants_->at(211);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_final_layer_norm_bias = constants_->at(212);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_fc1_bias = constants_->at(213);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_fc1_parametrizations_weight_original0 = constants_->at(214);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_fc1_parametrizations_weight_original1 = constants_->at(215);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_fc1_parametrizations_weight_original2 = constants_->at(216);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_fc2_bias = constants_->at(217);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_fc2_parametrizations_weight_original0 = constants_->at(218);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_fc2_parametrizations_weight_original1 = constants_->at(219);
	    [[maybe_unused]] auto& model_audio_tower_layers_7_fc2_parametrizations_weight_original2 = constants_->at(220);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_self_attn_layer_norm_weight = constants_->at(221);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_self_attn_layer_norm_bias = constants_->at(222);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_self_attn_q_proj_bias = constants_->at(223);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(224);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(225);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(226);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(227);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(228);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(229);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_self_attn_v_proj_bias = constants_->at(230);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(231);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(232);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(233);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_self_attn_out_proj_bias = constants_->at(234);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(235);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(236);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(237);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_final_layer_norm_weight = constants_->at(238);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_final_layer_norm_bias = constants_->at(239);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_fc1_bias = constants_->at(240);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_fc1_parametrizations_weight_original0 = constants_->at(241);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_fc1_parametrizations_weight_original1 = constants_->at(242);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_fc1_parametrizations_weight_original2 = constants_->at(243);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_fc2_bias = constants_->at(244);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_fc2_parametrizations_weight_original0 = constants_->at(245);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_fc2_parametrizations_weight_original1 = constants_->at(246);
	    [[maybe_unused]] auto& model_audio_tower_layers_8_fc2_parametrizations_weight_original2 = constants_->at(247);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_self_attn_layer_norm_weight = constants_->at(248);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_self_attn_layer_norm_bias = constants_->at(249);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_self_attn_q_proj_bias = constants_->at(250);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(251);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(252);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(253);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(254);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(255);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(256);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_self_attn_v_proj_bias = constants_->at(257);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(258);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(259);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(260);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_self_attn_out_proj_bias = constants_->at(261);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(262);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(263);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(264);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_final_layer_norm_weight = constants_->at(265);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_final_layer_norm_bias = constants_->at(266);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_fc1_bias = constants_->at(267);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_fc1_parametrizations_weight_original0 = constants_->at(268);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_fc1_parametrizations_weight_original1 = constants_->at(269);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_fc1_parametrizations_weight_original2 = constants_->at(270);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_fc2_bias = constants_->at(271);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_fc2_parametrizations_weight_original0 = constants_->at(272);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_fc2_parametrizations_weight_original1 = constants_->at(273);
	    [[maybe_unused]] auto& model_audio_tower_layers_9_fc2_parametrizations_weight_original2 = constants_->at(274);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_self_attn_layer_norm_weight = constants_->at(275);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_self_attn_layer_norm_bias = constants_->at(276);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_self_attn_q_proj_bias = constants_->at(277);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(278);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(279);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(280);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(281);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(282);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(283);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_self_attn_v_proj_bias = constants_->at(284);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(285);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(286);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(287);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_self_attn_out_proj_bias = constants_->at(288);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(289);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(290);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(291);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_final_layer_norm_weight = constants_->at(292);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_final_layer_norm_bias = constants_->at(293);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_fc1_bias = constants_->at(294);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_fc1_parametrizations_weight_original0 = constants_->at(295);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_fc1_parametrizations_weight_original1 = constants_->at(296);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_fc1_parametrizations_weight_original2 = constants_->at(297);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_fc2_bias = constants_->at(298);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_fc2_parametrizations_weight_original0 = constants_->at(299);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_fc2_parametrizations_weight_original1 = constants_->at(300);
	    [[maybe_unused]] auto& model_audio_tower_layers_10_fc2_parametrizations_weight_original2 = constants_->at(301);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_self_attn_layer_norm_weight = constants_->at(302);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_self_attn_layer_norm_bias = constants_->at(303);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_self_attn_q_proj_bias = constants_->at(304);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(305);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(306);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(307);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(308);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(309);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(310);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_self_attn_v_proj_bias = constants_->at(311);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(312);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(313);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(314);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_self_attn_out_proj_bias = constants_->at(315);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(316);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(317);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(318);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_final_layer_norm_weight = constants_->at(319);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_final_layer_norm_bias = constants_->at(320);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_fc1_bias = constants_->at(321);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_fc1_parametrizations_weight_original0 = constants_->at(322);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_fc1_parametrizations_weight_original1 = constants_->at(323);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_fc1_parametrizations_weight_original2 = constants_->at(324);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_fc2_bias = constants_->at(325);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_fc2_parametrizations_weight_original0 = constants_->at(326);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_fc2_parametrizations_weight_original1 = constants_->at(327);
	    [[maybe_unused]] auto& model_audio_tower_layers_11_fc2_parametrizations_weight_original2 = constants_->at(328);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_self_attn_layer_norm_weight = constants_->at(329);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_self_attn_layer_norm_bias = constants_->at(330);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_self_attn_q_proj_bias = constants_->at(331);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(332);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(333);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(334);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(335);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(336);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(337);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_self_attn_v_proj_bias = constants_->at(338);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(339);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(340);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(341);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_self_attn_out_proj_bias = constants_->at(342);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(343);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(344);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(345);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_final_layer_norm_weight = constants_->at(346);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_final_layer_norm_bias = constants_->at(347);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_fc1_bias = constants_->at(348);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_fc1_parametrizations_weight_original0 = constants_->at(349);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_fc1_parametrizations_weight_original1 = constants_->at(350);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_fc1_parametrizations_weight_original2 = constants_->at(351);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_fc2_bias = constants_->at(352);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_fc2_parametrizations_weight_original0 = constants_->at(353);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_fc2_parametrizations_weight_original1 = constants_->at(354);
	    [[maybe_unused]] auto& model_audio_tower_layers_12_fc2_parametrizations_weight_original2 = constants_->at(355);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_self_attn_layer_norm_weight = constants_->at(356);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_self_attn_layer_norm_bias = constants_->at(357);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_self_attn_q_proj_bias = constants_->at(358);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(359);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(360);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(361);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(362);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(363);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(364);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_self_attn_v_proj_bias = constants_->at(365);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(366);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(367);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(368);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_self_attn_out_proj_bias = constants_->at(369);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(370);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(371);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(372);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_final_layer_norm_weight = constants_->at(373);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_final_layer_norm_bias = constants_->at(374);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_fc1_bias = constants_->at(375);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_fc1_parametrizations_weight_original0 = constants_->at(376);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_fc1_parametrizations_weight_original1 = constants_->at(377);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_fc1_parametrizations_weight_original2 = constants_->at(378);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_fc2_bias = constants_->at(379);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_fc2_parametrizations_weight_original0 = constants_->at(380);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_fc2_parametrizations_weight_original1 = constants_->at(381);
	    [[maybe_unused]] auto& model_audio_tower_layers_13_fc2_parametrizations_weight_original2 = constants_->at(382);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_self_attn_layer_norm_weight = constants_->at(383);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_self_attn_layer_norm_bias = constants_->at(384);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_self_attn_q_proj_bias = constants_->at(385);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(386);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(387);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(388);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(389);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(390);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(391);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_self_attn_v_proj_bias = constants_->at(392);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(393);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(394);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(395);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_self_attn_out_proj_bias = constants_->at(396);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(397);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(398);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(399);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_final_layer_norm_weight = constants_->at(400);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_final_layer_norm_bias = constants_->at(401);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_fc1_bias = constants_->at(402);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_fc1_parametrizations_weight_original0 = constants_->at(403);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_fc1_parametrizations_weight_original1 = constants_->at(404);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_fc1_parametrizations_weight_original2 = constants_->at(405);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_fc2_bias = constants_->at(406);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_fc2_parametrizations_weight_original0 = constants_->at(407);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_fc2_parametrizations_weight_original1 = constants_->at(408);
	    [[maybe_unused]] auto& model_audio_tower_layers_14_fc2_parametrizations_weight_original2 = constants_->at(409);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_self_attn_layer_norm_weight = constants_->at(410);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_self_attn_layer_norm_bias = constants_->at(411);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_self_attn_q_proj_bias = constants_->at(412);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(413);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(414);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(415);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(416);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(417);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(418);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_self_attn_v_proj_bias = constants_->at(419);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(420);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(421);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(422);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_self_attn_out_proj_bias = constants_->at(423);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(424);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(425);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(426);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_final_layer_norm_weight = constants_->at(427);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_final_layer_norm_bias = constants_->at(428);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_fc1_bias = constants_->at(429);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_fc1_parametrizations_weight_original0 = constants_->at(430);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_fc1_parametrizations_weight_original1 = constants_->at(431);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_fc1_parametrizations_weight_original2 = constants_->at(432);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_fc2_bias = constants_->at(433);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_fc2_parametrizations_weight_original0 = constants_->at(434);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_fc2_parametrizations_weight_original1 = constants_->at(435);
	    [[maybe_unused]] auto& model_audio_tower_layers_15_fc2_parametrizations_weight_original2 = constants_->at(436);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_self_attn_layer_norm_weight = constants_->at(437);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_self_attn_layer_norm_bias = constants_->at(438);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_self_attn_q_proj_bias = constants_->at(439);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(440);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(441);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(442);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(443);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(444);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(445);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_self_attn_v_proj_bias = constants_->at(446);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(447);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(448);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(449);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_self_attn_out_proj_bias = constants_->at(450);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(451);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(452);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(453);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_final_layer_norm_weight = constants_->at(454);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_final_layer_norm_bias = constants_->at(455);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_fc1_bias = constants_->at(456);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_fc1_parametrizations_weight_original0 = constants_->at(457);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_fc1_parametrizations_weight_original1 = constants_->at(458);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_fc1_parametrizations_weight_original2 = constants_->at(459);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_fc2_bias = constants_->at(460);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_fc2_parametrizations_weight_original0 = constants_->at(461);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_fc2_parametrizations_weight_original1 = constants_->at(462);
	    [[maybe_unused]] auto& model_audio_tower_layers_16_fc2_parametrizations_weight_original2 = constants_->at(463);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_self_attn_layer_norm_weight = constants_->at(464);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_self_attn_layer_norm_bias = constants_->at(465);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_self_attn_q_proj_bias = constants_->at(466);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(467);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(468);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(469);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(470);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(471);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(472);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_self_attn_v_proj_bias = constants_->at(473);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(474);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(475);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(476);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_self_attn_out_proj_bias = constants_->at(477);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(478);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(479);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(480);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_final_layer_norm_weight = constants_->at(481);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_final_layer_norm_bias = constants_->at(482);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_fc1_bias = constants_->at(483);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_fc1_parametrizations_weight_original0 = constants_->at(484);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_fc1_parametrizations_weight_original1 = constants_->at(485);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_fc1_parametrizations_weight_original2 = constants_->at(486);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_fc2_bias = constants_->at(487);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_fc2_parametrizations_weight_original0 = constants_->at(488);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_fc2_parametrizations_weight_original1 = constants_->at(489);
	    [[maybe_unused]] auto& model_audio_tower_layers_17_fc2_parametrizations_weight_original2 = constants_->at(490);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_self_attn_layer_norm_weight = constants_->at(491);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_self_attn_layer_norm_bias = constants_->at(492);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_self_attn_q_proj_bias = constants_->at(493);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(494);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(495);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(496);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(497);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(498);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(499);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_self_attn_v_proj_bias = constants_->at(500);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(501);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(502);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(503);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_self_attn_out_proj_bias = constants_->at(504);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(505);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(506);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(507);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_final_layer_norm_weight = constants_->at(508);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_final_layer_norm_bias = constants_->at(509);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_fc1_bias = constants_->at(510);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_fc1_parametrizations_weight_original0 = constants_->at(511);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_fc1_parametrizations_weight_original1 = constants_->at(512);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_fc1_parametrizations_weight_original2 = constants_->at(513);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_fc2_bias = constants_->at(514);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_fc2_parametrizations_weight_original0 = constants_->at(515);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_fc2_parametrizations_weight_original1 = constants_->at(516);
	    [[maybe_unused]] auto& model_audio_tower_layers_18_fc2_parametrizations_weight_original2 = constants_->at(517);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_self_attn_layer_norm_weight = constants_->at(518);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_self_attn_layer_norm_bias = constants_->at(519);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_self_attn_q_proj_bias = constants_->at(520);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(521);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(522);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(523);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(524);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(525);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(526);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_self_attn_v_proj_bias = constants_->at(527);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(528);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(529);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(530);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_self_attn_out_proj_bias = constants_->at(531);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(532);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(533);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(534);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_final_layer_norm_weight = constants_->at(535);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_final_layer_norm_bias = constants_->at(536);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_fc1_bias = constants_->at(537);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_fc1_parametrizations_weight_original0 = constants_->at(538);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_fc1_parametrizations_weight_original1 = constants_->at(539);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_fc1_parametrizations_weight_original2 = constants_->at(540);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_fc2_bias = constants_->at(541);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_fc2_parametrizations_weight_original0 = constants_->at(542);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_fc2_parametrizations_weight_original1 = constants_->at(543);
	    [[maybe_unused]] auto& model_audio_tower_layers_19_fc2_parametrizations_weight_original2 = constants_->at(544);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_self_attn_layer_norm_weight = constants_->at(545);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_self_attn_layer_norm_bias = constants_->at(546);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_self_attn_q_proj_bias = constants_->at(547);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(548);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(549);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(550);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(551);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(552);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(553);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_self_attn_v_proj_bias = constants_->at(554);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(555);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(556);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(557);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_self_attn_out_proj_bias = constants_->at(558);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(559);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(560);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(561);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_final_layer_norm_weight = constants_->at(562);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_final_layer_norm_bias = constants_->at(563);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_fc1_bias = constants_->at(564);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_fc1_parametrizations_weight_original0 = constants_->at(565);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_fc1_parametrizations_weight_original1 = constants_->at(566);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_fc1_parametrizations_weight_original2 = constants_->at(567);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_fc2_bias = constants_->at(568);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_fc2_parametrizations_weight_original0 = constants_->at(569);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_fc2_parametrizations_weight_original1 = constants_->at(570);
	    [[maybe_unused]] auto& model_audio_tower_layers_20_fc2_parametrizations_weight_original2 = constants_->at(571);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_self_attn_layer_norm_weight = constants_->at(572);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_self_attn_layer_norm_bias = constants_->at(573);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_self_attn_q_proj_bias = constants_->at(574);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(575);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(576);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(577);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(578);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(579);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(580);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_self_attn_v_proj_bias = constants_->at(581);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(582);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(583);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(584);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_self_attn_out_proj_bias = constants_->at(585);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(586);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(587);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(588);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_final_layer_norm_weight = constants_->at(589);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_final_layer_norm_bias = constants_->at(590);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_fc1_bias = constants_->at(591);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_fc1_parametrizations_weight_original0 = constants_->at(592);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_fc1_parametrizations_weight_original1 = constants_->at(593);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_fc1_parametrizations_weight_original2 = constants_->at(594);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_fc2_bias = constants_->at(595);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_fc2_parametrizations_weight_original0 = constants_->at(596);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_fc2_parametrizations_weight_original1 = constants_->at(597);
	    [[maybe_unused]] auto& model_audio_tower_layers_21_fc2_parametrizations_weight_original2 = constants_->at(598);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_self_attn_layer_norm_weight = constants_->at(599);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_self_attn_layer_norm_bias = constants_->at(600);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_self_attn_q_proj_bias = constants_->at(601);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(602);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(603);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(604);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(605);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(606);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(607);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_self_attn_v_proj_bias = constants_->at(608);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(609);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(610);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(611);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_self_attn_out_proj_bias = constants_->at(612);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(613);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(614);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(615);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_final_layer_norm_weight = constants_->at(616);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_final_layer_norm_bias = constants_->at(617);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_fc1_bias = constants_->at(618);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_fc1_parametrizations_weight_original0 = constants_->at(619);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_fc1_parametrizations_weight_original1 = constants_->at(620);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_fc1_parametrizations_weight_original2 = constants_->at(621);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_fc2_bias = constants_->at(622);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_fc2_parametrizations_weight_original0 = constants_->at(623);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_fc2_parametrizations_weight_original1 = constants_->at(624);
	    [[maybe_unused]] auto& model_audio_tower_layers_22_fc2_parametrizations_weight_original2 = constants_->at(625);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_self_attn_layer_norm_weight = constants_->at(626);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_self_attn_layer_norm_bias = constants_->at(627);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_self_attn_q_proj_bias = constants_->at(628);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(629);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(630);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(631);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(632);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(633);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(634);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_self_attn_v_proj_bias = constants_->at(635);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(636);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(637);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(638);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_self_attn_out_proj_bias = constants_->at(639);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(640);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(641);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(642);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_final_layer_norm_weight = constants_->at(643);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_final_layer_norm_bias = constants_->at(644);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_fc1_bias = constants_->at(645);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_fc1_parametrizations_weight_original0 = constants_->at(646);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_fc1_parametrizations_weight_original1 = constants_->at(647);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_fc1_parametrizations_weight_original2 = constants_->at(648);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_fc2_bias = constants_->at(649);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_fc2_parametrizations_weight_original0 = constants_->at(650);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_fc2_parametrizations_weight_original1 = constants_->at(651);
	    [[maybe_unused]] auto& model_audio_tower_layers_23_fc2_parametrizations_weight_original2 = constants_->at(652);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_self_attn_layer_norm_weight = constants_->at(653);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_self_attn_layer_norm_bias = constants_->at(654);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_self_attn_q_proj_bias = constants_->at(655);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(656);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(657);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(658);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(659);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(660);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(661);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_self_attn_v_proj_bias = constants_->at(662);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(663);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(664);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(665);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_self_attn_out_proj_bias = constants_->at(666);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(667);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(668);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(669);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_final_layer_norm_weight = constants_->at(670);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_final_layer_norm_bias = constants_->at(671);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_fc1_bias = constants_->at(672);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_fc1_parametrizations_weight_original0 = constants_->at(673);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_fc1_parametrizations_weight_original1 = constants_->at(674);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_fc1_parametrizations_weight_original2 = constants_->at(675);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_fc2_bias = constants_->at(676);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_fc2_parametrizations_weight_original0 = constants_->at(677);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_fc2_parametrizations_weight_original1 = constants_->at(678);
	    [[maybe_unused]] auto& model_audio_tower_layers_24_fc2_parametrizations_weight_original2 = constants_->at(679);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_self_attn_layer_norm_weight = constants_->at(680);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_self_attn_layer_norm_bias = constants_->at(681);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_self_attn_q_proj_bias = constants_->at(682);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(683);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(684);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(685);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(686);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(687);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(688);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_self_attn_v_proj_bias = constants_->at(689);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(690);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(691);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(692);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_self_attn_out_proj_bias = constants_->at(693);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(694);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(695);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(696);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_final_layer_norm_weight = constants_->at(697);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_final_layer_norm_bias = constants_->at(698);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_fc1_bias = constants_->at(699);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_fc1_parametrizations_weight_original0 = constants_->at(700);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_fc1_parametrizations_weight_original1 = constants_->at(701);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_fc1_parametrizations_weight_original2 = constants_->at(702);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_fc2_bias = constants_->at(703);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_fc2_parametrizations_weight_original0 = constants_->at(704);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_fc2_parametrizations_weight_original1 = constants_->at(705);
	    [[maybe_unused]] auto& model_audio_tower_layers_25_fc2_parametrizations_weight_original2 = constants_->at(706);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_self_attn_layer_norm_weight = constants_->at(707);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_self_attn_layer_norm_bias = constants_->at(708);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_self_attn_q_proj_bias = constants_->at(709);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(710);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(711);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(712);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(713);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(714);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(715);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_self_attn_v_proj_bias = constants_->at(716);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(717);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(718);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(719);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_self_attn_out_proj_bias = constants_->at(720);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(721);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(722);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(723);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_final_layer_norm_weight = constants_->at(724);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_final_layer_norm_bias = constants_->at(725);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_fc1_bias = constants_->at(726);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_fc1_parametrizations_weight_original0 = constants_->at(727);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_fc1_parametrizations_weight_original1 = constants_->at(728);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_fc1_parametrizations_weight_original2 = constants_->at(729);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_fc2_bias = constants_->at(730);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_fc2_parametrizations_weight_original0 = constants_->at(731);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_fc2_parametrizations_weight_original1 = constants_->at(732);
	    [[maybe_unused]] auto& model_audio_tower_layers_26_fc2_parametrizations_weight_original2 = constants_->at(733);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_self_attn_layer_norm_weight = constants_->at(734);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_self_attn_layer_norm_bias = constants_->at(735);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_self_attn_q_proj_bias = constants_->at(736);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(737);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(738);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(739);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(740);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(741);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(742);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_self_attn_v_proj_bias = constants_->at(743);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(744);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(745);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(746);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_self_attn_out_proj_bias = constants_->at(747);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(748);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(749);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(750);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_final_layer_norm_weight = constants_->at(751);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_final_layer_norm_bias = constants_->at(752);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_fc1_bias = constants_->at(753);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_fc1_parametrizations_weight_original0 = constants_->at(754);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_fc1_parametrizations_weight_original1 = constants_->at(755);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_fc1_parametrizations_weight_original2 = constants_->at(756);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_fc2_bias = constants_->at(757);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_fc2_parametrizations_weight_original0 = constants_->at(758);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_fc2_parametrizations_weight_original1 = constants_->at(759);
	    [[maybe_unused]] auto& model_audio_tower_layers_27_fc2_parametrizations_weight_original2 = constants_->at(760);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_self_attn_layer_norm_weight = constants_->at(761);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_self_attn_layer_norm_bias = constants_->at(762);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_self_attn_q_proj_bias = constants_->at(763);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(764);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(765);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(766);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(767);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(768);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(769);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_self_attn_v_proj_bias = constants_->at(770);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(771);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(772);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(773);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_self_attn_out_proj_bias = constants_->at(774);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(775);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(776);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(777);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_final_layer_norm_weight = constants_->at(778);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_final_layer_norm_bias = constants_->at(779);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_fc1_bias = constants_->at(780);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_fc1_parametrizations_weight_original0 = constants_->at(781);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_fc1_parametrizations_weight_original1 = constants_->at(782);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_fc1_parametrizations_weight_original2 = constants_->at(783);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_fc2_bias = constants_->at(784);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_fc2_parametrizations_weight_original0 = constants_->at(785);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_fc2_parametrizations_weight_original1 = constants_->at(786);
	    [[maybe_unused]] auto& model_audio_tower_layers_28_fc2_parametrizations_weight_original2 = constants_->at(787);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_self_attn_layer_norm_weight = constants_->at(788);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_self_attn_layer_norm_bias = constants_->at(789);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_self_attn_q_proj_bias = constants_->at(790);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(791);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(792);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(793);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(794);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(795);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(796);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_self_attn_v_proj_bias = constants_->at(797);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(798);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(799);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(800);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_self_attn_out_proj_bias = constants_->at(801);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(802);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(803);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(804);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_final_layer_norm_weight = constants_->at(805);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_final_layer_norm_bias = constants_->at(806);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_fc1_bias = constants_->at(807);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_fc1_parametrizations_weight_original0 = constants_->at(808);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_fc1_parametrizations_weight_original1 = constants_->at(809);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_fc1_parametrizations_weight_original2 = constants_->at(810);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_fc2_bias = constants_->at(811);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_fc2_parametrizations_weight_original0 = constants_->at(812);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_fc2_parametrizations_weight_original1 = constants_->at(813);
	    [[maybe_unused]] auto& model_audio_tower_layers_29_fc2_parametrizations_weight_original2 = constants_->at(814);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_self_attn_layer_norm_weight = constants_->at(815);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_self_attn_layer_norm_bias = constants_->at(816);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_self_attn_q_proj_bias = constants_->at(817);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(818);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(819);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(820);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(821);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(822);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(823);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_self_attn_v_proj_bias = constants_->at(824);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(825);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(826);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(827);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_self_attn_out_proj_bias = constants_->at(828);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(829);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(830);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(831);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_final_layer_norm_weight = constants_->at(832);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_final_layer_norm_bias = constants_->at(833);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_fc1_bias = constants_->at(834);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_fc1_parametrizations_weight_original0 = constants_->at(835);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_fc1_parametrizations_weight_original1 = constants_->at(836);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_fc1_parametrizations_weight_original2 = constants_->at(837);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_fc2_bias = constants_->at(838);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_fc2_parametrizations_weight_original0 = constants_->at(839);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_fc2_parametrizations_weight_original1 = constants_->at(840);
	    [[maybe_unused]] auto& model_audio_tower_layers_30_fc2_parametrizations_weight_original2 = constants_->at(841);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_self_attn_layer_norm_weight = constants_->at(842);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_self_attn_layer_norm_bias = constants_->at(843);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_self_attn_q_proj_bias = constants_->at(844);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0 = constants_->at(845);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1 = constants_->at(846);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2 = constants_->at(847);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0 = constants_->at(848);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1 = constants_->at(849);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2 = constants_->at(850);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_self_attn_v_proj_bias = constants_->at(851);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0 = constants_->at(852);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1 = constants_->at(853);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2 = constants_->at(854);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_self_attn_out_proj_bias = constants_->at(855);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0 = constants_->at(856);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1 = constants_->at(857);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2 = constants_->at(858);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_final_layer_norm_weight = constants_->at(859);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_final_layer_norm_bias = constants_->at(860);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_fc1_bias = constants_->at(861);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_fc1_parametrizations_weight_original0 = constants_->at(862);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_fc1_parametrizations_weight_original1 = constants_->at(863);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_fc1_parametrizations_weight_original2 = constants_->at(864);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_fc2_bias = constants_->at(865);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_fc2_parametrizations_weight_original0 = constants_->at(866);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_fc2_parametrizations_weight_original1 = constants_->at(867);
	    [[maybe_unused]] auto& model_audio_tower_layers_31_fc2_parametrizations_weight_original2 = constants_->at(868);
	    [[maybe_unused]] auto& model_audio_tower_layer_norm_weight = constants_->at(869);
	    [[maybe_unused]] auto& model_audio_tower_layer_norm_bias = constants_->at(870);
	    [[maybe_unused]] auto& model_multi_modal_projector_linear_1_parametrizations_weight_original0 = constants_->at(871);
	    [[maybe_unused]] auto& model_multi_modal_projector_linear_1_parametrizations_weight_original1 = constants_->at(872);
	    [[maybe_unused]] auto& model_multi_modal_projector_linear_1_parametrizations_weight_original2 = constants_->at(873);
	    [[maybe_unused]] auto& model_multi_modal_projector_linear_2_parametrizations_weight_original0 = constants_->at(874);
	    [[maybe_unused]] auto& model_multi_modal_projector_linear_2_parametrizations_weight_original1 = constants_->at(875);
	    [[maybe_unused]] auto& model_multi_modal_projector_linear_2_parametrizations_weight_original2 = constants_->at(876);
	    auto arg877_1_size = arg877_1.sizes();
	    int64_t s6 = arg877_1_size[0];
	    inputs.clear();
	    [[maybe_unused]] auto& kernels = static_cast<AOTInductorModelKernels&>(*this->kernels_.get());
	
	    AOTICudaStreamGuard stream_guard(stream, this->device_idx_);
	    // Topologically Sorted Source Nodes: [conv1d], Original ATen: [aten.convolution]
	    AtenTensorHandle buf0_handle;
	    // [Provenance debug handles] aoti_torch_cuda_convolution:425
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_convolution(arg877_1, model_audio_tower_conv1_weight, nullptr, std::array<int64_t, 1>{1L}.cbegin(), 1, std::array<int64_t, 1>{1L}.cbegin(), 1, std::array<int64_t, 1>{1L}.cbegin(), 1, 0, std::array<int64_t, 1>{0L}.cbegin(), 1, 1L, &buf0_handle));
	    RAIIAtenTensorHandle buf0(buf0_handle);
	    arg877_1.reset();
	    auto buf1 = std::move(buf0);  // reuse
	    // Topologically Sorted Source Nodes: [conv1d, gelu, conv1d_1], Original ATen: [aten.convolution, aten.gelu]
	    // [Provenance debug handles] triton_poi_fused_convolution_gelu_0:1
	    int64_t triton_poi_fused_convolution_gelu_0_xnumel = 3840000L*s6;
	    call_triton_poi_fused_convolution_gelu_0(buf1, model_audio_tower_conv1_bias, triton_poi_fused_convolution_gelu_0_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    // Topologically Sorted Source Nodes: [conv1d, gelu, conv1d_1], Original ATen: [aten.convolution, aten.gelu]
	    AtenTensorHandle buf2_handle;
	    // [Provenance debug handles] aoti_torch_cuda_convolution:426
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_convolution(buf1, model_audio_tower_conv2_weight, nullptr, std::array<int64_t, 1>{2L}.cbegin(), 1, std::array<int64_t, 1>{1L}.cbegin(), 1, std::array<int64_t, 1>{1L}.cbegin(), 1, 0, std::array<int64_t, 1>{0L}.cbegin(), 1, 1L, &buf2_handle));
	    RAIIAtenTensorHandle buf2(buf2_handle);
	    buf1.reset();
	    const int64_t int_array_0[] = {s6, 1500L, 1L, 10L};
	    const int64_t int_array_1[] = {15008L, 10L, 15008L*s6, 1L};
	    AtenTensorHandle buf3_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(4, int_array_0, int_array_1, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf3_handle));
	    RAIIAtenTensorHandle buf3(buf3_handle);
	    AtenTensorHandle buf4_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(4, int_array_0, int_array_1, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf4_handle));
	    RAIIAtenTensorHandle buf4(buf4_handle);
	    AtenTensorHandle buf5_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(4, int_array_0, int_array_1, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf5_handle));
	    RAIIAtenTensorHandle buf5(buf5_handle);
	    // Topologically Sorted Source Nodes: [conv1d, gelu, conv1d_1, gelu_1, permute, add, layer_norm], Original ATen: [aten.convolution, aten.gelu, aten.permute, aten.add, aten.native_layer_norm]
	    // [Provenance debug handles] triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1:2
	    int64_t triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1_xnumel_1 = 15000L*s6; // manually changed for tlparse testing
	    call_triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1(buf2, model_audio_tower_conv2_bias, model_audio_tower_embed_positions_weight, buf3, buf4, buf5, triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1_xnumel_1, 128L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    const int64_t int_array_2[] = {s6, 1500L, 1L};
	    const int64_t int_array_3[] = {1504L, 1L, 1504L*s6};
	    AtenTensorHandle buf6_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_2, int_array_3, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf6_handle));
	    RAIIAtenTensorHandle buf6(buf6_handle);
	    AtenTensorHandle buf7_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_2, int_array_3, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf7_handle));
	    RAIIAtenTensorHandle buf7(buf7_handle);
	    // Topologically Sorted Source Nodes: [conv1d, gelu, conv1d_1, gelu_1, permute, add, layer_norm], Original ATen: [aten.convolution, aten.gelu, aten.permute, aten.add, aten.native_layer_norm]
	    // [Provenance debug handles] triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2:3
	    int64_t triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2_xnumel = 1500L*s6;
	    call_triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2(buf3, buf4, buf5, buf6, buf7, triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2_xnumel, 10L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf3.reset();
	    buf4.reset();
	    buf5.reset();
	    const int64_t int_array_4[] = {s6, 1500L, 1280L};
	    static constexpr int64_t int_array_5[] = {1920000L, 1280L, 1L};
	    AtenTensorHandle buf9_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf9_handle));
	    RAIIAtenTensorHandle buf9(buf9_handle);
	    AtenTensorHandle buf12_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf12_handle));
	    RAIIAtenTensorHandle buf12(buf12_handle);
	    auto buf19 = std::move(buf12);  // reuse
	    AtenTensorHandle buf15_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf15_handle));
	    RAIIAtenTensorHandle buf15(buf15_handle);
	    auto buf22 = std::move(buf15);  // reuse
	    AtenTensorHandle buf18_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf18_handle));
	    RAIIAtenTensorHandle buf18(buf18_handle);
	    auto buf25 = std::move(buf18);  // reuse
	    // Topologically Sorted Source Nodes: [conv1d, gelu, conv1d_1, gelu_1, permute, add, layer_norm, choose_qparams_affine_default, quantize_affine, dequantize_affine, choose_qparams_affine_default_1, quantize_affine_1, dequantize_affine_2, choose_qparams_affine_default_2, quantize_affine_2, dequantize_affine_4], Original ATen: [aten.convolution, aten.gelu, aten.permute, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.view, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4
	    int64_t triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3(buf19, buf22, buf25, buf2, model_audio_tower_conv2_bias, model_audio_tower_embed_positions_weight, buf6, buf7, model_audio_tower_layers_0_self_attn_layer_norm_weight, model_audio_tower_layers_0_self_attn_layer_norm_bias, buf9, triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf6.reset();
	    buf7.reset();
	    static constexpr int64_t int_array_6[] = {1280L, 40L, 32L};
	    static constexpr int64_t int_array_7[] = {1280L, 32L, 1L};
	    AtenTensorHandle buf20_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf20_handle));
	    RAIIAtenTensorHandle buf20(buf20_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_1], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:5
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_0_self_attn_q_proj_parametrizations_weight_original1, buf20, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    const int64_t int_array_8[] = {1500L*s6, 1280L};
	    static constexpr int64_t int_array_9[] = {1280L, 1L};
	    auto buf21 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf9, 2, int_array_8, int_array_9, 0L)); buf9.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default, dequantize_affine, dequantize_affine_1, linear, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    static constexpr int64_t int_array_10[] = {1280L, 1280L};
	    static constexpr int64_t int_array_11[] = {1L, 1280L};
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:427
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf21, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf19, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf20, 2, int_array_10, int_array_11, 0L))));
	    auto buf23 = std::move(buf20);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_3], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:6
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_0_self_attn_k_proj_parametrizations_weight_original1, buf23, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf24 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf19, 2, int_array_8, int_array_9, 0L)); buf19.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_1, dequantize_affine_2, dequantize_affine_3, linear_1], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:428
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf24, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf22, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf23, 2, int_array_10, int_array_11, 0L))));
	    auto buf26 = std::move(buf23);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_5], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:7
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_0_self_attn_v_proj_parametrizations_weight_original1, buf26, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf27 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf22, 2, int_array_8, int_array_9, 0L)); buf22.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_2, dequantize_affine_4, dequantize_affine_5, linear_2, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:429
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf27, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf25, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf26, 2, int_array_10, int_array_11, 0L))));
	    const int64_t int_array_12[] = {s6, 20L, 1500L, 64L};
	    static constexpr int64_t int_array_13[] = {1920000L, 96000L, 64L, 1L};
	    auto buf28 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf25, 4, int_array_12, int_array_13, 0L)); buf25.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear, mul_5, view, transpose, contiguous, linear_1, view_1, transpose_1, contiguous_1, linear_2, view_2, transpose_2, contiguous_2, scaled_dot_product_attention], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:8
	    int64_t triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf21, model_audio_tower_layers_0_self_attn_q_proj_bias, buf28, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf29 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf21, 4, int_array_12, int_array_13, 0L)); buf21.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear, mul_5, view, transpose, contiguous, linear_1, view_1, transpose_1, contiguous_1, linear_2, view_2, transpose_2, contiguous_2, scaled_dot_product_attention], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:9
	    int64_t triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf24, buf29, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf30 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf24, 4, int_array_12, int_array_13, 0L)); buf24.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear, mul_5, view, transpose, contiguous, linear_1, view_1, transpose_1, contiguous_1, linear_2, view_2, transpose_2, contiguous_2, scaled_dot_product_attention], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:10
	    int64_t triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf27, model_audio_tower_layers_0_self_attn_v_proj_bias, buf30, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf27.reset();
	    // Topologically Sorted Source Nodes: [, linear, mul_5, view, transpose, contiguous, linear_1, view_1, transpose_1, contiguous_1, linear_2, view_2, transpose_2, contiguous_2, scaled_dot_product_attention], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_0 = 1.0;
	    AtenTensorHandle buf32_handle;
	    AtenTensorHandle buf33_handle;
	    AtenTensorHandle buf34_handle;
	    AtenTensorHandle buf35_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:430
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf28, buf29, buf30, nullptr, 0, 0.0, 0, &var_0, &buf32_handle, &buf33_handle, &buf34_handle, &buf35_handle));
	    RAIIAtenTensorHandle buf32(buf32_handle);
	    RAIIAtenTensorHandle buf33(buf33_handle);
	    RAIIAtenTensorHandle buf34(buf34_handle);
	    RAIIAtenTensorHandle buf35(buf35_handle);
	    buf28.reset();
	    buf29.reset();
	
	    auto buf38 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf32, 3, int_array_4, int_array_5, 0L)); buf32.reset();  // reuse
	    auto buf39 = std::move(buf38);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_3, reshape, choose_qparams_affine_default_3, quantize_affine_3, dequantize_affine_6], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11
	    int64_t triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf39, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf40 = std::move(buf26);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_7], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:12
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_0_self_attn_out_proj_parametrizations_weight_original1, buf40, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf41 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf30, 2, int_array_8, int_array_9, 0L)); buf30.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_3, dequantize_affine_6, dequantize_affine_7, linear_3, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:431
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf41, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf39, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf40, 2, int_array_10, int_array_11, 0L))));
	    buf40.reset();
	    auto buf42 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf41, 3, int_array_4, int_array_5, 0L)); buf41.reset();  // reuse
	    auto buf46 = std::move(buf39);  // reuse
	    auto buf49 = std::move(buf46);  // reuse
	    auto buf50 = std::move(buf49);  // reuse
	    // Topologically Sorted Source Nodes: [conv1d, gelu, conv1d_1, gelu_1, permute, add, , linear_3, add_9, layer_norm_1, choose_qparams_affine_default_4, quantize_affine_4, dequantize_affine_8], Original ATen: [aten.convolution, aten.gelu, aten.permute, aten.add, aten.addmm, aten.view, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13
	    int64_t triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9(buf42, buf50, buf2, model_audio_tower_conv2_bias, model_audio_tower_embed_positions_weight, model_audio_tower_layers_0_self_attn_out_proj_bias, model_audio_tower_layers_0_final_layer_norm_weight, model_audio_tower_layers_0_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf2.reset();
	    static constexpr int64_t int_array_14[] = {5120L, 40L, 32L};
	    AtenTensorHandle buf51_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_14, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf51_handle));
	    RAIIAtenTensorHandle buf51(buf51_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_9], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:14
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_0_fc1_parametrizations_weight_original0, model_audio_tower_layers_0_fc1_parametrizations_weight_original2, model_audio_tower_layers_0_fc1_parametrizations_weight_original1, buf51, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    const int64_t int_array_15[] = {1500L*s6, 5120L};
	    static constexpr int64_t int_array_16[] = {5120L, 1L};
	    AtenTensorHandle buf52_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf52_handle));
	    RAIIAtenTensorHandle buf52(buf52_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_4, dequantize_affine_8, dequantize_affine_9, linear_4, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    static constexpr int64_t int_array_17[] = {1280L, 5120L};
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:432
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf52, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf50, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf51, 2, int_array_17, int_array_11, 0L))));
	    buf50.reset();
	    buf51.reset();
	    const int64_t int_array_18[] = {s6, 1500L, 5120L};
	    static constexpr int64_t int_array_19[] = {7680000L, 5120L, 1L};
	    auto buf55 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf52, 3, int_array_18, int_array_19, 0L)); buf52.reset();  // reuse
	    auto buf56 = std::move(buf55);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_4, gelu_2, choose_qparams_affine_default_5, quantize_affine_5, dequantize_affine_10], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15
	    int64_t triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf56, model_audio_tower_layers_0_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    static constexpr int64_t int_array_20[] = {1280L, 160L, 32L};
	    static constexpr int64_t int_array_21[] = {5120L, 32L, 1L};
	    AtenTensorHandle buf57_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf57_handle));
	    RAIIAtenTensorHandle buf57(buf57_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_11], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:16
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_0_fc2_parametrizations_weight_original0, model_audio_tower_layers_0_fc2_parametrizations_weight_original2, model_audio_tower_layers_0_fc2_parametrizations_weight_original1, buf57, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf58_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf58_handle));
	    RAIIAtenTensorHandle buf58(buf58_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_5, quantize_affine_5, dequantize_affine_10, dequantize_affine_11, linear_5, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    static constexpr int64_t int_array_22[] = {5120L, 1280L};
	    static constexpr int64_t int_array_23[] = {1L, 5120L};
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:433
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf58, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf56, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf57, 2, int_array_22, int_array_23, 0L))));
	    buf56.reset();
	    AtenTensorHandle buf62_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf62_handle));
	    RAIIAtenTensorHandle buf62(buf62_handle);
	    AtenTensorHandle buf65_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf65_handle));
	    RAIIAtenTensorHandle buf65(buf65_handle);
	    auto buf72 = std::move(buf65);  // reuse
	    AtenTensorHandle buf68_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf68_handle));
	    RAIIAtenTensorHandle buf68(buf68_handle);
	    auto buf75 = std::move(buf68);  // reuse
	    AtenTensorHandle buf71_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf71_handle));
	    RAIIAtenTensorHandle buf71(buf71_handle);
	    auto buf78 = std::move(buf71);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_5, add_14, layer_norm_2, choose_qparams_affine_default_6, quantize_affine_6, dequantize_affine_12, choose_qparams_affine_default_7, quantize_affine_7, dequantize_affine_14, choose_qparams_affine_default_8, quantize_affine_8, dequantize_affine_16], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17
	    int64_t triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf72, buf75, buf78, buf42, buf58, model_audio_tower_layers_0_fc2_bias, model_audio_tower_layers_1_self_attn_layer_norm_weight, model_audio_tower_layers_1_self_attn_layer_norm_bias, buf62, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf73_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf73_handle));
	    RAIIAtenTensorHandle buf73(buf73_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_13], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:18
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_1_self_attn_q_proj_parametrizations_weight_original1, buf73, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf74 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf62, 2, int_array_8, int_array_9, 0L)); buf62.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_6, dequantize_affine_12, dequantize_affine_13, linear_6, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:434
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf74, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf72, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf73, 2, int_array_10, int_array_11, 0L))));
	    auto buf76 = std::move(buf73);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_15], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:19
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_1_self_attn_k_proj_parametrizations_weight_original1, buf76, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf77 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf72, 2, int_array_8, int_array_9, 0L)); buf72.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_7, dequantize_affine_14, dequantize_affine_15, linear_7], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:435
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf77, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf75, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf76, 2, int_array_10, int_array_11, 0L))));
	    auto buf79 = std::move(buf76);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_17], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:20
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_1_self_attn_v_proj_parametrizations_weight_original1, buf79, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf80 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf75, 2, int_array_8, int_array_9, 0L)); buf75.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_8, dequantize_affine_16, dequantize_affine_17, linear_8, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:436
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf80, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf78, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf79, 2, int_array_10, int_array_11, 0L))));
	    auto buf81 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf78, 4, int_array_12, int_array_13, 0L)); buf78.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_6, mul_42, view_3, transpose_4, contiguous_4, linear_7, view_4, transpose_5, contiguous_5, linear_8, view_5, transpose_6, contiguous_6, scaled_dot_product_attention_1], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:21
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf74, model_audio_tower_layers_1_self_attn_q_proj_bias, buf81, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf82 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf74, 4, int_array_12, int_array_13, 0L)); buf74.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_6, mul_42, view_3, transpose_4, contiguous_4, linear_7, view_4, transpose_5, contiguous_5, linear_8, view_5, transpose_6, contiguous_6, scaled_dot_product_attention_1], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:22
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf77, buf82, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf83 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf77, 4, int_array_12, int_array_13, 0L)); buf77.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_6, mul_42, view_3, transpose_4, contiguous_4, linear_7, view_4, transpose_5, contiguous_5, linear_8, view_5, transpose_6, contiguous_6, scaled_dot_product_attention_1], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:23
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf80, model_audio_tower_layers_1_self_attn_v_proj_bias, buf83, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf80.reset();
	    // Topologically Sorted Source Nodes: [, linear_6, mul_42, view_3, transpose_4, contiguous_4, linear_7, view_4, transpose_5, contiguous_5, linear_8, view_5, transpose_6, contiguous_6, scaled_dot_product_attention_1], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_1 = 1.0;
	    AtenTensorHandle buf85_handle;
	    AtenTensorHandle buf86_handle;
	    AtenTensorHandle buf87_handle;
	    AtenTensorHandle buf88_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:437
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf81, buf82, buf83, nullptr, 0, 0.0, 0, &var_1, &buf85_handle, &buf86_handle, &buf87_handle, &buf88_handle));
	    RAIIAtenTensorHandle buf85(buf85_handle);
	    RAIIAtenTensorHandle buf86(buf86_handle);
	    RAIIAtenTensorHandle buf87(buf87_handle);
	    RAIIAtenTensorHandle buf88(buf88_handle);
	    buf81.reset();
	    buf82.reset();
	
	    auto buf91 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf85, 3, int_array_4, int_array_5, 0L)); buf85.reset();  // reuse
	    auto buf92 = std::move(buf91);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_7, reshape_1, choose_qparams_affine_default_9, quantize_affine_9, dequantize_affine_18], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf92, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf93 = std::move(buf79);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_19], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:25
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_1_self_attn_out_proj_parametrizations_weight_original1, buf93, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf94 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf83, 2, int_array_8, int_array_9, 0L)); buf83.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_9, dequantize_affine_18, dequantize_affine_19, linear_9, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:438
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf94, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf92, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf93, 2, int_array_10, int_array_11, 0L))));
	    buf93.reset();
	    auto buf95 = std::move(buf42);  // reuse
	    auto buf99 = std::move(buf92);  // reuse
	    auto buf102 = std::move(buf99);  // reuse
	    auto buf103 = std::move(buf102);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_5, add_14, linear_9, add_23, layer_norm_3, choose_qparams_affine_default_10, quantize_affine_10, dequantize_affine_20], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26
	    int64_t triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf95, buf103, buf58, model_audio_tower_layers_0_fc2_bias, buf94, model_audio_tower_layers_1_self_attn_out_proj_bias, model_audio_tower_layers_1_final_layer_norm_weight, model_audio_tower_layers_1_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf58.reset();
	    buf94.reset();
	    auto buf104 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf57, 3, int_array_14, int_array_7, 0L)); buf57.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_21], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:27
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_1_fc1_parametrizations_weight_original0, model_audio_tower_layers_1_fc1_parametrizations_weight_original2, model_audio_tower_layers_1_fc1_parametrizations_weight_original1, buf104, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf105_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf105_handle));
	    RAIIAtenTensorHandle buf105(buf105_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_10, dequantize_affine_20, dequantize_affine_21, linear_10, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:439
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf105, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf103, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf104, 2, int_array_17, int_array_11, 0L))));
	    buf103.reset();
	    buf104.reset();
	    auto buf108 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf105, 3, int_array_18, int_array_19, 0L)); buf105.reset();  // reuse
	    auto buf109 = std::move(buf108);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_10, gelu_3, choose_qparams_affine_default_11, quantize_affine_11, dequantize_affine_22], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf109, model_audio_tower_layers_1_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf110_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf110_handle));
	    RAIIAtenTensorHandle buf110(buf110_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_23], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:29
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_1_fc2_parametrizations_weight_original0, model_audio_tower_layers_1_fc2_parametrizations_weight_original2, model_audio_tower_layers_1_fc2_parametrizations_weight_original1, buf110, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf111_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf111_handle));
	    RAIIAtenTensorHandle buf111(buf111_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_11, quantize_affine_11, dequantize_affine_22, dequantize_affine_23, linear_11, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:440
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf111, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf109, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf110, 2, int_array_22, int_array_23, 0L))));
	    buf109.reset();
	    AtenTensorHandle buf115_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf115_handle));
	    RAIIAtenTensorHandle buf115(buf115_handle);
	    AtenTensorHandle buf118_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf118_handle));
	    RAIIAtenTensorHandle buf118(buf118_handle);
	    auto buf125 = std::move(buf118);  // reuse
	    AtenTensorHandle buf121_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf121_handle));
	    RAIIAtenTensorHandle buf121(buf121_handle);
	    auto buf128 = std::move(buf121);  // reuse
	    AtenTensorHandle buf124_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf124_handle));
	    RAIIAtenTensorHandle buf124(buf124_handle);
	    auto buf131 = std::move(buf124);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_11, add_28, layer_norm_4, choose_qparams_affine_default_12, quantize_affine_12, dequantize_affine_24, choose_qparams_affine_default_13, quantize_affine_13, dequantize_affine_26, choose_qparams_affine_default_14, quantize_affine_14, dequantize_affine_28], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf125, buf128, buf131, buf95, buf111, model_audio_tower_layers_1_fc2_bias, model_audio_tower_layers_2_self_attn_layer_norm_weight, model_audio_tower_layers_2_self_attn_layer_norm_bias, buf115, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf126_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf126_handle));
	    RAIIAtenTensorHandle buf126(buf126_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_25], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:31
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_2_self_attn_q_proj_parametrizations_weight_original1, buf126, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf127 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf115, 2, int_array_8, int_array_9, 0L)); buf115.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_12, dequantize_affine_24, dequantize_affine_25, linear_12, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:441
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf127, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf125, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf126, 2, int_array_10, int_array_11, 0L))));
	    auto buf129 = std::move(buf126);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_27], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:32
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_2_self_attn_k_proj_parametrizations_weight_original1, buf129, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf130 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf125, 2, int_array_8, int_array_9, 0L)); buf125.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_13, dequantize_affine_26, dequantize_affine_27, linear_13], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:442
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf130, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf128, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf129, 2, int_array_10, int_array_11, 0L))));
	    auto buf132 = std::move(buf129);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_29], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:33
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_2_self_attn_v_proj_parametrizations_weight_original1, buf132, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf133 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf128, 2, int_array_8, int_array_9, 0L)); buf128.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_14, dequantize_affine_28, dequantize_affine_29, linear_14, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:443
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf133, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf131, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf132, 2, int_array_10, int_array_11, 0L))));
	    auto buf134 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf131, 4, int_array_12, int_array_13, 0L)); buf131.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_12, mul_79, view_6, transpose_8, contiguous_8, linear_13, view_7, transpose_9, contiguous_9, linear_14, view_8, transpose_10, contiguous_10, scaled_dot_product_attention_2], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:34
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf127, model_audio_tower_layers_2_self_attn_q_proj_bias, buf134, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf135 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf127, 4, int_array_12, int_array_13, 0L)); buf127.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_12, mul_79, view_6, transpose_8, contiguous_8, linear_13, view_7, transpose_9, contiguous_9, linear_14, view_8, transpose_10, contiguous_10, scaled_dot_product_attention_2], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:35
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf130, buf135, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf136 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf130, 4, int_array_12, int_array_13, 0L)); buf130.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_12, mul_79, view_6, transpose_8, contiguous_8, linear_13, view_7, transpose_9, contiguous_9, linear_14, view_8, transpose_10, contiguous_10, scaled_dot_product_attention_2], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:36
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf133, model_audio_tower_layers_2_self_attn_v_proj_bias, buf136, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf133.reset();
	    // Topologically Sorted Source Nodes: [, linear_12, mul_79, view_6, transpose_8, contiguous_8, linear_13, view_7, transpose_9, contiguous_9, linear_14, view_8, transpose_10, contiguous_10, scaled_dot_product_attention_2], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_2 = 1.0;
	    AtenTensorHandle buf138_handle;
	    AtenTensorHandle buf139_handle;
	    AtenTensorHandle buf140_handle;
	    AtenTensorHandle buf141_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:444
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf134, buf135, buf136, nullptr, 0, 0.0, 0, &var_2, &buf138_handle, &buf139_handle, &buf140_handle, &buf141_handle));
	    RAIIAtenTensorHandle buf138(buf138_handle);
	    RAIIAtenTensorHandle buf139(buf139_handle);
	    RAIIAtenTensorHandle buf140(buf140_handle);
	    RAIIAtenTensorHandle buf141(buf141_handle);
	    buf134.reset();
	    buf135.reset();
	
	    auto buf144 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf138, 3, int_array_4, int_array_5, 0L)); buf138.reset();  // reuse
	    auto buf145 = std::move(buf144);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_11, reshape_2, choose_qparams_affine_default_15, quantize_affine_15, dequantize_affine_30], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf145, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf146 = std::move(buf132);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_31], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:38
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_2_self_attn_out_proj_parametrizations_weight_original1, buf146, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf147 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf136, 2, int_array_8, int_array_9, 0L)); buf136.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_15, dequantize_affine_30, dequantize_affine_31, linear_15, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:445
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf147, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf145, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf146, 2, int_array_10, int_array_11, 0L))));
	    buf146.reset();
	    auto buf148 = std::move(buf95);  // reuse
	    auto buf152 = std::move(buf145);  // reuse
	    auto buf155 = std::move(buf152);  // reuse
	    auto buf156 = std::move(buf155);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_11, add_28, linear_15, add_37, layer_norm_5, choose_qparams_affine_default_16, quantize_affine_16, dequantize_affine_32], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf148, buf156, buf111, model_audio_tower_layers_1_fc2_bias, buf147, model_audio_tower_layers_2_self_attn_out_proj_bias, model_audio_tower_layers_2_final_layer_norm_weight, model_audio_tower_layers_2_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf111.reset();
	    buf147.reset();
	    auto buf157 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf110, 3, int_array_14, int_array_7, 0L)); buf110.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_33], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:40
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_2_fc1_parametrizations_weight_original0, model_audio_tower_layers_2_fc1_parametrizations_weight_original2, model_audio_tower_layers_2_fc1_parametrizations_weight_original1, buf157, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf158_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf158_handle));
	    RAIIAtenTensorHandle buf158(buf158_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_16, dequantize_affine_32, dequantize_affine_33, linear_16, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:446
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf158, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf156, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf157, 2, int_array_17, int_array_11, 0L))));
	    buf156.reset();
	    buf157.reset();
	    auto buf161 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf158, 3, int_array_18, int_array_19, 0L)); buf158.reset();  // reuse
	    auto buf162 = std::move(buf161);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_16, gelu_4, choose_qparams_affine_default_17, quantize_affine_17, dequantize_affine_34], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf162, model_audio_tower_layers_2_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf163_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf163_handle));
	    RAIIAtenTensorHandle buf163(buf163_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_35], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:42
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_2_fc2_parametrizations_weight_original0, model_audio_tower_layers_2_fc2_parametrizations_weight_original2, model_audio_tower_layers_2_fc2_parametrizations_weight_original1, buf163, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf164_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf164_handle));
	    RAIIAtenTensorHandle buf164(buf164_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_17, quantize_affine_17, dequantize_affine_34, dequantize_affine_35, linear_17, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:447
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf164, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf162, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf163, 2, int_array_22, int_array_23, 0L))));
	    buf162.reset();
	    AtenTensorHandle buf168_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf168_handle));
	    RAIIAtenTensorHandle buf168(buf168_handle);
	    AtenTensorHandle buf171_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf171_handle));
	    RAIIAtenTensorHandle buf171(buf171_handle);
	    auto buf178 = std::move(buf171);  // reuse
	    AtenTensorHandle buf174_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf174_handle));
	    RAIIAtenTensorHandle buf174(buf174_handle);
	    auto buf181 = std::move(buf174);  // reuse
	    AtenTensorHandle buf177_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf177_handle));
	    RAIIAtenTensorHandle buf177(buf177_handle);
	    auto buf184 = std::move(buf177);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_17, add_42, layer_norm_6, choose_qparams_affine_default_18, quantize_affine_18, dequantize_affine_36, choose_qparams_affine_default_19, quantize_affine_19, dequantize_affine_38, choose_qparams_affine_default_20, quantize_affine_20, dequantize_affine_40], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf178, buf181, buf184, buf148, buf164, model_audio_tower_layers_2_fc2_bias, model_audio_tower_layers_3_self_attn_layer_norm_weight, model_audio_tower_layers_3_self_attn_layer_norm_bias, buf168, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf179_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf179_handle));
	    RAIIAtenTensorHandle buf179(buf179_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_37], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:44
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_3_self_attn_q_proj_parametrizations_weight_original1, buf179, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf180 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf168, 2, int_array_8, int_array_9, 0L)); buf168.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_18, dequantize_affine_36, dequantize_affine_37, linear_18, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:448
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf180, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf178, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf179, 2, int_array_10, int_array_11, 0L))));
	    auto buf182 = std::move(buf179);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_39], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:45
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_3_self_attn_k_proj_parametrizations_weight_original1, buf182, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf183 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf178, 2, int_array_8, int_array_9, 0L)); buf178.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_19, dequantize_affine_38, dequantize_affine_39, linear_19], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:449
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf183, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf181, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf182, 2, int_array_10, int_array_11, 0L))));
	    auto buf185 = std::move(buf182);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_41], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:46
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_3_self_attn_v_proj_parametrizations_weight_original1, buf185, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf186 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf181, 2, int_array_8, int_array_9, 0L)); buf181.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_20, dequantize_affine_40, dequantize_affine_41, linear_20, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:450
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf186, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf184, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf185, 2, int_array_10, int_array_11, 0L))));
	    auto buf187 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf184, 4, int_array_12, int_array_13, 0L)); buf184.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_18, mul_116, view_9, transpose_12, contiguous_12, linear_19, view_10, transpose_13, contiguous_13, linear_20, view_11, transpose_14, contiguous_14, scaled_dot_product_attention_3], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:47
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf180, model_audio_tower_layers_3_self_attn_q_proj_bias, buf187, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf188 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf180, 4, int_array_12, int_array_13, 0L)); buf180.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_18, mul_116, view_9, transpose_12, contiguous_12, linear_19, view_10, transpose_13, contiguous_13, linear_20, view_11, transpose_14, contiguous_14, scaled_dot_product_attention_3], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:48
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf183, buf188, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf189 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf183, 4, int_array_12, int_array_13, 0L)); buf183.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_18, mul_116, view_9, transpose_12, contiguous_12, linear_19, view_10, transpose_13, contiguous_13, linear_20, view_11, transpose_14, contiguous_14, scaled_dot_product_attention_3], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:49
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf186, model_audio_tower_layers_3_self_attn_v_proj_bias, buf189, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf186.reset();
	    // Topologically Sorted Source Nodes: [, linear_18, mul_116, view_9, transpose_12, contiguous_12, linear_19, view_10, transpose_13, contiguous_13, linear_20, view_11, transpose_14, contiguous_14, scaled_dot_product_attention_3], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_3 = 1.0;
	    AtenTensorHandle buf191_handle;
	    AtenTensorHandle buf192_handle;
	    AtenTensorHandle buf193_handle;
	    AtenTensorHandle buf194_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:451
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf187, buf188, buf189, nullptr, 0, 0.0, 0, &var_3, &buf191_handle, &buf192_handle, &buf193_handle, &buf194_handle));
	    RAIIAtenTensorHandle buf191(buf191_handle);
	    RAIIAtenTensorHandle buf192(buf192_handle);
	    RAIIAtenTensorHandle buf193(buf193_handle);
	    RAIIAtenTensorHandle buf194(buf194_handle);
	    buf187.reset();
	    buf188.reset();
	
	    auto buf197 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf191, 3, int_array_4, int_array_5, 0L)); buf191.reset();  // reuse
	    auto buf198 = std::move(buf197);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_15, reshape_3, choose_qparams_affine_default_21, quantize_affine_21, dequantize_affine_42], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf198, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf199 = std::move(buf185);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_43], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:51
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_3_self_attn_out_proj_parametrizations_weight_original1, buf199, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf200 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf189, 2, int_array_8, int_array_9, 0L)); buf189.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_21, dequantize_affine_42, dequantize_affine_43, linear_21, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:452
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf200, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf198, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf199, 2, int_array_10, int_array_11, 0L))));
	    buf199.reset();
	    auto buf201 = std::move(buf148);  // reuse
	    auto buf205 = std::move(buf198);  // reuse
	    auto buf208 = std::move(buf205);  // reuse
	    auto buf209 = std::move(buf208);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_17, add_42, linear_21, add_51, layer_norm_7, choose_qparams_affine_default_22, quantize_affine_22, dequantize_affine_44], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf201, buf209, buf164, model_audio_tower_layers_2_fc2_bias, buf200, model_audio_tower_layers_3_self_attn_out_proj_bias, model_audio_tower_layers_3_final_layer_norm_weight, model_audio_tower_layers_3_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf164.reset();
	    buf200.reset();
	    auto buf210 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf163, 3, int_array_14, int_array_7, 0L)); buf163.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_45], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:53
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_3_fc1_parametrizations_weight_original0, model_audio_tower_layers_3_fc1_parametrizations_weight_original2, model_audio_tower_layers_3_fc1_parametrizations_weight_original1, buf210, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf211_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf211_handle));
	    RAIIAtenTensorHandle buf211(buf211_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_22, dequantize_affine_44, dequantize_affine_45, linear_22, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:453
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf211, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf209, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf210, 2, int_array_17, int_array_11, 0L))));
	    buf209.reset();
	    buf210.reset();
	    auto buf214 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf211, 3, int_array_18, int_array_19, 0L)); buf211.reset();  // reuse
	    auto buf215 = std::move(buf214);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_22, gelu_5, choose_qparams_affine_default_23, quantize_affine_23, dequantize_affine_46], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf215, model_audio_tower_layers_3_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf216_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf216_handle));
	    RAIIAtenTensorHandle buf216(buf216_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_47], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:55
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_3_fc2_parametrizations_weight_original0, model_audio_tower_layers_3_fc2_parametrizations_weight_original2, model_audio_tower_layers_3_fc2_parametrizations_weight_original1, buf216, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf217_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf217_handle));
	    RAIIAtenTensorHandle buf217(buf217_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_23, quantize_affine_23, dequantize_affine_46, dequantize_affine_47, linear_23, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:454
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf217, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf215, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf216, 2, int_array_22, int_array_23, 0L))));
	    buf215.reset();
	    AtenTensorHandle buf221_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf221_handle));
	    RAIIAtenTensorHandle buf221(buf221_handle);
	    AtenTensorHandle buf224_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf224_handle));
	    RAIIAtenTensorHandle buf224(buf224_handle);
	    auto buf231 = std::move(buf224);  // reuse
	    AtenTensorHandle buf227_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf227_handle));
	    RAIIAtenTensorHandle buf227(buf227_handle);
	    auto buf234 = std::move(buf227);  // reuse
	    AtenTensorHandle buf230_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf230_handle));
	    RAIIAtenTensorHandle buf230(buf230_handle);
	    auto buf237 = std::move(buf230);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_23, add_56, layer_norm_8, choose_qparams_affine_default_24, quantize_affine_24, dequantize_affine_48, choose_qparams_affine_default_25, quantize_affine_25, dequantize_affine_50, choose_qparams_affine_default_26, quantize_affine_26, dequantize_affine_52], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf231, buf234, buf237, buf201, buf217, model_audio_tower_layers_3_fc2_bias, model_audio_tower_layers_4_self_attn_layer_norm_weight, model_audio_tower_layers_4_self_attn_layer_norm_bias, buf221, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf232_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf232_handle));
	    RAIIAtenTensorHandle buf232(buf232_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_49], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:57
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_4_self_attn_q_proj_parametrizations_weight_original1, buf232, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf233 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf221, 2, int_array_8, int_array_9, 0L)); buf221.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_24, dequantize_affine_48, dequantize_affine_49, linear_24, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:455
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf233, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf231, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf232, 2, int_array_10, int_array_11, 0L))));
	    auto buf235 = std::move(buf232);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_51], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:58
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_4_self_attn_k_proj_parametrizations_weight_original1, buf235, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf236 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf231, 2, int_array_8, int_array_9, 0L)); buf231.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_25, dequantize_affine_50, dequantize_affine_51, linear_25], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:456
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf236, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf234, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf235, 2, int_array_10, int_array_11, 0L))));
	    auto buf238 = std::move(buf235);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_53], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:59
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_4_self_attn_v_proj_parametrizations_weight_original1, buf238, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf239 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf234, 2, int_array_8, int_array_9, 0L)); buf234.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_26, dequantize_affine_52, dequantize_affine_53, linear_26, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:457
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf239, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf237, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf238, 2, int_array_10, int_array_11, 0L))));
	    auto buf240 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf237, 4, int_array_12, int_array_13, 0L)); buf237.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_24, mul_153, view_12, transpose_16, contiguous_16, linear_25, view_13, transpose_17, contiguous_17, linear_26, view_14, transpose_18, contiguous_18, scaled_dot_product_attention_4], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:60
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf233, model_audio_tower_layers_4_self_attn_q_proj_bias, buf240, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf241 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf233, 4, int_array_12, int_array_13, 0L)); buf233.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_24, mul_153, view_12, transpose_16, contiguous_16, linear_25, view_13, transpose_17, contiguous_17, linear_26, view_14, transpose_18, contiguous_18, scaled_dot_product_attention_4], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:61
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf236, buf241, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf242 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf236, 4, int_array_12, int_array_13, 0L)); buf236.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_24, mul_153, view_12, transpose_16, contiguous_16, linear_25, view_13, transpose_17, contiguous_17, linear_26, view_14, transpose_18, contiguous_18, scaled_dot_product_attention_4], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:62
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf239, model_audio_tower_layers_4_self_attn_v_proj_bias, buf242, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf239.reset();
	    // Topologically Sorted Source Nodes: [, linear_24, mul_153, view_12, transpose_16, contiguous_16, linear_25, view_13, transpose_17, contiguous_17, linear_26, view_14, transpose_18, contiguous_18, scaled_dot_product_attention_4], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_4 = 1.0;
	    AtenTensorHandle buf244_handle;
	    AtenTensorHandle buf245_handle;
	    AtenTensorHandle buf246_handle;
	    AtenTensorHandle buf247_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:458
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf240, buf241, buf242, nullptr, 0, 0.0, 0, &var_4, &buf244_handle, &buf245_handle, &buf246_handle, &buf247_handle));
	    RAIIAtenTensorHandle buf244(buf244_handle);
	    RAIIAtenTensorHandle buf245(buf245_handle);
	    RAIIAtenTensorHandle buf246(buf246_handle);
	    RAIIAtenTensorHandle buf247(buf247_handle);
	    buf240.reset();
	    buf241.reset();
	
	    auto buf250 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf244, 3, int_array_4, int_array_5, 0L)); buf244.reset();  // reuse
	    auto buf251 = std::move(buf250);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_19, reshape_4, choose_qparams_affine_default_27, quantize_affine_27, dequantize_affine_54], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf251, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf252 = std::move(buf238);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_55], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:64
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_4_self_attn_out_proj_parametrizations_weight_original1, buf252, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf253 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf242, 2, int_array_8, int_array_9, 0L)); buf242.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_27, dequantize_affine_54, dequantize_affine_55, linear_27, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:459
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf253, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf251, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf252, 2, int_array_10, int_array_11, 0L))));
	    buf252.reset();
	    auto buf254 = std::move(buf201);  // reuse
	    auto buf258 = std::move(buf251);  // reuse
	    auto buf261 = std::move(buf258);  // reuse
	    auto buf262 = std::move(buf261);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_23, add_56, linear_27, add_65, layer_norm_9, choose_qparams_affine_default_28, quantize_affine_28, dequantize_affine_56], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf254, buf262, buf217, model_audio_tower_layers_3_fc2_bias, buf253, model_audio_tower_layers_4_self_attn_out_proj_bias, model_audio_tower_layers_4_final_layer_norm_weight, model_audio_tower_layers_4_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf217.reset();
	    buf253.reset();
	    auto buf263 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf216, 3, int_array_14, int_array_7, 0L)); buf216.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_57], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:66
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_4_fc1_parametrizations_weight_original0, model_audio_tower_layers_4_fc1_parametrizations_weight_original2, model_audio_tower_layers_4_fc1_parametrizations_weight_original1, buf263, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf264_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf264_handle));
	    RAIIAtenTensorHandle buf264(buf264_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_28, dequantize_affine_56, dequantize_affine_57, linear_28, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:460
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf264, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf262, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf263, 2, int_array_17, int_array_11, 0L))));
	    buf262.reset();
	    buf263.reset();
	    auto buf267 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf264, 3, int_array_18, int_array_19, 0L)); buf264.reset();  // reuse
	    auto buf268 = std::move(buf267);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_28, gelu_6, choose_qparams_affine_default_29, quantize_affine_29, dequantize_affine_58], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf268, model_audio_tower_layers_4_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf269_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf269_handle));
	    RAIIAtenTensorHandle buf269(buf269_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_59], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:68
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_4_fc2_parametrizations_weight_original0, model_audio_tower_layers_4_fc2_parametrizations_weight_original2, model_audio_tower_layers_4_fc2_parametrizations_weight_original1, buf269, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf270_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf270_handle));
	    RAIIAtenTensorHandle buf270(buf270_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_29, quantize_affine_29, dequantize_affine_58, dequantize_affine_59, linear_29, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:461
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf270, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf268, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf269, 2, int_array_22, int_array_23, 0L))));
	    buf268.reset();
	    AtenTensorHandle buf274_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf274_handle));
	    RAIIAtenTensorHandle buf274(buf274_handle);
	    AtenTensorHandle buf277_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf277_handle));
	    RAIIAtenTensorHandle buf277(buf277_handle);
	    auto buf284 = std::move(buf277);  // reuse
	    AtenTensorHandle buf280_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf280_handle));
	    RAIIAtenTensorHandle buf280(buf280_handle);
	    auto buf287 = std::move(buf280);  // reuse
	    AtenTensorHandle buf283_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf283_handle));
	    RAIIAtenTensorHandle buf283(buf283_handle);
	    auto buf290 = std::move(buf283);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_29, add_70, layer_norm_10, choose_qparams_affine_default_30, quantize_affine_30, dequantize_affine_60, choose_qparams_affine_default_31, quantize_affine_31, dequantize_affine_62, choose_qparams_affine_default_32, quantize_affine_32, dequantize_affine_64], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf284, buf287, buf290, buf254, buf270, model_audio_tower_layers_4_fc2_bias, model_audio_tower_layers_5_self_attn_layer_norm_weight, model_audio_tower_layers_5_self_attn_layer_norm_bias, buf274, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf285_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf285_handle));
	    RAIIAtenTensorHandle buf285(buf285_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_61], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:70
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_5_self_attn_q_proj_parametrizations_weight_original1, buf285, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf286 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf274, 2, int_array_8, int_array_9, 0L)); buf274.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_30, dequantize_affine_60, dequantize_affine_61, linear_30, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:462
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf286, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf284, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf285, 2, int_array_10, int_array_11, 0L))));
	    auto buf288 = std::move(buf285);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_63], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:71
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_5_self_attn_k_proj_parametrizations_weight_original1, buf288, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf289 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf284, 2, int_array_8, int_array_9, 0L)); buf284.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_31, dequantize_affine_62, dequantize_affine_63, linear_31], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:463
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf289, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf287, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf288, 2, int_array_10, int_array_11, 0L))));
	    auto buf291 = std::move(buf288);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_65], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:72
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_5_self_attn_v_proj_parametrizations_weight_original1, buf291, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf292 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf287, 2, int_array_8, int_array_9, 0L)); buf287.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_32, dequantize_affine_64, dequantize_affine_65, linear_32, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:464
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf292, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf290, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf291, 2, int_array_10, int_array_11, 0L))));
	    auto buf293 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf290, 4, int_array_12, int_array_13, 0L)); buf290.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_30, mul_190, view_15, transpose_20, contiguous_20, linear_31, view_16, transpose_21, contiguous_21, linear_32, view_17, transpose_22, contiguous_22, scaled_dot_product_attention_5], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:73
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf286, model_audio_tower_layers_5_self_attn_q_proj_bias, buf293, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf294 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf286, 4, int_array_12, int_array_13, 0L)); buf286.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_30, mul_190, view_15, transpose_20, contiguous_20, linear_31, view_16, transpose_21, contiguous_21, linear_32, view_17, transpose_22, contiguous_22, scaled_dot_product_attention_5], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:74
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf289, buf294, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf295 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf289, 4, int_array_12, int_array_13, 0L)); buf289.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_30, mul_190, view_15, transpose_20, contiguous_20, linear_31, view_16, transpose_21, contiguous_21, linear_32, view_17, transpose_22, contiguous_22, scaled_dot_product_attention_5], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:75
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf292, model_audio_tower_layers_5_self_attn_v_proj_bias, buf295, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf292.reset();
	    // Topologically Sorted Source Nodes: [, linear_30, mul_190, view_15, transpose_20, contiguous_20, linear_31, view_16, transpose_21, contiguous_21, linear_32, view_17, transpose_22, contiguous_22, scaled_dot_product_attention_5], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_5 = 1.0;
	    AtenTensorHandle buf297_handle;
	    AtenTensorHandle buf298_handle;
	    AtenTensorHandle buf299_handle;
	    AtenTensorHandle buf300_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:465
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf293, buf294, buf295, nullptr, 0, 0.0, 0, &var_5, &buf297_handle, &buf298_handle, &buf299_handle, &buf300_handle));
	    RAIIAtenTensorHandle buf297(buf297_handle);
	    RAIIAtenTensorHandle buf298(buf298_handle);
	    RAIIAtenTensorHandle buf299(buf299_handle);
	    RAIIAtenTensorHandle buf300(buf300_handle);
	    buf293.reset();
	    buf294.reset();
	
	    auto buf303 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf297, 3, int_array_4, int_array_5, 0L)); buf297.reset();  // reuse
	    auto buf304 = std::move(buf303);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_23, reshape_5, choose_qparams_affine_default_33, quantize_affine_33, dequantize_affine_66], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf304, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf305 = std::move(buf291);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_67], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:77
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_5_self_attn_out_proj_parametrizations_weight_original1, buf305, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf306 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf295, 2, int_array_8, int_array_9, 0L)); buf295.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_33, dequantize_affine_66, dequantize_affine_67, linear_33, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:466
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf306, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf304, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf305, 2, int_array_10, int_array_11, 0L))));
	    buf305.reset();
	    auto buf307 = std::move(buf254);  // reuse
	    auto buf311 = std::move(buf304);  // reuse
	    auto buf314 = std::move(buf311);  // reuse
	    auto buf315 = std::move(buf314);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_29, add_70, linear_33, add_79, layer_norm_11, choose_qparams_affine_default_34, quantize_affine_34, dequantize_affine_68], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf307, buf315, buf270, model_audio_tower_layers_4_fc2_bias, buf306, model_audio_tower_layers_5_self_attn_out_proj_bias, model_audio_tower_layers_5_final_layer_norm_weight, model_audio_tower_layers_5_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf270.reset();
	    buf306.reset();
	    auto buf316 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf269, 3, int_array_14, int_array_7, 0L)); buf269.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_69], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:79
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_5_fc1_parametrizations_weight_original0, model_audio_tower_layers_5_fc1_parametrizations_weight_original2, model_audio_tower_layers_5_fc1_parametrizations_weight_original1, buf316, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf317_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf317_handle));
	    RAIIAtenTensorHandle buf317(buf317_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_34, dequantize_affine_68, dequantize_affine_69, linear_34, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:467
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf317, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf315, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf316, 2, int_array_17, int_array_11, 0L))));
	    buf315.reset();
	    buf316.reset();
	    auto buf320 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf317, 3, int_array_18, int_array_19, 0L)); buf317.reset();  // reuse
	    auto buf321 = std::move(buf320);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_34, gelu_7, choose_qparams_affine_default_35, quantize_affine_35, dequantize_affine_70], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf321, model_audio_tower_layers_5_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf322_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf322_handle));
	    RAIIAtenTensorHandle buf322(buf322_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_71], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:81
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_5_fc2_parametrizations_weight_original0, model_audio_tower_layers_5_fc2_parametrizations_weight_original2, model_audio_tower_layers_5_fc2_parametrizations_weight_original1, buf322, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf323_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf323_handle));
	    RAIIAtenTensorHandle buf323(buf323_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_35, quantize_affine_35, dequantize_affine_70, dequantize_affine_71, linear_35, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:468
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf323, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf321, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf322, 2, int_array_22, int_array_23, 0L))));
	    buf321.reset();
	    AtenTensorHandle buf327_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf327_handle));
	    RAIIAtenTensorHandle buf327(buf327_handle);
	    AtenTensorHandle buf330_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf330_handle));
	    RAIIAtenTensorHandle buf330(buf330_handle);
	    auto buf337 = std::move(buf330);  // reuse
	    AtenTensorHandle buf333_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf333_handle));
	    RAIIAtenTensorHandle buf333(buf333_handle);
	    auto buf340 = std::move(buf333);  // reuse
	    AtenTensorHandle buf336_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf336_handle));
	    RAIIAtenTensorHandle buf336(buf336_handle);
	    auto buf343 = std::move(buf336);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_35, add_84, layer_norm_12, choose_qparams_affine_default_36, quantize_affine_36, dequantize_affine_72, choose_qparams_affine_default_37, quantize_affine_37, dequantize_affine_74, choose_qparams_affine_default_38, quantize_affine_38, dequantize_affine_76], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf337, buf340, buf343, buf307, buf323, model_audio_tower_layers_5_fc2_bias, model_audio_tower_layers_6_self_attn_layer_norm_weight, model_audio_tower_layers_6_self_attn_layer_norm_bias, buf327, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf338_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf338_handle));
	    RAIIAtenTensorHandle buf338(buf338_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_73], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:83
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_6_self_attn_q_proj_parametrizations_weight_original1, buf338, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf339 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf327, 2, int_array_8, int_array_9, 0L)); buf327.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_36, dequantize_affine_72, dequantize_affine_73, linear_36, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:469
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf339, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf337, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf338, 2, int_array_10, int_array_11, 0L))));
	    auto buf341 = std::move(buf338);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_75], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:84
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_6_self_attn_k_proj_parametrizations_weight_original1, buf341, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf342 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf337, 2, int_array_8, int_array_9, 0L)); buf337.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_37, dequantize_affine_74, dequantize_affine_75, linear_37], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:470
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf342, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf340, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf341, 2, int_array_10, int_array_11, 0L))));
	    auto buf344 = std::move(buf341);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_77], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:85
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_6_self_attn_v_proj_parametrizations_weight_original1, buf344, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf345 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf340, 2, int_array_8, int_array_9, 0L)); buf340.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_38, dequantize_affine_76, dequantize_affine_77, linear_38, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:471
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf345, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf343, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf344, 2, int_array_10, int_array_11, 0L))));
	    auto buf346 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf343, 4, int_array_12, int_array_13, 0L)); buf343.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_36, mul_227, view_18, transpose_24, contiguous_24, linear_37, view_19, transpose_25, contiguous_25, linear_38, view_20, transpose_26, contiguous_26, scaled_dot_product_attention_6], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:86
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf339, model_audio_tower_layers_6_self_attn_q_proj_bias, buf346, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf347 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf339, 4, int_array_12, int_array_13, 0L)); buf339.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_36, mul_227, view_18, transpose_24, contiguous_24, linear_37, view_19, transpose_25, contiguous_25, linear_38, view_20, transpose_26, contiguous_26, scaled_dot_product_attention_6], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:87
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf342, buf347, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf348 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf342, 4, int_array_12, int_array_13, 0L)); buf342.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_36, mul_227, view_18, transpose_24, contiguous_24, linear_37, view_19, transpose_25, contiguous_25, linear_38, view_20, transpose_26, contiguous_26, scaled_dot_product_attention_6], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:88
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf345, model_audio_tower_layers_6_self_attn_v_proj_bias, buf348, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf345.reset();
	    // Topologically Sorted Source Nodes: [, linear_36, mul_227, view_18, transpose_24, contiguous_24, linear_37, view_19, transpose_25, contiguous_25, linear_38, view_20, transpose_26, contiguous_26, scaled_dot_product_attention_6], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_6 = 1.0;
	    AtenTensorHandle buf350_handle;
	    AtenTensorHandle buf351_handle;
	    AtenTensorHandle buf352_handle;
	    AtenTensorHandle buf353_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:472
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf346, buf347, buf348, nullptr, 0, 0.0, 0, &var_6, &buf350_handle, &buf351_handle, &buf352_handle, &buf353_handle));
	    RAIIAtenTensorHandle buf350(buf350_handle);
	    RAIIAtenTensorHandle buf351(buf351_handle);
	    RAIIAtenTensorHandle buf352(buf352_handle);
	    RAIIAtenTensorHandle buf353(buf353_handle);
	    buf346.reset();
	    buf347.reset();
	
	    auto buf356 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf350, 3, int_array_4, int_array_5, 0L)); buf350.reset();  // reuse
	    auto buf357 = std::move(buf356);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_27, reshape_6, choose_qparams_affine_default_39, quantize_affine_39, dequantize_affine_78], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf357, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf358 = std::move(buf344);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_79], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:90
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_6_self_attn_out_proj_parametrizations_weight_original1, buf358, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf359 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf348, 2, int_array_8, int_array_9, 0L)); buf348.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_39, dequantize_affine_78, dequantize_affine_79, linear_39, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:473
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf359, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf357, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf358, 2, int_array_10, int_array_11, 0L))));
	    buf358.reset();
	    auto buf360 = std::move(buf307);  // reuse
	    auto buf364 = std::move(buf357);  // reuse
	    auto buf367 = std::move(buf364);  // reuse
	    auto buf368 = std::move(buf367);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_35, add_84, linear_39, add_93, layer_norm_13, choose_qparams_affine_default_40, quantize_affine_40, dequantize_affine_80], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf360, buf368, buf323, model_audio_tower_layers_5_fc2_bias, buf359, model_audio_tower_layers_6_self_attn_out_proj_bias, model_audio_tower_layers_6_final_layer_norm_weight, model_audio_tower_layers_6_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf323.reset();
	    buf359.reset();
	    auto buf369 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf322, 3, int_array_14, int_array_7, 0L)); buf322.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_81], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:92
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_6_fc1_parametrizations_weight_original0, model_audio_tower_layers_6_fc1_parametrizations_weight_original2, model_audio_tower_layers_6_fc1_parametrizations_weight_original1, buf369, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf370_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf370_handle));
	    RAIIAtenTensorHandle buf370(buf370_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_40, dequantize_affine_80, dequantize_affine_81, linear_40, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:474
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf370, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf368, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf369, 2, int_array_17, int_array_11, 0L))));
	    buf368.reset();
	    buf369.reset();
	    auto buf373 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf370, 3, int_array_18, int_array_19, 0L)); buf370.reset();  // reuse
	    auto buf374 = std::move(buf373);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_40, gelu_8, choose_qparams_affine_default_41, quantize_affine_41, dequantize_affine_82], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf374, model_audio_tower_layers_6_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf375_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf375_handle));
	    RAIIAtenTensorHandle buf375(buf375_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_83], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:94
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_6_fc2_parametrizations_weight_original0, model_audio_tower_layers_6_fc2_parametrizations_weight_original2, model_audio_tower_layers_6_fc2_parametrizations_weight_original1, buf375, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf376_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf376_handle));
	    RAIIAtenTensorHandle buf376(buf376_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_41, quantize_affine_41, dequantize_affine_82, dequantize_affine_83, linear_41, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:475
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf376, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf374, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf375, 2, int_array_22, int_array_23, 0L))));
	    buf374.reset();
	    AtenTensorHandle buf380_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf380_handle));
	    RAIIAtenTensorHandle buf380(buf380_handle);
	    AtenTensorHandle buf383_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf383_handle));
	    RAIIAtenTensorHandle buf383(buf383_handle);
	    auto buf390 = std::move(buf383);  // reuse
	    AtenTensorHandle buf386_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf386_handle));
	    RAIIAtenTensorHandle buf386(buf386_handle);
	    auto buf393 = std::move(buf386);  // reuse
	    AtenTensorHandle buf389_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf389_handle));
	    RAIIAtenTensorHandle buf389(buf389_handle);
	    auto buf396 = std::move(buf389);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_41, add_98, layer_norm_14, choose_qparams_affine_default_42, quantize_affine_42, dequantize_affine_84, choose_qparams_affine_default_43, quantize_affine_43, dequantize_affine_86, choose_qparams_affine_default_44, quantize_affine_44, dequantize_affine_88], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf390, buf393, buf396, buf360, buf376, model_audio_tower_layers_6_fc2_bias, model_audio_tower_layers_7_self_attn_layer_norm_weight, model_audio_tower_layers_7_self_attn_layer_norm_bias, buf380, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf391_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf391_handle));
	    RAIIAtenTensorHandle buf391(buf391_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_85], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:96
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_7_self_attn_q_proj_parametrizations_weight_original1, buf391, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf392 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf380, 2, int_array_8, int_array_9, 0L)); buf380.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_42, dequantize_affine_84, dequantize_affine_85, linear_42, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:476
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf392, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf390, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf391, 2, int_array_10, int_array_11, 0L))));
	    auto buf394 = std::move(buf391);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_87], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:97
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_7_self_attn_k_proj_parametrizations_weight_original1, buf394, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf395 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf390, 2, int_array_8, int_array_9, 0L)); buf390.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_43, dequantize_affine_86, dequantize_affine_87, linear_43], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:477
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf395, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf393, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf394, 2, int_array_10, int_array_11, 0L))));
	    auto buf397 = std::move(buf394);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_89], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:98
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_7_self_attn_v_proj_parametrizations_weight_original1, buf397, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf398 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf393, 2, int_array_8, int_array_9, 0L)); buf393.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_44, dequantize_affine_88, dequantize_affine_89, linear_44, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:478
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf398, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf396, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf397, 2, int_array_10, int_array_11, 0L))));
	    auto buf399 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf396, 4, int_array_12, int_array_13, 0L)); buf396.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_42, mul_264, view_21, transpose_28, contiguous_28, linear_43, view_22, transpose_29, contiguous_29, linear_44, view_23, transpose_30, contiguous_30, scaled_dot_product_attention_7], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:99
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf392, model_audio_tower_layers_7_self_attn_q_proj_bias, buf399, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf400 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf392, 4, int_array_12, int_array_13, 0L)); buf392.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_42, mul_264, view_21, transpose_28, contiguous_28, linear_43, view_22, transpose_29, contiguous_29, linear_44, view_23, transpose_30, contiguous_30, scaled_dot_product_attention_7], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:100
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf395, buf400, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf401 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf395, 4, int_array_12, int_array_13, 0L)); buf395.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_42, mul_264, view_21, transpose_28, contiguous_28, linear_43, view_22, transpose_29, contiguous_29, linear_44, view_23, transpose_30, contiguous_30, scaled_dot_product_attention_7], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:101
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf398, model_audio_tower_layers_7_self_attn_v_proj_bias, buf401, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf398.reset();
	    // Topologically Sorted Source Nodes: [, linear_42, mul_264, view_21, transpose_28, contiguous_28, linear_43, view_22, transpose_29, contiguous_29, linear_44, view_23, transpose_30, contiguous_30, scaled_dot_product_attention_7], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_7 = 1.0;
	    AtenTensorHandle buf403_handle;
	    AtenTensorHandle buf404_handle;
	    AtenTensorHandle buf405_handle;
	    AtenTensorHandle buf406_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:479
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf399, buf400, buf401, nullptr, 0, 0.0, 0, &var_7, &buf403_handle, &buf404_handle, &buf405_handle, &buf406_handle));
	    RAIIAtenTensorHandle buf403(buf403_handle);
	    RAIIAtenTensorHandle buf404(buf404_handle);
	    RAIIAtenTensorHandle buf405(buf405_handle);
	    RAIIAtenTensorHandle buf406(buf406_handle);
	    buf399.reset();
	    buf400.reset();
	
	    auto buf409 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf403, 3, int_array_4, int_array_5, 0L)); buf403.reset();  // reuse
	    auto buf410 = std::move(buf409);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_31, reshape_7, choose_qparams_affine_default_45, quantize_affine_45, dequantize_affine_90], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf410, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf411 = std::move(buf397);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_91], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:103
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_7_self_attn_out_proj_parametrizations_weight_original1, buf411, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf412 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf401, 2, int_array_8, int_array_9, 0L)); buf401.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_45, dequantize_affine_90, dequantize_affine_91, linear_45, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:480
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf412, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf410, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf411, 2, int_array_10, int_array_11, 0L))));
	    buf411.reset();
	    auto buf413 = std::move(buf360);  // reuse
	    auto buf417 = std::move(buf410);  // reuse
	    auto buf420 = std::move(buf417);  // reuse
	    auto buf421 = std::move(buf420);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_41, add_98, linear_45, add_107, layer_norm_15, choose_qparams_affine_default_46, quantize_affine_46, dequantize_affine_92], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf413, buf421, buf376, model_audio_tower_layers_6_fc2_bias, buf412, model_audio_tower_layers_7_self_attn_out_proj_bias, model_audio_tower_layers_7_final_layer_norm_weight, model_audio_tower_layers_7_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf376.reset();
	    buf412.reset();
	    auto buf422 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf375, 3, int_array_14, int_array_7, 0L)); buf375.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_93], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:105
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_7_fc1_parametrizations_weight_original0, model_audio_tower_layers_7_fc1_parametrizations_weight_original2, model_audio_tower_layers_7_fc1_parametrizations_weight_original1, buf422, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf423_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf423_handle));
	    RAIIAtenTensorHandle buf423(buf423_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_46, dequantize_affine_92, dequantize_affine_93, linear_46, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:481
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf423, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf421, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf422, 2, int_array_17, int_array_11, 0L))));
	    buf421.reset();
	    buf422.reset();
	    auto buf426 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf423, 3, int_array_18, int_array_19, 0L)); buf423.reset();  // reuse
	    auto buf427 = std::move(buf426);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_46, gelu_9, choose_qparams_affine_default_47, quantize_affine_47, dequantize_affine_94], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf427, model_audio_tower_layers_7_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf428_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf428_handle));
	    RAIIAtenTensorHandle buf428(buf428_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_95], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:107
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_7_fc2_parametrizations_weight_original0, model_audio_tower_layers_7_fc2_parametrizations_weight_original2, model_audio_tower_layers_7_fc2_parametrizations_weight_original1, buf428, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf429_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf429_handle));
	    RAIIAtenTensorHandle buf429(buf429_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_47, quantize_affine_47, dequantize_affine_94, dequantize_affine_95, linear_47, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:482
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf429, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf427, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf428, 2, int_array_22, int_array_23, 0L))));
	    buf427.reset();
	    AtenTensorHandle buf433_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf433_handle));
	    RAIIAtenTensorHandle buf433(buf433_handle);
	    AtenTensorHandle buf436_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf436_handle));
	    RAIIAtenTensorHandle buf436(buf436_handle);
	    auto buf443 = std::move(buf436);  // reuse
	    AtenTensorHandle buf439_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf439_handle));
	    RAIIAtenTensorHandle buf439(buf439_handle);
	    auto buf446 = std::move(buf439);  // reuse
	    AtenTensorHandle buf442_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf442_handle));
	    RAIIAtenTensorHandle buf442(buf442_handle);
	    auto buf449 = std::move(buf442);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_47, add_112, layer_norm_16, choose_qparams_affine_default_48, quantize_affine_48, dequantize_affine_96, choose_qparams_affine_default_49, quantize_affine_49, dequantize_affine_98, choose_qparams_affine_default_50, quantize_affine_50, dequantize_affine_100], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf443, buf446, buf449, buf413, buf429, model_audio_tower_layers_7_fc2_bias, model_audio_tower_layers_8_self_attn_layer_norm_weight, model_audio_tower_layers_8_self_attn_layer_norm_bias, buf433, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf444_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf444_handle));
	    RAIIAtenTensorHandle buf444(buf444_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_97], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:109
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_8_self_attn_q_proj_parametrizations_weight_original1, buf444, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf445 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf433, 2, int_array_8, int_array_9, 0L)); buf433.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_48, dequantize_affine_96, dequantize_affine_97, linear_48, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:483
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf445, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf443, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf444, 2, int_array_10, int_array_11, 0L))));
	    auto buf447 = std::move(buf444);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_99], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:110
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_8_self_attn_k_proj_parametrizations_weight_original1, buf447, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf448 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf443, 2, int_array_8, int_array_9, 0L)); buf443.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_49, dequantize_affine_98, dequantize_affine_99, linear_49], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:484
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf448, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf446, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf447, 2, int_array_10, int_array_11, 0L))));
	    auto buf450 = std::move(buf447);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_101], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:111
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_8_self_attn_v_proj_parametrizations_weight_original1, buf450, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf451 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf446, 2, int_array_8, int_array_9, 0L)); buf446.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_50, dequantize_affine_100, dequantize_affine_101, linear_50, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:485
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf451, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf449, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf450, 2, int_array_10, int_array_11, 0L))));
	    auto buf452 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf449, 4, int_array_12, int_array_13, 0L)); buf449.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_48, mul_301, view_24, transpose_32, contiguous_32, linear_49, view_25, transpose_33, contiguous_33, linear_50, view_26, transpose_34, contiguous_34, scaled_dot_product_attention_8], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:112
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf445, model_audio_tower_layers_8_self_attn_q_proj_bias, buf452, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf453 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf445, 4, int_array_12, int_array_13, 0L)); buf445.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_48, mul_301, view_24, transpose_32, contiguous_32, linear_49, view_25, transpose_33, contiguous_33, linear_50, view_26, transpose_34, contiguous_34, scaled_dot_product_attention_8], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:113
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf448, buf453, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf454 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf448, 4, int_array_12, int_array_13, 0L)); buf448.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_48, mul_301, view_24, transpose_32, contiguous_32, linear_49, view_25, transpose_33, contiguous_33, linear_50, view_26, transpose_34, contiguous_34, scaled_dot_product_attention_8], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:114
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf451, model_audio_tower_layers_8_self_attn_v_proj_bias, buf454, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf451.reset();
	    // Topologically Sorted Source Nodes: [, linear_48, mul_301, view_24, transpose_32, contiguous_32, linear_49, view_25, transpose_33, contiguous_33, linear_50, view_26, transpose_34, contiguous_34, scaled_dot_product_attention_8], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_8 = 1.0;
	    AtenTensorHandle buf456_handle;
	    AtenTensorHandle buf457_handle;
	    AtenTensorHandle buf458_handle;
	    AtenTensorHandle buf459_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:486
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf452, buf453, buf454, nullptr, 0, 0.0, 0, &var_8, &buf456_handle, &buf457_handle, &buf458_handle, &buf459_handle));
	    RAIIAtenTensorHandle buf456(buf456_handle);
	    RAIIAtenTensorHandle buf457(buf457_handle);
	    RAIIAtenTensorHandle buf458(buf458_handle);
	    RAIIAtenTensorHandle buf459(buf459_handle);
	    buf452.reset();
	    buf453.reset();
	
	    auto buf462 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf456, 3, int_array_4, int_array_5, 0L)); buf456.reset();  // reuse
	    auto buf463 = std::move(buf462);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_35, reshape_8, choose_qparams_affine_default_51, quantize_affine_51, dequantize_affine_102], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf463, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf464 = std::move(buf450);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_103], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:116
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_8_self_attn_out_proj_parametrizations_weight_original1, buf464, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf465 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf454, 2, int_array_8, int_array_9, 0L)); buf454.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_51, dequantize_affine_102, dequantize_affine_103, linear_51, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:487
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf465, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf463, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf464, 2, int_array_10, int_array_11, 0L))));
	    buf464.reset();
	    auto buf466 = std::move(buf413);  // reuse
	    auto buf470 = std::move(buf463);  // reuse
	    auto buf473 = std::move(buf470);  // reuse
	    auto buf474 = std::move(buf473);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_47, add_112, linear_51, add_121, layer_norm_17, choose_qparams_affine_default_52, quantize_affine_52, dequantize_affine_104], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf466, buf474, buf429, model_audio_tower_layers_7_fc2_bias, buf465, model_audio_tower_layers_8_self_attn_out_proj_bias, model_audio_tower_layers_8_final_layer_norm_weight, model_audio_tower_layers_8_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf429.reset();
	    buf465.reset();
	    auto buf475 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf428, 3, int_array_14, int_array_7, 0L)); buf428.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_105], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:118
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_8_fc1_parametrizations_weight_original0, model_audio_tower_layers_8_fc1_parametrizations_weight_original2, model_audio_tower_layers_8_fc1_parametrizations_weight_original1, buf475, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf476_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf476_handle));
	    RAIIAtenTensorHandle buf476(buf476_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_52, dequantize_affine_104, dequantize_affine_105, linear_52, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:488
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf476, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf474, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf475, 2, int_array_17, int_array_11, 0L))));
	    buf474.reset();
	    buf475.reset();
	    auto buf479 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf476, 3, int_array_18, int_array_19, 0L)); buf476.reset();  // reuse
	    auto buf480 = std::move(buf479);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_52, gelu_10, choose_qparams_affine_default_53, quantize_affine_53, dequantize_affine_106], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf480, model_audio_tower_layers_8_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf481_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf481_handle));
	    RAIIAtenTensorHandle buf481(buf481_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_107], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:120
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_8_fc2_parametrizations_weight_original0, model_audio_tower_layers_8_fc2_parametrizations_weight_original2, model_audio_tower_layers_8_fc2_parametrizations_weight_original1, buf481, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf482_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf482_handle));
	    RAIIAtenTensorHandle buf482(buf482_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_53, quantize_affine_53, dequantize_affine_106, dequantize_affine_107, linear_53, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:489
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf482, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf480, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf481, 2, int_array_22, int_array_23, 0L))));
	    buf480.reset();
	    AtenTensorHandle buf486_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf486_handle));
	    RAIIAtenTensorHandle buf486(buf486_handle);
	    AtenTensorHandle buf489_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf489_handle));
	    RAIIAtenTensorHandle buf489(buf489_handle);
	    auto buf496 = std::move(buf489);  // reuse
	    AtenTensorHandle buf492_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf492_handle));
	    RAIIAtenTensorHandle buf492(buf492_handle);
	    auto buf499 = std::move(buf492);  // reuse
	    AtenTensorHandle buf495_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf495_handle));
	    RAIIAtenTensorHandle buf495(buf495_handle);
	    auto buf502 = std::move(buf495);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_53, add_126, layer_norm_18, choose_qparams_affine_default_54, quantize_affine_54, dequantize_affine_108, choose_qparams_affine_default_55, quantize_affine_55, dequantize_affine_110, choose_qparams_affine_default_56, quantize_affine_56, dequantize_affine_112], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf496, buf499, buf502, buf466, buf482, model_audio_tower_layers_8_fc2_bias, model_audio_tower_layers_9_self_attn_layer_norm_weight, model_audio_tower_layers_9_self_attn_layer_norm_bias, buf486, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf497_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf497_handle));
	    RAIIAtenTensorHandle buf497(buf497_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_109], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:122
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_9_self_attn_q_proj_parametrizations_weight_original1, buf497, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf498 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf486, 2, int_array_8, int_array_9, 0L)); buf486.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_54, dequantize_affine_108, dequantize_affine_109, linear_54, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:490
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf498, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf496, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf497, 2, int_array_10, int_array_11, 0L))));
	    auto buf500 = std::move(buf497);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_111], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:123
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_9_self_attn_k_proj_parametrizations_weight_original1, buf500, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf501 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf496, 2, int_array_8, int_array_9, 0L)); buf496.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_55, dequantize_affine_110, dequantize_affine_111, linear_55], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:491
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf501, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf499, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf500, 2, int_array_10, int_array_11, 0L))));
	    auto buf503 = std::move(buf500);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_113], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:124
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_9_self_attn_v_proj_parametrizations_weight_original1, buf503, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf504 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf499, 2, int_array_8, int_array_9, 0L)); buf499.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_56, dequantize_affine_112, dequantize_affine_113, linear_56, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:492
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf504, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf502, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf503, 2, int_array_10, int_array_11, 0L))));
	    auto buf505 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf502, 4, int_array_12, int_array_13, 0L)); buf502.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_54, mul_338, view_27, transpose_36, contiguous_36, linear_55, view_28, transpose_37, contiguous_37, linear_56, view_29, transpose_38, contiguous_38, scaled_dot_product_attention_9], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:125
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf498, model_audio_tower_layers_9_self_attn_q_proj_bias, buf505, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf506 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf498, 4, int_array_12, int_array_13, 0L)); buf498.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_54, mul_338, view_27, transpose_36, contiguous_36, linear_55, view_28, transpose_37, contiguous_37, linear_56, view_29, transpose_38, contiguous_38, scaled_dot_product_attention_9], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:126
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf501, buf506, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf507 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf501, 4, int_array_12, int_array_13, 0L)); buf501.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_54, mul_338, view_27, transpose_36, contiguous_36, linear_55, view_28, transpose_37, contiguous_37, linear_56, view_29, transpose_38, contiguous_38, scaled_dot_product_attention_9], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:127
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf504, model_audio_tower_layers_9_self_attn_v_proj_bias, buf507, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf504.reset();
	    // Topologically Sorted Source Nodes: [, linear_54, mul_338, view_27, transpose_36, contiguous_36, linear_55, view_28, transpose_37, contiguous_37, linear_56, view_29, transpose_38, contiguous_38, scaled_dot_product_attention_9], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_9 = 1.0;
	    AtenTensorHandle buf509_handle;
	    AtenTensorHandle buf510_handle;
	    AtenTensorHandle buf511_handle;
	    AtenTensorHandle buf512_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:493
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf505, buf506, buf507, nullptr, 0, 0.0, 0, &var_9, &buf509_handle, &buf510_handle, &buf511_handle, &buf512_handle));
	    RAIIAtenTensorHandle buf509(buf509_handle);
	    RAIIAtenTensorHandle buf510(buf510_handle);
	    RAIIAtenTensorHandle buf511(buf511_handle);
	    RAIIAtenTensorHandle buf512(buf512_handle);
	    buf505.reset();
	    buf506.reset();
	
	    auto buf515 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf509, 3, int_array_4, int_array_5, 0L)); buf509.reset();  // reuse
	    auto buf516 = std::move(buf515);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_39, reshape_9, choose_qparams_affine_default_57, quantize_affine_57, dequantize_affine_114], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf516, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf517 = std::move(buf503);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_115], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:129
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_9_self_attn_out_proj_parametrizations_weight_original1, buf517, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf518 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf507, 2, int_array_8, int_array_9, 0L)); buf507.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_57, dequantize_affine_114, dequantize_affine_115, linear_57, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:494
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf518, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf516, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf517, 2, int_array_10, int_array_11, 0L))));
	    buf517.reset();
	    auto buf519 = std::move(buf466);  // reuse
	    auto buf523 = std::move(buf516);  // reuse
	    auto buf526 = std::move(buf523);  // reuse
	    auto buf527 = std::move(buf526);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_53, add_126, linear_57, add_135, layer_norm_19, choose_qparams_affine_default_58, quantize_affine_58, dequantize_affine_116], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf519, buf527, buf482, model_audio_tower_layers_8_fc2_bias, buf518, model_audio_tower_layers_9_self_attn_out_proj_bias, model_audio_tower_layers_9_final_layer_norm_weight, model_audio_tower_layers_9_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf482.reset();
	    buf518.reset();
	    auto buf528 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf481, 3, int_array_14, int_array_7, 0L)); buf481.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_117], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:131
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_9_fc1_parametrizations_weight_original0, model_audio_tower_layers_9_fc1_parametrizations_weight_original2, model_audio_tower_layers_9_fc1_parametrizations_weight_original1, buf528, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf529_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf529_handle));
	    RAIIAtenTensorHandle buf529(buf529_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_58, dequantize_affine_116, dequantize_affine_117, linear_58, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:495
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf529, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf527, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf528, 2, int_array_17, int_array_11, 0L))));
	    buf527.reset();
	    buf528.reset();
	    auto buf532 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf529, 3, int_array_18, int_array_19, 0L)); buf529.reset();  // reuse
	    auto buf533 = std::move(buf532);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_58, gelu_11, choose_qparams_affine_default_59, quantize_affine_59, dequantize_affine_118], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf533, model_audio_tower_layers_9_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf534_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf534_handle));
	    RAIIAtenTensorHandle buf534(buf534_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_119], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:133
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_9_fc2_parametrizations_weight_original0, model_audio_tower_layers_9_fc2_parametrizations_weight_original2, model_audio_tower_layers_9_fc2_parametrizations_weight_original1, buf534, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf535_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf535_handle));
	    RAIIAtenTensorHandle buf535(buf535_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_59, quantize_affine_59, dequantize_affine_118, dequantize_affine_119, linear_59, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:496
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf535, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf533, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf534, 2, int_array_22, int_array_23, 0L))));
	    buf533.reset();
	    AtenTensorHandle buf539_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf539_handle));
	    RAIIAtenTensorHandle buf539(buf539_handle);
	    AtenTensorHandle buf542_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf542_handle));
	    RAIIAtenTensorHandle buf542(buf542_handle);
	    auto buf549 = std::move(buf542);  // reuse
	    AtenTensorHandle buf545_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf545_handle));
	    RAIIAtenTensorHandle buf545(buf545_handle);
	    auto buf552 = std::move(buf545);  // reuse
	    AtenTensorHandle buf548_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf548_handle));
	    RAIIAtenTensorHandle buf548(buf548_handle);
	    auto buf555 = std::move(buf548);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_59, add_140, layer_norm_20, choose_qparams_affine_default_60, quantize_affine_60, dequantize_affine_120, choose_qparams_affine_default_61, quantize_affine_61, dequantize_affine_122, choose_qparams_affine_default_62, quantize_affine_62, dequantize_affine_124], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf549, buf552, buf555, buf519, buf535, model_audio_tower_layers_9_fc2_bias, model_audio_tower_layers_10_self_attn_layer_norm_weight, model_audio_tower_layers_10_self_attn_layer_norm_bias, buf539, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf550_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf550_handle));
	    RAIIAtenTensorHandle buf550(buf550_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_121], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:135
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_10_self_attn_q_proj_parametrizations_weight_original1, buf550, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf551 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf539, 2, int_array_8, int_array_9, 0L)); buf539.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_60, dequantize_affine_120, dequantize_affine_121, linear_60, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:497
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf551, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf549, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf550, 2, int_array_10, int_array_11, 0L))));
	    auto buf553 = std::move(buf550);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_123], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:136
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_10_self_attn_k_proj_parametrizations_weight_original1, buf553, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf554 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf549, 2, int_array_8, int_array_9, 0L)); buf549.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_61, dequantize_affine_122, dequantize_affine_123, linear_61], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:498
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf554, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf552, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf553, 2, int_array_10, int_array_11, 0L))));
	    auto buf556 = std::move(buf553);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_125], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:137
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_10_self_attn_v_proj_parametrizations_weight_original1, buf556, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf557 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf552, 2, int_array_8, int_array_9, 0L)); buf552.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_62, dequantize_affine_124, dequantize_affine_125, linear_62, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:499
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf557, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf555, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf556, 2, int_array_10, int_array_11, 0L))));
	    auto buf558 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf555, 4, int_array_12, int_array_13, 0L)); buf555.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_60, mul_375, view_30, transpose_40, contiguous_40, linear_61, view_31, transpose_41, contiguous_41, linear_62, view_32, transpose_42, contiguous_42, scaled_dot_product_attention_10], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:138
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf551, model_audio_tower_layers_10_self_attn_q_proj_bias, buf558, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf559 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf551, 4, int_array_12, int_array_13, 0L)); buf551.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_60, mul_375, view_30, transpose_40, contiguous_40, linear_61, view_31, transpose_41, contiguous_41, linear_62, view_32, transpose_42, contiguous_42, scaled_dot_product_attention_10], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:139
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf554, buf559, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf560 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf554, 4, int_array_12, int_array_13, 0L)); buf554.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_60, mul_375, view_30, transpose_40, contiguous_40, linear_61, view_31, transpose_41, contiguous_41, linear_62, view_32, transpose_42, contiguous_42, scaled_dot_product_attention_10], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:140
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf557, model_audio_tower_layers_10_self_attn_v_proj_bias, buf560, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf557.reset();
	    // Topologically Sorted Source Nodes: [, linear_60, mul_375, view_30, transpose_40, contiguous_40, linear_61, view_31, transpose_41, contiguous_41, linear_62, view_32, transpose_42, contiguous_42, scaled_dot_product_attention_10], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_10 = 1.0;
	    AtenTensorHandle buf562_handle;
	    AtenTensorHandle buf563_handle;
	    AtenTensorHandle buf564_handle;
	    AtenTensorHandle buf565_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:500
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf558, buf559, buf560, nullptr, 0, 0.0, 0, &var_10, &buf562_handle, &buf563_handle, &buf564_handle, &buf565_handle));
	    RAIIAtenTensorHandle buf562(buf562_handle);
	    RAIIAtenTensorHandle buf563(buf563_handle);
	    RAIIAtenTensorHandle buf564(buf564_handle);
	    RAIIAtenTensorHandle buf565(buf565_handle);
	    buf558.reset();
	    buf559.reset();
	
	    auto buf568 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf562, 3, int_array_4, int_array_5, 0L)); buf562.reset();  // reuse
	    auto buf569 = std::move(buf568);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_43, reshape_10, choose_qparams_affine_default_63, quantize_affine_63, dequantize_affine_126], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf569, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf570 = std::move(buf556);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_127], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:142
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_10_self_attn_out_proj_parametrizations_weight_original1, buf570, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf571 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf560, 2, int_array_8, int_array_9, 0L)); buf560.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_63, dequantize_affine_126, dequantize_affine_127, linear_63, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:501
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf571, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf569, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf570, 2, int_array_10, int_array_11, 0L))));
	    buf570.reset();
	    auto buf572 = std::move(buf519);  // reuse
	    auto buf576 = std::move(buf569);  // reuse
	    auto buf579 = std::move(buf576);  // reuse
	    auto buf580 = std::move(buf579);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_59, add_140, linear_63, add_149, layer_norm_21, choose_qparams_affine_default_64, quantize_affine_64, dequantize_affine_128], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf572, buf580, buf535, model_audio_tower_layers_9_fc2_bias, buf571, model_audio_tower_layers_10_self_attn_out_proj_bias, model_audio_tower_layers_10_final_layer_norm_weight, model_audio_tower_layers_10_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf535.reset();
	    buf571.reset();
	    auto buf581 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf534, 3, int_array_14, int_array_7, 0L)); buf534.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_129], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:144
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_10_fc1_parametrizations_weight_original0, model_audio_tower_layers_10_fc1_parametrizations_weight_original2, model_audio_tower_layers_10_fc1_parametrizations_weight_original1, buf581, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf582_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf582_handle));
	    RAIIAtenTensorHandle buf582(buf582_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_64, dequantize_affine_128, dequantize_affine_129, linear_64, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:502
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf582, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf580, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf581, 2, int_array_17, int_array_11, 0L))));
	    buf580.reset();
	    buf581.reset();
	    auto buf585 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf582, 3, int_array_18, int_array_19, 0L)); buf582.reset();  // reuse
	    auto buf586 = std::move(buf585);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_64, gelu_12, choose_qparams_affine_default_65, quantize_affine_65, dequantize_affine_130], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf586, model_audio_tower_layers_10_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf587_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf587_handle));
	    RAIIAtenTensorHandle buf587(buf587_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_131], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:146
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_10_fc2_parametrizations_weight_original0, model_audio_tower_layers_10_fc2_parametrizations_weight_original2, model_audio_tower_layers_10_fc2_parametrizations_weight_original1, buf587, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf588_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf588_handle));
	    RAIIAtenTensorHandle buf588(buf588_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_65, quantize_affine_65, dequantize_affine_130, dequantize_affine_131, linear_65, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:503
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf588, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf586, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf587, 2, int_array_22, int_array_23, 0L))));
	    buf586.reset();
	    AtenTensorHandle buf592_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf592_handle));
	    RAIIAtenTensorHandle buf592(buf592_handle);
	    AtenTensorHandle buf595_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf595_handle));
	    RAIIAtenTensorHandle buf595(buf595_handle);
	    auto buf602 = std::move(buf595);  // reuse
	    AtenTensorHandle buf598_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf598_handle));
	    RAIIAtenTensorHandle buf598(buf598_handle);
	    auto buf605 = std::move(buf598);  // reuse
	    AtenTensorHandle buf601_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf601_handle));
	    RAIIAtenTensorHandle buf601(buf601_handle);
	    auto buf608 = std::move(buf601);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_65, add_154, layer_norm_22, choose_qparams_affine_default_66, quantize_affine_66, dequantize_affine_132, choose_qparams_affine_default_67, quantize_affine_67, dequantize_affine_134, choose_qparams_affine_default_68, quantize_affine_68, dequantize_affine_136], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf602, buf605, buf608, buf572, buf588, model_audio_tower_layers_10_fc2_bias, model_audio_tower_layers_11_self_attn_layer_norm_weight, model_audio_tower_layers_11_self_attn_layer_norm_bias, buf592, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf603_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf603_handle));
	    RAIIAtenTensorHandle buf603(buf603_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_133], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:148
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_11_self_attn_q_proj_parametrizations_weight_original1, buf603, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf604 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf592, 2, int_array_8, int_array_9, 0L)); buf592.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_66, dequantize_affine_132, dequantize_affine_133, linear_66, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:504
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf604, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf602, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf603, 2, int_array_10, int_array_11, 0L))));
	    auto buf606 = std::move(buf603);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_135], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:149
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_11_self_attn_k_proj_parametrizations_weight_original1, buf606, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf607 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf602, 2, int_array_8, int_array_9, 0L)); buf602.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_67, dequantize_affine_134, dequantize_affine_135, linear_67], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:505
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf607, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf605, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf606, 2, int_array_10, int_array_11, 0L))));
	    auto buf609 = std::move(buf606);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_137], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:150
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_11_self_attn_v_proj_parametrizations_weight_original1, buf609, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf610 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf605, 2, int_array_8, int_array_9, 0L)); buf605.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_68, dequantize_affine_136, dequantize_affine_137, linear_68, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:506
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf610, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf608, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf609, 2, int_array_10, int_array_11, 0L))));
	    auto buf611 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf608, 4, int_array_12, int_array_13, 0L)); buf608.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_66, mul_412, view_33, transpose_44, contiguous_44, linear_67, view_34, transpose_45, contiguous_45, linear_68, view_35, transpose_46, contiguous_46, scaled_dot_product_attention_11], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:151
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf604, model_audio_tower_layers_11_self_attn_q_proj_bias, buf611, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf612 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf604, 4, int_array_12, int_array_13, 0L)); buf604.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_66, mul_412, view_33, transpose_44, contiguous_44, linear_67, view_34, transpose_45, contiguous_45, linear_68, view_35, transpose_46, contiguous_46, scaled_dot_product_attention_11], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:152
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf607, buf612, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf613 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf607, 4, int_array_12, int_array_13, 0L)); buf607.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_66, mul_412, view_33, transpose_44, contiguous_44, linear_67, view_34, transpose_45, contiguous_45, linear_68, view_35, transpose_46, contiguous_46, scaled_dot_product_attention_11], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:153
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf610, model_audio_tower_layers_11_self_attn_v_proj_bias, buf613, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf610.reset();
	    // Topologically Sorted Source Nodes: [, linear_66, mul_412, view_33, transpose_44, contiguous_44, linear_67, view_34, transpose_45, contiguous_45, linear_68, view_35, transpose_46, contiguous_46, scaled_dot_product_attention_11], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_11 = 1.0;
	    AtenTensorHandle buf615_handle;
	    AtenTensorHandle buf616_handle;
	    AtenTensorHandle buf617_handle;
	    AtenTensorHandle buf618_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:507
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf611, buf612, buf613, nullptr, 0, 0.0, 0, &var_11, &buf615_handle, &buf616_handle, &buf617_handle, &buf618_handle));
	    RAIIAtenTensorHandle buf615(buf615_handle);
	    RAIIAtenTensorHandle buf616(buf616_handle);
	    RAIIAtenTensorHandle buf617(buf617_handle);
	    RAIIAtenTensorHandle buf618(buf618_handle);
	    buf611.reset();
	    buf612.reset();
	
	    auto buf621 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf615, 3, int_array_4, int_array_5, 0L)); buf615.reset();  // reuse
	    auto buf622 = std::move(buf621);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_47, reshape_11, choose_qparams_affine_default_69, quantize_affine_69, dequantize_affine_138], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf622, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf623 = std::move(buf609);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_139], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:155
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_11_self_attn_out_proj_parametrizations_weight_original1, buf623, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf624 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf613, 2, int_array_8, int_array_9, 0L)); buf613.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_69, dequantize_affine_138, dequantize_affine_139, linear_69, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:508
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf624, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf622, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf623, 2, int_array_10, int_array_11, 0L))));
	    buf623.reset();
	    auto buf625 = std::move(buf572);  // reuse
	    auto buf629 = std::move(buf622);  // reuse
	    auto buf632 = std::move(buf629);  // reuse
	    auto buf633 = std::move(buf632);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_65, add_154, linear_69, add_163, layer_norm_23, choose_qparams_affine_default_70, quantize_affine_70, dequantize_affine_140], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf625, buf633, buf588, model_audio_tower_layers_10_fc2_bias, buf624, model_audio_tower_layers_11_self_attn_out_proj_bias, model_audio_tower_layers_11_final_layer_norm_weight, model_audio_tower_layers_11_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf588.reset();
	    buf624.reset();
	    auto buf634 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf587, 3, int_array_14, int_array_7, 0L)); buf587.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_141], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:157
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_11_fc1_parametrizations_weight_original0, model_audio_tower_layers_11_fc1_parametrizations_weight_original2, model_audio_tower_layers_11_fc1_parametrizations_weight_original1, buf634, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf635_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf635_handle));
	    RAIIAtenTensorHandle buf635(buf635_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_70, dequantize_affine_140, dequantize_affine_141, linear_70, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:509
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf635, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf633, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf634, 2, int_array_17, int_array_11, 0L))));
	    buf633.reset();
	    buf634.reset();
	    auto buf638 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf635, 3, int_array_18, int_array_19, 0L)); buf635.reset();  // reuse
	    auto buf639 = std::move(buf638);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_70, gelu_13, choose_qparams_affine_default_71, quantize_affine_71, dequantize_affine_142], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf639, model_audio_tower_layers_11_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf640_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf640_handle));
	    RAIIAtenTensorHandle buf640(buf640_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_143], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:159
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_11_fc2_parametrizations_weight_original0, model_audio_tower_layers_11_fc2_parametrizations_weight_original2, model_audio_tower_layers_11_fc2_parametrizations_weight_original1, buf640, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf641_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf641_handle));
	    RAIIAtenTensorHandle buf641(buf641_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_71, quantize_affine_71, dequantize_affine_142, dequantize_affine_143, linear_71, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:510
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf641, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf639, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf640, 2, int_array_22, int_array_23, 0L))));
	    buf639.reset();
	    AtenTensorHandle buf645_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf645_handle));
	    RAIIAtenTensorHandle buf645(buf645_handle);
	    AtenTensorHandle buf648_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf648_handle));
	    RAIIAtenTensorHandle buf648(buf648_handle);
	    auto buf655 = std::move(buf648);  // reuse
	    AtenTensorHandle buf651_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf651_handle));
	    RAIIAtenTensorHandle buf651(buf651_handle);
	    auto buf658 = std::move(buf651);  // reuse
	    AtenTensorHandle buf654_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf654_handle));
	    RAIIAtenTensorHandle buf654(buf654_handle);
	    auto buf661 = std::move(buf654);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_71, add_168, layer_norm_24, choose_qparams_affine_default_72, quantize_affine_72, dequantize_affine_144, choose_qparams_affine_default_73, quantize_affine_73, dequantize_affine_146, choose_qparams_affine_default_74, quantize_affine_74, dequantize_affine_148], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf655, buf658, buf661, buf625, buf641, model_audio_tower_layers_11_fc2_bias, model_audio_tower_layers_12_self_attn_layer_norm_weight, model_audio_tower_layers_12_self_attn_layer_norm_bias, buf645, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf656_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf656_handle));
	    RAIIAtenTensorHandle buf656(buf656_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_145], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:161
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_12_self_attn_q_proj_parametrizations_weight_original1, buf656, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf657 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf645, 2, int_array_8, int_array_9, 0L)); buf645.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_72, dequantize_affine_144, dequantize_affine_145, linear_72, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:511
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf657, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf655, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf656, 2, int_array_10, int_array_11, 0L))));
	    auto buf659 = std::move(buf656);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_147], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:162
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_12_self_attn_k_proj_parametrizations_weight_original1, buf659, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf660 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf655, 2, int_array_8, int_array_9, 0L)); buf655.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_73, dequantize_affine_146, dequantize_affine_147, linear_73], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:512
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf660, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf658, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf659, 2, int_array_10, int_array_11, 0L))));
	    auto buf662 = std::move(buf659);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_149], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:163
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_12_self_attn_v_proj_parametrizations_weight_original1, buf662, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf663 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf658, 2, int_array_8, int_array_9, 0L)); buf658.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_74, dequantize_affine_148, dequantize_affine_149, linear_74, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:513
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf663, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf661, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf662, 2, int_array_10, int_array_11, 0L))));
	    auto buf664 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf661, 4, int_array_12, int_array_13, 0L)); buf661.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_72, mul_449, view_36, transpose_48, contiguous_48, linear_73, view_37, transpose_49, contiguous_49, linear_74, view_38, transpose_50, contiguous_50, scaled_dot_product_attention_12], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:164
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf657, model_audio_tower_layers_12_self_attn_q_proj_bias, buf664, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf665 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf657, 4, int_array_12, int_array_13, 0L)); buf657.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_72, mul_449, view_36, transpose_48, contiguous_48, linear_73, view_37, transpose_49, contiguous_49, linear_74, view_38, transpose_50, contiguous_50, scaled_dot_product_attention_12], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:165
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf660, buf665, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf666 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf660, 4, int_array_12, int_array_13, 0L)); buf660.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_72, mul_449, view_36, transpose_48, contiguous_48, linear_73, view_37, transpose_49, contiguous_49, linear_74, view_38, transpose_50, contiguous_50, scaled_dot_product_attention_12], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:166
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf663, model_audio_tower_layers_12_self_attn_v_proj_bias, buf666, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf663.reset();
	    // Topologically Sorted Source Nodes: [, linear_72, mul_449, view_36, transpose_48, contiguous_48, linear_73, view_37, transpose_49, contiguous_49, linear_74, view_38, transpose_50, contiguous_50, scaled_dot_product_attention_12], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_12 = 1.0;
	    AtenTensorHandle buf668_handle;
	    AtenTensorHandle buf669_handle;
	    AtenTensorHandle buf670_handle;
	    AtenTensorHandle buf671_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:514
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf664, buf665, buf666, nullptr, 0, 0.0, 0, &var_12, &buf668_handle, &buf669_handle, &buf670_handle, &buf671_handle));
	    RAIIAtenTensorHandle buf668(buf668_handle);
	    RAIIAtenTensorHandle buf669(buf669_handle);
	    RAIIAtenTensorHandle buf670(buf670_handle);
	    RAIIAtenTensorHandle buf671(buf671_handle);
	    buf664.reset();
	    buf665.reset();
	
	    auto buf674 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf668, 3, int_array_4, int_array_5, 0L)); buf668.reset();  // reuse
	    auto buf675 = std::move(buf674);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_51, reshape_12, choose_qparams_affine_default_75, quantize_affine_75, dequantize_affine_150], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf675, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf676 = std::move(buf662);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_151], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:168
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_12_self_attn_out_proj_parametrizations_weight_original1, buf676, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf677 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf666, 2, int_array_8, int_array_9, 0L)); buf666.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_75, dequantize_affine_150, dequantize_affine_151, linear_75, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:515
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf677, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf675, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf676, 2, int_array_10, int_array_11, 0L))));
	    buf676.reset();
	    auto buf678 = std::move(buf625);  // reuse
	    auto buf682 = std::move(buf675);  // reuse
	    auto buf685 = std::move(buf682);  // reuse
	    auto buf686 = std::move(buf685);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_71, add_168, linear_75, add_177, layer_norm_25, choose_qparams_affine_default_76, quantize_affine_76, dequantize_affine_152], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf678, buf686, buf641, model_audio_tower_layers_11_fc2_bias, buf677, model_audio_tower_layers_12_self_attn_out_proj_bias, model_audio_tower_layers_12_final_layer_norm_weight, model_audio_tower_layers_12_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf641.reset();
	    buf677.reset();
	    auto buf687 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf640, 3, int_array_14, int_array_7, 0L)); buf640.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_153], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:170
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_12_fc1_parametrizations_weight_original0, model_audio_tower_layers_12_fc1_parametrizations_weight_original2, model_audio_tower_layers_12_fc1_parametrizations_weight_original1, buf687, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf688_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf688_handle));
	    RAIIAtenTensorHandle buf688(buf688_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_76, dequantize_affine_152, dequantize_affine_153, linear_76, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:516
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf688, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf686, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf687, 2, int_array_17, int_array_11, 0L))));
	    buf686.reset();
	    buf687.reset();
	    auto buf691 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf688, 3, int_array_18, int_array_19, 0L)); buf688.reset();  // reuse
	    auto buf692 = std::move(buf691);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_76, gelu_14, choose_qparams_affine_default_77, quantize_affine_77, dequantize_affine_154], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf692, model_audio_tower_layers_12_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf693_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf693_handle));
	    RAIIAtenTensorHandle buf693(buf693_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_155], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:172
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_12_fc2_parametrizations_weight_original0, model_audio_tower_layers_12_fc2_parametrizations_weight_original2, model_audio_tower_layers_12_fc2_parametrizations_weight_original1, buf693, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf694_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf694_handle));
	    RAIIAtenTensorHandle buf694(buf694_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_77, quantize_affine_77, dequantize_affine_154, dequantize_affine_155, linear_77, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:517
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf694, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf692, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf693, 2, int_array_22, int_array_23, 0L))));
	    buf692.reset();
	    AtenTensorHandle buf698_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf698_handle));
	    RAIIAtenTensorHandle buf698(buf698_handle);
	    AtenTensorHandle buf701_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf701_handle));
	    RAIIAtenTensorHandle buf701(buf701_handle);
	    auto buf708 = std::move(buf701);  // reuse
	    AtenTensorHandle buf704_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf704_handle));
	    RAIIAtenTensorHandle buf704(buf704_handle);
	    auto buf711 = std::move(buf704);  // reuse
	    AtenTensorHandle buf707_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf707_handle));
	    RAIIAtenTensorHandle buf707(buf707_handle);
	    auto buf714 = std::move(buf707);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_77, add_182, layer_norm_26, choose_qparams_affine_default_78, quantize_affine_78, dequantize_affine_156, choose_qparams_affine_default_79, quantize_affine_79, dequantize_affine_158, choose_qparams_affine_default_80, quantize_affine_80, dequantize_affine_160], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf708, buf711, buf714, buf678, buf694, model_audio_tower_layers_12_fc2_bias, model_audio_tower_layers_13_self_attn_layer_norm_weight, model_audio_tower_layers_13_self_attn_layer_norm_bias, buf698, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf709_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf709_handle));
	    RAIIAtenTensorHandle buf709(buf709_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_157], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:174
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_13_self_attn_q_proj_parametrizations_weight_original1, buf709, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf710 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf698, 2, int_array_8, int_array_9, 0L)); buf698.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_78, dequantize_affine_156, dequantize_affine_157, linear_78, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:518
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf710, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf708, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf709, 2, int_array_10, int_array_11, 0L))));
	    auto buf712 = std::move(buf709);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_159], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:175
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_13_self_attn_k_proj_parametrizations_weight_original1, buf712, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf713 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf708, 2, int_array_8, int_array_9, 0L)); buf708.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_79, dequantize_affine_158, dequantize_affine_159, linear_79], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:519
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf713, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf711, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf712, 2, int_array_10, int_array_11, 0L))));
	    auto buf715 = std::move(buf712);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_161], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:176
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_13_self_attn_v_proj_parametrizations_weight_original1, buf715, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf716 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf711, 2, int_array_8, int_array_9, 0L)); buf711.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_80, dequantize_affine_160, dequantize_affine_161, linear_80, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:520
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf716, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf714, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf715, 2, int_array_10, int_array_11, 0L))));
	    auto buf717 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf714, 4, int_array_12, int_array_13, 0L)); buf714.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_78, mul_486, view_39, transpose_52, contiguous_52, linear_79, view_40, transpose_53, contiguous_53, linear_80, view_41, transpose_54, contiguous_54, scaled_dot_product_attention_13], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:177
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf710, model_audio_tower_layers_13_self_attn_q_proj_bias, buf717, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf718 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf710, 4, int_array_12, int_array_13, 0L)); buf710.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_78, mul_486, view_39, transpose_52, contiguous_52, linear_79, view_40, transpose_53, contiguous_53, linear_80, view_41, transpose_54, contiguous_54, scaled_dot_product_attention_13], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:178
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf713, buf718, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf719 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf713, 4, int_array_12, int_array_13, 0L)); buf713.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_78, mul_486, view_39, transpose_52, contiguous_52, linear_79, view_40, transpose_53, contiguous_53, linear_80, view_41, transpose_54, contiguous_54, scaled_dot_product_attention_13], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:179
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf716, model_audio_tower_layers_13_self_attn_v_proj_bias, buf719, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf716.reset();
	    // Topologically Sorted Source Nodes: [, linear_78, mul_486, view_39, transpose_52, contiguous_52, linear_79, view_40, transpose_53, contiguous_53, linear_80, view_41, transpose_54, contiguous_54, scaled_dot_product_attention_13], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_13 = 1.0;
	    AtenTensorHandle buf721_handle;
	    AtenTensorHandle buf722_handle;
	    AtenTensorHandle buf723_handle;
	    AtenTensorHandle buf724_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:521
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf717, buf718, buf719, nullptr, 0, 0.0, 0, &var_13, &buf721_handle, &buf722_handle, &buf723_handle, &buf724_handle));
	    RAIIAtenTensorHandle buf721(buf721_handle);
	    RAIIAtenTensorHandle buf722(buf722_handle);
	    RAIIAtenTensorHandle buf723(buf723_handle);
	    RAIIAtenTensorHandle buf724(buf724_handle);
	    buf717.reset();
	    buf718.reset();
	
	    auto buf727 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf721, 3, int_array_4, int_array_5, 0L)); buf721.reset();  // reuse
	    auto buf728 = std::move(buf727);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_55, reshape_13, choose_qparams_affine_default_81, quantize_affine_81, dequantize_affine_162], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf728, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf729 = std::move(buf715);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_163], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:181
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_13_self_attn_out_proj_parametrizations_weight_original1, buf729, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf730 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf719, 2, int_array_8, int_array_9, 0L)); buf719.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_81, dequantize_affine_162, dequantize_affine_163, linear_81, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:522
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf730, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf728, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf729, 2, int_array_10, int_array_11, 0L))));
	    buf729.reset();
	    auto buf731 = std::move(buf678);  // reuse
	    auto buf735 = std::move(buf728);  // reuse
	    auto buf738 = std::move(buf735);  // reuse
	    auto buf739 = std::move(buf738);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_77, add_182, linear_81, add_191, layer_norm_27, choose_qparams_affine_default_82, quantize_affine_82, dequantize_affine_164], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf731, buf739, buf694, model_audio_tower_layers_12_fc2_bias, buf730, model_audio_tower_layers_13_self_attn_out_proj_bias, model_audio_tower_layers_13_final_layer_norm_weight, model_audio_tower_layers_13_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf694.reset();
	    buf730.reset();
	    auto buf740 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf693, 3, int_array_14, int_array_7, 0L)); buf693.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_165], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:183
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_13_fc1_parametrizations_weight_original0, model_audio_tower_layers_13_fc1_parametrizations_weight_original2, model_audio_tower_layers_13_fc1_parametrizations_weight_original1, buf740, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf741_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf741_handle));
	    RAIIAtenTensorHandle buf741(buf741_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_82, dequantize_affine_164, dequantize_affine_165, linear_82, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:523
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf741, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf739, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf740, 2, int_array_17, int_array_11, 0L))));
	    buf739.reset();
	    buf740.reset();
	    auto buf744 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf741, 3, int_array_18, int_array_19, 0L)); buf741.reset();  // reuse
	    auto buf745 = std::move(buf744);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_82, gelu_15, choose_qparams_affine_default_83, quantize_affine_83, dequantize_affine_166], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf745, model_audio_tower_layers_13_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf746_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf746_handle));
	    RAIIAtenTensorHandle buf746(buf746_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_167], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:185
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_13_fc2_parametrizations_weight_original0, model_audio_tower_layers_13_fc2_parametrizations_weight_original2, model_audio_tower_layers_13_fc2_parametrizations_weight_original1, buf746, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf747_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf747_handle));
	    RAIIAtenTensorHandle buf747(buf747_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_83, quantize_affine_83, dequantize_affine_166, dequantize_affine_167, linear_83, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:524
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf747, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf745, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf746, 2, int_array_22, int_array_23, 0L))));
	    buf745.reset();
	    AtenTensorHandle buf751_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf751_handle));
	    RAIIAtenTensorHandle buf751(buf751_handle);
	    AtenTensorHandle buf754_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf754_handle));
	    RAIIAtenTensorHandle buf754(buf754_handle);
	    auto buf761 = std::move(buf754);  // reuse
	    AtenTensorHandle buf757_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf757_handle));
	    RAIIAtenTensorHandle buf757(buf757_handle);
	    auto buf764 = std::move(buf757);  // reuse
	    AtenTensorHandle buf760_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf760_handle));
	    RAIIAtenTensorHandle buf760(buf760_handle);
	    auto buf767 = std::move(buf760);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_83, add_196, layer_norm_28, choose_qparams_affine_default_84, quantize_affine_84, dequantize_affine_168, choose_qparams_affine_default_85, quantize_affine_85, dequantize_affine_170, choose_qparams_affine_default_86, quantize_affine_86, dequantize_affine_172], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf761, buf764, buf767, buf731, buf747, model_audio_tower_layers_13_fc2_bias, model_audio_tower_layers_14_self_attn_layer_norm_weight, model_audio_tower_layers_14_self_attn_layer_norm_bias, buf751, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf762_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf762_handle));
	    RAIIAtenTensorHandle buf762(buf762_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_169], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:187
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_14_self_attn_q_proj_parametrizations_weight_original1, buf762, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf763 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf751, 2, int_array_8, int_array_9, 0L)); buf751.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_84, dequantize_affine_168, dequantize_affine_169, linear_84, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:525
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf763, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf761, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf762, 2, int_array_10, int_array_11, 0L))));
	    auto buf765 = std::move(buf762);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_171], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:188
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_14_self_attn_k_proj_parametrizations_weight_original1, buf765, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf766 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf761, 2, int_array_8, int_array_9, 0L)); buf761.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_85, dequantize_affine_170, dequantize_affine_171, linear_85], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:526
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf766, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf764, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf765, 2, int_array_10, int_array_11, 0L))));
	    auto buf768 = std::move(buf765);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_173], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:189
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_14_self_attn_v_proj_parametrizations_weight_original1, buf768, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf769 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf764, 2, int_array_8, int_array_9, 0L)); buf764.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_86, dequantize_affine_172, dequantize_affine_173, linear_86, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:527
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf769, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf767, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf768, 2, int_array_10, int_array_11, 0L))));
	    auto buf770 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf767, 4, int_array_12, int_array_13, 0L)); buf767.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_84, mul_523, view_42, transpose_56, contiguous_56, linear_85, view_43, transpose_57, contiguous_57, linear_86, view_44, transpose_58, contiguous_58, scaled_dot_product_attention_14], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:190
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf763, model_audio_tower_layers_14_self_attn_q_proj_bias, buf770, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf771 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf763, 4, int_array_12, int_array_13, 0L)); buf763.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_84, mul_523, view_42, transpose_56, contiguous_56, linear_85, view_43, transpose_57, contiguous_57, linear_86, view_44, transpose_58, contiguous_58, scaled_dot_product_attention_14], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:191
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf766, buf771, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf772 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf766, 4, int_array_12, int_array_13, 0L)); buf766.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_84, mul_523, view_42, transpose_56, contiguous_56, linear_85, view_43, transpose_57, contiguous_57, linear_86, view_44, transpose_58, contiguous_58, scaled_dot_product_attention_14], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:192
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf769, model_audio_tower_layers_14_self_attn_v_proj_bias, buf772, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf769.reset();
	    // Topologically Sorted Source Nodes: [, linear_84, mul_523, view_42, transpose_56, contiguous_56, linear_85, view_43, transpose_57, contiguous_57, linear_86, view_44, transpose_58, contiguous_58, scaled_dot_product_attention_14], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_14 = 1.0;
	    AtenTensorHandle buf774_handle;
	    AtenTensorHandle buf775_handle;
	    AtenTensorHandle buf776_handle;
	    AtenTensorHandle buf777_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:528
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf770, buf771, buf772, nullptr, 0, 0.0, 0, &var_14, &buf774_handle, &buf775_handle, &buf776_handle, &buf777_handle));
	    RAIIAtenTensorHandle buf774(buf774_handle);
	    RAIIAtenTensorHandle buf775(buf775_handle);
	    RAIIAtenTensorHandle buf776(buf776_handle);
	    RAIIAtenTensorHandle buf777(buf777_handle);
	    buf770.reset();
	    buf771.reset();
	
	    auto buf780 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf774, 3, int_array_4, int_array_5, 0L)); buf774.reset();  // reuse
	    auto buf781 = std::move(buf780);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_59, reshape_14, choose_qparams_affine_default_87, quantize_affine_87, dequantize_affine_174], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf781, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf782 = std::move(buf768);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_175], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:194
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_14_self_attn_out_proj_parametrizations_weight_original1, buf782, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf783 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf772, 2, int_array_8, int_array_9, 0L)); buf772.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_87, dequantize_affine_174, dequantize_affine_175, linear_87, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:529
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf783, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf781, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf782, 2, int_array_10, int_array_11, 0L))));
	    buf782.reset();
	    auto buf784 = std::move(buf731);  // reuse
	    auto buf788 = std::move(buf781);  // reuse
	    auto buf791 = std::move(buf788);  // reuse
	    auto buf792 = std::move(buf791);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_83, add_196, linear_87, add_205, layer_norm_29, choose_qparams_affine_default_88, quantize_affine_88, dequantize_affine_176], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf784, buf792, buf747, model_audio_tower_layers_13_fc2_bias, buf783, model_audio_tower_layers_14_self_attn_out_proj_bias, model_audio_tower_layers_14_final_layer_norm_weight, model_audio_tower_layers_14_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf747.reset();
	    buf783.reset();
	    auto buf793 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf746, 3, int_array_14, int_array_7, 0L)); buf746.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_177], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:196
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_14_fc1_parametrizations_weight_original0, model_audio_tower_layers_14_fc1_parametrizations_weight_original2, model_audio_tower_layers_14_fc1_parametrizations_weight_original1, buf793, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf794_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf794_handle));
	    RAIIAtenTensorHandle buf794(buf794_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_88, dequantize_affine_176, dequantize_affine_177, linear_88, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:530
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf794, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf792, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf793, 2, int_array_17, int_array_11, 0L))));
	    buf792.reset();
	    buf793.reset();
	    auto buf797 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf794, 3, int_array_18, int_array_19, 0L)); buf794.reset();  // reuse
	    auto buf798 = std::move(buf797);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_88, gelu_16, choose_qparams_affine_default_89, quantize_affine_89, dequantize_affine_178], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf798, model_audio_tower_layers_14_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf799_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf799_handle));
	    RAIIAtenTensorHandle buf799(buf799_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_179], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:198
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_14_fc2_parametrizations_weight_original0, model_audio_tower_layers_14_fc2_parametrizations_weight_original2, model_audio_tower_layers_14_fc2_parametrizations_weight_original1, buf799, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf800_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf800_handle));
	    RAIIAtenTensorHandle buf800(buf800_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_89, quantize_affine_89, dequantize_affine_178, dequantize_affine_179, linear_89, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:531
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf800, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf798, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf799, 2, int_array_22, int_array_23, 0L))));
	    buf798.reset();
	    AtenTensorHandle buf804_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf804_handle));
	    RAIIAtenTensorHandle buf804(buf804_handle);
	    AtenTensorHandle buf807_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf807_handle));
	    RAIIAtenTensorHandle buf807(buf807_handle);
	    auto buf814 = std::move(buf807);  // reuse
	    AtenTensorHandle buf810_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf810_handle));
	    RAIIAtenTensorHandle buf810(buf810_handle);
	    auto buf817 = std::move(buf810);  // reuse
	    AtenTensorHandle buf813_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf813_handle));
	    RAIIAtenTensorHandle buf813(buf813_handle);
	    auto buf820 = std::move(buf813);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_89, add_210, layer_norm_30, choose_qparams_affine_default_90, quantize_affine_90, dequantize_affine_180, choose_qparams_affine_default_91, quantize_affine_91, dequantize_affine_182, choose_qparams_affine_default_92, quantize_affine_92, dequantize_affine_184], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf814, buf817, buf820, buf784, buf800, model_audio_tower_layers_14_fc2_bias, model_audio_tower_layers_15_self_attn_layer_norm_weight, model_audio_tower_layers_15_self_attn_layer_norm_bias, buf804, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf815_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf815_handle));
	    RAIIAtenTensorHandle buf815(buf815_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_181], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:200
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_15_self_attn_q_proj_parametrizations_weight_original1, buf815, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf816 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf804, 2, int_array_8, int_array_9, 0L)); buf804.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_90, dequantize_affine_180, dequantize_affine_181, linear_90, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:532
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf816, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf814, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf815, 2, int_array_10, int_array_11, 0L))));
	    auto buf818 = std::move(buf815);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_183], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:201
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_15_self_attn_k_proj_parametrizations_weight_original1, buf818, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf819 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf814, 2, int_array_8, int_array_9, 0L)); buf814.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_91, dequantize_affine_182, dequantize_affine_183, linear_91], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:533
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf819, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf817, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf818, 2, int_array_10, int_array_11, 0L))));
	    auto buf821 = std::move(buf818);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_185], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:202
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_15_self_attn_v_proj_parametrizations_weight_original1, buf821, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf822 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf817, 2, int_array_8, int_array_9, 0L)); buf817.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_92, dequantize_affine_184, dequantize_affine_185, linear_92, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:534
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf822, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf820, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf821, 2, int_array_10, int_array_11, 0L))));
	    auto buf823 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf820, 4, int_array_12, int_array_13, 0L)); buf820.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_90, mul_560, view_45, transpose_60, contiguous_60, linear_91, view_46, transpose_61, contiguous_61, linear_92, view_47, transpose_62, contiguous_62, scaled_dot_product_attention_15], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:203
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf816, model_audio_tower_layers_15_self_attn_q_proj_bias, buf823, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf824 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf816, 4, int_array_12, int_array_13, 0L)); buf816.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_90, mul_560, view_45, transpose_60, contiguous_60, linear_91, view_46, transpose_61, contiguous_61, linear_92, view_47, transpose_62, contiguous_62, scaled_dot_product_attention_15], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:204
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf819, buf824, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf825 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf819, 4, int_array_12, int_array_13, 0L)); buf819.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_90, mul_560, view_45, transpose_60, contiguous_60, linear_91, view_46, transpose_61, contiguous_61, linear_92, view_47, transpose_62, contiguous_62, scaled_dot_product_attention_15], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:205
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf822, model_audio_tower_layers_15_self_attn_v_proj_bias, buf825, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf822.reset();
	    // Topologically Sorted Source Nodes: [, linear_90, mul_560, view_45, transpose_60, contiguous_60, linear_91, view_46, transpose_61, contiguous_61, linear_92, view_47, transpose_62, contiguous_62, scaled_dot_product_attention_15], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_15 = 1.0;
	    AtenTensorHandle buf827_handle;
	    AtenTensorHandle buf828_handle;
	    AtenTensorHandle buf829_handle;
	    AtenTensorHandle buf830_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:535
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf823, buf824, buf825, nullptr, 0, 0.0, 0, &var_15, &buf827_handle, &buf828_handle, &buf829_handle, &buf830_handle));
	    RAIIAtenTensorHandle buf827(buf827_handle);
	    RAIIAtenTensorHandle buf828(buf828_handle);
	    RAIIAtenTensorHandle buf829(buf829_handle);
	    RAIIAtenTensorHandle buf830(buf830_handle);
	    buf823.reset();
	    buf824.reset();
	
	    auto buf833 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf827, 3, int_array_4, int_array_5, 0L)); buf827.reset();  // reuse
	    auto buf834 = std::move(buf833);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_63, reshape_15, choose_qparams_affine_default_93, quantize_affine_93, dequantize_affine_186], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf834, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf835 = std::move(buf821);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_187], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:207
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_15_self_attn_out_proj_parametrizations_weight_original1, buf835, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf836 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf825, 2, int_array_8, int_array_9, 0L)); buf825.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_93, dequantize_affine_186, dequantize_affine_187, linear_93, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:536
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf836, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf834, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf835, 2, int_array_10, int_array_11, 0L))));
	    buf835.reset();
	    auto buf837 = std::move(buf784);  // reuse
	    auto buf841 = std::move(buf834);  // reuse
	    auto buf844 = std::move(buf841);  // reuse
	    auto buf845 = std::move(buf844);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_89, add_210, linear_93, add_219, layer_norm_31, choose_qparams_affine_default_94, quantize_affine_94, dequantize_affine_188], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf837, buf845, buf800, model_audio_tower_layers_14_fc2_bias, buf836, model_audio_tower_layers_15_self_attn_out_proj_bias, model_audio_tower_layers_15_final_layer_norm_weight, model_audio_tower_layers_15_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf800.reset();
	    buf836.reset();
	    auto buf846 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf799, 3, int_array_14, int_array_7, 0L)); buf799.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_189], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:209
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_15_fc1_parametrizations_weight_original0, model_audio_tower_layers_15_fc1_parametrizations_weight_original2, model_audio_tower_layers_15_fc1_parametrizations_weight_original1, buf846, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf847_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf847_handle));
	    RAIIAtenTensorHandle buf847(buf847_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_94, dequantize_affine_188, dequantize_affine_189, linear_94, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:537
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf847, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf845, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf846, 2, int_array_17, int_array_11, 0L))));
	    buf845.reset();
	    buf846.reset();
	    auto buf850 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf847, 3, int_array_18, int_array_19, 0L)); buf847.reset();  // reuse
	    auto buf851 = std::move(buf850);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_94, gelu_17, choose_qparams_affine_default_95, quantize_affine_95, dequantize_affine_190], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf851, model_audio_tower_layers_15_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf852_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf852_handle));
	    RAIIAtenTensorHandle buf852(buf852_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_191], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:211
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_15_fc2_parametrizations_weight_original0, model_audio_tower_layers_15_fc2_parametrizations_weight_original2, model_audio_tower_layers_15_fc2_parametrizations_weight_original1, buf852, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf853_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf853_handle));
	    RAIIAtenTensorHandle buf853(buf853_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_95, quantize_affine_95, dequantize_affine_190, dequantize_affine_191, linear_95, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:538
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf853, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf851, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf852, 2, int_array_22, int_array_23, 0L))));
	    buf851.reset();
	    AtenTensorHandle buf857_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf857_handle));
	    RAIIAtenTensorHandle buf857(buf857_handle);
	    AtenTensorHandle buf860_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf860_handle));
	    RAIIAtenTensorHandle buf860(buf860_handle);
	    auto buf867 = std::move(buf860);  // reuse
	    AtenTensorHandle buf863_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf863_handle));
	    RAIIAtenTensorHandle buf863(buf863_handle);
	    auto buf870 = std::move(buf863);  // reuse
	    AtenTensorHandle buf866_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf866_handle));
	    RAIIAtenTensorHandle buf866(buf866_handle);
	    auto buf873 = std::move(buf866);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_95, add_224, layer_norm_32, choose_qparams_affine_default_96, quantize_affine_96, dequantize_affine_192, choose_qparams_affine_default_97, quantize_affine_97, dequantize_affine_194, choose_qparams_affine_default_98, quantize_affine_98, dequantize_affine_196], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf867, buf870, buf873, buf837, buf853, model_audio_tower_layers_15_fc2_bias, model_audio_tower_layers_16_self_attn_layer_norm_weight, model_audio_tower_layers_16_self_attn_layer_norm_bias, buf857, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf868_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf868_handle));
	    RAIIAtenTensorHandle buf868(buf868_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_193], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:213
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_16_self_attn_q_proj_parametrizations_weight_original1, buf868, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf869 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf857, 2, int_array_8, int_array_9, 0L)); buf857.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_96, dequantize_affine_192, dequantize_affine_193, linear_96, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:539
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf869, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf867, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf868, 2, int_array_10, int_array_11, 0L))));
	    auto buf871 = std::move(buf868);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_195], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:214
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_16_self_attn_k_proj_parametrizations_weight_original1, buf871, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf872 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf867, 2, int_array_8, int_array_9, 0L)); buf867.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_97, dequantize_affine_194, dequantize_affine_195, linear_97], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:540
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf872, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf870, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf871, 2, int_array_10, int_array_11, 0L))));
	    auto buf874 = std::move(buf871);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_197], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:215
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_16_self_attn_v_proj_parametrizations_weight_original1, buf874, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf875 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf870, 2, int_array_8, int_array_9, 0L)); buf870.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_98, dequantize_affine_196, dequantize_affine_197, linear_98, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:541
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf875, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf873, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf874, 2, int_array_10, int_array_11, 0L))));
	    auto buf876 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf873, 4, int_array_12, int_array_13, 0L)); buf873.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_96, mul_597, view_48, transpose_64, contiguous_64, linear_97, view_49, transpose_65, contiguous_65, linear_98, view_50, transpose_66, contiguous_66, scaled_dot_product_attention_16], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:216
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf869, model_audio_tower_layers_16_self_attn_q_proj_bias, buf876, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf877 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf869, 4, int_array_12, int_array_13, 0L)); buf869.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_96, mul_597, view_48, transpose_64, contiguous_64, linear_97, view_49, transpose_65, contiguous_65, linear_98, view_50, transpose_66, contiguous_66, scaled_dot_product_attention_16], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:217
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf872, buf877, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf878 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf872, 4, int_array_12, int_array_13, 0L)); buf872.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_96, mul_597, view_48, transpose_64, contiguous_64, linear_97, view_49, transpose_65, contiguous_65, linear_98, view_50, transpose_66, contiguous_66, scaled_dot_product_attention_16], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:218
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf875, model_audio_tower_layers_16_self_attn_v_proj_bias, buf878, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf875.reset();
	    // Topologically Sorted Source Nodes: [, linear_96, mul_597, view_48, transpose_64, contiguous_64, linear_97, view_49, transpose_65, contiguous_65, linear_98, view_50, transpose_66, contiguous_66, scaled_dot_product_attention_16], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_16 = 1.0;
	    AtenTensorHandle buf880_handle;
	    AtenTensorHandle buf881_handle;
	    AtenTensorHandle buf882_handle;
	    AtenTensorHandle buf883_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:542
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf876, buf877, buf878, nullptr, 0, 0.0, 0, &var_16, &buf880_handle, &buf881_handle, &buf882_handle, &buf883_handle));
	    RAIIAtenTensorHandle buf880(buf880_handle);
	    RAIIAtenTensorHandle buf881(buf881_handle);
	    RAIIAtenTensorHandle buf882(buf882_handle);
	    RAIIAtenTensorHandle buf883(buf883_handle);
	    buf876.reset();
	    buf877.reset();
	
	    auto buf886 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf880, 3, int_array_4, int_array_5, 0L)); buf880.reset();  // reuse
	    auto buf887 = std::move(buf886);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_67, reshape_16, choose_qparams_affine_default_99, quantize_affine_99, dequantize_affine_198], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf887, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf888 = std::move(buf874);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_199], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:220
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_16_self_attn_out_proj_parametrizations_weight_original1, buf888, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf889 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf878, 2, int_array_8, int_array_9, 0L)); buf878.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_99, dequantize_affine_198, dequantize_affine_199, linear_99, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:543
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf889, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf887, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf888, 2, int_array_10, int_array_11, 0L))));
	    buf888.reset();
	    auto buf890 = std::move(buf837);  // reuse
	    auto buf894 = std::move(buf887);  // reuse
	    auto buf897 = std::move(buf894);  // reuse
	    auto buf898 = std::move(buf897);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_95, add_224, linear_99, add_233, layer_norm_33, choose_qparams_affine_default_100, quantize_affine_100, dequantize_affine_200], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf890, buf898, buf853, model_audio_tower_layers_15_fc2_bias, buf889, model_audio_tower_layers_16_self_attn_out_proj_bias, model_audio_tower_layers_16_final_layer_norm_weight, model_audio_tower_layers_16_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf853.reset();
	    buf889.reset();
	    auto buf899 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf852, 3, int_array_14, int_array_7, 0L)); buf852.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_201], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:222
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_16_fc1_parametrizations_weight_original0, model_audio_tower_layers_16_fc1_parametrizations_weight_original2, model_audio_tower_layers_16_fc1_parametrizations_weight_original1, buf899, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf900_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf900_handle));
	    RAIIAtenTensorHandle buf900(buf900_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_100, dequantize_affine_200, dequantize_affine_201, linear_100, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:544
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf900, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf898, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf899, 2, int_array_17, int_array_11, 0L))));
	    buf898.reset();
	    buf899.reset();
	    auto buf903 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf900, 3, int_array_18, int_array_19, 0L)); buf900.reset();  // reuse
	    auto buf904 = std::move(buf903);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_100, gelu_18, choose_qparams_affine_default_101, quantize_affine_101, dequantize_affine_202], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf904, model_audio_tower_layers_16_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf905_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf905_handle));
	    RAIIAtenTensorHandle buf905(buf905_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_203], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:224
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_16_fc2_parametrizations_weight_original0, model_audio_tower_layers_16_fc2_parametrizations_weight_original2, model_audio_tower_layers_16_fc2_parametrizations_weight_original1, buf905, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf906_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf906_handle));
	    RAIIAtenTensorHandle buf906(buf906_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_101, quantize_affine_101, dequantize_affine_202, dequantize_affine_203, linear_101, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:545
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf906, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf904, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf905, 2, int_array_22, int_array_23, 0L))));
	    buf904.reset();
	    AtenTensorHandle buf910_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf910_handle));
	    RAIIAtenTensorHandle buf910(buf910_handle);
	    AtenTensorHandle buf913_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf913_handle));
	    RAIIAtenTensorHandle buf913(buf913_handle);
	    auto buf920 = std::move(buf913);  // reuse
	    AtenTensorHandle buf916_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf916_handle));
	    RAIIAtenTensorHandle buf916(buf916_handle);
	    auto buf923 = std::move(buf916);  // reuse
	    AtenTensorHandle buf919_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf919_handle));
	    RAIIAtenTensorHandle buf919(buf919_handle);
	    auto buf926 = std::move(buf919);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_101, add_238, layer_norm_34, choose_qparams_affine_default_102, quantize_affine_102, dequantize_affine_204, choose_qparams_affine_default_103, quantize_affine_103, dequantize_affine_206, choose_qparams_affine_default_104, quantize_affine_104, dequantize_affine_208], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf920, buf923, buf926, buf890, buf906, model_audio_tower_layers_16_fc2_bias, model_audio_tower_layers_17_self_attn_layer_norm_weight, model_audio_tower_layers_17_self_attn_layer_norm_bias, buf910, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf921_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf921_handle));
	    RAIIAtenTensorHandle buf921(buf921_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_205], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:226
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_17_self_attn_q_proj_parametrizations_weight_original1, buf921, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf922 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf910, 2, int_array_8, int_array_9, 0L)); buf910.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_102, dequantize_affine_204, dequantize_affine_205, linear_102, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:546
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf922, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf920, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf921, 2, int_array_10, int_array_11, 0L))));
	    auto buf924 = std::move(buf921);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_207], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:227
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_17_self_attn_k_proj_parametrizations_weight_original1, buf924, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf925 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf920, 2, int_array_8, int_array_9, 0L)); buf920.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_103, dequantize_affine_206, dequantize_affine_207, linear_103], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:547
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf925, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf923, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf924, 2, int_array_10, int_array_11, 0L))));
	    auto buf927 = std::move(buf924);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_209], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:228
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_17_self_attn_v_proj_parametrizations_weight_original1, buf927, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf928 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf923, 2, int_array_8, int_array_9, 0L)); buf923.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_104, dequantize_affine_208, dequantize_affine_209, linear_104, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:548
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf928, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf926, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf927, 2, int_array_10, int_array_11, 0L))));
	    auto buf929 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf926, 4, int_array_12, int_array_13, 0L)); buf926.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_102, mul_634, view_51, transpose_68, contiguous_68, linear_103, view_52, transpose_69, contiguous_69, linear_104, view_53, transpose_70, contiguous_70, scaled_dot_product_attention_17], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:229
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf922, model_audio_tower_layers_17_self_attn_q_proj_bias, buf929, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf930 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf922, 4, int_array_12, int_array_13, 0L)); buf922.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_102, mul_634, view_51, transpose_68, contiguous_68, linear_103, view_52, transpose_69, contiguous_69, linear_104, view_53, transpose_70, contiguous_70, scaled_dot_product_attention_17], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:230
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf925, buf930, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf931 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf925, 4, int_array_12, int_array_13, 0L)); buf925.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_102, mul_634, view_51, transpose_68, contiguous_68, linear_103, view_52, transpose_69, contiguous_69, linear_104, view_53, transpose_70, contiguous_70, scaled_dot_product_attention_17], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:231
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf928, model_audio_tower_layers_17_self_attn_v_proj_bias, buf931, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf928.reset();
	    // Topologically Sorted Source Nodes: [, linear_102, mul_634, view_51, transpose_68, contiguous_68, linear_103, view_52, transpose_69, contiguous_69, linear_104, view_53, transpose_70, contiguous_70, scaled_dot_product_attention_17], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_17 = 1.0;
	    AtenTensorHandle buf933_handle;
	    AtenTensorHandle buf934_handle;
	    AtenTensorHandle buf935_handle;
	    AtenTensorHandle buf936_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:549
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf929, buf930, buf931, nullptr, 0, 0.0, 0, &var_17, &buf933_handle, &buf934_handle, &buf935_handle, &buf936_handle));
	    RAIIAtenTensorHandle buf933(buf933_handle);
	    RAIIAtenTensorHandle buf934(buf934_handle);
	    RAIIAtenTensorHandle buf935(buf935_handle);
	    RAIIAtenTensorHandle buf936(buf936_handle);
	    buf929.reset();
	    buf930.reset();
	
	    auto buf939 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf933, 3, int_array_4, int_array_5, 0L)); buf933.reset();  // reuse
	    auto buf940 = std::move(buf939);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_71, reshape_17, choose_qparams_affine_default_105, quantize_affine_105, dequantize_affine_210], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf940, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf941 = std::move(buf927);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_211], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:233
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_17_self_attn_out_proj_parametrizations_weight_original1, buf941, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf942 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf931, 2, int_array_8, int_array_9, 0L)); buf931.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_105, dequantize_affine_210, dequantize_affine_211, linear_105, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:550
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf942, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf940, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf941, 2, int_array_10, int_array_11, 0L))));
	    buf941.reset();
	    auto buf943 = std::move(buf890);  // reuse
	    auto buf947 = std::move(buf940);  // reuse
	    auto buf950 = std::move(buf947);  // reuse
	    auto buf951 = std::move(buf950);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_101, add_238, linear_105, add_247, layer_norm_35, choose_qparams_affine_default_106, quantize_affine_106, dequantize_affine_212], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf943, buf951, buf906, model_audio_tower_layers_16_fc2_bias, buf942, model_audio_tower_layers_17_self_attn_out_proj_bias, model_audio_tower_layers_17_final_layer_norm_weight, model_audio_tower_layers_17_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf906.reset();
	    buf942.reset();
	    auto buf952 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf905, 3, int_array_14, int_array_7, 0L)); buf905.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_213], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:235
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_17_fc1_parametrizations_weight_original0, model_audio_tower_layers_17_fc1_parametrizations_weight_original2, model_audio_tower_layers_17_fc1_parametrizations_weight_original1, buf952, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf953_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf953_handle));
	    RAIIAtenTensorHandle buf953(buf953_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_106, dequantize_affine_212, dequantize_affine_213, linear_106, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:551
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf953, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf951, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf952, 2, int_array_17, int_array_11, 0L))));
	    buf951.reset();
	    buf952.reset();
	    auto buf956 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf953, 3, int_array_18, int_array_19, 0L)); buf953.reset();  // reuse
	    auto buf957 = std::move(buf956);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_106, gelu_19, choose_qparams_affine_default_107, quantize_affine_107, dequantize_affine_214], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf957, model_audio_tower_layers_17_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf958_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf958_handle));
	    RAIIAtenTensorHandle buf958(buf958_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_215], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:237
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_17_fc2_parametrizations_weight_original0, model_audio_tower_layers_17_fc2_parametrizations_weight_original2, model_audio_tower_layers_17_fc2_parametrizations_weight_original1, buf958, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf959_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf959_handle));
	    RAIIAtenTensorHandle buf959(buf959_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_107, quantize_affine_107, dequantize_affine_214, dequantize_affine_215, linear_107, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:552
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf959, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf957, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf958, 2, int_array_22, int_array_23, 0L))));
	    buf957.reset();
	    AtenTensorHandle buf963_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf963_handle));
	    RAIIAtenTensorHandle buf963(buf963_handle);
	    AtenTensorHandle buf966_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf966_handle));
	    RAIIAtenTensorHandle buf966(buf966_handle);
	    auto buf973 = std::move(buf966);  // reuse
	    AtenTensorHandle buf969_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf969_handle));
	    RAIIAtenTensorHandle buf969(buf969_handle);
	    auto buf976 = std::move(buf969);  // reuse
	    AtenTensorHandle buf972_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf972_handle));
	    RAIIAtenTensorHandle buf972(buf972_handle);
	    auto buf979 = std::move(buf972);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_107, add_252, layer_norm_36, choose_qparams_affine_default_108, quantize_affine_108, dequantize_affine_216, choose_qparams_affine_default_109, quantize_affine_109, dequantize_affine_218, choose_qparams_affine_default_110, quantize_affine_110, dequantize_affine_220], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf973, buf976, buf979, buf943, buf959, model_audio_tower_layers_17_fc2_bias, model_audio_tower_layers_18_self_attn_layer_norm_weight, model_audio_tower_layers_18_self_attn_layer_norm_bias, buf963, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf974_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf974_handle));
	    RAIIAtenTensorHandle buf974(buf974_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_217], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:239
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_18_self_attn_q_proj_parametrizations_weight_original1, buf974, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf975 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf963, 2, int_array_8, int_array_9, 0L)); buf963.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_108, dequantize_affine_216, dequantize_affine_217, linear_108, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:553
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf975, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf973, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf974, 2, int_array_10, int_array_11, 0L))));
	    auto buf977 = std::move(buf974);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_219], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:240
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_18_self_attn_k_proj_parametrizations_weight_original1, buf977, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf978 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf973, 2, int_array_8, int_array_9, 0L)); buf973.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_109, dequantize_affine_218, dequantize_affine_219, linear_109], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:554
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf978, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf976, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf977, 2, int_array_10, int_array_11, 0L))));
	    auto buf980 = std::move(buf977);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_221], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:241
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_18_self_attn_v_proj_parametrizations_weight_original1, buf980, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf981 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf976, 2, int_array_8, int_array_9, 0L)); buf976.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_110, dequantize_affine_220, dequantize_affine_221, linear_110, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:555
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf981, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf979, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf980, 2, int_array_10, int_array_11, 0L))));
	    auto buf982 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf979, 4, int_array_12, int_array_13, 0L)); buf979.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_108, mul_671, view_54, transpose_72, contiguous_72, linear_109, view_55, transpose_73, contiguous_73, linear_110, view_56, transpose_74, contiguous_74, scaled_dot_product_attention_18], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:242
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf975, model_audio_tower_layers_18_self_attn_q_proj_bias, buf982, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf983 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf975, 4, int_array_12, int_array_13, 0L)); buf975.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_108, mul_671, view_54, transpose_72, contiguous_72, linear_109, view_55, transpose_73, contiguous_73, linear_110, view_56, transpose_74, contiguous_74, scaled_dot_product_attention_18], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:243
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf978, buf983, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf984 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf978, 4, int_array_12, int_array_13, 0L)); buf978.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_108, mul_671, view_54, transpose_72, contiguous_72, linear_109, view_55, transpose_73, contiguous_73, linear_110, view_56, transpose_74, contiguous_74, scaled_dot_product_attention_18], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:244
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf981, model_audio_tower_layers_18_self_attn_v_proj_bias, buf984, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf981.reset();
	    // Topologically Sorted Source Nodes: [, linear_108, mul_671, view_54, transpose_72, contiguous_72, linear_109, view_55, transpose_73, contiguous_73, linear_110, view_56, transpose_74, contiguous_74, scaled_dot_product_attention_18], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_18 = 1.0;
	    AtenTensorHandle buf986_handle;
	    AtenTensorHandle buf987_handle;
	    AtenTensorHandle buf988_handle;
	    AtenTensorHandle buf989_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:556
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf982, buf983, buf984, nullptr, 0, 0.0, 0, &var_18, &buf986_handle, &buf987_handle, &buf988_handle, &buf989_handle));
	    RAIIAtenTensorHandle buf986(buf986_handle);
	    RAIIAtenTensorHandle buf987(buf987_handle);
	    RAIIAtenTensorHandle buf988(buf988_handle);
	    RAIIAtenTensorHandle buf989(buf989_handle);
	    buf982.reset();
	    buf983.reset();
	
	    auto buf992 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf986, 3, int_array_4, int_array_5, 0L)); buf986.reset();  // reuse
	    auto buf993 = std::move(buf992);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_75, reshape_18, choose_qparams_affine_default_111, quantize_affine_111, dequantize_affine_222], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf993, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf994 = std::move(buf980);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_223], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:246
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_18_self_attn_out_proj_parametrizations_weight_original1, buf994, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf995 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf984, 2, int_array_8, int_array_9, 0L)); buf984.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_111, dequantize_affine_222, dequantize_affine_223, linear_111, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:557
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf995, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf993, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf994, 2, int_array_10, int_array_11, 0L))));
	    buf994.reset();
	    auto buf996 = std::move(buf943);  // reuse
	    auto buf1000 = std::move(buf993);  // reuse
	    auto buf1003 = std::move(buf1000);  // reuse
	    auto buf1004 = std::move(buf1003);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_107, add_252, linear_111, add_261, layer_norm_37, choose_qparams_affine_default_112, quantize_affine_112, dequantize_affine_224], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf996, buf1004, buf959, model_audio_tower_layers_17_fc2_bias, buf995, model_audio_tower_layers_18_self_attn_out_proj_bias, model_audio_tower_layers_18_final_layer_norm_weight, model_audio_tower_layers_18_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf959.reset();
	    buf995.reset();
	    auto buf1005 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf958, 3, int_array_14, int_array_7, 0L)); buf958.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_225], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:248
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_18_fc1_parametrizations_weight_original0, model_audio_tower_layers_18_fc1_parametrizations_weight_original2, model_audio_tower_layers_18_fc1_parametrizations_weight_original1, buf1005, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1006_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1006_handle));
	    RAIIAtenTensorHandle buf1006(buf1006_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_112, dequantize_affine_224, dequantize_affine_225, linear_112, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:558
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1006, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1004, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1005, 2, int_array_17, int_array_11, 0L))));
	    buf1004.reset();
	    buf1005.reset();
	    auto buf1009 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1006, 3, int_array_18, int_array_19, 0L)); buf1006.reset();  // reuse
	    auto buf1010 = std::move(buf1009);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_112, gelu_20, choose_qparams_affine_default_113, quantize_affine_113, dequantize_affine_226], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf1010, model_audio_tower_layers_18_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1011_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1011_handle));
	    RAIIAtenTensorHandle buf1011(buf1011_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_227], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:250
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_18_fc2_parametrizations_weight_original0, model_audio_tower_layers_18_fc2_parametrizations_weight_original2, model_audio_tower_layers_18_fc2_parametrizations_weight_original1, buf1011, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1012_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1012_handle));
	    RAIIAtenTensorHandle buf1012(buf1012_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_113, quantize_affine_113, dequantize_affine_226, dequantize_affine_227, linear_113, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:559
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1012, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1010, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1011, 2, int_array_22, int_array_23, 0L))));
	    buf1010.reset();
	    AtenTensorHandle buf1016_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1016_handle));
	    RAIIAtenTensorHandle buf1016(buf1016_handle);
	    AtenTensorHandle buf1019_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1019_handle));
	    RAIIAtenTensorHandle buf1019(buf1019_handle);
	    auto buf1026 = std::move(buf1019);  // reuse
	    AtenTensorHandle buf1022_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1022_handle));
	    RAIIAtenTensorHandle buf1022(buf1022_handle);
	    auto buf1029 = std::move(buf1022);  // reuse
	    AtenTensorHandle buf1025_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1025_handle));
	    RAIIAtenTensorHandle buf1025(buf1025_handle);
	    auto buf1032 = std::move(buf1025);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_113, add_266, layer_norm_38, choose_qparams_affine_default_114, quantize_affine_114, dequantize_affine_228, choose_qparams_affine_default_115, quantize_affine_115, dequantize_affine_230, choose_qparams_affine_default_116, quantize_affine_116, dequantize_affine_232], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf1026, buf1029, buf1032, buf996, buf1012, model_audio_tower_layers_18_fc2_bias, model_audio_tower_layers_19_self_attn_layer_norm_weight, model_audio_tower_layers_19_self_attn_layer_norm_bias, buf1016, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1027_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1027_handle));
	    RAIIAtenTensorHandle buf1027(buf1027_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_229], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:252
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_19_self_attn_q_proj_parametrizations_weight_original1, buf1027, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1028 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1016, 2, int_array_8, int_array_9, 0L)); buf1016.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_114, dequantize_affine_228, dequantize_affine_229, linear_114, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:560
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1028, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1026, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1027, 2, int_array_10, int_array_11, 0L))));
	    auto buf1030 = std::move(buf1027);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_231], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:253
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_19_self_attn_k_proj_parametrizations_weight_original1, buf1030, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1031 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1026, 2, int_array_8, int_array_9, 0L)); buf1026.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_115, dequantize_affine_230, dequantize_affine_231, linear_115], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:561
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1031, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1029, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1030, 2, int_array_10, int_array_11, 0L))));
	    auto buf1033 = std::move(buf1030);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_233], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:254
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_19_self_attn_v_proj_parametrizations_weight_original1, buf1033, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1034 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1029, 2, int_array_8, int_array_9, 0L)); buf1029.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_116, dequantize_affine_232, dequantize_affine_233, linear_116, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:562
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1034, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1032, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1033, 2, int_array_10, int_array_11, 0L))));
	    auto buf1035 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1032, 4, int_array_12, int_array_13, 0L)); buf1032.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_114, mul_708, view_57, transpose_76, contiguous_76, linear_115, view_58, transpose_77, contiguous_77, linear_116, view_59, transpose_78, contiguous_78, scaled_dot_product_attention_19], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:255
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf1028, model_audio_tower_layers_19_self_attn_q_proj_bias, buf1035, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1036 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1028, 4, int_array_12, int_array_13, 0L)); buf1028.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_114, mul_708, view_57, transpose_76, contiguous_76, linear_115, view_58, transpose_77, contiguous_77, linear_116, view_59, transpose_78, contiguous_78, scaled_dot_product_attention_19], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:256
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf1031, buf1036, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1037 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1031, 4, int_array_12, int_array_13, 0L)); buf1031.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_114, mul_708, view_57, transpose_76, contiguous_76, linear_115, view_58, transpose_77, contiguous_77, linear_116, view_59, transpose_78, contiguous_78, scaled_dot_product_attention_19], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:257
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf1034, model_audio_tower_layers_19_self_attn_v_proj_bias, buf1037, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1034.reset();
	    // Topologically Sorted Source Nodes: [, linear_114, mul_708, view_57, transpose_76, contiguous_76, linear_115, view_58, transpose_77, contiguous_77, linear_116, view_59, transpose_78, contiguous_78, scaled_dot_product_attention_19], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_19 = 1.0;
	    AtenTensorHandle buf1039_handle;
	    AtenTensorHandle buf1040_handle;
	    AtenTensorHandle buf1041_handle;
	    AtenTensorHandle buf1042_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:563
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf1035, buf1036, buf1037, nullptr, 0, 0.0, 0, &var_19, &buf1039_handle, &buf1040_handle, &buf1041_handle, &buf1042_handle));
	    RAIIAtenTensorHandle buf1039(buf1039_handle);
	    RAIIAtenTensorHandle buf1040(buf1040_handle);
	    RAIIAtenTensorHandle buf1041(buf1041_handle);
	    RAIIAtenTensorHandle buf1042(buf1042_handle);
	    buf1035.reset();
	    buf1036.reset();
	
	    auto buf1045 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1039, 3, int_array_4, int_array_5, 0L)); buf1039.reset();  // reuse
	    auto buf1046 = std::move(buf1045);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_79, reshape_19, choose_qparams_affine_default_117, quantize_affine_117, dequantize_affine_234], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf1046, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1047 = std::move(buf1033);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_235], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:259
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_19_self_attn_out_proj_parametrizations_weight_original1, buf1047, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1048 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1037, 2, int_array_8, int_array_9, 0L)); buf1037.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_117, dequantize_affine_234, dequantize_affine_235, linear_117, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:564
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1048, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1046, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1047, 2, int_array_10, int_array_11, 0L))));
	    buf1047.reset();
	    auto buf1049 = std::move(buf996);  // reuse
	    auto buf1053 = std::move(buf1046);  // reuse
	    auto buf1056 = std::move(buf1053);  // reuse
	    auto buf1057 = std::move(buf1056);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_113, add_266, linear_117, add_275, layer_norm_39, choose_qparams_affine_default_118, quantize_affine_118, dequantize_affine_236], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf1049, buf1057, buf1012, model_audio_tower_layers_18_fc2_bias, buf1048, model_audio_tower_layers_19_self_attn_out_proj_bias, model_audio_tower_layers_19_final_layer_norm_weight, model_audio_tower_layers_19_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1012.reset();
	    buf1048.reset();
	    auto buf1058 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1011, 3, int_array_14, int_array_7, 0L)); buf1011.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_237], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:261
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_19_fc1_parametrizations_weight_original0, model_audio_tower_layers_19_fc1_parametrizations_weight_original2, model_audio_tower_layers_19_fc1_parametrizations_weight_original1, buf1058, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1059_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1059_handle));
	    RAIIAtenTensorHandle buf1059(buf1059_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_118, dequantize_affine_236, dequantize_affine_237, linear_118, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:565
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1059, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1057, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1058, 2, int_array_17, int_array_11, 0L))));
	    buf1057.reset();
	    buf1058.reset();
	    auto buf1062 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1059, 3, int_array_18, int_array_19, 0L)); buf1059.reset();  // reuse
	    auto buf1063 = std::move(buf1062);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_118, gelu_21, choose_qparams_affine_default_119, quantize_affine_119, dequantize_affine_238], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf1063, model_audio_tower_layers_19_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1064_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1064_handle));
	    RAIIAtenTensorHandle buf1064(buf1064_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_239], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:263
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_19_fc2_parametrizations_weight_original0, model_audio_tower_layers_19_fc2_parametrizations_weight_original2, model_audio_tower_layers_19_fc2_parametrizations_weight_original1, buf1064, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1065_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1065_handle));
	    RAIIAtenTensorHandle buf1065(buf1065_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_119, quantize_affine_119, dequantize_affine_238, dequantize_affine_239, linear_119, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:566
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1065, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1063, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1064, 2, int_array_22, int_array_23, 0L))));
	    buf1063.reset();
	    AtenTensorHandle buf1069_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1069_handle));
	    RAIIAtenTensorHandle buf1069(buf1069_handle);
	    AtenTensorHandle buf1072_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1072_handle));
	    RAIIAtenTensorHandle buf1072(buf1072_handle);
	    auto buf1079 = std::move(buf1072);  // reuse
	    AtenTensorHandle buf1075_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1075_handle));
	    RAIIAtenTensorHandle buf1075(buf1075_handle);
	    auto buf1082 = std::move(buf1075);  // reuse
	    AtenTensorHandle buf1078_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1078_handle));
	    RAIIAtenTensorHandle buf1078(buf1078_handle);
	    auto buf1085 = std::move(buf1078);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_119, add_280, layer_norm_40, choose_qparams_affine_default_120, quantize_affine_120, dequantize_affine_240, choose_qparams_affine_default_121, quantize_affine_121, dequantize_affine_242, choose_qparams_affine_default_122, quantize_affine_122, dequantize_affine_244], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf1079, buf1082, buf1085, buf1049, buf1065, model_audio_tower_layers_19_fc2_bias, model_audio_tower_layers_20_self_attn_layer_norm_weight, model_audio_tower_layers_20_self_attn_layer_norm_bias, buf1069, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1080_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1080_handle));
	    RAIIAtenTensorHandle buf1080(buf1080_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_241], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:265
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_20_self_attn_q_proj_parametrizations_weight_original1, buf1080, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1081 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1069, 2, int_array_8, int_array_9, 0L)); buf1069.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_120, dequantize_affine_240, dequantize_affine_241, linear_120, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:567
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1081, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1079, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1080, 2, int_array_10, int_array_11, 0L))));
	    auto buf1083 = std::move(buf1080);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_243], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:266
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_20_self_attn_k_proj_parametrizations_weight_original1, buf1083, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1084 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1079, 2, int_array_8, int_array_9, 0L)); buf1079.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_121, dequantize_affine_242, dequantize_affine_243, linear_121], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:568
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1084, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1082, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1083, 2, int_array_10, int_array_11, 0L))));
	    auto buf1086 = std::move(buf1083);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_245], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:267
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_20_self_attn_v_proj_parametrizations_weight_original1, buf1086, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1087 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1082, 2, int_array_8, int_array_9, 0L)); buf1082.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_122, dequantize_affine_244, dequantize_affine_245, linear_122, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:569
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1087, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1085, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1086, 2, int_array_10, int_array_11, 0L))));
	    auto buf1088 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1085, 4, int_array_12, int_array_13, 0L)); buf1085.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_120, mul_745, view_60, transpose_80, contiguous_80, linear_121, view_61, transpose_81, contiguous_81, linear_122, view_62, transpose_82, contiguous_82, scaled_dot_product_attention_20], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:268
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf1081, model_audio_tower_layers_20_self_attn_q_proj_bias, buf1088, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1089 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1081, 4, int_array_12, int_array_13, 0L)); buf1081.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_120, mul_745, view_60, transpose_80, contiguous_80, linear_121, view_61, transpose_81, contiguous_81, linear_122, view_62, transpose_82, contiguous_82, scaled_dot_product_attention_20], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:269
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf1084, buf1089, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1090 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1084, 4, int_array_12, int_array_13, 0L)); buf1084.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_120, mul_745, view_60, transpose_80, contiguous_80, linear_121, view_61, transpose_81, contiguous_81, linear_122, view_62, transpose_82, contiguous_82, scaled_dot_product_attention_20], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:270
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf1087, model_audio_tower_layers_20_self_attn_v_proj_bias, buf1090, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1087.reset();
	    // Topologically Sorted Source Nodes: [, linear_120, mul_745, view_60, transpose_80, contiguous_80, linear_121, view_61, transpose_81, contiguous_81, linear_122, view_62, transpose_82, contiguous_82, scaled_dot_product_attention_20], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_20 = 1.0;
	    AtenTensorHandle buf1092_handle;
	    AtenTensorHandle buf1093_handle;
	    AtenTensorHandle buf1094_handle;
	    AtenTensorHandle buf1095_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:570
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf1088, buf1089, buf1090, nullptr, 0, 0.0, 0, &var_20, &buf1092_handle, &buf1093_handle, &buf1094_handle, &buf1095_handle));
	    RAIIAtenTensorHandle buf1092(buf1092_handle);
	    RAIIAtenTensorHandle buf1093(buf1093_handle);
	    RAIIAtenTensorHandle buf1094(buf1094_handle);
	    RAIIAtenTensorHandle buf1095(buf1095_handle);
	    buf1088.reset();
	    buf1089.reset();
	
	    auto buf1098 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1092, 3, int_array_4, int_array_5, 0L)); buf1092.reset();  // reuse
	    auto buf1099 = std::move(buf1098);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_83, reshape_20, choose_qparams_affine_default_123, quantize_affine_123, dequantize_affine_246], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf1099, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1100 = std::move(buf1086);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_247], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:272
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_20_self_attn_out_proj_parametrizations_weight_original1, buf1100, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1101 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1090, 2, int_array_8, int_array_9, 0L)); buf1090.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_123, dequantize_affine_246, dequantize_affine_247, linear_123, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:571
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1101, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1099, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1100, 2, int_array_10, int_array_11, 0L))));
	    buf1100.reset();
	    auto buf1102 = std::move(buf1049);  // reuse
	    auto buf1106 = std::move(buf1099);  // reuse
	    auto buf1109 = std::move(buf1106);  // reuse
	    auto buf1110 = std::move(buf1109);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_119, add_280, linear_123, add_289, layer_norm_41, choose_qparams_affine_default_124, quantize_affine_124, dequantize_affine_248], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf1102, buf1110, buf1065, model_audio_tower_layers_19_fc2_bias, buf1101, model_audio_tower_layers_20_self_attn_out_proj_bias, model_audio_tower_layers_20_final_layer_norm_weight, model_audio_tower_layers_20_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1065.reset();
	    buf1101.reset();
	    auto buf1111 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1064, 3, int_array_14, int_array_7, 0L)); buf1064.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_249], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:274
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_20_fc1_parametrizations_weight_original0, model_audio_tower_layers_20_fc1_parametrizations_weight_original2, model_audio_tower_layers_20_fc1_parametrizations_weight_original1, buf1111, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1112_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1112_handle));
	    RAIIAtenTensorHandle buf1112(buf1112_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_124, dequantize_affine_248, dequantize_affine_249, linear_124, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:572
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1112, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1110, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1111, 2, int_array_17, int_array_11, 0L))));
	    buf1110.reset();
	    buf1111.reset();
	    auto buf1115 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1112, 3, int_array_18, int_array_19, 0L)); buf1112.reset();  // reuse
	    auto buf1116 = std::move(buf1115);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_124, gelu_22, choose_qparams_affine_default_125, quantize_affine_125, dequantize_affine_250], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf1116, model_audio_tower_layers_20_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1117_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1117_handle));
	    RAIIAtenTensorHandle buf1117(buf1117_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_251], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:276
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_20_fc2_parametrizations_weight_original0, model_audio_tower_layers_20_fc2_parametrizations_weight_original2, model_audio_tower_layers_20_fc2_parametrizations_weight_original1, buf1117, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1118_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1118_handle));
	    RAIIAtenTensorHandle buf1118(buf1118_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_125, quantize_affine_125, dequantize_affine_250, dequantize_affine_251, linear_125, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:573
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1118, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1116, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1117, 2, int_array_22, int_array_23, 0L))));
	    buf1116.reset();
	    AtenTensorHandle buf1122_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1122_handle));
	    RAIIAtenTensorHandle buf1122(buf1122_handle);
	    AtenTensorHandle buf1125_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1125_handle));
	    RAIIAtenTensorHandle buf1125(buf1125_handle);
	    auto buf1132 = std::move(buf1125);  // reuse
	    AtenTensorHandle buf1128_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1128_handle));
	    RAIIAtenTensorHandle buf1128(buf1128_handle);
	    auto buf1135 = std::move(buf1128);  // reuse
	    AtenTensorHandle buf1131_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1131_handle));
	    RAIIAtenTensorHandle buf1131(buf1131_handle);
	    auto buf1138 = std::move(buf1131);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_125, add_294, layer_norm_42, choose_qparams_affine_default_126, quantize_affine_126, dequantize_affine_252, choose_qparams_affine_default_127, quantize_affine_127, dequantize_affine_254, choose_qparams_affine_default_128, quantize_affine_128, dequantize_affine_256], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf1132, buf1135, buf1138, buf1102, buf1118, model_audio_tower_layers_20_fc2_bias, model_audio_tower_layers_21_self_attn_layer_norm_weight, model_audio_tower_layers_21_self_attn_layer_norm_bias, buf1122, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1133_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1133_handle));
	    RAIIAtenTensorHandle buf1133(buf1133_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_253], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:278
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_21_self_attn_q_proj_parametrizations_weight_original1, buf1133, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1134 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1122, 2, int_array_8, int_array_9, 0L)); buf1122.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_126, dequantize_affine_252, dequantize_affine_253, linear_126, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:574
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1134, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1132, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1133, 2, int_array_10, int_array_11, 0L))));
	    auto buf1136 = std::move(buf1133);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_255], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:279
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_21_self_attn_k_proj_parametrizations_weight_original1, buf1136, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1137 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1132, 2, int_array_8, int_array_9, 0L)); buf1132.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_127, dequantize_affine_254, dequantize_affine_255, linear_127], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:575
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1137, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1135, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1136, 2, int_array_10, int_array_11, 0L))));
	    auto buf1139 = std::move(buf1136);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_257], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:280
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_21_self_attn_v_proj_parametrizations_weight_original1, buf1139, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1140 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1135, 2, int_array_8, int_array_9, 0L)); buf1135.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_128, dequantize_affine_256, dequantize_affine_257, linear_128, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:576
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1140, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1138, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1139, 2, int_array_10, int_array_11, 0L))));
	    auto buf1141 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1138, 4, int_array_12, int_array_13, 0L)); buf1138.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_126, mul_782, view_63, transpose_84, contiguous_84, linear_127, view_64, transpose_85, contiguous_85, linear_128, view_65, transpose_86, contiguous_86, scaled_dot_product_attention_21], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:281
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf1134, model_audio_tower_layers_21_self_attn_q_proj_bias, buf1141, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1142 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1134, 4, int_array_12, int_array_13, 0L)); buf1134.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_126, mul_782, view_63, transpose_84, contiguous_84, linear_127, view_64, transpose_85, contiguous_85, linear_128, view_65, transpose_86, contiguous_86, scaled_dot_product_attention_21], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:282
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf1137, buf1142, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1143 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1137, 4, int_array_12, int_array_13, 0L)); buf1137.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_126, mul_782, view_63, transpose_84, contiguous_84, linear_127, view_64, transpose_85, contiguous_85, linear_128, view_65, transpose_86, contiguous_86, scaled_dot_product_attention_21], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:283
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf1140, model_audio_tower_layers_21_self_attn_v_proj_bias, buf1143, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1140.reset();
	    // Topologically Sorted Source Nodes: [, linear_126, mul_782, view_63, transpose_84, contiguous_84, linear_127, view_64, transpose_85, contiguous_85, linear_128, view_65, transpose_86, contiguous_86, scaled_dot_product_attention_21], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_21 = 1.0;
	    AtenTensorHandle buf1145_handle;
	    AtenTensorHandle buf1146_handle;
	    AtenTensorHandle buf1147_handle;
	    AtenTensorHandle buf1148_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:577
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf1141, buf1142, buf1143, nullptr, 0, 0.0, 0, &var_21, &buf1145_handle, &buf1146_handle, &buf1147_handle, &buf1148_handle));
	    RAIIAtenTensorHandle buf1145(buf1145_handle);
	    RAIIAtenTensorHandle buf1146(buf1146_handle);
	    RAIIAtenTensorHandle buf1147(buf1147_handle);
	    RAIIAtenTensorHandle buf1148(buf1148_handle);
	    buf1141.reset();
	    buf1142.reset();
	
	    auto buf1151 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1145, 3, int_array_4, int_array_5, 0L)); buf1145.reset();  // reuse
	    auto buf1152 = std::move(buf1151);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_87, reshape_21, choose_qparams_affine_default_129, quantize_affine_129, dequantize_affine_258], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf1152, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1153 = std::move(buf1139);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_259], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:285
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_21_self_attn_out_proj_parametrizations_weight_original1, buf1153, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1154 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1143, 2, int_array_8, int_array_9, 0L)); buf1143.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_129, dequantize_affine_258, dequantize_affine_259, linear_129, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:578
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1154, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1152, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1153, 2, int_array_10, int_array_11, 0L))));
	    buf1153.reset();
	    auto buf1155 = std::move(buf1102);  // reuse
	    auto buf1159 = std::move(buf1152);  // reuse
	    auto buf1162 = std::move(buf1159);  // reuse
	    auto buf1163 = std::move(buf1162);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_125, add_294, linear_129, add_303, layer_norm_43, choose_qparams_affine_default_130, quantize_affine_130, dequantize_affine_260], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf1155, buf1163, buf1118, model_audio_tower_layers_20_fc2_bias, buf1154, model_audio_tower_layers_21_self_attn_out_proj_bias, model_audio_tower_layers_21_final_layer_norm_weight, model_audio_tower_layers_21_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1118.reset();
	    buf1154.reset();
	    auto buf1164 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1117, 3, int_array_14, int_array_7, 0L)); buf1117.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_261], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:287
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_21_fc1_parametrizations_weight_original0, model_audio_tower_layers_21_fc1_parametrizations_weight_original2, model_audio_tower_layers_21_fc1_parametrizations_weight_original1, buf1164, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1165_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1165_handle));
	    RAIIAtenTensorHandle buf1165(buf1165_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_130, dequantize_affine_260, dequantize_affine_261, linear_130, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:579
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1165, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1163, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1164, 2, int_array_17, int_array_11, 0L))));
	    buf1163.reset();
	    buf1164.reset();
	    auto buf1168 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1165, 3, int_array_18, int_array_19, 0L)); buf1165.reset();  // reuse
	    auto buf1169 = std::move(buf1168);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_130, gelu_23, choose_qparams_affine_default_131, quantize_affine_131, dequantize_affine_262], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf1169, model_audio_tower_layers_21_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1170_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1170_handle));
	    RAIIAtenTensorHandle buf1170(buf1170_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_263], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:289
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_21_fc2_parametrizations_weight_original0, model_audio_tower_layers_21_fc2_parametrizations_weight_original2, model_audio_tower_layers_21_fc2_parametrizations_weight_original1, buf1170, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1171_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1171_handle));
	    RAIIAtenTensorHandle buf1171(buf1171_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_131, quantize_affine_131, dequantize_affine_262, dequantize_affine_263, linear_131, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:580
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1171, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1169, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1170, 2, int_array_22, int_array_23, 0L))));
	    buf1169.reset();
	    AtenTensorHandle buf1175_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1175_handle));
	    RAIIAtenTensorHandle buf1175(buf1175_handle);
	    AtenTensorHandle buf1178_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1178_handle));
	    RAIIAtenTensorHandle buf1178(buf1178_handle);
	    auto buf1185 = std::move(buf1178);  // reuse
	    AtenTensorHandle buf1181_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1181_handle));
	    RAIIAtenTensorHandle buf1181(buf1181_handle);
	    auto buf1188 = std::move(buf1181);  // reuse
	    AtenTensorHandle buf1184_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1184_handle));
	    RAIIAtenTensorHandle buf1184(buf1184_handle);
	    auto buf1191 = std::move(buf1184);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_131, add_308, layer_norm_44, choose_qparams_affine_default_132, quantize_affine_132, dequantize_affine_264, choose_qparams_affine_default_133, quantize_affine_133, dequantize_affine_266, choose_qparams_affine_default_134, quantize_affine_134, dequantize_affine_268], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf1185, buf1188, buf1191, buf1155, buf1171, model_audio_tower_layers_21_fc2_bias, model_audio_tower_layers_22_self_attn_layer_norm_weight, model_audio_tower_layers_22_self_attn_layer_norm_bias, buf1175, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1186_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1186_handle));
	    RAIIAtenTensorHandle buf1186(buf1186_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_265], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:291
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_22_self_attn_q_proj_parametrizations_weight_original1, buf1186, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1187 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1175, 2, int_array_8, int_array_9, 0L)); buf1175.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_132, dequantize_affine_264, dequantize_affine_265, linear_132, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:581
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1187, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1185, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1186, 2, int_array_10, int_array_11, 0L))));
	    auto buf1189 = std::move(buf1186);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_267], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:292
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_22_self_attn_k_proj_parametrizations_weight_original1, buf1189, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1190 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1185, 2, int_array_8, int_array_9, 0L)); buf1185.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_133, dequantize_affine_266, dequantize_affine_267, linear_133], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:582
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1190, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1188, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1189, 2, int_array_10, int_array_11, 0L))));
	    auto buf1192 = std::move(buf1189);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_269], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:293
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_22_self_attn_v_proj_parametrizations_weight_original1, buf1192, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1193 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1188, 2, int_array_8, int_array_9, 0L)); buf1188.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_134, dequantize_affine_268, dequantize_affine_269, linear_134, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:583
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1193, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1191, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1192, 2, int_array_10, int_array_11, 0L))));
	    auto buf1194 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1191, 4, int_array_12, int_array_13, 0L)); buf1191.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_132, mul_819, view_66, transpose_88, contiguous_88, linear_133, view_67, transpose_89, contiguous_89, linear_134, view_68, transpose_90, contiguous_90, scaled_dot_product_attention_22], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:294
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf1187, model_audio_tower_layers_22_self_attn_q_proj_bias, buf1194, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1195 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1187, 4, int_array_12, int_array_13, 0L)); buf1187.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_132, mul_819, view_66, transpose_88, contiguous_88, linear_133, view_67, transpose_89, contiguous_89, linear_134, view_68, transpose_90, contiguous_90, scaled_dot_product_attention_22], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:295
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf1190, buf1195, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1196 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1190, 4, int_array_12, int_array_13, 0L)); buf1190.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_132, mul_819, view_66, transpose_88, contiguous_88, linear_133, view_67, transpose_89, contiguous_89, linear_134, view_68, transpose_90, contiguous_90, scaled_dot_product_attention_22], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:296
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf1193, model_audio_tower_layers_22_self_attn_v_proj_bias, buf1196, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1193.reset();
	    // Topologically Sorted Source Nodes: [, linear_132, mul_819, view_66, transpose_88, contiguous_88, linear_133, view_67, transpose_89, contiguous_89, linear_134, view_68, transpose_90, contiguous_90, scaled_dot_product_attention_22], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_22 = 1.0;
	    AtenTensorHandle buf1198_handle;
	    AtenTensorHandle buf1199_handle;
	    AtenTensorHandle buf1200_handle;
	    AtenTensorHandle buf1201_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:584
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf1194, buf1195, buf1196, nullptr, 0, 0.0, 0, &var_22, &buf1198_handle, &buf1199_handle, &buf1200_handle, &buf1201_handle));
	    RAIIAtenTensorHandle buf1198(buf1198_handle);
	    RAIIAtenTensorHandle buf1199(buf1199_handle);
	    RAIIAtenTensorHandle buf1200(buf1200_handle);
	    RAIIAtenTensorHandle buf1201(buf1201_handle);
	    buf1194.reset();
	    buf1195.reset();
	
	    auto buf1204 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1198, 3, int_array_4, int_array_5, 0L)); buf1198.reset();  // reuse
	    auto buf1205 = std::move(buf1204);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_91, reshape_22, choose_qparams_affine_default_135, quantize_affine_135, dequantize_affine_270], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf1205, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1206 = std::move(buf1192);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_271], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:298
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_22_self_attn_out_proj_parametrizations_weight_original1, buf1206, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1207 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1196, 2, int_array_8, int_array_9, 0L)); buf1196.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_135, dequantize_affine_270, dequantize_affine_271, linear_135, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:585
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1207, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1205, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1206, 2, int_array_10, int_array_11, 0L))));
	    buf1206.reset();
	    auto buf1208 = std::move(buf1155);  // reuse
	    auto buf1212 = std::move(buf1205);  // reuse
	    auto buf1215 = std::move(buf1212);  // reuse
	    auto buf1216 = std::move(buf1215);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_131, add_308, linear_135, add_317, layer_norm_45, choose_qparams_affine_default_136, quantize_affine_136, dequantize_affine_272], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf1208, buf1216, buf1171, model_audio_tower_layers_21_fc2_bias, buf1207, model_audio_tower_layers_22_self_attn_out_proj_bias, model_audio_tower_layers_22_final_layer_norm_weight, model_audio_tower_layers_22_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1171.reset();
	    buf1207.reset();
	    auto buf1217 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1170, 3, int_array_14, int_array_7, 0L)); buf1170.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_273], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:300
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_22_fc1_parametrizations_weight_original0, model_audio_tower_layers_22_fc1_parametrizations_weight_original2, model_audio_tower_layers_22_fc1_parametrizations_weight_original1, buf1217, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1218_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1218_handle));
	    RAIIAtenTensorHandle buf1218(buf1218_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_136, dequantize_affine_272, dequantize_affine_273, linear_136, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:586
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1218, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1216, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1217, 2, int_array_17, int_array_11, 0L))));
	    buf1216.reset();
	    buf1217.reset();
	    auto buf1221 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1218, 3, int_array_18, int_array_19, 0L)); buf1218.reset();  // reuse
	    auto buf1222 = std::move(buf1221);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_136, gelu_24, choose_qparams_affine_default_137, quantize_affine_137, dequantize_affine_274], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf1222, model_audio_tower_layers_22_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1223_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1223_handle));
	    RAIIAtenTensorHandle buf1223(buf1223_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_275], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:302
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_22_fc2_parametrizations_weight_original0, model_audio_tower_layers_22_fc2_parametrizations_weight_original2, model_audio_tower_layers_22_fc2_parametrizations_weight_original1, buf1223, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1224_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1224_handle));
	    RAIIAtenTensorHandle buf1224(buf1224_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_137, quantize_affine_137, dequantize_affine_274, dequantize_affine_275, linear_137, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:587
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1224, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1222, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1223, 2, int_array_22, int_array_23, 0L))));
	    buf1222.reset();
	    AtenTensorHandle buf1228_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1228_handle));
	    RAIIAtenTensorHandle buf1228(buf1228_handle);
	    AtenTensorHandle buf1231_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1231_handle));
	    RAIIAtenTensorHandle buf1231(buf1231_handle);
	    auto buf1238 = std::move(buf1231);  // reuse
	    AtenTensorHandle buf1234_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1234_handle));
	    RAIIAtenTensorHandle buf1234(buf1234_handle);
	    auto buf1241 = std::move(buf1234);  // reuse
	    AtenTensorHandle buf1237_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1237_handle));
	    RAIIAtenTensorHandle buf1237(buf1237_handle);
	    auto buf1244 = std::move(buf1237);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_137, add_322, layer_norm_46, choose_qparams_affine_default_138, quantize_affine_138, dequantize_affine_276, choose_qparams_affine_default_139, quantize_affine_139, dequantize_affine_278, choose_qparams_affine_default_140, quantize_affine_140, dequantize_affine_280], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf1238, buf1241, buf1244, buf1208, buf1224, model_audio_tower_layers_22_fc2_bias, model_audio_tower_layers_23_self_attn_layer_norm_weight, model_audio_tower_layers_23_self_attn_layer_norm_bias, buf1228, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1239_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1239_handle));
	    RAIIAtenTensorHandle buf1239(buf1239_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_277], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:304
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_23_self_attn_q_proj_parametrizations_weight_original1, buf1239, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1240 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1228, 2, int_array_8, int_array_9, 0L)); buf1228.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_138, dequantize_affine_276, dequantize_affine_277, linear_138, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:588
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1240, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1238, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1239, 2, int_array_10, int_array_11, 0L))));
	    auto buf1242 = std::move(buf1239);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_279], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:305
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_23_self_attn_k_proj_parametrizations_weight_original1, buf1242, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1243 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1238, 2, int_array_8, int_array_9, 0L)); buf1238.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_139, dequantize_affine_278, dequantize_affine_279, linear_139], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:589
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1243, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1241, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1242, 2, int_array_10, int_array_11, 0L))));
	    auto buf1245 = std::move(buf1242);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_281], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:306
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_23_self_attn_v_proj_parametrizations_weight_original1, buf1245, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1246 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1241, 2, int_array_8, int_array_9, 0L)); buf1241.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_140, dequantize_affine_280, dequantize_affine_281, linear_140, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:590
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1246, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1244, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1245, 2, int_array_10, int_array_11, 0L))));
	    auto buf1247 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1244, 4, int_array_12, int_array_13, 0L)); buf1244.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_138, mul_856, view_69, transpose_92, contiguous_92, linear_139, view_70, transpose_93, contiguous_93, linear_140, view_71, transpose_94, contiguous_94, scaled_dot_product_attention_23], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:307
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf1240, model_audio_tower_layers_23_self_attn_q_proj_bias, buf1247, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1248 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1240, 4, int_array_12, int_array_13, 0L)); buf1240.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_138, mul_856, view_69, transpose_92, contiguous_92, linear_139, view_70, transpose_93, contiguous_93, linear_140, view_71, transpose_94, contiguous_94, scaled_dot_product_attention_23], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:308
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf1243, buf1248, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1249 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1243, 4, int_array_12, int_array_13, 0L)); buf1243.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_138, mul_856, view_69, transpose_92, contiguous_92, linear_139, view_70, transpose_93, contiguous_93, linear_140, view_71, transpose_94, contiguous_94, scaled_dot_product_attention_23], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:309
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf1246, model_audio_tower_layers_23_self_attn_v_proj_bias, buf1249, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1246.reset();
	    // Topologically Sorted Source Nodes: [, linear_138, mul_856, view_69, transpose_92, contiguous_92, linear_139, view_70, transpose_93, contiguous_93, linear_140, view_71, transpose_94, contiguous_94, scaled_dot_product_attention_23], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_23 = 1.0;
	    AtenTensorHandle buf1251_handle;
	    AtenTensorHandle buf1252_handle;
	    AtenTensorHandle buf1253_handle;
	    AtenTensorHandle buf1254_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:591
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf1247, buf1248, buf1249, nullptr, 0, 0.0, 0, &var_23, &buf1251_handle, &buf1252_handle, &buf1253_handle, &buf1254_handle));
	    RAIIAtenTensorHandle buf1251(buf1251_handle);
	    RAIIAtenTensorHandle buf1252(buf1252_handle);
	    RAIIAtenTensorHandle buf1253(buf1253_handle);
	    RAIIAtenTensorHandle buf1254(buf1254_handle);
	    buf1247.reset();
	    buf1248.reset();
	
	    auto buf1257 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1251, 3, int_array_4, int_array_5, 0L)); buf1251.reset();  // reuse
	    auto buf1258 = std::move(buf1257);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_95, reshape_23, choose_qparams_affine_default_141, quantize_affine_141, dequantize_affine_282], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf1258, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1259 = std::move(buf1245);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_283], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:311
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_23_self_attn_out_proj_parametrizations_weight_original1, buf1259, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1260 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1249, 2, int_array_8, int_array_9, 0L)); buf1249.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_141, dequantize_affine_282, dequantize_affine_283, linear_141, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:592
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1260, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1258, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1259, 2, int_array_10, int_array_11, 0L))));
	    buf1259.reset();
	    auto buf1261 = std::move(buf1208);  // reuse
	    auto buf1265 = std::move(buf1258);  // reuse
	    auto buf1268 = std::move(buf1265);  // reuse
	    auto buf1269 = std::move(buf1268);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_137, add_322, linear_141, add_331, layer_norm_47, choose_qparams_affine_default_142, quantize_affine_142, dequantize_affine_284], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf1261, buf1269, buf1224, model_audio_tower_layers_22_fc2_bias, buf1260, model_audio_tower_layers_23_self_attn_out_proj_bias, model_audio_tower_layers_23_final_layer_norm_weight, model_audio_tower_layers_23_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1224.reset();
	    buf1260.reset();
	    auto buf1270 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1223, 3, int_array_14, int_array_7, 0L)); buf1223.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_285], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:313
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_23_fc1_parametrizations_weight_original0, model_audio_tower_layers_23_fc1_parametrizations_weight_original2, model_audio_tower_layers_23_fc1_parametrizations_weight_original1, buf1270, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1271_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1271_handle));
	    RAIIAtenTensorHandle buf1271(buf1271_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_142, dequantize_affine_284, dequantize_affine_285, linear_142, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:593
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1271, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1269, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1270, 2, int_array_17, int_array_11, 0L))));
	    buf1269.reset();
	    buf1270.reset();
	    auto buf1274 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1271, 3, int_array_18, int_array_19, 0L)); buf1271.reset();  // reuse
	    auto buf1275 = std::move(buf1274);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_142, gelu_25, choose_qparams_affine_default_143, quantize_affine_143, dequantize_affine_286], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf1275, model_audio_tower_layers_23_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1276_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1276_handle));
	    RAIIAtenTensorHandle buf1276(buf1276_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_287], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:315
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_23_fc2_parametrizations_weight_original0, model_audio_tower_layers_23_fc2_parametrizations_weight_original2, model_audio_tower_layers_23_fc2_parametrizations_weight_original1, buf1276, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1277_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1277_handle));
	    RAIIAtenTensorHandle buf1277(buf1277_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_143, quantize_affine_143, dequantize_affine_286, dequantize_affine_287, linear_143, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:594
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1277, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1275, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1276, 2, int_array_22, int_array_23, 0L))));
	    buf1275.reset();
	    AtenTensorHandle buf1281_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1281_handle));
	    RAIIAtenTensorHandle buf1281(buf1281_handle);
	    AtenTensorHandle buf1284_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1284_handle));
	    RAIIAtenTensorHandle buf1284(buf1284_handle);
	    auto buf1291 = std::move(buf1284);  // reuse
	    AtenTensorHandle buf1287_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1287_handle));
	    RAIIAtenTensorHandle buf1287(buf1287_handle);
	    auto buf1294 = std::move(buf1287);  // reuse
	    AtenTensorHandle buf1290_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1290_handle));
	    RAIIAtenTensorHandle buf1290(buf1290_handle);
	    auto buf1297 = std::move(buf1290);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_143, add_336, layer_norm_48, choose_qparams_affine_default_144, quantize_affine_144, dequantize_affine_288, choose_qparams_affine_default_145, quantize_affine_145, dequantize_affine_290, choose_qparams_affine_default_146, quantize_affine_146, dequantize_affine_292], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf1291, buf1294, buf1297, buf1261, buf1277, model_audio_tower_layers_23_fc2_bias, model_audio_tower_layers_24_self_attn_layer_norm_weight, model_audio_tower_layers_24_self_attn_layer_norm_bias, buf1281, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1292_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1292_handle));
	    RAIIAtenTensorHandle buf1292(buf1292_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_289], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:317
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_24_self_attn_q_proj_parametrizations_weight_original1, buf1292, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1293 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1281, 2, int_array_8, int_array_9, 0L)); buf1281.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_144, dequantize_affine_288, dequantize_affine_289, linear_144, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:595
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1293, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1291, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1292, 2, int_array_10, int_array_11, 0L))));
	    auto buf1295 = std::move(buf1292);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_291], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:318
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_24_self_attn_k_proj_parametrizations_weight_original1, buf1295, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1296 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1291, 2, int_array_8, int_array_9, 0L)); buf1291.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_145, dequantize_affine_290, dequantize_affine_291, linear_145], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:596
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1296, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1294, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1295, 2, int_array_10, int_array_11, 0L))));
	    auto buf1298 = std::move(buf1295);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_293], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:319
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_24_self_attn_v_proj_parametrizations_weight_original1, buf1298, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1299 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1294, 2, int_array_8, int_array_9, 0L)); buf1294.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_146, dequantize_affine_292, dequantize_affine_293, linear_146, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:597
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1299, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1297, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1298, 2, int_array_10, int_array_11, 0L))));
	    auto buf1300 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1297, 4, int_array_12, int_array_13, 0L)); buf1297.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_144, mul_893, view_72, transpose_96, contiguous_96, linear_145, view_73, transpose_97, contiguous_97, linear_146, view_74, transpose_98, contiguous_98, scaled_dot_product_attention_24], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:320
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf1293, model_audio_tower_layers_24_self_attn_q_proj_bias, buf1300, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1301 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1293, 4, int_array_12, int_array_13, 0L)); buf1293.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_144, mul_893, view_72, transpose_96, contiguous_96, linear_145, view_73, transpose_97, contiguous_97, linear_146, view_74, transpose_98, contiguous_98, scaled_dot_product_attention_24], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:321
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf1296, buf1301, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1302 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1296, 4, int_array_12, int_array_13, 0L)); buf1296.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_144, mul_893, view_72, transpose_96, contiguous_96, linear_145, view_73, transpose_97, contiguous_97, linear_146, view_74, transpose_98, contiguous_98, scaled_dot_product_attention_24], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:322
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf1299, model_audio_tower_layers_24_self_attn_v_proj_bias, buf1302, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1299.reset();
	    // Topologically Sorted Source Nodes: [, linear_144, mul_893, view_72, transpose_96, contiguous_96, linear_145, view_73, transpose_97, contiguous_97, linear_146, view_74, transpose_98, contiguous_98, scaled_dot_product_attention_24], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_24 = 1.0;
	    AtenTensorHandle buf1304_handle;
	    AtenTensorHandle buf1305_handle;
	    AtenTensorHandle buf1306_handle;
	    AtenTensorHandle buf1307_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:598
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf1300, buf1301, buf1302, nullptr, 0, 0.0, 0, &var_24, &buf1304_handle, &buf1305_handle, &buf1306_handle, &buf1307_handle));
	    RAIIAtenTensorHandle buf1304(buf1304_handle);
	    RAIIAtenTensorHandle buf1305(buf1305_handle);
	    RAIIAtenTensorHandle buf1306(buf1306_handle);
	    RAIIAtenTensorHandle buf1307(buf1307_handle);
	    buf1300.reset();
	    buf1301.reset();
	
	    auto buf1310 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1304, 3, int_array_4, int_array_5, 0L)); buf1304.reset();  // reuse
	    auto buf1311 = std::move(buf1310);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_99, reshape_24, choose_qparams_affine_default_147, quantize_affine_147, dequantize_affine_294], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf1311, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1312 = std::move(buf1298);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_295], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:324
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_24_self_attn_out_proj_parametrizations_weight_original1, buf1312, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1313 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1302, 2, int_array_8, int_array_9, 0L)); buf1302.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_147, dequantize_affine_294, dequantize_affine_295, linear_147, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:599
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1313, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1311, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1312, 2, int_array_10, int_array_11, 0L))));
	    buf1312.reset();
	    auto buf1314 = std::move(buf1261);  // reuse
	    auto buf1318 = std::move(buf1311);  // reuse
	    auto buf1321 = std::move(buf1318);  // reuse
	    auto buf1322 = std::move(buf1321);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_143, add_336, linear_147, add_345, layer_norm_49, choose_qparams_affine_default_148, quantize_affine_148, dequantize_affine_296], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf1314, buf1322, buf1277, model_audio_tower_layers_23_fc2_bias, buf1313, model_audio_tower_layers_24_self_attn_out_proj_bias, model_audio_tower_layers_24_final_layer_norm_weight, model_audio_tower_layers_24_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1277.reset();
	    buf1313.reset();
	    auto buf1323 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1276, 3, int_array_14, int_array_7, 0L)); buf1276.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_297], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:326
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_24_fc1_parametrizations_weight_original0, model_audio_tower_layers_24_fc1_parametrizations_weight_original2, model_audio_tower_layers_24_fc1_parametrizations_weight_original1, buf1323, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1324_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1324_handle));
	    RAIIAtenTensorHandle buf1324(buf1324_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_148, dequantize_affine_296, dequantize_affine_297, linear_148, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:600
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1324, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1322, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1323, 2, int_array_17, int_array_11, 0L))));
	    buf1322.reset();
	    buf1323.reset();
	    auto buf1327 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1324, 3, int_array_18, int_array_19, 0L)); buf1324.reset();  // reuse
	    auto buf1328 = std::move(buf1327);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_148, gelu_26, choose_qparams_affine_default_149, quantize_affine_149, dequantize_affine_298], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf1328, model_audio_tower_layers_24_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1329_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1329_handle));
	    RAIIAtenTensorHandle buf1329(buf1329_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_299], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:328
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_24_fc2_parametrizations_weight_original0, model_audio_tower_layers_24_fc2_parametrizations_weight_original2, model_audio_tower_layers_24_fc2_parametrizations_weight_original1, buf1329, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1330_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1330_handle));
	    RAIIAtenTensorHandle buf1330(buf1330_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_149, quantize_affine_149, dequantize_affine_298, dequantize_affine_299, linear_149, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:601
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1330, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1328, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1329, 2, int_array_22, int_array_23, 0L))));
	    buf1328.reset();
	    AtenTensorHandle buf1334_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1334_handle));
	    RAIIAtenTensorHandle buf1334(buf1334_handle);
	    AtenTensorHandle buf1337_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1337_handle));
	    RAIIAtenTensorHandle buf1337(buf1337_handle);
	    auto buf1344 = std::move(buf1337);  // reuse
	    AtenTensorHandle buf1340_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1340_handle));
	    RAIIAtenTensorHandle buf1340(buf1340_handle);
	    auto buf1347 = std::move(buf1340);  // reuse
	    AtenTensorHandle buf1343_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1343_handle));
	    RAIIAtenTensorHandle buf1343(buf1343_handle);
	    auto buf1350 = std::move(buf1343);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_149, add_350, layer_norm_50, choose_qparams_affine_default_150, quantize_affine_150, dequantize_affine_300, choose_qparams_affine_default_151, quantize_affine_151, dequantize_affine_302, choose_qparams_affine_default_152, quantize_affine_152, dequantize_affine_304], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf1344, buf1347, buf1350, buf1314, buf1330, model_audio_tower_layers_24_fc2_bias, model_audio_tower_layers_25_self_attn_layer_norm_weight, model_audio_tower_layers_25_self_attn_layer_norm_bias, buf1334, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1345_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1345_handle));
	    RAIIAtenTensorHandle buf1345(buf1345_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_301], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:330
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_25_self_attn_q_proj_parametrizations_weight_original1, buf1345, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1346 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1334, 2, int_array_8, int_array_9, 0L)); buf1334.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_150, dequantize_affine_300, dequantize_affine_301, linear_150, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:602
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1346, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1344, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1345, 2, int_array_10, int_array_11, 0L))));
	    auto buf1348 = std::move(buf1345);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_303], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:331
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_25_self_attn_k_proj_parametrizations_weight_original1, buf1348, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1349 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1344, 2, int_array_8, int_array_9, 0L)); buf1344.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_151, dequantize_affine_302, dequantize_affine_303, linear_151], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:603
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1349, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1347, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1348, 2, int_array_10, int_array_11, 0L))));
	    auto buf1351 = std::move(buf1348);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_305], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:332
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_25_self_attn_v_proj_parametrizations_weight_original1, buf1351, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1352 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1347, 2, int_array_8, int_array_9, 0L)); buf1347.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_152, dequantize_affine_304, dequantize_affine_305, linear_152, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:604
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1352, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1350, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1351, 2, int_array_10, int_array_11, 0L))));
	    auto buf1353 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1350, 4, int_array_12, int_array_13, 0L)); buf1350.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_150, mul_930, view_75, transpose_100, contiguous_100, linear_151, view_76, transpose_101, contiguous_101, linear_152, view_77, transpose_102, contiguous_102, scaled_dot_product_attention_25], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:333
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf1346, model_audio_tower_layers_25_self_attn_q_proj_bias, buf1353, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1354 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1346, 4, int_array_12, int_array_13, 0L)); buf1346.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_150, mul_930, view_75, transpose_100, contiguous_100, linear_151, view_76, transpose_101, contiguous_101, linear_152, view_77, transpose_102, contiguous_102, scaled_dot_product_attention_25], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:334
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf1349, buf1354, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1355 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1349, 4, int_array_12, int_array_13, 0L)); buf1349.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_150, mul_930, view_75, transpose_100, contiguous_100, linear_151, view_76, transpose_101, contiguous_101, linear_152, view_77, transpose_102, contiguous_102, scaled_dot_product_attention_25], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:335
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf1352, model_audio_tower_layers_25_self_attn_v_proj_bias, buf1355, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1352.reset();
	    // Topologically Sorted Source Nodes: [, linear_150, mul_930, view_75, transpose_100, contiguous_100, linear_151, view_76, transpose_101, contiguous_101, linear_152, view_77, transpose_102, contiguous_102, scaled_dot_product_attention_25], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_25 = 1.0;
	    AtenTensorHandle buf1357_handle;
	    AtenTensorHandle buf1358_handle;
	    AtenTensorHandle buf1359_handle;
	    AtenTensorHandle buf1360_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:605
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf1353, buf1354, buf1355, nullptr, 0, 0.0, 0, &var_25, &buf1357_handle, &buf1358_handle, &buf1359_handle, &buf1360_handle));
	    RAIIAtenTensorHandle buf1357(buf1357_handle);
	    RAIIAtenTensorHandle buf1358(buf1358_handle);
	    RAIIAtenTensorHandle buf1359(buf1359_handle);
	    RAIIAtenTensorHandle buf1360(buf1360_handle);
	    buf1353.reset();
	    buf1354.reset();
	
	    auto buf1363 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1357, 3, int_array_4, int_array_5, 0L)); buf1357.reset();  // reuse
	    auto buf1364 = std::move(buf1363);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_103, reshape_25, choose_qparams_affine_default_153, quantize_affine_153, dequantize_affine_306], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf1364, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1365 = std::move(buf1351);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_307], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:337
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_25_self_attn_out_proj_parametrizations_weight_original1, buf1365, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1366 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1355, 2, int_array_8, int_array_9, 0L)); buf1355.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_153, dequantize_affine_306, dequantize_affine_307, linear_153, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:606
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1366, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1364, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1365, 2, int_array_10, int_array_11, 0L))));
	    buf1365.reset();
	    auto buf1367 = std::move(buf1314);  // reuse
	    auto buf1371 = std::move(buf1364);  // reuse
	    auto buf1374 = std::move(buf1371);  // reuse
	    auto buf1375 = std::move(buf1374);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_149, add_350, linear_153, add_359, layer_norm_51, choose_qparams_affine_default_154, quantize_affine_154, dequantize_affine_308], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf1367, buf1375, buf1330, model_audio_tower_layers_24_fc2_bias, buf1366, model_audio_tower_layers_25_self_attn_out_proj_bias, model_audio_tower_layers_25_final_layer_norm_weight, model_audio_tower_layers_25_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1330.reset();
	    buf1366.reset();
	    auto buf1376 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1329, 3, int_array_14, int_array_7, 0L)); buf1329.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_309], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:339
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_25_fc1_parametrizations_weight_original0, model_audio_tower_layers_25_fc1_parametrizations_weight_original2, model_audio_tower_layers_25_fc1_parametrizations_weight_original1, buf1376, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1377_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1377_handle));
	    RAIIAtenTensorHandle buf1377(buf1377_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_154, dequantize_affine_308, dequantize_affine_309, linear_154, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:607
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1377, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1375, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1376, 2, int_array_17, int_array_11, 0L))));
	    buf1375.reset();
	    buf1376.reset();
	    auto buf1380 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1377, 3, int_array_18, int_array_19, 0L)); buf1377.reset();  // reuse
	    auto buf1381 = std::move(buf1380);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_154, gelu_27, choose_qparams_affine_default_155, quantize_affine_155, dequantize_affine_310], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf1381, model_audio_tower_layers_25_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1382_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1382_handle));
	    RAIIAtenTensorHandle buf1382(buf1382_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_311], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:341
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_25_fc2_parametrizations_weight_original0, model_audio_tower_layers_25_fc2_parametrizations_weight_original2, model_audio_tower_layers_25_fc2_parametrizations_weight_original1, buf1382, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1383_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1383_handle));
	    RAIIAtenTensorHandle buf1383(buf1383_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_155, quantize_affine_155, dequantize_affine_310, dequantize_affine_311, linear_155, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:608
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1383, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1381, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1382, 2, int_array_22, int_array_23, 0L))));
	    buf1381.reset();
	    AtenTensorHandle buf1387_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1387_handle));
	    RAIIAtenTensorHandle buf1387(buf1387_handle);
	    AtenTensorHandle buf1390_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1390_handle));
	    RAIIAtenTensorHandle buf1390(buf1390_handle);
	    auto buf1397 = std::move(buf1390);  // reuse
	    AtenTensorHandle buf1393_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1393_handle));
	    RAIIAtenTensorHandle buf1393(buf1393_handle);
	    auto buf1400 = std::move(buf1393);  // reuse
	    AtenTensorHandle buf1396_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1396_handle));
	    RAIIAtenTensorHandle buf1396(buf1396_handle);
	    auto buf1403 = std::move(buf1396);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_155, add_364, layer_norm_52, choose_qparams_affine_default_156, quantize_affine_156, dequantize_affine_312, choose_qparams_affine_default_157, quantize_affine_157, dequantize_affine_314, choose_qparams_affine_default_158, quantize_affine_158, dequantize_affine_316], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf1397, buf1400, buf1403, buf1367, buf1383, model_audio_tower_layers_25_fc2_bias, model_audio_tower_layers_26_self_attn_layer_norm_weight, model_audio_tower_layers_26_self_attn_layer_norm_bias, buf1387, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1398_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1398_handle));
	    RAIIAtenTensorHandle buf1398(buf1398_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_313], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:343
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_26_self_attn_q_proj_parametrizations_weight_original1, buf1398, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1399 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1387, 2, int_array_8, int_array_9, 0L)); buf1387.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_156, dequantize_affine_312, dequantize_affine_313, linear_156, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:609
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1399, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1397, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1398, 2, int_array_10, int_array_11, 0L))));
	    auto buf1401 = std::move(buf1398);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_315], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:344
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_26_self_attn_k_proj_parametrizations_weight_original1, buf1401, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1402 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1397, 2, int_array_8, int_array_9, 0L)); buf1397.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_157, dequantize_affine_314, dequantize_affine_315, linear_157], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:610
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1402, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1400, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1401, 2, int_array_10, int_array_11, 0L))));
	    auto buf1404 = std::move(buf1401);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_317], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:345
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_26_self_attn_v_proj_parametrizations_weight_original1, buf1404, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1405 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1400, 2, int_array_8, int_array_9, 0L)); buf1400.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_158, dequantize_affine_316, dequantize_affine_317, linear_158, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:611
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1405, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1403, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1404, 2, int_array_10, int_array_11, 0L))));
	    auto buf1406 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1403, 4, int_array_12, int_array_13, 0L)); buf1403.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_156, mul_967, view_78, transpose_104, contiguous_104, linear_157, view_79, transpose_105, contiguous_105, linear_158, view_80, transpose_106, contiguous_106, scaled_dot_product_attention_26], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:346
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf1399, model_audio_tower_layers_26_self_attn_q_proj_bias, buf1406, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1407 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1399, 4, int_array_12, int_array_13, 0L)); buf1399.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_156, mul_967, view_78, transpose_104, contiguous_104, linear_157, view_79, transpose_105, contiguous_105, linear_158, view_80, transpose_106, contiguous_106, scaled_dot_product_attention_26], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:347
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf1402, buf1407, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1408 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1402, 4, int_array_12, int_array_13, 0L)); buf1402.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_156, mul_967, view_78, transpose_104, contiguous_104, linear_157, view_79, transpose_105, contiguous_105, linear_158, view_80, transpose_106, contiguous_106, scaled_dot_product_attention_26], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:348
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf1405, model_audio_tower_layers_26_self_attn_v_proj_bias, buf1408, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1405.reset();
	    // Topologically Sorted Source Nodes: [, linear_156, mul_967, view_78, transpose_104, contiguous_104, linear_157, view_79, transpose_105, contiguous_105, linear_158, view_80, transpose_106, contiguous_106, scaled_dot_product_attention_26], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_26 = 1.0;
	    AtenTensorHandle buf1410_handle;
	    AtenTensorHandle buf1411_handle;
	    AtenTensorHandle buf1412_handle;
	    AtenTensorHandle buf1413_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:612
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf1406, buf1407, buf1408, nullptr, 0, 0.0, 0, &var_26, &buf1410_handle, &buf1411_handle, &buf1412_handle, &buf1413_handle));
	    RAIIAtenTensorHandle buf1410(buf1410_handle);
	    RAIIAtenTensorHandle buf1411(buf1411_handle);
	    RAIIAtenTensorHandle buf1412(buf1412_handle);
	    RAIIAtenTensorHandle buf1413(buf1413_handle);
	    buf1406.reset();
	    buf1407.reset();
	
	    auto buf1416 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1410, 3, int_array_4, int_array_5, 0L)); buf1410.reset();  // reuse
	    auto buf1417 = std::move(buf1416);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_107, reshape_26, choose_qparams_affine_default_159, quantize_affine_159, dequantize_affine_318], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf1417, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1418 = std::move(buf1404);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_319], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:350
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_26_self_attn_out_proj_parametrizations_weight_original1, buf1418, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1419 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1408, 2, int_array_8, int_array_9, 0L)); buf1408.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_159, dequantize_affine_318, dequantize_affine_319, linear_159, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:613
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1419, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1417, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1418, 2, int_array_10, int_array_11, 0L))));
	    buf1418.reset();
	    auto buf1420 = std::move(buf1367);  // reuse
	    auto buf1424 = std::move(buf1417);  // reuse
	    auto buf1427 = std::move(buf1424);  // reuse
	    auto buf1428 = std::move(buf1427);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_155, add_364, linear_159, add_373, layer_norm_53, choose_qparams_affine_default_160, quantize_affine_160, dequantize_affine_320], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf1420, buf1428, buf1383, model_audio_tower_layers_25_fc2_bias, buf1419, model_audio_tower_layers_26_self_attn_out_proj_bias, model_audio_tower_layers_26_final_layer_norm_weight, model_audio_tower_layers_26_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1383.reset();
	    buf1419.reset();
	    auto buf1429 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1382, 3, int_array_14, int_array_7, 0L)); buf1382.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_321], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:352
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_26_fc1_parametrizations_weight_original0, model_audio_tower_layers_26_fc1_parametrizations_weight_original2, model_audio_tower_layers_26_fc1_parametrizations_weight_original1, buf1429, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1430_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1430_handle));
	    RAIIAtenTensorHandle buf1430(buf1430_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_160, dequantize_affine_320, dequantize_affine_321, linear_160, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:614
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1430, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1428, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1429, 2, int_array_17, int_array_11, 0L))));
	    buf1428.reset();
	    buf1429.reset();
	    auto buf1433 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1430, 3, int_array_18, int_array_19, 0L)); buf1430.reset();  // reuse
	    auto buf1434 = std::move(buf1433);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_160, gelu_28, choose_qparams_affine_default_161, quantize_affine_161, dequantize_affine_322], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf1434, model_audio_tower_layers_26_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1435_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1435_handle));
	    RAIIAtenTensorHandle buf1435(buf1435_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_323], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:354
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_26_fc2_parametrizations_weight_original0, model_audio_tower_layers_26_fc2_parametrizations_weight_original2, model_audio_tower_layers_26_fc2_parametrizations_weight_original1, buf1435, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1436_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1436_handle));
	    RAIIAtenTensorHandle buf1436(buf1436_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_161, quantize_affine_161, dequantize_affine_322, dequantize_affine_323, linear_161, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:615
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1436, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1434, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1435, 2, int_array_22, int_array_23, 0L))));
	    buf1434.reset();
	    AtenTensorHandle buf1440_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1440_handle));
	    RAIIAtenTensorHandle buf1440(buf1440_handle);
	    AtenTensorHandle buf1443_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1443_handle));
	    RAIIAtenTensorHandle buf1443(buf1443_handle);
	    auto buf1450 = std::move(buf1443);  // reuse
	    AtenTensorHandle buf1446_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1446_handle));
	    RAIIAtenTensorHandle buf1446(buf1446_handle);
	    auto buf1453 = std::move(buf1446);  // reuse
	    AtenTensorHandle buf1449_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1449_handle));
	    RAIIAtenTensorHandle buf1449(buf1449_handle);
	    auto buf1456 = std::move(buf1449);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_161, add_378, layer_norm_54, choose_qparams_affine_default_162, quantize_affine_162, dequantize_affine_324, choose_qparams_affine_default_163, quantize_affine_163, dequantize_affine_326, choose_qparams_affine_default_164, quantize_affine_164, dequantize_affine_328], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf1450, buf1453, buf1456, buf1420, buf1436, model_audio_tower_layers_26_fc2_bias, model_audio_tower_layers_27_self_attn_layer_norm_weight, model_audio_tower_layers_27_self_attn_layer_norm_bias, buf1440, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1451_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1451_handle));
	    RAIIAtenTensorHandle buf1451(buf1451_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_325], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:356
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_27_self_attn_q_proj_parametrizations_weight_original1, buf1451, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1452 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1440, 2, int_array_8, int_array_9, 0L)); buf1440.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_162, dequantize_affine_324, dequantize_affine_325, linear_162, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:616
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1452, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1450, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1451, 2, int_array_10, int_array_11, 0L))));
	    auto buf1454 = std::move(buf1451);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_327], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:357
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_27_self_attn_k_proj_parametrizations_weight_original1, buf1454, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1455 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1450, 2, int_array_8, int_array_9, 0L)); buf1450.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_163, dequantize_affine_326, dequantize_affine_327, linear_163], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:617
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1455, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1453, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1454, 2, int_array_10, int_array_11, 0L))));
	    auto buf1457 = std::move(buf1454);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_329], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:358
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_27_self_attn_v_proj_parametrizations_weight_original1, buf1457, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1458 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1453, 2, int_array_8, int_array_9, 0L)); buf1453.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_164, dequantize_affine_328, dequantize_affine_329, linear_164, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:618
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1458, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1456, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1457, 2, int_array_10, int_array_11, 0L))));
	    auto buf1459 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1456, 4, int_array_12, int_array_13, 0L)); buf1456.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_162, mul_1004, view_81, transpose_108, contiguous_108, linear_163, view_82, transpose_109, contiguous_109, linear_164, view_83, transpose_110, contiguous_110, scaled_dot_product_attention_27], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:359
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf1452, model_audio_tower_layers_27_self_attn_q_proj_bias, buf1459, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1460 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1452, 4, int_array_12, int_array_13, 0L)); buf1452.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_162, mul_1004, view_81, transpose_108, contiguous_108, linear_163, view_82, transpose_109, contiguous_109, linear_164, view_83, transpose_110, contiguous_110, scaled_dot_product_attention_27], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:360
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf1455, buf1460, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1461 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1455, 4, int_array_12, int_array_13, 0L)); buf1455.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_162, mul_1004, view_81, transpose_108, contiguous_108, linear_163, view_82, transpose_109, contiguous_109, linear_164, view_83, transpose_110, contiguous_110, scaled_dot_product_attention_27], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:361
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf1458, model_audio_tower_layers_27_self_attn_v_proj_bias, buf1461, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1458.reset();
	    // Topologically Sorted Source Nodes: [, linear_162, mul_1004, view_81, transpose_108, contiguous_108, linear_163, view_82, transpose_109, contiguous_109, linear_164, view_83, transpose_110, contiguous_110, scaled_dot_product_attention_27], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_27 = 1.0;
	    AtenTensorHandle buf1463_handle;
	    AtenTensorHandle buf1464_handle;
	    AtenTensorHandle buf1465_handle;
	    AtenTensorHandle buf1466_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:619
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf1459, buf1460, buf1461, nullptr, 0, 0.0, 0, &var_27, &buf1463_handle, &buf1464_handle, &buf1465_handle, &buf1466_handle));
	    RAIIAtenTensorHandle buf1463(buf1463_handle);
	    RAIIAtenTensorHandle buf1464(buf1464_handle);
	    RAIIAtenTensorHandle buf1465(buf1465_handle);
	    RAIIAtenTensorHandle buf1466(buf1466_handle);
	    buf1459.reset();
	    buf1460.reset();
	
	    auto buf1469 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1463, 3, int_array_4, int_array_5, 0L)); buf1463.reset();  // reuse
	    auto buf1470 = std::move(buf1469);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_111, reshape_27, choose_qparams_affine_default_165, quantize_affine_165, dequantize_affine_330], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf1470, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1471 = std::move(buf1457);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_331], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:363
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_27_self_attn_out_proj_parametrizations_weight_original1, buf1471, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1472 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1461, 2, int_array_8, int_array_9, 0L)); buf1461.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_165, dequantize_affine_330, dequantize_affine_331, linear_165, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:620
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1472, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1470, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1471, 2, int_array_10, int_array_11, 0L))));
	    buf1471.reset();
	    auto buf1473 = std::move(buf1420);  // reuse
	    auto buf1477 = std::move(buf1470);  // reuse
	    auto buf1480 = std::move(buf1477);  // reuse
	    auto buf1481 = std::move(buf1480);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_161, add_378, linear_165, add_387, layer_norm_55, choose_qparams_affine_default_166, quantize_affine_166, dequantize_affine_332], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf1473, buf1481, buf1436, model_audio_tower_layers_26_fc2_bias, buf1472, model_audio_tower_layers_27_self_attn_out_proj_bias, model_audio_tower_layers_27_final_layer_norm_weight, model_audio_tower_layers_27_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1436.reset();
	    buf1472.reset();
	    auto buf1482 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1435, 3, int_array_14, int_array_7, 0L)); buf1435.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_333], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:365
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_27_fc1_parametrizations_weight_original0, model_audio_tower_layers_27_fc1_parametrizations_weight_original2, model_audio_tower_layers_27_fc1_parametrizations_weight_original1, buf1482, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1483_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1483_handle));
	    RAIIAtenTensorHandle buf1483(buf1483_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_166, dequantize_affine_332, dequantize_affine_333, linear_166, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:621
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1483, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1481, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1482, 2, int_array_17, int_array_11, 0L))));
	    buf1481.reset();
	    buf1482.reset();
	    auto buf1486 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1483, 3, int_array_18, int_array_19, 0L)); buf1483.reset();  // reuse
	    auto buf1487 = std::move(buf1486);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_166, gelu_29, choose_qparams_affine_default_167, quantize_affine_167, dequantize_affine_334], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf1487, model_audio_tower_layers_27_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1488_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1488_handle));
	    RAIIAtenTensorHandle buf1488(buf1488_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_335], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:367
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_27_fc2_parametrizations_weight_original0, model_audio_tower_layers_27_fc2_parametrizations_weight_original2, model_audio_tower_layers_27_fc2_parametrizations_weight_original1, buf1488, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1489_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1489_handle));
	    RAIIAtenTensorHandle buf1489(buf1489_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_167, quantize_affine_167, dequantize_affine_334, dequantize_affine_335, linear_167, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:622
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1489, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1487, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1488, 2, int_array_22, int_array_23, 0L))));
	    buf1487.reset();
	    AtenTensorHandle buf1493_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1493_handle));
	    RAIIAtenTensorHandle buf1493(buf1493_handle);
	    AtenTensorHandle buf1496_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1496_handle));
	    RAIIAtenTensorHandle buf1496(buf1496_handle);
	    auto buf1503 = std::move(buf1496);  // reuse
	    AtenTensorHandle buf1499_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1499_handle));
	    RAIIAtenTensorHandle buf1499(buf1499_handle);
	    auto buf1506 = std::move(buf1499);  // reuse
	    AtenTensorHandle buf1502_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1502_handle));
	    RAIIAtenTensorHandle buf1502(buf1502_handle);
	    auto buf1509 = std::move(buf1502);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_167, add_392, layer_norm_56, choose_qparams_affine_default_168, quantize_affine_168, dequantize_affine_336, choose_qparams_affine_default_169, quantize_affine_169, dequantize_affine_338, choose_qparams_affine_default_170, quantize_affine_170, dequantize_affine_340], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf1503, buf1506, buf1509, buf1473, buf1489, model_audio_tower_layers_27_fc2_bias, model_audio_tower_layers_28_self_attn_layer_norm_weight, model_audio_tower_layers_28_self_attn_layer_norm_bias, buf1493, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1504_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1504_handle));
	    RAIIAtenTensorHandle buf1504(buf1504_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_337], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:369
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_28_self_attn_q_proj_parametrizations_weight_original1, buf1504, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1505 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1493, 2, int_array_8, int_array_9, 0L)); buf1493.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_168, dequantize_affine_336, dequantize_affine_337, linear_168, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:623
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1505, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1503, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1504, 2, int_array_10, int_array_11, 0L))));
	    auto buf1507 = std::move(buf1504);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_339], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:370
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_28_self_attn_k_proj_parametrizations_weight_original1, buf1507, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1508 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1503, 2, int_array_8, int_array_9, 0L)); buf1503.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_169, dequantize_affine_338, dequantize_affine_339, linear_169], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:624
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1508, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1506, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1507, 2, int_array_10, int_array_11, 0L))));
	    auto buf1510 = std::move(buf1507);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_341], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:371
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_28_self_attn_v_proj_parametrizations_weight_original1, buf1510, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1511 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1506, 2, int_array_8, int_array_9, 0L)); buf1506.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_170, dequantize_affine_340, dequantize_affine_341, linear_170, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:625
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1511, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1509, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1510, 2, int_array_10, int_array_11, 0L))));
	    auto buf1512 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1509, 4, int_array_12, int_array_13, 0L)); buf1509.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_168, mul_1041, view_84, transpose_112, contiguous_112, linear_169, view_85, transpose_113, contiguous_113, linear_170, view_86, transpose_114, contiguous_114, scaled_dot_product_attention_28], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:372
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf1505, model_audio_tower_layers_28_self_attn_q_proj_bias, buf1512, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1513 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1505, 4, int_array_12, int_array_13, 0L)); buf1505.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_168, mul_1041, view_84, transpose_112, contiguous_112, linear_169, view_85, transpose_113, contiguous_113, linear_170, view_86, transpose_114, contiguous_114, scaled_dot_product_attention_28], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:373
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf1508, buf1513, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1514 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1508, 4, int_array_12, int_array_13, 0L)); buf1508.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_168, mul_1041, view_84, transpose_112, contiguous_112, linear_169, view_85, transpose_113, contiguous_113, linear_170, view_86, transpose_114, contiguous_114, scaled_dot_product_attention_28], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:374
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf1511, model_audio_tower_layers_28_self_attn_v_proj_bias, buf1514, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1511.reset();
	    // Topologically Sorted Source Nodes: [, linear_168, mul_1041, view_84, transpose_112, contiguous_112, linear_169, view_85, transpose_113, contiguous_113, linear_170, view_86, transpose_114, contiguous_114, scaled_dot_product_attention_28], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_28 = 1.0;
	    AtenTensorHandle buf1516_handle;
	    AtenTensorHandle buf1517_handle;
	    AtenTensorHandle buf1518_handle;
	    AtenTensorHandle buf1519_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:626
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf1512, buf1513, buf1514, nullptr, 0, 0.0, 0, &var_28, &buf1516_handle, &buf1517_handle, &buf1518_handle, &buf1519_handle));
	    RAIIAtenTensorHandle buf1516(buf1516_handle);
	    RAIIAtenTensorHandle buf1517(buf1517_handle);
	    RAIIAtenTensorHandle buf1518(buf1518_handle);
	    RAIIAtenTensorHandle buf1519(buf1519_handle);
	    buf1512.reset();
	    buf1513.reset();
	
	    auto buf1522 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1516, 3, int_array_4, int_array_5, 0L)); buf1516.reset();  // reuse
	    auto buf1523 = std::move(buf1522);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_115, reshape_28, choose_qparams_affine_default_171, quantize_affine_171, dequantize_affine_342], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf1523, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1524 = std::move(buf1510);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_343], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:376
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_28_self_attn_out_proj_parametrizations_weight_original1, buf1524, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1525 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1514, 2, int_array_8, int_array_9, 0L)); buf1514.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_171, dequantize_affine_342, dequantize_affine_343, linear_171, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:627
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1525, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1523, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1524, 2, int_array_10, int_array_11, 0L))));
	    buf1524.reset();
	    auto buf1526 = std::move(buf1473);  // reuse
	    auto buf1530 = std::move(buf1523);  // reuse
	    auto buf1533 = std::move(buf1530);  // reuse
	    auto buf1534 = std::move(buf1533);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_167, add_392, linear_171, add_401, layer_norm_57, choose_qparams_affine_default_172, quantize_affine_172, dequantize_affine_344], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf1526, buf1534, buf1489, model_audio_tower_layers_27_fc2_bias, buf1525, model_audio_tower_layers_28_self_attn_out_proj_bias, model_audio_tower_layers_28_final_layer_norm_weight, model_audio_tower_layers_28_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1489.reset();
	    buf1525.reset();
	    auto buf1535 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1488, 3, int_array_14, int_array_7, 0L)); buf1488.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_345], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:378
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_28_fc1_parametrizations_weight_original0, model_audio_tower_layers_28_fc1_parametrizations_weight_original2, model_audio_tower_layers_28_fc1_parametrizations_weight_original1, buf1535, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1536_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1536_handle));
	    RAIIAtenTensorHandle buf1536(buf1536_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_172, dequantize_affine_344, dequantize_affine_345, linear_172, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:628
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1536, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1534, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1535, 2, int_array_17, int_array_11, 0L))));
	    buf1534.reset();
	    buf1535.reset();
	    auto buf1539 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1536, 3, int_array_18, int_array_19, 0L)); buf1536.reset();  // reuse
	    auto buf1540 = std::move(buf1539);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_172, gelu_30, choose_qparams_affine_default_173, quantize_affine_173, dequantize_affine_346], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf1540, model_audio_tower_layers_28_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1541_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1541_handle));
	    RAIIAtenTensorHandle buf1541(buf1541_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_347], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:380
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_28_fc2_parametrizations_weight_original0, model_audio_tower_layers_28_fc2_parametrizations_weight_original2, model_audio_tower_layers_28_fc2_parametrizations_weight_original1, buf1541, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1542_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1542_handle));
	    RAIIAtenTensorHandle buf1542(buf1542_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_173, quantize_affine_173, dequantize_affine_346, dequantize_affine_347, linear_173, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:629
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1542, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1540, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1541, 2, int_array_22, int_array_23, 0L))));
	    buf1540.reset();
	    AtenTensorHandle buf1546_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1546_handle));
	    RAIIAtenTensorHandle buf1546(buf1546_handle);
	    AtenTensorHandle buf1549_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1549_handle));
	    RAIIAtenTensorHandle buf1549(buf1549_handle);
	    auto buf1556 = std::move(buf1549);  // reuse
	    AtenTensorHandle buf1552_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1552_handle));
	    RAIIAtenTensorHandle buf1552(buf1552_handle);
	    auto buf1559 = std::move(buf1552);  // reuse
	    AtenTensorHandle buf1555_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1555_handle));
	    RAIIAtenTensorHandle buf1555(buf1555_handle);
	    auto buf1562 = std::move(buf1555);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_173, add_406, layer_norm_58, choose_qparams_affine_default_174, quantize_affine_174, dequantize_affine_348, choose_qparams_affine_default_175, quantize_affine_175, dequantize_affine_350, choose_qparams_affine_default_176, quantize_affine_176, dequantize_affine_352], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf1556, buf1559, buf1562, buf1526, buf1542, model_audio_tower_layers_28_fc2_bias, model_audio_tower_layers_29_self_attn_layer_norm_weight, model_audio_tower_layers_29_self_attn_layer_norm_bias, buf1546, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1557_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1557_handle));
	    RAIIAtenTensorHandle buf1557(buf1557_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_349], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:382
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_29_self_attn_q_proj_parametrizations_weight_original1, buf1557, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1558 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1546, 2, int_array_8, int_array_9, 0L)); buf1546.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_174, dequantize_affine_348, dequantize_affine_349, linear_174, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:630
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1558, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1556, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1557, 2, int_array_10, int_array_11, 0L))));
	    auto buf1560 = std::move(buf1557);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_351], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:383
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_29_self_attn_k_proj_parametrizations_weight_original1, buf1560, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1561 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1556, 2, int_array_8, int_array_9, 0L)); buf1556.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_175, dequantize_affine_350, dequantize_affine_351, linear_175], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:631
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1561, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1559, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1560, 2, int_array_10, int_array_11, 0L))));
	    auto buf1563 = std::move(buf1560);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_353], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:384
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_29_self_attn_v_proj_parametrizations_weight_original1, buf1563, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1564 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1559, 2, int_array_8, int_array_9, 0L)); buf1559.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_176, dequantize_affine_352, dequantize_affine_353, linear_176, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:632
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1564, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1562, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1563, 2, int_array_10, int_array_11, 0L))));
	    auto buf1565 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1562, 4, int_array_12, int_array_13, 0L)); buf1562.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_174, mul_1078, view_87, transpose_116, contiguous_116, linear_175, view_88, transpose_117, contiguous_117, linear_176, view_89, transpose_118, contiguous_118, scaled_dot_product_attention_29], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:385
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf1558, model_audio_tower_layers_29_self_attn_q_proj_bias, buf1565, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1566 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1558, 4, int_array_12, int_array_13, 0L)); buf1558.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_174, mul_1078, view_87, transpose_116, contiguous_116, linear_175, view_88, transpose_117, contiguous_117, linear_176, view_89, transpose_118, contiguous_118, scaled_dot_product_attention_29], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:386
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf1561, buf1566, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1567 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1561, 4, int_array_12, int_array_13, 0L)); buf1561.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_174, mul_1078, view_87, transpose_116, contiguous_116, linear_175, view_88, transpose_117, contiguous_117, linear_176, view_89, transpose_118, contiguous_118, scaled_dot_product_attention_29], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:387
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf1564, model_audio_tower_layers_29_self_attn_v_proj_bias, buf1567, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1564.reset();
	    // Topologically Sorted Source Nodes: [, linear_174, mul_1078, view_87, transpose_116, contiguous_116, linear_175, view_88, transpose_117, contiguous_117, linear_176, view_89, transpose_118, contiguous_118, scaled_dot_product_attention_29], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_29 = 1.0;
	    AtenTensorHandle buf1569_handle;
	    AtenTensorHandle buf1570_handle;
	    AtenTensorHandle buf1571_handle;
	    AtenTensorHandle buf1572_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:633
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf1565, buf1566, buf1567, nullptr, 0, 0.0, 0, &var_29, &buf1569_handle, &buf1570_handle, &buf1571_handle, &buf1572_handle));
	    RAIIAtenTensorHandle buf1569(buf1569_handle);
	    RAIIAtenTensorHandle buf1570(buf1570_handle);
	    RAIIAtenTensorHandle buf1571(buf1571_handle);
	    RAIIAtenTensorHandle buf1572(buf1572_handle);
	    buf1565.reset();
	    buf1566.reset();
	
	    auto buf1575 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1569, 3, int_array_4, int_array_5, 0L)); buf1569.reset();  // reuse
	    auto buf1576 = std::move(buf1575);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_119, reshape_29, choose_qparams_affine_default_177, quantize_affine_177, dequantize_affine_354], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf1576, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1577 = std::move(buf1563);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_355], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:389
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_29_self_attn_out_proj_parametrizations_weight_original1, buf1577, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1578 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1567, 2, int_array_8, int_array_9, 0L)); buf1567.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_177, dequantize_affine_354, dequantize_affine_355, linear_177, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:634
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1578, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1576, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1577, 2, int_array_10, int_array_11, 0L))));
	    buf1577.reset();
	    auto buf1579 = std::move(buf1526);  // reuse
	    auto buf1583 = std::move(buf1576);  // reuse
	    auto buf1586 = std::move(buf1583);  // reuse
	    auto buf1587 = std::move(buf1586);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_173, add_406, linear_177, add_415, layer_norm_59, choose_qparams_affine_default_178, quantize_affine_178, dequantize_affine_356], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf1579, buf1587, buf1542, model_audio_tower_layers_28_fc2_bias, buf1578, model_audio_tower_layers_29_self_attn_out_proj_bias, model_audio_tower_layers_29_final_layer_norm_weight, model_audio_tower_layers_29_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1542.reset();
	    buf1578.reset();
	    auto buf1588 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1541, 3, int_array_14, int_array_7, 0L)); buf1541.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_357], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:391
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_29_fc1_parametrizations_weight_original0, model_audio_tower_layers_29_fc1_parametrizations_weight_original2, model_audio_tower_layers_29_fc1_parametrizations_weight_original1, buf1588, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1589_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1589_handle));
	    RAIIAtenTensorHandle buf1589(buf1589_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_178, dequantize_affine_356, dequantize_affine_357, linear_178, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:635
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1589, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1587, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1588, 2, int_array_17, int_array_11, 0L))));
	    buf1587.reset();
	    buf1588.reset();
	    auto buf1592 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1589, 3, int_array_18, int_array_19, 0L)); buf1589.reset();  // reuse
	    auto buf1593 = std::move(buf1592);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_178, gelu_31, choose_qparams_affine_default_179, quantize_affine_179, dequantize_affine_358], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf1593, model_audio_tower_layers_29_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1594_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1594_handle));
	    RAIIAtenTensorHandle buf1594(buf1594_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_359], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:393
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_29_fc2_parametrizations_weight_original0, model_audio_tower_layers_29_fc2_parametrizations_weight_original2, model_audio_tower_layers_29_fc2_parametrizations_weight_original1, buf1594, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1595_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1595_handle));
	    RAIIAtenTensorHandle buf1595(buf1595_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_179, quantize_affine_179, dequantize_affine_358, dequantize_affine_359, linear_179, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:636
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1595, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1593, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1594, 2, int_array_22, int_array_23, 0L))));
	    buf1593.reset();
	    AtenTensorHandle buf1599_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1599_handle));
	    RAIIAtenTensorHandle buf1599(buf1599_handle);
	    AtenTensorHandle buf1602_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1602_handle));
	    RAIIAtenTensorHandle buf1602(buf1602_handle);
	    auto buf1609 = std::move(buf1602);  // reuse
	    AtenTensorHandle buf1605_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1605_handle));
	    RAIIAtenTensorHandle buf1605(buf1605_handle);
	    auto buf1612 = std::move(buf1605);  // reuse
	    AtenTensorHandle buf1608_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1608_handle));
	    RAIIAtenTensorHandle buf1608(buf1608_handle);
	    auto buf1615 = std::move(buf1608);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_179, add_420, layer_norm_60, choose_qparams_affine_default_180, quantize_affine_180, dequantize_affine_360, choose_qparams_affine_default_181, quantize_affine_181, dequantize_affine_362, choose_qparams_affine_default_182, quantize_affine_182, dequantize_affine_364], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf1609, buf1612, buf1615, buf1579, buf1595, model_audio_tower_layers_29_fc2_bias, model_audio_tower_layers_30_self_attn_layer_norm_weight, model_audio_tower_layers_30_self_attn_layer_norm_bias, buf1599, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1610_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1610_handle));
	    RAIIAtenTensorHandle buf1610(buf1610_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_361], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:395
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_30_self_attn_q_proj_parametrizations_weight_original1, buf1610, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1611 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1599, 2, int_array_8, int_array_9, 0L)); buf1599.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_180, dequantize_affine_360, dequantize_affine_361, linear_180, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:637
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1611, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1609, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1610, 2, int_array_10, int_array_11, 0L))));
	    auto buf1613 = std::move(buf1610);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_363], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:396
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_30_self_attn_k_proj_parametrizations_weight_original1, buf1613, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1614 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1609, 2, int_array_8, int_array_9, 0L)); buf1609.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_181, dequantize_affine_362, dequantize_affine_363, linear_181], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:638
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1614, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1612, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1613, 2, int_array_10, int_array_11, 0L))));
	    auto buf1616 = std::move(buf1613);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_365], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:397
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_30_self_attn_v_proj_parametrizations_weight_original1, buf1616, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1617 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1612, 2, int_array_8, int_array_9, 0L)); buf1612.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_182, dequantize_affine_364, dequantize_affine_365, linear_182, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:639
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1617, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1615, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1616, 2, int_array_10, int_array_11, 0L))));
	    auto buf1618 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1615, 4, int_array_12, int_array_13, 0L)); buf1615.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_180, mul_1115, view_90, transpose_120, contiguous_120, linear_181, view_91, transpose_121, contiguous_121, linear_182, view_92, transpose_122, contiguous_122, scaled_dot_product_attention_30], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:398
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf1611, model_audio_tower_layers_30_self_attn_q_proj_bias, buf1618, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1619 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1611, 4, int_array_12, int_array_13, 0L)); buf1611.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_180, mul_1115, view_90, transpose_120, contiguous_120, linear_181, view_91, transpose_121, contiguous_121, linear_182, view_92, transpose_122, contiguous_122, scaled_dot_product_attention_30], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:399
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf1614, buf1619, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1620 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1614, 4, int_array_12, int_array_13, 0L)); buf1614.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_180, mul_1115, view_90, transpose_120, contiguous_120, linear_181, view_91, transpose_121, contiguous_121, linear_182, view_92, transpose_122, contiguous_122, scaled_dot_product_attention_30], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:400
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf1617, model_audio_tower_layers_30_self_attn_v_proj_bias, buf1620, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1617.reset();
	    // Topologically Sorted Source Nodes: [, linear_180, mul_1115, view_90, transpose_120, contiguous_120, linear_181, view_91, transpose_121, contiguous_121, linear_182, view_92, transpose_122, contiguous_122, scaled_dot_product_attention_30], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_30 = 1.0;
	    AtenTensorHandle buf1622_handle;
	    AtenTensorHandle buf1623_handle;
	    AtenTensorHandle buf1624_handle;
	    AtenTensorHandle buf1625_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:640
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf1618, buf1619, buf1620, nullptr, 0, 0.0, 0, &var_30, &buf1622_handle, &buf1623_handle, &buf1624_handle, &buf1625_handle));
	    RAIIAtenTensorHandle buf1622(buf1622_handle);
	    RAIIAtenTensorHandle buf1623(buf1623_handle);
	    RAIIAtenTensorHandle buf1624(buf1624_handle);
	    RAIIAtenTensorHandle buf1625(buf1625_handle);
	    buf1618.reset();
	    buf1619.reset();
	
	    auto buf1628 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1622, 3, int_array_4, int_array_5, 0L)); buf1622.reset();  // reuse
	    auto buf1629 = std::move(buf1628);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_123, reshape_30, choose_qparams_affine_default_183, quantize_affine_183, dequantize_affine_366], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf1629, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1630 = std::move(buf1616);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_367], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:402
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_30_self_attn_out_proj_parametrizations_weight_original1, buf1630, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1631 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1620, 2, int_array_8, int_array_9, 0L)); buf1620.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_183, dequantize_affine_366, dequantize_affine_367, linear_183, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:641
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1631, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1629, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1630, 2, int_array_10, int_array_11, 0L))));
	    buf1630.reset();
	    auto buf1632 = std::move(buf1579);  // reuse
	    auto buf1636 = std::move(buf1629);  // reuse
	    auto buf1639 = std::move(buf1636);  // reuse
	    auto buf1640 = std::move(buf1639);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_179, add_420, linear_183, add_429, layer_norm_61, choose_qparams_affine_default_184, quantize_affine_184, dequantize_affine_368], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf1632, buf1640, buf1595, model_audio_tower_layers_29_fc2_bias, buf1631, model_audio_tower_layers_30_self_attn_out_proj_bias, model_audio_tower_layers_30_final_layer_norm_weight, model_audio_tower_layers_30_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1595.reset();
	    buf1631.reset();
	    auto buf1641 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1594, 3, int_array_14, int_array_7, 0L)); buf1594.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_369], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:404
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_30_fc1_parametrizations_weight_original0, model_audio_tower_layers_30_fc1_parametrizations_weight_original2, model_audio_tower_layers_30_fc1_parametrizations_weight_original1, buf1641, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1642_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1642_handle));
	    RAIIAtenTensorHandle buf1642(buf1642_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_184, dequantize_affine_368, dequantize_affine_369, linear_184, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:642
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1642, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1640, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1641, 2, int_array_17, int_array_11, 0L))));
	    buf1640.reset();
	    buf1641.reset();
	    auto buf1645 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1642, 3, int_array_18, int_array_19, 0L)); buf1642.reset();  // reuse
	    auto buf1646 = std::move(buf1645);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_184, gelu_32, choose_qparams_affine_default_185, quantize_affine_185, dequantize_affine_370], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf1646, model_audio_tower_layers_30_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1647_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1647_handle));
	    RAIIAtenTensorHandle buf1647(buf1647_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_371], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:406
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_30_fc2_parametrizations_weight_original0, model_audio_tower_layers_30_fc2_parametrizations_weight_original2, model_audio_tower_layers_30_fc2_parametrizations_weight_original1, buf1647, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1648_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1648_handle));
	    RAIIAtenTensorHandle buf1648(buf1648_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_185, quantize_affine_185, dequantize_affine_370, dequantize_affine_371, linear_185, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:643
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1648, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1646, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1647, 2, int_array_22, int_array_23, 0L))));
	    buf1646.reset();
	    AtenTensorHandle buf1652_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1652_handle));
	    RAIIAtenTensorHandle buf1652(buf1652_handle);
	    AtenTensorHandle buf1655_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1655_handle));
	    RAIIAtenTensorHandle buf1655(buf1655_handle);
	    auto buf1662 = std::move(buf1655);  // reuse
	    AtenTensorHandle buf1658_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1658_handle));
	    RAIIAtenTensorHandle buf1658(buf1658_handle);
	    auto buf1665 = std::move(buf1658);  // reuse
	    AtenTensorHandle buf1661_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_4, int_array_5, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1661_handle));
	    RAIIAtenTensorHandle buf1661(buf1661_handle);
	    auto buf1668 = std::move(buf1661);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_185, add_434, layer_norm_62, choose_qparams_affine_default_186, quantize_affine_186, dequantize_affine_372, choose_qparams_affine_default_187, quantize_affine_187, dequantize_affine_374, choose_qparams_affine_default_188, quantize_affine_188, dequantize_affine_376], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12(buf1662, buf1665, buf1668, buf1632, buf1648, model_audio_tower_layers_30_fc2_bias, model_audio_tower_layers_31_self_attn_layer_norm_weight, model_audio_tower_layers_31_self_attn_layer_norm_bias, buf1652, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1663_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_6, int_array_7, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1663_handle));
	    RAIIAtenTensorHandle buf1663(buf1663_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_373], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:408
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original0, model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original2, model_audio_tower_layers_31_self_attn_q_proj_parametrizations_weight_original1, buf1663, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1664 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1652, 2, int_array_8, int_array_9, 0L)); buf1652.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_186, dequantize_affine_372, dequantize_affine_373, linear_186, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:644
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1664, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1662, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1663, 2, int_array_10, int_array_11, 0L))));
	    auto buf1666 = std::move(buf1663);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_375], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:409
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original0, model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original2, model_audio_tower_layers_31_self_attn_k_proj_parametrizations_weight_original1, buf1666, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1667 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1662, 2, int_array_8, int_array_9, 0L)); buf1662.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_187, dequantize_affine_374, dequantize_affine_375, linear_187], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:645
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1667, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1665, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1666, 2, int_array_10, int_array_11, 0L))));
	    auto buf1669 = std::move(buf1666);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_377], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:410
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original0, model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original2, model_audio_tower_layers_31_self_attn_v_proj_parametrizations_weight_original1, buf1669, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1670 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1665, 2, int_array_8, int_array_9, 0L)); buf1665.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_188, dequantize_affine_376, dequantize_affine_377, linear_188, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:646
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1670, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1668, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1669, 2, int_array_10, int_array_11, 0L))));
	    auto buf1671 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1668, 4, int_array_12, int_array_13, 0L)); buf1668.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_186, mul_1152, view_93, transpose_124, contiguous_124, linear_187, view_94, transpose_125, contiguous_125, linear_188, view_95, transpose_126, contiguous_126, scaled_dot_product_attention_31], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:411
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5(buf1664, model_audio_tower_layers_31_self_attn_q_proj_bias, buf1671, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1672 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1664, 4, int_array_12, int_array_13, 0L)); buf1664.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_186, mul_1152, view_93, transpose_124, contiguous_124, linear_187, view_94, transpose_125, contiguous_125, linear_188, view_95, transpose_126, contiguous_126, scaled_dot_product_attention_31], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:412
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6(buf1667, buf1672, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1673 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1667, 4, int_array_12, int_array_13, 0L)); buf1667.reset();  // reuse
	    // Topologically Sorted Source Nodes: [, linear_186, mul_1152, view_93, transpose_124, contiguous_124, linear_187, view_94, transpose_125, contiguous_125, linear_188, view_95, transpose_126, contiguous_126, scaled_dot_product_attention_31], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    // [Provenance debug handles] triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:413
	    triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel = 1920000L*s6;
	    call_triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7(buf1670, model_audio_tower_layers_31_self_attn_v_proj_bias, buf1673, triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7_xnumel, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1670.reset();
	    // Topologically Sorted Source Nodes: [, linear_186, mul_1152, view_93, transpose_124, contiguous_124, linear_187, view_94, transpose_125, contiguous_125, linear_188, view_95, transpose_126, contiguous_126, scaled_dot_product_attention_31], Original ATen: [aten.addmm, aten.view, aten.mul, aten.transpose, aten.clone, aten._unsafe_view, aten._scaled_dot_product_efficient_attention]
	    double var_31 = 1.0;
	    AtenTensorHandle buf1675_handle;
	    AtenTensorHandle buf1676_handle;
	    AtenTensorHandle buf1677_handle;
	    AtenTensorHandle buf1678_handle;
	    // [Provenance debug handles] aoti_torch_cuda__scaled_dot_product_efficient_attention:647
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda__scaled_dot_product_efficient_attention(buf1671, buf1672, buf1673, nullptr, 0, 0.0, 0, &var_31, &buf1675_handle, &buf1676_handle, &buf1677_handle, &buf1678_handle));
	    RAIIAtenTensorHandle buf1675(buf1675_handle);
	    RAIIAtenTensorHandle buf1676(buf1676_handle);
	    RAIIAtenTensorHandle buf1677(buf1677_handle);
	    RAIIAtenTensorHandle buf1678(buf1678_handle);
	    buf1671.reset();
	    buf1672.reset();
	
	    auto buf1681 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1675, 3, int_array_4, int_array_5, 0L)); buf1675.reset();  // reuse
	    auto buf1682 = std::move(buf1681);  // reuse
	    // Topologically Sorted Source Nodes: [transpose_127, reshape_31, choose_qparams_affine_default_189, quantize_affine_189, dequantize_affine_378], Original ATen: [aten.transpose, aten.view, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414
	    triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8(buf1682, triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1683 = std::move(buf1669);  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_379], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_4:415
	    call_triton_poi_fused__to_copy_mul_sub_view_4(model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original0, model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original2, model_audio_tower_layers_31_self_attn_out_proj_parametrizations_weight_original1, buf1683, 1638400L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    auto buf1684 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1673, 2, int_array_8, int_array_9, 0L)); buf1673.reset();  // reuse
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_189, dequantize_affine_378, dequantize_affine_379, linear_189, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:648
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1684, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1682, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1683, 2, int_array_10, int_array_11, 0L))));
	    buf1683.reset();
	    auto buf1685 = std::move(buf1632);  // reuse
	    auto buf1689 = std::move(buf1682);  // reuse
	    auto buf1692 = std::move(buf1689);  // reuse
	    auto buf1693 = std::move(buf1692);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_185, add_434, linear_189, add_443, layer_norm_63, choose_qparams_affine_default_190, quantize_affine_190, dequantize_affine_380], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13(buf1685, buf1693, buf1648, model_audio_tower_layers_30_fc2_bias, buf1684, model_audio_tower_layers_31_self_attn_out_proj_bias, model_audio_tower_layers_31_final_layer_norm_weight, model_audio_tower_layers_31_final_layer_norm_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1648.reset();
	    buf1684.reset();
	    auto buf1694 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1647, 3, int_array_14, int_array_7, 0L)); buf1647.reset();  // reuse
	    // Topologically Sorted Source Nodes: [dequantize_affine_381], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:417
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_31_fc1_parametrizations_weight_original0, model_audio_tower_layers_31_fc1_parametrizations_weight_original2, model_audio_tower_layers_31_fc1_parametrizations_weight_original1, buf1694, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1695_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_15, int_array_16, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1695_handle));
	    RAIIAtenTensorHandle buf1695(buf1695_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_190, dequantize_affine_380, dequantize_affine_381, linear_190, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:649
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1695, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1693, 2, int_array_8, int_array_9, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1694, 2, int_array_17, int_array_11, 0L))));
	    buf1693.reset();
	    buf1694.reset();
	    auto buf1698 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1695, 3, int_array_18, int_array_19, 0L)); buf1695.reset();  // reuse
	    auto buf1699 = std::move(buf1698);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_190, gelu_33, choose_qparams_affine_default_191, quantize_affine_191, dequantize_affine_382], Original ATen: [aten.addmm, aten.view, aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418
	    triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel = 1500L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11(buf1699, model_audio_tower_layers_31_fc1_bias, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1700_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_20, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1700_handle));
	    RAIIAtenTensorHandle buf1700(buf1700_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_383], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_10:419
	    call_triton_poi_fused__to_copy_mul_sub_view_10(model_audio_tower_layers_31_fc2_parametrizations_weight_original0, model_audio_tower_layers_31_fc2_parametrizations_weight_original2, model_audio_tower_layers_31_fc2_parametrizations_weight_original1, buf1700, 6553600L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1701_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_8, int_array_9, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1701_handle));
	    RAIIAtenTensorHandle buf1701(buf1701_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_191, quantize_affine_191, dequantize_affine_382, dequantize_affine_383, linear_191, ], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.addmm]
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:650
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1701, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1699, 2, int_array_15, int_array_16, 0L)), wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1700, 2, int_array_22, int_array_23, 0L))));
	    buf1699.reset();
	    buf1700.reset();
	    auto buf1705 = std::move(buf1685);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_191, add_448, layer_norm_64], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm]
	    // [Provenance debug handles] triton_red_fused_add_addmm_native_layer_norm_view_14:420
	    int64_t triton_red_fused_add_addmm_native_layer_norm_view_14_xnumel = 1500L*s6;
	    call_triton_red_fused_add_addmm_native_layer_norm_view_14(buf1705, buf1701, model_audio_tower_layers_31_fc2_bias, model_audio_tower_layer_norm_weight, model_audio_tower_layer_norm_bias, triton_red_fused_add_addmm_native_layer_norm_view_14_xnumel, 1280L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    const int64_t int_array_24[] = {375L*s6, 5120L};
	    auto buf1708 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1701, 2, int_array_24, int_array_16, 0L)); buf1701.reset();  // reuse
	    auto buf1709 = std::move(buf1708);  // reuse
	    // Topologically Sorted Source Nodes: [, linear_191, add_448, layer_norm_64, reshape_32, choose_qparams_affine_default_192, quantize_affine_192, dequantize_affine_384], Original ATen: [aten.addmm, aten.view, aten.add, aten.native_layer_norm, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.reciprocal, aten.mul, aten._to_copy]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421
	    int64_t triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15_xnumel = 375L*s6;
	    call_triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15(buf1709, buf1705, s6, triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15_xnumel, 5120L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    buf1705.reset();
	    static constexpr int64_t int_array_25[] = {3072L, 160L, 32L};
	    AtenTensorHandle buf1710_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_25, int_array_21, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1710_handle));
	    RAIIAtenTensorHandle buf1710(buf1710_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_385], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_16:422
	    call_triton_poi_fused__to_copy_mul_sub_view_16(model_multi_modal_projector_linear_1_parametrizations_weight_original0, model_multi_modal_projector_linear_1_parametrizations_weight_original2, model_multi_modal_projector_linear_1_parametrizations_weight_original1, buf1710, 15728640L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    const int64_t int_array_26[] = {375L*s6, 3072L};
	    static constexpr int64_t int_array_27[] = {3072L, 1L};
	    AtenTensorHandle buf1711_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_26, int_array_27, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1711_handle));
	    RAIIAtenTensorHandle buf1711(buf1711_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_192, dequantize_affine_384, dequantize_affine_385, linear_192], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.view, aten.mul, aten._to_copy, aten.t, aten.mm]
	    static constexpr int64_t int_array_28[] = {5120L, 3072L};
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:651
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1711, buf1709, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1710, 2, int_array_28, int_array_23, 0L))));
	    buf1709.reset();
	    buf1710.reset();
	    auto buf1714 = std::move(buf1711);  // reuse
	    auto buf1715 = std::move(buf1714);  // reuse
	    // Topologically Sorted Source Nodes: [gelu_34, choose_qparams_affine_default_193, quantize_affine_193, dequantize_affine_386], Original ATen: [aten.gelu, aten.amax, aten.zeros_like, aten.maximum, aten.amin, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten.view, aten.reciprocal, aten.mul, aten._to_copy, aten.add]
	    // [Provenance debug handles] triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423
	    int64_t triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17_xnumel = 375L*s6;
	    call_triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17(buf1715, triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17_xnumel, 3072L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    static constexpr int64_t int_array_29[] = {3072L, 96L, 32L};
	    static constexpr int64_t int_array_30[] = {3072L, 32L, 1L};
	    AtenTensorHandle buf1716_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(3, int_array_29, int_array_30, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1716_handle));
	    RAIIAtenTensorHandle buf1716(buf1716_handle);
	    // Topologically Sorted Source Nodes: [dequantize_affine_387], Original ATen: [aten.view, aten._to_copy, aten.sub, aten.mul]
	    // [Provenance debug handles] triton_poi_fused__to_copy_mul_sub_view_18:424
	    call_triton_poi_fused__to_copy_mul_sub_view_18(model_multi_modal_projector_linear_2_parametrizations_weight_original0, model_multi_modal_projector_linear_2_parametrizations_weight_original2, model_multi_modal_projector_linear_2_parametrizations_weight_original1, buf1716, 9437184L, this->device_idx_, stream, kernels, this->cubin_dir_);
	    AtenTensorHandle buf1717_handle;
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_empty_strided(2, int_array_26, int_array_27, cached_torch_dtype_float32, cached_torch_device_type_cuda, this->device_idx_, &buf1717_handle));
	    RAIIAtenTensorHandle buf1717(buf1717_handle);
	    // Topologically Sorted Source Nodes: [choose_qparams_affine_default_193, quantize_affine_193, dequantize_affine_386, dequantize_affine_387, linear_193], Original ATen: [aten.zeros_like, aten.maximum, aten.minimum, aten.sub, aten.div, aten.clamp, aten.round, aten.rsub, aten._to_copy, aten.view, aten.mul, aten.t, aten.mm]
	    static constexpr int64_t int_array_31[] = {3072L, 3072L};
	    static constexpr int64_t int_array_32[] = {1L, 3072L};
	    // [Provenance debug handles] aoti_torch_cuda_mm_out:652
	    AOTI_TORCH_ERROR_CODE_CHECK(aoti_torch_cuda_mm_out(buf1717, buf1715, wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1716, 2, int_array_31, int_array_32, 0L))));
	    buf1715.reset();
	    buf1716.reset();
	    const int64_t int_array_33[] = {1L, 375L*s6, 3072L};
	    const int64_t int_array_34[] = {1152000L*s6, 3072L, 1L};
	    auto var_32 = wrap_with_raii_handle_if_needed(reinterpret_tensor_wrapper(buf1717, 3, int_array_33, int_array_34, 0L));
	    output_handles[0] = var_32.release();
	} // AOTInductorModel::run_impl
	} // namespace torch::aot_inductor
	
	
	
	
V0910 09:43:56.018000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_inductor/codecache.py:1822] {"graph_dump": {"name": "inductor_aot_kernel_code", "type": "cpp", "filename": "/var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/c56zjwltw67y3e4sfx4uih6pd3ir4goqbi6av3c2cwj5p57iab4p.kernel.cpp"}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2657, "name": "compile_fx", "filename": 19, "loc": "return inference_compiler(unlifted_gm, example_inputs_)"}, {"line": 1267, "name": "__call__", "filename": 33, "loc": "return self.compiler_fn(gm, example_inputs)"}, {"line": 2543, "name": "fw_compiler_base", "filename": 19, "loc": "return compile_fx_forward("}, {"line": 2260, "name": "compile_fx_forward", "filename": 19, "loc": "return inner_compile("}, {"line": 81, "name": "inner", "filename": 34, "loc": "return func(*args, **kwds)"}, {"line": 781, "name": "compile_fx_inner", "filename": 19, "loc": "return wrap_compiler_debug(_compile_fx_inner, compiler_name=\"inductor\")("}, {"line": 144, "name": "debug_wrapper", "filename": 35, "loc": "inner_compiled_fn = compiler_fn(gm, example_inputs)"}, {"line": 167, "name": "newFunction", "filename": 36, "loc": "return old_func(*args, **kwargs)"}, {"line": 962, "name": "_compile_fx_inner", "filename": 19, "loc": "mb_compiled_graph = fx_codegen_and_compile("}, {"line": 1694, "name": "fx_codegen_and_compile", "filename": 19, "loc": "return scheme.codegen_and_compile(gm, example_inputs, inputs_to_check, graph_kwargs)"}, {"line": 1486, "name": "codegen_and_compile", "filename": 19, "loc": "compiled_fn = AotCodeCompiler.compile("}, {"line": 1822, "name": "compile", "filename": 45, "loc": "trace_structured("}], "has_payload": "ec3a4a44b16d88c4fd55c2b9e0da0f46"}
	// Triton kernels are embedded as comments in /var/tmp/torchinductor_shangdiy/cetienwpviuyjxiyd6pah5cooivyuprspbupd3rhmsvj7mmmopkj/c352umhij7bqg7p3tqfyegchuzaz5aagiv7cyzfgjxn3nlod5dvn.wrapper.cpp
	
V0910 09:43:57.115000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "3ea4aeec498c369efc81b987d58fbe5c"}
	{
	"name": "compile_file",
	"ts": 1757522637115699.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:44:14.769000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "38f8fc186c8c75927ebe2941d931cf97"}
	{
	"name": "compile_file",
	"ts": 1757522654769476.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:44:14.771000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "d5f821f525b43fbf78d87a0bfc5b21b0"}
	{
	"name": "compile_file",
	"ts": 1757522654771800.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:44:14.789000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "9f2beedce8e98adc9bdcccfb1223477e"}
	{
	"name": "compile_file",
	"ts": 1757522654789242.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:44:14.793000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8c3daa353051f8919ef25ddb778e1e83"}
	{
	"name": "compile_file",
	"ts": 1757522654793922.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:44:15.448000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "069590c5f52edb5411f4c512d31e3a50"}
	{
	"name": "compile_file",
	"ts": 1757522655447980.2,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:44:15.735000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "10ac4eb117e774cdee5888658a0a3c66"}
	{
	"name": "compile_file",
	"ts": 1757522655734887.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "B",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:44:16.676000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "f2d74058bd69520fc2385aa9ad8cb6c7"}
	{
	"name": "compile_file",
	"ts": 1757522656675983.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:44:16.830000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "4efd6ad663285a5889cb128969f920a2"}
	{
	"name": "AotCodeCompiler.compile",
	"ts": 1757522656830767.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:44:16.835000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "2eb6950880c8205d537f10177bcc000e"}
	{
	"name": "GraphLowering.compile_to_fn",
	"ts": 1757522656835244.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:44:16.891000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_inductor/compile_fx.py:1520] {"artifact": {"name": "inductor_provenance_tracking_node_mappings", "encoding": "json"}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2657, "name": "compile_fx", "filename": 19, "loc": "return inference_compiler(unlifted_gm, example_inputs_)"}, {"line": 1267, "name": "__call__", "filename": 33, "loc": "return self.compiler_fn(gm, example_inputs)"}, {"line": 2543, "name": "fw_compiler_base", "filename": 19, "loc": "return compile_fx_forward("}, {"line": 2260, "name": "compile_fx_forward", "filename": 19, "loc": "return inner_compile("}, {"line": 81, "name": "inner", "filename": 34, "loc": "return func(*args, **kwds)"}, {"line": 781, "name": "compile_fx_inner", "filename": 19, "loc": "return wrap_compiler_debug(_compile_fx_inner, compiler_name=\"inductor\")("}, {"line": 144, "name": "debug_wrapper", "filename": 35, "loc": "inner_compiled_fn = compiler_fn(gm, example_inputs)"}, {"line": 167, "name": "newFunction", "filename": 36, "loc": "return old_func(*args, **kwargs)"}, {"line": 962, "name": "_compile_fx_inner", "filename": 19, "loc": "mb_compiled_graph = fx_codegen_and_compile("}, {"line": 1694, "name": "fx_codegen_and_compile", "filename": 19, "loc": "return scheme.codegen_and_compile(gm, example_inputs, inputs_to_check, graph_kwargs)"}, {"line": 1520, "name": "codegen_and_compile", "filename": 19, "loc": "trace_structured("}], "has_payload": "16c989fb1ead344d39bb7315422d7200"}
	{"preToPost": {"conv1d": ["convolution", "sym_size_int"], "gelu": ["mul_2", "mul_3", "erf", "add_4", "mul_4"], "conv1d_1": ["convolution_1"], "gelu_1": ["mul_9", "mul_10", "erf_1", "add_13", "mul_11"], "permute": ["permute"], "add": ["add_22"], "dropout": ["add_22"], "layer_norm": ["clone_1", "var_mean", "getitem", "getitem_1", "sub_7", "add_31", "rsqrt", "mul_20", "mul_21", "add_32"], "choose_qparams_affine_default": ["add_32", "amax", "full_1", "maximum", "amin", "full", "minimum", "sub_18", "div", "clamp_min", "div_1", "round_1", "sub_24", "clamp_min_1", "clamp_max", "convert_element_type"], "quantize_affine": ["add_32", "view_2", "reciprocal", "mul_69", "mul_72", "round_2", "view_3", "add_119", "clamp_min_2", "clamp_max_1", "convert_element_type_1"], "choose_qparams_affine_default_1": ["add_32", "amax_1", "full_3", "maximum_1", "amin_1", "full_2", "minimum_1", "sub_63", "div_2", "clamp_min_3", "div_3", "round_3", "sub_69", "clamp_min_4", "clamp_max_2", "convert_element_type_6"], "quantize_affine_1": ["add_32", "view_18", "reciprocal_1", "mul_165", "mul_168", "round_4", "view_19", "add_271", "clamp_min_5", "clamp_max_3", "convert_element_type_7"], "choose_qparams_affine_default_2": ["add_32", "amax_2", "full_5", "maximum_2", "amin_2", "full_4", "minimum_2", "sub_107", "div_4", "clamp_min_6", "div_5", "round_5", "sub_113", "clamp_min_7", "clamp_max_4", "convert_element_type_12"], "quantize_affine_2": ["add_32", "view_34", "reciprocal_2", "mul_264", "mul_267", "round_6", "view_35", "add_419", "clamp_min_8", "clamp_max_5", "convert_element_type_13"], "dequantize_affine": ["convert_element_type_1", "view_7", "convert_element_type_2", "convert_element_type_3", "sub_44", "view_6", "mul_94"], "dequantize_affine_1": ["view_9", "view_11", "convert_element_type_4", "convert_element_type_5", "sub_48", "view_10", "mul_99", "view_12"], "dequantize_affine_2": ["convert_element_type_7", "view_23", "convert_element_type_8", "convert_element_type_9", "sub_89", "view_22", "mul_190"], "dequantize_affine_3": ["view_25", "view_27", "convert_element_type_10", "convert_element_type_11", "sub_93", "view_26", "mul_195", "view_28"], "dequantize_affine_4": ["convert_element_type_13", "view_39", "convert_element_type_14", "convert_element_type_15", "sub_133", "view_38", "mul_289"], "dequantize_affine_5": ["view_41", "view_43", "convert_element_type_16", "convert_element_type_17", "sub_137", "view_42", "mul_294", "view_44"], "linear": ["mul_104", "view_13", "permute_1", "mm_default_159", "add_tensor_159", "view_14"], "mul_5": ["mul_111"], "view": ["view_15"], "transpose": ["permute_2"], "contiguous": ["clone_2"], "linear_1": ["mul_198", "view_29", "permute_3", "mm", "view_30"], "view_1": ["view_31"], "transpose_1": ["permute_4"], "contiguous_1": ["clone_3"], "linear_2": ["mul_299", "view_45", "permute_5", "mm_default_158", "add_tensor_158", "view_46"], "view_2": ["view_47"], "transpose_2": ["permute_6"], "contiguous_2": ["clone_4"], "scaled_dot_product_attention": ["_scaled_dot_product_efficient_attention", "getitem_2"], "transpose_3": ["permute_7"], "reshape": ["view_48"], "choose_qparams_affine_default_3": ["view_48", "amax_3", "full_7", "maximum_3", "amin_3", "full_6", "minimum_3", "sub_155", "div_6", "clamp_min_9", "div_7", "round_7", "sub_161", "clamp_min_10", "clamp_max_6", "convert_element_type_18"], "quantize_affine_3": ["view_48", "view_51", "reciprocal_3", "mul_369", "mul_372", "round_8", "view_52", "add_583", "clamp_min_11", "clamp_max_7", "convert_element_type_19"], "dequantize_affine_6": ["convert_element_type_19", "view_56", "convert_element_type_20", "convert_element_type_21", "sub_181", "view_55", "mul_394"], "dequantize_affine_7": ["view_58", "view_60", "convert_element_type_22", "convert_element_type_23", "sub_185", "view_59", "mul_399", "view_61"], "linear_3": ["mul_404", "view_62", "permute_8", "mm_default_157", "add_tensor_157", "view_63"], "dropout_1": ["view_63"], "add_9": ["add_646"], "layer_norm_1": ["clone_6", "var_mean_1", "getitem_6", "getitem_7", "sub_191", "add_651", "rsqrt_1", "mul_415", "mul_416", "add_652"], "choose_qparams_affine_default_4": ["add_652", "amax_4", "full_9", "maximum_4", "amin_4", "full_8", "minimum_4", "sub_202", "div_8", "clamp_min_12", "div_9", "round_9", "sub_208", "clamp_min_13", "clamp_max_8", "convert_element_type_24"], "quantize_affine_4": ["add_652", "view_66", "reciprocal_4", "mul_464", "mul_467", "round_10", "view_67", "add_739", "clamp_min_14", "clamp_max_9", "convert_element_type_25"], "dequantize_affine_8": ["convert_element_type_25", "view_71", "convert_element_type_26", "convert_element_type_27", "sub_228", "view_70", "mul_489"], "dequantize_affine_9": ["view_73", "view_75", "convert_element_type_28", "convert_element_type_29", "sub_232", "view_74", "mul_494", "view_76"], "linear_4": ["mul_499", "view_77", "permute_9", "mm_default_156", "add_tensor_156", "view_78"], "gelu_2": ["mul_506", "mul_507", "erf_2", "add_798", "mul_508"], "dropout_2": ["mul_508"], "choose_qparams_affine_default_5": ["mul_508", "amax_5", "full_11", "maximum_5", "amin_5", "full_10", "minimum_5", "sub_245", "div_10", "clamp_min_15", "div_11", "round_11", "sub_251", "clamp_min_16", "clamp_max_10", "convert_element_type_30"], "quantize_affine_5": ["mul_508", "view_81", "reciprocal_5", "mul_554", "mul_557", "round_12", "view_82", "add_881", "clamp_min_17", "clamp_max_11", "convert_element_type_31"], "dequantize_affine_10": ["convert_element_type_31", "view_86", "convert_element_type_32", "convert_element_type_33", "sub_271", "view_85", "mul_579"], "dequantize_affine_11": ["view_88", "view_90", "convert_element_type_34", "convert_element_type_35", "sub_275", "view_89", "mul_584", "view_91"], "linear_5": ["mul_589", "view_92", "permute_10", "mm_default_155", "add_tensor_155", "view_93"], "dropout_3": ["view_93"], "add_14": ["add_944"], "layer_norm_2": ["clone_9", "var_mean_2", "getitem_8", "getitem_9", "sub_281", "add_949", "rsqrt_2", "mul_600", "mul_601", "add_950"], "choose_qparams_affine_default_6": ["add_950", "amax_6", "full_13", "maximum_6", "amin_6", "full_12", "minimum_6", "sub_292", "div_12", "clamp_min_18", "div_13", "round_13", "sub_298", "clamp_min_19", "clamp_max_12", "convert_element_type_36"], "quantize_affine_6": ["add_950", "view_96", "reciprocal_6", "mul_649", "mul_652", "round_14", "view_97", "add_1037", "clamp_min_20", "clamp_max_13", "convert_element_type_37"], "choose_qparams_affine_default_7": ["add_950", "amax_7", "full_15", "maximum_7", "amin_7", "full_14", "minimum_7", "sub_337", "div_14", "clamp_min_21", "div_15", "round_15", "sub_343", "clamp_min_22", "clamp_max_14", "convert_element_type_42"], "quantize_affine_7": ["add_950", "view_112", "reciprocal_7", "mul_745", "mul_748", "round_16", "view_113", "add_1189", "clamp_min_23", "clamp_max_15", "convert_element_type_43"], "choose_qparams_affine_default_8": ["add_950", "amax_8", "full_17", "maximum_8", "amin_8", "full_16", "minimum_8", "sub_381", "div_16", "clamp_min_24", "div_17", "round_17", "sub_387", "clamp_min_25", "clamp_max_16", "convert_element_type_48"], "quantize_affine_8": ["add_950", "view_128", "reciprocal_8", "mul_844", "mul_847", "round_18", "view_129", "add_1337", "clamp_min_26", "clamp_max_17", "convert_element_type_49"], "dequantize_affine_12": ["convert_element_type_37", "view_101", "convert_element_type_38", "convert_element_type_39", "sub_318", "view_100", "mul_674"], "dequantize_affine_13": ["view_103", "view_105", "convert_element_type_40", "convert_element_type_41", "sub_322", "view_104", "mul_679", "view_106"], "dequantize_affine_14": ["convert_element_type_43", "view_117", "convert_element_type_44", "convert_element_type_45", "sub_363", "view_116", "mul_770"], "dequantize_affine_15": ["view_119", "view_121", "convert_element_type_46", "convert_element_type_47", "sub_367", "view_120", "mul_775", "view_122"], "dequantize_affine_16": ["convert_element_type_49", "view_133", "convert_element_type_50", "convert_element_type_51", "sub_407", "view_132", "mul_869"], "dequantize_affine_17": ["view_135", "view_137", "convert_element_type_52", "convert_element_type_53", "sub_411", "view_136", "mul_874", "view_138"], "linear_6": ["mul_684", "view_107", "permute_11", "mm_default_154", "add_tensor_154", "view_108"], "mul_42": ["mul_691"], "view_3": ["view_109"], "transpose_4": ["permute_12"], "contiguous_4": ["clone_10"], "linear_7": ["mul_778", "view_123", "permute_13", "mm_1", "view_124"], "view_4": ["view_125"], "transpose_5": ["permute_14"], "contiguous_5": ["clone_11"], "linear_8": ["mul_879", "view_139", "permute_15", "mm_default_153", "add_tensor_153", "view_140"], "view_5": ["view_141"], "transpose_6": ["permute_16"], "contiguous_6": ["clone_12"], "scaled_dot_product_attention_1": ["_scaled_dot_product_efficient_attention_1", "getitem_10"], "transpose_7": ["permute_17"], "reshape_1": ["view_142"], "choose_qparams_affine_default_9": ["view_142", "amax_9", "full_19", "maximum_9", "amin_9", "full_18", "minimum_9", "sub_429", "div_18", "clamp_min_27", "div_19", "round_19", "sub_435", "clamp_min_28", "clamp_max_18", "convert_element_type_54"], "quantize_affine_9": ["view_142", "view_145", "reciprocal_9", "mul_949", "mul_952", "round_20", "view_146", "add_1501", "clamp_min_29", "clamp_max_19", "convert_element_type_55"], "dequantize_affine_18": ["convert_element_type_55", "view_150", "convert_element_type_56", "convert_element_type_57", "sub_455", "view_149", "mul_974"], "dequantize_affine_19": ["view_152", "view_154", "convert_element_type_58", "convert_element_type_59", "sub_459", "view_153", "mul_979", "view_155"], "linear_9": ["mul_984", "view_156", "permute_18", "mm_default_152", "add_tensor_152", "view_157"], "dropout_4": ["view_157"], "add_23": ["add_1564"], "layer_norm_3": ["clone_14", "var_mean_3", "getitem_14", "getitem_15", "sub_465", "add_1569", "rsqrt_3", "mul_995", "mul_996", "add_1570"], "choose_qparams_affine_default_10": ["add_1570", "amax_10", "full_21", "maximum_10", "amin_10", "full_20", "minimum_10", "sub_476", "div_20", "clamp_min_30", "div_21", "round_21", "sub_482", "clamp_min_31", "clamp_max_20", "convert_element_type_60"], "quantize_affine_10": ["add_1570", "view_160", "reciprocal_10", "mul_1044", "mul_1047", "round_22", "view_161", "add_1657", "clamp_min_32", "clamp_max_21", "convert_element_type_61"], "dequantize_affine_20": ["convert_element_type_61", "view_165", "convert_element_type_62", "convert_element_type_63", "sub_502", "view_164", "mul_1069"], "dequantize_affine_21": ["view_167", "view_169", "convert_element_type_64", "convert_element_type_65", "sub_506", "view_168", "mul_1074", "view_170"], "linear_10": ["mul_1079", "view_171", "permute_19", "mm_default_151", "add_tensor_151", "view_172"], "gelu_3": ["mul_1086", "mul_1087", "erf_3", "add_1716", "mul_1088"], "dropout_5": ["mul_1088"], "choose_qparams_affine_default_11": ["mul_1088", "amax_11", "full_23", "maximum_11", "amin_11", "full_22", "minimum_11", "sub_519", "div_22", "clamp_min_33", "div_23", "round_23", "sub_525", "clamp_min_34", "clamp_max_22", "convert_element_type_66"], "quantize_affine_11": ["mul_1088", "view_175", "reciprocal_11", "mul_1134", "mul_1137", "round_24", "view_176", "add_1799", "clamp_min_35", "clamp_max_23", "convert_element_type_67"], "dequantize_affine_22": ["convert_element_type_67", "view_180", "convert_element_type_68", "convert_element_type_69", "sub_545", "view_179", "mul_1159"], "dequantize_affine_23": ["view_182", "view_184", "convert_element_type_70", "convert_element_type_71", "sub_549", "view_183", "mul_1164", "view_185"], "linear_11": ["mul_1169", "view_186", "permute_20", "mm_default_150", "add_tensor_150", "view_187"], "dropout_6": ["view_187"], "add_28": ["add_1862"], "layer_norm_4": ["clone_17", "var_mean_4", "getitem_16", "getitem_17", "sub_555", "add_1867", "rsqrt_4", "mul_1180", "mul_1181", "add_1868"], "choose_qparams_affine_default_12": ["add_1868", "amax_12", "full_25", "maximum_12", "amin_12", "full_24", "minimum_12", "sub_566", "div_24", "clamp_min_36", "div_25", "round_25", "sub_572", "clamp_min_37", "clamp_max_24", "convert_element_type_72"], "quantize_affine_12": ["add_1868", "view_190", "reciprocal_12", "mul_1229", "mul_1232", "round_26", "view_191", "add_1955", "clamp_min_38", "clamp_max_25", "convert_element_type_73"], "choose_qparams_affine_default_13": ["add_1868", "amax_13", "full_27", "maximum_13", "amin_13", "full_26", "minimum_13", "sub_611", "div_26", "clamp_min_39", "div_27", "round_27", "sub_617", "clamp_min_40", "clamp_max_26", "convert_element_type_78"], "quantize_affine_13": ["add_1868", "view_206", "reciprocal_13", "mul_1325", "mul_1328", "round_28", "view_207", "add_2107", "clamp_min_41", "clamp_max_27", "convert_element_type_79"], "choose_qparams_affine_default_14": ["add_1868", "amax_14", "full_29", "maximum_14", "amin_14", "full_28", "minimum_14", "sub_655", "div_28", "clamp_min_42", "div_29", "round_29", "sub_661", "clamp_min_43", "clamp_max_28", "convert_element_type_84"], "quantize_affine_14": ["add_1868", "view_222", "reciprocal_14", "mul_1424", "mul_1427", "round_30", "view_223", "add_2255", "clamp_min_44", "clamp_max_29", "convert_element_type_85"], "dequantize_affine_24": ["convert_element_type_73", "view_195", "convert_element_type_74", "convert_element_type_75", "sub_592", "view_194", "mul_1254"], "dequantize_affine_25": ["view_197", "view_199", "convert_element_type_76", "convert_element_type_77", "sub_596", "view_198", "mul_1259", "view_200"], "dequantize_affine_26": ["convert_element_type_79", "view_211", "convert_element_type_80", "convert_element_type_81", "sub_637", "view_210", "mul_1350"], "dequantize_affine_27": ["view_213", "view_215", "convert_element_type_82", "convert_element_type_83", "sub_641", "view_214", "mul_1355", "view_216"], "dequantize_affine_28": ["convert_element_type_85", "view_227", "convert_element_type_86", "convert_element_type_87", "sub_681", "view_226", "mul_1449"], "dequantize_affine_29": ["view_229", "view_231", "convert_element_type_88", "convert_element_type_89", "sub_685", "view_230", "mul_1454", "view_232"], "linear_12": ["mul_1264", "view_201", "permute_21", "mm_default_149", "add_tensor_149", "view_202"], "mul_79": ["mul_1271"], "view_6": ["view_203"], "transpose_8": ["permute_22"], "contiguous_8": ["clone_18"], "linear_13": ["mul_1358", "view_217", "permute_23", "mm_2", "view_218"], "view_7": ["view_219"], "transpose_9": ["permute_24"], "contiguous_9": ["clone_19"], "linear_14": ["mul_1459", "view_233", "permute_25", "mm_default_148", "add_tensor_148", "view_234"], "view_8": ["view_235"], "transpose_10": ["permute_26"], "contiguous_10": ["clone_20"], "scaled_dot_product_attention_2": ["_scaled_dot_product_efficient_attention_2", "getitem_18"], "transpose_11": ["permute_27"], "reshape_2": ["view_236"], "choose_qparams_affine_default_15": ["view_236", "amax_15", "full_31", "maximum_15", "amin_15", "full_30", "minimum_15", "sub_703", "div_30", "clamp_min_45", "div_31", "round_31", "sub_709", "clamp_min_46", "clamp_max_30", "convert_element_type_90"], "quantize_affine_15": ["view_236", "view_239", "reciprocal_15", "mul_1529", "mul_1532", "round_32", "view_240", "add_2419", "clamp_min_47", "clamp_max_31", "convert_element_type_91"], "dequantize_affine_30": ["convert_element_type_91", "view_244", "convert_element_type_92", "convert_element_type_93", "sub_729", "view_243", "mul_1554"], "dequantize_affine_31": ["view_246", "view_248", "convert_element_type_94", "convert_element_type_95", "sub_733", "view_247", "mul_1559", "view_249"], "linear_15": ["mul_1564", "view_250", "permute_28", "mm_default_147", "add_tensor_147", "view_251"], "dropout_7": ["view_251"], "add_37": ["add_2482"], "layer_norm_5": ["clone_22", "var_mean_5", "getitem_22", "getitem_23", "sub_739", "add_2487", "rsqrt_5", "mul_1575", "mul_1576", "add_2488"], "choose_qparams_affine_default_16": ["add_2488", "amax_16", "full_33", "maximum_16", "amin_16", "full_32", "minimum_16", "sub_750", "div_32", "clamp_min_48", "div_33", "round_33", "sub_756", "clamp_min_49", "clamp_max_32", "convert_element_type_96"], "quantize_affine_16": ["add_2488", "view_254", "reciprocal_16", "mul_1624", "mul_1627", "round_34", "view_255", "add_2575", "clamp_min_50", "clamp_max_33", "convert_element_type_97"], "dequantize_affine_32": ["convert_element_type_97", "view_259", "convert_element_type_98", "convert_element_type_99", "sub_776", "view_258", "mul_1649"], "dequantize_affine_33": ["view_261", "view_263", "convert_element_type_100", "convert_element_type_101", "sub_780", "view_262", "mul_1654", "view_264"], "linear_16": ["mul_1659", "view_265", "permute_29", "mm_default_146", "add_tensor_146", "view_266"], "gelu_4": ["mul_1666", "mul_1667", "erf_4", "add_2634", "mul_1668"], "dropout_8": ["mul_1668"], "choose_qparams_affine_default_17": ["mul_1668", "amax_17", "full_35", "maximum_17", "amin_17", "full_34", "minimum_17", "sub_793", "div_34", "clamp_min_51", "div_35", "round_35", "sub_799", "clamp_min_52", "clamp_max_34", "convert_element_type_102"], "quantize_affine_17": ["mul_1668", "view_269", "reciprocal_17", "mul_1714", "mul_1717", "round_36", "view_270", "add_2717", "clamp_min_53", "clamp_max_35", "convert_element_type_103"], "dequantize_affine_34": ["convert_element_type_103", "view_274", "convert_element_type_104", "convert_element_type_105", "sub_819", "view_273", "mul_1739"], "dequantize_affine_35": ["view_276", "view_278", "convert_element_type_106", "convert_element_type_107", "sub_823", "view_277", "mul_1744", "view_279"], "linear_17": ["mul_1749", "view_280", "permute_30", "mm_default_145", "add_tensor_145", "view_281"], "dropout_9": ["view_281"], "add_42": ["add_2780"], "layer_norm_6": ["clone_25", "var_mean_6", "getitem_24", "getitem_25", "sub_829", "add_2785", "rsqrt_6", "mul_1760", "mul_1761", "add_2786"], "choose_qparams_affine_default_18": ["add_2786", "amax_18", "full_37", "maximum_18", "amin_18", "full_36", "minimum_18", "sub_840", "div_36", "clamp_min_54", "div_37", "round_37", "sub_846", "clamp_min_55", "clamp_max_36", "convert_element_type_108"], "quantize_affine_18": ["add_2786", "view_284", "reciprocal_18", "mul_1809", "mul_1812", "round_38", "view_285", "add_2873", "clamp_min_56", "clamp_max_37", "convert_element_type_109"], "choose_qparams_affine_default_19": ["add_2786", "amax_19", "full_39", "maximum_19", "amin_19", "full_38", "minimum_19", "sub_885", "div_38", "clamp_min_57", "div_39", "round_39", "sub_891", "clamp_min_58", "clamp_max_38", "convert_element_type_114"], "quantize_affine_19": ["add_2786", "view_300", "reciprocal_19", "mul_1905", "mul_1908", "round_40", "view_301", "add_3025", "clamp_min_59", "clamp_max_39", "convert_element_type_115"], "choose_qparams_affine_default_20": ["add_2786", "amax_20", "full_41", "maximum_20", "amin_20", "full_40", "minimum_20", "sub_929", "div_40", "clamp_min_60", "div_41", "round_41", "sub_935", "clamp_min_61", "clamp_max_40", "convert_element_type_120"], "quantize_affine_20": ["add_2786", "view_316", "reciprocal_20", "mul_2004", "mul_2007", "round_42", "view_317", "add_3173", "clamp_min_62", "clamp_max_41", "convert_element_type_121"], "dequantize_affine_36": ["convert_element_type_109", "view_289", "convert_element_type_110", "convert_element_type_111", "sub_866", "view_288", "mul_1834"], "dequantize_affine_37": ["view_291", "view_293", "convert_element_type_112", "convert_element_type_113", "sub_870", "view_292", "mul_1839", "view_294"], "dequantize_affine_38": ["convert_element_type_115", "view_305", "convert_element_type_116", "convert_element_type_117", "sub_911", "view_304", "mul_1930"], "dequantize_affine_39": ["view_307", "view_309", "convert_element_type_118", "convert_element_type_119", "sub_915", "view_308", "mul_1935", "view_310"], "dequantize_affine_40": ["convert_element_type_121", "view_321", "convert_element_type_122", "convert_element_type_123", "sub_955", "view_320", "mul_2029"], "dequantize_affine_41": ["view_323", "view_325", "convert_element_type_124", "convert_element_type_125", "sub_959", "view_324", "mul_2034", "view_326"], "linear_18": ["mul_1844", "view_295", "permute_31", "mm_default_144", "add_tensor_144", "view_296"], "mul_116": ["mul_1851"], "view_9": ["view_297"], "transpose_12": ["permute_32"], "contiguous_12": ["clone_26"], "linear_19": ["mul_1938", "view_311", "permute_33", "mm_3", "view_312"], "view_10": ["view_313"], "transpose_13": ["permute_34"], "contiguous_13": ["clone_27"], "linear_20": ["mul_2039", "view_327", "permute_35", "mm_default_143", "add_tensor_143", "view_328"], "view_11": ["view_329"], "transpose_14": ["permute_36"], "contiguous_14": ["clone_28"], "scaled_dot_product_attention_3": ["_scaled_dot_product_efficient_attention_3", "getitem_26"], "transpose_15": ["permute_37"], "reshape_3": ["view_330"], "choose_qparams_affine_default_21": ["view_330", "amax_21", "full_43", "maximum_21", "amin_21", "full_42", "minimum_21", "sub_977", "div_42", "clamp_min_63", "div_43", "round_43", "sub_983", "clamp_min_64", "clamp_max_42", "convert_element_type_126"], "quantize_affine_21": ["view_330", "view_333", "reciprocal_21", "mul_2109", "mul_2112", "round_44", "view_334", "add_3337", "clamp_min_65", "clamp_max_43", "convert_element_type_127"], "dequantize_affine_42": ["convert_element_type_127", "view_338", "convert_element_type_128", "convert_element_type_129", "sub_1003", "view_337", "mul_2134"], "dequantize_affine_43": ["view_340", "view_342", "convert_element_type_130", "convert_element_type_131", "sub_1007", "view_341", "mul_2139", "view_343"], "linear_21": ["mul_2144", "view_344", "permute_38", "mm_default_142", "add_tensor_142", "view_345"], "dropout_10": ["view_345"], "add_51": ["add_3400"], "layer_norm_7": ["clone_30", "var_mean_7", "getitem_30", "getitem_31", "sub_1013", "add_3405", "rsqrt_7", "mul_2155", "mul_2156", "add_3406"], "choose_qparams_affine_default_22": ["add_3406", "amax_22", "full_45", "maximum_22", "amin_22", "full_44", "minimum_22", "sub_1024", "div_44", "clamp_min_66", "div_45", "round_45", "sub_1030", "clamp_min_67", "clamp_max_44", "convert_element_type_132"], "quantize_affine_22": ["add_3406", "view_348", "reciprocal_22", "mul_2204", "mul_2207", "round_46", "view_349", "add_3493", "clamp_min_68", "clamp_max_45", "convert_element_type_133"], "dequantize_affine_44": ["convert_element_type_133", "view_353", "convert_element_type_134", "convert_element_type_135", "sub_1050", "view_352", "mul_2229"], "dequantize_affine_45": ["view_355", "view_357", "convert_element_type_136", "convert_element_type_137", "sub_1054", "view_356", "mul_2234", "view_358"], "linear_22": ["mul_2239", "view_359", "permute_39", "mm_default_141", "add_tensor_141", "view_360"], "gelu_5": ["mul_2246", "mul_2247", "erf_5", "add_3552", "mul_2248"], "dropout_11": ["mul_2248"], "choose_qparams_affine_default_23": ["mul_2248", "amax_23", "full_47", "maximum_23", "amin_23", "full_46", "minimum_23", "sub_1067", "div_46", "clamp_min_69", "div_47", "round_47", "sub_1073", "clamp_min_70", "clamp_max_46", "convert_element_type_138"], "quantize_affine_23": ["mul_2248", "view_363", "reciprocal_23", "mul_2294", "mul_2297", "round_48", "view_364", "add_3635", "clamp_min_71", "clamp_max_47", "convert_element_type_139"], "dequantize_affine_46": ["convert_element_type_139", "view_368", "convert_element_type_140", "convert_element_type_141", "sub_1093", "view_367", "mul_2319"], "dequantize_affine_47": ["view_370", "view_372", "convert_element_type_142", "convert_element_type_143", "sub_1097", "view_371", "mul_2324", "view_373"], "linear_23": ["mul_2329", "view_374", "permute_40", "mm_default_140", "add_tensor_140", "view_375"], "dropout_12": ["view_375"], "add_56": ["add_3698"], "layer_norm_8": ["clone_33", "var_mean_8", "getitem_32", "getitem_33", "sub_1103", "add_3703", "rsqrt_8", "mul_2340", "mul_2341", "add_3704"], "choose_qparams_affine_default_24": ["add_3704", "amax_24", "full_49", "maximum_24", "amin_24", "full_48", "minimum_24", "sub_1114", "div_48", "clamp_min_72", "div_49", "round_49", "sub_1120", "clamp_min_73", "clamp_max_48", "convert_element_type_144"], "quantize_affine_24": ["add_3704", "view_378", "reciprocal_24", "mul_2389", "mul_2392", "round_50", "view_379", "add_3791", "clamp_min_74", "clamp_max_49", "convert_element_type_145"], "choose_qparams_affine_default_25": ["add_3704", "amax_25", "full_51", "maximum_25", "amin_25", "full_50", "minimum_25", "sub_1159", "div_50", "clamp_min_75", "div_51", "round_51", "sub_1165", "clamp_min_76", "clamp_max_50", "convert_element_type_150"], "quantize_affine_25": ["add_3704", "view_394", "reciprocal_25", "mul_2485", "mul_2488", "round_52", "view_395", "add_3943", "clamp_min_77", "clamp_max_51", "convert_element_type_151"], "choose_qparams_affine_default_26": ["add_3704", "amax_26", "full_53", "maximum_26", "amin_26", "full_52", "minimum_26", "sub_1203", "div_52", "clamp_min_78", "div_53", "round_53", "sub_1209", "clamp_min_79", "clamp_max_52", "convert_element_type_156"], "quantize_affine_26": ["add_3704", "view_410", "reciprocal_26", "mul_2584", "mul_2587", "round_54", "view_411", "add_4091", "clamp_min_80", "clamp_max_53", "convert_element_type_157"], "dequantize_affine_48": ["convert_element_type_145", "view_383", "convert_element_type_146", "convert_element_type_147", "sub_1140", "view_382", "mul_2414"], "dequantize_affine_49": ["view_385", "view_387", "convert_element_type_148", "convert_element_type_149", "sub_1144", "view_386", "mul_2419", "view_388"], "dequantize_affine_50": ["convert_element_type_151", "view_399", "convert_element_type_152", "convert_element_type_153", "sub_1185", "view_398", "mul_2510"], "dequantize_affine_51": ["view_401", "view_403", "convert_element_type_154", "convert_element_type_155", "sub_1189", "view_402", "mul_2515", "view_404"], "dequantize_affine_52": ["convert_element_type_157", "view_415", "convert_element_type_158", "convert_element_type_159", "sub_1229", "view_414", "mul_2609"], "dequantize_affine_53": ["view_417", "view_419", "convert_element_type_160", "convert_element_type_161", "sub_1233", "view_418", "mul_2614", "view_420"], "linear_24": ["mul_2424", "view_389", "permute_41", "mm_default_139", "add_tensor_139", "view_390"], "mul_153": ["mul_2431"], "view_12": ["view_391"], "transpose_16": ["permute_42"], "contiguous_16": ["clone_34"], "linear_25": ["mul_2518", "view_405", "permute_43", "mm_4", "view_406"], "view_13": ["view_407"], "transpose_17": ["permute_44"], "contiguous_17": ["clone_35"], "linear_26": ["mul_2619", "view_421", "permute_45", "mm_default_138", "add_tensor_138", "view_422"], "view_14": ["view_423"], "transpose_18": ["permute_46"], "contiguous_18": ["clone_36"], "scaled_dot_product_attention_4": ["_scaled_dot_product_efficient_attention_4", "getitem_34"], "transpose_19": ["permute_47"], "reshape_4": ["view_424"], "choose_qparams_affine_default_27": ["view_424", "amax_27", "full_55", "maximum_27", "amin_27", "full_54", "minimum_27", "sub_1251", "div_54", "clamp_min_81", "div_55", "round_55", "sub_1257", "clamp_min_82", "clamp_max_54", "convert_element_type_162"], "quantize_affine_27": ["view_424", "view_427", "reciprocal_27", "mul_2689", "mul_2692", "round_56", "view_428", "add_4255", "clamp_min_83", "clamp_max_55", "convert_element_type_163"], "dequantize_affine_54": ["convert_element_type_163", "view_432", "convert_element_type_164", "convert_element_type_165", "sub_1277", "view_431", "mul_2714"], "dequantize_affine_55": ["view_434", "view_436", "convert_element_type_166", "convert_element_type_167", "sub_1281", "view_435", "mul_2719", "view_437"], "linear_27": ["mul_2724", "view_438", "permute_48", "mm_default_137", "add_tensor_137", "view_439"], "dropout_13": ["view_439"], "add_65": ["add_4318"], "layer_norm_9": ["clone_38", "var_mean_9", "getitem_38", "getitem_39", "sub_1287", "add_4323", "rsqrt_9", "mul_2735", "mul_2736", "add_4324"], "choose_qparams_affine_default_28": ["add_4324", "amax_28", "full_57", "maximum_28", "amin_28", "full_56", "minimum_28", "sub_1298", "div_56", "clamp_min_84", "div_57", "round_57", "sub_1304", "clamp_min_85", "clamp_max_56", "convert_element_type_168"], "quantize_affine_28": ["add_4324", "view_442", "reciprocal_28", "mul_2784", "mul_2787", "round_58", "view_443", "add_4411", "clamp_min_86", "clamp_max_57", "convert_element_type_169"], "dequantize_affine_56": ["convert_element_type_169", "view_447", "convert_element_type_170", "convert_element_type_171", "sub_1324", "view_446", "mul_2809"], "dequantize_affine_57": ["view_449", "view_451", "convert_element_type_172", "convert_element_type_173", "sub_1328", "view_450", "mul_2814", "view_452"], "linear_28": ["mul_2819", "view_453", "permute_49", "mm_default_136", "add_tensor_136", "view_454"], "gelu_6": ["mul_2826", "mul_2827", "erf_6", "add_4470", "mul_2828"], "dropout_14": ["mul_2828"], "choose_qparams_affine_default_29": ["mul_2828", "amax_29", "full_59", "maximum_29", "amin_29", "full_58", "minimum_29", "sub_1341", "div_58", "clamp_min_87", "div_59", "round_59", "sub_1347", "clamp_min_88", "clamp_max_58", "convert_element_type_174"], "quantize_affine_29": ["mul_2828", "view_457", "reciprocal_29", "mul_2874", "mul_2877", "round_60", "view_458", "add_4553", "clamp_min_89", "clamp_max_59", "convert_element_type_175"], "dequantize_affine_58": ["convert_element_type_175", "view_462", "convert_element_type_176", "convert_element_type_177", "sub_1367", "view_461", "mul_2899"], "dequantize_affine_59": ["view_464", "view_466", "convert_element_type_178", "convert_element_type_179", "sub_1371", "view_465", "mul_2904", "view_467"], "linear_29": ["mul_2909", "view_468", "permute_50", "mm_default_135", "add_tensor_135", "view_469"], "dropout_15": ["view_469"], "add_70": ["add_4616"], "layer_norm_10": ["clone_41", "var_mean_10", "getitem_40", "getitem_41", "sub_1377", "add_4621", "rsqrt_10", "mul_2920", "mul_2921", "add_4622"], "choose_qparams_affine_default_30": ["add_4622", "amax_30", "full_61", "maximum_30", "amin_30", "full_60", "minimum_30", "sub_1388", "div_60", "clamp_min_90", "div_61", "round_61", "sub_1394", "clamp_min_91", "clamp_max_60", "convert_element_type_180"], "quantize_affine_30": ["add_4622", "view_472", "reciprocal_30", "mul_2969", "mul_2972", "round_62", "view_473", "add_4709", "clamp_min_92", "clamp_max_61", "convert_element_type_181"], "choose_qparams_affine_default_31": ["add_4622", "amax_31", "full_63", "maximum_31", "amin_31", "full_62", "minimum_31", "sub_1433", "div_62", "clamp_min_93", "div_63", "round_63", "sub_1439", "clamp_min_94", "clamp_max_62", "convert_element_type_186"], "quantize_affine_31": ["add_4622", "view_488", "reciprocal_31", "mul_3065", "mul_3068", "round_64", "view_489", "add_4861", "clamp_min_95", "clamp_max_63", "convert_element_type_187"], "choose_qparams_affine_default_32": ["add_4622", "amax_32", "full_65", "maximum_32", "amin_32", "full_64", "minimum_32", "sub_1477", "div_64", "clamp_min_96", "div_65", "round_65", "sub_1483", "clamp_min_97", "clamp_max_64", "convert_element_type_192"], "quantize_affine_32": ["add_4622", "view_504", "reciprocal_32", "mul_3164", "mul_3167", "round_66", "view_505", "add_5009", "clamp_min_98", "clamp_max_65", "convert_element_type_193"], "dequantize_affine_60": ["convert_element_type_181", "view_477", "convert_element_type_182", "convert_element_type_183", "sub_1414", "view_476", "mul_2994"], "dequantize_affine_61": ["view_479", "view_481", "convert_element_type_184", "convert_element_type_185", "sub_1418", "view_480", "mul_2999", "view_482"], "dequantize_affine_62": ["convert_element_type_187", "view_493", "convert_element_type_188", "convert_element_type_189", "sub_1459", "view_492", "mul_3090"], "dequantize_affine_63": ["view_495", "view_497", "convert_element_type_190", "convert_element_type_191", "sub_1463", "view_496", "mul_3095", "view_498"], "dequantize_affine_64": ["convert_element_type_193", "view_509", "convert_element_type_194", "convert_element_type_195", "sub_1503", "view_508", "mul_3189"], "dequantize_affine_65": ["view_511", "view_513", "convert_element_type_196", "convert_element_type_197", "sub_1507", "view_512", "mul_3194", "view_514"], "linear_30": ["mul_3004", "view_483", "permute_51", "mm_default_134", "add_tensor_134", "view_484"], "mul_190": ["mul_3011"], "view_15": ["view_485"], "transpose_20": ["permute_52"], "contiguous_20": ["clone_42"], "linear_31": ["mul_3098", "view_499", "permute_53", "mm_5", "view_500"], "view_16": ["view_501"], "transpose_21": ["permute_54"], "contiguous_21": ["clone_43"], "linear_32": ["mul_3199", "view_515", "permute_55", "mm_default_133", "add_tensor_133", "view_516"], "view_17": ["view_517"], "transpose_22": ["permute_56"], "contiguous_22": ["clone_44"], "scaled_dot_product_attention_5": ["_scaled_dot_product_efficient_attention_5", "getitem_42"], "transpose_23": ["permute_57"], "reshape_5": ["view_518"], "choose_qparams_affine_default_33": ["view_518", "amax_33", "full_67", "maximum_33", "amin_33", "full_66", "minimum_33", "sub_1525", "div_66", "clamp_min_99", "div_67", "round_67", "sub_1531", "clamp_min_100", "clamp_max_66", "convert_element_type_198"], "quantize_affine_33": ["view_518", "view_521", "reciprocal_33", "mul_3269", "mul_3272", "round_68", "view_522", "add_5173", "clamp_min_101", "clamp_max_67", "convert_element_type_199"], "dequantize_affine_66": ["convert_element_type_199", "view_526", "convert_element_type_200", "convert_element_type_201", "sub_1551", "view_525", "mul_3294"], "dequantize_affine_67": ["view_528", "view_530", "convert_element_type_202", "convert_element_type_203", "sub_1555", "view_529", "mul_3299", "view_531"], "linear_33": ["mul_3304", "view_532", "permute_58", "mm_default_132", "add_tensor_132", "view_533"], "dropout_16": ["view_533"], "add_79": ["add_5236"], "layer_norm_11": ["clone_46", "var_mean_11", "getitem_46", "getitem_47", "sub_1561", "add_5241", "rsqrt_11", "mul_3315", "mul_3316", "add_5242"], "choose_qparams_affine_default_34": ["add_5242", "amax_34", "full_69", "maximum_34", "amin_34", "full_68", "minimum_34", "sub_1572", "div_68", "clamp_min_102", "div_69", "round_69", "sub_1578", "clamp_min_103", "clamp_max_68", "convert_element_type_204"], "quantize_affine_34": ["add_5242", "view_536", "reciprocal_34", "mul_3364", "mul_3367", "round_70", "view_537", "add_5329", "clamp_min_104", "clamp_max_69", "convert_element_type_205"], "dequantize_affine_68": ["convert_element_type_205", "view_541", "convert_element_type_206", "convert_element_type_207", "sub_1598", "view_540", "mul_3389"], "dequantize_affine_69": ["view_543", "view_545", "convert_element_type_208", "convert_element_type_209", "sub_1602", "view_544", "mul_3394", "view_546"], "linear_34": ["mul_3399", "view_547", "permute_59", "mm_default_131", "add_tensor_131", "view_548"], "gelu_7": ["mul_3406", "mul_3407", "erf_7", "add_5388", "mul_3408"], "dropout_17": ["mul_3408"], "choose_qparams_affine_default_35": ["mul_3408", "amax_35", "full_71", "maximum_35", "amin_35", "full_70", "minimum_35", "sub_1615", "div_70", "clamp_min_105", "div_71", "round_71", "sub_1621", "clamp_min_106", "clamp_max_70", "convert_element_type_210"], "quantize_affine_35": ["mul_3408", "view_551", "reciprocal_35", "mul_3454", "mul_3457", "round_72", "view_552", "add_5471", "clamp_min_107", "clamp_max_71", "convert_element_type_211"], "dequantize_affine_70": ["convert_element_type_211", "view_556", "convert_element_type_212", "convert_element_type_213", "sub_1641", "view_555", "mul_3479"], "dequantize_affine_71": ["view_558", "view_560", "convert_element_type_214", "convert_element_type_215", "sub_1645", "view_559", "mul_3484", "view_561"], "linear_35": ["mul_3489", "view_562", "permute_60", "mm_default_130", "add_tensor_130", "view_563"], "dropout_18": ["view_563"], "add_84": ["add_5534"], "layer_norm_12": ["clone_49", "var_mean_12", "getitem_48", "getitem_49", "sub_1651", "add_5539", "rsqrt_12", "mul_3500", "mul_3501", "add_5540"], "choose_qparams_affine_default_36": ["add_5540", "amax_36", "full_73", "maximum_36", "amin_36", "full_72", "minimum_36", "sub_1662", "div_72", "clamp_min_108", "div_73", "round_73", "sub_1668", "clamp_min_109", "clamp_max_72", "convert_element_type_216"], "quantize_affine_36": ["add_5540", "view_566", "reciprocal_36", "mul_3549", "mul_3552", "round_74", "view_567", "add_5627", "clamp_min_110", "clamp_max_73", "convert_element_type_217"], "choose_qparams_affine_default_37": ["add_5540", "amax_37", "full_75", "maximum_37", "amin_37", "full_74", "minimum_37", "sub_1707", "div_74", "clamp_min_111", "div_75", "round_75", "sub_1713", "clamp_min_112", "clamp_max_74", "convert_element_type_222"], "quantize_affine_37": ["add_5540", "view_582", "reciprocal_37", "mul_3645", "mul_3648", "round_76", "view_583", "add_5779", "clamp_min_113", "clamp_max_75", "convert_element_type_223"], "choose_qparams_affine_default_38": ["add_5540", "amax_38", "full_77", "maximum_38", "amin_38", "full_76", "minimum_38", "sub_1751", "div_76", "clamp_min_114", "div_77", "round_77", "sub_1757", "clamp_min_115", "clamp_max_76", "convert_element_type_228"], "quantize_affine_38": ["add_5540", "view_598", "reciprocal_38", "mul_3744", "mul_3747", "round_78", "view_599", "add_5927", "clamp_min_116", "clamp_max_77", "convert_element_type_229"], "dequantize_affine_72": ["convert_element_type_217", "view_571", "convert_element_type_218", "convert_element_type_219", "sub_1688", "view_570", "mul_3574"], "dequantize_affine_73": ["view_573", "view_575", "convert_element_type_220", "convert_element_type_221", "sub_1692", "view_574", "mul_3579", "view_576"], "dequantize_affine_74": ["convert_element_type_223", "view_587", "convert_element_type_224", "convert_element_type_225", "sub_1733", "view_586", "mul_3670"], "dequantize_affine_75": ["view_589", "view_591", "convert_element_type_226", "convert_element_type_227", "sub_1737", "view_590", "mul_3675", "view_592"], "dequantize_affine_76": ["convert_element_type_229", "view_603", "convert_element_type_230", "convert_element_type_231", "sub_1777", "view_602", "mul_3769"], "dequantize_affine_77": ["view_605", "view_607", "convert_element_type_232", "convert_element_type_233", "sub_1781", "view_606", "mul_3774", "view_608"], "linear_36": ["mul_3584", "view_577", "permute_61", "mm_default_129", "add_tensor_129", "view_578"], "mul_227": ["mul_3591"], "view_18": ["view_579"], "transpose_24": ["permute_62"], "contiguous_24": ["clone_50"], "linear_37": ["mul_3678", "view_593", "permute_63", "mm_6", "view_594"], "view_19": ["view_595"], "transpose_25": ["permute_64"], "contiguous_25": ["clone_51"], "linear_38": ["mul_3779", "view_609", "permute_65", "mm_default_128", "add_tensor_128", "view_610"], "view_20": ["view_611"], "transpose_26": ["permute_66"], "contiguous_26": ["clone_52"], "scaled_dot_product_attention_6": ["_scaled_dot_product_efficient_attention_6", "getitem_50"], "transpose_27": ["permute_67"], "reshape_6": ["view_612"], "choose_qparams_affine_default_39": ["view_612", "amax_39", "full_79", "maximum_39", "amin_39", "full_78", "minimum_39", "sub_1799", "div_78", "clamp_min_117", "div_79", "round_79", "sub_1805", "clamp_min_118", "clamp_max_78", "convert_element_type_234"], "quantize_affine_39": ["view_612", "view_615", "reciprocal_39", "mul_3849", "mul_3852", "round_80", "view_616", "add_6091", "clamp_min_119", "clamp_max_79", "convert_element_type_235"], "dequantize_affine_78": ["convert_element_type_235", "view_620", "convert_element_type_236", "convert_element_type_237", "sub_1825", "view_619", "mul_3874"], "dequantize_affine_79": ["view_622", "view_624", "convert_element_type_238", "convert_element_type_239", "sub_1829", "view_623", "mul_3879", "view_625"], "linear_39": ["mul_3884", "view_626", "permute_68", "mm_default_127", "add_tensor_127", "view_627"], "dropout_19": ["view_627"], "add_93": ["add_6154"], "layer_norm_13": ["clone_54", "var_mean_13", "getitem_54", "getitem_55", "sub_1835", "add_6159", "rsqrt_13", "mul_3895", "mul_3896", "add_6160"], "choose_qparams_affine_default_40": ["add_6160", "amax_40", "full_81", "maximum_40", "amin_40", "full_80", "minimum_40", "sub_1846", "div_80", "clamp_min_120", "div_81", "round_81", "sub_1852", "clamp_min_121", "clamp_max_80", "convert_element_type_240"], "quantize_affine_40": ["add_6160", "view_630", "reciprocal_40", "mul_3944", "mul_3947", "round_82", "view_631", "add_6247", "clamp_min_122", "clamp_max_81", "convert_element_type_241"], "dequantize_affine_80": ["convert_element_type_241", "view_635", "convert_element_type_242", "convert_element_type_243", "sub_1872", "view_634", "mul_3969"], "dequantize_affine_81": ["view_637", "view_639", "convert_element_type_244", "convert_element_type_245", "sub_1876", "view_638", "mul_3974", "view_640"], "linear_40": ["mul_3979", "view_641", "permute_69", "mm_default_126", "add_tensor_126", "view_642"], "gelu_8": ["mul_3986", "mul_3987", "erf_8", "add_6306", "mul_3988"], "dropout_20": ["mul_3988"], "choose_qparams_affine_default_41": ["mul_3988", "amax_41", "full_83", "maximum_41", "amin_41", "full_82", "minimum_41", "sub_1889", "div_82", "clamp_min_123", "div_83", "round_83", "sub_1895", "clamp_min_124", "clamp_max_82", "convert_element_type_246"], "quantize_affine_41": ["mul_3988", "view_645", "reciprocal_41", "mul_4034", "mul_4037", "round_84", "view_646", "add_6389", "clamp_min_125", "clamp_max_83", "convert_element_type_247"], "dequantize_affine_82": ["convert_element_type_247", "view_650", "convert_element_type_248", "convert_element_type_249", "sub_1915", "view_649", "mul_4059"], "dequantize_affine_83": ["view_652", "view_654", "convert_element_type_250", "convert_element_type_251", "sub_1919", "view_653", "mul_4064", "view_655"], "linear_41": ["mul_4069", "view_656", "permute_70", "mm_default_125", "add_tensor_125", "view_657"], "dropout_21": ["view_657"], "add_98": ["add_6452"], "layer_norm_14": ["clone_57", "var_mean_14", "getitem_56", "getitem_57", "sub_1925", "add_6457", "rsqrt_14", "mul_4080", "mul_4081", "add_6458"], "choose_qparams_affine_default_42": ["add_6458", "amax_42", "full_85", "maximum_42", "amin_42", "full_84", "minimum_42", "sub_1936", "div_84", "clamp_min_126", "div_85", "round_85", "sub_1942", "clamp_min_127", "clamp_max_84", "convert_element_type_252"], "quantize_affine_42": ["add_6458", "view_660", "reciprocal_42", "mul_4129", "mul_4132", "round_86", "view_661", "add_6545", "clamp_min_128", "clamp_max_85", "convert_element_type_253"], "choose_qparams_affine_default_43": ["add_6458", "amax_43", "full_87", "maximum_43", "amin_43", "full_86", "minimum_43", "sub_1981", "div_86", "clamp_min_129", "div_87", "round_87", "sub_1987", "clamp_min_130", "clamp_max_86", "convert_element_type_258"], "quantize_affine_43": ["add_6458", "view_676", "reciprocal_43", "mul_4225", "mul_4228", "round_88", "view_677", "add_6697", "clamp_min_131", "clamp_max_87", "convert_element_type_259"], "choose_qparams_affine_default_44": ["add_6458", "amax_44", "full_89", "maximum_44", "amin_44", "full_88", "minimum_44", "sub_2025", "div_88", "clamp_min_132", "div_89", "round_89", "sub_2031", "clamp_min_133", "clamp_max_88", "convert_element_type_264"], "quantize_affine_44": ["add_6458", "view_692", "reciprocal_44", "mul_4324", "mul_4327", "round_90", "view_693", "add_6845", "clamp_min_134", "clamp_max_89", "convert_element_type_265"], "dequantize_affine_84": ["convert_element_type_253", "view_665", "convert_element_type_254", "convert_element_type_255", "sub_1962", "view_664", "mul_4154"], "dequantize_affine_85": ["view_667", "view_669", "convert_element_type_256", "convert_element_type_257", "sub_1966", "view_668", "mul_4159", "view_670"], "dequantize_affine_86": ["convert_element_type_259", "view_681", "convert_element_type_260", "convert_element_type_261", "sub_2007", "view_680", "mul_4250"], "dequantize_affine_87": ["view_683", "view_685", "convert_element_type_262", "convert_element_type_263", "sub_2011", "view_684", "mul_4255", "view_686"], "dequantize_affine_88": ["convert_element_type_265", "view_697", "convert_element_type_266", "convert_element_type_267", "sub_2051", "view_696", "mul_4349"], "dequantize_affine_89": ["view_699", "view_701", "convert_element_type_268", "convert_element_type_269", "sub_2055", "view_700", "mul_4354", "view_702"], "linear_42": ["mul_4164", "view_671", "permute_71", "mm_default_124", "add_tensor_124", "view_672"], "mul_264": ["mul_4171"], "view_21": ["view_673"], "transpose_28": ["permute_72"], "contiguous_28": ["clone_58"], "linear_43": ["mul_4258", "view_687", "permute_73", "mm_7", "view_688"], "view_22": ["view_689"], "transpose_29": ["permute_74"], "contiguous_29": ["clone_59"], "linear_44": ["mul_4359", "view_703", "permute_75", "mm_default_123", "add_tensor_123", "view_704"], "view_23": ["view_705"], "transpose_30": ["permute_76"], "contiguous_30": ["clone_60"], "scaled_dot_product_attention_7": ["_scaled_dot_product_efficient_attention_7", "getitem_58"], "transpose_31": ["permute_77"], "reshape_7": ["view_706"], "choose_qparams_affine_default_45": ["view_706", "amax_45", "full_91", "maximum_45", "amin_45", "full_90", "minimum_45", "sub_2073", "div_90", "clamp_min_135", "div_91", "round_91", "sub_2079", "clamp_min_136", "clamp_max_90", "convert_element_type_270"], "quantize_affine_45": ["view_706", "view_709", "reciprocal_45", "mul_4429", "mul_4432", "round_92", "view_710", "add_7009", "clamp_min_137", "clamp_max_91", "convert_element_type_271"], "dequantize_affine_90": ["convert_element_type_271", "view_714", "convert_element_type_272", "convert_element_type_273", "sub_2099", "view_713", "mul_4454"], "dequantize_affine_91": ["view_716", "view_718", "convert_element_type_274", "convert_element_type_275", "sub_2103", "view_717", "mul_4459", "view_719"], "linear_45": ["mul_4464", "view_720", "permute_78", "mm_default_122", "add_tensor_122", "view_721"], "dropout_22": ["view_721"], "add_107": ["add_7072"], "layer_norm_15": ["clone_62", "var_mean_15", "getitem_62", "getitem_63", "sub_2109", "add_7077", "rsqrt_15", "mul_4475", "mul_4476", "add_7078"], "choose_qparams_affine_default_46": ["add_7078", "amax_46", "full_93", "maximum_46", "amin_46", "full_92", "minimum_46", "sub_2120", "div_92", "clamp_min_138", "div_93", "round_93", "sub_2126", "clamp_min_139", "clamp_max_92", "convert_element_type_276"], "quantize_affine_46": ["add_7078", "view_724", "reciprocal_46", "mul_4524", "mul_4527", "round_94", "view_725", "add_7165", "clamp_min_140", "clamp_max_93", "convert_element_type_277"], "dequantize_affine_92": ["convert_element_type_277", "view_729", "convert_element_type_278", "convert_element_type_279", "sub_2146", "view_728", "mul_4549"], "dequantize_affine_93": ["view_731", "view_733", "convert_element_type_280", "convert_element_type_281", "sub_2150", "view_732", "mul_4554", "view_734"], "linear_46": ["mul_4559", "view_735", "permute_79", "mm_default_121", "add_tensor_121", "view_736"], "gelu_9": ["mul_4566", "mul_4567", "erf_9", "add_7224", "mul_4568"], "dropout_23": ["mul_4568"], "choose_qparams_affine_default_47": ["mul_4568", "amax_47", "full_95", "maximum_47", "amin_47", "full_94", "minimum_47", "sub_2163", "div_94", "clamp_min_141", "div_95", "round_95", "sub_2169", "clamp_min_142", "clamp_max_94", "convert_element_type_282"], "quantize_affine_47": ["mul_4568", "view_739", "reciprocal_47", "mul_4614", "mul_4617", "round_96", "view_740", "add_7307", "clamp_min_143", "clamp_max_95", "convert_element_type_283"], "dequantize_affine_94": ["convert_element_type_283", "view_744", "convert_element_type_284", "convert_element_type_285", "sub_2189", "view_743", "mul_4639"], "dequantize_affine_95": ["view_746", "view_748", "convert_element_type_286", "convert_element_type_287", "sub_2193", "view_747", "mul_4644", "view_749"], "linear_47": ["mul_4649", "view_750", "permute_80", "mm_default_120", "add_tensor_120", "view_751"], "dropout_24": ["view_751"], "add_112": ["add_7370"], "layer_norm_16": ["clone_65", "var_mean_16", "getitem_64", "getitem_65", "sub_2199", "add_7375", "rsqrt_16", "mul_4660", "mul_4661", "add_7376"], "choose_qparams_affine_default_48": ["add_7376", "amax_48", "full_97", "maximum_48", "amin_48", "full_96", "minimum_48", "sub_2210", "div_96", "clamp_min_144", "div_97", "round_97", "sub_2216", "clamp_min_145", "clamp_max_96", "convert_element_type_288"], "quantize_affine_48": ["add_7376", "view_754", "reciprocal_48", "mul_4709", "mul_4712", "round_98", "view_755", "add_7463", "clamp_min_146", "clamp_max_97", "convert_element_type_289"], "choose_qparams_affine_default_49": ["add_7376", "amax_49", "full_99", "maximum_49", "amin_49", "full_98", "minimum_49", "sub_2255", "div_98", "clamp_min_147", "div_99", "round_99", "sub_2261", "clamp_min_148", "clamp_max_98", "convert_element_type_294"], "quantize_affine_49": ["add_7376", "view_770", "reciprocal_49", "mul_4805", "mul_4808", "round_100", "view_771", "add_7615", "clamp_min_149", "clamp_max_99", "convert_element_type_295"], "choose_qparams_affine_default_50": ["add_7376", "amax_50", "full_101", "maximum_50", "amin_50", "full_100", "minimum_50", "sub_2299", "div_100", "clamp_min_150", "div_101", "round_101", "sub_2305", "clamp_min_151", "clamp_max_100", "convert_element_type_300"], "quantize_affine_50": ["add_7376", "view_786", "reciprocal_50", "mul_4904", "mul_4907", "round_102", "view_787", "add_7763", "clamp_min_152", "clamp_max_101", "convert_element_type_301"], "dequantize_affine_96": ["convert_element_type_289", "view_759", "convert_element_type_290", "convert_element_type_291", "sub_2236", "view_758", "mul_4734"], "dequantize_affine_97": ["view_761", "view_763", "convert_element_type_292", "convert_element_type_293", "sub_2240", "view_762", "mul_4739", "view_764"], "dequantize_affine_98": ["convert_element_type_295", "view_775", "convert_element_type_296", "convert_element_type_297", "sub_2281", "view_774", "mul_4830"], "dequantize_affine_99": ["view_777", "view_779", "convert_element_type_298", "convert_element_type_299", "sub_2285", "view_778", "mul_4835", "view_780"], "dequantize_affine_100": ["convert_element_type_301", "view_791", "convert_element_type_302", "convert_element_type_303", "sub_2325", "view_790", "mul_4929"], "dequantize_affine_101": ["view_793", "view_795", "convert_element_type_304", "convert_element_type_305", "sub_2329", "view_794", "mul_4934", "view_796"], "linear_48": ["mul_4744", "view_765", "permute_81", "mm_default_119", "add_tensor_119", "view_766"], "mul_301": ["mul_4751"], "view_24": ["view_767"], "transpose_32": ["permute_82"], "contiguous_32": ["clone_66"], "linear_49": ["mul_4838", "view_781", "permute_83", "mm_8", "view_782"], "view_25": ["view_783"], "transpose_33": ["permute_84"], "contiguous_33": ["clone_67"], "linear_50": ["mul_4939", "view_797", "permute_85", "mm_default_118", "add_tensor_118", "view_798"], "view_26": ["view_799"], "transpose_34": ["permute_86"], "contiguous_34": ["clone_68"], "scaled_dot_product_attention_8": ["_scaled_dot_product_efficient_attention_8", "getitem_66"], "transpose_35": ["permute_87"], "reshape_8": ["view_800"], "choose_qparams_affine_default_51": ["view_800", "amax_51", "full_103", "maximum_51", "amin_51", "full_102", "minimum_51", "sub_2347", "div_102", "clamp_min_153", "div_103", "round_103", "sub_2353", "clamp_min_154", "clamp_max_102", "convert_element_type_306"], "quantize_affine_51": ["view_800", "view_803", "reciprocal_51", "mul_5009", "mul_5012", "round_104", "view_804", "add_7927", "clamp_min_155", "clamp_max_103", "convert_element_type_307"], "dequantize_affine_102": ["convert_element_type_307", "view_808", "convert_element_type_308", "convert_element_type_309", "sub_2373", "view_807", "mul_5034"], "dequantize_affine_103": ["view_810", "view_812", "convert_element_type_310", "convert_element_type_311", "sub_2377", "view_811", "mul_5039", "view_813"], "linear_51": ["mul_5044", "view_814", "permute_88", "mm_default_117", "add_tensor_117", "view_815"], "dropout_25": ["view_815"], "add_121": ["add_7990"], "layer_norm_17": ["clone_70", "var_mean_17", "getitem_70", "getitem_71", "sub_2383", "add_7995", "rsqrt_17", "mul_5055", "mul_5056", "add_7996"], "choose_qparams_affine_default_52": ["add_7996", "amax_52", "full_105", "maximum_52", "amin_52", "full_104", "minimum_52", "sub_2394", "div_104", "clamp_min_156", "div_105", "round_105", "sub_2400", "clamp_min_157", "clamp_max_104", "convert_element_type_312"], "quantize_affine_52": ["add_7996", "view_818", "reciprocal_52", "mul_5104", "mul_5107", "round_106", "view_819", "add_8083", "clamp_min_158", "clamp_max_105", "convert_element_type_313"], "dequantize_affine_104": ["convert_element_type_313", "view_823", "convert_element_type_314", "convert_element_type_315", "sub_2420", "view_822", "mul_5129"], "dequantize_affine_105": ["view_825", "view_827", "convert_element_type_316", "convert_element_type_317", "sub_2424", "view_826", "mul_5134", "view_828"], "linear_52": ["mul_5139", "view_829", "permute_89", "mm_default_116", "add_tensor_116", "view_830"], "gelu_10": ["mul_5146", "mul_5147", "erf_10", "add_8142", "mul_5148"], "dropout_26": ["mul_5148"], "choose_qparams_affine_default_53": ["mul_5148", "amax_53", "full_107", "maximum_53", "amin_53", "full_106", "minimum_53", "sub_2437", "div_106", "clamp_min_159", "div_107", "round_107", "sub_2443", "clamp_min_160", "clamp_max_106", "convert_element_type_318"], "quantize_affine_53": ["mul_5148", "view_833", "reciprocal_53", "mul_5194", "mul_5197", "round_108", "view_834", "add_8225", "clamp_min_161", "clamp_max_107", "convert_element_type_319"], "dequantize_affine_106": ["convert_element_type_319", "view_838", "convert_element_type_320", "convert_element_type_321", "sub_2463", "view_837", "mul_5219"], "dequantize_affine_107": ["view_840", "view_842", "convert_element_type_322", "convert_element_type_323", "sub_2467", "view_841", "mul_5224", "view_843"], "linear_53": ["mul_5229", "view_844", "permute_90", "mm_default_115", "add_tensor_115", "view_845"], "dropout_27": ["view_845"], "add_126": ["add_8288"], "layer_norm_18": ["clone_73", "var_mean_18", "getitem_72", "getitem_73", "sub_2473", "add_8293", "rsqrt_18", "mul_5240", "mul_5241", "add_8294"], "choose_qparams_affine_default_54": ["add_8294", "amax_54", "full_109", "maximum_54", "amin_54", "full_108", "minimum_54", "sub_2484", "div_108", "clamp_min_162", "div_109", "round_109", "sub_2490", "clamp_min_163", "clamp_max_108", "convert_element_type_324"], "quantize_affine_54": ["add_8294", "view_848", "reciprocal_54", "mul_5289", "mul_5292", "round_110", "view_849", "add_8381", "clamp_min_164", "clamp_max_109", "convert_element_type_325"], "choose_qparams_affine_default_55": ["add_8294", "amax_55", "full_111", "maximum_55", "amin_55", "full_110", "minimum_55", "sub_2529", "div_110", "clamp_min_165", "div_111", "round_111", "sub_2535", "clamp_min_166", "clamp_max_110", "convert_element_type_330"], "quantize_affine_55": ["add_8294", "view_864", "reciprocal_55", "mul_5385", "mul_5388", "round_112", "view_865", "add_8533", "clamp_min_167", "clamp_max_111", "convert_element_type_331"], "choose_qparams_affine_default_56": ["add_8294", "amax_56", "full_113", "maximum_56", "amin_56", "full_112", "minimum_56", "sub_2573", "div_112", "clamp_min_168", "div_113", "round_113", "sub_2579", "clamp_min_169", "clamp_max_112", "convert_element_type_336"], "quantize_affine_56": ["add_8294", "view_880", "reciprocal_56", "mul_5484", "mul_5487", "round_114", "view_881", "add_8681", "clamp_min_170", "clamp_max_113", "convert_element_type_337"], "dequantize_affine_108": ["convert_element_type_325", "view_853", "convert_element_type_326", "convert_element_type_327", "sub_2510", "view_852", "mul_5314"], "dequantize_affine_109": ["view_855", "view_857", "convert_element_type_328", "convert_element_type_329", "sub_2514", "view_856", "mul_5319", "view_858"], "dequantize_affine_110": ["convert_element_type_331", "view_869", "convert_element_type_332", "convert_element_type_333", "sub_2555", "view_868", "mul_5410"], "dequantize_affine_111": ["view_871", "view_873", "convert_element_type_334", "convert_element_type_335", "sub_2559", "view_872", "mul_5415", "view_874"], "dequantize_affine_112": ["convert_element_type_337", "view_885", "convert_element_type_338", "convert_element_type_339", "sub_2599", "view_884", "mul_5509"], "dequantize_affine_113": ["view_887", "view_889", "convert_element_type_340", "convert_element_type_341", "sub_2603", "view_888", "mul_5514", "view_890"], "linear_54": ["mul_5324", "view_859", "permute_91", "mm_default_114", "add_tensor_114", "view_860"], "mul_338": ["mul_5331"], "view_27": ["view_861"], "transpose_36": ["permute_92"], "contiguous_36": ["clone_74"], "linear_55": ["mul_5418", "view_875", "permute_93", "mm_9", "view_876"], "view_28": ["view_877"], "transpose_37": ["permute_94"], "contiguous_37": ["clone_75"], "linear_56": ["mul_5519", "view_891", "permute_95", "mm_default_113", "add_tensor_113", "view_892"], "view_29": ["view_893"], "transpose_38": ["permute_96"], "contiguous_38": ["clone_76"], "scaled_dot_product_attention_9": ["_scaled_dot_product_efficient_attention_9", "getitem_74"], "transpose_39": ["permute_97"], "reshape_9": ["view_894"], "choose_qparams_affine_default_57": ["view_894", "amax_57", "full_115", "maximum_57", "amin_57", "full_114", "minimum_57", "sub_2621", "div_114", "clamp_min_171", "div_115", "round_115", "sub_2627", "clamp_min_172", "clamp_max_114", "convert_element_type_342"], "quantize_affine_57": ["view_894", "view_897", "reciprocal_57", "mul_5589", "mul_5592", "round_116", "view_898", "add_8845", "clamp_min_173", "clamp_max_115", "convert_element_type_343"], "dequantize_affine_114": ["convert_element_type_343", "view_902", "convert_element_type_344", "convert_element_type_345", "sub_2647", "view_901", "mul_5614"], "dequantize_affine_115": ["view_904", "view_906", "convert_element_type_346", "convert_element_type_347", "sub_2651", "view_905", "mul_5619", "view_907"], "linear_57": ["mul_5624", "view_908", "permute_98", "mm_default_112", "add_tensor_112", "view_909"], "dropout_28": ["view_909"], "add_135": ["add_8908"], "layer_norm_19": ["clone_78", "var_mean_19", "getitem_78", "getitem_79", "sub_2657", "add_8913", "rsqrt_19", "mul_5635", "mul_5636", "add_8914"], "choose_qparams_affine_default_58": ["add_8914", "amax_58", "full_117", "maximum_58", "amin_58", "full_116", "minimum_58", "sub_2668", "div_116", "clamp_min_174", "div_117", "round_117", "sub_2674", "clamp_min_175", "clamp_max_116", "convert_element_type_348"], "quantize_affine_58": ["add_8914", "view_912", "reciprocal_58", "mul_5684", "mul_5687", "round_118", "view_913", "add_9001", "clamp_min_176", "clamp_max_117", "convert_element_type_349"], "dequantize_affine_116": ["convert_element_type_349", "view_917", "convert_element_type_350", "convert_element_type_351", "sub_2694", "view_916", "mul_5709"], "dequantize_affine_117": ["view_919", "view_921", "convert_element_type_352", "convert_element_type_353", "sub_2698", "view_920", "mul_5714", "view_922"], "linear_58": ["mul_5719", "view_923", "permute_99", "mm_default_111", "add_tensor_111", "view_924"], "gelu_11": ["mul_5726", "mul_5727", "erf_11", "add_9060", "mul_5728"], "dropout_29": ["mul_5728"], "choose_qparams_affine_default_59": ["mul_5728", "amax_59", "full_119", "maximum_59", "amin_59", "full_118", "minimum_59", "sub_2711", "div_118", "clamp_min_177", "div_119", "round_119", "sub_2717", "clamp_min_178", "clamp_max_118", "convert_element_type_354"], "quantize_affine_59": ["mul_5728", "view_927", "reciprocal_59", "mul_5774", "mul_5777", "round_120", "view_928", "add_9143", "clamp_min_179", "clamp_max_119", "convert_element_type_355"], "dequantize_affine_118": ["convert_element_type_355", "view_932", "convert_element_type_356", "convert_element_type_357", "sub_2737", "view_931", "mul_5799"], "dequantize_affine_119": ["view_934", "view_936", "convert_element_type_358", "convert_element_type_359", "sub_2741", "view_935", "mul_5804", "view_937"], "linear_59": ["mul_5809", "view_938", "permute_100", "mm_default_110", "add_tensor_110", "view_939"], "dropout_30": ["view_939"], "add_140": ["add_9206"], "layer_norm_20": ["clone_81", "var_mean_20", "getitem_80", "getitem_81", "sub_2747", "add_9211", "rsqrt_20", "mul_5820", "mul_5821", "add_9212"], "choose_qparams_affine_default_60": ["add_9212", "amax_60", "full_121", "maximum_60", "amin_60", "full_120", "minimum_60", "sub_2758", "div_120", "clamp_min_180", "div_121", "round_121", "sub_2764", "clamp_min_181", "clamp_max_120", "convert_element_type_360"], "quantize_affine_60": ["add_9212", "view_942", "reciprocal_60", "mul_5869", "mul_5872", "round_122", "view_943", "add_9299", "clamp_min_182", "clamp_max_121", "convert_element_type_361"], "choose_qparams_affine_default_61": ["add_9212", "amax_61", "full_123", "maximum_61", "amin_61", "full_122", "minimum_61", "sub_2803", "div_122", "clamp_min_183", "div_123", "round_123", "sub_2809", "clamp_min_184", "clamp_max_122", "convert_element_type_366"], "quantize_affine_61": ["add_9212", "view_958", "reciprocal_61", "mul_5965", "mul_5968", "round_124", "view_959", "add_9451", "clamp_min_185", "clamp_max_123", "convert_element_type_367"], "choose_qparams_affine_default_62": ["add_9212", "amax_62", "full_125", "maximum_62", "amin_62", "full_124", "minimum_62", "sub_2847", "div_124", "clamp_min_186", "div_125", "round_125", "sub_2853", "clamp_min_187", "clamp_max_124", "convert_element_type_372"], "quantize_affine_62": ["add_9212", "view_974", "reciprocal_62", "mul_6064", "mul_6067", "round_126", "view_975", "add_9599", "clamp_min_188", "clamp_max_125", "convert_element_type_373"], "dequantize_affine_120": ["convert_element_type_361", "view_947", "convert_element_type_362", "convert_element_type_363", "sub_2784", "view_946", "mul_5894"], "dequantize_affine_121": ["view_949", "view_951", "convert_element_type_364", "convert_element_type_365", "sub_2788", "view_950", "mul_5899", "view_952"], "dequantize_affine_122": ["convert_element_type_367", "view_963", "convert_element_type_368", "convert_element_type_369", "sub_2829", "view_962", "mul_5990"], "dequantize_affine_123": ["view_965", "view_967", "convert_element_type_370", "convert_element_type_371", "sub_2833", "view_966", "mul_5995", "view_968"], "dequantize_affine_124": ["convert_element_type_373", "view_979", "convert_element_type_374", "convert_element_type_375", "sub_2873", "view_978", "mul_6089"], "dequantize_affine_125": ["view_981", "view_983", "convert_element_type_376", "convert_element_type_377", "sub_2877", "view_982", "mul_6094", "view_984"], "linear_60": ["mul_5904", "view_953", "permute_101", "mm_default_109", "add_tensor_109", "view_954"], "mul_375": ["mul_5911"], "view_30": ["view_955"], "transpose_40": ["permute_102"], "contiguous_40": ["clone_82"], "linear_61": ["mul_5998", "view_969", "permute_103", "mm_10", "view_970"], "view_31": ["view_971"], "transpose_41": ["permute_104"], "contiguous_41": ["clone_83"], "linear_62": ["mul_6099", "view_985", "permute_105", "mm_default_108", "add_tensor_108", "view_986"], "view_32": ["view_987"], "transpose_42": ["permute_106"], "contiguous_42": ["clone_84"], "scaled_dot_product_attention_10": ["_scaled_dot_product_efficient_attention_10", "getitem_82"], "transpose_43": ["permute_107"], "reshape_10": ["view_988"], "choose_qparams_affine_default_63": ["view_988", "amax_63", "full_127", "maximum_63", "amin_63", "full_126", "minimum_63", "sub_2895", "div_126", "clamp_min_189", "div_127", "round_127", "sub_2901", "clamp_min_190", "clamp_max_126", "convert_element_type_378"], "quantize_affine_63": ["view_988", "view_991", "reciprocal_63", "mul_6169", "mul_6172", "round_128", "view_992", "add_9763", "clamp_min_191", "clamp_max_127", "convert_element_type_379"], "dequantize_affine_126": ["convert_element_type_379", "view_996", "convert_element_type_380", "convert_element_type_381", "sub_2921", "view_995", "mul_6194"], "dequantize_affine_127": ["view_998", "view_1000", "convert_element_type_382", "convert_element_type_383", "sub_2925", "view_999", "mul_6199", "view_1001"], "linear_63": ["mul_6204", "view_1002", "permute_108", "mm_default_107", "add_tensor_107", "view_1003"], "dropout_31": ["view_1003"], "add_149": ["add_9826"], "layer_norm_21": ["clone_86", "var_mean_21", "getitem_86", "getitem_87", "sub_2931", "add_9831", "rsqrt_21", "mul_6215", "mul_6216", "add_9832"], "choose_qparams_affine_default_64": ["add_9832", "amax_64", "full_129", "maximum_64", "amin_64", "full_128", "minimum_64", "sub_2942", "div_128", "clamp_min_192", "div_129", "round_129", "sub_2948", "clamp_min_193", "clamp_max_128", "convert_element_type_384"], "quantize_affine_64": ["add_9832", "view_1006", "reciprocal_64", "mul_6264", "mul_6267", "round_130", "view_1007", "add_9919", "clamp_min_194", "clamp_max_129", "convert_element_type_385"], "dequantize_affine_128": ["convert_element_type_385", "view_1011", "convert_element_type_386", "convert_element_type_387", "sub_2968", "view_1010", "mul_6289"], "dequantize_affine_129": ["view_1013", "view_1015", "convert_element_type_388", "convert_element_type_389", "sub_2972", "view_1014", "mul_6294", "view_1016"], "linear_64": ["mul_6299", "view_1017", "permute_109", "mm_default_106", "add_tensor_106", "view_1018"], "gelu_12": ["mul_6306", "mul_6307", "erf_12", "add_9978", "mul_6308"], "dropout_32": ["mul_6308"], "choose_qparams_affine_default_65": ["mul_6308", "amax_65", "full_131", "maximum_65", "amin_65", "full_130", "minimum_65", "sub_2985", "div_130", "clamp_min_195", "div_131", "round_131", "sub_2991", "clamp_min_196", "clamp_max_130", "convert_element_type_390"], "quantize_affine_65": ["mul_6308", "view_1021", "reciprocal_65", "mul_6354", "mul_6357", "round_132", "view_1022", "add_10061", "clamp_min_197", "clamp_max_131", "convert_element_type_391"], "dequantize_affine_130": ["convert_element_type_391", "view_1026", "convert_element_type_392", "convert_element_type_393", "sub_3011", "view_1025", "mul_6379"], "dequantize_affine_131": ["view_1028", "view_1030", "convert_element_type_394", "convert_element_type_395", "sub_3015", "view_1029", "mul_6384", "view_1031"], "linear_65": ["mul_6389", "view_1032", "permute_110", "mm_default_105", "add_tensor_105", "view_1033"], "dropout_33": ["view_1033"], "add_154": ["add_10124"], "layer_norm_22": ["clone_89", "var_mean_22", "getitem_88", "getitem_89", "sub_3021", "add_10129", "rsqrt_22", "mul_6400", "mul_6401", "add_10130"], "choose_qparams_affine_default_66": ["add_10130", "amax_66", "full_133", "maximum_66", "amin_66", "full_132", "minimum_66", "sub_3032", "div_132", "clamp_min_198", "div_133", "round_133", "sub_3038", "clamp_min_199", "clamp_max_132", "convert_element_type_396"], "quantize_affine_66": ["add_10130", "view_1036", "reciprocal_66", "mul_6449", "mul_6452", "round_134", "view_1037", "add_10217", "clamp_min_200", "clamp_max_133", "convert_element_type_397"], "choose_qparams_affine_default_67": ["add_10130", "amax_67", "full_135", "maximum_67", "amin_67", "full_134", "minimum_67", "sub_3077", "div_134", "clamp_min_201", "div_135", "round_135", "sub_3083", "clamp_min_202", "clamp_max_134", "convert_element_type_402"], "quantize_affine_67": ["add_10130", "view_1052", "reciprocal_67", "mul_6545", "mul_6548", "round_136", "view_1053", "add_10369", "clamp_min_203", "clamp_max_135", "convert_element_type_403"], "choose_qparams_affine_default_68": ["add_10130", "amax_68", "full_137", "maximum_68", "amin_68", "full_136", "minimum_68", "sub_3121", "div_136", "clamp_min_204", "div_137", "round_137", "sub_3127", "clamp_min_205", "clamp_max_136", "convert_element_type_408"], "quantize_affine_68": ["add_10130", "view_1068", "reciprocal_68", "mul_6644", "mul_6647", "round_138", "view_1069", "add_10517", "clamp_min_206", "clamp_max_137", "convert_element_type_409"], "dequantize_affine_132": ["convert_element_type_397", "view_1041", "convert_element_type_398", "convert_element_type_399", "sub_3058", "view_1040", "mul_6474"], "dequantize_affine_133": ["view_1043", "view_1045", "convert_element_type_400", "convert_element_type_401", "sub_3062", "view_1044", "mul_6479", "view_1046"], "dequantize_affine_134": ["convert_element_type_403", "view_1057", "convert_element_type_404", "convert_element_type_405", "sub_3103", "view_1056", "mul_6570"], "dequantize_affine_135": ["view_1059", "view_1061", "convert_element_type_406", "convert_element_type_407", "sub_3107", "view_1060", "mul_6575", "view_1062"], "dequantize_affine_136": ["convert_element_type_409", "view_1073", "convert_element_type_410", "convert_element_type_411", "sub_3147", "view_1072", "mul_6669"], "dequantize_affine_137": ["view_1075", "view_1077", "convert_element_type_412", "convert_element_type_413", "sub_3151", "view_1076", "mul_6674", "view_1078"], "linear_66": ["mul_6484", "view_1047", "permute_111", "mm_default_104", "add_tensor_104", "view_1048"], "mul_412": ["mul_6491"], "view_33": ["view_1049"], "transpose_44": ["permute_112"], "contiguous_44": ["clone_90"], "linear_67": ["mul_6578", "view_1063", "permute_113", "mm_11", "view_1064"], "view_34": ["view_1065"], "transpose_45": ["permute_114"], "contiguous_45": ["clone_91"], "linear_68": ["mul_6679", "view_1079", "permute_115", "mm_default_103", "add_tensor_103", "view_1080"], "view_35": ["view_1081"], "transpose_46": ["permute_116"], "contiguous_46": ["clone_92"], "scaled_dot_product_attention_11": ["_scaled_dot_product_efficient_attention_11", "getitem_90"], "transpose_47": ["permute_117"], "reshape_11": ["view_1082"], "choose_qparams_affine_default_69": ["view_1082", "amax_69", "full_139", "maximum_69", "amin_69", "full_138", "minimum_69", "sub_3169", "div_138", "clamp_min_207", "div_139", "round_139", "sub_3175", "clamp_min_208", "clamp_max_138", "convert_element_type_414"], "quantize_affine_69": ["view_1082", "view_1085", "reciprocal_69", "mul_6749", "mul_6752", "round_140", "view_1086", "add_10681", "clamp_min_209", "clamp_max_139", "convert_element_type_415"], "dequantize_affine_138": ["convert_element_type_415", "view_1090", "convert_element_type_416", "convert_element_type_417", "sub_3195", "view_1089", "mul_6774"], "dequantize_affine_139": ["view_1092", "view_1094", "convert_element_type_418", "convert_element_type_419", "sub_3199", "view_1093", "mul_6779", "view_1095"], "linear_69": ["mul_6784", "view_1096", "permute_118", "mm_default_102", "add_tensor_102", "view_1097"], "dropout_34": ["view_1097"], "add_163": ["add_10744"], "layer_norm_23": ["clone_94", "var_mean_23", "getitem_94", "getitem_95", "sub_3205", "add_10749", "rsqrt_23", "mul_6795", "mul_6796", "add_10750"], "choose_qparams_affine_default_70": ["add_10750", "amax_70", "full_141", "maximum_70", "amin_70", "full_140", "minimum_70", "sub_3216", "div_140", "clamp_min_210", "div_141", "round_141", "sub_3222", "clamp_min_211", "clamp_max_140", "convert_element_type_420"], "quantize_affine_70": ["add_10750", "view_1100", "reciprocal_70", "mul_6844", "mul_6847", "round_142", "view_1101", "add_10837", "clamp_min_212", "clamp_max_141", "convert_element_type_421"], "dequantize_affine_140": ["convert_element_type_421", "view_1105", "convert_element_type_422", "convert_element_type_423", "sub_3242", "view_1104", "mul_6869"], "dequantize_affine_141": ["view_1107", "view_1109", "convert_element_type_424", "convert_element_type_425", "sub_3246", "view_1108", "mul_6874", "view_1110"], "linear_70": ["mul_6879", "view_1111", "permute_119", "mm_default_101", "add_tensor_101", "view_1112"], "gelu_13": ["mul_6886", "mul_6887", "erf_13", "add_10896", "mul_6888"], "dropout_35": ["mul_6888"], "choose_qparams_affine_default_71": ["mul_6888", "amax_71", "full_143", "maximum_71", "amin_71", "full_142", "minimum_71", "sub_3259", "div_142", "clamp_min_213", "div_143", "round_143", "sub_3265", "clamp_min_214", "clamp_max_142", "convert_element_type_426"], "quantize_affine_71": ["mul_6888", "view_1115", "reciprocal_71", "mul_6934", "mul_6937", "round_144", "view_1116", "add_10979", "clamp_min_215", "clamp_max_143", "convert_element_type_427"], "dequantize_affine_142": ["convert_element_type_427", "view_1120", "convert_element_type_428", "convert_element_type_429", "sub_3285", "view_1119", "mul_6959"], "dequantize_affine_143": ["view_1122", "view_1124", "convert_element_type_430", "convert_element_type_431", "sub_3289", "view_1123", "mul_6964", "view_1125"], "linear_71": ["mul_6969", "view_1126", "permute_120", "mm_default_100", "add_tensor_100", "view_1127"], "dropout_36": ["view_1127"], "add_168": ["add_11042"], "layer_norm_24": ["clone_97", "var_mean_24", "getitem_96", "getitem_97", "sub_3295", "add_11047", "rsqrt_24", "mul_6980", "mul_6981", "add_11048"], "choose_qparams_affine_default_72": ["add_11048", "amax_72", "full_145", "maximum_72", "amin_72", "full_144", "minimum_72", "sub_3306", "div_144", "clamp_min_216", "div_145", "round_145", "sub_3312", "clamp_min_217", "clamp_max_144", "convert_element_type_432"], "quantize_affine_72": ["add_11048", "view_1130", "reciprocal_72", "mul_7029", "mul_7032", "round_146", "view_1131", "add_11135", "clamp_min_218", "clamp_max_145", "convert_element_type_433"], "choose_qparams_affine_default_73": ["add_11048", "amax_73", "full_147", "maximum_73", "amin_73", "full_146", "minimum_73", "sub_3351", "div_146", "clamp_min_219", "div_147", "round_147", "sub_3357", "clamp_min_220", "clamp_max_146", "convert_element_type_438"], "quantize_affine_73": ["add_11048", "view_1146", "reciprocal_73", "mul_7125", "mul_7128", "round_148", "view_1147", "add_11287", "clamp_min_221", "clamp_max_147", "convert_element_type_439"], "choose_qparams_affine_default_74": ["add_11048", "amax_74", "full_149", "maximum_74", "amin_74", "full_148", "minimum_74", "sub_3395", "div_148", "clamp_min_222", "div_149", "round_149", "sub_3401", "clamp_min_223", "clamp_max_148", "convert_element_type_444"], "quantize_affine_74": ["add_11048", "view_1162", "reciprocal_74", "mul_7224", "mul_7227", "round_150", "view_1163", "add_11435", "clamp_min_224", "clamp_max_149", "convert_element_type_445"], "dequantize_affine_144": ["convert_element_type_433", "view_1135", "convert_element_type_434", "convert_element_type_435", "sub_3332", "view_1134", "mul_7054"], "dequantize_affine_145": ["view_1137", "view_1139", "convert_element_type_436", "convert_element_type_437", "sub_3336", "view_1138", "mul_7059", "view_1140"], "dequantize_affine_146": ["convert_element_type_439", "view_1151", "convert_element_type_440", "convert_element_type_441", "sub_3377", "view_1150", "mul_7150"], "dequantize_affine_147": ["view_1153", "view_1155", "convert_element_type_442", "convert_element_type_443", "sub_3381", "view_1154", "mul_7155", "view_1156"], "dequantize_affine_148": ["convert_element_type_445", "view_1167", "convert_element_type_446", "convert_element_type_447", "sub_3421", "view_1166", "mul_7249"], "dequantize_affine_149": ["view_1169", "view_1171", "convert_element_type_448", "convert_element_type_449", "sub_3425", "view_1170", "mul_7254", "view_1172"], "linear_72": ["mul_7064", "view_1141", "permute_121", "mm_default_99", "add_tensor_99", "view_1142"], "mul_449": ["mul_7071"], "view_36": ["view_1143"], "transpose_48": ["permute_122"], "contiguous_48": ["clone_98"], "linear_73": ["mul_7158", "view_1157", "permute_123", "mm_12", "view_1158"], "view_37": ["view_1159"], "transpose_49": ["permute_124"], "contiguous_49": ["clone_99"], "linear_74": ["mul_7259", "view_1173", "permute_125", "mm_default_98", "add_tensor_98", "view_1174"], "view_38": ["view_1175"], "transpose_50": ["permute_126"], "contiguous_50": ["clone_100"], "scaled_dot_product_attention_12": ["_scaled_dot_product_efficient_attention_12", "getitem_98"], "transpose_51": ["permute_127"], "reshape_12": ["view_1176"], "choose_qparams_affine_default_75": ["view_1176", "amax_75", "full_151", "maximum_75", "amin_75", "full_150", "minimum_75", "sub_3443", "div_150", "clamp_min_225", "div_151", "round_151", "sub_3449", "clamp_min_226", "clamp_max_150", "convert_element_type_450"], "quantize_affine_75": ["view_1176", "view_1179", "reciprocal_75", "mul_7329", "mul_7332", "round_152", "view_1180", "add_11599", "clamp_min_227", "clamp_max_151", "convert_element_type_451"], "dequantize_affine_150": ["convert_element_type_451", "view_1184", "convert_element_type_452", "convert_element_type_453", "sub_3469", "view_1183", "mul_7354"], "dequantize_affine_151": ["view_1186", "view_1188", "convert_element_type_454", "convert_element_type_455", "sub_3473", "view_1187", "mul_7359", "view_1189"], "linear_75": ["mul_7364", "view_1190", "permute_128", "mm_default_97", "add_tensor_97", "view_1191"], "dropout_37": ["view_1191"], "add_177": ["add_11662"], "layer_norm_25": ["clone_102", "var_mean_25", "getitem_102", "getitem_103", "sub_3479", "add_11667", "rsqrt_25", "mul_7375", "mul_7376", "add_11668"], "choose_qparams_affine_default_76": ["add_11668", "amax_76", "full_153", "maximum_76", "amin_76", "full_152", "minimum_76", "sub_3490", "div_152", "clamp_min_228", "div_153", "round_153", "sub_3496", "clamp_min_229", "clamp_max_152", "convert_element_type_456"], "quantize_affine_76": ["add_11668", "view_1194", "reciprocal_76", "mul_7424", "mul_7427", "round_154", "view_1195", "add_11755", "clamp_min_230", "clamp_max_153", "convert_element_type_457"], "dequantize_affine_152": ["convert_element_type_457", "view_1199", "convert_element_type_458", "convert_element_type_459", "sub_3516", "view_1198", "mul_7449"], "dequantize_affine_153": ["view_1201", "view_1203", "convert_element_type_460", "convert_element_type_461", "sub_3520", "view_1202", "mul_7454", "view_1204"], "linear_76": ["mul_7459", "view_1205", "permute_129", "mm_default_96", "add_tensor_96", "view_1206"], "gelu_14": ["mul_7466", "mul_7467", "erf_14", "add_11814", "mul_7468"], "dropout_38": ["mul_7468"], "choose_qparams_affine_default_77": ["mul_7468", "amax_77", "full_155", "maximum_77", "amin_77", "full_154", "minimum_77", "sub_3533", "div_154", "clamp_min_231", "div_155", "round_155", "sub_3539", "clamp_min_232", "clamp_max_154", "convert_element_type_462"], "quantize_affine_77": ["mul_7468", "view_1209", "reciprocal_77", "mul_7514", "mul_7517", "round_156", "view_1210", "add_11897", "clamp_min_233", "clamp_max_155", "convert_element_type_463"], "dequantize_affine_154": ["convert_element_type_463", "view_1214", "convert_element_type_464", "convert_element_type_465", "sub_3559", "view_1213", "mul_7539"], "dequantize_affine_155": ["view_1216", "view_1218", "convert_element_type_466", "convert_element_type_467", "sub_3563", "view_1217", "mul_7544", "view_1219"], "linear_77": ["mul_7549", "view_1220", "permute_130", "mm_default_95", "add_tensor_95", "view_1221"], "dropout_39": ["view_1221"], "add_182": ["add_11960"], "layer_norm_26": ["clone_105", "var_mean_26", "getitem_104", "getitem_105", "sub_3569", "add_11965", "rsqrt_26", "mul_7560", "mul_7561", "add_11966"], "choose_qparams_affine_default_78": ["add_11966", "amax_78", "full_157", "maximum_78", "amin_78", "full_156", "minimum_78", "sub_3580", "div_156", "clamp_min_234", "div_157", "round_157", "sub_3586", "clamp_min_235", "clamp_max_156", "convert_element_type_468"], "quantize_affine_78": ["add_11966", "view_1224", "reciprocal_78", "mul_7609", "mul_7612", "round_158", "view_1225", "add_12053", "clamp_min_236", "clamp_max_157", "convert_element_type_469"], "choose_qparams_affine_default_79": ["add_11966", "amax_79", "full_159", "maximum_79", "amin_79", "full_158", "minimum_79", "sub_3625", "div_158", "clamp_min_237", "div_159", "round_159", "sub_3631", "clamp_min_238", "clamp_max_158", "convert_element_type_474"], "quantize_affine_79": ["add_11966", "view_1240", "reciprocal_79", "mul_7705", "mul_7708", "round_160", "view_1241", "add_12205", "clamp_min_239", "clamp_max_159", "convert_element_type_475"], "choose_qparams_affine_default_80": ["add_11966", "amax_80", "full_161", "maximum_80", "amin_80", "full_160", "minimum_80", "sub_3669", "div_160", "clamp_min_240", "div_161", "round_161", "sub_3675", "clamp_min_241", "clamp_max_160", "convert_element_type_480"], "quantize_affine_80": ["add_11966", "view_1256", "reciprocal_80", "mul_7804", "mul_7807", "round_162", "view_1257", "add_12353", "clamp_min_242", "clamp_max_161", "convert_element_type_481"], "dequantize_affine_156": ["convert_element_type_469", "view_1229", "convert_element_type_470", "convert_element_type_471", "sub_3606", "view_1228", "mul_7634"], "dequantize_affine_157": ["view_1231", "view_1233", "convert_element_type_472", "convert_element_type_473", "sub_3610", "view_1232", "mul_7639", "view_1234"], "dequantize_affine_158": ["convert_element_type_475", "view_1245", "convert_element_type_476", "convert_element_type_477", "sub_3651", "view_1244", "mul_7730"], "dequantize_affine_159": ["view_1247", "view_1249", "convert_element_type_478", "convert_element_type_479", "sub_3655", "view_1248", "mul_7735", "view_1250"], "dequantize_affine_160": ["convert_element_type_481", "view_1261", "convert_element_type_482", "convert_element_type_483", "sub_3695", "view_1260", "mul_7829"], "dequantize_affine_161": ["view_1263", "view_1265", "convert_element_type_484", "convert_element_type_485", "sub_3699", "view_1264", "mul_7834", "view_1266"], "linear_78": ["mul_7644", "view_1235", "permute_131", "mm_default_94", "add_tensor_94", "view_1236"], "mul_486": ["mul_7651"], "view_39": ["view_1237"], "transpose_52": ["permute_132"], "contiguous_52": ["clone_106"], "linear_79": ["mul_7738", "view_1251", "permute_133", "mm_13", "view_1252"], "view_40": ["view_1253"], "transpose_53": ["permute_134"], "contiguous_53": ["clone_107"], "linear_80": ["mul_7839", "view_1267", "permute_135", "mm_default_93", "add_tensor_93", "view_1268"], "view_41": ["view_1269"], "transpose_54": ["permute_136"], "contiguous_54": ["clone_108"], "scaled_dot_product_attention_13": ["_scaled_dot_product_efficient_attention_13", "getitem_106"], "transpose_55": ["permute_137"], "reshape_13": ["view_1270"], "choose_qparams_affine_default_81": ["view_1270", "amax_81", "full_163", "maximum_81", "amin_81", "full_162", "minimum_81", "sub_3717", "div_162", "clamp_min_243", "div_163", "round_163", "sub_3723", "clamp_min_244", "clamp_max_162", "convert_element_type_486"], "quantize_affine_81": ["view_1270", "view_1273", "reciprocal_81", "mul_7909", "mul_7912", "round_164", "view_1274", "add_12517", "clamp_min_245", "clamp_max_163", "convert_element_type_487"], "dequantize_affine_162": ["convert_element_type_487", "view_1278", "convert_element_type_488", "convert_element_type_489", "sub_3743", "view_1277", "mul_7934"], "dequantize_affine_163": ["view_1280", "view_1282", "convert_element_type_490", "convert_element_type_491", "sub_3747", "view_1281", "mul_7939", "view_1283"], "linear_81": ["mul_7944", "view_1284", "permute_138", "mm_default_92", "add_tensor_92", "view_1285"], "dropout_40": ["view_1285"], "add_191": ["add_12580"], "layer_norm_27": ["clone_110", "var_mean_27", "getitem_110", "getitem_111", "sub_3753", "add_12585", "rsqrt_27", "mul_7955", "mul_7956", "add_12586"], "choose_qparams_affine_default_82": ["add_12586", "amax_82", "full_165", "maximum_82", "amin_82", "full_164", "minimum_82", "sub_3764", "div_164", "clamp_min_246", "div_165", "round_165", "sub_3770", "clamp_min_247", "clamp_max_164", "convert_element_type_492"], "quantize_affine_82": ["add_12586", "view_1288", "reciprocal_82", "mul_8004", "mul_8007", "round_166", "view_1289", "add_12673", "clamp_min_248", "clamp_max_165", "convert_element_type_493"], "dequantize_affine_164": ["convert_element_type_493", "view_1293", "convert_element_type_494", "convert_element_type_495", "sub_3790", "view_1292", "mul_8029"], "dequantize_affine_165": ["view_1295", "view_1297", "convert_element_type_496", "convert_element_type_497", "sub_3794", "view_1296", "mul_8034", "view_1298"], "linear_82": ["mul_8039", "view_1299", "permute_139", "mm_default_91", "add_tensor_91", "view_1300"], "gelu_15": ["mul_8046", "mul_8047", "erf_15", "add_12732", "mul_8048"], "dropout_41": ["mul_8048"], "choose_qparams_affine_default_83": ["mul_8048", "amax_83", "full_167", "maximum_83", "amin_83", "full_166", "minimum_83", "sub_3807", "div_166", "clamp_min_249", "div_167", "round_167", "sub_3813", "clamp_min_250", "clamp_max_166", "convert_element_type_498"], "quantize_affine_83": ["mul_8048", "view_1303", "reciprocal_83", "mul_8094", "mul_8097", "round_168", "view_1304", "add_12815", "clamp_min_251", "clamp_max_167", "convert_element_type_499"], "dequantize_affine_166": ["convert_element_type_499", "view_1308", "convert_element_type_500", "convert_element_type_501", "sub_3833", "view_1307", "mul_8119"], "dequantize_affine_167": ["view_1310", "view_1312", "convert_element_type_502", "convert_element_type_503", "sub_3837", "view_1311", "mul_8124", "view_1313"], "linear_83": ["mul_8129", "view_1314", "permute_140", "mm_default_90", "add_tensor_90", "view_1315"], "dropout_42": ["view_1315"], "add_196": ["add_12878"], "layer_norm_28": ["clone_113", "var_mean_28", "getitem_112", "getitem_113", "sub_3843", "add_12883", "rsqrt_28", "mul_8140", "mul_8141", "add_12884"], "choose_qparams_affine_default_84": ["add_12884", "amax_84", "full_169", "maximum_84", "amin_84", "full_168", "minimum_84", "sub_3854", "div_168", "clamp_min_252", "div_169", "round_169", "sub_3860", "clamp_min_253", "clamp_max_168", "convert_element_type_504"], "quantize_affine_84": ["add_12884", "view_1318", "reciprocal_84", "mul_8189", "mul_8192", "round_170", "view_1319", "add_12971", "clamp_min_254", "clamp_max_169", "convert_element_type_505"], "choose_qparams_affine_default_85": ["add_12884", "amax_85", "full_171", "maximum_85", "amin_85", "full_170", "minimum_85", "sub_3899", "div_170", "clamp_min_255", "div_171", "round_171", "sub_3905", "clamp_min_256", "clamp_max_170", "convert_element_type_510"], "quantize_affine_85": ["add_12884", "view_1334", "reciprocal_85", "mul_8285", "mul_8288", "round_172", "view_1335", "add_13123", "clamp_min_257", "clamp_max_171", "convert_element_type_511"], "choose_qparams_affine_default_86": ["add_12884", "amax_86", "full_173", "maximum_86", "amin_86", "full_172", "minimum_86", "sub_3943", "div_172", "clamp_min_258", "div_173", "round_173", "sub_3949", "clamp_min_259", "clamp_max_172", "convert_element_type_516"], "quantize_affine_86": ["add_12884", "view_1350", "reciprocal_86", "mul_8384", "mul_8387", "round_174", "view_1351", "add_13271", "clamp_min_260", "clamp_max_173", "convert_element_type_517"], "dequantize_affine_168": ["convert_element_type_505", "view_1323", "convert_element_type_506", "convert_element_type_507", "sub_3880", "view_1322", "mul_8214"], "dequantize_affine_169": ["view_1325", "view_1327", "convert_element_type_508", "convert_element_type_509", "sub_3884", "view_1326", "mul_8219", "view_1328"], "dequantize_affine_170": ["convert_element_type_511", "view_1339", "convert_element_type_512", "convert_element_type_513", "sub_3925", "view_1338", "mul_8310"], "dequantize_affine_171": ["view_1341", "view_1343", "convert_element_type_514", "convert_element_type_515", "sub_3929", "view_1342", "mul_8315", "view_1344"], "dequantize_affine_172": ["convert_element_type_517", "view_1355", "convert_element_type_518", "convert_element_type_519", "sub_3969", "view_1354", "mul_8409"], "dequantize_affine_173": ["view_1357", "view_1359", "convert_element_type_520", "convert_element_type_521", "sub_3973", "view_1358", "mul_8414", "view_1360"], "linear_84": ["mul_8224", "view_1329", "permute_141", "mm_default_89", "add_tensor_89", "view_1330"], "mul_523": ["mul_8231"], "view_42": ["view_1331"], "transpose_56": ["permute_142"], "contiguous_56": ["clone_114"], "linear_85": ["mul_8318", "view_1345", "permute_143", "mm_14", "view_1346"], "view_43": ["view_1347"], "transpose_57": ["permute_144"], "contiguous_57": ["clone_115"], "linear_86": ["mul_8419", "view_1361", "permute_145", "mm_default_88", "add_tensor_88", "view_1362"], "view_44": ["view_1363"], "transpose_58": ["permute_146"], "contiguous_58": ["clone_116"], "scaled_dot_product_attention_14": ["_scaled_dot_product_efficient_attention_14", "getitem_114"], "transpose_59": ["permute_147"], "reshape_14": ["view_1364"], "choose_qparams_affine_default_87": ["view_1364", "amax_87", "full_175", "maximum_87", "amin_87", "full_174", "minimum_87", "sub_3991", "div_174", "clamp_min_261", "div_175", "round_175", "sub_3997", "clamp_min_262", "clamp_max_174", "convert_element_type_522"], "quantize_affine_87": ["view_1364", "view_1367", "reciprocal_87", "mul_8489", "mul_8492", "round_176", "view_1368", "add_13435", "clamp_min_263", "clamp_max_175", "convert_element_type_523"], "dequantize_affine_174": ["convert_element_type_523", "view_1372", "convert_element_type_524", "convert_element_type_525", "sub_4017", "view_1371", "mul_8514"], "dequantize_affine_175": ["view_1374", "view_1376", "convert_element_type_526", "convert_element_type_527", "sub_4021", "view_1375", "mul_8519", "view_1377"], "linear_87": ["mul_8524", "view_1378", "permute_148", "mm_default_87", "add_tensor_87", "view_1379"], "dropout_43": ["view_1379"], "add_205": ["add_13498"], "layer_norm_29": ["clone_118", "var_mean_29", "getitem_118", "getitem_119", "sub_4027", "add_13503", "rsqrt_29", "mul_8535", "mul_8536", "add_13504"], "choose_qparams_affine_default_88": ["add_13504", "amax_88", "full_177", "maximum_88", "amin_88", "full_176", "minimum_88", "sub_4038", "div_176", "clamp_min_264", "div_177", "round_177", "sub_4044", "clamp_min_265", "clamp_max_176", "convert_element_type_528"], "quantize_affine_88": ["add_13504", "view_1382", "reciprocal_88", "mul_8584", "mul_8587", "round_178", "view_1383", "add_13591", "clamp_min_266", "clamp_max_177", "convert_element_type_529"], "dequantize_affine_176": ["convert_element_type_529", "view_1387", "convert_element_type_530", "convert_element_type_531", "sub_4064", "view_1386", "mul_8609"], "dequantize_affine_177": ["view_1389", "view_1391", "convert_element_type_532", "convert_element_type_533", "sub_4068", "view_1390", "mul_8614", "view_1392"], "linear_88": ["mul_8619", "view_1393", "permute_149", "mm_default_86", "add_tensor_86", "view_1394"], "gelu_16": ["mul_8626", "mul_8627", "erf_16", "add_13650", "mul_8628"], "dropout_44": ["mul_8628"], "choose_qparams_affine_default_89": ["mul_8628", "amax_89", "full_179", "maximum_89", "amin_89", "full_178", "minimum_89", "sub_4081", "div_178", "clamp_min_267", "div_179", "round_179", "sub_4087", "clamp_min_268", "clamp_max_178", "convert_element_type_534"], "quantize_affine_89": ["mul_8628", "view_1397", "reciprocal_89", "mul_8674", "mul_8677", "round_180", "view_1398", "add_13733", "clamp_min_269", "clamp_max_179", "convert_element_type_535"], "dequantize_affine_178": ["convert_element_type_535", "view_1402", "convert_element_type_536", "convert_element_type_537", "sub_4107", "view_1401", "mul_8699"], "dequantize_affine_179": ["view_1404", "view_1406", "convert_element_type_538", "convert_element_type_539", "sub_4111", "view_1405", "mul_8704", "view_1407"], "linear_89": ["mul_8709", "view_1408", "permute_150", "mm_default_85", "add_tensor_85", "view_1409"], "dropout_45": ["view_1409"], "add_210": ["add_13796"], "layer_norm_30": ["clone_121", "var_mean_30", "getitem_120", "getitem_121", "sub_4117", "add_13801", "rsqrt_30", "mul_8720", "mul_8721", "add_13802"], "choose_qparams_affine_default_90": ["add_13802", "amax_90", "full_181", "maximum_90", "amin_90", "full_180", "minimum_90", "sub_4128", "div_180", "clamp_min_270", "div_181", "round_181", "sub_4134", "clamp_min_271", "clamp_max_180", "convert_element_type_540"], "quantize_affine_90": ["add_13802", "view_1412", "reciprocal_90", "mul_8769", "mul_8772", "round_182", "view_1413", "add_13889", "clamp_min_272", "clamp_max_181", "convert_element_type_541"], "choose_qparams_affine_default_91": ["add_13802", "amax_91", "full_183", "maximum_91", "amin_91", "full_182", "minimum_91", "sub_4173", "div_182", "clamp_min_273", "div_183", "round_183", "sub_4179", "clamp_min_274", "clamp_max_182", "convert_element_type_546"], "quantize_affine_91": ["add_13802", "view_1428", "reciprocal_91", "mul_8865", "mul_8868", "round_184", "view_1429", "add_14041", "clamp_min_275", "clamp_max_183", "convert_element_type_547"], "choose_qparams_affine_default_92": ["add_13802", "amax_92", "full_185", "maximum_92", "amin_92", "full_184", "minimum_92", "sub_4217", "div_184", "clamp_min_276", "div_185", "round_185", "sub_4223", "clamp_min_277", "clamp_max_184", "convert_element_type_552"], "quantize_affine_92": ["add_13802", "view_1444", "reciprocal_92", "mul_8964", "mul_8967", "round_186", "view_1445", "add_14189", "clamp_min_278", "clamp_max_185", "convert_element_type_553"], "dequantize_affine_180": ["convert_element_type_541", "view_1417", "convert_element_type_542", "convert_element_type_543", "sub_4154", "view_1416", "mul_8794"], "dequantize_affine_181": ["view_1419", "view_1421", "convert_element_type_544", "convert_element_type_545", "sub_4158", "view_1420", "mul_8799", "view_1422"], "dequantize_affine_182": ["convert_element_type_547", "view_1433", "convert_element_type_548", "convert_element_type_549", "sub_4199", "view_1432", "mul_8890"], "dequantize_affine_183": ["view_1435", "view_1437", "convert_element_type_550", "convert_element_type_551", "sub_4203", "view_1436", "mul_8895", "view_1438"], "dequantize_affine_184": ["convert_element_type_553", "view_1449", "convert_element_type_554", "convert_element_type_555", "sub_4243", "view_1448", "mul_8989"], "dequantize_affine_185": ["view_1451", "view_1453", "convert_element_type_556", "convert_element_type_557", "sub_4247", "view_1452", "mul_8994", "view_1454"], "linear_90": ["mul_8804", "view_1423", "permute_151", "mm_default_84", "add_tensor_84", "view_1424"], "mul_560": ["mul_8811"], "view_45": ["view_1425"], "transpose_60": ["permute_152"], "contiguous_60": ["clone_122"], "linear_91": ["mul_8898", "view_1439", "permute_153", "mm_15", "view_1440"], "view_46": ["view_1441"], "transpose_61": ["permute_154"], "contiguous_61": ["clone_123"], "linear_92": ["mul_8999", "view_1455", "permute_155", "mm_default_83", "add_tensor_83", "view_1456"], "view_47": ["view_1457"], "transpose_62": ["permute_156"], "contiguous_62": ["clone_124"], "scaled_dot_product_attention_15": ["_scaled_dot_product_efficient_attention_15", "getitem_122"], "transpose_63": ["permute_157"], "reshape_15": ["view_1458"], "choose_qparams_affine_default_93": ["view_1458", "amax_93", "full_187", "maximum_93", "amin_93", "full_186", "minimum_93", "sub_4265", "div_186", "clamp_min_279", "div_187", "round_187", "sub_4271", "clamp_min_280", "clamp_max_186", "convert_element_type_558"], "quantize_affine_93": ["view_1458", "view_1461", "reciprocal_93", "mul_9069", "mul_9072", "round_188", "view_1462", "add_14353", "clamp_min_281", "clamp_max_187", "convert_element_type_559"], "dequantize_affine_186": ["convert_element_type_559", "view_1466", "convert_element_type_560", "convert_element_type_561", "sub_4291", "view_1465", "mul_9094"], "dequantize_affine_187": ["view_1468", "view_1470", "convert_element_type_562", "convert_element_type_563", "sub_4295", "view_1469", "mul_9099", "view_1471"], "linear_93": ["mul_9104", "view_1472", "permute_158", "mm_default_82", "add_tensor_82", "view_1473"], "dropout_46": ["view_1473"], "add_219": ["add_14416"], "layer_norm_31": ["clone_126", "var_mean_31", "getitem_126", "getitem_127", "sub_4301", "add_14421", "rsqrt_31", "mul_9115", "mul_9116", "add_14422"], "choose_qparams_affine_default_94": ["add_14422", "amax_94", "full_189", "maximum_94", "amin_94", "full_188", "minimum_94", "sub_4312", "div_188", "clamp_min_282", "div_189", "round_189", "sub_4318", "clamp_min_283", "clamp_max_188", "convert_element_type_564"], "quantize_affine_94": ["add_14422", "view_1476", "reciprocal_94", "mul_9164", "mul_9167", "round_190", "view_1477", "add_14509", "clamp_min_284", "clamp_max_189", "convert_element_type_565"], "dequantize_affine_188": ["convert_element_type_565", "view_1481", "convert_element_type_566", "convert_element_type_567", "sub_4338", "view_1480", "mul_9189"], "dequantize_affine_189": ["view_1483", "view_1485", "convert_element_type_568", "convert_element_type_569", "sub_4342", "view_1484", "mul_9194", "view_1486"], "linear_94": ["mul_9199", "view_1487", "permute_159", "mm_default_81", "add_tensor_81", "view_1488"], "gelu_17": ["mul_9206", "mul_9207", "erf_17", "add_14568", "mul_9208"], "dropout_47": ["mul_9208"], "choose_qparams_affine_default_95": ["mul_9208", "amax_95", "full_191", "maximum_95", "amin_95", "full_190", "minimum_95", "sub_4355", "div_190", "clamp_min_285", "div_191", "round_191", "sub_4361", "clamp_min_286", "clamp_max_190", "convert_element_type_570"], "quantize_affine_95": ["mul_9208", "view_1491", "reciprocal_95", "mul_9254", "mul_9257", "round_192", "view_1492", "add_14651", "clamp_min_287", "clamp_max_191", "convert_element_type_571"], "dequantize_affine_190": ["convert_element_type_571", "view_1496", "convert_element_type_572", "convert_element_type_573", "sub_4381", "view_1495", "mul_9279"], "dequantize_affine_191": ["view_1498", "view_1500", "convert_element_type_574", "convert_element_type_575", "sub_4385", "view_1499", "mul_9284", "view_1501"], "linear_95": ["mul_9289", "view_1502", "permute_160", "mm_default_80", "add_tensor_80", "view_1503"], "dropout_48": ["view_1503"], "add_224": ["add_14714"], "layer_norm_32": ["clone_129", "var_mean_32", "getitem_128", "getitem_129", "sub_4391", "add_14719", "rsqrt_32", "mul_9300", "mul_9301", "add_14720"], "choose_qparams_affine_default_96": ["add_14720", "amax_96", "full_193", "maximum_96", "amin_96", "full_192", "minimum_96", "sub_4402", "div_192", "clamp_min_288", "div_193", "round_193", "sub_4408", "clamp_min_289", "clamp_max_192", "convert_element_type_576"], "quantize_affine_96": ["add_14720", "view_1506", "reciprocal_96", "mul_9349", "mul_9352", "round_194", "view_1507", "add_14807", "clamp_min_290", "clamp_max_193", "convert_element_type_577"], "choose_qparams_affine_default_97": ["add_14720", "amax_97", "full_195", "maximum_97", "amin_97", "full_194", "minimum_97", "sub_4447", "div_194", "clamp_min_291", "div_195", "round_195", "sub_4453", "clamp_min_292", "clamp_max_194", "convert_element_type_582"], "quantize_affine_97": ["add_14720", "view_1522", "reciprocal_97", "mul_9445", "mul_9448", "round_196", "view_1523", "add_14959", "clamp_min_293", "clamp_max_195", "convert_element_type_583"], "choose_qparams_affine_default_98": ["add_14720", "amax_98", "full_197", "maximum_98", "amin_98", "full_196", "minimum_98", "sub_4491", "div_196", "clamp_min_294", "div_197", "round_197", "sub_4497", "clamp_min_295", "clamp_max_196", "convert_element_type_588"], "quantize_affine_98": ["add_14720", "view_1538", "reciprocal_98", "mul_9544", "mul_9547", "round_198", "view_1539", "add_15107", "clamp_min_296", "clamp_max_197", "convert_element_type_589"], "dequantize_affine_192": ["convert_element_type_577", "view_1511", "convert_element_type_578", "convert_element_type_579", "sub_4428", "view_1510", "mul_9374"], "dequantize_affine_193": ["view_1513", "view_1515", "convert_element_type_580", "convert_element_type_581", "sub_4432", "view_1514", "mul_9379", "view_1516"], "dequantize_affine_194": ["convert_element_type_583", "view_1527", "convert_element_type_584", "convert_element_type_585", "sub_4473", "view_1526", "mul_9470"], "dequantize_affine_195": ["view_1529", "view_1531", "convert_element_type_586", "convert_element_type_587", "sub_4477", "view_1530", "mul_9475", "view_1532"], "dequantize_affine_196": ["convert_element_type_589", "view_1543", "convert_element_type_590", "convert_element_type_591", "sub_4517", "view_1542", "mul_9569"], "dequantize_affine_197": ["view_1545", "view_1547", "convert_element_type_592", "convert_element_type_593", "sub_4521", "view_1546", "mul_9574", "view_1548"], "linear_96": ["mul_9384", "view_1517", "permute_161", "mm_default_79", "add_tensor_79", "view_1518"], "mul_597": ["mul_9391"], "view_48": ["view_1519"], "transpose_64": ["permute_162"], "contiguous_64": ["clone_130"], "linear_97": ["mul_9478", "view_1533", "permute_163", "mm_16", "view_1534"], "view_49": ["view_1535"], "transpose_65": ["permute_164"], "contiguous_65": ["clone_131"], "linear_98": ["mul_9579", "view_1549", "permute_165", "mm_default_78", "add_tensor_78", "view_1550"], "view_50": ["view_1551"], "transpose_66": ["permute_166"], "contiguous_66": ["clone_132"], "scaled_dot_product_attention_16": ["_scaled_dot_product_efficient_attention_16", "getitem_130"], "transpose_67": ["permute_167"], "reshape_16": ["view_1552"], "choose_qparams_affine_default_99": ["view_1552", "amax_99", "full_199", "maximum_99", "amin_99", "full_198", "minimum_99", "sub_4539", "div_198", "clamp_min_297", "div_199", "round_199", "sub_4545", "clamp_min_298", "clamp_max_198", "convert_element_type_594"], "quantize_affine_99": ["view_1552", "view_1555", "reciprocal_99", "mul_9649", "mul_9652", "round_200", "view_1556", "add_15271", "clamp_min_299", "clamp_max_199", "convert_element_type_595"], "dequantize_affine_198": ["convert_element_type_595", "view_1560", "convert_element_type_596", "convert_element_type_597", "sub_4565", "view_1559", "mul_9674"], "dequantize_affine_199": ["view_1562", "view_1564", "convert_element_type_598", "convert_element_type_599", "sub_4569", "view_1563", "mul_9679", "view_1565"], "linear_99": ["mul_9684", "view_1566", "permute_168", "mm_default_77", "add_tensor_77", "view_1567"], "dropout_49": ["view_1567"], "add_233": ["add_15334"], "layer_norm_33": ["clone_134", "var_mean_33", "getitem_134", "getitem_135", "sub_4575", "add_15339", "rsqrt_33", "mul_9695", "mul_9696", "add_15340"], "choose_qparams_affine_default_100": ["add_15340", "amax_100", "full_201", "maximum_100", "amin_100", "full_200", "minimum_100", "sub_4586", "div_200", "clamp_min_300", "div_201", "round_201", "sub_4592", "clamp_min_301", "clamp_max_200", "convert_element_type_600"], "quantize_affine_100": ["add_15340", "view_1570", "reciprocal_100", "mul_9744", "mul_9747", "round_202", "view_1571", "add_15427", "clamp_min_302", "clamp_max_201", "convert_element_type_601"], "dequantize_affine_200": ["convert_element_type_601", "view_1575", "convert_element_type_602", "convert_element_type_603", "sub_4612", "view_1574", "mul_9769"], "dequantize_affine_201": ["view_1577", "view_1579", "convert_element_type_604", "convert_element_type_605", "sub_4616", "view_1578", "mul_9774", "view_1580"], "linear_100": ["mul_9779", "view_1581", "permute_169", "mm_default_76", "add_tensor_76", "view_1582"], "gelu_18": ["mul_9786", "mul_9787", "erf_18", "add_15486", "mul_9788"], "dropout_50": ["mul_9788"], "choose_qparams_affine_default_101": ["mul_9788", "amax_101", "full_203", "maximum_101", "amin_101", "full_202", "minimum_101", "sub_4629", "div_202", "clamp_min_303", "div_203", "round_203", "sub_4635", "clamp_min_304", "clamp_max_202", "convert_element_type_606"], "quantize_affine_101": ["mul_9788", "view_1585", "reciprocal_101", "mul_9834", "mul_9837", "round_204", "view_1586", "add_15569", "clamp_min_305", "clamp_max_203", "convert_element_type_607"], "dequantize_affine_202": ["convert_element_type_607", "view_1590", "convert_element_type_608", "convert_element_type_609", "sub_4655", "view_1589", "mul_9859"], "dequantize_affine_203": ["view_1592", "view_1594", "convert_element_type_610", "convert_element_type_611", "sub_4659", "view_1593", "mul_9864", "view_1595"], "linear_101": ["mul_9869", "view_1596", "permute_170", "mm_default_75", "add_tensor_75", "view_1597"], "dropout_51": ["view_1597"], "add_238": ["add_15632"], "layer_norm_34": ["clone_137", "var_mean_34", "getitem_136", "getitem_137", "sub_4665", "add_15637", "rsqrt_34", "mul_9880", "mul_9881", "add_15638"], "choose_qparams_affine_default_102": ["add_15638", "amax_102", "full_205", "maximum_102", "amin_102", "full_204", "minimum_102", "sub_4676", "div_204", "clamp_min_306", "div_205", "round_205", "sub_4682", "clamp_min_307", "clamp_max_204", "convert_element_type_612"], "quantize_affine_102": ["add_15638", "view_1600", "reciprocal_102", "mul_9929", "mul_9932", "round_206", "view_1601", "add_15725", "clamp_min_308", "clamp_max_205", "convert_element_type_613"], "choose_qparams_affine_default_103": ["add_15638", "amax_103", "full_207", "maximum_103", "amin_103", "full_206", "minimum_103", "sub_4721", "div_206", "clamp_min_309", "div_207", "round_207", "sub_4727", "clamp_min_310", "clamp_max_206", "convert_element_type_618"], "quantize_affine_103": ["add_15638", "view_1616", "reciprocal_103", "mul_10025", "mul_10028", "round_208", "view_1617", "add_15877", "clamp_min_311", "clamp_max_207", "convert_element_type_619"], "choose_qparams_affine_default_104": ["add_15638", "amax_104", "full_209", "maximum_104", "amin_104", "full_208", "minimum_104", "sub_4765", "div_208", "clamp_min_312", "div_209", "round_209", "sub_4771", "clamp_min_313", "clamp_max_208", "convert_element_type_624"], "quantize_affine_104": ["add_15638", "view_1632", "reciprocal_104", "mul_10124", "mul_10127", "round_210", "view_1633", "add_16025", "clamp_min_314", "clamp_max_209", "convert_element_type_625"], "dequantize_affine_204": ["convert_element_type_613", "view_1605", "convert_element_type_614", "convert_element_type_615", "sub_4702", "view_1604", "mul_9954"], "dequantize_affine_205": ["view_1607", "view_1609", "convert_element_type_616", "convert_element_type_617", "sub_4706", "view_1608", "mul_9959", "view_1610"], "dequantize_affine_206": ["convert_element_type_619", "view_1621", "convert_element_type_620", "convert_element_type_621", "sub_4747", "view_1620", "mul_10050"], "dequantize_affine_207": ["view_1623", "view_1625", "convert_element_type_622", "convert_element_type_623", "sub_4751", "view_1624", "mul_10055", "view_1626"], "dequantize_affine_208": ["convert_element_type_625", "view_1637", "convert_element_type_626", "convert_element_type_627", "sub_4791", "view_1636", "mul_10149"], "dequantize_affine_209": ["view_1639", "view_1641", "convert_element_type_628", "convert_element_type_629", "sub_4795", "view_1640", "mul_10154", "view_1642"], "linear_102": ["mul_9964", "view_1611", "permute_171", "mm_default_74", "add_tensor_74", "view_1612"], "mul_634": ["mul_9971"], "view_51": ["view_1613"], "transpose_68": ["permute_172"], "contiguous_68": ["clone_138"], "linear_103": ["mul_10058", "view_1627", "permute_173", "mm_17", "view_1628"], "view_52": ["view_1629"], "transpose_69": ["permute_174"], "contiguous_69": ["clone_139"], "linear_104": ["mul_10159", "view_1643", "permute_175", "mm_default_73", "add_tensor_73", "view_1644"], "view_53": ["view_1645"], "transpose_70": ["permute_176"], "contiguous_70": ["clone_140"], "scaled_dot_product_attention_17": ["_scaled_dot_product_efficient_attention_17", "getitem_138"], "transpose_71": ["permute_177"], "reshape_17": ["view_1646"], "choose_qparams_affine_default_105": ["view_1646", "amax_105", "full_211", "maximum_105", "amin_105", "full_210", "minimum_105", "sub_4813", "div_210", "clamp_min_315", "div_211", "round_211", "sub_4819", "clamp_min_316", "clamp_max_210", "convert_element_type_630"], "quantize_affine_105": ["view_1646", "view_1649", "reciprocal_105", "mul_10229", "mul_10232", "round_212", "view_1650", "add_16189", "clamp_min_317", "clamp_max_211", "convert_element_type_631"], "dequantize_affine_210": ["convert_element_type_631", "view_1654", "convert_element_type_632", "convert_element_type_633", "sub_4839", "view_1653", "mul_10254"], "dequantize_affine_211": ["view_1656", "view_1658", "convert_element_type_634", "convert_element_type_635", "sub_4843", "view_1657", "mul_10259", "view_1659"], "linear_105": ["mul_10264", "view_1660", "permute_178", "mm_default_72", "add_tensor_72", "view_1661"], "dropout_52": ["view_1661"], "add_247": ["add_16252"], "layer_norm_35": ["clone_142", "var_mean_35", "getitem_142", "getitem_143", "sub_4849", "add_16257", "rsqrt_35", "mul_10275", "mul_10276", "add_16258"], "choose_qparams_affine_default_106": ["add_16258", "amax_106", "full_213", "maximum_106", "amin_106", "full_212", "minimum_106", "sub_4860", "div_212", "clamp_min_318", "div_213", "round_213", "sub_4866", "clamp_min_319", "clamp_max_212", "convert_element_type_636"], "quantize_affine_106": ["add_16258", "view_1664", "reciprocal_106", "mul_10324", "mul_10327", "round_214", "view_1665", "add_16345", "clamp_min_320", "clamp_max_213", "convert_element_type_637"], "dequantize_affine_212": ["convert_element_type_637", "view_1669", "convert_element_type_638", "convert_element_type_639", "sub_4886", "view_1668", "mul_10349"], "dequantize_affine_213": ["view_1671", "view_1673", "convert_element_type_640", "convert_element_type_641", "sub_4890", "view_1672", "mul_10354", "view_1674"], "linear_106": ["mul_10359", "view_1675", "permute_179", "mm_default_71", "add_tensor_71", "view_1676"], "gelu_19": ["mul_10366", "mul_10367", "erf_19", "add_16404", "mul_10368"], "dropout_53": ["mul_10368"], "choose_qparams_affine_default_107": ["mul_10368", "amax_107", "full_215", "maximum_107", "amin_107", "full_214", "minimum_107", "sub_4903", "div_214", "clamp_min_321", "div_215", "round_215", "sub_4909", "clamp_min_322", "clamp_max_214", "convert_element_type_642"], "quantize_affine_107": ["mul_10368", "view_1679", "reciprocal_107", "mul_10414", "mul_10417", "round_216", "view_1680", "add_16487", "clamp_min_323", "clamp_max_215", "convert_element_type_643"], "dequantize_affine_214": ["convert_element_type_643", "view_1684", "convert_element_type_644", "convert_element_type_645", "sub_4929", "view_1683", "mul_10439"], "dequantize_affine_215": ["view_1686", "view_1688", "convert_element_type_646", "convert_element_type_647", "sub_4933", "view_1687", "mul_10444", "view_1689"], "linear_107": ["mul_10449", "view_1690", "permute_180", "mm_default_70", "add_tensor_70", "view_1691"], "dropout_54": ["view_1691"], "add_252": ["add_16550"], "layer_norm_36": ["clone_145", "var_mean_36", "getitem_144", "getitem_145", "sub_4939", "add_16555", "rsqrt_36", "mul_10460", "mul_10461", "add_16556"], "choose_qparams_affine_default_108": ["add_16556", "amax_108", "full_217", "maximum_108", "amin_108", "full_216", "minimum_108", "sub_4950", "div_216", "clamp_min_324", "div_217", "round_217", "sub_4956", "clamp_min_325", "clamp_max_216", "convert_element_type_648"], "quantize_affine_108": ["add_16556", "view_1694", "reciprocal_108", "mul_10509", "mul_10512", "round_218", "view_1695", "add_16643", "clamp_min_326", "clamp_max_217", "convert_element_type_649"], "choose_qparams_affine_default_109": ["add_16556", "amax_109", "full_219", "maximum_109", "amin_109", "full_218", "minimum_109", "sub_4995", "div_218", "clamp_min_327", "div_219", "round_219", "sub_5001", "clamp_min_328", "clamp_max_218", "convert_element_type_654"], "quantize_affine_109": ["add_16556", "view_1710", "reciprocal_109", "mul_10605", "mul_10608", "round_220", "view_1711", "add_16795", "clamp_min_329", "clamp_max_219", "convert_element_type_655"], "choose_qparams_affine_default_110": ["add_16556", "amax_110", "full_221", "maximum_110", "amin_110", "full_220", "minimum_110", "sub_5039", "div_220", "clamp_min_330", "div_221", "round_221", "sub_5045", "clamp_min_331", "clamp_max_220", "convert_element_type_660"], "quantize_affine_110": ["add_16556", "view_1726", "reciprocal_110", "mul_10704", "mul_10707", "round_222", "view_1727", "add_16943", "clamp_min_332", "clamp_max_221", "convert_element_type_661"], "dequantize_affine_216": ["convert_element_type_649", "view_1699", "convert_element_type_650", "convert_element_type_651", "sub_4976", "view_1698", "mul_10534"], "dequantize_affine_217": ["view_1701", "view_1703", "convert_element_type_652", "convert_element_type_653", "sub_4980", "view_1702", "mul_10539", "view_1704"], "dequantize_affine_218": ["convert_element_type_655", "view_1715", "convert_element_type_656", "convert_element_type_657", "sub_5021", "view_1714", "mul_10630"], "dequantize_affine_219": ["view_1717", "view_1719", "convert_element_type_658", "convert_element_type_659", "sub_5025", "view_1718", "mul_10635", "view_1720"], "dequantize_affine_220": ["convert_element_type_661", "view_1731", "convert_element_type_662", "convert_element_type_663", "sub_5065", "view_1730", "mul_10729"], "dequantize_affine_221": ["view_1733", "view_1735", "convert_element_type_664", "convert_element_type_665", "sub_5069", "view_1734", "mul_10734", "view_1736"], "linear_108": ["mul_10544", "view_1705", "permute_181", "mm_default_69", "add_tensor_69", "view_1706"], "mul_671": ["mul_10551"], "view_54": ["view_1707"], "transpose_72": ["permute_182"], "contiguous_72": ["clone_146"], "linear_109": ["mul_10638", "view_1721", "permute_183", "mm_18", "view_1722"], "view_55": ["view_1723"], "transpose_73": ["permute_184"], "contiguous_73": ["clone_147"], "linear_110": ["mul_10739", "view_1737", "permute_185", "mm_default_68", "add_tensor_68", "view_1738"], "view_56": ["view_1739"], "transpose_74": ["permute_186"], "contiguous_74": ["clone_148"], "scaled_dot_product_attention_18": ["_scaled_dot_product_efficient_attention_18", "getitem_146"], "transpose_75": ["permute_187"], "reshape_18": ["view_1740"], "choose_qparams_affine_default_111": ["view_1740", "amax_111", "full_223", "maximum_111", "amin_111", "full_222", "minimum_111", "sub_5087", "div_222", "clamp_min_333", "div_223", "round_223", "sub_5093", "clamp_min_334", "clamp_max_222", "convert_element_type_666"], "quantize_affine_111": ["view_1740", "view_1743", "reciprocal_111", "mul_10809", "mul_10812", "round_224", "view_1744", "add_17107", "clamp_min_335", "clamp_max_223", "convert_element_type_667"], "dequantize_affine_222": ["convert_element_type_667", "view_1748", "convert_element_type_668", "convert_element_type_669", "sub_5113", "view_1747", "mul_10834"], "dequantize_affine_223": ["view_1750", "view_1752", "convert_element_type_670", "convert_element_type_671", "sub_5117", "view_1751", "mul_10839", "view_1753"], "linear_111": ["mul_10844", "view_1754", "permute_188", "mm_default_67", "add_tensor_67", "view_1755"], "dropout_55": ["view_1755"], "add_261": ["add_17170"], "layer_norm_37": ["clone_150", "var_mean_37", "getitem_150", "getitem_151", "sub_5123", "add_17175", "rsqrt_37", "mul_10855", "mul_10856", "add_17176"], "choose_qparams_affine_default_112": ["add_17176", "amax_112", "full_225", "maximum_112", "amin_112", "full_224", "minimum_112", "sub_5134", "div_224", "clamp_min_336", "div_225", "round_225", "sub_5140", "clamp_min_337", "clamp_max_224", "convert_element_type_672"], "quantize_affine_112": ["add_17176", "view_1758", "reciprocal_112", "mul_10904", "mul_10907", "round_226", "view_1759", "add_17263", "clamp_min_338", "clamp_max_225", "convert_element_type_673"], "dequantize_affine_224": ["convert_element_type_673", "view_1763", "convert_element_type_674", "convert_element_type_675", "sub_5160", "view_1762", "mul_10929"], "dequantize_affine_225": ["view_1765", "view_1767", "convert_element_type_676", "convert_element_type_677", "sub_5164", "view_1766", "mul_10934", "view_1768"], "linear_112": ["mul_10939", "view_1769", "permute_189", "mm_default_66", "add_tensor_66", "view_1770"], "gelu_20": ["mul_10946", "mul_10947", "erf_20", "add_17322", "mul_10948"], "dropout_56": ["mul_10948"], "choose_qparams_affine_default_113": ["mul_10948", "amax_113", "full_227", "maximum_113", "amin_113", "full_226", "minimum_113", "sub_5177", "div_226", "clamp_min_339", "div_227", "round_227", "sub_5183", "clamp_min_340", "clamp_max_226", "convert_element_type_678"], "quantize_affine_113": ["mul_10948", "view_1773", "reciprocal_113", "mul_10994", "mul_10997", "round_228", "view_1774", "add_17405", "clamp_min_341", "clamp_max_227", "convert_element_type_679"], "dequantize_affine_226": ["convert_element_type_679", "view_1778", "convert_element_type_680", "convert_element_type_681", "sub_5203", "view_1777", "mul_11019"], "dequantize_affine_227": ["view_1780", "view_1782", "convert_element_type_682", "convert_element_type_683", "sub_5207", "view_1781", "mul_11024", "view_1783"], "linear_113": ["mul_11029", "view_1784", "permute_190", "mm_default_65", "add_tensor_65", "view_1785"], "dropout_57": ["view_1785"], "add_266": ["add_17468"], "layer_norm_38": ["clone_153", "var_mean_38", "getitem_152", "getitem_153", "sub_5213", "add_17473", "rsqrt_38", "mul_11040", "mul_11041", "add_17474"], "choose_qparams_affine_default_114": ["add_17474", "amax_114", "full_229", "maximum_114", "amin_114", "full_228", "minimum_114", "sub_5224", "div_228", "clamp_min_342", "div_229", "round_229", "sub_5230", "clamp_min_343", "clamp_max_228", "convert_element_type_684"], "quantize_affine_114": ["add_17474", "view_1788", "reciprocal_114", "mul_11089", "mul_11092", "round_230", "view_1789", "add_17561", "clamp_min_344", "clamp_max_229", "convert_element_type_685"], "choose_qparams_affine_default_115": ["add_17474", "amax_115", "full_231", "maximum_115", "amin_115", "full_230", "minimum_115", "sub_5269", "div_230", "clamp_min_345", "div_231", "round_231", "sub_5275", "clamp_min_346", "clamp_max_230", "convert_element_type_690"], "quantize_affine_115": ["add_17474", "view_1804", "reciprocal_115", "mul_11185", "mul_11188", "round_232", "view_1805", "add_17713", "clamp_min_347", "clamp_max_231", "convert_element_type_691"], "choose_qparams_affine_default_116": ["add_17474", "amax_116", "full_233", "maximum_116", "amin_116", "full_232", "minimum_116", "sub_5313", "div_232", "clamp_min_348", "div_233", "round_233", "sub_5319", "clamp_min_349", "clamp_max_232", "convert_element_type_696"], "quantize_affine_116": ["add_17474", "view_1820", "reciprocal_116", "mul_11284", "mul_11287", "round_234", "view_1821", "add_17861", "clamp_min_350", "clamp_max_233", "convert_element_type_697"], "dequantize_affine_228": ["convert_element_type_685", "view_1793", "convert_element_type_686", "convert_element_type_687", "sub_5250", "view_1792", "mul_11114"], "dequantize_affine_229": ["view_1795", "view_1797", "convert_element_type_688", "convert_element_type_689", "sub_5254", "view_1796", "mul_11119", "view_1798"], "dequantize_affine_230": ["convert_element_type_691", "view_1809", "convert_element_type_692", "convert_element_type_693", "sub_5295", "view_1808", "mul_11210"], "dequantize_affine_231": ["view_1811", "view_1813", "convert_element_type_694", "convert_element_type_695", "sub_5299", "view_1812", "mul_11215", "view_1814"], "dequantize_affine_232": ["convert_element_type_697", "view_1825", "convert_element_type_698", "convert_element_type_699", "sub_5339", "view_1824", "mul_11309"], "dequantize_affine_233": ["view_1827", "view_1829", "convert_element_type_700", "convert_element_type_701", "sub_5343", "view_1828", "mul_11314", "view_1830"], "linear_114": ["mul_11124", "view_1799", "permute_191", "mm_default_64", "add_tensor_64", "view_1800"], "mul_708": ["mul_11131"], "view_57": ["view_1801"], "transpose_76": ["permute_192"], "contiguous_76": ["clone_154"], "linear_115": ["mul_11218", "view_1815", "permute_193", "mm_19", "view_1816"], "view_58": ["view_1817"], "transpose_77": ["permute_194"], "contiguous_77": ["clone_155"], "linear_116": ["mul_11319", "view_1831", "permute_195", "mm_default_63", "add_tensor_63", "view_1832"], "view_59": ["view_1833"], "transpose_78": ["permute_196"], "contiguous_78": ["clone_156"], "scaled_dot_product_attention_19": ["_scaled_dot_product_efficient_attention_19", "getitem_154"], "transpose_79": ["permute_197"], "reshape_19": ["view_1834"], "choose_qparams_affine_default_117": ["view_1834", "amax_117", "full_235", "maximum_117", "amin_117", "full_234", "minimum_117", "sub_5361", "div_234", "clamp_min_351", "div_235", "round_235", "sub_5367", "clamp_min_352", "clamp_max_234", "convert_element_type_702"], "quantize_affine_117": ["view_1834", "view_1837", "reciprocal_117", "mul_11389", "mul_11392", "round_236", "view_1838", "add_18025", "clamp_min_353", "clamp_max_235", "convert_element_type_703"], "dequantize_affine_234": ["convert_element_type_703", "view_1842", "convert_element_type_704", "convert_element_type_705", "sub_5387", "view_1841", "mul_11414"], "dequantize_affine_235": ["view_1844", "view_1846", "convert_element_type_706", "convert_element_type_707", "sub_5391", "view_1845", "mul_11419", "view_1847"], "linear_117": ["mul_11424", "view_1848", "permute_198", "mm_default_62", "add_tensor_62", "view_1849"], "dropout_58": ["view_1849"], "add_275": ["add_18088"], "layer_norm_39": ["clone_158", "var_mean_39", "getitem_158", "getitem_159", "sub_5397", "add_18093", "rsqrt_39", "mul_11435", "mul_11436", "add_18094"], "choose_qparams_affine_default_118": ["add_18094", "amax_118", "full_237", "maximum_118", "amin_118", "full_236", "minimum_118", "sub_5408", "div_236", "clamp_min_354", "div_237", "round_237", "sub_5414", "clamp_min_355", "clamp_max_236", "convert_element_type_708"], "quantize_affine_118": ["add_18094", "view_1852", "reciprocal_118", "mul_11484", "mul_11487", "round_238", "view_1853", "add_18181", "clamp_min_356", "clamp_max_237", "convert_element_type_709"], "dequantize_affine_236": ["convert_element_type_709", "view_1857", "convert_element_type_710", "convert_element_type_711", "sub_5434", "view_1856", "mul_11509"], "dequantize_affine_237": ["view_1859", "view_1861", "convert_element_type_712", "convert_element_type_713", "sub_5438", "view_1860", "mul_11514", "view_1862"], "linear_118": ["mul_11519", "view_1863", "permute_199", "mm_default_61", "add_tensor_61", "view_1864"], "gelu_21": ["mul_11526", "mul_11527", "erf_21", "add_18240", "mul_11528"], "dropout_59": ["mul_11528"], "choose_qparams_affine_default_119": ["mul_11528", "amax_119", "full_239", "maximum_119", "amin_119", "full_238", "minimum_119", "sub_5451", "div_238", "clamp_min_357", "div_239", "round_239", "sub_5457", "clamp_min_358", "clamp_max_238", "convert_element_type_714"], "quantize_affine_119": ["mul_11528", "view_1867", "reciprocal_119", "mul_11574", "mul_11577", "round_240", "view_1868", "add_18323", "clamp_min_359", "clamp_max_239", "convert_element_type_715"], "dequantize_affine_238": ["convert_element_type_715", "view_1872", "convert_element_type_716", "convert_element_type_717", "sub_5477", "view_1871", "mul_11599"], "dequantize_affine_239": ["view_1874", "view_1876", "convert_element_type_718", "convert_element_type_719", "sub_5481", "view_1875", "mul_11604", "view_1877"], "linear_119": ["mul_11609", "view_1878", "permute_200", "mm_default_60", "add_tensor_60", "view_1879"], "dropout_60": ["view_1879"], "add_280": ["add_18386"], "layer_norm_40": ["clone_161", "var_mean_40", "getitem_160", "getitem_161", "sub_5487", "add_18391", "rsqrt_40", "mul_11620", "mul_11621", "add_18392"], "choose_qparams_affine_default_120": ["add_18392", "amax_120", "full_241", "maximum_120", "amin_120", "full_240", "minimum_120", "sub_5498", "div_240", "clamp_min_360", "div_241", "round_241", "sub_5504", "clamp_min_361", "clamp_max_240", "convert_element_type_720"], "quantize_affine_120": ["add_18392", "view_1882", "reciprocal_120", "mul_11669", "mul_11672", "round_242", "view_1883", "add_18479", "clamp_min_362", "clamp_max_241", "convert_element_type_721"], "choose_qparams_affine_default_121": ["add_18392", "amax_121", "full_243", "maximum_121", "amin_121", "full_242", "minimum_121", "sub_5543", "div_242", "clamp_min_363", "div_243", "round_243", "sub_5549", "clamp_min_364", "clamp_max_242", "convert_element_type_726"], "quantize_affine_121": ["add_18392", "view_1898", "reciprocal_121", "mul_11765", "mul_11768", "round_244", "view_1899", "add_18631", "clamp_min_365", "clamp_max_243", "convert_element_type_727"], "choose_qparams_affine_default_122": ["add_18392", "amax_122", "full_245", "maximum_122", "amin_122", "full_244", "minimum_122", "sub_5587", "div_244", "clamp_min_366", "div_245", "round_245", "sub_5593", "clamp_min_367", "clamp_max_244", "convert_element_type_732"], "quantize_affine_122": ["add_18392", "view_1914", "reciprocal_122", "mul_11864", "mul_11867", "round_246", "view_1915", "add_18779", "clamp_min_368", "clamp_max_245", "convert_element_type_733"], "dequantize_affine_240": ["convert_element_type_721", "view_1887", "convert_element_type_722", "convert_element_type_723", "sub_5524", "view_1886", "mul_11694"], "dequantize_affine_241": ["view_1889", "view_1891", "convert_element_type_724", "convert_element_type_725", "sub_5528", "view_1890", "mul_11699", "view_1892"], "dequantize_affine_242": ["convert_element_type_727", "view_1903", "convert_element_type_728", "convert_element_type_729", "sub_5569", "view_1902", "mul_11790"], "dequantize_affine_243": ["view_1905", "view_1907", "convert_element_type_730", "convert_element_type_731", "sub_5573", "view_1906", "mul_11795", "view_1908"], "dequantize_affine_244": ["convert_element_type_733", "view_1919", "convert_element_type_734", "convert_element_type_735", "sub_5613", "view_1918", "mul_11889"], "dequantize_affine_245": ["view_1921", "view_1923", "convert_element_type_736", "convert_element_type_737", "sub_5617", "view_1922", "mul_11894", "view_1924"], "linear_120": ["mul_11704", "view_1893", "permute_201", "mm_default_59", "add_tensor_59", "view_1894"], "mul_745": ["mul_11711"], "view_60": ["view_1895"], "transpose_80": ["permute_202"], "contiguous_80": ["clone_162"], "linear_121": ["mul_11798", "view_1909", "permute_203", "mm_20", "view_1910"], "view_61": ["view_1911"], "transpose_81": ["permute_204"], "contiguous_81": ["clone_163"], "linear_122": ["mul_11899", "view_1925", "permute_205", "mm_default_58", "add_tensor_58", "view_1926"], "view_62": ["view_1927"], "transpose_82": ["permute_206"], "contiguous_82": ["clone_164"], "scaled_dot_product_attention_20": ["_scaled_dot_product_efficient_attention_20", "getitem_162"], "transpose_83": ["permute_207"], "reshape_20": ["view_1928"], "choose_qparams_affine_default_123": ["view_1928", "amax_123", "full_247", "maximum_123", "amin_123", "full_246", "minimum_123", "sub_5635", "div_246", "clamp_min_369", "div_247", "round_247", "sub_5641", "clamp_min_370", "clamp_max_246", "convert_element_type_738"], "quantize_affine_123": ["view_1928", "view_1931", "reciprocal_123", "mul_11969", "mul_11972", "round_248", "view_1932", "add_18943", "clamp_min_371", "clamp_max_247", "convert_element_type_739"], "dequantize_affine_246": ["convert_element_type_739", "view_1936", "convert_element_type_740", "convert_element_type_741", "sub_5661", "view_1935", "mul_11994"], "dequantize_affine_247": ["view_1938", "view_1940", "convert_element_type_742", "convert_element_type_743", "sub_5665", "view_1939", "mul_11999", "view_1941"], "linear_123": ["mul_12004", "view_1942", "permute_208", "mm_default_57", "add_tensor_57", "view_1943"], "dropout_61": ["view_1943"], "add_289": ["add_19006"], "layer_norm_41": ["clone_166", "var_mean_41", "getitem_166", "getitem_167", "sub_5671", "add_19011", "rsqrt_41", "mul_12015", "mul_12016", "add_19012"], "choose_qparams_affine_default_124": ["add_19012", "amax_124", "full_249", "maximum_124", "amin_124", "full_248", "minimum_124", "sub_5682", "div_248", "clamp_min_372", "div_249", "round_249", "sub_5688", "clamp_min_373", "clamp_max_248", "convert_element_type_744"], "quantize_affine_124": ["add_19012", "view_1946", "reciprocal_124", "mul_12064", "mul_12067", "round_250", "view_1947", "add_19099", "clamp_min_374", "clamp_max_249", "convert_element_type_745"], "dequantize_affine_248": ["convert_element_type_745", "view_1951", "convert_element_type_746", "convert_element_type_747", "sub_5708", "view_1950", "mul_12089"], "dequantize_affine_249": ["view_1953", "view_1955", "convert_element_type_748", "convert_element_type_749", "sub_5712", "view_1954", "mul_12094", "view_1956"], "linear_124": ["mul_12099", "view_1957", "permute_209", "mm_default_56", "add_tensor_56", "view_1958"], "gelu_22": ["mul_12106", "mul_12107", "erf_22", "add_19158", "mul_12108"], "dropout_62": ["mul_12108"], "choose_qparams_affine_default_125": ["mul_12108", "amax_125", "full_251", "maximum_125", "amin_125", "full_250", "minimum_125", "sub_5725", "div_250", "clamp_min_375", "div_251", "round_251", "sub_5731", "clamp_min_376", "clamp_max_250", "convert_element_type_750"], "quantize_affine_125": ["mul_12108", "view_1961", "reciprocal_125", "mul_12154", "mul_12157", "round_252", "view_1962", "add_19241", "clamp_min_377", "clamp_max_251", "convert_element_type_751"], "dequantize_affine_250": ["convert_element_type_751", "view_1966", "convert_element_type_752", "convert_element_type_753", "sub_5751", "view_1965", "mul_12179"], "dequantize_affine_251": ["view_1968", "view_1970", "convert_element_type_754", "convert_element_type_755", "sub_5755", "view_1969", "mul_12184", "view_1971"], "linear_125": ["mul_12189", "view_1972", "permute_210", "mm_default_55", "add_tensor_55", "view_1973"], "dropout_63": ["view_1973"], "add_294": ["add_19304"], "layer_norm_42": ["clone_169", "var_mean_42", "getitem_168", "getitem_169", "sub_5761", "add_19309", "rsqrt_42", "mul_12200", "mul_12201", "add_19310"], "choose_qparams_affine_default_126": ["add_19310", "amax_126", "full_253", "maximum_126", "amin_126", "full_252", "minimum_126", "sub_5772", "div_252", "clamp_min_378", "div_253", "round_253", "sub_5778", "clamp_min_379", "clamp_max_252", "convert_element_type_756"], "quantize_affine_126": ["add_19310", "view_1976", "reciprocal_126", "mul_12249", "mul_12252", "round_254", "view_1977", "add_19397", "clamp_min_380", "clamp_max_253", "convert_element_type_757"], "choose_qparams_affine_default_127": ["add_19310", "amax_127", "full_255", "maximum_127", "amin_127", "full_254", "minimum_127", "sub_5817", "div_254", "clamp_min_381", "div_255", "round_255", "sub_5823", "clamp_min_382", "clamp_max_254", "convert_element_type_762"], "quantize_affine_127": ["add_19310", "view_1992", "reciprocal_127", "mul_12345", "mul_12348", "round_256", "view_1993", "add_19549", "clamp_min_383", "clamp_max_255", "convert_element_type_763"], "choose_qparams_affine_default_128": ["add_19310", "amax_128", "full_257", "maximum_128", "amin_128", "full_256", "minimum_128", "sub_5861", "div_256", "clamp_min_384", "div_257", "round_257", "sub_5867", "clamp_min_385", "clamp_max_256", "convert_element_type_768"], "quantize_affine_128": ["add_19310", "view_2008", "reciprocal_128", "mul_12444", "mul_12447", "round_258", "view_2009", "add_19697", "clamp_min_386", "clamp_max_257", "convert_element_type_769"], "dequantize_affine_252": ["convert_element_type_757", "view_1981", "convert_element_type_758", "convert_element_type_759", "sub_5798", "view_1980", "mul_12274"], "dequantize_affine_253": ["view_1983", "view_1985", "convert_element_type_760", "convert_element_type_761", "sub_5802", "view_1984", "mul_12279", "view_1986"], "dequantize_affine_254": ["convert_element_type_763", "view_1997", "convert_element_type_764", "convert_element_type_765", "sub_5843", "view_1996", "mul_12370"], "dequantize_affine_255": ["view_1999", "view_2001", "convert_element_type_766", "convert_element_type_767", "sub_5847", "view_2000", "mul_12375", "view_2002"], "dequantize_affine_256": ["convert_element_type_769", "view_2013", "convert_element_type_770", "convert_element_type_771", "sub_5887", "view_2012", "mul_12469"], "dequantize_affine_257": ["view_2015", "view_2017", "convert_element_type_772", "convert_element_type_773", "sub_5891", "view_2016", "mul_12474", "view_2018"], "linear_126": ["mul_12284", "view_1987", "permute_211", "mm_default_54", "add_tensor_54", "view_1988"], "mul_782": ["mul_12291"], "view_63": ["view_1989"], "transpose_84": ["permute_212"], "contiguous_84": ["clone_170"], "linear_127": ["mul_12378", "view_2003", "permute_213", "mm_21", "view_2004"], "view_64": ["view_2005"], "transpose_85": ["permute_214"], "contiguous_85": ["clone_171"], "linear_128": ["mul_12479", "view_2019", "permute_215", "mm_default_53", "add_tensor_53", "view_2020"], "view_65": ["view_2021"], "transpose_86": ["permute_216"], "contiguous_86": ["clone_172"], "scaled_dot_product_attention_21": ["_scaled_dot_product_efficient_attention_21", "getitem_170"], "transpose_87": ["permute_217"], "reshape_21": ["view_2022"], "choose_qparams_affine_default_129": ["view_2022", "amax_129", "full_259", "maximum_129", "amin_129", "full_258", "minimum_129", "sub_5909", "div_258", "clamp_min_387", "div_259", "round_259", "sub_5915", "clamp_min_388", "clamp_max_258", "convert_element_type_774"], "quantize_affine_129": ["view_2022", "view_2025", "reciprocal_129", "mul_12549", "mul_12552", "round_260", "view_2026", "add_19861", "clamp_min_389", "clamp_max_259", "convert_element_type_775"], "dequantize_affine_258": ["convert_element_type_775", "view_2030", "convert_element_type_776", "convert_element_type_777", "sub_5935", "view_2029", "mul_12574"], "dequantize_affine_259": ["view_2032", "view_2034", "convert_element_type_778", "convert_element_type_779", "sub_5939", "view_2033", "mul_12579", "view_2035"], "linear_129": ["mul_12584", "view_2036", "permute_218", "mm_default_52", "add_tensor_52", "view_2037"], "dropout_64": ["view_2037"], "add_303": ["add_19924"], "layer_norm_43": ["clone_174", "var_mean_43", "getitem_174", "getitem_175", "sub_5945", "add_19929", "rsqrt_43", "mul_12595", "mul_12596", "add_19930"], "choose_qparams_affine_default_130": ["add_19930", "amax_130", "full_261", "maximum_130", "amin_130", "full_260", "minimum_130", "sub_5956", "div_260", "clamp_min_390", "div_261", "round_261", "sub_5962", "clamp_min_391", "clamp_max_260", "convert_element_type_780"], "quantize_affine_130": ["add_19930", "view_2040", "reciprocal_130", "mul_12644", "mul_12647", "round_262", "view_2041", "add_20017", "clamp_min_392", "clamp_max_261", "convert_element_type_781"], "dequantize_affine_260": ["convert_element_type_781", "view_2045", "convert_element_type_782", "convert_element_type_783", "sub_5982", "view_2044", "mul_12669"], "dequantize_affine_261": ["view_2047", "view_2049", "convert_element_type_784", "convert_element_type_785", "sub_5986", "view_2048", "mul_12674", "view_2050"], "linear_130": ["mul_12679", "view_2051", "permute_219", "mm_default_51", "add_tensor_51", "view_2052"], "gelu_23": ["mul_12686", "mul_12687", "erf_23", "add_20076", "mul_12688"], "dropout_65": ["mul_12688"], "choose_qparams_affine_default_131": ["mul_12688", "amax_131", "full_263", "maximum_131", "amin_131", "full_262", "minimum_131", "sub_5999", "div_262", "clamp_min_393", "div_263", "round_263", "sub_6005", "clamp_min_394", "clamp_max_262", "convert_element_type_786"], "quantize_affine_131": ["mul_12688", "view_2055", "reciprocal_131", "mul_12734", "mul_12737", "round_264", "view_2056", "add_20159", "clamp_min_395", "clamp_max_263", "convert_element_type_787"], "dequantize_affine_262": ["convert_element_type_787", "view_2060", "convert_element_type_788", "convert_element_type_789", "sub_6025", "view_2059", "mul_12759"], "dequantize_affine_263": ["view_2062", "view_2064", "convert_element_type_790", "convert_element_type_791", "sub_6029", "view_2063", "mul_12764", "view_2065"], "linear_131": ["mul_12769", "view_2066", "permute_220", "mm_default_50", "add_tensor_50", "view_2067"], "dropout_66": ["view_2067"], "add_308": ["add_20222"], "layer_norm_44": ["clone_177", "var_mean_44", "getitem_176", "getitem_177", "sub_6035", "add_20227", "rsqrt_44", "mul_12780", "mul_12781", "add_20228"], "choose_qparams_affine_default_132": ["add_20228", "amax_132", "full_265", "maximum_132", "amin_132", "full_264", "minimum_132", "sub_6046", "div_264", "clamp_min_396", "div_265", "round_265", "sub_6052", "clamp_min_397", "clamp_max_264", "convert_element_type_792"], "quantize_affine_132": ["add_20228", "view_2070", "reciprocal_132", "mul_12829", "mul_12832", "round_266", "view_2071", "add_20315", "clamp_min_398", "clamp_max_265", "convert_element_type_793"], "choose_qparams_affine_default_133": ["add_20228", "amax_133", "full_267", "maximum_133", "amin_133", "full_266", "minimum_133", "sub_6091", "div_266", "clamp_min_399", "div_267", "round_267", "sub_6097", "clamp_min_400", "clamp_max_266", "convert_element_type_798"], "quantize_affine_133": ["add_20228", "view_2086", "reciprocal_133", "mul_12925", "mul_12928", "round_268", "view_2087", "add_20467", "clamp_min_401", "clamp_max_267", "convert_element_type_799"], "choose_qparams_affine_default_134": ["add_20228", "amax_134", "full_269", "maximum_134", "amin_134", "full_268", "minimum_134", "sub_6135", "div_268", "clamp_min_402", "div_269", "round_269", "sub_6141", "clamp_min_403", "clamp_max_268", "convert_element_type_804"], "quantize_affine_134": ["add_20228", "view_2102", "reciprocal_134", "mul_13024", "mul_13027", "round_270", "view_2103", "add_20615", "clamp_min_404", "clamp_max_269", "convert_element_type_805"], "dequantize_affine_264": ["convert_element_type_793", "view_2075", "convert_element_type_794", "convert_element_type_795", "sub_6072", "view_2074", "mul_12854"], "dequantize_affine_265": ["view_2077", "view_2079", "convert_element_type_796", "convert_element_type_797", "sub_6076", "view_2078", "mul_12859", "view_2080"], "dequantize_affine_266": ["convert_element_type_799", "view_2091", "convert_element_type_800", "convert_element_type_801", "sub_6117", "view_2090", "mul_12950"], "dequantize_affine_267": ["view_2093", "view_2095", "convert_element_type_802", "convert_element_type_803", "sub_6121", "view_2094", "mul_12955", "view_2096"], "dequantize_affine_268": ["convert_element_type_805", "view_2107", "convert_element_type_806", "convert_element_type_807", "sub_6161", "view_2106", "mul_13049"], "dequantize_affine_269": ["view_2109", "view_2111", "convert_element_type_808", "convert_element_type_809", "sub_6165", "view_2110", "mul_13054", "view_2112"], "linear_132": ["mul_12864", "view_2081", "permute_221", "mm_default_49", "add_tensor_49", "view_2082"], "mul_819": ["mul_12871"], "view_66": ["view_2083"], "transpose_88": ["permute_222"], "contiguous_88": ["clone_178"], "linear_133": ["mul_12958", "view_2097", "permute_223", "mm_22", "view_2098"], "view_67": ["view_2099"], "transpose_89": ["permute_224"], "contiguous_89": ["clone_179"], "linear_134": ["mul_13059", "view_2113", "permute_225", "mm_default_48", "add_tensor_48", "view_2114"], "view_68": ["view_2115"], "transpose_90": ["permute_226"], "contiguous_90": ["clone_180"], "scaled_dot_product_attention_22": ["_scaled_dot_product_efficient_attention_22", "getitem_178"], "transpose_91": ["permute_227"], "reshape_22": ["view_2116"], "choose_qparams_affine_default_135": ["view_2116", "amax_135", "full_271", "maximum_135", "amin_135", "full_270", "minimum_135", "sub_6183", "div_270", "clamp_min_405", "div_271", "round_271", "sub_6189", "clamp_min_406", "clamp_max_270", "convert_element_type_810"], "quantize_affine_135": ["view_2116", "view_2119", "reciprocal_135", "mul_13129", "mul_13132", "round_272", "view_2120", "add_20779", "clamp_min_407", "clamp_max_271", "convert_element_type_811"], "dequantize_affine_270": ["convert_element_type_811", "view_2124", "convert_element_type_812", "convert_element_type_813", "sub_6209", "view_2123", "mul_13154"], "dequantize_affine_271": ["view_2126", "view_2128", "convert_element_type_814", "convert_element_type_815", "sub_6213", "view_2127", "mul_13159", "view_2129"], "linear_135": ["mul_13164", "view_2130", "permute_228", "mm_default_47", "add_tensor_47", "view_2131"], "dropout_67": ["view_2131"], "add_317": ["add_20842"], "layer_norm_45": ["clone_182", "var_mean_45", "getitem_182", "getitem_183", "sub_6219", "add_20847", "rsqrt_45", "mul_13175", "mul_13176", "add_20848"], "choose_qparams_affine_default_136": ["add_20848", "amax_136", "full_273", "maximum_136", "amin_136", "full_272", "minimum_136", "sub_6230", "div_272", "clamp_min_408", "div_273", "round_273", "sub_6236", "clamp_min_409", "clamp_max_272", "convert_element_type_816"], "quantize_affine_136": ["add_20848", "view_2134", "reciprocal_136", "mul_13224", "mul_13227", "round_274", "view_2135", "add_20935", "clamp_min_410", "clamp_max_273", "convert_element_type_817"], "dequantize_affine_272": ["convert_element_type_817", "view_2139", "convert_element_type_818", "convert_element_type_819", "sub_6256", "view_2138", "mul_13249"], "dequantize_affine_273": ["view_2141", "view_2143", "convert_element_type_820", "convert_element_type_821", "sub_6260", "view_2142", "mul_13254", "view_2144"], "linear_136": ["mul_13259", "view_2145", "permute_229", "mm_default_46", "add_tensor_46", "view_2146"], "gelu_24": ["mul_13266", "mul_13267", "erf_24", "add_20994", "mul_13268"], "dropout_68": ["mul_13268"], "choose_qparams_affine_default_137": ["mul_13268", "amax_137", "full_275", "maximum_137", "amin_137", "full_274", "minimum_137", "sub_6273", "div_274", "clamp_min_411", "div_275", "round_275", "sub_6279", "clamp_min_412", "clamp_max_274", "convert_element_type_822"], "quantize_affine_137": ["mul_13268", "view_2149", "reciprocal_137", "mul_13314", "mul_13317", "round_276", "view_2150", "add_21077", "clamp_min_413", "clamp_max_275", "convert_element_type_823"], "dequantize_affine_274": ["convert_element_type_823", "view_2154", "convert_element_type_824", "convert_element_type_825", "sub_6299", "view_2153", "mul_13339"], "dequantize_affine_275": ["view_2156", "view_2158", "convert_element_type_826", "convert_element_type_827", "sub_6303", "view_2157", "mul_13344", "view_2159"], "linear_137": ["mul_13349", "view_2160", "permute_230", "mm_default_45", "add_tensor_45", "view_2161"], "dropout_69": ["view_2161"], "add_322": ["add_21140"], "layer_norm_46": ["clone_185", "var_mean_46", "getitem_184", "getitem_185", "sub_6309", "add_21145", "rsqrt_46", "mul_13360", "mul_13361", "add_21146"], "choose_qparams_affine_default_138": ["add_21146", "amax_138", "full_277", "maximum_138", "amin_138", "full_276", "minimum_138", "sub_6320", "div_276", "clamp_min_414", "div_277", "round_277", "sub_6326", "clamp_min_415", "clamp_max_276", "convert_element_type_828"], "quantize_affine_138": ["add_21146", "view_2164", "reciprocal_138", "mul_13409", "mul_13412", "round_278", "view_2165", "add_21233", "clamp_min_416", "clamp_max_277", "convert_element_type_829"], "choose_qparams_affine_default_139": ["add_21146", "amax_139", "full_279", "maximum_139", "amin_139", "full_278", "minimum_139", "sub_6365", "div_278", "clamp_min_417", "div_279", "round_279", "sub_6371", "clamp_min_418", "clamp_max_278", "convert_element_type_834"], "quantize_affine_139": ["add_21146", "view_2180", "reciprocal_139", "mul_13505", "mul_13508", "round_280", "view_2181", "add_21385", "clamp_min_419", "clamp_max_279", "convert_element_type_835"], "choose_qparams_affine_default_140": ["add_21146", "amax_140", "full_281", "maximum_140", "amin_140", "full_280", "minimum_140", "sub_6409", "div_280", "clamp_min_420", "div_281", "round_281", "sub_6415", "clamp_min_421", "clamp_max_280", "convert_element_type_840"], "quantize_affine_140": ["add_21146", "view_2196", "reciprocal_140", "mul_13604", "mul_13607", "round_282", "view_2197", "add_21533", "clamp_min_422", "clamp_max_281", "convert_element_type_841"], "dequantize_affine_276": ["convert_element_type_829", "view_2169", "convert_element_type_830", "convert_element_type_831", "sub_6346", "view_2168", "mul_13434"], "dequantize_affine_277": ["view_2171", "view_2173", "convert_element_type_832", "convert_element_type_833", "sub_6350", "view_2172", "mul_13439", "view_2174"], "dequantize_affine_278": ["convert_element_type_835", "view_2185", "convert_element_type_836", "convert_element_type_837", "sub_6391", "view_2184", "mul_13530"], "dequantize_affine_279": ["view_2187", "view_2189", "convert_element_type_838", "convert_element_type_839", "sub_6395", "view_2188", "mul_13535", "view_2190"], "dequantize_affine_280": ["convert_element_type_841", "view_2201", "convert_element_type_842", "convert_element_type_843", "sub_6435", "view_2200", "mul_13629"], "dequantize_affine_281": ["view_2203", "view_2205", "convert_element_type_844", "convert_element_type_845", "sub_6439", "view_2204", "mul_13634", "view_2206"], "linear_138": ["mul_13444", "view_2175", "permute_231", "mm_default_44", "add_tensor_44", "view_2176"], "mul_856": ["mul_13451"], "view_69": ["view_2177"], "transpose_92": ["permute_232"], "contiguous_92": ["clone_186"], "linear_139": ["mul_13538", "view_2191", "permute_233", "mm_23", "view_2192"], "view_70": ["view_2193"], "transpose_93": ["permute_234"], "contiguous_93": ["clone_187"], "linear_140": ["mul_13639", "view_2207", "permute_235", "mm_default_43", "add_tensor_43", "view_2208"], "view_71": ["view_2209"], "transpose_94": ["permute_236"], "contiguous_94": ["clone_188"], "scaled_dot_product_attention_23": ["_scaled_dot_product_efficient_attention_23", "getitem_186"], "transpose_95": ["permute_237"], "reshape_23": ["view_2210"], "choose_qparams_affine_default_141": ["view_2210", "amax_141", "full_283", "maximum_141", "amin_141", "full_282", "minimum_141", "sub_6457", "div_282", "clamp_min_423", "div_283", "round_283", "sub_6463", "clamp_min_424", "clamp_max_282", "convert_element_type_846"], "quantize_affine_141": ["view_2210", "view_2213", "reciprocal_141", "mul_13709", "mul_13712", "round_284", "view_2214", "add_21697", "clamp_min_425", "clamp_max_283", "convert_element_type_847"], "dequantize_affine_282": ["convert_element_type_847", "view_2218", "convert_element_type_848", "convert_element_type_849", "sub_6483", "view_2217", "mul_13734"], "dequantize_affine_283": ["view_2220", "view_2222", "convert_element_type_850", "convert_element_type_851", "sub_6487", "view_2221", "mul_13739", "view_2223"], "linear_141": ["mul_13744", "view_2224", "permute_238", "mm_default_42", "add_tensor_42", "view_2225"], "dropout_70": ["view_2225"], "add_331": ["add_21760"], "layer_norm_47": ["clone_190", "var_mean_47", "getitem_190", "getitem_191", "sub_6493", "add_21765", "rsqrt_47", "mul_13755", "mul_13756", "add_21766"], "choose_qparams_affine_default_142": ["add_21766", "amax_142", "full_285", "maximum_142", "amin_142", "full_284", "minimum_142", "sub_6504", "div_284", "clamp_min_426", "div_285", "round_285", "sub_6510", "clamp_min_427", "clamp_max_284", "convert_element_type_852"], "quantize_affine_142": ["add_21766", "view_2228", "reciprocal_142", "mul_13804", "mul_13807", "round_286", "view_2229", "add_21853", "clamp_min_428", "clamp_max_285", "convert_element_type_853"], "dequantize_affine_284": ["convert_element_type_853", "view_2233", "convert_element_type_854", "convert_element_type_855", "sub_6530", "view_2232", "mul_13829"], "dequantize_affine_285": ["view_2235", "view_2237", "convert_element_type_856", "convert_element_type_857", "sub_6534", "view_2236", "mul_13834", "view_2238"], "linear_142": ["mul_13839", "view_2239", "permute_239", "mm_default_41", "add_tensor_41", "view_2240"], "gelu_25": ["mul_13846", "mul_13847", "erf_25", "add_21912", "mul_13848"], "dropout_71": ["mul_13848"], "choose_qparams_affine_default_143": ["mul_13848", "amax_143", "full_287", "maximum_143", "amin_143", "full_286", "minimum_143", "sub_6547", "div_286", "clamp_min_429", "div_287", "round_287", "sub_6553", "clamp_min_430", "clamp_max_286", "convert_element_type_858"], "quantize_affine_143": ["mul_13848", "view_2243", "reciprocal_143", "mul_13894", "mul_13897", "round_288", "view_2244", "add_21995", "clamp_min_431", "clamp_max_287", "convert_element_type_859"], "dequantize_affine_286": ["convert_element_type_859", "view_2248", "convert_element_type_860", "convert_element_type_861", "sub_6573", "view_2247", "mul_13919"], "dequantize_affine_287": ["view_2250", "view_2252", "convert_element_type_862", "convert_element_type_863", "sub_6577", "view_2251", "mul_13924", "view_2253"], "linear_143": ["mul_13929", "view_2254", "permute_240", "mm_default_40", "add_tensor_40", "view_2255"], "dropout_72": ["view_2255"], "add_336": ["add_22058"], "layer_norm_48": ["clone_193", "var_mean_48", "getitem_192", "getitem_193", "sub_6583", "add_22063", "rsqrt_48", "mul_13940", "mul_13941", "add_22064"], "choose_qparams_affine_default_144": ["add_22064", "amax_144", "full_289", "maximum_144", "amin_144", "full_288", "minimum_144", "sub_6594", "div_288", "clamp_min_432", "div_289", "round_289", "sub_6600", "clamp_min_433", "clamp_max_288", "convert_element_type_864"], "quantize_affine_144": ["add_22064", "view_2258", "reciprocal_144", "mul_13989", "mul_13992", "round_290", "view_2259", "add_22151", "clamp_min_434", "clamp_max_289", "convert_element_type_865"], "choose_qparams_affine_default_145": ["add_22064", "amax_145", "full_291", "maximum_145", "amin_145", "full_290", "minimum_145", "sub_6639", "div_290", "clamp_min_435", "div_291", "round_291", "sub_6645", "clamp_min_436", "clamp_max_290", "convert_element_type_870"], "quantize_affine_145": ["add_22064", "view_2274", "reciprocal_145", "mul_14085", "mul_14088", "round_292", "view_2275", "add_22303", "clamp_min_437", "clamp_max_291", "convert_element_type_871"], "choose_qparams_affine_default_146": ["add_22064", "amax_146", "full_293", "maximum_146", "amin_146", "full_292", "minimum_146", "sub_6683", "div_292", "clamp_min_438", "div_293", "round_293", "sub_6689", "clamp_min_439", "clamp_max_292", "convert_element_type_876"], "quantize_affine_146": ["add_22064", "view_2290", "reciprocal_146", "mul_14184", "mul_14187", "round_294", "view_2291", "add_22451", "clamp_min_440", "clamp_max_293", "convert_element_type_877"], "dequantize_affine_288": ["convert_element_type_865", "view_2263", "convert_element_type_866", "convert_element_type_867", "sub_6620", "view_2262", "mul_14014"], "dequantize_affine_289": ["view_2265", "view_2267", "convert_element_type_868", "convert_element_type_869", "sub_6624", "view_2266", "mul_14019", "view_2268"], "dequantize_affine_290": ["convert_element_type_871", "view_2279", "convert_element_type_872", "convert_element_type_873", "sub_6665", "view_2278", "mul_14110"], "dequantize_affine_291": ["view_2281", "view_2283", "convert_element_type_874", "convert_element_type_875", "sub_6669", "view_2282", "mul_14115", "view_2284"], "dequantize_affine_292": ["convert_element_type_877", "view_2295", "convert_element_type_878", "convert_element_type_879", "sub_6709", "view_2294", "mul_14209"], "dequantize_affine_293": ["view_2297", "view_2299", "convert_element_type_880", "convert_element_type_881", "sub_6713", "view_2298", "mul_14214", "view_2300"], "linear_144": ["mul_14024", "view_2269", "permute_241", "mm_default_39", "add_tensor_39", "view_2270"], "mul_893": ["mul_14031"], "view_72": ["view_2271"], "transpose_96": ["permute_242"], "contiguous_96": ["clone_194"], "linear_145": ["mul_14118", "view_2285", "permute_243", "mm_24", "view_2286"], "view_73": ["view_2287"], "transpose_97": ["permute_244"], "contiguous_97": ["clone_195"], "linear_146": ["mul_14219", "view_2301", "permute_245", "mm_default_38", "add_tensor_38", "view_2302"], "view_74": ["view_2303"], "transpose_98": ["permute_246"], "contiguous_98": ["clone_196"], "scaled_dot_product_attention_24": ["_scaled_dot_product_efficient_attention_24", "getitem_194"], "transpose_99": ["permute_247"], "reshape_24": ["view_2304"], "choose_qparams_affine_default_147": ["view_2304", "amax_147", "full_295", "maximum_147", "amin_147", "full_294", "minimum_147", "sub_6731", "div_294", "clamp_min_441", "div_295", "round_295", "sub_6737", "clamp_min_442", "clamp_max_294", "convert_element_type_882"], "quantize_affine_147": ["view_2304", "view_2307", "reciprocal_147", "mul_14289", "mul_14292", "round_296", "view_2308", "add_22615", "clamp_min_443", "clamp_max_295", "convert_element_type_883"], "dequantize_affine_294": ["convert_element_type_883", "view_2312", "convert_element_type_884", "convert_element_type_885", "sub_6757", "view_2311", "mul_14314"], "dequantize_affine_295": ["view_2314", "view_2316", "convert_element_type_886", "convert_element_type_887", "sub_6761", "view_2315", "mul_14319", "view_2317"], "linear_147": ["mul_14324", "view_2318", "permute_248", "mm_default_37", "add_tensor_37", "view_2319"], "dropout_73": ["view_2319"], "add_345": ["add_22678"], "layer_norm_49": ["clone_198", "var_mean_49", "getitem_198", "getitem_199", "sub_6767", "add_22683", "rsqrt_49", "mul_14335", "mul_14336", "add_22684"], "choose_qparams_affine_default_148": ["add_22684", "amax_148", "full_297", "maximum_148", "amin_148", "full_296", "minimum_148", "sub_6778", "div_296", "clamp_min_444", "div_297", "round_297", "sub_6784", "clamp_min_445", "clamp_max_296", "convert_element_type_888"], "quantize_affine_148": ["add_22684", "view_2322", "reciprocal_148", "mul_14384", "mul_14387", "round_298", "view_2323", "add_22771", "clamp_min_446", "clamp_max_297", "convert_element_type_889"], "dequantize_affine_296": ["convert_element_type_889", "view_2327", "convert_element_type_890", "convert_element_type_891", "sub_6804", "view_2326", "mul_14409"], "dequantize_affine_297": ["view_2329", "view_2331", "convert_element_type_892", "convert_element_type_893", "sub_6808", "view_2330", "mul_14414", "view_2332"], "linear_148": ["mul_14419", "view_2333", "permute_249", "mm_default_36", "add_tensor_36", "view_2334"], "gelu_26": ["mul_14426", "mul_14427", "erf_26", "add_22830", "mul_14428"], "dropout_74": ["mul_14428"], "choose_qparams_affine_default_149": ["mul_14428", "amax_149", "full_299", "maximum_149", "amin_149", "full_298", "minimum_149", "sub_6821", "div_298", "clamp_min_447", "div_299", "round_299", "sub_6827", "clamp_min_448", "clamp_max_298", "convert_element_type_894"], "quantize_affine_149": ["mul_14428", "view_2337", "reciprocal_149", "mul_14474", "mul_14477", "round_300", "view_2338", "add_22913", "clamp_min_449", "clamp_max_299", "convert_element_type_895"], "dequantize_affine_298": ["convert_element_type_895", "view_2342", "convert_element_type_896", "convert_element_type_897", "sub_6847", "view_2341", "mul_14499"], "dequantize_affine_299": ["view_2344", "view_2346", "convert_element_type_898", "convert_element_type_899", "sub_6851", "view_2345", "mul_14504", "view_2347"], "linear_149": ["mul_14509", "view_2348", "permute_250", "mm_default_35", "add_tensor_35", "view_2349"], "dropout_75": ["view_2349"], "add_350": ["add_22976"], "layer_norm_50": ["clone_201", "var_mean_50", "getitem_200", "getitem_201", "sub_6857", "add_22981", "rsqrt_50", "mul_14520", "mul_14521", "add_22982"], "choose_qparams_affine_default_150": ["add_22982", "amax_150", "full_301", "maximum_150", "amin_150", "full_300", "minimum_150", "sub_6868", "div_300", "clamp_min_450", "div_301", "round_301", "sub_6874", "clamp_min_451", "clamp_max_300", "convert_element_type_900"], "quantize_affine_150": ["add_22982", "view_2352", "reciprocal_150", "mul_14569", "mul_14572", "round_302", "view_2353", "add_23069", "clamp_min_452", "clamp_max_301", "convert_element_type_901"], "choose_qparams_affine_default_151": ["add_22982", "amax_151", "full_303", "maximum_151", "amin_151", "full_302", "minimum_151", "sub_6913", "div_302", "clamp_min_453", "div_303", "round_303", "sub_6919", "clamp_min_454", "clamp_max_302", "convert_element_type_906"], "quantize_affine_151": ["add_22982", "view_2368", "reciprocal_151", "mul_14665", "mul_14668", "round_304", "view_2369", "add_23221", "clamp_min_455", "clamp_max_303", "convert_element_type_907"], "choose_qparams_affine_default_152": ["add_22982", "amax_152", "full_305", "maximum_152", "amin_152", "full_304", "minimum_152", "sub_6957", "div_304", "clamp_min_456", "div_305", "round_305", "sub_6963", "clamp_min_457", "clamp_max_304", "convert_element_type_912"], "quantize_affine_152": ["add_22982", "view_2384", "reciprocal_152", "mul_14764", "mul_14767", "round_306", "view_2385", "add_23369", "clamp_min_458", "clamp_max_305", "convert_element_type_913"], "dequantize_affine_300": ["convert_element_type_901", "view_2357", "convert_element_type_902", "convert_element_type_903", "sub_6894", "view_2356", "mul_14594"], "dequantize_affine_301": ["view_2359", "view_2361", "convert_element_type_904", "convert_element_type_905", "sub_6898", "view_2360", "mul_14599", "view_2362"], "dequantize_affine_302": ["convert_element_type_907", "view_2373", "convert_element_type_908", "convert_element_type_909", "sub_6939", "view_2372", "mul_14690"], "dequantize_affine_303": ["view_2375", "view_2377", "convert_element_type_910", "convert_element_type_911", "sub_6943", "view_2376", "mul_14695", "view_2378"], "dequantize_affine_304": ["convert_element_type_913", "view_2389", "convert_element_type_914", "convert_element_type_915", "sub_6983", "view_2388", "mul_14789"], "dequantize_affine_305": ["view_2391", "view_2393", "convert_element_type_916", "convert_element_type_917", "sub_6987", "view_2392", "mul_14794", "view_2394"], "linear_150": ["mul_14604", "view_2363", "permute_251", "mm_default_34", "add_tensor_34", "view_2364"], "mul_930": ["mul_14611"], "view_75": ["view_2365"], "transpose_100": ["permute_252"], "contiguous_100": ["clone_202"], "linear_151": ["mul_14698", "view_2379", "permute_253", "mm_25", "view_2380"], "view_76": ["view_2381"], "transpose_101": ["permute_254"], "contiguous_101": ["clone_203"], "linear_152": ["mul_14799", "view_2395", "permute_255", "mm_default_33", "add_tensor_33", "view_2396"], "view_77": ["view_2397"], "transpose_102": ["permute_256"], "contiguous_102": ["clone_204"], "scaled_dot_product_attention_25": ["_scaled_dot_product_efficient_attention_25", "getitem_202"], "transpose_103": ["permute_257"], "reshape_25": ["view_2398"], "choose_qparams_affine_default_153": ["view_2398", "amax_153", "full_307", "maximum_153", "amin_153", "full_306", "minimum_153", "sub_7005", "div_306", "clamp_min_459", "div_307", "round_307", "sub_7011", "clamp_min_460", "clamp_max_306", "convert_element_type_918"], "quantize_affine_153": ["view_2398", "view_2401", "reciprocal_153", "mul_14869", "mul_14872", "round_308", "view_2402", "add_23533", "clamp_min_461", "clamp_max_307", "convert_element_type_919"], "dequantize_affine_306": ["convert_element_type_919", "view_2406", "convert_element_type_920", "convert_element_type_921", "sub_7031", "view_2405", "mul_14894"], "dequantize_affine_307": ["view_2408", "view_2410", "convert_element_type_922", "convert_element_type_923", "sub_7035", "view_2409", "mul_14899", "view_2411"], "linear_153": ["mul_14904", "view_2412", "permute_258", "mm_default_32", "add_tensor_32", "view_2413"], "dropout_76": ["view_2413"], "add_359": ["add_23596"], "layer_norm_51": ["clone_206", "var_mean_51", "getitem_206", "getitem_207", "sub_7041", "add_23601", "rsqrt_51", "mul_14915", "mul_14916", "add_23602"], "choose_qparams_affine_default_154": ["add_23602", "amax_154", "full_309", "maximum_154", "amin_154", "full_308", "minimum_154", "sub_7052", "div_308", "clamp_min_462", "div_309", "round_309", "sub_7058", "clamp_min_463", "clamp_max_308", "convert_element_type_924"], "quantize_affine_154": ["add_23602", "view_2416", "reciprocal_154", "mul_14964", "mul_14967", "round_310", "view_2417", "add_23689", "clamp_min_464", "clamp_max_309", "convert_element_type_925"], "dequantize_affine_308": ["convert_element_type_925", "view_2421", "convert_element_type_926", "convert_element_type_927", "sub_7078", "view_2420", "mul_14989"], "dequantize_affine_309": ["view_2423", "view_2425", "convert_element_type_928", "convert_element_type_929", "sub_7082", "view_2424", "mul_14994", "view_2426"], "linear_154": ["mul_14999", "view_2427", "permute_259", "mm_default_31", "add_tensor_31", "view_2428"], "gelu_27": ["mul_15006", "mul_15007", "erf_27", "add_23748", "mul_15008"], "dropout_77": ["mul_15008"], "choose_qparams_affine_default_155": ["mul_15008", "amax_155", "full_311", "maximum_155", "amin_155", "full_310", "minimum_155", "sub_7095", "div_310", "clamp_min_465", "div_311", "round_311", "sub_7101", "clamp_min_466", "clamp_max_310", "convert_element_type_930"], "quantize_affine_155": ["mul_15008", "view_2431", "reciprocal_155", "mul_15054", "mul_15057", "round_312", "view_2432", "add_23831", "clamp_min_467", "clamp_max_311", "convert_element_type_931"], "dequantize_affine_310": ["convert_element_type_931", "view_2436", "convert_element_type_932", "convert_element_type_933", "sub_7121", "view_2435", "mul_15079"], "dequantize_affine_311": ["view_2438", "view_2440", "convert_element_type_934", "convert_element_type_935", "sub_7125", "view_2439", "mul_15084", "view_2441"], "linear_155": ["mul_15089", "view_2442", "permute_260", "mm_default_30", "add_tensor_30", "view_2443"], "dropout_78": ["view_2443"], "add_364": ["add_23894"], "layer_norm_52": ["clone_209", "var_mean_52", "getitem_208", "getitem_209", "sub_7131", "add_23899", "rsqrt_52", "mul_15100", "mul_15101", "add_23900"], "choose_qparams_affine_default_156": ["add_23900", "amax_156", "full_313", "maximum_156", "amin_156", "full_312", "minimum_156", "sub_7142", "div_312", "clamp_min_468", "div_313", "round_313", "sub_7148", "clamp_min_469", "clamp_max_312", "convert_element_type_936"], "quantize_affine_156": ["add_23900", "view_2446", "reciprocal_156", "mul_15149", "mul_15152", "round_314", "view_2447", "add_23987", "clamp_min_470", "clamp_max_313", "convert_element_type_937"], "choose_qparams_affine_default_157": ["add_23900", "amax_157", "full_315", "maximum_157", "amin_157", "full_314", "minimum_157", "sub_7187", "div_314", "clamp_min_471", "div_315", "round_315", "sub_7193", "clamp_min_472", "clamp_max_314", "convert_element_type_942"], "quantize_affine_157": ["add_23900", "view_2462", "reciprocal_157", "mul_15245", "mul_15248", "round_316", "view_2463", "add_24139", "clamp_min_473", "clamp_max_315", "convert_element_type_943"], "choose_qparams_affine_default_158": ["add_23900", "amax_158", "full_317", "maximum_158", "amin_158", "full_316", "minimum_158", "sub_7231", "div_316", "clamp_min_474", "div_317", "round_317", "sub_7237", "clamp_min_475", "clamp_max_316", "convert_element_type_948"], "quantize_affine_158": ["add_23900", "view_2478", "reciprocal_158", "mul_15344", "mul_15347", "round_318", "view_2479", "add_24287", "clamp_min_476", "clamp_max_317", "convert_element_type_949"], "dequantize_affine_312": ["convert_element_type_937", "view_2451", "convert_element_type_938", "convert_element_type_939", "sub_7168", "view_2450", "mul_15174"], "dequantize_affine_313": ["view_2453", "view_2455", "convert_element_type_940", "convert_element_type_941", "sub_7172", "view_2454", "mul_15179", "view_2456"], "dequantize_affine_314": ["convert_element_type_943", "view_2467", "convert_element_type_944", "convert_element_type_945", "sub_7213", "view_2466", "mul_15270"], "dequantize_affine_315": ["view_2469", "view_2471", "convert_element_type_946", "convert_element_type_947", "sub_7217", "view_2470", "mul_15275", "view_2472"], "dequantize_affine_316": ["convert_element_type_949", "view_2483", "convert_element_type_950", "convert_element_type_951", "sub_7257", "view_2482", "mul_15369"], "dequantize_affine_317": ["view_2485", "view_2487", "convert_element_type_952", "convert_element_type_953", "sub_7261", "view_2486", "mul_15374", "view_2488"], "linear_156": ["mul_15184", "view_2457", "permute_261", "mm_default_29", "add_tensor_29", "view_2458"], "mul_967": ["mul_15191"], "view_78": ["view_2459"], "transpose_104": ["permute_262"], "contiguous_104": ["clone_210"], "linear_157": ["mul_15278", "view_2473", "permute_263", "mm_26", "view_2474"], "view_79": ["view_2475"], "transpose_105": ["permute_264"], "contiguous_105": ["clone_211"], "linear_158": ["mul_15379", "view_2489", "permute_265", "mm_default_28", "add_tensor_28", "view_2490"], "view_80": ["view_2491"], "transpose_106": ["permute_266"], "contiguous_106": ["clone_212"], "scaled_dot_product_attention_26": ["_scaled_dot_product_efficient_attention_26", "getitem_210"], "transpose_107": ["permute_267"], "reshape_26": ["view_2492"], "choose_qparams_affine_default_159": ["view_2492", "amax_159", "full_319", "maximum_159", "amin_159", "full_318", "minimum_159", "sub_7279", "div_318", "clamp_min_477", "div_319", "round_319", "sub_7285", "clamp_min_478", "clamp_max_318", "convert_element_type_954"], "quantize_affine_159": ["view_2492", "view_2495", "reciprocal_159", "mul_15449", "mul_15452", "round_320", "view_2496", "add_24451", "clamp_min_479", "clamp_max_319", "convert_element_type_955"], "dequantize_affine_318": ["convert_element_type_955", "view_2500", "convert_element_type_956", "convert_element_type_957", "sub_7305", "view_2499", "mul_15474"], "dequantize_affine_319": ["view_2502", "view_2504", "convert_element_type_958", "convert_element_type_959", "sub_7309", "view_2503", "mul_15479", "view_2505"], "linear_159": ["mul_15484", "view_2506", "permute_268", "mm_default_27", "add_tensor_27", "view_2507"], "dropout_79": ["view_2507"], "add_373": ["add_24514"], "layer_norm_53": ["clone_214", "var_mean_53", "getitem_214", "getitem_215", "sub_7315", "add_24519", "rsqrt_53", "mul_15495", "mul_15496", "add_24520"], "choose_qparams_affine_default_160": ["add_24520", "amax_160", "full_321", "maximum_160", "amin_160", "full_320", "minimum_160", "sub_7326", "div_320", "clamp_min_480", "div_321", "round_321", "sub_7332", "clamp_min_481", "clamp_max_320", "convert_element_type_960"], "quantize_affine_160": ["add_24520", "view_2510", "reciprocal_160", "mul_15544", "mul_15547", "round_322", "view_2511", "add_24607", "clamp_min_482", "clamp_max_321", "convert_element_type_961"], "dequantize_affine_320": ["convert_element_type_961", "view_2515", "convert_element_type_962", "convert_element_type_963", "sub_7352", "view_2514", "mul_15569"], "dequantize_affine_321": ["view_2517", "view_2519", "convert_element_type_964", "convert_element_type_965", "sub_7356", "view_2518", "mul_15574", "view_2520"], "linear_160": ["mul_15579", "view_2521", "permute_269", "mm_default_26", "add_tensor_26", "view_2522"], "gelu_28": ["mul_15586", "mul_15587", "erf_28", "add_24666", "mul_15588"], "dropout_80": ["mul_15588"], "choose_qparams_affine_default_161": ["mul_15588", "amax_161", "full_323", "maximum_161", "amin_161", "full_322", "minimum_161", "sub_7369", "div_322", "clamp_min_483", "div_323", "round_323", "sub_7375", "clamp_min_484", "clamp_max_322", "convert_element_type_966"], "quantize_affine_161": ["mul_15588", "view_2525", "reciprocal_161", "mul_15634", "mul_15637", "round_324", "view_2526", "add_24749", "clamp_min_485", "clamp_max_323", "convert_element_type_967"], "dequantize_affine_322": ["convert_element_type_967", "view_2530", "convert_element_type_968", "convert_element_type_969", "sub_7395", "view_2529", "mul_15659"], "dequantize_affine_323": ["view_2532", "view_2534", "convert_element_type_970", "convert_element_type_971", "sub_7399", "view_2533", "mul_15664", "view_2535"], "linear_161": ["mul_15669", "view_2536", "permute_270", "mm_default_25", "add_tensor_25", "view_2537"], "dropout_81": ["view_2537"], "add_378": ["add_24812"], "layer_norm_54": ["clone_217", "var_mean_54", "getitem_216", "getitem_217", "sub_7405", "add_24817", "rsqrt_54", "mul_15680", "mul_15681", "add_24818"], "choose_qparams_affine_default_162": ["add_24818", "amax_162", "full_325", "maximum_162", "amin_162", "full_324", "minimum_162", "sub_7416", "div_324", "clamp_min_486", "div_325", "round_325", "sub_7422", "clamp_min_487", "clamp_max_324", "convert_element_type_972"], "quantize_affine_162": ["add_24818", "view_2540", "reciprocal_162", "mul_15729", "mul_15732", "round_326", "view_2541", "add_24905", "clamp_min_488", "clamp_max_325", "convert_element_type_973"], "choose_qparams_affine_default_163": ["add_24818", "amax_163", "full_327", "maximum_163", "amin_163", "full_326", "minimum_163", "sub_7461", "div_326", "clamp_min_489", "div_327", "round_327", "sub_7467", "clamp_min_490", "clamp_max_326", "convert_element_type_978"], "quantize_affine_163": ["add_24818", "view_2556", "reciprocal_163", "mul_15825", "mul_15828", "round_328", "view_2557", "add_25057", "clamp_min_491", "clamp_max_327", "convert_element_type_979"], "choose_qparams_affine_default_164": ["add_24818", "amax_164", "full_329", "maximum_164", "amin_164", "full_328", "minimum_164", "sub_7505", "div_328", "clamp_min_492", "div_329", "round_329", "sub_7511", "clamp_min_493", "clamp_max_328", "convert_element_type_984"], "quantize_affine_164": ["add_24818", "view_2572", "reciprocal_164", "mul_15924", "mul_15927", "round_330", "view_2573", "add_25205", "clamp_min_494", "clamp_max_329", "convert_element_type_985"], "dequantize_affine_324": ["convert_element_type_973", "view_2545", "convert_element_type_974", "convert_element_type_975", "sub_7442", "view_2544", "mul_15754"], "dequantize_affine_325": ["view_2547", "view_2549", "convert_element_type_976", "convert_element_type_977", "sub_7446", "view_2548", "mul_15759", "view_2550"], "dequantize_affine_326": ["convert_element_type_979", "view_2561", "convert_element_type_980", "convert_element_type_981", "sub_7487", "view_2560", "mul_15850"], "dequantize_affine_327": ["view_2563", "view_2565", "convert_element_type_982", "convert_element_type_983", "sub_7491", "view_2564", "mul_15855", "view_2566"], "dequantize_affine_328": ["convert_element_type_985", "view_2577", "convert_element_type_986", "convert_element_type_987", "sub_7531", "view_2576", "mul_15949"], "dequantize_affine_329": ["view_2579", "view_2581", "convert_element_type_988", "convert_element_type_989", "sub_7535", "view_2580", "mul_15954", "view_2582"], "linear_162": ["mul_15764", "view_2551", "permute_271", "mm_default_24", "add_tensor_24", "view_2552"], "mul_1004": ["mul_15771"], "view_81": ["view_2553"], "transpose_108": ["permute_272"], "contiguous_108": ["clone_218"], "linear_163": ["mul_15858", "view_2567", "permute_273", "mm_27", "view_2568"], "view_82": ["view_2569"], "transpose_109": ["permute_274"], "contiguous_109": ["clone_219"], "linear_164": ["mul_15959", "view_2583", "permute_275", "mm_default_23", "add_tensor_23", "view_2584"], "view_83": ["view_2585"], "transpose_110": ["permute_276"], "contiguous_110": ["clone_220"], "scaled_dot_product_attention_27": ["_scaled_dot_product_efficient_attention_27", "getitem_218"], "transpose_111": ["permute_277"], "reshape_27": ["view_2586"], "choose_qparams_affine_default_165": ["view_2586", "amax_165", "full_331", "maximum_165", "amin_165", "full_330", "minimum_165", "sub_7553", "div_330", "clamp_min_495", "div_331", "round_331", "sub_7559", "clamp_min_496", "clamp_max_330", "convert_element_type_990"], "quantize_affine_165": ["view_2586", "view_2589", "reciprocal_165", "mul_16029", "mul_16032", "round_332", "view_2590", "add_25369", "clamp_min_497", "clamp_max_331", "convert_element_type_991"], "dequantize_affine_330": ["convert_element_type_991", "view_2594", "convert_element_type_992", "convert_element_type_993", "sub_7579", "view_2593", "mul_16054"], "dequantize_affine_331": ["view_2596", "view_2598", "convert_element_type_994", "convert_element_type_995", "sub_7583", "view_2597", "mul_16059", "view_2599"], "linear_165": ["mul_16064", "view_2600", "permute_278", "mm_default_22", "add_tensor_22", "view_2601"], "dropout_82": ["view_2601"], "add_387": ["add_25432"], "layer_norm_55": ["clone_222", "var_mean_55", "getitem_222", "getitem_223", "sub_7589", "add_25437", "rsqrt_55", "mul_16075", "mul_16076", "add_25438"], "choose_qparams_affine_default_166": ["add_25438", "amax_166", "full_333", "maximum_166", "amin_166", "full_332", "minimum_166", "sub_7600", "div_332", "clamp_min_498", "div_333", "round_333", "sub_7606", "clamp_min_499", "clamp_max_332", "convert_element_type_996"], "quantize_affine_166": ["add_25438", "view_2604", "reciprocal_166", "mul_16124", "mul_16127", "round_334", "view_2605", "add_25525", "clamp_min_500", "clamp_max_333", "convert_element_type_997"], "dequantize_affine_332": ["convert_element_type_997", "view_2609", "convert_element_type_998", "convert_element_type_999", "sub_7626", "view_2608", "mul_16149"], "dequantize_affine_333": ["view_2611", "view_2613", "convert_element_type_1000", "convert_element_type_1001", "sub_7630", "view_2612", "mul_16154", "view_2614"], "linear_166": ["mul_16159", "view_2615", "permute_279", "mm_default_21", "add_tensor_21", "view_2616"], "gelu_29": ["mul_16166", "mul_16167", "erf_29", "add_25584", "mul_16168"], "dropout_83": ["mul_16168"], "choose_qparams_affine_default_167": ["mul_16168", "amax_167", "full_335", "maximum_167", "amin_167", "full_334", "minimum_167", "sub_7643", "div_334", "clamp_min_501", "div_335", "round_335", "sub_7649", "clamp_min_502", "clamp_max_334", "convert_element_type_1002"], "quantize_affine_167": ["mul_16168", "view_2619", "reciprocal_167", "mul_16214", "mul_16217", "round_336", "view_2620", "add_25667", "clamp_min_503", "clamp_max_335", "convert_element_type_1003"], "dequantize_affine_334": ["convert_element_type_1003", "view_2624", "convert_element_type_1004", "convert_element_type_1005", "sub_7669", "view_2623", "mul_16239"], "dequantize_affine_335": ["view_2626", "view_2628", "convert_element_type_1006", "convert_element_type_1007", "sub_7673", "view_2627", "mul_16244", "view_2629"], "linear_167": ["mul_16249", "view_2630", "permute_280", "mm_default_20", "add_tensor_20", "view_2631"], "dropout_84": ["view_2631"], "add_392": ["add_25730"], "layer_norm_56": ["clone_225", "var_mean_56", "getitem_224", "getitem_225", "sub_7679", "add_25735", "rsqrt_56", "mul_16260", "mul_16261", "add_25736"], "choose_qparams_affine_default_168": ["add_25736", "amax_168", "full_337", "maximum_168", "amin_168", "full_336", "minimum_168", "sub_7690", "div_336", "clamp_min_504", "div_337", "round_337", "sub_7696", "clamp_min_505", "clamp_max_336", "convert_element_type_1008"], "quantize_affine_168": ["add_25736", "view_2634", "reciprocal_168", "mul_16309", "mul_16312", "round_338", "view_2635", "add_25823", "clamp_min_506", "clamp_max_337", "convert_element_type_1009"], "choose_qparams_affine_default_169": ["add_25736", "amax_169", "full_339", "maximum_169", "amin_169", "full_338", "minimum_169", "sub_7735", "div_338", "clamp_min_507", "div_339", "round_339", "sub_7741", "clamp_min_508", "clamp_max_338", "convert_element_type_1014"], "quantize_affine_169": ["add_25736", "view_2650", "reciprocal_169", "mul_16405", "mul_16408", "round_340", "view_2651", "add_25975", "clamp_min_509", "clamp_max_339", "convert_element_type_1015"], "choose_qparams_affine_default_170": ["add_25736", "amax_170", "full_341", "maximum_170", "amin_170", "full_340", "minimum_170", "sub_7779", "div_340", "clamp_min_510", "div_341", "round_341", "sub_7785", "clamp_min_511", "clamp_max_340", "convert_element_type_1020"], "quantize_affine_170": ["add_25736", "view_2666", "reciprocal_170", "mul_16504", "mul_16507", "round_342", "view_2667", "add_26123", "clamp_min_512", "clamp_max_341", "convert_element_type_1021"], "dequantize_affine_336": ["convert_element_type_1009", "view_2639", "convert_element_type_1010", "convert_element_type_1011", "sub_7716", "view_2638", "mul_16334"], "dequantize_affine_337": ["view_2641", "view_2643", "convert_element_type_1012", "convert_element_type_1013", "sub_7720", "view_2642", "mul_16339", "view_2644"], "dequantize_affine_338": ["convert_element_type_1015", "view_2655", "convert_element_type_1016", "convert_element_type_1017", "sub_7761", "view_2654", "mul_16430"], "dequantize_affine_339": ["view_2657", "view_2659", "convert_element_type_1018", "convert_element_type_1019", "sub_7765", "view_2658", "mul_16435", "view_2660"], "dequantize_affine_340": ["convert_element_type_1021", "view_2671", "convert_element_type_1022", "convert_element_type_1023", "sub_7805", "view_2670", "mul_16529"], "dequantize_affine_341": ["view_2673", "view_2675", "convert_element_type_1024", "convert_element_type_1025", "sub_7809", "view_2674", "mul_16534", "view_2676"], "linear_168": ["mul_16344", "view_2645", "permute_281", "mm_default_19", "add_tensor_19", "view_2646"], "mul_1041": ["mul_16351"], "view_84": ["view_2647"], "transpose_112": ["permute_282"], "contiguous_112": ["clone_226"], "linear_169": ["mul_16438", "view_2661", "permute_283", "mm_28", "view_2662"], "view_85": ["view_2663"], "transpose_113": ["permute_284"], "contiguous_113": ["clone_227"], "linear_170": ["mul_16539", "view_2677", "permute_285", "mm_default_18", "add_tensor_18", "view_2678"], "view_86": ["view_2679"], "transpose_114": ["permute_286"], "contiguous_114": ["clone_228"], "scaled_dot_product_attention_28": ["_scaled_dot_product_efficient_attention_28", "getitem_226"], "transpose_115": ["permute_287"], "reshape_28": ["view_2680"], "choose_qparams_affine_default_171": ["view_2680", "amax_171", "full_343", "maximum_171", "amin_171", "full_342", "minimum_171", "sub_7827", "div_342", "clamp_min_513", "div_343", "round_343", "sub_7833", "clamp_min_514", "clamp_max_342", "convert_element_type_1026"], "quantize_affine_171": ["view_2680", "view_2683", "reciprocal_171", "mul_16609", "mul_16612", "round_344", "view_2684", "add_26287", "clamp_min_515", "clamp_max_343", "convert_element_type_1027"], "dequantize_affine_342": ["convert_element_type_1027", "view_2688", "convert_element_type_1028", "convert_element_type_1029", "sub_7853", "view_2687", "mul_16634"], "dequantize_affine_343": ["view_2690", "view_2692", "convert_element_type_1030", "convert_element_type_1031", "sub_7857", "view_2691", "mul_16639", "view_2693"], "linear_171": ["mul_16644", "view_2694", "permute_288", "mm_default_17", "add_tensor_17", "view_2695"], "dropout_85": ["view_2695"], "add_401": ["add_26350"], "layer_norm_57": ["clone_230", "var_mean_57", "getitem_230", "getitem_231", "sub_7863", "add_26355", "rsqrt_57", "mul_16655", "mul_16656", "add_26356"], "choose_qparams_affine_default_172": ["add_26356", "amax_172", "full_345", "maximum_172", "amin_172", "full_344", "minimum_172", "sub_7874", "div_344", "clamp_min_516", "div_345", "round_345", "sub_7880", "clamp_min_517", "clamp_max_344", "convert_element_type_1032"], "quantize_affine_172": ["add_26356", "view_2698", "reciprocal_172", "mul_16704", "mul_16707", "round_346", "view_2699", "add_26443", "clamp_min_518", "clamp_max_345", "convert_element_type_1033"], "dequantize_affine_344": ["convert_element_type_1033", "view_2703", "convert_element_type_1034", "convert_element_type_1035", "sub_7900", "view_2702", "mul_16729"], "dequantize_affine_345": ["view_2705", "view_2707", "convert_element_type_1036", "convert_element_type_1037", "sub_7904", "view_2706", "mul_16734", "view_2708"], "linear_172": ["mul_16739", "view_2709", "permute_289", "mm_default_16", "add_tensor_16", "view_2710"], "gelu_30": ["mul_16746", "mul_16747", "erf_30", "add_26502", "mul_16748"], "dropout_86": ["mul_16748"], "choose_qparams_affine_default_173": ["mul_16748", "amax_173", "full_347", "maximum_173", "amin_173", "full_346", "minimum_173", "sub_7917", "div_346", "clamp_min_519", "div_347", "round_347", "sub_7923", "clamp_min_520", "clamp_max_346", "convert_element_type_1038"], "quantize_affine_173": ["mul_16748", "view_2713", "reciprocal_173", "mul_16794", "mul_16797", "round_348", "view_2714", "add_26585", "clamp_min_521", "clamp_max_347", "convert_element_type_1039"], "dequantize_affine_346": ["convert_element_type_1039", "view_2718", "convert_element_type_1040", "convert_element_type_1041", "sub_7943", "view_2717", "mul_16819"], "dequantize_affine_347": ["view_2720", "view_2722", "convert_element_type_1042", "convert_element_type_1043", "sub_7947", "view_2721", "mul_16824", "view_2723"], "linear_173": ["mul_16829", "view_2724", "permute_290", "mm_default_15", "add_tensor_15", "view_2725"], "dropout_87": ["view_2725"], "add_406": ["add_26648"], "layer_norm_58": ["clone_233", "var_mean_58", "getitem_232", "getitem_233", "sub_7953", "add_26653", "rsqrt_58", "mul_16840", "mul_16841", "add_26654"], "choose_qparams_affine_default_174": ["add_26654", "amax_174", "full_349", "maximum_174", "amin_174", "full_348", "minimum_174", "sub_7964", "div_348", "clamp_min_522", "div_349", "round_349", "sub_7970", "clamp_min_523", "clamp_max_348", "convert_element_type_1044"], "quantize_affine_174": ["add_26654", "view_2728", "reciprocal_174", "mul_16889", "mul_16892", "round_350", "view_2729", "add_26741", "clamp_min_524", "clamp_max_349", "convert_element_type_1045"], "choose_qparams_affine_default_175": ["add_26654", "amax_175", "full_351", "maximum_175", "amin_175", "full_350", "minimum_175", "sub_8009", "div_350", "clamp_min_525", "div_351", "round_351", "sub_8015", "clamp_min_526", "clamp_max_350", "convert_element_type_1050"], "quantize_affine_175": ["add_26654", "view_2744", "reciprocal_175", "mul_16985", "mul_16988", "round_352", "view_2745", "add_26893", "clamp_min_527", "clamp_max_351", "convert_element_type_1051"], "choose_qparams_affine_default_176": ["add_26654", "amax_176", "full_353", "maximum_176", "amin_176", "full_352", "minimum_176", "sub_8053", "div_352", "clamp_min_528", "div_353", "round_353", "sub_8059", "clamp_min_529", "clamp_max_352", "convert_element_type_1056"], "quantize_affine_176": ["add_26654", "view_2760", "reciprocal_176", "mul_17084", "mul_17087", "round_354", "view_2761", "add_27041", "clamp_min_530", "clamp_max_353", "convert_element_type_1057"], "dequantize_affine_348": ["convert_element_type_1045", "view_2733", "convert_element_type_1046", "convert_element_type_1047", "sub_7990", "view_2732", "mul_16914"], "dequantize_affine_349": ["view_2735", "view_2737", "convert_element_type_1048", "convert_element_type_1049", "sub_7994", "view_2736", "mul_16919", "view_2738"], "dequantize_affine_350": ["convert_element_type_1051", "view_2749", "convert_element_type_1052", "convert_element_type_1053", "sub_8035", "view_2748", "mul_17010"], "dequantize_affine_351": ["view_2751", "view_2753", "convert_element_type_1054", "convert_element_type_1055", "sub_8039", "view_2752", "mul_17015", "view_2754"], "dequantize_affine_352": ["convert_element_type_1057", "view_2765", "convert_element_type_1058", "convert_element_type_1059", "sub_8079", "view_2764", "mul_17109"], "dequantize_affine_353": ["view_2767", "view_2769", "convert_element_type_1060", "convert_element_type_1061", "sub_8083", "view_2768", "mul_17114", "view_2770"], "linear_174": ["mul_16924", "view_2739", "permute_291", "mm_default_14", "add_tensor_14", "view_2740"], "mul_1078": ["mul_16931"], "view_87": ["view_2741"], "transpose_116": ["permute_292"], "contiguous_116": ["clone_234"], "linear_175": ["mul_17018", "view_2755", "permute_293", "mm_29", "view_2756"], "view_88": ["view_2757"], "transpose_117": ["permute_294"], "contiguous_117": ["clone_235"], "linear_176": ["mul_17119", "view_2771", "permute_295", "mm_default_13", "add_tensor_13", "view_2772"], "view_89": ["view_2773"], "transpose_118": ["permute_296"], "contiguous_118": ["clone_236"], "scaled_dot_product_attention_29": ["_scaled_dot_product_efficient_attention_29", "getitem_234"], "transpose_119": ["permute_297"], "reshape_29": ["view_2774"], "choose_qparams_affine_default_177": ["view_2774", "amax_177", "full_355", "maximum_177", "amin_177", "full_354", "minimum_177", "sub_8101", "div_354", "clamp_min_531", "div_355", "round_355", "sub_8107", "clamp_min_532", "clamp_max_354", "convert_element_type_1062"], "quantize_affine_177": ["view_2774", "view_2777", "reciprocal_177", "mul_17189", "mul_17192", "round_356", "view_2778", "add_27205", "clamp_min_533", "clamp_max_355", "convert_element_type_1063"], "dequantize_affine_354": ["convert_element_type_1063", "view_2782", "convert_element_type_1064", "convert_element_type_1065", "sub_8127", "view_2781", "mul_17214"], "dequantize_affine_355": ["view_2784", "view_2786", "convert_element_type_1066", "convert_element_type_1067", "sub_8131", "view_2785", "mul_17219", "view_2787"], "linear_177": ["mul_17224", "view_2788", "permute_298", "mm_default_12", "add_tensor_12", "view_2789"], "dropout_88": ["view_2789"], "add_415": ["add_27268"], "layer_norm_59": ["clone_238", "var_mean_59", "getitem_238", "getitem_239", "sub_8137", "add_27273", "rsqrt_59", "mul_17235", "mul_17236", "add_27274"], "choose_qparams_affine_default_178": ["add_27274", "amax_178", "full_357", "maximum_178", "amin_178", "full_356", "minimum_178", "sub_8148", "div_356", "clamp_min_534", "div_357", "round_357", "sub_8154", "clamp_min_535", "clamp_max_356", "convert_element_type_1068"], "quantize_affine_178": ["add_27274", "view_2792", "reciprocal_178", "mul_17284", "mul_17287", "round_358", "view_2793", "add_27361", "clamp_min_536", "clamp_max_357", "convert_element_type_1069"], "dequantize_affine_356": ["convert_element_type_1069", "view_2797", "convert_element_type_1070", "convert_element_type_1071", "sub_8174", "view_2796", "mul_17309"], "dequantize_affine_357": ["view_2799", "view_2801", "convert_element_type_1072", "convert_element_type_1073", "sub_8178", "view_2800", "mul_17314", "view_2802"], "linear_178": ["mul_17319", "view_2803", "permute_299", "mm_default_11", "add_tensor_11", "view_2804"], "gelu_31": ["mul_17326", "mul_17327", "erf_31", "add_27420", "mul_17328"], "dropout_89": ["mul_17328"], "choose_qparams_affine_default_179": ["mul_17328", "amax_179", "full_359", "maximum_179", "amin_179", "full_358", "minimum_179", "sub_8191", "div_358", "clamp_min_537", "div_359", "round_359", "sub_8197", "clamp_min_538", "clamp_max_358", "convert_element_type_1074"], "quantize_affine_179": ["mul_17328", "view_2807", "reciprocal_179", "mul_17374", "mul_17377", "round_360", "view_2808", "add_27503", "clamp_min_539", "clamp_max_359", "convert_element_type_1075"], "dequantize_affine_358": ["convert_element_type_1075", "view_2812", "convert_element_type_1076", "convert_element_type_1077", "sub_8217", "view_2811", "mul_17399"], "dequantize_affine_359": ["view_2814", "view_2816", "convert_element_type_1078", "convert_element_type_1079", "sub_8221", "view_2815", "mul_17404", "view_2817"], "linear_179": ["mul_17409", "view_2818", "permute_300", "mm_default_10", "add_tensor_10", "view_2819"], "dropout_90": ["view_2819"], "add_420": ["add_27566"], "layer_norm_60": ["clone_241", "var_mean_60", "getitem_240", "getitem_241", "sub_8227", "add_27571", "rsqrt_60", "mul_17420", "mul_17421", "add_27572"], "choose_qparams_affine_default_180": ["add_27572", "amax_180", "full_361", "maximum_180", "amin_180", "full_360", "minimum_180", "sub_8238", "div_360", "clamp_min_540", "div_361", "round_361", "sub_8244", "clamp_min_541", "clamp_max_360", "convert_element_type_1080"], "quantize_affine_180": ["add_27572", "view_2822", "reciprocal_180", "mul_17469", "mul_17472", "round_362", "view_2823", "add_27659", "clamp_min_542", "clamp_max_361", "convert_element_type_1081"], "choose_qparams_affine_default_181": ["add_27572", "amax_181", "full_363", "maximum_181", "amin_181", "full_362", "minimum_181", "sub_8283", "div_362", "clamp_min_543", "div_363", "round_363", "sub_8289", "clamp_min_544", "clamp_max_362", "convert_element_type_1086"], "quantize_affine_181": ["add_27572", "view_2838", "reciprocal_181", "mul_17565", "mul_17568", "round_364", "view_2839", "add_27811", "clamp_min_545", "clamp_max_363", "convert_element_type_1087"], "choose_qparams_affine_default_182": ["add_27572", "amax_182", "full_365", "maximum_182", "amin_182", "full_364", "minimum_182", "sub_8327", "div_364", "clamp_min_546", "div_365", "round_365", "sub_8333", "clamp_min_547", "clamp_max_364", "convert_element_type_1092"], "quantize_affine_182": ["add_27572", "view_2854", "reciprocal_182", "mul_17664", "mul_17667", "round_366", "view_2855", "add_27959", "clamp_min_548", "clamp_max_365", "convert_element_type_1093"], "dequantize_affine_360": ["convert_element_type_1081", "view_2827", "convert_element_type_1082", "convert_element_type_1083", "sub_8264", "view_2826", "mul_17494"], "dequantize_affine_361": ["view_2829", "view_2831", "convert_element_type_1084", "convert_element_type_1085", "sub_8268", "view_2830", "mul_17499", "view_2832"], "dequantize_affine_362": ["convert_element_type_1087", "view_2843", "convert_element_type_1088", "convert_element_type_1089", "sub_8309", "view_2842", "mul_17590"], "dequantize_affine_363": ["view_2845", "view_2847", "convert_element_type_1090", "convert_element_type_1091", "sub_8313", "view_2846", "mul_17595", "view_2848"], "dequantize_affine_364": ["convert_element_type_1093", "view_2859", "convert_element_type_1094", "convert_element_type_1095", "sub_8353", "view_2858", "mul_17689"], "dequantize_affine_365": ["view_2861", "view_2863", "convert_element_type_1096", "convert_element_type_1097", "sub_8357", "view_2862", "mul_17694", "view_2864"], "linear_180": ["mul_17504", "view_2833", "permute_301", "mm_default_9", "add_tensor_9", "view_2834"], "mul_1115": ["mul_17511"], "view_90": ["view_2835"], "transpose_120": ["permute_302"], "contiguous_120": ["clone_242"], "linear_181": ["mul_17598", "view_2849", "permute_303", "mm_30", "view_2850"], "view_91": ["view_2851"], "transpose_121": ["permute_304"], "contiguous_121": ["clone_243"], "linear_182": ["mul_17699", "view_2865", "permute_305", "mm_default_8", "add_tensor_8", "view_2866"], "view_92": ["view_2867"], "transpose_122": ["permute_306"], "contiguous_122": ["clone_244"], "scaled_dot_product_attention_30": ["_scaled_dot_product_efficient_attention_30", "getitem_242"], "transpose_123": ["permute_307"], "reshape_30": ["view_2868"], "choose_qparams_affine_default_183": ["view_2868", "amax_183", "full_367", "maximum_183", "amin_183", "full_366", "minimum_183", "sub_8375", "div_366", "clamp_min_549", "div_367", "round_367", "sub_8381", "clamp_min_550", "clamp_max_366", "convert_element_type_1098"], "quantize_affine_183": ["view_2868", "view_2871", "reciprocal_183", "mul_17769", "mul_17772", "round_368", "view_2872", "add_28123", "clamp_min_551", "clamp_max_367", "convert_element_type_1099"], "dequantize_affine_366": ["convert_element_type_1099", "view_2876", "convert_element_type_1100", "convert_element_type_1101", "sub_8401", "view_2875", "mul_17794"], "dequantize_affine_367": ["view_2878", "view_2880", "convert_element_type_1102", "convert_element_type_1103", "sub_8405", "view_2879", "mul_17799", "view_2881"], "linear_183": ["mul_17804", "view_2882", "permute_308", "mm_default_7", "add_tensor_7", "view_2883"], "dropout_91": ["view_2883"], "add_429": ["add_28186"], "layer_norm_61": ["clone_246", "var_mean_61", "getitem_246", "getitem_247", "sub_8411", "add_28191", "rsqrt_61", "mul_17815", "mul_17816", "add_28192"], "choose_qparams_affine_default_184": ["add_28192", "amax_184", "full_369", "maximum_184", "amin_184", "full_368", "minimum_184", "sub_8422", "div_368", "clamp_min_552", "div_369", "round_369", "sub_8428", "clamp_min_553", "clamp_max_368", "convert_element_type_1104"], "quantize_affine_184": ["add_28192", "view_2886", "reciprocal_184", "mul_17864", "mul_17867", "round_370", "view_2887", "add_28279", "clamp_min_554", "clamp_max_369", "convert_element_type_1105"], "dequantize_affine_368": ["convert_element_type_1105", "view_2891", "convert_element_type_1106", "convert_element_type_1107", "sub_8448", "view_2890", "mul_17889"], "dequantize_affine_369": ["view_2893", "view_2895", "convert_element_type_1108", "convert_element_type_1109", "sub_8452", "view_2894", "mul_17894", "view_2896"], "linear_184": ["mul_17899", "view_2897", "permute_309", "mm_default_6", "add_tensor_6", "view_2898"], "gelu_32": ["mul_17906", "mul_17907", "erf_32", "add_28338", "mul_17908"], "dropout_92": ["mul_17908"], "choose_qparams_affine_default_185": ["mul_17908", "amax_185", "full_371", "maximum_185", "amin_185", "full_370", "minimum_185", "sub_8465", "div_370", "clamp_min_555", "div_371", "round_371", "sub_8471", "clamp_min_556", "clamp_max_370", "convert_element_type_1110"], "quantize_affine_185": ["mul_17908", "view_2901", "reciprocal_185", "mul_17954", "mul_17957", "round_372", "view_2902", "add_28421", "clamp_min_557", "clamp_max_371", "convert_element_type_1111"], "dequantize_affine_370": ["convert_element_type_1111", "view_2906", "convert_element_type_1112", "convert_element_type_1113", "sub_8491", "view_2905", "mul_17979"], "dequantize_affine_371": ["view_2908", "view_2910", "convert_element_type_1114", "convert_element_type_1115", "sub_8495", "view_2909", "mul_17984", "view_2911"], "linear_185": ["mul_17989", "view_2912", "permute_310", "mm_default_5", "add_tensor_5", "view_2913"], "dropout_93": ["view_2913"], "add_434": ["add_28484"], "layer_norm_62": ["clone_249", "var_mean_62", "getitem_248", "getitem_249", "sub_8501", "add_28489", "rsqrt_62", "mul_18000", "mul_18001", "add_28490"], "choose_qparams_affine_default_186": ["add_28490", "amax_186", "full_373", "maximum_186", "amin_186", "full_372", "minimum_186", "sub_8512", "div_372", "clamp_min_558", "div_373", "round_373", "sub_8518", "clamp_min_559", "clamp_max_372", "convert_element_type_1116"], "quantize_affine_186": ["add_28490", "view_2916", "reciprocal_186", "mul_18049", "mul_18052", "round_374", "view_2917", "add_28577", "clamp_min_560", "clamp_max_373", "convert_element_type_1117"], "choose_qparams_affine_default_187": ["add_28490", "amax_187", "full_375", "maximum_187", "amin_187", "full_374", "minimum_187", "sub_8557", "div_374", "clamp_min_561", "div_375", "round_375", "sub_8563", "clamp_min_562", "clamp_max_374", "convert_element_type_1122"], "quantize_affine_187": ["add_28490", "view_2932", "reciprocal_187", "mul_18145", "mul_18148", "round_376", "view_2933", "add_28729", "clamp_min_563", "clamp_max_375", "convert_element_type_1123"], "choose_qparams_affine_default_188": ["add_28490", "amax_188", "full_377", "maximum_188", "amin_188", "full_376", "minimum_188", "sub_8601", "div_376", "clamp_min_564", "div_377", "round_377", "sub_8607", "clamp_min_565", "clamp_max_376", "convert_element_type_1128"], "quantize_affine_188": ["add_28490", "view_2948", "reciprocal_188", "mul_18244", "mul_18247", "round_378", "view_2949", "add_28877", "clamp_min_566", "clamp_max_377", "convert_element_type_1129"], "dequantize_affine_372": ["convert_element_type_1117", "view_2921", "convert_element_type_1118", "convert_element_type_1119", "sub_8538", "view_2920", "mul_18074"], "dequantize_affine_373": ["view_2923", "view_2925", "convert_element_type_1120", "convert_element_type_1121", "sub_8542", "view_2924", "mul_18079", "view_2926"], "dequantize_affine_374": ["convert_element_type_1123", "view_2937", "convert_element_type_1124", "convert_element_type_1125", "sub_8583", "view_2936", "mul_18170"], "dequantize_affine_375": ["view_2939", "view_2941", "convert_element_type_1126", "convert_element_type_1127", "sub_8587", "view_2940", "mul_18175", "view_2942"], "dequantize_affine_376": ["convert_element_type_1129", "view_2953", "convert_element_type_1130", "convert_element_type_1131", "sub_8627", "view_2952", "mul_18269"], "dequantize_affine_377": ["view_2955", "view_2957", "convert_element_type_1132", "convert_element_type_1133", "sub_8631", "view_2956", "mul_18274", "view_2958"], "linear_186": ["mul_18084", "view_2927", "permute_311", "mm_default_4", "add_tensor_4", "view_2928"], "mul_1152": ["mul_18091"], "view_93": ["view_2929"], "transpose_124": ["permute_312"], "contiguous_124": ["clone_250"], "linear_187": ["mul_18178", "view_2943", "permute_313", "mm_31", "view_2944"], "view_94": ["view_2945"], "transpose_125": ["permute_314"], "contiguous_125": ["clone_251"], "linear_188": ["mul_18279", "view_2959", "permute_315", "mm_default_3", "add_tensor_3", "view_2960"], "view_95": ["view_2961"], "transpose_126": ["permute_316"], "contiguous_126": ["clone_252"], "scaled_dot_product_attention_31": ["_scaled_dot_product_efficient_attention_31", "getitem_250"], "transpose_127": ["permute_317"], "reshape_31": ["view_2962"], "choose_qparams_affine_default_189": ["view_2962", "amax_189", "full_379", "maximum_189", "amin_189", "full_378", "minimum_189", "sub_8649", "div_378", "clamp_min_567", "div_379", "round_379", "sub_8655", "clamp_min_568", "clamp_max_378", "convert_element_type_1134"], "quantize_affine_189": ["view_2962", "view_2965", "reciprocal_189", "mul_18349", "mul_18352", "round_380", "view_2966", "add_29041", "clamp_min_569", "clamp_max_379", "convert_element_type_1135"], "dequantize_affine_378": ["convert_element_type_1135", "view_2970", "convert_element_type_1136", "convert_element_type_1137", "sub_8675", "view_2969", "mul_18374"], "dequantize_affine_379": ["view_2972", "view_2974", "convert_element_type_1138", "convert_element_type_1139", "sub_8679", "view_2973", "mul_18379", "view_2975"], "linear_189": ["mul_18384", "view_2976", "permute_318", "mm_default_2", "add_tensor_2", "view_2977"], "dropout_94": ["view_2977"], "add_443": ["add_29104"], "layer_norm_63": ["clone_254", "var_mean_63", "getitem_254", "getitem_255", "sub_8685", "add_29109", "rsqrt_63", "mul_18395", "mul_18396", "add_29110"], "choose_qparams_affine_default_190": ["add_29110", "amax_190", "full_381", "maximum_190", "amin_190", "full_380", "minimum_190", "sub_8696", "div_380", "clamp_min_570", "div_381", "round_381", "sub_8702", "clamp_min_571", "clamp_max_380", "convert_element_type_1140"], "quantize_affine_190": ["add_29110", "view_2980", "reciprocal_190", "mul_18444", "mul_18447", "round_382", "view_2981", "add_29197", "clamp_min_572", "clamp_max_381", "convert_element_type_1141"], "dequantize_affine_380": ["convert_element_type_1141", "view_2985", "convert_element_type_1142", "convert_element_type_1143", "sub_8722", "view_2984", "mul_18469"], "dequantize_affine_381": ["view_2987", "view_2989", "convert_element_type_1144", "convert_element_type_1145", "sub_8726", "view_2988", "mul_18474", "view_2990"], "linear_190": ["mul_18479", "view_2991", "permute_319", "mm_default_1", "add_tensor_1", "view_2992"], "gelu_33": ["mul_18486", "mul_18487", "erf_33", "add_29256", "mul_18488"], "dropout_95": ["mul_18488"], "choose_qparams_affine_default_191": ["mul_18488", "amax_191", "full_383", "maximum_191", "amin_191", "full_382", "minimum_191", "sub_8739", "div_382", "clamp_min_573", "div_383", "round_383", "sub_8745", "clamp_min_574", "clamp_max_382", "convert_element_type_1146"], "quantize_affine_191": ["mul_18488", "view_2995", "reciprocal_191", "mul_18534", "mul_18537", "round_384", "view_2996", "add_29339", "clamp_min_575", "clamp_max_383", "convert_element_type_1147"], "dequantize_affine_382": ["convert_element_type_1147", "view_3000", "convert_element_type_1148", "convert_element_type_1149", "sub_8765", "view_2999", "mul_18559"], "dequantize_affine_383": ["view_3002", "view_3004", "convert_element_type_1150", "convert_element_type_1151", "sub_8769", "view_3003", "mul_18564", "view_3005"], "linear_191": ["mul_18569", "view_3006", "permute_320", "mm_default", "add_tensor", "view_3007"], "dropout_96": ["view_3007"], "add_448": ["add_29402"], "layer_norm_64": ["clone_257", "var_mean_64", "getitem_256", "getitem_257", "sub_8775", "add_29407", "rsqrt_64", "mul_18580", "mul_18581", "add_29408"], "reshape_32": ["view_3008", "sym_size_int_193"], "choose_qparams_affine_default_192": ["view_3008", "amax_192", "full_385", "maximum_192", "amin_192", "full_384", "minimum_192", "sub_8787", "div_384", "clamp_min_576", "div_385", "round_385", "sub_8793", "clamp_min_577", "clamp_max_384", "convert_element_type_1152"], "quantize_affine_192": ["view_3008", "view_3011", "reciprocal_192", "mul_18613", "mul_18615", "round_386", "view_3012", "add_29476", "clamp_min_578", "clamp_max_385", "convert_element_type_1153"], "dequantize_affine_384": ["convert_element_type_1153", "view_3016", "convert_element_type_1154", "convert_element_type_1155", "sub_8813", "view_3015", "mul_18634"], "dequantize_affine_385": ["view_3018", "view_3020", "convert_element_type_1156", "convert_element_type_1157", "sub_8817", "view_3019", "mul_18639", "view_3021"], "linear_192": ["permute_321", "mm_32"], "gelu_34": ["mul_18642", "mul_18643", "erf_34", "add_29516", "mul_18644"], "choose_qparams_affine_default_193": ["mul_18644", "amax_193", "full_387", "maximum_193", "amin_193", "full_386", "minimum_193", "sub_8827", "div_386", "clamp_min_579", "div_387", "round_387", "sub_8833", "clamp_min_580", "clamp_max_386", "convert_element_type_1158"], "quantize_affine_193": ["mul_18644", "view_3024", "reciprocal_193", "mul_18666", "mul_18668", "round_388", "view_3025", "add_29572", "clamp_min_581", "clamp_max_387", "convert_element_type_1159"], "dequantize_affine_386": ["convert_element_type_1159", "view_3029", "convert_element_type_1160", "convert_element_type_1161", "sub_8853", "view_3028", "mul_18687"], "dequantize_affine_387": ["view_3031", "view_3033", "convert_element_type_1162", "convert_element_type_1163", "sub_8857", "view_3032", "mul_18692", "view_3034"], "linear_193": ["permute_322", "mm_33"], "unsqueeze": ["unsqueeze"]}, "postToPre": {"convolution": ["conv1d"], "mul_2": ["gelu"], "mul_3": ["gelu"], "erf": ["gelu"], "add_4": ["gelu"], "mul_4": ["gelu"], "convolution_1": ["conv1d_1"], "mul_9": ["gelu_1"], "mul_10": ["gelu_1"], "erf_1": ["gelu_1"], "add_13": ["gelu_1"], "mul_11": ["gelu_1"], "permute": ["permute"], "add_22": ["add", "dropout"], "clone_1": ["layer_norm"], "var_mean": ["layer_norm"], "getitem": ["layer_norm"], "getitem_1": ["layer_norm"], "sub_7": ["layer_norm"], "add_31": ["layer_norm"], "rsqrt": ["layer_norm"], "mul_20": ["layer_norm"], "mul_21": ["layer_norm"], "add_32": ["layer_norm", "choose_qparams_affine_default", "quantize_affine", "choose_qparams_affine_default_1", "quantize_affine_1", "choose_qparams_affine_default_2", "quantize_affine_2"], "amax": ["choose_qparams_affine_default"], "sym_size_int": ["conv1d"], "full_1": ["choose_qparams_affine_default"], "maximum": ["choose_qparams_affine_default"], "amin": ["choose_qparams_affine_default"], "full": ["choose_qparams_affine_default"], "minimum": ["choose_qparams_affine_default"], "sub_18": ["choose_qparams_affine_default"], "div": ["choose_qparams_affine_default"], "clamp_min": ["choose_qparams_affine_default"], "div_1": ["choose_qparams_affine_default"], "round_1": ["choose_qparams_affine_default"], "sub_24": ["choose_qparams_affine_default"], "clamp_min_1": ["choose_qparams_affine_default"], "clamp_max": ["choose_qparams_affine_default"], "view_2": ["quantize_affine"], "reciprocal": ["quantize_affine"], "mul_69": ["quantize_affine"], "mul_72": ["quantize_affine"], "round_2": ["quantize_affine"], "convert_element_type": ["choose_qparams_affine_default"], "view_3": ["quantize_affine"], "add_119": ["quantize_affine"], "clamp_min_2": ["quantize_affine"], "clamp_max_1": ["quantize_affine"], "convert_element_type_1": ["quantize_affine", "dequantize_affine"], "view_7": ["dequantize_affine"], "convert_element_type_2": ["dequantize_affine"], "convert_element_type_3": ["dequantize_affine"], "sub_44": ["dequantize_affine"], "view_6": ["dequantize_affine"], "mul_94": ["dequantize_affine"], "view_9": ["dequantize_affine_1"], "view_11": ["dequantize_affine_1"], "convert_element_type_4": ["dequantize_affine_1"], "convert_element_type_5": ["dequantize_affine_1"], "sub_48": ["dequantize_affine_1"], "view_10": ["dequantize_affine_1"], "mul_99": ["dequantize_affine_1"], "view_12": ["dequantize_affine_1"], "amax_1": ["choose_qparams_affine_default_1"], "full_3": ["choose_qparams_affine_default_1"], "maximum_1": ["choose_qparams_affine_default_1"], "amin_1": ["choose_qparams_affine_default_1"], "full_2": ["choose_qparams_affine_default_1"], "minimum_1": ["choose_qparams_affine_default_1"], "sub_63": ["choose_qparams_affine_default_1"], "div_2": ["choose_qparams_affine_default_1"], "clamp_min_3": ["choose_qparams_affine_default_1"], "div_3": ["choose_qparams_affine_default_1"], "round_3": ["choose_qparams_affine_default_1"], "sub_69": ["choose_qparams_affine_default_1"], "clamp_min_4": ["choose_qparams_affine_default_1"], "clamp_max_2": ["choose_qparams_affine_default_1"], "view_18": ["quantize_affine_1"], "reciprocal_1": ["quantize_affine_1"], "mul_165": ["quantize_affine_1"], "mul_168": ["quantize_affine_1"], "round_4": ["quantize_affine_1"], "convert_element_type_6": ["choose_qparams_affine_default_1"], "view_19": ["quantize_affine_1"], "add_271": ["quantize_affine_1"], "clamp_min_5": ["quantize_affine_1"], "clamp_max_3": ["quantize_affine_1"], "convert_element_type_7": ["quantize_affine_1", "dequantize_affine_2"], "view_23": ["dequantize_affine_2"], "convert_element_type_8": ["dequantize_affine_2"], "convert_element_type_9": ["dequantize_affine_2"], "sub_89": ["dequantize_affine_2"], "view_22": ["dequantize_affine_2"], "mul_190": ["dequantize_affine_2"], "view_25": ["dequantize_affine_3"], "view_27": ["dequantize_affine_3"], "convert_element_type_10": ["dequantize_affine_3"], "convert_element_type_11": ["dequantize_affine_3"], "sub_93": ["dequantize_affine_3"], "view_26": ["dequantize_affine_3"], "mul_195": ["dequantize_affine_3"], "view_28": ["dequantize_affine_3"], "amax_2": ["choose_qparams_affine_default_2"], "full_5": ["choose_qparams_affine_default_2"], "maximum_2": ["choose_qparams_affine_default_2"], "amin_2": ["choose_qparams_affine_default_2"], "full_4": ["choose_qparams_affine_default_2"], "minimum_2": ["choose_qparams_affine_default_2"], "sub_107": ["choose_qparams_affine_default_2"], "div_4": ["choose_qparams_affine_default_2"], "clamp_min_6": ["choose_qparams_affine_default_2"], "div_5": ["choose_qparams_affine_default_2"], "round_5": ["choose_qparams_affine_default_2"], "sub_113": ["choose_qparams_affine_default_2"], "clamp_min_7": ["choose_qparams_affine_default_2"], "clamp_max_4": ["choose_qparams_affine_default_2"], "view_34": ["quantize_affine_2"], "reciprocal_2": ["quantize_affine_2"], "mul_264": ["quantize_affine_2"], "mul_267": ["quantize_affine_2"], "round_6": ["quantize_affine_2"], "convert_element_type_12": ["choose_qparams_affine_default_2"], "view_35": ["quantize_affine_2"], "add_419": ["quantize_affine_2"], "clamp_min_8": ["quantize_affine_2"], "clamp_max_5": ["quantize_affine_2"], "convert_element_type_13": ["quantize_affine_2", "dequantize_affine_4"], "view_39": ["dequantize_affine_4"], "convert_element_type_14": ["dequantize_affine_4"], "convert_element_type_15": ["dequantize_affine_4"], "sub_133": ["dequantize_affine_4"], "view_38": ["dequantize_affine_4"], "mul_289": ["dequantize_affine_4"], "view_41": ["dequantize_affine_5"], "view_43": ["dequantize_affine_5"], "convert_element_type_16": ["dequantize_affine_5"], "convert_element_type_17": ["dequantize_affine_5"], "sub_137": ["dequantize_affine_5"], "view_42": ["dequantize_affine_5"], "mul_294": ["dequantize_affine_5"], "view_44": ["dequantize_affine_5"], "mul_104": ["linear"], "view_13": ["linear"], "permute_1": ["linear"], "mm_default_159": ["linear"], "add_tensor_159": ["linear"], "view_14": ["linear"], "mul_111": ["mul_5"], "view_15": ["view"], "permute_2": ["transpose"], "clone_2": ["contiguous"], "mul_198": ["linear_1"], "view_29": ["linear_1"], "permute_3": ["linear_1"], "mm": ["linear_1"], "view_30": ["linear_1"], "view_31": ["view_1"], "permute_4": ["transpose_1"], "clone_3": ["contiguous_1"], "mul_299": ["linear_2"], "view_45": ["linear_2"], "permute_5": ["linear_2"], "mm_default_158": ["linear_2"], "add_tensor_158": ["linear_2"], "view_46": ["linear_2"], "view_47": ["view_2"], "permute_6": ["transpose_2"], "clone_4": ["contiguous_2"], "_scaled_dot_product_efficient_attention": ["scaled_dot_product_attention"], "getitem_2": ["scaled_dot_product_attention"], "permute_7": ["transpose_3"], "view_48": ["reshape", "choose_qparams_affine_default_3", "quantize_affine_3"], "amax_3": ["choose_qparams_affine_default_3"], "full_7": ["choose_qparams_affine_default_3"], "maximum_3": ["choose_qparams_affine_default_3"], "amin_3": ["choose_qparams_affine_default_3"], "full_6": ["choose_qparams_affine_default_3"], "minimum_3": ["choose_qparams_affine_default_3"], "sub_155": ["choose_qparams_affine_default_3"], "div_6": ["choose_qparams_affine_default_3"], "clamp_min_9": ["choose_qparams_affine_default_3"], "div_7": ["choose_qparams_affine_default_3"], "round_7": ["choose_qparams_affine_default_3"], "sub_161": ["choose_qparams_affine_default_3"], "clamp_min_10": ["choose_qparams_affine_default_3"], "clamp_max_6": ["choose_qparams_affine_default_3"], "view_51": ["quantize_affine_3"], "reciprocal_3": ["quantize_affine_3"], "mul_369": ["quantize_affine_3"], "mul_372": ["quantize_affine_3"], "round_8": ["quantize_affine_3"], "convert_element_type_18": ["choose_qparams_affine_default_3"], "view_52": ["quantize_affine_3"], "add_583": ["quantize_affine_3"], "clamp_min_11": ["quantize_affine_3"], "clamp_max_7": ["quantize_affine_3"], "convert_element_type_19": ["quantize_affine_3", "dequantize_affine_6"], "view_56": ["dequantize_affine_6"], "convert_element_type_20": ["dequantize_affine_6"], "convert_element_type_21": ["dequantize_affine_6"], "sub_181": ["dequantize_affine_6"], "view_55": ["dequantize_affine_6"], "mul_394": ["dequantize_affine_6"], "view_58": ["dequantize_affine_7"], "view_60": ["dequantize_affine_7"], "convert_element_type_22": ["dequantize_affine_7"], "convert_element_type_23": ["dequantize_affine_7"], "sub_185": ["dequantize_affine_7"], "view_59": ["dequantize_affine_7"], "mul_399": ["dequantize_affine_7"], "view_61": ["dequantize_affine_7"], "mul_404": ["linear_3"], "view_62": ["linear_3"], "permute_8": ["linear_3"], "mm_default_157": ["linear_3"], "add_tensor_157": ["linear_3"], "view_63": ["linear_3", "dropout_1"], "add_646": ["add_9"], "clone_6": ["layer_norm_1"], "var_mean_1": ["layer_norm_1"], "getitem_6": ["layer_norm_1"], "getitem_7": ["layer_norm_1"], "sub_191": ["layer_norm_1"], "add_651": ["layer_norm_1"], "rsqrt_1": ["layer_norm_1"], "mul_415": ["layer_norm_1"], "mul_416": ["layer_norm_1"], "add_652": ["layer_norm_1", "choose_qparams_affine_default_4", "quantize_affine_4"], "amax_4": ["choose_qparams_affine_default_4"], "full_9": ["choose_qparams_affine_default_4"], "maximum_4": ["choose_qparams_affine_default_4"], "amin_4": ["choose_qparams_affine_default_4"], "full_8": ["choose_qparams_affine_default_4"], "minimum_4": ["choose_qparams_affine_default_4"], "sub_202": ["choose_qparams_affine_default_4"], "div_8": ["choose_qparams_affine_default_4"], "clamp_min_12": ["choose_qparams_affine_default_4"], "div_9": ["choose_qparams_affine_default_4"], "round_9": ["choose_qparams_affine_default_4"], "sub_208": ["choose_qparams_affine_default_4"], "clamp_min_13": ["choose_qparams_affine_default_4"], "clamp_max_8": ["choose_qparams_affine_default_4"], "view_66": ["quantize_affine_4"], "reciprocal_4": ["quantize_affine_4"], "mul_464": ["quantize_affine_4"], "mul_467": ["quantize_affine_4"], "round_10": ["quantize_affine_4"], "convert_element_type_24": ["choose_qparams_affine_default_4"], "view_67": ["quantize_affine_4"], "add_739": ["quantize_affine_4"], "clamp_min_14": ["quantize_affine_4"], "clamp_max_9": ["quantize_affine_4"], "convert_element_type_25": ["quantize_affine_4", "dequantize_affine_8"], "view_71": ["dequantize_affine_8"], "convert_element_type_26": ["dequantize_affine_8"], "convert_element_type_27": ["dequantize_affine_8"], "sub_228": ["dequantize_affine_8"], "view_70": ["dequantize_affine_8"], "mul_489": ["dequantize_affine_8"], "view_73": ["dequantize_affine_9"], "view_75": ["dequantize_affine_9"], "convert_element_type_28": ["dequantize_affine_9"], "convert_element_type_29": ["dequantize_affine_9"], "sub_232": ["dequantize_affine_9"], "view_74": ["dequantize_affine_9"], "mul_494": ["dequantize_affine_9"], "view_76": ["dequantize_affine_9"], "mul_499": ["linear_4"], "view_77": ["linear_4"], "permute_9": ["linear_4"], "mm_default_156": ["linear_4"], "add_tensor_156": ["linear_4"], "view_78": ["linear_4"], "mul_506": ["gelu_2"], "mul_507": ["gelu_2"], "erf_2": ["gelu_2"], "add_798": ["gelu_2"], "mul_508": ["gelu_2", "dropout_2", "choose_qparams_affine_default_5", "quantize_affine_5"], "amax_5": ["choose_qparams_affine_default_5"], "full_11": ["choose_qparams_affine_default_5"], "maximum_5": ["choose_qparams_affine_default_5"], "amin_5": ["choose_qparams_affine_default_5"], "full_10": ["choose_qparams_affine_default_5"], "minimum_5": ["choose_qparams_affine_default_5"], "sub_245": ["choose_qparams_affine_default_5"], "div_10": ["choose_qparams_affine_default_5"], "clamp_min_15": ["choose_qparams_affine_default_5"], "div_11": ["choose_qparams_affine_default_5"], "round_11": ["choose_qparams_affine_default_5"], "sub_251": ["choose_qparams_affine_default_5"], "clamp_min_16": ["choose_qparams_affine_default_5"], "clamp_max_10": ["choose_qparams_affine_default_5"], "view_81": ["quantize_affine_5"], "reciprocal_5": ["quantize_affine_5"], "mul_554": ["quantize_affine_5"], "mul_557": ["quantize_affine_5"], "round_12": ["quantize_affine_5"], "convert_element_type_30": ["choose_qparams_affine_default_5"], "view_82": ["quantize_affine_5"], "add_881": ["quantize_affine_5"], "clamp_min_17": ["quantize_affine_5"], "clamp_max_11": ["quantize_affine_5"], "convert_element_type_31": ["quantize_affine_5", "dequantize_affine_10"], "view_86": ["dequantize_affine_10"], "convert_element_type_32": ["dequantize_affine_10"], "convert_element_type_33": ["dequantize_affine_10"], "sub_271": ["dequantize_affine_10"], "view_85": ["dequantize_affine_10"], "mul_579": ["dequantize_affine_10"], "view_88": ["dequantize_affine_11"], "view_90": ["dequantize_affine_11"], "convert_element_type_34": ["dequantize_affine_11"], "convert_element_type_35": ["dequantize_affine_11"], "sub_275": ["dequantize_affine_11"], "view_89": ["dequantize_affine_11"], "mul_584": ["dequantize_affine_11"], "view_91": ["dequantize_affine_11"], "mul_589": ["linear_5"], "view_92": ["linear_5"], "permute_10": ["linear_5"], "mm_default_155": ["linear_5"], "add_tensor_155": ["linear_5"], "view_93": ["linear_5", "dropout_3"], "add_944": ["add_14"], "clone_9": ["layer_norm_2"], "var_mean_2": ["layer_norm_2"], "getitem_8": ["layer_norm_2"], "getitem_9": ["layer_norm_2"], "sub_281": ["layer_norm_2"], "add_949": ["layer_norm_2"], "rsqrt_2": ["layer_norm_2"], "mul_600": ["layer_norm_2"], "mul_601": ["layer_norm_2"], "add_950": ["layer_norm_2", "choose_qparams_affine_default_6", "quantize_affine_6", "choose_qparams_affine_default_7", "quantize_affine_7", "choose_qparams_affine_default_8", "quantize_affine_8"], "amax_6": ["choose_qparams_affine_default_6"], "full_13": ["choose_qparams_affine_default_6"], "maximum_6": ["choose_qparams_affine_default_6"], "amin_6": ["choose_qparams_affine_default_6"], "full_12": ["choose_qparams_affine_default_6"], "minimum_6": ["choose_qparams_affine_default_6"], "sub_292": ["choose_qparams_affine_default_6"], "div_12": ["choose_qparams_affine_default_6"], "clamp_min_18": ["choose_qparams_affine_default_6"], "div_13": ["choose_qparams_affine_default_6"], "round_13": ["choose_qparams_affine_default_6"], "sub_298": ["choose_qparams_affine_default_6"], "clamp_min_19": ["choose_qparams_affine_default_6"], "clamp_max_12": ["choose_qparams_affine_default_6"], "view_96": ["quantize_affine_6"], "reciprocal_6": ["quantize_affine_6"], "mul_649": ["quantize_affine_6"], "mul_652": ["quantize_affine_6"], "round_14": ["quantize_affine_6"], "convert_element_type_36": ["choose_qparams_affine_default_6"], "view_97": ["quantize_affine_6"], "add_1037": ["quantize_affine_6"], "clamp_min_20": ["quantize_affine_6"], "clamp_max_13": ["quantize_affine_6"], "convert_element_type_37": ["quantize_affine_6", "dequantize_affine_12"], "view_101": ["dequantize_affine_12"], "convert_element_type_38": ["dequantize_affine_12"], "convert_element_type_39": ["dequantize_affine_12"], "sub_318": ["dequantize_affine_12"], "view_100": ["dequantize_affine_12"], "mul_674": ["dequantize_affine_12"], "view_103": ["dequantize_affine_13"], "view_105": ["dequantize_affine_13"], "convert_element_type_40": ["dequantize_affine_13"], "convert_element_type_41": ["dequantize_affine_13"], "sub_322": ["dequantize_affine_13"], "view_104": ["dequantize_affine_13"], "mul_679": ["dequantize_affine_13"], "view_106": ["dequantize_affine_13"], "amax_7": ["choose_qparams_affine_default_7"], "full_15": ["choose_qparams_affine_default_7"], "maximum_7": ["choose_qparams_affine_default_7"], "amin_7": ["choose_qparams_affine_default_7"], "full_14": ["choose_qparams_affine_default_7"], "minimum_7": ["choose_qparams_affine_default_7"], "sub_337": ["choose_qparams_affine_default_7"], "div_14": ["choose_qparams_affine_default_7"], "clamp_min_21": ["choose_qparams_affine_default_7"], "div_15": ["choose_qparams_affine_default_7"], "round_15": ["choose_qparams_affine_default_7"], "sub_343": ["choose_qparams_affine_default_7"], "clamp_min_22": ["choose_qparams_affine_default_7"], "clamp_max_14": ["choose_qparams_affine_default_7"], "view_112": ["quantize_affine_7"], "reciprocal_7": ["quantize_affine_7"], "mul_745": ["quantize_affine_7"], "mul_748": ["quantize_affine_7"], "round_16": ["quantize_affine_7"], "convert_element_type_42": ["choose_qparams_affine_default_7"], "view_113": ["quantize_affine_7"], "add_1189": ["quantize_affine_7"], "clamp_min_23": ["quantize_affine_7"], "clamp_max_15": ["quantize_affine_7"], "convert_element_type_43": ["quantize_affine_7", "dequantize_affine_14"], "view_117": ["dequantize_affine_14"], "convert_element_type_44": ["dequantize_affine_14"], "convert_element_type_45": ["dequantize_affine_14"], "sub_363": ["dequantize_affine_14"], "view_116": ["dequantize_affine_14"], "mul_770": ["dequantize_affine_14"], "view_119": ["dequantize_affine_15"], "view_121": ["dequantize_affine_15"], "convert_element_type_46": ["dequantize_affine_15"], "convert_element_type_47": ["dequantize_affine_15"], "sub_367": ["dequantize_affine_15"], "view_120": ["dequantize_affine_15"], "mul_775": ["dequantize_affine_15"], "view_122": ["dequantize_affine_15"], "amax_8": ["choose_qparams_affine_default_8"], "full_17": ["choose_qparams_affine_default_8"], "maximum_8": ["choose_qparams_affine_default_8"], "amin_8": ["choose_qparams_affine_default_8"], "full_16": ["choose_qparams_affine_default_8"], "minimum_8": ["choose_qparams_affine_default_8"], "sub_381": ["choose_qparams_affine_default_8"], "div_16": ["choose_qparams_affine_default_8"], "clamp_min_24": ["choose_qparams_affine_default_8"], "div_17": ["choose_qparams_affine_default_8"], "round_17": ["choose_qparams_affine_default_8"], "sub_387": ["choose_qparams_affine_default_8"], "clamp_min_25": ["choose_qparams_affine_default_8"], "clamp_max_16": ["choose_qparams_affine_default_8"], "view_128": ["quantize_affine_8"], "reciprocal_8": ["quantize_affine_8"], "mul_844": ["quantize_affine_8"], "mul_847": ["quantize_affine_8"], "round_18": ["quantize_affine_8"], "convert_element_type_48": ["choose_qparams_affine_default_8"], "view_129": ["quantize_affine_8"], "add_1337": ["quantize_affine_8"], "clamp_min_26": ["quantize_affine_8"], "clamp_max_17": ["quantize_affine_8"], "convert_element_type_49": ["quantize_affine_8", "dequantize_affine_16"], "view_133": ["dequantize_affine_16"], "convert_element_type_50": ["dequantize_affine_16"], "convert_element_type_51": ["dequantize_affine_16"], "sub_407": ["dequantize_affine_16"], "view_132": ["dequantize_affine_16"], "mul_869": ["dequantize_affine_16"], "view_135": ["dequantize_affine_17"], "view_137": ["dequantize_affine_17"], "convert_element_type_52": ["dequantize_affine_17"], "convert_element_type_53": ["dequantize_affine_17"], "sub_411": ["dequantize_affine_17"], "view_136": ["dequantize_affine_17"], "mul_874": ["dequantize_affine_17"], "view_138": ["dequantize_affine_17"], "mul_684": ["linear_6"], "view_107": ["linear_6"], "permute_11": ["linear_6"], "mm_default_154": ["linear_6"], "add_tensor_154": ["linear_6"], "view_108": ["linear_6"], "mul_691": ["mul_42"], "view_109": ["view_3"], "permute_12": ["transpose_4"], "clone_10": ["contiguous_4"], "mul_778": ["linear_7"], "view_123": ["linear_7"], "permute_13": ["linear_7"], "mm_1": ["linear_7"], "view_124": ["linear_7"], "view_125": ["view_4"], "permute_14": ["transpose_5"], "clone_11": ["contiguous_5"], "mul_879": ["linear_8"], "view_139": ["linear_8"], "permute_15": ["linear_8"], "mm_default_153": ["linear_8"], "add_tensor_153": ["linear_8"], "view_140": ["linear_8"], "view_141": ["view_5"], "permute_16": ["transpose_6"], "clone_12": ["contiguous_6"], "_scaled_dot_product_efficient_attention_1": ["scaled_dot_product_attention_1"], "getitem_10": ["scaled_dot_product_attention_1"], "permute_17": ["transpose_7"], "view_142": ["reshape_1", "choose_qparams_affine_default_9", "quantize_affine_9"], "amax_9": ["choose_qparams_affine_default_9"], "full_19": ["choose_qparams_affine_default_9"], "maximum_9": ["choose_qparams_affine_default_9"], "amin_9": ["choose_qparams_affine_default_9"], "full_18": ["choose_qparams_affine_default_9"], "minimum_9": ["choose_qparams_affine_default_9"], "sub_429": ["choose_qparams_affine_default_9"], "div_18": ["choose_qparams_affine_default_9"], "clamp_min_27": ["choose_qparams_affine_default_9"], "div_19": ["choose_qparams_affine_default_9"], "round_19": ["choose_qparams_affine_default_9"], "sub_435": ["choose_qparams_affine_default_9"], "clamp_min_28": ["choose_qparams_affine_default_9"], "clamp_max_18": ["choose_qparams_affine_default_9"], "view_145": ["quantize_affine_9"], "reciprocal_9": ["quantize_affine_9"], "mul_949": ["quantize_affine_9"], "mul_952": ["quantize_affine_9"], "round_20": ["quantize_affine_9"], "convert_element_type_54": ["choose_qparams_affine_default_9"], "view_146": ["quantize_affine_9"], "add_1501": ["quantize_affine_9"], "clamp_min_29": ["quantize_affine_9"], "clamp_max_19": ["quantize_affine_9"], "convert_element_type_55": ["quantize_affine_9", "dequantize_affine_18"], "view_150": ["dequantize_affine_18"], "convert_element_type_56": ["dequantize_affine_18"], "convert_element_type_57": ["dequantize_affine_18"], "sub_455": ["dequantize_affine_18"], "view_149": ["dequantize_affine_18"], "mul_974": ["dequantize_affine_18"], "view_152": ["dequantize_affine_19"], "view_154": ["dequantize_affine_19"], "convert_element_type_58": ["dequantize_affine_19"], "convert_element_type_59": ["dequantize_affine_19"], "sub_459": ["dequantize_affine_19"], "view_153": ["dequantize_affine_19"], "mul_979": ["dequantize_affine_19"], "view_155": ["dequantize_affine_19"], "mul_984": ["linear_9"], "view_156": ["linear_9"], "permute_18": ["linear_9"], "mm_default_152": ["linear_9"], "add_tensor_152": ["linear_9"], "view_157": ["linear_9", "dropout_4"], "add_1564": ["add_23"], "clone_14": ["layer_norm_3"], "var_mean_3": ["layer_norm_3"], "getitem_14": ["layer_norm_3"], "getitem_15": ["layer_norm_3"], "sub_465": ["layer_norm_3"], "add_1569": ["layer_norm_3"], "rsqrt_3": ["layer_norm_3"], "mul_995": ["layer_norm_3"], "mul_996": ["layer_norm_3"], "add_1570": ["layer_norm_3", "choose_qparams_affine_default_10", "quantize_affine_10"], "amax_10": ["choose_qparams_affine_default_10"], "full_21": ["choose_qparams_affine_default_10"], "maximum_10": ["choose_qparams_affine_default_10"], "amin_10": ["choose_qparams_affine_default_10"], "full_20": ["choose_qparams_affine_default_10"], "minimum_10": ["choose_qparams_affine_default_10"], "sub_476": ["choose_qparams_affine_default_10"], "div_20": ["choose_qparams_affine_default_10"], "clamp_min_30": ["choose_qparams_affine_default_10"], "div_21": ["choose_qparams_affine_default_10"], "round_21": ["choose_qparams_affine_default_10"], "sub_482": ["choose_qparams_affine_default_10"], "clamp_min_31": ["choose_qparams_affine_default_10"], "clamp_max_20": ["choose_qparams_affine_default_10"], "view_160": ["quantize_affine_10"], "reciprocal_10": ["quantize_affine_10"], "mul_1044": ["quantize_affine_10"], "mul_1047": ["quantize_affine_10"], "round_22": ["quantize_affine_10"], "convert_element_type_60": ["choose_qparams_affine_default_10"], "view_161": ["quantize_affine_10"], "add_1657": ["quantize_affine_10"], "clamp_min_32": ["quantize_affine_10"], "clamp_max_21": ["quantize_affine_10"], "convert_element_type_61": ["quantize_affine_10", "dequantize_affine_20"], "view_165": ["dequantize_affine_20"], "convert_element_type_62": ["dequantize_affine_20"], "convert_element_type_63": ["dequantize_affine_20"], "sub_502": ["dequantize_affine_20"], "view_164": ["dequantize_affine_20"], "mul_1069": ["dequantize_affine_20"], "view_167": ["dequantize_affine_21"], "view_169": ["dequantize_affine_21"], "convert_element_type_64": ["dequantize_affine_21"], "convert_element_type_65": ["dequantize_affine_21"], "sub_506": ["dequantize_affine_21"], "view_168": ["dequantize_affine_21"], "mul_1074": ["dequantize_affine_21"], "view_170": ["dequantize_affine_21"], "mul_1079": ["linear_10"], "view_171": ["linear_10"], "permute_19": ["linear_10"], "mm_default_151": ["linear_10"], "add_tensor_151": ["linear_10"], "view_172": ["linear_10"], "mul_1086": ["gelu_3"], "mul_1087": ["gelu_3"], "erf_3": ["gelu_3"], "add_1716": ["gelu_3"], "mul_1088": ["gelu_3", "dropout_5", "choose_qparams_affine_default_11", "quantize_affine_11"], "amax_11": ["choose_qparams_affine_default_11"], "full_23": ["choose_qparams_affine_default_11"], "maximum_11": ["choose_qparams_affine_default_11"], "amin_11": ["choose_qparams_affine_default_11"], "full_22": ["choose_qparams_affine_default_11"], "minimum_11": ["choose_qparams_affine_default_11"], "sub_519": ["choose_qparams_affine_default_11"], "div_22": ["choose_qparams_affine_default_11"], "clamp_min_33": ["choose_qparams_affine_default_11"], "div_23": ["choose_qparams_affine_default_11"], "round_23": ["choose_qparams_affine_default_11"], "sub_525": ["choose_qparams_affine_default_11"], "clamp_min_34": ["choose_qparams_affine_default_11"], "clamp_max_22": ["choose_qparams_affine_default_11"], "view_175": ["quantize_affine_11"], "reciprocal_11": ["quantize_affine_11"], "mul_1134": ["quantize_affine_11"], "mul_1137": ["quantize_affine_11"], "round_24": ["quantize_affine_11"], "convert_element_type_66": ["choose_qparams_affine_default_11"], "view_176": ["quantize_affine_11"], "add_1799": ["quantize_affine_11"], "clamp_min_35": ["quantize_affine_11"], "clamp_max_23": ["quantize_affine_11"], "convert_element_type_67": ["quantize_affine_11", "dequantize_affine_22"], "view_180": ["dequantize_affine_22"], "convert_element_type_68": ["dequantize_affine_22"], "convert_element_type_69": ["dequantize_affine_22"], "sub_545": ["dequantize_affine_22"], "view_179": ["dequantize_affine_22"], "mul_1159": ["dequantize_affine_22"], "view_182": ["dequantize_affine_23"], "view_184": ["dequantize_affine_23"], "convert_element_type_70": ["dequantize_affine_23"], "convert_element_type_71": ["dequantize_affine_23"], "sub_549": ["dequantize_affine_23"], "view_183": ["dequantize_affine_23"], "mul_1164": ["dequantize_affine_23"], "view_185": ["dequantize_affine_23"], "mul_1169": ["linear_11"], "view_186": ["linear_11"], "permute_20": ["linear_11"], "mm_default_150": ["linear_11"], "add_tensor_150": ["linear_11"], "view_187": ["linear_11", "dropout_6"], "add_1862": ["add_28"], "clone_17": ["layer_norm_4"], "var_mean_4": ["layer_norm_4"], "getitem_16": ["layer_norm_4"], "getitem_17": ["layer_norm_4"], "sub_555": ["layer_norm_4"], "add_1867": ["layer_norm_4"], "rsqrt_4": ["layer_norm_4"], "mul_1180": ["layer_norm_4"], "mul_1181": ["layer_norm_4"], "add_1868": ["layer_norm_4", "choose_qparams_affine_default_12", "quantize_affine_12", "choose_qparams_affine_default_13", "quantize_affine_13", "choose_qparams_affine_default_14", "quantize_affine_14"], "amax_12": ["choose_qparams_affine_default_12"], "full_25": ["choose_qparams_affine_default_12"], "maximum_12": ["choose_qparams_affine_default_12"], "amin_12": ["choose_qparams_affine_default_12"], "full_24": ["choose_qparams_affine_default_12"], "minimum_12": ["choose_qparams_affine_default_12"], "sub_566": ["choose_qparams_affine_default_12"], "div_24": ["choose_qparams_affine_default_12"], "clamp_min_36": ["choose_qparams_affine_default_12"], "div_25": ["choose_qparams_affine_default_12"], "round_25": ["choose_qparams_affine_default_12"], "sub_572": ["choose_qparams_affine_default_12"], "clamp_min_37": ["choose_qparams_affine_default_12"], "clamp_max_24": ["choose_qparams_affine_default_12"], "view_190": ["quantize_affine_12"], "reciprocal_12": ["quantize_affine_12"], "mul_1229": ["quantize_affine_12"], "mul_1232": ["quantize_affine_12"], "round_26": ["quantize_affine_12"], "convert_element_type_72": ["choose_qparams_affine_default_12"], "view_191": ["quantize_affine_12"], "add_1955": ["quantize_affine_12"], "clamp_min_38": ["quantize_affine_12"], "clamp_max_25": ["quantize_affine_12"], "convert_element_type_73": ["quantize_affine_12", "dequantize_affine_24"], "view_195": ["dequantize_affine_24"], "convert_element_type_74": ["dequantize_affine_24"], "convert_element_type_75": ["dequantize_affine_24"], "sub_592": ["dequantize_affine_24"], "view_194": ["dequantize_affine_24"], "mul_1254": ["dequantize_affine_24"], "view_197": ["dequantize_affine_25"], "view_199": ["dequantize_affine_25"], "convert_element_type_76": ["dequantize_affine_25"], "convert_element_type_77": ["dequantize_affine_25"], "sub_596": ["dequantize_affine_25"], "view_198": ["dequantize_affine_25"], "mul_1259": ["dequantize_affine_25"], "view_200": ["dequantize_affine_25"], "amax_13": ["choose_qparams_affine_default_13"], "full_27": ["choose_qparams_affine_default_13"], "maximum_13": ["choose_qparams_affine_default_13"], "amin_13": ["choose_qparams_affine_default_13"], "full_26": ["choose_qparams_affine_default_13"], "minimum_13": ["choose_qparams_affine_default_13"], "sub_611": ["choose_qparams_affine_default_13"], "div_26": ["choose_qparams_affine_default_13"], "clamp_min_39": ["choose_qparams_affine_default_13"], "div_27": ["choose_qparams_affine_default_13"], "round_27": ["choose_qparams_affine_default_13"], "sub_617": ["choose_qparams_affine_default_13"], "clamp_min_40": ["choose_qparams_affine_default_13"], "clamp_max_26": ["choose_qparams_affine_default_13"], "view_206": ["quantize_affine_13"], "reciprocal_13": ["quantize_affine_13"], "mul_1325": ["quantize_affine_13"], "mul_1328": ["quantize_affine_13"], "round_28": ["quantize_affine_13"], "convert_element_type_78": ["choose_qparams_affine_default_13"], "view_207": ["quantize_affine_13"], "add_2107": ["quantize_affine_13"], "clamp_min_41": ["quantize_affine_13"], "clamp_max_27": ["quantize_affine_13"], "convert_element_type_79": ["quantize_affine_13", "dequantize_affine_26"], "view_211": ["dequantize_affine_26"], "convert_element_type_80": ["dequantize_affine_26"], "convert_element_type_81": ["dequantize_affine_26"], "sub_637": ["dequantize_affine_26"], "view_210": ["dequantize_affine_26"], "mul_1350": ["dequantize_affine_26"], "view_213": ["dequantize_affine_27"], "view_215": ["dequantize_affine_27"], "convert_element_type_82": ["dequantize_affine_27"], "convert_element_type_83": ["dequantize_affine_27"], "sub_641": ["dequantize_affine_27"], "view_214": ["dequantize_affine_27"], "mul_1355": ["dequantize_affine_27"], "view_216": ["dequantize_affine_27"], "amax_14": ["choose_qparams_affine_default_14"], "full_29": ["choose_qparams_affine_default_14"], "maximum_14": ["choose_qparams_affine_default_14"], "amin_14": ["choose_qparams_affine_default_14"], "full_28": ["choose_qparams_affine_default_14"], "minimum_14": ["choose_qparams_affine_default_14"], "sub_655": ["choose_qparams_affine_default_14"], "div_28": ["choose_qparams_affine_default_14"], "clamp_min_42": ["choose_qparams_affine_default_14"], "div_29": ["choose_qparams_affine_default_14"], "round_29": ["choose_qparams_affine_default_14"], "sub_661": ["choose_qparams_affine_default_14"], "clamp_min_43": ["choose_qparams_affine_default_14"], "clamp_max_28": ["choose_qparams_affine_default_14"], "view_222": ["quantize_affine_14"], "reciprocal_14": ["quantize_affine_14"], "mul_1424": ["quantize_affine_14"], "mul_1427": ["quantize_affine_14"], "round_30": ["quantize_affine_14"], "convert_element_type_84": ["choose_qparams_affine_default_14"], "view_223": ["quantize_affine_14"], "add_2255": ["quantize_affine_14"], "clamp_min_44": ["quantize_affine_14"], "clamp_max_29": ["quantize_affine_14"], "convert_element_type_85": ["quantize_affine_14", "dequantize_affine_28"], "view_227": ["dequantize_affine_28"], "convert_element_type_86": ["dequantize_affine_28"], "convert_element_type_87": ["dequantize_affine_28"], "sub_681": ["dequantize_affine_28"], "view_226": ["dequantize_affine_28"], "mul_1449": ["dequantize_affine_28"], "view_229": ["dequantize_affine_29"], "view_231": ["dequantize_affine_29"], "convert_element_type_88": ["dequantize_affine_29"], "convert_element_type_89": ["dequantize_affine_29"], "sub_685": ["dequantize_affine_29"], "view_230": ["dequantize_affine_29"], "mul_1454": ["dequantize_affine_29"], "view_232": ["dequantize_affine_29"], "mul_1264": ["linear_12"], "view_201": ["linear_12"], "permute_21": ["linear_12"], "mm_default_149": ["linear_12"], "add_tensor_149": ["linear_12"], "view_202": ["linear_12"], "mul_1271": ["mul_79"], "view_203": ["view_6"], "permute_22": ["transpose_8"], "clone_18": ["contiguous_8"], "mul_1358": ["linear_13"], "view_217": ["linear_13"], "permute_23": ["linear_13"], "mm_2": ["linear_13"], "view_218": ["linear_13"], "view_219": ["view_7"], "permute_24": ["transpose_9"], "clone_19": ["contiguous_9"], "mul_1459": ["linear_14"], "view_233": ["linear_14"], "permute_25": ["linear_14"], "mm_default_148": ["linear_14"], "add_tensor_148": ["linear_14"], "view_234": ["linear_14"], "view_235": ["view_8"], "permute_26": ["transpose_10"], "clone_20": ["contiguous_10"], "_scaled_dot_product_efficient_attention_2": ["scaled_dot_product_attention_2"], "getitem_18": ["scaled_dot_product_attention_2"], "permute_27": ["transpose_11"], "view_236": ["reshape_2", "choose_qparams_affine_default_15", "quantize_affine_15"], "amax_15": ["choose_qparams_affine_default_15"], "full_31": ["choose_qparams_affine_default_15"], "maximum_15": ["choose_qparams_affine_default_15"], "amin_15": ["choose_qparams_affine_default_15"], "full_30": ["choose_qparams_affine_default_15"], "minimum_15": ["choose_qparams_affine_default_15"], "sub_703": ["choose_qparams_affine_default_15"], "div_30": ["choose_qparams_affine_default_15"], "clamp_min_45": ["choose_qparams_affine_default_15"], "div_31": ["choose_qparams_affine_default_15"], "round_31": ["choose_qparams_affine_default_15"], "sub_709": ["choose_qparams_affine_default_15"], "clamp_min_46": ["choose_qparams_affine_default_15"], "clamp_max_30": ["choose_qparams_affine_default_15"], "view_239": ["quantize_affine_15"], "reciprocal_15": ["quantize_affine_15"], "mul_1529": ["quantize_affine_15"], "mul_1532": ["quantize_affine_15"], "round_32": ["quantize_affine_15"], "convert_element_type_90": ["choose_qparams_affine_default_15"], "view_240": ["quantize_affine_15"], "add_2419": ["quantize_affine_15"], "clamp_min_47": ["quantize_affine_15"], "clamp_max_31": ["quantize_affine_15"], "convert_element_type_91": ["quantize_affine_15", "dequantize_affine_30"], "view_244": ["dequantize_affine_30"], "convert_element_type_92": ["dequantize_affine_30"], "convert_element_type_93": ["dequantize_affine_30"], "sub_729": ["dequantize_affine_30"], "view_243": ["dequantize_affine_30"], "mul_1554": ["dequantize_affine_30"], "view_246": ["dequantize_affine_31"], "view_248": ["dequantize_affine_31"], "convert_element_type_94": ["dequantize_affine_31"], "convert_element_type_95": ["dequantize_affine_31"], "sub_733": ["dequantize_affine_31"], "view_247": ["dequantize_affine_31"], "mul_1559": ["dequantize_affine_31"], "view_249": ["dequantize_affine_31"], "mul_1564": ["linear_15"], "view_250": ["linear_15"], "permute_28": ["linear_15"], "mm_default_147": ["linear_15"], "add_tensor_147": ["linear_15"], "view_251": ["linear_15", "dropout_7"], "add_2482": ["add_37"], "clone_22": ["layer_norm_5"], "var_mean_5": ["layer_norm_5"], "getitem_22": ["layer_norm_5"], "getitem_23": ["layer_norm_5"], "sub_739": ["layer_norm_5"], "add_2487": ["layer_norm_5"], "rsqrt_5": ["layer_norm_5"], "mul_1575": ["layer_norm_5"], "mul_1576": ["layer_norm_5"], "add_2488": ["layer_norm_5", "choose_qparams_affine_default_16", "quantize_affine_16"], "amax_16": ["choose_qparams_affine_default_16"], "full_33": ["choose_qparams_affine_default_16"], "maximum_16": ["choose_qparams_affine_default_16"], "amin_16": ["choose_qparams_affine_default_16"], "full_32": ["choose_qparams_affine_default_16"], "minimum_16": ["choose_qparams_affine_default_16"], "sub_750": ["choose_qparams_affine_default_16"], "div_32": ["choose_qparams_affine_default_16"], "clamp_min_48": ["choose_qparams_affine_default_16"], "div_33": ["choose_qparams_affine_default_16"], "round_33": ["choose_qparams_affine_default_16"], "sub_756": ["choose_qparams_affine_default_16"], "clamp_min_49": ["choose_qparams_affine_default_16"], "clamp_max_32": ["choose_qparams_affine_default_16"], "view_254": ["quantize_affine_16"], "reciprocal_16": ["quantize_affine_16"], "mul_1624": ["quantize_affine_16"], "mul_1627": ["quantize_affine_16"], "round_34": ["quantize_affine_16"], "convert_element_type_96": ["choose_qparams_affine_default_16"], "view_255": ["quantize_affine_16"], "add_2575": ["quantize_affine_16"], "clamp_min_50": ["quantize_affine_16"], "clamp_max_33": ["quantize_affine_16"], "convert_element_type_97": ["quantize_affine_16", "dequantize_affine_32"], "view_259": ["dequantize_affine_32"], "convert_element_type_98": ["dequantize_affine_32"], "convert_element_type_99": ["dequantize_affine_32"], "sub_776": ["dequantize_affine_32"], "view_258": ["dequantize_affine_32"], "mul_1649": ["dequantize_affine_32"], "view_261": ["dequantize_affine_33"], "view_263": ["dequantize_affine_33"], "convert_element_type_100": ["dequantize_affine_33"], "convert_element_type_101": ["dequantize_affine_33"], "sub_780": ["dequantize_affine_33"], "view_262": ["dequantize_affine_33"], "mul_1654": ["dequantize_affine_33"], "view_264": ["dequantize_affine_33"], "mul_1659": ["linear_16"], "view_265": ["linear_16"], "permute_29": ["linear_16"], "mm_default_146": ["linear_16"], "add_tensor_146": ["linear_16"], "view_266": ["linear_16"], "mul_1666": ["gelu_4"], "mul_1667": ["gelu_4"], "erf_4": ["gelu_4"], "add_2634": ["gelu_4"], "mul_1668": ["gelu_4", "dropout_8", "choose_qparams_affine_default_17", "quantize_affine_17"], "amax_17": ["choose_qparams_affine_default_17"], "full_35": ["choose_qparams_affine_default_17"], "maximum_17": ["choose_qparams_affine_default_17"], "amin_17": ["choose_qparams_affine_default_17"], "full_34": ["choose_qparams_affine_default_17"], "minimum_17": ["choose_qparams_affine_default_17"], "sub_793": ["choose_qparams_affine_default_17"], "div_34": ["choose_qparams_affine_default_17"], "clamp_min_51": ["choose_qparams_affine_default_17"], "div_35": ["choose_qparams_affine_default_17"], "round_35": ["choose_qparams_affine_default_17"], "sub_799": ["choose_qparams_affine_default_17"], "clamp_min_52": ["choose_qparams_affine_default_17"], "clamp_max_34": ["choose_qparams_affine_default_17"], "view_269": ["quantize_affine_17"], "reciprocal_17": ["quantize_affine_17"], "mul_1714": ["quantize_affine_17"], "mul_1717": ["quantize_affine_17"], "round_36": ["quantize_affine_17"], "convert_element_type_102": ["choose_qparams_affine_default_17"], "view_270": ["quantize_affine_17"], "add_2717": ["quantize_affine_17"], "clamp_min_53": ["quantize_affine_17"], "clamp_max_35": ["quantize_affine_17"], "convert_element_type_103": ["quantize_affine_17", "dequantize_affine_34"], "view_274": ["dequantize_affine_34"], "convert_element_type_104": ["dequantize_affine_34"], "convert_element_type_105": ["dequantize_affine_34"], "sub_819": ["dequantize_affine_34"], "view_273": ["dequantize_affine_34"], "mul_1739": ["dequantize_affine_34"], "view_276": ["dequantize_affine_35"], "view_278": ["dequantize_affine_35"], "convert_element_type_106": ["dequantize_affine_35"], "convert_element_type_107": ["dequantize_affine_35"], "sub_823": ["dequantize_affine_35"], "view_277": ["dequantize_affine_35"], "mul_1744": ["dequantize_affine_35"], "view_279": ["dequantize_affine_35"], "mul_1749": ["linear_17"], "view_280": ["linear_17"], "permute_30": ["linear_17"], "mm_default_145": ["linear_17"], "add_tensor_145": ["linear_17"], "view_281": ["linear_17", "dropout_9"], "add_2780": ["add_42"], "clone_25": ["layer_norm_6"], "var_mean_6": ["layer_norm_6"], "getitem_24": ["layer_norm_6"], "getitem_25": ["layer_norm_6"], "sub_829": ["layer_norm_6"], "add_2785": ["layer_norm_6"], "rsqrt_6": ["layer_norm_6"], "mul_1760": ["layer_norm_6"], "mul_1761": ["layer_norm_6"], "add_2786": ["layer_norm_6", "choose_qparams_affine_default_18", "quantize_affine_18", "choose_qparams_affine_default_19", "quantize_affine_19", "choose_qparams_affine_default_20", "quantize_affine_20"], "amax_18": ["choose_qparams_affine_default_18"], "full_37": ["choose_qparams_affine_default_18"], "maximum_18": ["choose_qparams_affine_default_18"], "amin_18": ["choose_qparams_affine_default_18"], "full_36": ["choose_qparams_affine_default_18"], "minimum_18": ["choose_qparams_affine_default_18"], "sub_840": ["choose_qparams_affine_default_18"], "div_36": ["choose_qparams_affine_default_18"], "clamp_min_54": ["choose_qparams_affine_default_18"], "div_37": ["choose_qparams_affine_default_18"], "round_37": ["choose_qparams_affine_default_18"], "sub_846": ["choose_qparams_affine_default_18"], "clamp_min_55": ["choose_qparams_affine_default_18"], "clamp_max_36": ["choose_qparams_affine_default_18"], "view_284": ["quantize_affine_18"], "reciprocal_18": ["quantize_affine_18"], "mul_1809": ["quantize_affine_18"], "mul_1812": ["quantize_affine_18"], "round_38": ["quantize_affine_18"], "convert_element_type_108": ["choose_qparams_affine_default_18"], "view_285": ["quantize_affine_18"], "add_2873": ["quantize_affine_18"], "clamp_min_56": ["quantize_affine_18"], "clamp_max_37": ["quantize_affine_18"], "convert_element_type_109": ["quantize_affine_18", "dequantize_affine_36"], "view_289": ["dequantize_affine_36"], "convert_element_type_110": ["dequantize_affine_36"], "convert_element_type_111": ["dequantize_affine_36"], "sub_866": ["dequantize_affine_36"], "view_288": ["dequantize_affine_36"], "mul_1834": ["dequantize_affine_36"], "view_291": ["dequantize_affine_37"], "view_293": ["dequantize_affine_37"], "convert_element_type_112": ["dequantize_affine_37"], "convert_element_type_113": ["dequantize_affine_37"], "sub_870": ["dequantize_affine_37"], "view_292": ["dequantize_affine_37"], "mul_1839": ["dequantize_affine_37"], "view_294": ["dequantize_affine_37"], "amax_19": ["choose_qparams_affine_default_19"], "full_39": ["choose_qparams_affine_default_19"], "maximum_19": ["choose_qparams_affine_default_19"], "amin_19": ["choose_qparams_affine_default_19"], "full_38": ["choose_qparams_affine_default_19"], "minimum_19": ["choose_qparams_affine_default_19"], "sub_885": ["choose_qparams_affine_default_19"], "div_38": ["choose_qparams_affine_default_19"], "clamp_min_57": ["choose_qparams_affine_default_19"], "div_39": ["choose_qparams_affine_default_19"], "round_39": ["choose_qparams_affine_default_19"], "sub_891": ["choose_qparams_affine_default_19"], "clamp_min_58": ["choose_qparams_affine_default_19"], "clamp_max_38": ["choose_qparams_affine_default_19"], "view_300": ["quantize_affine_19"], "reciprocal_19": ["quantize_affine_19"], "mul_1905": ["quantize_affine_19"], "mul_1908": ["quantize_affine_19"], "round_40": ["quantize_affine_19"], "convert_element_type_114": ["choose_qparams_affine_default_19"], "view_301": ["quantize_affine_19"], "add_3025": ["quantize_affine_19"], "clamp_min_59": ["quantize_affine_19"], "clamp_max_39": ["quantize_affine_19"], "convert_element_type_115": ["quantize_affine_19", "dequantize_affine_38"], "view_305": ["dequantize_affine_38"], "convert_element_type_116": ["dequantize_affine_38"], "convert_element_type_117": ["dequantize_affine_38"], "sub_911": ["dequantize_affine_38"], "view_304": ["dequantize_affine_38"], "mul_1930": ["dequantize_affine_38"], "view_307": ["dequantize_affine_39"], "view_309": ["dequantize_affine_39"], "convert_element_type_118": ["dequantize_affine_39"], "convert_element_type_119": ["dequantize_affine_39"], "sub_915": ["dequantize_affine_39"], "view_308": ["dequantize_affine_39"], "mul_1935": ["dequantize_affine_39"], "view_310": ["dequantize_affine_39"], "amax_20": ["choose_qparams_affine_default_20"], "full_41": ["choose_qparams_affine_default_20"], "maximum_20": ["choose_qparams_affine_default_20"], "amin_20": ["choose_qparams_affine_default_20"], "full_40": ["choose_qparams_affine_default_20"], "minimum_20": ["choose_qparams_affine_default_20"], "sub_929": ["choose_qparams_affine_default_20"], "div_40": ["choose_qparams_affine_default_20"], "clamp_min_60": ["choose_qparams_affine_default_20"], "div_41": ["choose_qparams_affine_default_20"], "round_41": ["choose_qparams_affine_default_20"], "sub_935": ["choose_qparams_affine_default_20"], "clamp_min_61": ["choose_qparams_affine_default_20"], "clamp_max_40": ["choose_qparams_affine_default_20"], "view_316": ["quantize_affine_20"], "reciprocal_20": ["quantize_affine_20"], "mul_2004": ["quantize_affine_20"], "mul_2007": ["quantize_affine_20"], "round_42": ["quantize_affine_20"], "convert_element_type_120": ["choose_qparams_affine_default_20"], "view_317": ["quantize_affine_20"], "add_3173": ["quantize_affine_20"], "clamp_min_62": ["quantize_affine_20"], "clamp_max_41": ["quantize_affine_20"], "convert_element_type_121": ["quantize_affine_20", "dequantize_affine_40"], "view_321": ["dequantize_affine_40"], "convert_element_type_122": ["dequantize_affine_40"], "convert_element_type_123": ["dequantize_affine_40"], "sub_955": ["dequantize_affine_40"], "view_320": ["dequantize_affine_40"], "mul_2029": ["dequantize_affine_40"], "view_323": ["dequantize_affine_41"], "view_325": ["dequantize_affine_41"], "convert_element_type_124": ["dequantize_affine_41"], "convert_element_type_125": ["dequantize_affine_41"], "sub_959": ["dequantize_affine_41"], "view_324": ["dequantize_affine_41"], "mul_2034": ["dequantize_affine_41"], "view_326": ["dequantize_affine_41"], "mul_1844": ["linear_18"], "view_295": ["linear_18"], "permute_31": ["linear_18"], "mm_default_144": ["linear_18"], "add_tensor_144": ["linear_18"], "view_296": ["linear_18"], "mul_1851": ["mul_116"], "view_297": ["view_9"], "permute_32": ["transpose_12"], "clone_26": ["contiguous_12"], "mul_1938": ["linear_19"], "view_311": ["linear_19"], "permute_33": ["linear_19"], "mm_3": ["linear_19"], "view_312": ["linear_19"], "view_313": ["view_10"], "permute_34": ["transpose_13"], "clone_27": ["contiguous_13"], "mul_2039": ["linear_20"], "view_327": ["linear_20"], "permute_35": ["linear_20"], "mm_default_143": ["linear_20"], "add_tensor_143": ["linear_20"], "view_328": ["linear_20"], "view_329": ["view_11"], "permute_36": ["transpose_14"], "clone_28": ["contiguous_14"], "_scaled_dot_product_efficient_attention_3": ["scaled_dot_product_attention_3"], "getitem_26": ["scaled_dot_product_attention_3"], "permute_37": ["transpose_15"], "view_330": ["reshape_3", "choose_qparams_affine_default_21", "quantize_affine_21"], "amax_21": ["choose_qparams_affine_default_21"], "full_43": ["choose_qparams_affine_default_21"], "maximum_21": ["choose_qparams_affine_default_21"], "amin_21": ["choose_qparams_affine_default_21"], "full_42": ["choose_qparams_affine_default_21"], "minimum_21": ["choose_qparams_affine_default_21"], "sub_977": ["choose_qparams_affine_default_21"], "div_42": ["choose_qparams_affine_default_21"], "clamp_min_63": ["choose_qparams_affine_default_21"], "div_43": ["choose_qparams_affine_default_21"], "round_43": ["choose_qparams_affine_default_21"], "sub_983": ["choose_qparams_affine_default_21"], "clamp_min_64": ["choose_qparams_affine_default_21"], "clamp_max_42": ["choose_qparams_affine_default_21"], "view_333": ["quantize_affine_21"], "reciprocal_21": ["quantize_affine_21"], "mul_2109": ["quantize_affine_21"], "mul_2112": ["quantize_affine_21"], "round_44": ["quantize_affine_21"], "convert_element_type_126": ["choose_qparams_affine_default_21"], "view_334": ["quantize_affine_21"], "add_3337": ["quantize_affine_21"], "clamp_min_65": ["quantize_affine_21"], "clamp_max_43": ["quantize_affine_21"], "convert_element_type_127": ["quantize_affine_21", "dequantize_affine_42"], "view_338": ["dequantize_affine_42"], "convert_element_type_128": ["dequantize_affine_42"], "convert_element_type_129": ["dequantize_affine_42"], "sub_1003": ["dequantize_affine_42"], "view_337": ["dequantize_affine_42"], "mul_2134": ["dequantize_affine_42"], "view_340": ["dequantize_affine_43"], "view_342": ["dequantize_affine_43"], "convert_element_type_130": ["dequantize_affine_43"], "convert_element_type_131": ["dequantize_affine_43"], "sub_1007": ["dequantize_affine_43"], "view_341": ["dequantize_affine_43"], "mul_2139": ["dequantize_affine_43"], "view_343": ["dequantize_affine_43"], "mul_2144": ["linear_21"], "view_344": ["linear_21"], "permute_38": ["linear_21"], "mm_default_142": ["linear_21"], "add_tensor_142": ["linear_21"], "view_345": ["linear_21", "dropout_10"], "add_3400": ["add_51"], "clone_30": ["layer_norm_7"], "var_mean_7": ["layer_norm_7"], "getitem_30": ["layer_norm_7"], "getitem_31": ["layer_norm_7"], "sub_1013": ["layer_norm_7"], "add_3405": ["layer_norm_7"], "rsqrt_7": ["layer_norm_7"], "mul_2155": ["layer_norm_7"], "mul_2156": ["layer_norm_7"], "add_3406": ["layer_norm_7", "choose_qparams_affine_default_22", "quantize_affine_22"], "amax_22": ["choose_qparams_affine_default_22"], "full_45": ["choose_qparams_affine_default_22"], "maximum_22": ["choose_qparams_affine_default_22"], "amin_22": ["choose_qparams_affine_default_22"], "full_44": ["choose_qparams_affine_default_22"], "minimum_22": ["choose_qparams_affine_default_22"], "sub_1024": ["choose_qparams_affine_default_22"], "div_44": ["choose_qparams_affine_default_22"], "clamp_min_66": ["choose_qparams_affine_default_22"], "div_45": ["choose_qparams_affine_default_22"], "round_45": ["choose_qparams_affine_default_22"], "sub_1030": ["choose_qparams_affine_default_22"], "clamp_min_67": ["choose_qparams_affine_default_22"], "clamp_max_44": ["choose_qparams_affine_default_22"], "view_348": ["quantize_affine_22"], "reciprocal_22": ["quantize_affine_22"], "mul_2204": ["quantize_affine_22"], "mul_2207": ["quantize_affine_22"], "round_46": ["quantize_affine_22"], "convert_element_type_132": ["choose_qparams_affine_default_22"], "view_349": ["quantize_affine_22"], "add_3493": ["quantize_affine_22"], "clamp_min_68": ["quantize_affine_22"], "clamp_max_45": ["quantize_affine_22"], "convert_element_type_133": ["quantize_affine_22", "dequantize_affine_44"], "view_353": ["dequantize_affine_44"], "convert_element_type_134": ["dequantize_affine_44"], "convert_element_type_135": ["dequantize_affine_44"], "sub_1050": ["dequantize_affine_44"], "view_352": ["dequantize_affine_44"], "mul_2229": ["dequantize_affine_44"], "view_355": ["dequantize_affine_45"], "view_357": ["dequantize_affine_45"], "convert_element_type_136": ["dequantize_affine_45"], "convert_element_type_137": ["dequantize_affine_45"], "sub_1054": ["dequantize_affine_45"], "view_356": ["dequantize_affine_45"], "mul_2234": ["dequantize_affine_45"], "view_358": ["dequantize_affine_45"], "mul_2239": ["linear_22"], "view_359": ["linear_22"], "permute_39": ["linear_22"], "mm_default_141": ["linear_22"], "add_tensor_141": ["linear_22"], "view_360": ["linear_22"], "mul_2246": ["gelu_5"], "mul_2247": ["gelu_5"], "erf_5": ["gelu_5"], "add_3552": ["gelu_5"], "mul_2248": ["gelu_5", "dropout_11", "choose_qparams_affine_default_23", "quantize_affine_23"], "amax_23": ["choose_qparams_affine_default_23"], "full_47": ["choose_qparams_affine_default_23"], "maximum_23": ["choose_qparams_affine_default_23"], "amin_23": ["choose_qparams_affine_default_23"], "full_46": ["choose_qparams_affine_default_23"], "minimum_23": ["choose_qparams_affine_default_23"], "sub_1067": ["choose_qparams_affine_default_23"], "div_46": ["choose_qparams_affine_default_23"], "clamp_min_69": ["choose_qparams_affine_default_23"], "div_47": ["choose_qparams_affine_default_23"], "round_47": ["choose_qparams_affine_default_23"], "sub_1073": ["choose_qparams_affine_default_23"], "clamp_min_70": ["choose_qparams_affine_default_23"], "clamp_max_46": ["choose_qparams_affine_default_23"], "view_363": ["quantize_affine_23"], "reciprocal_23": ["quantize_affine_23"], "mul_2294": ["quantize_affine_23"], "mul_2297": ["quantize_affine_23"], "round_48": ["quantize_affine_23"], "convert_element_type_138": ["choose_qparams_affine_default_23"], "view_364": ["quantize_affine_23"], "add_3635": ["quantize_affine_23"], "clamp_min_71": ["quantize_affine_23"], "clamp_max_47": ["quantize_affine_23"], "convert_element_type_139": ["quantize_affine_23", "dequantize_affine_46"], "view_368": ["dequantize_affine_46"], "convert_element_type_140": ["dequantize_affine_46"], "convert_element_type_141": ["dequantize_affine_46"], "sub_1093": ["dequantize_affine_46"], "view_367": ["dequantize_affine_46"], "mul_2319": ["dequantize_affine_46"], "view_370": ["dequantize_affine_47"], "view_372": ["dequantize_affine_47"], "convert_element_type_142": ["dequantize_affine_47"], "convert_element_type_143": ["dequantize_affine_47"], "sub_1097": ["dequantize_affine_47"], "view_371": ["dequantize_affine_47"], "mul_2324": ["dequantize_affine_47"], "view_373": ["dequantize_affine_47"], "mul_2329": ["linear_23"], "view_374": ["linear_23"], "permute_40": ["linear_23"], "mm_default_140": ["linear_23"], "add_tensor_140": ["linear_23"], "view_375": ["linear_23", "dropout_12"], "add_3698": ["add_56"], "clone_33": ["layer_norm_8"], "var_mean_8": ["layer_norm_8"], "getitem_32": ["layer_norm_8"], "getitem_33": ["layer_norm_8"], "sub_1103": ["layer_norm_8"], "add_3703": ["layer_norm_8"], "rsqrt_8": ["layer_norm_8"], "mul_2340": ["layer_norm_8"], "mul_2341": ["layer_norm_8"], "add_3704": ["layer_norm_8", "choose_qparams_affine_default_24", "quantize_affine_24", "choose_qparams_affine_default_25", "quantize_affine_25", "choose_qparams_affine_default_26", "quantize_affine_26"], "amax_24": ["choose_qparams_affine_default_24"], "full_49": ["choose_qparams_affine_default_24"], "maximum_24": ["choose_qparams_affine_default_24"], "amin_24": ["choose_qparams_affine_default_24"], "full_48": ["choose_qparams_affine_default_24"], "minimum_24": ["choose_qparams_affine_default_24"], "sub_1114": ["choose_qparams_affine_default_24"], "div_48": ["choose_qparams_affine_default_24"], "clamp_min_72": ["choose_qparams_affine_default_24"], "div_49": ["choose_qparams_affine_default_24"], "round_49": ["choose_qparams_affine_default_24"], "sub_1120": ["choose_qparams_affine_default_24"], "clamp_min_73": ["choose_qparams_affine_default_24"], "clamp_max_48": ["choose_qparams_affine_default_24"], "view_378": ["quantize_affine_24"], "reciprocal_24": ["quantize_affine_24"], "mul_2389": ["quantize_affine_24"], "mul_2392": ["quantize_affine_24"], "round_50": ["quantize_affine_24"], "convert_element_type_144": ["choose_qparams_affine_default_24"], "view_379": ["quantize_affine_24"], "add_3791": ["quantize_affine_24"], "clamp_min_74": ["quantize_affine_24"], "clamp_max_49": ["quantize_affine_24"], "convert_element_type_145": ["quantize_affine_24", "dequantize_affine_48"], "view_383": ["dequantize_affine_48"], "convert_element_type_146": ["dequantize_affine_48"], "convert_element_type_147": ["dequantize_affine_48"], "sub_1140": ["dequantize_affine_48"], "view_382": ["dequantize_affine_48"], "mul_2414": ["dequantize_affine_48"], "view_385": ["dequantize_affine_49"], "view_387": ["dequantize_affine_49"], "convert_element_type_148": ["dequantize_affine_49"], "convert_element_type_149": ["dequantize_affine_49"], "sub_1144": ["dequantize_affine_49"], "view_386": ["dequantize_affine_49"], "mul_2419": ["dequantize_affine_49"], "view_388": ["dequantize_affine_49"], "amax_25": ["choose_qparams_affine_default_25"], "full_51": ["choose_qparams_affine_default_25"], "maximum_25": ["choose_qparams_affine_default_25"], "amin_25": ["choose_qparams_affine_default_25"], "full_50": ["choose_qparams_affine_default_25"], "minimum_25": ["choose_qparams_affine_default_25"], "sub_1159": ["choose_qparams_affine_default_25"], "div_50": ["choose_qparams_affine_default_25"], "clamp_min_75": ["choose_qparams_affine_default_25"], "div_51": ["choose_qparams_affine_default_25"], "round_51": ["choose_qparams_affine_default_25"], "sub_1165": ["choose_qparams_affine_default_25"], "clamp_min_76": ["choose_qparams_affine_default_25"], "clamp_max_50": ["choose_qparams_affine_default_25"], "view_394": ["quantize_affine_25"], "reciprocal_25": ["quantize_affine_25"], "mul_2485": ["quantize_affine_25"], "mul_2488": ["quantize_affine_25"], "round_52": ["quantize_affine_25"], "convert_element_type_150": ["choose_qparams_affine_default_25"], "view_395": ["quantize_affine_25"], "add_3943": ["quantize_affine_25"], "clamp_min_77": ["quantize_affine_25"], "clamp_max_51": ["quantize_affine_25"], "convert_element_type_151": ["quantize_affine_25", "dequantize_affine_50"], "view_399": ["dequantize_affine_50"], "convert_element_type_152": ["dequantize_affine_50"], "convert_element_type_153": ["dequantize_affine_50"], "sub_1185": ["dequantize_affine_50"], "view_398": ["dequantize_affine_50"], "mul_2510": ["dequantize_affine_50"], "view_401": ["dequantize_affine_51"], "view_403": ["dequantize_affine_51"], "convert_element_type_154": ["dequantize_affine_51"], "convert_element_type_155": ["dequantize_affine_51"], "sub_1189": ["dequantize_affine_51"], "view_402": ["dequantize_affine_51"], "mul_2515": ["dequantize_affine_51"], "view_404": ["dequantize_affine_51"], "amax_26": ["choose_qparams_affine_default_26"], "full_53": ["choose_qparams_affine_default_26"], "maximum_26": ["choose_qparams_affine_default_26"], "amin_26": ["choose_qparams_affine_default_26"], "full_52": ["choose_qparams_affine_default_26"], "minimum_26": ["choose_qparams_affine_default_26"], "sub_1203": ["choose_qparams_affine_default_26"], "div_52": ["choose_qparams_affine_default_26"], "clamp_min_78": ["choose_qparams_affine_default_26"], "div_53": ["choose_qparams_affine_default_26"], "round_53": ["choose_qparams_affine_default_26"], "sub_1209": ["choose_qparams_affine_default_26"], "clamp_min_79": ["choose_qparams_affine_default_26"], "clamp_max_52": ["choose_qparams_affine_default_26"], "view_410": ["quantize_affine_26"], "reciprocal_26": ["quantize_affine_26"], "mul_2584": ["quantize_affine_26"], "mul_2587": ["quantize_affine_26"], "round_54": ["quantize_affine_26"], "convert_element_type_156": ["choose_qparams_affine_default_26"], "view_411": ["quantize_affine_26"], "add_4091": ["quantize_affine_26"], "clamp_min_80": ["quantize_affine_26"], "clamp_max_53": ["quantize_affine_26"], "convert_element_type_157": ["quantize_affine_26", "dequantize_affine_52"], "view_415": ["dequantize_affine_52"], "convert_element_type_158": ["dequantize_affine_52"], "convert_element_type_159": ["dequantize_affine_52"], "sub_1229": ["dequantize_affine_52"], "view_414": ["dequantize_affine_52"], "mul_2609": ["dequantize_affine_52"], "view_417": ["dequantize_affine_53"], "view_419": ["dequantize_affine_53"], "convert_element_type_160": ["dequantize_affine_53"], "convert_element_type_161": ["dequantize_affine_53"], "sub_1233": ["dequantize_affine_53"], "view_418": ["dequantize_affine_53"], "mul_2614": ["dequantize_affine_53"], "view_420": ["dequantize_affine_53"], "mul_2424": ["linear_24"], "view_389": ["linear_24"], "permute_41": ["linear_24"], "mm_default_139": ["linear_24"], "add_tensor_139": ["linear_24"], "view_390": ["linear_24"], "mul_2431": ["mul_153"], "view_391": ["view_12"], "permute_42": ["transpose_16"], "clone_34": ["contiguous_16"], "mul_2518": ["linear_25"], "view_405": ["linear_25"], "permute_43": ["linear_25"], "mm_4": ["linear_25"], "view_406": ["linear_25"], "view_407": ["view_13"], "permute_44": ["transpose_17"], "clone_35": ["contiguous_17"], "mul_2619": ["linear_26"], "view_421": ["linear_26"], "permute_45": ["linear_26"], "mm_default_138": ["linear_26"], "add_tensor_138": ["linear_26"], "view_422": ["linear_26"], "view_423": ["view_14"], "permute_46": ["transpose_18"], "clone_36": ["contiguous_18"], "_scaled_dot_product_efficient_attention_4": ["scaled_dot_product_attention_4"], "getitem_34": ["scaled_dot_product_attention_4"], "permute_47": ["transpose_19"], "view_424": ["reshape_4", "choose_qparams_affine_default_27", "quantize_affine_27"], "amax_27": ["choose_qparams_affine_default_27"], "full_55": ["choose_qparams_affine_default_27"], "maximum_27": ["choose_qparams_affine_default_27"], "amin_27": ["choose_qparams_affine_default_27"], "full_54": ["choose_qparams_affine_default_27"], "minimum_27": ["choose_qparams_affine_default_27"], "sub_1251": ["choose_qparams_affine_default_27"], "div_54": ["choose_qparams_affine_default_27"], "clamp_min_81": ["choose_qparams_affine_default_27"], "div_55": ["choose_qparams_affine_default_27"], "round_55": ["choose_qparams_affine_default_27"], "sub_1257": ["choose_qparams_affine_default_27"], "clamp_min_82": ["choose_qparams_affine_default_27"], "clamp_max_54": ["choose_qparams_affine_default_27"], "view_427": ["quantize_affine_27"], "reciprocal_27": ["quantize_affine_27"], "mul_2689": ["quantize_affine_27"], "mul_2692": ["quantize_affine_27"], "round_56": ["quantize_affine_27"], "convert_element_type_162": ["choose_qparams_affine_default_27"], "view_428": ["quantize_affine_27"], "add_4255": ["quantize_affine_27"], "clamp_min_83": ["quantize_affine_27"], "clamp_max_55": ["quantize_affine_27"], "convert_element_type_163": ["quantize_affine_27", "dequantize_affine_54"], "view_432": ["dequantize_affine_54"], "convert_element_type_164": ["dequantize_affine_54"], "convert_element_type_165": ["dequantize_affine_54"], "sub_1277": ["dequantize_affine_54"], "view_431": ["dequantize_affine_54"], "mul_2714": ["dequantize_affine_54"], "view_434": ["dequantize_affine_55"], "view_436": ["dequantize_affine_55"], "convert_element_type_166": ["dequantize_affine_55"], "convert_element_type_167": ["dequantize_affine_55"], "sub_1281": ["dequantize_affine_55"], "view_435": ["dequantize_affine_55"], "mul_2719": ["dequantize_affine_55"], "view_437": ["dequantize_affine_55"], "mul_2724": ["linear_27"], "view_438": ["linear_27"], "permute_48": ["linear_27"], "mm_default_137": ["linear_27"], "add_tensor_137": ["linear_27"], "view_439": ["linear_27", "dropout_13"], "add_4318": ["add_65"], "clone_38": ["layer_norm_9"], "var_mean_9": ["layer_norm_9"], "getitem_38": ["layer_norm_9"], "getitem_39": ["layer_norm_9"], "sub_1287": ["layer_norm_9"], "add_4323": ["layer_norm_9"], "rsqrt_9": ["layer_norm_9"], "mul_2735": ["layer_norm_9"], "mul_2736": ["layer_norm_9"], "add_4324": ["layer_norm_9", "choose_qparams_affine_default_28", "quantize_affine_28"], "amax_28": ["choose_qparams_affine_default_28"], "full_57": ["choose_qparams_affine_default_28"], "maximum_28": ["choose_qparams_affine_default_28"], "amin_28": ["choose_qparams_affine_default_28"], "full_56": ["choose_qparams_affine_default_28"], "minimum_28": ["choose_qparams_affine_default_28"], "sub_1298": ["choose_qparams_affine_default_28"], "div_56": ["choose_qparams_affine_default_28"], "clamp_min_84": ["choose_qparams_affine_default_28"], "div_57": ["choose_qparams_affine_default_28"], "round_57": ["choose_qparams_affine_default_28"], "sub_1304": ["choose_qparams_affine_default_28"], "clamp_min_85": ["choose_qparams_affine_default_28"], "clamp_max_56": ["choose_qparams_affine_default_28"], "view_442": ["quantize_affine_28"], "reciprocal_28": ["quantize_affine_28"], "mul_2784": ["quantize_affine_28"], "mul_2787": ["quantize_affine_28"], "round_58": ["quantize_affine_28"], "convert_element_type_168": ["choose_qparams_affine_default_28"], "view_443": ["quantize_affine_28"], "add_4411": ["quantize_affine_28"], "clamp_min_86": ["quantize_affine_28"], "clamp_max_57": ["quantize_affine_28"], "convert_element_type_169": ["quantize_affine_28", "dequantize_affine_56"], "view_447": ["dequantize_affine_56"], "convert_element_type_170": ["dequantize_affine_56"], "convert_element_type_171": ["dequantize_affine_56"], "sub_1324": ["dequantize_affine_56"], "view_446": ["dequantize_affine_56"], "mul_2809": ["dequantize_affine_56"], "view_449": ["dequantize_affine_57"], "view_451": ["dequantize_affine_57"], "convert_element_type_172": ["dequantize_affine_57"], "convert_element_type_173": ["dequantize_affine_57"], "sub_1328": ["dequantize_affine_57"], "view_450": ["dequantize_affine_57"], "mul_2814": ["dequantize_affine_57"], "view_452": ["dequantize_affine_57"], "mul_2819": ["linear_28"], "view_453": ["linear_28"], "permute_49": ["linear_28"], "mm_default_136": ["linear_28"], "add_tensor_136": ["linear_28"], "view_454": ["linear_28"], "mul_2826": ["gelu_6"], "mul_2827": ["gelu_6"], "erf_6": ["gelu_6"], "add_4470": ["gelu_6"], "mul_2828": ["gelu_6", "dropout_14", "choose_qparams_affine_default_29", "quantize_affine_29"], "amax_29": ["choose_qparams_affine_default_29"], "full_59": ["choose_qparams_affine_default_29"], "maximum_29": ["choose_qparams_affine_default_29"], "amin_29": ["choose_qparams_affine_default_29"], "full_58": ["choose_qparams_affine_default_29"], "minimum_29": ["choose_qparams_affine_default_29"], "sub_1341": ["choose_qparams_affine_default_29"], "div_58": ["choose_qparams_affine_default_29"], "clamp_min_87": ["choose_qparams_affine_default_29"], "div_59": ["choose_qparams_affine_default_29"], "round_59": ["choose_qparams_affine_default_29"], "sub_1347": ["choose_qparams_affine_default_29"], "clamp_min_88": ["choose_qparams_affine_default_29"], "clamp_max_58": ["choose_qparams_affine_default_29"], "view_457": ["quantize_affine_29"], "reciprocal_29": ["quantize_affine_29"], "mul_2874": ["quantize_affine_29"], "mul_2877": ["quantize_affine_29"], "round_60": ["quantize_affine_29"], "convert_element_type_174": ["choose_qparams_affine_default_29"], "view_458": ["quantize_affine_29"], "add_4553": ["quantize_affine_29"], "clamp_min_89": ["quantize_affine_29"], "clamp_max_59": ["quantize_affine_29"], "convert_element_type_175": ["quantize_affine_29", "dequantize_affine_58"], "view_462": ["dequantize_affine_58"], "convert_element_type_176": ["dequantize_affine_58"], "convert_element_type_177": ["dequantize_affine_58"], "sub_1367": ["dequantize_affine_58"], "view_461": ["dequantize_affine_58"], "mul_2899": ["dequantize_affine_58"], "view_464": ["dequantize_affine_59"], "view_466": ["dequantize_affine_59"], "convert_element_type_178": ["dequantize_affine_59"], "convert_element_type_179": ["dequantize_affine_59"], "sub_1371": ["dequantize_affine_59"], "view_465": ["dequantize_affine_59"], "mul_2904": ["dequantize_affine_59"], "view_467": ["dequantize_affine_59"], "mul_2909": ["linear_29"], "view_468": ["linear_29"], "permute_50": ["linear_29"], "mm_default_135": ["linear_29"], "add_tensor_135": ["linear_29"], "view_469": ["linear_29", "dropout_15"], "add_4616": ["add_70"], "clone_41": ["layer_norm_10"], "var_mean_10": ["layer_norm_10"], "getitem_40": ["layer_norm_10"], "getitem_41": ["layer_norm_10"], "sub_1377": ["layer_norm_10"], "add_4621": ["layer_norm_10"], "rsqrt_10": ["layer_norm_10"], "mul_2920": ["layer_norm_10"], "mul_2921": ["layer_norm_10"], "add_4622": ["layer_norm_10", "choose_qparams_affine_default_30", "quantize_affine_30", "choose_qparams_affine_default_31", "quantize_affine_31", "choose_qparams_affine_default_32", "quantize_affine_32"], "amax_30": ["choose_qparams_affine_default_30"], "full_61": ["choose_qparams_affine_default_30"], "maximum_30": ["choose_qparams_affine_default_30"], "amin_30": ["choose_qparams_affine_default_30"], "full_60": ["choose_qparams_affine_default_30"], "minimum_30": ["choose_qparams_affine_default_30"], "sub_1388": ["choose_qparams_affine_default_30"], "div_60": ["choose_qparams_affine_default_30"], "clamp_min_90": ["choose_qparams_affine_default_30"], "div_61": ["choose_qparams_affine_default_30"], "round_61": ["choose_qparams_affine_default_30"], "sub_1394": ["choose_qparams_affine_default_30"], "clamp_min_91": ["choose_qparams_affine_default_30"], "clamp_max_60": ["choose_qparams_affine_default_30"], "view_472": ["quantize_affine_30"], "reciprocal_30": ["quantize_affine_30"], "mul_2969": ["quantize_affine_30"], "mul_2972": ["quantize_affine_30"], "round_62": ["quantize_affine_30"], "convert_element_type_180": ["choose_qparams_affine_default_30"], "view_473": ["quantize_affine_30"], "add_4709": ["quantize_affine_30"], "clamp_min_92": ["quantize_affine_30"], "clamp_max_61": ["quantize_affine_30"], "convert_element_type_181": ["quantize_affine_30", "dequantize_affine_60"], "view_477": ["dequantize_affine_60"], "convert_element_type_182": ["dequantize_affine_60"], "convert_element_type_183": ["dequantize_affine_60"], "sub_1414": ["dequantize_affine_60"], "view_476": ["dequantize_affine_60"], "mul_2994": ["dequantize_affine_60"], "view_479": ["dequantize_affine_61"], "view_481": ["dequantize_affine_61"], "convert_element_type_184": ["dequantize_affine_61"], "convert_element_type_185": ["dequantize_affine_61"], "sub_1418": ["dequantize_affine_61"], "view_480": ["dequantize_affine_61"], "mul_2999": ["dequantize_affine_61"], "view_482": ["dequantize_affine_61"], "amax_31": ["choose_qparams_affine_default_31"], "full_63": ["choose_qparams_affine_default_31"], "maximum_31": ["choose_qparams_affine_default_31"], "amin_31": ["choose_qparams_affine_default_31"], "full_62": ["choose_qparams_affine_default_31"], "minimum_31": ["choose_qparams_affine_default_31"], "sub_1433": ["choose_qparams_affine_default_31"], "div_62": ["choose_qparams_affine_default_31"], "clamp_min_93": ["choose_qparams_affine_default_31"], "div_63": ["choose_qparams_affine_default_31"], "round_63": ["choose_qparams_affine_default_31"], "sub_1439": ["choose_qparams_affine_default_31"], "clamp_min_94": ["choose_qparams_affine_default_31"], "clamp_max_62": ["choose_qparams_affine_default_31"], "view_488": ["quantize_affine_31"], "reciprocal_31": ["quantize_affine_31"], "mul_3065": ["quantize_affine_31"], "mul_3068": ["quantize_affine_31"], "round_64": ["quantize_affine_31"], "convert_element_type_186": ["choose_qparams_affine_default_31"], "view_489": ["quantize_affine_31"], "add_4861": ["quantize_affine_31"], "clamp_min_95": ["quantize_affine_31"], "clamp_max_63": ["quantize_affine_31"], "convert_element_type_187": ["quantize_affine_31", "dequantize_affine_62"], "view_493": ["dequantize_affine_62"], "convert_element_type_188": ["dequantize_affine_62"], "convert_element_type_189": ["dequantize_affine_62"], "sub_1459": ["dequantize_affine_62"], "view_492": ["dequantize_affine_62"], "mul_3090": ["dequantize_affine_62"], "view_495": ["dequantize_affine_63"], "view_497": ["dequantize_affine_63"], "convert_element_type_190": ["dequantize_affine_63"], "convert_element_type_191": ["dequantize_affine_63"], "sub_1463": ["dequantize_affine_63"], "view_496": ["dequantize_affine_63"], "mul_3095": ["dequantize_affine_63"], "view_498": ["dequantize_affine_63"], "amax_32": ["choose_qparams_affine_default_32"], "full_65": ["choose_qparams_affine_default_32"], "maximum_32": ["choose_qparams_affine_default_32"], "amin_32": ["choose_qparams_affine_default_32"], "full_64": ["choose_qparams_affine_default_32"], "minimum_32": ["choose_qparams_affine_default_32"], "sub_1477": ["choose_qparams_affine_default_32"], "div_64": ["choose_qparams_affine_default_32"], "clamp_min_96": ["choose_qparams_affine_default_32"], "div_65": ["choose_qparams_affine_default_32"], "round_65": ["choose_qparams_affine_default_32"], "sub_1483": ["choose_qparams_affine_default_32"], "clamp_min_97": ["choose_qparams_affine_default_32"], "clamp_max_64": ["choose_qparams_affine_default_32"], "view_504": ["quantize_affine_32"], "reciprocal_32": ["quantize_affine_32"], "mul_3164": ["quantize_affine_32"], "mul_3167": ["quantize_affine_32"], "round_66": ["quantize_affine_32"], "convert_element_type_192": ["choose_qparams_affine_default_32"], "view_505": ["quantize_affine_32"], "add_5009": ["quantize_affine_32"], "clamp_min_98": ["quantize_affine_32"], "clamp_max_65": ["quantize_affine_32"], "convert_element_type_193": ["quantize_affine_32", "dequantize_affine_64"], "view_509": ["dequantize_affine_64"], "convert_element_type_194": ["dequantize_affine_64"], "convert_element_type_195": ["dequantize_affine_64"], "sub_1503": ["dequantize_affine_64"], "view_508": ["dequantize_affine_64"], "mul_3189": ["dequantize_affine_64"], "view_511": ["dequantize_affine_65"], "view_513": ["dequantize_affine_65"], "convert_element_type_196": ["dequantize_affine_65"], "convert_element_type_197": ["dequantize_affine_65"], "sub_1507": ["dequantize_affine_65"], "view_512": ["dequantize_affine_65"], "mul_3194": ["dequantize_affine_65"], "view_514": ["dequantize_affine_65"], "mul_3004": ["linear_30"], "view_483": ["linear_30"], "permute_51": ["linear_30"], "mm_default_134": ["linear_30"], "add_tensor_134": ["linear_30"], "view_484": ["linear_30"], "mul_3011": ["mul_190"], "view_485": ["view_15"], "permute_52": ["transpose_20"], "clone_42": ["contiguous_20"], "mul_3098": ["linear_31"], "view_499": ["linear_31"], "permute_53": ["linear_31"], "mm_5": ["linear_31"], "view_500": ["linear_31"], "view_501": ["view_16"], "permute_54": ["transpose_21"], "clone_43": ["contiguous_21"], "mul_3199": ["linear_32"], "view_515": ["linear_32"], "permute_55": ["linear_32"], "mm_default_133": ["linear_32"], "add_tensor_133": ["linear_32"], "view_516": ["linear_32"], "view_517": ["view_17"], "permute_56": ["transpose_22"], "clone_44": ["contiguous_22"], "_scaled_dot_product_efficient_attention_5": ["scaled_dot_product_attention_5"], "getitem_42": ["scaled_dot_product_attention_5"], "permute_57": ["transpose_23"], "view_518": ["reshape_5", "choose_qparams_affine_default_33", "quantize_affine_33"], "amax_33": ["choose_qparams_affine_default_33"], "full_67": ["choose_qparams_affine_default_33"], "maximum_33": ["choose_qparams_affine_default_33"], "amin_33": ["choose_qparams_affine_default_33"], "full_66": ["choose_qparams_affine_default_33"], "minimum_33": ["choose_qparams_affine_default_33"], "sub_1525": ["choose_qparams_affine_default_33"], "div_66": ["choose_qparams_affine_default_33"], "clamp_min_99": ["choose_qparams_affine_default_33"], "div_67": ["choose_qparams_affine_default_33"], "round_67": ["choose_qparams_affine_default_33"], "sub_1531": ["choose_qparams_affine_default_33"], "clamp_min_100": ["choose_qparams_affine_default_33"], "clamp_max_66": ["choose_qparams_affine_default_33"], "view_521": ["quantize_affine_33"], "reciprocal_33": ["quantize_affine_33"], "mul_3269": ["quantize_affine_33"], "mul_3272": ["quantize_affine_33"], "round_68": ["quantize_affine_33"], "convert_element_type_198": ["choose_qparams_affine_default_33"], "view_522": ["quantize_affine_33"], "add_5173": ["quantize_affine_33"], "clamp_min_101": ["quantize_affine_33"], "clamp_max_67": ["quantize_affine_33"], "convert_element_type_199": ["quantize_affine_33", "dequantize_affine_66"], "view_526": ["dequantize_affine_66"], "convert_element_type_200": ["dequantize_affine_66"], "convert_element_type_201": ["dequantize_affine_66"], "sub_1551": ["dequantize_affine_66"], "view_525": ["dequantize_affine_66"], "mul_3294": ["dequantize_affine_66"], "view_528": ["dequantize_affine_67"], "view_530": ["dequantize_affine_67"], "convert_element_type_202": ["dequantize_affine_67"], "convert_element_type_203": ["dequantize_affine_67"], "sub_1555": ["dequantize_affine_67"], "view_529": ["dequantize_affine_67"], "mul_3299": ["dequantize_affine_67"], "view_531": ["dequantize_affine_67"], "mul_3304": ["linear_33"], "view_532": ["linear_33"], "permute_58": ["linear_33"], "mm_default_132": ["linear_33"], "add_tensor_132": ["linear_33"], "view_533": ["linear_33", "dropout_16"], "add_5236": ["add_79"], "clone_46": ["layer_norm_11"], "var_mean_11": ["layer_norm_11"], "getitem_46": ["layer_norm_11"], "getitem_47": ["layer_norm_11"], "sub_1561": ["layer_norm_11"], "add_5241": ["layer_norm_11"], "rsqrt_11": ["layer_norm_11"], "mul_3315": ["layer_norm_11"], "mul_3316": ["layer_norm_11"], "add_5242": ["layer_norm_11", "choose_qparams_affine_default_34", "quantize_affine_34"], "amax_34": ["choose_qparams_affine_default_34"], "full_69": ["choose_qparams_affine_default_34"], "maximum_34": ["choose_qparams_affine_default_34"], "amin_34": ["choose_qparams_affine_default_34"], "full_68": ["choose_qparams_affine_default_34"], "minimum_34": ["choose_qparams_affine_default_34"], "sub_1572": ["choose_qparams_affine_default_34"], "div_68": ["choose_qparams_affine_default_34"], "clamp_min_102": ["choose_qparams_affine_default_34"], "div_69": ["choose_qparams_affine_default_34"], "round_69": ["choose_qparams_affine_default_34"], "sub_1578": ["choose_qparams_affine_default_34"], "clamp_min_103": ["choose_qparams_affine_default_34"], "clamp_max_68": ["choose_qparams_affine_default_34"], "view_536": ["quantize_affine_34"], "reciprocal_34": ["quantize_affine_34"], "mul_3364": ["quantize_affine_34"], "mul_3367": ["quantize_affine_34"], "round_70": ["quantize_affine_34"], "convert_element_type_204": ["choose_qparams_affine_default_34"], "view_537": ["quantize_affine_34"], "add_5329": ["quantize_affine_34"], "clamp_min_104": ["quantize_affine_34"], "clamp_max_69": ["quantize_affine_34"], "convert_element_type_205": ["quantize_affine_34", "dequantize_affine_68"], "view_541": ["dequantize_affine_68"], "convert_element_type_206": ["dequantize_affine_68"], "convert_element_type_207": ["dequantize_affine_68"], "sub_1598": ["dequantize_affine_68"], "view_540": ["dequantize_affine_68"], "mul_3389": ["dequantize_affine_68"], "view_543": ["dequantize_affine_69"], "view_545": ["dequantize_affine_69"], "convert_element_type_208": ["dequantize_affine_69"], "convert_element_type_209": ["dequantize_affine_69"], "sub_1602": ["dequantize_affine_69"], "view_544": ["dequantize_affine_69"], "mul_3394": ["dequantize_affine_69"], "view_546": ["dequantize_affine_69"], "mul_3399": ["linear_34"], "view_547": ["linear_34"], "permute_59": ["linear_34"], "mm_default_131": ["linear_34"], "add_tensor_131": ["linear_34"], "view_548": ["linear_34"], "mul_3406": ["gelu_7"], "mul_3407": ["gelu_7"], "erf_7": ["gelu_7"], "add_5388": ["gelu_7"], "mul_3408": ["gelu_7", "dropout_17", "choose_qparams_affine_default_35", "quantize_affine_35"], "amax_35": ["choose_qparams_affine_default_35"], "full_71": ["choose_qparams_affine_default_35"], "maximum_35": ["choose_qparams_affine_default_35"], "amin_35": ["choose_qparams_affine_default_35"], "full_70": ["choose_qparams_affine_default_35"], "minimum_35": ["choose_qparams_affine_default_35"], "sub_1615": ["choose_qparams_affine_default_35"], "div_70": ["choose_qparams_affine_default_35"], "clamp_min_105": ["choose_qparams_affine_default_35"], "div_71": ["choose_qparams_affine_default_35"], "round_71": ["choose_qparams_affine_default_35"], "sub_1621": ["choose_qparams_affine_default_35"], "clamp_min_106": ["choose_qparams_affine_default_35"], "clamp_max_70": ["choose_qparams_affine_default_35"], "view_551": ["quantize_affine_35"], "reciprocal_35": ["quantize_affine_35"], "mul_3454": ["quantize_affine_35"], "mul_3457": ["quantize_affine_35"], "round_72": ["quantize_affine_35"], "convert_element_type_210": ["choose_qparams_affine_default_35"], "view_552": ["quantize_affine_35"], "add_5471": ["quantize_affine_35"], "clamp_min_107": ["quantize_affine_35"], "clamp_max_71": ["quantize_affine_35"], "convert_element_type_211": ["quantize_affine_35", "dequantize_affine_70"], "view_556": ["dequantize_affine_70"], "convert_element_type_212": ["dequantize_affine_70"], "convert_element_type_213": ["dequantize_affine_70"], "sub_1641": ["dequantize_affine_70"], "view_555": ["dequantize_affine_70"], "mul_3479": ["dequantize_affine_70"], "view_558": ["dequantize_affine_71"], "view_560": ["dequantize_affine_71"], "convert_element_type_214": ["dequantize_affine_71"], "convert_element_type_215": ["dequantize_affine_71"], "sub_1645": ["dequantize_affine_71"], "view_559": ["dequantize_affine_71"], "mul_3484": ["dequantize_affine_71"], "view_561": ["dequantize_affine_71"], "mul_3489": ["linear_35"], "view_562": ["linear_35"], "permute_60": ["linear_35"], "mm_default_130": ["linear_35"], "add_tensor_130": ["linear_35"], "view_563": ["linear_35", "dropout_18"], "add_5534": ["add_84"], "clone_49": ["layer_norm_12"], "var_mean_12": ["layer_norm_12"], "getitem_48": ["layer_norm_12"], "getitem_49": ["layer_norm_12"], "sub_1651": ["layer_norm_12"], "add_5539": ["layer_norm_12"], "rsqrt_12": ["layer_norm_12"], "mul_3500": ["layer_norm_12"], "mul_3501": ["layer_norm_12"], "add_5540": ["layer_norm_12", "choose_qparams_affine_default_36", "quantize_affine_36", "choose_qparams_affine_default_37", "quantize_affine_37", "choose_qparams_affine_default_38", "quantize_affine_38"], "amax_36": ["choose_qparams_affine_default_36"], "full_73": ["choose_qparams_affine_default_36"], "maximum_36": ["choose_qparams_affine_default_36"], "amin_36": ["choose_qparams_affine_default_36"], "full_72": ["choose_qparams_affine_default_36"], "minimum_36": ["choose_qparams_affine_default_36"], "sub_1662": ["choose_qparams_affine_default_36"], "div_72": ["choose_qparams_affine_default_36"], "clamp_min_108": ["choose_qparams_affine_default_36"], "div_73": ["choose_qparams_affine_default_36"], "round_73": ["choose_qparams_affine_default_36"], "sub_1668": ["choose_qparams_affine_default_36"], "clamp_min_109": ["choose_qparams_affine_default_36"], "clamp_max_72": ["choose_qparams_affine_default_36"], "view_566": ["quantize_affine_36"], "reciprocal_36": ["quantize_affine_36"], "mul_3549": ["quantize_affine_36"], "mul_3552": ["quantize_affine_36"], "round_74": ["quantize_affine_36"], "convert_element_type_216": ["choose_qparams_affine_default_36"], "view_567": ["quantize_affine_36"], "add_5627": ["quantize_affine_36"], "clamp_min_110": ["quantize_affine_36"], "clamp_max_73": ["quantize_affine_36"], "convert_element_type_217": ["quantize_affine_36", "dequantize_affine_72"], "view_571": ["dequantize_affine_72"], "convert_element_type_218": ["dequantize_affine_72"], "convert_element_type_219": ["dequantize_affine_72"], "sub_1688": ["dequantize_affine_72"], "view_570": ["dequantize_affine_72"], "mul_3574": ["dequantize_affine_72"], "view_573": ["dequantize_affine_73"], "view_575": ["dequantize_affine_73"], "convert_element_type_220": ["dequantize_affine_73"], "convert_element_type_221": ["dequantize_affine_73"], "sub_1692": ["dequantize_affine_73"], "view_574": ["dequantize_affine_73"], "mul_3579": ["dequantize_affine_73"], "view_576": ["dequantize_affine_73"], "amax_37": ["choose_qparams_affine_default_37"], "full_75": ["choose_qparams_affine_default_37"], "maximum_37": ["choose_qparams_affine_default_37"], "amin_37": ["choose_qparams_affine_default_37"], "full_74": ["choose_qparams_affine_default_37"], "minimum_37": ["choose_qparams_affine_default_37"], "sub_1707": ["choose_qparams_affine_default_37"], "div_74": ["choose_qparams_affine_default_37"], "clamp_min_111": ["choose_qparams_affine_default_37"], "div_75": ["choose_qparams_affine_default_37"], "round_75": ["choose_qparams_affine_default_37"], "sub_1713": ["choose_qparams_affine_default_37"], "clamp_min_112": ["choose_qparams_affine_default_37"], "clamp_max_74": ["choose_qparams_affine_default_37"], "view_582": ["quantize_affine_37"], "reciprocal_37": ["quantize_affine_37"], "mul_3645": ["quantize_affine_37"], "mul_3648": ["quantize_affine_37"], "round_76": ["quantize_affine_37"], "convert_element_type_222": ["choose_qparams_affine_default_37"], "view_583": ["quantize_affine_37"], "add_5779": ["quantize_affine_37"], "clamp_min_113": ["quantize_affine_37"], "clamp_max_75": ["quantize_affine_37"], "convert_element_type_223": ["quantize_affine_37", "dequantize_affine_74"], "view_587": ["dequantize_affine_74"], "convert_element_type_224": ["dequantize_affine_74"], "convert_element_type_225": ["dequantize_affine_74"], "sub_1733": ["dequantize_affine_74"], "view_586": ["dequantize_affine_74"], "mul_3670": ["dequantize_affine_74"], "view_589": ["dequantize_affine_75"], "view_591": ["dequantize_affine_75"], "convert_element_type_226": ["dequantize_affine_75"], "convert_element_type_227": ["dequantize_affine_75"], "sub_1737": ["dequantize_affine_75"], "view_590": ["dequantize_affine_75"], "mul_3675": ["dequantize_affine_75"], "view_592": ["dequantize_affine_75"], "amax_38": ["choose_qparams_affine_default_38"], "full_77": ["choose_qparams_affine_default_38"], "maximum_38": ["choose_qparams_affine_default_38"], "amin_38": ["choose_qparams_affine_default_38"], "full_76": ["choose_qparams_affine_default_38"], "minimum_38": ["choose_qparams_affine_default_38"], "sub_1751": ["choose_qparams_affine_default_38"], "div_76": ["choose_qparams_affine_default_38"], "clamp_min_114": ["choose_qparams_affine_default_38"], "div_77": ["choose_qparams_affine_default_38"], "round_77": ["choose_qparams_affine_default_38"], "sub_1757": ["choose_qparams_affine_default_38"], "clamp_min_115": ["choose_qparams_affine_default_38"], "clamp_max_76": ["choose_qparams_affine_default_38"], "view_598": ["quantize_affine_38"], "reciprocal_38": ["quantize_affine_38"], "mul_3744": ["quantize_affine_38"], "mul_3747": ["quantize_affine_38"], "round_78": ["quantize_affine_38"], "convert_element_type_228": ["choose_qparams_affine_default_38"], "view_599": ["quantize_affine_38"], "add_5927": ["quantize_affine_38"], "clamp_min_116": ["quantize_affine_38"], "clamp_max_77": ["quantize_affine_38"], "convert_element_type_229": ["quantize_affine_38", "dequantize_affine_76"], "view_603": ["dequantize_affine_76"], "convert_element_type_230": ["dequantize_affine_76"], "convert_element_type_231": ["dequantize_affine_76"], "sub_1777": ["dequantize_affine_76"], "view_602": ["dequantize_affine_76"], "mul_3769": ["dequantize_affine_76"], "view_605": ["dequantize_affine_77"], "view_607": ["dequantize_affine_77"], "convert_element_type_232": ["dequantize_affine_77"], "convert_element_type_233": ["dequantize_affine_77"], "sub_1781": ["dequantize_affine_77"], "view_606": ["dequantize_affine_77"], "mul_3774": ["dequantize_affine_77"], "view_608": ["dequantize_affine_77"], "mul_3584": ["linear_36"], "view_577": ["linear_36"], "permute_61": ["linear_36"], "mm_default_129": ["linear_36"], "add_tensor_129": ["linear_36"], "view_578": ["linear_36"], "mul_3591": ["mul_227"], "view_579": ["view_18"], "permute_62": ["transpose_24"], "clone_50": ["contiguous_24"], "mul_3678": ["linear_37"], "view_593": ["linear_37"], "permute_63": ["linear_37"], "mm_6": ["linear_37"], "view_594": ["linear_37"], "view_595": ["view_19"], "permute_64": ["transpose_25"], "clone_51": ["contiguous_25"], "mul_3779": ["linear_38"], "view_609": ["linear_38"], "permute_65": ["linear_38"], "mm_default_128": ["linear_38"], "add_tensor_128": ["linear_38"], "view_610": ["linear_38"], "view_611": ["view_20"], "permute_66": ["transpose_26"], "clone_52": ["contiguous_26"], "_scaled_dot_product_efficient_attention_6": ["scaled_dot_product_attention_6"], "getitem_50": ["scaled_dot_product_attention_6"], "permute_67": ["transpose_27"], "view_612": ["reshape_6", "choose_qparams_affine_default_39", "quantize_affine_39"], "amax_39": ["choose_qparams_affine_default_39"], "full_79": ["choose_qparams_affine_default_39"], "maximum_39": ["choose_qparams_affine_default_39"], "amin_39": ["choose_qparams_affine_default_39"], "full_78": ["choose_qparams_affine_default_39"], "minimum_39": ["choose_qparams_affine_default_39"], "sub_1799": ["choose_qparams_affine_default_39"], "div_78": ["choose_qparams_affine_default_39"], "clamp_min_117": ["choose_qparams_affine_default_39"], "div_79": ["choose_qparams_affine_default_39"], "round_79": ["choose_qparams_affine_default_39"], "sub_1805": ["choose_qparams_affine_default_39"], "clamp_min_118": ["choose_qparams_affine_default_39"], "clamp_max_78": ["choose_qparams_affine_default_39"], "view_615": ["quantize_affine_39"], "reciprocal_39": ["quantize_affine_39"], "mul_3849": ["quantize_affine_39"], "mul_3852": ["quantize_affine_39"], "round_80": ["quantize_affine_39"], "convert_element_type_234": ["choose_qparams_affine_default_39"], "view_616": ["quantize_affine_39"], "add_6091": ["quantize_affine_39"], "clamp_min_119": ["quantize_affine_39"], "clamp_max_79": ["quantize_affine_39"], "convert_element_type_235": ["quantize_affine_39", "dequantize_affine_78"], "view_620": ["dequantize_affine_78"], "convert_element_type_236": ["dequantize_affine_78"], "convert_element_type_237": ["dequantize_affine_78"], "sub_1825": ["dequantize_affine_78"], "view_619": ["dequantize_affine_78"], "mul_3874": ["dequantize_affine_78"], "view_622": ["dequantize_affine_79"], "view_624": ["dequantize_affine_79"], "convert_element_type_238": ["dequantize_affine_79"], "convert_element_type_239": ["dequantize_affine_79"], "sub_1829": ["dequantize_affine_79"], "view_623": ["dequantize_affine_79"], "mul_3879": ["dequantize_affine_79"], "view_625": ["dequantize_affine_79"], "mul_3884": ["linear_39"], "view_626": ["linear_39"], "permute_68": ["linear_39"], "mm_default_127": ["linear_39"], "add_tensor_127": ["linear_39"], "view_627": ["linear_39", "dropout_19"], "add_6154": ["add_93"], "clone_54": ["layer_norm_13"], "var_mean_13": ["layer_norm_13"], "getitem_54": ["layer_norm_13"], "getitem_55": ["layer_norm_13"], "sub_1835": ["layer_norm_13"], "add_6159": ["layer_norm_13"], "rsqrt_13": ["layer_norm_13"], "mul_3895": ["layer_norm_13"], "mul_3896": ["layer_norm_13"], "add_6160": ["layer_norm_13", "choose_qparams_affine_default_40", "quantize_affine_40"], "amax_40": ["choose_qparams_affine_default_40"], "full_81": ["choose_qparams_affine_default_40"], "maximum_40": ["choose_qparams_affine_default_40"], "amin_40": ["choose_qparams_affine_default_40"], "full_80": ["choose_qparams_affine_default_40"], "minimum_40": ["choose_qparams_affine_default_40"], "sub_1846": ["choose_qparams_affine_default_40"], "div_80": ["choose_qparams_affine_default_40"], "clamp_min_120": ["choose_qparams_affine_default_40"], "div_81": ["choose_qparams_affine_default_40"], "round_81": ["choose_qparams_affine_default_40"], "sub_1852": ["choose_qparams_affine_default_40"], "clamp_min_121": ["choose_qparams_affine_default_40"], "clamp_max_80": ["choose_qparams_affine_default_40"], "view_630": ["quantize_affine_40"], "reciprocal_40": ["quantize_affine_40"], "mul_3944": ["quantize_affine_40"], "mul_3947": ["quantize_affine_40"], "round_82": ["quantize_affine_40"], "convert_element_type_240": ["choose_qparams_affine_default_40"], "view_631": ["quantize_affine_40"], "add_6247": ["quantize_affine_40"], "clamp_min_122": ["quantize_affine_40"], "clamp_max_81": ["quantize_affine_40"], "convert_element_type_241": ["quantize_affine_40", "dequantize_affine_80"], "view_635": ["dequantize_affine_80"], "convert_element_type_242": ["dequantize_affine_80"], "convert_element_type_243": ["dequantize_affine_80"], "sub_1872": ["dequantize_affine_80"], "view_634": ["dequantize_affine_80"], "mul_3969": ["dequantize_affine_80"], "view_637": ["dequantize_affine_81"], "view_639": ["dequantize_affine_81"], "convert_element_type_244": ["dequantize_affine_81"], "convert_element_type_245": ["dequantize_affine_81"], "sub_1876": ["dequantize_affine_81"], "view_638": ["dequantize_affine_81"], "mul_3974": ["dequantize_affine_81"], "view_640": ["dequantize_affine_81"], "mul_3979": ["linear_40"], "view_641": ["linear_40"], "permute_69": ["linear_40"], "mm_default_126": ["linear_40"], "add_tensor_126": ["linear_40"], "view_642": ["linear_40"], "mul_3986": ["gelu_8"], "mul_3987": ["gelu_8"], "erf_8": ["gelu_8"], "add_6306": ["gelu_8"], "mul_3988": ["gelu_8", "dropout_20", "choose_qparams_affine_default_41", "quantize_affine_41"], "amax_41": ["choose_qparams_affine_default_41"], "full_83": ["choose_qparams_affine_default_41"], "maximum_41": ["choose_qparams_affine_default_41"], "amin_41": ["choose_qparams_affine_default_41"], "full_82": ["choose_qparams_affine_default_41"], "minimum_41": ["choose_qparams_affine_default_41"], "sub_1889": ["choose_qparams_affine_default_41"], "div_82": ["choose_qparams_affine_default_41"], "clamp_min_123": ["choose_qparams_affine_default_41"], "div_83": ["choose_qparams_affine_default_41"], "round_83": ["choose_qparams_affine_default_41"], "sub_1895": ["choose_qparams_affine_default_41"], "clamp_min_124": ["choose_qparams_affine_default_41"], "clamp_max_82": ["choose_qparams_affine_default_41"], "view_645": ["quantize_affine_41"], "reciprocal_41": ["quantize_affine_41"], "mul_4034": ["quantize_affine_41"], "mul_4037": ["quantize_affine_41"], "round_84": ["quantize_affine_41"], "convert_element_type_246": ["choose_qparams_affine_default_41"], "view_646": ["quantize_affine_41"], "add_6389": ["quantize_affine_41"], "clamp_min_125": ["quantize_affine_41"], "clamp_max_83": ["quantize_affine_41"], "convert_element_type_247": ["quantize_affine_41", "dequantize_affine_82"], "view_650": ["dequantize_affine_82"], "convert_element_type_248": ["dequantize_affine_82"], "convert_element_type_249": ["dequantize_affine_82"], "sub_1915": ["dequantize_affine_82"], "view_649": ["dequantize_affine_82"], "mul_4059": ["dequantize_affine_82"], "view_652": ["dequantize_affine_83"], "view_654": ["dequantize_affine_83"], "convert_element_type_250": ["dequantize_affine_83"], "convert_element_type_251": ["dequantize_affine_83"], "sub_1919": ["dequantize_affine_83"], "view_653": ["dequantize_affine_83"], "mul_4064": ["dequantize_affine_83"], "view_655": ["dequantize_affine_83"], "mul_4069": ["linear_41"], "view_656": ["linear_41"], "permute_70": ["linear_41"], "mm_default_125": ["linear_41"], "add_tensor_125": ["linear_41"], "view_657": ["linear_41", "dropout_21"], "add_6452": ["add_98"], "clone_57": ["layer_norm_14"], "var_mean_14": ["layer_norm_14"], "getitem_56": ["layer_norm_14"], "getitem_57": ["layer_norm_14"], "sub_1925": ["layer_norm_14"], "add_6457": ["layer_norm_14"], "rsqrt_14": ["layer_norm_14"], "mul_4080": ["layer_norm_14"], "mul_4081": ["layer_norm_14"], "add_6458": ["layer_norm_14", "choose_qparams_affine_default_42", "quantize_affine_42", "choose_qparams_affine_default_43", "quantize_affine_43", "choose_qparams_affine_default_44", "quantize_affine_44"], "amax_42": ["choose_qparams_affine_default_42"], "full_85": ["choose_qparams_affine_default_42"], "maximum_42": ["choose_qparams_affine_default_42"], "amin_42": ["choose_qparams_affine_default_42"], "full_84": ["choose_qparams_affine_default_42"], "minimum_42": ["choose_qparams_affine_default_42"], "sub_1936": ["choose_qparams_affine_default_42"], "div_84": ["choose_qparams_affine_default_42"], "clamp_min_126": ["choose_qparams_affine_default_42"], "div_85": ["choose_qparams_affine_default_42"], "round_85": ["choose_qparams_affine_default_42"], "sub_1942": ["choose_qparams_affine_default_42"], "clamp_min_127": ["choose_qparams_affine_default_42"], "clamp_max_84": ["choose_qparams_affine_default_42"], "view_660": ["quantize_affine_42"], "reciprocal_42": ["quantize_affine_42"], "mul_4129": ["quantize_affine_42"], "mul_4132": ["quantize_affine_42"], "round_86": ["quantize_affine_42"], "convert_element_type_252": ["choose_qparams_affine_default_42"], "view_661": ["quantize_affine_42"], "add_6545": ["quantize_affine_42"], "clamp_min_128": ["quantize_affine_42"], "clamp_max_85": ["quantize_affine_42"], "convert_element_type_253": ["quantize_affine_42", "dequantize_affine_84"], "view_665": ["dequantize_affine_84"], "convert_element_type_254": ["dequantize_affine_84"], "convert_element_type_255": ["dequantize_affine_84"], "sub_1962": ["dequantize_affine_84"], "view_664": ["dequantize_affine_84"], "mul_4154": ["dequantize_affine_84"], "view_667": ["dequantize_affine_85"], "view_669": ["dequantize_affine_85"], "convert_element_type_256": ["dequantize_affine_85"], "convert_element_type_257": ["dequantize_affine_85"], "sub_1966": ["dequantize_affine_85"], "view_668": ["dequantize_affine_85"], "mul_4159": ["dequantize_affine_85"], "view_670": ["dequantize_affine_85"], "amax_43": ["choose_qparams_affine_default_43"], "full_87": ["choose_qparams_affine_default_43"], "maximum_43": ["choose_qparams_affine_default_43"], "amin_43": ["choose_qparams_affine_default_43"], "full_86": ["choose_qparams_affine_default_43"], "minimum_43": ["choose_qparams_affine_default_43"], "sub_1981": ["choose_qparams_affine_default_43"], "div_86": ["choose_qparams_affine_default_43"], "clamp_min_129": ["choose_qparams_affine_default_43"], "div_87": ["choose_qparams_affine_default_43"], "round_87": ["choose_qparams_affine_default_43"], "sub_1987": ["choose_qparams_affine_default_43"], "clamp_min_130": ["choose_qparams_affine_default_43"], "clamp_max_86": ["choose_qparams_affine_default_43"], "view_676": ["quantize_affine_43"], "reciprocal_43": ["quantize_affine_43"], "mul_4225": ["quantize_affine_43"], "mul_4228": ["quantize_affine_43"], "round_88": ["quantize_affine_43"], "convert_element_type_258": ["choose_qparams_affine_default_43"], "view_677": ["quantize_affine_43"], "add_6697": ["quantize_affine_43"], "clamp_min_131": ["quantize_affine_43"], "clamp_max_87": ["quantize_affine_43"], "convert_element_type_259": ["quantize_affine_43", "dequantize_affine_86"], "view_681": ["dequantize_affine_86"], "convert_element_type_260": ["dequantize_affine_86"], "convert_element_type_261": ["dequantize_affine_86"], "sub_2007": ["dequantize_affine_86"], "view_680": ["dequantize_affine_86"], "mul_4250": ["dequantize_affine_86"], "view_683": ["dequantize_affine_87"], "view_685": ["dequantize_affine_87"], "convert_element_type_262": ["dequantize_affine_87"], "convert_element_type_263": ["dequantize_affine_87"], "sub_2011": ["dequantize_affine_87"], "view_684": ["dequantize_affine_87"], "mul_4255": ["dequantize_affine_87"], "view_686": ["dequantize_affine_87"], "amax_44": ["choose_qparams_affine_default_44"], "full_89": ["choose_qparams_affine_default_44"], "maximum_44": ["choose_qparams_affine_default_44"], "amin_44": ["choose_qparams_affine_default_44"], "full_88": ["choose_qparams_affine_default_44"], "minimum_44": ["choose_qparams_affine_default_44"], "sub_2025": ["choose_qparams_affine_default_44"], "div_88": ["choose_qparams_affine_default_44"], "clamp_min_132": ["choose_qparams_affine_default_44"], "div_89": ["choose_qparams_affine_default_44"], "round_89": ["choose_qparams_affine_default_44"], "sub_2031": ["choose_qparams_affine_default_44"], "clamp_min_133": ["choose_qparams_affine_default_44"], "clamp_max_88": ["choose_qparams_affine_default_44"], "view_692": ["quantize_affine_44"], "reciprocal_44": ["quantize_affine_44"], "mul_4324": ["quantize_affine_44"], "mul_4327": ["quantize_affine_44"], "round_90": ["quantize_affine_44"], "convert_element_type_264": ["choose_qparams_affine_default_44"], "view_693": ["quantize_affine_44"], "add_6845": ["quantize_affine_44"], "clamp_min_134": ["quantize_affine_44"], "clamp_max_89": ["quantize_affine_44"], "convert_element_type_265": ["quantize_affine_44", "dequantize_affine_88"], "view_697": ["dequantize_affine_88"], "convert_element_type_266": ["dequantize_affine_88"], "convert_element_type_267": ["dequantize_affine_88"], "sub_2051": ["dequantize_affine_88"], "view_696": ["dequantize_affine_88"], "mul_4349": ["dequantize_affine_88"], "view_699": ["dequantize_affine_89"], "view_701": ["dequantize_affine_89"], "convert_element_type_268": ["dequantize_affine_89"], "convert_element_type_269": ["dequantize_affine_89"], "sub_2055": ["dequantize_affine_89"], "view_700": ["dequantize_affine_89"], "mul_4354": ["dequantize_affine_89"], "view_702": ["dequantize_affine_89"], "mul_4164": ["linear_42"], "view_671": ["linear_42"], "permute_71": ["linear_42"], "mm_default_124": ["linear_42"], "add_tensor_124": ["linear_42"], "view_672": ["linear_42"], "mul_4171": ["mul_264"], "view_673": ["view_21"], "permute_72": ["transpose_28"], "clone_58": ["contiguous_28"], "mul_4258": ["linear_43"], "view_687": ["linear_43"], "permute_73": ["linear_43"], "mm_7": ["linear_43"], "view_688": ["linear_43"], "view_689": ["view_22"], "permute_74": ["transpose_29"], "clone_59": ["contiguous_29"], "mul_4359": ["linear_44"], "view_703": ["linear_44"], "permute_75": ["linear_44"], "mm_default_123": ["linear_44"], "add_tensor_123": ["linear_44"], "view_704": ["linear_44"], "view_705": ["view_23"], "permute_76": ["transpose_30"], "clone_60": ["contiguous_30"], "_scaled_dot_product_efficient_attention_7": ["scaled_dot_product_attention_7"], "getitem_58": ["scaled_dot_product_attention_7"], "permute_77": ["transpose_31"], "view_706": ["reshape_7", "choose_qparams_affine_default_45", "quantize_affine_45"], "amax_45": ["choose_qparams_affine_default_45"], "full_91": ["choose_qparams_affine_default_45"], "maximum_45": ["choose_qparams_affine_default_45"], "amin_45": ["choose_qparams_affine_default_45"], "full_90": ["choose_qparams_affine_default_45"], "minimum_45": ["choose_qparams_affine_default_45"], "sub_2073": ["choose_qparams_affine_default_45"], "div_90": ["choose_qparams_affine_default_45"], "clamp_min_135": ["choose_qparams_affine_default_45"], "div_91": ["choose_qparams_affine_default_45"], "round_91": ["choose_qparams_affine_default_45"], "sub_2079": ["choose_qparams_affine_default_45"], "clamp_min_136": ["choose_qparams_affine_default_45"], "clamp_max_90": ["choose_qparams_affine_default_45"], "view_709": ["quantize_affine_45"], "reciprocal_45": ["quantize_affine_45"], "mul_4429": ["quantize_affine_45"], "mul_4432": ["quantize_affine_45"], "round_92": ["quantize_affine_45"], "convert_element_type_270": ["choose_qparams_affine_default_45"], "view_710": ["quantize_affine_45"], "add_7009": ["quantize_affine_45"], "clamp_min_137": ["quantize_affine_45"], "clamp_max_91": ["quantize_affine_45"], "convert_element_type_271": ["quantize_affine_45", "dequantize_affine_90"], "view_714": ["dequantize_affine_90"], "convert_element_type_272": ["dequantize_affine_90"], "convert_element_type_273": ["dequantize_affine_90"], "sub_2099": ["dequantize_affine_90"], "view_713": ["dequantize_affine_90"], "mul_4454": ["dequantize_affine_90"], "view_716": ["dequantize_affine_91"], "view_718": ["dequantize_affine_91"], "convert_element_type_274": ["dequantize_affine_91"], "convert_element_type_275": ["dequantize_affine_91"], "sub_2103": ["dequantize_affine_91"], "view_717": ["dequantize_affine_91"], "mul_4459": ["dequantize_affine_91"], "view_719": ["dequantize_affine_91"], "mul_4464": ["linear_45"], "view_720": ["linear_45"], "permute_78": ["linear_45"], "mm_default_122": ["linear_45"], "add_tensor_122": ["linear_45"], "view_721": ["linear_45", "dropout_22"], "add_7072": ["add_107"], "clone_62": ["layer_norm_15"], "var_mean_15": ["layer_norm_15"], "getitem_62": ["layer_norm_15"], "getitem_63": ["layer_norm_15"], "sub_2109": ["layer_norm_15"], "add_7077": ["layer_norm_15"], "rsqrt_15": ["layer_norm_15"], "mul_4475": ["layer_norm_15"], "mul_4476": ["layer_norm_15"], "add_7078": ["layer_norm_15", "choose_qparams_affine_default_46", "quantize_affine_46"], "amax_46": ["choose_qparams_affine_default_46"], "full_93": ["choose_qparams_affine_default_46"], "maximum_46": ["choose_qparams_affine_default_46"], "amin_46": ["choose_qparams_affine_default_46"], "full_92": ["choose_qparams_affine_default_46"], "minimum_46": ["choose_qparams_affine_default_46"], "sub_2120": ["choose_qparams_affine_default_46"], "div_92": ["choose_qparams_affine_default_46"], "clamp_min_138": ["choose_qparams_affine_default_46"], "div_93": ["choose_qparams_affine_default_46"], "round_93": ["choose_qparams_affine_default_46"], "sub_2126": ["choose_qparams_affine_default_46"], "clamp_min_139": ["choose_qparams_affine_default_46"], "clamp_max_92": ["choose_qparams_affine_default_46"], "view_724": ["quantize_affine_46"], "reciprocal_46": ["quantize_affine_46"], "mul_4524": ["quantize_affine_46"], "mul_4527": ["quantize_affine_46"], "round_94": ["quantize_affine_46"], "convert_element_type_276": ["choose_qparams_affine_default_46"], "view_725": ["quantize_affine_46"], "add_7165": ["quantize_affine_46"], "clamp_min_140": ["quantize_affine_46"], "clamp_max_93": ["quantize_affine_46"], "convert_element_type_277": ["quantize_affine_46", "dequantize_affine_92"], "view_729": ["dequantize_affine_92"], "convert_element_type_278": ["dequantize_affine_92"], "convert_element_type_279": ["dequantize_affine_92"], "sub_2146": ["dequantize_affine_92"], "view_728": ["dequantize_affine_92"], "mul_4549": ["dequantize_affine_92"], "view_731": ["dequantize_affine_93"], "view_733": ["dequantize_affine_93"], "convert_element_type_280": ["dequantize_affine_93"], "convert_element_type_281": ["dequantize_affine_93"], "sub_2150": ["dequantize_affine_93"], "view_732": ["dequantize_affine_93"], "mul_4554": ["dequantize_affine_93"], "view_734": ["dequantize_affine_93"], "mul_4559": ["linear_46"], "view_735": ["linear_46"], "permute_79": ["linear_46"], "mm_default_121": ["linear_46"], "add_tensor_121": ["linear_46"], "view_736": ["linear_46"], "mul_4566": ["gelu_9"], "mul_4567": ["gelu_9"], "erf_9": ["gelu_9"], "add_7224": ["gelu_9"], "mul_4568": ["gelu_9", "dropout_23", "choose_qparams_affine_default_47", "quantize_affine_47"], "amax_47": ["choose_qparams_affine_default_47"], "full_95": ["choose_qparams_affine_default_47"], "maximum_47": ["choose_qparams_affine_default_47"], "amin_47": ["choose_qparams_affine_default_47"], "full_94": ["choose_qparams_affine_default_47"], "minimum_47": ["choose_qparams_affine_default_47"], "sub_2163": ["choose_qparams_affine_default_47"], "div_94": ["choose_qparams_affine_default_47"], "clamp_min_141": ["choose_qparams_affine_default_47"], "div_95": ["choose_qparams_affine_default_47"], "round_95": ["choose_qparams_affine_default_47"], "sub_2169": ["choose_qparams_affine_default_47"], "clamp_min_142": ["choose_qparams_affine_default_47"], "clamp_max_94": ["choose_qparams_affine_default_47"], "view_739": ["quantize_affine_47"], "reciprocal_47": ["quantize_affine_47"], "mul_4614": ["quantize_affine_47"], "mul_4617": ["quantize_affine_47"], "round_96": ["quantize_affine_47"], "convert_element_type_282": ["choose_qparams_affine_default_47"], "view_740": ["quantize_affine_47"], "add_7307": ["quantize_affine_47"], "clamp_min_143": ["quantize_affine_47"], "clamp_max_95": ["quantize_affine_47"], "convert_element_type_283": ["quantize_affine_47", "dequantize_affine_94"], "view_744": ["dequantize_affine_94"], "convert_element_type_284": ["dequantize_affine_94"], "convert_element_type_285": ["dequantize_affine_94"], "sub_2189": ["dequantize_affine_94"], "view_743": ["dequantize_affine_94"], "mul_4639": ["dequantize_affine_94"], "view_746": ["dequantize_affine_95"], "view_748": ["dequantize_affine_95"], "convert_element_type_286": ["dequantize_affine_95"], "convert_element_type_287": ["dequantize_affine_95"], "sub_2193": ["dequantize_affine_95"], "view_747": ["dequantize_affine_95"], "mul_4644": ["dequantize_affine_95"], "view_749": ["dequantize_affine_95"], "mul_4649": ["linear_47"], "view_750": ["linear_47"], "permute_80": ["linear_47"], "mm_default_120": ["linear_47"], "add_tensor_120": ["linear_47"], "view_751": ["linear_47", "dropout_24"], "add_7370": ["add_112"], "clone_65": ["layer_norm_16"], "var_mean_16": ["layer_norm_16"], "getitem_64": ["layer_norm_16"], "getitem_65": ["layer_norm_16"], "sub_2199": ["layer_norm_16"], "add_7375": ["layer_norm_16"], "rsqrt_16": ["layer_norm_16"], "mul_4660": ["layer_norm_16"], "mul_4661": ["layer_norm_16"], "add_7376": ["layer_norm_16", "choose_qparams_affine_default_48", "quantize_affine_48", "choose_qparams_affine_default_49", "quantize_affine_49", "choose_qparams_affine_default_50", "quantize_affine_50"], "amax_48": ["choose_qparams_affine_default_48"], "full_97": ["choose_qparams_affine_default_48"], "maximum_48": ["choose_qparams_affine_default_48"], "amin_48": ["choose_qparams_affine_default_48"], "full_96": ["choose_qparams_affine_default_48"], "minimum_48": ["choose_qparams_affine_default_48"], "sub_2210": ["choose_qparams_affine_default_48"], "div_96": ["choose_qparams_affine_default_48"], "clamp_min_144": ["choose_qparams_affine_default_48"], "div_97": ["choose_qparams_affine_default_48"], "round_97": ["choose_qparams_affine_default_48"], "sub_2216": ["choose_qparams_affine_default_48"], "clamp_min_145": ["choose_qparams_affine_default_48"], "clamp_max_96": ["choose_qparams_affine_default_48"], "view_754": ["quantize_affine_48"], "reciprocal_48": ["quantize_affine_48"], "mul_4709": ["quantize_affine_48"], "mul_4712": ["quantize_affine_48"], "round_98": ["quantize_affine_48"], "convert_element_type_288": ["choose_qparams_affine_default_48"], "view_755": ["quantize_affine_48"], "add_7463": ["quantize_affine_48"], "clamp_min_146": ["quantize_affine_48"], "clamp_max_97": ["quantize_affine_48"], "convert_element_type_289": ["quantize_affine_48", "dequantize_affine_96"], "view_759": ["dequantize_affine_96"], "convert_element_type_290": ["dequantize_affine_96"], "convert_element_type_291": ["dequantize_affine_96"], "sub_2236": ["dequantize_affine_96"], "view_758": ["dequantize_affine_96"], "mul_4734": ["dequantize_affine_96"], "view_761": ["dequantize_affine_97"], "view_763": ["dequantize_affine_97"], "convert_element_type_292": ["dequantize_affine_97"], "convert_element_type_293": ["dequantize_affine_97"], "sub_2240": ["dequantize_affine_97"], "view_762": ["dequantize_affine_97"], "mul_4739": ["dequantize_affine_97"], "view_764": ["dequantize_affine_97"], "amax_49": ["choose_qparams_affine_default_49"], "full_99": ["choose_qparams_affine_default_49"], "maximum_49": ["choose_qparams_affine_default_49"], "amin_49": ["choose_qparams_affine_default_49"], "full_98": ["choose_qparams_affine_default_49"], "minimum_49": ["choose_qparams_affine_default_49"], "sub_2255": ["choose_qparams_affine_default_49"], "div_98": ["choose_qparams_affine_default_49"], "clamp_min_147": ["choose_qparams_affine_default_49"], "div_99": ["choose_qparams_affine_default_49"], "round_99": ["choose_qparams_affine_default_49"], "sub_2261": ["choose_qparams_affine_default_49"], "clamp_min_148": ["choose_qparams_affine_default_49"], "clamp_max_98": ["choose_qparams_affine_default_49"], "view_770": ["quantize_affine_49"], "reciprocal_49": ["quantize_affine_49"], "mul_4805": ["quantize_affine_49"], "mul_4808": ["quantize_affine_49"], "round_100": ["quantize_affine_49"], "convert_element_type_294": ["choose_qparams_affine_default_49"], "view_771": ["quantize_affine_49"], "add_7615": ["quantize_affine_49"], "clamp_min_149": ["quantize_affine_49"], "clamp_max_99": ["quantize_affine_49"], "convert_element_type_295": ["quantize_affine_49", "dequantize_affine_98"], "view_775": ["dequantize_affine_98"], "convert_element_type_296": ["dequantize_affine_98"], "convert_element_type_297": ["dequantize_affine_98"], "sub_2281": ["dequantize_affine_98"], "view_774": ["dequantize_affine_98"], "mul_4830": ["dequantize_affine_98"], "view_777": ["dequantize_affine_99"], "view_779": ["dequantize_affine_99"], "convert_element_type_298": ["dequantize_affine_99"], "convert_element_type_299": ["dequantize_affine_99"], "sub_2285": ["dequantize_affine_99"], "view_778": ["dequantize_affine_99"], "mul_4835": ["dequantize_affine_99"], "view_780": ["dequantize_affine_99"], "amax_50": ["choose_qparams_affine_default_50"], "full_101": ["choose_qparams_affine_default_50"], "maximum_50": ["choose_qparams_affine_default_50"], "amin_50": ["choose_qparams_affine_default_50"], "full_100": ["choose_qparams_affine_default_50"], "minimum_50": ["choose_qparams_affine_default_50"], "sub_2299": ["choose_qparams_affine_default_50"], "div_100": ["choose_qparams_affine_default_50"], "clamp_min_150": ["choose_qparams_affine_default_50"], "div_101": ["choose_qparams_affine_default_50"], "round_101": ["choose_qparams_affine_default_50"], "sub_2305": ["choose_qparams_affine_default_50"], "clamp_min_151": ["choose_qparams_affine_default_50"], "clamp_max_100": ["choose_qparams_affine_default_50"], "view_786": ["quantize_affine_50"], "reciprocal_50": ["quantize_affine_50"], "mul_4904": ["quantize_affine_50"], "mul_4907": ["quantize_affine_50"], "round_102": ["quantize_affine_50"], "convert_element_type_300": ["choose_qparams_affine_default_50"], "view_787": ["quantize_affine_50"], "add_7763": ["quantize_affine_50"], "clamp_min_152": ["quantize_affine_50"], "clamp_max_101": ["quantize_affine_50"], "convert_element_type_301": ["quantize_affine_50", "dequantize_affine_100"], "view_791": ["dequantize_affine_100"], "convert_element_type_302": ["dequantize_affine_100"], "convert_element_type_303": ["dequantize_affine_100"], "sub_2325": ["dequantize_affine_100"], "view_790": ["dequantize_affine_100"], "mul_4929": ["dequantize_affine_100"], "view_793": ["dequantize_affine_101"], "view_795": ["dequantize_affine_101"], "convert_element_type_304": ["dequantize_affine_101"], "convert_element_type_305": ["dequantize_affine_101"], "sub_2329": ["dequantize_affine_101"], "view_794": ["dequantize_affine_101"], "mul_4934": ["dequantize_affine_101"], "view_796": ["dequantize_affine_101"], "mul_4744": ["linear_48"], "view_765": ["linear_48"], "permute_81": ["linear_48"], "mm_default_119": ["linear_48"], "add_tensor_119": ["linear_48"], "view_766": ["linear_48"], "mul_4751": ["mul_301"], "view_767": ["view_24"], "permute_82": ["transpose_32"], "clone_66": ["contiguous_32"], "mul_4838": ["linear_49"], "view_781": ["linear_49"], "permute_83": ["linear_49"], "mm_8": ["linear_49"], "view_782": ["linear_49"], "view_783": ["view_25"], "permute_84": ["transpose_33"], "clone_67": ["contiguous_33"], "mul_4939": ["linear_50"], "view_797": ["linear_50"], "permute_85": ["linear_50"], "mm_default_118": ["linear_50"], "add_tensor_118": ["linear_50"], "view_798": ["linear_50"], "view_799": ["view_26"], "permute_86": ["transpose_34"], "clone_68": ["contiguous_34"], "_scaled_dot_product_efficient_attention_8": ["scaled_dot_product_attention_8"], "getitem_66": ["scaled_dot_product_attention_8"], "permute_87": ["transpose_35"], "view_800": ["reshape_8", "choose_qparams_affine_default_51", "quantize_affine_51"], "amax_51": ["choose_qparams_affine_default_51"], "full_103": ["choose_qparams_affine_default_51"], "maximum_51": ["choose_qparams_affine_default_51"], "amin_51": ["choose_qparams_affine_default_51"], "full_102": ["choose_qparams_affine_default_51"], "minimum_51": ["choose_qparams_affine_default_51"], "sub_2347": ["choose_qparams_affine_default_51"], "div_102": ["choose_qparams_affine_default_51"], "clamp_min_153": ["choose_qparams_affine_default_51"], "div_103": ["choose_qparams_affine_default_51"], "round_103": ["choose_qparams_affine_default_51"], "sub_2353": ["choose_qparams_affine_default_51"], "clamp_min_154": ["choose_qparams_affine_default_51"], "clamp_max_102": ["choose_qparams_affine_default_51"], "view_803": ["quantize_affine_51"], "reciprocal_51": ["quantize_affine_51"], "mul_5009": ["quantize_affine_51"], "mul_5012": ["quantize_affine_51"], "round_104": ["quantize_affine_51"], "convert_element_type_306": ["choose_qparams_affine_default_51"], "view_804": ["quantize_affine_51"], "add_7927": ["quantize_affine_51"], "clamp_min_155": ["quantize_affine_51"], "clamp_max_103": ["quantize_affine_51"], "convert_element_type_307": ["quantize_affine_51", "dequantize_affine_102"], "view_808": ["dequantize_affine_102"], "convert_element_type_308": ["dequantize_affine_102"], "convert_element_type_309": ["dequantize_affine_102"], "sub_2373": ["dequantize_affine_102"], "view_807": ["dequantize_affine_102"], "mul_5034": ["dequantize_affine_102"], "view_810": ["dequantize_affine_103"], "view_812": ["dequantize_affine_103"], "convert_element_type_310": ["dequantize_affine_103"], "convert_element_type_311": ["dequantize_affine_103"], "sub_2377": ["dequantize_affine_103"], "view_811": ["dequantize_affine_103"], "mul_5039": ["dequantize_affine_103"], "view_813": ["dequantize_affine_103"], "mul_5044": ["linear_51"], "view_814": ["linear_51"], "permute_88": ["linear_51"], "mm_default_117": ["linear_51"], "add_tensor_117": ["linear_51"], "view_815": ["linear_51", "dropout_25"], "add_7990": ["add_121"], "clone_70": ["layer_norm_17"], "var_mean_17": ["layer_norm_17"], "getitem_70": ["layer_norm_17"], "getitem_71": ["layer_norm_17"], "sub_2383": ["layer_norm_17"], "add_7995": ["layer_norm_17"], "rsqrt_17": ["layer_norm_17"], "mul_5055": ["layer_norm_17"], "mul_5056": ["layer_norm_17"], "add_7996": ["layer_norm_17", "choose_qparams_affine_default_52", "quantize_affine_52"], "amax_52": ["choose_qparams_affine_default_52"], "full_105": ["choose_qparams_affine_default_52"], "maximum_52": ["choose_qparams_affine_default_52"], "amin_52": ["choose_qparams_affine_default_52"], "full_104": ["choose_qparams_affine_default_52"], "minimum_52": ["choose_qparams_affine_default_52"], "sub_2394": ["choose_qparams_affine_default_52"], "div_104": ["choose_qparams_affine_default_52"], "clamp_min_156": ["choose_qparams_affine_default_52"], "div_105": ["choose_qparams_affine_default_52"], "round_105": ["choose_qparams_affine_default_52"], "sub_2400": ["choose_qparams_affine_default_52"], "clamp_min_157": ["choose_qparams_affine_default_52"], "clamp_max_104": ["choose_qparams_affine_default_52"], "view_818": ["quantize_affine_52"], "reciprocal_52": ["quantize_affine_52"], "mul_5104": ["quantize_affine_52"], "mul_5107": ["quantize_affine_52"], "round_106": ["quantize_affine_52"], "convert_element_type_312": ["choose_qparams_affine_default_52"], "view_819": ["quantize_affine_52"], "add_8083": ["quantize_affine_52"], "clamp_min_158": ["quantize_affine_52"], "clamp_max_105": ["quantize_affine_52"], "convert_element_type_313": ["quantize_affine_52", "dequantize_affine_104"], "view_823": ["dequantize_affine_104"], "convert_element_type_314": ["dequantize_affine_104"], "convert_element_type_315": ["dequantize_affine_104"], "sub_2420": ["dequantize_affine_104"], "view_822": ["dequantize_affine_104"], "mul_5129": ["dequantize_affine_104"], "view_825": ["dequantize_affine_105"], "view_827": ["dequantize_affine_105"], "convert_element_type_316": ["dequantize_affine_105"], "convert_element_type_317": ["dequantize_affine_105"], "sub_2424": ["dequantize_affine_105"], "view_826": ["dequantize_affine_105"], "mul_5134": ["dequantize_affine_105"], "view_828": ["dequantize_affine_105"], "mul_5139": ["linear_52"], "view_829": ["linear_52"], "permute_89": ["linear_52"], "mm_default_116": ["linear_52"], "add_tensor_116": ["linear_52"], "view_830": ["linear_52"], "mul_5146": ["gelu_10"], "mul_5147": ["gelu_10"], "erf_10": ["gelu_10"], "add_8142": ["gelu_10"], "mul_5148": ["gelu_10", "dropout_26", "choose_qparams_affine_default_53", "quantize_affine_53"], "amax_53": ["choose_qparams_affine_default_53"], "full_107": ["choose_qparams_affine_default_53"], "maximum_53": ["choose_qparams_affine_default_53"], "amin_53": ["choose_qparams_affine_default_53"], "full_106": ["choose_qparams_affine_default_53"], "minimum_53": ["choose_qparams_affine_default_53"], "sub_2437": ["choose_qparams_affine_default_53"], "div_106": ["choose_qparams_affine_default_53"], "clamp_min_159": ["choose_qparams_affine_default_53"], "div_107": ["choose_qparams_affine_default_53"], "round_107": ["choose_qparams_affine_default_53"], "sub_2443": ["choose_qparams_affine_default_53"], "clamp_min_160": ["choose_qparams_affine_default_53"], "clamp_max_106": ["choose_qparams_affine_default_53"], "view_833": ["quantize_affine_53"], "reciprocal_53": ["quantize_affine_53"], "mul_5194": ["quantize_affine_53"], "mul_5197": ["quantize_affine_53"], "round_108": ["quantize_affine_53"], "convert_element_type_318": ["choose_qparams_affine_default_53"], "view_834": ["quantize_affine_53"], "add_8225": ["quantize_affine_53"], "clamp_min_161": ["quantize_affine_53"], "clamp_max_107": ["quantize_affine_53"], "convert_element_type_319": ["quantize_affine_53", "dequantize_affine_106"], "view_838": ["dequantize_affine_106"], "convert_element_type_320": ["dequantize_affine_106"], "convert_element_type_321": ["dequantize_affine_106"], "sub_2463": ["dequantize_affine_106"], "view_837": ["dequantize_affine_106"], "mul_5219": ["dequantize_affine_106"], "view_840": ["dequantize_affine_107"], "view_842": ["dequantize_affine_107"], "convert_element_type_322": ["dequantize_affine_107"], "convert_element_type_323": ["dequantize_affine_107"], "sub_2467": ["dequantize_affine_107"], "view_841": ["dequantize_affine_107"], "mul_5224": ["dequantize_affine_107"], "view_843": ["dequantize_affine_107"], "mul_5229": ["linear_53"], "view_844": ["linear_53"], "permute_90": ["linear_53"], "mm_default_115": ["linear_53"], "add_tensor_115": ["linear_53"], "view_845": ["linear_53", "dropout_27"], "add_8288": ["add_126"], "clone_73": ["layer_norm_18"], "var_mean_18": ["layer_norm_18"], "getitem_72": ["layer_norm_18"], "getitem_73": ["layer_norm_18"], "sub_2473": ["layer_norm_18"], "add_8293": ["layer_norm_18"], "rsqrt_18": ["layer_norm_18"], "mul_5240": ["layer_norm_18"], "mul_5241": ["layer_norm_18"], "add_8294": ["layer_norm_18", "choose_qparams_affine_default_54", "quantize_affine_54", "choose_qparams_affine_default_55", "quantize_affine_55", "choose_qparams_affine_default_56", "quantize_affine_56"], "amax_54": ["choose_qparams_affine_default_54"], "full_109": ["choose_qparams_affine_default_54"], "maximum_54": ["choose_qparams_affine_default_54"], "amin_54": ["choose_qparams_affine_default_54"], "full_108": ["choose_qparams_affine_default_54"], "minimum_54": ["choose_qparams_affine_default_54"], "sub_2484": ["choose_qparams_affine_default_54"], "div_108": ["choose_qparams_affine_default_54"], "clamp_min_162": ["choose_qparams_affine_default_54"], "div_109": ["choose_qparams_affine_default_54"], "round_109": ["choose_qparams_affine_default_54"], "sub_2490": ["choose_qparams_affine_default_54"], "clamp_min_163": ["choose_qparams_affine_default_54"], "clamp_max_108": ["choose_qparams_affine_default_54"], "view_848": ["quantize_affine_54"], "reciprocal_54": ["quantize_affine_54"], "mul_5289": ["quantize_affine_54"], "mul_5292": ["quantize_affine_54"], "round_110": ["quantize_affine_54"], "convert_element_type_324": ["choose_qparams_affine_default_54"], "view_849": ["quantize_affine_54"], "add_8381": ["quantize_affine_54"], "clamp_min_164": ["quantize_affine_54"], "clamp_max_109": ["quantize_affine_54"], "convert_element_type_325": ["quantize_affine_54", "dequantize_affine_108"], "view_853": ["dequantize_affine_108"], "convert_element_type_326": ["dequantize_affine_108"], "convert_element_type_327": ["dequantize_affine_108"], "sub_2510": ["dequantize_affine_108"], "view_852": ["dequantize_affine_108"], "mul_5314": ["dequantize_affine_108"], "view_855": ["dequantize_affine_109"], "view_857": ["dequantize_affine_109"], "convert_element_type_328": ["dequantize_affine_109"], "convert_element_type_329": ["dequantize_affine_109"], "sub_2514": ["dequantize_affine_109"], "view_856": ["dequantize_affine_109"], "mul_5319": ["dequantize_affine_109"], "view_858": ["dequantize_affine_109"], "amax_55": ["choose_qparams_affine_default_55"], "full_111": ["choose_qparams_affine_default_55"], "maximum_55": ["choose_qparams_affine_default_55"], "amin_55": ["choose_qparams_affine_default_55"], "full_110": ["choose_qparams_affine_default_55"], "minimum_55": ["choose_qparams_affine_default_55"], "sub_2529": ["choose_qparams_affine_default_55"], "div_110": ["choose_qparams_affine_default_55"], "clamp_min_165": ["choose_qparams_affine_default_55"], "div_111": ["choose_qparams_affine_default_55"], "round_111": ["choose_qparams_affine_default_55"], "sub_2535": ["choose_qparams_affine_default_55"], "clamp_min_166": ["choose_qparams_affine_default_55"], "clamp_max_110": ["choose_qparams_affine_default_55"], "view_864": ["quantize_affine_55"], "reciprocal_55": ["quantize_affine_55"], "mul_5385": ["quantize_affine_55"], "mul_5388": ["quantize_affine_55"], "round_112": ["quantize_affine_55"], "convert_element_type_330": ["choose_qparams_affine_default_55"], "view_865": ["quantize_affine_55"], "add_8533": ["quantize_affine_55"], "clamp_min_167": ["quantize_affine_55"], "clamp_max_111": ["quantize_affine_55"], "convert_element_type_331": ["quantize_affine_55", "dequantize_affine_110"], "view_869": ["dequantize_affine_110"], "convert_element_type_332": ["dequantize_affine_110"], "convert_element_type_333": ["dequantize_affine_110"], "sub_2555": ["dequantize_affine_110"], "view_868": ["dequantize_affine_110"], "mul_5410": ["dequantize_affine_110"], "view_871": ["dequantize_affine_111"], "view_873": ["dequantize_affine_111"], "convert_element_type_334": ["dequantize_affine_111"], "convert_element_type_335": ["dequantize_affine_111"], "sub_2559": ["dequantize_affine_111"], "view_872": ["dequantize_affine_111"], "mul_5415": ["dequantize_affine_111"], "view_874": ["dequantize_affine_111"], "amax_56": ["choose_qparams_affine_default_56"], "full_113": ["choose_qparams_affine_default_56"], "maximum_56": ["choose_qparams_affine_default_56"], "amin_56": ["choose_qparams_affine_default_56"], "full_112": ["choose_qparams_affine_default_56"], "minimum_56": ["choose_qparams_affine_default_56"], "sub_2573": ["choose_qparams_affine_default_56"], "div_112": ["choose_qparams_affine_default_56"], "clamp_min_168": ["choose_qparams_affine_default_56"], "div_113": ["choose_qparams_affine_default_56"], "round_113": ["choose_qparams_affine_default_56"], "sub_2579": ["choose_qparams_affine_default_56"], "clamp_min_169": ["choose_qparams_affine_default_56"], "clamp_max_112": ["choose_qparams_affine_default_56"], "view_880": ["quantize_affine_56"], "reciprocal_56": ["quantize_affine_56"], "mul_5484": ["quantize_affine_56"], "mul_5487": ["quantize_affine_56"], "round_114": ["quantize_affine_56"], "convert_element_type_336": ["choose_qparams_affine_default_56"], "view_881": ["quantize_affine_56"], "add_8681": ["quantize_affine_56"], "clamp_min_170": ["quantize_affine_56"], "clamp_max_113": ["quantize_affine_56"], "convert_element_type_337": ["quantize_affine_56", "dequantize_affine_112"], "view_885": ["dequantize_affine_112"], "convert_element_type_338": ["dequantize_affine_112"], "convert_element_type_339": ["dequantize_affine_112"], "sub_2599": ["dequantize_affine_112"], "view_884": ["dequantize_affine_112"], "mul_5509": ["dequantize_affine_112"], "view_887": ["dequantize_affine_113"], "view_889": ["dequantize_affine_113"], "convert_element_type_340": ["dequantize_affine_113"], "convert_element_type_341": ["dequantize_affine_113"], "sub_2603": ["dequantize_affine_113"], "view_888": ["dequantize_affine_113"], "mul_5514": ["dequantize_affine_113"], "view_890": ["dequantize_affine_113"], "mul_5324": ["linear_54"], "view_859": ["linear_54"], "permute_91": ["linear_54"], "mm_default_114": ["linear_54"], "add_tensor_114": ["linear_54"], "view_860": ["linear_54"], "mul_5331": ["mul_338"], "view_861": ["view_27"], "permute_92": ["transpose_36"], "clone_74": ["contiguous_36"], "mul_5418": ["linear_55"], "view_875": ["linear_55"], "permute_93": ["linear_55"], "mm_9": ["linear_55"], "view_876": ["linear_55"], "view_877": ["view_28"], "permute_94": ["transpose_37"], "clone_75": ["contiguous_37"], "mul_5519": ["linear_56"], "view_891": ["linear_56"], "permute_95": ["linear_56"], "mm_default_113": ["linear_56"], "add_tensor_113": ["linear_56"], "view_892": ["linear_56"], "view_893": ["view_29"], "permute_96": ["transpose_38"], "clone_76": ["contiguous_38"], "_scaled_dot_product_efficient_attention_9": ["scaled_dot_product_attention_9"], "getitem_74": ["scaled_dot_product_attention_9"], "permute_97": ["transpose_39"], "view_894": ["reshape_9", "choose_qparams_affine_default_57", "quantize_affine_57"], "amax_57": ["choose_qparams_affine_default_57"], "full_115": ["choose_qparams_affine_default_57"], "maximum_57": ["choose_qparams_affine_default_57"], "amin_57": ["choose_qparams_affine_default_57"], "full_114": ["choose_qparams_affine_default_57"], "minimum_57": ["choose_qparams_affine_default_57"], "sub_2621": ["choose_qparams_affine_default_57"], "div_114": ["choose_qparams_affine_default_57"], "clamp_min_171": ["choose_qparams_affine_default_57"], "div_115": ["choose_qparams_affine_default_57"], "round_115": ["choose_qparams_affine_default_57"], "sub_2627": ["choose_qparams_affine_default_57"], "clamp_min_172": ["choose_qparams_affine_default_57"], "clamp_max_114": ["choose_qparams_affine_default_57"], "view_897": ["quantize_affine_57"], "reciprocal_57": ["quantize_affine_57"], "mul_5589": ["quantize_affine_57"], "mul_5592": ["quantize_affine_57"], "round_116": ["quantize_affine_57"], "convert_element_type_342": ["choose_qparams_affine_default_57"], "view_898": ["quantize_affine_57"], "add_8845": ["quantize_affine_57"], "clamp_min_173": ["quantize_affine_57"], "clamp_max_115": ["quantize_affine_57"], "convert_element_type_343": ["quantize_affine_57", "dequantize_affine_114"], "view_902": ["dequantize_affine_114"], "convert_element_type_344": ["dequantize_affine_114"], "convert_element_type_345": ["dequantize_affine_114"], "sub_2647": ["dequantize_affine_114"], "view_901": ["dequantize_affine_114"], "mul_5614": ["dequantize_affine_114"], "view_904": ["dequantize_affine_115"], "view_906": ["dequantize_affine_115"], "convert_element_type_346": ["dequantize_affine_115"], "convert_element_type_347": ["dequantize_affine_115"], "sub_2651": ["dequantize_affine_115"], "view_905": ["dequantize_affine_115"], "mul_5619": ["dequantize_affine_115"], "view_907": ["dequantize_affine_115"], "mul_5624": ["linear_57"], "view_908": ["linear_57"], "permute_98": ["linear_57"], "mm_default_112": ["linear_57"], "add_tensor_112": ["linear_57"], "view_909": ["linear_57", "dropout_28"], "add_8908": ["add_135"], "clone_78": ["layer_norm_19"], "var_mean_19": ["layer_norm_19"], "getitem_78": ["layer_norm_19"], "getitem_79": ["layer_norm_19"], "sub_2657": ["layer_norm_19"], "add_8913": ["layer_norm_19"], "rsqrt_19": ["layer_norm_19"], "mul_5635": ["layer_norm_19"], "mul_5636": ["layer_norm_19"], "add_8914": ["layer_norm_19", "choose_qparams_affine_default_58", "quantize_affine_58"], "amax_58": ["choose_qparams_affine_default_58"], "full_117": ["choose_qparams_affine_default_58"], "maximum_58": ["choose_qparams_affine_default_58"], "amin_58": ["choose_qparams_affine_default_58"], "full_116": ["choose_qparams_affine_default_58"], "minimum_58": ["choose_qparams_affine_default_58"], "sub_2668": ["choose_qparams_affine_default_58"], "div_116": ["choose_qparams_affine_default_58"], "clamp_min_174": ["choose_qparams_affine_default_58"], "div_117": ["choose_qparams_affine_default_58"], "round_117": ["choose_qparams_affine_default_58"], "sub_2674": ["choose_qparams_affine_default_58"], "clamp_min_175": ["choose_qparams_affine_default_58"], "clamp_max_116": ["choose_qparams_affine_default_58"], "view_912": ["quantize_affine_58"], "reciprocal_58": ["quantize_affine_58"], "mul_5684": ["quantize_affine_58"], "mul_5687": ["quantize_affine_58"], "round_118": ["quantize_affine_58"], "convert_element_type_348": ["choose_qparams_affine_default_58"], "view_913": ["quantize_affine_58"], "add_9001": ["quantize_affine_58"], "clamp_min_176": ["quantize_affine_58"], "clamp_max_117": ["quantize_affine_58"], "convert_element_type_349": ["quantize_affine_58", "dequantize_affine_116"], "view_917": ["dequantize_affine_116"], "convert_element_type_350": ["dequantize_affine_116"], "convert_element_type_351": ["dequantize_affine_116"], "sub_2694": ["dequantize_affine_116"], "view_916": ["dequantize_affine_116"], "mul_5709": ["dequantize_affine_116"], "view_919": ["dequantize_affine_117"], "view_921": ["dequantize_affine_117"], "convert_element_type_352": ["dequantize_affine_117"], "convert_element_type_353": ["dequantize_affine_117"], "sub_2698": ["dequantize_affine_117"], "view_920": ["dequantize_affine_117"], "mul_5714": ["dequantize_affine_117"], "view_922": ["dequantize_affine_117"], "mul_5719": ["linear_58"], "view_923": ["linear_58"], "permute_99": ["linear_58"], "mm_default_111": ["linear_58"], "add_tensor_111": ["linear_58"], "view_924": ["linear_58"], "mul_5726": ["gelu_11"], "mul_5727": ["gelu_11"], "erf_11": ["gelu_11"], "add_9060": ["gelu_11"], "mul_5728": ["gelu_11", "dropout_29", "choose_qparams_affine_default_59", "quantize_affine_59"], "amax_59": ["choose_qparams_affine_default_59"], "full_119": ["choose_qparams_affine_default_59"], "maximum_59": ["choose_qparams_affine_default_59"], "amin_59": ["choose_qparams_affine_default_59"], "full_118": ["choose_qparams_affine_default_59"], "minimum_59": ["choose_qparams_affine_default_59"], "sub_2711": ["choose_qparams_affine_default_59"], "div_118": ["choose_qparams_affine_default_59"], "clamp_min_177": ["choose_qparams_affine_default_59"], "div_119": ["choose_qparams_affine_default_59"], "round_119": ["choose_qparams_affine_default_59"], "sub_2717": ["choose_qparams_affine_default_59"], "clamp_min_178": ["choose_qparams_affine_default_59"], "clamp_max_118": ["choose_qparams_affine_default_59"], "view_927": ["quantize_affine_59"], "reciprocal_59": ["quantize_affine_59"], "mul_5774": ["quantize_affine_59"], "mul_5777": ["quantize_affine_59"], "round_120": ["quantize_affine_59"], "convert_element_type_354": ["choose_qparams_affine_default_59"], "view_928": ["quantize_affine_59"], "add_9143": ["quantize_affine_59"], "clamp_min_179": ["quantize_affine_59"], "clamp_max_119": ["quantize_affine_59"], "convert_element_type_355": ["quantize_affine_59", "dequantize_affine_118"], "view_932": ["dequantize_affine_118"], "convert_element_type_356": ["dequantize_affine_118"], "convert_element_type_357": ["dequantize_affine_118"], "sub_2737": ["dequantize_affine_118"], "view_931": ["dequantize_affine_118"], "mul_5799": ["dequantize_affine_118"], "view_934": ["dequantize_affine_119"], "view_936": ["dequantize_affine_119"], "convert_element_type_358": ["dequantize_affine_119"], "convert_element_type_359": ["dequantize_affine_119"], "sub_2741": ["dequantize_affine_119"], "view_935": ["dequantize_affine_119"], "mul_5804": ["dequantize_affine_119"], "view_937": ["dequantize_affine_119"], "mul_5809": ["linear_59"], "view_938": ["linear_59"], "permute_100": ["linear_59"], "mm_default_110": ["linear_59"], "add_tensor_110": ["linear_59"], "view_939": ["linear_59", "dropout_30"], "add_9206": ["add_140"], "clone_81": ["layer_norm_20"], "var_mean_20": ["layer_norm_20"], "getitem_80": ["layer_norm_20"], "getitem_81": ["layer_norm_20"], "sub_2747": ["layer_norm_20"], "add_9211": ["layer_norm_20"], "rsqrt_20": ["layer_norm_20"], "mul_5820": ["layer_norm_20"], "mul_5821": ["layer_norm_20"], "add_9212": ["layer_norm_20", "choose_qparams_affine_default_60", "quantize_affine_60", "choose_qparams_affine_default_61", "quantize_affine_61", "choose_qparams_affine_default_62", "quantize_affine_62"], "amax_60": ["choose_qparams_affine_default_60"], "full_121": ["choose_qparams_affine_default_60"], "maximum_60": ["choose_qparams_affine_default_60"], "amin_60": ["choose_qparams_affine_default_60"], "full_120": ["choose_qparams_affine_default_60"], "minimum_60": ["choose_qparams_affine_default_60"], "sub_2758": ["choose_qparams_affine_default_60"], "div_120": ["choose_qparams_affine_default_60"], "clamp_min_180": ["choose_qparams_affine_default_60"], "div_121": ["choose_qparams_affine_default_60"], "round_121": ["choose_qparams_affine_default_60"], "sub_2764": ["choose_qparams_affine_default_60"], "clamp_min_181": ["choose_qparams_affine_default_60"], "clamp_max_120": ["choose_qparams_affine_default_60"], "view_942": ["quantize_affine_60"], "reciprocal_60": ["quantize_affine_60"], "mul_5869": ["quantize_affine_60"], "mul_5872": ["quantize_affine_60"], "round_122": ["quantize_affine_60"], "convert_element_type_360": ["choose_qparams_affine_default_60"], "view_943": ["quantize_affine_60"], "add_9299": ["quantize_affine_60"], "clamp_min_182": ["quantize_affine_60"], "clamp_max_121": ["quantize_affine_60"], "convert_element_type_361": ["quantize_affine_60", "dequantize_affine_120"], "view_947": ["dequantize_affine_120"], "convert_element_type_362": ["dequantize_affine_120"], "convert_element_type_363": ["dequantize_affine_120"], "sub_2784": ["dequantize_affine_120"], "view_946": ["dequantize_affine_120"], "mul_5894": ["dequantize_affine_120"], "view_949": ["dequantize_affine_121"], "view_951": ["dequantize_affine_121"], "convert_element_type_364": ["dequantize_affine_121"], "convert_element_type_365": ["dequantize_affine_121"], "sub_2788": ["dequantize_affine_121"], "view_950": ["dequantize_affine_121"], "mul_5899": ["dequantize_affine_121"], "view_952": ["dequantize_affine_121"], "amax_61": ["choose_qparams_affine_default_61"], "full_123": ["choose_qparams_affine_default_61"], "maximum_61": ["choose_qparams_affine_default_61"], "amin_61": ["choose_qparams_affine_default_61"], "full_122": ["choose_qparams_affine_default_61"], "minimum_61": ["choose_qparams_affine_default_61"], "sub_2803": ["choose_qparams_affine_default_61"], "div_122": ["choose_qparams_affine_default_61"], "clamp_min_183": ["choose_qparams_affine_default_61"], "div_123": ["choose_qparams_affine_default_61"], "round_123": ["choose_qparams_affine_default_61"], "sub_2809": ["choose_qparams_affine_default_61"], "clamp_min_184": ["choose_qparams_affine_default_61"], "clamp_max_122": ["choose_qparams_affine_default_61"], "view_958": ["quantize_affine_61"], "reciprocal_61": ["quantize_affine_61"], "mul_5965": ["quantize_affine_61"], "mul_5968": ["quantize_affine_61"], "round_124": ["quantize_affine_61"], "convert_element_type_366": ["choose_qparams_affine_default_61"], "view_959": ["quantize_affine_61"], "add_9451": ["quantize_affine_61"], "clamp_min_185": ["quantize_affine_61"], "clamp_max_123": ["quantize_affine_61"], "convert_element_type_367": ["quantize_affine_61", "dequantize_affine_122"], "view_963": ["dequantize_affine_122"], "convert_element_type_368": ["dequantize_affine_122"], "convert_element_type_369": ["dequantize_affine_122"], "sub_2829": ["dequantize_affine_122"], "view_962": ["dequantize_affine_122"], "mul_5990": ["dequantize_affine_122"], "view_965": ["dequantize_affine_123"], "view_967": ["dequantize_affine_123"], "convert_element_type_370": ["dequantize_affine_123"], "convert_element_type_371": ["dequantize_affine_123"], "sub_2833": ["dequantize_affine_123"], "view_966": ["dequantize_affine_123"], "mul_5995": ["dequantize_affine_123"], "view_968": ["dequantize_affine_123"], "amax_62": ["choose_qparams_affine_default_62"], "full_125": ["choose_qparams_affine_default_62"], "maximum_62": ["choose_qparams_affine_default_62"], "amin_62": ["choose_qparams_affine_default_62"], "full_124": ["choose_qparams_affine_default_62"], "minimum_62": ["choose_qparams_affine_default_62"], "sub_2847": ["choose_qparams_affine_default_62"], "div_124": ["choose_qparams_affine_default_62"], "clamp_min_186": ["choose_qparams_affine_default_62"], "div_125": ["choose_qparams_affine_default_62"], "round_125": ["choose_qparams_affine_default_62"], "sub_2853": ["choose_qparams_affine_default_62"], "clamp_min_187": ["choose_qparams_affine_default_62"], "clamp_max_124": ["choose_qparams_affine_default_62"], "view_974": ["quantize_affine_62"], "reciprocal_62": ["quantize_affine_62"], "mul_6064": ["quantize_affine_62"], "mul_6067": ["quantize_affine_62"], "round_126": ["quantize_affine_62"], "convert_element_type_372": ["choose_qparams_affine_default_62"], "view_975": ["quantize_affine_62"], "add_9599": ["quantize_affine_62"], "clamp_min_188": ["quantize_affine_62"], "clamp_max_125": ["quantize_affine_62"], "convert_element_type_373": ["quantize_affine_62", "dequantize_affine_124"], "view_979": ["dequantize_affine_124"], "convert_element_type_374": ["dequantize_affine_124"], "convert_element_type_375": ["dequantize_affine_124"], "sub_2873": ["dequantize_affine_124"], "view_978": ["dequantize_affine_124"], "mul_6089": ["dequantize_affine_124"], "view_981": ["dequantize_affine_125"], "view_983": ["dequantize_affine_125"], "convert_element_type_376": ["dequantize_affine_125"], "convert_element_type_377": ["dequantize_affine_125"], "sub_2877": ["dequantize_affine_125"], "view_982": ["dequantize_affine_125"], "mul_6094": ["dequantize_affine_125"], "view_984": ["dequantize_affine_125"], "mul_5904": ["linear_60"], "view_953": ["linear_60"], "permute_101": ["linear_60"], "mm_default_109": ["linear_60"], "add_tensor_109": ["linear_60"], "view_954": ["linear_60"], "mul_5911": ["mul_375"], "view_955": ["view_30"], "permute_102": ["transpose_40"], "clone_82": ["contiguous_40"], "mul_5998": ["linear_61"], "view_969": ["linear_61"], "permute_103": ["linear_61"], "mm_10": ["linear_61"], "view_970": ["linear_61"], "view_971": ["view_31"], "permute_104": ["transpose_41"], "clone_83": ["contiguous_41"], "mul_6099": ["linear_62"], "view_985": ["linear_62"], "permute_105": ["linear_62"], "mm_default_108": ["linear_62"], "add_tensor_108": ["linear_62"], "view_986": ["linear_62"], "view_987": ["view_32"], "permute_106": ["transpose_42"], "clone_84": ["contiguous_42"], "_scaled_dot_product_efficient_attention_10": ["scaled_dot_product_attention_10"], "getitem_82": ["scaled_dot_product_attention_10"], "permute_107": ["transpose_43"], "view_988": ["reshape_10", "choose_qparams_affine_default_63", "quantize_affine_63"], "amax_63": ["choose_qparams_affine_default_63"], "full_127": ["choose_qparams_affine_default_63"], "maximum_63": ["choose_qparams_affine_default_63"], "amin_63": ["choose_qparams_affine_default_63"], "full_126": ["choose_qparams_affine_default_63"], "minimum_63": ["choose_qparams_affine_default_63"], "sub_2895": ["choose_qparams_affine_default_63"], "div_126": ["choose_qparams_affine_default_63"], "clamp_min_189": ["choose_qparams_affine_default_63"], "div_127": ["choose_qparams_affine_default_63"], "round_127": ["choose_qparams_affine_default_63"], "sub_2901": ["choose_qparams_affine_default_63"], "clamp_min_190": ["choose_qparams_affine_default_63"], "clamp_max_126": ["choose_qparams_affine_default_63"], "view_991": ["quantize_affine_63"], "reciprocal_63": ["quantize_affine_63"], "mul_6169": ["quantize_affine_63"], "mul_6172": ["quantize_affine_63"], "round_128": ["quantize_affine_63"], "convert_element_type_378": ["choose_qparams_affine_default_63"], "view_992": ["quantize_affine_63"], "add_9763": ["quantize_affine_63"], "clamp_min_191": ["quantize_affine_63"], "clamp_max_127": ["quantize_affine_63"], "convert_element_type_379": ["quantize_affine_63", "dequantize_affine_126"], "view_996": ["dequantize_affine_126"], "convert_element_type_380": ["dequantize_affine_126"], "convert_element_type_381": ["dequantize_affine_126"], "sub_2921": ["dequantize_affine_126"], "view_995": ["dequantize_affine_126"], "mul_6194": ["dequantize_affine_126"], "view_998": ["dequantize_affine_127"], "view_1000": ["dequantize_affine_127"], "convert_element_type_382": ["dequantize_affine_127"], "convert_element_type_383": ["dequantize_affine_127"], "sub_2925": ["dequantize_affine_127"], "view_999": ["dequantize_affine_127"], "mul_6199": ["dequantize_affine_127"], "view_1001": ["dequantize_affine_127"], "mul_6204": ["linear_63"], "view_1002": ["linear_63"], "permute_108": ["linear_63"], "mm_default_107": ["linear_63"], "add_tensor_107": ["linear_63"], "view_1003": ["linear_63", "dropout_31"], "add_9826": ["add_149"], "clone_86": ["layer_norm_21"], "var_mean_21": ["layer_norm_21"], "getitem_86": ["layer_norm_21"], "getitem_87": ["layer_norm_21"], "sub_2931": ["layer_norm_21"], "add_9831": ["layer_norm_21"], "rsqrt_21": ["layer_norm_21"], "mul_6215": ["layer_norm_21"], "mul_6216": ["layer_norm_21"], "add_9832": ["layer_norm_21", "choose_qparams_affine_default_64", "quantize_affine_64"], "amax_64": ["choose_qparams_affine_default_64"], "full_129": ["choose_qparams_affine_default_64"], "maximum_64": ["choose_qparams_affine_default_64"], "amin_64": ["choose_qparams_affine_default_64"], "full_128": ["choose_qparams_affine_default_64"], "minimum_64": ["choose_qparams_affine_default_64"], "sub_2942": ["choose_qparams_affine_default_64"], "div_128": ["choose_qparams_affine_default_64"], "clamp_min_192": ["choose_qparams_affine_default_64"], "div_129": ["choose_qparams_affine_default_64"], "round_129": ["choose_qparams_affine_default_64"], "sub_2948": ["choose_qparams_affine_default_64"], "clamp_min_193": ["choose_qparams_affine_default_64"], "clamp_max_128": ["choose_qparams_affine_default_64"], "view_1006": ["quantize_affine_64"], "reciprocal_64": ["quantize_affine_64"], "mul_6264": ["quantize_affine_64"], "mul_6267": ["quantize_affine_64"], "round_130": ["quantize_affine_64"], "convert_element_type_384": ["choose_qparams_affine_default_64"], "view_1007": ["quantize_affine_64"], "add_9919": ["quantize_affine_64"], "clamp_min_194": ["quantize_affine_64"], "clamp_max_129": ["quantize_affine_64"], "convert_element_type_385": ["quantize_affine_64", "dequantize_affine_128"], "view_1011": ["dequantize_affine_128"], "convert_element_type_386": ["dequantize_affine_128"], "convert_element_type_387": ["dequantize_affine_128"], "sub_2968": ["dequantize_affine_128"], "view_1010": ["dequantize_affine_128"], "mul_6289": ["dequantize_affine_128"], "view_1013": ["dequantize_affine_129"], "view_1015": ["dequantize_affine_129"], "convert_element_type_388": ["dequantize_affine_129"], "convert_element_type_389": ["dequantize_affine_129"], "sub_2972": ["dequantize_affine_129"], "view_1014": ["dequantize_affine_129"], "mul_6294": ["dequantize_affine_129"], "view_1016": ["dequantize_affine_129"], "mul_6299": ["linear_64"], "view_1017": ["linear_64"], "permute_109": ["linear_64"], "mm_default_106": ["linear_64"], "add_tensor_106": ["linear_64"], "view_1018": ["linear_64"], "mul_6306": ["gelu_12"], "mul_6307": ["gelu_12"], "erf_12": ["gelu_12"], "add_9978": ["gelu_12"], "mul_6308": ["gelu_12", "dropout_32", "choose_qparams_affine_default_65", "quantize_affine_65"], "amax_65": ["choose_qparams_affine_default_65"], "full_131": ["choose_qparams_affine_default_65"], "maximum_65": ["choose_qparams_affine_default_65"], "amin_65": ["choose_qparams_affine_default_65"], "full_130": ["choose_qparams_affine_default_65"], "minimum_65": ["choose_qparams_affine_default_65"], "sub_2985": ["choose_qparams_affine_default_65"], "div_130": ["choose_qparams_affine_default_65"], "clamp_min_195": ["choose_qparams_affine_default_65"], "div_131": ["choose_qparams_affine_default_65"], "round_131": ["choose_qparams_affine_default_65"], "sub_2991": ["choose_qparams_affine_default_65"], "clamp_min_196": ["choose_qparams_affine_default_65"], "clamp_max_130": ["choose_qparams_affine_default_65"], "view_1021": ["quantize_affine_65"], "reciprocal_65": ["quantize_affine_65"], "mul_6354": ["quantize_affine_65"], "mul_6357": ["quantize_affine_65"], "round_132": ["quantize_affine_65"], "convert_element_type_390": ["choose_qparams_affine_default_65"], "view_1022": ["quantize_affine_65"], "add_10061": ["quantize_affine_65"], "clamp_min_197": ["quantize_affine_65"], "clamp_max_131": ["quantize_affine_65"], "convert_element_type_391": ["quantize_affine_65", "dequantize_affine_130"], "view_1026": ["dequantize_affine_130"], "convert_element_type_392": ["dequantize_affine_130"], "convert_element_type_393": ["dequantize_affine_130"], "sub_3011": ["dequantize_affine_130"], "view_1025": ["dequantize_affine_130"], "mul_6379": ["dequantize_affine_130"], "view_1028": ["dequantize_affine_131"], "view_1030": ["dequantize_affine_131"], "convert_element_type_394": ["dequantize_affine_131"], "convert_element_type_395": ["dequantize_affine_131"], "sub_3015": ["dequantize_affine_131"], "view_1029": ["dequantize_affine_131"], "mul_6384": ["dequantize_affine_131"], "view_1031": ["dequantize_affine_131"], "mul_6389": ["linear_65"], "view_1032": ["linear_65"], "permute_110": ["linear_65"], "mm_default_105": ["linear_65"], "add_tensor_105": ["linear_65"], "view_1033": ["linear_65", "dropout_33"], "add_10124": ["add_154"], "clone_89": ["layer_norm_22"], "var_mean_22": ["layer_norm_22"], "getitem_88": ["layer_norm_22"], "getitem_89": ["layer_norm_22"], "sub_3021": ["layer_norm_22"], "add_10129": ["layer_norm_22"], "rsqrt_22": ["layer_norm_22"], "mul_6400": ["layer_norm_22"], "mul_6401": ["layer_norm_22"], "add_10130": ["layer_norm_22", "choose_qparams_affine_default_66", "quantize_affine_66", "choose_qparams_affine_default_67", "quantize_affine_67", "choose_qparams_affine_default_68", "quantize_affine_68"], "amax_66": ["choose_qparams_affine_default_66"], "full_133": ["choose_qparams_affine_default_66"], "maximum_66": ["choose_qparams_affine_default_66"], "amin_66": ["choose_qparams_affine_default_66"], "full_132": ["choose_qparams_affine_default_66"], "minimum_66": ["choose_qparams_affine_default_66"], "sub_3032": ["choose_qparams_affine_default_66"], "div_132": ["choose_qparams_affine_default_66"], "clamp_min_198": ["choose_qparams_affine_default_66"], "div_133": ["choose_qparams_affine_default_66"], "round_133": ["choose_qparams_affine_default_66"], "sub_3038": ["choose_qparams_affine_default_66"], "clamp_min_199": ["choose_qparams_affine_default_66"], "clamp_max_132": ["choose_qparams_affine_default_66"], "view_1036": ["quantize_affine_66"], "reciprocal_66": ["quantize_affine_66"], "mul_6449": ["quantize_affine_66"], "mul_6452": ["quantize_affine_66"], "round_134": ["quantize_affine_66"], "convert_element_type_396": ["choose_qparams_affine_default_66"], "view_1037": ["quantize_affine_66"], "add_10217": ["quantize_affine_66"], "clamp_min_200": ["quantize_affine_66"], "clamp_max_133": ["quantize_affine_66"], "convert_element_type_397": ["quantize_affine_66", "dequantize_affine_132"], "view_1041": ["dequantize_affine_132"], "convert_element_type_398": ["dequantize_affine_132"], "convert_element_type_399": ["dequantize_affine_132"], "sub_3058": ["dequantize_affine_132"], "view_1040": ["dequantize_affine_132"], "mul_6474": ["dequantize_affine_132"], "view_1043": ["dequantize_affine_133"], "view_1045": ["dequantize_affine_133"], "convert_element_type_400": ["dequantize_affine_133"], "convert_element_type_401": ["dequantize_affine_133"], "sub_3062": ["dequantize_affine_133"], "view_1044": ["dequantize_affine_133"], "mul_6479": ["dequantize_affine_133"], "view_1046": ["dequantize_affine_133"], "amax_67": ["choose_qparams_affine_default_67"], "full_135": ["choose_qparams_affine_default_67"], "maximum_67": ["choose_qparams_affine_default_67"], "amin_67": ["choose_qparams_affine_default_67"], "full_134": ["choose_qparams_affine_default_67"], "minimum_67": ["choose_qparams_affine_default_67"], "sub_3077": ["choose_qparams_affine_default_67"], "div_134": ["choose_qparams_affine_default_67"], "clamp_min_201": ["choose_qparams_affine_default_67"], "div_135": ["choose_qparams_affine_default_67"], "round_135": ["choose_qparams_affine_default_67"], "sub_3083": ["choose_qparams_affine_default_67"], "clamp_min_202": ["choose_qparams_affine_default_67"], "clamp_max_134": ["choose_qparams_affine_default_67"], "view_1052": ["quantize_affine_67"], "reciprocal_67": ["quantize_affine_67"], "mul_6545": ["quantize_affine_67"], "mul_6548": ["quantize_affine_67"], "round_136": ["quantize_affine_67"], "convert_element_type_402": ["choose_qparams_affine_default_67"], "view_1053": ["quantize_affine_67"], "add_10369": ["quantize_affine_67"], "clamp_min_203": ["quantize_affine_67"], "clamp_max_135": ["quantize_affine_67"], "convert_element_type_403": ["quantize_affine_67", "dequantize_affine_134"], "view_1057": ["dequantize_affine_134"], "convert_element_type_404": ["dequantize_affine_134"], "convert_element_type_405": ["dequantize_affine_134"], "sub_3103": ["dequantize_affine_134"], "view_1056": ["dequantize_affine_134"], "mul_6570": ["dequantize_affine_134"], "view_1059": ["dequantize_affine_135"], "view_1061": ["dequantize_affine_135"], "convert_element_type_406": ["dequantize_affine_135"], "convert_element_type_407": ["dequantize_affine_135"], "sub_3107": ["dequantize_affine_135"], "view_1060": ["dequantize_affine_135"], "mul_6575": ["dequantize_affine_135"], "view_1062": ["dequantize_affine_135"], "amax_68": ["choose_qparams_affine_default_68"], "full_137": ["choose_qparams_affine_default_68"], "maximum_68": ["choose_qparams_affine_default_68"], "amin_68": ["choose_qparams_affine_default_68"], "full_136": ["choose_qparams_affine_default_68"], "minimum_68": ["choose_qparams_affine_default_68"], "sub_3121": ["choose_qparams_affine_default_68"], "div_136": ["choose_qparams_affine_default_68"], "clamp_min_204": ["choose_qparams_affine_default_68"], "div_137": ["choose_qparams_affine_default_68"], "round_137": ["choose_qparams_affine_default_68"], "sub_3127": ["choose_qparams_affine_default_68"], "clamp_min_205": ["choose_qparams_affine_default_68"], "clamp_max_136": ["choose_qparams_affine_default_68"], "view_1068": ["quantize_affine_68"], "reciprocal_68": ["quantize_affine_68"], "mul_6644": ["quantize_affine_68"], "mul_6647": ["quantize_affine_68"], "round_138": ["quantize_affine_68"], "convert_element_type_408": ["choose_qparams_affine_default_68"], "view_1069": ["quantize_affine_68"], "add_10517": ["quantize_affine_68"], "clamp_min_206": ["quantize_affine_68"], "clamp_max_137": ["quantize_affine_68"], "convert_element_type_409": ["quantize_affine_68", "dequantize_affine_136"], "view_1073": ["dequantize_affine_136"], "convert_element_type_410": ["dequantize_affine_136"], "convert_element_type_411": ["dequantize_affine_136"], "sub_3147": ["dequantize_affine_136"], "view_1072": ["dequantize_affine_136"], "mul_6669": ["dequantize_affine_136"], "view_1075": ["dequantize_affine_137"], "view_1077": ["dequantize_affine_137"], "convert_element_type_412": ["dequantize_affine_137"], "convert_element_type_413": ["dequantize_affine_137"], "sub_3151": ["dequantize_affine_137"], "view_1076": ["dequantize_affine_137"], "mul_6674": ["dequantize_affine_137"], "view_1078": ["dequantize_affine_137"], "mul_6484": ["linear_66"], "view_1047": ["linear_66"], "permute_111": ["linear_66"], "mm_default_104": ["linear_66"], "add_tensor_104": ["linear_66"], "view_1048": ["linear_66"], "mul_6491": ["mul_412"], "view_1049": ["view_33"], "permute_112": ["transpose_44"], "clone_90": ["contiguous_44"], "mul_6578": ["linear_67"], "view_1063": ["linear_67"], "permute_113": ["linear_67"], "mm_11": ["linear_67"], "view_1064": ["linear_67"], "view_1065": ["view_34"], "permute_114": ["transpose_45"], "clone_91": ["contiguous_45"], "mul_6679": ["linear_68"], "view_1079": ["linear_68"], "permute_115": ["linear_68"], "mm_default_103": ["linear_68"], "add_tensor_103": ["linear_68"], "view_1080": ["linear_68"], "view_1081": ["view_35"], "permute_116": ["transpose_46"], "clone_92": ["contiguous_46"], "_scaled_dot_product_efficient_attention_11": ["scaled_dot_product_attention_11"], "getitem_90": ["scaled_dot_product_attention_11"], "permute_117": ["transpose_47"], "view_1082": ["reshape_11", "choose_qparams_affine_default_69", "quantize_affine_69"], "amax_69": ["choose_qparams_affine_default_69"], "full_139": ["choose_qparams_affine_default_69"], "maximum_69": ["choose_qparams_affine_default_69"], "amin_69": ["choose_qparams_affine_default_69"], "full_138": ["choose_qparams_affine_default_69"], "minimum_69": ["choose_qparams_affine_default_69"], "sub_3169": ["choose_qparams_affine_default_69"], "div_138": ["choose_qparams_affine_default_69"], "clamp_min_207": ["choose_qparams_affine_default_69"], "div_139": ["choose_qparams_affine_default_69"], "round_139": ["choose_qparams_affine_default_69"], "sub_3175": ["choose_qparams_affine_default_69"], "clamp_min_208": ["choose_qparams_affine_default_69"], "clamp_max_138": ["choose_qparams_affine_default_69"], "view_1085": ["quantize_affine_69"], "reciprocal_69": ["quantize_affine_69"], "mul_6749": ["quantize_affine_69"], "mul_6752": ["quantize_affine_69"], "round_140": ["quantize_affine_69"], "convert_element_type_414": ["choose_qparams_affine_default_69"], "view_1086": ["quantize_affine_69"], "add_10681": ["quantize_affine_69"], "clamp_min_209": ["quantize_affine_69"], "clamp_max_139": ["quantize_affine_69"], "convert_element_type_415": ["quantize_affine_69", "dequantize_affine_138"], "view_1090": ["dequantize_affine_138"], "convert_element_type_416": ["dequantize_affine_138"], "convert_element_type_417": ["dequantize_affine_138"], "sub_3195": ["dequantize_affine_138"], "view_1089": ["dequantize_affine_138"], "mul_6774": ["dequantize_affine_138"], "view_1092": ["dequantize_affine_139"], "view_1094": ["dequantize_affine_139"], "convert_element_type_418": ["dequantize_affine_139"], "convert_element_type_419": ["dequantize_affine_139"], "sub_3199": ["dequantize_affine_139"], "view_1093": ["dequantize_affine_139"], "mul_6779": ["dequantize_affine_139"], "view_1095": ["dequantize_affine_139"], "mul_6784": ["linear_69"], "view_1096": ["linear_69"], "permute_118": ["linear_69"], "mm_default_102": ["linear_69"], "add_tensor_102": ["linear_69"], "view_1097": ["linear_69", "dropout_34"], "add_10744": ["add_163"], "clone_94": ["layer_norm_23"], "var_mean_23": ["layer_norm_23"], "getitem_94": ["layer_norm_23"], "getitem_95": ["layer_norm_23"], "sub_3205": ["layer_norm_23"], "add_10749": ["layer_norm_23"], "rsqrt_23": ["layer_norm_23"], "mul_6795": ["layer_norm_23"], "mul_6796": ["layer_norm_23"], "add_10750": ["layer_norm_23", "choose_qparams_affine_default_70", "quantize_affine_70"], "amax_70": ["choose_qparams_affine_default_70"], "full_141": ["choose_qparams_affine_default_70"], "maximum_70": ["choose_qparams_affine_default_70"], "amin_70": ["choose_qparams_affine_default_70"], "full_140": ["choose_qparams_affine_default_70"], "minimum_70": ["choose_qparams_affine_default_70"], "sub_3216": ["choose_qparams_affine_default_70"], "div_140": ["choose_qparams_affine_default_70"], "clamp_min_210": ["choose_qparams_affine_default_70"], "div_141": ["choose_qparams_affine_default_70"], "round_141": ["choose_qparams_affine_default_70"], "sub_3222": ["choose_qparams_affine_default_70"], "clamp_min_211": ["choose_qparams_affine_default_70"], "clamp_max_140": ["choose_qparams_affine_default_70"], "view_1100": ["quantize_affine_70"], "reciprocal_70": ["quantize_affine_70"], "mul_6844": ["quantize_affine_70"], "mul_6847": ["quantize_affine_70"], "round_142": ["quantize_affine_70"], "convert_element_type_420": ["choose_qparams_affine_default_70"], "view_1101": ["quantize_affine_70"], "add_10837": ["quantize_affine_70"], "clamp_min_212": ["quantize_affine_70"], "clamp_max_141": ["quantize_affine_70"], "convert_element_type_421": ["quantize_affine_70", "dequantize_affine_140"], "view_1105": ["dequantize_affine_140"], "convert_element_type_422": ["dequantize_affine_140"], "convert_element_type_423": ["dequantize_affine_140"], "sub_3242": ["dequantize_affine_140"], "view_1104": ["dequantize_affine_140"], "mul_6869": ["dequantize_affine_140"], "view_1107": ["dequantize_affine_141"], "view_1109": ["dequantize_affine_141"], "convert_element_type_424": ["dequantize_affine_141"], "convert_element_type_425": ["dequantize_affine_141"], "sub_3246": ["dequantize_affine_141"], "view_1108": ["dequantize_affine_141"], "mul_6874": ["dequantize_affine_141"], "view_1110": ["dequantize_affine_141"], "mul_6879": ["linear_70"], "view_1111": ["linear_70"], "permute_119": ["linear_70"], "mm_default_101": ["linear_70"], "add_tensor_101": ["linear_70"], "view_1112": ["linear_70"], "mul_6886": ["gelu_13"], "mul_6887": ["gelu_13"], "erf_13": ["gelu_13"], "add_10896": ["gelu_13"], "mul_6888": ["gelu_13", "dropout_35", "choose_qparams_affine_default_71", "quantize_affine_71"], "amax_71": ["choose_qparams_affine_default_71"], "full_143": ["choose_qparams_affine_default_71"], "maximum_71": ["choose_qparams_affine_default_71"], "amin_71": ["choose_qparams_affine_default_71"], "full_142": ["choose_qparams_affine_default_71"], "minimum_71": ["choose_qparams_affine_default_71"], "sub_3259": ["choose_qparams_affine_default_71"], "div_142": ["choose_qparams_affine_default_71"], "clamp_min_213": ["choose_qparams_affine_default_71"], "div_143": ["choose_qparams_affine_default_71"], "round_143": ["choose_qparams_affine_default_71"], "sub_3265": ["choose_qparams_affine_default_71"], "clamp_min_214": ["choose_qparams_affine_default_71"], "clamp_max_142": ["choose_qparams_affine_default_71"], "view_1115": ["quantize_affine_71"], "reciprocal_71": ["quantize_affine_71"], "mul_6934": ["quantize_affine_71"], "mul_6937": ["quantize_affine_71"], "round_144": ["quantize_affine_71"], "convert_element_type_426": ["choose_qparams_affine_default_71"], "view_1116": ["quantize_affine_71"], "add_10979": ["quantize_affine_71"], "clamp_min_215": ["quantize_affine_71"], "clamp_max_143": ["quantize_affine_71"], "convert_element_type_427": ["quantize_affine_71", "dequantize_affine_142"], "view_1120": ["dequantize_affine_142"], "convert_element_type_428": ["dequantize_affine_142"], "convert_element_type_429": ["dequantize_affine_142"], "sub_3285": ["dequantize_affine_142"], "view_1119": ["dequantize_affine_142"], "mul_6959": ["dequantize_affine_142"], "view_1122": ["dequantize_affine_143"], "view_1124": ["dequantize_affine_143"], "convert_element_type_430": ["dequantize_affine_143"], "convert_element_type_431": ["dequantize_affine_143"], "sub_3289": ["dequantize_affine_143"], "view_1123": ["dequantize_affine_143"], "mul_6964": ["dequantize_affine_143"], "view_1125": ["dequantize_affine_143"], "mul_6969": ["linear_71"], "view_1126": ["linear_71"], "permute_120": ["linear_71"], "mm_default_100": ["linear_71"], "add_tensor_100": ["linear_71"], "view_1127": ["linear_71", "dropout_36"], "add_11042": ["add_168"], "clone_97": ["layer_norm_24"], "var_mean_24": ["layer_norm_24"], "getitem_96": ["layer_norm_24"], "getitem_97": ["layer_norm_24"], "sub_3295": ["layer_norm_24"], "add_11047": ["layer_norm_24"], "rsqrt_24": ["layer_norm_24"], "mul_6980": ["layer_norm_24"], "mul_6981": ["layer_norm_24"], "add_11048": ["layer_norm_24", "choose_qparams_affine_default_72", "quantize_affine_72", "choose_qparams_affine_default_73", "quantize_affine_73", "choose_qparams_affine_default_74", "quantize_affine_74"], "amax_72": ["choose_qparams_affine_default_72"], "full_145": ["choose_qparams_affine_default_72"], "maximum_72": ["choose_qparams_affine_default_72"], "amin_72": ["choose_qparams_affine_default_72"], "full_144": ["choose_qparams_affine_default_72"], "minimum_72": ["choose_qparams_affine_default_72"], "sub_3306": ["choose_qparams_affine_default_72"], "div_144": ["choose_qparams_affine_default_72"], "clamp_min_216": ["choose_qparams_affine_default_72"], "div_145": ["choose_qparams_affine_default_72"], "round_145": ["choose_qparams_affine_default_72"], "sub_3312": ["choose_qparams_affine_default_72"], "clamp_min_217": ["choose_qparams_affine_default_72"], "clamp_max_144": ["choose_qparams_affine_default_72"], "view_1130": ["quantize_affine_72"], "reciprocal_72": ["quantize_affine_72"], "mul_7029": ["quantize_affine_72"], "mul_7032": ["quantize_affine_72"], "round_146": ["quantize_affine_72"], "convert_element_type_432": ["choose_qparams_affine_default_72"], "view_1131": ["quantize_affine_72"], "add_11135": ["quantize_affine_72"], "clamp_min_218": ["quantize_affine_72"], "clamp_max_145": ["quantize_affine_72"], "convert_element_type_433": ["quantize_affine_72", "dequantize_affine_144"], "view_1135": ["dequantize_affine_144"], "convert_element_type_434": ["dequantize_affine_144"], "convert_element_type_435": ["dequantize_affine_144"], "sub_3332": ["dequantize_affine_144"], "view_1134": ["dequantize_affine_144"], "mul_7054": ["dequantize_affine_144"], "view_1137": ["dequantize_affine_145"], "view_1139": ["dequantize_affine_145"], "convert_element_type_436": ["dequantize_affine_145"], "convert_element_type_437": ["dequantize_affine_145"], "sub_3336": ["dequantize_affine_145"], "view_1138": ["dequantize_affine_145"], "mul_7059": ["dequantize_affine_145"], "view_1140": ["dequantize_affine_145"], "amax_73": ["choose_qparams_affine_default_73"], "full_147": ["choose_qparams_affine_default_73"], "maximum_73": ["choose_qparams_affine_default_73"], "amin_73": ["choose_qparams_affine_default_73"], "full_146": ["choose_qparams_affine_default_73"], "minimum_73": ["choose_qparams_affine_default_73"], "sub_3351": ["choose_qparams_affine_default_73"], "div_146": ["choose_qparams_affine_default_73"], "clamp_min_219": ["choose_qparams_affine_default_73"], "div_147": ["choose_qparams_affine_default_73"], "round_147": ["choose_qparams_affine_default_73"], "sub_3357": ["choose_qparams_affine_default_73"], "clamp_min_220": ["choose_qparams_affine_default_73"], "clamp_max_146": ["choose_qparams_affine_default_73"], "view_1146": ["quantize_affine_73"], "reciprocal_73": ["quantize_affine_73"], "mul_7125": ["quantize_affine_73"], "mul_7128": ["quantize_affine_73"], "round_148": ["quantize_affine_73"], "convert_element_type_438": ["choose_qparams_affine_default_73"], "view_1147": ["quantize_affine_73"], "add_11287": ["quantize_affine_73"], "clamp_min_221": ["quantize_affine_73"], "clamp_max_147": ["quantize_affine_73"], "convert_element_type_439": ["quantize_affine_73", "dequantize_affine_146"], "view_1151": ["dequantize_affine_146"], "convert_element_type_440": ["dequantize_affine_146"], "convert_element_type_441": ["dequantize_affine_146"], "sub_3377": ["dequantize_affine_146"], "view_1150": ["dequantize_affine_146"], "mul_7150": ["dequantize_affine_146"], "view_1153": ["dequantize_affine_147"], "view_1155": ["dequantize_affine_147"], "convert_element_type_442": ["dequantize_affine_147"], "convert_element_type_443": ["dequantize_affine_147"], "sub_3381": ["dequantize_affine_147"], "view_1154": ["dequantize_affine_147"], "mul_7155": ["dequantize_affine_147"], "view_1156": ["dequantize_affine_147"], "amax_74": ["choose_qparams_affine_default_74"], "full_149": ["choose_qparams_affine_default_74"], "maximum_74": ["choose_qparams_affine_default_74"], "amin_74": ["choose_qparams_affine_default_74"], "full_148": ["choose_qparams_affine_default_74"], "minimum_74": ["choose_qparams_affine_default_74"], "sub_3395": ["choose_qparams_affine_default_74"], "div_148": ["choose_qparams_affine_default_74"], "clamp_min_222": ["choose_qparams_affine_default_74"], "div_149": ["choose_qparams_affine_default_74"], "round_149": ["choose_qparams_affine_default_74"], "sub_3401": ["choose_qparams_affine_default_74"], "clamp_min_223": ["choose_qparams_affine_default_74"], "clamp_max_148": ["choose_qparams_affine_default_74"], "view_1162": ["quantize_affine_74"], "reciprocal_74": ["quantize_affine_74"], "mul_7224": ["quantize_affine_74"], "mul_7227": ["quantize_affine_74"], "round_150": ["quantize_affine_74"], "convert_element_type_444": ["choose_qparams_affine_default_74"], "view_1163": ["quantize_affine_74"], "add_11435": ["quantize_affine_74"], "clamp_min_224": ["quantize_affine_74"], "clamp_max_149": ["quantize_affine_74"], "convert_element_type_445": ["quantize_affine_74", "dequantize_affine_148"], "view_1167": ["dequantize_affine_148"], "convert_element_type_446": ["dequantize_affine_148"], "convert_element_type_447": ["dequantize_affine_148"], "sub_3421": ["dequantize_affine_148"], "view_1166": ["dequantize_affine_148"], "mul_7249": ["dequantize_affine_148"], "view_1169": ["dequantize_affine_149"], "view_1171": ["dequantize_affine_149"], "convert_element_type_448": ["dequantize_affine_149"], "convert_element_type_449": ["dequantize_affine_149"], "sub_3425": ["dequantize_affine_149"], "view_1170": ["dequantize_affine_149"], "mul_7254": ["dequantize_affine_149"], "view_1172": ["dequantize_affine_149"], "mul_7064": ["linear_72"], "view_1141": ["linear_72"], "permute_121": ["linear_72"], "mm_default_99": ["linear_72"], "add_tensor_99": ["linear_72"], "view_1142": ["linear_72"], "mul_7071": ["mul_449"], "view_1143": ["view_36"], "permute_122": ["transpose_48"], "clone_98": ["contiguous_48"], "mul_7158": ["linear_73"], "view_1157": ["linear_73"], "permute_123": ["linear_73"], "mm_12": ["linear_73"], "view_1158": ["linear_73"], "view_1159": ["view_37"], "permute_124": ["transpose_49"], "clone_99": ["contiguous_49"], "mul_7259": ["linear_74"], "view_1173": ["linear_74"], "permute_125": ["linear_74"], "mm_default_98": ["linear_74"], "add_tensor_98": ["linear_74"], "view_1174": ["linear_74"], "view_1175": ["view_38"], "permute_126": ["transpose_50"], "clone_100": ["contiguous_50"], "_scaled_dot_product_efficient_attention_12": ["scaled_dot_product_attention_12"], "getitem_98": ["scaled_dot_product_attention_12"], "permute_127": ["transpose_51"], "view_1176": ["reshape_12", "choose_qparams_affine_default_75", "quantize_affine_75"], "amax_75": ["choose_qparams_affine_default_75"], "full_151": ["choose_qparams_affine_default_75"], "maximum_75": ["choose_qparams_affine_default_75"], "amin_75": ["choose_qparams_affine_default_75"], "full_150": ["choose_qparams_affine_default_75"], "minimum_75": ["choose_qparams_affine_default_75"], "sub_3443": ["choose_qparams_affine_default_75"], "div_150": ["choose_qparams_affine_default_75"], "clamp_min_225": ["choose_qparams_affine_default_75"], "div_151": ["choose_qparams_affine_default_75"], "round_151": ["choose_qparams_affine_default_75"], "sub_3449": ["choose_qparams_affine_default_75"], "clamp_min_226": ["choose_qparams_affine_default_75"], "clamp_max_150": ["choose_qparams_affine_default_75"], "view_1179": ["quantize_affine_75"], "reciprocal_75": ["quantize_affine_75"], "mul_7329": ["quantize_affine_75"], "mul_7332": ["quantize_affine_75"], "round_152": ["quantize_affine_75"], "convert_element_type_450": ["choose_qparams_affine_default_75"], "view_1180": ["quantize_affine_75"], "add_11599": ["quantize_affine_75"], "clamp_min_227": ["quantize_affine_75"], "clamp_max_151": ["quantize_affine_75"], "convert_element_type_451": ["quantize_affine_75", "dequantize_affine_150"], "view_1184": ["dequantize_affine_150"], "convert_element_type_452": ["dequantize_affine_150"], "convert_element_type_453": ["dequantize_affine_150"], "sub_3469": ["dequantize_affine_150"], "view_1183": ["dequantize_affine_150"], "mul_7354": ["dequantize_affine_150"], "view_1186": ["dequantize_affine_151"], "view_1188": ["dequantize_affine_151"], "convert_element_type_454": ["dequantize_affine_151"], "convert_element_type_455": ["dequantize_affine_151"], "sub_3473": ["dequantize_affine_151"], "view_1187": ["dequantize_affine_151"], "mul_7359": ["dequantize_affine_151"], "view_1189": ["dequantize_affine_151"], "mul_7364": ["linear_75"], "view_1190": ["linear_75"], "permute_128": ["linear_75"], "mm_default_97": ["linear_75"], "add_tensor_97": ["linear_75"], "view_1191": ["linear_75", "dropout_37"], "add_11662": ["add_177"], "clone_102": ["layer_norm_25"], "var_mean_25": ["layer_norm_25"], "getitem_102": ["layer_norm_25"], "getitem_103": ["layer_norm_25"], "sub_3479": ["layer_norm_25"], "add_11667": ["layer_norm_25"], "rsqrt_25": ["layer_norm_25"], "mul_7375": ["layer_norm_25"], "mul_7376": ["layer_norm_25"], "add_11668": ["layer_norm_25", "choose_qparams_affine_default_76", "quantize_affine_76"], "amax_76": ["choose_qparams_affine_default_76"], "full_153": ["choose_qparams_affine_default_76"], "maximum_76": ["choose_qparams_affine_default_76"], "amin_76": ["choose_qparams_affine_default_76"], "full_152": ["choose_qparams_affine_default_76"], "minimum_76": ["choose_qparams_affine_default_76"], "sub_3490": ["choose_qparams_affine_default_76"], "div_152": ["choose_qparams_affine_default_76"], "clamp_min_228": ["choose_qparams_affine_default_76"], "div_153": ["choose_qparams_affine_default_76"], "round_153": ["choose_qparams_affine_default_76"], "sub_3496": ["choose_qparams_affine_default_76"], "clamp_min_229": ["choose_qparams_affine_default_76"], "clamp_max_152": ["choose_qparams_affine_default_76"], "view_1194": ["quantize_affine_76"], "reciprocal_76": ["quantize_affine_76"], "mul_7424": ["quantize_affine_76"], "mul_7427": ["quantize_affine_76"], "round_154": ["quantize_affine_76"], "convert_element_type_456": ["choose_qparams_affine_default_76"], "view_1195": ["quantize_affine_76"], "add_11755": ["quantize_affine_76"], "clamp_min_230": ["quantize_affine_76"], "clamp_max_153": ["quantize_affine_76"], "convert_element_type_457": ["quantize_affine_76", "dequantize_affine_152"], "view_1199": ["dequantize_affine_152"], "convert_element_type_458": ["dequantize_affine_152"], "convert_element_type_459": ["dequantize_affine_152"], "sub_3516": ["dequantize_affine_152"], "view_1198": ["dequantize_affine_152"], "mul_7449": ["dequantize_affine_152"], "view_1201": ["dequantize_affine_153"], "view_1203": ["dequantize_affine_153"], "convert_element_type_460": ["dequantize_affine_153"], "convert_element_type_461": ["dequantize_affine_153"], "sub_3520": ["dequantize_affine_153"], "view_1202": ["dequantize_affine_153"], "mul_7454": ["dequantize_affine_153"], "view_1204": ["dequantize_affine_153"], "mul_7459": ["linear_76"], "view_1205": ["linear_76"], "permute_129": ["linear_76"], "mm_default_96": ["linear_76"], "add_tensor_96": ["linear_76"], "view_1206": ["linear_76"], "mul_7466": ["gelu_14"], "mul_7467": ["gelu_14"], "erf_14": ["gelu_14"], "add_11814": ["gelu_14"], "mul_7468": ["gelu_14", "dropout_38", "choose_qparams_affine_default_77", "quantize_affine_77"], "amax_77": ["choose_qparams_affine_default_77"], "full_155": ["choose_qparams_affine_default_77"], "maximum_77": ["choose_qparams_affine_default_77"], "amin_77": ["choose_qparams_affine_default_77"], "full_154": ["choose_qparams_affine_default_77"], "minimum_77": ["choose_qparams_affine_default_77"], "sub_3533": ["choose_qparams_affine_default_77"], "div_154": ["choose_qparams_affine_default_77"], "clamp_min_231": ["choose_qparams_affine_default_77"], "div_155": ["choose_qparams_affine_default_77"], "round_155": ["choose_qparams_affine_default_77"], "sub_3539": ["choose_qparams_affine_default_77"], "clamp_min_232": ["choose_qparams_affine_default_77"], "clamp_max_154": ["choose_qparams_affine_default_77"], "view_1209": ["quantize_affine_77"], "reciprocal_77": ["quantize_affine_77"], "mul_7514": ["quantize_affine_77"], "mul_7517": ["quantize_affine_77"], "round_156": ["quantize_affine_77"], "convert_element_type_462": ["choose_qparams_affine_default_77"], "view_1210": ["quantize_affine_77"], "add_11897": ["quantize_affine_77"], "clamp_min_233": ["quantize_affine_77"], "clamp_max_155": ["quantize_affine_77"], "convert_element_type_463": ["quantize_affine_77", "dequantize_affine_154"], "view_1214": ["dequantize_affine_154"], "convert_element_type_464": ["dequantize_affine_154"], "convert_element_type_465": ["dequantize_affine_154"], "sub_3559": ["dequantize_affine_154"], "view_1213": ["dequantize_affine_154"], "mul_7539": ["dequantize_affine_154"], "view_1216": ["dequantize_affine_155"], "view_1218": ["dequantize_affine_155"], "convert_element_type_466": ["dequantize_affine_155"], "convert_element_type_467": ["dequantize_affine_155"], "sub_3563": ["dequantize_affine_155"], "view_1217": ["dequantize_affine_155"], "mul_7544": ["dequantize_affine_155"], "view_1219": ["dequantize_affine_155"], "mul_7549": ["linear_77"], "view_1220": ["linear_77"], "permute_130": ["linear_77"], "mm_default_95": ["linear_77"], "add_tensor_95": ["linear_77"], "view_1221": ["linear_77", "dropout_39"], "add_11960": ["add_182"], "clone_105": ["layer_norm_26"], "var_mean_26": ["layer_norm_26"], "getitem_104": ["layer_norm_26"], "getitem_105": ["layer_norm_26"], "sub_3569": ["layer_norm_26"], "add_11965": ["layer_norm_26"], "rsqrt_26": ["layer_norm_26"], "mul_7560": ["layer_norm_26"], "mul_7561": ["layer_norm_26"], "add_11966": ["layer_norm_26", "choose_qparams_affine_default_78", "quantize_affine_78", "choose_qparams_affine_default_79", "quantize_affine_79", "choose_qparams_affine_default_80", "quantize_affine_80"], "amax_78": ["choose_qparams_affine_default_78"], "full_157": ["choose_qparams_affine_default_78"], "maximum_78": ["choose_qparams_affine_default_78"], "amin_78": ["choose_qparams_affine_default_78"], "full_156": ["choose_qparams_affine_default_78"], "minimum_78": ["choose_qparams_affine_default_78"], "sub_3580": ["choose_qparams_affine_default_78"], "div_156": ["choose_qparams_affine_default_78"], "clamp_min_234": ["choose_qparams_affine_default_78"], "div_157": ["choose_qparams_affine_default_78"], "round_157": ["choose_qparams_affine_default_78"], "sub_3586": ["choose_qparams_affine_default_78"], "clamp_min_235": ["choose_qparams_affine_default_78"], "clamp_max_156": ["choose_qparams_affine_default_78"], "view_1224": ["quantize_affine_78"], "reciprocal_78": ["quantize_affine_78"], "mul_7609": ["quantize_affine_78"], "mul_7612": ["quantize_affine_78"], "round_158": ["quantize_affine_78"], "convert_element_type_468": ["choose_qparams_affine_default_78"], "view_1225": ["quantize_affine_78"], "add_12053": ["quantize_affine_78"], "clamp_min_236": ["quantize_affine_78"], "clamp_max_157": ["quantize_affine_78"], "convert_element_type_469": ["quantize_affine_78", "dequantize_affine_156"], "view_1229": ["dequantize_affine_156"], "convert_element_type_470": ["dequantize_affine_156"], "convert_element_type_471": ["dequantize_affine_156"], "sub_3606": ["dequantize_affine_156"], "view_1228": ["dequantize_affine_156"], "mul_7634": ["dequantize_affine_156"], "view_1231": ["dequantize_affine_157"], "view_1233": ["dequantize_affine_157"], "convert_element_type_472": ["dequantize_affine_157"], "convert_element_type_473": ["dequantize_affine_157"], "sub_3610": ["dequantize_affine_157"], "view_1232": ["dequantize_affine_157"], "mul_7639": ["dequantize_affine_157"], "view_1234": ["dequantize_affine_157"], "amax_79": ["choose_qparams_affine_default_79"], "full_159": ["choose_qparams_affine_default_79"], "maximum_79": ["choose_qparams_affine_default_79"], "amin_79": ["choose_qparams_affine_default_79"], "full_158": ["choose_qparams_affine_default_79"], "minimum_79": ["choose_qparams_affine_default_79"], "sub_3625": ["choose_qparams_affine_default_79"], "div_158": ["choose_qparams_affine_default_79"], "clamp_min_237": ["choose_qparams_affine_default_79"], "div_159": ["choose_qparams_affine_default_79"], "round_159": ["choose_qparams_affine_default_79"], "sub_3631": ["choose_qparams_affine_default_79"], "clamp_min_238": ["choose_qparams_affine_default_79"], "clamp_max_158": ["choose_qparams_affine_default_79"], "view_1240": ["quantize_affine_79"], "reciprocal_79": ["quantize_affine_79"], "mul_7705": ["quantize_affine_79"], "mul_7708": ["quantize_affine_79"], "round_160": ["quantize_affine_79"], "convert_element_type_474": ["choose_qparams_affine_default_79"], "view_1241": ["quantize_affine_79"], "add_12205": ["quantize_affine_79"], "clamp_min_239": ["quantize_affine_79"], "clamp_max_159": ["quantize_affine_79"], "convert_element_type_475": ["quantize_affine_79", "dequantize_affine_158"], "view_1245": ["dequantize_affine_158"], "convert_element_type_476": ["dequantize_affine_158"], "convert_element_type_477": ["dequantize_affine_158"], "sub_3651": ["dequantize_affine_158"], "view_1244": ["dequantize_affine_158"], "mul_7730": ["dequantize_affine_158"], "view_1247": ["dequantize_affine_159"], "view_1249": ["dequantize_affine_159"], "convert_element_type_478": ["dequantize_affine_159"], "convert_element_type_479": ["dequantize_affine_159"], "sub_3655": ["dequantize_affine_159"], "view_1248": ["dequantize_affine_159"], "mul_7735": ["dequantize_affine_159"], "view_1250": ["dequantize_affine_159"], "amax_80": ["choose_qparams_affine_default_80"], "full_161": ["choose_qparams_affine_default_80"], "maximum_80": ["choose_qparams_affine_default_80"], "amin_80": ["choose_qparams_affine_default_80"], "full_160": ["choose_qparams_affine_default_80"], "minimum_80": ["choose_qparams_affine_default_80"], "sub_3669": ["choose_qparams_affine_default_80"], "div_160": ["choose_qparams_affine_default_80"], "clamp_min_240": ["choose_qparams_affine_default_80"], "div_161": ["choose_qparams_affine_default_80"], "round_161": ["choose_qparams_affine_default_80"], "sub_3675": ["choose_qparams_affine_default_80"], "clamp_min_241": ["choose_qparams_affine_default_80"], "clamp_max_160": ["choose_qparams_affine_default_80"], "view_1256": ["quantize_affine_80"], "reciprocal_80": ["quantize_affine_80"], "mul_7804": ["quantize_affine_80"], "mul_7807": ["quantize_affine_80"], "round_162": ["quantize_affine_80"], "convert_element_type_480": ["choose_qparams_affine_default_80"], "view_1257": ["quantize_affine_80"], "add_12353": ["quantize_affine_80"], "clamp_min_242": ["quantize_affine_80"], "clamp_max_161": ["quantize_affine_80"], "convert_element_type_481": ["quantize_affine_80", "dequantize_affine_160"], "view_1261": ["dequantize_affine_160"], "convert_element_type_482": ["dequantize_affine_160"], "convert_element_type_483": ["dequantize_affine_160"], "sub_3695": ["dequantize_affine_160"], "view_1260": ["dequantize_affine_160"], "mul_7829": ["dequantize_affine_160"], "view_1263": ["dequantize_affine_161"], "view_1265": ["dequantize_affine_161"], "convert_element_type_484": ["dequantize_affine_161"], "convert_element_type_485": ["dequantize_affine_161"], "sub_3699": ["dequantize_affine_161"], "view_1264": ["dequantize_affine_161"], "mul_7834": ["dequantize_affine_161"], "view_1266": ["dequantize_affine_161"], "mul_7644": ["linear_78"], "view_1235": ["linear_78"], "permute_131": ["linear_78"], "mm_default_94": ["linear_78"], "add_tensor_94": ["linear_78"], "view_1236": ["linear_78"], "mul_7651": ["mul_486"], "view_1237": ["view_39"], "permute_132": ["transpose_52"], "clone_106": ["contiguous_52"], "mul_7738": ["linear_79"], "view_1251": ["linear_79"], "permute_133": ["linear_79"], "mm_13": ["linear_79"], "view_1252": ["linear_79"], "view_1253": ["view_40"], "permute_134": ["transpose_53"], "clone_107": ["contiguous_53"], "mul_7839": ["linear_80"], "view_1267": ["linear_80"], "permute_135": ["linear_80"], "mm_default_93": ["linear_80"], "add_tensor_93": ["linear_80"], "view_1268": ["linear_80"], "view_1269": ["view_41"], "permute_136": ["transpose_54"], "clone_108": ["contiguous_54"], "_scaled_dot_product_efficient_attention_13": ["scaled_dot_product_attention_13"], "getitem_106": ["scaled_dot_product_attention_13"], "permute_137": ["transpose_55"], "view_1270": ["reshape_13", "choose_qparams_affine_default_81", "quantize_affine_81"], "amax_81": ["choose_qparams_affine_default_81"], "full_163": ["choose_qparams_affine_default_81"], "maximum_81": ["choose_qparams_affine_default_81"], "amin_81": ["choose_qparams_affine_default_81"], "full_162": ["choose_qparams_affine_default_81"], "minimum_81": ["choose_qparams_affine_default_81"], "sub_3717": ["choose_qparams_affine_default_81"], "div_162": ["choose_qparams_affine_default_81"], "clamp_min_243": ["choose_qparams_affine_default_81"], "div_163": ["choose_qparams_affine_default_81"], "round_163": ["choose_qparams_affine_default_81"], "sub_3723": ["choose_qparams_affine_default_81"], "clamp_min_244": ["choose_qparams_affine_default_81"], "clamp_max_162": ["choose_qparams_affine_default_81"], "view_1273": ["quantize_affine_81"], "reciprocal_81": ["quantize_affine_81"], "mul_7909": ["quantize_affine_81"], "mul_7912": ["quantize_affine_81"], "round_164": ["quantize_affine_81"], "convert_element_type_486": ["choose_qparams_affine_default_81"], "view_1274": ["quantize_affine_81"], "add_12517": ["quantize_affine_81"], "clamp_min_245": ["quantize_affine_81"], "clamp_max_163": ["quantize_affine_81"], "convert_element_type_487": ["quantize_affine_81", "dequantize_affine_162"], "view_1278": ["dequantize_affine_162"], "convert_element_type_488": ["dequantize_affine_162"], "convert_element_type_489": ["dequantize_affine_162"], "sub_3743": ["dequantize_affine_162"], "view_1277": ["dequantize_affine_162"], "mul_7934": ["dequantize_affine_162"], "view_1280": ["dequantize_affine_163"], "view_1282": ["dequantize_affine_163"], "convert_element_type_490": ["dequantize_affine_163"], "convert_element_type_491": ["dequantize_affine_163"], "sub_3747": ["dequantize_affine_163"], "view_1281": ["dequantize_affine_163"], "mul_7939": ["dequantize_affine_163"], "view_1283": ["dequantize_affine_163"], "mul_7944": ["linear_81"], "view_1284": ["linear_81"], "permute_138": ["linear_81"], "mm_default_92": ["linear_81"], "add_tensor_92": ["linear_81"], "view_1285": ["linear_81", "dropout_40"], "add_12580": ["add_191"], "clone_110": ["layer_norm_27"], "var_mean_27": ["layer_norm_27"], "getitem_110": ["layer_norm_27"], "getitem_111": ["layer_norm_27"], "sub_3753": ["layer_norm_27"], "add_12585": ["layer_norm_27"], "rsqrt_27": ["layer_norm_27"], "mul_7955": ["layer_norm_27"], "mul_7956": ["layer_norm_27"], "add_12586": ["layer_norm_27", "choose_qparams_affine_default_82", "quantize_affine_82"], "amax_82": ["choose_qparams_affine_default_82"], "full_165": ["choose_qparams_affine_default_82"], "maximum_82": ["choose_qparams_affine_default_82"], "amin_82": ["choose_qparams_affine_default_82"], "full_164": ["choose_qparams_affine_default_82"], "minimum_82": ["choose_qparams_affine_default_82"], "sub_3764": ["choose_qparams_affine_default_82"], "div_164": ["choose_qparams_affine_default_82"], "clamp_min_246": ["choose_qparams_affine_default_82"], "div_165": ["choose_qparams_affine_default_82"], "round_165": ["choose_qparams_affine_default_82"], "sub_3770": ["choose_qparams_affine_default_82"], "clamp_min_247": ["choose_qparams_affine_default_82"], "clamp_max_164": ["choose_qparams_affine_default_82"], "view_1288": ["quantize_affine_82"], "reciprocal_82": ["quantize_affine_82"], "mul_8004": ["quantize_affine_82"], "mul_8007": ["quantize_affine_82"], "round_166": ["quantize_affine_82"], "convert_element_type_492": ["choose_qparams_affine_default_82"], "view_1289": ["quantize_affine_82"], "add_12673": ["quantize_affine_82"], "clamp_min_248": ["quantize_affine_82"], "clamp_max_165": ["quantize_affine_82"], "convert_element_type_493": ["quantize_affine_82", "dequantize_affine_164"], "view_1293": ["dequantize_affine_164"], "convert_element_type_494": ["dequantize_affine_164"], "convert_element_type_495": ["dequantize_affine_164"], "sub_3790": ["dequantize_affine_164"], "view_1292": ["dequantize_affine_164"], "mul_8029": ["dequantize_affine_164"], "view_1295": ["dequantize_affine_165"], "view_1297": ["dequantize_affine_165"], "convert_element_type_496": ["dequantize_affine_165"], "convert_element_type_497": ["dequantize_affine_165"], "sub_3794": ["dequantize_affine_165"], "view_1296": ["dequantize_affine_165"], "mul_8034": ["dequantize_affine_165"], "view_1298": ["dequantize_affine_165"], "mul_8039": ["linear_82"], "view_1299": ["linear_82"], "permute_139": ["linear_82"], "mm_default_91": ["linear_82"], "add_tensor_91": ["linear_82"], "view_1300": ["linear_82"], "mul_8046": ["gelu_15"], "mul_8047": ["gelu_15"], "erf_15": ["gelu_15"], "add_12732": ["gelu_15"], "mul_8048": ["gelu_15", "dropout_41", "choose_qparams_affine_default_83", "quantize_affine_83"], "amax_83": ["choose_qparams_affine_default_83"], "full_167": ["choose_qparams_affine_default_83"], "maximum_83": ["choose_qparams_affine_default_83"], "amin_83": ["choose_qparams_affine_default_83"], "full_166": ["choose_qparams_affine_default_83"], "minimum_83": ["choose_qparams_affine_default_83"], "sub_3807": ["choose_qparams_affine_default_83"], "div_166": ["choose_qparams_affine_default_83"], "clamp_min_249": ["choose_qparams_affine_default_83"], "div_167": ["choose_qparams_affine_default_83"], "round_167": ["choose_qparams_affine_default_83"], "sub_3813": ["choose_qparams_affine_default_83"], "clamp_min_250": ["choose_qparams_affine_default_83"], "clamp_max_166": ["choose_qparams_affine_default_83"], "view_1303": ["quantize_affine_83"], "reciprocal_83": ["quantize_affine_83"], "mul_8094": ["quantize_affine_83"], "mul_8097": ["quantize_affine_83"], "round_168": ["quantize_affine_83"], "convert_element_type_498": ["choose_qparams_affine_default_83"], "view_1304": ["quantize_affine_83"], "add_12815": ["quantize_affine_83"], "clamp_min_251": ["quantize_affine_83"], "clamp_max_167": ["quantize_affine_83"], "convert_element_type_499": ["quantize_affine_83", "dequantize_affine_166"], "view_1308": ["dequantize_affine_166"], "convert_element_type_500": ["dequantize_affine_166"], "convert_element_type_501": ["dequantize_affine_166"], "sub_3833": ["dequantize_affine_166"], "view_1307": ["dequantize_affine_166"], "mul_8119": ["dequantize_affine_166"], "view_1310": ["dequantize_affine_167"], "view_1312": ["dequantize_affine_167"], "convert_element_type_502": ["dequantize_affine_167"], "convert_element_type_503": ["dequantize_affine_167"], "sub_3837": ["dequantize_affine_167"], "view_1311": ["dequantize_affine_167"], "mul_8124": ["dequantize_affine_167"], "view_1313": ["dequantize_affine_167"], "mul_8129": ["linear_83"], "view_1314": ["linear_83"], "permute_140": ["linear_83"], "mm_default_90": ["linear_83"], "add_tensor_90": ["linear_83"], "view_1315": ["linear_83", "dropout_42"], "add_12878": ["add_196"], "clone_113": ["layer_norm_28"], "var_mean_28": ["layer_norm_28"], "getitem_112": ["layer_norm_28"], "getitem_113": ["layer_norm_28"], "sub_3843": ["layer_norm_28"], "add_12883": ["layer_norm_28"], "rsqrt_28": ["layer_norm_28"], "mul_8140": ["layer_norm_28"], "mul_8141": ["layer_norm_28"], "add_12884": ["layer_norm_28", "choose_qparams_affine_default_84", "quantize_affine_84", "choose_qparams_affine_default_85", "quantize_affine_85", "choose_qparams_affine_default_86", "quantize_affine_86"], "amax_84": ["choose_qparams_affine_default_84"], "full_169": ["choose_qparams_affine_default_84"], "maximum_84": ["choose_qparams_affine_default_84"], "amin_84": ["choose_qparams_affine_default_84"], "full_168": ["choose_qparams_affine_default_84"], "minimum_84": ["choose_qparams_affine_default_84"], "sub_3854": ["choose_qparams_affine_default_84"], "div_168": ["choose_qparams_affine_default_84"], "clamp_min_252": ["choose_qparams_affine_default_84"], "div_169": ["choose_qparams_affine_default_84"], "round_169": ["choose_qparams_affine_default_84"], "sub_3860": ["choose_qparams_affine_default_84"], "clamp_min_253": ["choose_qparams_affine_default_84"], "clamp_max_168": ["choose_qparams_affine_default_84"], "view_1318": ["quantize_affine_84"], "reciprocal_84": ["quantize_affine_84"], "mul_8189": ["quantize_affine_84"], "mul_8192": ["quantize_affine_84"], "round_170": ["quantize_affine_84"], "convert_element_type_504": ["choose_qparams_affine_default_84"], "view_1319": ["quantize_affine_84"], "add_12971": ["quantize_affine_84"], "clamp_min_254": ["quantize_affine_84"], "clamp_max_169": ["quantize_affine_84"], "convert_element_type_505": ["quantize_affine_84", "dequantize_affine_168"], "view_1323": ["dequantize_affine_168"], "convert_element_type_506": ["dequantize_affine_168"], "convert_element_type_507": ["dequantize_affine_168"], "sub_3880": ["dequantize_affine_168"], "view_1322": ["dequantize_affine_168"], "mul_8214": ["dequantize_affine_168"], "view_1325": ["dequantize_affine_169"], "view_1327": ["dequantize_affine_169"], "convert_element_type_508": ["dequantize_affine_169"], "convert_element_type_509": ["dequantize_affine_169"], "sub_3884": ["dequantize_affine_169"], "view_1326": ["dequantize_affine_169"], "mul_8219": ["dequantize_affine_169"], "view_1328": ["dequantize_affine_169"], "amax_85": ["choose_qparams_affine_default_85"], "full_171": ["choose_qparams_affine_default_85"], "maximum_85": ["choose_qparams_affine_default_85"], "amin_85": ["choose_qparams_affine_default_85"], "full_170": ["choose_qparams_affine_default_85"], "minimum_85": ["choose_qparams_affine_default_85"], "sub_3899": ["choose_qparams_affine_default_85"], "div_170": ["choose_qparams_affine_default_85"], "clamp_min_255": ["choose_qparams_affine_default_85"], "div_171": ["choose_qparams_affine_default_85"], "round_171": ["choose_qparams_affine_default_85"], "sub_3905": ["choose_qparams_affine_default_85"], "clamp_min_256": ["choose_qparams_affine_default_85"], "clamp_max_170": ["choose_qparams_affine_default_85"], "view_1334": ["quantize_affine_85"], "reciprocal_85": ["quantize_affine_85"], "mul_8285": ["quantize_affine_85"], "mul_8288": ["quantize_affine_85"], "round_172": ["quantize_affine_85"], "convert_element_type_510": ["choose_qparams_affine_default_85"], "view_1335": ["quantize_affine_85"], "add_13123": ["quantize_affine_85"], "clamp_min_257": ["quantize_affine_85"], "clamp_max_171": ["quantize_affine_85"], "convert_element_type_511": ["quantize_affine_85", "dequantize_affine_170"], "view_1339": ["dequantize_affine_170"], "convert_element_type_512": ["dequantize_affine_170"], "convert_element_type_513": ["dequantize_affine_170"], "sub_3925": ["dequantize_affine_170"], "view_1338": ["dequantize_affine_170"], "mul_8310": ["dequantize_affine_170"], "view_1341": ["dequantize_affine_171"], "view_1343": ["dequantize_affine_171"], "convert_element_type_514": ["dequantize_affine_171"], "convert_element_type_515": ["dequantize_affine_171"], "sub_3929": ["dequantize_affine_171"], "view_1342": ["dequantize_affine_171"], "mul_8315": ["dequantize_affine_171"], "view_1344": ["dequantize_affine_171"], "amax_86": ["choose_qparams_affine_default_86"], "full_173": ["choose_qparams_affine_default_86"], "maximum_86": ["choose_qparams_affine_default_86"], "amin_86": ["choose_qparams_affine_default_86"], "full_172": ["choose_qparams_affine_default_86"], "minimum_86": ["choose_qparams_affine_default_86"], "sub_3943": ["choose_qparams_affine_default_86"], "div_172": ["choose_qparams_affine_default_86"], "clamp_min_258": ["choose_qparams_affine_default_86"], "div_173": ["choose_qparams_affine_default_86"], "round_173": ["choose_qparams_affine_default_86"], "sub_3949": ["choose_qparams_affine_default_86"], "clamp_min_259": ["choose_qparams_affine_default_86"], "clamp_max_172": ["choose_qparams_affine_default_86"], "view_1350": ["quantize_affine_86"], "reciprocal_86": ["quantize_affine_86"], "mul_8384": ["quantize_affine_86"], "mul_8387": ["quantize_affine_86"], "round_174": ["quantize_affine_86"], "convert_element_type_516": ["choose_qparams_affine_default_86"], "view_1351": ["quantize_affine_86"], "add_13271": ["quantize_affine_86"], "clamp_min_260": ["quantize_affine_86"], "clamp_max_173": ["quantize_affine_86"], "convert_element_type_517": ["quantize_affine_86", "dequantize_affine_172"], "view_1355": ["dequantize_affine_172"], "convert_element_type_518": ["dequantize_affine_172"], "convert_element_type_519": ["dequantize_affine_172"], "sub_3969": ["dequantize_affine_172"], "view_1354": ["dequantize_affine_172"], "mul_8409": ["dequantize_affine_172"], "view_1357": ["dequantize_affine_173"], "view_1359": ["dequantize_affine_173"], "convert_element_type_520": ["dequantize_affine_173"], "convert_element_type_521": ["dequantize_affine_173"], "sub_3973": ["dequantize_affine_173"], "view_1358": ["dequantize_affine_173"], "mul_8414": ["dequantize_affine_173"], "view_1360": ["dequantize_affine_173"], "mul_8224": ["linear_84"], "view_1329": ["linear_84"], "permute_141": ["linear_84"], "mm_default_89": ["linear_84"], "add_tensor_89": ["linear_84"], "view_1330": ["linear_84"], "mul_8231": ["mul_523"], "view_1331": ["view_42"], "permute_142": ["transpose_56"], "clone_114": ["contiguous_56"], "mul_8318": ["linear_85"], "view_1345": ["linear_85"], "permute_143": ["linear_85"], "mm_14": ["linear_85"], "view_1346": ["linear_85"], "view_1347": ["view_43"], "permute_144": ["transpose_57"], "clone_115": ["contiguous_57"], "mul_8419": ["linear_86"], "view_1361": ["linear_86"], "permute_145": ["linear_86"], "mm_default_88": ["linear_86"], "add_tensor_88": ["linear_86"], "view_1362": ["linear_86"], "view_1363": ["view_44"], "permute_146": ["transpose_58"], "clone_116": ["contiguous_58"], "_scaled_dot_product_efficient_attention_14": ["scaled_dot_product_attention_14"], "getitem_114": ["scaled_dot_product_attention_14"], "permute_147": ["transpose_59"], "view_1364": ["reshape_14", "choose_qparams_affine_default_87", "quantize_affine_87"], "amax_87": ["choose_qparams_affine_default_87"], "full_175": ["choose_qparams_affine_default_87"], "maximum_87": ["choose_qparams_affine_default_87"], "amin_87": ["choose_qparams_affine_default_87"], "full_174": ["choose_qparams_affine_default_87"], "minimum_87": ["choose_qparams_affine_default_87"], "sub_3991": ["choose_qparams_affine_default_87"], "div_174": ["choose_qparams_affine_default_87"], "clamp_min_261": ["choose_qparams_affine_default_87"], "div_175": ["choose_qparams_affine_default_87"], "round_175": ["choose_qparams_affine_default_87"], "sub_3997": ["choose_qparams_affine_default_87"], "clamp_min_262": ["choose_qparams_affine_default_87"], "clamp_max_174": ["choose_qparams_affine_default_87"], "view_1367": ["quantize_affine_87"], "reciprocal_87": ["quantize_affine_87"], "mul_8489": ["quantize_affine_87"], "mul_8492": ["quantize_affine_87"], "round_176": ["quantize_affine_87"], "convert_element_type_522": ["choose_qparams_affine_default_87"], "view_1368": ["quantize_affine_87"], "add_13435": ["quantize_affine_87"], "clamp_min_263": ["quantize_affine_87"], "clamp_max_175": ["quantize_affine_87"], "convert_element_type_523": ["quantize_affine_87", "dequantize_affine_174"], "view_1372": ["dequantize_affine_174"], "convert_element_type_524": ["dequantize_affine_174"], "convert_element_type_525": ["dequantize_affine_174"], "sub_4017": ["dequantize_affine_174"], "view_1371": ["dequantize_affine_174"], "mul_8514": ["dequantize_affine_174"], "view_1374": ["dequantize_affine_175"], "view_1376": ["dequantize_affine_175"], "convert_element_type_526": ["dequantize_affine_175"], "convert_element_type_527": ["dequantize_affine_175"], "sub_4021": ["dequantize_affine_175"], "view_1375": ["dequantize_affine_175"], "mul_8519": ["dequantize_affine_175"], "view_1377": ["dequantize_affine_175"], "mul_8524": ["linear_87"], "view_1378": ["linear_87"], "permute_148": ["linear_87"], "mm_default_87": ["linear_87"], "add_tensor_87": ["linear_87"], "view_1379": ["linear_87", "dropout_43"], "add_13498": ["add_205"], "clone_118": ["layer_norm_29"], "var_mean_29": ["layer_norm_29"], "getitem_118": ["layer_norm_29"], "getitem_119": ["layer_norm_29"], "sub_4027": ["layer_norm_29"], "add_13503": ["layer_norm_29"], "rsqrt_29": ["layer_norm_29"], "mul_8535": ["layer_norm_29"], "mul_8536": ["layer_norm_29"], "add_13504": ["layer_norm_29", "choose_qparams_affine_default_88", "quantize_affine_88"], "amax_88": ["choose_qparams_affine_default_88"], "full_177": ["choose_qparams_affine_default_88"], "maximum_88": ["choose_qparams_affine_default_88"], "amin_88": ["choose_qparams_affine_default_88"], "full_176": ["choose_qparams_affine_default_88"], "minimum_88": ["choose_qparams_affine_default_88"], "sub_4038": ["choose_qparams_affine_default_88"], "div_176": ["choose_qparams_affine_default_88"], "clamp_min_264": ["choose_qparams_affine_default_88"], "div_177": ["choose_qparams_affine_default_88"], "round_177": ["choose_qparams_affine_default_88"], "sub_4044": ["choose_qparams_affine_default_88"], "clamp_min_265": ["choose_qparams_affine_default_88"], "clamp_max_176": ["choose_qparams_affine_default_88"], "view_1382": ["quantize_affine_88"], "reciprocal_88": ["quantize_affine_88"], "mul_8584": ["quantize_affine_88"], "mul_8587": ["quantize_affine_88"], "round_178": ["quantize_affine_88"], "convert_element_type_528": ["choose_qparams_affine_default_88"], "view_1383": ["quantize_affine_88"], "add_13591": ["quantize_affine_88"], "clamp_min_266": ["quantize_affine_88"], "clamp_max_177": ["quantize_affine_88"], "convert_element_type_529": ["quantize_affine_88", "dequantize_affine_176"], "view_1387": ["dequantize_affine_176"], "convert_element_type_530": ["dequantize_affine_176"], "convert_element_type_531": ["dequantize_affine_176"], "sub_4064": ["dequantize_affine_176"], "view_1386": ["dequantize_affine_176"], "mul_8609": ["dequantize_affine_176"], "view_1389": ["dequantize_affine_177"], "view_1391": ["dequantize_affine_177"], "convert_element_type_532": ["dequantize_affine_177"], "convert_element_type_533": ["dequantize_affine_177"], "sub_4068": ["dequantize_affine_177"], "view_1390": ["dequantize_affine_177"], "mul_8614": ["dequantize_affine_177"], "view_1392": ["dequantize_affine_177"], "mul_8619": ["linear_88"], "view_1393": ["linear_88"], "permute_149": ["linear_88"], "mm_default_86": ["linear_88"], "add_tensor_86": ["linear_88"], "view_1394": ["linear_88"], "mul_8626": ["gelu_16"], "mul_8627": ["gelu_16"], "erf_16": ["gelu_16"], "add_13650": ["gelu_16"], "mul_8628": ["gelu_16", "dropout_44", "choose_qparams_affine_default_89", "quantize_affine_89"], "amax_89": ["choose_qparams_affine_default_89"], "full_179": ["choose_qparams_affine_default_89"], "maximum_89": ["choose_qparams_affine_default_89"], "amin_89": ["choose_qparams_affine_default_89"], "full_178": ["choose_qparams_affine_default_89"], "minimum_89": ["choose_qparams_affine_default_89"], "sub_4081": ["choose_qparams_affine_default_89"], "div_178": ["choose_qparams_affine_default_89"], "clamp_min_267": ["choose_qparams_affine_default_89"], "div_179": ["choose_qparams_affine_default_89"], "round_179": ["choose_qparams_affine_default_89"], "sub_4087": ["choose_qparams_affine_default_89"], "clamp_min_268": ["choose_qparams_affine_default_89"], "clamp_max_178": ["choose_qparams_affine_default_89"], "view_1397": ["quantize_affine_89"], "reciprocal_89": ["quantize_affine_89"], "mul_8674": ["quantize_affine_89"], "mul_8677": ["quantize_affine_89"], "round_180": ["quantize_affine_89"], "convert_element_type_534": ["choose_qparams_affine_default_89"], "view_1398": ["quantize_affine_89"], "add_13733": ["quantize_affine_89"], "clamp_min_269": ["quantize_affine_89"], "clamp_max_179": ["quantize_affine_89"], "convert_element_type_535": ["quantize_affine_89", "dequantize_affine_178"], "view_1402": ["dequantize_affine_178"], "convert_element_type_536": ["dequantize_affine_178"], "convert_element_type_537": ["dequantize_affine_178"], "sub_4107": ["dequantize_affine_178"], "view_1401": ["dequantize_affine_178"], "mul_8699": ["dequantize_affine_178"], "view_1404": ["dequantize_affine_179"], "view_1406": ["dequantize_affine_179"], "convert_element_type_538": ["dequantize_affine_179"], "convert_element_type_539": ["dequantize_affine_179"], "sub_4111": ["dequantize_affine_179"], "view_1405": ["dequantize_affine_179"], "mul_8704": ["dequantize_affine_179"], "view_1407": ["dequantize_affine_179"], "mul_8709": ["linear_89"], "view_1408": ["linear_89"], "permute_150": ["linear_89"], "mm_default_85": ["linear_89"], "add_tensor_85": ["linear_89"], "view_1409": ["linear_89", "dropout_45"], "add_13796": ["add_210"], "clone_121": ["layer_norm_30"], "var_mean_30": ["layer_norm_30"], "getitem_120": ["layer_norm_30"], "getitem_121": ["layer_norm_30"], "sub_4117": ["layer_norm_30"], "add_13801": ["layer_norm_30"], "rsqrt_30": ["layer_norm_30"], "mul_8720": ["layer_norm_30"], "mul_8721": ["layer_norm_30"], "add_13802": ["layer_norm_30", "choose_qparams_affine_default_90", "quantize_affine_90", "choose_qparams_affine_default_91", "quantize_affine_91", "choose_qparams_affine_default_92", "quantize_affine_92"], "amax_90": ["choose_qparams_affine_default_90"], "full_181": ["choose_qparams_affine_default_90"], "maximum_90": ["choose_qparams_affine_default_90"], "amin_90": ["choose_qparams_affine_default_90"], "full_180": ["choose_qparams_affine_default_90"], "minimum_90": ["choose_qparams_affine_default_90"], "sub_4128": ["choose_qparams_affine_default_90"], "div_180": ["choose_qparams_affine_default_90"], "clamp_min_270": ["choose_qparams_affine_default_90"], "div_181": ["choose_qparams_affine_default_90"], "round_181": ["choose_qparams_affine_default_90"], "sub_4134": ["choose_qparams_affine_default_90"], "clamp_min_271": ["choose_qparams_affine_default_90"], "clamp_max_180": ["choose_qparams_affine_default_90"], "view_1412": ["quantize_affine_90"], "reciprocal_90": ["quantize_affine_90"], "mul_8769": ["quantize_affine_90"], "mul_8772": ["quantize_affine_90"], "round_182": ["quantize_affine_90"], "convert_element_type_540": ["choose_qparams_affine_default_90"], "view_1413": ["quantize_affine_90"], "add_13889": ["quantize_affine_90"], "clamp_min_272": ["quantize_affine_90"], "clamp_max_181": ["quantize_affine_90"], "convert_element_type_541": ["quantize_affine_90", "dequantize_affine_180"], "view_1417": ["dequantize_affine_180"], "convert_element_type_542": ["dequantize_affine_180"], "convert_element_type_543": ["dequantize_affine_180"], "sub_4154": ["dequantize_affine_180"], "view_1416": ["dequantize_affine_180"], "mul_8794": ["dequantize_affine_180"], "view_1419": ["dequantize_affine_181"], "view_1421": ["dequantize_affine_181"], "convert_element_type_544": ["dequantize_affine_181"], "convert_element_type_545": ["dequantize_affine_181"], "sub_4158": ["dequantize_affine_181"], "view_1420": ["dequantize_affine_181"], "mul_8799": ["dequantize_affine_181"], "view_1422": ["dequantize_affine_181"], "amax_91": ["choose_qparams_affine_default_91"], "full_183": ["choose_qparams_affine_default_91"], "maximum_91": ["choose_qparams_affine_default_91"], "amin_91": ["choose_qparams_affine_default_91"], "full_182": ["choose_qparams_affine_default_91"], "minimum_91": ["choose_qparams_affine_default_91"], "sub_4173": ["choose_qparams_affine_default_91"], "div_182": ["choose_qparams_affine_default_91"], "clamp_min_273": ["choose_qparams_affine_default_91"], "div_183": ["choose_qparams_affine_default_91"], "round_183": ["choose_qparams_affine_default_91"], "sub_4179": ["choose_qparams_affine_default_91"], "clamp_min_274": ["choose_qparams_affine_default_91"], "clamp_max_182": ["choose_qparams_affine_default_91"], "view_1428": ["quantize_affine_91"], "reciprocal_91": ["quantize_affine_91"], "mul_8865": ["quantize_affine_91"], "mul_8868": ["quantize_affine_91"], "round_184": ["quantize_affine_91"], "convert_element_type_546": ["choose_qparams_affine_default_91"], "view_1429": ["quantize_affine_91"], "add_14041": ["quantize_affine_91"], "clamp_min_275": ["quantize_affine_91"], "clamp_max_183": ["quantize_affine_91"], "convert_element_type_547": ["quantize_affine_91", "dequantize_affine_182"], "view_1433": ["dequantize_affine_182"], "convert_element_type_548": ["dequantize_affine_182"], "convert_element_type_549": ["dequantize_affine_182"], "sub_4199": ["dequantize_affine_182"], "view_1432": ["dequantize_affine_182"], "mul_8890": ["dequantize_affine_182"], "view_1435": ["dequantize_affine_183"], "view_1437": ["dequantize_affine_183"], "convert_element_type_550": ["dequantize_affine_183"], "convert_element_type_551": ["dequantize_affine_183"], "sub_4203": ["dequantize_affine_183"], "view_1436": ["dequantize_affine_183"], "mul_8895": ["dequantize_affine_183"], "view_1438": ["dequantize_affine_183"], "amax_92": ["choose_qparams_affine_default_92"], "full_185": ["choose_qparams_affine_default_92"], "maximum_92": ["choose_qparams_affine_default_92"], "amin_92": ["choose_qparams_affine_default_92"], "full_184": ["choose_qparams_affine_default_92"], "minimum_92": ["choose_qparams_affine_default_92"], "sub_4217": ["choose_qparams_affine_default_92"], "div_184": ["choose_qparams_affine_default_92"], "clamp_min_276": ["choose_qparams_affine_default_92"], "div_185": ["choose_qparams_affine_default_92"], "round_185": ["choose_qparams_affine_default_92"], "sub_4223": ["choose_qparams_affine_default_92"], "clamp_min_277": ["choose_qparams_affine_default_92"], "clamp_max_184": ["choose_qparams_affine_default_92"], "view_1444": ["quantize_affine_92"], "reciprocal_92": ["quantize_affine_92"], "mul_8964": ["quantize_affine_92"], "mul_8967": ["quantize_affine_92"], "round_186": ["quantize_affine_92"], "convert_element_type_552": ["choose_qparams_affine_default_92"], "view_1445": ["quantize_affine_92"], "add_14189": ["quantize_affine_92"], "clamp_min_278": ["quantize_affine_92"], "clamp_max_185": ["quantize_affine_92"], "convert_element_type_553": ["quantize_affine_92", "dequantize_affine_184"], "view_1449": ["dequantize_affine_184"], "convert_element_type_554": ["dequantize_affine_184"], "convert_element_type_555": ["dequantize_affine_184"], "sub_4243": ["dequantize_affine_184"], "view_1448": ["dequantize_affine_184"], "mul_8989": ["dequantize_affine_184"], "view_1451": ["dequantize_affine_185"], "view_1453": ["dequantize_affine_185"], "convert_element_type_556": ["dequantize_affine_185"], "convert_element_type_557": ["dequantize_affine_185"], "sub_4247": ["dequantize_affine_185"], "view_1452": ["dequantize_affine_185"], "mul_8994": ["dequantize_affine_185"], "view_1454": ["dequantize_affine_185"], "mul_8804": ["linear_90"], "view_1423": ["linear_90"], "permute_151": ["linear_90"], "mm_default_84": ["linear_90"], "add_tensor_84": ["linear_90"], "view_1424": ["linear_90"], "mul_8811": ["mul_560"], "view_1425": ["view_45"], "permute_152": ["transpose_60"], "clone_122": ["contiguous_60"], "mul_8898": ["linear_91"], "view_1439": ["linear_91"], "permute_153": ["linear_91"], "mm_15": ["linear_91"], "view_1440": ["linear_91"], "view_1441": ["view_46"], "permute_154": ["transpose_61"], "clone_123": ["contiguous_61"], "mul_8999": ["linear_92"], "view_1455": ["linear_92"], "permute_155": ["linear_92"], "mm_default_83": ["linear_92"], "add_tensor_83": ["linear_92"], "view_1456": ["linear_92"], "view_1457": ["view_47"], "permute_156": ["transpose_62"], "clone_124": ["contiguous_62"], "_scaled_dot_product_efficient_attention_15": ["scaled_dot_product_attention_15"], "getitem_122": ["scaled_dot_product_attention_15"], "permute_157": ["transpose_63"], "view_1458": ["reshape_15", "choose_qparams_affine_default_93", "quantize_affine_93"], "amax_93": ["choose_qparams_affine_default_93"], "full_187": ["choose_qparams_affine_default_93"], "maximum_93": ["choose_qparams_affine_default_93"], "amin_93": ["choose_qparams_affine_default_93"], "full_186": ["choose_qparams_affine_default_93"], "minimum_93": ["choose_qparams_affine_default_93"], "sub_4265": ["choose_qparams_affine_default_93"], "div_186": ["choose_qparams_affine_default_93"], "clamp_min_279": ["choose_qparams_affine_default_93"], "div_187": ["choose_qparams_affine_default_93"], "round_187": ["choose_qparams_affine_default_93"], "sub_4271": ["choose_qparams_affine_default_93"], "clamp_min_280": ["choose_qparams_affine_default_93"], "clamp_max_186": ["choose_qparams_affine_default_93"], "view_1461": ["quantize_affine_93"], "reciprocal_93": ["quantize_affine_93"], "mul_9069": ["quantize_affine_93"], "mul_9072": ["quantize_affine_93"], "round_188": ["quantize_affine_93"], "convert_element_type_558": ["choose_qparams_affine_default_93"], "view_1462": ["quantize_affine_93"], "add_14353": ["quantize_affine_93"], "clamp_min_281": ["quantize_affine_93"], "clamp_max_187": ["quantize_affine_93"], "convert_element_type_559": ["quantize_affine_93", "dequantize_affine_186"], "view_1466": ["dequantize_affine_186"], "convert_element_type_560": ["dequantize_affine_186"], "convert_element_type_561": ["dequantize_affine_186"], "sub_4291": ["dequantize_affine_186"], "view_1465": ["dequantize_affine_186"], "mul_9094": ["dequantize_affine_186"], "view_1468": ["dequantize_affine_187"], "view_1470": ["dequantize_affine_187"], "convert_element_type_562": ["dequantize_affine_187"], "convert_element_type_563": ["dequantize_affine_187"], "sub_4295": ["dequantize_affine_187"], "view_1469": ["dequantize_affine_187"], "mul_9099": ["dequantize_affine_187"], "view_1471": ["dequantize_affine_187"], "mul_9104": ["linear_93"], "view_1472": ["linear_93"], "permute_158": ["linear_93"], "mm_default_82": ["linear_93"], "add_tensor_82": ["linear_93"], "view_1473": ["linear_93", "dropout_46"], "add_14416": ["add_219"], "clone_126": ["layer_norm_31"], "var_mean_31": ["layer_norm_31"], "getitem_126": ["layer_norm_31"], "getitem_127": ["layer_norm_31"], "sub_4301": ["layer_norm_31"], "add_14421": ["layer_norm_31"], "rsqrt_31": ["layer_norm_31"], "mul_9115": ["layer_norm_31"], "mul_9116": ["layer_norm_31"], "add_14422": ["layer_norm_31", "choose_qparams_affine_default_94", "quantize_affine_94"], "amax_94": ["choose_qparams_affine_default_94"], "full_189": ["choose_qparams_affine_default_94"], "maximum_94": ["choose_qparams_affine_default_94"], "amin_94": ["choose_qparams_affine_default_94"], "full_188": ["choose_qparams_affine_default_94"], "minimum_94": ["choose_qparams_affine_default_94"], "sub_4312": ["choose_qparams_affine_default_94"], "div_188": ["choose_qparams_affine_default_94"], "clamp_min_282": ["choose_qparams_affine_default_94"], "div_189": ["choose_qparams_affine_default_94"], "round_189": ["choose_qparams_affine_default_94"], "sub_4318": ["choose_qparams_affine_default_94"], "clamp_min_283": ["choose_qparams_affine_default_94"], "clamp_max_188": ["choose_qparams_affine_default_94"], "view_1476": ["quantize_affine_94"], "reciprocal_94": ["quantize_affine_94"], "mul_9164": ["quantize_affine_94"], "mul_9167": ["quantize_affine_94"], "round_190": ["quantize_affine_94"], "convert_element_type_564": ["choose_qparams_affine_default_94"], "view_1477": ["quantize_affine_94"], "add_14509": ["quantize_affine_94"], "clamp_min_284": ["quantize_affine_94"], "clamp_max_189": ["quantize_affine_94"], "convert_element_type_565": ["quantize_affine_94", "dequantize_affine_188"], "view_1481": ["dequantize_affine_188"], "convert_element_type_566": ["dequantize_affine_188"], "convert_element_type_567": ["dequantize_affine_188"], "sub_4338": ["dequantize_affine_188"], "view_1480": ["dequantize_affine_188"], "mul_9189": ["dequantize_affine_188"], "view_1483": ["dequantize_affine_189"], "view_1485": ["dequantize_affine_189"], "convert_element_type_568": ["dequantize_affine_189"], "convert_element_type_569": ["dequantize_affine_189"], "sub_4342": ["dequantize_affine_189"], "view_1484": ["dequantize_affine_189"], "mul_9194": ["dequantize_affine_189"], "view_1486": ["dequantize_affine_189"], "mul_9199": ["linear_94"], "view_1487": ["linear_94"], "permute_159": ["linear_94"], "mm_default_81": ["linear_94"], "add_tensor_81": ["linear_94"], "view_1488": ["linear_94"], "mul_9206": ["gelu_17"], "mul_9207": ["gelu_17"], "erf_17": ["gelu_17"], "add_14568": ["gelu_17"], "mul_9208": ["gelu_17", "dropout_47", "choose_qparams_affine_default_95", "quantize_affine_95"], "amax_95": ["choose_qparams_affine_default_95"], "full_191": ["choose_qparams_affine_default_95"], "maximum_95": ["choose_qparams_affine_default_95"], "amin_95": ["choose_qparams_affine_default_95"], "full_190": ["choose_qparams_affine_default_95"], "minimum_95": ["choose_qparams_affine_default_95"], "sub_4355": ["choose_qparams_affine_default_95"], "div_190": ["choose_qparams_affine_default_95"], "clamp_min_285": ["choose_qparams_affine_default_95"], "div_191": ["choose_qparams_affine_default_95"], "round_191": ["choose_qparams_affine_default_95"], "sub_4361": ["choose_qparams_affine_default_95"], "clamp_min_286": ["choose_qparams_affine_default_95"], "clamp_max_190": ["choose_qparams_affine_default_95"], "view_1491": ["quantize_affine_95"], "reciprocal_95": ["quantize_affine_95"], "mul_9254": ["quantize_affine_95"], "mul_9257": ["quantize_affine_95"], "round_192": ["quantize_affine_95"], "convert_element_type_570": ["choose_qparams_affine_default_95"], "view_1492": ["quantize_affine_95"], "add_14651": ["quantize_affine_95"], "clamp_min_287": ["quantize_affine_95"], "clamp_max_191": ["quantize_affine_95"], "convert_element_type_571": ["quantize_affine_95", "dequantize_affine_190"], "view_1496": ["dequantize_affine_190"], "convert_element_type_572": ["dequantize_affine_190"], "convert_element_type_573": ["dequantize_affine_190"], "sub_4381": ["dequantize_affine_190"], "view_1495": ["dequantize_affine_190"], "mul_9279": ["dequantize_affine_190"], "view_1498": ["dequantize_affine_191"], "view_1500": ["dequantize_affine_191"], "convert_element_type_574": ["dequantize_affine_191"], "convert_element_type_575": ["dequantize_affine_191"], "sub_4385": ["dequantize_affine_191"], "view_1499": ["dequantize_affine_191"], "mul_9284": ["dequantize_affine_191"], "view_1501": ["dequantize_affine_191"], "mul_9289": ["linear_95"], "view_1502": ["linear_95"], "permute_160": ["linear_95"], "mm_default_80": ["linear_95"], "add_tensor_80": ["linear_95"], "view_1503": ["linear_95", "dropout_48"], "add_14714": ["add_224"], "clone_129": ["layer_norm_32"], "var_mean_32": ["layer_norm_32"], "getitem_128": ["layer_norm_32"], "getitem_129": ["layer_norm_32"], "sub_4391": ["layer_norm_32"], "add_14719": ["layer_norm_32"], "rsqrt_32": ["layer_norm_32"], "mul_9300": ["layer_norm_32"], "mul_9301": ["layer_norm_32"], "add_14720": ["layer_norm_32", "choose_qparams_affine_default_96", "quantize_affine_96", "choose_qparams_affine_default_97", "quantize_affine_97", "choose_qparams_affine_default_98", "quantize_affine_98"], "amax_96": ["choose_qparams_affine_default_96"], "full_193": ["choose_qparams_affine_default_96"], "maximum_96": ["choose_qparams_affine_default_96"], "amin_96": ["choose_qparams_affine_default_96"], "full_192": ["choose_qparams_affine_default_96"], "minimum_96": ["choose_qparams_affine_default_96"], "sub_4402": ["choose_qparams_affine_default_96"], "div_192": ["choose_qparams_affine_default_96"], "clamp_min_288": ["choose_qparams_affine_default_96"], "div_193": ["choose_qparams_affine_default_96"], "round_193": ["choose_qparams_affine_default_96"], "sub_4408": ["choose_qparams_affine_default_96"], "clamp_min_289": ["choose_qparams_affine_default_96"], "clamp_max_192": ["choose_qparams_affine_default_96"], "view_1506": ["quantize_affine_96"], "reciprocal_96": ["quantize_affine_96"], "mul_9349": ["quantize_affine_96"], "mul_9352": ["quantize_affine_96"], "round_194": ["quantize_affine_96"], "convert_element_type_576": ["choose_qparams_affine_default_96"], "view_1507": ["quantize_affine_96"], "add_14807": ["quantize_affine_96"], "clamp_min_290": ["quantize_affine_96"], "clamp_max_193": ["quantize_affine_96"], "convert_element_type_577": ["quantize_affine_96", "dequantize_affine_192"], "view_1511": ["dequantize_affine_192"], "convert_element_type_578": ["dequantize_affine_192"], "convert_element_type_579": ["dequantize_affine_192"], "sub_4428": ["dequantize_affine_192"], "view_1510": ["dequantize_affine_192"], "mul_9374": ["dequantize_affine_192"], "view_1513": ["dequantize_affine_193"], "view_1515": ["dequantize_affine_193"], "convert_element_type_580": ["dequantize_affine_193"], "convert_element_type_581": ["dequantize_affine_193"], "sub_4432": ["dequantize_affine_193"], "view_1514": ["dequantize_affine_193"], "mul_9379": ["dequantize_affine_193"], "view_1516": ["dequantize_affine_193"], "amax_97": ["choose_qparams_affine_default_97"], "full_195": ["choose_qparams_affine_default_97"], "maximum_97": ["choose_qparams_affine_default_97"], "amin_97": ["choose_qparams_affine_default_97"], "full_194": ["choose_qparams_affine_default_97"], "minimum_97": ["choose_qparams_affine_default_97"], "sub_4447": ["choose_qparams_affine_default_97"], "div_194": ["choose_qparams_affine_default_97"], "clamp_min_291": ["choose_qparams_affine_default_97"], "div_195": ["choose_qparams_affine_default_97"], "round_195": ["choose_qparams_affine_default_97"], "sub_4453": ["choose_qparams_affine_default_97"], "clamp_min_292": ["choose_qparams_affine_default_97"], "clamp_max_194": ["choose_qparams_affine_default_97"], "view_1522": ["quantize_affine_97"], "reciprocal_97": ["quantize_affine_97"], "mul_9445": ["quantize_affine_97"], "mul_9448": ["quantize_affine_97"], "round_196": ["quantize_affine_97"], "convert_element_type_582": ["choose_qparams_affine_default_97"], "view_1523": ["quantize_affine_97"], "add_14959": ["quantize_affine_97"], "clamp_min_293": ["quantize_affine_97"], "clamp_max_195": ["quantize_affine_97"], "convert_element_type_583": ["quantize_affine_97", "dequantize_affine_194"], "view_1527": ["dequantize_affine_194"], "convert_element_type_584": ["dequantize_affine_194"], "convert_element_type_585": ["dequantize_affine_194"], "sub_4473": ["dequantize_affine_194"], "view_1526": ["dequantize_affine_194"], "mul_9470": ["dequantize_affine_194"], "view_1529": ["dequantize_affine_195"], "view_1531": ["dequantize_affine_195"], "convert_element_type_586": ["dequantize_affine_195"], "convert_element_type_587": ["dequantize_affine_195"], "sub_4477": ["dequantize_affine_195"], "view_1530": ["dequantize_affine_195"], "mul_9475": ["dequantize_affine_195"], "view_1532": ["dequantize_affine_195"], "amax_98": ["choose_qparams_affine_default_98"], "full_197": ["choose_qparams_affine_default_98"], "maximum_98": ["choose_qparams_affine_default_98"], "amin_98": ["choose_qparams_affine_default_98"], "full_196": ["choose_qparams_affine_default_98"], "minimum_98": ["choose_qparams_affine_default_98"], "sub_4491": ["choose_qparams_affine_default_98"], "div_196": ["choose_qparams_affine_default_98"], "clamp_min_294": ["choose_qparams_affine_default_98"], "div_197": ["choose_qparams_affine_default_98"], "round_197": ["choose_qparams_affine_default_98"], "sub_4497": ["choose_qparams_affine_default_98"], "clamp_min_295": ["choose_qparams_affine_default_98"], "clamp_max_196": ["choose_qparams_affine_default_98"], "view_1538": ["quantize_affine_98"], "reciprocal_98": ["quantize_affine_98"], "mul_9544": ["quantize_affine_98"], "mul_9547": ["quantize_affine_98"], "round_198": ["quantize_affine_98"], "convert_element_type_588": ["choose_qparams_affine_default_98"], "view_1539": ["quantize_affine_98"], "add_15107": ["quantize_affine_98"], "clamp_min_296": ["quantize_affine_98"], "clamp_max_197": ["quantize_affine_98"], "convert_element_type_589": ["quantize_affine_98", "dequantize_affine_196"], "view_1543": ["dequantize_affine_196"], "convert_element_type_590": ["dequantize_affine_196"], "convert_element_type_591": ["dequantize_affine_196"], "sub_4517": ["dequantize_affine_196"], "view_1542": ["dequantize_affine_196"], "mul_9569": ["dequantize_affine_196"], "view_1545": ["dequantize_affine_197"], "view_1547": ["dequantize_affine_197"], "convert_element_type_592": ["dequantize_affine_197"], "convert_element_type_593": ["dequantize_affine_197"], "sub_4521": ["dequantize_affine_197"], "view_1546": ["dequantize_affine_197"], "mul_9574": ["dequantize_affine_197"], "view_1548": ["dequantize_affine_197"], "mul_9384": ["linear_96"], "view_1517": ["linear_96"], "permute_161": ["linear_96"], "mm_default_79": ["linear_96"], "add_tensor_79": ["linear_96"], "view_1518": ["linear_96"], "mul_9391": ["mul_597"], "view_1519": ["view_48"], "permute_162": ["transpose_64"], "clone_130": ["contiguous_64"], "mul_9478": ["linear_97"], "view_1533": ["linear_97"], "permute_163": ["linear_97"], "mm_16": ["linear_97"], "view_1534": ["linear_97"], "view_1535": ["view_49"], "permute_164": ["transpose_65"], "clone_131": ["contiguous_65"], "mul_9579": ["linear_98"], "view_1549": ["linear_98"], "permute_165": ["linear_98"], "mm_default_78": ["linear_98"], "add_tensor_78": ["linear_98"], "view_1550": ["linear_98"], "view_1551": ["view_50"], "permute_166": ["transpose_66"], "clone_132": ["contiguous_66"], "_scaled_dot_product_efficient_attention_16": ["scaled_dot_product_attention_16"], "getitem_130": ["scaled_dot_product_attention_16"], "permute_167": ["transpose_67"], "view_1552": ["reshape_16", "choose_qparams_affine_default_99", "quantize_affine_99"], "amax_99": ["choose_qparams_affine_default_99"], "full_199": ["choose_qparams_affine_default_99"], "maximum_99": ["choose_qparams_affine_default_99"], "amin_99": ["choose_qparams_affine_default_99"], "full_198": ["choose_qparams_affine_default_99"], "minimum_99": ["choose_qparams_affine_default_99"], "sub_4539": ["choose_qparams_affine_default_99"], "div_198": ["choose_qparams_affine_default_99"], "clamp_min_297": ["choose_qparams_affine_default_99"], "div_199": ["choose_qparams_affine_default_99"], "round_199": ["choose_qparams_affine_default_99"], "sub_4545": ["choose_qparams_affine_default_99"], "clamp_min_298": ["choose_qparams_affine_default_99"], "clamp_max_198": ["choose_qparams_affine_default_99"], "view_1555": ["quantize_affine_99"], "reciprocal_99": ["quantize_affine_99"], "mul_9649": ["quantize_affine_99"], "mul_9652": ["quantize_affine_99"], "round_200": ["quantize_affine_99"], "convert_element_type_594": ["choose_qparams_affine_default_99"], "view_1556": ["quantize_affine_99"], "add_15271": ["quantize_affine_99"], "clamp_min_299": ["quantize_affine_99"], "clamp_max_199": ["quantize_affine_99"], "convert_element_type_595": ["quantize_affine_99", "dequantize_affine_198"], "view_1560": ["dequantize_affine_198"], "convert_element_type_596": ["dequantize_affine_198"], "convert_element_type_597": ["dequantize_affine_198"], "sub_4565": ["dequantize_affine_198"], "view_1559": ["dequantize_affine_198"], "mul_9674": ["dequantize_affine_198"], "view_1562": ["dequantize_affine_199"], "view_1564": ["dequantize_affine_199"], "convert_element_type_598": ["dequantize_affine_199"], "convert_element_type_599": ["dequantize_affine_199"], "sub_4569": ["dequantize_affine_199"], "view_1563": ["dequantize_affine_199"], "mul_9679": ["dequantize_affine_199"], "view_1565": ["dequantize_affine_199"], "mul_9684": ["linear_99"], "view_1566": ["linear_99"], "permute_168": ["linear_99"], "mm_default_77": ["linear_99"], "add_tensor_77": ["linear_99"], "view_1567": ["linear_99", "dropout_49"], "add_15334": ["add_233"], "clone_134": ["layer_norm_33"], "var_mean_33": ["layer_norm_33"], "getitem_134": ["layer_norm_33"], "getitem_135": ["layer_norm_33"], "sub_4575": ["layer_norm_33"], "add_15339": ["layer_norm_33"], "rsqrt_33": ["layer_norm_33"], "mul_9695": ["layer_norm_33"], "mul_9696": ["layer_norm_33"], "add_15340": ["layer_norm_33", "choose_qparams_affine_default_100", "quantize_affine_100"], "amax_100": ["choose_qparams_affine_default_100"], "full_201": ["choose_qparams_affine_default_100"], "maximum_100": ["choose_qparams_affine_default_100"], "amin_100": ["choose_qparams_affine_default_100"], "full_200": ["choose_qparams_affine_default_100"], "minimum_100": ["choose_qparams_affine_default_100"], "sub_4586": ["choose_qparams_affine_default_100"], "div_200": ["choose_qparams_affine_default_100"], "clamp_min_300": ["choose_qparams_affine_default_100"], "div_201": ["choose_qparams_affine_default_100"], "round_201": ["choose_qparams_affine_default_100"], "sub_4592": ["choose_qparams_affine_default_100"], "clamp_min_301": ["choose_qparams_affine_default_100"], "clamp_max_200": ["choose_qparams_affine_default_100"], "view_1570": ["quantize_affine_100"], "reciprocal_100": ["quantize_affine_100"], "mul_9744": ["quantize_affine_100"], "mul_9747": ["quantize_affine_100"], "round_202": ["quantize_affine_100"], "convert_element_type_600": ["choose_qparams_affine_default_100"], "view_1571": ["quantize_affine_100"], "add_15427": ["quantize_affine_100"], "clamp_min_302": ["quantize_affine_100"], "clamp_max_201": ["quantize_affine_100"], "convert_element_type_601": ["quantize_affine_100", "dequantize_affine_200"], "view_1575": ["dequantize_affine_200"], "convert_element_type_602": ["dequantize_affine_200"], "convert_element_type_603": ["dequantize_affine_200"], "sub_4612": ["dequantize_affine_200"], "view_1574": ["dequantize_affine_200"], "mul_9769": ["dequantize_affine_200"], "view_1577": ["dequantize_affine_201"], "view_1579": ["dequantize_affine_201"], "convert_element_type_604": ["dequantize_affine_201"], "convert_element_type_605": ["dequantize_affine_201"], "sub_4616": ["dequantize_affine_201"], "view_1578": ["dequantize_affine_201"], "mul_9774": ["dequantize_affine_201"], "view_1580": ["dequantize_affine_201"], "mul_9779": ["linear_100"], "view_1581": ["linear_100"], "permute_169": ["linear_100"], "mm_default_76": ["linear_100"], "add_tensor_76": ["linear_100"], "view_1582": ["linear_100"], "mul_9786": ["gelu_18"], "mul_9787": ["gelu_18"], "erf_18": ["gelu_18"], "add_15486": ["gelu_18"], "mul_9788": ["gelu_18", "dropout_50", "choose_qparams_affine_default_101", "quantize_affine_101"], "amax_101": ["choose_qparams_affine_default_101"], "full_203": ["choose_qparams_affine_default_101"], "maximum_101": ["choose_qparams_affine_default_101"], "amin_101": ["choose_qparams_affine_default_101"], "full_202": ["choose_qparams_affine_default_101"], "minimum_101": ["choose_qparams_affine_default_101"], "sub_4629": ["choose_qparams_affine_default_101"], "div_202": ["choose_qparams_affine_default_101"], "clamp_min_303": ["choose_qparams_affine_default_101"], "div_203": ["choose_qparams_affine_default_101"], "round_203": ["choose_qparams_affine_default_101"], "sub_4635": ["choose_qparams_affine_default_101"], "clamp_min_304": ["choose_qparams_affine_default_101"], "clamp_max_202": ["choose_qparams_affine_default_101"], "view_1585": ["quantize_affine_101"], "reciprocal_101": ["quantize_affine_101"], "mul_9834": ["quantize_affine_101"], "mul_9837": ["quantize_affine_101"], "round_204": ["quantize_affine_101"], "convert_element_type_606": ["choose_qparams_affine_default_101"], "view_1586": ["quantize_affine_101"], "add_15569": ["quantize_affine_101"], "clamp_min_305": ["quantize_affine_101"], "clamp_max_203": ["quantize_affine_101"], "convert_element_type_607": ["quantize_affine_101", "dequantize_affine_202"], "view_1590": ["dequantize_affine_202"], "convert_element_type_608": ["dequantize_affine_202"], "convert_element_type_609": ["dequantize_affine_202"], "sub_4655": ["dequantize_affine_202"], "view_1589": ["dequantize_affine_202"], "mul_9859": ["dequantize_affine_202"], "view_1592": ["dequantize_affine_203"], "view_1594": ["dequantize_affine_203"], "convert_element_type_610": ["dequantize_affine_203"], "convert_element_type_611": ["dequantize_affine_203"], "sub_4659": ["dequantize_affine_203"], "view_1593": ["dequantize_affine_203"], "mul_9864": ["dequantize_affine_203"], "view_1595": ["dequantize_affine_203"], "mul_9869": ["linear_101"], "view_1596": ["linear_101"], "permute_170": ["linear_101"], "mm_default_75": ["linear_101"], "add_tensor_75": ["linear_101"], "view_1597": ["linear_101", "dropout_51"], "add_15632": ["add_238"], "clone_137": ["layer_norm_34"], "var_mean_34": ["layer_norm_34"], "getitem_136": ["layer_norm_34"], "getitem_137": ["layer_norm_34"], "sub_4665": ["layer_norm_34"], "add_15637": ["layer_norm_34"], "rsqrt_34": ["layer_norm_34"], "mul_9880": ["layer_norm_34"], "mul_9881": ["layer_norm_34"], "add_15638": ["layer_norm_34", "choose_qparams_affine_default_102", "quantize_affine_102", "choose_qparams_affine_default_103", "quantize_affine_103", "choose_qparams_affine_default_104", "quantize_affine_104"], "amax_102": ["choose_qparams_affine_default_102"], "full_205": ["choose_qparams_affine_default_102"], "maximum_102": ["choose_qparams_affine_default_102"], "amin_102": ["choose_qparams_affine_default_102"], "full_204": ["choose_qparams_affine_default_102"], "minimum_102": ["choose_qparams_affine_default_102"], "sub_4676": ["choose_qparams_affine_default_102"], "div_204": ["choose_qparams_affine_default_102"], "clamp_min_306": ["choose_qparams_affine_default_102"], "div_205": ["choose_qparams_affine_default_102"], "round_205": ["choose_qparams_affine_default_102"], "sub_4682": ["choose_qparams_affine_default_102"], "clamp_min_307": ["choose_qparams_affine_default_102"], "clamp_max_204": ["choose_qparams_affine_default_102"], "view_1600": ["quantize_affine_102"], "reciprocal_102": ["quantize_affine_102"], "mul_9929": ["quantize_affine_102"], "mul_9932": ["quantize_affine_102"], "round_206": ["quantize_affine_102"], "convert_element_type_612": ["choose_qparams_affine_default_102"], "view_1601": ["quantize_affine_102"], "add_15725": ["quantize_affine_102"], "clamp_min_308": ["quantize_affine_102"], "clamp_max_205": ["quantize_affine_102"], "convert_element_type_613": ["quantize_affine_102", "dequantize_affine_204"], "view_1605": ["dequantize_affine_204"], "convert_element_type_614": ["dequantize_affine_204"], "convert_element_type_615": ["dequantize_affine_204"], "sub_4702": ["dequantize_affine_204"], "view_1604": ["dequantize_affine_204"], "mul_9954": ["dequantize_affine_204"], "view_1607": ["dequantize_affine_205"], "view_1609": ["dequantize_affine_205"], "convert_element_type_616": ["dequantize_affine_205"], "convert_element_type_617": ["dequantize_affine_205"], "sub_4706": ["dequantize_affine_205"], "view_1608": ["dequantize_affine_205"], "mul_9959": ["dequantize_affine_205"], "view_1610": ["dequantize_affine_205"], "amax_103": ["choose_qparams_affine_default_103"], "full_207": ["choose_qparams_affine_default_103"], "maximum_103": ["choose_qparams_affine_default_103"], "amin_103": ["choose_qparams_affine_default_103"], "full_206": ["choose_qparams_affine_default_103"], "minimum_103": ["choose_qparams_affine_default_103"], "sub_4721": ["choose_qparams_affine_default_103"], "div_206": ["choose_qparams_affine_default_103"], "clamp_min_309": ["choose_qparams_affine_default_103"], "div_207": ["choose_qparams_affine_default_103"], "round_207": ["choose_qparams_affine_default_103"], "sub_4727": ["choose_qparams_affine_default_103"], "clamp_min_310": ["choose_qparams_affine_default_103"], "clamp_max_206": ["choose_qparams_affine_default_103"], "view_1616": ["quantize_affine_103"], "reciprocal_103": ["quantize_affine_103"], "mul_10025": ["quantize_affine_103"], "mul_10028": ["quantize_affine_103"], "round_208": ["quantize_affine_103"], "convert_element_type_618": ["choose_qparams_affine_default_103"], "view_1617": ["quantize_affine_103"], "add_15877": ["quantize_affine_103"], "clamp_min_311": ["quantize_affine_103"], "clamp_max_207": ["quantize_affine_103"], "convert_element_type_619": ["quantize_affine_103", "dequantize_affine_206"], "view_1621": ["dequantize_affine_206"], "convert_element_type_620": ["dequantize_affine_206"], "convert_element_type_621": ["dequantize_affine_206"], "sub_4747": ["dequantize_affine_206"], "view_1620": ["dequantize_affine_206"], "mul_10050": ["dequantize_affine_206"], "view_1623": ["dequantize_affine_207"], "view_1625": ["dequantize_affine_207"], "convert_element_type_622": ["dequantize_affine_207"], "convert_element_type_623": ["dequantize_affine_207"], "sub_4751": ["dequantize_affine_207"], "view_1624": ["dequantize_affine_207"], "mul_10055": ["dequantize_affine_207"], "view_1626": ["dequantize_affine_207"], "amax_104": ["choose_qparams_affine_default_104"], "full_209": ["choose_qparams_affine_default_104"], "maximum_104": ["choose_qparams_affine_default_104"], "amin_104": ["choose_qparams_affine_default_104"], "full_208": ["choose_qparams_affine_default_104"], "minimum_104": ["choose_qparams_affine_default_104"], "sub_4765": ["choose_qparams_affine_default_104"], "div_208": ["choose_qparams_affine_default_104"], "clamp_min_312": ["choose_qparams_affine_default_104"], "div_209": ["choose_qparams_affine_default_104"], "round_209": ["choose_qparams_affine_default_104"], "sub_4771": ["choose_qparams_affine_default_104"], "clamp_min_313": ["choose_qparams_affine_default_104"], "clamp_max_208": ["choose_qparams_affine_default_104"], "view_1632": ["quantize_affine_104"], "reciprocal_104": ["quantize_affine_104"], "mul_10124": ["quantize_affine_104"], "mul_10127": ["quantize_affine_104"], "round_210": ["quantize_affine_104"], "convert_element_type_624": ["choose_qparams_affine_default_104"], "view_1633": ["quantize_affine_104"], "add_16025": ["quantize_affine_104"], "clamp_min_314": ["quantize_affine_104"], "clamp_max_209": ["quantize_affine_104"], "convert_element_type_625": ["quantize_affine_104", "dequantize_affine_208"], "view_1637": ["dequantize_affine_208"], "convert_element_type_626": ["dequantize_affine_208"], "convert_element_type_627": ["dequantize_affine_208"], "sub_4791": ["dequantize_affine_208"], "view_1636": ["dequantize_affine_208"], "mul_10149": ["dequantize_affine_208"], "view_1639": ["dequantize_affine_209"], "view_1641": ["dequantize_affine_209"], "convert_element_type_628": ["dequantize_affine_209"], "convert_element_type_629": ["dequantize_affine_209"], "sub_4795": ["dequantize_affine_209"], "view_1640": ["dequantize_affine_209"], "mul_10154": ["dequantize_affine_209"], "view_1642": ["dequantize_affine_209"], "mul_9964": ["linear_102"], "view_1611": ["linear_102"], "permute_171": ["linear_102"], "mm_default_74": ["linear_102"], "add_tensor_74": ["linear_102"], "view_1612": ["linear_102"], "mul_9971": ["mul_634"], "view_1613": ["view_51"], "permute_172": ["transpose_68"], "clone_138": ["contiguous_68"], "mul_10058": ["linear_103"], "view_1627": ["linear_103"], "permute_173": ["linear_103"], "mm_17": ["linear_103"], "view_1628": ["linear_103"], "view_1629": ["view_52"], "permute_174": ["transpose_69"], "clone_139": ["contiguous_69"], "mul_10159": ["linear_104"], "view_1643": ["linear_104"], "permute_175": ["linear_104"], "mm_default_73": ["linear_104"], "add_tensor_73": ["linear_104"], "view_1644": ["linear_104"], "view_1645": ["view_53"], "permute_176": ["transpose_70"], "clone_140": ["contiguous_70"], "_scaled_dot_product_efficient_attention_17": ["scaled_dot_product_attention_17"], "getitem_138": ["scaled_dot_product_attention_17"], "permute_177": ["transpose_71"], "view_1646": ["reshape_17", "choose_qparams_affine_default_105", "quantize_affine_105"], "amax_105": ["choose_qparams_affine_default_105"], "full_211": ["choose_qparams_affine_default_105"], "maximum_105": ["choose_qparams_affine_default_105"], "amin_105": ["choose_qparams_affine_default_105"], "full_210": ["choose_qparams_affine_default_105"], "minimum_105": ["choose_qparams_affine_default_105"], "sub_4813": ["choose_qparams_affine_default_105"], "div_210": ["choose_qparams_affine_default_105"], "clamp_min_315": ["choose_qparams_affine_default_105"], "div_211": ["choose_qparams_affine_default_105"], "round_211": ["choose_qparams_affine_default_105"], "sub_4819": ["choose_qparams_affine_default_105"], "clamp_min_316": ["choose_qparams_affine_default_105"], "clamp_max_210": ["choose_qparams_affine_default_105"], "view_1649": ["quantize_affine_105"], "reciprocal_105": ["quantize_affine_105"], "mul_10229": ["quantize_affine_105"], "mul_10232": ["quantize_affine_105"], "round_212": ["quantize_affine_105"], "convert_element_type_630": ["choose_qparams_affine_default_105"], "view_1650": ["quantize_affine_105"], "add_16189": ["quantize_affine_105"], "clamp_min_317": ["quantize_affine_105"], "clamp_max_211": ["quantize_affine_105"], "convert_element_type_631": ["quantize_affine_105", "dequantize_affine_210"], "view_1654": ["dequantize_affine_210"], "convert_element_type_632": ["dequantize_affine_210"], "convert_element_type_633": ["dequantize_affine_210"], "sub_4839": ["dequantize_affine_210"], "view_1653": ["dequantize_affine_210"], "mul_10254": ["dequantize_affine_210"], "view_1656": ["dequantize_affine_211"], "view_1658": ["dequantize_affine_211"], "convert_element_type_634": ["dequantize_affine_211"], "convert_element_type_635": ["dequantize_affine_211"], "sub_4843": ["dequantize_affine_211"], "view_1657": ["dequantize_affine_211"], "mul_10259": ["dequantize_affine_211"], "view_1659": ["dequantize_affine_211"], "mul_10264": ["linear_105"], "view_1660": ["linear_105"], "permute_178": ["linear_105"], "mm_default_72": ["linear_105"], "add_tensor_72": ["linear_105"], "view_1661": ["linear_105", "dropout_52"], "add_16252": ["add_247"], "clone_142": ["layer_norm_35"], "var_mean_35": ["layer_norm_35"], "getitem_142": ["layer_norm_35"], "getitem_143": ["layer_norm_35"], "sub_4849": ["layer_norm_35"], "add_16257": ["layer_norm_35"], "rsqrt_35": ["layer_norm_35"], "mul_10275": ["layer_norm_35"], "mul_10276": ["layer_norm_35"], "add_16258": ["layer_norm_35", "choose_qparams_affine_default_106", "quantize_affine_106"], "amax_106": ["choose_qparams_affine_default_106"], "full_213": ["choose_qparams_affine_default_106"], "maximum_106": ["choose_qparams_affine_default_106"], "amin_106": ["choose_qparams_affine_default_106"], "full_212": ["choose_qparams_affine_default_106"], "minimum_106": ["choose_qparams_affine_default_106"], "sub_4860": ["choose_qparams_affine_default_106"], "div_212": ["choose_qparams_affine_default_106"], "clamp_min_318": ["choose_qparams_affine_default_106"], "div_213": ["choose_qparams_affine_default_106"], "round_213": ["choose_qparams_affine_default_106"], "sub_4866": ["choose_qparams_affine_default_106"], "clamp_min_319": ["choose_qparams_affine_default_106"], "clamp_max_212": ["choose_qparams_affine_default_106"], "view_1664": ["quantize_affine_106"], "reciprocal_106": ["quantize_affine_106"], "mul_10324": ["quantize_affine_106"], "mul_10327": ["quantize_affine_106"], "round_214": ["quantize_affine_106"], "convert_element_type_636": ["choose_qparams_affine_default_106"], "view_1665": ["quantize_affine_106"], "add_16345": ["quantize_affine_106"], "clamp_min_320": ["quantize_affine_106"], "clamp_max_213": ["quantize_affine_106"], "convert_element_type_637": ["quantize_affine_106", "dequantize_affine_212"], "view_1669": ["dequantize_affine_212"], "convert_element_type_638": ["dequantize_affine_212"], "convert_element_type_639": ["dequantize_affine_212"], "sub_4886": ["dequantize_affine_212"], "view_1668": ["dequantize_affine_212"], "mul_10349": ["dequantize_affine_212"], "view_1671": ["dequantize_affine_213"], "view_1673": ["dequantize_affine_213"], "convert_element_type_640": ["dequantize_affine_213"], "convert_element_type_641": ["dequantize_affine_213"], "sub_4890": ["dequantize_affine_213"], "view_1672": ["dequantize_affine_213"], "mul_10354": ["dequantize_affine_213"], "view_1674": ["dequantize_affine_213"], "mul_10359": ["linear_106"], "view_1675": ["linear_106"], "permute_179": ["linear_106"], "mm_default_71": ["linear_106"], "add_tensor_71": ["linear_106"], "view_1676": ["linear_106"], "mul_10366": ["gelu_19"], "mul_10367": ["gelu_19"], "erf_19": ["gelu_19"], "add_16404": ["gelu_19"], "mul_10368": ["gelu_19", "dropout_53", "choose_qparams_affine_default_107", "quantize_affine_107"], "amax_107": ["choose_qparams_affine_default_107"], "full_215": ["choose_qparams_affine_default_107"], "maximum_107": ["choose_qparams_affine_default_107"], "amin_107": ["choose_qparams_affine_default_107"], "full_214": ["choose_qparams_affine_default_107"], "minimum_107": ["choose_qparams_affine_default_107"], "sub_4903": ["choose_qparams_affine_default_107"], "div_214": ["choose_qparams_affine_default_107"], "clamp_min_321": ["choose_qparams_affine_default_107"], "div_215": ["choose_qparams_affine_default_107"], "round_215": ["choose_qparams_affine_default_107"], "sub_4909": ["choose_qparams_affine_default_107"], "clamp_min_322": ["choose_qparams_affine_default_107"], "clamp_max_214": ["choose_qparams_affine_default_107"], "view_1679": ["quantize_affine_107"], "reciprocal_107": ["quantize_affine_107"], "mul_10414": ["quantize_affine_107"], "mul_10417": ["quantize_affine_107"], "round_216": ["quantize_affine_107"], "convert_element_type_642": ["choose_qparams_affine_default_107"], "view_1680": ["quantize_affine_107"], "add_16487": ["quantize_affine_107"], "clamp_min_323": ["quantize_affine_107"], "clamp_max_215": ["quantize_affine_107"], "convert_element_type_643": ["quantize_affine_107", "dequantize_affine_214"], "view_1684": ["dequantize_affine_214"], "convert_element_type_644": ["dequantize_affine_214"], "convert_element_type_645": ["dequantize_affine_214"], "sub_4929": ["dequantize_affine_214"], "view_1683": ["dequantize_affine_214"], "mul_10439": ["dequantize_affine_214"], "view_1686": ["dequantize_affine_215"], "view_1688": ["dequantize_affine_215"], "convert_element_type_646": ["dequantize_affine_215"], "convert_element_type_647": ["dequantize_affine_215"], "sub_4933": ["dequantize_affine_215"], "view_1687": ["dequantize_affine_215"], "mul_10444": ["dequantize_affine_215"], "view_1689": ["dequantize_affine_215"], "mul_10449": ["linear_107"], "view_1690": ["linear_107"], "permute_180": ["linear_107"], "mm_default_70": ["linear_107"], "add_tensor_70": ["linear_107"], "view_1691": ["linear_107", "dropout_54"], "add_16550": ["add_252"], "clone_145": ["layer_norm_36"], "var_mean_36": ["layer_norm_36"], "getitem_144": ["layer_norm_36"], "getitem_145": ["layer_norm_36"], "sub_4939": ["layer_norm_36"], "add_16555": ["layer_norm_36"], "rsqrt_36": ["layer_norm_36"], "mul_10460": ["layer_norm_36"], "mul_10461": ["layer_norm_36"], "add_16556": ["layer_norm_36", "choose_qparams_affine_default_108", "quantize_affine_108", "choose_qparams_affine_default_109", "quantize_affine_109", "choose_qparams_affine_default_110", "quantize_affine_110"], "amax_108": ["choose_qparams_affine_default_108"], "full_217": ["choose_qparams_affine_default_108"], "maximum_108": ["choose_qparams_affine_default_108"], "amin_108": ["choose_qparams_affine_default_108"], "full_216": ["choose_qparams_affine_default_108"], "minimum_108": ["choose_qparams_affine_default_108"], "sub_4950": ["choose_qparams_affine_default_108"], "div_216": ["choose_qparams_affine_default_108"], "clamp_min_324": ["choose_qparams_affine_default_108"], "div_217": ["choose_qparams_affine_default_108"], "round_217": ["choose_qparams_affine_default_108"], "sub_4956": ["choose_qparams_affine_default_108"], "clamp_min_325": ["choose_qparams_affine_default_108"], "clamp_max_216": ["choose_qparams_affine_default_108"], "view_1694": ["quantize_affine_108"], "reciprocal_108": ["quantize_affine_108"], "mul_10509": ["quantize_affine_108"], "mul_10512": ["quantize_affine_108"], "round_218": ["quantize_affine_108"], "convert_element_type_648": ["choose_qparams_affine_default_108"], "view_1695": ["quantize_affine_108"], "add_16643": ["quantize_affine_108"], "clamp_min_326": ["quantize_affine_108"], "clamp_max_217": ["quantize_affine_108"], "convert_element_type_649": ["quantize_affine_108", "dequantize_affine_216"], "view_1699": ["dequantize_affine_216"], "convert_element_type_650": ["dequantize_affine_216"], "convert_element_type_651": ["dequantize_affine_216"], "sub_4976": ["dequantize_affine_216"], "view_1698": ["dequantize_affine_216"], "mul_10534": ["dequantize_affine_216"], "view_1701": ["dequantize_affine_217"], "view_1703": ["dequantize_affine_217"], "convert_element_type_652": ["dequantize_affine_217"], "convert_element_type_653": ["dequantize_affine_217"], "sub_4980": ["dequantize_affine_217"], "view_1702": ["dequantize_affine_217"], "mul_10539": ["dequantize_affine_217"], "view_1704": ["dequantize_affine_217"], "amax_109": ["choose_qparams_affine_default_109"], "full_219": ["choose_qparams_affine_default_109"], "maximum_109": ["choose_qparams_affine_default_109"], "amin_109": ["choose_qparams_affine_default_109"], "full_218": ["choose_qparams_affine_default_109"], "minimum_109": ["choose_qparams_affine_default_109"], "sub_4995": ["choose_qparams_affine_default_109"], "div_218": ["choose_qparams_affine_default_109"], "clamp_min_327": ["choose_qparams_affine_default_109"], "div_219": ["choose_qparams_affine_default_109"], "round_219": ["choose_qparams_affine_default_109"], "sub_5001": ["choose_qparams_affine_default_109"], "clamp_min_328": ["choose_qparams_affine_default_109"], "clamp_max_218": ["choose_qparams_affine_default_109"], "view_1710": ["quantize_affine_109"], "reciprocal_109": ["quantize_affine_109"], "mul_10605": ["quantize_affine_109"], "mul_10608": ["quantize_affine_109"], "round_220": ["quantize_affine_109"], "convert_element_type_654": ["choose_qparams_affine_default_109"], "view_1711": ["quantize_affine_109"], "add_16795": ["quantize_affine_109"], "clamp_min_329": ["quantize_affine_109"], "clamp_max_219": ["quantize_affine_109"], "convert_element_type_655": ["quantize_affine_109", "dequantize_affine_218"], "view_1715": ["dequantize_affine_218"], "convert_element_type_656": ["dequantize_affine_218"], "convert_element_type_657": ["dequantize_affine_218"], "sub_5021": ["dequantize_affine_218"], "view_1714": ["dequantize_affine_218"], "mul_10630": ["dequantize_affine_218"], "view_1717": ["dequantize_affine_219"], "view_1719": ["dequantize_affine_219"], "convert_element_type_658": ["dequantize_affine_219"], "convert_element_type_659": ["dequantize_affine_219"], "sub_5025": ["dequantize_affine_219"], "view_1718": ["dequantize_affine_219"], "mul_10635": ["dequantize_affine_219"], "view_1720": ["dequantize_affine_219"], "amax_110": ["choose_qparams_affine_default_110"], "full_221": ["choose_qparams_affine_default_110"], "maximum_110": ["choose_qparams_affine_default_110"], "amin_110": ["choose_qparams_affine_default_110"], "full_220": ["choose_qparams_affine_default_110"], "minimum_110": ["choose_qparams_affine_default_110"], "sub_5039": ["choose_qparams_affine_default_110"], "div_220": ["choose_qparams_affine_default_110"], "clamp_min_330": ["choose_qparams_affine_default_110"], "div_221": ["choose_qparams_affine_default_110"], "round_221": ["choose_qparams_affine_default_110"], "sub_5045": ["choose_qparams_affine_default_110"], "clamp_min_331": ["choose_qparams_affine_default_110"], "clamp_max_220": ["choose_qparams_affine_default_110"], "view_1726": ["quantize_affine_110"], "reciprocal_110": ["quantize_affine_110"], "mul_10704": ["quantize_affine_110"], "mul_10707": ["quantize_affine_110"], "round_222": ["quantize_affine_110"], "convert_element_type_660": ["choose_qparams_affine_default_110"], "view_1727": ["quantize_affine_110"], "add_16943": ["quantize_affine_110"], "clamp_min_332": ["quantize_affine_110"], "clamp_max_221": ["quantize_affine_110"], "convert_element_type_661": ["quantize_affine_110", "dequantize_affine_220"], "view_1731": ["dequantize_affine_220"], "convert_element_type_662": ["dequantize_affine_220"], "convert_element_type_663": ["dequantize_affine_220"], "sub_5065": ["dequantize_affine_220"], "view_1730": ["dequantize_affine_220"], "mul_10729": ["dequantize_affine_220"], "view_1733": ["dequantize_affine_221"], "view_1735": ["dequantize_affine_221"], "convert_element_type_664": ["dequantize_affine_221"], "convert_element_type_665": ["dequantize_affine_221"], "sub_5069": ["dequantize_affine_221"], "view_1734": ["dequantize_affine_221"], "mul_10734": ["dequantize_affine_221"], "view_1736": ["dequantize_affine_221"], "mul_10544": ["linear_108"], "view_1705": ["linear_108"], "permute_181": ["linear_108"], "mm_default_69": ["linear_108"], "add_tensor_69": ["linear_108"], "view_1706": ["linear_108"], "mul_10551": ["mul_671"], "view_1707": ["view_54"], "permute_182": ["transpose_72"], "clone_146": ["contiguous_72"], "mul_10638": ["linear_109"], "view_1721": ["linear_109"], "permute_183": ["linear_109"], "mm_18": ["linear_109"], "view_1722": ["linear_109"], "view_1723": ["view_55"], "permute_184": ["transpose_73"], "clone_147": ["contiguous_73"], "mul_10739": ["linear_110"], "view_1737": ["linear_110"], "permute_185": ["linear_110"], "mm_default_68": ["linear_110"], "add_tensor_68": ["linear_110"], "view_1738": ["linear_110"], "view_1739": ["view_56"], "permute_186": ["transpose_74"], "clone_148": ["contiguous_74"], "_scaled_dot_product_efficient_attention_18": ["scaled_dot_product_attention_18"], "getitem_146": ["scaled_dot_product_attention_18"], "permute_187": ["transpose_75"], "view_1740": ["reshape_18", "choose_qparams_affine_default_111", "quantize_affine_111"], "amax_111": ["choose_qparams_affine_default_111"], "full_223": ["choose_qparams_affine_default_111"], "maximum_111": ["choose_qparams_affine_default_111"], "amin_111": ["choose_qparams_affine_default_111"], "full_222": ["choose_qparams_affine_default_111"], "minimum_111": ["choose_qparams_affine_default_111"], "sub_5087": ["choose_qparams_affine_default_111"], "div_222": ["choose_qparams_affine_default_111"], "clamp_min_333": ["choose_qparams_affine_default_111"], "div_223": ["choose_qparams_affine_default_111"], "round_223": ["choose_qparams_affine_default_111"], "sub_5093": ["choose_qparams_affine_default_111"], "clamp_min_334": ["choose_qparams_affine_default_111"], "clamp_max_222": ["choose_qparams_affine_default_111"], "view_1743": ["quantize_affine_111"], "reciprocal_111": ["quantize_affine_111"], "mul_10809": ["quantize_affine_111"], "mul_10812": ["quantize_affine_111"], "round_224": ["quantize_affine_111"], "convert_element_type_666": ["choose_qparams_affine_default_111"], "view_1744": ["quantize_affine_111"], "add_17107": ["quantize_affine_111"], "clamp_min_335": ["quantize_affine_111"], "clamp_max_223": ["quantize_affine_111"], "convert_element_type_667": ["quantize_affine_111", "dequantize_affine_222"], "view_1748": ["dequantize_affine_222"], "convert_element_type_668": ["dequantize_affine_222"], "convert_element_type_669": ["dequantize_affine_222"], "sub_5113": ["dequantize_affine_222"], "view_1747": ["dequantize_affine_222"], "mul_10834": ["dequantize_affine_222"], "view_1750": ["dequantize_affine_223"], "view_1752": ["dequantize_affine_223"], "convert_element_type_670": ["dequantize_affine_223"], "convert_element_type_671": ["dequantize_affine_223"], "sub_5117": ["dequantize_affine_223"], "view_1751": ["dequantize_affine_223"], "mul_10839": ["dequantize_affine_223"], "view_1753": ["dequantize_affine_223"], "mul_10844": ["linear_111"], "view_1754": ["linear_111"], "permute_188": ["linear_111"], "mm_default_67": ["linear_111"], "add_tensor_67": ["linear_111"], "view_1755": ["linear_111", "dropout_55"], "add_17170": ["add_261"], "clone_150": ["layer_norm_37"], "var_mean_37": ["layer_norm_37"], "getitem_150": ["layer_norm_37"], "getitem_151": ["layer_norm_37"], "sub_5123": ["layer_norm_37"], "add_17175": ["layer_norm_37"], "rsqrt_37": ["layer_norm_37"], "mul_10855": ["layer_norm_37"], "mul_10856": ["layer_norm_37"], "add_17176": ["layer_norm_37", "choose_qparams_affine_default_112", "quantize_affine_112"], "amax_112": ["choose_qparams_affine_default_112"], "full_225": ["choose_qparams_affine_default_112"], "maximum_112": ["choose_qparams_affine_default_112"], "amin_112": ["choose_qparams_affine_default_112"], "full_224": ["choose_qparams_affine_default_112"], "minimum_112": ["choose_qparams_affine_default_112"], "sub_5134": ["choose_qparams_affine_default_112"], "div_224": ["choose_qparams_affine_default_112"], "clamp_min_336": ["choose_qparams_affine_default_112"], "div_225": ["choose_qparams_affine_default_112"], "round_225": ["choose_qparams_affine_default_112"], "sub_5140": ["choose_qparams_affine_default_112"], "clamp_min_337": ["choose_qparams_affine_default_112"], "clamp_max_224": ["choose_qparams_affine_default_112"], "view_1758": ["quantize_affine_112"], "reciprocal_112": ["quantize_affine_112"], "mul_10904": ["quantize_affine_112"], "mul_10907": ["quantize_affine_112"], "round_226": ["quantize_affine_112"], "convert_element_type_672": ["choose_qparams_affine_default_112"], "view_1759": ["quantize_affine_112"], "add_17263": ["quantize_affine_112"], "clamp_min_338": ["quantize_affine_112"], "clamp_max_225": ["quantize_affine_112"], "convert_element_type_673": ["quantize_affine_112", "dequantize_affine_224"], "view_1763": ["dequantize_affine_224"], "convert_element_type_674": ["dequantize_affine_224"], "convert_element_type_675": ["dequantize_affine_224"], "sub_5160": ["dequantize_affine_224"], "view_1762": ["dequantize_affine_224"], "mul_10929": ["dequantize_affine_224"], "view_1765": ["dequantize_affine_225"], "view_1767": ["dequantize_affine_225"], "convert_element_type_676": ["dequantize_affine_225"], "convert_element_type_677": ["dequantize_affine_225"], "sub_5164": ["dequantize_affine_225"], "view_1766": ["dequantize_affine_225"], "mul_10934": ["dequantize_affine_225"], "view_1768": ["dequantize_affine_225"], "mul_10939": ["linear_112"], "view_1769": ["linear_112"], "permute_189": ["linear_112"], "mm_default_66": ["linear_112"], "add_tensor_66": ["linear_112"], "view_1770": ["linear_112"], "mul_10946": ["gelu_20"], "mul_10947": ["gelu_20"], "erf_20": ["gelu_20"], "add_17322": ["gelu_20"], "mul_10948": ["gelu_20", "dropout_56", "choose_qparams_affine_default_113", "quantize_affine_113"], "amax_113": ["choose_qparams_affine_default_113"], "full_227": ["choose_qparams_affine_default_113"], "maximum_113": ["choose_qparams_affine_default_113"], "amin_113": ["choose_qparams_affine_default_113"], "full_226": ["choose_qparams_affine_default_113"], "minimum_113": ["choose_qparams_affine_default_113"], "sub_5177": ["choose_qparams_affine_default_113"], "div_226": ["choose_qparams_affine_default_113"], "clamp_min_339": ["choose_qparams_affine_default_113"], "div_227": ["choose_qparams_affine_default_113"], "round_227": ["choose_qparams_affine_default_113"], "sub_5183": ["choose_qparams_affine_default_113"], "clamp_min_340": ["choose_qparams_affine_default_113"], "clamp_max_226": ["choose_qparams_affine_default_113"], "view_1773": ["quantize_affine_113"], "reciprocal_113": ["quantize_affine_113"], "mul_10994": ["quantize_affine_113"], "mul_10997": ["quantize_affine_113"], "round_228": ["quantize_affine_113"], "convert_element_type_678": ["choose_qparams_affine_default_113"], "view_1774": ["quantize_affine_113"], "add_17405": ["quantize_affine_113"], "clamp_min_341": ["quantize_affine_113"], "clamp_max_227": ["quantize_affine_113"], "convert_element_type_679": ["quantize_affine_113", "dequantize_affine_226"], "view_1778": ["dequantize_affine_226"], "convert_element_type_680": ["dequantize_affine_226"], "convert_element_type_681": ["dequantize_affine_226"], "sub_5203": ["dequantize_affine_226"], "view_1777": ["dequantize_affine_226"], "mul_11019": ["dequantize_affine_226"], "view_1780": ["dequantize_affine_227"], "view_1782": ["dequantize_affine_227"], "convert_element_type_682": ["dequantize_affine_227"], "convert_element_type_683": ["dequantize_affine_227"], "sub_5207": ["dequantize_affine_227"], "view_1781": ["dequantize_affine_227"], "mul_11024": ["dequantize_affine_227"], "view_1783": ["dequantize_affine_227"], "mul_11029": ["linear_113"], "view_1784": ["linear_113"], "permute_190": ["linear_113"], "mm_default_65": ["linear_113"], "add_tensor_65": ["linear_113"], "view_1785": ["linear_113", "dropout_57"], "add_17468": ["add_266"], "clone_153": ["layer_norm_38"], "var_mean_38": ["layer_norm_38"], "getitem_152": ["layer_norm_38"], "getitem_153": ["layer_norm_38"], "sub_5213": ["layer_norm_38"], "add_17473": ["layer_norm_38"], "rsqrt_38": ["layer_norm_38"], "mul_11040": ["layer_norm_38"], "mul_11041": ["layer_norm_38"], "add_17474": ["layer_norm_38", "choose_qparams_affine_default_114", "quantize_affine_114", "choose_qparams_affine_default_115", "quantize_affine_115", "choose_qparams_affine_default_116", "quantize_affine_116"], "amax_114": ["choose_qparams_affine_default_114"], "full_229": ["choose_qparams_affine_default_114"], "maximum_114": ["choose_qparams_affine_default_114"], "amin_114": ["choose_qparams_affine_default_114"], "full_228": ["choose_qparams_affine_default_114"], "minimum_114": ["choose_qparams_affine_default_114"], "sub_5224": ["choose_qparams_affine_default_114"], "div_228": ["choose_qparams_affine_default_114"], "clamp_min_342": ["choose_qparams_affine_default_114"], "div_229": ["choose_qparams_affine_default_114"], "round_229": ["choose_qparams_affine_default_114"], "sub_5230": ["choose_qparams_affine_default_114"], "clamp_min_343": ["choose_qparams_affine_default_114"], "clamp_max_228": ["choose_qparams_affine_default_114"], "view_1788": ["quantize_affine_114"], "reciprocal_114": ["quantize_affine_114"], "mul_11089": ["quantize_affine_114"], "mul_11092": ["quantize_affine_114"], "round_230": ["quantize_affine_114"], "convert_element_type_684": ["choose_qparams_affine_default_114"], "view_1789": ["quantize_affine_114"], "add_17561": ["quantize_affine_114"], "clamp_min_344": ["quantize_affine_114"], "clamp_max_229": ["quantize_affine_114"], "convert_element_type_685": ["quantize_affine_114", "dequantize_affine_228"], "view_1793": ["dequantize_affine_228"], "convert_element_type_686": ["dequantize_affine_228"], "convert_element_type_687": ["dequantize_affine_228"], "sub_5250": ["dequantize_affine_228"], "view_1792": ["dequantize_affine_228"], "mul_11114": ["dequantize_affine_228"], "view_1795": ["dequantize_affine_229"], "view_1797": ["dequantize_affine_229"], "convert_element_type_688": ["dequantize_affine_229"], "convert_element_type_689": ["dequantize_affine_229"], "sub_5254": ["dequantize_affine_229"], "view_1796": ["dequantize_affine_229"], "mul_11119": ["dequantize_affine_229"], "view_1798": ["dequantize_affine_229"], "amax_115": ["choose_qparams_affine_default_115"], "full_231": ["choose_qparams_affine_default_115"], "maximum_115": ["choose_qparams_affine_default_115"], "amin_115": ["choose_qparams_affine_default_115"], "full_230": ["choose_qparams_affine_default_115"], "minimum_115": ["choose_qparams_affine_default_115"], "sub_5269": ["choose_qparams_affine_default_115"], "div_230": ["choose_qparams_affine_default_115"], "clamp_min_345": ["choose_qparams_affine_default_115"], "div_231": ["choose_qparams_affine_default_115"], "round_231": ["choose_qparams_affine_default_115"], "sub_5275": ["choose_qparams_affine_default_115"], "clamp_min_346": ["choose_qparams_affine_default_115"], "clamp_max_230": ["choose_qparams_affine_default_115"], "view_1804": ["quantize_affine_115"], "reciprocal_115": ["quantize_affine_115"], "mul_11185": ["quantize_affine_115"], "mul_11188": ["quantize_affine_115"], "round_232": ["quantize_affine_115"], "convert_element_type_690": ["choose_qparams_affine_default_115"], "view_1805": ["quantize_affine_115"], "add_17713": ["quantize_affine_115"], "clamp_min_347": ["quantize_affine_115"], "clamp_max_231": ["quantize_affine_115"], "convert_element_type_691": ["quantize_affine_115", "dequantize_affine_230"], "view_1809": ["dequantize_affine_230"], "convert_element_type_692": ["dequantize_affine_230"], "convert_element_type_693": ["dequantize_affine_230"], "sub_5295": ["dequantize_affine_230"], "view_1808": ["dequantize_affine_230"], "mul_11210": ["dequantize_affine_230"], "view_1811": ["dequantize_affine_231"], "view_1813": ["dequantize_affine_231"], "convert_element_type_694": ["dequantize_affine_231"], "convert_element_type_695": ["dequantize_affine_231"], "sub_5299": ["dequantize_affine_231"], "view_1812": ["dequantize_affine_231"], "mul_11215": ["dequantize_affine_231"], "view_1814": ["dequantize_affine_231"], "amax_116": ["choose_qparams_affine_default_116"], "full_233": ["choose_qparams_affine_default_116"], "maximum_116": ["choose_qparams_affine_default_116"], "amin_116": ["choose_qparams_affine_default_116"], "full_232": ["choose_qparams_affine_default_116"], "minimum_116": ["choose_qparams_affine_default_116"], "sub_5313": ["choose_qparams_affine_default_116"], "div_232": ["choose_qparams_affine_default_116"], "clamp_min_348": ["choose_qparams_affine_default_116"], "div_233": ["choose_qparams_affine_default_116"], "round_233": ["choose_qparams_affine_default_116"], "sub_5319": ["choose_qparams_affine_default_116"], "clamp_min_349": ["choose_qparams_affine_default_116"], "clamp_max_232": ["choose_qparams_affine_default_116"], "view_1820": ["quantize_affine_116"], "reciprocal_116": ["quantize_affine_116"], "mul_11284": ["quantize_affine_116"], "mul_11287": ["quantize_affine_116"], "round_234": ["quantize_affine_116"], "convert_element_type_696": ["choose_qparams_affine_default_116"], "view_1821": ["quantize_affine_116"], "add_17861": ["quantize_affine_116"], "clamp_min_350": ["quantize_affine_116"], "clamp_max_233": ["quantize_affine_116"], "convert_element_type_697": ["quantize_affine_116", "dequantize_affine_232"], "view_1825": ["dequantize_affine_232"], "convert_element_type_698": ["dequantize_affine_232"], "convert_element_type_699": ["dequantize_affine_232"], "sub_5339": ["dequantize_affine_232"], "view_1824": ["dequantize_affine_232"], "mul_11309": ["dequantize_affine_232"], "view_1827": ["dequantize_affine_233"], "view_1829": ["dequantize_affine_233"], "convert_element_type_700": ["dequantize_affine_233"], "convert_element_type_701": ["dequantize_affine_233"], "sub_5343": ["dequantize_affine_233"], "view_1828": ["dequantize_affine_233"], "mul_11314": ["dequantize_affine_233"], "view_1830": ["dequantize_affine_233"], "mul_11124": ["linear_114"], "view_1799": ["linear_114"], "permute_191": ["linear_114"], "mm_default_64": ["linear_114"], "add_tensor_64": ["linear_114"], "view_1800": ["linear_114"], "mul_11131": ["mul_708"], "view_1801": ["view_57"], "permute_192": ["transpose_76"], "clone_154": ["contiguous_76"], "mul_11218": ["linear_115"], "view_1815": ["linear_115"], "permute_193": ["linear_115"], "mm_19": ["linear_115"], "view_1816": ["linear_115"], "view_1817": ["view_58"], "permute_194": ["transpose_77"], "clone_155": ["contiguous_77"], "mul_11319": ["linear_116"], "view_1831": ["linear_116"], "permute_195": ["linear_116"], "mm_default_63": ["linear_116"], "add_tensor_63": ["linear_116"], "view_1832": ["linear_116"], "view_1833": ["view_59"], "permute_196": ["transpose_78"], "clone_156": ["contiguous_78"], "_scaled_dot_product_efficient_attention_19": ["scaled_dot_product_attention_19"], "getitem_154": ["scaled_dot_product_attention_19"], "permute_197": ["transpose_79"], "view_1834": ["reshape_19", "choose_qparams_affine_default_117", "quantize_affine_117"], "amax_117": ["choose_qparams_affine_default_117"], "full_235": ["choose_qparams_affine_default_117"], "maximum_117": ["choose_qparams_affine_default_117"], "amin_117": ["choose_qparams_affine_default_117"], "full_234": ["choose_qparams_affine_default_117"], "minimum_117": ["choose_qparams_affine_default_117"], "sub_5361": ["choose_qparams_affine_default_117"], "div_234": ["choose_qparams_affine_default_117"], "clamp_min_351": ["choose_qparams_affine_default_117"], "div_235": ["choose_qparams_affine_default_117"], "round_235": ["choose_qparams_affine_default_117"], "sub_5367": ["choose_qparams_affine_default_117"], "clamp_min_352": ["choose_qparams_affine_default_117"], "clamp_max_234": ["choose_qparams_affine_default_117"], "view_1837": ["quantize_affine_117"], "reciprocal_117": ["quantize_affine_117"], "mul_11389": ["quantize_affine_117"], "mul_11392": ["quantize_affine_117"], "round_236": ["quantize_affine_117"], "convert_element_type_702": ["choose_qparams_affine_default_117"], "view_1838": ["quantize_affine_117"], "add_18025": ["quantize_affine_117"], "clamp_min_353": ["quantize_affine_117"], "clamp_max_235": ["quantize_affine_117"], "convert_element_type_703": ["quantize_affine_117", "dequantize_affine_234"], "view_1842": ["dequantize_affine_234"], "convert_element_type_704": ["dequantize_affine_234"], "convert_element_type_705": ["dequantize_affine_234"], "sub_5387": ["dequantize_affine_234"], "view_1841": ["dequantize_affine_234"], "mul_11414": ["dequantize_affine_234"], "view_1844": ["dequantize_affine_235"], "view_1846": ["dequantize_affine_235"], "convert_element_type_706": ["dequantize_affine_235"], "convert_element_type_707": ["dequantize_affine_235"], "sub_5391": ["dequantize_affine_235"], "view_1845": ["dequantize_affine_235"], "mul_11419": ["dequantize_affine_235"], "view_1847": ["dequantize_affine_235"], "mul_11424": ["linear_117"], "view_1848": ["linear_117"], "permute_198": ["linear_117"], "mm_default_62": ["linear_117"], "add_tensor_62": ["linear_117"], "view_1849": ["linear_117", "dropout_58"], "add_18088": ["add_275"], "clone_158": ["layer_norm_39"], "var_mean_39": ["layer_norm_39"], "getitem_158": ["layer_norm_39"], "getitem_159": ["layer_norm_39"], "sub_5397": ["layer_norm_39"], "add_18093": ["layer_norm_39"], "rsqrt_39": ["layer_norm_39"], "mul_11435": ["layer_norm_39"], "mul_11436": ["layer_norm_39"], "add_18094": ["layer_norm_39", "choose_qparams_affine_default_118", "quantize_affine_118"], "amax_118": ["choose_qparams_affine_default_118"], "full_237": ["choose_qparams_affine_default_118"], "maximum_118": ["choose_qparams_affine_default_118"], "amin_118": ["choose_qparams_affine_default_118"], "full_236": ["choose_qparams_affine_default_118"], "minimum_118": ["choose_qparams_affine_default_118"], "sub_5408": ["choose_qparams_affine_default_118"], "div_236": ["choose_qparams_affine_default_118"], "clamp_min_354": ["choose_qparams_affine_default_118"], "div_237": ["choose_qparams_affine_default_118"], "round_237": ["choose_qparams_affine_default_118"], "sub_5414": ["choose_qparams_affine_default_118"], "clamp_min_355": ["choose_qparams_affine_default_118"], "clamp_max_236": ["choose_qparams_affine_default_118"], "view_1852": ["quantize_affine_118"], "reciprocal_118": ["quantize_affine_118"], "mul_11484": ["quantize_affine_118"], "mul_11487": ["quantize_affine_118"], "round_238": ["quantize_affine_118"], "convert_element_type_708": ["choose_qparams_affine_default_118"], "view_1853": ["quantize_affine_118"], "add_18181": ["quantize_affine_118"], "clamp_min_356": ["quantize_affine_118"], "clamp_max_237": ["quantize_affine_118"], "convert_element_type_709": ["quantize_affine_118", "dequantize_affine_236"], "view_1857": ["dequantize_affine_236"], "convert_element_type_710": ["dequantize_affine_236"], "convert_element_type_711": ["dequantize_affine_236"], "sub_5434": ["dequantize_affine_236"], "view_1856": ["dequantize_affine_236"], "mul_11509": ["dequantize_affine_236"], "view_1859": ["dequantize_affine_237"], "view_1861": ["dequantize_affine_237"], "convert_element_type_712": ["dequantize_affine_237"], "convert_element_type_713": ["dequantize_affine_237"], "sub_5438": ["dequantize_affine_237"], "view_1860": ["dequantize_affine_237"], "mul_11514": ["dequantize_affine_237"], "view_1862": ["dequantize_affine_237"], "mul_11519": ["linear_118"], "view_1863": ["linear_118"], "permute_199": ["linear_118"], "mm_default_61": ["linear_118"], "add_tensor_61": ["linear_118"], "view_1864": ["linear_118"], "mul_11526": ["gelu_21"], "mul_11527": ["gelu_21"], "erf_21": ["gelu_21"], "add_18240": ["gelu_21"], "mul_11528": ["gelu_21", "dropout_59", "choose_qparams_affine_default_119", "quantize_affine_119"], "amax_119": ["choose_qparams_affine_default_119"], "full_239": ["choose_qparams_affine_default_119"], "maximum_119": ["choose_qparams_affine_default_119"], "amin_119": ["choose_qparams_affine_default_119"], "full_238": ["choose_qparams_affine_default_119"], "minimum_119": ["choose_qparams_affine_default_119"], "sub_5451": ["choose_qparams_affine_default_119"], "div_238": ["choose_qparams_affine_default_119"], "clamp_min_357": ["choose_qparams_affine_default_119"], "div_239": ["choose_qparams_affine_default_119"], "round_239": ["choose_qparams_affine_default_119"], "sub_5457": ["choose_qparams_affine_default_119"], "clamp_min_358": ["choose_qparams_affine_default_119"], "clamp_max_238": ["choose_qparams_affine_default_119"], "view_1867": ["quantize_affine_119"], "reciprocal_119": ["quantize_affine_119"], "mul_11574": ["quantize_affine_119"], "mul_11577": ["quantize_affine_119"], "round_240": ["quantize_affine_119"], "convert_element_type_714": ["choose_qparams_affine_default_119"], "view_1868": ["quantize_affine_119"], "add_18323": ["quantize_affine_119"], "clamp_min_359": ["quantize_affine_119"], "clamp_max_239": ["quantize_affine_119"], "convert_element_type_715": ["quantize_affine_119", "dequantize_affine_238"], "view_1872": ["dequantize_affine_238"], "convert_element_type_716": ["dequantize_affine_238"], "convert_element_type_717": ["dequantize_affine_238"], "sub_5477": ["dequantize_affine_238"], "view_1871": ["dequantize_affine_238"], "mul_11599": ["dequantize_affine_238"], "view_1874": ["dequantize_affine_239"], "view_1876": ["dequantize_affine_239"], "convert_element_type_718": ["dequantize_affine_239"], "convert_element_type_719": ["dequantize_affine_239"], "sub_5481": ["dequantize_affine_239"], "view_1875": ["dequantize_affine_239"], "mul_11604": ["dequantize_affine_239"], "view_1877": ["dequantize_affine_239"], "mul_11609": ["linear_119"], "view_1878": ["linear_119"], "permute_200": ["linear_119"], "mm_default_60": ["linear_119"], "add_tensor_60": ["linear_119"], "view_1879": ["linear_119", "dropout_60"], "add_18386": ["add_280"], "clone_161": ["layer_norm_40"], "var_mean_40": ["layer_norm_40"], "getitem_160": ["layer_norm_40"], "getitem_161": ["layer_norm_40"], "sub_5487": ["layer_norm_40"], "add_18391": ["layer_norm_40"], "rsqrt_40": ["layer_norm_40"], "mul_11620": ["layer_norm_40"], "mul_11621": ["layer_norm_40"], "add_18392": ["layer_norm_40", "choose_qparams_affine_default_120", "quantize_affine_120", "choose_qparams_affine_default_121", "quantize_affine_121", "choose_qparams_affine_default_122", "quantize_affine_122"], "amax_120": ["choose_qparams_affine_default_120"], "full_241": ["choose_qparams_affine_default_120"], "maximum_120": ["choose_qparams_affine_default_120"], "amin_120": ["choose_qparams_affine_default_120"], "full_240": ["choose_qparams_affine_default_120"], "minimum_120": ["choose_qparams_affine_default_120"], "sub_5498": ["choose_qparams_affine_default_120"], "div_240": ["choose_qparams_affine_default_120"], "clamp_min_360": ["choose_qparams_affine_default_120"], "div_241": ["choose_qparams_affine_default_120"], "round_241": ["choose_qparams_affine_default_120"], "sub_5504": ["choose_qparams_affine_default_120"], "clamp_min_361": ["choose_qparams_affine_default_120"], "clamp_max_240": ["choose_qparams_affine_default_120"], "view_1882": ["quantize_affine_120"], "reciprocal_120": ["quantize_affine_120"], "mul_11669": ["quantize_affine_120"], "mul_11672": ["quantize_affine_120"], "round_242": ["quantize_affine_120"], "convert_element_type_720": ["choose_qparams_affine_default_120"], "view_1883": ["quantize_affine_120"], "add_18479": ["quantize_affine_120"], "clamp_min_362": ["quantize_affine_120"], "clamp_max_241": ["quantize_affine_120"], "convert_element_type_721": ["quantize_affine_120", "dequantize_affine_240"], "view_1887": ["dequantize_affine_240"], "convert_element_type_722": ["dequantize_affine_240"], "convert_element_type_723": ["dequantize_affine_240"], "sub_5524": ["dequantize_affine_240"], "view_1886": ["dequantize_affine_240"], "mul_11694": ["dequantize_affine_240"], "view_1889": ["dequantize_affine_241"], "view_1891": ["dequantize_affine_241"], "convert_element_type_724": ["dequantize_affine_241"], "convert_element_type_725": ["dequantize_affine_241"], "sub_5528": ["dequantize_affine_241"], "view_1890": ["dequantize_affine_241"], "mul_11699": ["dequantize_affine_241"], "view_1892": ["dequantize_affine_241"], "amax_121": ["choose_qparams_affine_default_121"], "full_243": ["choose_qparams_affine_default_121"], "maximum_121": ["choose_qparams_affine_default_121"], "amin_121": ["choose_qparams_affine_default_121"], "full_242": ["choose_qparams_affine_default_121"], "minimum_121": ["choose_qparams_affine_default_121"], "sub_5543": ["choose_qparams_affine_default_121"], "div_242": ["choose_qparams_affine_default_121"], "clamp_min_363": ["choose_qparams_affine_default_121"], "div_243": ["choose_qparams_affine_default_121"], "round_243": ["choose_qparams_affine_default_121"], "sub_5549": ["choose_qparams_affine_default_121"], "clamp_min_364": ["choose_qparams_affine_default_121"], "clamp_max_242": ["choose_qparams_affine_default_121"], "view_1898": ["quantize_affine_121"], "reciprocal_121": ["quantize_affine_121"], "mul_11765": ["quantize_affine_121"], "mul_11768": ["quantize_affine_121"], "round_244": ["quantize_affine_121"], "convert_element_type_726": ["choose_qparams_affine_default_121"], "view_1899": ["quantize_affine_121"], "add_18631": ["quantize_affine_121"], "clamp_min_365": ["quantize_affine_121"], "clamp_max_243": ["quantize_affine_121"], "convert_element_type_727": ["quantize_affine_121", "dequantize_affine_242"], "view_1903": ["dequantize_affine_242"], "convert_element_type_728": ["dequantize_affine_242"], "convert_element_type_729": ["dequantize_affine_242"], "sub_5569": ["dequantize_affine_242"], "view_1902": ["dequantize_affine_242"], "mul_11790": ["dequantize_affine_242"], "view_1905": ["dequantize_affine_243"], "view_1907": ["dequantize_affine_243"], "convert_element_type_730": ["dequantize_affine_243"], "convert_element_type_731": ["dequantize_affine_243"], "sub_5573": ["dequantize_affine_243"], "view_1906": ["dequantize_affine_243"], "mul_11795": ["dequantize_affine_243"], "view_1908": ["dequantize_affine_243"], "amax_122": ["choose_qparams_affine_default_122"], "full_245": ["choose_qparams_affine_default_122"], "maximum_122": ["choose_qparams_affine_default_122"], "amin_122": ["choose_qparams_affine_default_122"], "full_244": ["choose_qparams_affine_default_122"], "minimum_122": ["choose_qparams_affine_default_122"], "sub_5587": ["choose_qparams_affine_default_122"], "div_244": ["choose_qparams_affine_default_122"], "clamp_min_366": ["choose_qparams_affine_default_122"], "div_245": ["choose_qparams_affine_default_122"], "round_245": ["choose_qparams_affine_default_122"], "sub_5593": ["choose_qparams_affine_default_122"], "clamp_min_367": ["choose_qparams_affine_default_122"], "clamp_max_244": ["choose_qparams_affine_default_122"], "view_1914": ["quantize_affine_122"], "reciprocal_122": ["quantize_affine_122"], "mul_11864": ["quantize_affine_122"], "mul_11867": ["quantize_affine_122"], "round_246": ["quantize_affine_122"], "convert_element_type_732": ["choose_qparams_affine_default_122"], "view_1915": ["quantize_affine_122"], "add_18779": ["quantize_affine_122"], "clamp_min_368": ["quantize_affine_122"], "clamp_max_245": ["quantize_affine_122"], "convert_element_type_733": ["quantize_affine_122", "dequantize_affine_244"], "view_1919": ["dequantize_affine_244"], "convert_element_type_734": ["dequantize_affine_244"], "convert_element_type_735": ["dequantize_affine_244"], "sub_5613": ["dequantize_affine_244"], "view_1918": ["dequantize_affine_244"], "mul_11889": ["dequantize_affine_244"], "view_1921": ["dequantize_affine_245"], "view_1923": ["dequantize_affine_245"], "convert_element_type_736": ["dequantize_affine_245"], "convert_element_type_737": ["dequantize_affine_245"], "sub_5617": ["dequantize_affine_245"], "view_1922": ["dequantize_affine_245"], "mul_11894": ["dequantize_affine_245"], "view_1924": ["dequantize_affine_245"], "mul_11704": ["linear_120"], "view_1893": ["linear_120"], "permute_201": ["linear_120"], "mm_default_59": ["linear_120"], "add_tensor_59": ["linear_120"], "view_1894": ["linear_120"], "mul_11711": ["mul_745"], "view_1895": ["view_60"], "permute_202": ["transpose_80"], "clone_162": ["contiguous_80"], "mul_11798": ["linear_121"], "view_1909": ["linear_121"], "permute_203": ["linear_121"], "mm_20": ["linear_121"], "view_1910": ["linear_121"], "view_1911": ["view_61"], "permute_204": ["transpose_81"], "clone_163": ["contiguous_81"], "mul_11899": ["linear_122"], "view_1925": ["linear_122"], "permute_205": ["linear_122"], "mm_default_58": ["linear_122"], "add_tensor_58": ["linear_122"], "view_1926": ["linear_122"], "view_1927": ["view_62"], "permute_206": ["transpose_82"], "clone_164": ["contiguous_82"], "_scaled_dot_product_efficient_attention_20": ["scaled_dot_product_attention_20"], "getitem_162": ["scaled_dot_product_attention_20"], "permute_207": ["transpose_83"], "view_1928": ["reshape_20", "choose_qparams_affine_default_123", "quantize_affine_123"], "amax_123": ["choose_qparams_affine_default_123"], "full_247": ["choose_qparams_affine_default_123"], "maximum_123": ["choose_qparams_affine_default_123"], "amin_123": ["choose_qparams_affine_default_123"], "full_246": ["choose_qparams_affine_default_123"], "minimum_123": ["choose_qparams_affine_default_123"], "sub_5635": ["choose_qparams_affine_default_123"], "div_246": ["choose_qparams_affine_default_123"], "clamp_min_369": ["choose_qparams_affine_default_123"], "div_247": ["choose_qparams_affine_default_123"], "round_247": ["choose_qparams_affine_default_123"], "sub_5641": ["choose_qparams_affine_default_123"], "clamp_min_370": ["choose_qparams_affine_default_123"], "clamp_max_246": ["choose_qparams_affine_default_123"], "view_1931": ["quantize_affine_123"], "reciprocal_123": ["quantize_affine_123"], "mul_11969": ["quantize_affine_123"], "mul_11972": ["quantize_affine_123"], "round_248": ["quantize_affine_123"], "convert_element_type_738": ["choose_qparams_affine_default_123"], "view_1932": ["quantize_affine_123"], "add_18943": ["quantize_affine_123"], "clamp_min_371": ["quantize_affine_123"], "clamp_max_247": ["quantize_affine_123"], "convert_element_type_739": ["quantize_affine_123", "dequantize_affine_246"], "view_1936": ["dequantize_affine_246"], "convert_element_type_740": ["dequantize_affine_246"], "convert_element_type_741": ["dequantize_affine_246"], "sub_5661": ["dequantize_affine_246"], "view_1935": ["dequantize_affine_246"], "mul_11994": ["dequantize_affine_246"], "view_1938": ["dequantize_affine_247"], "view_1940": ["dequantize_affine_247"], "convert_element_type_742": ["dequantize_affine_247"], "convert_element_type_743": ["dequantize_affine_247"], "sub_5665": ["dequantize_affine_247"], "view_1939": ["dequantize_affine_247"], "mul_11999": ["dequantize_affine_247"], "view_1941": ["dequantize_affine_247"], "mul_12004": ["linear_123"], "view_1942": ["linear_123"], "permute_208": ["linear_123"], "mm_default_57": ["linear_123"], "add_tensor_57": ["linear_123"], "view_1943": ["linear_123", "dropout_61"], "add_19006": ["add_289"], "clone_166": ["layer_norm_41"], "var_mean_41": ["layer_norm_41"], "getitem_166": ["layer_norm_41"], "getitem_167": ["layer_norm_41"], "sub_5671": ["layer_norm_41"], "add_19011": ["layer_norm_41"], "rsqrt_41": ["layer_norm_41"], "mul_12015": ["layer_norm_41"], "mul_12016": ["layer_norm_41"], "add_19012": ["layer_norm_41", "choose_qparams_affine_default_124", "quantize_affine_124"], "amax_124": ["choose_qparams_affine_default_124"], "full_249": ["choose_qparams_affine_default_124"], "maximum_124": ["choose_qparams_affine_default_124"], "amin_124": ["choose_qparams_affine_default_124"], "full_248": ["choose_qparams_affine_default_124"], "minimum_124": ["choose_qparams_affine_default_124"], "sub_5682": ["choose_qparams_affine_default_124"], "div_248": ["choose_qparams_affine_default_124"], "clamp_min_372": ["choose_qparams_affine_default_124"], "div_249": ["choose_qparams_affine_default_124"], "round_249": ["choose_qparams_affine_default_124"], "sub_5688": ["choose_qparams_affine_default_124"], "clamp_min_373": ["choose_qparams_affine_default_124"], "clamp_max_248": ["choose_qparams_affine_default_124"], "view_1946": ["quantize_affine_124"], "reciprocal_124": ["quantize_affine_124"], "mul_12064": ["quantize_affine_124"], "mul_12067": ["quantize_affine_124"], "round_250": ["quantize_affine_124"], "convert_element_type_744": ["choose_qparams_affine_default_124"], "view_1947": ["quantize_affine_124"], "add_19099": ["quantize_affine_124"], "clamp_min_374": ["quantize_affine_124"], "clamp_max_249": ["quantize_affine_124"], "convert_element_type_745": ["quantize_affine_124", "dequantize_affine_248"], "view_1951": ["dequantize_affine_248"], "convert_element_type_746": ["dequantize_affine_248"], "convert_element_type_747": ["dequantize_affine_248"], "sub_5708": ["dequantize_affine_248"], "view_1950": ["dequantize_affine_248"], "mul_12089": ["dequantize_affine_248"], "view_1953": ["dequantize_affine_249"], "view_1955": ["dequantize_affine_249"], "convert_element_type_748": ["dequantize_affine_249"], "convert_element_type_749": ["dequantize_affine_249"], "sub_5712": ["dequantize_affine_249"], "view_1954": ["dequantize_affine_249"], "mul_12094": ["dequantize_affine_249"], "view_1956": ["dequantize_affine_249"], "mul_12099": ["linear_124"], "view_1957": ["linear_124"], "permute_209": ["linear_124"], "mm_default_56": ["linear_124"], "add_tensor_56": ["linear_124"], "view_1958": ["linear_124"], "mul_12106": ["gelu_22"], "mul_12107": ["gelu_22"], "erf_22": ["gelu_22"], "add_19158": ["gelu_22"], "mul_12108": ["gelu_22", "dropout_62", "choose_qparams_affine_default_125", "quantize_affine_125"], "amax_125": ["choose_qparams_affine_default_125"], "full_251": ["choose_qparams_affine_default_125"], "maximum_125": ["choose_qparams_affine_default_125"], "amin_125": ["choose_qparams_affine_default_125"], "full_250": ["choose_qparams_affine_default_125"], "minimum_125": ["choose_qparams_affine_default_125"], "sub_5725": ["choose_qparams_affine_default_125"], "div_250": ["choose_qparams_affine_default_125"], "clamp_min_375": ["choose_qparams_affine_default_125"], "div_251": ["choose_qparams_affine_default_125"], "round_251": ["choose_qparams_affine_default_125"], "sub_5731": ["choose_qparams_affine_default_125"], "clamp_min_376": ["choose_qparams_affine_default_125"], "clamp_max_250": ["choose_qparams_affine_default_125"], "view_1961": ["quantize_affine_125"], "reciprocal_125": ["quantize_affine_125"], "mul_12154": ["quantize_affine_125"], "mul_12157": ["quantize_affine_125"], "round_252": ["quantize_affine_125"], "convert_element_type_750": ["choose_qparams_affine_default_125"], "view_1962": ["quantize_affine_125"], "add_19241": ["quantize_affine_125"], "clamp_min_377": ["quantize_affine_125"], "clamp_max_251": ["quantize_affine_125"], "convert_element_type_751": ["quantize_affine_125", "dequantize_affine_250"], "view_1966": ["dequantize_affine_250"], "convert_element_type_752": ["dequantize_affine_250"], "convert_element_type_753": ["dequantize_affine_250"], "sub_5751": ["dequantize_affine_250"], "view_1965": ["dequantize_affine_250"], "mul_12179": ["dequantize_affine_250"], "view_1968": ["dequantize_affine_251"], "view_1970": ["dequantize_affine_251"], "convert_element_type_754": ["dequantize_affine_251"], "convert_element_type_755": ["dequantize_affine_251"], "sub_5755": ["dequantize_affine_251"], "view_1969": ["dequantize_affine_251"], "mul_12184": ["dequantize_affine_251"], "view_1971": ["dequantize_affine_251"], "mul_12189": ["linear_125"], "view_1972": ["linear_125"], "permute_210": ["linear_125"], "mm_default_55": ["linear_125"], "add_tensor_55": ["linear_125"], "view_1973": ["linear_125", "dropout_63"], "add_19304": ["add_294"], "clone_169": ["layer_norm_42"], "var_mean_42": ["layer_norm_42"], "getitem_168": ["layer_norm_42"], "getitem_169": ["layer_norm_42"], "sub_5761": ["layer_norm_42"], "add_19309": ["layer_norm_42"], "rsqrt_42": ["layer_norm_42"], "mul_12200": ["layer_norm_42"], "mul_12201": ["layer_norm_42"], "add_19310": ["layer_norm_42", "choose_qparams_affine_default_126", "quantize_affine_126", "choose_qparams_affine_default_127", "quantize_affine_127", "choose_qparams_affine_default_128", "quantize_affine_128"], "amax_126": ["choose_qparams_affine_default_126"], "full_253": ["choose_qparams_affine_default_126"], "maximum_126": ["choose_qparams_affine_default_126"], "amin_126": ["choose_qparams_affine_default_126"], "full_252": ["choose_qparams_affine_default_126"], "minimum_126": ["choose_qparams_affine_default_126"], "sub_5772": ["choose_qparams_affine_default_126"], "div_252": ["choose_qparams_affine_default_126"], "clamp_min_378": ["choose_qparams_affine_default_126"], "div_253": ["choose_qparams_affine_default_126"], "round_253": ["choose_qparams_affine_default_126"], "sub_5778": ["choose_qparams_affine_default_126"], "clamp_min_379": ["choose_qparams_affine_default_126"], "clamp_max_252": ["choose_qparams_affine_default_126"], "view_1976": ["quantize_affine_126"], "reciprocal_126": ["quantize_affine_126"], "mul_12249": ["quantize_affine_126"], "mul_12252": ["quantize_affine_126"], "round_254": ["quantize_affine_126"], "convert_element_type_756": ["choose_qparams_affine_default_126"], "view_1977": ["quantize_affine_126"], "add_19397": ["quantize_affine_126"], "clamp_min_380": ["quantize_affine_126"], "clamp_max_253": ["quantize_affine_126"], "convert_element_type_757": ["quantize_affine_126", "dequantize_affine_252"], "view_1981": ["dequantize_affine_252"], "convert_element_type_758": ["dequantize_affine_252"], "convert_element_type_759": ["dequantize_affine_252"], "sub_5798": ["dequantize_affine_252"], "view_1980": ["dequantize_affine_252"], "mul_12274": ["dequantize_affine_252"], "view_1983": ["dequantize_affine_253"], "view_1985": ["dequantize_affine_253"], "convert_element_type_760": ["dequantize_affine_253"], "convert_element_type_761": ["dequantize_affine_253"], "sub_5802": ["dequantize_affine_253"], "view_1984": ["dequantize_affine_253"], "mul_12279": ["dequantize_affine_253"], "view_1986": ["dequantize_affine_253"], "amax_127": ["choose_qparams_affine_default_127"], "full_255": ["choose_qparams_affine_default_127"], "maximum_127": ["choose_qparams_affine_default_127"], "amin_127": ["choose_qparams_affine_default_127"], "full_254": ["choose_qparams_affine_default_127"], "minimum_127": ["choose_qparams_affine_default_127"], "sub_5817": ["choose_qparams_affine_default_127"], "div_254": ["choose_qparams_affine_default_127"], "clamp_min_381": ["choose_qparams_affine_default_127"], "div_255": ["choose_qparams_affine_default_127"], "round_255": ["choose_qparams_affine_default_127"], "sub_5823": ["choose_qparams_affine_default_127"], "clamp_min_382": ["choose_qparams_affine_default_127"], "clamp_max_254": ["choose_qparams_affine_default_127"], "view_1992": ["quantize_affine_127"], "reciprocal_127": ["quantize_affine_127"], "mul_12345": ["quantize_affine_127"], "mul_12348": ["quantize_affine_127"], "round_256": ["quantize_affine_127"], "convert_element_type_762": ["choose_qparams_affine_default_127"], "view_1993": ["quantize_affine_127"], "add_19549": ["quantize_affine_127"], "clamp_min_383": ["quantize_affine_127"], "clamp_max_255": ["quantize_affine_127"], "convert_element_type_763": ["quantize_affine_127", "dequantize_affine_254"], "view_1997": ["dequantize_affine_254"], "convert_element_type_764": ["dequantize_affine_254"], "convert_element_type_765": ["dequantize_affine_254"], "sub_5843": ["dequantize_affine_254"], "view_1996": ["dequantize_affine_254"], "mul_12370": ["dequantize_affine_254"], "view_1999": ["dequantize_affine_255"], "view_2001": ["dequantize_affine_255"], "convert_element_type_766": ["dequantize_affine_255"], "convert_element_type_767": ["dequantize_affine_255"], "sub_5847": ["dequantize_affine_255"], "view_2000": ["dequantize_affine_255"], "mul_12375": ["dequantize_affine_255"], "view_2002": ["dequantize_affine_255"], "amax_128": ["choose_qparams_affine_default_128"], "full_257": ["choose_qparams_affine_default_128"], "maximum_128": ["choose_qparams_affine_default_128"], "amin_128": ["choose_qparams_affine_default_128"], "full_256": ["choose_qparams_affine_default_128"], "minimum_128": ["choose_qparams_affine_default_128"], "sub_5861": ["choose_qparams_affine_default_128"], "div_256": ["choose_qparams_affine_default_128"], "clamp_min_384": ["choose_qparams_affine_default_128"], "div_257": ["choose_qparams_affine_default_128"], "round_257": ["choose_qparams_affine_default_128"], "sub_5867": ["choose_qparams_affine_default_128"], "clamp_min_385": ["choose_qparams_affine_default_128"], "clamp_max_256": ["choose_qparams_affine_default_128"], "view_2008": ["quantize_affine_128"], "reciprocal_128": ["quantize_affine_128"], "mul_12444": ["quantize_affine_128"], "mul_12447": ["quantize_affine_128"], "round_258": ["quantize_affine_128"], "convert_element_type_768": ["choose_qparams_affine_default_128"], "view_2009": ["quantize_affine_128"], "add_19697": ["quantize_affine_128"], "clamp_min_386": ["quantize_affine_128"], "clamp_max_257": ["quantize_affine_128"], "convert_element_type_769": ["quantize_affine_128", "dequantize_affine_256"], "view_2013": ["dequantize_affine_256"], "convert_element_type_770": ["dequantize_affine_256"], "convert_element_type_771": ["dequantize_affine_256"], "sub_5887": ["dequantize_affine_256"], "view_2012": ["dequantize_affine_256"], "mul_12469": ["dequantize_affine_256"], "view_2015": ["dequantize_affine_257"], "view_2017": ["dequantize_affine_257"], "convert_element_type_772": ["dequantize_affine_257"], "convert_element_type_773": ["dequantize_affine_257"], "sub_5891": ["dequantize_affine_257"], "view_2016": ["dequantize_affine_257"], "mul_12474": ["dequantize_affine_257"], "view_2018": ["dequantize_affine_257"], "mul_12284": ["linear_126"], "view_1987": ["linear_126"], "permute_211": ["linear_126"], "mm_default_54": ["linear_126"], "add_tensor_54": ["linear_126"], "view_1988": ["linear_126"], "mul_12291": ["mul_782"], "view_1989": ["view_63"], "permute_212": ["transpose_84"], "clone_170": ["contiguous_84"], "mul_12378": ["linear_127"], "view_2003": ["linear_127"], "permute_213": ["linear_127"], "mm_21": ["linear_127"], "view_2004": ["linear_127"], "view_2005": ["view_64"], "permute_214": ["transpose_85"], "clone_171": ["contiguous_85"], "mul_12479": ["linear_128"], "view_2019": ["linear_128"], "permute_215": ["linear_128"], "mm_default_53": ["linear_128"], "add_tensor_53": ["linear_128"], "view_2020": ["linear_128"], "view_2021": ["view_65"], "permute_216": ["transpose_86"], "clone_172": ["contiguous_86"], "_scaled_dot_product_efficient_attention_21": ["scaled_dot_product_attention_21"], "getitem_170": ["scaled_dot_product_attention_21"], "permute_217": ["transpose_87"], "view_2022": ["reshape_21", "choose_qparams_affine_default_129", "quantize_affine_129"], "amax_129": ["choose_qparams_affine_default_129"], "full_259": ["choose_qparams_affine_default_129"], "maximum_129": ["choose_qparams_affine_default_129"], "amin_129": ["choose_qparams_affine_default_129"], "full_258": ["choose_qparams_affine_default_129"], "minimum_129": ["choose_qparams_affine_default_129"], "sub_5909": ["choose_qparams_affine_default_129"], "div_258": ["choose_qparams_affine_default_129"], "clamp_min_387": ["choose_qparams_affine_default_129"], "div_259": ["choose_qparams_affine_default_129"], "round_259": ["choose_qparams_affine_default_129"], "sub_5915": ["choose_qparams_affine_default_129"], "clamp_min_388": ["choose_qparams_affine_default_129"], "clamp_max_258": ["choose_qparams_affine_default_129"], "view_2025": ["quantize_affine_129"], "reciprocal_129": ["quantize_affine_129"], "mul_12549": ["quantize_affine_129"], "mul_12552": ["quantize_affine_129"], "round_260": ["quantize_affine_129"], "convert_element_type_774": ["choose_qparams_affine_default_129"], "view_2026": ["quantize_affine_129"], "add_19861": ["quantize_affine_129"], "clamp_min_389": ["quantize_affine_129"], "clamp_max_259": ["quantize_affine_129"], "convert_element_type_775": ["quantize_affine_129", "dequantize_affine_258"], "view_2030": ["dequantize_affine_258"], "convert_element_type_776": ["dequantize_affine_258"], "convert_element_type_777": ["dequantize_affine_258"], "sub_5935": ["dequantize_affine_258"], "view_2029": ["dequantize_affine_258"], "mul_12574": ["dequantize_affine_258"], "view_2032": ["dequantize_affine_259"], "view_2034": ["dequantize_affine_259"], "convert_element_type_778": ["dequantize_affine_259"], "convert_element_type_779": ["dequantize_affine_259"], "sub_5939": ["dequantize_affine_259"], "view_2033": ["dequantize_affine_259"], "mul_12579": ["dequantize_affine_259"], "view_2035": ["dequantize_affine_259"], "mul_12584": ["linear_129"], "view_2036": ["linear_129"], "permute_218": ["linear_129"], "mm_default_52": ["linear_129"], "add_tensor_52": ["linear_129"], "view_2037": ["linear_129", "dropout_64"], "add_19924": ["add_303"], "clone_174": ["layer_norm_43"], "var_mean_43": ["layer_norm_43"], "getitem_174": ["layer_norm_43"], "getitem_175": ["layer_norm_43"], "sub_5945": ["layer_norm_43"], "add_19929": ["layer_norm_43"], "rsqrt_43": ["layer_norm_43"], "mul_12595": ["layer_norm_43"], "mul_12596": ["layer_norm_43"], "add_19930": ["layer_norm_43", "choose_qparams_affine_default_130", "quantize_affine_130"], "amax_130": ["choose_qparams_affine_default_130"], "full_261": ["choose_qparams_affine_default_130"], "maximum_130": ["choose_qparams_affine_default_130"], "amin_130": ["choose_qparams_affine_default_130"], "full_260": ["choose_qparams_affine_default_130"], "minimum_130": ["choose_qparams_affine_default_130"], "sub_5956": ["choose_qparams_affine_default_130"], "div_260": ["choose_qparams_affine_default_130"], "clamp_min_390": ["choose_qparams_affine_default_130"], "div_261": ["choose_qparams_affine_default_130"], "round_261": ["choose_qparams_affine_default_130"], "sub_5962": ["choose_qparams_affine_default_130"], "clamp_min_391": ["choose_qparams_affine_default_130"], "clamp_max_260": ["choose_qparams_affine_default_130"], "view_2040": ["quantize_affine_130"], "reciprocal_130": ["quantize_affine_130"], "mul_12644": ["quantize_affine_130"], "mul_12647": ["quantize_affine_130"], "round_262": ["quantize_affine_130"], "convert_element_type_780": ["choose_qparams_affine_default_130"], "view_2041": ["quantize_affine_130"], "add_20017": ["quantize_affine_130"], "clamp_min_392": ["quantize_affine_130"], "clamp_max_261": ["quantize_affine_130"], "convert_element_type_781": ["quantize_affine_130", "dequantize_affine_260"], "view_2045": ["dequantize_affine_260"], "convert_element_type_782": ["dequantize_affine_260"], "convert_element_type_783": ["dequantize_affine_260"], "sub_5982": ["dequantize_affine_260"], "view_2044": ["dequantize_affine_260"], "mul_12669": ["dequantize_affine_260"], "view_2047": ["dequantize_affine_261"], "view_2049": ["dequantize_affine_261"], "convert_element_type_784": ["dequantize_affine_261"], "convert_element_type_785": ["dequantize_affine_261"], "sub_5986": ["dequantize_affine_261"], "view_2048": ["dequantize_affine_261"], "mul_12674": ["dequantize_affine_261"], "view_2050": ["dequantize_affine_261"], "mul_12679": ["linear_130"], "view_2051": ["linear_130"], "permute_219": ["linear_130"], "mm_default_51": ["linear_130"], "add_tensor_51": ["linear_130"], "view_2052": ["linear_130"], "mul_12686": ["gelu_23"], "mul_12687": ["gelu_23"], "erf_23": ["gelu_23"], "add_20076": ["gelu_23"], "mul_12688": ["gelu_23", "dropout_65", "choose_qparams_affine_default_131", "quantize_affine_131"], "amax_131": ["choose_qparams_affine_default_131"], "full_263": ["choose_qparams_affine_default_131"], "maximum_131": ["choose_qparams_affine_default_131"], "amin_131": ["choose_qparams_affine_default_131"], "full_262": ["choose_qparams_affine_default_131"], "minimum_131": ["choose_qparams_affine_default_131"], "sub_5999": ["choose_qparams_affine_default_131"], "div_262": ["choose_qparams_affine_default_131"], "clamp_min_393": ["choose_qparams_affine_default_131"], "div_263": ["choose_qparams_affine_default_131"], "round_263": ["choose_qparams_affine_default_131"], "sub_6005": ["choose_qparams_affine_default_131"], "clamp_min_394": ["choose_qparams_affine_default_131"], "clamp_max_262": ["choose_qparams_affine_default_131"], "view_2055": ["quantize_affine_131"], "reciprocal_131": ["quantize_affine_131"], "mul_12734": ["quantize_affine_131"], "mul_12737": ["quantize_affine_131"], "round_264": ["quantize_affine_131"], "convert_element_type_786": ["choose_qparams_affine_default_131"], "view_2056": ["quantize_affine_131"], "add_20159": ["quantize_affine_131"], "clamp_min_395": ["quantize_affine_131"], "clamp_max_263": ["quantize_affine_131"], "convert_element_type_787": ["quantize_affine_131", "dequantize_affine_262"], "view_2060": ["dequantize_affine_262"], "convert_element_type_788": ["dequantize_affine_262"], "convert_element_type_789": ["dequantize_affine_262"], "sub_6025": ["dequantize_affine_262"], "view_2059": ["dequantize_affine_262"], "mul_12759": ["dequantize_affine_262"], "view_2062": ["dequantize_affine_263"], "view_2064": ["dequantize_affine_263"], "convert_element_type_790": ["dequantize_affine_263"], "convert_element_type_791": ["dequantize_affine_263"], "sub_6029": ["dequantize_affine_263"], "view_2063": ["dequantize_affine_263"], "mul_12764": ["dequantize_affine_263"], "view_2065": ["dequantize_affine_263"], "mul_12769": ["linear_131"], "view_2066": ["linear_131"], "permute_220": ["linear_131"], "mm_default_50": ["linear_131"], "add_tensor_50": ["linear_131"], "view_2067": ["linear_131", "dropout_66"], "add_20222": ["add_308"], "clone_177": ["layer_norm_44"], "var_mean_44": ["layer_norm_44"], "getitem_176": ["layer_norm_44"], "getitem_177": ["layer_norm_44"], "sub_6035": ["layer_norm_44"], "add_20227": ["layer_norm_44"], "rsqrt_44": ["layer_norm_44"], "mul_12780": ["layer_norm_44"], "mul_12781": ["layer_norm_44"], "add_20228": ["layer_norm_44", "choose_qparams_affine_default_132", "quantize_affine_132", "choose_qparams_affine_default_133", "quantize_affine_133", "choose_qparams_affine_default_134", "quantize_affine_134"], "amax_132": ["choose_qparams_affine_default_132"], "full_265": ["choose_qparams_affine_default_132"], "maximum_132": ["choose_qparams_affine_default_132"], "amin_132": ["choose_qparams_affine_default_132"], "full_264": ["choose_qparams_affine_default_132"], "minimum_132": ["choose_qparams_affine_default_132"], "sub_6046": ["choose_qparams_affine_default_132"], "div_264": ["choose_qparams_affine_default_132"], "clamp_min_396": ["choose_qparams_affine_default_132"], "div_265": ["choose_qparams_affine_default_132"], "round_265": ["choose_qparams_affine_default_132"], "sub_6052": ["choose_qparams_affine_default_132"], "clamp_min_397": ["choose_qparams_affine_default_132"], "clamp_max_264": ["choose_qparams_affine_default_132"], "view_2070": ["quantize_affine_132"], "reciprocal_132": ["quantize_affine_132"], "mul_12829": ["quantize_affine_132"], "mul_12832": ["quantize_affine_132"], "round_266": ["quantize_affine_132"], "convert_element_type_792": ["choose_qparams_affine_default_132"], "view_2071": ["quantize_affine_132"], "add_20315": ["quantize_affine_132"], "clamp_min_398": ["quantize_affine_132"], "clamp_max_265": ["quantize_affine_132"], "convert_element_type_793": ["quantize_affine_132", "dequantize_affine_264"], "view_2075": ["dequantize_affine_264"], "convert_element_type_794": ["dequantize_affine_264"], "convert_element_type_795": ["dequantize_affine_264"], "sub_6072": ["dequantize_affine_264"], "view_2074": ["dequantize_affine_264"], "mul_12854": ["dequantize_affine_264"], "view_2077": ["dequantize_affine_265"], "view_2079": ["dequantize_affine_265"], "convert_element_type_796": ["dequantize_affine_265"], "convert_element_type_797": ["dequantize_affine_265"], "sub_6076": ["dequantize_affine_265"], "view_2078": ["dequantize_affine_265"], "mul_12859": ["dequantize_affine_265"], "view_2080": ["dequantize_affine_265"], "amax_133": ["choose_qparams_affine_default_133"], "full_267": ["choose_qparams_affine_default_133"], "maximum_133": ["choose_qparams_affine_default_133"], "amin_133": ["choose_qparams_affine_default_133"], "full_266": ["choose_qparams_affine_default_133"], "minimum_133": ["choose_qparams_affine_default_133"], "sub_6091": ["choose_qparams_affine_default_133"], "div_266": ["choose_qparams_affine_default_133"], "clamp_min_399": ["choose_qparams_affine_default_133"], "div_267": ["choose_qparams_affine_default_133"], "round_267": ["choose_qparams_affine_default_133"], "sub_6097": ["choose_qparams_affine_default_133"], "clamp_min_400": ["choose_qparams_affine_default_133"], "clamp_max_266": ["choose_qparams_affine_default_133"], "view_2086": ["quantize_affine_133"], "reciprocal_133": ["quantize_affine_133"], "mul_12925": ["quantize_affine_133"], "mul_12928": ["quantize_affine_133"], "round_268": ["quantize_affine_133"], "convert_element_type_798": ["choose_qparams_affine_default_133"], "view_2087": ["quantize_affine_133"], "add_20467": ["quantize_affine_133"], "clamp_min_401": ["quantize_affine_133"], "clamp_max_267": ["quantize_affine_133"], "convert_element_type_799": ["quantize_affine_133", "dequantize_affine_266"], "view_2091": ["dequantize_affine_266"], "convert_element_type_800": ["dequantize_affine_266"], "convert_element_type_801": ["dequantize_affine_266"], "sub_6117": ["dequantize_affine_266"], "view_2090": ["dequantize_affine_266"], "mul_12950": ["dequantize_affine_266"], "view_2093": ["dequantize_affine_267"], "view_2095": ["dequantize_affine_267"], "convert_element_type_802": ["dequantize_affine_267"], "convert_element_type_803": ["dequantize_affine_267"], "sub_6121": ["dequantize_affine_267"], "view_2094": ["dequantize_affine_267"], "mul_12955": ["dequantize_affine_267"], "view_2096": ["dequantize_affine_267"], "amax_134": ["choose_qparams_affine_default_134"], "full_269": ["choose_qparams_affine_default_134"], "maximum_134": ["choose_qparams_affine_default_134"], "amin_134": ["choose_qparams_affine_default_134"], "full_268": ["choose_qparams_affine_default_134"], "minimum_134": ["choose_qparams_affine_default_134"], "sub_6135": ["choose_qparams_affine_default_134"], "div_268": ["choose_qparams_affine_default_134"], "clamp_min_402": ["choose_qparams_affine_default_134"], "div_269": ["choose_qparams_affine_default_134"], "round_269": ["choose_qparams_affine_default_134"], "sub_6141": ["choose_qparams_affine_default_134"], "clamp_min_403": ["choose_qparams_affine_default_134"], "clamp_max_268": ["choose_qparams_affine_default_134"], "view_2102": ["quantize_affine_134"], "reciprocal_134": ["quantize_affine_134"], "mul_13024": ["quantize_affine_134"], "mul_13027": ["quantize_affine_134"], "round_270": ["quantize_affine_134"], "convert_element_type_804": ["choose_qparams_affine_default_134"], "view_2103": ["quantize_affine_134"], "add_20615": ["quantize_affine_134"], "clamp_min_404": ["quantize_affine_134"], "clamp_max_269": ["quantize_affine_134"], "convert_element_type_805": ["quantize_affine_134", "dequantize_affine_268"], "view_2107": ["dequantize_affine_268"], "convert_element_type_806": ["dequantize_affine_268"], "convert_element_type_807": ["dequantize_affine_268"], "sub_6161": ["dequantize_affine_268"], "view_2106": ["dequantize_affine_268"], "mul_13049": ["dequantize_affine_268"], "view_2109": ["dequantize_affine_269"], "view_2111": ["dequantize_affine_269"], "convert_element_type_808": ["dequantize_affine_269"], "convert_element_type_809": ["dequantize_affine_269"], "sub_6165": ["dequantize_affine_269"], "view_2110": ["dequantize_affine_269"], "mul_13054": ["dequantize_affine_269"], "view_2112": ["dequantize_affine_269"], "mul_12864": ["linear_132"], "view_2081": ["linear_132"], "permute_221": ["linear_132"], "mm_default_49": ["linear_132"], "add_tensor_49": ["linear_132"], "view_2082": ["linear_132"], "mul_12871": ["mul_819"], "view_2083": ["view_66"], "permute_222": ["transpose_88"], "clone_178": ["contiguous_88"], "mul_12958": ["linear_133"], "view_2097": ["linear_133"], "permute_223": ["linear_133"], "mm_22": ["linear_133"], "view_2098": ["linear_133"], "view_2099": ["view_67"], "permute_224": ["transpose_89"], "clone_179": ["contiguous_89"], "mul_13059": ["linear_134"], "view_2113": ["linear_134"], "permute_225": ["linear_134"], "mm_default_48": ["linear_134"], "add_tensor_48": ["linear_134"], "view_2114": ["linear_134"], "view_2115": ["view_68"], "permute_226": ["transpose_90"], "clone_180": ["contiguous_90"], "_scaled_dot_product_efficient_attention_22": ["scaled_dot_product_attention_22"], "getitem_178": ["scaled_dot_product_attention_22"], "permute_227": ["transpose_91"], "view_2116": ["reshape_22", "choose_qparams_affine_default_135", "quantize_affine_135"], "amax_135": ["choose_qparams_affine_default_135"], "full_271": ["choose_qparams_affine_default_135"], "maximum_135": ["choose_qparams_affine_default_135"], "amin_135": ["choose_qparams_affine_default_135"], "full_270": ["choose_qparams_affine_default_135"], "minimum_135": ["choose_qparams_affine_default_135"], "sub_6183": ["choose_qparams_affine_default_135"], "div_270": ["choose_qparams_affine_default_135"], "clamp_min_405": ["choose_qparams_affine_default_135"], "div_271": ["choose_qparams_affine_default_135"], "round_271": ["choose_qparams_affine_default_135"], "sub_6189": ["choose_qparams_affine_default_135"], "clamp_min_406": ["choose_qparams_affine_default_135"], "clamp_max_270": ["choose_qparams_affine_default_135"], "view_2119": ["quantize_affine_135"], "reciprocal_135": ["quantize_affine_135"], "mul_13129": ["quantize_affine_135"], "mul_13132": ["quantize_affine_135"], "round_272": ["quantize_affine_135"], "convert_element_type_810": ["choose_qparams_affine_default_135"], "view_2120": ["quantize_affine_135"], "add_20779": ["quantize_affine_135"], "clamp_min_407": ["quantize_affine_135"], "clamp_max_271": ["quantize_affine_135"], "convert_element_type_811": ["quantize_affine_135", "dequantize_affine_270"], "view_2124": ["dequantize_affine_270"], "convert_element_type_812": ["dequantize_affine_270"], "convert_element_type_813": ["dequantize_affine_270"], "sub_6209": ["dequantize_affine_270"], "view_2123": ["dequantize_affine_270"], "mul_13154": ["dequantize_affine_270"], "view_2126": ["dequantize_affine_271"], "view_2128": ["dequantize_affine_271"], "convert_element_type_814": ["dequantize_affine_271"], "convert_element_type_815": ["dequantize_affine_271"], "sub_6213": ["dequantize_affine_271"], "view_2127": ["dequantize_affine_271"], "mul_13159": ["dequantize_affine_271"], "view_2129": ["dequantize_affine_271"], "mul_13164": ["linear_135"], "view_2130": ["linear_135"], "permute_228": ["linear_135"], "mm_default_47": ["linear_135"], "add_tensor_47": ["linear_135"], "view_2131": ["linear_135", "dropout_67"], "add_20842": ["add_317"], "clone_182": ["layer_norm_45"], "var_mean_45": ["layer_norm_45"], "getitem_182": ["layer_norm_45"], "getitem_183": ["layer_norm_45"], "sub_6219": ["layer_norm_45"], "add_20847": ["layer_norm_45"], "rsqrt_45": ["layer_norm_45"], "mul_13175": ["layer_norm_45"], "mul_13176": ["layer_norm_45"], "add_20848": ["layer_norm_45", "choose_qparams_affine_default_136", "quantize_affine_136"], "amax_136": ["choose_qparams_affine_default_136"], "full_273": ["choose_qparams_affine_default_136"], "maximum_136": ["choose_qparams_affine_default_136"], "amin_136": ["choose_qparams_affine_default_136"], "full_272": ["choose_qparams_affine_default_136"], "minimum_136": ["choose_qparams_affine_default_136"], "sub_6230": ["choose_qparams_affine_default_136"], "div_272": ["choose_qparams_affine_default_136"], "clamp_min_408": ["choose_qparams_affine_default_136"], "div_273": ["choose_qparams_affine_default_136"], "round_273": ["choose_qparams_affine_default_136"], "sub_6236": ["choose_qparams_affine_default_136"], "clamp_min_409": ["choose_qparams_affine_default_136"], "clamp_max_272": ["choose_qparams_affine_default_136"], "view_2134": ["quantize_affine_136"], "reciprocal_136": ["quantize_affine_136"], "mul_13224": ["quantize_affine_136"], "mul_13227": ["quantize_affine_136"], "round_274": ["quantize_affine_136"], "convert_element_type_816": ["choose_qparams_affine_default_136"], "view_2135": ["quantize_affine_136"], "add_20935": ["quantize_affine_136"], "clamp_min_410": ["quantize_affine_136"], "clamp_max_273": ["quantize_affine_136"], "convert_element_type_817": ["quantize_affine_136", "dequantize_affine_272"], "view_2139": ["dequantize_affine_272"], "convert_element_type_818": ["dequantize_affine_272"], "convert_element_type_819": ["dequantize_affine_272"], "sub_6256": ["dequantize_affine_272"], "view_2138": ["dequantize_affine_272"], "mul_13249": ["dequantize_affine_272"], "view_2141": ["dequantize_affine_273"], "view_2143": ["dequantize_affine_273"], "convert_element_type_820": ["dequantize_affine_273"], "convert_element_type_821": ["dequantize_affine_273"], "sub_6260": ["dequantize_affine_273"], "view_2142": ["dequantize_affine_273"], "mul_13254": ["dequantize_affine_273"], "view_2144": ["dequantize_affine_273"], "mul_13259": ["linear_136"], "view_2145": ["linear_136"], "permute_229": ["linear_136"], "mm_default_46": ["linear_136"], "add_tensor_46": ["linear_136"], "view_2146": ["linear_136"], "mul_13266": ["gelu_24"], "mul_13267": ["gelu_24"], "erf_24": ["gelu_24"], "add_20994": ["gelu_24"], "mul_13268": ["gelu_24", "dropout_68", "choose_qparams_affine_default_137", "quantize_affine_137"], "amax_137": ["choose_qparams_affine_default_137"], "full_275": ["choose_qparams_affine_default_137"], "maximum_137": ["choose_qparams_affine_default_137"], "amin_137": ["choose_qparams_affine_default_137"], "full_274": ["choose_qparams_affine_default_137"], "minimum_137": ["choose_qparams_affine_default_137"], "sub_6273": ["choose_qparams_affine_default_137"], "div_274": ["choose_qparams_affine_default_137"], "clamp_min_411": ["choose_qparams_affine_default_137"], "div_275": ["choose_qparams_affine_default_137"], "round_275": ["choose_qparams_affine_default_137"], "sub_6279": ["choose_qparams_affine_default_137"], "clamp_min_412": ["choose_qparams_affine_default_137"], "clamp_max_274": ["choose_qparams_affine_default_137"], "view_2149": ["quantize_affine_137"], "reciprocal_137": ["quantize_affine_137"], "mul_13314": ["quantize_affine_137"], "mul_13317": ["quantize_affine_137"], "round_276": ["quantize_affine_137"], "convert_element_type_822": ["choose_qparams_affine_default_137"], "view_2150": ["quantize_affine_137"], "add_21077": ["quantize_affine_137"], "clamp_min_413": ["quantize_affine_137"], "clamp_max_275": ["quantize_affine_137"], "convert_element_type_823": ["quantize_affine_137", "dequantize_affine_274"], "view_2154": ["dequantize_affine_274"], "convert_element_type_824": ["dequantize_affine_274"], "convert_element_type_825": ["dequantize_affine_274"], "sub_6299": ["dequantize_affine_274"], "view_2153": ["dequantize_affine_274"], "mul_13339": ["dequantize_affine_274"], "view_2156": ["dequantize_affine_275"], "view_2158": ["dequantize_affine_275"], "convert_element_type_826": ["dequantize_affine_275"], "convert_element_type_827": ["dequantize_affine_275"], "sub_6303": ["dequantize_affine_275"], "view_2157": ["dequantize_affine_275"], "mul_13344": ["dequantize_affine_275"], "view_2159": ["dequantize_affine_275"], "mul_13349": ["linear_137"], "view_2160": ["linear_137"], "permute_230": ["linear_137"], "mm_default_45": ["linear_137"], "add_tensor_45": ["linear_137"], "view_2161": ["linear_137", "dropout_69"], "add_21140": ["add_322"], "clone_185": ["layer_norm_46"], "var_mean_46": ["layer_norm_46"], "getitem_184": ["layer_norm_46"], "getitem_185": ["layer_norm_46"], "sub_6309": ["layer_norm_46"], "add_21145": ["layer_norm_46"], "rsqrt_46": ["layer_norm_46"], "mul_13360": ["layer_norm_46"], "mul_13361": ["layer_norm_46"], "add_21146": ["layer_norm_46", "choose_qparams_affine_default_138", "quantize_affine_138", "choose_qparams_affine_default_139", "quantize_affine_139", "choose_qparams_affine_default_140", "quantize_affine_140"], "amax_138": ["choose_qparams_affine_default_138"], "full_277": ["choose_qparams_affine_default_138"], "maximum_138": ["choose_qparams_affine_default_138"], "amin_138": ["choose_qparams_affine_default_138"], "full_276": ["choose_qparams_affine_default_138"], "minimum_138": ["choose_qparams_affine_default_138"], "sub_6320": ["choose_qparams_affine_default_138"], "div_276": ["choose_qparams_affine_default_138"], "clamp_min_414": ["choose_qparams_affine_default_138"], "div_277": ["choose_qparams_affine_default_138"], "round_277": ["choose_qparams_affine_default_138"], "sub_6326": ["choose_qparams_affine_default_138"], "clamp_min_415": ["choose_qparams_affine_default_138"], "clamp_max_276": ["choose_qparams_affine_default_138"], "view_2164": ["quantize_affine_138"], "reciprocal_138": ["quantize_affine_138"], "mul_13409": ["quantize_affine_138"], "mul_13412": ["quantize_affine_138"], "round_278": ["quantize_affine_138"], "convert_element_type_828": ["choose_qparams_affine_default_138"], "view_2165": ["quantize_affine_138"], "add_21233": ["quantize_affine_138"], "clamp_min_416": ["quantize_affine_138"], "clamp_max_277": ["quantize_affine_138"], "convert_element_type_829": ["quantize_affine_138", "dequantize_affine_276"], "view_2169": ["dequantize_affine_276"], "convert_element_type_830": ["dequantize_affine_276"], "convert_element_type_831": ["dequantize_affine_276"], "sub_6346": ["dequantize_affine_276"], "view_2168": ["dequantize_affine_276"], "mul_13434": ["dequantize_affine_276"], "view_2171": ["dequantize_affine_277"], "view_2173": ["dequantize_affine_277"], "convert_element_type_832": ["dequantize_affine_277"], "convert_element_type_833": ["dequantize_affine_277"], "sub_6350": ["dequantize_affine_277"], "view_2172": ["dequantize_affine_277"], "mul_13439": ["dequantize_affine_277"], "view_2174": ["dequantize_affine_277"], "amax_139": ["choose_qparams_affine_default_139"], "full_279": ["choose_qparams_affine_default_139"], "maximum_139": ["choose_qparams_affine_default_139"], "amin_139": ["choose_qparams_affine_default_139"], "full_278": ["choose_qparams_affine_default_139"], "minimum_139": ["choose_qparams_affine_default_139"], "sub_6365": ["choose_qparams_affine_default_139"], "div_278": ["choose_qparams_affine_default_139"], "clamp_min_417": ["choose_qparams_affine_default_139"], "div_279": ["choose_qparams_affine_default_139"], "round_279": ["choose_qparams_affine_default_139"], "sub_6371": ["choose_qparams_affine_default_139"], "clamp_min_418": ["choose_qparams_affine_default_139"], "clamp_max_278": ["choose_qparams_affine_default_139"], "view_2180": ["quantize_affine_139"], "reciprocal_139": ["quantize_affine_139"], "mul_13505": ["quantize_affine_139"], "mul_13508": ["quantize_affine_139"], "round_280": ["quantize_affine_139"], "convert_element_type_834": ["choose_qparams_affine_default_139"], "view_2181": ["quantize_affine_139"], "add_21385": ["quantize_affine_139"], "clamp_min_419": ["quantize_affine_139"], "clamp_max_279": ["quantize_affine_139"], "convert_element_type_835": ["quantize_affine_139", "dequantize_affine_278"], "view_2185": ["dequantize_affine_278"], "convert_element_type_836": ["dequantize_affine_278"], "convert_element_type_837": ["dequantize_affine_278"], "sub_6391": ["dequantize_affine_278"], "view_2184": ["dequantize_affine_278"], "mul_13530": ["dequantize_affine_278"], "view_2187": ["dequantize_affine_279"], "view_2189": ["dequantize_affine_279"], "convert_element_type_838": ["dequantize_affine_279"], "convert_element_type_839": ["dequantize_affine_279"], "sub_6395": ["dequantize_affine_279"], "view_2188": ["dequantize_affine_279"], "mul_13535": ["dequantize_affine_279"], "view_2190": ["dequantize_affine_279"], "amax_140": ["choose_qparams_affine_default_140"], "full_281": ["choose_qparams_affine_default_140"], "maximum_140": ["choose_qparams_affine_default_140"], "amin_140": ["choose_qparams_affine_default_140"], "full_280": ["choose_qparams_affine_default_140"], "minimum_140": ["choose_qparams_affine_default_140"], "sub_6409": ["choose_qparams_affine_default_140"], "div_280": ["choose_qparams_affine_default_140"], "clamp_min_420": ["choose_qparams_affine_default_140"], "div_281": ["choose_qparams_affine_default_140"], "round_281": ["choose_qparams_affine_default_140"], "sub_6415": ["choose_qparams_affine_default_140"], "clamp_min_421": ["choose_qparams_affine_default_140"], "clamp_max_280": ["choose_qparams_affine_default_140"], "view_2196": ["quantize_affine_140"], "reciprocal_140": ["quantize_affine_140"], "mul_13604": ["quantize_affine_140"], "mul_13607": ["quantize_affine_140"], "round_282": ["quantize_affine_140"], "convert_element_type_840": ["choose_qparams_affine_default_140"], "view_2197": ["quantize_affine_140"], "add_21533": ["quantize_affine_140"], "clamp_min_422": ["quantize_affine_140"], "clamp_max_281": ["quantize_affine_140"], "convert_element_type_841": ["quantize_affine_140", "dequantize_affine_280"], "view_2201": ["dequantize_affine_280"], "convert_element_type_842": ["dequantize_affine_280"], "convert_element_type_843": ["dequantize_affine_280"], "sub_6435": ["dequantize_affine_280"], "view_2200": ["dequantize_affine_280"], "mul_13629": ["dequantize_affine_280"], "view_2203": ["dequantize_affine_281"], "view_2205": ["dequantize_affine_281"], "convert_element_type_844": ["dequantize_affine_281"], "convert_element_type_845": ["dequantize_affine_281"], "sub_6439": ["dequantize_affine_281"], "view_2204": ["dequantize_affine_281"], "mul_13634": ["dequantize_affine_281"], "view_2206": ["dequantize_affine_281"], "mul_13444": ["linear_138"], "view_2175": ["linear_138"], "permute_231": ["linear_138"], "mm_default_44": ["linear_138"], "add_tensor_44": ["linear_138"], "view_2176": ["linear_138"], "mul_13451": ["mul_856"], "view_2177": ["view_69"], "permute_232": ["transpose_92"], "clone_186": ["contiguous_92"], "mul_13538": ["linear_139"], "view_2191": ["linear_139"], "permute_233": ["linear_139"], "mm_23": ["linear_139"], "view_2192": ["linear_139"], "view_2193": ["view_70"], "permute_234": ["transpose_93"], "clone_187": ["contiguous_93"], "mul_13639": ["linear_140"], "view_2207": ["linear_140"], "permute_235": ["linear_140"], "mm_default_43": ["linear_140"], "add_tensor_43": ["linear_140"], "view_2208": ["linear_140"], "view_2209": ["view_71"], "permute_236": ["transpose_94"], "clone_188": ["contiguous_94"], "_scaled_dot_product_efficient_attention_23": ["scaled_dot_product_attention_23"], "getitem_186": ["scaled_dot_product_attention_23"], "permute_237": ["transpose_95"], "view_2210": ["reshape_23", "choose_qparams_affine_default_141", "quantize_affine_141"], "amax_141": ["choose_qparams_affine_default_141"], "full_283": ["choose_qparams_affine_default_141"], "maximum_141": ["choose_qparams_affine_default_141"], "amin_141": ["choose_qparams_affine_default_141"], "full_282": ["choose_qparams_affine_default_141"], "minimum_141": ["choose_qparams_affine_default_141"], "sub_6457": ["choose_qparams_affine_default_141"], "div_282": ["choose_qparams_affine_default_141"], "clamp_min_423": ["choose_qparams_affine_default_141"], "div_283": ["choose_qparams_affine_default_141"], "round_283": ["choose_qparams_affine_default_141"], "sub_6463": ["choose_qparams_affine_default_141"], "clamp_min_424": ["choose_qparams_affine_default_141"], "clamp_max_282": ["choose_qparams_affine_default_141"], "view_2213": ["quantize_affine_141"], "reciprocal_141": ["quantize_affine_141"], "mul_13709": ["quantize_affine_141"], "mul_13712": ["quantize_affine_141"], "round_284": ["quantize_affine_141"], "convert_element_type_846": ["choose_qparams_affine_default_141"], "view_2214": ["quantize_affine_141"], "add_21697": ["quantize_affine_141"], "clamp_min_425": ["quantize_affine_141"], "clamp_max_283": ["quantize_affine_141"], "convert_element_type_847": ["quantize_affine_141", "dequantize_affine_282"], "view_2218": ["dequantize_affine_282"], "convert_element_type_848": ["dequantize_affine_282"], "convert_element_type_849": ["dequantize_affine_282"], "sub_6483": ["dequantize_affine_282"], "view_2217": ["dequantize_affine_282"], "mul_13734": ["dequantize_affine_282"], "view_2220": ["dequantize_affine_283"], "view_2222": ["dequantize_affine_283"], "convert_element_type_850": ["dequantize_affine_283"], "convert_element_type_851": ["dequantize_affine_283"], "sub_6487": ["dequantize_affine_283"], "view_2221": ["dequantize_affine_283"], "mul_13739": ["dequantize_affine_283"], "view_2223": ["dequantize_affine_283"], "mul_13744": ["linear_141"], "view_2224": ["linear_141"], "permute_238": ["linear_141"], "mm_default_42": ["linear_141"], "add_tensor_42": ["linear_141"], "view_2225": ["linear_141", "dropout_70"], "add_21760": ["add_331"], "clone_190": ["layer_norm_47"], "var_mean_47": ["layer_norm_47"], "getitem_190": ["layer_norm_47"], "getitem_191": ["layer_norm_47"], "sub_6493": ["layer_norm_47"], "add_21765": ["layer_norm_47"], "rsqrt_47": ["layer_norm_47"], "mul_13755": ["layer_norm_47"], "mul_13756": ["layer_norm_47"], "add_21766": ["layer_norm_47", "choose_qparams_affine_default_142", "quantize_affine_142"], "amax_142": ["choose_qparams_affine_default_142"], "full_285": ["choose_qparams_affine_default_142"], "maximum_142": ["choose_qparams_affine_default_142"], "amin_142": ["choose_qparams_affine_default_142"], "full_284": ["choose_qparams_affine_default_142"], "minimum_142": ["choose_qparams_affine_default_142"], "sub_6504": ["choose_qparams_affine_default_142"], "div_284": ["choose_qparams_affine_default_142"], "clamp_min_426": ["choose_qparams_affine_default_142"], "div_285": ["choose_qparams_affine_default_142"], "round_285": ["choose_qparams_affine_default_142"], "sub_6510": ["choose_qparams_affine_default_142"], "clamp_min_427": ["choose_qparams_affine_default_142"], "clamp_max_284": ["choose_qparams_affine_default_142"], "view_2228": ["quantize_affine_142"], "reciprocal_142": ["quantize_affine_142"], "mul_13804": ["quantize_affine_142"], "mul_13807": ["quantize_affine_142"], "round_286": ["quantize_affine_142"], "convert_element_type_852": ["choose_qparams_affine_default_142"], "view_2229": ["quantize_affine_142"], "add_21853": ["quantize_affine_142"], "clamp_min_428": ["quantize_affine_142"], "clamp_max_285": ["quantize_affine_142"], "convert_element_type_853": ["quantize_affine_142", "dequantize_affine_284"], "view_2233": ["dequantize_affine_284"], "convert_element_type_854": ["dequantize_affine_284"], "convert_element_type_855": ["dequantize_affine_284"], "sub_6530": ["dequantize_affine_284"], "view_2232": ["dequantize_affine_284"], "mul_13829": ["dequantize_affine_284"], "view_2235": ["dequantize_affine_285"], "view_2237": ["dequantize_affine_285"], "convert_element_type_856": ["dequantize_affine_285"], "convert_element_type_857": ["dequantize_affine_285"], "sub_6534": ["dequantize_affine_285"], "view_2236": ["dequantize_affine_285"], "mul_13834": ["dequantize_affine_285"], "view_2238": ["dequantize_affine_285"], "mul_13839": ["linear_142"], "view_2239": ["linear_142"], "permute_239": ["linear_142"], "mm_default_41": ["linear_142"], "add_tensor_41": ["linear_142"], "view_2240": ["linear_142"], "mul_13846": ["gelu_25"], "mul_13847": ["gelu_25"], "erf_25": ["gelu_25"], "add_21912": ["gelu_25"], "mul_13848": ["gelu_25", "dropout_71", "choose_qparams_affine_default_143", "quantize_affine_143"], "amax_143": ["choose_qparams_affine_default_143"], "full_287": ["choose_qparams_affine_default_143"], "maximum_143": ["choose_qparams_affine_default_143"], "amin_143": ["choose_qparams_affine_default_143"], "full_286": ["choose_qparams_affine_default_143"], "minimum_143": ["choose_qparams_affine_default_143"], "sub_6547": ["choose_qparams_affine_default_143"], "div_286": ["choose_qparams_affine_default_143"], "clamp_min_429": ["choose_qparams_affine_default_143"], "div_287": ["choose_qparams_affine_default_143"], "round_287": ["choose_qparams_affine_default_143"], "sub_6553": ["choose_qparams_affine_default_143"], "clamp_min_430": ["choose_qparams_affine_default_143"], "clamp_max_286": ["choose_qparams_affine_default_143"], "view_2243": ["quantize_affine_143"], "reciprocal_143": ["quantize_affine_143"], "mul_13894": ["quantize_affine_143"], "mul_13897": ["quantize_affine_143"], "round_288": ["quantize_affine_143"], "convert_element_type_858": ["choose_qparams_affine_default_143"], "view_2244": ["quantize_affine_143"], "add_21995": ["quantize_affine_143"], "clamp_min_431": ["quantize_affine_143"], "clamp_max_287": ["quantize_affine_143"], "convert_element_type_859": ["quantize_affine_143", "dequantize_affine_286"], "view_2248": ["dequantize_affine_286"], "convert_element_type_860": ["dequantize_affine_286"], "convert_element_type_861": ["dequantize_affine_286"], "sub_6573": ["dequantize_affine_286"], "view_2247": ["dequantize_affine_286"], "mul_13919": ["dequantize_affine_286"], "view_2250": ["dequantize_affine_287"], "view_2252": ["dequantize_affine_287"], "convert_element_type_862": ["dequantize_affine_287"], "convert_element_type_863": ["dequantize_affine_287"], "sub_6577": ["dequantize_affine_287"], "view_2251": ["dequantize_affine_287"], "mul_13924": ["dequantize_affine_287"], "view_2253": ["dequantize_affine_287"], "mul_13929": ["linear_143"], "view_2254": ["linear_143"], "permute_240": ["linear_143"], "mm_default_40": ["linear_143"], "add_tensor_40": ["linear_143"], "view_2255": ["linear_143", "dropout_72"], "add_22058": ["add_336"], "clone_193": ["layer_norm_48"], "var_mean_48": ["layer_norm_48"], "getitem_192": ["layer_norm_48"], "getitem_193": ["layer_norm_48"], "sub_6583": ["layer_norm_48"], "add_22063": ["layer_norm_48"], "rsqrt_48": ["layer_norm_48"], "mul_13940": ["layer_norm_48"], "mul_13941": ["layer_norm_48"], "add_22064": ["layer_norm_48", "choose_qparams_affine_default_144", "quantize_affine_144", "choose_qparams_affine_default_145", "quantize_affine_145", "choose_qparams_affine_default_146", "quantize_affine_146"], "amax_144": ["choose_qparams_affine_default_144"], "full_289": ["choose_qparams_affine_default_144"], "maximum_144": ["choose_qparams_affine_default_144"], "amin_144": ["choose_qparams_affine_default_144"], "full_288": ["choose_qparams_affine_default_144"], "minimum_144": ["choose_qparams_affine_default_144"], "sub_6594": ["choose_qparams_affine_default_144"], "div_288": ["choose_qparams_affine_default_144"], "clamp_min_432": ["choose_qparams_affine_default_144"], "div_289": ["choose_qparams_affine_default_144"], "round_289": ["choose_qparams_affine_default_144"], "sub_6600": ["choose_qparams_affine_default_144"], "clamp_min_433": ["choose_qparams_affine_default_144"], "clamp_max_288": ["choose_qparams_affine_default_144"], "view_2258": ["quantize_affine_144"], "reciprocal_144": ["quantize_affine_144"], "mul_13989": ["quantize_affine_144"], "mul_13992": ["quantize_affine_144"], "round_290": ["quantize_affine_144"], "convert_element_type_864": ["choose_qparams_affine_default_144"], "view_2259": ["quantize_affine_144"], "add_22151": ["quantize_affine_144"], "clamp_min_434": ["quantize_affine_144"], "clamp_max_289": ["quantize_affine_144"], "convert_element_type_865": ["quantize_affine_144", "dequantize_affine_288"], "view_2263": ["dequantize_affine_288"], "convert_element_type_866": ["dequantize_affine_288"], "convert_element_type_867": ["dequantize_affine_288"], "sub_6620": ["dequantize_affine_288"], "view_2262": ["dequantize_affine_288"], "mul_14014": ["dequantize_affine_288"], "view_2265": ["dequantize_affine_289"], "view_2267": ["dequantize_affine_289"], "convert_element_type_868": ["dequantize_affine_289"], "convert_element_type_869": ["dequantize_affine_289"], "sub_6624": ["dequantize_affine_289"], "view_2266": ["dequantize_affine_289"], "mul_14019": ["dequantize_affine_289"], "view_2268": ["dequantize_affine_289"], "amax_145": ["choose_qparams_affine_default_145"], "full_291": ["choose_qparams_affine_default_145"], "maximum_145": ["choose_qparams_affine_default_145"], "amin_145": ["choose_qparams_affine_default_145"], "full_290": ["choose_qparams_affine_default_145"], "minimum_145": ["choose_qparams_affine_default_145"], "sub_6639": ["choose_qparams_affine_default_145"], "div_290": ["choose_qparams_affine_default_145"], "clamp_min_435": ["choose_qparams_affine_default_145"], "div_291": ["choose_qparams_affine_default_145"], "round_291": ["choose_qparams_affine_default_145"], "sub_6645": ["choose_qparams_affine_default_145"], "clamp_min_436": ["choose_qparams_affine_default_145"], "clamp_max_290": ["choose_qparams_affine_default_145"], "view_2274": ["quantize_affine_145"], "reciprocal_145": ["quantize_affine_145"], "mul_14085": ["quantize_affine_145"], "mul_14088": ["quantize_affine_145"], "round_292": ["quantize_affine_145"], "convert_element_type_870": ["choose_qparams_affine_default_145"], "view_2275": ["quantize_affine_145"], "add_22303": ["quantize_affine_145"], "clamp_min_437": ["quantize_affine_145"], "clamp_max_291": ["quantize_affine_145"], "convert_element_type_871": ["quantize_affine_145", "dequantize_affine_290"], "view_2279": ["dequantize_affine_290"], "convert_element_type_872": ["dequantize_affine_290"], "convert_element_type_873": ["dequantize_affine_290"], "sub_6665": ["dequantize_affine_290"], "view_2278": ["dequantize_affine_290"], "mul_14110": ["dequantize_affine_290"], "view_2281": ["dequantize_affine_291"], "view_2283": ["dequantize_affine_291"], "convert_element_type_874": ["dequantize_affine_291"], "convert_element_type_875": ["dequantize_affine_291"], "sub_6669": ["dequantize_affine_291"], "view_2282": ["dequantize_affine_291"], "mul_14115": ["dequantize_affine_291"], "view_2284": ["dequantize_affine_291"], "amax_146": ["choose_qparams_affine_default_146"], "full_293": ["choose_qparams_affine_default_146"], "maximum_146": ["choose_qparams_affine_default_146"], "amin_146": ["choose_qparams_affine_default_146"], "full_292": ["choose_qparams_affine_default_146"], "minimum_146": ["choose_qparams_affine_default_146"], "sub_6683": ["choose_qparams_affine_default_146"], "div_292": ["choose_qparams_affine_default_146"], "clamp_min_438": ["choose_qparams_affine_default_146"], "div_293": ["choose_qparams_affine_default_146"], "round_293": ["choose_qparams_affine_default_146"], "sub_6689": ["choose_qparams_affine_default_146"], "clamp_min_439": ["choose_qparams_affine_default_146"], "clamp_max_292": ["choose_qparams_affine_default_146"], "view_2290": ["quantize_affine_146"], "reciprocal_146": ["quantize_affine_146"], "mul_14184": ["quantize_affine_146"], "mul_14187": ["quantize_affine_146"], "round_294": ["quantize_affine_146"], "convert_element_type_876": ["choose_qparams_affine_default_146"], "view_2291": ["quantize_affine_146"], "add_22451": ["quantize_affine_146"], "clamp_min_440": ["quantize_affine_146"], "clamp_max_293": ["quantize_affine_146"], "convert_element_type_877": ["quantize_affine_146", "dequantize_affine_292"], "view_2295": ["dequantize_affine_292"], "convert_element_type_878": ["dequantize_affine_292"], "convert_element_type_879": ["dequantize_affine_292"], "sub_6709": ["dequantize_affine_292"], "view_2294": ["dequantize_affine_292"], "mul_14209": ["dequantize_affine_292"], "view_2297": ["dequantize_affine_293"], "view_2299": ["dequantize_affine_293"], "convert_element_type_880": ["dequantize_affine_293"], "convert_element_type_881": ["dequantize_affine_293"], "sub_6713": ["dequantize_affine_293"], "view_2298": ["dequantize_affine_293"], "mul_14214": ["dequantize_affine_293"], "view_2300": ["dequantize_affine_293"], "mul_14024": ["linear_144"], "view_2269": ["linear_144"], "permute_241": ["linear_144"], "mm_default_39": ["linear_144"], "add_tensor_39": ["linear_144"], "view_2270": ["linear_144"], "mul_14031": ["mul_893"], "view_2271": ["view_72"], "permute_242": ["transpose_96"], "clone_194": ["contiguous_96"], "mul_14118": ["linear_145"], "view_2285": ["linear_145"], "permute_243": ["linear_145"], "mm_24": ["linear_145"], "view_2286": ["linear_145"], "view_2287": ["view_73"], "permute_244": ["transpose_97"], "clone_195": ["contiguous_97"], "mul_14219": ["linear_146"], "view_2301": ["linear_146"], "permute_245": ["linear_146"], "mm_default_38": ["linear_146"], "add_tensor_38": ["linear_146"], "view_2302": ["linear_146"], "view_2303": ["view_74"], "permute_246": ["transpose_98"], "clone_196": ["contiguous_98"], "_scaled_dot_product_efficient_attention_24": ["scaled_dot_product_attention_24"], "getitem_194": ["scaled_dot_product_attention_24"], "permute_247": ["transpose_99"], "view_2304": ["reshape_24", "choose_qparams_affine_default_147", "quantize_affine_147"], "amax_147": ["choose_qparams_affine_default_147"], "full_295": ["choose_qparams_affine_default_147"], "maximum_147": ["choose_qparams_affine_default_147"], "amin_147": ["choose_qparams_affine_default_147"], "full_294": ["choose_qparams_affine_default_147"], "minimum_147": ["choose_qparams_affine_default_147"], "sub_6731": ["choose_qparams_affine_default_147"], "div_294": ["choose_qparams_affine_default_147"], "clamp_min_441": ["choose_qparams_affine_default_147"], "div_295": ["choose_qparams_affine_default_147"], "round_295": ["choose_qparams_affine_default_147"], "sub_6737": ["choose_qparams_affine_default_147"], "clamp_min_442": ["choose_qparams_affine_default_147"], "clamp_max_294": ["choose_qparams_affine_default_147"], "view_2307": ["quantize_affine_147"], "reciprocal_147": ["quantize_affine_147"], "mul_14289": ["quantize_affine_147"], "mul_14292": ["quantize_affine_147"], "round_296": ["quantize_affine_147"], "convert_element_type_882": ["choose_qparams_affine_default_147"], "view_2308": ["quantize_affine_147"], "add_22615": ["quantize_affine_147"], "clamp_min_443": ["quantize_affine_147"], "clamp_max_295": ["quantize_affine_147"], "convert_element_type_883": ["quantize_affine_147", "dequantize_affine_294"], "view_2312": ["dequantize_affine_294"], "convert_element_type_884": ["dequantize_affine_294"], "convert_element_type_885": ["dequantize_affine_294"], "sub_6757": ["dequantize_affine_294"], "view_2311": ["dequantize_affine_294"], "mul_14314": ["dequantize_affine_294"], "view_2314": ["dequantize_affine_295"], "view_2316": ["dequantize_affine_295"], "convert_element_type_886": ["dequantize_affine_295"], "convert_element_type_887": ["dequantize_affine_295"], "sub_6761": ["dequantize_affine_295"], "view_2315": ["dequantize_affine_295"], "mul_14319": ["dequantize_affine_295"], "view_2317": ["dequantize_affine_295"], "mul_14324": ["linear_147"], "view_2318": ["linear_147"], "permute_248": ["linear_147"], "mm_default_37": ["linear_147"], "add_tensor_37": ["linear_147"], "view_2319": ["linear_147", "dropout_73"], "add_22678": ["add_345"], "clone_198": ["layer_norm_49"], "var_mean_49": ["layer_norm_49"], "getitem_198": ["layer_norm_49"], "getitem_199": ["layer_norm_49"], "sub_6767": ["layer_norm_49"], "add_22683": ["layer_norm_49"], "rsqrt_49": ["layer_norm_49"], "mul_14335": ["layer_norm_49"], "mul_14336": ["layer_norm_49"], "add_22684": ["layer_norm_49", "choose_qparams_affine_default_148", "quantize_affine_148"], "amax_148": ["choose_qparams_affine_default_148"], "full_297": ["choose_qparams_affine_default_148"], "maximum_148": ["choose_qparams_affine_default_148"], "amin_148": ["choose_qparams_affine_default_148"], "full_296": ["choose_qparams_affine_default_148"], "minimum_148": ["choose_qparams_affine_default_148"], "sub_6778": ["choose_qparams_affine_default_148"], "div_296": ["choose_qparams_affine_default_148"], "clamp_min_444": ["choose_qparams_affine_default_148"], "div_297": ["choose_qparams_affine_default_148"], "round_297": ["choose_qparams_affine_default_148"], "sub_6784": ["choose_qparams_affine_default_148"], "clamp_min_445": ["choose_qparams_affine_default_148"], "clamp_max_296": ["choose_qparams_affine_default_148"], "view_2322": ["quantize_affine_148"], "reciprocal_148": ["quantize_affine_148"], "mul_14384": ["quantize_affine_148"], "mul_14387": ["quantize_affine_148"], "round_298": ["quantize_affine_148"], "convert_element_type_888": ["choose_qparams_affine_default_148"], "view_2323": ["quantize_affine_148"], "add_22771": ["quantize_affine_148"], "clamp_min_446": ["quantize_affine_148"], "clamp_max_297": ["quantize_affine_148"], "convert_element_type_889": ["quantize_affine_148", "dequantize_affine_296"], "view_2327": ["dequantize_affine_296"], "convert_element_type_890": ["dequantize_affine_296"], "convert_element_type_891": ["dequantize_affine_296"], "sub_6804": ["dequantize_affine_296"], "view_2326": ["dequantize_affine_296"], "mul_14409": ["dequantize_affine_296"], "view_2329": ["dequantize_affine_297"], "view_2331": ["dequantize_affine_297"], "convert_element_type_892": ["dequantize_affine_297"], "convert_element_type_893": ["dequantize_affine_297"], "sub_6808": ["dequantize_affine_297"], "view_2330": ["dequantize_affine_297"], "mul_14414": ["dequantize_affine_297"], "view_2332": ["dequantize_affine_297"], "mul_14419": ["linear_148"], "view_2333": ["linear_148"], "permute_249": ["linear_148"], "mm_default_36": ["linear_148"], "add_tensor_36": ["linear_148"], "view_2334": ["linear_148"], "mul_14426": ["gelu_26"], "mul_14427": ["gelu_26"], "erf_26": ["gelu_26"], "add_22830": ["gelu_26"], "mul_14428": ["gelu_26", "dropout_74", "choose_qparams_affine_default_149", "quantize_affine_149"], "amax_149": ["choose_qparams_affine_default_149"], "full_299": ["choose_qparams_affine_default_149"], "maximum_149": ["choose_qparams_affine_default_149"], "amin_149": ["choose_qparams_affine_default_149"], "full_298": ["choose_qparams_affine_default_149"], "minimum_149": ["choose_qparams_affine_default_149"], "sub_6821": ["choose_qparams_affine_default_149"], "div_298": ["choose_qparams_affine_default_149"], "clamp_min_447": ["choose_qparams_affine_default_149"], "div_299": ["choose_qparams_affine_default_149"], "round_299": ["choose_qparams_affine_default_149"], "sub_6827": ["choose_qparams_affine_default_149"], "clamp_min_448": ["choose_qparams_affine_default_149"], "clamp_max_298": ["choose_qparams_affine_default_149"], "view_2337": ["quantize_affine_149"], "reciprocal_149": ["quantize_affine_149"], "mul_14474": ["quantize_affine_149"], "mul_14477": ["quantize_affine_149"], "round_300": ["quantize_affine_149"], "convert_element_type_894": ["choose_qparams_affine_default_149"], "view_2338": ["quantize_affine_149"], "add_22913": ["quantize_affine_149"], "clamp_min_449": ["quantize_affine_149"], "clamp_max_299": ["quantize_affine_149"], "convert_element_type_895": ["quantize_affine_149", "dequantize_affine_298"], "view_2342": ["dequantize_affine_298"], "convert_element_type_896": ["dequantize_affine_298"], "convert_element_type_897": ["dequantize_affine_298"], "sub_6847": ["dequantize_affine_298"], "view_2341": ["dequantize_affine_298"], "mul_14499": ["dequantize_affine_298"], "view_2344": ["dequantize_affine_299"], "view_2346": ["dequantize_affine_299"], "convert_element_type_898": ["dequantize_affine_299"], "convert_element_type_899": ["dequantize_affine_299"], "sub_6851": ["dequantize_affine_299"], "view_2345": ["dequantize_affine_299"], "mul_14504": ["dequantize_affine_299"], "view_2347": ["dequantize_affine_299"], "mul_14509": ["linear_149"], "view_2348": ["linear_149"], "permute_250": ["linear_149"], "mm_default_35": ["linear_149"], "add_tensor_35": ["linear_149"], "view_2349": ["linear_149", "dropout_75"], "add_22976": ["add_350"], "clone_201": ["layer_norm_50"], "var_mean_50": ["layer_norm_50"], "getitem_200": ["layer_norm_50"], "getitem_201": ["layer_norm_50"], "sub_6857": ["layer_norm_50"], "add_22981": ["layer_norm_50"], "rsqrt_50": ["layer_norm_50"], "mul_14520": ["layer_norm_50"], "mul_14521": ["layer_norm_50"], "add_22982": ["layer_norm_50", "choose_qparams_affine_default_150", "quantize_affine_150", "choose_qparams_affine_default_151", "quantize_affine_151", "choose_qparams_affine_default_152", "quantize_affine_152"], "amax_150": ["choose_qparams_affine_default_150"], "full_301": ["choose_qparams_affine_default_150"], "maximum_150": ["choose_qparams_affine_default_150"], "amin_150": ["choose_qparams_affine_default_150"], "full_300": ["choose_qparams_affine_default_150"], "minimum_150": ["choose_qparams_affine_default_150"], "sub_6868": ["choose_qparams_affine_default_150"], "div_300": ["choose_qparams_affine_default_150"], "clamp_min_450": ["choose_qparams_affine_default_150"], "div_301": ["choose_qparams_affine_default_150"], "round_301": ["choose_qparams_affine_default_150"], "sub_6874": ["choose_qparams_affine_default_150"], "clamp_min_451": ["choose_qparams_affine_default_150"], "clamp_max_300": ["choose_qparams_affine_default_150"], "view_2352": ["quantize_affine_150"], "reciprocal_150": ["quantize_affine_150"], "mul_14569": ["quantize_affine_150"], "mul_14572": ["quantize_affine_150"], "round_302": ["quantize_affine_150"], "convert_element_type_900": ["choose_qparams_affine_default_150"], "view_2353": ["quantize_affine_150"], "add_23069": ["quantize_affine_150"], "clamp_min_452": ["quantize_affine_150"], "clamp_max_301": ["quantize_affine_150"], "convert_element_type_901": ["quantize_affine_150", "dequantize_affine_300"], "view_2357": ["dequantize_affine_300"], "convert_element_type_902": ["dequantize_affine_300"], "convert_element_type_903": ["dequantize_affine_300"], "sub_6894": ["dequantize_affine_300"], "view_2356": ["dequantize_affine_300"], "mul_14594": ["dequantize_affine_300"], "view_2359": ["dequantize_affine_301"], "view_2361": ["dequantize_affine_301"], "convert_element_type_904": ["dequantize_affine_301"], "convert_element_type_905": ["dequantize_affine_301"], "sub_6898": ["dequantize_affine_301"], "view_2360": ["dequantize_affine_301"], "mul_14599": ["dequantize_affine_301"], "view_2362": ["dequantize_affine_301"], "amax_151": ["choose_qparams_affine_default_151"], "full_303": ["choose_qparams_affine_default_151"], "maximum_151": ["choose_qparams_affine_default_151"], "amin_151": ["choose_qparams_affine_default_151"], "full_302": ["choose_qparams_affine_default_151"], "minimum_151": ["choose_qparams_affine_default_151"], "sub_6913": ["choose_qparams_affine_default_151"], "div_302": ["choose_qparams_affine_default_151"], "clamp_min_453": ["choose_qparams_affine_default_151"], "div_303": ["choose_qparams_affine_default_151"], "round_303": ["choose_qparams_affine_default_151"], "sub_6919": ["choose_qparams_affine_default_151"], "clamp_min_454": ["choose_qparams_affine_default_151"], "clamp_max_302": ["choose_qparams_affine_default_151"], "view_2368": ["quantize_affine_151"], "reciprocal_151": ["quantize_affine_151"], "mul_14665": ["quantize_affine_151"], "mul_14668": ["quantize_affine_151"], "round_304": ["quantize_affine_151"], "convert_element_type_906": ["choose_qparams_affine_default_151"], "view_2369": ["quantize_affine_151"], "add_23221": ["quantize_affine_151"], "clamp_min_455": ["quantize_affine_151"], "clamp_max_303": ["quantize_affine_151"], "convert_element_type_907": ["quantize_affine_151", "dequantize_affine_302"], "view_2373": ["dequantize_affine_302"], "convert_element_type_908": ["dequantize_affine_302"], "convert_element_type_909": ["dequantize_affine_302"], "sub_6939": ["dequantize_affine_302"], "view_2372": ["dequantize_affine_302"], "mul_14690": ["dequantize_affine_302"], "view_2375": ["dequantize_affine_303"], "view_2377": ["dequantize_affine_303"], "convert_element_type_910": ["dequantize_affine_303"], "convert_element_type_911": ["dequantize_affine_303"], "sub_6943": ["dequantize_affine_303"], "view_2376": ["dequantize_affine_303"], "mul_14695": ["dequantize_affine_303"], "view_2378": ["dequantize_affine_303"], "amax_152": ["choose_qparams_affine_default_152"], "full_305": ["choose_qparams_affine_default_152"], "maximum_152": ["choose_qparams_affine_default_152"], "amin_152": ["choose_qparams_affine_default_152"], "full_304": ["choose_qparams_affine_default_152"], "minimum_152": ["choose_qparams_affine_default_152"], "sub_6957": ["choose_qparams_affine_default_152"], "div_304": ["choose_qparams_affine_default_152"], "clamp_min_456": ["choose_qparams_affine_default_152"], "div_305": ["choose_qparams_affine_default_152"], "round_305": ["choose_qparams_affine_default_152"], "sub_6963": ["choose_qparams_affine_default_152"], "clamp_min_457": ["choose_qparams_affine_default_152"], "clamp_max_304": ["choose_qparams_affine_default_152"], "view_2384": ["quantize_affine_152"], "reciprocal_152": ["quantize_affine_152"], "mul_14764": ["quantize_affine_152"], "mul_14767": ["quantize_affine_152"], "round_306": ["quantize_affine_152"], "convert_element_type_912": ["choose_qparams_affine_default_152"], "view_2385": ["quantize_affine_152"], "add_23369": ["quantize_affine_152"], "clamp_min_458": ["quantize_affine_152"], "clamp_max_305": ["quantize_affine_152"], "convert_element_type_913": ["quantize_affine_152", "dequantize_affine_304"], "view_2389": ["dequantize_affine_304"], "convert_element_type_914": ["dequantize_affine_304"], "convert_element_type_915": ["dequantize_affine_304"], "sub_6983": ["dequantize_affine_304"], "view_2388": ["dequantize_affine_304"], "mul_14789": ["dequantize_affine_304"], "view_2391": ["dequantize_affine_305"], "view_2393": ["dequantize_affine_305"], "convert_element_type_916": ["dequantize_affine_305"], "convert_element_type_917": ["dequantize_affine_305"], "sub_6987": ["dequantize_affine_305"], "view_2392": ["dequantize_affine_305"], "mul_14794": ["dequantize_affine_305"], "view_2394": ["dequantize_affine_305"], "mul_14604": ["linear_150"], "view_2363": ["linear_150"], "permute_251": ["linear_150"], "mm_default_34": ["linear_150"], "add_tensor_34": ["linear_150"], "view_2364": ["linear_150"], "mul_14611": ["mul_930"], "view_2365": ["view_75"], "permute_252": ["transpose_100"], "clone_202": ["contiguous_100"], "mul_14698": ["linear_151"], "view_2379": ["linear_151"], "permute_253": ["linear_151"], "mm_25": ["linear_151"], "view_2380": ["linear_151"], "view_2381": ["view_76"], "permute_254": ["transpose_101"], "clone_203": ["contiguous_101"], "mul_14799": ["linear_152"], "view_2395": ["linear_152"], "permute_255": ["linear_152"], "mm_default_33": ["linear_152"], "add_tensor_33": ["linear_152"], "view_2396": ["linear_152"], "view_2397": ["view_77"], "permute_256": ["transpose_102"], "clone_204": ["contiguous_102"], "_scaled_dot_product_efficient_attention_25": ["scaled_dot_product_attention_25"], "getitem_202": ["scaled_dot_product_attention_25"], "permute_257": ["transpose_103"], "view_2398": ["reshape_25", "choose_qparams_affine_default_153", "quantize_affine_153"], "amax_153": ["choose_qparams_affine_default_153"], "full_307": ["choose_qparams_affine_default_153"], "maximum_153": ["choose_qparams_affine_default_153"], "amin_153": ["choose_qparams_affine_default_153"], "full_306": ["choose_qparams_affine_default_153"], "minimum_153": ["choose_qparams_affine_default_153"], "sub_7005": ["choose_qparams_affine_default_153"], "div_306": ["choose_qparams_affine_default_153"], "clamp_min_459": ["choose_qparams_affine_default_153"], "div_307": ["choose_qparams_affine_default_153"], "round_307": ["choose_qparams_affine_default_153"], "sub_7011": ["choose_qparams_affine_default_153"], "clamp_min_460": ["choose_qparams_affine_default_153"], "clamp_max_306": ["choose_qparams_affine_default_153"], "view_2401": ["quantize_affine_153"], "reciprocal_153": ["quantize_affine_153"], "mul_14869": ["quantize_affine_153"], "mul_14872": ["quantize_affine_153"], "round_308": ["quantize_affine_153"], "convert_element_type_918": ["choose_qparams_affine_default_153"], "view_2402": ["quantize_affine_153"], "add_23533": ["quantize_affine_153"], "clamp_min_461": ["quantize_affine_153"], "clamp_max_307": ["quantize_affine_153"], "convert_element_type_919": ["quantize_affine_153", "dequantize_affine_306"], "view_2406": ["dequantize_affine_306"], "convert_element_type_920": ["dequantize_affine_306"], "convert_element_type_921": ["dequantize_affine_306"], "sub_7031": ["dequantize_affine_306"], "view_2405": ["dequantize_affine_306"], "mul_14894": ["dequantize_affine_306"], "view_2408": ["dequantize_affine_307"], "view_2410": ["dequantize_affine_307"], "convert_element_type_922": ["dequantize_affine_307"], "convert_element_type_923": ["dequantize_affine_307"], "sub_7035": ["dequantize_affine_307"], "view_2409": ["dequantize_affine_307"], "mul_14899": ["dequantize_affine_307"], "view_2411": ["dequantize_affine_307"], "mul_14904": ["linear_153"], "view_2412": ["linear_153"], "permute_258": ["linear_153"], "mm_default_32": ["linear_153"], "add_tensor_32": ["linear_153"], "view_2413": ["linear_153", "dropout_76"], "add_23596": ["add_359"], "clone_206": ["layer_norm_51"], "var_mean_51": ["layer_norm_51"], "getitem_206": ["layer_norm_51"], "getitem_207": ["layer_norm_51"], "sub_7041": ["layer_norm_51"], "add_23601": ["layer_norm_51"], "rsqrt_51": ["layer_norm_51"], "mul_14915": ["layer_norm_51"], "mul_14916": ["layer_norm_51"], "add_23602": ["layer_norm_51", "choose_qparams_affine_default_154", "quantize_affine_154"], "amax_154": ["choose_qparams_affine_default_154"], "full_309": ["choose_qparams_affine_default_154"], "maximum_154": ["choose_qparams_affine_default_154"], "amin_154": ["choose_qparams_affine_default_154"], "full_308": ["choose_qparams_affine_default_154"], "minimum_154": ["choose_qparams_affine_default_154"], "sub_7052": ["choose_qparams_affine_default_154"], "div_308": ["choose_qparams_affine_default_154"], "clamp_min_462": ["choose_qparams_affine_default_154"], "div_309": ["choose_qparams_affine_default_154"], "round_309": ["choose_qparams_affine_default_154"], "sub_7058": ["choose_qparams_affine_default_154"], "clamp_min_463": ["choose_qparams_affine_default_154"], "clamp_max_308": ["choose_qparams_affine_default_154"], "view_2416": ["quantize_affine_154"], "reciprocal_154": ["quantize_affine_154"], "mul_14964": ["quantize_affine_154"], "mul_14967": ["quantize_affine_154"], "round_310": ["quantize_affine_154"], "convert_element_type_924": ["choose_qparams_affine_default_154"], "view_2417": ["quantize_affine_154"], "add_23689": ["quantize_affine_154"], "clamp_min_464": ["quantize_affine_154"], "clamp_max_309": ["quantize_affine_154"], "convert_element_type_925": ["quantize_affine_154", "dequantize_affine_308"], "view_2421": ["dequantize_affine_308"], "convert_element_type_926": ["dequantize_affine_308"], "convert_element_type_927": ["dequantize_affine_308"], "sub_7078": ["dequantize_affine_308"], "view_2420": ["dequantize_affine_308"], "mul_14989": ["dequantize_affine_308"], "view_2423": ["dequantize_affine_309"], "view_2425": ["dequantize_affine_309"], "convert_element_type_928": ["dequantize_affine_309"], "convert_element_type_929": ["dequantize_affine_309"], "sub_7082": ["dequantize_affine_309"], "view_2424": ["dequantize_affine_309"], "mul_14994": ["dequantize_affine_309"], "view_2426": ["dequantize_affine_309"], "mul_14999": ["linear_154"], "view_2427": ["linear_154"], "permute_259": ["linear_154"], "mm_default_31": ["linear_154"], "add_tensor_31": ["linear_154"], "view_2428": ["linear_154"], "mul_15006": ["gelu_27"], "mul_15007": ["gelu_27"], "erf_27": ["gelu_27"], "add_23748": ["gelu_27"], "mul_15008": ["gelu_27", "dropout_77", "choose_qparams_affine_default_155", "quantize_affine_155"], "amax_155": ["choose_qparams_affine_default_155"], "full_311": ["choose_qparams_affine_default_155"], "maximum_155": ["choose_qparams_affine_default_155"], "amin_155": ["choose_qparams_affine_default_155"], "full_310": ["choose_qparams_affine_default_155"], "minimum_155": ["choose_qparams_affine_default_155"], "sub_7095": ["choose_qparams_affine_default_155"], "div_310": ["choose_qparams_affine_default_155"], "clamp_min_465": ["choose_qparams_affine_default_155"], "div_311": ["choose_qparams_affine_default_155"], "round_311": ["choose_qparams_affine_default_155"], "sub_7101": ["choose_qparams_affine_default_155"], "clamp_min_466": ["choose_qparams_affine_default_155"], "clamp_max_310": ["choose_qparams_affine_default_155"], "view_2431": ["quantize_affine_155"], "reciprocal_155": ["quantize_affine_155"], "mul_15054": ["quantize_affine_155"], "mul_15057": ["quantize_affine_155"], "round_312": ["quantize_affine_155"], "convert_element_type_930": ["choose_qparams_affine_default_155"], "view_2432": ["quantize_affine_155"], "add_23831": ["quantize_affine_155"], "clamp_min_467": ["quantize_affine_155"], "clamp_max_311": ["quantize_affine_155"], "convert_element_type_931": ["quantize_affine_155", "dequantize_affine_310"], "view_2436": ["dequantize_affine_310"], "convert_element_type_932": ["dequantize_affine_310"], "convert_element_type_933": ["dequantize_affine_310"], "sub_7121": ["dequantize_affine_310"], "view_2435": ["dequantize_affine_310"], "mul_15079": ["dequantize_affine_310"], "view_2438": ["dequantize_affine_311"], "view_2440": ["dequantize_affine_311"], "convert_element_type_934": ["dequantize_affine_311"], "convert_element_type_935": ["dequantize_affine_311"], "sub_7125": ["dequantize_affine_311"], "view_2439": ["dequantize_affine_311"], "mul_15084": ["dequantize_affine_311"], "view_2441": ["dequantize_affine_311"], "mul_15089": ["linear_155"], "view_2442": ["linear_155"], "permute_260": ["linear_155"], "mm_default_30": ["linear_155"], "add_tensor_30": ["linear_155"], "view_2443": ["linear_155", "dropout_78"], "add_23894": ["add_364"], "clone_209": ["layer_norm_52"], "var_mean_52": ["layer_norm_52"], "getitem_208": ["layer_norm_52"], "getitem_209": ["layer_norm_52"], "sub_7131": ["layer_norm_52"], "add_23899": ["layer_norm_52"], "rsqrt_52": ["layer_norm_52"], "mul_15100": ["layer_norm_52"], "mul_15101": ["layer_norm_52"], "add_23900": ["layer_norm_52", "choose_qparams_affine_default_156", "quantize_affine_156", "choose_qparams_affine_default_157", "quantize_affine_157", "choose_qparams_affine_default_158", "quantize_affine_158"], "amax_156": ["choose_qparams_affine_default_156"], "full_313": ["choose_qparams_affine_default_156"], "maximum_156": ["choose_qparams_affine_default_156"], "amin_156": ["choose_qparams_affine_default_156"], "full_312": ["choose_qparams_affine_default_156"], "minimum_156": ["choose_qparams_affine_default_156"], "sub_7142": ["choose_qparams_affine_default_156"], "div_312": ["choose_qparams_affine_default_156"], "clamp_min_468": ["choose_qparams_affine_default_156"], "div_313": ["choose_qparams_affine_default_156"], "round_313": ["choose_qparams_affine_default_156"], "sub_7148": ["choose_qparams_affine_default_156"], "clamp_min_469": ["choose_qparams_affine_default_156"], "clamp_max_312": ["choose_qparams_affine_default_156"], "view_2446": ["quantize_affine_156"], "reciprocal_156": ["quantize_affine_156"], "mul_15149": ["quantize_affine_156"], "mul_15152": ["quantize_affine_156"], "round_314": ["quantize_affine_156"], "convert_element_type_936": ["choose_qparams_affine_default_156"], "view_2447": ["quantize_affine_156"], "add_23987": ["quantize_affine_156"], "clamp_min_470": ["quantize_affine_156"], "clamp_max_313": ["quantize_affine_156"], "convert_element_type_937": ["quantize_affine_156", "dequantize_affine_312"], "view_2451": ["dequantize_affine_312"], "convert_element_type_938": ["dequantize_affine_312"], "convert_element_type_939": ["dequantize_affine_312"], "sub_7168": ["dequantize_affine_312"], "view_2450": ["dequantize_affine_312"], "mul_15174": ["dequantize_affine_312"], "view_2453": ["dequantize_affine_313"], "view_2455": ["dequantize_affine_313"], "convert_element_type_940": ["dequantize_affine_313"], "convert_element_type_941": ["dequantize_affine_313"], "sub_7172": ["dequantize_affine_313"], "view_2454": ["dequantize_affine_313"], "mul_15179": ["dequantize_affine_313"], "view_2456": ["dequantize_affine_313"], "amax_157": ["choose_qparams_affine_default_157"], "full_315": ["choose_qparams_affine_default_157"], "maximum_157": ["choose_qparams_affine_default_157"], "amin_157": ["choose_qparams_affine_default_157"], "full_314": ["choose_qparams_affine_default_157"], "minimum_157": ["choose_qparams_affine_default_157"], "sub_7187": ["choose_qparams_affine_default_157"], "div_314": ["choose_qparams_affine_default_157"], "clamp_min_471": ["choose_qparams_affine_default_157"], "div_315": ["choose_qparams_affine_default_157"], "round_315": ["choose_qparams_affine_default_157"], "sub_7193": ["choose_qparams_affine_default_157"], "clamp_min_472": ["choose_qparams_affine_default_157"], "clamp_max_314": ["choose_qparams_affine_default_157"], "view_2462": ["quantize_affine_157"], "reciprocal_157": ["quantize_affine_157"], "mul_15245": ["quantize_affine_157"], "mul_15248": ["quantize_affine_157"], "round_316": ["quantize_affine_157"], "convert_element_type_942": ["choose_qparams_affine_default_157"], "view_2463": ["quantize_affine_157"], "add_24139": ["quantize_affine_157"], "clamp_min_473": ["quantize_affine_157"], "clamp_max_315": ["quantize_affine_157"], "convert_element_type_943": ["quantize_affine_157", "dequantize_affine_314"], "view_2467": ["dequantize_affine_314"], "convert_element_type_944": ["dequantize_affine_314"], "convert_element_type_945": ["dequantize_affine_314"], "sub_7213": ["dequantize_affine_314"], "view_2466": ["dequantize_affine_314"], "mul_15270": ["dequantize_affine_314"], "view_2469": ["dequantize_affine_315"], "view_2471": ["dequantize_affine_315"], "convert_element_type_946": ["dequantize_affine_315"], "convert_element_type_947": ["dequantize_affine_315"], "sub_7217": ["dequantize_affine_315"], "view_2470": ["dequantize_affine_315"], "mul_15275": ["dequantize_affine_315"], "view_2472": ["dequantize_affine_315"], "amax_158": ["choose_qparams_affine_default_158"], "full_317": ["choose_qparams_affine_default_158"], "maximum_158": ["choose_qparams_affine_default_158"], "amin_158": ["choose_qparams_affine_default_158"], "full_316": ["choose_qparams_affine_default_158"], "minimum_158": ["choose_qparams_affine_default_158"], "sub_7231": ["choose_qparams_affine_default_158"], "div_316": ["choose_qparams_affine_default_158"], "clamp_min_474": ["choose_qparams_affine_default_158"], "div_317": ["choose_qparams_affine_default_158"], "round_317": ["choose_qparams_affine_default_158"], "sub_7237": ["choose_qparams_affine_default_158"], "clamp_min_475": ["choose_qparams_affine_default_158"], "clamp_max_316": ["choose_qparams_affine_default_158"], "view_2478": ["quantize_affine_158"], "reciprocal_158": ["quantize_affine_158"], "mul_15344": ["quantize_affine_158"], "mul_15347": ["quantize_affine_158"], "round_318": ["quantize_affine_158"], "convert_element_type_948": ["choose_qparams_affine_default_158"], "view_2479": ["quantize_affine_158"], "add_24287": ["quantize_affine_158"], "clamp_min_476": ["quantize_affine_158"], "clamp_max_317": ["quantize_affine_158"], "convert_element_type_949": ["quantize_affine_158", "dequantize_affine_316"], "view_2483": ["dequantize_affine_316"], "convert_element_type_950": ["dequantize_affine_316"], "convert_element_type_951": ["dequantize_affine_316"], "sub_7257": ["dequantize_affine_316"], "view_2482": ["dequantize_affine_316"], "mul_15369": ["dequantize_affine_316"], "view_2485": ["dequantize_affine_317"], "view_2487": ["dequantize_affine_317"], "convert_element_type_952": ["dequantize_affine_317"], "convert_element_type_953": ["dequantize_affine_317"], "sub_7261": ["dequantize_affine_317"], "view_2486": ["dequantize_affine_317"], "mul_15374": ["dequantize_affine_317"], "view_2488": ["dequantize_affine_317"], "mul_15184": ["linear_156"], "view_2457": ["linear_156"], "permute_261": ["linear_156"], "mm_default_29": ["linear_156"], "add_tensor_29": ["linear_156"], "view_2458": ["linear_156"], "mul_15191": ["mul_967"], "view_2459": ["view_78"], "permute_262": ["transpose_104"], "clone_210": ["contiguous_104"], "mul_15278": ["linear_157"], "view_2473": ["linear_157"], "permute_263": ["linear_157"], "mm_26": ["linear_157"], "view_2474": ["linear_157"], "view_2475": ["view_79"], "permute_264": ["transpose_105"], "clone_211": ["contiguous_105"], "mul_15379": ["linear_158"], "view_2489": ["linear_158"], "permute_265": ["linear_158"], "mm_default_28": ["linear_158"], "add_tensor_28": ["linear_158"], "view_2490": ["linear_158"], "view_2491": ["view_80"], "permute_266": ["transpose_106"], "clone_212": ["contiguous_106"], "_scaled_dot_product_efficient_attention_26": ["scaled_dot_product_attention_26"], "getitem_210": ["scaled_dot_product_attention_26"], "permute_267": ["transpose_107"], "view_2492": ["reshape_26", "choose_qparams_affine_default_159", "quantize_affine_159"], "amax_159": ["choose_qparams_affine_default_159"], "full_319": ["choose_qparams_affine_default_159"], "maximum_159": ["choose_qparams_affine_default_159"], "amin_159": ["choose_qparams_affine_default_159"], "full_318": ["choose_qparams_affine_default_159"], "minimum_159": ["choose_qparams_affine_default_159"], "sub_7279": ["choose_qparams_affine_default_159"], "div_318": ["choose_qparams_affine_default_159"], "clamp_min_477": ["choose_qparams_affine_default_159"], "div_319": ["choose_qparams_affine_default_159"], "round_319": ["choose_qparams_affine_default_159"], "sub_7285": ["choose_qparams_affine_default_159"], "clamp_min_478": ["choose_qparams_affine_default_159"], "clamp_max_318": ["choose_qparams_affine_default_159"], "view_2495": ["quantize_affine_159"], "reciprocal_159": ["quantize_affine_159"], "mul_15449": ["quantize_affine_159"], "mul_15452": ["quantize_affine_159"], "round_320": ["quantize_affine_159"], "convert_element_type_954": ["choose_qparams_affine_default_159"], "view_2496": ["quantize_affine_159"], "add_24451": ["quantize_affine_159"], "clamp_min_479": ["quantize_affine_159"], "clamp_max_319": ["quantize_affine_159"], "convert_element_type_955": ["quantize_affine_159", "dequantize_affine_318"], "view_2500": ["dequantize_affine_318"], "convert_element_type_956": ["dequantize_affine_318"], "convert_element_type_957": ["dequantize_affine_318"], "sub_7305": ["dequantize_affine_318"], "view_2499": ["dequantize_affine_318"], "mul_15474": ["dequantize_affine_318"], "view_2502": ["dequantize_affine_319"], "view_2504": ["dequantize_affine_319"], "convert_element_type_958": ["dequantize_affine_319"], "convert_element_type_959": ["dequantize_affine_319"], "sub_7309": ["dequantize_affine_319"], "view_2503": ["dequantize_affine_319"], "mul_15479": ["dequantize_affine_319"], "view_2505": ["dequantize_affine_319"], "mul_15484": ["linear_159"], "view_2506": ["linear_159"], "permute_268": ["linear_159"], "mm_default_27": ["linear_159"], "add_tensor_27": ["linear_159"], "view_2507": ["linear_159", "dropout_79"], "add_24514": ["add_373"], "clone_214": ["layer_norm_53"], "var_mean_53": ["layer_norm_53"], "getitem_214": ["layer_norm_53"], "getitem_215": ["layer_norm_53"], "sub_7315": ["layer_norm_53"], "add_24519": ["layer_norm_53"], "rsqrt_53": ["layer_norm_53"], "mul_15495": ["layer_norm_53"], "mul_15496": ["layer_norm_53"], "add_24520": ["layer_norm_53", "choose_qparams_affine_default_160", "quantize_affine_160"], "amax_160": ["choose_qparams_affine_default_160"], "full_321": ["choose_qparams_affine_default_160"], "maximum_160": ["choose_qparams_affine_default_160"], "amin_160": ["choose_qparams_affine_default_160"], "full_320": ["choose_qparams_affine_default_160"], "minimum_160": ["choose_qparams_affine_default_160"], "sub_7326": ["choose_qparams_affine_default_160"], "div_320": ["choose_qparams_affine_default_160"], "clamp_min_480": ["choose_qparams_affine_default_160"], "div_321": ["choose_qparams_affine_default_160"], "round_321": ["choose_qparams_affine_default_160"], "sub_7332": ["choose_qparams_affine_default_160"], "clamp_min_481": ["choose_qparams_affine_default_160"], "clamp_max_320": ["choose_qparams_affine_default_160"], "view_2510": ["quantize_affine_160"], "reciprocal_160": ["quantize_affine_160"], "mul_15544": ["quantize_affine_160"], "mul_15547": ["quantize_affine_160"], "round_322": ["quantize_affine_160"], "convert_element_type_960": ["choose_qparams_affine_default_160"], "view_2511": ["quantize_affine_160"], "add_24607": ["quantize_affine_160"], "clamp_min_482": ["quantize_affine_160"], "clamp_max_321": ["quantize_affine_160"], "convert_element_type_961": ["quantize_affine_160", "dequantize_affine_320"], "view_2515": ["dequantize_affine_320"], "convert_element_type_962": ["dequantize_affine_320"], "convert_element_type_963": ["dequantize_affine_320"], "sub_7352": ["dequantize_affine_320"], "view_2514": ["dequantize_affine_320"], "mul_15569": ["dequantize_affine_320"], "view_2517": ["dequantize_affine_321"], "view_2519": ["dequantize_affine_321"], "convert_element_type_964": ["dequantize_affine_321"], "convert_element_type_965": ["dequantize_affine_321"], "sub_7356": ["dequantize_affine_321"], "view_2518": ["dequantize_affine_321"], "mul_15574": ["dequantize_affine_321"], "view_2520": ["dequantize_affine_321"], "mul_15579": ["linear_160"], "view_2521": ["linear_160"], "permute_269": ["linear_160"], "mm_default_26": ["linear_160"], "add_tensor_26": ["linear_160"], "view_2522": ["linear_160"], "mul_15586": ["gelu_28"], "mul_15587": ["gelu_28"], "erf_28": ["gelu_28"], "add_24666": ["gelu_28"], "mul_15588": ["gelu_28", "dropout_80", "choose_qparams_affine_default_161", "quantize_affine_161"], "amax_161": ["choose_qparams_affine_default_161"], "full_323": ["choose_qparams_affine_default_161"], "maximum_161": ["choose_qparams_affine_default_161"], "amin_161": ["choose_qparams_affine_default_161"], "full_322": ["choose_qparams_affine_default_161"], "minimum_161": ["choose_qparams_affine_default_161"], "sub_7369": ["choose_qparams_affine_default_161"], "div_322": ["choose_qparams_affine_default_161"], "clamp_min_483": ["choose_qparams_affine_default_161"], "div_323": ["choose_qparams_affine_default_161"], "round_323": ["choose_qparams_affine_default_161"], "sub_7375": ["choose_qparams_affine_default_161"], "clamp_min_484": ["choose_qparams_affine_default_161"], "clamp_max_322": ["choose_qparams_affine_default_161"], "view_2525": ["quantize_affine_161"], "reciprocal_161": ["quantize_affine_161"], "mul_15634": ["quantize_affine_161"], "mul_15637": ["quantize_affine_161"], "round_324": ["quantize_affine_161"], "convert_element_type_966": ["choose_qparams_affine_default_161"], "view_2526": ["quantize_affine_161"], "add_24749": ["quantize_affine_161"], "clamp_min_485": ["quantize_affine_161"], "clamp_max_323": ["quantize_affine_161"], "convert_element_type_967": ["quantize_affine_161", "dequantize_affine_322"], "view_2530": ["dequantize_affine_322"], "convert_element_type_968": ["dequantize_affine_322"], "convert_element_type_969": ["dequantize_affine_322"], "sub_7395": ["dequantize_affine_322"], "view_2529": ["dequantize_affine_322"], "mul_15659": ["dequantize_affine_322"], "view_2532": ["dequantize_affine_323"], "view_2534": ["dequantize_affine_323"], "convert_element_type_970": ["dequantize_affine_323"], "convert_element_type_971": ["dequantize_affine_323"], "sub_7399": ["dequantize_affine_323"], "view_2533": ["dequantize_affine_323"], "mul_15664": ["dequantize_affine_323"], "view_2535": ["dequantize_affine_323"], "mul_15669": ["linear_161"], "view_2536": ["linear_161"], "permute_270": ["linear_161"], "mm_default_25": ["linear_161"], "add_tensor_25": ["linear_161"], "view_2537": ["linear_161", "dropout_81"], "add_24812": ["add_378"], "clone_217": ["layer_norm_54"], "var_mean_54": ["layer_norm_54"], "getitem_216": ["layer_norm_54"], "getitem_217": ["layer_norm_54"], "sub_7405": ["layer_norm_54"], "add_24817": ["layer_norm_54"], "rsqrt_54": ["layer_norm_54"], "mul_15680": ["layer_norm_54"], "mul_15681": ["layer_norm_54"], "add_24818": ["layer_norm_54", "choose_qparams_affine_default_162", "quantize_affine_162", "choose_qparams_affine_default_163", "quantize_affine_163", "choose_qparams_affine_default_164", "quantize_affine_164"], "amax_162": ["choose_qparams_affine_default_162"], "full_325": ["choose_qparams_affine_default_162"], "maximum_162": ["choose_qparams_affine_default_162"], "amin_162": ["choose_qparams_affine_default_162"], "full_324": ["choose_qparams_affine_default_162"], "minimum_162": ["choose_qparams_affine_default_162"], "sub_7416": ["choose_qparams_affine_default_162"], "div_324": ["choose_qparams_affine_default_162"], "clamp_min_486": ["choose_qparams_affine_default_162"], "div_325": ["choose_qparams_affine_default_162"], "round_325": ["choose_qparams_affine_default_162"], "sub_7422": ["choose_qparams_affine_default_162"], "clamp_min_487": ["choose_qparams_affine_default_162"], "clamp_max_324": ["choose_qparams_affine_default_162"], "view_2540": ["quantize_affine_162"], "reciprocal_162": ["quantize_affine_162"], "mul_15729": ["quantize_affine_162"], "mul_15732": ["quantize_affine_162"], "round_326": ["quantize_affine_162"], "convert_element_type_972": ["choose_qparams_affine_default_162"], "view_2541": ["quantize_affine_162"], "add_24905": ["quantize_affine_162"], "clamp_min_488": ["quantize_affine_162"], "clamp_max_325": ["quantize_affine_162"], "convert_element_type_973": ["quantize_affine_162", "dequantize_affine_324"], "view_2545": ["dequantize_affine_324"], "convert_element_type_974": ["dequantize_affine_324"], "convert_element_type_975": ["dequantize_affine_324"], "sub_7442": ["dequantize_affine_324"], "view_2544": ["dequantize_affine_324"], "mul_15754": ["dequantize_affine_324"], "view_2547": ["dequantize_affine_325"], "view_2549": ["dequantize_affine_325"], "convert_element_type_976": ["dequantize_affine_325"], "convert_element_type_977": ["dequantize_affine_325"], "sub_7446": ["dequantize_affine_325"], "view_2548": ["dequantize_affine_325"], "mul_15759": ["dequantize_affine_325"], "view_2550": ["dequantize_affine_325"], "amax_163": ["choose_qparams_affine_default_163"], "full_327": ["choose_qparams_affine_default_163"], "maximum_163": ["choose_qparams_affine_default_163"], "amin_163": ["choose_qparams_affine_default_163"], "full_326": ["choose_qparams_affine_default_163"], "minimum_163": ["choose_qparams_affine_default_163"], "sub_7461": ["choose_qparams_affine_default_163"], "div_326": ["choose_qparams_affine_default_163"], "clamp_min_489": ["choose_qparams_affine_default_163"], "div_327": ["choose_qparams_affine_default_163"], "round_327": ["choose_qparams_affine_default_163"], "sub_7467": ["choose_qparams_affine_default_163"], "clamp_min_490": ["choose_qparams_affine_default_163"], "clamp_max_326": ["choose_qparams_affine_default_163"], "view_2556": ["quantize_affine_163"], "reciprocal_163": ["quantize_affine_163"], "mul_15825": ["quantize_affine_163"], "mul_15828": ["quantize_affine_163"], "round_328": ["quantize_affine_163"], "convert_element_type_978": ["choose_qparams_affine_default_163"], "view_2557": ["quantize_affine_163"], "add_25057": ["quantize_affine_163"], "clamp_min_491": ["quantize_affine_163"], "clamp_max_327": ["quantize_affine_163"], "convert_element_type_979": ["quantize_affine_163", "dequantize_affine_326"], "view_2561": ["dequantize_affine_326"], "convert_element_type_980": ["dequantize_affine_326"], "convert_element_type_981": ["dequantize_affine_326"], "sub_7487": ["dequantize_affine_326"], "view_2560": ["dequantize_affine_326"], "mul_15850": ["dequantize_affine_326"], "view_2563": ["dequantize_affine_327"], "view_2565": ["dequantize_affine_327"], "convert_element_type_982": ["dequantize_affine_327"], "convert_element_type_983": ["dequantize_affine_327"], "sub_7491": ["dequantize_affine_327"], "view_2564": ["dequantize_affine_327"], "mul_15855": ["dequantize_affine_327"], "view_2566": ["dequantize_affine_327"], "amax_164": ["choose_qparams_affine_default_164"], "full_329": ["choose_qparams_affine_default_164"], "maximum_164": ["choose_qparams_affine_default_164"], "amin_164": ["choose_qparams_affine_default_164"], "full_328": ["choose_qparams_affine_default_164"], "minimum_164": ["choose_qparams_affine_default_164"], "sub_7505": ["choose_qparams_affine_default_164"], "div_328": ["choose_qparams_affine_default_164"], "clamp_min_492": ["choose_qparams_affine_default_164"], "div_329": ["choose_qparams_affine_default_164"], "round_329": ["choose_qparams_affine_default_164"], "sub_7511": ["choose_qparams_affine_default_164"], "clamp_min_493": ["choose_qparams_affine_default_164"], "clamp_max_328": ["choose_qparams_affine_default_164"], "view_2572": ["quantize_affine_164"], "reciprocal_164": ["quantize_affine_164"], "mul_15924": ["quantize_affine_164"], "mul_15927": ["quantize_affine_164"], "round_330": ["quantize_affine_164"], "convert_element_type_984": ["choose_qparams_affine_default_164"], "view_2573": ["quantize_affine_164"], "add_25205": ["quantize_affine_164"], "clamp_min_494": ["quantize_affine_164"], "clamp_max_329": ["quantize_affine_164"], "convert_element_type_985": ["quantize_affine_164", "dequantize_affine_328"], "view_2577": ["dequantize_affine_328"], "convert_element_type_986": ["dequantize_affine_328"], "convert_element_type_987": ["dequantize_affine_328"], "sub_7531": ["dequantize_affine_328"], "view_2576": ["dequantize_affine_328"], "mul_15949": ["dequantize_affine_328"], "view_2579": ["dequantize_affine_329"], "view_2581": ["dequantize_affine_329"], "convert_element_type_988": ["dequantize_affine_329"], "convert_element_type_989": ["dequantize_affine_329"], "sub_7535": ["dequantize_affine_329"], "view_2580": ["dequantize_affine_329"], "mul_15954": ["dequantize_affine_329"], "view_2582": ["dequantize_affine_329"], "mul_15764": ["linear_162"], "view_2551": ["linear_162"], "permute_271": ["linear_162"], "mm_default_24": ["linear_162"], "add_tensor_24": ["linear_162"], "view_2552": ["linear_162"], "mul_15771": ["mul_1004"], "view_2553": ["view_81"], "permute_272": ["transpose_108"], "clone_218": ["contiguous_108"], "mul_15858": ["linear_163"], "view_2567": ["linear_163"], "permute_273": ["linear_163"], "mm_27": ["linear_163"], "view_2568": ["linear_163"], "view_2569": ["view_82"], "permute_274": ["transpose_109"], "clone_219": ["contiguous_109"], "mul_15959": ["linear_164"], "view_2583": ["linear_164"], "permute_275": ["linear_164"], "mm_default_23": ["linear_164"], "add_tensor_23": ["linear_164"], "view_2584": ["linear_164"], "view_2585": ["view_83"], "permute_276": ["transpose_110"], "clone_220": ["contiguous_110"], "_scaled_dot_product_efficient_attention_27": ["scaled_dot_product_attention_27"], "getitem_218": ["scaled_dot_product_attention_27"], "permute_277": ["transpose_111"], "view_2586": ["reshape_27", "choose_qparams_affine_default_165", "quantize_affine_165"], "amax_165": ["choose_qparams_affine_default_165"], "full_331": ["choose_qparams_affine_default_165"], "maximum_165": ["choose_qparams_affine_default_165"], "amin_165": ["choose_qparams_affine_default_165"], "full_330": ["choose_qparams_affine_default_165"], "minimum_165": ["choose_qparams_affine_default_165"], "sub_7553": ["choose_qparams_affine_default_165"], "div_330": ["choose_qparams_affine_default_165"], "clamp_min_495": ["choose_qparams_affine_default_165"], "div_331": ["choose_qparams_affine_default_165"], "round_331": ["choose_qparams_affine_default_165"], "sub_7559": ["choose_qparams_affine_default_165"], "clamp_min_496": ["choose_qparams_affine_default_165"], "clamp_max_330": ["choose_qparams_affine_default_165"], "view_2589": ["quantize_affine_165"], "reciprocal_165": ["quantize_affine_165"], "mul_16029": ["quantize_affine_165"], "mul_16032": ["quantize_affine_165"], "round_332": ["quantize_affine_165"], "convert_element_type_990": ["choose_qparams_affine_default_165"], "view_2590": ["quantize_affine_165"], "add_25369": ["quantize_affine_165"], "clamp_min_497": ["quantize_affine_165"], "clamp_max_331": ["quantize_affine_165"], "convert_element_type_991": ["quantize_affine_165", "dequantize_affine_330"], "view_2594": ["dequantize_affine_330"], "convert_element_type_992": ["dequantize_affine_330"], "convert_element_type_993": ["dequantize_affine_330"], "sub_7579": ["dequantize_affine_330"], "view_2593": ["dequantize_affine_330"], "mul_16054": ["dequantize_affine_330"], "view_2596": ["dequantize_affine_331"], "view_2598": ["dequantize_affine_331"], "convert_element_type_994": ["dequantize_affine_331"], "convert_element_type_995": ["dequantize_affine_331"], "sub_7583": ["dequantize_affine_331"], "view_2597": ["dequantize_affine_331"], "mul_16059": ["dequantize_affine_331"], "view_2599": ["dequantize_affine_331"], "mul_16064": ["linear_165"], "view_2600": ["linear_165"], "permute_278": ["linear_165"], "mm_default_22": ["linear_165"], "add_tensor_22": ["linear_165"], "view_2601": ["linear_165", "dropout_82"], "add_25432": ["add_387"], "clone_222": ["layer_norm_55"], "var_mean_55": ["layer_norm_55"], "getitem_222": ["layer_norm_55"], "getitem_223": ["layer_norm_55"], "sub_7589": ["layer_norm_55"], "add_25437": ["layer_norm_55"], "rsqrt_55": ["layer_norm_55"], "mul_16075": ["layer_norm_55"], "mul_16076": ["layer_norm_55"], "add_25438": ["layer_norm_55", "choose_qparams_affine_default_166", "quantize_affine_166"], "amax_166": ["choose_qparams_affine_default_166"], "full_333": ["choose_qparams_affine_default_166"], "maximum_166": ["choose_qparams_affine_default_166"], "amin_166": ["choose_qparams_affine_default_166"], "full_332": ["choose_qparams_affine_default_166"], "minimum_166": ["choose_qparams_affine_default_166"], "sub_7600": ["choose_qparams_affine_default_166"], "div_332": ["choose_qparams_affine_default_166"], "clamp_min_498": ["choose_qparams_affine_default_166"], "div_333": ["choose_qparams_affine_default_166"], "round_333": ["choose_qparams_affine_default_166"], "sub_7606": ["choose_qparams_affine_default_166"], "clamp_min_499": ["choose_qparams_affine_default_166"], "clamp_max_332": ["choose_qparams_affine_default_166"], "view_2604": ["quantize_affine_166"], "reciprocal_166": ["quantize_affine_166"], "mul_16124": ["quantize_affine_166"], "mul_16127": ["quantize_affine_166"], "round_334": ["quantize_affine_166"], "convert_element_type_996": ["choose_qparams_affine_default_166"], "view_2605": ["quantize_affine_166"], "add_25525": ["quantize_affine_166"], "clamp_min_500": ["quantize_affine_166"], "clamp_max_333": ["quantize_affine_166"], "convert_element_type_997": ["quantize_affine_166", "dequantize_affine_332"], "view_2609": ["dequantize_affine_332"], "convert_element_type_998": ["dequantize_affine_332"], "convert_element_type_999": ["dequantize_affine_332"], "sub_7626": ["dequantize_affine_332"], "view_2608": ["dequantize_affine_332"], "mul_16149": ["dequantize_affine_332"], "view_2611": ["dequantize_affine_333"], "view_2613": ["dequantize_affine_333"], "convert_element_type_1000": ["dequantize_affine_333"], "convert_element_type_1001": ["dequantize_affine_333"], "sub_7630": ["dequantize_affine_333"], "view_2612": ["dequantize_affine_333"], "mul_16154": ["dequantize_affine_333"], "view_2614": ["dequantize_affine_333"], "mul_16159": ["linear_166"], "view_2615": ["linear_166"], "permute_279": ["linear_166"], "mm_default_21": ["linear_166"], "add_tensor_21": ["linear_166"], "view_2616": ["linear_166"], "mul_16166": ["gelu_29"], "mul_16167": ["gelu_29"], "erf_29": ["gelu_29"], "add_25584": ["gelu_29"], "mul_16168": ["gelu_29", "dropout_83", "choose_qparams_affine_default_167", "quantize_affine_167"], "amax_167": ["choose_qparams_affine_default_167"], "full_335": ["choose_qparams_affine_default_167"], "maximum_167": ["choose_qparams_affine_default_167"], "amin_167": ["choose_qparams_affine_default_167"], "full_334": ["choose_qparams_affine_default_167"], "minimum_167": ["choose_qparams_affine_default_167"], "sub_7643": ["choose_qparams_affine_default_167"], "div_334": ["choose_qparams_affine_default_167"], "clamp_min_501": ["choose_qparams_affine_default_167"], "div_335": ["choose_qparams_affine_default_167"], "round_335": ["choose_qparams_affine_default_167"], "sub_7649": ["choose_qparams_affine_default_167"], "clamp_min_502": ["choose_qparams_affine_default_167"], "clamp_max_334": ["choose_qparams_affine_default_167"], "view_2619": ["quantize_affine_167"], "reciprocal_167": ["quantize_affine_167"], "mul_16214": ["quantize_affine_167"], "mul_16217": ["quantize_affine_167"], "round_336": ["quantize_affine_167"], "convert_element_type_1002": ["choose_qparams_affine_default_167"], "view_2620": ["quantize_affine_167"], "add_25667": ["quantize_affine_167"], "clamp_min_503": ["quantize_affine_167"], "clamp_max_335": ["quantize_affine_167"], "convert_element_type_1003": ["quantize_affine_167", "dequantize_affine_334"], "view_2624": ["dequantize_affine_334"], "convert_element_type_1004": ["dequantize_affine_334"], "convert_element_type_1005": ["dequantize_affine_334"], "sub_7669": ["dequantize_affine_334"], "view_2623": ["dequantize_affine_334"], "mul_16239": ["dequantize_affine_334"], "view_2626": ["dequantize_affine_335"], "view_2628": ["dequantize_affine_335"], "convert_element_type_1006": ["dequantize_affine_335"], "convert_element_type_1007": ["dequantize_affine_335"], "sub_7673": ["dequantize_affine_335"], "view_2627": ["dequantize_affine_335"], "mul_16244": ["dequantize_affine_335"], "view_2629": ["dequantize_affine_335"], "mul_16249": ["linear_167"], "view_2630": ["linear_167"], "permute_280": ["linear_167"], "mm_default_20": ["linear_167"], "add_tensor_20": ["linear_167"], "view_2631": ["linear_167", "dropout_84"], "add_25730": ["add_392"], "clone_225": ["layer_norm_56"], "var_mean_56": ["layer_norm_56"], "getitem_224": ["layer_norm_56"], "getitem_225": ["layer_norm_56"], "sub_7679": ["layer_norm_56"], "add_25735": ["layer_norm_56"], "rsqrt_56": ["layer_norm_56"], "mul_16260": ["layer_norm_56"], "mul_16261": ["layer_norm_56"], "add_25736": ["layer_norm_56", "choose_qparams_affine_default_168", "quantize_affine_168", "choose_qparams_affine_default_169", "quantize_affine_169", "choose_qparams_affine_default_170", "quantize_affine_170"], "amax_168": ["choose_qparams_affine_default_168"], "full_337": ["choose_qparams_affine_default_168"], "maximum_168": ["choose_qparams_affine_default_168"], "amin_168": ["choose_qparams_affine_default_168"], "full_336": ["choose_qparams_affine_default_168"], "minimum_168": ["choose_qparams_affine_default_168"], "sub_7690": ["choose_qparams_affine_default_168"], "div_336": ["choose_qparams_affine_default_168"], "clamp_min_504": ["choose_qparams_affine_default_168"], "div_337": ["choose_qparams_affine_default_168"], "round_337": ["choose_qparams_affine_default_168"], "sub_7696": ["choose_qparams_affine_default_168"], "clamp_min_505": ["choose_qparams_affine_default_168"], "clamp_max_336": ["choose_qparams_affine_default_168"], "view_2634": ["quantize_affine_168"], "reciprocal_168": ["quantize_affine_168"], "mul_16309": ["quantize_affine_168"], "mul_16312": ["quantize_affine_168"], "round_338": ["quantize_affine_168"], "convert_element_type_1008": ["choose_qparams_affine_default_168"], "view_2635": ["quantize_affine_168"], "add_25823": ["quantize_affine_168"], "clamp_min_506": ["quantize_affine_168"], "clamp_max_337": ["quantize_affine_168"], "convert_element_type_1009": ["quantize_affine_168", "dequantize_affine_336"], "view_2639": ["dequantize_affine_336"], "convert_element_type_1010": ["dequantize_affine_336"], "convert_element_type_1011": ["dequantize_affine_336"], "sub_7716": ["dequantize_affine_336"], "view_2638": ["dequantize_affine_336"], "mul_16334": ["dequantize_affine_336"], "view_2641": ["dequantize_affine_337"], "view_2643": ["dequantize_affine_337"], "convert_element_type_1012": ["dequantize_affine_337"], "convert_element_type_1013": ["dequantize_affine_337"], "sub_7720": ["dequantize_affine_337"], "view_2642": ["dequantize_affine_337"], "mul_16339": ["dequantize_affine_337"], "view_2644": ["dequantize_affine_337"], "amax_169": ["choose_qparams_affine_default_169"], "full_339": ["choose_qparams_affine_default_169"], "maximum_169": ["choose_qparams_affine_default_169"], "amin_169": ["choose_qparams_affine_default_169"], "full_338": ["choose_qparams_affine_default_169"], "minimum_169": ["choose_qparams_affine_default_169"], "sub_7735": ["choose_qparams_affine_default_169"], "div_338": ["choose_qparams_affine_default_169"], "clamp_min_507": ["choose_qparams_affine_default_169"], "div_339": ["choose_qparams_affine_default_169"], "round_339": ["choose_qparams_affine_default_169"], "sub_7741": ["choose_qparams_affine_default_169"], "clamp_min_508": ["choose_qparams_affine_default_169"], "clamp_max_338": ["choose_qparams_affine_default_169"], "view_2650": ["quantize_affine_169"], "reciprocal_169": ["quantize_affine_169"], "mul_16405": ["quantize_affine_169"], "mul_16408": ["quantize_affine_169"], "round_340": ["quantize_affine_169"], "convert_element_type_1014": ["choose_qparams_affine_default_169"], "view_2651": ["quantize_affine_169"], "add_25975": ["quantize_affine_169"], "clamp_min_509": ["quantize_affine_169"], "clamp_max_339": ["quantize_affine_169"], "convert_element_type_1015": ["quantize_affine_169", "dequantize_affine_338"], "view_2655": ["dequantize_affine_338"], "convert_element_type_1016": ["dequantize_affine_338"], "convert_element_type_1017": ["dequantize_affine_338"], "sub_7761": ["dequantize_affine_338"], "view_2654": ["dequantize_affine_338"], "mul_16430": ["dequantize_affine_338"], "view_2657": ["dequantize_affine_339"], "view_2659": ["dequantize_affine_339"], "convert_element_type_1018": ["dequantize_affine_339"], "convert_element_type_1019": ["dequantize_affine_339"], "sub_7765": ["dequantize_affine_339"], "view_2658": ["dequantize_affine_339"], "mul_16435": ["dequantize_affine_339"], "view_2660": ["dequantize_affine_339"], "amax_170": ["choose_qparams_affine_default_170"], "full_341": ["choose_qparams_affine_default_170"], "maximum_170": ["choose_qparams_affine_default_170"], "amin_170": ["choose_qparams_affine_default_170"], "full_340": ["choose_qparams_affine_default_170"], "minimum_170": ["choose_qparams_affine_default_170"], "sub_7779": ["choose_qparams_affine_default_170"], "div_340": ["choose_qparams_affine_default_170"], "clamp_min_510": ["choose_qparams_affine_default_170"], "div_341": ["choose_qparams_affine_default_170"], "round_341": ["choose_qparams_affine_default_170"], "sub_7785": ["choose_qparams_affine_default_170"], "clamp_min_511": ["choose_qparams_affine_default_170"], "clamp_max_340": ["choose_qparams_affine_default_170"], "view_2666": ["quantize_affine_170"], "reciprocal_170": ["quantize_affine_170"], "mul_16504": ["quantize_affine_170"], "mul_16507": ["quantize_affine_170"], "round_342": ["quantize_affine_170"], "convert_element_type_1020": ["choose_qparams_affine_default_170"], "view_2667": ["quantize_affine_170"], "add_26123": ["quantize_affine_170"], "clamp_min_512": ["quantize_affine_170"], "clamp_max_341": ["quantize_affine_170"], "convert_element_type_1021": ["quantize_affine_170", "dequantize_affine_340"], "view_2671": ["dequantize_affine_340"], "convert_element_type_1022": ["dequantize_affine_340"], "convert_element_type_1023": ["dequantize_affine_340"], "sub_7805": ["dequantize_affine_340"], "view_2670": ["dequantize_affine_340"], "mul_16529": ["dequantize_affine_340"], "view_2673": ["dequantize_affine_341"], "view_2675": ["dequantize_affine_341"], "convert_element_type_1024": ["dequantize_affine_341"], "convert_element_type_1025": ["dequantize_affine_341"], "sub_7809": ["dequantize_affine_341"], "view_2674": ["dequantize_affine_341"], "mul_16534": ["dequantize_affine_341"], "view_2676": ["dequantize_affine_341"], "mul_16344": ["linear_168"], "view_2645": ["linear_168"], "permute_281": ["linear_168"], "mm_default_19": ["linear_168"], "add_tensor_19": ["linear_168"], "view_2646": ["linear_168"], "mul_16351": ["mul_1041"], "view_2647": ["view_84"], "permute_282": ["transpose_112"], "clone_226": ["contiguous_112"], "mul_16438": ["linear_169"], "view_2661": ["linear_169"], "permute_283": ["linear_169"], "mm_28": ["linear_169"], "view_2662": ["linear_169"], "view_2663": ["view_85"], "permute_284": ["transpose_113"], "clone_227": ["contiguous_113"], "mul_16539": ["linear_170"], "view_2677": ["linear_170"], "permute_285": ["linear_170"], "mm_default_18": ["linear_170"], "add_tensor_18": ["linear_170"], "view_2678": ["linear_170"], "view_2679": ["view_86"], "permute_286": ["transpose_114"], "clone_228": ["contiguous_114"], "_scaled_dot_product_efficient_attention_28": ["scaled_dot_product_attention_28"], "getitem_226": ["scaled_dot_product_attention_28"], "permute_287": ["transpose_115"], "view_2680": ["reshape_28", "choose_qparams_affine_default_171", "quantize_affine_171"], "amax_171": ["choose_qparams_affine_default_171"], "full_343": ["choose_qparams_affine_default_171"], "maximum_171": ["choose_qparams_affine_default_171"], "amin_171": ["choose_qparams_affine_default_171"], "full_342": ["choose_qparams_affine_default_171"], "minimum_171": ["choose_qparams_affine_default_171"], "sub_7827": ["choose_qparams_affine_default_171"], "div_342": ["choose_qparams_affine_default_171"], "clamp_min_513": ["choose_qparams_affine_default_171"], "div_343": ["choose_qparams_affine_default_171"], "round_343": ["choose_qparams_affine_default_171"], "sub_7833": ["choose_qparams_affine_default_171"], "clamp_min_514": ["choose_qparams_affine_default_171"], "clamp_max_342": ["choose_qparams_affine_default_171"], "view_2683": ["quantize_affine_171"], "reciprocal_171": ["quantize_affine_171"], "mul_16609": ["quantize_affine_171"], "mul_16612": ["quantize_affine_171"], "round_344": ["quantize_affine_171"], "convert_element_type_1026": ["choose_qparams_affine_default_171"], "view_2684": ["quantize_affine_171"], "add_26287": ["quantize_affine_171"], "clamp_min_515": ["quantize_affine_171"], "clamp_max_343": ["quantize_affine_171"], "convert_element_type_1027": ["quantize_affine_171", "dequantize_affine_342"], "view_2688": ["dequantize_affine_342"], "convert_element_type_1028": ["dequantize_affine_342"], "convert_element_type_1029": ["dequantize_affine_342"], "sub_7853": ["dequantize_affine_342"], "view_2687": ["dequantize_affine_342"], "mul_16634": ["dequantize_affine_342"], "view_2690": ["dequantize_affine_343"], "view_2692": ["dequantize_affine_343"], "convert_element_type_1030": ["dequantize_affine_343"], "convert_element_type_1031": ["dequantize_affine_343"], "sub_7857": ["dequantize_affine_343"], "view_2691": ["dequantize_affine_343"], "mul_16639": ["dequantize_affine_343"], "view_2693": ["dequantize_affine_343"], "mul_16644": ["linear_171"], "view_2694": ["linear_171"], "permute_288": ["linear_171"], "mm_default_17": ["linear_171"], "add_tensor_17": ["linear_171"], "view_2695": ["linear_171", "dropout_85"], "add_26350": ["add_401"], "clone_230": ["layer_norm_57"], "var_mean_57": ["layer_norm_57"], "getitem_230": ["layer_norm_57"], "getitem_231": ["layer_norm_57"], "sub_7863": ["layer_norm_57"], "add_26355": ["layer_norm_57"], "rsqrt_57": ["layer_norm_57"], "mul_16655": ["layer_norm_57"], "mul_16656": ["layer_norm_57"], "add_26356": ["layer_norm_57", "choose_qparams_affine_default_172", "quantize_affine_172"], "amax_172": ["choose_qparams_affine_default_172"], "full_345": ["choose_qparams_affine_default_172"], "maximum_172": ["choose_qparams_affine_default_172"], "amin_172": ["choose_qparams_affine_default_172"], "full_344": ["choose_qparams_affine_default_172"], "minimum_172": ["choose_qparams_affine_default_172"], "sub_7874": ["choose_qparams_affine_default_172"], "div_344": ["choose_qparams_affine_default_172"], "clamp_min_516": ["choose_qparams_affine_default_172"], "div_345": ["choose_qparams_affine_default_172"], "round_345": ["choose_qparams_affine_default_172"], "sub_7880": ["choose_qparams_affine_default_172"], "clamp_min_517": ["choose_qparams_affine_default_172"], "clamp_max_344": ["choose_qparams_affine_default_172"], "view_2698": ["quantize_affine_172"], "reciprocal_172": ["quantize_affine_172"], "mul_16704": ["quantize_affine_172"], "mul_16707": ["quantize_affine_172"], "round_346": ["quantize_affine_172"], "convert_element_type_1032": ["choose_qparams_affine_default_172"], "view_2699": ["quantize_affine_172"], "add_26443": ["quantize_affine_172"], "clamp_min_518": ["quantize_affine_172"], "clamp_max_345": ["quantize_affine_172"], "convert_element_type_1033": ["quantize_affine_172", "dequantize_affine_344"], "view_2703": ["dequantize_affine_344"], "convert_element_type_1034": ["dequantize_affine_344"], "convert_element_type_1035": ["dequantize_affine_344"], "sub_7900": ["dequantize_affine_344"], "view_2702": ["dequantize_affine_344"], "mul_16729": ["dequantize_affine_344"], "view_2705": ["dequantize_affine_345"], "view_2707": ["dequantize_affine_345"], "convert_element_type_1036": ["dequantize_affine_345"], "convert_element_type_1037": ["dequantize_affine_345"], "sub_7904": ["dequantize_affine_345"], "view_2706": ["dequantize_affine_345"], "mul_16734": ["dequantize_affine_345"], "view_2708": ["dequantize_affine_345"], "mul_16739": ["linear_172"], "view_2709": ["linear_172"], "permute_289": ["linear_172"], "mm_default_16": ["linear_172"], "add_tensor_16": ["linear_172"], "view_2710": ["linear_172"], "mul_16746": ["gelu_30"], "mul_16747": ["gelu_30"], "erf_30": ["gelu_30"], "add_26502": ["gelu_30"], "mul_16748": ["gelu_30", "dropout_86", "choose_qparams_affine_default_173", "quantize_affine_173"], "amax_173": ["choose_qparams_affine_default_173"], "full_347": ["choose_qparams_affine_default_173"], "maximum_173": ["choose_qparams_affine_default_173"], "amin_173": ["choose_qparams_affine_default_173"], "full_346": ["choose_qparams_affine_default_173"], "minimum_173": ["choose_qparams_affine_default_173"], "sub_7917": ["choose_qparams_affine_default_173"], "div_346": ["choose_qparams_affine_default_173"], "clamp_min_519": ["choose_qparams_affine_default_173"], "div_347": ["choose_qparams_affine_default_173"], "round_347": ["choose_qparams_affine_default_173"], "sub_7923": ["choose_qparams_affine_default_173"], "clamp_min_520": ["choose_qparams_affine_default_173"], "clamp_max_346": ["choose_qparams_affine_default_173"], "view_2713": ["quantize_affine_173"], "reciprocal_173": ["quantize_affine_173"], "mul_16794": ["quantize_affine_173"], "mul_16797": ["quantize_affine_173"], "round_348": ["quantize_affine_173"], "convert_element_type_1038": ["choose_qparams_affine_default_173"], "view_2714": ["quantize_affine_173"], "add_26585": ["quantize_affine_173"], "clamp_min_521": ["quantize_affine_173"], "clamp_max_347": ["quantize_affine_173"], "convert_element_type_1039": ["quantize_affine_173", "dequantize_affine_346"], "view_2718": ["dequantize_affine_346"], "convert_element_type_1040": ["dequantize_affine_346"], "convert_element_type_1041": ["dequantize_affine_346"], "sub_7943": ["dequantize_affine_346"], "view_2717": ["dequantize_affine_346"], "mul_16819": ["dequantize_affine_346"], "view_2720": ["dequantize_affine_347"], "view_2722": ["dequantize_affine_347"], "convert_element_type_1042": ["dequantize_affine_347"], "convert_element_type_1043": ["dequantize_affine_347"], "sub_7947": ["dequantize_affine_347"], "view_2721": ["dequantize_affine_347"], "mul_16824": ["dequantize_affine_347"], "view_2723": ["dequantize_affine_347"], "mul_16829": ["linear_173"], "view_2724": ["linear_173"], "permute_290": ["linear_173"], "mm_default_15": ["linear_173"], "add_tensor_15": ["linear_173"], "view_2725": ["linear_173", "dropout_87"], "add_26648": ["add_406"], "clone_233": ["layer_norm_58"], "var_mean_58": ["layer_norm_58"], "getitem_232": ["layer_norm_58"], "getitem_233": ["layer_norm_58"], "sub_7953": ["layer_norm_58"], "add_26653": ["layer_norm_58"], "rsqrt_58": ["layer_norm_58"], "mul_16840": ["layer_norm_58"], "mul_16841": ["layer_norm_58"], "add_26654": ["layer_norm_58", "choose_qparams_affine_default_174", "quantize_affine_174", "choose_qparams_affine_default_175", "quantize_affine_175", "choose_qparams_affine_default_176", "quantize_affine_176"], "amax_174": ["choose_qparams_affine_default_174"], "full_349": ["choose_qparams_affine_default_174"], "maximum_174": ["choose_qparams_affine_default_174"], "amin_174": ["choose_qparams_affine_default_174"], "full_348": ["choose_qparams_affine_default_174"], "minimum_174": ["choose_qparams_affine_default_174"], "sub_7964": ["choose_qparams_affine_default_174"], "div_348": ["choose_qparams_affine_default_174"], "clamp_min_522": ["choose_qparams_affine_default_174"], "div_349": ["choose_qparams_affine_default_174"], "round_349": ["choose_qparams_affine_default_174"], "sub_7970": ["choose_qparams_affine_default_174"], "clamp_min_523": ["choose_qparams_affine_default_174"], "clamp_max_348": ["choose_qparams_affine_default_174"], "view_2728": ["quantize_affine_174"], "reciprocal_174": ["quantize_affine_174"], "mul_16889": ["quantize_affine_174"], "mul_16892": ["quantize_affine_174"], "round_350": ["quantize_affine_174"], "convert_element_type_1044": ["choose_qparams_affine_default_174"], "view_2729": ["quantize_affine_174"], "add_26741": ["quantize_affine_174"], "clamp_min_524": ["quantize_affine_174"], "clamp_max_349": ["quantize_affine_174"], "convert_element_type_1045": ["quantize_affine_174", "dequantize_affine_348"], "view_2733": ["dequantize_affine_348"], "convert_element_type_1046": ["dequantize_affine_348"], "convert_element_type_1047": ["dequantize_affine_348"], "sub_7990": ["dequantize_affine_348"], "view_2732": ["dequantize_affine_348"], "mul_16914": ["dequantize_affine_348"], "view_2735": ["dequantize_affine_349"], "view_2737": ["dequantize_affine_349"], "convert_element_type_1048": ["dequantize_affine_349"], "convert_element_type_1049": ["dequantize_affine_349"], "sub_7994": ["dequantize_affine_349"], "view_2736": ["dequantize_affine_349"], "mul_16919": ["dequantize_affine_349"], "view_2738": ["dequantize_affine_349"], "amax_175": ["choose_qparams_affine_default_175"], "full_351": ["choose_qparams_affine_default_175"], "maximum_175": ["choose_qparams_affine_default_175"], "amin_175": ["choose_qparams_affine_default_175"], "full_350": ["choose_qparams_affine_default_175"], "minimum_175": ["choose_qparams_affine_default_175"], "sub_8009": ["choose_qparams_affine_default_175"], "div_350": ["choose_qparams_affine_default_175"], "clamp_min_525": ["choose_qparams_affine_default_175"], "div_351": ["choose_qparams_affine_default_175"], "round_351": ["choose_qparams_affine_default_175"], "sub_8015": ["choose_qparams_affine_default_175"], "clamp_min_526": ["choose_qparams_affine_default_175"], "clamp_max_350": ["choose_qparams_affine_default_175"], "view_2744": ["quantize_affine_175"], "reciprocal_175": ["quantize_affine_175"], "mul_16985": ["quantize_affine_175"], "mul_16988": ["quantize_affine_175"], "round_352": ["quantize_affine_175"], "convert_element_type_1050": ["choose_qparams_affine_default_175"], "view_2745": ["quantize_affine_175"], "add_26893": ["quantize_affine_175"], "clamp_min_527": ["quantize_affine_175"], "clamp_max_351": ["quantize_affine_175"], "convert_element_type_1051": ["quantize_affine_175", "dequantize_affine_350"], "view_2749": ["dequantize_affine_350"], "convert_element_type_1052": ["dequantize_affine_350"], "convert_element_type_1053": ["dequantize_affine_350"], "sub_8035": ["dequantize_affine_350"], "view_2748": ["dequantize_affine_350"], "mul_17010": ["dequantize_affine_350"], "view_2751": ["dequantize_affine_351"], "view_2753": ["dequantize_affine_351"], "convert_element_type_1054": ["dequantize_affine_351"], "convert_element_type_1055": ["dequantize_affine_351"], "sub_8039": ["dequantize_affine_351"], "view_2752": ["dequantize_affine_351"], "mul_17015": ["dequantize_affine_351"], "view_2754": ["dequantize_affine_351"], "amax_176": ["choose_qparams_affine_default_176"], "full_353": ["choose_qparams_affine_default_176"], "maximum_176": ["choose_qparams_affine_default_176"], "amin_176": ["choose_qparams_affine_default_176"], "full_352": ["choose_qparams_affine_default_176"], "minimum_176": ["choose_qparams_affine_default_176"], "sub_8053": ["choose_qparams_affine_default_176"], "div_352": ["choose_qparams_affine_default_176"], "clamp_min_528": ["choose_qparams_affine_default_176"], "div_353": ["choose_qparams_affine_default_176"], "round_353": ["choose_qparams_affine_default_176"], "sub_8059": ["choose_qparams_affine_default_176"], "clamp_min_529": ["choose_qparams_affine_default_176"], "clamp_max_352": ["choose_qparams_affine_default_176"], "view_2760": ["quantize_affine_176"], "reciprocal_176": ["quantize_affine_176"], "mul_17084": ["quantize_affine_176"], "mul_17087": ["quantize_affine_176"], "round_354": ["quantize_affine_176"], "convert_element_type_1056": ["choose_qparams_affine_default_176"], "view_2761": ["quantize_affine_176"], "add_27041": ["quantize_affine_176"], "clamp_min_530": ["quantize_affine_176"], "clamp_max_353": ["quantize_affine_176"], "convert_element_type_1057": ["quantize_affine_176", "dequantize_affine_352"], "view_2765": ["dequantize_affine_352"], "convert_element_type_1058": ["dequantize_affine_352"], "convert_element_type_1059": ["dequantize_affine_352"], "sub_8079": ["dequantize_affine_352"], "view_2764": ["dequantize_affine_352"], "mul_17109": ["dequantize_affine_352"], "view_2767": ["dequantize_affine_353"], "view_2769": ["dequantize_affine_353"], "convert_element_type_1060": ["dequantize_affine_353"], "convert_element_type_1061": ["dequantize_affine_353"], "sub_8083": ["dequantize_affine_353"], "view_2768": ["dequantize_affine_353"], "mul_17114": ["dequantize_affine_353"], "view_2770": ["dequantize_affine_353"], "mul_16924": ["linear_174"], "view_2739": ["linear_174"], "permute_291": ["linear_174"], "mm_default_14": ["linear_174"], "add_tensor_14": ["linear_174"], "view_2740": ["linear_174"], "mul_16931": ["mul_1078"], "view_2741": ["view_87"], "permute_292": ["transpose_116"], "clone_234": ["contiguous_116"], "mul_17018": ["linear_175"], "view_2755": ["linear_175"], "permute_293": ["linear_175"], "mm_29": ["linear_175"], "view_2756": ["linear_175"], "view_2757": ["view_88"], "permute_294": ["transpose_117"], "clone_235": ["contiguous_117"], "mul_17119": ["linear_176"], "view_2771": ["linear_176"], "permute_295": ["linear_176"], "mm_default_13": ["linear_176"], "add_tensor_13": ["linear_176"], "view_2772": ["linear_176"], "view_2773": ["view_89"], "permute_296": ["transpose_118"], "clone_236": ["contiguous_118"], "_scaled_dot_product_efficient_attention_29": ["scaled_dot_product_attention_29"], "getitem_234": ["scaled_dot_product_attention_29"], "permute_297": ["transpose_119"], "view_2774": ["reshape_29", "choose_qparams_affine_default_177", "quantize_affine_177"], "amax_177": ["choose_qparams_affine_default_177"], "full_355": ["choose_qparams_affine_default_177"], "maximum_177": ["choose_qparams_affine_default_177"], "amin_177": ["choose_qparams_affine_default_177"], "full_354": ["choose_qparams_affine_default_177"], "minimum_177": ["choose_qparams_affine_default_177"], "sub_8101": ["choose_qparams_affine_default_177"], "div_354": ["choose_qparams_affine_default_177"], "clamp_min_531": ["choose_qparams_affine_default_177"], "div_355": ["choose_qparams_affine_default_177"], "round_355": ["choose_qparams_affine_default_177"], "sub_8107": ["choose_qparams_affine_default_177"], "clamp_min_532": ["choose_qparams_affine_default_177"], "clamp_max_354": ["choose_qparams_affine_default_177"], "view_2777": ["quantize_affine_177"], "reciprocal_177": ["quantize_affine_177"], "mul_17189": ["quantize_affine_177"], "mul_17192": ["quantize_affine_177"], "round_356": ["quantize_affine_177"], "convert_element_type_1062": ["choose_qparams_affine_default_177"], "view_2778": ["quantize_affine_177"], "add_27205": ["quantize_affine_177"], "clamp_min_533": ["quantize_affine_177"], "clamp_max_355": ["quantize_affine_177"], "convert_element_type_1063": ["quantize_affine_177", "dequantize_affine_354"], "view_2782": ["dequantize_affine_354"], "convert_element_type_1064": ["dequantize_affine_354"], "convert_element_type_1065": ["dequantize_affine_354"], "sub_8127": ["dequantize_affine_354"], "view_2781": ["dequantize_affine_354"], "mul_17214": ["dequantize_affine_354"], "view_2784": ["dequantize_affine_355"], "view_2786": ["dequantize_affine_355"], "convert_element_type_1066": ["dequantize_affine_355"], "convert_element_type_1067": ["dequantize_affine_355"], "sub_8131": ["dequantize_affine_355"], "view_2785": ["dequantize_affine_355"], "mul_17219": ["dequantize_affine_355"], "view_2787": ["dequantize_affine_355"], "mul_17224": ["linear_177"], "view_2788": ["linear_177"], "permute_298": ["linear_177"], "mm_default_12": ["linear_177"], "add_tensor_12": ["linear_177"], "view_2789": ["linear_177", "dropout_88"], "add_27268": ["add_415"], "clone_238": ["layer_norm_59"], "var_mean_59": ["layer_norm_59"], "getitem_238": ["layer_norm_59"], "getitem_239": ["layer_norm_59"], "sub_8137": ["layer_norm_59"], "add_27273": ["layer_norm_59"], "rsqrt_59": ["layer_norm_59"], "mul_17235": ["layer_norm_59"], "mul_17236": ["layer_norm_59"], "add_27274": ["layer_norm_59", "choose_qparams_affine_default_178", "quantize_affine_178"], "amax_178": ["choose_qparams_affine_default_178"], "full_357": ["choose_qparams_affine_default_178"], "maximum_178": ["choose_qparams_affine_default_178"], "amin_178": ["choose_qparams_affine_default_178"], "full_356": ["choose_qparams_affine_default_178"], "minimum_178": ["choose_qparams_affine_default_178"], "sub_8148": ["choose_qparams_affine_default_178"], "div_356": ["choose_qparams_affine_default_178"], "clamp_min_534": ["choose_qparams_affine_default_178"], "div_357": ["choose_qparams_affine_default_178"], "round_357": ["choose_qparams_affine_default_178"], "sub_8154": ["choose_qparams_affine_default_178"], "clamp_min_535": ["choose_qparams_affine_default_178"], "clamp_max_356": ["choose_qparams_affine_default_178"], "view_2792": ["quantize_affine_178"], "reciprocal_178": ["quantize_affine_178"], "mul_17284": ["quantize_affine_178"], "mul_17287": ["quantize_affine_178"], "round_358": ["quantize_affine_178"], "convert_element_type_1068": ["choose_qparams_affine_default_178"], "view_2793": ["quantize_affine_178"], "add_27361": ["quantize_affine_178"], "clamp_min_536": ["quantize_affine_178"], "clamp_max_357": ["quantize_affine_178"], "convert_element_type_1069": ["quantize_affine_178", "dequantize_affine_356"], "view_2797": ["dequantize_affine_356"], "convert_element_type_1070": ["dequantize_affine_356"], "convert_element_type_1071": ["dequantize_affine_356"], "sub_8174": ["dequantize_affine_356"], "view_2796": ["dequantize_affine_356"], "mul_17309": ["dequantize_affine_356"], "view_2799": ["dequantize_affine_357"], "view_2801": ["dequantize_affine_357"], "convert_element_type_1072": ["dequantize_affine_357"], "convert_element_type_1073": ["dequantize_affine_357"], "sub_8178": ["dequantize_affine_357"], "view_2800": ["dequantize_affine_357"], "mul_17314": ["dequantize_affine_357"], "view_2802": ["dequantize_affine_357"], "mul_17319": ["linear_178"], "view_2803": ["linear_178"], "permute_299": ["linear_178"], "mm_default_11": ["linear_178"], "add_tensor_11": ["linear_178"], "view_2804": ["linear_178"], "mul_17326": ["gelu_31"], "mul_17327": ["gelu_31"], "erf_31": ["gelu_31"], "add_27420": ["gelu_31"], "mul_17328": ["gelu_31", "dropout_89", "choose_qparams_affine_default_179", "quantize_affine_179"], "amax_179": ["choose_qparams_affine_default_179"], "full_359": ["choose_qparams_affine_default_179"], "maximum_179": ["choose_qparams_affine_default_179"], "amin_179": ["choose_qparams_affine_default_179"], "full_358": ["choose_qparams_affine_default_179"], "minimum_179": ["choose_qparams_affine_default_179"], "sub_8191": ["choose_qparams_affine_default_179"], "div_358": ["choose_qparams_affine_default_179"], "clamp_min_537": ["choose_qparams_affine_default_179"], "div_359": ["choose_qparams_affine_default_179"], "round_359": ["choose_qparams_affine_default_179"], "sub_8197": ["choose_qparams_affine_default_179"], "clamp_min_538": ["choose_qparams_affine_default_179"], "clamp_max_358": ["choose_qparams_affine_default_179"], "view_2807": ["quantize_affine_179"], "reciprocal_179": ["quantize_affine_179"], "mul_17374": ["quantize_affine_179"], "mul_17377": ["quantize_affine_179"], "round_360": ["quantize_affine_179"], "convert_element_type_1074": ["choose_qparams_affine_default_179"], "view_2808": ["quantize_affine_179"], "add_27503": ["quantize_affine_179"], "clamp_min_539": ["quantize_affine_179"], "clamp_max_359": ["quantize_affine_179"], "convert_element_type_1075": ["quantize_affine_179", "dequantize_affine_358"], "view_2812": ["dequantize_affine_358"], "convert_element_type_1076": ["dequantize_affine_358"], "convert_element_type_1077": ["dequantize_affine_358"], "sub_8217": ["dequantize_affine_358"], "view_2811": ["dequantize_affine_358"], "mul_17399": ["dequantize_affine_358"], "view_2814": ["dequantize_affine_359"], "view_2816": ["dequantize_affine_359"], "convert_element_type_1078": ["dequantize_affine_359"], "convert_element_type_1079": ["dequantize_affine_359"], "sub_8221": ["dequantize_affine_359"], "view_2815": ["dequantize_affine_359"], "mul_17404": ["dequantize_affine_359"], "view_2817": ["dequantize_affine_359"], "mul_17409": ["linear_179"], "view_2818": ["linear_179"], "permute_300": ["linear_179"], "mm_default_10": ["linear_179"], "add_tensor_10": ["linear_179"], "view_2819": ["linear_179", "dropout_90"], "add_27566": ["add_420"], "clone_241": ["layer_norm_60"], "var_mean_60": ["layer_norm_60"], "getitem_240": ["layer_norm_60"], "getitem_241": ["layer_norm_60"], "sub_8227": ["layer_norm_60"], "add_27571": ["layer_norm_60"], "rsqrt_60": ["layer_norm_60"], "mul_17420": ["layer_norm_60"], "mul_17421": ["layer_norm_60"], "add_27572": ["layer_norm_60", "choose_qparams_affine_default_180", "quantize_affine_180", "choose_qparams_affine_default_181", "quantize_affine_181", "choose_qparams_affine_default_182", "quantize_affine_182"], "amax_180": ["choose_qparams_affine_default_180"], "full_361": ["choose_qparams_affine_default_180"], "maximum_180": ["choose_qparams_affine_default_180"], "amin_180": ["choose_qparams_affine_default_180"], "full_360": ["choose_qparams_affine_default_180"], "minimum_180": ["choose_qparams_affine_default_180"], "sub_8238": ["choose_qparams_affine_default_180"], "div_360": ["choose_qparams_affine_default_180"], "clamp_min_540": ["choose_qparams_affine_default_180"], "div_361": ["choose_qparams_affine_default_180"], "round_361": ["choose_qparams_affine_default_180"], "sub_8244": ["choose_qparams_affine_default_180"], "clamp_min_541": ["choose_qparams_affine_default_180"], "clamp_max_360": ["choose_qparams_affine_default_180"], "view_2822": ["quantize_affine_180"], "reciprocal_180": ["quantize_affine_180"], "mul_17469": ["quantize_affine_180"], "mul_17472": ["quantize_affine_180"], "round_362": ["quantize_affine_180"], "convert_element_type_1080": ["choose_qparams_affine_default_180"], "view_2823": ["quantize_affine_180"], "add_27659": ["quantize_affine_180"], "clamp_min_542": ["quantize_affine_180"], "clamp_max_361": ["quantize_affine_180"], "convert_element_type_1081": ["quantize_affine_180", "dequantize_affine_360"], "view_2827": ["dequantize_affine_360"], "convert_element_type_1082": ["dequantize_affine_360"], "convert_element_type_1083": ["dequantize_affine_360"], "sub_8264": ["dequantize_affine_360"], "view_2826": ["dequantize_affine_360"], "mul_17494": ["dequantize_affine_360"], "view_2829": ["dequantize_affine_361"], "view_2831": ["dequantize_affine_361"], "convert_element_type_1084": ["dequantize_affine_361"], "convert_element_type_1085": ["dequantize_affine_361"], "sub_8268": ["dequantize_affine_361"], "view_2830": ["dequantize_affine_361"], "mul_17499": ["dequantize_affine_361"], "view_2832": ["dequantize_affine_361"], "amax_181": ["choose_qparams_affine_default_181"], "full_363": ["choose_qparams_affine_default_181"], "maximum_181": ["choose_qparams_affine_default_181"], "amin_181": ["choose_qparams_affine_default_181"], "full_362": ["choose_qparams_affine_default_181"], "minimum_181": ["choose_qparams_affine_default_181"], "sub_8283": ["choose_qparams_affine_default_181"], "div_362": ["choose_qparams_affine_default_181"], "clamp_min_543": ["choose_qparams_affine_default_181"], "div_363": ["choose_qparams_affine_default_181"], "round_363": ["choose_qparams_affine_default_181"], "sub_8289": ["choose_qparams_affine_default_181"], "clamp_min_544": ["choose_qparams_affine_default_181"], "clamp_max_362": ["choose_qparams_affine_default_181"], "view_2838": ["quantize_affine_181"], "reciprocal_181": ["quantize_affine_181"], "mul_17565": ["quantize_affine_181"], "mul_17568": ["quantize_affine_181"], "round_364": ["quantize_affine_181"], "convert_element_type_1086": ["choose_qparams_affine_default_181"], "view_2839": ["quantize_affine_181"], "add_27811": ["quantize_affine_181"], "clamp_min_545": ["quantize_affine_181"], "clamp_max_363": ["quantize_affine_181"], "convert_element_type_1087": ["quantize_affine_181", "dequantize_affine_362"], "view_2843": ["dequantize_affine_362"], "convert_element_type_1088": ["dequantize_affine_362"], "convert_element_type_1089": ["dequantize_affine_362"], "sub_8309": ["dequantize_affine_362"], "view_2842": ["dequantize_affine_362"], "mul_17590": ["dequantize_affine_362"], "view_2845": ["dequantize_affine_363"], "view_2847": ["dequantize_affine_363"], "convert_element_type_1090": ["dequantize_affine_363"], "convert_element_type_1091": ["dequantize_affine_363"], "sub_8313": ["dequantize_affine_363"], "view_2846": ["dequantize_affine_363"], "mul_17595": ["dequantize_affine_363"], "view_2848": ["dequantize_affine_363"], "amax_182": ["choose_qparams_affine_default_182"], "full_365": ["choose_qparams_affine_default_182"], "maximum_182": ["choose_qparams_affine_default_182"], "amin_182": ["choose_qparams_affine_default_182"], "full_364": ["choose_qparams_affine_default_182"], "minimum_182": ["choose_qparams_affine_default_182"], "sub_8327": ["choose_qparams_affine_default_182"], "div_364": ["choose_qparams_affine_default_182"], "clamp_min_546": ["choose_qparams_affine_default_182"], "div_365": ["choose_qparams_affine_default_182"], "round_365": ["choose_qparams_affine_default_182"], "sub_8333": ["choose_qparams_affine_default_182"], "clamp_min_547": ["choose_qparams_affine_default_182"], "clamp_max_364": ["choose_qparams_affine_default_182"], "view_2854": ["quantize_affine_182"], "reciprocal_182": ["quantize_affine_182"], "mul_17664": ["quantize_affine_182"], "mul_17667": ["quantize_affine_182"], "round_366": ["quantize_affine_182"], "convert_element_type_1092": ["choose_qparams_affine_default_182"], "view_2855": ["quantize_affine_182"], "add_27959": ["quantize_affine_182"], "clamp_min_548": ["quantize_affine_182"], "clamp_max_365": ["quantize_affine_182"], "convert_element_type_1093": ["quantize_affine_182", "dequantize_affine_364"], "view_2859": ["dequantize_affine_364"], "convert_element_type_1094": ["dequantize_affine_364"], "convert_element_type_1095": ["dequantize_affine_364"], "sub_8353": ["dequantize_affine_364"], "view_2858": ["dequantize_affine_364"], "mul_17689": ["dequantize_affine_364"], "view_2861": ["dequantize_affine_365"], "view_2863": ["dequantize_affine_365"], "convert_element_type_1096": ["dequantize_affine_365"], "convert_element_type_1097": ["dequantize_affine_365"], "sub_8357": ["dequantize_affine_365"], "view_2862": ["dequantize_affine_365"], "mul_17694": ["dequantize_affine_365"], "view_2864": ["dequantize_affine_365"], "mul_17504": ["linear_180"], "view_2833": ["linear_180"], "permute_301": ["linear_180"], "mm_default_9": ["linear_180"], "add_tensor_9": ["linear_180"], "view_2834": ["linear_180"], "mul_17511": ["mul_1115"], "view_2835": ["view_90"], "permute_302": ["transpose_120"], "clone_242": ["contiguous_120"], "mul_17598": ["linear_181"], "view_2849": ["linear_181"], "permute_303": ["linear_181"], "mm_30": ["linear_181"], "view_2850": ["linear_181"], "view_2851": ["view_91"], "permute_304": ["transpose_121"], "clone_243": ["contiguous_121"], "mul_17699": ["linear_182"], "view_2865": ["linear_182"], "permute_305": ["linear_182"], "mm_default_8": ["linear_182"], "add_tensor_8": ["linear_182"], "view_2866": ["linear_182"], "view_2867": ["view_92"], "permute_306": ["transpose_122"], "clone_244": ["contiguous_122"], "_scaled_dot_product_efficient_attention_30": ["scaled_dot_product_attention_30"], "getitem_242": ["scaled_dot_product_attention_30"], "permute_307": ["transpose_123"], "view_2868": ["reshape_30", "choose_qparams_affine_default_183", "quantize_affine_183"], "amax_183": ["choose_qparams_affine_default_183"], "full_367": ["choose_qparams_affine_default_183"], "maximum_183": ["choose_qparams_affine_default_183"], "amin_183": ["choose_qparams_affine_default_183"], "full_366": ["choose_qparams_affine_default_183"], "minimum_183": ["choose_qparams_affine_default_183"], "sub_8375": ["choose_qparams_affine_default_183"], "div_366": ["choose_qparams_affine_default_183"], "clamp_min_549": ["choose_qparams_affine_default_183"], "div_367": ["choose_qparams_affine_default_183"], "round_367": ["choose_qparams_affine_default_183"], "sub_8381": ["choose_qparams_affine_default_183"], "clamp_min_550": ["choose_qparams_affine_default_183"], "clamp_max_366": ["choose_qparams_affine_default_183"], "view_2871": ["quantize_affine_183"], "reciprocal_183": ["quantize_affine_183"], "mul_17769": ["quantize_affine_183"], "mul_17772": ["quantize_affine_183"], "round_368": ["quantize_affine_183"], "convert_element_type_1098": ["choose_qparams_affine_default_183"], "view_2872": ["quantize_affine_183"], "add_28123": ["quantize_affine_183"], "clamp_min_551": ["quantize_affine_183"], "clamp_max_367": ["quantize_affine_183"], "convert_element_type_1099": ["quantize_affine_183", "dequantize_affine_366"], "view_2876": ["dequantize_affine_366"], "convert_element_type_1100": ["dequantize_affine_366"], "convert_element_type_1101": ["dequantize_affine_366"], "sub_8401": ["dequantize_affine_366"], "view_2875": ["dequantize_affine_366"], "mul_17794": ["dequantize_affine_366"], "view_2878": ["dequantize_affine_367"], "view_2880": ["dequantize_affine_367"], "convert_element_type_1102": ["dequantize_affine_367"], "convert_element_type_1103": ["dequantize_affine_367"], "sub_8405": ["dequantize_affine_367"], "view_2879": ["dequantize_affine_367"], "mul_17799": ["dequantize_affine_367"], "view_2881": ["dequantize_affine_367"], "mul_17804": ["linear_183"], "view_2882": ["linear_183"], "permute_308": ["linear_183"], "mm_default_7": ["linear_183"], "add_tensor_7": ["linear_183"], "view_2883": ["linear_183", "dropout_91"], "add_28186": ["add_429"], "clone_246": ["layer_norm_61"], "var_mean_61": ["layer_norm_61"], "getitem_246": ["layer_norm_61"], "getitem_247": ["layer_norm_61"], "sub_8411": ["layer_norm_61"], "add_28191": ["layer_norm_61"], "rsqrt_61": ["layer_norm_61"], "mul_17815": ["layer_norm_61"], "mul_17816": ["layer_norm_61"], "add_28192": ["layer_norm_61", "choose_qparams_affine_default_184", "quantize_affine_184"], "amax_184": ["choose_qparams_affine_default_184"], "full_369": ["choose_qparams_affine_default_184"], "maximum_184": ["choose_qparams_affine_default_184"], "amin_184": ["choose_qparams_affine_default_184"], "full_368": ["choose_qparams_affine_default_184"], "minimum_184": ["choose_qparams_affine_default_184"], "sub_8422": ["choose_qparams_affine_default_184"], "div_368": ["choose_qparams_affine_default_184"], "clamp_min_552": ["choose_qparams_affine_default_184"], "div_369": ["choose_qparams_affine_default_184"], "round_369": ["choose_qparams_affine_default_184"], "sub_8428": ["choose_qparams_affine_default_184"], "clamp_min_553": ["choose_qparams_affine_default_184"], "clamp_max_368": ["choose_qparams_affine_default_184"], "view_2886": ["quantize_affine_184"], "reciprocal_184": ["quantize_affine_184"], "mul_17864": ["quantize_affine_184"], "mul_17867": ["quantize_affine_184"], "round_370": ["quantize_affine_184"], "convert_element_type_1104": ["choose_qparams_affine_default_184"], "view_2887": ["quantize_affine_184"], "add_28279": ["quantize_affine_184"], "clamp_min_554": ["quantize_affine_184"], "clamp_max_369": ["quantize_affine_184"], "convert_element_type_1105": ["quantize_affine_184", "dequantize_affine_368"], "view_2891": ["dequantize_affine_368"], "convert_element_type_1106": ["dequantize_affine_368"], "convert_element_type_1107": ["dequantize_affine_368"], "sub_8448": ["dequantize_affine_368"], "view_2890": ["dequantize_affine_368"], "mul_17889": ["dequantize_affine_368"], "view_2893": ["dequantize_affine_369"], "view_2895": ["dequantize_affine_369"], "convert_element_type_1108": ["dequantize_affine_369"], "convert_element_type_1109": ["dequantize_affine_369"], "sub_8452": ["dequantize_affine_369"], "view_2894": ["dequantize_affine_369"], "mul_17894": ["dequantize_affine_369"], "view_2896": ["dequantize_affine_369"], "mul_17899": ["linear_184"], "view_2897": ["linear_184"], "permute_309": ["linear_184"], "mm_default_6": ["linear_184"], "add_tensor_6": ["linear_184"], "view_2898": ["linear_184"], "mul_17906": ["gelu_32"], "mul_17907": ["gelu_32"], "erf_32": ["gelu_32"], "add_28338": ["gelu_32"], "mul_17908": ["gelu_32", "dropout_92", "choose_qparams_affine_default_185", "quantize_affine_185"], "amax_185": ["choose_qparams_affine_default_185"], "full_371": ["choose_qparams_affine_default_185"], "maximum_185": ["choose_qparams_affine_default_185"], "amin_185": ["choose_qparams_affine_default_185"], "full_370": ["choose_qparams_affine_default_185"], "minimum_185": ["choose_qparams_affine_default_185"], "sub_8465": ["choose_qparams_affine_default_185"], "div_370": ["choose_qparams_affine_default_185"], "clamp_min_555": ["choose_qparams_affine_default_185"], "div_371": ["choose_qparams_affine_default_185"], "round_371": ["choose_qparams_affine_default_185"], "sub_8471": ["choose_qparams_affine_default_185"], "clamp_min_556": ["choose_qparams_affine_default_185"], "clamp_max_370": ["choose_qparams_affine_default_185"], "view_2901": ["quantize_affine_185"], "reciprocal_185": ["quantize_affine_185"], "mul_17954": ["quantize_affine_185"], "mul_17957": ["quantize_affine_185"], "round_372": ["quantize_affine_185"], "convert_element_type_1110": ["choose_qparams_affine_default_185"], "view_2902": ["quantize_affine_185"], "add_28421": ["quantize_affine_185"], "clamp_min_557": ["quantize_affine_185"], "clamp_max_371": ["quantize_affine_185"], "convert_element_type_1111": ["quantize_affine_185", "dequantize_affine_370"], "view_2906": ["dequantize_affine_370"], "convert_element_type_1112": ["dequantize_affine_370"], "convert_element_type_1113": ["dequantize_affine_370"], "sub_8491": ["dequantize_affine_370"], "view_2905": ["dequantize_affine_370"], "mul_17979": ["dequantize_affine_370"], "view_2908": ["dequantize_affine_371"], "view_2910": ["dequantize_affine_371"], "convert_element_type_1114": ["dequantize_affine_371"], "convert_element_type_1115": ["dequantize_affine_371"], "sub_8495": ["dequantize_affine_371"], "view_2909": ["dequantize_affine_371"], "mul_17984": ["dequantize_affine_371"], "view_2911": ["dequantize_affine_371"], "mul_17989": ["linear_185"], "view_2912": ["linear_185"], "permute_310": ["linear_185"], "mm_default_5": ["linear_185"], "add_tensor_5": ["linear_185"], "view_2913": ["linear_185", "dropout_93"], "add_28484": ["add_434"], "clone_249": ["layer_norm_62"], "var_mean_62": ["layer_norm_62"], "getitem_248": ["layer_norm_62"], "getitem_249": ["layer_norm_62"], "sub_8501": ["layer_norm_62"], "add_28489": ["layer_norm_62"], "rsqrt_62": ["layer_norm_62"], "mul_18000": ["layer_norm_62"], "mul_18001": ["layer_norm_62"], "add_28490": ["layer_norm_62", "choose_qparams_affine_default_186", "quantize_affine_186", "choose_qparams_affine_default_187", "quantize_affine_187", "choose_qparams_affine_default_188", "quantize_affine_188"], "amax_186": ["choose_qparams_affine_default_186"], "full_373": ["choose_qparams_affine_default_186"], "maximum_186": ["choose_qparams_affine_default_186"], "amin_186": ["choose_qparams_affine_default_186"], "full_372": ["choose_qparams_affine_default_186"], "minimum_186": ["choose_qparams_affine_default_186"], "sub_8512": ["choose_qparams_affine_default_186"], "div_372": ["choose_qparams_affine_default_186"], "clamp_min_558": ["choose_qparams_affine_default_186"], "div_373": ["choose_qparams_affine_default_186"], "round_373": ["choose_qparams_affine_default_186"], "sub_8518": ["choose_qparams_affine_default_186"], "clamp_min_559": ["choose_qparams_affine_default_186"], "clamp_max_372": ["choose_qparams_affine_default_186"], "view_2916": ["quantize_affine_186"], "reciprocal_186": ["quantize_affine_186"], "mul_18049": ["quantize_affine_186"], "mul_18052": ["quantize_affine_186"], "round_374": ["quantize_affine_186"], "convert_element_type_1116": ["choose_qparams_affine_default_186"], "view_2917": ["quantize_affine_186"], "add_28577": ["quantize_affine_186"], "clamp_min_560": ["quantize_affine_186"], "clamp_max_373": ["quantize_affine_186"], "convert_element_type_1117": ["quantize_affine_186", "dequantize_affine_372"], "view_2921": ["dequantize_affine_372"], "convert_element_type_1118": ["dequantize_affine_372"], "convert_element_type_1119": ["dequantize_affine_372"], "sub_8538": ["dequantize_affine_372"], "view_2920": ["dequantize_affine_372"], "mul_18074": ["dequantize_affine_372"], "view_2923": ["dequantize_affine_373"], "view_2925": ["dequantize_affine_373"], "convert_element_type_1120": ["dequantize_affine_373"], "convert_element_type_1121": ["dequantize_affine_373"], "sub_8542": ["dequantize_affine_373"], "view_2924": ["dequantize_affine_373"], "mul_18079": ["dequantize_affine_373"], "view_2926": ["dequantize_affine_373"], "amax_187": ["choose_qparams_affine_default_187"], "full_375": ["choose_qparams_affine_default_187"], "maximum_187": ["choose_qparams_affine_default_187"], "amin_187": ["choose_qparams_affine_default_187"], "full_374": ["choose_qparams_affine_default_187"], "minimum_187": ["choose_qparams_affine_default_187"], "sub_8557": ["choose_qparams_affine_default_187"], "div_374": ["choose_qparams_affine_default_187"], "clamp_min_561": ["choose_qparams_affine_default_187"], "div_375": ["choose_qparams_affine_default_187"], "round_375": ["choose_qparams_affine_default_187"], "sub_8563": ["choose_qparams_affine_default_187"], "clamp_min_562": ["choose_qparams_affine_default_187"], "clamp_max_374": ["choose_qparams_affine_default_187"], "view_2932": ["quantize_affine_187"], "reciprocal_187": ["quantize_affine_187"], "mul_18145": ["quantize_affine_187"], "mul_18148": ["quantize_affine_187"], "round_376": ["quantize_affine_187"], "convert_element_type_1122": ["choose_qparams_affine_default_187"], "view_2933": ["quantize_affine_187"], "add_28729": ["quantize_affine_187"], "clamp_min_563": ["quantize_affine_187"], "clamp_max_375": ["quantize_affine_187"], "convert_element_type_1123": ["quantize_affine_187", "dequantize_affine_374"], "view_2937": ["dequantize_affine_374"], "convert_element_type_1124": ["dequantize_affine_374"], "convert_element_type_1125": ["dequantize_affine_374"], "sub_8583": ["dequantize_affine_374"], "view_2936": ["dequantize_affine_374"], "mul_18170": ["dequantize_affine_374"], "view_2939": ["dequantize_affine_375"], "view_2941": ["dequantize_affine_375"], "convert_element_type_1126": ["dequantize_affine_375"], "convert_element_type_1127": ["dequantize_affine_375"], "sub_8587": ["dequantize_affine_375"], "view_2940": ["dequantize_affine_375"], "mul_18175": ["dequantize_affine_375"], "view_2942": ["dequantize_affine_375"], "amax_188": ["choose_qparams_affine_default_188"], "full_377": ["choose_qparams_affine_default_188"], "maximum_188": ["choose_qparams_affine_default_188"], "amin_188": ["choose_qparams_affine_default_188"], "full_376": ["choose_qparams_affine_default_188"], "minimum_188": ["choose_qparams_affine_default_188"], "sub_8601": ["choose_qparams_affine_default_188"], "div_376": ["choose_qparams_affine_default_188"], "clamp_min_564": ["choose_qparams_affine_default_188"], "div_377": ["choose_qparams_affine_default_188"], "round_377": ["choose_qparams_affine_default_188"], "sub_8607": ["choose_qparams_affine_default_188"], "clamp_min_565": ["choose_qparams_affine_default_188"], "clamp_max_376": ["choose_qparams_affine_default_188"], "view_2948": ["quantize_affine_188"], "reciprocal_188": ["quantize_affine_188"], "mul_18244": ["quantize_affine_188"], "mul_18247": ["quantize_affine_188"], "round_378": ["quantize_affine_188"], "convert_element_type_1128": ["choose_qparams_affine_default_188"], "view_2949": ["quantize_affine_188"], "add_28877": ["quantize_affine_188"], "clamp_min_566": ["quantize_affine_188"], "clamp_max_377": ["quantize_affine_188"], "convert_element_type_1129": ["quantize_affine_188", "dequantize_affine_376"], "view_2953": ["dequantize_affine_376"], "convert_element_type_1130": ["dequantize_affine_376"], "convert_element_type_1131": ["dequantize_affine_376"], "sub_8627": ["dequantize_affine_376"], "view_2952": ["dequantize_affine_376"], "mul_18269": ["dequantize_affine_376"], "view_2955": ["dequantize_affine_377"], "view_2957": ["dequantize_affine_377"], "convert_element_type_1132": ["dequantize_affine_377"], "convert_element_type_1133": ["dequantize_affine_377"], "sub_8631": ["dequantize_affine_377"], "view_2956": ["dequantize_affine_377"], "mul_18274": ["dequantize_affine_377"], "view_2958": ["dequantize_affine_377"], "mul_18084": ["linear_186"], "view_2927": ["linear_186"], "permute_311": ["linear_186"], "mm_default_4": ["linear_186"], "add_tensor_4": ["linear_186"], "view_2928": ["linear_186"], "mul_18091": ["mul_1152"], "view_2929": ["view_93"], "permute_312": ["transpose_124"], "clone_250": ["contiguous_124"], "mul_18178": ["linear_187"], "view_2943": ["linear_187"], "permute_313": ["linear_187"], "mm_31": ["linear_187"], "view_2944": ["linear_187"], "view_2945": ["view_94"], "permute_314": ["transpose_125"], "clone_251": ["contiguous_125"], "mul_18279": ["linear_188"], "view_2959": ["linear_188"], "permute_315": ["linear_188"], "mm_default_3": ["linear_188"], "add_tensor_3": ["linear_188"], "view_2960": ["linear_188"], "view_2961": ["view_95"], "permute_316": ["transpose_126"], "clone_252": ["contiguous_126"], "_scaled_dot_product_efficient_attention_31": ["scaled_dot_product_attention_31"], "getitem_250": ["scaled_dot_product_attention_31"], "permute_317": ["transpose_127"], "view_2962": ["reshape_31", "choose_qparams_affine_default_189", "quantize_affine_189"], "amax_189": ["choose_qparams_affine_default_189"], "full_379": ["choose_qparams_affine_default_189"], "maximum_189": ["choose_qparams_affine_default_189"], "amin_189": ["choose_qparams_affine_default_189"], "full_378": ["choose_qparams_affine_default_189"], "minimum_189": ["choose_qparams_affine_default_189"], "sub_8649": ["choose_qparams_affine_default_189"], "div_378": ["choose_qparams_affine_default_189"], "clamp_min_567": ["choose_qparams_affine_default_189"], "div_379": ["choose_qparams_affine_default_189"], "round_379": ["choose_qparams_affine_default_189"], "sub_8655": ["choose_qparams_affine_default_189"], "clamp_min_568": ["choose_qparams_affine_default_189"], "clamp_max_378": ["choose_qparams_affine_default_189"], "view_2965": ["quantize_affine_189"], "reciprocal_189": ["quantize_affine_189"], "mul_18349": ["quantize_affine_189"], "mul_18352": ["quantize_affine_189"], "round_380": ["quantize_affine_189"], "convert_element_type_1134": ["choose_qparams_affine_default_189"], "view_2966": ["quantize_affine_189"], "add_29041": ["quantize_affine_189"], "clamp_min_569": ["quantize_affine_189"], "clamp_max_379": ["quantize_affine_189"], "convert_element_type_1135": ["quantize_affine_189", "dequantize_affine_378"], "view_2970": ["dequantize_affine_378"], "convert_element_type_1136": ["dequantize_affine_378"], "convert_element_type_1137": ["dequantize_affine_378"], "sub_8675": ["dequantize_affine_378"], "view_2969": ["dequantize_affine_378"], "mul_18374": ["dequantize_affine_378"], "view_2972": ["dequantize_affine_379"], "view_2974": ["dequantize_affine_379"], "convert_element_type_1138": ["dequantize_affine_379"], "convert_element_type_1139": ["dequantize_affine_379"], "sub_8679": ["dequantize_affine_379"], "view_2973": ["dequantize_affine_379"], "mul_18379": ["dequantize_affine_379"], "view_2975": ["dequantize_affine_379"], "mul_18384": ["linear_189"], "view_2976": ["linear_189"], "permute_318": ["linear_189"], "mm_default_2": ["linear_189"], "add_tensor_2": ["linear_189"], "view_2977": ["linear_189", "dropout_94"], "add_29104": ["add_443"], "clone_254": ["layer_norm_63"], "var_mean_63": ["layer_norm_63"], "getitem_254": ["layer_norm_63"], "getitem_255": ["layer_norm_63"], "sub_8685": ["layer_norm_63"], "add_29109": ["layer_norm_63"], "rsqrt_63": ["layer_norm_63"], "mul_18395": ["layer_norm_63"], "mul_18396": ["layer_norm_63"], "add_29110": ["layer_norm_63", "choose_qparams_affine_default_190", "quantize_affine_190"], "amax_190": ["choose_qparams_affine_default_190"], "full_381": ["choose_qparams_affine_default_190"], "maximum_190": ["choose_qparams_affine_default_190"], "amin_190": ["choose_qparams_affine_default_190"], "full_380": ["choose_qparams_affine_default_190"], "minimum_190": ["choose_qparams_affine_default_190"], "sub_8696": ["choose_qparams_affine_default_190"], "div_380": ["choose_qparams_affine_default_190"], "clamp_min_570": ["choose_qparams_affine_default_190"], "div_381": ["choose_qparams_affine_default_190"], "round_381": ["choose_qparams_affine_default_190"], "sub_8702": ["choose_qparams_affine_default_190"], "clamp_min_571": ["choose_qparams_affine_default_190"], "clamp_max_380": ["choose_qparams_affine_default_190"], "view_2980": ["quantize_affine_190"], "reciprocal_190": ["quantize_affine_190"], "mul_18444": ["quantize_affine_190"], "mul_18447": ["quantize_affine_190"], "round_382": ["quantize_affine_190"], "convert_element_type_1140": ["choose_qparams_affine_default_190"], "view_2981": ["quantize_affine_190"], "add_29197": ["quantize_affine_190"], "clamp_min_572": ["quantize_affine_190"], "clamp_max_381": ["quantize_affine_190"], "convert_element_type_1141": ["quantize_affine_190", "dequantize_affine_380"], "view_2985": ["dequantize_affine_380"], "convert_element_type_1142": ["dequantize_affine_380"], "convert_element_type_1143": ["dequantize_affine_380"], "sub_8722": ["dequantize_affine_380"], "view_2984": ["dequantize_affine_380"], "mul_18469": ["dequantize_affine_380"], "view_2987": ["dequantize_affine_381"], "view_2989": ["dequantize_affine_381"], "convert_element_type_1144": ["dequantize_affine_381"], "convert_element_type_1145": ["dequantize_affine_381"], "sub_8726": ["dequantize_affine_381"], "view_2988": ["dequantize_affine_381"], "mul_18474": ["dequantize_affine_381"], "view_2990": ["dequantize_affine_381"], "mul_18479": ["linear_190"], "view_2991": ["linear_190"], "permute_319": ["linear_190"], "mm_default_1": ["linear_190"], "add_tensor_1": ["linear_190"], "view_2992": ["linear_190"], "mul_18486": ["gelu_33"], "mul_18487": ["gelu_33"], "erf_33": ["gelu_33"], "add_29256": ["gelu_33"], "mul_18488": ["gelu_33", "dropout_95", "choose_qparams_affine_default_191", "quantize_affine_191"], "amax_191": ["choose_qparams_affine_default_191"], "full_383": ["choose_qparams_affine_default_191"], "maximum_191": ["choose_qparams_affine_default_191"], "amin_191": ["choose_qparams_affine_default_191"], "full_382": ["choose_qparams_affine_default_191"], "minimum_191": ["choose_qparams_affine_default_191"], "sub_8739": ["choose_qparams_affine_default_191"], "div_382": ["choose_qparams_affine_default_191"], "clamp_min_573": ["choose_qparams_affine_default_191"], "div_383": ["choose_qparams_affine_default_191"], "round_383": ["choose_qparams_affine_default_191"], "sub_8745": ["choose_qparams_affine_default_191"], "clamp_min_574": ["choose_qparams_affine_default_191"], "clamp_max_382": ["choose_qparams_affine_default_191"], "view_2995": ["quantize_affine_191"], "reciprocal_191": ["quantize_affine_191"], "mul_18534": ["quantize_affine_191"], "mul_18537": ["quantize_affine_191"], "round_384": ["quantize_affine_191"], "convert_element_type_1146": ["choose_qparams_affine_default_191"], "view_2996": ["quantize_affine_191"], "add_29339": ["quantize_affine_191"], "clamp_min_575": ["quantize_affine_191"], "clamp_max_383": ["quantize_affine_191"], "convert_element_type_1147": ["quantize_affine_191", "dequantize_affine_382"], "view_3000": ["dequantize_affine_382"], "convert_element_type_1148": ["dequantize_affine_382"], "convert_element_type_1149": ["dequantize_affine_382"], "sub_8765": ["dequantize_affine_382"], "view_2999": ["dequantize_affine_382"], "mul_18559": ["dequantize_affine_382"], "view_3002": ["dequantize_affine_383"], "view_3004": ["dequantize_affine_383"], "convert_element_type_1150": ["dequantize_affine_383"], "convert_element_type_1151": ["dequantize_affine_383"], "sub_8769": ["dequantize_affine_383"], "view_3003": ["dequantize_affine_383"], "mul_18564": ["dequantize_affine_383"], "view_3005": ["dequantize_affine_383"], "mul_18569": ["linear_191"], "view_3006": ["linear_191"], "permute_320": ["linear_191"], "mm_default": ["linear_191"], "add_tensor": ["linear_191"], "view_3007": ["linear_191", "dropout_96"], "add_29402": ["add_448"], "clone_257": ["layer_norm_64"], "var_mean_64": ["layer_norm_64"], "getitem_256": ["layer_norm_64"], "getitem_257": ["layer_norm_64"], "sub_8775": ["layer_norm_64"], "add_29407": ["layer_norm_64"], "rsqrt_64": ["layer_norm_64"], "mul_18580": ["layer_norm_64"], "mul_18581": ["layer_norm_64"], "add_29408": ["layer_norm_64"], "view_3008": ["reshape_32", "choose_qparams_affine_default_192", "quantize_affine_192"], "amax_192": ["choose_qparams_affine_default_192"], "sym_size_int_193": ["reshape_32"], "full_385": ["choose_qparams_affine_default_192"], "maximum_192": ["choose_qparams_affine_default_192"], "amin_192": ["choose_qparams_affine_default_192"], "full_384": ["choose_qparams_affine_default_192"], "minimum_192": ["choose_qparams_affine_default_192"], "sub_8787": ["choose_qparams_affine_default_192"], "div_384": ["choose_qparams_affine_default_192"], "clamp_min_576": ["choose_qparams_affine_default_192"], "div_385": ["choose_qparams_affine_default_192"], "round_385": ["choose_qparams_affine_default_192"], "sub_8793": ["choose_qparams_affine_default_192"], "clamp_min_577": ["choose_qparams_affine_default_192"], "clamp_max_384": ["choose_qparams_affine_default_192"], "view_3011": ["quantize_affine_192"], "reciprocal_192": ["quantize_affine_192"], "mul_18613": ["quantize_affine_192"], "mul_18615": ["quantize_affine_192"], "round_386": ["quantize_affine_192"], "convert_element_type_1152": ["choose_qparams_affine_default_192"], "view_3012": ["quantize_affine_192"], "add_29476": ["quantize_affine_192"], "clamp_min_578": ["quantize_affine_192"], "clamp_max_385": ["quantize_affine_192"], "convert_element_type_1153": ["quantize_affine_192", "dequantize_affine_384"], "view_3016": ["dequantize_affine_384"], "convert_element_type_1154": ["dequantize_affine_384"], "convert_element_type_1155": ["dequantize_affine_384"], "sub_8813": ["dequantize_affine_384"], "view_3015": ["dequantize_affine_384"], "mul_18634": ["dequantize_affine_384"], "view_3018": ["dequantize_affine_385"], "view_3020": ["dequantize_affine_385"], "convert_element_type_1156": ["dequantize_affine_385"], "convert_element_type_1157": ["dequantize_affine_385"], "sub_8817": ["dequantize_affine_385"], "view_3019": ["dequantize_affine_385"], "mul_18639": ["dequantize_affine_385"], "view_3021": ["dequantize_affine_385"], "permute_321": ["linear_192"], "mm_32": ["linear_192"], "mul_18642": ["gelu_34"], "mul_18643": ["gelu_34"], "erf_34": ["gelu_34"], "add_29516": ["gelu_34"], "mul_18644": ["gelu_34", "choose_qparams_affine_default_193", "quantize_affine_193"], "amax_193": ["choose_qparams_affine_default_193"], "full_387": ["choose_qparams_affine_default_193"], "maximum_193": ["choose_qparams_affine_default_193"], "amin_193": ["choose_qparams_affine_default_193"], "full_386": ["choose_qparams_affine_default_193"], "minimum_193": ["choose_qparams_affine_default_193"], "sub_8827": ["choose_qparams_affine_default_193"], "div_386": ["choose_qparams_affine_default_193"], "clamp_min_579": ["choose_qparams_affine_default_193"], "div_387": ["choose_qparams_affine_default_193"], "round_387": ["choose_qparams_affine_default_193"], "sub_8833": ["choose_qparams_affine_default_193"], "clamp_min_580": ["choose_qparams_affine_default_193"], "clamp_max_386": ["choose_qparams_affine_default_193"], "view_3024": ["quantize_affine_193"], "reciprocal_193": ["quantize_affine_193"], "mul_18666": ["quantize_affine_193"], "mul_18668": ["quantize_affine_193"], "round_388": ["quantize_affine_193"], "convert_element_type_1158": ["choose_qparams_affine_default_193"], "view_3025": ["quantize_affine_193"], "add_29572": ["quantize_affine_193"], "clamp_min_581": ["quantize_affine_193"], "clamp_max_387": ["quantize_affine_193"], "convert_element_type_1159": ["quantize_affine_193", "dequantize_affine_386"], "view_3029": ["dequantize_affine_386"], "convert_element_type_1160": ["dequantize_affine_386"], "convert_element_type_1161": ["dequantize_affine_386"], "sub_8853": ["dequantize_affine_386"], "view_3028": ["dequantize_affine_386"], "mul_18687": ["dequantize_affine_386"], "view_3031": ["dequantize_affine_387"], "view_3033": ["dequantize_affine_387"], "convert_element_type_1162": ["dequantize_affine_387"], "convert_element_type_1163": ["dequantize_affine_387"], "sub_8857": ["dequantize_affine_387"], "view_3032": ["dequantize_affine_387"], "mul_18692": ["dequantize_affine_387"], "view_3034": ["dequantize_affine_387"], "permute_322": ["linear_193"], "mm_33": ["linear_193"], "unsqueeze": ["unsqueeze"]}, "cppCodeToPost": {"triton_poi_fused_convolution_gelu_0:1": ["convolution_1", "mul_4", "mul_2", "convolution", "add_4", "erf", "mul_3"], "triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1:2": ["var_mean", "clone_1", "add_22", "permute", "mul_11", "mul_9", "convolution_1", "mul_4", "mul_2", "convolution", "add_4", "erf", "mul_3", "add_13", "erf_1", "mul_10"], "triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2:3": ["var_mean", "clone_1", "add_22", "permute", "mul_11", "mul_9", "convolution_1", "mul_4", "mul_2", "convolution", "add_4", "erf", "mul_3", "add_13", "erf_1", "mul_10"], "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4": ["add_32", "mul_21", "mul_20", "sub_7", "clone_1", "add_22", "permute", "mul_11", "mul_9", "convolution_1", "mul_4", "mul_2", "convolution", "add_4", "erf", "mul_3", "add_13", "erf_1", "mul_10", "rsqrt", "add_31", "var_mean", "amax", "amin", "amax_1", "amin_1", "amax_2", "amin_2", "sub_44", "convert_element_type_2", "convert_element_type_1", "clamp_max_1", "clamp_min_2", "add_119", "round_2", "mul_72", "mul_69", "reciprocal", "view_2", "clamp_min", "div", "sub_18", "maximum", "full_1", "minimum", "full", "view_3", "convert_element_type", "clamp_max", "clamp_min_1", "sub_24", "round_1", "div_1", "convert_element_type_3", "view_7", "mul_94", "view_6", "sub_89", "convert_element_type_8", "convert_element_type_7", "clamp_max_3", "clamp_min_5", "add_271", "round_4", "mul_168", "mul_165", "reciprocal_1", "view_18", "clamp_min_3", "div_2", "sub_63", "maximum_1", "full_3", "minimum_1", "full_2", "view_19", "convert_element_type_6", "clamp_max_2", "clamp_min_4", "sub_69", "round_3", "div_3", "convert_element_type_9", "view_23", "mul_190", "view_22", "sub_133", "convert_element_type_14", "convert_element_type_13", "clamp_max_5", "clamp_min_8", "add_419", "round_6", "mul_267", "mul_264", "reciprocal_2", "view_34", "clamp_min_6", "div_4", "sub_107", "maximum_2", "full_5", "minimum_2", "full_4", "view_35", "convert_element_type_12", "clamp_max_4", "clamp_min_7", "sub_113", "round_5", "div_5", "convert_element_type_15", "view_39", "mul_289", "view_38"], "triton_poi_fused__to_copy_mul_sub_view_4:5": ["mul_99", "sub_48", "convert_element_type_4", "view_9", "convert_element_type_5", "view_11", "view_10"], "triton_poi_fused__to_copy_mul_sub_view_4:6": ["mul_195", "sub_93", "convert_element_type_10", "view_25", "convert_element_type_11", "view_27", "view_26"], "triton_poi_fused__to_copy_mul_sub_view_4:7": ["mul_294", "sub_137", "convert_element_type_16", "view_41", "convert_element_type_17", "view_43", "view_42"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:8": ["_scaled_dot_product_efficient_attention", "clone_2", "permute_2", "view_15", "mul_111", "view_14", "add_tensor_159", "clone_3", "permute_4", "view_31", "view_30", "clone_4", "permute_6", "view_47", "view_46", "add_tensor_158"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:9": ["_scaled_dot_product_efficient_attention", "clone_2", "permute_2", "view_15", "mul_111", "view_14", "add_tensor_159", "clone_3", "permute_4", "view_31", "view_30", "clone_4", "permute_6", "view_47", "view_46", "add_tensor_158"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:10": ["_scaled_dot_product_efficient_attention", "clone_2", "permute_2", "view_15", "mul_111", "view_14", "add_tensor_159", "clone_3", "permute_4", "view_31", "view_30", "clone_4", "permute_6", "view_47", "view_46", "add_tensor_158"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11": ["amax_3", "view_48", "permute_7", "amin_3", "sub_181", "convert_element_type_20", "convert_element_type_19", "clamp_max_7", "clamp_min_11", "add_583", "round_8", "mul_372", "mul_369", "reciprocal_3", "view_51", "clamp_min_9", "div_6", "sub_155", "maximum_3", "full_7", "minimum_3", "full_6", "view_52", "convert_element_type_18", "clamp_max_6", "clamp_min_10", "sub_161", "round_7", "div_7", "convert_element_type_21", "view_56", "mul_394", "view_55"], "triton_poi_fused__to_copy_mul_sub_view_4:12": ["mul_399", "sub_185", "convert_element_type_22", "view_58", "convert_element_type_23", "view_60", "view_59"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13": ["add_646", "add_22", "permute", "mul_11", "mul_9", "convolution_1", "mul_4", "mul_2", "convolution", "add_4", "erf", "mul_3", "add_13", "erf_1", "mul_10", "view_63", "add_tensor_157", "var_mean_1", "clone_6", "add_652", "mul_416", "mul_415", "sub_191", "rsqrt_1", "add_651", "amax_4", "amin_4", "sub_228", "convert_element_type_26", "convert_element_type_25", "clamp_max_9", "clamp_min_14", "add_739", "round_10", "mul_467", "mul_464", "reciprocal_4", "view_66", "clamp_min_12", "div_8", "sub_202", "maximum_4", "full_9", "minimum_4", "full_8", "view_67", "convert_element_type_24", "clamp_max_8", "clamp_min_13", "sub_208", "round_9", "div_9", "convert_element_type_27", "view_71", "mul_489", "view_70"], "triton_poi_fused__to_copy_mul_sub_view_10:14": ["mul_494", "sub_232", "convert_element_type_28", "view_73", "convert_element_type_29", "view_75", "view_74"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15": ["amax_5", "mul_508", "mul_506", "view_78", "add_tensor_156", "add_798", "erf_2", "mul_507", "amin_5", "add_881", "round_12", "mul_557", "mul_554", "reciprocal_5", "view_81", "clamp_min_15", "div_10", "sub_245", "maximum_5", "full_11", "minimum_5", "full_10", "view_82", "convert_element_type_30", "clamp_max_10", "clamp_min_16", "sub_251", "round_11", "div_11", "mul_579", "sub_271", "convert_element_type_32", "convert_element_type_31", "clamp_max_11", "clamp_min_17", "convert_element_type_33", "view_86", "view_85"], "triton_poi_fused__to_copy_mul_sub_view_10:16": ["mul_584", "sub_275", "convert_element_type_34", "view_88", "convert_element_type_35", "view_90", "view_89"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17": ["var_mean_2", "clone_9", "add_944", "view_93", "add_tensor_155", "add_950", "mul_601", "mul_600", "sub_281", "rsqrt_2", "add_949", "amax_6", "amin_6", "amax_7", "amin_7", "amax_8", "amin_8", "sub_318", "convert_element_type_38", "convert_element_type_37", "clamp_max_13", "clamp_min_20", "add_1037", "round_14", "mul_652", "mul_649", "reciprocal_6", "view_96", "clamp_min_18", "div_12", "sub_292", "maximum_6", "full_13", "minimum_6", "full_12", "view_97", "convert_element_type_36", "clamp_max_12", "clamp_min_19", "sub_298", "round_13", "div_13", "convert_element_type_39", "view_101", "mul_674", "view_100", "sub_363", "convert_element_type_44", "convert_element_type_43", "clamp_max_15", "clamp_min_23", "add_1189", "round_16", "mul_748", "mul_745", "reciprocal_7", "view_112", "clamp_min_21", "div_14", "sub_337", "maximum_7", "full_15", "minimum_7", "full_14", "view_113", "convert_element_type_42", "clamp_max_14", "clamp_min_22", "sub_343", "round_15", "div_15", "convert_element_type_45", "view_117", "mul_770", "view_116", "sub_407", "convert_element_type_50", "convert_element_type_49", "clamp_max_17", "clamp_min_26", "add_1337", "round_18", "mul_847", "mul_844", "reciprocal_8", "view_128", "clamp_min_24", "div_16", "sub_381", "maximum_8", "full_17", "minimum_8", "full_16", "view_129", "convert_element_type_48", "clamp_max_16", "clamp_min_25", "sub_387", "round_17", "div_17", "convert_element_type_51", "view_133", "mul_869", "view_132"], "triton_poi_fused__to_copy_mul_sub_view_4:18": ["mul_679", "sub_322", "convert_element_type_40", "view_103", "convert_element_type_41", "view_105", "view_104"], "triton_poi_fused__to_copy_mul_sub_view_4:19": ["mul_775", "sub_367", "convert_element_type_46", "view_119", "convert_element_type_47", "view_121", "view_120"], "triton_poi_fused__to_copy_mul_sub_view_4:20": ["mul_874", "sub_411", "convert_element_type_52", "view_135", "convert_element_type_53", "view_137", "view_136"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:21": ["_scaled_dot_product_efficient_attention_1", "clone_10", "permute_12", "view_109", "mul_691", "view_108", "add_tensor_154", "clone_11", "permute_14", "view_125", "view_124", "clone_12", "permute_16", "view_141", "view_140", "add_tensor_153"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:22": ["_scaled_dot_product_efficient_attention_1", "clone_10", "permute_12", "view_109", "mul_691", "view_108", "add_tensor_154", "clone_11", "permute_14", "view_125", "view_124", "clone_12", "permute_16", "view_141", "view_140", "add_tensor_153"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:23": ["_scaled_dot_product_efficient_attention_1", "clone_10", "permute_12", "view_109", "mul_691", "view_108", "add_tensor_154", "clone_11", "permute_14", "view_125", "view_124", "clone_12", "permute_16", "view_141", "view_140", "add_tensor_153"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24": ["amax_9", "view_142", "permute_17", "amin_9", "sub_455", "convert_element_type_56", "convert_element_type_55", "clamp_max_19", "clamp_min_29", "add_1501", "round_20", "mul_952", "mul_949", "reciprocal_9", "view_145", "clamp_min_27", "div_18", "sub_429", "maximum_9", "full_19", "minimum_9", "full_18", "view_146", "convert_element_type_54", "clamp_max_18", "clamp_min_28", "sub_435", "round_19", "div_19", "convert_element_type_57", "view_150", "mul_974", "view_149"], "triton_poi_fused__to_copy_mul_sub_view_4:25": ["mul_979", "sub_459", "convert_element_type_58", "view_152", "convert_element_type_59", "view_154", "view_153"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26": ["add_1564", "add_944", "view_93", "add_tensor_155", "view_157", "add_tensor_152", "var_mean_3", "clone_14", "add_1570", "mul_996", "mul_995", "sub_465", "rsqrt_3", "add_1569", "amax_10", "amin_10", "sub_502", "convert_element_type_62", "convert_element_type_61", "clamp_max_21", "clamp_min_32", "add_1657", "round_22", "mul_1047", "mul_1044", "reciprocal_10", "view_160", "clamp_min_30", "div_20", "sub_476", "maximum_10", "full_21", "minimum_10", "full_20", "view_161", "convert_element_type_60", "clamp_max_20", "clamp_min_31", "sub_482", "round_21", "div_21", "convert_element_type_63", "view_165", "mul_1069", "view_164"], "triton_poi_fused__to_copy_mul_sub_view_10:27": ["mul_1074", "sub_506", "convert_element_type_64", "view_167", "convert_element_type_65", "view_169", "view_168"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28": ["amax_11", "mul_1088", "mul_1086", "view_172", "add_tensor_151", "add_1716", "erf_3", "mul_1087", "amin_11", "add_1799", "round_24", "mul_1137", "mul_1134", "reciprocal_11", "view_175", "clamp_min_33", "div_22", "sub_519", "maximum_11", "full_23", "minimum_11", "full_22", "view_176", "convert_element_type_66", "clamp_max_22", "clamp_min_34", "sub_525", "round_23", "div_23", "mul_1159", "sub_545", "convert_element_type_68", "convert_element_type_67", "clamp_max_23", "clamp_min_35", "convert_element_type_69", "view_180", "view_179"], "triton_poi_fused__to_copy_mul_sub_view_10:29": ["mul_1164", "sub_549", "convert_element_type_70", "view_182", "convert_element_type_71", "view_184", "view_183"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30": ["var_mean_4", "clone_17", "add_1862", "view_187", "add_tensor_150", "add_1868", "mul_1181", "mul_1180", "sub_555", "rsqrt_4", "add_1867", "amax_12", "amin_12", "amax_13", "amin_13", "amax_14", "amin_14", "sub_592", "convert_element_type_74", "convert_element_type_73", "clamp_max_25", "clamp_min_38", "add_1955", "round_26", "mul_1232", "mul_1229", "reciprocal_12", "view_190", "clamp_min_36", "div_24", "sub_566", "maximum_12", "full_25", "minimum_12", "full_24", "view_191", "convert_element_type_72", "clamp_max_24", "clamp_min_37", "sub_572", "round_25", "div_25", "convert_element_type_75", "view_195", "mul_1254", "view_194", "sub_637", "convert_element_type_80", "convert_element_type_79", "clamp_max_27", "clamp_min_41", "add_2107", "round_28", "mul_1328", "mul_1325", "reciprocal_13", "view_206", "clamp_min_39", "div_26", "sub_611", "maximum_13", "full_27", "minimum_13", "full_26", "view_207", "convert_element_type_78", "clamp_max_26", "clamp_min_40", "sub_617", "round_27", "div_27", "convert_element_type_81", "view_211", "mul_1350", "view_210", "sub_681", "convert_element_type_86", "convert_element_type_85", "clamp_max_29", "clamp_min_44", "add_2255", "round_30", "mul_1427", "mul_1424", "reciprocal_14", "view_222", "clamp_min_42", "div_28", "sub_655", "maximum_14", "full_29", "minimum_14", "full_28", "view_223", "convert_element_type_84", "clamp_max_28", "clamp_min_43", "sub_661", "round_29", "div_29", "convert_element_type_87", "view_227", "mul_1449", "view_226"], "triton_poi_fused__to_copy_mul_sub_view_4:31": ["mul_1259", "sub_596", "convert_element_type_76", "view_197", "convert_element_type_77", "view_199", "view_198"], "triton_poi_fused__to_copy_mul_sub_view_4:32": ["mul_1355", "sub_641", "convert_element_type_82", "view_213", "convert_element_type_83", "view_215", "view_214"], "triton_poi_fused__to_copy_mul_sub_view_4:33": ["mul_1454", "sub_685", "convert_element_type_88", "view_229", "convert_element_type_89", "view_231", "view_230"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:34": ["_scaled_dot_product_efficient_attention_2", "clone_18", "permute_22", "view_203", "mul_1271", "view_202", "add_tensor_149", "clone_19", "permute_24", "view_219", "view_218", "clone_20", "permute_26", "view_235", "view_234", "add_tensor_148"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:35": ["_scaled_dot_product_efficient_attention_2", "clone_18", "permute_22", "view_203", "mul_1271", "view_202", "add_tensor_149", "clone_19", "permute_24", "view_219", "view_218", "clone_20", "permute_26", "view_235", "view_234", "add_tensor_148"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:36": ["_scaled_dot_product_efficient_attention_2", "clone_18", "permute_22", "view_203", "mul_1271", "view_202", "add_tensor_149", "clone_19", "permute_24", "view_219", "view_218", "clone_20", "permute_26", "view_235", "view_234", "add_tensor_148"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37": ["amax_15", "view_236", "permute_27", "amin_15", "sub_729", "convert_element_type_92", "convert_element_type_91", "clamp_max_31", "clamp_min_47", "add_2419", "round_32", "mul_1532", "mul_1529", "reciprocal_15", "view_239", "clamp_min_45", "div_30", "sub_703", "maximum_15", "full_31", "minimum_15", "full_30", "view_240", "convert_element_type_90", "clamp_max_30", "clamp_min_46", "sub_709", "round_31", "div_31", "convert_element_type_93", "view_244", "mul_1554", "view_243"], "triton_poi_fused__to_copy_mul_sub_view_4:38": ["mul_1559", "sub_733", "convert_element_type_94", "view_246", "convert_element_type_95", "view_248", "view_247"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39": ["add_2482", "add_1862", "view_187", "add_tensor_150", "view_251", "add_tensor_147", "var_mean_5", "clone_22", "add_2488", "mul_1576", "mul_1575", "sub_739", "rsqrt_5", "add_2487", "amax_16", "amin_16", "sub_776", "convert_element_type_98", "convert_element_type_97", "clamp_max_33", "clamp_min_50", "add_2575", "round_34", "mul_1627", "mul_1624", "reciprocal_16", "view_254", "clamp_min_48", "div_32", "sub_750", "maximum_16", "full_33", "minimum_16", "full_32", "view_255", "convert_element_type_96", "clamp_max_32", "clamp_min_49", "sub_756", "round_33", "div_33", "convert_element_type_99", "view_259", "mul_1649", "view_258"], "triton_poi_fused__to_copy_mul_sub_view_10:40": ["mul_1654", "sub_780", "convert_element_type_100", "view_261", "convert_element_type_101", "view_263", "view_262"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41": ["amax_17", "mul_1668", "mul_1666", "view_266", "add_tensor_146", "add_2634", "erf_4", "mul_1667", "amin_17", "add_2717", "round_36", "mul_1717", "mul_1714", "reciprocal_17", "view_269", "clamp_min_51", "div_34", "sub_793", "maximum_17", "full_35", "minimum_17", "full_34", "view_270", "convert_element_type_102", "clamp_max_34", "clamp_min_52", "sub_799", "round_35", "div_35", "mul_1739", "sub_819", "convert_element_type_104", "convert_element_type_103", "clamp_max_35", "clamp_min_53", "convert_element_type_105", "view_274", "view_273"], "triton_poi_fused__to_copy_mul_sub_view_10:42": ["mul_1744", "sub_823", "convert_element_type_106", "view_276", "convert_element_type_107", "view_278", "view_277"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43": ["var_mean_6", "clone_25", "add_2780", "view_281", "add_tensor_145", "add_2786", "mul_1761", "mul_1760", "sub_829", "rsqrt_6", "add_2785", "amax_18", "amin_18", "amax_19", "amin_19", "amax_20", "amin_20", "sub_866", "convert_element_type_110", "convert_element_type_109", "clamp_max_37", "clamp_min_56", "add_2873", "round_38", "mul_1812", "mul_1809", "reciprocal_18", "view_284", "clamp_min_54", "div_36", "sub_840", "maximum_18", "full_37", "minimum_18", "full_36", "view_285", "convert_element_type_108", "clamp_max_36", "clamp_min_55", "sub_846", "round_37", "div_37", "convert_element_type_111", "view_289", "mul_1834", "view_288", "sub_911", "convert_element_type_116", "convert_element_type_115", "clamp_max_39", "clamp_min_59", "add_3025", "round_40", "mul_1908", "mul_1905", "reciprocal_19", "view_300", "clamp_min_57", "div_38", "sub_885", "maximum_19", "full_39", "minimum_19", "full_38", "view_301", "convert_element_type_114", "clamp_max_38", "clamp_min_58", "sub_891", "round_39", "div_39", "convert_element_type_117", "view_305", "mul_1930", "view_304", "sub_955", "convert_element_type_122", "convert_element_type_121", "clamp_max_41", "clamp_min_62", "add_3173", "round_42", "mul_2007", "mul_2004", "reciprocal_20", "view_316", "clamp_min_60", "div_40", "sub_929", "maximum_20", "full_41", "minimum_20", "full_40", "view_317", "convert_element_type_120", "clamp_max_40", "clamp_min_61", "sub_935", "round_41", "div_41", "convert_element_type_123", "view_321", "mul_2029", "view_320"], "triton_poi_fused__to_copy_mul_sub_view_4:44": ["mul_1839", "sub_870", "convert_element_type_112", "view_291", "convert_element_type_113", "view_293", "view_292"], "triton_poi_fused__to_copy_mul_sub_view_4:45": ["mul_1935", "sub_915", "convert_element_type_118", "view_307", "convert_element_type_119", "view_309", "view_308"], "triton_poi_fused__to_copy_mul_sub_view_4:46": ["mul_2034", "sub_959", "convert_element_type_124", "view_323", "convert_element_type_125", "view_325", "view_324"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:47": ["_scaled_dot_product_efficient_attention_3", "clone_26", "permute_32", "view_297", "mul_1851", "view_296", "add_tensor_144", "clone_27", "permute_34", "view_313", "view_312", "clone_28", "permute_36", "view_329", "view_328", "add_tensor_143"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:48": ["_scaled_dot_product_efficient_attention_3", "clone_26", "permute_32", "view_297", "mul_1851", "view_296", "add_tensor_144", "clone_27", "permute_34", "view_313", "view_312", "clone_28", "permute_36", "view_329", "view_328", "add_tensor_143"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:49": ["_scaled_dot_product_efficient_attention_3", "clone_26", "permute_32", "view_297", "mul_1851", "view_296", "add_tensor_144", "clone_27", "permute_34", "view_313", "view_312", "clone_28", "permute_36", "view_329", "view_328", "add_tensor_143"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50": ["amax_21", "view_330", "permute_37", "amin_21", "sub_1003", "convert_element_type_128", "convert_element_type_127", "clamp_max_43", "clamp_min_65", "add_3337", "round_44", "mul_2112", "mul_2109", "reciprocal_21", "view_333", "clamp_min_63", "div_42", "sub_977", "maximum_21", "full_43", "minimum_21", "full_42", "view_334", "convert_element_type_126", "clamp_max_42", "clamp_min_64", "sub_983", "round_43", "div_43", "convert_element_type_129", "view_338", "mul_2134", "view_337"], "triton_poi_fused__to_copy_mul_sub_view_4:51": ["mul_2139", "sub_1007", "convert_element_type_130", "view_340", "convert_element_type_131", "view_342", "view_341"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52": ["add_3400", "add_2780", "view_281", "add_tensor_145", "view_345", "add_tensor_142", "var_mean_7", "clone_30", "add_3406", "mul_2156", "mul_2155", "sub_1013", "rsqrt_7", "add_3405", "amax_22", "amin_22", "sub_1050", "convert_element_type_134", "convert_element_type_133", "clamp_max_45", "clamp_min_68", "add_3493", "round_46", "mul_2207", "mul_2204", "reciprocal_22", "view_348", "clamp_min_66", "div_44", "sub_1024", "maximum_22", "full_45", "minimum_22", "full_44", "view_349", "convert_element_type_132", "clamp_max_44", "clamp_min_67", "sub_1030", "round_45", "div_45", "convert_element_type_135", "view_353", "mul_2229", "view_352"], "triton_poi_fused__to_copy_mul_sub_view_10:53": ["mul_2234", "sub_1054", "convert_element_type_136", "view_355", "convert_element_type_137", "view_357", "view_356"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54": ["amax_23", "mul_2248", "mul_2246", "view_360", "add_tensor_141", "add_3552", "erf_5", "mul_2247", "amin_23", "add_3635", "round_48", "mul_2297", "mul_2294", "reciprocal_23", "view_363", "clamp_min_69", "div_46", "sub_1067", "maximum_23", "full_47", "minimum_23", "full_46", "view_364", "convert_element_type_138", "clamp_max_46", "clamp_min_70", "sub_1073", "round_47", "div_47", "mul_2319", "sub_1093", "convert_element_type_140", "convert_element_type_139", "clamp_max_47", "clamp_min_71", "convert_element_type_141", "view_368", "view_367"], "triton_poi_fused__to_copy_mul_sub_view_10:55": ["mul_2324", "sub_1097", "convert_element_type_142", "view_370", "convert_element_type_143", "view_372", "view_371"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56": ["var_mean_8", "clone_33", "add_3698", "view_375", "add_tensor_140", "add_3704", "mul_2341", "mul_2340", "sub_1103", "rsqrt_8", "add_3703", "amax_24", "amin_24", "amax_25", "amin_25", "amax_26", "amin_26", "sub_1140", "convert_element_type_146", "convert_element_type_145", "clamp_max_49", "clamp_min_74", "add_3791", "round_50", "mul_2392", "mul_2389", "reciprocal_24", "view_378", "clamp_min_72", "div_48", "sub_1114", "maximum_24", "full_49", "minimum_24", "full_48", "view_379", "convert_element_type_144", "clamp_max_48", "clamp_min_73", "sub_1120", "round_49", "div_49", "convert_element_type_147", "view_383", "mul_2414", "view_382", "sub_1185", "convert_element_type_152", "convert_element_type_151", "clamp_max_51", "clamp_min_77", "add_3943", "round_52", "mul_2488", "mul_2485", "reciprocal_25", "view_394", "clamp_min_75", "div_50", "sub_1159", "maximum_25", "full_51", "minimum_25", "full_50", "view_395", "convert_element_type_150", "clamp_max_50", "clamp_min_76", "sub_1165", "round_51", "div_51", "convert_element_type_153", "view_399", "mul_2510", "view_398", "sub_1229", "convert_element_type_158", "convert_element_type_157", "clamp_max_53", "clamp_min_80", "add_4091", "round_54", "mul_2587", "mul_2584", "reciprocal_26", "view_410", "clamp_min_78", "div_52", "sub_1203", "maximum_26", "full_53", "minimum_26", "full_52", "view_411", "convert_element_type_156", "clamp_max_52", "clamp_min_79", "sub_1209", "round_53", "div_53", "convert_element_type_159", "view_415", "mul_2609", "view_414"], "triton_poi_fused__to_copy_mul_sub_view_4:57": ["mul_2419", "sub_1144", "convert_element_type_148", "view_385", "convert_element_type_149", "view_387", "view_386"], "triton_poi_fused__to_copy_mul_sub_view_4:58": ["mul_2515", "sub_1189", "convert_element_type_154", "view_401", "convert_element_type_155", "view_403", "view_402"], "triton_poi_fused__to_copy_mul_sub_view_4:59": ["mul_2614", "sub_1233", "convert_element_type_160", "view_417", "convert_element_type_161", "view_419", "view_418"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:60": ["_scaled_dot_product_efficient_attention_4", "clone_34", "permute_42", "view_391", "mul_2431", "view_390", "add_tensor_139", "clone_35", "permute_44", "view_407", "view_406", "clone_36", "permute_46", "view_423", "view_422", "add_tensor_138"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:61": ["_scaled_dot_product_efficient_attention_4", "clone_34", "permute_42", "view_391", "mul_2431", "view_390", "add_tensor_139", "clone_35", "permute_44", "view_407", "view_406", "clone_36", "permute_46", "view_423", "view_422", "add_tensor_138"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:62": ["_scaled_dot_product_efficient_attention_4", "clone_34", "permute_42", "view_391", "mul_2431", "view_390", "add_tensor_139", "clone_35", "permute_44", "view_407", "view_406", "clone_36", "permute_46", "view_423", "view_422", "add_tensor_138"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63": ["amax_27", "view_424", "permute_47", "amin_27", "sub_1277", "convert_element_type_164", "convert_element_type_163", "clamp_max_55", "clamp_min_83", "add_4255", "round_56", "mul_2692", "mul_2689", "reciprocal_27", "view_427", "clamp_min_81", "div_54", "sub_1251", "maximum_27", "full_55", "minimum_27", "full_54", "view_428", "convert_element_type_162", "clamp_max_54", "clamp_min_82", "sub_1257", "round_55", "div_55", "convert_element_type_165", "view_432", "mul_2714", "view_431"], "triton_poi_fused__to_copy_mul_sub_view_4:64": ["mul_2719", "sub_1281", "convert_element_type_166", "view_434", "convert_element_type_167", "view_436", "view_435"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65": ["add_4318", "add_3698", "view_375", "add_tensor_140", "view_439", "add_tensor_137", "var_mean_9", "clone_38", "add_4324", "mul_2736", "mul_2735", "sub_1287", "rsqrt_9", "add_4323", "amax_28", "amin_28", "sub_1324", "convert_element_type_170", "convert_element_type_169", "clamp_max_57", "clamp_min_86", "add_4411", "round_58", "mul_2787", "mul_2784", "reciprocal_28", "view_442", "clamp_min_84", "div_56", "sub_1298", "maximum_28", "full_57", "minimum_28", "full_56", "view_443", "convert_element_type_168", "clamp_max_56", "clamp_min_85", "sub_1304", "round_57", "div_57", "convert_element_type_171", "view_447", "mul_2809", "view_446"], "triton_poi_fused__to_copy_mul_sub_view_10:66": ["mul_2814", "sub_1328", "convert_element_type_172", "view_449", "convert_element_type_173", "view_451", "view_450"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67": ["amax_29", "mul_2828", "mul_2826", "view_454", "add_tensor_136", "add_4470", "erf_6", "mul_2827", "amin_29", "add_4553", "round_60", "mul_2877", "mul_2874", "reciprocal_29", "view_457", "clamp_min_87", "div_58", "sub_1341", "maximum_29", "full_59", "minimum_29", "full_58", "view_458", "convert_element_type_174", "clamp_max_58", "clamp_min_88", "sub_1347", "round_59", "div_59", "mul_2899", "sub_1367", "convert_element_type_176", "convert_element_type_175", "clamp_max_59", "clamp_min_89", "convert_element_type_177", "view_462", "view_461"], "triton_poi_fused__to_copy_mul_sub_view_10:68": ["mul_2904", "sub_1371", "convert_element_type_178", "view_464", "convert_element_type_179", "view_466", "view_465"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69": ["var_mean_10", "clone_41", "add_4616", "view_469", "add_tensor_135", "add_4622", "mul_2921", "mul_2920", "sub_1377", "rsqrt_10", "add_4621", "amax_30", "amin_30", "amax_31", "amin_31", "amax_32", "amin_32", "sub_1414", "convert_element_type_182", "convert_element_type_181", "clamp_max_61", "clamp_min_92", "add_4709", "round_62", "mul_2972", "mul_2969", "reciprocal_30", "view_472", "clamp_min_90", "div_60", "sub_1388", "maximum_30", "full_61", "minimum_30", "full_60", "view_473", "convert_element_type_180", "clamp_max_60", "clamp_min_91", "sub_1394", "round_61", "div_61", "convert_element_type_183", "view_477", "mul_2994", "view_476", "sub_1459", "convert_element_type_188", "convert_element_type_187", "clamp_max_63", "clamp_min_95", "add_4861", "round_64", "mul_3068", "mul_3065", "reciprocal_31", "view_488", "clamp_min_93", "div_62", "sub_1433", "maximum_31", "full_63", "minimum_31", "full_62", "view_489", "convert_element_type_186", "clamp_max_62", "clamp_min_94", "sub_1439", "round_63", "div_63", "convert_element_type_189", "view_493", "mul_3090", "view_492", "sub_1503", "convert_element_type_194", "convert_element_type_193", "clamp_max_65", "clamp_min_98", "add_5009", "round_66", "mul_3167", "mul_3164", "reciprocal_32", "view_504", "clamp_min_96", "div_64", "sub_1477", "maximum_32", "full_65", "minimum_32", "full_64", "view_505", "convert_element_type_192", "clamp_max_64", "clamp_min_97", "sub_1483", "round_65", "div_65", "convert_element_type_195", "view_509", "mul_3189", "view_508"], "triton_poi_fused__to_copy_mul_sub_view_4:70": ["mul_2999", "sub_1418", "convert_element_type_184", "view_479", "convert_element_type_185", "view_481", "view_480"], "triton_poi_fused__to_copy_mul_sub_view_4:71": ["mul_3095", "sub_1463", "convert_element_type_190", "view_495", "convert_element_type_191", "view_497", "view_496"], "triton_poi_fused__to_copy_mul_sub_view_4:72": ["mul_3194", "sub_1507", "convert_element_type_196", "view_511", "convert_element_type_197", "view_513", "view_512"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:73": ["_scaled_dot_product_efficient_attention_5", "clone_42", "permute_52", "view_485", "mul_3011", "view_484", "add_tensor_134", "clone_43", "permute_54", "view_501", "view_500", "clone_44", "permute_56", "view_517", "view_516", "add_tensor_133"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:74": ["_scaled_dot_product_efficient_attention_5", "clone_42", "permute_52", "view_485", "mul_3011", "view_484", "add_tensor_134", "clone_43", "permute_54", "view_501", "view_500", "clone_44", "permute_56", "view_517", "view_516", "add_tensor_133"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:75": ["_scaled_dot_product_efficient_attention_5", "clone_42", "permute_52", "view_485", "mul_3011", "view_484", "add_tensor_134", "clone_43", "permute_54", "view_501", "view_500", "clone_44", "permute_56", "view_517", "view_516", "add_tensor_133"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76": ["amax_33", "view_518", "permute_57", "amin_33", "sub_1551", "convert_element_type_200", "convert_element_type_199", "clamp_max_67", "clamp_min_101", "add_5173", "round_68", "mul_3272", "mul_3269", "reciprocal_33", "view_521", "clamp_min_99", "div_66", "sub_1525", "maximum_33", "full_67", "minimum_33", "full_66", "view_522", "convert_element_type_198", "clamp_max_66", "clamp_min_100", "sub_1531", "round_67", "div_67", "convert_element_type_201", "view_526", "mul_3294", "view_525"], "triton_poi_fused__to_copy_mul_sub_view_4:77": ["mul_3299", "sub_1555", "convert_element_type_202", "view_528", "convert_element_type_203", "view_530", "view_529"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78": ["add_5236", "add_4616", "view_469", "add_tensor_135", "view_533", "add_tensor_132", "var_mean_11", "clone_46", "add_5242", "mul_3316", "mul_3315", "sub_1561", "rsqrt_11", "add_5241", "amax_34", "amin_34", "sub_1598", "convert_element_type_206", "convert_element_type_205", "clamp_max_69", "clamp_min_104", "add_5329", "round_70", "mul_3367", "mul_3364", "reciprocal_34", "view_536", "clamp_min_102", "div_68", "sub_1572", "maximum_34", "full_69", "minimum_34", "full_68", "view_537", "convert_element_type_204", "clamp_max_68", "clamp_min_103", "sub_1578", "round_69", "div_69", "convert_element_type_207", "view_541", "mul_3389", "view_540"], "triton_poi_fused__to_copy_mul_sub_view_10:79": ["mul_3394", "sub_1602", "convert_element_type_208", "view_543", "convert_element_type_209", "view_545", "view_544"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80": ["amax_35", "mul_3408", "mul_3406", "view_548", "add_tensor_131", "add_5388", "erf_7", "mul_3407", "amin_35", "add_5471", "round_72", "mul_3457", "mul_3454", "reciprocal_35", "view_551", "clamp_min_105", "div_70", "sub_1615", "maximum_35", "full_71", "minimum_35", "full_70", "view_552", "convert_element_type_210", "clamp_max_70", "clamp_min_106", "sub_1621", "round_71", "div_71", "mul_3479", "sub_1641", "convert_element_type_212", "convert_element_type_211", "clamp_max_71", "clamp_min_107", "convert_element_type_213", "view_556", "view_555"], "triton_poi_fused__to_copy_mul_sub_view_10:81": ["mul_3484", "sub_1645", "convert_element_type_214", "view_558", "convert_element_type_215", "view_560", "view_559"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82": ["var_mean_12", "clone_49", "add_5534", "view_563", "add_tensor_130", "add_5540", "mul_3501", "mul_3500", "sub_1651", "rsqrt_12", "add_5539", "amax_36", "amin_36", "amax_37", "amin_37", "amax_38", "amin_38", "sub_1688", "convert_element_type_218", "convert_element_type_217", "clamp_max_73", "clamp_min_110", "add_5627", "round_74", "mul_3552", "mul_3549", "reciprocal_36", "view_566", "clamp_min_108", "div_72", "sub_1662", "maximum_36", "full_73", "minimum_36", "full_72", "view_567", "convert_element_type_216", "clamp_max_72", "clamp_min_109", "sub_1668", "round_73", "div_73", "convert_element_type_219", "view_571", "mul_3574", "view_570", "sub_1733", "convert_element_type_224", "convert_element_type_223", "clamp_max_75", "clamp_min_113", "add_5779", "round_76", "mul_3648", "mul_3645", "reciprocal_37", "view_582", "clamp_min_111", "div_74", "sub_1707", "maximum_37", "full_75", "minimum_37", "full_74", "view_583", "convert_element_type_222", "clamp_max_74", "clamp_min_112", "sub_1713", "round_75", "div_75", "convert_element_type_225", "view_587", "mul_3670", "view_586", "sub_1777", "convert_element_type_230", "convert_element_type_229", "clamp_max_77", "clamp_min_116", "add_5927", "round_78", "mul_3747", "mul_3744", "reciprocal_38", "view_598", "clamp_min_114", "div_76", "sub_1751", "maximum_38", "full_77", "minimum_38", "full_76", "view_599", "convert_element_type_228", "clamp_max_76", "clamp_min_115", "sub_1757", "round_77", "div_77", "convert_element_type_231", "view_603", "mul_3769", "view_602"], "triton_poi_fused__to_copy_mul_sub_view_4:83": ["mul_3579", "sub_1692", "convert_element_type_220", "view_573", "convert_element_type_221", "view_575", "view_574"], "triton_poi_fused__to_copy_mul_sub_view_4:84": ["mul_3675", "sub_1737", "convert_element_type_226", "view_589", "convert_element_type_227", "view_591", "view_590"], "triton_poi_fused__to_copy_mul_sub_view_4:85": ["mul_3774", "sub_1781", "convert_element_type_232", "view_605", "convert_element_type_233", "view_607", "view_606"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:86": ["_scaled_dot_product_efficient_attention_6", "clone_50", "permute_62", "view_579", "mul_3591", "view_578", "add_tensor_129", "clone_51", "permute_64", "view_595", "view_594", "clone_52", "permute_66", "view_611", "view_610", "add_tensor_128"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:87": ["_scaled_dot_product_efficient_attention_6", "clone_50", "permute_62", "view_579", "mul_3591", "view_578", "add_tensor_129", "clone_51", "permute_64", "view_595", "view_594", "clone_52", "permute_66", "view_611", "view_610", "add_tensor_128"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:88": ["_scaled_dot_product_efficient_attention_6", "clone_50", "permute_62", "view_579", "mul_3591", "view_578", "add_tensor_129", "clone_51", "permute_64", "view_595", "view_594", "clone_52", "permute_66", "view_611", "view_610", "add_tensor_128"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89": ["amax_39", "view_612", "permute_67", "amin_39", "sub_1825", "convert_element_type_236", "convert_element_type_235", "clamp_max_79", "clamp_min_119", "add_6091", "round_80", "mul_3852", "mul_3849", "reciprocal_39", "view_615", "clamp_min_117", "div_78", "sub_1799", "maximum_39", "full_79", "minimum_39", "full_78", "view_616", "convert_element_type_234", "clamp_max_78", "clamp_min_118", "sub_1805", "round_79", "div_79", "convert_element_type_237", "view_620", "mul_3874", "view_619"], "triton_poi_fused__to_copy_mul_sub_view_4:90": ["mul_3879", "sub_1829", "convert_element_type_238", "view_622", "convert_element_type_239", "view_624", "view_623"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91": ["add_6154", "add_5534", "view_563", "add_tensor_130", "view_627", "add_tensor_127", "var_mean_13", "clone_54", "add_6160", "mul_3896", "mul_3895", "sub_1835", "rsqrt_13", "add_6159", "amax_40", "amin_40", "sub_1872", "convert_element_type_242", "convert_element_type_241", "clamp_max_81", "clamp_min_122", "add_6247", "round_82", "mul_3947", "mul_3944", "reciprocal_40", "view_630", "clamp_min_120", "div_80", "sub_1846", "maximum_40", "full_81", "minimum_40", "full_80", "view_631", "convert_element_type_240", "clamp_max_80", "clamp_min_121", "sub_1852", "round_81", "div_81", "convert_element_type_243", "view_635", "mul_3969", "view_634"], "triton_poi_fused__to_copy_mul_sub_view_10:92": ["mul_3974", "sub_1876", "convert_element_type_244", "view_637", "convert_element_type_245", "view_639", "view_638"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93": ["amax_41", "mul_3988", "mul_3986", "view_642", "add_tensor_126", "add_6306", "erf_8", "mul_3987", "amin_41", "add_6389", "round_84", "mul_4037", "mul_4034", "reciprocal_41", "view_645", "clamp_min_123", "div_82", "sub_1889", "maximum_41", "full_83", "minimum_41", "full_82", "view_646", "convert_element_type_246", "clamp_max_82", "clamp_min_124", "sub_1895", "round_83", "div_83", "mul_4059", "sub_1915", "convert_element_type_248", "convert_element_type_247", "clamp_max_83", "clamp_min_125", "convert_element_type_249", "view_650", "view_649"], "triton_poi_fused__to_copy_mul_sub_view_10:94": ["mul_4064", "sub_1919", "convert_element_type_250", "view_652", "convert_element_type_251", "view_654", "view_653"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95": ["var_mean_14", "clone_57", "add_6452", "view_657", "add_tensor_125", "add_6458", "mul_4081", "mul_4080", "sub_1925", "rsqrt_14", "add_6457", "amax_42", "amin_42", "amax_43", "amin_43", "amax_44", "amin_44", "sub_1962", "convert_element_type_254", "convert_element_type_253", "clamp_max_85", "clamp_min_128", "add_6545", "round_86", "mul_4132", "mul_4129", "reciprocal_42", "view_660", "clamp_min_126", "div_84", "sub_1936", "maximum_42", "full_85", "minimum_42", "full_84", "view_661", "convert_element_type_252", "clamp_max_84", "clamp_min_127", "sub_1942", "round_85", "div_85", "convert_element_type_255", "view_665", "mul_4154", "view_664", "sub_2007", "convert_element_type_260", "convert_element_type_259", "clamp_max_87", "clamp_min_131", "add_6697", "round_88", "mul_4228", "mul_4225", "reciprocal_43", "view_676", "clamp_min_129", "div_86", "sub_1981", "maximum_43", "full_87", "minimum_43", "full_86", "view_677", "convert_element_type_258", "clamp_max_86", "clamp_min_130", "sub_1987", "round_87", "div_87", "convert_element_type_261", "view_681", "mul_4250", "view_680", "sub_2051", "convert_element_type_266", "convert_element_type_265", "clamp_max_89", "clamp_min_134", "add_6845", "round_90", "mul_4327", "mul_4324", "reciprocal_44", "view_692", "clamp_min_132", "div_88", "sub_2025", "maximum_44", "full_89", "minimum_44", "full_88", "view_693", "convert_element_type_264", "clamp_max_88", "clamp_min_133", "sub_2031", "round_89", "div_89", "convert_element_type_267", "view_697", "mul_4349", "view_696"], "triton_poi_fused__to_copy_mul_sub_view_4:96": ["mul_4159", "sub_1966", "convert_element_type_256", "view_667", "convert_element_type_257", "view_669", "view_668"], "triton_poi_fused__to_copy_mul_sub_view_4:97": ["mul_4255", "sub_2011", "convert_element_type_262", "view_683", "convert_element_type_263", "view_685", "view_684"], "triton_poi_fused__to_copy_mul_sub_view_4:98": ["mul_4354", "sub_2055", "convert_element_type_268", "view_699", "convert_element_type_269", "view_701", "view_700"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:99": ["_scaled_dot_product_efficient_attention_7", "clone_58", "permute_72", "view_673", "mul_4171", "view_672", "add_tensor_124", "clone_59", "permute_74", "view_689", "view_688", "clone_60", "permute_76", "view_705", "view_704", "add_tensor_123"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:100": ["_scaled_dot_product_efficient_attention_7", "clone_58", "permute_72", "view_673", "mul_4171", "view_672", "add_tensor_124", "clone_59", "permute_74", "view_689", "view_688", "clone_60", "permute_76", "view_705", "view_704", "add_tensor_123"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:101": ["_scaled_dot_product_efficient_attention_7", "clone_58", "permute_72", "view_673", "mul_4171", "view_672", "add_tensor_124", "clone_59", "permute_74", "view_689", "view_688", "clone_60", "permute_76", "view_705", "view_704", "add_tensor_123"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102": ["amax_45", "view_706", "permute_77", "amin_45", "sub_2099", "convert_element_type_272", "convert_element_type_271", "clamp_max_91", "clamp_min_137", "add_7009", "round_92", "mul_4432", "mul_4429", "reciprocal_45", "view_709", "clamp_min_135", "div_90", "sub_2073", "maximum_45", "full_91", "minimum_45", "full_90", "view_710", "convert_element_type_270", "clamp_max_90", "clamp_min_136", "sub_2079", "round_91", "div_91", "convert_element_type_273", "view_714", "mul_4454", "view_713"], "triton_poi_fused__to_copy_mul_sub_view_4:103": ["mul_4459", "sub_2103", "convert_element_type_274", "view_716", "convert_element_type_275", "view_718", "view_717"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104": ["add_7072", "add_6452", "view_657", "add_tensor_125", "view_721", "add_tensor_122", "var_mean_15", "clone_62", "add_7078", "mul_4476", "mul_4475", "sub_2109", "rsqrt_15", "add_7077", "amax_46", "amin_46", "sub_2146", "convert_element_type_278", "convert_element_type_277", "clamp_max_93", "clamp_min_140", "add_7165", "round_94", "mul_4527", "mul_4524", "reciprocal_46", "view_724", "clamp_min_138", "div_92", "sub_2120", "maximum_46", "full_93", "minimum_46", "full_92", "view_725", "convert_element_type_276", "clamp_max_92", "clamp_min_139", "sub_2126", "round_93", "div_93", "convert_element_type_279", "view_729", "mul_4549", "view_728"], "triton_poi_fused__to_copy_mul_sub_view_10:105": ["mul_4554", "sub_2150", "convert_element_type_280", "view_731", "convert_element_type_281", "view_733", "view_732"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106": ["amax_47", "mul_4568", "mul_4566", "view_736", "add_tensor_121", "add_7224", "erf_9", "mul_4567", "amin_47", "add_7307", "round_96", "mul_4617", "mul_4614", "reciprocal_47", "view_739", "clamp_min_141", "div_94", "sub_2163", "maximum_47", "full_95", "minimum_47", "full_94", "view_740", "convert_element_type_282", "clamp_max_94", "clamp_min_142", "sub_2169", "round_95", "div_95", "mul_4639", "sub_2189", "convert_element_type_284", "convert_element_type_283", "clamp_max_95", "clamp_min_143", "convert_element_type_285", "view_744", "view_743"], "triton_poi_fused__to_copy_mul_sub_view_10:107": ["mul_4644", "sub_2193", "convert_element_type_286", "view_746", "convert_element_type_287", "view_748", "view_747"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108": ["var_mean_16", "clone_65", "add_7370", "view_751", "add_tensor_120", "add_7376", "mul_4661", "mul_4660", "sub_2199", "rsqrt_16", "add_7375", "amax_48", "amin_48", "amax_49", "amin_49", "amax_50", "amin_50", "sub_2236", "convert_element_type_290", "convert_element_type_289", "clamp_max_97", "clamp_min_146", "add_7463", "round_98", "mul_4712", "mul_4709", "reciprocal_48", "view_754", "clamp_min_144", "div_96", "sub_2210", "maximum_48", "full_97", "minimum_48", "full_96", "view_755", "convert_element_type_288", "clamp_max_96", "clamp_min_145", "sub_2216", "round_97", "div_97", "convert_element_type_291", "view_759", "mul_4734", "view_758", "sub_2281", "convert_element_type_296", "convert_element_type_295", "clamp_max_99", "clamp_min_149", "add_7615", "round_100", "mul_4808", "mul_4805", "reciprocal_49", "view_770", "clamp_min_147", "div_98", "sub_2255", "maximum_49", "full_99", "minimum_49", "full_98", "view_771", "convert_element_type_294", "clamp_max_98", "clamp_min_148", "sub_2261", "round_99", "div_99", "convert_element_type_297", "view_775", "mul_4830", "view_774", "sub_2325", "convert_element_type_302", "convert_element_type_301", "clamp_max_101", "clamp_min_152", "add_7763", "round_102", "mul_4907", "mul_4904", "reciprocal_50", "view_786", "clamp_min_150", "div_100", "sub_2299", "maximum_50", "full_101", "minimum_50", "full_100", "view_787", "convert_element_type_300", "clamp_max_100", "clamp_min_151", "sub_2305", "round_101", "div_101", "convert_element_type_303", "view_791", "mul_4929", "view_790"], "triton_poi_fused__to_copy_mul_sub_view_4:109": ["mul_4739", "sub_2240", "convert_element_type_292", "view_761", "convert_element_type_293", "view_763", "view_762"], "triton_poi_fused__to_copy_mul_sub_view_4:110": ["mul_4835", "sub_2285", "convert_element_type_298", "view_777", "convert_element_type_299", "view_779", "view_778"], "triton_poi_fused__to_copy_mul_sub_view_4:111": ["mul_4934", "sub_2329", "convert_element_type_304", "view_793", "convert_element_type_305", "view_795", "view_794"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:112": ["_scaled_dot_product_efficient_attention_8", "clone_66", "permute_82", "view_767", "mul_4751", "view_766", "add_tensor_119", "clone_67", "permute_84", "view_783", "view_782", "clone_68", "permute_86", "view_799", "view_798", "add_tensor_118"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:113": ["_scaled_dot_product_efficient_attention_8", "clone_66", "permute_82", "view_767", "mul_4751", "view_766", "add_tensor_119", "clone_67", "permute_84", "view_783", "view_782", "clone_68", "permute_86", "view_799", "view_798", "add_tensor_118"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:114": ["_scaled_dot_product_efficient_attention_8", "clone_66", "permute_82", "view_767", "mul_4751", "view_766", "add_tensor_119", "clone_67", "permute_84", "view_783", "view_782", "clone_68", "permute_86", "view_799", "view_798", "add_tensor_118"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115": ["amax_51", "view_800", "permute_87", "amin_51", "sub_2373", "convert_element_type_308", "convert_element_type_307", "clamp_max_103", "clamp_min_155", "add_7927", "round_104", "mul_5012", "mul_5009", "reciprocal_51", "view_803", "clamp_min_153", "div_102", "sub_2347", "maximum_51", "full_103", "minimum_51", "full_102", "view_804", "convert_element_type_306", "clamp_max_102", "clamp_min_154", "sub_2353", "round_103", "div_103", "convert_element_type_309", "view_808", "mul_5034", "view_807"], "triton_poi_fused__to_copy_mul_sub_view_4:116": ["mul_5039", "sub_2377", "convert_element_type_310", "view_810", "convert_element_type_311", "view_812", "view_811"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117": ["add_7990", "add_7370", "view_751", "add_tensor_120", "view_815", "add_tensor_117", "var_mean_17", "clone_70", "add_7996", "mul_5056", "mul_5055", "sub_2383", "rsqrt_17", "add_7995", "amax_52", "amin_52", "sub_2420", "convert_element_type_314", "convert_element_type_313", "clamp_max_105", "clamp_min_158", "add_8083", "round_106", "mul_5107", "mul_5104", "reciprocal_52", "view_818", "clamp_min_156", "div_104", "sub_2394", "maximum_52", "full_105", "minimum_52", "full_104", "view_819", "convert_element_type_312", "clamp_max_104", "clamp_min_157", "sub_2400", "round_105", "div_105", "convert_element_type_315", "view_823", "mul_5129", "view_822"], "triton_poi_fused__to_copy_mul_sub_view_10:118": ["mul_5134", "sub_2424", "convert_element_type_316", "view_825", "convert_element_type_317", "view_827", "view_826"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119": ["amax_53", "mul_5148", "mul_5146", "view_830", "add_tensor_116", "add_8142", "erf_10", "mul_5147", "amin_53", "add_8225", "round_108", "mul_5197", "mul_5194", "reciprocal_53", "view_833", "clamp_min_159", "div_106", "sub_2437", "maximum_53", "full_107", "minimum_53", "full_106", "view_834", "convert_element_type_318", "clamp_max_106", "clamp_min_160", "sub_2443", "round_107", "div_107", "mul_5219", "sub_2463", "convert_element_type_320", "convert_element_type_319", "clamp_max_107", "clamp_min_161", "convert_element_type_321", "view_838", "view_837"], "triton_poi_fused__to_copy_mul_sub_view_10:120": ["mul_5224", "sub_2467", "convert_element_type_322", "view_840", "convert_element_type_323", "view_842", "view_841"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121": ["var_mean_18", "clone_73", "add_8288", "view_845", "add_tensor_115", "add_8294", "mul_5241", "mul_5240", "sub_2473", "rsqrt_18", "add_8293", "amax_54", "amin_54", "amax_55", "amin_55", "amax_56", "amin_56", "sub_2510", "convert_element_type_326", "convert_element_type_325", "clamp_max_109", "clamp_min_164", "add_8381", "round_110", "mul_5292", "mul_5289", "reciprocal_54", "view_848", "clamp_min_162", "div_108", "sub_2484", "maximum_54", "full_109", "minimum_54", "full_108", "view_849", "convert_element_type_324", "clamp_max_108", "clamp_min_163", "sub_2490", "round_109", "div_109", "convert_element_type_327", "view_853", "mul_5314", "view_852", "sub_2555", "convert_element_type_332", "convert_element_type_331", "clamp_max_111", "clamp_min_167", "add_8533", "round_112", "mul_5388", "mul_5385", "reciprocal_55", "view_864", "clamp_min_165", "div_110", "sub_2529", "maximum_55", "full_111", "minimum_55", "full_110", "view_865", "convert_element_type_330", "clamp_max_110", "clamp_min_166", "sub_2535", "round_111", "div_111", "convert_element_type_333", "view_869", "mul_5410", "view_868", "sub_2599", "convert_element_type_338", "convert_element_type_337", "clamp_max_113", "clamp_min_170", "add_8681", "round_114", "mul_5487", "mul_5484", "reciprocal_56", "view_880", "clamp_min_168", "div_112", "sub_2573", "maximum_56", "full_113", "minimum_56", "full_112", "view_881", "convert_element_type_336", "clamp_max_112", "clamp_min_169", "sub_2579", "round_113", "div_113", "convert_element_type_339", "view_885", "mul_5509", "view_884"], "triton_poi_fused__to_copy_mul_sub_view_4:122": ["mul_5319", "sub_2514", "convert_element_type_328", "view_855", "convert_element_type_329", "view_857", "view_856"], "triton_poi_fused__to_copy_mul_sub_view_4:123": ["mul_5415", "sub_2559", "convert_element_type_334", "view_871", "convert_element_type_335", "view_873", "view_872"], "triton_poi_fused__to_copy_mul_sub_view_4:124": ["mul_5514", "sub_2603", "convert_element_type_340", "view_887", "convert_element_type_341", "view_889", "view_888"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:125": ["_scaled_dot_product_efficient_attention_9", "clone_74", "permute_92", "view_861", "mul_5331", "view_860", "add_tensor_114", "clone_75", "permute_94", "view_877", "view_876", "clone_76", "permute_96", "view_893", "view_892", "add_tensor_113"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:126": ["_scaled_dot_product_efficient_attention_9", "clone_74", "permute_92", "view_861", "mul_5331", "view_860", "add_tensor_114", "clone_75", "permute_94", "view_877", "view_876", "clone_76", "permute_96", "view_893", "view_892", "add_tensor_113"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:127": ["_scaled_dot_product_efficient_attention_9", "clone_74", "permute_92", "view_861", "mul_5331", "view_860", "add_tensor_114", "clone_75", "permute_94", "view_877", "view_876", "clone_76", "permute_96", "view_893", "view_892", "add_tensor_113"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128": ["amax_57", "view_894", "permute_97", "amin_57", "sub_2647", "convert_element_type_344", "convert_element_type_343", "clamp_max_115", "clamp_min_173", "add_8845", "round_116", "mul_5592", "mul_5589", "reciprocal_57", "view_897", "clamp_min_171", "div_114", "sub_2621", "maximum_57", "full_115", "minimum_57", "full_114", "view_898", "convert_element_type_342", "clamp_max_114", "clamp_min_172", "sub_2627", "round_115", "div_115", "convert_element_type_345", "view_902", "mul_5614", "view_901"], "triton_poi_fused__to_copy_mul_sub_view_4:129": ["mul_5619", "sub_2651", "convert_element_type_346", "view_904", "convert_element_type_347", "view_906", "view_905"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130": ["add_8908", "add_8288", "view_845", "add_tensor_115", "view_909", "add_tensor_112", "var_mean_19", "clone_78", "add_8914", "mul_5636", "mul_5635", "sub_2657", "rsqrt_19", "add_8913", "amax_58", "amin_58", "sub_2694", "convert_element_type_350", "convert_element_type_349", "clamp_max_117", "clamp_min_176", "add_9001", "round_118", "mul_5687", "mul_5684", "reciprocal_58", "view_912", "clamp_min_174", "div_116", "sub_2668", "maximum_58", "full_117", "minimum_58", "full_116", "view_913", "convert_element_type_348", "clamp_max_116", "clamp_min_175", "sub_2674", "round_117", "div_117", "convert_element_type_351", "view_917", "mul_5709", "view_916"], "triton_poi_fused__to_copy_mul_sub_view_10:131": ["mul_5714", "sub_2698", "convert_element_type_352", "view_919", "convert_element_type_353", "view_921", "view_920"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132": ["amax_59", "mul_5728", "mul_5726", "view_924", "add_tensor_111", "add_9060", "erf_11", "mul_5727", "amin_59", "add_9143", "round_120", "mul_5777", "mul_5774", "reciprocal_59", "view_927", "clamp_min_177", "div_118", "sub_2711", "maximum_59", "full_119", "minimum_59", "full_118", "view_928", "convert_element_type_354", "clamp_max_118", "clamp_min_178", "sub_2717", "round_119", "div_119", "mul_5799", "sub_2737", "convert_element_type_356", "convert_element_type_355", "clamp_max_119", "clamp_min_179", "convert_element_type_357", "view_932", "view_931"], "triton_poi_fused__to_copy_mul_sub_view_10:133": ["mul_5804", "sub_2741", "convert_element_type_358", "view_934", "convert_element_type_359", "view_936", "view_935"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134": ["var_mean_20", "clone_81", "add_9206", "view_939", "add_tensor_110", "add_9212", "mul_5821", "mul_5820", "sub_2747", "rsqrt_20", "add_9211", "amax_60", "amin_60", "amax_61", "amin_61", "amax_62", "amin_62", "sub_2784", "convert_element_type_362", "convert_element_type_361", "clamp_max_121", "clamp_min_182", "add_9299", "round_122", "mul_5872", "mul_5869", "reciprocal_60", "view_942", "clamp_min_180", "div_120", "sub_2758", "maximum_60", "full_121", "minimum_60", "full_120", "view_943", "convert_element_type_360", "clamp_max_120", "clamp_min_181", "sub_2764", "round_121", "div_121", "convert_element_type_363", "view_947", "mul_5894", "view_946", "sub_2829", "convert_element_type_368", "convert_element_type_367", "clamp_max_123", "clamp_min_185", "add_9451", "round_124", "mul_5968", "mul_5965", "reciprocal_61", "view_958", "clamp_min_183", "div_122", "sub_2803", "maximum_61", "full_123", "minimum_61", "full_122", "view_959", "convert_element_type_366", "clamp_max_122", "clamp_min_184", "sub_2809", "round_123", "div_123", "convert_element_type_369", "view_963", "mul_5990", "view_962", "sub_2873", "convert_element_type_374", "convert_element_type_373", "clamp_max_125", "clamp_min_188", "add_9599", "round_126", "mul_6067", "mul_6064", "reciprocal_62", "view_974", "clamp_min_186", "div_124", "sub_2847", "maximum_62", "full_125", "minimum_62", "full_124", "view_975", "convert_element_type_372", "clamp_max_124", "clamp_min_187", "sub_2853", "round_125", "div_125", "convert_element_type_375", "view_979", "mul_6089", "view_978"], "triton_poi_fused__to_copy_mul_sub_view_4:135": ["mul_5899", "sub_2788", "convert_element_type_364", "view_949", "convert_element_type_365", "view_951", "view_950"], "triton_poi_fused__to_copy_mul_sub_view_4:136": ["mul_5995", "sub_2833", "convert_element_type_370", "view_965", "convert_element_type_371", "view_967", "view_966"], "triton_poi_fused__to_copy_mul_sub_view_4:137": ["mul_6094", "sub_2877", "convert_element_type_376", "view_981", "convert_element_type_377", "view_983", "view_982"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:138": ["_scaled_dot_product_efficient_attention_10", "clone_82", "permute_102", "view_955", "mul_5911", "view_954", "add_tensor_109", "clone_83", "permute_104", "view_971", "view_970", "clone_84", "permute_106", "view_987", "view_986", "add_tensor_108"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:139": ["_scaled_dot_product_efficient_attention_10", "clone_82", "permute_102", "view_955", "mul_5911", "view_954", "add_tensor_109", "clone_83", "permute_104", "view_971", "view_970", "clone_84", "permute_106", "view_987", "view_986", "add_tensor_108"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:140": ["_scaled_dot_product_efficient_attention_10", "clone_82", "permute_102", "view_955", "mul_5911", "view_954", "add_tensor_109", "clone_83", "permute_104", "view_971", "view_970", "clone_84", "permute_106", "view_987", "view_986", "add_tensor_108"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141": ["amax_63", "view_988", "permute_107", "amin_63", "sub_2921", "convert_element_type_380", "convert_element_type_379", "clamp_max_127", "clamp_min_191", "add_9763", "round_128", "mul_6172", "mul_6169", "reciprocal_63", "view_991", "clamp_min_189", "div_126", "sub_2895", "maximum_63", "full_127", "minimum_63", "full_126", "view_992", "convert_element_type_378", "clamp_max_126", "clamp_min_190", "sub_2901", "round_127", "div_127", "convert_element_type_381", "view_996", "mul_6194", "view_995"], "triton_poi_fused__to_copy_mul_sub_view_4:142": ["mul_6199", "sub_2925", "convert_element_type_382", "view_998", "convert_element_type_383", "view_1000", "view_999"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143": ["add_9826", "add_9206", "view_939", "add_tensor_110", "view_1003", "add_tensor_107", "var_mean_21", "clone_86", "add_9832", "mul_6216", "mul_6215", "sub_2931", "rsqrt_21", "add_9831", "amax_64", "amin_64", "sub_2968", "convert_element_type_386", "convert_element_type_385", "clamp_max_129", "clamp_min_194", "add_9919", "round_130", "mul_6267", "mul_6264", "reciprocal_64", "view_1006", "clamp_min_192", "div_128", "sub_2942", "maximum_64", "full_129", "minimum_64", "full_128", "view_1007", "convert_element_type_384", "clamp_max_128", "clamp_min_193", "sub_2948", "round_129", "div_129", "convert_element_type_387", "view_1011", "mul_6289", "view_1010"], "triton_poi_fused__to_copy_mul_sub_view_10:144": ["mul_6294", "sub_2972", "convert_element_type_388", "view_1013", "convert_element_type_389", "view_1015", "view_1014"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145": ["amax_65", "mul_6308", "mul_6306", "view_1018", "add_tensor_106", "add_9978", "erf_12", "mul_6307", "amin_65", "add_10061", "round_132", "mul_6357", "mul_6354", "reciprocal_65", "view_1021", "clamp_min_195", "div_130", "sub_2985", "maximum_65", "full_131", "minimum_65", "full_130", "view_1022", "convert_element_type_390", "clamp_max_130", "clamp_min_196", "sub_2991", "round_131", "div_131", "mul_6379", "sub_3011", "convert_element_type_392", "convert_element_type_391", "clamp_max_131", "clamp_min_197", "convert_element_type_393", "view_1026", "view_1025"], "triton_poi_fused__to_copy_mul_sub_view_10:146": ["mul_6384", "sub_3015", "convert_element_type_394", "view_1028", "convert_element_type_395", "view_1030", "view_1029"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147": ["var_mean_22", "clone_89", "add_10124", "view_1033", "add_tensor_105", "add_10130", "mul_6401", "mul_6400", "sub_3021", "rsqrt_22", "add_10129", "amax_66", "amin_66", "amax_67", "amin_67", "amax_68", "amin_68", "sub_3058", "convert_element_type_398", "convert_element_type_397", "clamp_max_133", "clamp_min_200", "add_10217", "round_134", "mul_6452", "mul_6449", "reciprocal_66", "view_1036", "clamp_min_198", "div_132", "sub_3032", "maximum_66", "full_133", "minimum_66", "full_132", "view_1037", "convert_element_type_396", "clamp_max_132", "clamp_min_199", "sub_3038", "round_133", "div_133", "convert_element_type_399", "view_1041", "mul_6474", "view_1040", "sub_3103", "convert_element_type_404", "convert_element_type_403", "clamp_max_135", "clamp_min_203", "add_10369", "round_136", "mul_6548", "mul_6545", "reciprocal_67", "view_1052", "clamp_min_201", "div_134", "sub_3077", "maximum_67", "full_135", "minimum_67", "full_134", "view_1053", "convert_element_type_402", "clamp_max_134", "clamp_min_202", "sub_3083", "round_135", "div_135", "convert_element_type_405", "view_1057", "mul_6570", "view_1056", "sub_3147", "convert_element_type_410", "convert_element_type_409", "clamp_max_137", "clamp_min_206", "add_10517", "round_138", "mul_6647", "mul_6644", "reciprocal_68", "view_1068", "clamp_min_204", "div_136", "sub_3121", "maximum_68", "full_137", "minimum_68", "full_136", "view_1069", "convert_element_type_408", "clamp_max_136", "clamp_min_205", "sub_3127", "round_137", "div_137", "convert_element_type_411", "view_1073", "mul_6669", "view_1072"], "triton_poi_fused__to_copy_mul_sub_view_4:148": ["mul_6479", "sub_3062", "convert_element_type_400", "view_1043", "convert_element_type_401", "view_1045", "view_1044"], "triton_poi_fused__to_copy_mul_sub_view_4:149": ["mul_6575", "sub_3107", "convert_element_type_406", "view_1059", "convert_element_type_407", "view_1061", "view_1060"], "triton_poi_fused__to_copy_mul_sub_view_4:150": ["mul_6674", "sub_3151", "convert_element_type_412", "view_1075", "convert_element_type_413", "view_1077", "view_1076"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:151": ["_scaled_dot_product_efficient_attention_11", "clone_90", "permute_112", "view_1049", "mul_6491", "view_1048", "add_tensor_104", "clone_91", "permute_114", "view_1065", "view_1064", "clone_92", "permute_116", "view_1081", "view_1080", "add_tensor_103"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:152": ["_scaled_dot_product_efficient_attention_11", "clone_90", "permute_112", "view_1049", "mul_6491", "view_1048", "add_tensor_104", "clone_91", "permute_114", "view_1065", "view_1064", "clone_92", "permute_116", "view_1081", "view_1080", "add_tensor_103"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:153": ["_scaled_dot_product_efficient_attention_11", "clone_90", "permute_112", "view_1049", "mul_6491", "view_1048", "add_tensor_104", "clone_91", "permute_114", "view_1065", "view_1064", "clone_92", "permute_116", "view_1081", "view_1080", "add_tensor_103"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154": ["amax_69", "view_1082", "permute_117", "amin_69", "sub_3195", "convert_element_type_416", "convert_element_type_415", "clamp_max_139", "clamp_min_209", "add_10681", "round_140", "mul_6752", "mul_6749", "reciprocal_69", "view_1085", "clamp_min_207", "div_138", "sub_3169", "maximum_69", "full_139", "minimum_69", "full_138", "view_1086", "convert_element_type_414", "clamp_max_138", "clamp_min_208", "sub_3175", "round_139", "div_139", "convert_element_type_417", "view_1090", "mul_6774", "view_1089"], "triton_poi_fused__to_copy_mul_sub_view_4:155": ["mul_6779", "sub_3199", "convert_element_type_418", "view_1092", "convert_element_type_419", "view_1094", "view_1093"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156": ["add_10744", "add_10124", "view_1033", "add_tensor_105", "view_1097", "add_tensor_102", "var_mean_23", "clone_94", "add_10750", "mul_6796", "mul_6795", "sub_3205", "rsqrt_23", "add_10749", "amax_70", "amin_70", "sub_3242", "convert_element_type_422", "convert_element_type_421", "clamp_max_141", "clamp_min_212", "add_10837", "round_142", "mul_6847", "mul_6844", "reciprocal_70", "view_1100", "clamp_min_210", "div_140", "sub_3216", "maximum_70", "full_141", "minimum_70", "full_140", "view_1101", "convert_element_type_420", "clamp_max_140", "clamp_min_211", "sub_3222", "round_141", "div_141", "convert_element_type_423", "view_1105", "mul_6869", "view_1104"], "triton_poi_fused__to_copy_mul_sub_view_10:157": ["mul_6874", "sub_3246", "convert_element_type_424", "view_1107", "convert_element_type_425", "view_1109", "view_1108"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158": ["amax_71", "mul_6888", "mul_6886", "view_1112", "add_tensor_101", "add_10896", "erf_13", "mul_6887", "amin_71", "add_10979", "round_144", "mul_6937", "mul_6934", "reciprocal_71", "view_1115", "clamp_min_213", "div_142", "sub_3259", "maximum_71", "full_143", "minimum_71", "full_142", "view_1116", "convert_element_type_426", "clamp_max_142", "clamp_min_214", "sub_3265", "round_143", "div_143", "mul_6959", "sub_3285", "convert_element_type_428", "convert_element_type_427", "clamp_max_143", "clamp_min_215", "convert_element_type_429", "view_1120", "view_1119"], "triton_poi_fused__to_copy_mul_sub_view_10:159": ["mul_6964", "sub_3289", "convert_element_type_430", "view_1122", "convert_element_type_431", "view_1124", "view_1123"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160": ["var_mean_24", "clone_97", "add_11042", "view_1127", "add_tensor_100", "add_11048", "mul_6981", "mul_6980", "sub_3295", "rsqrt_24", "add_11047", "amax_72", "amin_72", "amax_73", "amin_73", "amax_74", "amin_74", "sub_3332", "convert_element_type_434", "convert_element_type_433", "clamp_max_145", "clamp_min_218", "add_11135", "round_146", "mul_7032", "mul_7029", "reciprocal_72", "view_1130", "clamp_min_216", "div_144", "sub_3306", "maximum_72", "full_145", "minimum_72", "full_144", "view_1131", "convert_element_type_432", "clamp_max_144", "clamp_min_217", "sub_3312", "round_145", "div_145", "convert_element_type_435", "view_1135", "mul_7054", "view_1134", "sub_3377", "convert_element_type_440", "convert_element_type_439", "clamp_max_147", "clamp_min_221", "add_11287", "round_148", "mul_7128", "mul_7125", "reciprocal_73", "view_1146", "clamp_min_219", "div_146", "sub_3351", "maximum_73", "full_147", "minimum_73", "full_146", "view_1147", "convert_element_type_438", "clamp_max_146", "clamp_min_220", "sub_3357", "round_147", "div_147", "convert_element_type_441", "view_1151", "mul_7150", "view_1150", "sub_3421", "convert_element_type_446", "convert_element_type_445", "clamp_max_149", "clamp_min_224", "add_11435", "round_150", "mul_7227", "mul_7224", "reciprocal_74", "view_1162", "clamp_min_222", "div_148", "sub_3395", "maximum_74", "full_149", "minimum_74", "full_148", "view_1163", "convert_element_type_444", "clamp_max_148", "clamp_min_223", "sub_3401", "round_149", "div_149", "convert_element_type_447", "view_1167", "mul_7249", "view_1166"], "triton_poi_fused__to_copy_mul_sub_view_4:161": ["mul_7059", "sub_3336", "convert_element_type_436", "view_1137", "convert_element_type_437", "view_1139", "view_1138"], "triton_poi_fused__to_copy_mul_sub_view_4:162": ["mul_7155", "sub_3381", "convert_element_type_442", "view_1153", "convert_element_type_443", "view_1155", "view_1154"], "triton_poi_fused__to_copy_mul_sub_view_4:163": ["mul_7254", "sub_3425", "convert_element_type_448", "view_1169", "convert_element_type_449", "view_1171", "view_1170"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:164": ["_scaled_dot_product_efficient_attention_12", "clone_98", "permute_122", "view_1143", "mul_7071", "view_1142", "add_tensor_99", "clone_99", "permute_124", "view_1159", "view_1158", "clone_100", "permute_126", "view_1175", "view_1174", "add_tensor_98"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:165": ["_scaled_dot_product_efficient_attention_12", "clone_98", "permute_122", "view_1143", "mul_7071", "view_1142", "add_tensor_99", "clone_99", "permute_124", "view_1159", "view_1158", "clone_100", "permute_126", "view_1175", "view_1174", "add_tensor_98"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:166": ["_scaled_dot_product_efficient_attention_12", "clone_98", "permute_122", "view_1143", "mul_7071", "view_1142", "add_tensor_99", "clone_99", "permute_124", "view_1159", "view_1158", "clone_100", "permute_126", "view_1175", "view_1174", "add_tensor_98"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167": ["amax_75", "view_1176", "permute_127", "amin_75", "sub_3469", "convert_element_type_452", "convert_element_type_451", "clamp_max_151", "clamp_min_227", "add_11599", "round_152", "mul_7332", "mul_7329", "reciprocal_75", "view_1179", "clamp_min_225", "div_150", "sub_3443", "maximum_75", "full_151", "minimum_75", "full_150", "view_1180", "convert_element_type_450", "clamp_max_150", "clamp_min_226", "sub_3449", "round_151", "div_151", "convert_element_type_453", "view_1184", "mul_7354", "view_1183"], "triton_poi_fused__to_copy_mul_sub_view_4:168": ["mul_7359", "sub_3473", "convert_element_type_454", "view_1186", "convert_element_type_455", "view_1188", "view_1187"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169": ["add_11662", "add_11042", "view_1127", "add_tensor_100", "view_1191", "add_tensor_97", "var_mean_25", "clone_102", "add_11668", "mul_7376", "mul_7375", "sub_3479", "rsqrt_25", "add_11667", "amax_76", "amin_76", "sub_3516", "convert_element_type_458", "convert_element_type_457", "clamp_max_153", "clamp_min_230", "add_11755", "round_154", "mul_7427", "mul_7424", "reciprocal_76", "view_1194", "clamp_min_228", "div_152", "sub_3490", "maximum_76", "full_153", "minimum_76", "full_152", "view_1195", "convert_element_type_456", "clamp_max_152", "clamp_min_229", "sub_3496", "round_153", "div_153", "convert_element_type_459", "view_1199", "mul_7449", "view_1198"], "triton_poi_fused__to_copy_mul_sub_view_10:170": ["mul_7454", "sub_3520", "convert_element_type_460", "view_1201", "convert_element_type_461", "view_1203", "view_1202"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171": ["amax_77", "mul_7468", "mul_7466", "view_1206", "add_tensor_96", "add_11814", "erf_14", "mul_7467", "amin_77", "add_11897", "round_156", "mul_7517", "mul_7514", "reciprocal_77", "view_1209", "clamp_min_231", "div_154", "sub_3533", "maximum_77", "full_155", "minimum_77", "full_154", "view_1210", "convert_element_type_462", "clamp_max_154", "clamp_min_232", "sub_3539", "round_155", "div_155", "mul_7539", "sub_3559", "convert_element_type_464", "convert_element_type_463", "clamp_max_155", "clamp_min_233", "convert_element_type_465", "view_1214", "view_1213"], "triton_poi_fused__to_copy_mul_sub_view_10:172": ["mul_7544", "sub_3563", "convert_element_type_466", "view_1216", "convert_element_type_467", "view_1218", "view_1217"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173": ["var_mean_26", "clone_105", "add_11960", "view_1221", "add_tensor_95", "add_11966", "mul_7561", "mul_7560", "sub_3569", "rsqrt_26", "add_11965", "amax_78", "amin_78", "amax_79", "amin_79", "amax_80", "amin_80", "sub_3606", "convert_element_type_470", "convert_element_type_469", "clamp_max_157", "clamp_min_236", "add_12053", "round_158", "mul_7612", "mul_7609", "reciprocal_78", "view_1224", "clamp_min_234", "div_156", "sub_3580", "maximum_78", "full_157", "minimum_78", "full_156", "view_1225", "convert_element_type_468", "clamp_max_156", "clamp_min_235", "sub_3586", "round_157", "div_157", "convert_element_type_471", "view_1229", "mul_7634", "view_1228", "sub_3651", "convert_element_type_476", "convert_element_type_475", "clamp_max_159", "clamp_min_239", "add_12205", "round_160", "mul_7708", "mul_7705", "reciprocal_79", "view_1240", "clamp_min_237", "div_158", "sub_3625", "maximum_79", "full_159", "minimum_79", "full_158", "view_1241", "convert_element_type_474", "clamp_max_158", "clamp_min_238", "sub_3631", "round_159", "div_159", "convert_element_type_477", "view_1245", "mul_7730", "view_1244", "sub_3695", "convert_element_type_482", "convert_element_type_481", "clamp_max_161", "clamp_min_242", "add_12353", "round_162", "mul_7807", "mul_7804", "reciprocal_80", "view_1256", "clamp_min_240", "div_160", "sub_3669", "maximum_80", "full_161", "minimum_80", "full_160", "view_1257", "convert_element_type_480", "clamp_max_160", "clamp_min_241", "sub_3675", "round_161", "div_161", "convert_element_type_483", "view_1261", "mul_7829", "view_1260"], "triton_poi_fused__to_copy_mul_sub_view_4:174": ["mul_7639", "sub_3610", "convert_element_type_472", "view_1231", "convert_element_type_473", "view_1233", "view_1232"], "triton_poi_fused__to_copy_mul_sub_view_4:175": ["mul_7735", "sub_3655", "convert_element_type_478", "view_1247", "convert_element_type_479", "view_1249", "view_1248"], "triton_poi_fused__to_copy_mul_sub_view_4:176": ["mul_7834", "sub_3699", "convert_element_type_484", "view_1263", "convert_element_type_485", "view_1265", "view_1264"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:177": ["_scaled_dot_product_efficient_attention_13", "clone_106", "permute_132", "view_1237", "mul_7651", "view_1236", "add_tensor_94", "clone_107", "permute_134", "view_1253", "view_1252", "clone_108", "permute_136", "view_1269", "view_1268", "add_tensor_93"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:178": ["_scaled_dot_product_efficient_attention_13", "clone_106", "permute_132", "view_1237", "mul_7651", "view_1236", "add_tensor_94", "clone_107", "permute_134", "view_1253", "view_1252", "clone_108", "permute_136", "view_1269", "view_1268", "add_tensor_93"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:179": ["_scaled_dot_product_efficient_attention_13", "clone_106", "permute_132", "view_1237", "mul_7651", "view_1236", "add_tensor_94", "clone_107", "permute_134", "view_1253", "view_1252", "clone_108", "permute_136", "view_1269", "view_1268", "add_tensor_93"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180": ["amax_81", "view_1270", "permute_137", "amin_81", "sub_3743", "convert_element_type_488", "convert_element_type_487", "clamp_max_163", "clamp_min_245", "add_12517", "round_164", "mul_7912", "mul_7909", "reciprocal_81", "view_1273", "clamp_min_243", "div_162", "sub_3717", "maximum_81", "full_163", "minimum_81", "full_162", "view_1274", "convert_element_type_486", "clamp_max_162", "clamp_min_244", "sub_3723", "round_163", "div_163", "convert_element_type_489", "view_1278", "mul_7934", "view_1277"], "triton_poi_fused__to_copy_mul_sub_view_4:181": ["mul_7939", "sub_3747", "convert_element_type_490", "view_1280", "convert_element_type_491", "view_1282", "view_1281"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182": ["add_12580", "add_11960", "view_1221", "add_tensor_95", "view_1285", "add_tensor_92", "var_mean_27", "clone_110", "add_12586", "mul_7956", "mul_7955", "sub_3753", "rsqrt_27", "add_12585", "amax_82", "amin_82", "sub_3790", "convert_element_type_494", "convert_element_type_493", "clamp_max_165", "clamp_min_248", "add_12673", "round_166", "mul_8007", "mul_8004", "reciprocal_82", "view_1288", "clamp_min_246", "div_164", "sub_3764", "maximum_82", "full_165", "minimum_82", "full_164", "view_1289", "convert_element_type_492", "clamp_max_164", "clamp_min_247", "sub_3770", "round_165", "div_165", "convert_element_type_495", "view_1293", "mul_8029", "view_1292"], "triton_poi_fused__to_copy_mul_sub_view_10:183": ["mul_8034", "sub_3794", "convert_element_type_496", "view_1295", "convert_element_type_497", "view_1297", "view_1296"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184": ["amax_83", "mul_8048", "mul_8046", "view_1300", "add_tensor_91", "add_12732", "erf_15", "mul_8047", "amin_83", "add_12815", "round_168", "mul_8097", "mul_8094", "reciprocal_83", "view_1303", "clamp_min_249", "div_166", "sub_3807", "maximum_83", "full_167", "minimum_83", "full_166", "view_1304", "convert_element_type_498", "clamp_max_166", "clamp_min_250", "sub_3813", "round_167", "div_167", "mul_8119", "sub_3833", "convert_element_type_500", "convert_element_type_499", "clamp_max_167", "clamp_min_251", "convert_element_type_501", "view_1308", "view_1307"], "triton_poi_fused__to_copy_mul_sub_view_10:185": ["mul_8124", "sub_3837", "convert_element_type_502", "view_1310", "convert_element_type_503", "view_1312", "view_1311"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186": ["var_mean_28", "clone_113", "add_12878", "view_1315", "add_tensor_90", "add_12884", "mul_8141", "mul_8140", "sub_3843", "rsqrt_28", "add_12883", "amax_84", "amin_84", "amax_85", "amin_85", "amax_86", "amin_86", "sub_3880", "convert_element_type_506", "convert_element_type_505", "clamp_max_169", "clamp_min_254", "add_12971", "round_170", "mul_8192", "mul_8189", "reciprocal_84", "view_1318", "clamp_min_252", "div_168", "sub_3854", "maximum_84", "full_169", "minimum_84", "full_168", "view_1319", "convert_element_type_504", "clamp_max_168", "clamp_min_253", "sub_3860", "round_169", "div_169", "convert_element_type_507", "view_1323", "mul_8214", "view_1322", "sub_3925", "convert_element_type_512", "convert_element_type_511", "clamp_max_171", "clamp_min_257", "add_13123", "round_172", "mul_8288", "mul_8285", "reciprocal_85", "view_1334", "clamp_min_255", "div_170", "sub_3899", "maximum_85", "full_171", "minimum_85", "full_170", "view_1335", "convert_element_type_510", "clamp_max_170", "clamp_min_256", "sub_3905", "round_171", "div_171", "convert_element_type_513", "view_1339", "mul_8310", "view_1338", "sub_3969", "convert_element_type_518", "convert_element_type_517", "clamp_max_173", "clamp_min_260", "add_13271", "round_174", "mul_8387", "mul_8384", "reciprocal_86", "view_1350", "clamp_min_258", "div_172", "sub_3943", "maximum_86", "full_173", "minimum_86", "full_172", "view_1351", "convert_element_type_516", "clamp_max_172", "clamp_min_259", "sub_3949", "round_173", "div_173", "convert_element_type_519", "view_1355", "mul_8409", "view_1354"], "triton_poi_fused__to_copy_mul_sub_view_4:187": ["mul_8219", "sub_3884", "convert_element_type_508", "view_1325", "convert_element_type_509", "view_1327", "view_1326"], "triton_poi_fused__to_copy_mul_sub_view_4:188": ["mul_8315", "sub_3929", "convert_element_type_514", "view_1341", "convert_element_type_515", "view_1343", "view_1342"], "triton_poi_fused__to_copy_mul_sub_view_4:189": ["mul_8414", "sub_3973", "convert_element_type_520", "view_1357", "convert_element_type_521", "view_1359", "view_1358"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:190": ["_scaled_dot_product_efficient_attention_14", "clone_114", "permute_142", "view_1331", "mul_8231", "view_1330", "add_tensor_89", "clone_115", "permute_144", "view_1347", "view_1346", "clone_116", "permute_146", "view_1363", "view_1362", "add_tensor_88"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:191": ["_scaled_dot_product_efficient_attention_14", "clone_114", "permute_142", "view_1331", "mul_8231", "view_1330", "add_tensor_89", "clone_115", "permute_144", "view_1347", "view_1346", "clone_116", "permute_146", "view_1363", "view_1362", "add_tensor_88"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:192": ["_scaled_dot_product_efficient_attention_14", "clone_114", "permute_142", "view_1331", "mul_8231", "view_1330", "add_tensor_89", "clone_115", "permute_144", "view_1347", "view_1346", "clone_116", "permute_146", "view_1363", "view_1362", "add_tensor_88"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193": ["amax_87", "view_1364", "permute_147", "amin_87", "sub_4017", "convert_element_type_524", "convert_element_type_523", "clamp_max_175", "clamp_min_263", "add_13435", "round_176", "mul_8492", "mul_8489", "reciprocal_87", "view_1367", "clamp_min_261", "div_174", "sub_3991", "maximum_87", "full_175", "minimum_87", "full_174", "view_1368", "convert_element_type_522", "clamp_max_174", "clamp_min_262", "sub_3997", "round_175", "div_175", "convert_element_type_525", "view_1372", "mul_8514", "view_1371"], "triton_poi_fused__to_copy_mul_sub_view_4:194": ["mul_8519", "sub_4021", "convert_element_type_526", "view_1374", "convert_element_type_527", "view_1376", "view_1375"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195": ["add_13498", "add_12878", "view_1315", "add_tensor_90", "view_1379", "add_tensor_87", "var_mean_29", "clone_118", "add_13504", "mul_8536", "mul_8535", "sub_4027", "rsqrt_29", "add_13503", "amax_88", "amin_88", "sub_4064", "convert_element_type_530", "convert_element_type_529", "clamp_max_177", "clamp_min_266", "add_13591", "round_178", "mul_8587", "mul_8584", "reciprocal_88", "view_1382", "clamp_min_264", "div_176", "sub_4038", "maximum_88", "full_177", "minimum_88", "full_176", "view_1383", "convert_element_type_528", "clamp_max_176", "clamp_min_265", "sub_4044", "round_177", "div_177", "convert_element_type_531", "view_1387", "mul_8609", "view_1386"], "triton_poi_fused__to_copy_mul_sub_view_10:196": ["mul_8614", "sub_4068", "convert_element_type_532", "view_1389", "convert_element_type_533", "view_1391", "view_1390"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197": ["amax_89", "mul_8628", "mul_8626", "view_1394", "add_tensor_86", "add_13650", "erf_16", "mul_8627", "amin_89", "add_13733", "round_180", "mul_8677", "mul_8674", "reciprocal_89", "view_1397", "clamp_min_267", "div_178", "sub_4081", "maximum_89", "full_179", "minimum_89", "full_178", "view_1398", "convert_element_type_534", "clamp_max_178", "clamp_min_268", "sub_4087", "round_179", "div_179", "mul_8699", "sub_4107", "convert_element_type_536", "convert_element_type_535", "clamp_max_179", "clamp_min_269", "convert_element_type_537", "view_1402", "view_1401"], "triton_poi_fused__to_copy_mul_sub_view_10:198": ["mul_8704", "sub_4111", "convert_element_type_538", "view_1404", "convert_element_type_539", "view_1406", "view_1405"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199": ["var_mean_30", "clone_121", "add_13796", "view_1409", "add_tensor_85", "add_13802", "mul_8721", "mul_8720", "sub_4117", "rsqrt_30", "add_13801", "amax_90", "amin_90", "amax_91", "amin_91", "amax_92", "amin_92", "sub_4154", "convert_element_type_542", "convert_element_type_541", "clamp_max_181", "clamp_min_272", "add_13889", "round_182", "mul_8772", "mul_8769", "reciprocal_90", "view_1412", "clamp_min_270", "div_180", "sub_4128", "maximum_90", "full_181", "minimum_90", "full_180", "view_1413", "convert_element_type_540", "clamp_max_180", "clamp_min_271", "sub_4134", "round_181", "div_181", "convert_element_type_543", "view_1417", "mul_8794", "view_1416", "sub_4199", "convert_element_type_548", "convert_element_type_547", "clamp_max_183", "clamp_min_275", "add_14041", "round_184", "mul_8868", "mul_8865", "reciprocal_91", "view_1428", "clamp_min_273", "div_182", "sub_4173", "maximum_91", "full_183", "minimum_91", "full_182", "view_1429", "convert_element_type_546", "clamp_max_182", "clamp_min_274", "sub_4179", "round_183", "div_183", "convert_element_type_549", "view_1433", "mul_8890", "view_1432", "sub_4243", "convert_element_type_554", "convert_element_type_553", "clamp_max_185", "clamp_min_278", "add_14189", "round_186", "mul_8967", "mul_8964", "reciprocal_92", "view_1444", "clamp_min_276", "div_184", "sub_4217", "maximum_92", "full_185", "minimum_92", "full_184", "view_1445", "convert_element_type_552", "clamp_max_184", "clamp_min_277", "sub_4223", "round_185", "div_185", "convert_element_type_555", "view_1449", "mul_8989", "view_1448"], "triton_poi_fused__to_copy_mul_sub_view_4:200": ["mul_8799", "sub_4158", "convert_element_type_544", "view_1419", "convert_element_type_545", "view_1421", "view_1420"], "triton_poi_fused__to_copy_mul_sub_view_4:201": ["mul_8895", "sub_4203", "convert_element_type_550", "view_1435", "convert_element_type_551", "view_1437", "view_1436"], "triton_poi_fused__to_copy_mul_sub_view_4:202": ["mul_8994", "sub_4247", "convert_element_type_556", "view_1451", "convert_element_type_557", "view_1453", "view_1452"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:203": ["_scaled_dot_product_efficient_attention_15", "clone_122", "permute_152", "view_1425", "mul_8811", "view_1424", "add_tensor_84", "clone_123", "permute_154", "view_1441", "view_1440", "clone_124", "permute_156", "view_1457", "view_1456", "add_tensor_83"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:204": ["_scaled_dot_product_efficient_attention_15", "clone_122", "permute_152", "view_1425", "mul_8811", "view_1424", "add_tensor_84", "clone_123", "permute_154", "view_1441", "view_1440", "clone_124", "permute_156", "view_1457", "view_1456", "add_tensor_83"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:205": ["_scaled_dot_product_efficient_attention_15", "clone_122", "permute_152", "view_1425", "mul_8811", "view_1424", "add_tensor_84", "clone_123", "permute_154", "view_1441", "view_1440", "clone_124", "permute_156", "view_1457", "view_1456", "add_tensor_83"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206": ["amax_93", "view_1458", "permute_157", "amin_93", "sub_4291", "convert_element_type_560", "convert_element_type_559", "clamp_max_187", "clamp_min_281", "add_14353", "round_188", "mul_9072", "mul_9069", "reciprocal_93", "view_1461", "clamp_min_279", "div_186", "sub_4265", "maximum_93", "full_187", "minimum_93", "full_186", "view_1462", "convert_element_type_558", "clamp_max_186", "clamp_min_280", "sub_4271", "round_187", "div_187", "convert_element_type_561", "view_1466", "mul_9094", "view_1465"], "triton_poi_fused__to_copy_mul_sub_view_4:207": ["mul_9099", "sub_4295", "convert_element_type_562", "view_1468", "convert_element_type_563", "view_1470", "view_1469"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208": ["add_14416", "add_13796", "view_1409", "add_tensor_85", "view_1473", "add_tensor_82", "var_mean_31", "clone_126", "add_14422", "mul_9116", "mul_9115", "sub_4301", "rsqrt_31", "add_14421", "amax_94", "amin_94", "sub_4338", "convert_element_type_566", "convert_element_type_565", "clamp_max_189", "clamp_min_284", "add_14509", "round_190", "mul_9167", "mul_9164", "reciprocal_94", "view_1476", "clamp_min_282", "div_188", "sub_4312", "maximum_94", "full_189", "minimum_94", "full_188", "view_1477", "convert_element_type_564", "clamp_max_188", "clamp_min_283", "sub_4318", "round_189", "div_189", "convert_element_type_567", "view_1481", "mul_9189", "view_1480"], "triton_poi_fused__to_copy_mul_sub_view_10:209": ["mul_9194", "sub_4342", "convert_element_type_568", "view_1483", "convert_element_type_569", "view_1485", "view_1484"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210": ["amax_95", "mul_9208", "mul_9206", "view_1488", "add_tensor_81", "add_14568", "erf_17", "mul_9207", "amin_95", "add_14651", "round_192", "mul_9257", "mul_9254", "reciprocal_95", "view_1491", "clamp_min_285", "div_190", "sub_4355", "maximum_95", "full_191", "minimum_95", "full_190", "view_1492", "convert_element_type_570", "clamp_max_190", "clamp_min_286", "sub_4361", "round_191", "div_191", "mul_9279", "sub_4381", "convert_element_type_572", "convert_element_type_571", "clamp_max_191", "clamp_min_287", "convert_element_type_573", "view_1496", "view_1495"], "triton_poi_fused__to_copy_mul_sub_view_10:211": ["mul_9284", "sub_4385", "convert_element_type_574", "view_1498", "convert_element_type_575", "view_1500", "view_1499"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212": ["var_mean_32", "clone_129", "add_14714", "view_1503", "add_tensor_80", "add_14720", "mul_9301", "mul_9300", "sub_4391", "rsqrt_32", "add_14719", "amax_96", "amin_96", "amax_97", "amin_97", "amax_98", "amin_98", "sub_4428", "convert_element_type_578", "convert_element_type_577", "clamp_max_193", "clamp_min_290", "add_14807", "round_194", "mul_9352", "mul_9349", "reciprocal_96", "view_1506", "clamp_min_288", "div_192", "sub_4402", "maximum_96", "full_193", "minimum_96", "full_192", "view_1507", "convert_element_type_576", "clamp_max_192", "clamp_min_289", "sub_4408", "round_193", "div_193", "convert_element_type_579", "view_1511", "mul_9374", "view_1510", "sub_4473", "convert_element_type_584", "convert_element_type_583", "clamp_max_195", "clamp_min_293", "add_14959", "round_196", "mul_9448", "mul_9445", "reciprocal_97", "view_1522", "clamp_min_291", "div_194", "sub_4447", "maximum_97", "full_195", "minimum_97", "full_194", "view_1523", "convert_element_type_582", "clamp_max_194", "clamp_min_292", "sub_4453", "round_195", "div_195", "convert_element_type_585", "view_1527", "mul_9470", "view_1526", "sub_4517", "convert_element_type_590", "convert_element_type_589", "clamp_max_197", "clamp_min_296", "add_15107", "round_198", "mul_9547", "mul_9544", "reciprocal_98", "view_1538", "clamp_min_294", "div_196", "sub_4491", "maximum_98", "full_197", "minimum_98", "full_196", "view_1539", "convert_element_type_588", "clamp_max_196", "clamp_min_295", "sub_4497", "round_197", "div_197", "convert_element_type_591", "view_1543", "mul_9569", "view_1542"], "triton_poi_fused__to_copy_mul_sub_view_4:213": ["mul_9379", "sub_4432", "convert_element_type_580", "view_1513", "convert_element_type_581", "view_1515", "view_1514"], "triton_poi_fused__to_copy_mul_sub_view_4:214": ["mul_9475", "sub_4477", "convert_element_type_586", "view_1529", "convert_element_type_587", "view_1531", "view_1530"], "triton_poi_fused__to_copy_mul_sub_view_4:215": ["mul_9574", "sub_4521", "convert_element_type_592", "view_1545", "convert_element_type_593", "view_1547", "view_1546"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:216": ["_scaled_dot_product_efficient_attention_16", "clone_130", "permute_162", "view_1519", "mul_9391", "view_1518", "add_tensor_79", "clone_131", "permute_164", "view_1535", "view_1534", "clone_132", "permute_166", "view_1551", "view_1550", "add_tensor_78"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:217": ["_scaled_dot_product_efficient_attention_16", "clone_130", "permute_162", "view_1519", "mul_9391", "view_1518", "add_tensor_79", "clone_131", "permute_164", "view_1535", "view_1534", "clone_132", "permute_166", "view_1551", "view_1550", "add_tensor_78"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:218": ["_scaled_dot_product_efficient_attention_16", "clone_130", "permute_162", "view_1519", "mul_9391", "view_1518", "add_tensor_79", "clone_131", "permute_164", "view_1535", "view_1534", "clone_132", "permute_166", "view_1551", "view_1550", "add_tensor_78"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219": ["amax_99", "view_1552", "permute_167", "amin_99", "sub_4565", "convert_element_type_596", "convert_element_type_595", "clamp_max_199", "clamp_min_299", "add_15271", "round_200", "mul_9652", "mul_9649", "reciprocal_99", "view_1555", "clamp_min_297", "div_198", "sub_4539", "maximum_99", "full_199", "minimum_99", "full_198", "view_1556", "convert_element_type_594", "clamp_max_198", "clamp_min_298", "sub_4545", "round_199", "div_199", "convert_element_type_597", "view_1560", "mul_9674", "view_1559"], "triton_poi_fused__to_copy_mul_sub_view_4:220": ["mul_9679", "sub_4569", "convert_element_type_598", "view_1562", "convert_element_type_599", "view_1564", "view_1563"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221": ["add_15334", "add_14714", "view_1503", "add_tensor_80", "view_1567", "add_tensor_77", "var_mean_33", "clone_134", "add_15340", "mul_9696", "mul_9695", "sub_4575", "rsqrt_33", "add_15339", "amax_100", "amin_100", "sub_4612", "convert_element_type_602", "convert_element_type_601", "clamp_max_201", "clamp_min_302", "add_15427", "round_202", "mul_9747", "mul_9744", "reciprocal_100", "view_1570", "clamp_min_300", "div_200", "sub_4586", "maximum_100", "full_201", "minimum_100", "full_200", "view_1571", "convert_element_type_600", "clamp_max_200", "clamp_min_301", "sub_4592", "round_201", "div_201", "convert_element_type_603", "view_1575", "mul_9769", "view_1574"], "triton_poi_fused__to_copy_mul_sub_view_10:222": ["mul_9774", "sub_4616", "convert_element_type_604", "view_1577", "convert_element_type_605", "view_1579", "view_1578"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223": ["amax_101", "mul_9788", "mul_9786", "view_1582", "add_tensor_76", "add_15486", "erf_18", "mul_9787", "amin_101", "add_15569", "round_204", "mul_9837", "mul_9834", "reciprocal_101", "view_1585", "clamp_min_303", "div_202", "sub_4629", "maximum_101", "full_203", "minimum_101", "full_202", "view_1586", "convert_element_type_606", "clamp_max_202", "clamp_min_304", "sub_4635", "round_203", "div_203", "mul_9859", "sub_4655", "convert_element_type_608", "convert_element_type_607", "clamp_max_203", "clamp_min_305", "convert_element_type_609", "view_1590", "view_1589"], "triton_poi_fused__to_copy_mul_sub_view_10:224": ["mul_9864", "sub_4659", "convert_element_type_610", "view_1592", "convert_element_type_611", "view_1594", "view_1593"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225": ["var_mean_34", "clone_137", "add_15632", "view_1597", "add_tensor_75", "add_15638", "mul_9881", "mul_9880", "sub_4665", "rsqrt_34", "add_15637", "amax_102", "amin_102", "amax_103", "amin_103", "amax_104", "amin_104", "sub_4702", "convert_element_type_614", "convert_element_type_613", "clamp_max_205", "clamp_min_308", "add_15725", "round_206", "mul_9932", "mul_9929", "reciprocal_102", "view_1600", "clamp_min_306", "div_204", "sub_4676", "maximum_102", "full_205", "minimum_102", "full_204", "view_1601", "convert_element_type_612", "clamp_max_204", "clamp_min_307", "sub_4682", "round_205", "div_205", "convert_element_type_615", "view_1605", "mul_9954", "view_1604", "sub_4747", "convert_element_type_620", "convert_element_type_619", "clamp_max_207", "clamp_min_311", "add_15877", "round_208", "mul_10028", "mul_10025", "reciprocal_103", "view_1616", "clamp_min_309", "div_206", "sub_4721", "maximum_103", "full_207", "minimum_103", "full_206", "view_1617", "convert_element_type_618", "clamp_max_206", "clamp_min_310", "sub_4727", "round_207", "div_207", "convert_element_type_621", "view_1621", "mul_10050", "view_1620", "sub_4791", "convert_element_type_626", "convert_element_type_625", "clamp_max_209", "clamp_min_314", "add_16025", "round_210", "mul_10127", "mul_10124", "reciprocal_104", "view_1632", "clamp_min_312", "div_208", "sub_4765", "maximum_104", "full_209", "minimum_104", "full_208", "view_1633", "convert_element_type_624", "clamp_max_208", "clamp_min_313", "sub_4771", "round_209", "div_209", "convert_element_type_627", "view_1637", "mul_10149", "view_1636"], "triton_poi_fused__to_copy_mul_sub_view_4:226": ["mul_9959", "sub_4706", "convert_element_type_616", "view_1607", "convert_element_type_617", "view_1609", "view_1608"], "triton_poi_fused__to_copy_mul_sub_view_4:227": ["mul_10055", "sub_4751", "convert_element_type_622", "view_1623", "convert_element_type_623", "view_1625", "view_1624"], "triton_poi_fused__to_copy_mul_sub_view_4:228": ["mul_10154", "sub_4795", "convert_element_type_628", "view_1639", "convert_element_type_629", "view_1641", "view_1640"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:229": ["_scaled_dot_product_efficient_attention_17", "clone_138", "permute_172", "view_1613", "mul_9971", "view_1612", "add_tensor_74", "clone_139", "permute_174", "view_1629", "view_1628", "clone_140", "permute_176", "view_1645", "view_1644", "add_tensor_73"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:230": ["_scaled_dot_product_efficient_attention_17", "clone_138", "permute_172", "view_1613", "mul_9971", "view_1612", "add_tensor_74", "clone_139", "permute_174", "view_1629", "view_1628", "clone_140", "permute_176", "view_1645", "view_1644", "add_tensor_73"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:231": ["_scaled_dot_product_efficient_attention_17", "clone_138", "permute_172", "view_1613", "mul_9971", "view_1612", "add_tensor_74", "clone_139", "permute_174", "view_1629", "view_1628", "clone_140", "permute_176", "view_1645", "view_1644", "add_tensor_73"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232": ["amax_105", "view_1646", "permute_177", "amin_105", "sub_4839", "convert_element_type_632", "convert_element_type_631", "clamp_max_211", "clamp_min_317", "add_16189", "round_212", "mul_10232", "mul_10229", "reciprocal_105", "view_1649", "clamp_min_315", "div_210", "sub_4813", "maximum_105", "full_211", "minimum_105", "full_210", "view_1650", "convert_element_type_630", "clamp_max_210", "clamp_min_316", "sub_4819", "round_211", "div_211", "convert_element_type_633", "view_1654", "mul_10254", "view_1653"], "triton_poi_fused__to_copy_mul_sub_view_4:233": ["mul_10259", "sub_4843", "convert_element_type_634", "view_1656", "convert_element_type_635", "view_1658", "view_1657"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234": ["add_16252", "add_15632", "view_1597", "add_tensor_75", "view_1661", "add_tensor_72", "var_mean_35", "clone_142", "add_16258", "mul_10276", "mul_10275", "sub_4849", "rsqrt_35", "add_16257", "amax_106", "amin_106", "sub_4886", "convert_element_type_638", "convert_element_type_637", "clamp_max_213", "clamp_min_320", "add_16345", "round_214", "mul_10327", "mul_10324", "reciprocal_106", "view_1664", "clamp_min_318", "div_212", "sub_4860", "maximum_106", "full_213", "minimum_106", "full_212", "view_1665", "convert_element_type_636", "clamp_max_212", "clamp_min_319", "sub_4866", "round_213", "div_213", "convert_element_type_639", "view_1669", "mul_10349", "view_1668"], "triton_poi_fused__to_copy_mul_sub_view_10:235": ["mul_10354", "sub_4890", "convert_element_type_640", "view_1671", "convert_element_type_641", "view_1673", "view_1672"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236": ["amax_107", "mul_10368", "mul_10366", "view_1676", "add_tensor_71", "add_16404", "erf_19", "mul_10367", "amin_107", "add_16487", "round_216", "mul_10417", "mul_10414", "reciprocal_107", "view_1679", "clamp_min_321", "div_214", "sub_4903", "maximum_107", "full_215", "minimum_107", "full_214", "view_1680", "convert_element_type_642", "clamp_max_214", "clamp_min_322", "sub_4909", "round_215", "div_215", "mul_10439", "sub_4929", "convert_element_type_644", "convert_element_type_643", "clamp_max_215", "clamp_min_323", "convert_element_type_645", "view_1684", "view_1683"], "triton_poi_fused__to_copy_mul_sub_view_10:237": ["mul_10444", "sub_4933", "convert_element_type_646", "view_1686", "convert_element_type_647", "view_1688", "view_1687"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238": ["var_mean_36", "clone_145", "add_16550", "view_1691", "add_tensor_70", "add_16556", "mul_10461", "mul_10460", "sub_4939", "rsqrt_36", "add_16555", "amax_108", "amin_108", "amax_109", "amin_109", "amax_110", "amin_110", "sub_4976", "convert_element_type_650", "convert_element_type_649", "clamp_max_217", "clamp_min_326", "add_16643", "round_218", "mul_10512", "mul_10509", "reciprocal_108", "view_1694", "clamp_min_324", "div_216", "sub_4950", "maximum_108", "full_217", "minimum_108", "full_216", "view_1695", "convert_element_type_648", "clamp_max_216", "clamp_min_325", "sub_4956", "round_217", "div_217", "convert_element_type_651", "view_1699", "mul_10534", "view_1698", "sub_5021", "convert_element_type_656", "convert_element_type_655", "clamp_max_219", "clamp_min_329", "add_16795", "round_220", "mul_10608", "mul_10605", "reciprocal_109", "view_1710", "clamp_min_327", "div_218", "sub_4995", "maximum_109", "full_219", "minimum_109", "full_218", "view_1711", "convert_element_type_654", "clamp_max_218", "clamp_min_328", "sub_5001", "round_219", "div_219", "convert_element_type_657", "view_1715", "mul_10630", "view_1714", "sub_5065", "convert_element_type_662", "convert_element_type_661", "clamp_max_221", "clamp_min_332", "add_16943", "round_222", "mul_10707", "mul_10704", "reciprocal_110", "view_1726", "clamp_min_330", "div_220", "sub_5039", "maximum_110", "full_221", "minimum_110", "full_220", "view_1727", "convert_element_type_660", "clamp_max_220", "clamp_min_331", "sub_5045", "round_221", "div_221", "convert_element_type_663", "view_1731", "mul_10729", "view_1730"], "triton_poi_fused__to_copy_mul_sub_view_4:239": ["mul_10539", "sub_4980", "convert_element_type_652", "view_1701", "convert_element_type_653", "view_1703", "view_1702"], "triton_poi_fused__to_copy_mul_sub_view_4:240": ["mul_10635", "sub_5025", "convert_element_type_658", "view_1717", "convert_element_type_659", "view_1719", "view_1718"], "triton_poi_fused__to_copy_mul_sub_view_4:241": ["mul_10734", "sub_5069", "convert_element_type_664", "view_1733", "convert_element_type_665", "view_1735", "view_1734"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:242": ["_scaled_dot_product_efficient_attention_18", "clone_146", "permute_182", "view_1707", "mul_10551", "view_1706", "add_tensor_69", "clone_147", "permute_184", "view_1723", "view_1722", "clone_148", "permute_186", "view_1739", "view_1738", "add_tensor_68"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:243": ["_scaled_dot_product_efficient_attention_18", "clone_146", "permute_182", "view_1707", "mul_10551", "view_1706", "add_tensor_69", "clone_147", "permute_184", "view_1723", "view_1722", "clone_148", "permute_186", "view_1739", "view_1738", "add_tensor_68"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:244": ["_scaled_dot_product_efficient_attention_18", "clone_146", "permute_182", "view_1707", "mul_10551", "view_1706", "add_tensor_69", "clone_147", "permute_184", "view_1723", "view_1722", "clone_148", "permute_186", "view_1739", "view_1738", "add_tensor_68"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245": ["amax_111", "view_1740", "permute_187", "amin_111", "sub_5113", "convert_element_type_668", "convert_element_type_667", "clamp_max_223", "clamp_min_335", "add_17107", "round_224", "mul_10812", "mul_10809", "reciprocal_111", "view_1743", "clamp_min_333", "div_222", "sub_5087", "maximum_111", "full_223", "minimum_111", "full_222", "view_1744", "convert_element_type_666", "clamp_max_222", "clamp_min_334", "sub_5093", "round_223", "div_223", "convert_element_type_669", "view_1748", "mul_10834", "view_1747"], "triton_poi_fused__to_copy_mul_sub_view_4:246": ["mul_10839", "sub_5117", "convert_element_type_670", "view_1750", "convert_element_type_671", "view_1752", "view_1751"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247": ["add_17170", "add_16550", "view_1691", "add_tensor_70", "view_1755", "add_tensor_67", "var_mean_37", "clone_150", "add_17176", "mul_10856", "mul_10855", "sub_5123", "rsqrt_37", "add_17175", "amax_112", "amin_112", "sub_5160", "convert_element_type_674", "convert_element_type_673", "clamp_max_225", "clamp_min_338", "add_17263", "round_226", "mul_10907", "mul_10904", "reciprocal_112", "view_1758", "clamp_min_336", "div_224", "sub_5134", "maximum_112", "full_225", "minimum_112", "full_224", "view_1759", "convert_element_type_672", "clamp_max_224", "clamp_min_337", "sub_5140", "round_225", "div_225", "convert_element_type_675", "view_1763", "mul_10929", "view_1762"], "triton_poi_fused__to_copy_mul_sub_view_10:248": ["mul_10934", "sub_5164", "convert_element_type_676", "view_1765", "convert_element_type_677", "view_1767", "view_1766"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249": ["amax_113", "mul_10948", "mul_10946", "view_1770", "add_tensor_66", "add_17322", "erf_20", "mul_10947", "amin_113", "add_17405", "round_228", "mul_10997", "mul_10994", "reciprocal_113", "view_1773", "clamp_min_339", "div_226", "sub_5177", "maximum_113", "full_227", "minimum_113", "full_226", "view_1774", "convert_element_type_678", "clamp_max_226", "clamp_min_340", "sub_5183", "round_227", "div_227", "mul_11019", "sub_5203", "convert_element_type_680", "convert_element_type_679", "clamp_max_227", "clamp_min_341", "convert_element_type_681", "view_1778", "view_1777"], "triton_poi_fused__to_copy_mul_sub_view_10:250": ["mul_11024", "sub_5207", "convert_element_type_682", "view_1780", "convert_element_type_683", "view_1782", "view_1781"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251": ["var_mean_38", "clone_153", "add_17468", "view_1785", "add_tensor_65", "add_17474", "mul_11041", "mul_11040", "sub_5213", "rsqrt_38", "add_17473", "amax_114", "amin_114", "amax_115", "amin_115", "amax_116", "amin_116", "sub_5250", "convert_element_type_686", "convert_element_type_685", "clamp_max_229", "clamp_min_344", "add_17561", "round_230", "mul_11092", "mul_11089", "reciprocal_114", "view_1788", "clamp_min_342", "div_228", "sub_5224", "maximum_114", "full_229", "minimum_114", "full_228", "view_1789", "convert_element_type_684", "clamp_max_228", "clamp_min_343", "sub_5230", "round_229", "div_229", "convert_element_type_687", "view_1793", "mul_11114", "view_1792", "sub_5295", "convert_element_type_692", "convert_element_type_691", "clamp_max_231", "clamp_min_347", "add_17713", "round_232", "mul_11188", "mul_11185", "reciprocal_115", "view_1804", "clamp_min_345", "div_230", "sub_5269", "maximum_115", "full_231", "minimum_115", "full_230", "view_1805", "convert_element_type_690", "clamp_max_230", "clamp_min_346", "sub_5275", "round_231", "div_231", "convert_element_type_693", "view_1809", "mul_11210", "view_1808", "sub_5339", "convert_element_type_698", "convert_element_type_697", "clamp_max_233", "clamp_min_350", "add_17861", "round_234", "mul_11287", "mul_11284", "reciprocal_116", "view_1820", "clamp_min_348", "div_232", "sub_5313", "maximum_116", "full_233", "minimum_116", "full_232", "view_1821", "convert_element_type_696", "clamp_max_232", "clamp_min_349", "sub_5319", "round_233", "div_233", "convert_element_type_699", "view_1825", "mul_11309", "view_1824"], "triton_poi_fused__to_copy_mul_sub_view_4:252": ["mul_11119", "sub_5254", "convert_element_type_688", "view_1795", "convert_element_type_689", "view_1797", "view_1796"], "triton_poi_fused__to_copy_mul_sub_view_4:253": ["mul_11215", "sub_5299", "convert_element_type_694", "view_1811", "convert_element_type_695", "view_1813", "view_1812"], "triton_poi_fused__to_copy_mul_sub_view_4:254": ["mul_11314", "sub_5343", "convert_element_type_700", "view_1827", "convert_element_type_701", "view_1829", "view_1828"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:255": ["_scaled_dot_product_efficient_attention_19", "clone_154", "permute_192", "view_1801", "mul_11131", "view_1800", "add_tensor_64", "clone_155", "permute_194", "view_1817", "view_1816", "clone_156", "permute_196", "view_1833", "view_1832", "add_tensor_63"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:256": ["_scaled_dot_product_efficient_attention_19", "clone_154", "permute_192", "view_1801", "mul_11131", "view_1800", "add_tensor_64", "clone_155", "permute_194", "view_1817", "view_1816", "clone_156", "permute_196", "view_1833", "view_1832", "add_tensor_63"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:257": ["_scaled_dot_product_efficient_attention_19", "clone_154", "permute_192", "view_1801", "mul_11131", "view_1800", "add_tensor_64", "clone_155", "permute_194", "view_1817", "view_1816", "clone_156", "permute_196", "view_1833", "view_1832", "add_tensor_63"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258": ["amax_117", "view_1834", "permute_197", "amin_117", "sub_5387", "convert_element_type_704", "convert_element_type_703", "clamp_max_235", "clamp_min_353", "add_18025", "round_236", "mul_11392", "mul_11389", "reciprocal_117", "view_1837", "clamp_min_351", "div_234", "sub_5361", "maximum_117", "full_235", "minimum_117", "full_234", "view_1838", "convert_element_type_702", "clamp_max_234", "clamp_min_352", "sub_5367", "round_235", "div_235", "convert_element_type_705", "view_1842", "mul_11414", "view_1841"], "triton_poi_fused__to_copy_mul_sub_view_4:259": ["mul_11419", "sub_5391", "convert_element_type_706", "view_1844", "convert_element_type_707", "view_1846", "view_1845"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260": ["add_18088", "add_17468", "view_1785", "add_tensor_65", "view_1849", "add_tensor_62", "var_mean_39", "clone_158", "add_18094", "mul_11436", "mul_11435", "sub_5397", "rsqrt_39", "add_18093", "amax_118", "amin_118", "sub_5434", "convert_element_type_710", "convert_element_type_709", "clamp_max_237", "clamp_min_356", "add_18181", "round_238", "mul_11487", "mul_11484", "reciprocal_118", "view_1852", "clamp_min_354", "div_236", "sub_5408", "maximum_118", "full_237", "minimum_118", "full_236", "view_1853", "convert_element_type_708", "clamp_max_236", "clamp_min_355", "sub_5414", "round_237", "div_237", "convert_element_type_711", "view_1857", "mul_11509", "view_1856"], "triton_poi_fused__to_copy_mul_sub_view_10:261": ["mul_11514", "sub_5438", "convert_element_type_712", "view_1859", "convert_element_type_713", "view_1861", "view_1860"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262": ["amax_119", "mul_11528", "mul_11526", "view_1864", "add_tensor_61", "add_18240", "erf_21", "mul_11527", "amin_119", "add_18323", "round_240", "mul_11577", "mul_11574", "reciprocal_119", "view_1867", "clamp_min_357", "div_238", "sub_5451", "maximum_119", "full_239", "minimum_119", "full_238", "view_1868", "convert_element_type_714", "clamp_max_238", "clamp_min_358", "sub_5457", "round_239", "div_239", "mul_11599", "sub_5477", "convert_element_type_716", "convert_element_type_715", "clamp_max_239", "clamp_min_359", "convert_element_type_717", "view_1872", "view_1871"], "triton_poi_fused__to_copy_mul_sub_view_10:263": ["mul_11604", "sub_5481", "convert_element_type_718", "view_1874", "convert_element_type_719", "view_1876", "view_1875"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264": ["var_mean_40", "clone_161", "add_18386", "view_1879", "add_tensor_60", "add_18392", "mul_11621", "mul_11620", "sub_5487", "rsqrt_40", "add_18391", "amax_120", "amin_120", "amax_121", "amin_121", "amax_122", "amin_122", "sub_5524", "convert_element_type_722", "convert_element_type_721", "clamp_max_241", "clamp_min_362", "add_18479", "round_242", "mul_11672", "mul_11669", "reciprocal_120", "view_1882", "clamp_min_360", "div_240", "sub_5498", "maximum_120", "full_241", "minimum_120", "full_240", "view_1883", "convert_element_type_720", "clamp_max_240", "clamp_min_361", "sub_5504", "round_241", "div_241", "convert_element_type_723", "view_1887", "mul_11694", "view_1886", "sub_5569", "convert_element_type_728", "convert_element_type_727", "clamp_max_243", "clamp_min_365", "add_18631", "round_244", "mul_11768", "mul_11765", "reciprocal_121", "view_1898", "clamp_min_363", "div_242", "sub_5543", "maximum_121", "full_243", "minimum_121", "full_242", "view_1899", "convert_element_type_726", "clamp_max_242", "clamp_min_364", "sub_5549", "round_243", "div_243", "convert_element_type_729", "view_1903", "mul_11790", "view_1902", "sub_5613", "convert_element_type_734", "convert_element_type_733", "clamp_max_245", "clamp_min_368", "add_18779", "round_246", "mul_11867", "mul_11864", "reciprocal_122", "view_1914", "clamp_min_366", "div_244", "sub_5587", "maximum_122", "full_245", "minimum_122", "full_244", "view_1915", "convert_element_type_732", "clamp_max_244", "clamp_min_367", "sub_5593", "round_245", "div_245", "convert_element_type_735", "view_1919", "mul_11889", "view_1918"], "triton_poi_fused__to_copy_mul_sub_view_4:265": ["mul_11699", "sub_5528", "convert_element_type_724", "view_1889", "convert_element_type_725", "view_1891", "view_1890"], "triton_poi_fused__to_copy_mul_sub_view_4:266": ["mul_11795", "sub_5573", "convert_element_type_730", "view_1905", "convert_element_type_731", "view_1907", "view_1906"], "triton_poi_fused__to_copy_mul_sub_view_4:267": ["mul_11894", "sub_5617", "convert_element_type_736", "view_1921", "convert_element_type_737", "view_1923", "view_1922"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:268": ["_scaled_dot_product_efficient_attention_20", "clone_162", "permute_202", "view_1895", "mul_11711", "view_1894", "add_tensor_59", "clone_163", "permute_204", "view_1911", "view_1910", "clone_164", "permute_206", "view_1927", "view_1926", "add_tensor_58"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:269": ["_scaled_dot_product_efficient_attention_20", "clone_162", "permute_202", "view_1895", "mul_11711", "view_1894", "add_tensor_59", "clone_163", "permute_204", "view_1911", "view_1910", "clone_164", "permute_206", "view_1927", "view_1926", "add_tensor_58"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:270": ["_scaled_dot_product_efficient_attention_20", "clone_162", "permute_202", "view_1895", "mul_11711", "view_1894", "add_tensor_59", "clone_163", "permute_204", "view_1911", "view_1910", "clone_164", "permute_206", "view_1927", "view_1926", "add_tensor_58"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271": ["amax_123", "view_1928", "permute_207", "amin_123", "sub_5661", "convert_element_type_740", "convert_element_type_739", "clamp_max_247", "clamp_min_371", "add_18943", "round_248", "mul_11972", "mul_11969", "reciprocal_123", "view_1931", "clamp_min_369", "div_246", "sub_5635", "maximum_123", "full_247", "minimum_123", "full_246", "view_1932", "convert_element_type_738", "clamp_max_246", "clamp_min_370", "sub_5641", "round_247", "div_247", "convert_element_type_741", "view_1936", "mul_11994", "view_1935"], "triton_poi_fused__to_copy_mul_sub_view_4:272": ["mul_11999", "sub_5665", "convert_element_type_742", "view_1938", "convert_element_type_743", "view_1940", "view_1939"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273": ["add_19006", "add_18386", "view_1879", "add_tensor_60", "view_1943", "add_tensor_57", "var_mean_41", "clone_166", "add_19012", "mul_12016", "mul_12015", "sub_5671", "rsqrt_41", "add_19011", "amax_124", "amin_124", "sub_5708", "convert_element_type_746", "convert_element_type_745", "clamp_max_249", "clamp_min_374", "add_19099", "round_250", "mul_12067", "mul_12064", "reciprocal_124", "view_1946", "clamp_min_372", "div_248", "sub_5682", "maximum_124", "full_249", "minimum_124", "full_248", "view_1947", "convert_element_type_744", "clamp_max_248", "clamp_min_373", "sub_5688", "round_249", "div_249", "convert_element_type_747", "view_1951", "mul_12089", "view_1950"], "triton_poi_fused__to_copy_mul_sub_view_10:274": ["mul_12094", "sub_5712", "convert_element_type_748", "view_1953", "convert_element_type_749", "view_1955", "view_1954"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275": ["amax_125", "mul_12108", "mul_12106", "view_1958", "add_tensor_56", "add_19158", "erf_22", "mul_12107", "amin_125", "add_19241", "round_252", "mul_12157", "mul_12154", "reciprocal_125", "view_1961", "clamp_min_375", "div_250", "sub_5725", "maximum_125", "full_251", "minimum_125", "full_250", "view_1962", "convert_element_type_750", "clamp_max_250", "clamp_min_376", "sub_5731", "round_251", "div_251", "mul_12179", "sub_5751", "convert_element_type_752", "convert_element_type_751", "clamp_max_251", "clamp_min_377", "convert_element_type_753", "view_1966", "view_1965"], "triton_poi_fused__to_copy_mul_sub_view_10:276": ["mul_12184", "sub_5755", "convert_element_type_754", "view_1968", "convert_element_type_755", "view_1970", "view_1969"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277": ["var_mean_42", "clone_169", "add_19304", "view_1973", "add_tensor_55", "add_19310", "mul_12201", "mul_12200", "sub_5761", "rsqrt_42", "add_19309", "amax_126", "amin_126", "amax_127", "amin_127", "amax_128", "amin_128", "sub_5798", "convert_element_type_758", "convert_element_type_757", "clamp_max_253", "clamp_min_380", "add_19397", "round_254", "mul_12252", "mul_12249", "reciprocal_126", "view_1976", "clamp_min_378", "div_252", "sub_5772", "maximum_126", "full_253", "minimum_126", "full_252", "view_1977", "convert_element_type_756", "clamp_max_252", "clamp_min_379", "sub_5778", "round_253", "div_253", "convert_element_type_759", "view_1981", "mul_12274", "view_1980", "sub_5843", "convert_element_type_764", "convert_element_type_763", "clamp_max_255", "clamp_min_383", "add_19549", "round_256", "mul_12348", "mul_12345", "reciprocal_127", "view_1992", "clamp_min_381", "div_254", "sub_5817", "maximum_127", "full_255", "minimum_127", "full_254", "view_1993", "convert_element_type_762", "clamp_max_254", "clamp_min_382", "sub_5823", "round_255", "div_255", "convert_element_type_765", "view_1997", "mul_12370", "view_1996", "sub_5887", "convert_element_type_770", "convert_element_type_769", "clamp_max_257", "clamp_min_386", "add_19697", "round_258", "mul_12447", "mul_12444", "reciprocal_128", "view_2008", "clamp_min_384", "div_256", "sub_5861", "maximum_128", "full_257", "minimum_128", "full_256", "view_2009", "convert_element_type_768", "clamp_max_256", "clamp_min_385", "sub_5867", "round_257", "div_257", "convert_element_type_771", "view_2013", "mul_12469", "view_2012"], "triton_poi_fused__to_copy_mul_sub_view_4:278": ["mul_12279", "sub_5802", "convert_element_type_760", "view_1983", "convert_element_type_761", "view_1985", "view_1984"], "triton_poi_fused__to_copy_mul_sub_view_4:279": ["mul_12375", "sub_5847", "convert_element_type_766", "view_1999", "convert_element_type_767", "view_2001", "view_2000"], "triton_poi_fused__to_copy_mul_sub_view_4:280": ["mul_12474", "sub_5891", "convert_element_type_772", "view_2015", "convert_element_type_773", "view_2017", "view_2016"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:281": ["_scaled_dot_product_efficient_attention_21", "clone_170", "permute_212", "view_1989", "mul_12291", "view_1988", "add_tensor_54", "clone_171", "permute_214", "view_2005", "view_2004", "clone_172", "permute_216", "view_2021", "view_2020", "add_tensor_53"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:282": ["_scaled_dot_product_efficient_attention_21", "clone_170", "permute_212", "view_1989", "mul_12291", "view_1988", "add_tensor_54", "clone_171", "permute_214", "view_2005", "view_2004", "clone_172", "permute_216", "view_2021", "view_2020", "add_tensor_53"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:283": ["_scaled_dot_product_efficient_attention_21", "clone_170", "permute_212", "view_1989", "mul_12291", "view_1988", "add_tensor_54", "clone_171", "permute_214", "view_2005", "view_2004", "clone_172", "permute_216", "view_2021", "view_2020", "add_tensor_53"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284": ["amax_129", "view_2022", "permute_217", "amin_129", "sub_5935", "convert_element_type_776", "convert_element_type_775", "clamp_max_259", "clamp_min_389", "add_19861", "round_260", "mul_12552", "mul_12549", "reciprocal_129", "view_2025", "clamp_min_387", "div_258", "sub_5909", "maximum_129", "full_259", "minimum_129", "full_258", "view_2026", "convert_element_type_774", "clamp_max_258", "clamp_min_388", "sub_5915", "round_259", "div_259", "convert_element_type_777", "view_2030", "mul_12574", "view_2029"], "triton_poi_fused__to_copy_mul_sub_view_4:285": ["mul_12579", "sub_5939", "convert_element_type_778", "view_2032", "convert_element_type_779", "view_2034", "view_2033"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286": ["add_19924", "add_19304", "view_1973", "add_tensor_55", "view_2037", "add_tensor_52", "var_mean_43", "clone_174", "add_19930", "mul_12596", "mul_12595", "sub_5945", "rsqrt_43", "add_19929", "amax_130", "amin_130", "sub_5982", "convert_element_type_782", "convert_element_type_781", "clamp_max_261", "clamp_min_392", "add_20017", "round_262", "mul_12647", "mul_12644", "reciprocal_130", "view_2040", "clamp_min_390", "div_260", "sub_5956", "maximum_130", "full_261", "minimum_130", "full_260", "view_2041", "convert_element_type_780", "clamp_max_260", "clamp_min_391", "sub_5962", "round_261", "div_261", "convert_element_type_783", "view_2045", "mul_12669", "view_2044"], "triton_poi_fused__to_copy_mul_sub_view_10:287": ["mul_12674", "sub_5986", "convert_element_type_784", "view_2047", "convert_element_type_785", "view_2049", "view_2048"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288": ["amax_131", "mul_12688", "mul_12686", "view_2052", "add_tensor_51", "add_20076", "erf_23", "mul_12687", "amin_131", "add_20159", "round_264", "mul_12737", "mul_12734", "reciprocal_131", "view_2055", "clamp_min_393", "div_262", "sub_5999", "maximum_131", "full_263", "minimum_131", "full_262", "view_2056", "convert_element_type_786", "clamp_max_262", "clamp_min_394", "sub_6005", "round_263", "div_263", "mul_12759", "sub_6025", "convert_element_type_788", "convert_element_type_787", "clamp_max_263", "clamp_min_395", "convert_element_type_789", "view_2060", "view_2059"], "triton_poi_fused__to_copy_mul_sub_view_10:289": ["mul_12764", "sub_6029", "convert_element_type_790", "view_2062", "convert_element_type_791", "view_2064", "view_2063"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290": ["var_mean_44", "clone_177", "add_20222", "view_2067", "add_tensor_50", "add_20228", "mul_12781", "mul_12780", "sub_6035", "rsqrt_44", "add_20227", "amax_132", "amin_132", "amax_133", "amin_133", "amax_134", "amin_134", "sub_6072", "convert_element_type_794", "convert_element_type_793", "clamp_max_265", "clamp_min_398", "add_20315", "round_266", "mul_12832", "mul_12829", "reciprocal_132", "view_2070", "clamp_min_396", "div_264", "sub_6046", "maximum_132", "full_265", "minimum_132", "full_264", "view_2071", "convert_element_type_792", "clamp_max_264", "clamp_min_397", "sub_6052", "round_265", "div_265", "convert_element_type_795", "view_2075", "mul_12854", "view_2074", "sub_6117", "convert_element_type_800", "convert_element_type_799", "clamp_max_267", "clamp_min_401", "add_20467", "round_268", "mul_12928", "mul_12925", "reciprocal_133", "view_2086", "clamp_min_399", "div_266", "sub_6091", "maximum_133", "full_267", "minimum_133", "full_266", "view_2087", "convert_element_type_798", "clamp_max_266", "clamp_min_400", "sub_6097", "round_267", "div_267", "convert_element_type_801", "view_2091", "mul_12950", "view_2090", "sub_6161", "convert_element_type_806", "convert_element_type_805", "clamp_max_269", "clamp_min_404", "add_20615", "round_270", "mul_13027", "mul_13024", "reciprocal_134", "view_2102", "clamp_min_402", "div_268", "sub_6135", "maximum_134", "full_269", "minimum_134", "full_268", "view_2103", "convert_element_type_804", "clamp_max_268", "clamp_min_403", "sub_6141", "round_269", "div_269", "convert_element_type_807", "view_2107", "mul_13049", "view_2106"], "triton_poi_fused__to_copy_mul_sub_view_4:291": ["mul_12859", "sub_6076", "convert_element_type_796", "view_2077", "convert_element_type_797", "view_2079", "view_2078"], "triton_poi_fused__to_copy_mul_sub_view_4:292": ["mul_12955", "sub_6121", "convert_element_type_802", "view_2093", "convert_element_type_803", "view_2095", "view_2094"], "triton_poi_fused__to_copy_mul_sub_view_4:293": ["mul_13054", "sub_6165", "convert_element_type_808", "view_2109", "convert_element_type_809", "view_2111", "view_2110"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:294": ["_scaled_dot_product_efficient_attention_22", "clone_178", "permute_222", "view_2083", "mul_12871", "view_2082", "add_tensor_49", "clone_179", "permute_224", "view_2099", "view_2098", "clone_180", "permute_226", "view_2115", "view_2114", "add_tensor_48"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:295": ["_scaled_dot_product_efficient_attention_22", "clone_178", "permute_222", "view_2083", "mul_12871", "view_2082", "add_tensor_49", "clone_179", "permute_224", "view_2099", "view_2098", "clone_180", "permute_226", "view_2115", "view_2114", "add_tensor_48"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:296": ["_scaled_dot_product_efficient_attention_22", "clone_178", "permute_222", "view_2083", "mul_12871", "view_2082", "add_tensor_49", "clone_179", "permute_224", "view_2099", "view_2098", "clone_180", "permute_226", "view_2115", "view_2114", "add_tensor_48"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297": ["amax_135", "view_2116", "permute_227", "amin_135", "sub_6209", "convert_element_type_812", "convert_element_type_811", "clamp_max_271", "clamp_min_407", "add_20779", "round_272", "mul_13132", "mul_13129", "reciprocal_135", "view_2119", "clamp_min_405", "div_270", "sub_6183", "maximum_135", "full_271", "minimum_135", "full_270", "view_2120", "convert_element_type_810", "clamp_max_270", "clamp_min_406", "sub_6189", "round_271", "div_271", "convert_element_type_813", "view_2124", "mul_13154", "view_2123"], "triton_poi_fused__to_copy_mul_sub_view_4:298": ["mul_13159", "sub_6213", "convert_element_type_814", "view_2126", "convert_element_type_815", "view_2128", "view_2127"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299": ["add_20842", "add_20222", "view_2067", "add_tensor_50", "view_2131", "add_tensor_47", "var_mean_45", "clone_182", "add_20848", "mul_13176", "mul_13175", "sub_6219", "rsqrt_45", "add_20847", "amax_136", "amin_136", "sub_6256", "convert_element_type_818", "convert_element_type_817", "clamp_max_273", "clamp_min_410", "add_20935", "round_274", "mul_13227", "mul_13224", "reciprocal_136", "view_2134", "clamp_min_408", "div_272", "sub_6230", "maximum_136", "full_273", "minimum_136", "full_272", "view_2135", "convert_element_type_816", "clamp_max_272", "clamp_min_409", "sub_6236", "round_273", "div_273", "convert_element_type_819", "view_2139", "mul_13249", "view_2138"], "triton_poi_fused__to_copy_mul_sub_view_10:300": ["mul_13254", "sub_6260", "convert_element_type_820", "view_2141", "convert_element_type_821", "view_2143", "view_2142"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301": ["amax_137", "mul_13268", "mul_13266", "view_2146", "add_tensor_46", "add_20994", "erf_24", "mul_13267", "amin_137", "add_21077", "round_276", "mul_13317", "mul_13314", "reciprocal_137", "view_2149", "clamp_min_411", "div_274", "sub_6273", "maximum_137", "full_275", "minimum_137", "full_274", "view_2150", "convert_element_type_822", "clamp_max_274", "clamp_min_412", "sub_6279", "round_275", "div_275", "mul_13339", "sub_6299", "convert_element_type_824", "convert_element_type_823", "clamp_max_275", "clamp_min_413", "convert_element_type_825", "view_2154", "view_2153"], "triton_poi_fused__to_copy_mul_sub_view_10:302": ["mul_13344", "sub_6303", "convert_element_type_826", "view_2156", "convert_element_type_827", "view_2158", "view_2157"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303": ["var_mean_46", "clone_185", "add_21140", "view_2161", "add_tensor_45", "add_21146", "mul_13361", "mul_13360", "sub_6309", "rsqrt_46", "add_21145", "amax_138", "amin_138", "amax_139", "amin_139", "amax_140", "amin_140", "sub_6346", "convert_element_type_830", "convert_element_type_829", "clamp_max_277", "clamp_min_416", "add_21233", "round_278", "mul_13412", "mul_13409", "reciprocal_138", "view_2164", "clamp_min_414", "div_276", "sub_6320", "maximum_138", "full_277", "minimum_138", "full_276", "view_2165", "convert_element_type_828", "clamp_max_276", "clamp_min_415", "sub_6326", "round_277", "div_277", "convert_element_type_831", "view_2169", "mul_13434", "view_2168", "sub_6391", "convert_element_type_836", "convert_element_type_835", "clamp_max_279", "clamp_min_419", "add_21385", "round_280", "mul_13508", "mul_13505", "reciprocal_139", "view_2180", "clamp_min_417", "div_278", "sub_6365", "maximum_139", "full_279", "minimum_139", "full_278", "view_2181", "convert_element_type_834", "clamp_max_278", "clamp_min_418", "sub_6371", "round_279", "div_279", "convert_element_type_837", "view_2185", "mul_13530", "view_2184", "sub_6435", "convert_element_type_842", "convert_element_type_841", "clamp_max_281", "clamp_min_422", "add_21533", "round_282", "mul_13607", "mul_13604", "reciprocal_140", "view_2196", "clamp_min_420", "div_280", "sub_6409", "maximum_140", "full_281", "minimum_140", "full_280", "view_2197", "convert_element_type_840", "clamp_max_280", "clamp_min_421", "sub_6415", "round_281", "div_281", "convert_element_type_843", "view_2201", "mul_13629", "view_2200"], "triton_poi_fused__to_copy_mul_sub_view_4:304": ["mul_13439", "sub_6350", "convert_element_type_832", "view_2171", "convert_element_type_833", "view_2173", "view_2172"], "triton_poi_fused__to_copy_mul_sub_view_4:305": ["mul_13535", "sub_6395", "convert_element_type_838", "view_2187", "convert_element_type_839", "view_2189", "view_2188"], "triton_poi_fused__to_copy_mul_sub_view_4:306": ["mul_13634", "sub_6439", "convert_element_type_844", "view_2203", "convert_element_type_845", "view_2205", "view_2204"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:307": ["_scaled_dot_product_efficient_attention_23", "clone_186", "permute_232", "view_2177", "mul_13451", "view_2176", "add_tensor_44", "clone_187", "permute_234", "view_2193", "view_2192", "clone_188", "permute_236", "view_2209", "view_2208", "add_tensor_43"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:308": ["_scaled_dot_product_efficient_attention_23", "clone_186", "permute_232", "view_2177", "mul_13451", "view_2176", "add_tensor_44", "clone_187", "permute_234", "view_2193", "view_2192", "clone_188", "permute_236", "view_2209", "view_2208", "add_tensor_43"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:309": ["_scaled_dot_product_efficient_attention_23", "clone_186", "permute_232", "view_2177", "mul_13451", "view_2176", "add_tensor_44", "clone_187", "permute_234", "view_2193", "view_2192", "clone_188", "permute_236", "view_2209", "view_2208", "add_tensor_43"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310": ["amax_141", "view_2210", "permute_237", "amin_141", "sub_6483", "convert_element_type_848", "convert_element_type_847", "clamp_max_283", "clamp_min_425", "add_21697", "round_284", "mul_13712", "mul_13709", "reciprocal_141", "view_2213", "clamp_min_423", "div_282", "sub_6457", "maximum_141", "full_283", "minimum_141", "full_282", "view_2214", "convert_element_type_846", "clamp_max_282", "clamp_min_424", "sub_6463", "round_283", "div_283", "convert_element_type_849", "view_2218", "mul_13734", "view_2217"], "triton_poi_fused__to_copy_mul_sub_view_4:311": ["mul_13739", "sub_6487", "convert_element_type_850", "view_2220", "convert_element_type_851", "view_2222", "view_2221"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312": ["add_21760", "add_21140", "view_2161", "add_tensor_45", "view_2225", "add_tensor_42", "var_mean_47", "clone_190", "add_21766", "mul_13756", "mul_13755", "sub_6493", "rsqrt_47", "add_21765", "amax_142", "amin_142", "sub_6530", "convert_element_type_854", "convert_element_type_853", "clamp_max_285", "clamp_min_428", "add_21853", "round_286", "mul_13807", "mul_13804", "reciprocal_142", "view_2228", "clamp_min_426", "div_284", "sub_6504", "maximum_142", "full_285", "minimum_142", "full_284", "view_2229", "convert_element_type_852", "clamp_max_284", "clamp_min_427", "sub_6510", "round_285", "div_285", "convert_element_type_855", "view_2233", "mul_13829", "view_2232"], "triton_poi_fused__to_copy_mul_sub_view_10:313": ["mul_13834", "sub_6534", "convert_element_type_856", "view_2235", "convert_element_type_857", "view_2237", "view_2236"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314": ["amax_143", "mul_13848", "mul_13846", "view_2240", "add_tensor_41", "add_21912", "erf_25", "mul_13847", "amin_143", "add_21995", "round_288", "mul_13897", "mul_13894", "reciprocal_143", "view_2243", "clamp_min_429", "div_286", "sub_6547", "maximum_143", "full_287", "minimum_143", "full_286", "view_2244", "convert_element_type_858", "clamp_max_286", "clamp_min_430", "sub_6553", "round_287", "div_287", "mul_13919", "sub_6573", "convert_element_type_860", "convert_element_type_859", "clamp_max_287", "clamp_min_431", "convert_element_type_861", "view_2248", "view_2247"], "triton_poi_fused__to_copy_mul_sub_view_10:315": ["mul_13924", "sub_6577", "convert_element_type_862", "view_2250", "convert_element_type_863", "view_2252", "view_2251"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316": ["var_mean_48", "clone_193", "add_22058", "view_2255", "add_tensor_40", "add_22064", "mul_13941", "mul_13940", "sub_6583", "rsqrt_48", "add_22063", "amax_144", "amin_144", "amax_145", "amin_145", "amax_146", "amin_146", "sub_6620", "convert_element_type_866", "convert_element_type_865", "clamp_max_289", "clamp_min_434", "add_22151", "round_290", "mul_13992", "mul_13989", "reciprocal_144", "view_2258", "clamp_min_432", "div_288", "sub_6594", "maximum_144", "full_289", "minimum_144", "full_288", "view_2259", "convert_element_type_864", "clamp_max_288", "clamp_min_433", "sub_6600", "round_289", "div_289", "convert_element_type_867", "view_2263", "mul_14014", "view_2262", "sub_6665", "convert_element_type_872", "convert_element_type_871", "clamp_max_291", "clamp_min_437", "add_22303", "round_292", "mul_14088", "mul_14085", "reciprocal_145", "view_2274", "clamp_min_435", "div_290", "sub_6639", "maximum_145", "full_291", "minimum_145", "full_290", "view_2275", "convert_element_type_870", "clamp_max_290", "clamp_min_436", "sub_6645", "round_291", "div_291", "convert_element_type_873", "view_2279", "mul_14110", "view_2278", "sub_6709", "convert_element_type_878", "convert_element_type_877", "clamp_max_293", "clamp_min_440", "add_22451", "round_294", "mul_14187", "mul_14184", "reciprocal_146", "view_2290", "clamp_min_438", "div_292", "sub_6683", "maximum_146", "full_293", "minimum_146", "full_292", "view_2291", "convert_element_type_876", "clamp_max_292", "clamp_min_439", "sub_6689", "round_293", "div_293", "convert_element_type_879", "view_2295", "mul_14209", "view_2294"], "triton_poi_fused__to_copy_mul_sub_view_4:317": ["mul_14019", "sub_6624", "convert_element_type_868", "view_2265", "convert_element_type_869", "view_2267", "view_2266"], "triton_poi_fused__to_copy_mul_sub_view_4:318": ["mul_14115", "sub_6669", "convert_element_type_874", "view_2281", "convert_element_type_875", "view_2283", "view_2282"], "triton_poi_fused__to_copy_mul_sub_view_4:319": ["mul_14214", "sub_6713", "convert_element_type_880", "view_2297", "convert_element_type_881", "view_2299", "view_2298"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:320": ["_scaled_dot_product_efficient_attention_24", "clone_194", "permute_242", "view_2271", "mul_14031", "view_2270", "add_tensor_39", "clone_195", "permute_244", "view_2287", "view_2286", "clone_196", "permute_246", "view_2303", "view_2302", "add_tensor_38"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:321": ["_scaled_dot_product_efficient_attention_24", "clone_194", "permute_242", "view_2271", "mul_14031", "view_2270", "add_tensor_39", "clone_195", "permute_244", "view_2287", "view_2286", "clone_196", "permute_246", "view_2303", "view_2302", "add_tensor_38"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:322": ["_scaled_dot_product_efficient_attention_24", "clone_194", "permute_242", "view_2271", "mul_14031", "view_2270", "add_tensor_39", "clone_195", "permute_244", "view_2287", "view_2286", "clone_196", "permute_246", "view_2303", "view_2302", "add_tensor_38"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323": ["amax_147", "view_2304", "permute_247", "amin_147", "sub_6757", "convert_element_type_884", "convert_element_type_883", "clamp_max_295", "clamp_min_443", "add_22615", "round_296", "mul_14292", "mul_14289", "reciprocal_147", "view_2307", "clamp_min_441", "div_294", "sub_6731", "maximum_147", "full_295", "minimum_147", "full_294", "view_2308", "convert_element_type_882", "clamp_max_294", "clamp_min_442", "sub_6737", "round_295", "div_295", "convert_element_type_885", "view_2312", "mul_14314", "view_2311"], "triton_poi_fused__to_copy_mul_sub_view_4:324": ["mul_14319", "sub_6761", "convert_element_type_886", "view_2314", "convert_element_type_887", "view_2316", "view_2315"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325": ["add_22678", "add_22058", "view_2255", "add_tensor_40", "view_2319", "add_tensor_37", "var_mean_49", "clone_198", "add_22684", "mul_14336", "mul_14335", "sub_6767", "rsqrt_49", "add_22683", "amax_148", "amin_148", "sub_6804", "convert_element_type_890", "convert_element_type_889", "clamp_max_297", "clamp_min_446", "add_22771", "round_298", "mul_14387", "mul_14384", "reciprocal_148", "view_2322", "clamp_min_444", "div_296", "sub_6778", "maximum_148", "full_297", "minimum_148", "full_296", "view_2323", "convert_element_type_888", "clamp_max_296", "clamp_min_445", "sub_6784", "round_297", "div_297", "convert_element_type_891", "view_2327", "mul_14409", "view_2326"], "triton_poi_fused__to_copy_mul_sub_view_10:326": ["mul_14414", "sub_6808", "convert_element_type_892", "view_2329", "convert_element_type_893", "view_2331", "view_2330"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327": ["amax_149", "mul_14428", "mul_14426", "view_2334", "add_tensor_36", "add_22830", "erf_26", "mul_14427", "amin_149", "add_22913", "round_300", "mul_14477", "mul_14474", "reciprocal_149", "view_2337", "clamp_min_447", "div_298", "sub_6821", "maximum_149", "full_299", "minimum_149", "full_298", "view_2338", "convert_element_type_894", "clamp_max_298", "clamp_min_448", "sub_6827", "round_299", "div_299", "mul_14499", "sub_6847", "convert_element_type_896", "convert_element_type_895", "clamp_max_299", "clamp_min_449", "convert_element_type_897", "view_2342", "view_2341"], "triton_poi_fused__to_copy_mul_sub_view_10:328": ["mul_14504", "sub_6851", "convert_element_type_898", "view_2344", "convert_element_type_899", "view_2346", "view_2345"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329": ["var_mean_50", "clone_201", "add_22976", "view_2349", "add_tensor_35", "add_22982", "mul_14521", "mul_14520", "sub_6857", "rsqrt_50", "add_22981", "amax_150", "amin_150", "amax_151", "amin_151", "amax_152", "amin_152", "sub_6894", "convert_element_type_902", "convert_element_type_901", "clamp_max_301", "clamp_min_452", "add_23069", "round_302", "mul_14572", "mul_14569", "reciprocal_150", "view_2352", "clamp_min_450", "div_300", "sub_6868", "maximum_150", "full_301", "minimum_150", "full_300", "view_2353", "convert_element_type_900", "clamp_max_300", "clamp_min_451", "sub_6874", "round_301", "div_301", "convert_element_type_903", "view_2357", "mul_14594", "view_2356", "sub_6939", "convert_element_type_908", "convert_element_type_907", "clamp_max_303", "clamp_min_455", "add_23221", "round_304", "mul_14668", "mul_14665", "reciprocal_151", "view_2368", "clamp_min_453", "div_302", "sub_6913", "maximum_151", "full_303", "minimum_151", "full_302", "view_2369", "convert_element_type_906", "clamp_max_302", "clamp_min_454", "sub_6919", "round_303", "div_303", "convert_element_type_909", "view_2373", "mul_14690", "view_2372", "sub_6983", "convert_element_type_914", "convert_element_type_913", "clamp_max_305", "clamp_min_458", "add_23369", "round_306", "mul_14767", "mul_14764", "reciprocal_152", "view_2384", "clamp_min_456", "div_304", "sub_6957", "maximum_152", "full_305", "minimum_152", "full_304", "view_2385", "convert_element_type_912", "clamp_max_304", "clamp_min_457", "sub_6963", "round_305", "div_305", "convert_element_type_915", "view_2389", "mul_14789", "view_2388"], "triton_poi_fused__to_copy_mul_sub_view_4:330": ["mul_14599", "sub_6898", "convert_element_type_904", "view_2359", "convert_element_type_905", "view_2361", "view_2360"], "triton_poi_fused__to_copy_mul_sub_view_4:331": ["mul_14695", "sub_6943", "convert_element_type_910", "view_2375", "convert_element_type_911", "view_2377", "view_2376"], "triton_poi_fused__to_copy_mul_sub_view_4:332": ["mul_14794", "sub_6987", "convert_element_type_916", "view_2391", "convert_element_type_917", "view_2393", "view_2392"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:333": ["_scaled_dot_product_efficient_attention_25", "clone_202", "permute_252", "view_2365", "mul_14611", "view_2364", "add_tensor_34", "clone_203", "permute_254", "view_2381", "view_2380", "clone_204", "permute_256", "view_2397", "view_2396", "add_tensor_33"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:334": ["_scaled_dot_product_efficient_attention_25", "clone_202", "permute_252", "view_2365", "mul_14611", "view_2364", "add_tensor_34", "clone_203", "permute_254", "view_2381", "view_2380", "clone_204", "permute_256", "view_2397", "view_2396", "add_tensor_33"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:335": ["_scaled_dot_product_efficient_attention_25", "clone_202", "permute_252", "view_2365", "mul_14611", "view_2364", "add_tensor_34", "clone_203", "permute_254", "view_2381", "view_2380", "clone_204", "permute_256", "view_2397", "view_2396", "add_tensor_33"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336": ["amax_153", "view_2398", "permute_257", "amin_153", "sub_7031", "convert_element_type_920", "convert_element_type_919", "clamp_max_307", "clamp_min_461", "add_23533", "round_308", "mul_14872", "mul_14869", "reciprocal_153", "view_2401", "clamp_min_459", "div_306", "sub_7005", "maximum_153", "full_307", "minimum_153", "full_306", "view_2402", "convert_element_type_918", "clamp_max_306", "clamp_min_460", "sub_7011", "round_307", "div_307", "convert_element_type_921", "view_2406", "mul_14894", "view_2405"], "triton_poi_fused__to_copy_mul_sub_view_4:337": ["mul_14899", "sub_7035", "convert_element_type_922", "view_2408", "convert_element_type_923", "view_2410", "view_2409"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338": ["add_23596", "add_22976", "view_2349", "add_tensor_35", "view_2413", "add_tensor_32", "var_mean_51", "clone_206", "add_23602", "mul_14916", "mul_14915", "sub_7041", "rsqrt_51", "add_23601", "amax_154", "amin_154", "sub_7078", "convert_element_type_926", "convert_element_type_925", "clamp_max_309", "clamp_min_464", "add_23689", "round_310", "mul_14967", "mul_14964", "reciprocal_154", "view_2416", "clamp_min_462", "div_308", "sub_7052", "maximum_154", "full_309", "minimum_154", "full_308", "view_2417", "convert_element_type_924", "clamp_max_308", "clamp_min_463", "sub_7058", "round_309", "div_309", "convert_element_type_927", "view_2421", "mul_14989", "view_2420"], "triton_poi_fused__to_copy_mul_sub_view_10:339": ["mul_14994", "sub_7082", "convert_element_type_928", "view_2423", "convert_element_type_929", "view_2425", "view_2424"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340": ["amax_155", "mul_15008", "mul_15006", "view_2428", "add_tensor_31", "add_23748", "erf_27", "mul_15007", "amin_155", "add_23831", "round_312", "mul_15057", "mul_15054", "reciprocal_155", "view_2431", "clamp_min_465", "div_310", "sub_7095", "maximum_155", "full_311", "minimum_155", "full_310", "view_2432", "convert_element_type_930", "clamp_max_310", "clamp_min_466", "sub_7101", "round_311", "div_311", "mul_15079", "sub_7121", "convert_element_type_932", "convert_element_type_931", "clamp_max_311", "clamp_min_467", "convert_element_type_933", "view_2436", "view_2435"], "triton_poi_fused__to_copy_mul_sub_view_10:341": ["mul_15084", "sub_7125", "convert_element_type_934", "view_2438", "convert_element_type_935", "view_2440", "view_2439"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342": ["var_mean_52", "clone_209", "add_23894", "view_2443", "add_tensor_30", "add_23900", "mul_15101", "mul_15100", "sub_7131", "rsqrt_52", "add_23899", "amax_156", "amin_156", "amax_157", "amin_157", "amax_158", "amin_158", "sub_7168", "convert_element_type_938", "convert_element_type_937", "clamp_max_313", "clamp_min_470", "add_23987", "round_314", "mul_15152", "mul_15149", "reciprocal_156", "view_2446", "clamp_min_468", "div_312", "sub_7142", "maximum_156", "full_313", "minimum_156", "full_312", "view_2447", "convert_element_type_936", "clamp_max_312", "clamp_min_469", "sub_7148", "round_313", "div_313", "convert_element_type_939", "view_2451", "mul_15174", "view_2450", "sub_7213", "convert_element_type_944", "convert_element_type_943", "clamp_max_315", "clamp_min_473", "add_24139", "round_316", "mul_15248", "mul_15245", "reciprocal_157", "view_2462", "clamp_min_471", "div_314", "sub_7187", "maximum_157", "full_315", "minimum_157", "full_314", "view_2463", "convert_element_type_942", "clamp_max_314", "clamp_min_472", "sub_7193", "round_315", "div_315", "convert_element_type_945", "view_2467", "mul_15270", "view_2466", "sub_7257", "convert_element_type_950", "convert_element_type_949", "clamp_max_317", "clamp_min_476", "add_24287", "round_318", "mul_15347", "mul_15344", "reciprocal_158", "view_2478", "clamp_min_474", "div_316", "sub_7231", "maximum_158", "full_317", "minimum_158", "full_316", "view_2479", "convert_element_type_948", "clamp_max_316", "clamp_min_475", "sub_7237", "round_317", "div_317", "convert_element_type_951", "view_2483", "mul_15369", "view_2482"], "triton_poi_fused__to_copy_mul_sub_view_4:343": ["mul_15179", "sub_7172", "convert_element_type_940", "view_2453", "convert_element_type_941", "view_2455", "view_2454"], "triton_poi_fused__to_copy_mul_sub_view_4:344": ["mul_15275", "sub_7217", "convert_element_type_946", "view_2469", "convert_element_type_947", "view_2471", "view_2470"], "triton_poi_fused__to_copy_mul_sub_view_4:345": ["mul_15374", "sub_7261", "convert_element_type_952", "view_2485", "convert_element_type_953", "view_2487", "view_2486"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:346": ["_scaled_dot_product_efficient_attention_26", "clone_210", "permute_262", "view_2459", "mul_15191", "view_2458", "add_tensor_29", "clone_211", "permute_264", "view_2475", "view_2474", "clone_212", "permute_266", "view_2491", "view_2490", "add_tensor_28"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:347": ["_scaled_dot_product_efficient_attention_26", "clone_210", "permute_262", "view_2459", "mul_15191", "view_2458", "add_tensor_29", "clone_211", "permute_264", "view_2475", "view_2474", "clone_212", "permute_266", "view_2491", "view_2490", "add_tensor_28"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:348": ["_scaled_dot_product_efficient_attention_26", "clone_210", "permute_262", "view_2459", "mul_15191", "view_2458", "add_tensor_29", "clone_211", "permute_264", "view_2475", "view_2474", "clone_212", "permute_266", "view_2491", "view_2490", "add_tensor_28"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349": ["amax_159", "view_2492", "permute_267", "amin_159", "sub_7305", "convert_element_type_956", "convert_element_type_955", "clamp_max_319", "clamp_min_479", "add_24451", "round_320", "mul_15452", "mul_15449", "reciprocal_159", "view_2495", "clamp_min_477", "div_318", "sub_7279", "maximum_159", "full_319", "minimum_159", "full_318", "view_2496", "convert_element_type_954", "clamp_max_318", "clamp_min_478", "sub_7285", "round_319", "div_319", "convert_element_type_957", "view_2500", "mul_15474", "view_2499"], "triton_poi_fused__to_copy_mul_sub_view_4:350": ["mul_15479", "sub_7309", "convert_element_type_958", "view_2502", "convert_element_type_959", "view_2504", "view_2503"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351": ["add_24514", "add_23894", "view_2443", "add_tensor_30", "view_2507", "add_tensor_27", "var_mean_53", "clone_214", "add_24520", "mul_15496", "mul_15495", "sub_7315", "rsqrt_53", "add_24519", "amax_160", "amin_160", "sub_7352", "convert_element_type_962", "convert_element_type_961", "clamp_max_321", "clamp_min_482", "add_24607", "round_322", "mul_15547", "mul_15544", "reciprocal_160", "view_2510", "clamp_min_480", "div_320", "sub_7326", "maximum_160", "full_321", "minimum_160", "full_320", "view_2511", "convert_element_type_960", "clamp_max_320", "clamp_min_481", "sub_7332", "round_321", "div_321", "convert_element_type_963", "view_2515", "mul_15569", "view_2514"], "triton_poi_fused__to_copy_mul_sub_view_10:352": ["mul_15574", "sub_7356", "convert_element_type_964", "view_2517", "convert_element_type_965", "view_2519", "view_2518"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353": ["amax_161", "mul_15588", "mul_15586", "view_2522", "add_tensor_26", "add_24666", "erf_28", "mul_15587", "amin_161", "add_24749", "round_324", "mul_15637", "mul_15634", "reciprocal_161", "view_2525", "clamp_min_483", "div_322", "sub_7369", "maximum_161", "full_323", "minimum_161", "full_322", "view_2526", "convert_element_type_966", "clamp_max_322", "clamp_min_484", "sub_7375", "round_323", "div_323", "mul_15659", "sub_7395", "convert_element_type_968", "convert_element_type_967", "clamp_max_323", "clamp_min_485", "convert_element_type_969", "view_2530", "view_2529"], "triton_poi_fused__to_copy_mul_sub_view_10:354": ["mul_15664", "sub_7399", "convert_element_type_970", "view_2532", "convert_element_type_971", "view_2534", "view_2533"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355": ["var_mean_54", "clone_217", "add_24812", "view_2537", "add_tensor_25", "add_24818", "mul_15681", "mul_15680", "sub_7405", "rsqrt_54", "add_24817", "amax_162", "amin_162", "amax_163", "amin_163", "amax_164", "amin_164", "sub_7442", "convert_element_type_974", "convert_element_type_973", "clamp_max_325", "clamp_min_488", "add_24905", "round_326", "mul_15732", "mul_15729", "reciprocal_162", "view_2540", "clamp_min_486", "div_324", "sub_7416", "maximum_162", "full_325", "minimum_162", "full_324", "view_2541", "convert_element_type_972", "clamp_max_324", "clamp_min_487", "sub_7422", "round_325", "div_325", "convert_element_type_975", "view_2545", "mul_15754", "view_2544", "sub_7487", "convert_element_type_980", "convert_element_type_979", "clamp_max_327", "clamp_min_491", "add_25057", "round_328", "mul_15828", "mul_15825", "reciprocal_163", "view_2556", "clamp_min_489", "div_326", "sub_7461", "maximum_163", "full_327", "minimum_163", "full_326", "view_2557", "convert_element_type_978", "clamp_max_326", "clamp_min_490", "sub_7467", "round_327", "div_327", "convert_element_type_981", "view_2561", "mul_15850", "view_2560", "sub_7531", "convert_element_type_986", "convert_element_type_985", "clamp_max_329", "clamp_min_494", "add_25205", "round_330", "mul_15927", "mul_15924", "reciprocal_164", "view_2572", "clamp_min_492", "div_328", "sub_7505", "maximum_164", "full_329", "minimum_164", "full_328", "view_2573", "convert_element_type_984", "clamp_max_328", "clamp_min_493", "sub_7511", "round_329", "div_329", "convert_element_type_987", "view_2577", "mul_15949", "view_2576"], "triton_poi_fused__to_copy_mul_sub_view_4:356": ["mul_15759", "sub_7446", "convert_element_type_976", "view_2547", "convert_element_type_977", "view_2549", "view_2548"], "triton_poi_fused__to_copy_mul_sub_view_4:357": ["mul_15855", "sub_7491", "convert_element_type_982", "view_2563", "convert_element_type_983", "view_2565", "view_2564"], "triton_poi_fused__to_copy_mul_sub_view_4:358": ["mul_15954", "sub_7535", "convert_element_type_988", "view_2579", "convert_element_type_989", "view_2581", "view_2580"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:359": ["_scaled_dot_product_efficient_attention_27", "clone_218", "permute_272", "view_2553", "mul_15771", "view_2552", "add_tensor_24", "clone_219", "permute_274", "view_2569", "view_2568", "clone_220", "permute_276", "view_2585", "view_2584", "add_tensor_23"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:360": ["_scaled_dot_product_efficient_attention_27", "clone_218", "permute_272", "view_2553", "mul_15771", "view_2552", "add_tensor_24", "clone_219", "permute_274", "view_2569", "view_2568", "clone_220", "permute_276", "view_2585", "view_2584", "add_tensor_23"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:361": ["_scaled_dot_product_efficient_attention_27", "clone_218", "permute_272", "view_2553", "mul_15771", "view_2552", "add_tensor_24", "clone_219", "permute_274", "view_2569", "view_2568", "clone_220", "permute_276", "view_2585", "view_2584", "add_tensor_23"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362": ["amax_165", "view_2586", "permute_277", "amin_165", "sub_7579", "convert_element_type_992", "convert_element_type_991", "clamp_max_331", "clamp_min_497", "add_25369", "round_332", "mul_16032", "mul_16029", "reciprocal_165", "view_2589", "clamp_min_495", "div_330", "sub_7553", "maximum_165", "full_331", "minimum_165", "full_330", "view_2590", "convert_element_type_990", "clamp_max_330", "clamp_min_496", "sub_7559", "round_331", "div_331", "convert_element_type_993", "view_2594", "mul_16054", "view_2593"], "triton_poi_fused__to_copy_mul_sub_view_4:363": ["mul_16059", "sub_7583", "convert_element_type_994", "view_2596", "convert_element_type_995", "view_2598", "view_2597"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364": ["add_25432", "add_24812", "view_2537", "add_tensor_25", "view_2601", "add_tensor_22", "var_mean_55", "clone_222", "add_25438", "mul_16076", "mul_16075", "sub_7589", "rsqrt_55", "add_25437", "amax_166", "amin_166", "sub_7626", "convert_element_type_998", "convert_element_type_997", "clamp_max_333", "clamp_min_500", "add_25525", "round_334", "mul_16127", "mul_16124", "reciprocal_166", "view_2604", "clamp_min_498", "div_332", "sub_7600", "maximum_166", "full_333", "minimum_166", "full_332", "view_2605", "convert_element_type_996", "clamp_max_332", "clamp_min_499", "sub_7606", "round_333", "div_333", "convert_element_type_999", "view_2609", "mul_16149", "view_2608"], "triton_poi_fused__to_copy_mul_sub_view_10:365": ["mul_16154", "sub_7630", "convert_element_type_1000", "view_2611", "convert_element_type_1001", "view_2613", "view_2612"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366": ["amax_167", "mul_16168", "mul_16166", "view_2616", "add_tensor_21", "add_25584", "erf_29", "mul_16167", "amin_167", "add_25667", "round_336", "mul_16217", "mul_16214", "reciprocal_167", "view_2619", "clamp_min_501", "div_334", "sub_7643", "maximum_167", "full_335", "minimum_167", "full_334", "view_2620", "convert_element_type_1002", "clamp_max_334", "clamp_min_502", "sub_7649", "round_335", "div_335", "mul_16239", "sub_7669", "convert_element_type_1004", "convert_element_type_1003", "clamp_max_335", "clamp_min_503", "convert_element_type_1005", "view_2624", "view_2623"], "triton_poi_fused__to_copy_mul_sub_view_10:367": ["mul_16244", "sub_7673", "convert_element_type_1006", "view_2626", "convert_element_type_1007", "view_2628", "view_2627"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368": ["var_mean_56", "clone_225", "add_25730", "view_2631", "add_tensor_20", "add_25736", "mul_16261", "mul_16260", "sub_7679", "rsqrt_56", "add_25735", "amax_168", "amin_168", "amax_169", "amin_169", "amax_170", "amin_170", "sub_7716", "convert_element_type_1010", "convert_element_type_1009", "clamp_max_337", "clamp_min_506", "add_25823", "round_338", "mul_16312", "mul_16309", "reciprocal_168", "view_2634", "clamp_min_504", "div_336", "sub_7690", "maximum_168", "full_337", "minimum_168", "full_336", "view_2635", "convert_element_type_1008", "clamp_max_336", "clamp_min_505", "sub_7696", "round_337", "div_337", "convert_element_type_1011", "view_2639", "mul_16334", "view_2638", "sub_7761", "convert_element_type_1016", "convert_element_type_1015", "clamp_max_339", "clamp_min_509", "add_25975", "round_340", "mul_16408", "mul_16405", "reciprocal_169", "view_2650", "clamp_min_507", "div_338", "sub_7735", "maximum_169", "full_339", "minimum_169", "full_338", "view_2651", "convert_element_type_1014", "clamp_max_338", "clamp_min_508", "sub_7741", "round_339", "div_339", "convert_element_type_1017", "view_2655", "mul_16430", "view_2654", "sub_7805", "convert_element_type_1022", "convert_element_type_1021", "clamp_max_341", "clamp_min_512", "add_26123", "round_342", "mul_16507", "mul_16504", "reciprocal_170", "view_2666", "clamp_min_510", "div_340", "sub_7779", "maximum_170", "full_341", "minimum_170", "full_340", "view_2667", "convert_element_type_1020", "clamp_max_340", "clamp_min_511", "sub_7785", "round_341", "div_341", "convert_element_type_1023", "view_2671", "mul_16529", "view_2670"], "triton_poi_fused__to_copy_mul_sub_view_4:369": ["mul_16339", "sub_7720", "convert_element_type_1012", "view_2641", "convert_element_type_1013", "view_2643", "view_2642"], "triton_poi_fused__to_copy_mul_sub_view_4:370": ["mul_16435", "sub_7765", "convert_element_type_1018", "view_2657", "convert_element_type_1019", "view_2659", "view_2658"], "triton_poi_fused__to_copy_mul_sub_view_4:371": ["mul_16534", "sub_7809", "convert_element_type_1024", "view_2673", "convert_element_type_1025", "view_2675", "view_2674"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:372": ["_scaled_dot_product_efficient_attention_28", "clone_226", "permute_282", "view_2647", "mul_16351", "view_2646", "add_tensor_19", "clone_227", "permute_284", "view_2663", "view_2662", "clone_228", "permute_286", "view_2679", "view_2678", "add_tensor_18"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:373": ["_scaled_dot_product_efficient_attention_28", "clone_226", "permute_282", "view_2647", "mul_16351", "view_2646", "add_tensor_19", "clone_227", "permute_284", "view_2663", "view_2662", "clone_228", "permute_286", "view_2679", "view_2678", "add_tensor_18"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:374": ["_scaled_dot_product_efficient_attention_28", "clone_226", "permute_282", "view_2647", "mul_16351", "view_2646", "add_tensor_19", "clone_227", "permute_284", "view_2663", "view_2662", "clone_228", "permute_286", "view_2679", "view_2678", "add_tensor_18"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375": ["amax_171", "view_2680", "permute_287", "amin_171", "sub_7853", "convert_element_type_1028", "convert_element_type_1027", "clamp_max_343", "clamp_min_515", "add_26287", "round_344", "mul_16612", "mul_16609", "reciprocal_171", "view_2683", "clamp_min_513", "div_342", "sub_7827", "maximum_171", "full_343", "minimum_171", "full_342", "view_2684", "convert_element_type_1026", "clamp_max_342", "clamp_min_514", "sub_7833", "round_343", "div_343", "convert_element_type_1029", "view_2688", "mul_16634", "view_2687"], "triton_poi_fused__to_copy_mul_sub_view_4:376": ["mul_16639", "sub_7857", "convert_element_type_1030", "view_2690", "convert_element_type_1031", "view_2692", "view_2691"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377": ["add_26350", "add_25730", "view_2631", "add_tensor_20", "view_2695", "add_tensor_17", "var_mean_57", "clone_230", "add_26356", "mul_16656", "mul_16655", "sub_7863", "rsqrt_57", "add_26355", "amax_172", "amin_172", "sub_7900", "convert_element_type_1034", "convert_element_type_1033", "clamp_max_345", "clamp_min_518", "add_26443", "round_346", "mul_16707", "mul_16704", "reciprocal_172", "view_2698", "clamp_min_516", "div_344", "sub_7874", "maximum_172", "full_345", "minimum_172", "full_344", "view_2699", "convert_element_type_1032", "clamp_max_344", "clamp_min_517", "sub_7880", "round_345", "div_345", "convert_element_type_1035", "view_2703", "mul_16729", "view_2702"], "triton_poi_fused__to_copy_mul_sub_view_10:378": ["mul_16734", "sub_7904", "convert_element_type_1036", "view_2705", "convert_element_type_1037", "view_2707", "view_2706"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379": ["amax_173", "mul_16748", "mul_16746", "view_2710", "add_tensor_16", "add_26502", "erf_30", "mul_16747", "amin_173", "add_26585", "round_348", "mul_16797", "mul_16794", "reciprocal_173", "view_2713", "clamp_min_519", "div_346", "sub_7917", "maximum_173", "full_347", "minimum_173", "full_346", "view_2714", "convert_element_type_1038", "clamp_max_346", "clamp_min_520", "sub_7923", "round_347", "div_347", "mul_16819", "sub_7943", "convert_element_type_1040", "convert_element_type_1039", "clamp_max_347", "clamp_min_521", "convert_element_type_1041", "view_2718", "view_2717"], "triton_poi_fused__to_copy_mul_sub_view_10:380": ["mul_16824", "sub_7947", "convert_element_type_1042", "view_2720", "convert_element_type_1043", "view_2722", "view_2721"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381": ["var_mean_58", "clone_233", "add_26648", "view_2725", "add_tensor_15", "add_26654", "mul_16841", "mul_16840", "sub_7953", "rsqrt_58", "add_26653", "amax_174", "amin_174", "amax_175", "amin_175", "amax_176", "amin_176", "sub_7990", "convert_element_type_1046", "convert_element_type_1045", "clamp_max_349", "clamp_min_524", "add_26741", "round_350", "mul_16892", "mul_16889", "reciprocal_174", "view_2728", "clamp_min_522", "div_348", "sub_7964", "maximum_174", "full_349", "minimum_174", "full_348", "view_2729", "convert_element_type_1044", "clamp_max_348", "clamp_min_523", "sub_7970", "round_349", "div_349", "convert_element_type_1047", "view_2733", "mul_16914", "view_2732", "sub_8035", "convert_element_type_1052", "convert_element_type_1051", "clamp_max_351", "clamp_min_527", "add_26893", "round_352", "mul_16988", "mul_16985", "reciprocal_175", "view_2744", "clamp_min_525", "div_350", "sub_8009", "maximum_175", "full_351", "minimum_175", "full_350", "view_2745", "convert_element_type_1050", "clamp_max_350", "clamp_min_526", "sub_8015", "round_351", "div_351", "convert_element_type_1053", "view_2749", "mul_17010", "view_2748", "sub_8079", "convert_element_type_1058", "convert_element_type_1057", "clamp_max_353", "clamp_min_530", "add_27041", "round_354", "mul_17087", "mul_17084", "reciprocal_176", "view_2760", "clamp_min_528", "div_352", "sub_8053", "maximum_176", "full_353", "minimum_176", "full_352", "view_2761", "convert_element_type_1056", "clamp_max_352", "clamp_min_529", "sub_8059", "round_353", "div_353", "convert_element_type_1059", "view_2765", "mul_17109", "view_2764"], "triton_poi_fused__to_copy_mul_sub_view_4:382": ["mul_16919", "sub_7994", "convert_element_type_1048", "view_2735", "convert_element_type_1049", "view_2737", "view_2736"], "triton_poi_fused__to_copy_mul_sub_view_4:383": ["mul_17015", "sub_8039", "convert_element_type_1054", "view_2751", "convert_element_type_1055", "view_2753", "view_2752"], "triton_poi_fused__to_copy_mul_sub_view_4:384": ["mul_17114", "sub_8083", "convert_element_type_1060", "view_2767", "convert_element_type_1061", "view_2769", "view_2768"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:385": ["_scaled_dot_product_efficient_attention_29", "clone_234", "permute_292", "view_2741", "mul_16931", "view_2740", "add_tensor_14", "clone_235", "permute_294", "view_2757", "view_2756", "clone_236", "permute_296", "view_2773", "view_2772", "add_tensor_13"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:386": ["_scaled_dot_product_efficient_attention_29", "clone_234", "permute_292", "view_2741", "mul_16931", "view_2740", "add_tensor_14", "clone_235", "permute_294", "view_2757", "view_2756", "clone_236", "permute_296", "view_2773", "view_2772", "add_tensor_13"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:387": ["_scaled_dot_product_efficient_attention_29", "clone_234", "permute_292", "view_2741", "mul_16931", "view_2740", "add_tensor_14", "clone_235", "permute_294", "view_2757", "view_2756", "clone_236", "permute_296", "view_2773", "view_2772", "add_tensor_13"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388": ["amax_177", "view_2774", "permute_297", "amin_177", "sub_8127", "convert_element_type_1064", "convert_element_type_1063", "clamp_max_355", "clamp_min_533", "add_27205", "round_356", "mul_17192", "mul_17189", "reciprocal_177", "view_2777", "clamp_min_531", "div_354", "sub_8101", "maximum_177", "full_355", "minimum_177", "full_354", "view_2778", "convert_element_type_1062", "clamp_max_354", "clamp_min_532", "sub_8107", "round_355", "div_355", "convert_element_type_1065", "view_2782", "mul_17214", "view_2781"], "triton_poi_fused__to_copy_mul_sub_view_4:389": ["mul_17219", "sub_8131", "convert_element_type_1066", "view_2784", "convert_element_type_1067", "view_2786", "view_2785"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390": ["add_27268", "add_26648", "view_2725", "add_tensor_15", "view_2789", "add_tensor_12", "var_mean_59", "clone_238", "add_27274", "mul_17236", "mul_17235", "sub_8137", "rsqrt_59", "add_27273", "amax_178", "amin_178", "sub_8174", "convert_element_type_1070", "convert_element_type_1069", "clamp_max_357", "clamp_min_536", "add_27361", "round_358", "mul_17287", "mul_17284", "reciprocal_178", "view_2792", "clamp_min_534", "div_356", "sub_8148", "maximum_178", "full_357", "minimum_178", "full_356", "view_2793", "convert_element_type_1068", "clamp_max_356", "clamp_min_535", "sub_8154", "round_357", "div_357", "convert_element_type_1071", "view_2797", "mul_17309", "view_2796"], "triton_poi_fused__to_copy_mul_sub_view_10:391": ["mul_17314", "sub_8178", "convert_element_type_1072", "view_2799", "convert_element_type_1073", "view_2801", "view_2800"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392": ["amax_179", "mul_17328", "mul_17326", "view_2804", "add_tensor_11", "add_27420", "erf_31", "mul_17327", "amin_179", "add_27503", "round_360", "mul_17377", "mul_17374", "reciprocal_179", "view_2807", "clamp_min_537", "div_358", "sub_8191", "maximum_179", "full_359", "minimum_179", "full_358", "view_2808", "convert_element_type_1074", "clamp_max_358", "clamp_min_538", "sub_8197", "round_359", "div_359", "mul_17399", "sub_8217", "convert_element_type_1076", "convert_element_type_1075", "clamp_max_359", "clamp_min_539", "convert_element_type_1077", "view_2812", "view_2811"], "triton_poi_fused__to_copy_mul_sub_view_10:393": ["mul_17404", "sub_8221", "convert_element_type_1078", "view_2814", "convert_element_type_1079", "view_2816", "view_2815"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394": ["var_mean_60", "clone_241", "add_27566", "view_2819", "add_tensor_10", "add_27572", "mul_17421", "mul_17420", "sub_8227", "rsqrt_60", "add_27571", "amax_180", "amin_180", "amax_181", "amin_181", "amax_182", "amin_182", "sub_8264", "convert_element_type_1082", "convert_element_type_1081", "clamp_max_361", "clamp_min_542", "add_27659", "round_362", "mul_17472", "mul_17469", "reciprocal_180", "view_2822", "clamp_min_540", "div_360", "sub_8238", "maximum_180", "full_361", "minimum_180", "full_360", "view_2823", "convert_element_type_1080", "clamp_max_360", "clamp_min_541", "sub_8244", "round_361", "div_361", "convert_element_type_1083", "view_2827", "mul_17494", "view_2826", "sub_8309", "convert_element_type_1088", "convert_element_type_1087", "clamp_max_363", "clamp_min_545", "add_27811", "round_364", "mul_17568", "mul_17565", "reciprocal_181", "view_2838", "clamp_min_543", "div_362", "sub_8283", "maximum_181", "full_363", "minimum_181", "full_362", "view_2839", "convert_element_type_1086", "clamp_max_362", "clamp_min_544", "sub_8289", "round_363", "div_363", "convert_element_type_1089", "view_2843", "mul_17590", "view_2842", "sub_8353", "convert_element_type_1094", "convert_element_type_1093", "clamp_max_365", "clamp_min_548", "add_27959", "round_366", "mul_17667", "mul_17664", "reciprocal_182", "view_2854", "clamp_min_546", "div_364", "sub_8327", "maximum_182", "full_365", "minimum_182", "full_364", "view_2855", "convert_element_type_1092", "clamp_max_364", "clamp_min_547", "sub_8333", "round_365", "div_365", "convert_element_type_1095", "view_2859", "mul_17689", "view_2858"], "triton_poi_fused__to_copy_mul_sub_view_4:395": ["mul_17499", "sub_8268", "convert_element_type_1084", "view_2829", "convert_element_type_1085", "view_2831", "view_2830"], "triton_poi_fused__to_copy_mul_sub_view_4:396": ["mul_17595", "sub_8313", "convert_element_type_1090", "view_2845", "convert_element_type_1091", "view_2847", "view_2846"], "triton_poi_fused__to_copy_mul_sub_view_4:397": ["mul_17694", "sub_8357", "convert_element_type_1096", "view_2861", "convert_element_type_1097", "view_2863", "view_2862"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:398": ["_scaled_dot_product_efficient_attention_30", "clone_242", "permute_302", "view_2835", "mul_17511", "view_2834", "add_tensor_9", "clone_243", "permute_304", "view_2851", "view_2850", "clone_244", "permute_306", "view_2867", "view_2866", "add_tensor_8"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:399": ["_scaled_dot_product_efficient_attention_30", "clone_242", "permute_302", "view_2835", "mul_17511", "view_2834", "add_tensor_9", "clone_243", "permute_304", "view_2851", "view_2850", "clone_244", "permute_306", "view_2867", "view_2866", "add_tensor_8"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:400": ["_scaled_dot_product_efficient_attention_30", "clone_242", "permute_302", "view_2835", "mul_17511", "view_2834", "add_tensor_9", "clone_243", "permute_304", "view_2851", "view_2850", "clone_244", "permute_306", "view_2867", "view_2866", "add_tensor_8"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401": ["amax_183", "view_2868", "permute_307", "amin_183", "sub_8401", "convert_element_type_1100", "convert_element_type_1099", "clamp_max_367", "clamp_min_551", "add_28123", "round_368", "mul_17772", "mul_17769", "reciprocal_183", "view_2871", "clamp_min_549", "div_366", "sub_8375", "maximum_183", "full_367", "minimum_183", "full_366", "view_2872", "convert_element_type_1098", "clamp_max_366", "clamp_min_550", "sub_8381", "round_367", "div_367", "convert_element_type_1101", "view_2876", "mul_17794", "view_2875"], "triton_poi_fused__to_copy_mul_sub_view_4:402": ["mul_17799", "sub_8405", "convert_element_type_1102", "view_2878", "convert_element_type_1103", "view_2880", "view_2879"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403": ["add_28186", "add_27566", "view_2819", "add_tensor_10", "view_2883", "add_tensor_7", "var_mean_61", "clone_246", "add_28192", "mul_17816", "mul_17815", "sub_8411", "rsqrt_61", "add_28191", "amax_184", "amin_184", "sub_8448", "convert_element_type_1106", "convert_element_type_1105", "clamp_max_369", "clamp_min_554", "add_28279", "round_370", "mul_17867", "mul_17864", "reciprocal_184", "view_2886", "clamp_min_552", "div_368", "sub_8422", "maximum_184", "full_369", "minimum_184", "full_368", "view_2887", "convert_element_type_1104", "clamp_max_368", "clamp_min_553", "sub_8428", "round_369", "div_369", "convert_element_type_1107", "view_2891", "mul_17889", "view_2890"], "triton_poi_fused__to_copy_mul_sub_view_10:404": ["mul_17894", "sub_8452", "convert_element_type_1108", "view_2893", "convert_element_type_1109", "view_2895", "view_2894"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405": ["amax_185", "mul_17908", "mul_17906", "view_2898", "add_tensor_6", "add_28338", "erf_32", "mul_17907", "amin_185", "add_28421", "round_372", "mul_17957", "mul_17954", "reciprocal_185", "view_2901", "clamp_min_555", "div_370", "sub_8465", "maximum_185", "full_371", "minimum_185", "full_370", "view_2902", "convert_element_type_1110", "clamp_max_370", "clamp_min_556", "sub_8471", "round_371", "div_371", "mul_17979", "sub_8491", "convert_element_type_1112", "convert_element_type_1111", "clamp_max_371", "clamp_min_557", "convert_element_type_1113", "view_2906", "view_2905"], "triton_poi_fused__to_copy_mul_sub_view_10:406": ["mul_17984", "sub_8495", "convert_element_type_1114", "view_2908", "convert_element_type_1115", "view_2910", "view_2909"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407": ["var_mean_62", "clone_249", "add_28484", "view_2913", "add_tensor_5", "add_28490", "mul_18001", "mul_18000", "sub_8501", "rsqrt_62", "add_28489", "amax_186", "amin_186", "amax_187", "amin_187", "amax_188", "amin_188", "sub_8538", "convert_element_type_1118", "convert_element_type_1117", "clamp_max_373", "clamp_min_560", "add_28577", "round_374", "mul_18052", "mul_18049", "reciprocal_186", "view_2916", "clamp_min_558", "div_372", "sub_8512", "maximum_186", "full_373", "minimum_186", "full_372", "view_2917", "convert_element_type_1116", "clamp_max_372", "clamp_min_559", "sub_8518", "round_373", "div_373", "convert_element_type_1119", "view_2921", "mul_18074", "view_2920", "sub_8583", "convert_element_type_1124", "convert_element_type_1123", "clamp_max_375", "clamp_min_563", "add_28729", "round_376", "mul_18148", "mul_18145", "reciprocal_187", "view_2932", "clamp_min_561", "div_374", "sub_8557", "maximum_187", "full_375", "minimum_187", "full_374", "view_2933", "convert_element_type_1122", "clamp_max_374", "clamp_min_562", "sub_8563", "round_375", "div_375", "convert_element_type_1125", "view_2937", "mul_18170", "view_2936", "sub_8627", "convert_element_type_1130", "convert_element_type_1129", "clamp_max_377", "clamp_min_566", "add_28877", "round_378", "mul_18247", "mul_18244", "reciprocal_188", "view_2948", "clamp_min_564", "div_376", "sub_8601", "maximum_188", "full_377", "minimum_188", "full_376", "view_2949", "convert_element_type_1128", "clamp_max_376", "clamp_min_565", "sub_8607", "round_377", "div_377", "convert_element_type_1131", "view_2953", "mul_18269", "view_2952"], "triton_poi_fused__to_copy_mul_sub_view_4:408": ["mul_18079", "sub_8542", "convert_element_type_1120", "view_2923", "convert_element_type_1121", "view_2925", "view_2924"], "triton_poi_fused__to_copy_mul_sub_view_4:409": ["mul_18175", "sub_8587", "convert_element_type_1126", "view_2939", "convert_element_type_1127", "view_2941", "view_2940"], "triton_poi_fused__to_copy_mul_sub_view_4:410": ["mul_18274", "sub_8631", "convert_element_type_1132", "view_2955", "convert_element_type_1133", "view_2957", "view_2956"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:411": ["_scaled_dot_product_efficient_attention_31", "clone_250", "permute_312", "view_2929", "mul_18091", "view_2928", "add_tensor_4", "clone_251", "permute_314", "view_2945", "view_2944", "clone_252", "permute_316", "view_2961", "view_2960", "add_tensor_3"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:412": ["_scaled_dot_product_efficient_attention_31", "clone_250", "permute_312", "view_2929", "mul_18091", "view_2928", "add_tensor_4", "clone_251", "permute_314", "view_2945", "view_2944", "clone_252", "permute_316", "view_2961", "view_2960", "add_tensor_3"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:413": ["_scaled_dot_product_efficient_attention_31", "clone_250", "permute_312", "view_2929", "mul_18091", "view_2928", "add_tensor_4", "clone_251", "permute_314", "view_2945", "view_2944", "clone_252", "permute_316", "view_2961", "view_2960", "add_tensor_3"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414": ["amax_189", "view_2962", "permute_317", "amin_189", "sub_8675", "convert_element_type_1136", "convert_element_type_1135", "clamp_max_379", "clamp_min_569", "add_29041", "round_380", "mul_18352", "mul_18349", "reciprocal_189", "view_2965", "clamp_min_567", "div_378", "sub_8649", "maximum_189", "full_379", "minimum_189", "full_378", "view_2966", "convert_element_type_1134", "clamp_max_378", "clamp_min_568", "sub_8655", "round_379", "div_379", "convert_element_type_1137", "view_2970", "mul_18374", "view_2969"], "triton_poi_fused__to_copy_mul_sub_view_4:415": ["mul_18379", "sub_8679", "convert_element_type_1138", "view_2972", "convert_element_type_1139", "view_2974", "view_2973"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416": ["add_29104", "add_28484", "view_2913", "add_tensor_5", "view_2977", "add_tensor_2", "var_mean_63", "clone_254", "add_29110", "mul_18396", "mul_18395", "sub_8685", "rsqrt_63", "add_29109", "amax_190", "amin_190", "sub_8722", "convert_element_type_1142", "convert_element_type_1141", "clamp_max_381", "clamp_min_572", "add_29197", "round_382", "mul_18447", "mul_18444", "reciprocal_190", "view_2980", "clamp_min_570", "div_380", "sub_8696", "maximum_190", "full_381", "minimum_190", "full_380", "view_2981", "convert_element_type_1140", "clamp_max_380", "clamp_min_571", "sub_8702", "round_381", "div_381", "convert_element_type_1143", "view_2985", "mul_18469", "view_2984"], "triton_poi_fused__to_copy_mul_sub_view_10:417": ["mul_18474", "sub_8726", "convert_element_type_1144", "view_2987", "convert_element_type_1145", "view_2989", "view_2988"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418": ["amax_191", "mul_18488", "mul_18486", "view_2992", "add_tensor_1", "add_29256", "erf_33", "mul_18487", "amin_191", "add_29339", "round_384", "mul_18537", "mul_18534", "reciprocal_191", "view_2995", "clamp_min_573", "div_382", "sub_8739", "maximum_191", "full_383", "minimum_191", "full_382", "view_2996", "convert_element_type_1146", "clamp_max_382", "clamp_min_574", "sub_8745", "round_383", "div_383", "mul_18559", "sub_8765", "convert_element_type_1148", "convert_element_type_1147", "clamp_max_383", "clamp_min_575", "convert_element_type_1149", "view_3000", "view_2999"], "triton_poi_fused__to_copy_mul_sub_view_10:419": ["mul_18564", "sub_8769", "convert_element_type_1150", "view_3002", "convert_element_type_1151", "view_3004", "view_3003"], "triton_red_fused_add_addmm_native_layer_norm_view_14:420": ["var_mean_64", "clone_257", "add_29402", "view_3007", "add_tensor", "add_29408", "mul_18581", "mul_18580", "sub_8775", "rsqrt_64", "add_29407"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421": ["amax_192", "view_3008", "add_29408", "mul_18581", "mul_18580", "sub_8775", "clone_257", "add_29402", "view_3007", "add_tensor", "rsqrt_64", "add_29407", "var_mean_64", "amin_192", "sub_8813", "convert_element_type_1154", "convert_element_type_1153", "clamp_max_385", "clamp_min_578", "add_29476", "round_386", "mul_18615", "mul_18613", "reciprocal_192", "view_3011", "clamp_min_576", "div_384", "sub_8787", "maximum_192", "full_385", "minimum_192", "full_384", "view_3012", "convert_element_type_1152", "clamp_max_384", "clamp_min_577", "sub_8793", "round_385", "div_385", "convert_element_type_1155", "view_3016", "mul_18634", "view_3015"], "triton_poi_fused__to_copy_mul_sub_view_16:422": ["mul_18639", "sub_8817", "convert_element_type_1156", "view_3018", "convert_element_type_1157", "view_3020", "view_3019"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423": ["amax_193", "mul_18644", "mul_18642", "add_29516", "erf_34", "mul_18643", "amin_193", "add_29572", "round_388", "mul_18668", "mul_18666", "reciprocal_193", "view_3024", "clamp_min_579", "div_386", "sub_8827", "maximum_193", "full_387", "minimum_193", "full_386", "view_3025", "convert_element_type_1158", "clamp_max_386", "clamp_min_580", "sub_8833", "round_387", "div_387", "mul_18687", "sub_8853", "convert_element_type_1160", "convert_element_type_1159", "clamp_max_387", "clamp_min_581", "convert_element_type_1161", "view_3029", "view_3028"], "triton_poi_fused__to_copy_mul_sub_view_18:424": ["mul_18692", "sub_8857", "convert_element_type_1162", "view_3031", "convert_element_type_1163", "view_3033", "view_3032"], "aoti_torch_cuda_convolution:425": ["convolution"], "aoti_torch_cuda_convolution:426": ["convolution_1", "mul_4", "mul_2", "convolution", "add_4", "erf", "mul_3"], "aoti_torch_cuda_mm_out:427": ["mm_default_159"], "aoti_torch_cuda_mm_out:428": ["mm"], "aoti_torch_cuda_mm_out:429": ["mm_default_158"], "at::_ops::_scaled_dot_product_efficient_attention::call:430": ["_scaled_dot_product_efficient_attention", "clone_2", "permute_2", "view_15", "mul_111", "view_14", "add_tensor_159", "clone_3", "permute_4", "view_31", "view_30", "clone_4", "permute_6", "view_47", "view_46", "add_tensor_158"], "aoti_torch_cuda_mm_out:431": ["mm_default_157"], "aoti_torch_cuda_mm_out:432": ["mm_default_156"], "aoti_torch_cuda_mm_out:433": ["mm_default_155"], "aoti_torch_cuda_mm_out:434": ["mm_default_154"], "aoti_torch_cuda_mm_out:435": ["mm_1"], "aoti_torch_cuda_mm_out:436": ["mm_default_153"], "at::_ops::_scaled_dot_product_efficient_attention::call:437": ["_scaled_dot_product_efficient_attention_1", "clone_10", "permute_12", "view_109", "mul_691", "view_108", "add_tensor_154", "clone_11", "permute_14", "view_125", "view_124", "clone_12", "permute_16", "view_141", "view_140", "add_tensor_153"], "aoti_torch_cuda_mm_out:438": ["mm_default_152"], "aoti_torch_cuda_mm_out:439": ["mm_default_151"], "aoti_torch_cuda_mm_out:440": ["mm_default_150"], "aoti_torch_cuda_mm_out:441": ["mm_default_149"], "aoti_torch_cuda_mm_out:442": ["mm_2"], "aoti_torch_cuda_mm_out:443": ["mm_default_148"], "at::_ops::_scaled_dot_product_efficient_attention::call:444": ["_scaled_dot_product_efficient_attention_2", "clone_18", "permute_22", "view_203", "mul_1271", "view_202", "add_tensor_149", "clone_19", "permute_24", "view_219", "view_218", "clone_20", "permute_26", "view_235", "view_234", "add_tensor_148"], "aoti_torch_cuda_mm_out:445": ["mm_default_147"], "aoti_torch_cuda_mm_out:446": ["mm_default_146"], "aoti_torch_cuda_mm_out:447": ["mm_default_145"], "aoti_torch_cuda_mm_out:448": ["mm_default_144"], "aoti_torch_cuda_mm_out:449": ["mm_3"], "aoti_torch_cuda_mm_out:450": ["mm_default_143"], "at::_ops::_scaled_dot_product_efficient_attention::call:451": ["_scaled_dot_product_efficient_attention_3", "clone_26", "permute_32", "view_297", "mul_1851", "view_296", "add_tensor_144", "clone_27", "permute_34", "view_313", "view_312", "clone_28", "permute_36", "view_329", "view_328", "add_tensor_143"], "aoti_torch_cuda_mm_out:452": ["mm_default_142"], "aoti_torch_cuda_mm_out:453": ["mm_default_141"], "aoti_torch_cuda_mm_out:454": ["mm_default_140"], "aoti_torch_cuda_mm_out:455": ["mm_default_139"], "aoti_torch_cuda_mm_out:456": ["mm_4"], "aoti_torch_cuda_mm_out:457": ["mm_default_138"], "at::_ops::_scaled_dot_product_efficient_attention::call:458": ["_scaled_dot_product_efficient_attention_4", "clone_34", "permute_42", "view_391", "mul_2431", "view_390", "add_tensor_139", "clone_35", "permute_44", "view_407", "view_406", "clone_36", "permute_46", "view_423", "view_422", "add_tensor_138"], "aoti_torch_cuda_mm_out:459": ["mm_default_137"], "aoti_torch_cuda_mm_out:460": ["mm_default_136"], "aoti_torch_cuda_mm_out:461": ["mm_default_135"], "aoti_torch_cuda_mm_out:462": ["mm_default_134"], "aoti_torch_cuda_mm_out:463": ["mm_5"], "aoti_torch_cuda_mm_out:464": ["mm_default_133"], "at::_ops::_scaled_dot_product_efficient_attention::call:465": ["_scaled_dot_product_efficient_attention_5", "clone_42", "permute_52", "view_485", "mul_3011", "view_484", "add_tensor_134", "clone_43", "permute_54", "view_501", "view_500", "clone_44", "permute_56", "view_517", "view_516", "add_tensor_133"], "aoti_torch_cuda_mm_out:466": ["mm_default_132"], "aoti_torch_cuda_mm_out:467": ["mm_default_131"], "aoti_torch_cuda_mm_out:468": ["mm_default_130"], "aoti_torch_cuda_mm_out:469": ["mm_default_129"], "aoti_torch_cuda_mm_out:470": ["mm_6"], "aoti_torch_cuda_mm_out:471": ["mm_default_128"], "at::_ops::_scaled_dot_product_efficient_attention::call:472": ["_scaled_dot_product_efficient_attention_6", "clone_50", "permute_62", "view_579", "mul_3591", "view_578", "add_tensor_129", "clone_51", "permute_64", "view_595", "view_594", "clone_52", "permute_66", "view_611", "view_610", "add_tensor_128"], "aoti_torch_cuda_mm_out:473": ["mm_default_127"], "aoti_torch_cuda_mm_out:474": ["mm_default_126"], "aoti_torch_cuda_mm_out:475": ["mm_default_125"], "aoti_torch_cuda_mm_out:476": ["mm_default_124"], "aoti_torch_cuda_mm_out:477": ["mm_7"], "aoti_torch_cuda_mm_out:478": ["mm_default_123"], "at::_ops::_scaled_dot_product_efficient_attention::call:479": ["_scaled_dot_product_efficient_attention_7", "clone_58", "permute_72", "view_673", "mul_4171", "view_672", "add_tensor_124", "clone_59", "permute_74", "view_689", "view_688", "clone_60", "permute_76", "view_705", "view_704", "add_tensor_123"], "aoti_torch_cuda_mm_out:480": ["mm_default_122"], "aoti_torch_cuda_mm_out:481": ["mm_default_121"], "aoti_torch_cuda_mm_out:482": ["mm_default_120"], "aoti_torch_cuda_mm_out:483": ["mm_default_119"], "aoti_torch_cuda_mm_out:484": ["mm_8"], "aoti_torch_cuda_mm_out:485": ["mm_default_118"], "at::_ops::_scaled_dot_product_efficient_attention::call:486": ["_scaled_dot_product_efficient_attention_8", "clone_66", "permute_82", "view_767", "mul_4751", "view_766", "add_tensor_119", "clone_67", "permute_84", "view_783", "view_782", "clone_68", "permute_86", "view_799", "view_798", "add_tensor_118"], "aoti_torch_cuda_mm_out:487": ["mm_default_117"], "aoti_torch_cuda_mm_out:488": ["mm_default_116"], "aoti_torch_cuda_mm_out:489": ["mm_default_115"], "aoti_torch_cuda_mm_out:490": ["mm_default_114"], "aoti_torch_cuda_mm_out:491": ["mm_9"], "aoti_torch_cuda_mm_out:492": ["mm_default_113"], "at::_ops::_scaled_dot_product_efficient_attention::call:493": ["_scaled_dot_product_efficient_attention_9", "clone_74", "permute_92", "view_861", "mul_5331", "view_860", "add_tensor_114", "clone_75", "permute_94", "view_877", "view_876", "clone_76", "permute_96", "view_893", "view_892", "add_tensor_113"], "aoti_torch_cuda_mm_out:494": ["mm_default_112"], "aoti_torch_cuda_mm_out:495": ["mm_default_111"], "aoti_torch_cuda_mm_out:496": ["mm_default_110"], "aoti_torch_cuda_mm_out:497": ["mm_default_109"], "aoti_torch_cuda_mm_out:498": ["mm_10"], "aoti_torch_cuda_mm_out:499": ["mm_default_108"], "at::_ops::_scaled_dot_product_efficient_attention::call:500": ["_scaled_dot_product_efficient_attention_10", "clone_82", "permute_102", "view_955", "mul_5911", "view_954", "add_tensor_109", "clone_83", "permute_104", "view_971", "view_970", "clone_84", "permute_106", "view_987", "view_986", "add_tensor_108"], "aoti_torch_cuda_mm_out:501": ["mm_default_107"], "aoti_torch_cuda_mm_out:502": ["mm_default_106"], "aoti_torch_cuda_mm_out:503": ["mm_default_105"], "aoti_torch_cuda_mm_out:504": ["mm_default_104"], "aoti_torch_cuda_mm_out:505": ["mm_11"], "aoti_torch_cuda_mm_out:506": ["mm_default_103"], "at::_ops::_scaled_dot_product_efficient_attention::call:507": ["_scaled_dot_product_efficient_attention_11", "clone_90", "permute_112", "view_1049", "mul_6491", "view_1048", "add_tensor_104", "clone_91", "permute_114", "view_1065", "view_1064", "clone_92", "permute_116", "view_1081", "view_1080", "add_tensor_103"], "aoti_torch_cuda_mm_out:508": ["mm_default_102"], "aoti_torch_cuda_mm_out:509": ["mm_default_101"], "aoti_torch_cuda_mm_out:510": ["mm_default_100"], "aoti_torch_cuda_mm_out:511": ["mm_default_99"], "aoti_torch_cuda_mm_out:512": ["mm_12"], "aoti_torch_cuda_mm_out:513": ["mm_default_98"], "at::_ops::_scaled_dot_product_efficient_attention::call:514": ["_scaled_dot_product_efficient_attention_12", "clone_98", "permute_122", "view_1143", "mul_7071", "view_1142", "add_tensor_99", "clone_99", "permute_124", "view_1159", "view_1158", "clone_100", "permute_126", "view_1175", "view_1174", "add_tensor_98"], "aoti_torch_cuda_mm_out:515": ["mm_default_97"], "aoti_torch_cuda_mm_out:516": ["mm_default_96"], "aoti_torch_cuda_mm_out:517": ["mm_default_95"], "aoti_torch_cuda_mm_out:518": ["mm_default_94"], "aoti_torch_cuda_mm_out:519": ["mm_13"], "aoti_torch_cuda_mm_out:520": ["mm_default_93"], "at::_ops::_scaled_dot_product_efficient_attention::call:521": ["_scaled_dot_product_efficient_attention_13", "clone_106", "permute_132", "view_1237", "mul_7651", "view_1236", "add_tensor_94", "clone_107", "permute_134", "view_1253", "view_1252", "clone_108", "permute_136", "view_1269", "view_1268", "add_tensor_93"], "aoti_torch_cuda_mm_out:522": ["mm_default_92"], "aoti_torch_cuda_mm_out:523": ["mm_default_91"], "aoti_torch_cuda_mm_out:524": ["mm_default_90"], "aoti_torch_cuda_mm_out:525": ["mm_default_89"], "aoti_torch_cuda_mm_out:526": ["mm_14"], "aoti_torch_cuda_mm_out:527": ["mm_default_88"], "at::_ops::_scaled_dot_product_efficient_attention::call:528": ["_scaled_dot_product_efficient_attention_14", "clone_114", "permute_142", "view_1331", "mul_8231", "view_1330", "add_tensor_89", "clone_115", "permute_144", "view_1347", "view_1346", "clone_116", "permute_146", "view_1363", "view_1362", "add_tensor_88"], "aoti_torch_cuda_mm_out:529": ["mm_default_87"], "aoti_torch_cuda_mm_out:530": ["mm_default_86"], "aoti_torch_cuda_mm_out:531": ["mm_default_85"], "aoti_torch_cuda_mm_out:532": ["mm_default_84"], "aoti_torch_cuda_mm_out:533": ["mm_15"], "aoti_torch_cuda_mm_out:534": ["mm_default_83"], "at::_ops::_scaled_dot_product_efficient_attention::call:535": ["_scaled_dot_product_efficient_attention_15", "clone_122", "permute_152", "view_1425", "mul_8811", "view_1424", "add_tensor_84", "clone_123", "permute_154", "view_1441", "view_1440", "clone_124", "permute_156", "view_1457", "view_1456", "add_tensor_83"], "aoti_torch_cuda_mm_out:536": ["mm_default_82"], "aoti_torch_cuda_mm_out:537": ["mm_default_81"], "aoti_torch_cuda_mm_out:538": ["mm_default_80"], "aoti_torch_cuda_mm_out:539": ["mm_default_79"], "aoti_torch_cuda_mm_out:540": ["mm_16"], "aoti_torch_cuda_mm_out:541": ["mm_default_78"], "at::_ops::_scaled_dot_product_efficient_attention::call:542": ["_scaled_dot_product_efficient_attention_16", "clone_130", "permute_162", "view_1519", "mul_9391", "view_1518", "add_tensor_79", "clone_131", "permute_164", "view_1535", "view_1534", "clone_132", "permute_166", "view_1551", "view_1550", "add_tensor_78"], "aoti_torch_cuda_mm_out:543": ["mm_default_77"], "aoti_torch_cuda_mm_out:544": ["mm_default_76"], "aoti_torch_cuda_mm_out:545": ["mm_default_75"], "aoti_torch_cuda_mm_out:546": ["mm_default_74"], "aoti_torch_cuda_mm_out:547": ["mm_17"], "aoti_torch_cuda_mm_out:548": ["mm_default_73"], "at::_ops::_scaled_dot_product_efficient_attention::call:549": ["_scaled_dot_product_efficient_attention_17", "clone_138", "permute_172", "view_1613", "mul_9971", "view_1612", "add_tensor_74", "clone_139", "permute_174", "view_1629", "view_1628", "clone_140", "permute_176", "view_1645", "view_1644", "add_tensor_73"], "aoti_torch_cuda_mm_out:550": ["mm_default_72"], "aoti_torch_cuda_mm_out:551": ["mm_default_71"], "aoti_torch_cuda_mm_out:552": ["mm_default_70"], "aoti_torch_cuda_mm_out:553": ["mm_default_69"], "aoti_torch_cuda_mm_out:554": ["mm_18"], "aoti_torch_cuda_mm_out:555": ["mm_default_68"], "at::_ops::_scaled_dot_product_efficient_attention::call:556": ["_scaled_dot_product_efficient_attention_18", "clone_146", "permute_182", "view_1707", "mul_10551", "view_1706", "add_tensor_69", "clone_147", "permute_184", "view_1723", "view_1722", "clone_148", "permute_186", "view_1739", "view_1738", "add_tensor_68"], "aoti_torch_cuda_mm_out:557": ["mm_default_67"], "aoti_torch_cuda_mm_out:558": ["mm_default_66"], "aoti_torch_cuda_mm_out:559": ["mm_default_65"], "aoti_torch_cuda_mm_out:560": ["mm_default_64"], "aoti_torch_cuda_mm_out:561": ["mm_19"], "aoti_torch_cuda_mm_out:562": ["mm_default_63"], "at::_ops::_scaled_dot_product_efficient_attention::call:563": ["_scaled_dot_product_efficient_attention_19", "clone_154", "permute_192", "view_1801", "mul_11131", "view_1800", "add_tensor_64", "clone_155", "permute_194", "view_1817", "view_1816", "clone_156", "permute_196", "view_1833", "view_1832", "add_tensor_63"], "aoti_torch_cuda_mm_out:564": ["mm_default_62"], "aoti_torch_cuda_mm_out:565": ["mm_default_61"], "aoti_torch_cuda_mm_out:566": ["mm_default_60"], "aoti_torch_cuda_mm_out:567": ["mm_default_59"], "aoti_torch_cuda_mm_out:568": ["mm_20"], "aoti_torch_cuda_mm_out:569": ["mm_default_58"], "at::_ops::_scaled_dot_product_efficient_attention::call:570": ["_scaled_dot_product_efficient_attention_20", "clone_162", "permute_202", "view_1895", "mul_11711", "view_1894", "add_tensor_59", "clone_163", "permute_204", "view_1911", "view_1910", "clone_164", "permute_206", "view_1927", "view_1926", "add_tensor_58"], "aoti_torch_cuda_mm_out:571": ["mm_default_57"], "aoti_torch_cuda_mm_out:572": ["mm_default_56"], "aoti_torch_cuda_mm_out:573": ["mm_default_55"], "aoti_torch_cuda_mm_out:574": ["mm_default_54"], "aoti_torch_cuda_mm_out:575": ["mm_21"], "aoti_torch_cuda_mm_out:576": ["mm_default_53"], "at::_ops::_scaled_dot_product_efficient_attention::call:577": ["_scaled_dot_product_efficient_attention_21", "clone_170", "permute_212", "view_1989", "mul_12291", "view_1988", "add_tensor_54", "clone_171", "permute_214", "view_2005", "view_2004", "clone_172", "permute_216", "view_2021", "view_2020", "add_tensor_53"], "aoti_torch_cuda_mm_out:578": ["mm_default_52"], "aoti_torch_cuda_mm_out:579": ["mm_default_51"], "aoti_torch_cuda_mm_out:580": ["mm_default_50"], "aoti_torch_cuda_mm_out:581": ["mm_default_49"], "aoti_torch_cuda_mm_out:582": ["mm_22"], "aoti_torch_cuda_mm_out:583": ["mm_default_48"], "at::_ops::_scaled_dot_product_efficient_attention::call:584": ["_scaled_dot_product_efficient_attention_22", "clone_178", "permute_222", "view_2083", "mul_12871", "view_2082", "add_tensor_49", "clone_179", "permute_224", "view_2099", "view_2098", "clone_180", "permute_226", "view_2115", "view_2114", "add_tensor_48"], "aoti_torch_cuda_mm_out:585": ["mm_default_47"], "aoti_torch_cuda_mm_out:586": ["mm_default_46"], "aoti_torch_cuda_mm_out:587": ["mm_default_45"], "aoti_torch_cuda_mm_out:588": ["mm_default_44"], "aoti_torch_cuda_mm_out:589": ["mm_23"], "aoti_torch_cuda_mm_out:590": ["mm_default_43"], "at::_ops::_scaled_dot_product_efficient_attention::call:591": ["_scaled_dot_product_efficient_attention_23", "clone_186", "permute_232", "view_2177", "mul_13451", "view_2176", "add_tensor_44", "clone_187", "permute_234", "view_2193", "view_2192", "clone_188", "permute_236", "view_2209", "view_2208", "add_tensor_43"], "aoti_torch_cuda_mm_out:592": ["mm_default_42"], "aoti_torch_cuda_mm_out:593": ["mm_default_41"], "aoti_torch_cuda_mm_out:594": ["mm_default_40"], "aoti_torch_cuda_mm_out:595": ["mm_default_39"], "aoti_torch_cuda_mm_out:596": ["mm_24"], "aoti_torch_cuda_mm_out:597": ["mm_default_38"], "at::_ops::_scaled_dot_product_efficient_attention::call:598": ["_scaled_dot_product_efficient_attention_24", "clone_194", "permute_242", "view_2271", "mul_14031", "view_2270", "add_tensor_39", "clone_195", "permute_244", "view_2287", "view_2286", "clone_196", "permute_246", "view_2303", "view_2302", "add_tensor_38"], "aoti_torch_cuda_mm_out:599": ["mm_default_37"], "aoti_torch_cuda_mm_out:600": ["mm_default_36"], "aoti_torch_cuda_mm_out:601": ["mm_default_35"], "aoti_torch_cuda_mm_out:602": ["mm_default_34"], "aoti_torch_cuda_mm_out:603": ["mm_25"], "aoti_torch_cuda_mm_out:604": ["mm_default_33"], "at::_ops::_scaled_dot_product_efficient_attention::call:605": ["_scaled_dot_product_efficient_attention_25", "clone_202", "permute_252", "view_2365", "mul_14611", "view_2364", "add_tensor_34", "clone_203", "permute_254", "view_2381", "view_2380", "clone_204", "permute_256", "view_2397", "view_2396", "add_tensor_33"], "aoti_torch_cuda_mm_out:606": ["mm_default_32"], "aoti_torch_cuda_mm_out:607": ["mm_default_31"], "aoti_torch_cuda_mm_out:608": ["mm_default_30"], "aoti_torch_cuda_mm_out:609": ["mm_default_29"], "aoti_torch_cuda_mm_out:610": ["mm_26"], "aoti_torch_cuda_mm_out:611": ["mm_default_28"], "at::_ops::_scaled_dot_product_efficient_attention::call:612": ["_scaled_dot_product_efficient_attention_26", "clone_210", "permute_262", "view_2459", "mul_15191", "view_2458", "add_tensor_29", "clone_211", "permute_264", "view_2475", "view_2474", "clone_212", "permute_266", "view_2491", "view_2490", "add_tensor_28"], "aoti_torch_cuda_mm_out:613": ["mm_default_27"], "aoti_torch_cuda_mm_out:614": ["mm_default_26"], "aoti_torch_cuda_mm_out:615": ["mm_default_25"], "aoti_torch_cuda_mm_out:616": ["mm_default_24"], "aoti_torch_cuda_mm_out:617": ["mm_27"], "aoti_torch_cuda_mm_out:618": ["mm_default_23"], "at::_ops::_scaled_dot_product_efficient_attention::call:619": ["_scaled_dot_product_efficient_attention_27", "clone_218", "permute_272", "view_2553", "mul_15771", "view_2552", "add_tensor_24", "clone_219", "permute_274", "view_2569", "view_2568", "clone_220", "permute_276", "view_2585", "view_2584", "add_tensor_23"], "aoti_torch_cuda_mm_out:620": ["mm_default_22"], "aoti_torch_cuda_mm_out:621": ["mm_default_21"], "aoti_torch_cuda_mm_out:622": ["mm_default_20"], "aoti_torch_cuda_mm_out:623": ["mm_default_19"], "aoti_torch_cuda_mm_out:624": ["mm_28"], "aoti_torch_cuda_mm_out:625": ["mm_default_18"], "at::_ops::_scaled_dot_product_efficient_attention::call:626": ["_scaled_dot_product_efficient_attention_28", "clone_226", "permute_282", "view_2647", "mul_16351", "view_2646", "add_tensor_19", "clone_227", "permute_284", "view_2663", "view_2662", "clone_228", "permute_286", "view_2679", "view_2678", "add_tensor_18"], "aoti_torch_cuda_mm_out:627": ["mm_default_17"], "aoti_torch_cuda_mm_out:628": ["mm_default_16"], "aoti_torch_cuda_mm_out:629": ["mm_default_15"], "aoti_torch_cuda_mm_out:630": ["mm_default_14"], "aoti_torch_cuda_mm_out:631": ["mm_29"], "aoti_torch_cuda_mm_out:632": ["mm_default_13"], "at::_ops::_scaled_dot_product_efficient_attention::call:633": ["_scaled_dot_product_efficient_attention_29", "clone_234", "permute_292", "view_2741", "mul_16931", "view_2740", "add_tensor_14", "clone_235", "permute_294", "view_2757", "view_2756", "clone_236", "permute_296", "view_2773", "view_2772", "add_tensor_13"], "aoti_torch_cuda_mm_out:634": ["mm_default_12"], "aoti_torch_cuda_mm_out:635": ["mm_default_11"], "aoti_torch_cuda_mm_out:636": ["mm_default_10"], "aoti_torch_cuda_mm_out:637": ["mm_default_9"], "aoti_torch_cuda_mm_out:638": ["mm_30"], "aoti_torch_cuda_mm_out:639": ["mm_default_8"], "at::_ops::_scaled_dot_product_efficient_attention::call:640": ["_scaled_dot_product_efficient_attention_30", "clone_242", "permute_302", "view_2835", "mul_17511", "view_2834", "add_tensor_9", "clone_243", "permute_304", "view_2851", "view_2850", "clone_244", "permute_306", "view_2867", "view_2866", "add_tensor_8"], "aoti_torch_cuda_mm_out:641": ["mm_default_7"], "aoti_torch_cuda_mm_out:642": ["mm_default_6"], "aoti_torch_cuda_mm_out:643": ["mm_default_5"], "aoti_torch_cuda_mm_out:644": ["mm_default_4"], "aoti_torch_cuda_mm_out:645": ["mm_31"], "aoti_torch_cuda_mm_out:646": ["mm_default_3"], "at::_ops::_scaled_dot_product_efficient_attention::call:647": ["_scaled_dot_product_efficient_attention_31", "clone_250", "permute_312", "view_2929", "mul_18091", "view_2928", "add_tensor_4", "clone_251", "permute_314", "view_2945", "view_2944", "clone_252", "permute_316", "view_2961", "view_2960", "add_tensor_3"], "aoti_torch_cuda_mm_out:648": ["mm_default_2"], "aoti_torch_cuda_mm_out:649": ["mm_default_1"], "aoti_torch_cuda_mm_out:650": ["mm_default"], "aoti_torch_cuda_mm_out:651": ["mm_32"], "aoti_torch_cuda_mm_out:652": ["mm_33"]}, "postToCppCode": {"convolution_1": ["triton_poi_fused_convolution_gelu_0:1", "triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1:2", "triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2:3", "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13", "aoti_torch_cuda_convolution:426"], "mul_4": ["triton_poi_fused_convolution_gelu_0:1", "triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1:2", "triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2:3", "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13", "aoti_torch_cuda_convolution:426"], "mul_2": ["triton_poi_fused_convolution_gelu_0:1", "triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1:2", "triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2:3", "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13", "aoti_torch_cuda_convolution:426"], "convolution": ["triton_poi_fused_convolution_gelu_0:1", "triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1:2", "triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2:3", "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13", "aoti_torch_cuda_convolution:425", "aoti_torch_cuda_convolution:426"], "add_4": ["triton_poi_fused_convolution_gelu_0:1", "triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1:2", "triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2:3", "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13", "aoti_torch_cuda_convolution:426"], "erf": ["triton_poi_fused_convolution_gelu_0:1", "triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1:2", "triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2:3", "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13", "aoti_torch_cuda_convolution:426"], "mul_3": ["triton_poi_fused_convolution_gelu_0:1", "triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1:2", "triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2:3", "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13", "aoti_torch_cuda_convolution:426"], "var_mean": ["triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1:2", "triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2:3", "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "clone_1": ["triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1:2", "triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2:3", "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "add_22": ["triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1:2", "triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2:3", "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "permute": ["triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1:2", "triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2:3", "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "mul_11": ["triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1:2", "triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2:3", "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "mul_9": ["triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1:2", "triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2:3", "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "add_13": ["triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1:2", "triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2:3", "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "erf_1": ["triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1:2", "triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2:3", "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "mul_10": ["triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1:2", "triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2:3", "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "add_32": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "mul_21": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "mul_20": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "sub_7": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "rsqrt": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "add_31": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "amax": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "amin": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "amax_1": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "amin_1": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "amax_2": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "amin_2": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "sub_44": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "convert_element_type_2": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "convert_element_type_1": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "clamp_max_1": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "clamp_min_2": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "add_119": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "round_2": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "mul_72": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "mul_69": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "reciprocal": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "view_2": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "clamp_min": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "div": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "sub_18": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "maximum": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "full_1": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "minimum": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "full": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "view_3": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "convert_element_type": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "clamp_max": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "clamp_min_1": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "sub_24": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "round_1": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "div_1": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "convert_element_type_3": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "view_7": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "mul_94": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "view_6": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "sub_89": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "convert_element_type_8": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "convert_element_type_7": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "clamp_max_3": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "clamp_min_5": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "add_271": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "round_4": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "mul_168": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "mul_165": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "reciprocal_1": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "view_18": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "clamp_min_3": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "div_2": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "sub_63": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "maximum_1": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "full_3": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "minimum_1": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "full_2": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "view_19": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "convert_element_type_6": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "clamp_max_2": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "clamp_min_4": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "sub_69": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "round_3": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "div_3": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "convert_element_type_9": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "view_23": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "mul_190": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "view_22": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "sub_133": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "convert_element_type_14": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "convert_element_type_13": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "clamp_max_5": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "clamp_min_8": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "add_419": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "round_6": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "mul_267": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "mul_264": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "reciprocal_2": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "view_34": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "clamp_min_6": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "div_4": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "sub_107": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "maximum_2": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "full_5": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "minimum_2": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "full_4": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "view_35": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "convert_element_type_12": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "clamp_max_4": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "clamp_min_7": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "sub_113": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "round_5": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "div_5": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "convert_element_type_15": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "view_39": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "mul_289": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "view_38": ["triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4"], "mul_99": ["triton_poi_fused__to_copy_mul_sub_view_4:5"], "sub_48": ["triton_poi_fused__to_copy_mul_sub_view_4:5"], "convert_element_type_4": ["triton_poi_fused__to_copy_mul_sub_view_4:5"], "view_9": ["triton_poi_fused__to_copy_mul_sub_view_4:5"], "convert_element_type_5": ["triton_poi_fused__to_copy_mul_sub_view_4:5"], "view_11": ["triton_poi_fused__to_copy_mul_sub_view_4:5"], "view_10": ["triton_poi_fused__to_copy_mul_sub_view_4:5"], "mul_195": ["triton_poi_fused__to_copy_mul_sub_view_4:6"], "sub_93": ["triton_poi_fused__to_copy_mul_sub_view_4:6"], "convert_element_type_10": ["triton_poi_fused__to_copy_mul_sub_view_4:6"], "view_25": ["triton_poi_fused__to_copy_mul_sub_view_4:6"], "convert_element_type_11": ["triton_poi_fused__to_copy_mul_sub_view_4:6"], "view_27": ["triton_poi_fused__to_copy_mul_sub_view_4:6"], "view_26": ["triton_poi_fused__to_copy_mul_sub_view_4:6"], "mul_294": ["triton_poi_fused__to_copy_mul_sub_view_4:7"], "sub_137": ["triton_poi_fused__to_copy_mul_sub_view_4:7"], "convert_element_type_16": ["triton_poi_fused__to_copy_mul_sub_view_4:7"], "view_41": ["triton_poi_fused__to_copy_mul_sub_view_4:7"], "convert_element_type_17": ["triton_poi_fused__to_copy_mul_sub_view_4:7"], "view_43": ["triton_poi_fused__to_copy_mul_sub_view_4:7"], "view_42": ["triton_poi_fused__to_copy_mul_sub_view_4:7"], "_scaled_dot_product_efficient_attention": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:8", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:9", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:10", "at::_ops::_scaled_dot_product_efficient_attention::call:430"], "clone_2": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:8", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:9", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:10", "at::_ops::_scaled_dot_product_efficient_attention::call:430"], "permute_2": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:8", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:9", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:10", "at::_ops::_scaled_dot_product_efficient_attention::call:430"], "view_15": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:8", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:9", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:10", "at::_ops::_scaled_dot_product_efficient_attention::call:430"], "mul_111": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:8", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:9", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:10", "at::_ops::_scaled_dot_product_efficient_attention::call:430"], "view_14": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:8", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:9", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:10", "at::_ops::_scaled_dot_product_efficient_attention::call:430"], "add_tensor_159": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:8", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:9", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:10", "at::_ops::_scaled_dot_product_efficient_attention::call:430"], "clone_3": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:8", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:9", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:10", "at::_ops::_scaled_dot_product_efficient_attention::call:430"], "permute_4": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:8", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:9", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:10", "at::_ops::_scaled_dot_product_efficient_attention::call:430"], "view_31": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:8", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:9", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:10", "at::_ops::_scaled_dot_product_efficient_attention::call:430"], "view_30": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:8", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:9", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:10", "at::_ops::_scaled_dot_product_efficient_attention::call:430"], "clone_4": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:8", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:9", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:10", "at::_ops::_scaled_dot_product_efficient_attention::call:430"], "permute_6": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:8", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:9", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:10", "at::_ops::_scaled_dot_product_efficient_attention::call:430"], "view_47": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:8", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:9", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:10", "at::_ops::_scaled_dot_product_efficient_attention::call:430"], "view_46": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:8", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:9", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:10", "at::_ops::_scaled_dot_product_efficient_attention::call:430"], "add_tensor_158": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:8", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:9", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:10", "at::_ops::_scaled_dot_product_efficient_attention::call:430"], "amax_3": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "view_48": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "permute_7": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "amin_3": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "sub_181": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "convert_element_type_20": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "convert_element_type_19": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "clamp_max_7": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "clamp_min_11": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "add_583": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "round_8": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "mul_372": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "mul_369": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "reciprocal_3": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "view_51": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "clamp_min_9": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "div_6": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "sub_155": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "maximum_3": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "full_7": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "minimum_3": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "full_6": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "view_52": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "convert_element_type_18": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "clamp_max_6": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "clamp_min_10": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "sub_161": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "round_7": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "div_7": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "convert_element_type_21": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "view_56": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "mul_394": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "view_55": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11"], "mul_399": ["triton_poi_fused__to_copy_mul_sub_view_4:12"], "sub_185": ["triton_poi_fused__to_copy_mul_sub_view_4:12"], "convert_element_type_22": ["triton_poi_fused__to_copy_mul_sub_view_4:12"], "view_58": ["triton_poi_fused__to_copy_mul_sub_view_4:12"], "convert_element_type_23": ["triton_poi_fused__to_copy_mul_sub_view_4:12"], "view_60": ["triton_poi_fused__to_copy_mul_sub_view_4:12"], "view_59": ["triton_poi_fused__to_copy_mul_sub_view_4:12"], "add_646": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "view_63": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "add_tensor_157": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "var_mean_1": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "clone_6": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "add_652": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "mul_416": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "mul_415": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "sub_191": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "rsqrt_1": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "add_651": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "amax_4": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "amin_4": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "sub_228": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "convert_element_type_26": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "convert_element_type_25": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "clamp_max_9": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "clamp_min_14": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "add_739": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "round_10": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "mul_467": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "mul_464": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "reciprocal_4": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "view_66": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "clamp_min_12": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "div_8": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "sub_202": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "maximum_4": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "full_9": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "minimum_4": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "full_8": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "view_67": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "convert_element_type_24": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "clamp_max_8": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "clamp_min_13": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "sub_208": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "round_9": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "div_9": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "convert_element_type_27": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "view_71": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "mul_489": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "view_70": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13"], "mul_494": ["triton_poi_fused__to_copy_mul_sub_view_10:14"], "sub_232": ["triton_poi_fused__to_copy_mul_sub_view_10:14"], "convert_element_type_28": ["triton_poi_fused__to_copy_mul_sub_view_10:14"], "view_73": ["triton_poi_fused__to_copy_mul_sub_view_10:14"], "convert_element_type_29": ["triton_poi_fused__to_copy_mul_sub_view_10:14"], "view_75": ["triton_poi_fused__to_copy_mul_sub_view_10:14"], "view_74": ["triton_poi_fused__to_copy_mul_sub_view_10:14"], "amax_5": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "mul_508": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "mul_506": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "view_78": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "add_tensor_156": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "add_798": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "erf_2": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "mul_507": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "amin_5": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "add_881": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "round_12": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "mul_557": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "mul_554": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "reciprocal_5": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "view_81": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "clamp_min_15": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "div_10": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "sub_245": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "maximum_5": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "full_11": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "minimum_5": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "full_10": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "view_82": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "convert_element_type_30": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "clamp_max_10": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "clamp_min_16": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "sub_251": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "round_11": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "div_11": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "mul_579": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "sub_271": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "convert_element_type_32": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "convert_element_type_31": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "clamp_max_11": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "clamp_min_17": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "convert_element_type_33": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "view_86": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "view_85": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15"], "mul_584": ["triton_poi_fused__to_copy_mul_sub_view_10:16"], "sub_275": ["triton_poi_fused__to_copy_mul_sub_view_10:16"], "convert_element_type_34": ["triton_poi_fused__to_copy_mul_sub_view_10:16"], "view_88": ["triton_poi_fused__to_copy_mul_sub_view_10:16"], "convert_element_type_35": ["triton_poi_fused__to_copy_mul_sub_view_10:16"], "view_90": ["triton_poi_fused__to_copy_mul_sub_view_10:16"], "view_89": ["triton_poi_fused__to_copy_mul_sub_view_10:16"], "var_mean_2": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "clone_9": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "add_944": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "view_93": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "add_tensor_155": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "add_950": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "mul_601": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "mul_600": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "sub_281": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "rsqrt_2": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "add_949": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "amax_6": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "amin_6": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "amax_7": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "amin_7": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "amax_8": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "amin_8": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "sub_318": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "convert_element_type_38": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "convert_element_type_37": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "clamp_max_13": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "clamp_min_20": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "add_1037": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "round_14": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "mul_652": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "mul_649": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "reciprocal_6": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "view_96": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "clamp_min_18": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "div_12": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "sub_292": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "maximum_6": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "full_13": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "minimum_6": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "full_12": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "view_97": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "convert_element_type_36": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "clamp_max_12": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "clamp_min_19": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "sub_298": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "round_13": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "div_13": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "convert_element_type_39": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "view_101": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "mul_674": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "view_100": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "sub_363": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "convert_element_type_44": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "convert_element_type_43": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "clamp_max_15": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "clamp_min_23": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "add_1189": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "round_16": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "mul_748": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "mul_745": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "reciprocal_7": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "view_112": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "clamp_min_21": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "div_14": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "sub_337": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "maximum_7": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "full_15": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "minimum_7": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "full_14": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "view_113": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "convert_element_type_42": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "clamp_max_14": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "clamp_min_22": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "sub_343": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "round_15": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "div_15": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "convert_element_type_45": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "view_117": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "mul_770": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "view_116": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "sub_407": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "convert_element_type_50": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "convert_element_type_49": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "clamp_max_17": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "clamp_min_26": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "add_1337": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "round_18": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "mul_847": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "mul_844": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "reciprocal_8": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "view_128": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "clamp_min_24": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "div_16": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "sub_381": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "maximum_8": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "full_17": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "minimum_8": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "full_16": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "view_129": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "convert_element_type_48": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "clamp_max_16": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "clamp_min_25": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "sub_387": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "round_17": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "div_17": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "convert_element_type_51": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "view_133": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "mul_869": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "view_132": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17"], "mul_679": ["triton_poi_fused__to_copy_mul_sub_view_4:18"], "sub_322": ["triton_poi_fused__to_copy_mul_sub_view_4:18"], "convert_element_type_40": ["triton_poi_fused__to_copy_mul_sub_view_4:18"], "view_103": ["triton_poi_fused__to_copy_mul_sub_view_4:18"], "convert_element_type_41": ["triton_poi_fused__to_copy_mul_sub_view_4:18"], "view_105": ["triton_poi_fused__to_copy_mul_sub_view_4:18"], "view_104": ["triton_poi_fused__to_copy_mul_sub_view_4:18"], "mul_775": ["triton_poi_fused__to_copy_mul_sub_view_4:19"], "sub_367": ["triton_poi_fused__to_copy_mul_sub_view_4:19"], "convert_element_type_46": ["triton_poi_fused__to_copy_mul_sub_view_4:19"], "view_119": ["triton_poi_fused__to_copy_mul_sub_view_4:19"], "convert_element_type_47": ["triton_poi_fused__to_copy_mul_sub_view_4:19"], "view_121": ["triton_poi_fused__to_copy_mul_sub_view_4:19"], "view_120": ["triton_poi_fused__to_copy_mul_sub_view_4:19"], "mul_874": ["triton_poi_fused__to_copy_mul_sub_view_4:20"], "sub_411": ["triton_poi_fused__to_copy_mul_sub_view_4:20"], "convert_element_type_52": ["triton_poi_fused__to_copy_mul_sub_view_4:20"], "view_135": ["triton_poi_fused__to_copy_mul_sub_view_4:20"], "convert_element_type_53": ["triton_poi_fused__to_copy_mul_sub_view_4:20"], "view_137": ["triton_poi_fused__to_copy_mul_sub_view_4:20"], "view_136": ["triton_poi_fused__to_copy_mul_sub_view_4:20"], "_scaled_dot_product_efficient_attention_1": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:21", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:22", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:23", "at::_ops::_scaled_dot_product_efficient_attention::call:437"], "clone_10": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:21", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:22", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:23", "at::_ops::_scaled_dot_product_efficient_attention::call:437"], "permute_12": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:21", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:22", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:23", "at::_ops::_scaled_dot_product_efficient_attention::call:437"], "view_109": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:21", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:22", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:23", "at::_ops::_scaled_dot_product_efficient_attention::call:437"], "mul_691": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:21", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:22", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:23", "at::_ops::_scaled_dot_product_efficient_attention::call:437"], "view_108": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:21", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:22", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:23", "at::_ops::_scaled_dot_product_efficient_attention::call:437"], "add_tensor_154": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:21", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:22", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:23", "at::_ops::_scaled_dot_product_efficient_attention::call:437"], "clone_11": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:21", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:22", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:23", "at::_ops::_scaled_dot_product_efficient_attention::call:437"], "permute_14": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:21", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:22", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:23", "at::_ops::_scaled_dot_product_efficient_attention::call:437"], "view_125": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:21", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:22", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:23", "at::_ops::_scaled_dot_product_efficient_attention::call:437"], "view_124": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:21", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:22", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:23", "at::_ops::_scaled_dot_product_efficient_attention::call:437"], "clone_12": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:21", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:22", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:23", "at::_ops::_scaled_dot_product_efficient_attention::call:437"], "permute_16": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:21", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:22", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:23", "at::_ops::_scaled_dot_product_efficient_attention::call:437"], "view_141": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:21", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:22", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:23", "at::_ops::_scaled_dot_product_efficient_attention::call:437"], "view_140": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:21", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:22", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:23", "at::_ops::_scaled_dot_product_efficient_attention::call:437"], "add_tensor_153": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:21", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:22", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:23", "at::_ops::_scaled_dot_product_efficient_attention::call:437"], "amax_9": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "view_142": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "permute_17": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "amin_9": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "sub_455": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "convert_element_type_56": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "convert_element_type_55": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "clamp_max_19": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "clamp_min_29": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "add_1501": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "round_20": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "mul_952": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "mul_949": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "reciprocal_9": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "view_145": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "clamp_min_27": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "div_18": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "sub_429": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "maximum_9": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "full_19": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "minimum_9": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "full_18": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "view_146": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "convert_element_type_54": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "clamp_max_18": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "clamp_min_28": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "sub_435": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "round_19": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "div_19": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "convert_element_type_57": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "view_150": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "mul_974": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "view_149": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24"], "mul_979": ["triton_poi_fused__to_copy_mul_sub_view_4:25"], "sub_459": ["triton_poi_fused__to_copy_mul_sub_view_4:25"], "convert_element_type_58": ["triton_poi_fused__to_copy_mul_sub_view_4:25"], "view_152": ["triton_poi_fused__to_copy_mul_sub_view_4:25"], "convert_element_type_59": ["triton_poi_fused__to_copy_mul_sub_view_4:25"], "view_154": ["triton_poi_fused__to_copy_mul_sub_view_4:25"], "view_153": ["triton_poi_fused__to_copy_mul_sub_view_4:25"], "add_1564": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "view_157": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "add_tensor_152": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "var_mean_3": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "clone_14": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "add_1570": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "mul_996": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "mul_995": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "sub_465": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "rsqrt_3": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "add_1569": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "amax_10": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "amin_10": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "sub_502": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "convert_element_type_62": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "convert_element_type_61": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "clamp_max_21": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "clamp_min_32": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "add_1657": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "round_22": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "mul_1047": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "mul_1044": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "reciprocal_10": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "view_160": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "clamp_min_30": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "div_20": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "sub_476": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "maximum_10": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "full_21": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "minimum_10": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "full_20": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "view_161": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "convert_element_type_60": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "clamp_max_20": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "clamp_min_31": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "sub_482": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "round_21": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "div_21": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "convert_element_type_63": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "view_165": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "mul_1069": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "view_164": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26"], "mul_1074": ["triton_poi_fused__to_copy_mul_sub_view_10:27"], "sub_506": ["triton_poi_fused__to_copy_mul_sub_view_10:27"], "convert_element_type_64": ["triton_poi_fused__to_copy_mul_sub_view_10:27"], "view_167": ["triton_poi_fused__to_copy_mul_sub_view_10:27"], "convert_element_type_65": ["triton_poi_fused__to_copy_mul_sub_view_10:27"], "view_169": ["triton_poi_fused__to_copy_mul_sub_view_10:27"], "view_168": ["triton_poi_fused__to_copy_mul_sub_view_10:27"], "amax_11": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "mul_1088": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "mul_1086": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "view_172": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "add_tensor_151": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "add_1716": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "erf_3": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "mul_1087": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "amin_11": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "add_1799": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "round_24": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "mul_1137": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "mul_1134": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "reciprocal_11": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "view_175": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "clamp_min_33": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "div_22": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "sub_519": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "maximum_11": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "full_23": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "minimum_11": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "full_22": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "view_176": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "convert_element_type_66": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "clamp_max_22": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "clamp_min_34": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "sub_525": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "round_23": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "div_23": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "mul_1159": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "sub_545": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "convert_element_type_68": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "convert_element_type_67": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "clamp_max_23": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "clamp_min_35": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "convert_element_type_69": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "view_180": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "view_179": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28"], "mul_1164": ["triton_poi_fused__to_copy_mul_sub_view_10:29"], "sub_549": ["triton_poi_fused__to_copy_mul_sub_view_10:29"], "convert_element_type_70": ["triton_poi_fused__to_copy_mul_sub_view_10:29"], "view_182": ["triton_poi_fused__to_copy_mul_sub_view_10:29"], "convert_element_type_71": ["triton_poi_fused__to_copy_mul_sub_view_10:29"], "view_184": ["triton_poi_fused__to_copy_mul_sub_view_10:29"], "view_183": ["triton_poi_fused__to_copy_mul_sub_view_10:29"], "var_mean_4": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "clone_17": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "add_1862": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "view_187": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "add_tensor_150": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "add_1868": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "mul_1181": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "mul_1180": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "sub_555": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "rsqrt_4": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "add_1867": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "amax_12": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "amin_12": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "amax_13": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "amin_13": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "amax_14": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "amin_14": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "sub_592": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "convert_element_type_74": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "convert_element_type_73": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "clamp_max_25": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "clamp_min_38": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "add_1955": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "round_26": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "mul_1232": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "mul_1229": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "reciprocal_12": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "view_190": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "clamp_min_36": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "div_24": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "sub_566": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "maximum_12": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "full_25": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "minimum_12": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "full_24": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "view_191": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "convert_element_type_72": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "clamp_max_24": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "clamp_min_37": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "sub_572": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "round_25": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "div_25": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "convert_element_type_75": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "view_195": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "mul_1254": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "view_194": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "sub_637": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "convert_element_type_80": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "convert_element_type_79": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "clamp_max_27": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "clamp_min_41": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "add_2107": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "round_28": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "mul_1328": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "mul_1325": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "reciprocal_13": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "view_206": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "clamp_min_39": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "div_26": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "sub_611": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "maximum_13": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "full_27": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "minimum_13": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "full_26": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "view_207": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "convert_element_type_78": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "clamp_max_26": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "clamp_min_40": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "sub_617": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "round_27": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "div_27": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "convert_element_type_81": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "view_211": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "mul_1350": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "view_210": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "sub_681": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "convert_element_type_86": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "convert_element_type_85": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "clamp_max_29": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "clamp_min_44": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "add_2255": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "round_30": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "mul_1427": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "mul_1424": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "reciprocal_14": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "view_222": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "clamp_min_42": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "div_28": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "sub_655": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "maximum_14": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "full_29": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "minimum_14": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "full_28": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "view_223": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "convert_element_type_84": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "clamp_max_28": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "clamp_min_43": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "sub_661": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "round_29": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "div_29": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "convert_element_type_87": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "view_227": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "mul_1449": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "view_226": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30"], "mul_1259": ["triton_poi_fused__to_copy_mul_sub_view_4:31"], "sub_596": ["triton_poi_fused__to_copy_mul_sub_view_4:31"], "convert_element_type_76": ["triton_poi_fused__to_copy_mul_sub_view_4:31"], "view_197": ["triton_poi_fused__to_copy_mul_sub_view_4:31"], "convert_element_type_77": ["triton_poi_fused__to_copy_mul_sub_view_4:31"], "view_199": ["triton_poi_fused__to_copy_mul_sub_view_4:31"], "view_198": ["triton_poi_fused__to_copy_mul_sub_view_4:31"], "mul_1355": ["triton_poi_fused__to_copy_mul_sub_view_4:32"], "sub_641": ["triton_poi_fused__to_copy_mul_sub_view_4:32"], "convert_element_type_82": ["triton_poi_fused__to_copy_mul_sub_view_4:32"], "view_213": ["triton_poi_fused__to_copy_mul_sub_view_4:32"], "convert_element_type_83": ["triton_poi_fused__to_copy_mul_sub_view_4:32"], "view_215": ["triton_poi_fused__to_copy_mul_sub_view_4:32"], "view_214": ["triton_poi_fused__to_copy_mul_sub_view_4:32"], "mul_1454": ["triton_poi_fused__to_copy_mul_sub_view_4:33"], "sub_685": ["triton_poi_fused__to_copy_mul_sub_view_4:33"], "convert_element_type_88": ["triton_poi_fused__to_copy_mul_sub_view_4:33"], "view_229": ["triton_poi_fused__to_copy_mul_sub_view_4:33"], "convert_element_type_89": ["triton_poi_fused__to_copy_mul_sub_view_4:33"], "view_231": ["triton_poi_fused__to_copy_mul_sub_view_4:33"], "view_230": ["triton_poi_fused__to_copy_mul_sub_view_4:33"], "_scaled_dot_product_efficient_attention_2": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:34", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:35", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:36", "at::_ops::_scaled_dot_product_efficient_attention::call:444"], "clone_18": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:34", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:35", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:36", "at::_ops::_scaled_dot_product_efficient_attention::call:444"], "permute_22": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:34", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:35", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:36", "at::_ops::_scaled_dot_product_efficient_attention::call:444"], "view_203": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:34", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:35", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:36", "at::_ops::_scaled_dot_product_efficient_attention::call:444"], "mul_1271": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:34", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:35", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:36", "at::_ops::_scaled_dot_product_efficient_attention::call:444"], "view_202": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:34", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:35", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:36", "at::_ops::_scaled_dot_product_efficient_attention::call:444"], "add_tensor_149": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:34", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:35", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:36", "at::_ops::_scaled_dot_product_efficient_attention::call:444"], "clone_19": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:34", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:35", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:36", "at::_ops::_scaled_dot_product_efficient_attention::call:444"], "permute_24": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:34", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:35", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:36", "at::_ops::_scaled_dot_product_efficient_attention::call:444"], "view_219": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:34", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:35", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:36", "at::_ops::_scaled_dot_product_efficient_attention::call:444"], "view_218": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:34", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:35", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:36", "at::_ops::_scaled_dot_product_efficient_attention::call:444"], "clone_20": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:34", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:35", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:36", "at::_ops::_scaled_dot_product_efficient_attention::call:444"], "permute_26": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:34", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:35", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:36", "at::_ops::_scaled_dot_product_efficient_attention::call:444"], "view_235": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:34", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:35", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:36", "at::_ops::_scaled_dot_product_efficient_attention::call:444"], "view_234": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:34", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:35", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:36", "at::_ops::_scaled_dot_product_efficient_attention::call:444"], "add_tensor_148": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:34", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:35", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:36", "at::_ops::_scaled_dot_product_efficient_attention::call:444"], "amax_15": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "view_236": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "permute_27": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "amin_15": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "sub_729": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "convert_element_type_92": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "convert_element_type_91": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "clamp_max_31": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "clamp_min_47": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "add_2419": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "round_32": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "mul_1532": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "mul_1529": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "reciprocal_15": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "view_239": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "clamp_min_45": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "div_30": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "sub_703": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "maximum_15": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "full_31": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "minimum_15": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "full_30": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "view_240": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "convert_element_type_90": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "clamp_max_30": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "clamp_min_46": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "sub_709": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "round_31": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "div_31": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "convert_element_type_93": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "view_244": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "mul_1554": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "view_243": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37"], "mul_1559": ["triton_poi_fused__to_copy_mul_sub_view_4:38"], "sub_733": ["triton_poi_fused__to_copy_mul_sub_view_4:38"], "convert_element_type_94": ["triton_poi_fused__to_copy_mul_sub_view_4:38"], "view_246": ["triton_poi_fused__to_copy_mul_sub_view_4:38"], "convert_element_type_95": ["triton_poi_fused__to_copy_mul_sub_view_4:38"], "view_248": ["triton_poi_fused__to_copy_mul_sub_view_4:38"], "view_247": ["triton_poi_fused__to_copy_mul_sub_view_4:38"], "add_2482": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "view_251": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "add_tensor_147": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "var_mean_5": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "clone_22": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "add_2488": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "mul_1576": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "mul_1575": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "sub_739": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "rsqrt_5": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "add_2487": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "amax_16": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "amin_16": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "sub_776": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "convert_element_type_98": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "convert_element_type_97": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "clamp_max_33": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "clamp_min_50": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "add_2575": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "round_34": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "mul_1627": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "mul_1624": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "reciprocal_16": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "view_254": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "clamp_min_48": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "div_32": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "sub_750": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "maximum_16": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "full_33": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "minimum_16": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "full_32": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "view_255": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "convert_element_type_96": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "clamp_max_32": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "clamp_min_49": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "sub_756": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "round_33": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "div_33": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "convert_element_type_99": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "view_259": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "mul_1649": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "view_258": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39"], "mul_1654": ["triton_poi_fused__to_copy_mul_sub_view_10:40"], "sub_780": ["triton_poi_fused__to_copy_mul_sub_view_10:40"], "convert_element_type_100": ["triton_poi_fused__to_copy_mul_sub_view_10:40"], "view_261": ["triton_poi_fused__to_copy_mul_sub_view_10:40"], "convert_element_type_101": ["triton_poi_fused__to_copy_mul_sub_view_10:40"], "view_263": ["triton_poi_fused__to_copy_mul_sub_view_10:40"], "view_262": ["triton_poi_fused__to_copy_mul_sub_view_10:40"], "amax_17": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "mul_1668": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "mul_1666": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "view_266": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "add_tensor_146": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "add_2634": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "erf_4": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "mul_1667": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "amin_17": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "add_2717": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "round_36": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "mul_1717": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "mul_1714": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "reciprocal_17": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "view_269": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "clamp_min_51": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "div_34": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "sub_793": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "maximum_17": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "full_35": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "minimum_17": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "full_34": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "view_270": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "convert_element_type_102": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "clamp_max_34": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "clamp_min_52": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "sub_799": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "round_35": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "div_35": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "mul_1739": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "sub_819": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "convert_element_type_104": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "convert_element_type_103": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "clamp_max_35": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "clamp_min_53": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "convert_element_type_105": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "view_274": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "view_273": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41"], "mul_1744": ["triton_poi_fused__to_copy_mul_sub_view_10:42"], "sub_823": ["triton_poi_fused__to_copy_mul_sub_view_10:42"], "convert_element_type_106": ["triton_poi_fused__to_copy_mul_sub_view_10:42"], "view_276": ["triton_poi_fused__to_copy_mul_sub_view_10:42"], "convert_element_type_107": ["triton_poi_fused__to_copy_mul_sub_view_10:42"], "view_278": ["triton_poi_fused__to_copy_mul_sub_view_10:42"], "view_277": ["triton_poi_fused__to_copy_mul_sub_view_10:42"], "var_mean_6": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "clone_25": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "add_2780": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "view_281": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "add_tensor_145": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "add_2786": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "mul_1761": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "mul_1760": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "sub_829": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "rsqrt_6": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "add_2785": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "amax_18": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "amin_18": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "amax_19": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "amin_19": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "amax_20": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "amin_20": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "sub_866": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "convert_element_type_110": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "convert_element_type_109": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "clamp_max_37": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "clamp_min_56": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "add_2873": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "round_38": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "mul_1812": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "mul_1809": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "reciprocal_18": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "view_284": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "clamp_min_54": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "div_36": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "sub_840": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "maximum_18": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "full_37": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "minimum_18": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "full_36": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "view_285": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "convert_element_type_108": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "clamp_max_36": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "clamp_min_55": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "sub_846": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "round_37": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "div_37": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "convert_element_type_111": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "view_289": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "mul_1834": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "view_288": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "sub_911": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "convert_element_type_116": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "convert_element_type_115": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "clamp_max_39": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "clamp_min_59": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "add_3025": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "round_40": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "mul_1908": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "mul_1905": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "reciprocal_19": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "view_300": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "clamp_min_57": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "div_38": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "sub_885": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "maximum_19": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "full_39": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "minimum_19": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "full_38": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "view_301": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "convert_element_type_114": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "clamp_max_38": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "clamp_min_58": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "sub_891": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "round_39": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "div_39": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "convert_element_type_117": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "view_305": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "mul_1930": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "view_304": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "sub_955": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "convert_element_type_122": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "convert_element_type_121": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "clamp_max_41": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "clamp_min_62": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "add_3173": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "round_42": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "mul_2007": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "mul_2004": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "reciprocal_20": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "view_316": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "clamp_min_60": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "div_40": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "sub_929": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "maximum_20": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "full_41": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "minimum_20": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "full_40": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "view_317": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "convert_element_type_120": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "clamp_max_40": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "clamp_min_61": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "sub_935": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "round_41": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "div_41": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "convert_element_type_123": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "view_321": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "mul_2029": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "view_320": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43"], "mul_1839": ["triton_poi_fused__to_copy_mul_sub_view_4:44"], "sub_870": ["triton_poi_fused__to_copy_mul_sub_view_4:44"], "convert_element_type_112": ["triton_poi_fused__to_copy_mul_sub_view_4:44"], "view_291": ["triton_poi_fused__to_copy_mul_sub_view_4:44"], "convert_element_type_113": ["triton_poi_fused__to_copy_mul_sub_view_4:44"], "view_293": ["triton_poi_fused__to_copy_mul_sub_view_4:44"], "view_292": ["triton_poi_fused__to_copy_mul_sub_view_4:44"], "mul_1935": ["triton_poi_fused__to_copy_mul_sub_view_4:45"], "sub_915": ["triton_poi_fused__to_copy_mul_sub_view_4:45"], "convert_element_type_118": ["triton_poi_fused__to_copy_mul_sub_view_4:45"], "view_307": ["triton_poi_fused__to_copy_mul_sub_view_4:45"], "convert_element_type_119": ["triton_poi_fused__to_copy_mul_sub_view_4:45"], "view_309": ["triton_poi_fused__to_copy_mul_sub_view_4:45"], "view_308": ["triton_poi_fused__to_copy_mul_sub_view_4:45"], "mul_2034": ["triton_poi_fused__to_copy_mul_sub_view_4:46"], "sub_959": ["triton_poi_fused__to_copy_mul_sub_view_4:46"], "convert_element_type_124": ["triton_poi_fused__to_copy_mul_sub_view_4:46"], "view_323": ["triton_poi_fused__to_copy_mul_sub_view_4:46"], "convert_element_type_125": ["triton_poi_fused__to_copy_mul_sub_view_4:46"], "view_325": ["triton_poi_fused__to_copy_mul_sub_view_4:46"], "view_324": ["triton_poi_fused__to_copy_mul_sub_view_4:46"], "_scaled_dot_product_efficient_attention_3": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:47", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:48", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:49", "at::_ops::_scaled_dot_product_efficient_attention::call:451"], "clone_26": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:47", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:48", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:49", "at::_ops::_scaled_dot_product_efficient_attention::call:451"], "permute_32": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:47", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:48", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:49", "at::_ops::_scaled_dot_product_efficient_attention::call:451"], "view_297": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:47", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:48", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:49", "at::_ops::_scaled_dot_product_efficient_attention::call:451"], "mul_1851": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:47", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:48", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:49", "at::_ops::_scaled_dot_product_efficient_attention::call:451"], "view_296": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:47", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:48", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:49", "at::_ops::_scaled_dot_product_efficient_attention::call:451"], "add_tensor_144": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:47", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:48", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:49", "at::_ops::_scaled_dot_product_efficient_attention::call:451"], "clone_27": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:47", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:48", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:49", "at::_ops::_scaled_dot_product_efficient_attention::call:451"], "permute_34": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:47", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:48", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:49", "at::_ops::_scaled_dot_product_efficient_attention::call:451"], "view_313": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:47", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:48", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:49", "at::_ops::_scaled_dot_product_efficient_attention::call:451"], "view_312": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:47", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:48", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:49", "at::_ops::_scaled_dot_product_efficient_attention::call:451"], "clone_28": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:47", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:48", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:49", "at::_ops::_scaled_dot_product_efficient_attention::call:451"], "permute_36": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:47", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:48", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:49", "at::_ops::_scaled_dot_product_efficient_attention::call:451"], "view_329": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:47", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:48", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:49", "at::_ops::_scaled_dot_product_efficient_attention::call:451"], "view_328": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:47", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:48", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:49", "at::_ops::_scaled_dot_product_efficient_attention::call:451"], "add_tensor_143": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:47", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:48", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:49", "at::_ops::_scaled_dot_product_efficient_attention::call:451"], "amax_21": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "view_330": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "permute_37": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "amin_21": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "sub_1003": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "convert_element_type_128": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "convert_element_type_127": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "clamp_max_43": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "clamp_min_65": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "add_3337": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "round_44": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "mul_2112": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "mul_2109": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "reciprocal_21": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "view_333": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "clamp_min_63": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "div_42": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "sub_977": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "maximum_21": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "full_43": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "minimum_21": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "full_42": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "view_334": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "convert_element_type_126": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "clamp_max_42": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "clamp_min_64": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "sub_983": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "round_43": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "div_43": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "convert_element_type_129": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "view_338": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "mul_2134": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "view_337": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50"], "mul_2139": ["triton_poi_fused__to_copy_mul_sub_view_4:51"], "sub_1007": ["triton_poi_fused__to_copy_mul_sub_view_4:51"], "convert_element_type_130": ["triton_poi_fused__to_copy_mul_sub_view_4:51"], "view_340": ["triton_poi_fused__to_copy_mul_sub_view_4:51"], "convert_element_type_131": ["triton_poi_fused__to_copy_mul_sub_view_4:51"], "view_342": ["triton_poi_fused__to_copy_mul_sub_view_4:51"], "view_341": ["triton_poi_fused__to_copy_mul_sub_view_4:51"], "add_3400": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "view_345": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "add_tensor_142": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "var_mean_7": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "clone_30": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "add_3406": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "mul_2156": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "mul_2155": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "sub_1013": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "rsqrt_7": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "add_3405": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "amax_22": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "amin_22": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "sub_1050": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "convert_element_type_134": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "convert_element_type_133": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "clamp_max_45": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "clamp_min_68": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "add_3493": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "round_46": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "mul_2207": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "mul_2204": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "reciprocal_22": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "view_348": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "clamp_min_66": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "div_44": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "sub_1024": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "maximum_22": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "full_45": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "minimum_22": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "full_44": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "view_349": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "convert_element_type_132": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "clamp_max_44": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "clamp_min_67": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "sub_1030": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "round_45": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "div_45": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "convert_element_type_135": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "view_353": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "mul_2229": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "view_352": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52"], "mul_2234": ["triton_poi_fused__to_copy_mul_sub_view_10:53"], "sub_1054": ["triton_poi_fused__to_copy_mul_sub_view_10:53"], "convert_element_type_136": ["triton_poi_fused__to_copy_mul_sub_view_10:53"], "view_355": ["triton_poi_fused__to_copy_mul_sub_view_10:53"], "convert_element_type_137": ["triton_poi_fused__to_copy_mul_sub_view_10:53"], "view_357": ["triton_poi_fused__to_copy_mul_sub_view_10:53"], "view_356": ["triton_poi_fused__to_copy_mul_sub_view_10:53"], "amax_23": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "mul_2248": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "mul_2246": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "view_360": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "add_tensor_141": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "add_3552": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "erf_5": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "mul_2247": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "amin_23": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "add_3635": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "round_48": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "mul_2297": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "mul_2294": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "reciprocal_23": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "view_363": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "clamp_min_69": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "div_46": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "sub_1067": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "maximum_23": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "full_47": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "minimum_23": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "full_46": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "view_364": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "convert_element_type_138": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "clamp_max_46": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "clamp_min_70": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "sub_1073": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "round_47": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "div_47": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "mul_2319": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "sub_1093": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "convert_element_type_140": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "convert_element_type_139": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "clamp_max_47": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "clamp_min_71": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "convert_element_type_141": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "view_368": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "view_367": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54"], "mul_2324": ["triton_poi_fused__to_copy_mul_sub_view_10:55"], "sub_1097": ["triton_poi_fused__to_copy_mul_sub_view_10:55"], "convert_element_type_142": ["triton_poi_fused__to_copy_mul_sub_view_10:55"], "view_370": ["triton_poi_fused__to_copy_mul_sub_view_10:55"], "convert_element_type_143": ["triton_poi_fused__to_copy_mul_sub_view_10:55"], "view_372": ["triton_poi_fused__to_copy_mul_sub_view_10:55"], "view_371": ["triton_poi_fused__to_copy_mul_sub_view_10:55"], "var_mean_8": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "clone_33": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "add_3698": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "view_375": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "add_tensor_140": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "add_3704": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "mul_2341": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "mul_2340": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "sub_1103": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "rsqrt_8": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "add_3703": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "amax_24": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "amin_24": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "amax_25": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "amin_25": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "amax_26": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "amin_26": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "sub_1140": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "convert_element_type_146": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "convert_element_type_145": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "clamp_max_49": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "clamp_min_74": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "add_3791": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "round_50": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "mul_2392": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "mul_2389": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "reciprocal_24": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "view_378": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "clamp_min_72": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "div_48": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "sub_1114": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "maximum_24": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "full_49": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "minimum_24": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "full_48": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "view_379": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "convert_element_type_144": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "clamp_max_48": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "clamp_min_73": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "sub_1120": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "round_49": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "div_49": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "convert_element_type_147": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "view_383": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "mul_2414": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "view_382": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "sub_1185": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "convert_element_type_152": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "convert_element_type_151": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "clamp_max_51": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "clamp_min_77": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "add_3943": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "round_52": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "mul_2488": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "mul_2485": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "reciprocal_25": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "view_394": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "clamp_min_75": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "div_50": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "sub_1159": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "maximum_25": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "full_51": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "minimum_25": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "full_50": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "view_395": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "convert_element_type_150": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "clamp_max_50": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "clamp_min_76": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "sub_1165": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "round_51": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "div_51": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "convert_element_type_153": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "view_399": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "mul_2510": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "view_398": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "sub_1229": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "convert_element_type_158": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "convert_element_type_157": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "clamp_max_53": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "clamp_min_80": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "add_4091": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "round_54": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "mul_2587": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "mul_2584": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "reciprocal_26": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "view_410": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "clamp_min_78": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "div_52": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "sub_1203": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "maximum_26": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "full_53": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "minimum_26": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "full_52": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "view_411": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "convert_element_type_156": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "clamp_max_52": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "clamp_min_79": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "sub_1209": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "round_53": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "div_53": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "convert_element_type_159": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "view_415": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "mul_2609": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "view_414": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56"], "mul_2419": ["triton_poi_fused__to_copy_mul_sub_view_4:57"], "sub_1144": ["triton_poi_fused__to_copy_mul_sub_view_4:57"], "convert_element_type_148": ["triton_poi_fused__to_copy_mul_sub_view_4:57"], "view_385": ["triton_poi_fused__to_copy_mul_sub_view_4:57"], "convert_element_type_149": ["triton_poi_fused__to_copy_mul_sub_view_4:57"], "view_387": ["triton_poi_fused__to_copy_mul_sub_view_4:57"], "view_386": ["triton_poi_fused__to_copy_mul_sub_view_4:57"], "mul_2515": ["triton_poi_fused__to_copy_mul_sub_view_4:58"], "sub_1189": ["triton_poi_fused__to_copy_mul_sub_view_4:58"], "convert_element_type_154": ["triton_poi_fused__to_copy_mul_sub_view_4:58"], "view_401": ["triton_poi_fused__to_copy_mul_sub_view_4:58"], "convert_element_type_155": ["triton_poi_fused__to_copy_mul_sub_view_4:58"], "view_403": ["triton_poi_fused__to_copy_mul_sub_view_4:58"], "view_402": ["triton_poi_fused__to_copy_mul_sub_view_4:58"], "mul_2614": ["triton_poi_fused__to_copy_mul_sub_view_4:59"], "sub_1233": ["triton_poi_fused__to_copy_mul_sub_view_4:59"], "convert_element_type_160": ["triton_poi_fused__to_copy_mul_sub_view_4:59"], "view_417": ["triton_poi_fused__to_copy_mul_sub_view_4:59"], "convert_element_type_161": ["triton_poi_fused__to_copy_mul_sub_view_4:59"], "view_419": ["triton_poi_fused__to_copy_mul_sub_view_4:59"], "view_418": ["triton_poi_fused__to_copy_mul_sub_view_4:59"], "_scaled_dot_product_efficient_attention_4": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:60", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:61", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:62", "at::_ops::_scaled_dot_product_efficient_attention::call:458"], "clone_34": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:60", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:61", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:62", "at::_ops::_scaled_dot_product_efficient_attention::call:458"], "permute_42": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:60", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:61", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:62", "at::_ops::_scaled_dot_product_efficient_attention::call:458"], "view_391": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:60", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:61", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:62", "at::_ops::_scaled_dot_product_efficient_attention::call:458"], "mul_2431": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:60", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:61", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:62", "at::_ops::_scaled_dot_product_efficient_attention::call:458"], "view_390": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:60", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:61", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:62", "at::_ops::_scaled_dot_product_efficient_attention::call:458"], "add_tensor_139": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:60", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:61", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:62", "at::_ops::_scaled_dot_product_efficient_attention::call:458"], "clone_35": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:60", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:61", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:62", "at::_ops::_scaled_dot_product_efficient_attention::call:458"], "permute_44": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:60", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:61", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:62", "at::_ops::_scaled_dot_product_efficient_attention::call:458"], "view_407": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:60", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:61", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:62", "at::_ops::_scaled_dot_product_efficient_attention::call:458"], "view_406": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:60", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:61", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:62", "at::_ops::_scaled_dot_product_efficient_attention::call:458"], "clone_36": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:60", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:61", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:62", "at::_ops::_scaled_dot_product_efficient_attention::call:458"], "permute_46": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:60", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:61", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:62", "at::_ops::_scaled_dot_product_efficient_attention::call:458"], "view_423": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:60", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:61", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:62", "at::_ops::_scaled_dot_product_efficient_attention::call:458"], "view_422": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:60", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:61", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:62", "at::_ops::_scaled_dot_product_efficient_attention::call:458"], "add_tensor_138": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:60", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:61", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:62", "at::_ops::_scaled_dot_product_efficient_attention::call:458"], "amax_27": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "view_424": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "permute_47": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "amin_27": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "sub_1277": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "convert_element_type_164": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "convert_element_type_163": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "clamp_max_55": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "clamp_min_83": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "add_4255": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "round_56": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "mul_2692": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "mul_2689": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "reciprocal_27": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "view_427": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "clamp_min_81": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "div_54": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "sub_1251": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "maximum_27": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "full_55": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "minimum_27": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "full_54": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "view_428": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "convert_element_type_162": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "clamp_max_54": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "clamp_min_82": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "sub_1257": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "round_55": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "div_55": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "convert_element_type_165": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "view_432": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "mul_2714": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "view_431": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63"], "mul_2719": ["triton_poi_fused__to_copy_mul_sub_view_4:64"], "sub_1281": ["triton_poi_fused__to_copy_mul_sub_view_4:64"], "convert_element_type_166": ["triton_poi_fused__to_copy_mul_sub_view_4:64"], "view_434": ["triton_poi_fused__to_copy_mul_sub_view_4:64"], "convert_element_type_167": ["triton_poi_fused__to_copy_mul_sub_view_4:64"], "view_436": ["triton_poi_fused__to_copy_mul_sub_view_4:64"], "view_435": ["triton_poi_fused__to_copy_mul_sub_view_4:64"], "add_4318": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "view_439": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "add_tensor_137": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "var_mean_9": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "clone_38": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "add_4324": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "mul_2736": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "mul_2735": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "sub_1287": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "rsqrt_9": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "add_4323": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "amax_28": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "amin_28": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "sub_1324": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "convert_element_type_170": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "convert_element_type_169": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "clamp_max_57": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "clamp_min_86": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "add_4411": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "round_58": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "mul_2787": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "mul_2784": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "reciprocal_28": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "view_442": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "clamp_min_84": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "div_56": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "sub_1298": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "maximum_28": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "full_57": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "minimum_28": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "full_56": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "view_443": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "convert_element_type_168": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "clamp_max_56": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "clamp_min_85": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "sub_1304": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "round_57": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "div_57": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "convert_element_type_171": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "view_447": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "mul_2809": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "view_446": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65"], "mul_2814": ["triton_poi_fused__to_copy_mul_sub_view_10:66"], "sub_1328": ["triton_poi_fused__to_copy_mul_sub_view_10:66"], "convert_element_type_172": ["triton_poi_fused__to_copy_mul_sub_view_10:66"], "view_449": ["triton_poi_fused__to_copy_mul_sub_view_10:66"], "convert_element_type_173": ["triton_poi_fused__to_copy_mul_sub_view_10:66"], "view_451": ["triton_poi_fused__to_copy_mul_sub_view_10:66"], "view_450": ["triton_poi_fused__to_copy_mul_sub_view_10:66"], "amax_29": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "mul_2828": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "mul_2826": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "view_454": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "add_tensor_136": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "add_4470": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "erf_6": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "mul_2827": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "amin_29": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "add_4553": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "round_60": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "mul_2877": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "mul_2874": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "reciprocal_29": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "view_457": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "clamp_min_87": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "div_58": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "sub_1341": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "maximum_29": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "full_59": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "minimum_29": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "full_58": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "view_458": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "convert_element_type_174": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "clamp_max_58": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "clamp_min_88": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "sub_1347": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "round_59": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "div_59": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "mul_2899": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "sub_1367": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "convert_element_type_176": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "convert_element_type_175": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "clamp_max_59": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "clamp_min_89": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "convert_element_type_177": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "view_462": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "view_461": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67"], "mul_2904": ["triton_poi_fused__to_copy_mul_sub_view_10:68"], "sub_1371": ["triton_poi_fused__to_copy_mul_sub_view_10:68"], "convert_element_type_178": ["triton_poi_fused__to_copy_mul_sub_view_10:68"], "view_464": ["triton_poi_fused__to_copy_mul_sub_view_10:68"], "convert_element_type_179": ["triton_poi_fused__to_copy_mul_sub_view_10:68"], "view_466": ["triton_poi_fused__to_copy_mul_sub_view_10:68"], "view_465": ["triton_poi_fused__to_copy_mul_sub_view_10:68"], "var_mean_10": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "clone_41": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "add_4616": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "view_469": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "add_tensor_135": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "add_4622": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "mul_2921": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "mul_2920": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "sub_1377": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "rsqrt_10": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "add_4621": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "amax_30": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "amin_30": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "amax_31": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "amin_31": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "amax_32": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "amin_32": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "sub_1414": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "convert_element_type_182": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "convert_element_type_181": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "clamp_max_61": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "clamp_min_92": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "add_4709": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "round_62": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "mul_2972": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "mul_2969": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "reciprocal_30": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "view_472": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "clamp_min_90": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "div_60": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "sub_1388": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "maximum_30": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "full_61": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "minimum_30": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "full_60": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "view_473": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "convert_element_type_180": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "clamp_max_60": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "clamp_min_91": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "sub_1394": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "round_61": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "div_61": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "convert_element_type_183": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "view_477": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "mul_2994": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "view_476": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "sub_1459": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "convert_element_type_188": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "convert_element_type_187": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "clamp_max_63": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "clamp_min_95": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "add_4861": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "round_64": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "mul_3068": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "mul_3065": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "reciprocal_31": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "view_488": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "clamp_min_93": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "div_62": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "sub_1433": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "maximum_31": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "full_63": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "minimum_31": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "full_62": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "view_489": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "convert_element_type_186": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "clamp_max_62": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "clamp_min_94": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "sub_1439": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "round_63": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "div_63": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "convert_element_type_189": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "view_493": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "mul_3090": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "view_492": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "sub_1503": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "convert_element_type_194": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "convert_element_type_193": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "clamp_max_65": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "clamp_min_98": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "add_5009": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "round_66": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "mul_3167": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "mul_3164": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "reciprocal_32": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "view_504": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "clamp_min_96": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "div_64": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "sub_1477": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "maximum_32": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "full_65": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "minimum_32": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "full_64": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "view_505": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "convert_element_type_192": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "clamp_max_64": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "clamp_min_97": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "sub_1483": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "round_65": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "div_65": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "convert_element_type_195": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "view_509": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "mul_3189": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "view_508": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69"], "mul_2999": ["triton_poi_fused__to_copy_mul_sub_view_4:70"], "sub_1418": ["triton_poi_fused__to_copy_mul_sub_view_4:70"], "convert_element_type_184": ["triton_poi_fused__to_copy_mul_sub_view_4:70"], "view_479": ["triton_poi_fused__to_copy_mul_sub_view_4:70"], "convert_element_type_185": ["triton_poi_fused__to_copy_mul_sub_view_4:70"], "view_481": ["triton_poi_fused__to_copy_mul_sub_view_4:70"], "view_480": ["triton_poi_fused__to_copy_mul_sub_view_4:70"], "mul_3095": ["triton_poi_fused__to_copy_mul_sub_view_4:71"], "sub_1463": ["triton_poi_fused__to_copy_mul_sub_view_4:71"], "convert_element_type_190": ["triton_poi_fused__to_copy_mul_sub_view_4:71"], "view_495": ["triton_poi_fused__to_copy_mul_sub_view_4:71"], "convert_element_type_191": ["triton_poi_fused__to_copy_mul_sub_view_4:71"], "view_497": ["triton_poi_fused__to_copy_mul_sub_view_4:71"], "view_496": ["triton_poi_fused__to_copy_mul_sub_view_4:71"], "mul_3194": ["triton_poi_fused__to_copy_mul_sub_view_4:72"], "sub_1507": ["triton_poi_fused__to_copy_mul_sub_view_4:72"], "convert_element_type_196": ["triton_poi_fused__to_copy_mul_sub_view_4:72"], "view_511": ["triton_poi_fused__to_copy_mul_sub_view_4:72"], "convert_element_type_197": ["triton_poi_fused__to_copy_mul_sub_view_4:72"], "view_513": ["triton_poi_fused__to_copy_mul_sub_view_4:72"], "view_512": ["triton_poi_fused__to_copy_mul_sub_view_4:72"], "_scaled_dot_product_efficient_attention_5": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:73", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:74", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:75", "at::_ops::_scaled_dot_product_efficient_attention::call:465"], "clone_42": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:73", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:74", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:75", "at::_ops::_scaled_dot_product_efficient_attention::call:465"], "permute_52": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:73", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:74", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:75", "at::_ops::_scaled_dot_product_efficient_attention::call:465"], "view_485": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:73", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:74", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:75", "at::_ops::_scaled_dot_product_efficient_attention::call:465"], "mul_3011": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:73", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:74", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:75", "at::_ops::_scaled_dot_product_efficient_attention::call:465"], "view_484": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:73", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:74", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:75", "at::_ops::_scaled_dot_product_efficient_attention::call:465"], "add_tensor_134": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:73", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:74", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:75", "at::_ops::_scaled_dot_product_efficient_attention::call:465"], "clone_43": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:73", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:74", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:75", "at::_ops::_scaled_dot_product_efficient_attention::call:465"], "permute_54": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:73", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:74", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:75", "at::_ops::_scaled_dot_product_efficient_attention::call:465"], "view_501": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:73", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:74", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:75", "at::_ops::_scaled_dot_product_efficient_attention::call:465"], "view_500": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:73", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:74", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:75", "at::_ops::_scaled_dot_product_efficient_attention::call:465"], "clone_44": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:73", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:74", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:75", "at::_ops::_scaled_dot_product_efficient_attention::call:465"], "permute_56": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:73", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:74", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:75", "at::_ops::_scaled_dot_product_efficient_attention::call:465"], "view_517": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:73", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:74", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:75", "at::_ops::_scaled_dot_product_efficient_attention::call:465"], "view_516": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:73", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:74", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:75", "at::_ops::_scaled_dot_product_efficient_attention::call:465"], "add_tensor_133": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:73", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:74", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:75", "at::_ops::_scaled_dot_product_efficient_attention::call:465"], "amax_33": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "view_518": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "permute_57": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "amin_33": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "sub_1551": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "convert_element_type_200": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "convert_element_type_199": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "clamp_max_67": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "clamp_min_101": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "add_5173": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "round_68": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "mul_3272": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "mul_3269": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "reciprocal_33": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "view_521": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "clamp_min_99": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "div_66": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "sub_1525": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "maximum_33": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "full_67": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "minimum_33": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "full_66": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "view_522": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "convert_element_type_198": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "clamp_max_66": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "clamp_min_100": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "sub_1531": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "round_67": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "div_67": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "convert_element_type_201": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "view_526": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "mul_3294": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "view_525": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76"], "mul_3299": ["triton_poi_fused__to_copy_mul_sub_view_4:77"], "sub_1555": ["triton_poi_fused__to_copy_mul_sub_view_4:77"], "convert_element_type_202": ["triton_poi_fused__to_copy_mul_sub_view_4:77"], "view_528": ["triton_poi_fused__to_copy_mul_sub_view_4:77"], "convert_element_type_203": ["triton_poi_fused__to_copy_mul_sub_view_4:77"], "view_530": ["triton_poi_fused__to_copy_mul_sub_view_4:77"], "view_529": ["triton_poi_fused__to_copy_mul_sub_view_4:77"], "add_5236": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "view_533": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "add_tensor_132": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "var_mean_11": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "clone_46": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "add_5242": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "mul_3316": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "mul_3315": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "sub_1561": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "rsqrt_11": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "add_5241": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "amax_34": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "amin_34": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "sub_1598": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "convert_element_type_206": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "convert_element_type_205": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "clamp_max_69": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "clamp_min_104": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "add_5329": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "round_70": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "mul_3367": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "mul_3364": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "reciprocal_34": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "view_536": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "clamp_min_102": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "div_68": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "sub_1572": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "maximum_34": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "full_69": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "minimum_34": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "full_68": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "view_537": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "convert_element_type_204": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "clamp_max_68": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "clamp_min_103": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "sub_1578": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "round_69": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "div_69": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "convert_element_type_207": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "view_541": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "mul_3389": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "view_540": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78"], "mul_3394": ["triton_poi_fused__to_copy_mul_sub_view_10:79"], "sub_1602": ["triton_poi_fused__to_copy_mul_sub_view_10:79"], "convert_element_type_208": ["triton_poi_fused__to_copy_mul_sub_view_10:79"], "view_543": ["triton_poi_fused__to_copy_mul_sub_view_10:79"], "convert_element_type_209": ["triton_poi_fused__to_copy_mul_sub_view_10:79"], "view_545": ["triton_poi_fused__to_copy_mul_sub_view_10:79"], "view_544": ["triton_poi_fused__to_copy_mul_sub_view_10:79"], "amax_35": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "mul_3408": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "mul_3406": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "view_548": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "add_tensor_131": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "add_5388": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "erf_7": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "mul_3407": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "amin_35": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "add_5471": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "round_72": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "mul_3457": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "mul_3454": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "reciprocal_35": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "view_551": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "clamp_min_105": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "div_70": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "sub_1615": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "maximum_35": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "full_71": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "minimum_35": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "full_70": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "view_552": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "convert_element_type_210": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "clamp_max_70": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "clamp_min_106": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "sub_1621": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "round_71": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "div_71": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "mul_3479": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "sub_1641": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "convert_element_type_212": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "convert_element_type_211": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "clamp_max_71": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "clamp_min_107": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "convert_element_type_213": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "view_556": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "view_555": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80"], "mul_3484": ["triton_poi_fused__to_copy_mul_sub_view_10:81"], "sub_1645": ["triton_poi_fused__to_copy_mul_sub_view_10:81"], "convert_element_type_214": ["triton_poi_fused__to_copy_mul_sub_view_10:81"], "view_558": ["triton_poi_fused__to_copy_mul_sub_view_10:81"], "convert_element_type_215": ["triton_poi_fused__to_copy_mul_sub_view_10:81"], "view_560": ["triton_poi_fused__to_copy_mul_sub_view_10:81"], "view_559": ["triton_poi_fused__to_copy_mul_sub_view_10:81"], "var_mean_12": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "clone_49": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "add_5534": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "view_563": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "add_tensor_130": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "add_5540": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "mul_3501": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "mul_3500": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "sub_1651": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "rsqrt_12": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "add_5539": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "amax_36": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "amin_36": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "amax_37": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "amin_37": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "amax_38": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "amin_38": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "sub_1688": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "convert_element_type_218": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "convert_element_type_217": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "clamp_max_73": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "clamp_min_110": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "add_5627": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "round_74": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "mul_3552": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "mul_3549": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "reciprocal_36": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "view_566": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "clamp_min_108": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "div_72": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "sub_1662": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "maximum_36": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "full_73": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "minimum_36": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "full_72": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "view_567": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "convert_element_type_216": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "clamp_max_72": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "clamp_min_109": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "sub_1668": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "round_73": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "div_73": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "convert_element_type_219": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "view_571": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "mul_3574": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "view_570": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "sub_1733": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "convert_element_type_224": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "convert_element_type_223": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "clamp_max_75": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "clamp_min_113": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "add_5779": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "round_76": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "mul_3648": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "mul_3645": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "reciprocal_37": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "view_582": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "clamp_min_111": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "div_74": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "sub_1707": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "maximum_37": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "full_75": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "minimum_37": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "full_74": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "view_583": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "convert_element_type_222": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "clamp_max_74": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "clamp_min_112": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "sub_1713": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "round_75": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "div_75": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "convert_element_type_225": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "view_587": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "mul_3670": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "view_586": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "sub_1777": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "convert_element_type_230": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "convert_element_type_229": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "clamp_max_77": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "clamp_min_116": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "add_5927": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "round_78": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "mul_3747": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "mul_3744": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "reciprocal_38": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "view_598": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "clamp_min_114": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "div_76": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "sub_1751": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "maximum_38": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "full_77": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "minimum_38": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "full_76": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "view_599": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "convert_element_type_228": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "clamp_max_76": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "clamp_min_115": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "sub_1757": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "round_77": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "div_77": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "convert_element_type_231": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "view_603": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "mul_3769": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "view_602": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82"], "mul_3579": ["triton_poi_fused__to_copy_mul_sub_view_4:83"], "sub_1692": ["triton_poi_fused__to_copy_mul_sub_view_4:83"], "convert_element_type_220": ["triton_poi_fused__to_copy_mul_sub_view_4:83"], "view_573": ["triton_poi_fused__to_copy_mul_sub_view_4:83"], "convert_element_type_221": ["triton_poi_fused__to_copy_mul_sub_view_4:83"], "view_575": ["triton_poi_fused__to_copy_mul_sub_view_4:83"], "view_574": ["triton_poi_fused__to_copy_mul_sub_view_4:83"], "mul_3675": ["triton_poi_fused__to_copy_mul_sub_view_4:84"], "sub_1737": ["triton_poi_fused__to_copy_mul_sub_view_4:84"], "convert_element_type_226": ["triton_poi_fused__to_copy_mul_sub_view_4:84"], "view_589": ["triton_poi_fused__to_copy_mul_sub_view_4:84"], "convert_element_type_227": ["triton_poi_fused__to_copy_mul_sub_view_4:84"], "view_591": ["triton_poi_fused__to_copy_mul_sub_view_4:84"], "view_590": ["triton_poi_fused__to_copy_mul_sub_view_4:84"], "mul_3774": ["triton_poi_fused__to_copy_mul_sub_view_4:85"], "sub_1781": ["triton_poi_fused__to_copy_mul_sub_view_4:85"], "convert_element_type_232": ["triton_poi_fused__to_copy_mul_sub_view_4:85"], "view_605": ["triton_poi_fused__to_copy_mul_sub_view_4:85"], "convert_element_type_233": ["triton_poi_fused__to_copy_mul_sub_view_4:85"], "view_607": ["triton_poi_fused__to_copy_mul_sub_view_4:85"], "view_606": ["triton_poi_fused__to_copy_mul_sub_view_4:85"], "_scaled_dot_product_efficient_attention_6": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:86", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:87", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:88", "at::_ops::_scaled_dot_product_efficient_attention::call:472"], "clone_50": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:86", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:87", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:88", "at::_ops::_scaled_dot_product_efficient_attention::call:472"], "permute_62": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:86", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:87", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:88", "at::_ops::_scaled_dot_product_efficient_attention::call:472"], "view_579": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:86", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:87", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:88", "at::_ops::_scaled_dot_product_efficient_attention::call:472"], "mul_3591": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:86", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:87", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:88", "at::_ops::_scaled_dot_product_efficient_attention::call:472"], "view_578": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:86", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:87", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:88", "at::_ops::_scaled_dot_product_efficient_attention::call:472"], "add_tensor_129": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:86", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:87", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:88", "at::_ops::_scaled_dot_product_efficient_attention::call:472"], "clone_51": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:86", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:87", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:88", "at::_ops::_scaled_dot_product_efficient_attention::call:472"], "permute_64": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:86", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:87", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:88", "at::_ops::_scaled_dot_product_efficient_attention::call:472"], "view_595": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:86", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:87", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:88", "at::_ops::_scaled_dot_product_efficient_attention::call:472"], "view_594": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:86", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:87", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:88", "at::_ops::_scaled_dot_product_efficient_attention::call:472"], "clone_52": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:86", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:87", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:88", "at::_ops::_scaled_dot_product_efficient_attention::call:472"], "permute_66": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:86", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:87", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:88", "at::_ops::_scaled_dot_product_efficient_attention::call:472"], "view_611": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:86", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:87", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:88", "at::_ops::_scaled_dot_product_efficient_attention::call:472"], "view_610": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:86", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:87", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:88", "at::_ops::_scaled_dot_product_efficient_attention::call:472"], "add_tensor_128": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:86", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:87", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:88", "at::_ops::_scaled_dot_product_efficient_attention::call:472"], "amax_39": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "view_612": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "permute_67": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "amin_39": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "sub_1825": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "convert_element_type_236": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "convert_element_type_235": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "clamp_max_79": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "clamp_min_119": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "add_6091": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "round_80": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "mul_3852": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "mul_3849": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "reciprocal_39": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "view_615": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "clamp_min_117": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "div_78": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "sub_1799": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "maximum_39": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "full_79": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "minimum_39": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "full_78": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "view_616": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "convert_element_type_234": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "clamp_max_78": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "clamp_min_118": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "sub_1805": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "round_79": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "div_79": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "convert_element_type_237": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "view_620": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "mul_3874": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "view_619": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89"], "mul_3879": ["triton_poi_fused__to_copy_mul_sub_view_4:90"], "sub_1829": ["triton_poi_fused__to_copy_mul_sub_view_4:90"], "convert_element_type_238": ["triton_poi_fused__to_copy_mul_sub_view_4:90"], "view_622": ["triton_poi_fused__to_copy_mul_sub_view_4:90"], "convert_element_type_239": ["triton_poi_fused__to_copy_mul_sub_view_4:90"], "view_624": ["triton_poi_fused__to_copy_mul_sub_view_4:90"], "view_623": ["triton_poi_fused__to_copy_mul_sub_view_4:90"], "add_6154": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "view_627": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "add_tensor_127": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "var_mean_13": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "clone_54": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "add_6160": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "mul_3896": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "mul_3895": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "sub_1835": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "rsqrt_13": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "add_6159": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "amax_40": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "amin_40": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "sub_1872": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "convert_element_type_242": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "convert_element_type_241": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "clamp_max_81": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "clamp_min_122": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "add_6247": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "round_82": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "mul_3947": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "mul_3944": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "reciprocal_40": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "view_630": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "clamp_min_120": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "div_80": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "sub_1846": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "maximum_40": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "full_81": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "minimum_40": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "full_80": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "view_631": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "convert_element_type_240": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "clamp_max_80": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "clamp_min_121": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "sub_1852": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "round_81": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "div_81": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "convert_element_type_243": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "view_635": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "mul_3969": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "view_634": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91"], "mul_3974": ["triton_poi_fused__to_copy_mul_sub_view_10:92"], "sub_1876": ["triton_poi_fused__to_copy_mul_sub_view_10:92"], "convert_element_type_244": ["triton_poi_fused__to_copy_mul_sub_view_10:92"], "view_637": ["triton_poi_fused__to_copy_mul_sub_view_10:92"], "convert_element_type_245": ["triton_poi_fused__to_copy_mul_sub_view_10:92"], "view_639": ["triton_poi_fused__to_copy_mul_sub_view_10:92"], "view_638": ["triton_poi_fused__to_copy_mul_sub_view_10:92"], "amax_41": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "mul_3988": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "mul_3986": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "view_642": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "add_tensor_126": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "add_6306": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "erf_8": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "mul_3987": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "amin_41": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "add_6389": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "round_84": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "mul_4037": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "mul_4034": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "reciprocal_41": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "view_645": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "clamp_min_123": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "div_82": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "sub_1889": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "maximum_41": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "full_83": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "minimum_41": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "full_82": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "view_646": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "convert_element_type_246": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "clamp_max_82": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "clamp_min_124": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "sub_1895": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "round_83": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "div_83": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "mul_4059": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "sub_1915": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "convert_element_type_248": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "convert_element_type_247": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "clamp_max_83": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "clamp_min_125": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "convert_element_type_249": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "view_650": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "view_649": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93"], "mul_4064": ["triton_poi_fused__to_copy_mul_sub_view_10:94"], "sub_1919": ["triton_poi_fused__to_copy_mul_sub_view_10:94"], "convert_element_type_250": ["triton_poi_fused__to_copy_mul_sub_view_10:94"], "view_652": ["triton_poi_fused__to_copy_mul_sub_view_10:94"], "convert_element_type_251": ["triton_poi_fused__to_copy_mul_sub_view_10:94"], "view_654": ["triton_poi_fused__to_copy_mul_sub_view_10:94"], "view_653": ["triton_poi_fused__to_copy_mul_sub_view_10:94"], "var_mean_14": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "clone_57": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "add_6452": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "view_657": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "add_tensor_125": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "add_6458": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "mul_4081": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "mul_4080": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "sub_1925": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "rsqrt_14": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "add_6457": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "amax_42": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "amin_42": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "amax_43": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "amin_43": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "amax_44": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "amin_44": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "sub_1962": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "convert_element_type_254": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "convert_element_type_253": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "clamp_max_85": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "clamp_min_128": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "add_6545": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "round_86": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "mul_4132": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "mul_4129": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "reciprocal_42": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "view_660": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "clamp_min_126": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "div_84": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "sub_1936": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "maximum_42": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "full_85": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "minimum_42": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "full_84": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "view_661": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "convert_element_type_252": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "clamp_max_84": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "clamp_min_127": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "sub_1942": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "round_85": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "div_85": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "convert_element_type_255": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "view_665": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "mul_4154": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "view_664": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "sub_2007": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "convert_element_type_260": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "convert_element_type_259": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "clamp_max_87": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "clamp_min_131": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "add_6697": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "round_88": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "mul_4228": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "mul_4225": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "reciprocal_43": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "view_676": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "clamp_min_129": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "div_86": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "sub_1981": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "maximum_43": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "full_87": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "minimum_43": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "full_86": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "view_677": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "convert_element_type_258": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "clamp_max_86": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "clamp_min_130": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "sub_1987": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "round_87": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "div_87": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "convert_element_type_261": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "view_681": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "mul_4250": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "view_680": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "sub_2051": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "convert_element_type_266": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "convert_element_type_265": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "clamp_max_89": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "clamp_min_134": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "add_6845": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "round_90": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "mul_4327": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "mul_4324": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "reciprocal_44": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "view_692": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "clamp_min_132": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "div_88": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "sub_2025": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "maximum_44": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "full_89": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "minimum_44": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "full_88": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "view_693": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "convert_element_type_264": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "clamp_max_88": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "clamp_min_133": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "sub_2031": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "round_89": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "div_89": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "convert_element_type_267": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "view_697": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "mul_4349": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "view_696": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95"], "mul_4159": ["triton_poi_fused__to_copy_mul_sub_view_4:96"], "sub_1966": ["triton_poi_fused__to_copy_mul_sub_view_4:96"], "convert_element_type_256": ["triton_poi_fused__to_copy_mul_sub_view_4:96"], "view_667": ["triton_poi_fused__to_copy_mul_sub_view_4:96"], "convert_element_type_257": ["triton_poi_fused__to_copy_mul_sub_view_4:96"], "view_669": ["triton_poi_fused__to_copy_mul_sub_view_4:96"], "view_668": ["triton_poi_fused__to_copy_mul_sub_view_4:96"], "mul_4255": ["triton_poi_fused__to_copy_mul_sub_view_4:97"], "sub_2011": ["triton_poi_fused__to_copy_mul_sub_view_4:97"], "convert_element_type_262": ["triton_poi_fused__to_copy_mul_sub_view_4:97"], "view_683": ["triton_poi_fused__to_copy_mul_sub_view_4:97"], "convert_element_type_263": ["triton_poi_fused__to_copy_mul_sub_view_4:97"], "view_685": ["triton_poi_fused__to_copy_mul_sub_view_4:97"], "view_684": ["triton_poi_fused__to_copy_mul_sub_view_4:97"], "mul_4354": ["triton_poi_fused__to_copy_mul_sub_view_4:98"], "sub_2055": ["triton_poi_fused__to_copy_mul_sub_view_4:98"], "convert_element_type_268": ["triton_poi_fused__to_copy_mul_sub_view_4:98"], "view_699": ["triton_poi_fused__to_copy_mul_sub_view_4:98"], "convert_element_type_269": ["triton_poi_fused__to_copy_mul_sub_view_4:98"], "view_701": ["triton_poi_fused__to_copy_mul_sub_view_4:98"], "view_700": ["triton_poi_fused__to_copy_mul_sub_view_4:98"], "_scaled_dot_product_efficient_attention_7": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:99", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:100", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:101", "at::_ops::_scaled_dot_product_efficient_attention::call:479"], "clone_58": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:99", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:100", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:101", "at::_ops::_scaled_dot_product_efficient_attention::call:479"], "permute_72": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:99", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:100", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:101", "at::_ops::_scaled_dot_product_efficient_attention::call:479"], "view_673": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:99", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:100", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:101", "at::_ops::_scaled_dot_product_efficient_attention::call:479"], "mul_4171": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:99", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:100", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:101", "at::_ops::_scaled_dot_product_efficient_attention::call:479"], "view_672": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:99", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:100", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:101", "at::_ops::_scaled_dot_product_efficient_attention::call:479"], "add_tensor_124": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:99", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:100", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:101", "at::_ops::_scaled_dot_product_efficient_attention::call:479"], "clone_59": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:99", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:100", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:101", "at::_ops::_scaled_dot_product_efficient_attention::call:479"], "permute_74": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:99", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:100", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:101", "at::_ops::_scaled_dot_product_efficient_attention::call:479"], "view_689": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:99", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:100", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:101", "at::_ops::_scaled_dot_product_efficient_attention::call:479"], "view_688": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:99", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:100", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:101", "at::_ops::_scaled_dot_product_efficient_attention::call:479"], "clone_60": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:99", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:100", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:101", "at::_ops::_scaled_dot_product_efficient_attention::call:479"], "permute_76": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:99", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:100", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:101", "at::_ops::_scaled_dot_product_efficient_attention::call:479"], "view_705": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:99", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:100", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:101", "at::_ops::_scaled_dot_product_efficient_attention::call:479"], "view_704": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:99", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:100", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:101", "at::_ops::_scaled_dot_product_efficient_attention::call:479"], "add_tensor_123": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:99", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:100", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:101", "at::_ops::_scaled_dot_product_efficient_attention::call:479"], "amax_45": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "view_706": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "permute_77": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "amin_45": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "sub_2099": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "convert_element_type_272": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "convert_element_type_271": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "clamp_max_91": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "clamp_min_137": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "add_7009": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "round_92": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "mul_4432": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "mul_4429": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "reciprocal_45": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "view_709": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "clamp_min_135": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "div_90": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "sub_2073": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "maximum_45": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "full_91": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "minimum_45": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "full_90": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "view_710": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "convert_element_type_270": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "clamp_max_90": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "clamp_min_136": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "sub_2079": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "round_91": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "div_91": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "convert_element_type_273": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "view_714": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "mul_4454": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "view_713": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102"], "mul_4459": ["triton_poi_fused__to_copy_mul_sub_view_4:103"], "sub_2103": ["triton_poi_fused__to_copy_mul_sub_view_4:103"], "convert_element_type_274": ["triton_poi_fused__to_copy_mul_sub_view_4:103"], "view_716": ["triton_poi_fused__to_copy_mul_sub_view_4:103"], "convert_element_type_275": ["triton_poi_fused__to_copy_mul_sub_view_4:103"], "view_718": ["triton_poi_fused__to_copy_mul_sub_view_4:103"], "view_717": ["triton_poi_fused__to_copy_mul_sub_view_4:103"], "add_7072": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "view_721": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "add_tensor_122": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "var_mean_15": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "clone_62": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "add_7078": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "mul_4476": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "mul_4475": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "sub_2109": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "rsqrt_15": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "add_7077": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "amax_46": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "amin_46": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "sub_2146": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "convert_element_type_278": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "convert_element_type_277": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "clamp_max_93": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "clamp_min_140": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "add_7165": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "round_94": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "mul_4527": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "mul_4524": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "reciprocal_46": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "view_724": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "clamp_min_138": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "div_92": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "sub_2120": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "maximum_46": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "full_93": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "minimum_46": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "full_92": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "view_725": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "convert_element_type_276": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "clamp_max_92": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "clamp_min_139": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "sub_2126": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "round_93": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "div_93": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "convert_element_type_279": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "view_729": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "mul_4549": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "view_728": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104"], "mul_4554": ["triton_poi_fused__to_copy_mul_sub_view_10:105"], "sub_2150": ["triton_poi_fused__to_copy_mul_sub_view_10:105"], "convert_element_type_280": ["triton_poi_fused__to_copy_mul_sub_view_10:105"], "view_731": ["triton_poi_fused__to_copy_mul_sub_view_10:105"], "convert_element_type_281": ["triton_poi_fused__to_copy_mul_sub_view_10:105"], "view_733": ["triton_poi_fused__to_copy_mul_sub_view_10:105"], "view_732": ["triton_poi_fused__to_copy_mul_sub_view_10:105"], "amax_47": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "mul_4568": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "mul_4566": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "view_736": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "add_tensor_121": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "add_7224": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "erf_9": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "mul_4567": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "amin_47": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "add_7307": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "round_96": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "mul_4617": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "mul_4614": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "reciprocal_47": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "view_739": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "clamp_min_141": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "div_94": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "sub_2163": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "maximum_47": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "full_95": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "minimum_47": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "full_94": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "view_740": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "convert_element_type_282": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "clamp_max_94": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "clamp_min_142": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "sub_2169": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "round_95": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "div_95": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "mul_4639": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "sub_2189": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "convert_element_type_284": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "convert_element_type_283": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "clamp_max_95": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "clamp_min_143": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "convert_element_type_285": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "view_744": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "view_743": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106"], "mul_4644": ["triton_poi_fused__to_copy_mul_sub_view_10:107"], "sub_2193": ["triton_poi_fused__to_copy_mul_sub_view_10:107"], "convert_element_type_286": ["triton_poi_fused__to_copy_mul_sub_view_10:107"], "view_746": ["triton_poi_fused__to_copy_mul_sub_view_10:107"], "convert_element_type_287": ["triton_poi_fused__to_copy_mul_sub_view_10:107"], "view_748": ["triton_poi_fused__to_copy_mul_sub_view_10:107"], "view_747": ["triton_poi_fused__to_copy_mul_sub_view_10:107"], "var_mean_16": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "clone_65": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "add_7370": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "view_751": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "add_tensor_120": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "add_7376": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "mul_4661": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "mul_4660": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "sub_2199": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "rsqrt_16": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "add_7375": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "amax_48": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "amin_48": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "amax_49": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "amin_49": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "amax_50": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "amin_50": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "sub_2236": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "convert_element_type_290": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "convert_element_type_289": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "clamp_max_97": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "clamp_min_146": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "add_7463": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "round_98": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "mul_4712": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "mul_4709": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "reciprocal_48": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "view_754": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "clamp_min_144": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "div_96": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "sub_2210": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "maximum_48": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "full_97": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "minimum_48": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "full_96": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "view_755": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "convert_element_type_288": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "clamp_max_96": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "clamp_min_145": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "sub_2216": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "round_97": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "div_97": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "convert_element_type_291": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "view_759": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "mul_4734": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "view_758": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "sub_2281": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "convert_element_type_296": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "convert_element_type_295": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "clamp_max_99": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "clamp_min_149": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "add_7615": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "round_100": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "mul_4808": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "mul_4805": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "reciprocal_49": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "view_770": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "clamp_min_147": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "div_98": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "sub_2255": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "maximum_49": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "full_99": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "minimum_49": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "full_98": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "view_771": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "convert_element_type_294": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "clamp_max_98": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "clamp_min_148": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "sub_2261": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "round_99": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "div_99": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "convert_element_type_297": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "view_775": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "mul_4830": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "view_774": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "sub_2325": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "convert_element_type_302": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "convert_element_type_301": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "clamp_max_101": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "clamp_min_152": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "add_7763": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "round_102": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "mul_4907": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "mul_4904": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "reciprocal_50": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "view_786": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "clamp_min_150": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "div_100": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "sub_2299": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "maximum_50": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "full_101": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "minimum_50": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "full_100": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "view_787": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "convert_element_type_300": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "clamp_max_100": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "clamp_min_151": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "sub_2305": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "round_101": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "div_101": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "convert_element_type_303": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "view_791": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "mul_4929": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "view_790": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108"], "mul_4739": ["triton_poi_fused__to_copy_mul_sub_view_4:109"], "sub_2240": ["triton_poi_fused__to_copy_mul_sub_view_4:109"], "convert_element_type_292": ["triton_poi_fused__to_copy_mul_sub_view_4:109"], "view_761": ["triton_poi_fused__to_copy_mul_sub_view_4:109"], "convert_element_type_293": ["triton_poi_fused__to_copy_mul_sub_view_4:109"], "view_763": ["triton_poi_fused__to_copy_mul_sub_view_4:109"], "view_762": ["triton_poi_fused__to_copy_mul_sub_view_4:109"], "mul_4835": ["triton_poi_fused__to_copy_mul_sub_view_4:110"], "sub_2285": ["triton_poi_fused__to_copy_mul_sub_view_4:110"], "convert_element_type_298": ["triton_poi_fused__to_copy_mul_sub_view_4:110"], "view_777": ["triton_poi_fused__to_copy_mul_sub_view_4:110"], "convert_element_type_299": ["triton_poi_fused__to_copy_mul_sub_view_4:110"], "view_779": ["triton_poi_fused__to_copy_mul_sub_view_4:110"], "view_778": ["triton_poi_fused__to_copy_mul_sub_view_4:110"], "mul_4934": ["triton_poi_fused__to_copy_mul_sub_view_4:111"], "sub_2329": ["triton_poi_fused__to_copy_mul_sub_view_4:111"], "convert_element_type_304": ["triton_poi_fused__to_copy_mul_sub_view_4:111"], "view_793": ["triton_poi_fused__to_copy_mul_sub_view_4:111"], "convert_element_type_305": ["triton_poi_fused__to_copy_mul_sub_view_4:111"], "view_795": ["triton_poi_fused__to_copy_mul_sub_view_4:111"], "view_794": ["triton_poi_fused__to_copy_mul_sub_view_4:111"], "_scaled_dot_product_efficient_attention_8": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:112", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:113", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:114", "at::_ops::_scaled_dot_product_efficient_attention::call:486"], "clone_66": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:112", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:113", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:114", "at::_ops::_scaled_dot_product_efficient_attention::call:486"], "permute_82": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:112", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:113", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:114", "at::_ops::_scaled_dot_product_efficient_attention::call:486"], "view_767": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:112", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:113", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:114", "at::_ops::_scaled_dot_product_efficient_attention::call:486"], "mul_4751": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:112", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:113", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:114", "at::_ops::_scaled_dot_product_efficient_attention::call:486"], "view_766": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:112", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:113", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:114", "at::_ops::_scaled_dot_product_efficient_attention::call:486"], "add_tensor_119": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:112", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:113", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:114", "at::_ops::_scaled_dot_product_efficient_attention::call:486"], "clone_67": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:112", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:113", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:114", "at::_ops::_scaled_dot_product_efficient_attention::call:486"], "permute_84": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:112", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:113", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:114", "at::_ops::_scaled_dot_product_efficient_attention::call:486"], "view_783": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:112", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:113", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:114", "at::_ops::_scaled_dot_product_efficient_attention::call:486"], "view_782": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:112", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:113", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:114", "at::_ops::_scaled_dot_product_efficient_attention::call:486"], "clone_68": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:112", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:113", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:114", "at::_ops::_scaled_dot_product_efficient_attention::call:486"], "permute_86": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:112", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:113", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:114", "at::_ops::_scaled_dot_product_efficient_attention::call:486"], "view_799": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:112", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:113", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:114", "at::_ops::_scaled_dot_product_efficient_attention::call:486"], "view_798": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:112", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:113", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:114", "at::_ops::_scaled_dot_product_efficient_attention::call:486"], "add_tensor_118": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:112", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:113", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:114", "at::_ops::_scaled_dot_product_efficient_attention::call:486"], "amax_51": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "view_800": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "permute_87": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "amin_51": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "sub_2373": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "convert_element_type_308": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "convert_element_type_307": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "clamp_max_103": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "clamp_min_155": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "add_7927": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "round_104": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "mul_5012": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "mul_5009": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "reciprocal_51": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "view_803": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "clamp_min_153": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "div_102": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "sub_2347": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "maximum_51": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "full_103": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "minimum_51": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "full_102": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "view_804": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "convert_element_type_306": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "clamp_max_102": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "clamp_min_154": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "sub_2353": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "round_103": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "div_103": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "convert_element_type_309": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "view_808": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "mul_5034": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "view_807": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115"], "mul_5039": ["triton_poi_fused__to_copy_mul_sub_view_4:116"], "sub_2377": ["triton_poi_fused__to_copy_mul_sub_view_4:116"], "convert_element_type_310": ["triton_poi_fused__to_copy_mul_sub_view_4:116"], "view_810": ["triton_poi_fused__to_copy_mul_sub_view_4:116"], "convert_element_type_311": ["triton_poi_fused__to_copy_mul_sub_view_4:116"], "view_812": ["triton_poi_fused__to_copy_mul_sub_view_4:116"], "view_811": ["triton_poi_fused__to_copy_mul_sub_view_4:116"], "add_7990": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "view_815": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "add_tensor_117": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "var_mean_17": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "clone_70": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "add_7996": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "mul_5056": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "mul_5055": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "sub_2383": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "rsqrt_17": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "add_7995": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "amax_52": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "amin_52": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "sub_2420": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "convert_element_type_314": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "convert_element_type_313": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "clamp_max_105": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "clamp_min_158": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "add_8083": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "round_106": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "mul_5107": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "mul_5104": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "reciprocal_52": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "view_818": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "clamp_min_156": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "div_104": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "sub_2394": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "maximum_52": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "full_105": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "minimum_52": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "full_104": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "view_819": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "convert_element_type_312": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "clamp_max_104": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "clamp_min_157": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "sub_2400": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "round_105": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "div_105": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "convert_element_type_315": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "view_823": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "mul_5129": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "view_822": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117"], "mul_5134": ["triton_poi_fused__to_copy_mul_sub_view_10:118"], "sub_2424": ["triton_poi_fused__to_copy_mul_sub_view_10:118"], "convert_element_type_316": ["triton_poi_fused__to_copy_mul_sub_view_10:118"], "view_825": ["triton_poi_fused__to_copy_mul_sub_view_10:118"], "convert_element_type_317": ["triton_poi_fused__to_copy_mul_sub_view_10:118"], "view_827": ["triton_poi_fused__to_copy_mul_sub_view_10:118"], "view_826": ["triton_poi_fused__to_copy_mul_sub_view_10:118"], "amax_53": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "mul_5148": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "mul_5146": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "view_830": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "add_tensor_116": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "add_8142": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "erf_10": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "mul_5147": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "amin_53": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "add_8225": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "round_108": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "mul_5197": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "mul_5194": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "reciprocal_53": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "view_833": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "clamp_min_159": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "div_106": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "sub_2437": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "maximum_53": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "full_107": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "minimum_53": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "full_106": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "view_834": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "convert_element_type_318": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "clamp_max_106": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "clamp_min_160": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "sub_2443": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "round_107": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "div_107": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "mul_5219": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "sub_2463": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "convert_element_type_320": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "convert_element_type_319": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "clamp_max_107": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "clamp_min_161": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "convert_element_type_321": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "view_838": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "view_837": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119"], "mul_5224": ["triton_poi_fused__to_copy_mul_sub_view_10:120"], "sub_2467": ["triton_poi_fused__to_copy_mul_sub_view_10:120"], "convert_element_type_322": ["triton_poi_fused__to_copy_mul_sub_view_10:120"], "view_840": ["triton_poi_fused__to_copy_mul_sub_view_10:120"], "convert_element_type_323": ["triton_poi_fused__to_copy_mul_sub_view_10:120"], "view_842": ["triton_poi_fused__to_copy_mul_sub_view_10:120"], "view_841": ["triton_poi_fused__to_copy_mul_sub_view_10:120"], "var_mean_18": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "clone_73": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "add_8288": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "view_845": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "add_tensor_115": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "add_8294": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "mul_5241": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "mul_5240": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "sub_2473": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "rsqrt_18": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "add_8293": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "amax_54": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "amin_54": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "amax_55": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "amin_55": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "amax_56": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "amin_56": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "sub_2510": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "convert_element_type_326": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "convert_element_type_325": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "clamp_max_109": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "clamp_min_164": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "add_8381": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "round_110": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "mul_5292": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "mul_5289": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "reciprocal_54": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "view_848": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "clamp_min_162": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "div_108": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "sub_2484": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "maximum_54": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "full_109": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "minimum_54": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "full_108": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "view_849": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "convert_element_type_324": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "clamp_max_108": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "clamp_min_163": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "sub_2490": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "round_109": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "div_109": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "convert_element_type_327": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "view_853": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "mul_5314": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "view_852": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "sub_2555": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "convert_element_type_332": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "convert_element_type_331": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "clamp_max_111": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "clamp_min_167": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "add_8533": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "round_112": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "mul_5388": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "mul_5385": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "reciprocal_55": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "view_864": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "clamp_min_165": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "div_110": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "sub_2529": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "maximum_55": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "full_111": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "minimum_55": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "full_110": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "view_865": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "convert_element_type_330": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "clamp_max_110": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "clamp_min_166": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "sub_2535": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "round_111": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "div_111": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "convert_element_type_333": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "view_869": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "mul_5410": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "view_868": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "sub_2599": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "convert_element_type_338": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "convert_element_type_337": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "clamp_max_113": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "clamp_min_170": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "add_8681": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "round_114": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "mul_5487": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "mul_5484": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "reciprocal_56": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "view_880": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "clamp_min_168": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "div_112": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "sub_2573": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "maximum_56": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "full_113": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "minimum_56": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "full_112": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "view_881": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "convert_element_type_336": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "clamp_max_112": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "clamp_min_169": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "sub_2579": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "round_113": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "div_113": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "convert_element_type_339": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "view_885": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "mul_5509": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "view_884": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121"], "mul_5319": ["triton_poi_fused__to_copy_mul_sub_view_4:122"], "sub_2514": ["triton_poi_fused__to_copy_mul_sub_view_4:122"], "convert_element_type_328": ["triton_poi_fused__to_copy_mul_sub_view_4:122"], "view_855": ["triton_poi_fused__to_copy_mul_sub_view_4:122"], "convert_element_type_329": ["triton_poi_fused__to_copy_mul_sub_view_4:122"], "view_857": ["triton_poi_fused__to_copy_mul_sub_view_4:122"], "view_856": ["triton_poi_fused__to_copy_mul_sub_view_4:122"], "mul_5415": ["triton_poi_fused__to_copy_mul_sub_view_4:123"], "sub_2559": ["triton_poi_fused__to_copy_mul_sub_view_4:123"], "convert_element_type_334": ["triton_poi_fused__to_copy_mul_sub_view_4:123"], "view_871": ["triton_poi_fused__to_copy_mul_sub_view_4:123"], "convert_element_type_335": ["triton_poi_fused__to_copy_mul_sub_view_4:123"], "view_873": ["triton_poi_fused__to_copy_mul_sub_view_4:123"], "view_872": ["triton_poi_fused__to_copy_mul_sub_view_4:123"], "mul_5514": ["triton_poi_fused__to_copy_mul_sub_view_4:124"], "sub_2603": ["triton_poi_fused__to_copy_mul_sub_view_4:124"], "convert_element_type_340": ["triton_poi_fused__to_copy_mul_sub_view_4:124"], "view_887": ["triton_poi_fused__to_copy_mul_sub_view_4:124"], "convert_element_type_341": ["triton_poi_fused__to_copy_mul_sub_view_4:124"], "view_889": ["triton_poi_fused__to_copy_mul_sub_view_4:124"], "view_888": ["triton_poi_fused__to_copy_mul_sub_view_4:124"], "_scaled_dot_product_efficient_attention_9": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:125", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:126", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:127", "at::_ops::_scaled_dot_product_efficient_attention::call:493"], "clone_74": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:125", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:126", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:127", "at::_ops::_scaled_dot_product_efficient_attention::call:493"], "permute_92": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:125", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:126", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:127", "at::_ops::_scaled_dot_product_efficient_attention::call:493"], "view_861": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:125", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:126", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:127", "at::_ops::_scaled_dot_product_efficient_attention::call:493"], "mul_5331": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:125", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:126", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:127", "at::_ops::_scaled_dot_product_efficient_attention::call:493"], "view_860": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:125", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:126", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:127", "at::_ops::_scaled_dot_product_efficient_attention::call:493"], "add_tensor_114": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:125", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:126", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:127", "at::_ops::_scaled_dot_product_efficient_attention::call:493"], "clone_75": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:125", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:126", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:127", "at::_ops::_scaled_dot_product_efficient_attention::call:493"], "permute_94": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:125", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:126", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:127", "at::_ops::_scaled_dot_product_efficient_attention::call:493"], "view_877": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:125", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:126", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:127", "at::_ops::_scaled_dot_product_efficient_attention::call:493"], "view_876": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:125", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:126", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:127", "at::_ops::_scaled_dot_product_efficient_attention::call:493"], "clone_76": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:125", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:126", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:127", "at::_ops::_scaled_dot_product_efficient_attention::call:493"], "permute_96": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:125", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:126", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:127", "at::_ops::_scaled_dot_product_efficient_attention::call:493"], "view_893": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:125", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:126", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:127", "at::_ops::_scaled_dot_product_efficient_attention::call:493"], "view_892": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:125", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:126", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:127", "at::_ops::_scaled_dot_product_efficient_attention::call:493"], "add_tensor_113": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:125", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:126", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:127", "at::_ops::_scaled_dot_product_efficient_attention::call:493"], "amax_57": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "view_894": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "permute_97": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "amin_57": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "sub_2647": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "convert_element_type_344": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "convert_element_type_343": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "clamp_max_115": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "clamp_min_173": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "add_8845": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "round_116": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "mul_5592": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "mul_5589": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "reciprocal_57": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "view_897": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "clamp_min_171": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "div_114": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "sub_2621": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "maximum_57": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "full_115": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "minimum_57": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "full_114": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "view_898": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "convert_element_type_342": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "clamp_max_114": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "clamp_min_172": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "sub_2627": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "round_115": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "div_115": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "convert_element_type_345": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "view_902": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "mul_5614": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "view_901": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128"], "mul_5619": ["triton_poi_fused__to_copy_mul_sub_view_4:129"], "sub_2651": ["triton_poi_fused__to_copy_mul_sub_view_4:129"], "convert_element_type_346": ["triton_poi_fused__to_copy_mul_sub_view_4:129"], "view_904": ["triton_poi_fused__to_copy_mul_sub_view_4:129"], "convert_element_type_347": ["triton_poi_fused__to_copy_mul_sub_view_4:129"], "view_906": ["triton_poi_fused__to_copy_mul_sub_view_4:129"], "view_905": ["triton_poi_fused__to_copy_mul_sub_view_4:129"], "add_8908": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "view_909": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "add_tensor_112": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "var_mean_19": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "clone_78": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "add_8914": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "mul_5636": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "mul_5635": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "sub_2657": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "rsqrt_19": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "add_8913": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "amax_58": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "amin_58": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "sub_2694": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "convert_element_type_350": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "convert_element_type_349": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "clamp_max_117": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "clamp_min_176": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "add_9001": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "round_118": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "mul_5687": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "mul_5684": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "reciprocal_58": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "view_912": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "clamp_min_174": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "div_116": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "sub_2668": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "maximum_58": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "full_117": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "minimum_58": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "full_116": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "view_913": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "convert_element_type_348": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "clamp_max_116": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "clamp_min_175": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "sub_2674": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "round_117": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "div_117": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "convert_element_type_351": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "view_917": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "mul_5709": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "view_916": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130"], "mul_5714": ["triton_poi_fused__to_copy_mul_sub_view_10:131"], "sub_2698": ["triton_poi_fused__to_copy_mul_sub_view_10:131"], "convert_element_type_352": ["triton_poi_fused__to_copy_mul_sub_view_10:131"], "view_919": ["triton_poi_fused__to_copy_mul_sub_view_10:131"], "convert_element_type_353": ["triton_poi_fused__to_copy_mul_sub_view_10:131"], "view_921": ["triton_poi_fused__to_copy_mul_sub_view_10:131"], "view_920": ["triton_poi_fused__to_copy_mul_sub_view_10:131"], "amax_59": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "mul_5728": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "mul_5726": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "view_924": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "add_tensor_111": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "add_9060": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "erf_11": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "mul_5727": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "amin_59": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "add_9143": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "round_120": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "mul_5777": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "mul_5774": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "reciprocal_59": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "view_927": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "clamp_min_177": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "div_118": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "sub_2711": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "maximum_59": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "full_119": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "minimum_59": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "full_118": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "view_928": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "convert_element_type_354": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "clamp_max_118": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "clamp_min_178": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "sub_2717": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "round_119": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "div_119": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "mul_5799": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "sub_2737": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "convert_element_type_356": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "convert_element_type_355": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "clamp_max_119": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "clamp_min_179": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "convert_element_type_357": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "view_932": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "view_931": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132"], "mul_5804": ["triton_poi_fused__to_copy_mul_sub_view_10:133"], "sub_2741": ["triton_poi_fused__to_copy_mul_sub_view_10:133"], "convert_element_type_358": ["triton_poi_fused__to_copy_mul_sub_view_10:133"], "view_934": ["triton_poi_fused__to_copy_mul_sub_view_10:133"], "convert_element_type_359": ["triton_poi_fused__to_copy_mul_sub_view_10:133"], "view_936": ["triton_poi_fused__to_copy_mul_sub_view_10:133"], "view_935": ["triton_poi_fused__to_copy_mul_sub_view_10:133"], "var_mean_20": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "clone_81": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "add_9206": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "view_939": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "add_tensor_110": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "add_9212": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "mul_5821": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "mul_5820": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "sub_2747": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "rsqrt_20": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "add_9211": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "amax_60": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "amin_60": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "amax_61": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "amin_61": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "amax_62": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "amin_62": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "sub_2784": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "convert_element_type_362": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "convert_element_type_361": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "clamp_max_121": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "clamp_min_182": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "add_9299": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "round_122": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "mul_5872": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "mul_5869": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "reciprocal_60": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "view_942": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "clamp_min_180": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "div_120": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "sub_2758": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "maximum_60": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "full_121": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "minimum_60": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "full_120": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "view_943": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "convert_element_type_360": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "clamp_max_120": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "clamp_min_181": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "sub_2764": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "round_121": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "div_121": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "convert_element_type_363": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "view_947": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "mul_5894": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "view_946": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "sub_2829": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "convert_element_type_368": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "convert_element_type_367": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "clamp_max_123": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "clamp_min_185": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "add_9451": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "round_124": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "mul_5968": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "mul_5965": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "reciprocal_61": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "view_958": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "clamp_min_183": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "div_122": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "sub_2803": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "maximum_61": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "full_123": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "minimum_61": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "full_122": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "view_959": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "convert_element_type_366": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "clamp_max_122": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "clamp_min_184": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "sub_2809": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "round_123": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "div_123": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "convert_element_type_369": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "view_963": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "mul_5990": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "view_962": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "sub_2873": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "convert_element_type_374": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "convert_element_type_373": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "clamp_max_125": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "clamp_min_188": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "add_9599": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "round_126": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "mul_6067": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "mul_6064": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "reciprocal_62": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "view_974": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "clamp_min_186": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "div_124": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "sub_2847": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "maximum_62": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "full_125": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "minimum_62": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "full_124": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "view_975": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "convert_element_type_372": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "clamp_max_124": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "clamp_min_187": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "sub_2853": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "round_125": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "div_125": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "convert_element_type_375": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "view_979": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "mul_6089": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "view_978": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134"], "mul_5899": ["triton_poi_fused__to_copy_mul_sub_view_4:135"], "sub_2788": ["triton_poi_fused__to_copy_mul_sub_view_4:135"], "convert_element_type_364": ["triton_poi_fused__to_copy_mul_sub_view_4:135"], "view_949": ["triton_poi_fused__to_copy_mul_sub_view_4:135"], "convert_element_type_365": ["triton_poi_fused__to_copy_mul_sub_view_4:135"], "view_951": ["triton_poi_fused__to_copy_mul_sub_view_4:135"], "view_950": ["triton_poi_fused__to_copy_mul_sub_view_4:135"], "mul_5995": ["triton_poi_fused__to_copy_mul_sub_view_4:136"], "sub_2833": ["triton_poi_fused__to_copy_mul_sub_view_4:136"], "convert_element_type_370": ["triton_poi_fused__to_copy_mul_sub_view_4:136"], "view_965": ["triton_poi_fused__to_copy_mul_sub_view_4:136"], "convert_element_type_371": ["triton_poi_fused__to_copy_mul_sub_view_4:136"], "view_967": ["triton_poi_fused__to_copy_mul_sub_view_4:136"], "view_966": ["triton_poi_fused__to_copy_mul_sub_view_4:136"], "mul_6094": ["triton_poi_fused__to_copy_mul_sub_view_4:137"], "sub_2877": ["triton_poi_fused__to_copy_mul_sub_view_4:137"], "convert_element_type_376": ["triton_poi_fused__to_copy_mul_sub_view_4:137"], "view_981": ["triton_poi_fused__to_copy_mul_sub_view_4:137"], "convert_element_type_377": ["triton_poi_fused__to_copy_mul_sub_view_4:137"], "view_983": ["triton_poi_fused__to_copy_mul_sub_view_4:137"], "view_982": ["triton_poi_fused__to_copy_mul_sub_view_4:137"], "_scaled_dot_product_efficient_attention_10": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:138", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:139", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:140", "at::_ops::_scaled_dot_product_efficient_attention::call:500"], "clone_82": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:138", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:139", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:140", "at::_ops::_scaled_dot_product_efficient_attention::call:500"], "permute_102": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:138", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:139", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:140", "at::_ops::_scaled_dot_product_efficient_attention::call:500"], "view_955": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:138", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:139", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:140", "at::_ops::_scaled_dot_product_efficient_attention::call:500"], "mul_5911": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:138", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:139", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:140", "at::_ops::_scaled_dot_product_efficient_attention::call:500"], "view_954": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:138", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:139", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:140", "at::_ops::_scaled_dot_product_efficient_attention::call:500"], "add_tensor_109": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:138", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:139", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:140", "at::_ops::_scaled_dot_product_efficient_attention::call:500"], "clone_83": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:138", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:139", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:140", "at::_ops::_scaled_dot_product_efficient_attention::call:500"], "permute_104": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:138", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:139", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:140", "at::_ops::_scaled_dot_product_efficient_attention::call:500"], "view_971": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:138", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:139", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:140", "at::_ops::_scaled_dot_product_efficient_attention::call:500"], "view_970": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:138", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:139", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:140", "at::_ops::_scaled_dot_product_efficient_attention::call:500"], "clone_84": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:138", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:139", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:140", "at::_ops::_scaled_dot_product_efficient_attention::call:500"], "permute_106": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:138", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:139", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:140", "at::_ops::_scaled_dot_product_efficient_attention::call:500"], "view_987": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:138", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:139", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:140", "at::_ops::_scaled_dot_product_efficient_attention::call:500"], "view_986": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:138", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:139", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:140", "at::_ops::_scaled_dot_product_efficient_attention::call:500"], "add_tensor_108": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:138", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:139", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:140", "at::_ops::_scaled_dot_product_efficient_attention::call:500"], "amax_63": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "view_988": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "permute_107": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "amin_63": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "sub_2921": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "convert_element_type_380": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "convert_element_type_379": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "clamp_max_127": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "clamp_min_191": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "add_9763": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "round_128": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "mul_6172": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "mul_6169": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "reciprocal_63": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "view_991": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "clamp_min_189": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "div_126": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "sub_2895": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "maximum_63": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "full_127": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "minimum_63": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "full_126": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "view_992": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "convert_element_type_378": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "clamp_max_126": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "clamp_min_190": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "sub_2901": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "round_127": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "div_127": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "convert_element_type_381": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "view_996": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "mul_6194": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "view_995": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141"], "mul_6199": ["triton_poi_fused__to_copy_mul_sub_view_4:142"], "sub_2925": ["triton_poi_fused__to_copy_mul_sub_view_4:142"], "convert_element_type_382": ["triton_poi_fused__to_copy_mul_sub_view_4:142"], "view_998": ["triton_poi_fused__to_copy_mul_sub_view_4:142"], "convert_element_type_383": ["triton_poi_fused__to_copy_mul_sub_view_4:142"], "view_1000": ["triton_poi_fused__to_copy_mul_sub_view_4:142"], "view_999": ["triton_poi_fused__to_copy_mul_sub_view_4:142"], "add_9826": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "view_1003": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "add_tensor_107": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "var_mean_21": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "clone_86": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "add_9832": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "mul_6216": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "mul_6215": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "sub_2931": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "rsqrt_21": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "add_9831": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "amax_64": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "amin_64": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "sub_2968": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "convert_element_type_386": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "convert_element_type_385": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "clamp_max_129": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "clamp_min_194": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "add_9919": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "round_130": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "mul_6267": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "mul_6264": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "reciprocal_64": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "view_1006": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "clamp_min_192": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "div_128": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "sub_2942": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "maximum_64": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "full_129": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "minimum_64": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "full_128": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "view_1007": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "convert_element_type_384": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "clamp_max_128": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "clamp_min_193": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "sub_2948": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "round_129": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "div_129": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "convert_element_type_387": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "view_1011": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "mul_6289": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "view_1010": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143"], "mul_6294": ["triton_poi_fused__to_copy_mul_sub_view_10:144"], "sub_2972": ["triton_poi_fused__to_copy_mul_sub_view_10:144"], "convert_element_type_388": ["triton_poi_fused__to_copy_mul_sub_view_10:144"], "view_1013": ["triton_poi_fused__to_copy_mul_sub_view_10:144"], "convert_element_type_389": ["triton_poi_fused__to_copy_mul_sub_view_10:144"], "view_1015": ["triton_poi_fused__to_copy_mul_sub_view_10:144"], "view_1014": ["triton_poi_fused__to_copy_mul_sub_view_10:144"], "amax_65": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "mul_6308": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "mul_6306": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "view_1018": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "add_tensor_106": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "add_9978": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "erf_12": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "mul_6307": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "amin_65": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "add_10061": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "round_132": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "mul_6357": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "mul_6354": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "reciprocal_65": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "view_1021": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "clamp_min_195": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "div_130": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "sub_2985": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "maximum_65": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "full_131": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "minimum_65": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "full_130": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "view_1022": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "convert_element_type_390": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "clamp_max_130": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "clamp_min_196": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "sub_2991": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "round_131": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "div_131": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "mul_6379": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "sub_3011": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "convert_element_type_392": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "convert_element_type_391": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "clamp_max_131": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "clamp_min_197": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "convert_element_type_393": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "view_1026": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "view_1025": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145"], "mul_6384": ["triton_poi_fused__to_copy_mul_sub_view_10:146"], "sub_3015": ["triton_poi_fused__to_copy_mul_sub_view_10:146"], "convert_element_type_394": ["triton_poi_fused__to_copy_mul_sub_view_10:146"], "view_1028": ["triton_poi_fused__to_copy_mul_sub_view_10:146"], "convert_element_type_395": ["triton_poi_fused__to_copy_mul_sub_view_10:146"], "view_1030": ["triton_poi_fused__to_copy_mul_sub_view_10:146"], "view_1029": ["triton_poi_fused__to_copy_mul_sub_view_10:146"], "var_mean_22": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "clone_89": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "add_10124": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "view_1033": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "add_tensor_105": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "add_10130": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "mul_6401": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "mul_6400": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "sub_3021": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "rsqrt_22": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "add_10129": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "amax_66": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "amin_66": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "amax_67": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "amin_67": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "amax_68": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "amin_68": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "sub_3058": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "convert_element_type_398": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "convert_element_type_397": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "clamp_max_133": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "clamp_min_200": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "add_10217": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "round_134": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "mul_6452": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "mul_6449": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "reciprocal_66": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "view_1036": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "clamp_min_198": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "div_132": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "sub_3032": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "maximum_66": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "full_133": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "minimum_66": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "full_132": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "view_1037": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "convert_element_type_396": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "clamp_max_132": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "clamp_min_199": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "sub_3038": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "round_133": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "div_133": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "convert_element_type_399": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "view_1041": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "mul_6474": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "view_1040": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "sub_3103": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "convert_element_type_404": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "convert_element_type_403": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "clamp_max_135": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "clamp_min_203": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "add_10369": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "round_136": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "mul_6548": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "mul_6545": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "reciprocal_67": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "view_1052": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "clamp_min_201": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "div_134": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "sub_3077": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "maximum_67": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "full_135": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "minimum_67": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "full_134": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "view_1053": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "convert_element_type_402": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "clamp_max_134": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "clamp_min_202": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "sub_3083": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "round_135": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "div_135": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "convert_element_type_405": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "view_1057": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "mul_6570": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "view_1056": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "sub_3147": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "convert_element_type_410": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "convert_element_type_409": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "clamp_max_137": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "clamp_min_206": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "add_10517": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "round_138": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "mul_6647": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "mul_6644": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "reciprocal_68": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "view_1068": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "clamp_min_204": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "div_136": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "sub_3121": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "maximum_68": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "full_137": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "minimum_68": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "full_136": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "view_1069": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "convert_element_type_408": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "clamp_max_136": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "clamp_min_205": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "sub_3127": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "round_137": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "div_137": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "convert_element_type_411": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "view_1073": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "mul_6669": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "view_1072": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147"], "mul_6479": ["triton_poi_fused__to_copy_mul_sub_view_4:148"], "sub_3062": ["triton_poi_fused__to_copy_mul_sub_view_4:148"], "convert_element_type_400": ["triton_poi_fused__to_copy_mul_sub_view_4:148"], "view_1043": ["triton_poi_fused__to_copy_mul_sub_view_4:148"], "convert_element_type_401": ["triton_poi_fused__to_copy_mul_sub_view_4:148"], "view_1045": ["triton_poi_fused__to_copy_mul_sub_view_4:148"], "view_1044": ["triton_poi_fused__to_copy_mul_sub_view_4:148"], "mul_6575": ["triton_poi_fused__to_copy_mul_sub_view_4:149"], "sub_3107": ["triton_poi_fused__to_copy_mul_sub_view_4:149"], "convert_element_type_406": ["triton_poi_fused__to_copy_mul_sub_view_4:149"], "view_1059": ["triton_poi_fused__to_copy_mul_sub_view_4:149"], "convert_element_type_407": ["triton_poi_fused__to_copy_mul_sub_view_4:149"], "view_1061": ["triton_poi_fused__to_copy_mul_sub_view_4:149"], "view_1060": ["triton_poi_fused__to_copy_mul_sub_view_4:149"], "mul_6674": ["triton_poi_fused__to_copy_mul_sub_view_4:150"], "sub_3151": ["triton_poi_fused__to_copy_mul_sub_view_4:150"], "convert_element_type_412": ["triton_poi_fused__to_copy_mul_sub_view_4:150"], "view_1075": ["triton_poi_fused__to_copy_mul_sub_view_4:150"], "convert_element_type_413": ["triton_poi_fused__to_copy_mul_sub_view_4:150"], "view_1077": ["triton_poi_fused__to_copy_mul_sub_view_4:150"], "view_1076": ["triton_poi_fused__to_copy_mul_sub_view_4:150"], "_scaled_dot_product_efficient_attention_11": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:151", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:152", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:153", "at::_ops::_scaled_dot_product_efficient_attention::call:507"], "clone_90": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:151", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:152", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:153", "at::_ops::_scaled_dot_product_efficient_attention::call:507"], "permute_112": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:151", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:152", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:153", "at::_ops::_scaled_dot_product_efficient_attention::call:507"], "view_1049": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:151", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:152", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:153", "at::_ops::_scaled_dot_product_efficient_attention::call:507"], "mul_6491": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:151", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:152", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:153", "at::_ops::_scaled_dot_product_efficient_attention::call:507"], "view_1048": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:151", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:152", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:153", "at::_ops::_scaled_dot_product_efficient_attention::call:507"], "add_tensor_104": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:151", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:152", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:153", "at::_ops::_scaled_dot_product_efficient_attention::call:507"], "clone_91": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:151", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:152", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:153", "at::_ops::_scaled_dot_product_efficient_attention::call:507"], "permute_114": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:151", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:152", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:153", "at::_ops::_scaled_dot_product_efficient_attention::call:507"], "view_1065": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:151", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:152", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:153", "at::_ops::_scaled_dot_product_efficient_attention::call:507"], "view_1064": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:151", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:152", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:153", "at::_ops::_scaled_dot_product_efficient_attention::call:507"], "clone_92": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:151", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:152", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:153", "at::_ops::_scaled_dot_product_efficient_attention::call:507"], "permute_116": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:151", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:152", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:153", "at::_ops::_scaled_dot_product_efficient_attention::call:507"], "view_1081": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:151", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:152", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:153", "at::_ops::_scaled_dot_product_efficient_attention::call:507"], "view_1080": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:151", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:152", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:153", "at::_ops::_scaled_dot_product_efficient_attention::call:507"], "add_tensor_103": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:151", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:152", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:153", "at::_ops::_scaled_dot_product_efficient_attention::call:507"], "amax_69": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "view_1082": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "permute_117": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "amin_69": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "sub_3195": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "convert_element_type_416": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "convert_element_type_415": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "clamp_max_139": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "clamp_min_209": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "add_10681": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "round_140": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "mul_6752": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "mul_6749": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "reciprocal_69": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "view_1085": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "clamp_min_207": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "div_138": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "sub_3169": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "maximum_69": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "full_139": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "minimum_69": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "full_138": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "view_1086": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "convert_element_type_414": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "clamp_max_138": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "clamp_min_208": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "sub_3175": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "round_139": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "div_139": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "convert_element_type_417": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "view_1090": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "mul_6774": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "view_1089": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154"], "mul_6779": ["triton_poi_fused__to_copy_mul_sub_view_4:155"], "sub_3199": ["triton_poi_fused__to_copy_mul_sub_view_4:155"], "convert_element_type_418": ["triton_poi_fused__to_copy_mul_sub_view_4:155"], "view_1092": ["triton_poi_fused__to_copy_mul_sub_view_4:155"], "convert_element_type_419": ["triton_poi_fused__to_copy_mul_sub_view_4:155"], "view_1094": ["triton_poi_fused__to_copy_mul_sub_view_4:155"], "view_1093": ["triton_poi_fused__to_copy_mul_sub_view_4:155"], "add_10744": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "view_1097": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "add_tensor_102": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "var_mean_23": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "clone_94": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "add_10750": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "mul_6796": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "mul_6795": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "sub_3205": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "rsqrt_23": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "add_10749": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "amax_70": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "amin_70": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "sub_3242": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "convert_element_type_422": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "convert_element_type_421": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "clamp_max_141": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "clamp_min_212": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "add_10837": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "round_142": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "mul_6847": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "mul_6844": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "reciprocal_70": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "view_1100": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "clamp_min_210": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "div_140": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "sub_3216": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "maximum_70": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "full_141": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "minimum_70": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "full_140": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "view_1101": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "convert_element_type_420": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "clamp_max_140": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "clamp_min_211": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "sub_3222": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "round_141": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "div_141": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "convert_element_type_423": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "view_1105": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "mul_6869": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "view_1104": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156"], "mul_6874": ["triton_poi_fused__to_copy_mul_sub_view_10:157"], "sub_3246": ["triton_poi_fused__to_copy_mul_sub_view_10:157"], "convert_element_type_424": ["triton_poi_fused__to_copy_mul_sub_view_10:157"], "view_1107": ["triton_poi_fused__to_copy_mul_sub_view_10:157"], "convert_element_type_425": ["triton_poi_fused__to_copy_mul_sub_view_10:157"], "view_1109": ["triton_poi_fused__to_copy_mul_sub_view_10:157"], "view_1108": ["triton_poi_fused__to_copy_mul_sub_view_10:157"], "amax_71": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "mul_6888": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "mul_6886": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "view_1112": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "add_tensor_101": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "add_10896": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "erf_13": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "mul_6887": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "amin_71": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "add_10979": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "round_144": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "mul_6937": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "mul_6934": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "reciprocal_71": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "view_1115": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "clamp_min_213": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "div_142": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "sub_3259": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "maximum_71": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "full_143": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "minimum_71": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "full_142": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "view_1116": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "convert_element_type_426": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "clamp_max_142": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "clamp_min_214": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "sub_3265": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "round_143": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "div_143": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "mul_6959": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "sub_3285": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "convert_element_type_428": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "convert_element_type_427": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "clamp_max_143": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "clamp_min_215": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "convert_element_type_429": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "view_1120": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "view_1119": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158"], "mul_6964": ["triton_poi_fused__to_copy_mul_sub_view_10:159"], "sub_3289": ["triton_poi_fused__to_copy_mul_sub_view_10:159"], "convert_element_type_430": ["triton_poi_fused__to_copy_mul_sub_view_10:159"], "view_1122": ["triton_poi_fused__to_copy_mul_sub_view_10:159"], "convert_element_type_431": ["triton_poi_fused__to_copy_mul_sub_view_10:159"], "view_1124": ["triton_poi_fused__to_copy_mul_sub_view_10:159"], "view_1123": ["triton_poi_fused__to_copy_mul_sub_view_10:159"], "var_mean_24": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "clone_97": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "add_11042": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "view_1127": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "add_tensor_100": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "add_11048": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "mul_6981": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "mul_6980": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "sub_3295": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "rsqrt_24": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "add_11047": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "amax_72": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "amin_72": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "amax_73": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "amin_73": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "amax_74": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "amin_74": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "sub_3332": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "convert_element_type_434": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "convert_element_type_433": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "clamp_max_145": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "clamp_min_218": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "add_11135": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "round_146": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "mul_7032": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "mul_7029": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "reciprocal_72": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "view_1130": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "clamp_min_216": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "div_144": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "sub_3306": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "maximum_72": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "full_145": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "minimum_72": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "full_144": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "view_1131": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "convert_element_type_432": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "clamp_max_144": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "clamp_min_217": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "sub_3312": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "round_145": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "div_145": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "convert_element_type_435": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "view_1135": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "mul_7054": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "view_1134": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "sub_3377": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "convert_element_type_440": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "convert_element_type_439": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "clamp_max_147": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "clamp_min_221": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "add_11287": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "round_148": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "mul_7128": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "mul_7125": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "reciprocal_73": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "view_1146": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "clamp_min_219": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "div_146": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "sub_3351": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "maximum_73": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "full_147": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "minimum_73": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "full_146": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "view_1147": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "convert_element_type_438": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "clamp_max_146": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "clamp_min_220": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "sub_3357": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "round_147": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "div_147": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "convert_element_type_441": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "view_1151": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "mul_7150": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "view_1150": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "sub_3421": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "convert_element_type_446": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "convert_element_type_445": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "clamp_max_149": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "clamp_min_224": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "add_11435": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "round_150": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "mul_7227": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "mul_7224": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "reciprocal_74": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "view_1162": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "clamp_min_222": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "div_148": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "sub_3395": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "maximum_74": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "full_149": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "minimum_74": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "full_148": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "view_1163": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "convert_element_type_444": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "clamp_max_148": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "clamp_min_223": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "sub_3401": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "round_149": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "div_149": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "convert_element_type_447": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "view_1167": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "mul_7249": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "view_1166": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160"], "mul_7059": ["triton_poi_fused__to_copy_mul_sub_view_4:161"], "sub_3336": ["triton_poi_fused__to_copy_mul_sub_view_4:161"], "convert_element_type_436": ["triton_poi_fused__to_copy_mul_sub_view_4:161"], "view_1137": ["triton_poi_fused__to_copy_mul_sub_view_4:161"], "convert_element_type_437": ["triton_poi_fused__to_copy_mul_sub_view_4:161"], "view_1139": ["triton_poi_fused__to_copy_mul_sub_view_4:161"], "view_1138": ["triton_poi_fused__to_copy_mul_sub_view_4:161"], "mul_7155": ["triton_poi_fused__to_copy_mul_sub_view_4:162"], "sub_3381": ["triton_poi_fused__to_copy_mul_sub_view_4:162"], "convert_element_type_442": ["triton_poi_fused__to_copy_mul_sub_view_4:162"], "view_1153": ["triton_poi_fused__to_copy_mul_sub_view_4:162"], "convert_element_type_443": ["triton_poi_fused__to_copy_mul_sub_view_4:162"], "view_1155": ["triton_poi_fused__to_copy_mul_sub_view_4:162"], "view_1154": ["triton_poi_fused__to_copy_mul_sub_view_4:162"], "mul_7254": ["triton_poi_fused__to_copy_mul_sub_view_4:163"], "sub_3425": ["triton_poi_fused__to_copy_mul_sub_view_4:163"], "convert_element_type_448": ["triton_poi_fused__to_copy_mul_sub_view_4:163"], "view_1169": ["triton_poi_fused__to_copy_mul_sub_view_4:163"], "convert_element_type_449": ["triton_poi_fused__to_copy_mul_sub_view_4:163"], "view_1171": ["triton_poi_fused__to_copy_mul_sub_view_4:163"], "view_1170": ["triton_poi_fused__to_copy_mul_sub_view_4:163"], "_scaled_dot_product_efficient_attention_12": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:164", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:165", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:166", "at::_ops::_scaled_dot_product_efficient_attention::call:514"], "clone_98": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:164", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:165", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:166", "at::_ops::_scaled_dot_product_efficient_attention::call:514"], "permute_122": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:164", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:165", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:166", "at::_ops::_scaled_dot_product_efficient_attention::call:514"], "view_1143": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:164", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:165", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:166", "at::_ops::_scaled_dot_product_efficient_attention::call:514"], "mul_7071": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:164", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:165", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:166", "at::_ops::_scaled_dot_product_efficient_attention::call:514"], "view_1142": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:164", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:165", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:166", "at::_ops::_scaled_dot_product_efficient_attention::call:514"], "add_tensor_99": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:164", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:165", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:166", "at::_ops::_scaled_dot_product_efficient_attention::call:514"], "clone_99": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:164", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:165", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:166", "at::_ops::_scaled_dot_product_efficient_attention::call:514"], "permute_124": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:164", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:165", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:166", "at::_ops::_scaled_dot_product_efficient_attention::call:514"], "view_1159": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:164", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:165", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:166", "at::_ops::_scaled_dot_product_efficient_attention::call:514"], "view_1158": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:164", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:165", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:166", "at::_ops::_scaled_dot_product_efficient_attention::call:514"], "clone_100": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:164", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:165", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:166", "at::_ops::_scaled_dot_product_efficient_attention::call:514"], "permute_126": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:164", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:165", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:166", "at::_ops::_scaled_dot_product_efficient_attention::call:514"], "view_1175": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:164", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:165", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:166", "at::_ops::_scaled_dot_product_efficient_attention::call:514"], "view_1174": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:164", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:165", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:166", "at::_ops::_scaled_dot_product_efficient_attention::call:514"], "add_tensor_98": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:164", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:165", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:166", "at::_ops::_scaled_dot_product_efficient_attention::call:514"], "amax_75": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "view_1176": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "permute_127": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "amin_75": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "sub_3469": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "convert_element_type_452": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "convert_element_type_451": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "clamp_max_151": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "clamp_min_227": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "add_11599": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "round_152": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "mul_7332": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "mul_7329": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "reciprocal_75": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "view_1179": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "clamp_min_225": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "div_150": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "sub_3443": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "maximum_75": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "full_151": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "minimum_75": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "full_150": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "view_1180": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "convert_element_type_450": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "clamp_max_150": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "clamp_min_226": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "sub_3449": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "round_151": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "div_151": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "convert_element_type_453": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "view_1184": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "mul_7354": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "view_1183": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167"], "mul_7359": ["triton_poi_fused__to_copy_mul_sub_view_4:168"], "sub_3473": ["triton_poi_fused__to_copy_mul_sub_view_4:168"], "convert_element_type_454": ["triton_poi_fused__to_copy_mul_sub_view_4:168"], "view_1186": ["triton_poi_fused__to_copy_mul_sub_view_4:168"], "convert_element_type_455": ["triton_poi_fused__to_copy_mul_sub_view_4:168"], "view_1188": ["triton_poi_fused__to_copy_mul_sub_view_4:168"], "view_1187": ["triton_poi_fused__to_copy_mul_sub_view_4:168"], "add_11662": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "view_1191": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "add_tensor_97": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "var_mean_25": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "clone_102": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "add_11668": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "mul_7376": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "mul_7375": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "sub_3479": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "rsqrt_25": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "add_11667": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "amax_76": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "amin_76": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "sub_3516": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "convert_element_type_458": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "convert_element_type_457": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "clamp_max_153": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "clamp_min_230": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "add_11755": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "round_154": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "mul_7427": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "mul_7424": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "reciprocal_76": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "view_1194": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "clamp_min_228": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "div_152": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "sub_3490": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "maximum_76": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "full_153": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "minimum_76": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "full_152": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "view_1195": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "convert_element_type_456": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "clamp_max_152": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "clamp_min_229": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "sub_3496": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "round_153": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "div_153": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "convert_element_type_459": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "view_1199": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "mul_7449": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "view_1198": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169"], "mul_7454": ["triton_poi_fused__to_copy_mul_sub_view_10:170"], "sub_3520": ["triton_poi_fused__to_copy_mul_sub_view_10:170"], "convert_element_type_460": ["triton_poi_fused__to_copy_mul_sub_view_10:170"], "view_1201": ["triton_poi_fused__to_copy_mul_sub_view_10:170"], "convert_element_type_461": ["triton_poi_fused__to_copy_mul_sub_view_10:170"], "view_1203": ["triton_poi_fused__to_copy_mul_sub_view_10:170"], "view_1202": ["triton_poi_fused__to_copy_mul_sub_view_10:170"], "amax_77": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "mul_7468": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "mul_7466": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "view_1206": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "add_tensor_96": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "add_11814": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "erf_14": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "mul_7467": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "amin_77": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "add_11897": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "round_156": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "mul_7517": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "mul_7514": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "reciprocal_77": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "view_1209": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "clamp_min_231": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "div_154": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "sub_3533": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "maximum_77": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "full_155": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "minimum_77": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "full_154": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "view_1210": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "convert_element_type_462": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "clamp_max_154": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "clamp_min_232": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "sub_3539": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "round_155": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "div_155": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "mul_7539": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "sub_3559": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "convert_element_type_464": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "convert_element_type_463": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "clamp_max_155": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "clamp_min_233": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "convert_element_type_465": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "view_1214": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "view_1213": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171"], "mul_7544": ["triton_poi_fused__to_copy_mul_sub_view_10:172"], "sub_3563": ["triton_poi_fused__to_copy_mul_sub_view_10:172"], "convert_element_type_466": ["triton_poi_fused__to_copy_mul_sub_view_10:172"], "view_1216": ["triton_poi_fused__to_copy_mul_sub_view_10:172"], "convert_element_type_467": ["triton_poi_fused__to_copy_mul_sub_view_10:172"], "view_1218": ["triton_poi_fused__to_copy_mul_sub_view_10:172"], "view_1217": ["triton_poi_fused__to_copy_mul_sub_view_10:172"], "var_mean_26": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "clone_105": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "add_11960": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "view_1221": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "add_tensor_95": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "add_11966": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "mul_7561": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "mul_7560": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "sub_3569": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "rsqrt_26": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "add_11965": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "amax_78": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "amin_78": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "amax_79": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "amin_79": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "amax_80": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "amin_80": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "sub_3606": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "convert_element_type_470": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "convert_element_type_469": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "clamp_max_157": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "clamp_min_236": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "add_12053": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "round_158": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "mul_7612": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "mul_7609": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "reciprocal_78": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "view_1224": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "clamp_min_234": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "div_156": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "sub_3580": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "maximum_78": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "full_157": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "minimum_78": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "full_156": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "view_1225": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "convert_element_type_468": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "clamp_max_156": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "clamp_min_235": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "sub_3586": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "round_157": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "div_157": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "convert_element_type_471": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "view_1229": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "mul_7634": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "view_1228": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "sub_3651": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "convert_element_type_476": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "convert_element_type_475": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "clamp_max_159": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "clamp_min_239": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "add_12205": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "round_160": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "mul_7708": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "mul_7705": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "reciprocal_79": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "view_1240": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "clamp_min_237": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "div_158": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "sub_3625": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "maximum_79": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "full_159": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "minimum_79": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "full_158": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "view_1241": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "convert_element_type_474": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "clamp_max_158": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "clamp_min_238": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "sub_3631": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "round_159": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "div_159": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "convert_element_type_477": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "view_1245": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "mul_7730": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "view_1244": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "sub_3695": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "convert_element_type_482": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "convert_element_type_481": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "clamp_max_161": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "clamp_min_242": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "add_12353": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "round_162": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "mul_7807": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "mul_7804": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "reciprocal_80": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "view_1256": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "clamp_min_240": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "div_160": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "sub_3669": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "maximum_80": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "full_161": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "minimum_80": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "full_160": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "view_1257": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "convert_element_type_480": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "clamp_max_160": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "clamp_min_241": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "sub_3675": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "round_161": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "div_161": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "convert_element_type_483": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "view_1261": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "mul_7829": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "view_1260": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173"], "mul_7639": ["triton_poi_fused__to_copy_mul_sub_view_4:174"], "sub_3610": ["triton_poi_fused__to_copy_mul_sub_view_4:174"], "convert_element_type_472": ["triton_poi_fused__to_copy_mul_sub_view_4:174"], "view_1231": ["triton_poi_fused__to_copy_mul_sub_view_4:174"], "convert_element_type_473": ["triton_poi_fused__to_copy_mul_sub_view_4:174"], "view_1233": ["triton_poi_fused__to_copy_mul_sub_view_4:174"], "view_1232": ["triton_poi_fused__to_copy_mul_sub_view_4:174"], "mul_7735": ["triton_poi_fused__to_copy_mul_sub_view_4:175"], "sub_3655": ["triton_poi_fused__to_copy_mul_sub_view_4:175"], "convert_element_type_478": ["triton_poi_fused__to_copy_mul_sub_view_4:175"], "view_1247": ["triton_poi_fused__to_copy_mul_sub_view_4:175"], "convert_element_type_479": ["triton_poi_fused__to_copy_mul_sub_view_4:175"], "view_1249": ["triton_poi_fused__to_copy_mul_sub_view_4:175"], "view_1248": ["triton_poi_fused__to_copy_mul_sub_view_4:175"], "mul_7834": ["triton_poi_fused__to_copy_mul_sub_view_4:176"], "sub_3699": ["triton_poi_fused__to_copy_mul_sub_view_4:176"], "convert_element_type_484": ["triton_poi_fused__to_copy_mul_sub_view_4:176"], "view_1263": ["triton_poi_fused__to_copy_mul_sub_view_4:176"], "convert_element_type_485": ["triton_poi_fused__to_copy_mul_sub_view_4:176"], "view_1265": ["triton_poi_fused__to_copy_mul_sub_view_4:176"], "view_1264": ["triton_poi_fused__to_copy_mul_sub_view_4:176"], "_scaled_dot_product_efficient_attention_13": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:177", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:178", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:179", "at::_ops::_scaled_dot_product_efficient_attention::call:521"], "clone_106": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:177", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:178", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:179", "at::_ops::_scaled_dot_product_efficient_attention::call:521"], "permute_132": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:177", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:178", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:179", "at::_ops::_scaled_dot_product_efficient_attention::call:521"], "view_1237": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:177", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:178", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:179", "at::_ops::_scaled_dot_product_efficient_attention::call:521"], "mul_7651": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:177", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:178", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:179", "at::_ops::_scaled_dot_product_efficient_attention::call:521"], "view_1236": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:177", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:178", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:179", "at::_ops::_scaled_dot_product_efficient_attention::call:521"], "add_tensor_94": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:177", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:178", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:179", "at::_ops::_scaled_dot_product_efficient_attention::call:521"], "clone_107": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:177", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:178", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:179", "at::_ops::_scaled_dot_product_efficient_attention::call:521"], "permute_134": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:177", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:178", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:179", "at::_ops::_scaled_dot_product_efficient_attention::call:521"], "view_1253": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:177", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:178", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:179", "at::_ops::_scaled_dot_product_efficient_attention::call:521"], "view_1252": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:177", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:178", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:179", "at::_ops::_scaled_dot_product_efficient_attention::call:521"], "clone_108": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:177", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:178", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:179", "at::_ops::_scaled_dot_product_efficient_attention::call:521"], "permute_136": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:177", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:178", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:179", "at::_ops::_scaled_dot_product_efficient_attention::call:521"], "view_1269": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:177", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:178", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:179", "at::_ops::_scaled_dot_product_efficient_attention::call:521"], "view_1268": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:177", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:178", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:179", "at::_ops::_scaled_dot_product_efficient_attention::call:521"], "add_tensor_93": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:177", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:178", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:179", "at::_ops::_scaled_dot_product_efficient_attention::call:521"], "amax_81": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "view_1270": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "permute_137": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "amin_81": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "sub_3743": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "convert_element_type_488": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "convert_element_type_487": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "clamp_max_163": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "clamp_min_245": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "add_12517": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "round_164": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "mul_7912": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "mul_7909": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "reciprocal_81": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "view_1273": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "clamp_min_243": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "div_162": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "sub_3717": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "maximum_81": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "full_163": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "minimum_81": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "full_162": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "view_1274": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "convert_element_type_486": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "clamp_max_162": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "clamp_min_244": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "sub_3723": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "round_163": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "div_163": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "convert_element_type_489": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "view_1278": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "mul_7934": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "view_1277": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180"], "mul_7939": ["triton_poi_fused__to_copy_mul_sub_view_4:181"], "sub_3747": ["triton_poi_fused__to_copy_mul_sub_view_4:181"], "convert_element_type_490": ["triton_poi_fused__to_copy_mul_sub_view_4:181"], "view_1280": ["triton_poi_fused__to_copy_mul_sub_view_4:181"], "convert_element_type_491": ["triton_poi_fused__to_copy_mul_sub_view_4:181"], "view_1282": ["triton_poi_fused__to_copy_mul_sub_view_4:181"], "view_1281": ["triton_poi_fused__to_copy_mul_sub_view_4:181"], "add_12580": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "view_1285": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "add_tensor_92": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "var_mean_27": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "clone_110": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "add_12586": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "mul_7956": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "mul_7955": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "sub_3753": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "rsqrt_27": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "add_12585": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "amax_82": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "amin_82": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "sub_3790": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "convert_element_type_494": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "convert_element_type_493": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "clamp_max_165": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "clamp_min_248": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "add_12673": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "round_166": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "mul_8007": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "mul_8004": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "reciprocal_82": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "view_1288": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "clamp_min_246": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "div_164": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "sub_3764": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "maximum_82": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "full_165": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "minimum_82": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "full_164": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "view_1289": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "convert_element_type_492": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "clamp_max_164": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "clamp_min_247": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "sub_3770": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "round_165": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "div_165": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "convert_element_type_495": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "view_1293": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "mul_8029": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "view_1292": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182"], "mul_8034": ["triton_poi_fused__to_copy_mul_sub_view_10:183"], "sub_3794": ["triton_poi_fused__to_copy_mul_sub_view_10:183"], "convert_element_type_496": ["triton_poi_fused__to_copy_mul_sub_view_10:183"], "view_1295": ["triton_poi_fused__to_copy_mul_sub_view_10:183"], "convert_element_type_497": ["triton_poi_fused__to_copy_mul_sub_view_10:183"], "view_1297": ["triton_poi_fused__to_copy_mul_sub_view_10:183"], "view_1296": ["triton_poi_fused__to_copy_mul_sub_view_10:183"], "amax_83": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "mul_8048": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "mul_8046": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "view_1300": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "add_tensor_91": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "add_12732": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "erf_15": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "mul_8047": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "amin_83": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "add_12815": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "round_168": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "mul_8097": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "mul_8094": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "reciprocal_83": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "view_1303": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "clamp_min_249": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "div_166": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "sub_3807": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "maximum_83": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "full_167": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "minimum_83": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "full_166": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "view_1304": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "convert_element_type_498": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "clamp_max_166": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "clamp_min_250": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "sub_3813": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "round_167": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "div_167": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "mul_8119": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "sub_3833": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "convert_element_type_500": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "convert_element_type_499": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "clamp_max_167": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "clamp_min_251": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "convert_element_type_501": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "view_1308": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "view_1307": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184"], "mul_8124": ["triton_poi_fused__to_copy_mul_sub_view_10:185"], "sub_3837": ["triton_poi_fused__to_copy_mul_sub_view_10:185"], "convert_element_type_502": ["triton_poi_fused__to_copy_mul_sub_view_10:185"], "view_1310": ["triton_poi_fused__to_copy_mul_sub_view_10:185"], "convert_element_type_503": ["triton_poi_fused__to_copy_mul_sub_view_10:185"], "view_1312": ["triton_poi_fused__to_copy_mul_sub_view_10:185"], "view_1311": ["triton_poi_fused__to_copy_mul_sub_view_10:185"], "var_mean_28": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "clone_113": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "add_12878": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "view_1315": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "add_tensor_90": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "add_12884": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "mul_8141": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "mul_8140": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "sub_3843": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "rsqrt_28": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "add_12883": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "amax_84": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "amin_84": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "amax_85": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "amin_85": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "amax_86": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "amin_86": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "sub_3880": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "convert_element_type_506": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "convert_element_type_505": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "clamp_max_169": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "clamp_min_254": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "add_12971": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "round_170": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "mul_8192": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "mul_8189": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "reciprocal_84": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "view_1318": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "clamp_min_252": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "div_168": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "sub_3854": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "maximum_84": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "full_169": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "minimum_84": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "full_168": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "view_1319": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "convert_element_type_504": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "clamp_max_168": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "clamp_min_253": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "sub_3860": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "round_169": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "div_169": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "convert_element_type_507": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "view_1323": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "mul_8214": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "view_1322": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "sub_3925": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "convert_element_type_512": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "convert_element_type_511": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "clamp_max_171": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "clamp_min_257": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "add_13123": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "round_172": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "mul_8288": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "mul_8285": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "reciprocal_85": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "view_1334": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "clamp_min_255": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "div_170": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "sub_3899": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "maximum_85": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "full_171": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "minimum_85": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "full_170": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "view_1335": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "convert_element_type_510": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "clamp_max_170": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "clamp_min_256": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "sub_3905": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "round_171": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "div_171": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "convert_element_type_513": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "view_1339": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "mul_8310": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "view_1338": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "sub_3969": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "convert_element_type_518": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "convert_element_type_517": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "clamp_max_173": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "clamp_min_260": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "add_13271": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "round_174": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "mul_8387": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "mul_8384": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "reciprocal_86": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "view_1350": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "clamp_min_258": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "div_172": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "sub_3943": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "maximum_86": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "full_173": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "minimum_86": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "full_172": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "view_1351": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "convert_element_type_516": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "clamp_max_172": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "clamp_min_259": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "sub_3949": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "round_173": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "div_173": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "convert_element_type_519": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "view_1355": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "mul_8409": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "view_1354": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186"], "mul_8219": ["triton_poi_fused__to_copy_mul_sub_view_4:187"], "sub_3884": ["triton_poi_fused__to_copy_mul_sub_view_4:187"], "convert_element_type_508": ["triton_poi_fused__to_copy_mul_sub_view_4:187"], "view_1325": ["triton_poi_fused__to_copy_mul_sub_view_4:187"], "convert_element_type_509": ["triton_poi_fused__to_copy_mul_sub_view_4:187"], "view_1327": ["triton_poi_fused__to_copy_mul_sub_view_4:187"], "view_1326": ["triton_poi_fused__to_copy_mul_sub_view_4:187"], "mul_8315": ["triton_poi_fused__to_copy_mul_sub_view_4:188"], "sub_3929": ["triton_poi_fused__to_copy_mul_sub_view_4:188"], "convert_element_type_514": ["triton_poi_fused__to_copy_mul_sub_view_4:188"], "view_1341": ["triton_poi_fused__to_copy_mul_sub_view_4:188"], "convert_element_type_515": ["triton_poi_fused__to_copy_mul_sub_view_4:188"], "view_1343": ["triton_poi_fused__to_copy_mul_sub_view_4:188"], "view_1342": ["triton_poi_fused__to_copy_mul_sub_view_4:188"], "mul_8414": ["triton_poi_fused__to_copy_mul_sub_view_4:189"], "sub_3973": ["triton_poi_fused__to_copy_mul_sub_view_4:189"], "convert_element_type_520": ["triton_poi_fused__to_copy_mul_sub_view_4:189"], "view_1357": ["triton_poi_fused__to_copy_mul_sub_view_4:189"], "convert_element_type_521": ["triton_poi_fused__to_copy_mul_sub_view_4:189"], "view_1359": ["triton_poi_fused__to_copy_mul_sub_view_4:189"], "view_1358": ["triton_poi_fused__to_copy_mul_sub_view_4:189"], "_scaled_dot_product_efficient_attention_14": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:190", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:191", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:192", "at::_ops::_scaled_dot_product_efficient_attention::call:528"], "clone_114": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:190", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:191", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:192", "at::_ops::_scaled_dot_product_efficient_attention::call:528"], "permute_142": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:190", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:191", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:192", "at::_ops::_scaled_dot_product_efficient_attention::call:528"], "view_1331": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:190", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:191", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:192", "at::_ops::_scaled_dot_product_efficient_attention::call:528"], "mul_8231": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:190", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:191", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:192", "at::_ops::_scaled_dot_product_efficient_attention::call:528"], "view_1330": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:190", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:191", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:192", "at::_ops::_scaled_dot_product_efficient_attention::call:528"], "add_tensor_89": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:190", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:191", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:192", "at::_ops::_scaled_dot_product_efficient_attention::call:528"], "clone_115": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:190", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:191", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:192", "at::_ops::_scaled_dot_product_efficient_attention::call:528"], "permute_144": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:190", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:191", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:192", "at::_ops::_scaled_dot_product_efficient_attention::call:528"], "view_1347": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:190", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:191", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:192", "at::_ops::_scaled_dot_product_efficient_attention::call:528"], "view_1346": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:190", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:191", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:192", "at::_ops::_scaled_dot_product_efficient_attention::call:528"], "clone_116": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:190", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:191", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:192", "at::_ops::_scaled_dot_product_efficient_attention::call:528"], "permute_146": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:190", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:191", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:192", "at::_ops::_scaled_dot_product_efficient_attention::call:528"], "view_1363": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:190", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:191", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:192", "at::_ops::_scaled_dot_product_efficient_attention::call:528"], "view_1362": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:190", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:191", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:192", "at::_ops::_scaled_dot_product_efficient_attention::call:528"], "add_tensor_88": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:190", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:191", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:192", "at::_ops::_scaled_dot_product_efficient_attention::call:528"], "amax_87": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "view_1364": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "permute_147": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "amin_87": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "sub_4017": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "convert_element_type_524": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "convert_element_type_523": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "clamp_max_175": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "clamp_min_263": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "add_13435": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "round_176": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "mul_8492": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "mul_8489": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "reciprocal_87": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "view_1367": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "clamp_min_261": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "div_174": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "sub_3991": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "maximum_87": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "full_175": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "minimum_87": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "full_174": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "view_1368": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "convert_element_type_522": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "clamp_max_174": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "clamp_min_262": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "sub_3997": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "round_175": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "div_175": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "convert_element_type_525": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "view_1372": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "mul_8514": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "view_1371": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193"], "mul_8519": ["triton_poi_fused__to_copy_mul_sub_view_4:194"], "sub_4021": ["triton_poi_fused__to_copy_mul_sub_view_4:194"], "convert_element_type_526": ["triton_poi_fused__to_copy_mul_sub_view_4:194"], "view_1374": ["triton_poi_fused__to_copy_mul_sub_view_4:194"], "convert_element_type_527": ["triton_poi_fused__to_copy_mul_sub_view_4:194"], "view_1376": ["triton_poi_fused__to_copy_mul_sub_view_4:194"], "view_1375": ["triton_poi_fused__to_copy_mul_sub_view_4:194"], "add_13498": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "view_1379": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "add_tensor_87": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "var_mean_29": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "clone_118": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "add_13504": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "mul_8536": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "mul_8535": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "sub_4027": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "rsqrt_29": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "add_13503": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "amax_88": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "amin_88": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "sub_4064": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "convert_element_type_530": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "convert_element_type_529": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "clamp_max_177": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "clamp_min_266": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "add_13591": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "round_178": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "mul_8587": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "mul_8584": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "reciprocal_88": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "view_1382": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "clamp_min_264": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "div_176": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "sub_4038": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "maximum_88": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "full_177": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "minimum_88": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "full_176": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "view_1383": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "convert_element_type_528": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "clamp_max_176": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "clamp_min_265": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "sub_4044": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "round_177": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "div_177": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "convert_element_type_531": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "view_1387": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "mul_8609": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "view_1386": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195"], "mul_8614": ["triton_poi_fused__to_copy_mul_sub_view_10:196"], "sub_4068": ["triton_poi_fused__to_copy_mul_sub_view_10:196"], "convert_element_type_532": ["triton_poi_fused__to_copy_mul_sub_view_10:196"], "view_1389": ["triton_poi_fused__to_copy_mul_sub_view_10:196"], "convert_element_type_533": ["triton_poi_fused__to_copy_mul_sub_view_10:196"], "view_1391": ["triton_poi_fused__to_copy_mul_sub_view_10:196"], "view_1390": ["triton_poi_fused__to_copy_mul_sub_view_10:196"], "amax_89": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "mul_8628": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "mul_8626": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "view_1394": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "add_tensor_86": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "add_13650": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "erf_16": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "mul_8627": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "amin_89": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "add_13733": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "round_180": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "mul_8677": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "mul_8674": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "reciprocal_89": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "view_1397": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "clamp_min_267": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "div_178": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "sub_4081": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "maximum_89": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "full_179": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "minimum_89": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "full_178": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "view_1398": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "convert_element_type_534": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "clamp_max_178": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "clamp_min_268": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "sub_4087": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "round_179": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "div_179": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "mul_8699": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "sub_4107": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "convert_element_type_536": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "convert_element_type_535": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "clamp_max_179": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "clamp_min_269": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "convert_element_type_537": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "view_1402": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "view_1401": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197"], "mul_8704": ["triton_poi_fused__to_copy_mul_sub_view_10:198"], "sub_4111": ["triton_poi_fused__to_copy_mul_sub_view_10:198"], "convert_element_type_538": ["triton_poi_fused__to_copy_mul_sub_view_10:198"], "view_1404": ["triton_poi_fused__to_copy_mul_sub_view_10:198"], "convert_element_type_539": ["triton_poi_fused__to_copy_mul_sub_view_10:198"], "view_1406": ["triton_poi_fused__to_copy_mul_sub_view_10:198"], "view_1405": ["triton_poi_fused__to_copy_mul_sub_view_10:198"], "var_mean_30": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "clone_121": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "add_13796": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "view_1409": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "add_tensor_85": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "add_13802": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "mul_8721": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "mul_8720": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "sub_4117": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "rsqrt_30": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "add_13801": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "amax_90": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "amin_90": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "amax_91": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "amin_91": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "amax_92": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "amin_92": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "sub_4154": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "convert_element_type_542": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "convert_element_type_541": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "clamp_max_181": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "clamp_min_272": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "add_13889": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "round_182": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "mul_8772": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "mul_8769": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "reciprocal_90": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "view_1412": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "clamp_min_270": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "div_180": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "sub_4128": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "maximum_90": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "full_181": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "minimum_90": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "full_180": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "view_1413": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "convert_element_type_540": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "clamp_max_180": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "clamp_min_271": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "sub_4134": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "round_181": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "div_181": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "convert_element_type_543": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "view_1417": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "mul_8794": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "view_1416": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "sub_4199": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "convert_element_type_548": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "convert_element_type_547": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "clamp_max_183": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "clamp_min_275": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "add_14041": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "round_184": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "mul_8868": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "mul_8865": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "reciprocal_91": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "view_1428": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "clamp_min_273": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "div_182": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "sub_4173": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "maximum_91": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "full_183": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "minimum_91": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "full_182": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "view_1429": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "convert_element_type_546": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "clamp_max_182": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "clamp_min_274": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "sub_4179": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "round_183": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "div_183": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "convert_element_type_549": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "view_1433": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "mul_8890": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "view_1432": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "sub_4243": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "convert_element_type_554": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "convert_element_type_553": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "clamp_max_185": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "clamp_min_278": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "add_14189": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "round_186": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "mul_8967": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "mul_8964": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "reciprocal_92": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "view_1444": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "clamp_min_276": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "div_184": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "sub_4217": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "maximum_92": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "full_185": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "minimum_92": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "full_184": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "view_1445": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "convert_element_type_552": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "clamp_max_184": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "clamp_min_277": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "sub_4223": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "round_185": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "div_185": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "convert_element_type_555": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "view_1449": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "mul_8989": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "view_1448": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199"], "mul_8799": ["triton_poi_fused__to_copy_mul_sub_view_4:200"], "sub_4158": ["triton_poi_fused__to_copy_mul_sub_view_4:200"], "convert_element_type_544": ["triton_poi_fused__to_copy_mul_sub_view_4:200"], "view_1419": ["triton_poi_fused__to_copy_mul_sub_view_4:200"], "convert_element_type_545": ["triton_poi_fused__to_copy_mul_sub_view_4:200"], "view_1421": ["triton_poi_fused__to_copy_mul_sub_view_4:200"], "view_1420": ["triton_poi_fused__to_copy_mul_sub_view_4:200"], "mul_8895": ["triton_poi_fused__to_copy_mul_sub_view_4:201"], "sub_4203": ["triton_poi_fused__to_copy_mul_sub_view_4:201"], "convert_element_type_550": ["triton_poi_fused__to_copy_mul_sub_view_4:201"], "view_1435": ["triton_poi_fused__to_copy_mul_sub_view_4:201"], "convert_element_type_551": ["triton_poi_fused__to_copy_mul_sub_view_4:201"], "view_1437": ["triton_poi_fused__to_copy_mul_sub_view_4:201"], "view_1436": ["triton_poi_fused__to_copy_mul_sub_view_4:201"], "mul_8994": ["triton_poi_fused__to_copy_mul_sub_view_4:202"], "sub_4247": ["triton_poi_fused__to_copy_mul_sub_view_4:202"], "convert_element_type_556": ["triton_poi_fused__to_copy_mul_sub_view_4:202"], "view_1451": ["triton_poi_fused__to_copy_mul_sub_view_4:202"], "convert_element_type_557": ["triton_poi_fused__to_copy_mul_sub_view_4:202"], "view_1453": ["triton_poi_fused__to_copy_mul_sub_view_4:202"], "view_1452": ["triton_poi_fused__to_copy_mul_sub_view_4:202"], "_scaled_dot_product_efficient_attention_15": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:203", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:204", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:205", "at::_ops::_scaled_dot_product_efficient_attention::call:535"], "clone_122": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:203", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:204", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:205", "at::_ops::_scaled_dot_product_efficient_attention::call:535"], "permute_152": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:203", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:204", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:205", "at::_ops::_scaled_dot_product_efficient_attention::call:535"], "view_1425": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:203", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:204", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:205", "at::_ops::_scaled_dot_product_efficient_attention::call:535"], "mul_8811": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:203", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:204", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:205", "at::_ops::_scaled_dot_product_efficient_attention::call:535"], "view_1424": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:203", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:204", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:205", "at::_ops::_scaled_dot_product_efficient_attention::call:535"], "add_tensor_84": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:203", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:204", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:205", "at::_ops::_scaled_dot_product_efficient_attention::call:535"], "clone_123": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:203", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:204", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:205", "at::_ops::_scaled_dot_product_efficient_attention::call:535"], "permute_154": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:203", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:204", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:205", "at::_ops::_scaled_dot_product_efficient_attention::call:535"], "view_1441": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:203", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:204", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:205", "at::_ops::_scaled_dot_product_efficient_attention::call:535"], "view_1440": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:203", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:204", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:205", "at::_ops::_scaled_dot_product_efficient_attention::call:535"], "clone_124": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:203", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:204", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:205", "at::_ops::_scaled_dot_product_efficient_attention::call:535"], "permute_156": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:203", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:204", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:205", "at::_ops::_scaled_dot_product_efficient_attention::call:535"], "view_1457": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:203", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:204", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:205", "at::_ops::_scaled_dot_product_efficient_attention::call:535"], "view_1456": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:203", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:204", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:205", "at::_ops::_scaled_dot_product_efficient_attention::call:535"], "add_tensor_83": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:203", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:204", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:205", "at::_ops::_scaled_dot_product_efficient_attention::call:535"], "amax_93": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "view_1458": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "permute_157": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "amin_93": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "sub_4291": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "convert_element_type_560": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "convert_element_type_559": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "clamp_max_187": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "clamp_min_281": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "add_14353": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "round_188": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "mul_9072": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "mul_9069": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "reciprocal_93": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "view_1461": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "clamp_min_279": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "div_186": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "sub_4265": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "maximum_93": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "full_187": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "minimum_93": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "full_186": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "view_1462": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "convert_element_type_558": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "clamp_max_186": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "clamp_min_280": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "sub_4271": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "round_187": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "div_187": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "convert_element_type_561": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "view_1466": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "mul_9094": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "view_1465": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206"], "mul_9099": ["triton_poi_fused__to_copy_mul_sub_view_4:207"], "sub_4295": ["triton_poi_fused__to_copy_mul_sub_view_4:207"], "convert_element_type_562": ["triton_poi_fused__to_copy_mul_sub_view_4:207"], "view_1468": ["triton_poi_fused__to_copy_mul_sub_view_4:207"], "convert_element_type_563": ["triton_poi_fused__to_copy_mul_sub_view_4:207"], "view_1470": ["triton_poi_fused__to_copy_mul_sub_view_4:207"], "view_1469": ["triton_poi_fused__to_copy_mul_sub_view_4:207"], "add_14416": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "view_1473": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "add_tensor_82": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "var_mean_31": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "clone_126": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "add_14422": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "mul_9116": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "mul_9115": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "sub_4301": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "rsqrt_31": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "add_14421": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "amax_94": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "amin_94": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "sub_4338": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "convert_element_type_566": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "convert_element_type_565": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "clamp_max_189": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "clamp_min_284": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "add_14509": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "round_190": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "mul_9167": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "mul_9164": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "reciprocal_94": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "view_1476": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "clamp_min_282": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "div_188": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "sub_4312": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "maximum_94": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "full_189": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "minimum_94": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "full_188": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "view_1477": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "convert_element_type_564": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "clamp_max_188": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "clamp_min_283": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "sub_4318": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "round_189": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "div_189": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "convert_element_type_567": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "view_1481": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "mul_9189": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "view_1480": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208"], "mul_9194": ["triton_poi_fused__to_copy_mul_sub_view_10:209"], "sub_4342": ["triton_poi_fused__to_copy_mul_sub_view_10:209"], "convert_element_type_568": ["triton_poi_fused__to_copy_mul_sub_view_10:209"], "view_1483": ["triton_poi_fused__to_copy_mul_sub_view_10:209"], "convert_element_type_569": ["triton_poi_fused__to_copy_mul_sub_view_10:209"], "view_1485": ["triton_poi_fused__to_copy_mul_sub_view_10:209"], "view_1484": ["triton_poi_fused__to_copy_mul_sub_view_10:209"], "amax_95": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "mul_9208": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "mul_9206": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "view_1488": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "add_tensor_81": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "add_14568": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "erf_17": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "mul_9207": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "amin_95": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "add_14651": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "round_192": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "mul_9257": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "mul_9254": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "reciprocal_95": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "view_1491": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "clamp_min_285": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "div_190": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "sub_4355": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "maximum_95": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "full_191": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "minimum_95": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "full_190": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "view_1492": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "convert_element_type_570": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "clamp_max_190": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "clamp_min_286": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "sub_4361": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "round_191": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "div_191": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "mul_9279": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "sub_4381": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "convert_element_type_572": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "convert_element_type_571": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "clamp_max_191": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "clamp_min_287": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "convert_element_type_573": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "view_1496": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "view_1495": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210"], "mul_9284": ["triton_poi_fused__to_copy_mul_sub_view_10:211"], "sub_4385": ["triton_poi_fused__to_copy_mul_sub_view_10:211"], "convert_element_type_574": ["triton_poi_fused__to_copy_mul_sub_view_10:211"], "view_1498": ["triton_poi_fused__to_copy_mul_sub_view_10:211"], "convert_element_type_575": ["triton_poi_fused__to_copy_mul_sub_view_10:211"], "view_1500": ["triton_poi_fused__to_copy_mul_sub_view_10:211"], "view_1499": ["triton_poi_fused__to_copy_mul_sub_view_10:211"], "var_mean_32": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "clone_129": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "add_14714": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "view_1503": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "add_tensor_80": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "add_14720": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "mul_9301": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "mul_9300": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "sub_4391": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "rsqrt_32": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "add_14719": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "amax_96": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "amin_96": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "amax_97": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "amin_97": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "amax_98": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "amin_98": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "sub_4428": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "convert_element_type_578": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "convert_element_type_577": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "clamp_max_193": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "clamp_min_290": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "add_14807": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "round_194": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "mul_9352": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "mul_9349": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "reciprocal_96": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "view_1506": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "clamp_min_288": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "div_192": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "sub_4402": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "maximum_96": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "full_193": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "minimum_96": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "full_192": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "view_1507": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "convert_element_type_576": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "clamp_max_192": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "clamp_min_289": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "sub_4408": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "round_193": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "div_193": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "convert_element_type_579": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "view_1511": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "mul_9374": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "view_1510": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "sub_4473": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "convert_element_type_584": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "convert_element_type_583": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "clamp_max_195": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "clamp_min_293": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "add_14959": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "round_196": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "mul_9448": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "mul_9445": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "reciprocal_97": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "view_1522": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "clamp_min_291": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "div_194": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "sub_4447": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "maximum_97": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "full_195": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "minimum_97": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "full_194": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "view_1523": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "convert_element_type_582": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "clamp_max_194": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "clamp_min_292": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "sub_4453": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "round_195": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "div_195": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "convert_element_type_585": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "view_1527": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "mul_9470": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "view_1526": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "sub_4517": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "convert_element_type_590": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "convert_element_type_589": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "clamp_max_197": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "clamp_min_296": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "add_15107": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "round_198": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "mul_9547": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "mul_9544": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "reciprocal_98": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "view_1538": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "clamp_min_294": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "div_196": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "sub_4491": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "maximum_98": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "full_197": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "minimum_98": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "full_196": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "view_1539": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "convert_element_type_588": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "clamp_max_196": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "clamp_min_295": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "sub_4497": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "round_197": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "div_197": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "convert_element_type_591": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "view_1543": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "mul_9569": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "view_1542": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212"], "mul_9379": ["triton_poi_fused__to_copy_mul_sub_view_4:213"], "sub_4432": ["triton_poi_fused__to_copy_mul_sub_view_4:213"], "convert_element_type_580": ["triton_poi_fused__to_copy_mul_sub_view_4:213"], "view_1513": ["triton_poi_fused__to_copy_mul_sub_view_4:213"], "convert_element_type_581": ["triton_poi_fused__to_copy_mul_sub_view_4:213"], "view_1515": ["triton_poi_fused__to_copy_mul_sub_view_4:213"], "view_1514": ["triton_poi_fused__to_copy_mul_sub_view_4:213"], "mul_9475": ["triton_poi_fused__to_copy_mul_sub_view_4:214"], "sub_4477": ["triton_poi_fused__to_copy_mul_sub_view_4:214"], "convert_element_type_586": ["triton_poi_fused__to_copy_mul_sub_view_4:214"], "view_1529": ["triton_poi_fused__to_copy_mul_sub_view_4:214"], "convert_element_type_587": ["triton_poi_fused__to_copy_mul_sub_view_4:214"], "view_1531": ["triton_poi_fused__to_copy_mul_sub_view_4:214"], "view_1530": ["triton_poi_fused__to_copy_mul_sub_view_4:214"], "mul_9574": ["triton_poi_fused__to_copy_mul_sub_view_4:215"], "sub_4521": ["triton_poi_fused__to_copy_mul_sub_view_4:215"], "convert_element_type_592": ["triton_poi_fused__to_copy_mul_sub_view_4:215"], "view_1545": ["triton_poi_fused__to_copy_mul_sub_view_4:215"], "convert_element_type_593": ["triton_poi_fused__to_copy_mul_sub_view_4:215"], "view_1547": ["triton_poi_fused__to_copy_mul_sub_view_4:215"], "view_1546": ["triton_poi_fused__to_copy_mul_sub_view_4:215"], "_scaled_dot_product_efficient_attention_16": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:216", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:217", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:218", "at::_ops::_scaled_dot_product_efficient_attention::call:542"], "clone_130": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:216", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:217", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:218", "at::_ops::_scaled_dot_product_efficient_attention::call:542"], "permute_162": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:216", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:217", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:218", "at::_ops::_scaled_dot_product_efficient_attention::call:542"], "view_1519": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:216", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:217", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:218", "at::_ops::_scaled_dot_product_efficient_attention::call:542"], "mul_9391": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:216", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:217", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:218", "at::_ops::_scaled_dot_product_efficient_attention::call:542"], "view_1518": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:216", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:217", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:218", "at::_ops::_scaled_dot_product_efficient_attention::call:542"], "add_tensor_79": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:216", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:217", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:218", "at::_ops::_scaled_dot_product_efficient_attention::call:542"], "clone_131": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:216", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:217", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:218", "at::_ops::_scaled_dot_product_efficient_attention::call:542"], "permute_164": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:216", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:217", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:218", "at::_ops::_scaled_dot_product_efficient_attention::call:542"], "view_1535": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:216", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:217", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:218", "at::_ops::_scaled_dot_product_efficient_attention::call:542"], "view_1534": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:216", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:217", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:218", "at::_ops::_scaled_dot_product_efficient_attention::call:542"], "clone_132": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:216", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:217", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:218", "at::_ops::_scaled_dot_product_efficient_attention::call:542"], "permute_166": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:216", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:217", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:218", "at::_ops::_scaled_dot_product_efficient_attention::call:542"], "view_1551": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:216", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:217", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:218", "at::_ops::_scaled_dot_product_efficient_attention::call:542"], "view_1550": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:216", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:217", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:218", "at::_ops::_scaled_dot_product_efficient_attention::call:542"], "add_tensor_78": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:216", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:217", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:218", "at::_ops::_scaled_dot_product_efficient_attention::call:542"], "amax_99": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "view_1552": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "permute_167": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "amin_99": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "sub_4565": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "convert_element_type_596": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "convert_element_type_595": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "clamp_max_199": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "clamp_min_299": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "add_15271": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "round_200": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "mul_9652": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "mul_9649": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "reciprocal_99": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "view_1555": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "clamp_min_297": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "div_198": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "sub_4539": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "maximum_99": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "full_199": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "minimum_99": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "full_198": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "view_1556": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "convert_element_type_594": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "clamp_max_198": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "clamp_min_298": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "sub_4545": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "round_199": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "div_199": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "convert_element_type_597": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "view_1560": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "mul_9674": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "view_1559": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219"], "mul_9679": ["triton_poi_fused__to_copy_mul_sub_view_4:220"], "sub_4569": ["triton_poi_fused__to_copy_mul_sub_view_4:220"], "convert_element_type_598": ["triton_poi_fused__to_copy_mul_sub_view_4:220"], "view_1562": ["triton_poi_fused__to_copy_mul_sub_view_4:220"], "convert_element_type_599": ["triton_poi_fused__to_copy_mul_sub_view_4:220"], "view_1564": ["triton_poi_fused__to_copy_mul_sub_view_4:220"], "view_1563": ["triton_poi_fused__to_copy_mul_sub_view_4:220"], "add_15334": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "view_1567": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "add_tensor_77": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "var_mean_33": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "clone_134": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "add_15340": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "mul_9696": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "mul_9695": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "sub_4575": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "rsqrt_33": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "add_15339": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "amax_100": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "amin_100": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "sub_4612": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "convert_element_type_602": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "convert_element_type_601": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "clamp_max_201": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "clamp_min_302": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "add_15427": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "round_202": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "mul_9747": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "mul_9744": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "reciprocal_100": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "view_1570": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "clamp_min_300": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "div_200": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "sub_4586": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "maximum_100": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "full_201": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "minimum_100": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "full_200": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "view_1571": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "convert_element_type_600": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "clamp_max_200": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "clamp_min_301": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "sub_4592": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "round_201": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "div_201": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "convert_element_type_603": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "view_1575": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "mul_9769": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "view_1574": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221"], "mul_9774": ["triton_poi_fused__to_copy_mul_sub_view_10:222"], "sub_4616": ["triton_poi_fused__to_copy_mul_sub_view_10:222"], "convert_element_type_604": ["triton_poi_fused__to_copy_mul_sub_view_10:222"], "view_1577": ["triton_poi_fused__to_copy_mul_sub_view_10:222"], "convert_element_type_605": ["triton_poi_fused__to_copy_mul_sub_view_10:222"], "view_1579": ["triton_poi_fused__to_copy_mul_sub_view_10:222"], "view_1578": ["triton_poi_fused__to_copy_mul_sub_view_10:222"], "amax_101": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "mul_9788": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "mul_9786": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "view_1582": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "add_tensor_76": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "add_15486": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "erf_18": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "mul_9787": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "amin_101": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "add_15569": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "round_204": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "mul_9837": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "mul_9834": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "reciprocal_101": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "view_1585": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "clamp_min_303": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "div_202": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "sub_4629": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "maximum_101": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "full_203": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "minimum_101": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "full_202": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "view_1586": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "convert_element_type_606": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "clamp_max_202": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "clamp_min_304": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "sub_4635": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "round_203": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "div_203": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "mul_9859": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "sub_4655": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "convert_element_type_608": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "convert_element_type_607": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "clamp_max_203": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "clamp_min_305": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "convert_element_type_609": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "view_1590": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "view_1589": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223"], "mul_9864": ["triton_poi_fused__to_copy_mul_sub_view_10:224"], "sub_4659": ["triton_poi_fused__to_copy_mul_sub_view_10:224"], "convert_element_type_610": ["triton_poi_fused__to_copy_mul_sub_view_10:224"], "view_1592": ["triton_poi_fused__to_copy_mul_sub_view_10:224"], "convert_element_type_611": ["triton_poi_fused__to_copy_mul_sub_view_10:224"], "view_1594": ["triton_poi_fused__to_copy_mul_sub_view_10:224"], "view_1593": ["triton_poi_fused__to_copy_mul_sub_view_10:224"], "var_mean_34": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "clone_137": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "add_15632": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "view_1597": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "add_tensor_75": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "add_15638": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "mul_9881": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "mul_9880": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "sub_4665": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "rsqrt_34": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "add_15637": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "amax_102": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "amin_102": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "amax_103": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "amin_103": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "amax_104": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "amin_104": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "sub_4702": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "convert_element_type_614": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "convert_element_type_613": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "clamp_max_205": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "clamp_min_308": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "add_15725": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "round_206": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "mul_9932": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "mul_9929": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "reciprocal_102": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "view_1600": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "clamp_min_306": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "div_204": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "sub_4676": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "maximum_102": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "full_205": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "minimum_102": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "full_204": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "view_1601": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "convert_element_type_612": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "clamp_max_204": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "clamp_min_307": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "sub_4682": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "round_205": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "div_205": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "convert_element_type_615": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "view_1605": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "mul_9954": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "view_1604": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "sub_4747": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "convert_element_type_620": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "convert_element_type_619": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "clamp_max_207": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "clamp_min_311": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "add_15877": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "round_208": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "mul_10028": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "mul_10025": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "reciprocal_103": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "view_1616": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "clamp_min_309": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "div_206": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "sub_4721": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "maximum_103": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "full_207": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "minimum_103": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "full_206": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "view_1617": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "convert_element_type_618": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "clamp_max_206": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "clamp_min_310": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "sub_4727": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "round_207": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "div_207": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "convert_element_type_621": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "view_1621": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "mul_10050": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "view_1620": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "sub_4791": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "convert_element_type_626": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "convert_element_type_625": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "clamp_max_209": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "clamp_min_314": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "add_16025": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "round_210": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "mul_10127": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "mul_10124": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "reciprocal_104": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "view_1632": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "clamp_min_312": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "div_208": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "sub_4765": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "maximum_104": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "full_209": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "minimum_104": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "full_208": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "view_1633": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "convert_element_type_624": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "clamp_max_208": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "clamp_min_313": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "sub_4771": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "round_209": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "div_209": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "convert_element_type_627": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "view_1637": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "mul_10149": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "view_1636": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225"], "mul_9959": ["triton_poi_fused__to_copy_mul_sub_view_4:226"], "sub_4706": ["triton_poi_fused__to_copy_mul_sub_view_4:226"], "convert_element_type_616": ["triton_poi_fused__to_copy_mul_sub_view_4:226"], "view_1607": ["triton_poi_fused__to_copy_mul_sub_view_4:226"], "convert_element_type_617": ["triton_poi_fused__to_copy_mul_sub_view_4:226"], "view_1609": ["triton_poi_fused__to_copy_mul_sub_view_4:226"], "view_1608": ["triton_poi_fused__to_copy_mul_sub_view_4:226"], "mul_10055": ["triton_poi_fused__to_copy_mul_sub_view_4:227"], "sub_4751": ["triton_poi_fused__to_copy_mul_sub_view_4:227"], "convert_element_type_622": ["triton_poi_fused__to_copy_mul_sub_view_4:227"], "view_1623": ["triton_poi_fused__to_copy_mul_sub_view_4:227"], "convert_element_type_623": ["triton_poi_fused__to_copy_mul_sub_view_4:227"], "view_1625": ["triton_poi_fused__to_copy_mul_sub_view_4:227"], "view_1624": ["triton_poi_fused__to_copy_mul_sub_view_4:227"], "mul_10154": ["triton_poi_fused__to_copy_mul_sub_view_4:228"], "sub_4795": ["triton_poi_fused__to_copy_mul_sub_view_4:228"], "convert_element_type_628": ["triton_poi_fused__to_copy_mul_sub_view_4:228"], "view_1639": ["triton_poi_fused__to_copy_mul_sub_view_4:228"], "convert_element_type_629": ["triton_poi_fused__to_copy_mul_sub_view_4:228"], "view_1641": ["triton_poi_fused__to_copy_mul_sub_view_4:228"], "view_1640": ["triton_poi_fused__to_copy_mul_sub_view_4:228"], "_scaled_dot_product_efficient_attention_17": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:229", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:230", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:231", "at::_ops::_scaled_dot_product_efficient_attention::call:549"], "clone_138": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:229", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:230", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:231", "at::_ops::_scaled_dot_product_efficient_attention::call:549"], "permute_172": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:229", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:230", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:231", "at::_ops::_scaled_dot_product_efficient_attention::call:549"], "view_1613": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:229", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:230", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:231", "at::_ops::_scaled_dot_product_efficient_attention::call:549"], "mul_9971": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:229", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:230", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:231", "at::_ops::_scaled_dot_product_efficient_attention::call:549"], "view_1612": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:229", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:230", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:231", "at::_ops::_scaled_dot_product_efficient_attention::call:549"], "add_tensor_74": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:229", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:230", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:231", "at::_ops::_scaled_dot_product_efficient_attention::call:549"], "clone_139": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:229", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:230", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:231", "at::_ops::_scaled_dot_product_efficient_attention::call:549"], "permute_174": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:229", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:230", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:231", "at::_ops::_scaled_dot_product_efficient_attention::call:549"], "view_1629": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:229", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:230", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:231", "at::_ops::_scaled_dot_product_efficient_attention::call:549"], "view_1628": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:229", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:230", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:231", "at::_ops::_scaled_dot_product_efficient_attention::call:549"], "clone_140": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:229", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:230", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:231", "at::_ops::_scaled_dot_product_efficient_attention::call:549"], "permute_176": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:229", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:230", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:231", "at::_ops::_scaled_dot_product_efficient_attention::call:549"], "view_1645": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:229", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:230", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:231", "at::_ops::_scaled_dot_product_efficient_attention::call:549"], "view_1644": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:229", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:230", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:231", "at::_ops::_scaled_dot_product_efficient_attention::call:549"], "add_tensor_73": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:229", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:230", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:231", "at::_ops::_scaled_dot_product_efficient_attention::call:549"], "amax_105": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "view_1646": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "permute_177": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "amin_105": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "sub_4839": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "convert_element_type_632": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "convert_element_type_631": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "clamp_max_211": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "clamp_min_317": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "add_16189": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "round_212": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "mul_10232": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "mul_10229": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "reciprocal_105": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "view_1649": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "clamp_min_315": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "div_210": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "sub_4813": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "maximum_105": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "full_211": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "minimum_105": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "full_210": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "view_1650": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "convert_element_type_630": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "clamp_max_210": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "clamp_min_316": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "sub_4819": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "round_211": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "div_211": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "convert_element_type_633": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "view_1654": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "mul_10254": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "view_1653": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232"], "mul_10259": ["triton_poi_fused__to_copy_mul_sub_view_4:233"], "sub_4843": ["triton_poi_fused__to_copy_mul_sub_view_4:233"], "convert_element_type_634": ["triton_poi_fused__to_copy_mul_sub_view_4:233"], "view_1656": ["triton_poi_fused__to_copy_mul_sub_view_4:233"], "convert_element_type_635": ["triton_poi_fused__to_copy_mul_sub_view_4:233"], "view_1658": ["triton_poi_fused__to_copy_mul_sub_view_4:233"], "view_1657": ["triton_poi_fused__to_copy_mul_sub_view_4:233"], "add_16252": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "view_1661": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "add_tensor_72": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "var_mean_35": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "clone_142": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "add_16258": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "mul_10276": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "mul_10275": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "sub_4849": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "rsqrt_35": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "add_16257": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "amax_106": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "amin_106": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "sub_4886": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "convert_element_type_638": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "convert_element_type_637": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "clamp_max_213": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "clamp_min_320": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "add_16345": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "round_214": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "mul_10327": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "mul_10324": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "reciprocal_106": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "view_1664": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "clamp_min_318": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "div_212": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "sub_4860": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "maximum_106": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "full_213": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "minimum_106": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "full_212": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "view_1665": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "convert_element_type_636": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "clamp_max_212": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "clamp_min_319": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "sub_4866": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "round_213": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "div_213": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "convert_element_type_639": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "view_1669": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "mul_10349": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "view_1668": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234"], "mul_10354": ["triton_poi_fused__to_copy_mul_sub_view_10:235"], "sub_4890": ["triton_poi_fused__to_copy_mul_sub_view_10:235"], "convert_element_type_640": ["triton_poi_fused__to_copy_mul_sub_view_10:235"], "view_1671": ["triton_poi_fused__to_copy_mul_sub_view_10:235"], "convert_element_type_641": ["triton_poi_fused__to_copy_mul_sub_view_10:235"], "view_1673": ["triton_poi_fused__to_copy_mul_sub_view_10:235"], "view_1672": ["triton_poi_fused__to_copy_mul_sub_view_10:235"], "amax_107": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "mul_10368": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "mul_10366": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "view_1676": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "add_tensor_71": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "add_16404": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "erf_19": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "mul_10367": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "amin_107": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "add_16487": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "round_216": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "mul_10417": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "mul_10414": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "reciprocal_107": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "view_1679": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "clamp_min_321": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "div_214": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "sub_4903": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "maximum_107": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "full_215": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "minimum_107": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "full_214": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "view_1680": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "convert_element_type_642": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "clamp_max_214": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "clamp_min_322": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "sub_4909": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "round_215": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "div_215": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "mul_10439": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "sub_4929": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "convert_element_type_644": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "convert_element_type_643": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "clamp_max_215": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "clamp_min_323": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "convert_element_type_645": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "view_1684": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "view_1683": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236"], "mul_10444": ["triton_poi_fused__to_copy_mul_sub_view_10:237"], "sub_4933": ["triton_poi_fused__to_copy_mul_sub_view_10:237"], "convert_element_type_646": ["triton_poi_fused__to_copy_mul_sub_view_10:237"], "view_1686": ["triton_poi_fused__to_copy_mul_sub_view_10:237"], "convert_element_type_647": ["triton_poi_fused__to_copy_mul_sub_view_10:237"], "view_1688": ["triton_poi_fused__to_copy_mul_sub_view_10:237"], "view_1687": ["triton_poi_fused__to_copy_mul_sub_view_10:237"], "var_mean_36": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "clone_145": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "add_16550": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "view_1691": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "add_tensor_70": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "add_16556": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "mul_10461": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "mul_10460": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "sub_4939": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "rsqrt_36": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "add_16555": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "amax_108": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "amin_108": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "amax_109": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "amin_109": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "amax_110": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "amin_110": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "sub_4976": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "convert_element_type_650": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "convert_element_type_649": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "clamp_max_217": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "clamp_min_326": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "add_16643": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "round_218": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "mul_10512": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "mul_10509": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "reciprocal_108": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "view_1694": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "clamp_min_324": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "div_216": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "sub_4950": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "maximum_108": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "full_217": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "minimum_108": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "full_216": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "view_1695": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "convert_element_type_648": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "clamp_max_216": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "clamp_min_325": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "sub_4956": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "round_217": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "div_217": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "convert_element_type_651": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "view_1699": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "mul_10534": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "view_1698": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "sub_5021": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "convert_element_type_656": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "convert_element_type_655": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "clamp_max_219": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "clamp_min_329": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "add_16795": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "round_220": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "mul_10608": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "mul_10605": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "reciprocal_109": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "view_1710": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "clamp_min_327": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "div_218": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "sub_4995": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "maximum_109": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "full_219": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "minimum_109": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "full_218": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "view_1711": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "convert_element_type_654": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "clamp_max_218": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "clamp_min_328": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "sub_5001": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "round_219": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "div_219": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "convert_element_type_657": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "view_1715": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "mul_10630": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "view_1714": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "sub_5065": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "convert_element_type_662": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "convert_element_type_661": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "clamp_max_221": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "clamp_min_332": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "add_16943": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "round_222": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "mul_10707": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "mul_10704": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "reciprocal_110": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "view_1726": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "clamp_min_330": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "div_220": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "sub_5039": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "maximum_110": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "full_221": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "minimum_110": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "full_220": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "view_1727": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "convert_element_type_660": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "clamp_max_220": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "clamp_min_331": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "sub_5045": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "round_221": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "div_221": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "convert_element_type_663": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "view_1731": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "mul_10729": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "view_1730": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238"], "mul_10539": ["triton_poi_fused__to_copy_mul_sub_view_4:239"], "sub_4980": ["triton_poi_fused__to_copy_mul_sub_view_4:239"], "convert_element_type_652": ["triton_poi_fused__to_copy_mul_sub_view_4:239"], "view_1701": ["triton_poi_fused__to_copy_mul_sub_view_4:239"], "convert_element_type_653": ["triton_poi_fused__to_copy_mul_sub_view_4:239"], "view_1703": ["triton_poi_fused__to_copy_mul_sub_view_4:239"], "view_1702": ["triton_poi_fused__to_copy_mul_sub_view_4:239"], "mul_10635": ["triton_poi_fused__to_copy_mul_sub_view_4:240"], "sub_5025": ["triton_poi_fused__to_copy_mul_sub_view_4:240"], "convert_element_type_658": ["triton_poi_fused__to_copy_mul_sub_view_4:240"], "view_1717": ["triton_poi_fused__to_copy_mul_sub_view_4:240"], "convert_element_type_659": ["triton_poi_fused__to_copy_mul_sub_view_4:240"], "view_1719": ["triton_poi_fused__to_copy_mul_sub_view_4:240"], "view_1718": ["triton_poi_fused__to_copy_mul_sub_view_4:240"], "mul_10734": ["triton_poi_fused__to_copy_mul_sub_view_4:241"], "sub_5069": ["triton_poi_fused__to_copy_mul_sub_view_4:241"], "convert_element_type_664": ["triton_poi_fused__to_copy_mul_sub_view_4:241"], "view_1733": ["triton_poi_fused__to_copy_mul_sub_view_4:241"], "convert_element_type_665": ["triton_poi_fused__to_copy_mul_sub_view_4:241"], "view_1735": ["triton_poi_fused__to_copy_mul_sub_view_4:241"], "view_1734": ["triton_poi_fused__to_copy_mul_sub_view_4:241"], "_scaled_dot_product_efficient_attention_18": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:242", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:243", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:244", "at::_ops::_scaled_dot_product_efficient_attention::call:556"], "clone_146": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:242", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:243", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:244", "at::_ops::_scaled_dot_product_efficient_attention::call:556"], "permute_182": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:242", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:243", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:244", "at::_ops::_scaled_dot_product_efficient_attention::call:556"], "view_1707": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:242", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:243", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:244", "at::_ops::_scaled_dot_product_efficient_attention::call:556"], "mul_10551": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:242", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:243", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:244", "at::_ops::_scaled_dot_product_efficient_attention::call:556"], "view_1706": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:242", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:243", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:244", "at::_ops::_scaled_dot_product_efficient_attention::call:556"], "add_tensor_69": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:242", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:243", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:244", "at::_ops::_scaled_dot_product_efficient_attention::call:556"], "clone_147": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:242", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:243", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:244", "at::_ops::_scaled_dot_product_efficient_attention::call:556"], "permute_184": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:242", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:243", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:244", "at::_ops::_scaled_dot_product_efficient_attention::call:556"], "view_1723": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:242", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:243", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:244", "at::_ops::_scaled_dot_product_efficient_attention::call:556"], "view_1722": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:242", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:243", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:244", "at::_ops::_scaled_dot_product_efficient_attention::call:556"], "clone_148": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:242", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:243", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:244", "at::_ops::_scaled_dot_product_efficient_attention::call:556"], "permute_186": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:242", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:243", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:244", "at::_ops::_scaled_dot_product_efficient_attention::call:556"], "view_1739": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:242", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:243", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:244", "at::_ops::_scaled_dot_product_efficient_attention::call:556"], "view_1738": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:242", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:243", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:244", "at::_ops::_scaled_dot_product_efficient_attention::call:556"], "add_tensor_68": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:242", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:243", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:244", "at::_ops::_scaled_dot_product_efficient_attention::call:556"], "amax_111": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "view_1740": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "permute_187": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "amin_111": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "sub_5113": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "convert_element_type_668": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "convert_element_type_667": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "clamp_max_223": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "clamp_min_335": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "add_17107": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "round_224": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "mul_10812": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "mul_10809": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "reciprocal_111": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "view_1743": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "clamp_min_333": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "div_222": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "sub_5087": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "maximum_111": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "full_223": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "minimum_111": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "full_222": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "view_1744": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "convert_element_type_666": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "clamp_max_222": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "clamp_min_334": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "sub_5093": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "round_223": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "div_223": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "convert_element_type_669": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "view_1748": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "mul_10834": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "view_1747": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245"], "mul_10839": ["triton_poi_fused__to_copy_mul_sub_view_4:246"], "sub_5117": ["triton_poi_fused__to_copy_mul_sub_view_4:246"], "convert_element_type_670": ["triton_poi_fused__to_copy_mul_sub_view_4:246"], "view_1750": ["triton_poi_fused__to_copy_mul_sub_view_4:246"], "convert_element_type_671": ["triton_poi_fused__to_copy_mul_sub_view_4:246"], "view_1752": ["triton_poi_fused__to_copy_mul_sub_view_4:246"], "view_1751": ["triton_poi_fused__to_copy_mul_sub_view_4:246"], "add_17170": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "view_1755": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "add_tensor_67": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "var_mean_37": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "clone_150": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "add_17176": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "mul_10856": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "mul_10855": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "sub_5123": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "rsqrt_37": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "add_17175": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "amax_112": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "amin_112": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "sub_5160": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "convert_element_type_674": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "convert_element_type_673": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "clamp_max_225": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "clamp_min_338": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "add_17263": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "round_226": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "mul_10907": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "mul_10904": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "reciprocal_112": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "view_1758": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "clamp_min_336": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "div_224": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "sub_5134": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "maximum_112": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "full_225": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "minimum_112": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "full_224": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "view_1759": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "convert_element_type_672": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "clamp_max_224": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "clamp_min_337": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "sub_5140": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "round_225": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "div_225": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "convert_element_type_675": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "view_1763": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "mul_10929": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "view_1762": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247"], "mul_10934": ["triton_poi_fused__to_copy_mul_sub_view_10:248"], "sub_5164": ["triton_poi_fused__to_copy_mul_sub_view_10:248"], "convert_element_type_676": ["triton_poi_fused__to_copy_mul_sub_view_10:248"], "view_1765": ["triton_poi_fused__to_copy_mul_sub_view_10:248"], "convert_element_type_677": ["triton_poi_fused__to_copy_mul_sub_view_10:248"], "view_1767": ["triton_poi_fused__to_copy_mul_sub_view_10:248"], "view_1766": ["triton_poi_fused__to_copy_mul_sub_view_10:248"], "amax_113": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "mul_10948": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "mul_10946": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "view_1770": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "add_tensor_66": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "add_17322": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "erf_20": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "mul_10947": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "amin_113": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "add_17405": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "round_228": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "mul_10997": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "mul_10994": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "reciprocal_113": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "view_1773": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "clamp_min_339": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "div_226": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "sub_5177": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "maximum_113": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "full_227": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "minimum_113": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "full_226": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "view_1774": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "convert_element_type_678": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "clamp_max_226": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "clamp_min_340": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "sub_5183": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "round_227": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "div_227": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "mul_11019": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "sub_5203": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "convert_element_type_680": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "convert_element_type_679": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "clamp_max_227": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "clamp_min_341": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "convert_element_type_681": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "view_1778": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "view_1777": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249"], "mul_11024": ["triton_poi_fused__to_copy_mul_sub_view_10:250"], "sub_5207": ["triton_poi_fused__to_copy_mul_sub_view_10:250"], "convert_element_type_682": ["triton_poi_fused__to_copy_mul_sub_view_10:250"], "view_1780": ["triton_poi_fused__to_copy_mul_sub_view_10:250"], "convert_element_type_683": ["triton_poi_fused__to_copy_mul_sub_view_10:250"], "view_1782": ["triton_poi_fused__to_copy_mul_sub_view_10:250"], "view_1781": ["triton_poi_fused__to_copy_mul_sub_view_10:250"], "var_mean_38": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "clone_153": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "add_17468": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "view_1785": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "add_tensor_65": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "add_17474": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "mul_11041": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "mul_11040": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "sub_5213": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "rsqrt_38": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "add_17473": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "amax_114": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "amin_114": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "amax_115": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "amin_115": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "amax_116": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "amin_116": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "sub_5250": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "convert_element_type_686": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "convert_element_type_685": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "clamp_max_229": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "clamp_min_344": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "add_17561": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "round_230": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "mul_11092": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "mul_11089": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "reciprocal_114": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "view_1788": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "clamp_min_342": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "div_228": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "sub_5224": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "maximum_114": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "full_229": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "minimum_114": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "full_228": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "view_1789": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "convert_element_type_684": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "clamp_max_228": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "clamp_min_343": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "sub_5230": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "round_229": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "div_229": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "convert_element_type_687": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "view_1793": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "mul_11114": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "view_1792": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "sub_5295": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "convert_element_type_692": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "convert_element_type_691": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "clamp_max_231": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "clamp_min_347": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "add_17713": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "round_232": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "mul_11188": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "mul_11185": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "reciprocal_115": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "view_1804": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "clamp_min_345": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "div_230": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "sub_5269": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "maximum_115": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "full_231": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "minimum_115": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "full_230": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "view_1805": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "convert_element_type_690": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "clamp_max_230": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "clamp_min_346": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "sub_5275": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "round_231": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "div_231": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "convert_element_type_693": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "view_1809": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "mul_11210": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "view_1808": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "sub_5339": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "convert_element_type_698": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "convert_element_type_697": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "clamp_max_233": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "clamp_min_350": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "add_17861": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "round_234": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "mul_11287": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "mul_11284": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "reciprocal_116": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "view_1820": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "clamp_min_348": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "div_232": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "sub_5313": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "maximum_116": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "full_233": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "minimum_116": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "full_232": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "view_1821": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "convert_element_type_696": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "clamp_max_232": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "clamp_min_349": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "sub_5319": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "round_233": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "div_233": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "convert_element_type_699": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "view_1825": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "mul_11309": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "view_1824": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251"], "mul_11119": ["triton_poi_fused__to_copy_mul_sub_view_4:252"], "sub_5254": ["triton_poi_fused__to_copy_mul_sub_view_4:252"], "convert_element_type_688": ["triton_poi_fused__to_copy_mul_sub_view_4:252"], "view_1795": ["triton_poi_fused__to_copy_mul_sub_view_4:252"], "convert_element_type_689": ["triton_poi_fused__to_copy_mul_sub_view_4:252"], "view_1797": ["triton_poi_fused__to_copy_mul_sub_view_4:252"], "view_1796": ["triton_poi_fused__to_copy_mul_sub_view_4:252"], "mul_11215": ["triton_poi_fused__to_copy_mul_sub_view_4:253"], "sub_5299": ["triton_poi_fused__to_copy_mul_sub_view_4:253"], "convert_element_type_694": ["triton_poi_fused__to_copy_mul_sub_view_4:253"], "view_1811": ["triton_poi_fused__to_copy_mul_sub_view_4:253"], "convert_element_type_695": ["triton_poi_fused__to_copy_mul_sub_view_4:253"], "view_1813": ["triton_poi_fused__to_copy_mul_sub_view_4:253"], "view_1812": ["triton_poi_fused__to_copy_mul_sub_view_4:253"], "mul_11314": ["triton_poi_fused__to_copy_mul_sub_view_4:254"], "sub_5343": ["triton_poi_fused__to_copy_mul_sub_view_4:254"], "convert_element_type_700": ["triton_poi_fused__to_copy_mul_sub_view_4:254"], "view_1827": ["triton_poi_fused__to_copy_mul_sub_view_4:254"], "convert_element_type_701": ["triton_poi_fused__to_copy_mul_sub_view_4:254"], "view_1829": ["triton_poi_fused__to_copy_mul_sub_view_4:254"], "view_1828": ["triton_poi_fused__to_copy_mul_sub_view_4:254"], "_scaled_dot_product_efficient_attention_19": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:255", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:256", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:257", "at::_ops::_scaled_dot_product_efficient_attention::call:563"], "clone_154": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:255", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:256", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:257", "at::_ops::_scaled_dot_product_efficient_attention::call:563"], "permute_192": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:255", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:256", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:257", "at::_ops::_scaled_dot_product_efficient_attention::call:563"], "view_1801": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:255", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:256", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:257", "at::_ops::_scaled_dot_product_efficient_attention::call:563"], "mul_11131": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:255", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:256", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:257", "at::_ops::_scaled_dot_product_efficient_attention::call:563"], "view_1800": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:255", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:256", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:257", "at::_ops::_scaled_dot_product_efficient_attention::call:563"], "add_tensor_64": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:255", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:256", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:257", "at::_ops::_scaled_dot_product_efficient_attention::call:563"], "clone_155": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:255", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:256", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:257", "at::_ops::_scaled_dot_product_efficient_attention::call:563"], "permute_194": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:255", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:256", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:257", "at::_ops::_scaled_dot_product_efficient_attention::call:563"], "view_1817": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:255", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:256", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:257", "at::_ops::_scaled_dot_product_efficient_attention::call:563"], "view_1816": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:255", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:256", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:257", "at::_ops::_scaled_dot_product_efficient_attention::call:563"], "clone_156": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:255", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:256", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:257", "at::_ops::_scaled_dot_product_efficient_attention::call:563"], "permute_196": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:255", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:256", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:257", "at::_ops::_scaled_dot_product_efficient_attention::call:563"], "view_1833": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:255", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:256", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:257", "at::_ops::_scaled_dot_product_efficient_attention::call:563"], "view_1832": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:255", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:256", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:257", "at::_ops::_scaled_dot_product_efficient_attention::call:563"], "add_tensor_63": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:255", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:256", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:257", "at::_ops::_scaled_dot_product_efficient_attention::call:563"], "amax_117": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "view_1834": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "permute_197": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "amin_117": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "sub_5387": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "convert_element_type_704": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "convert_element_type_703": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "clamp_max_235": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "clamp_min_353": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "add_18025": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "round_236": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "mul_11392": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "mul_11389": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "reciprocal_117": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "view_1837": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "clamp_min_351": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "div_234": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "sub_5361": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "maximum_117": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "full_235": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "minimum_117": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "full_234": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "view_1838": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "convert_element_type_702": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "clamp_max_234": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "clamp_min_352": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "sub_5367": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "round_235": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "div_235": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "convert_element_type_705": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "view_1842": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "mul_11414": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "view_1841": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258"], "mul_11419": ["triton_poi_fused__to_copy_mul_sub_view_4:259"], "sub_5391": ["triton_poi_fused__to_copy_mul_sub_view_4:259"], "convert_element_type_706": ["triton_poi_fused__to_copy_mul_sub_view_4:259"], "view_1844": ["triton_poi_fused__to_copy_mul_sub_view_4:259"], "convert_element_type_707": ["triton_poi_fused__to_copy_mul_sub_view_4:259"], "view_1846": ["triton_poi_fused__to_copy_mul_sub_view_4:259"], "view_1845": ["triton_poi_fused__to_copy_mul_sub_view_4:259"], "add_18088": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "view_1849": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "add_tensor_62": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "var_mean_39": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "clone_158": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "add_18094": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "mul_11436": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "mul_11435": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "sub_5397": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "rsqrt_39": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "add_18093": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "amax_118": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "amin_118": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "sub_5434": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "convert_element_type_710": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "convert_element_type_709": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "clamp_max_237": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "clamp_min_356": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "add_18181": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "round_238": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "mul_11487": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "mul_11484": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "reciprocal_118": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "view_1852": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "clamp_min_354": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "div_236": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "sub_5408": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "maximum_118": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "full_237": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "minimum_118": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "full_236": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "view_1853": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "convert_element_type_708": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "clamp_max_236": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "clamp_min_355": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "sub_5414": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "round_237": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "div_237": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "convert_element_type_711": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "view_1857": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "mul_11509": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "view_1856": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260"], "mul_11514": ["triton_poi_fused__to_copy_mul_sub_view_10:261"], "sub_5438": ["triton_poi_fused__to_copy_mul_sub_view_10:261"], "convert_element_type_712": ["triton_poi_fused__to_copy_mul_sub_view_10:261"], "view_1859": ["triton_poi_fused__to_copy_mul_sub_view_10:261"], "convert_element_type_713": ["triton_poi_fused__to_copy_mul_sub_view_10:261"], "view_1861": ["triton_poi_fused__to_copy_mul_sub_view_10:261"], "view_1860": ["triton_poi_fused__to_copy_mul_sub_view_10:261"], "amax_119": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "mul_11528": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "mul_11526": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "view_1864": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "add_tensor_61": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "add_18240": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "erf_21": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "mul_11527": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "amin_119": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "add_18323": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "round_240": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "mul_11577": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "mul_11574": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "reciprocal_119": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "view_1867": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "clamp_min_357": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "div_238": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "sub_5451": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "maximum_119": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "full_239": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "minimum_119": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "full_238": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "view_1868": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "convert_element_type_714": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "clamp_max_238": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "clamp_min_358": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "sub_5457": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "round_239": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "div_239": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "mul_11599": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "sub_5477": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "convert_element_type_716": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "convert_element_type_715": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "clamp_max_239": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "clamp_min_359": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "convert_element_type_717": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "view_1872": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "view_1871": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262"], "mul_11604": ["triton_poi_fused__to_copy_mul_sub_view_10:263"], "sub_5481": ["triton_poi_fused__to_copy_mul_sub_view_10:263"], "convert_element_type_718": ["triton_poi_fused__to_copy_mul_sub_view_10:263"], "view_1874": ["triton_poi_fused__to_copy_mul_sub_view_10:263"], "convert_element_type_719": ["triton_poi_fused__to_copy_mul_sub_view_10:263"], "view_1876": ["triton_poi_fused__to_copy_mul_sub_view_10:263"], "view_1875": ["triton_poi_fused__to_copy_mul_sub_view_10:263"], "var_mean_40": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "clone_161": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "add_18386": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "view_1879": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "add_tensor_60": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "add_18392": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "mul_11621": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "mul_11620": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "sub_5487": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "rsqrt_40": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "add_18391": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "amax_120": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "amin_120": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "amax_121": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "amin_121": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "amax_122": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "amin_122": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "sub_5524": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "convert_element_type_722": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "convert_element_type_721": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "clamp_max_241": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "clamp_min_362": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "add_18479": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "round_242": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "mul_11672": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "mul_11669": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "reciprocal_120": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "view_1882": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "clamp_min_360": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "div_240": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "sub_5498": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "maximum_120": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "full_241": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "minimum_120": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "full_240": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "view_1883": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "convert_element_type_720": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "clamp_max_240": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "clamp_min_361": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "sub_5504": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "round_241": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "div_241": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "convert_element_type_723": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "view_1887": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "mul_11694": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "view_1886": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "sub_5569": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "convert_element_type_728": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "convert_element_type_727": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "clamp_max_243": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "clamp_min_365": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "add_18631": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "round_244": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "mul_11768": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "mul_11765": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "reciprocal_121": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "view_1898": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "clamp_min_363": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "div_242": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "sub_5543": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "maximum_121": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "full_243": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "minimum_121": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "full_242": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "view_1899": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "convert_element_type_726": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "clamp_max_242": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "clamp_min_364": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "sub_5549": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "round_243": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "div_243": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "convert_element_type_729": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "view_1903": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "mul_11790": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "view_1902": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "sub_5613": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "convert_element_type_734": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "convert_element_type_733": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "clamp_max_245": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "clamp_min_368": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "add_18779": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "round_246": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "mul_11867": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "mul_11864": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "reciprocal_122": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "view_1914": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "clamp_min_366": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "div_244": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "sub_5587": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "maximum_122": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "full_245": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "minimum_122": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "full_244": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "view_1915": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "convert_element_type_732": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "clamp_max_244": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "clamp_min_367": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "sub_5593": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "round_245": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "div_245": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "convert_element_type_735": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "view_1919": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "mul_11889": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "view_1918": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264"], "mul_11699": ["triton_poi_fused__to_copy_mul_sub_view_4:265"], "sub_5528": ["triton_poi_fused__to_copy_mul_sub_view_4:265"], "convert_element_type_724": ["triton_poi_fused__to_copy_mul_sub_view_4:265"], "view_1889": ["triton_poi_fused__to_copy_mul_sub_view_4:265"], "convert_element_type_725": ["triton_poi_fused__to_copy_mul_sub_view_4:265"], "view_1891": ["triton_poi_fused__to_copy_mul_sub_view_4:265"], "view_1890": ["triton_poi_fused__to_copy_mul_sub_view_4:265"], "mul_11795": ["triton_poi_fused__to_copy_mul_sub_view_4:266"], "sub_5573": ["triton_poi_fused__to_copy_mul_sub_view_4:266"], "convert_element_type_730": ["triton_poi_fused__to_copy_mul_sub_view_4:266"], "view_1905": ["triton_poi_fused__to_copy_mul_sub_view_4:266"], "convert_element_type_731": ["triton_poi_fused__to_copy_mul_sub_view_4:266"], "view_1907": ["triton_poi_fused__to_copy_mul_sub_view_4:266"], "view_1906": ["triton_poi_fused__to_copy_mul_sub_view_4:266"], "mul_11894": ["triton_poi_fused__to_copy_mul_sub_view_4:267"], "sub_5617": ["triton_poi_fused__to_copy_mul_sub_view_4:267"], "convert_element_type_736": ["triton_poi_fused__to_copy_mul_sub_view_4:267"], "view_1921": ["triton_poi_fused__to_copy_mul_sub_view_4:267"], "convert_element_type_737": ["triton_poi_fused__to_copy_mul_sub_view_4:267"], "view_1923": ["triton_poi_fused__to_copy_mul_sub_view_4:267"], "view_1922": ["triton_poi_fused__to_copy_mul_sub_view_4:267"], "_scaled_dot_product_efficient_attention_20": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:268", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:269", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:270", "at::_ops::_scaled_dot_product_efficient_attention::call:570"], "clone_162": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:268", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:269", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:270", "at::_ops::_scaled_dot_product_efficient_attention::call:570"], "permute_202": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:268", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:269", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:270", "at::_ops::_scaled_dot_product_efficient_attention::call:570"], "view_1895": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:268", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:269", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:270", "at::_ops::_scaled_dot_product_efficient_attention::call:570"], "mul_11711": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:268", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:269", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:270", "at::_ops::_scaled_dot_product_efficient_attention::call:570"], "view_1894": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:268", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:269", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:270", "at::_ops::_scaled_dot_product_efficient_attention::call:570"], "add_tensor_59": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:268", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:269", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:270", "at::_ops::_scaled_dot_product_efficient_attention::call:570"], "clone_163": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:268", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:269", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:270", "at::_ops::_scaled_dot_product_efficient_attention::call:570"], "permute_204": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:268", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:269", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:270", "at::_ops::_scaled_dot_product_efficient_attention::call:570"], "view_1911": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:268", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:269", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:270", "at::_ops::_scaled_dot_product_efficient_attention::call:570"], "view_1910": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:268", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:269", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:270", "at::_ops::_scaled_dot_product_efficient_attention::call:570"], "clone_164": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:268", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:269", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:270", "at::_ops::_scaled_dot_product_efficient_attention::call:570"], "permute_206": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:268", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:269", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:270", "at::_ops::_scaled_dot_product_efficient_attention::call:570"], "view_1927": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:268", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:269", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:270", "at::_ops::_scaled_dot_product_efficient_attention::call:570"], "view_1926": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:268", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:269", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:270", "at::_ops::_scaled_dot_product_efficient_attention::call:570"], "add_tensor_58": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:268", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:269", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:270", "at::_ops::_scaled_dot_product_efficient_attention::call:570"], "amax_123": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "view_1928": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "permute_207": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "amin_123": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "sub_5661": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "convert_element_type_740": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "convert_element_type_739": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "clamp_max_247": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "clamp_min_371": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "add_18943": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "round_248": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "mul_11972": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "mul_11969": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "reciprocal_123": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "view_1931": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "clamp_min_369": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "div_246": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "sub_5635": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "maximum_123": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "full_247": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "minimum_123": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "full_246": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "view_1932": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "convert_element_type_738": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "clamp_max_246": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "clamp_min_370": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "sub_5641": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "round_247": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "div_247": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "convert_element_type_741": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "view_1936": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "mul_11994": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "view_1935": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271"], "mul_11999": ["triton_poi_fused__to_copy_mul_sub_view_4:272"], "sub_5665": ["triton_poi_fused__to_copy_mul_sub_view_4:272"], "convert_element_type_742": ["triton_poi_fused__to_copy_mul_sub_view_4:272"], "view_1938": ["triton_poi_fused__to_copy_mul_sub_view_4:272"], "convert_element_type_743": ["triton_poi_fused__to_copy_mul_sub_view_4:272"], "view_1940": ["triton_poi_fused__to_copy_mul_sub_view_4:272"], "view_1939": ["triton_poi_fused__to_copy_mul_sub_view_4:272"], "add_19006": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "view_1943": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "add_tensor_57": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "var_mean_41": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "clone_166": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "add_19012": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "mul_12016": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "mul_12015": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "sub_5671": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "rsqrt_41": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "add_19011": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "amax_124": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "amin_124": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "sub_5708": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "convert_element_type_746": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "convert_element_type_745": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "clamp_max_249": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "clamp_min_374": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "add_19099": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "round_250": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "mul_12067": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "mul_12064": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "reciprocal_124": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "view_1946": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "clamp_min_372": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "div_248": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "sub_5682": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "maximum_124": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "full_249": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "minimum_124": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "full_248": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "view_1947": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "convert_element_type_744": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "clamp_max_248": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "clamp_min_373": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "sub_5688": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "round_249": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "div_249": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "convert_element_type_747": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "view_1951": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "mul_12089": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "view_1950": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273"], "mul_12094": ["triton_poi_fused__to_copy_mul_sub_view_10:274"], "sub_5712": ["triton_poi_fused__to_copy_mul_sub_view_10:274"], "convert_element_type_748": ["triton_poi_fused__to_copy_mul_sub_view_10:274"], "view_1953": ["triton_poi_fused__to_copy_mul_sub_view_10:274"], "convert_element_type_749": ["triton_poi_fused__to_copy_mul_sub_view_10:274"], "view_1955": ["triton_poi_fused__to_copy_mul_sub_view_10:274"], "view_1954": ["triton_poi_fused__to_copy_mul_sub_view_10:274"], "amax_125": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "mul_12108": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "mul_12106": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "view_1958": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "add_tensor_56": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "add_19158": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "erf_22": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "mul_12107": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "amin_125": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "add_19241": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "round_252": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "mul_12157": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "mul_12154": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "reciprocal_125": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "view_1961": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "clamp_min_375": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "div_250": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "sub_5725": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "maximum_125": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "full_251": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "minimum_125": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "full_250": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "view_1962": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "convert_element_type_750": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "clamp_max_250": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "clamp_min_376": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "sub_5731": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "round_251": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "div_251": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "mul_12179": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "sub_5751": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "convert_element_type_752": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "convert_element_type_751": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "clamp_max_251": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "clamp_min_377": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "convert_element_type_753": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "view_1966": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "view_1965": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275"], "mul_12184": ["triton_poi_fused__to_copy_mul_sub_view_10:276"], "sub_5755": ["triton_poi_fused__to_copy_mul_sub_view_10:276"], "convert_element_type_754": ["triton_poi_fused__to_copy_mul_sub_view_10:276"], "view_1968": ["triton_poi_fused__to_copy_mul_sub_view_10:276"], "convert_element_type_755": ["triton_poi_fused__to_copy_mul_sub_view_10:276"], "view_1970": ["triton_poi_fused__to_copy_mul_sub_view_10:276"], "view_1969": ["triton_poi_fused__to_copy_mul_sub_view_10:276"], "var_mean_42": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "clone_169": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "add_19304": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "view_1973": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "add_tensor_55": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "add_19310": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "mul_12201": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "mul_12200": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "sub_5761": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "rsqrt_42": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "add_19309": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "amax_126": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "amin_126": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "amax_127": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "amin_127": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "amax_128": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "amin_128": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "sub_5798": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "convert_element_type_758": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "convert_element_type_757": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "clamp_max_253": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "clamp_min_380": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "add_19397": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "round_254": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "mul_12252": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "mul_12249": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "reciprocal_126": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "view_1976": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "clamp_min_378": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "div_252": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "sub_5772": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "maximum_126": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "full_253": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "minimum_126": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "full_252": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "view_1977": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "convert_element_type_756": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "clamp_max_252": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "clamp_min_379": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "sub_5778": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "round_253": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "div_253": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "convert_element_type_759": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "view_1981": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "mul_12274": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "view_1980": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "sub_5843": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "convert_element_type_764": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "convert_element_type_763": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "clamp_max_255": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "clamp_min_383": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "add_19549": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "round_256": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "mul_12348": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "mul_12345": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "reciprocal_127": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "view_1992": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "clamp_min_381": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "div_254": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "sub_5817": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "maximum_127": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "full_255": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "minimum_127": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "full_254": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "view_1993": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "convert_element_type_762": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "clamp_max_254": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "clamp_min_382": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "sub_5823": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "round_255": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "div_255": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "convert_element_type_765": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "view_1997": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "mul_12370": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "view_1996": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "sub_5887": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "convert_element_type_770": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "convert_element_type_769": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "clamp_max_257": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "clamp_min_386": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "add_19697": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "round_258": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "mul_12447": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "mul_12444": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "reciprocal_128": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "view_2008": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "clamp_min_384": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "div_256": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "sub_5861": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "maximum_128": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "full_257": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "minimum_128": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "full_256": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "view_2009": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "convert_element_type_768": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "clamp_max_256": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "clamp_min_385": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "sub_5867": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "round_257": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "div_257": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "convert_element_type_771": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "view_2013": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "mul_12469": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "view_2012": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277"], "mul_12279": ["triton_poi_fused__to_copy_mul_sub_view_4:278"], "sub_5802": ["triton_poi_fused__to_copy_mul_sub_view_4:278"], "convert_element_type_760": ["triton_poi_fused__to_copy_mul_sub_view_4:278"], "view_1983": ["triton_poi_fused__to_copy_mul_sub_view_4:278"], "convert_element_type_761": ["triton_poi_fused__to_copy_mul_sub_view_4:278"], "view_1985": ["triton_poi_fused__to_copy_mul_sub_view_4:278"], "view_1984": ["triton_poi_fused__to_copy_mul_sub_view_4:278"], "mul_12375": ["triton_poi_fused__to_copy_mul_sub_view_4:279"], "sub_5847": ["triton_poi_fused__to_copy_mul_sub_view_4:279"], "convert_element_type_766": ["triton_poi_fused__to_copy_mul_sub_view_4:279"], "view_1999": ["triton_poi_fused__to_copy_mul_sub_view_4:279"], "convert_element_type_767": ["triton_poi_fused__to_copy_mul_sub_view_4:279"], "view_2001": ["triton_poi_fused__to_copy_mul_sub_view_4:279"], "view_2000": ["triton_poi_fused__to_copy_mul_sub_view_4:279"], "mul_12474": ["triton_poi_fused__to_copy_mul_sub_view_4:280"], "sub_5891": ["triton_poi_fused__to_copy_mul_sub_view_4:280"], "convert_element_type_772": ["triton_poi_fused__to_copy_mul_sub_view_4:280"], "view_2015": ["triton_poi_fused__to_copy_mul_sub_view_4:280"], "convert_element_type_773": ["triton_poi_fused__to_copy_mul_sub_view_4:280"], "view_2017": ["triton_poi_fused__to_copy_mul_sub_view_4:280"], "view_2016": ["triton_poi_fused__to_copy_mul_sub_view_4:280"], "_scaled_dot_product_efficient_attention_21": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:281", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:282", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:283", "at::_ops::_scaled_dot_product_efficient_attention::call:577"], "clone_170": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:281", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:282", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:283", "at::_ops::_scaled_dot_product_efficient_attention::call:577"], "permute_212": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:281", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:282", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:283", "at::_ops::_scaled_dot_product_efficient_attention::call:577"], "view_1989": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:281", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:282", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:283", "at::_ops::_scaled_dot_product_efficient_attention::call:577"], "mul_12291": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:281", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:282", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:283", "at::_ops::_scaled_dot_product_efficient_attention::call:577"], "view_1988": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:281", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:282", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:283", "at::_ops::_scaled_dot_product_efficient_attention::call:577"], "add_tensor_54": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:281", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:282", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:283", "at::_ops::_scaled_dot_product_efficient_attention::call:577"], "clone_171": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:281", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:282", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:283", "at::_ops::_scaled_dot_product_efficient_attention::call:577"], "permute_214": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:281", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:282", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:283", "at::_ops::_scaled_dot_product_efficient_attention::call:577"], "view_2005": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:281", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:282", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:283", "at::_ops::_scaled_dot_product_efficient_attention::call:577"], "view_2004": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:281", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:282", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:283", "at::_ops::_scaled_dot_product_efficient_attention::call:577"], "clone_172": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:281", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:282", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:283", "at::_ops::_scaled_dot_product_efficient_attention::call:577"], "permute_216": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:281", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:282", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:283", "at::_ops::_scaled_dot_product_efficient_attention::call:577"], "view_2021": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:281", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:282", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:283", "at::_ops::_scaled_dot_product_efficient_attention::call:577"], "view_2020": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:281", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:282", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:283", "at::_ops::_scaled_dot_product_efficient_attention::call:577"], "add_tensor_53": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:281", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:282", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:283", "at::_ops::_scaled_dot_product_efficient_attention::call:577"], "amax_129": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "view_2022": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "permute_217": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "amin_129": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "sub_5935": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "convert_element_type_776": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "convert_element_type_775": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "clamp_max_259": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "clamp_min_389": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "add_19861": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "round_260": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "mul_12552": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "mul_12549": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "reciprocal_129": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "view_2025": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "clamp_min_387": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "div_258": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "sub_5909": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "maximum_129": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "full_259": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "minimum_129": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "full_258": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "view_2026": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "convert_element_type_774": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "clamp_max_258": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "clamp_min_388": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "sub_5915": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "round_259": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "div_259": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "convert_element_type_777": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "view_2030": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "mul_12574": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "view_2029": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284"], "mul_12579": ["triton_poi_fused__to_copy_mul_sub_view_4:285"], "sub_5939": ["triton_poi_fused__to_copy_mul_sub_view_4:285"], "convert_element_type_778": ["triton_poi_fused__to_copy_mul_sub_view_4:285"], "view_2032": ["triton_poi_fused__to_copy_mul_sub_view_4:285"], "convert_element_type_779": ["triton_poi_fused__to_copy_mul_sub_view_4:285"], "view_2034": ["triton_poi_fused__to_copy_mul_sub_view_4:285"], "view_2033": ["triton_poi_fused__to_copy_mul_sub_view_4:285"], "add_19924": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "view_2037": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "add_tensor_52": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "var_mean_43": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "clone_174": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "add_19930": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "mul_12596": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "mul_12595": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "sub_5945": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "rsqrt_43": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "add_19929": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "amax_130": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "amin_130": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "sub_5982": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "convert_element_type_782": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "convert_element_type_781": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "clamp_max_261": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "clamp_min_392": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "add_20017": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "round_262": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "mul_12647": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "mul_12644": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "reciprocal_130": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "view_2040": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "clamp_min_390": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "div_260": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "sub_5956": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "maximum_130": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "full_261": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "minimum_130": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "full_260": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "view_2041": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "convert_element_type_780": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "clamp_max_260": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "clamp_min_391": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "sub_5962": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "round_261": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "div_261": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "convert_element_type_783": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "view_2045": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "mul_12669": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "view_2044": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286"], "mul_12674": ["triton_poi_fused__to_copy_mul_sub_view_10:287"], "sub_5986": ["triton_poi_fused__to_copy_mul_sub_view_10:287"], "convert_element_type_784": ["triton_poi_fused__to_copy_mul_sub_view_10:287"], "view_2047": ["triton_poi_fused__to_copy_mul_sub_view_10:287"], "convert_element_type_785": ["triton_poi_fused__to_copy_mul_sub_view_10:287"], "view_2049": ["triton_poi_fused__to_copy_mul_sub_view_10:287"], "view_2048": ["triton_poi_fused__to_copy_mul_sub_view_10:287"], "amax_131": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "mul_12688": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "mul_12686": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "view_2052": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "add_tensor_51": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "add_20076": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "erf_23": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "mul_12687": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "amin_131": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "add_20159": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "round_264": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "mul_12737": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "mul_12734": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "reciprocal_131": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "view_2055": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "clamp_min_393": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "div_262": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "sub_5999": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "maximum_131": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "full_263": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "minimum_131": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "full_262": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "view_2056": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "convert_element_type_786": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "clamp_max_262": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "clamp_min_394": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "sub_6005": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "round_263": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "div_263": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "mul_12759": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "sub_6025": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "convert_element_type_788": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "convert_element_type_787": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "clamp_max_263": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "clamp_min_395": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "convert_element_type_789": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "view_2060": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "view_2059": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288"], "mul_12764": ["triton_poi_fused__to_copy_mul_sub_view_10:289"], "sub_6029": ["triton_poi_fused__to_copy_mul_sub_view_10:289"], "convert_element_type_790": ["triton_poi_fused__to_copy_mul_sub_view_10:289"], "view_2062": ["triton_poi_fused__to_copy_mul_sub_view_10:289"], "convert_element_type_791": ["triton_poi_fused__to_copy_mul_sub_view_10:289"], "view_2064": ["triton_poi_fused__to_copy_mul_sub_view_10:289"], "view_2063": ["triton_poi_fused__to_copy_mul_sub_view_10:289"], "var_mean_44": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "clone_177": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "add_20222": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "view_2067": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "add_tensor_50": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "add_20228": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "mul_12781": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "mul_12780": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "sub_6035": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "rsqrt_44": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "add_20227": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "amax_132": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "amin_132": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "amax_133": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "amin_133": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "amax_134": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "amin_134": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "sub_6072": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "convert_element_type_794": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "convert_element_type_793": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "clamp_max_265": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "clamp_min_398": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "add_20315": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "round_266": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "mul_12832": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "mul_12829": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "reciprocal_132": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "view_2070": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "clamp_min_396": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "div_264": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "sub_6046": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "maximum_132": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "full_265": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "minimum_132": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "full_264": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "view_2071": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "convert_element_type_792": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "clamp_max_264": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "clamp_min_397": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "sub_6052": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "round_265": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "div_265": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "convert_element_type_795": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "view_2075": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "mul_12854": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "view_2074": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "sub_6117": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "convert_element_type_800": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "convert_element_type_799": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "clamp_max_267": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "clamp_min_401": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "add_20467": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "round_268": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "mul_12928": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "mul_12925": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "reciprocal_133": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "view_2086": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "clamp_min_399": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "div_266": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "sub_6091": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "maximum_133": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "full_267": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "minimum_133": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "full_266": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "view_2087": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "convert_element_type_798": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "clamp_max_266": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "clamp_min_400": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "sub_6097": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "round_267": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "div_267": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "convert_element_type_801": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "view_2091": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "mul_12950": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "view_2090": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "sub_6161": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "convert_element_type_806": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "convert_element_type_805": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "clamp_max_269": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "clamp_min_404": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "add_20615": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "round_270": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "mul_13027": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "mul_13024": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "reciprocal_134": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "view_2102": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "clamp_min_402": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "div_268": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "sub_6135": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "maximum_134": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "full_269": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "minimum_134": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "full_268": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "view_2103": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "convert_element_type_804": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "clamp_max_268": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "clamp_min_403": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "sub_6141": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "round_269": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "div_269": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "convert_element_type_807": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "view_2107": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "mul_13049": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "view_2106": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290"], "mul_12859": ["triton_poi_fused__to_copy_mul_sub_view_4:291"], "sub_6076": ["triton_poi_fused__to_copy_mul_sub_view_4:291"], "convert_element_type_796": ["triton_poi_fused__to_copy_mul_sub_view_4:291"], "view_2077": ["triton_poi_fused__to_copy_mul_sub_view_4:291"], "convert_element_type_797": ["triton_poi_fused__to_copy_mul_sub_view_4:291"], "view_2079": ["triton_poi_fused__to_copy_mul_sub_view_4:291"], "view_2078": ["triton_poi_fused__to_copy_mul_sub_view_4:291"], "mul_12955": ["triton_poi_fused__to_copy_mul_sub_view_4:292"], "sub_6121": ["triton_poi_fused__to_copy_mul_sub_view_4:292"], "convert_element_type_802": ["triton_poi_fused__to_copy_mul_sub_view_4:292"], "view_2093": ["triton_poi_fused__to_copy_mul_sub_view_4:292"], "convert_element_type_803": ["triton_poi_fused__to_copy_mul_sub_view_4:292"], "view_2095": ["triton_poi_fused__to_copy_mul_sub_view_4:292"], "view_2094": ["triton_poi_fused__to_copy_mul_sub_view_4:292"], "mul_13054": ["triton_poi_fused__to_copy_mul_sub_view_4:293"], "sub_6165": ["triton_poi_fused__to_copy_mul_sub_view_4:293"], "convert_element_type_808": ["triton_poi_fused__to_copy_mul_sub_view_4:293"], "view_2109": ["triton_poi_fused__to_copy_mul_sub_view_4:293"], "convert_element_type_809": ["triton_poi_fused__to_copy_mul_sub_view_4:293"], "view_2111": ["triton_poi_fused__to_copy_mul_sub_view_4:293"], "view_2110": ["triton_poi_fused__to_copy_mul_sub_view_4:293"], "_scaled_dot_product_efficient_attention_22": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:294", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:295", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:296", "at::_ops::_scaled_dot_product_efficient_attention::call:584"], "clone_178": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:294", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:295", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:296", "at::_ops::_scaled_dot_product_efficient_attention::call:584"], "permute_222": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:294", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:295", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:296", "at::_ops::_scaled_dot_product_efficient_attention::call:584"], "view_2083": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:294", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:295", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:296", "at::_ops::_scaled_dot_product_efficient_attention::call:584"], "mul_12871": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:294", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:295", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:296", "at::_ops::_scaled_dot_product_efficient_attention::call:584"], "view_2082": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:294", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:295", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:296", "at::_ops::_scaled_dot_product_efficient_attention::call:584"], "add_tensor_49": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:294", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:295", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:296", "at::_ops::_scaled_dot_product_efficient_attention::call:584"], "clone_179": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:294", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:295", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:296", "at::_ops::_scaled_dot_product_efficient_attention::call:584"], "permute_224": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:294", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:295", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:296", "at::_ops::_scaled_dot_product_efficient_attention::call:584"], "view_2099": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:294", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:295", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:296", "at::_ops::_scaled_dot_product_efficient_attention::call:584"], "view_2098": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:294", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:295", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:296", "at::_ops::_scaled_dot_product_efficient_attention::call:584"], "clone_180": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:294", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:295", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:296", "at::_ops::_scaled_dot_product_efficient_attention::call:584"], "permute_226": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:294", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:295", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:296", "at::_ops::_scaled_dot_product_efficient_attention::call:584"], "view_2115": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:294", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:295", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:296", "at::_ops::_scaled_dot_product_efficient_attention::call:584"], "view_2114": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:294", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:295", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:296", "at::_ops::_scaled_dot_product_efficient_attention::call:584"], "add_tensor_48": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:294", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:295", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:296", "at::_ops::_scaled_dot_product_efficient_attention::call:584"], "amax_135": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "view_2116": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "permute_227": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "amin_135": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "sub_6209": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "convert_element_type_812": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "convert_element_type_811": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "clamp_max_271": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "clamp_min_407": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "add_20779": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "round_272": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "mul_13132": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "mul_13129": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "reciprocal_135": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "view_2119": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "clamp_min_405": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "div_270": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "sub_6183": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "maximum_135": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "full_271": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "minimum_135": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "full_270": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "view_2120": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "convert_element_type_810": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "clamp_max_270": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "clamp_min_406": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "sub_6189": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "round_271": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "div_271": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "convert_element_type_813": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "view_2124": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "mul_13154": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "view_2123": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297"], "mul_13159": ["triton_poi_fused__to_copy_mul_sub_view_4:298"], "sub_6213": ["triton_poi_fused__to_copy_mul_sub_view_4:298"], "convert_element_type_814": ["triton_poi_fused__to_copy_mul_sub_view_4:298"], "view_2126": ["triton_poi_fused__to_copy_mul_sub_view_4:298"], "convert_element_type_815": ["triton_poi_fused__to_copy_mul_sub_view_4:298"], "view_2128": ["triton_poi_fused__to_copy_mul_sub_view_4:298"], "view_2127": ["triton_poi_fused__to_copy_mul_sub_view_4:298"], "add_20842": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "view_2131": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "add_tensor_47": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "var_mean_45": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "clone_182": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "add_20848": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "mul_13176": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "mul_13175": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "sub_6219": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "rsqrt_45": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "add_20847": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "amax_136": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "amin_136": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "sub_6256": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "convert_element_type_818": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "convert_element_type_817": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "clamp_max_273": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "clamp_min_410": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "add_20935": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "round_274": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "mul_13227": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "mul_13224": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "reciprocal_136": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "view_2134": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "clamp_min_408": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "div_272": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "sub_6230": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "maximum_136": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "full_273": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "minimum_136": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "full_272": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "view_2135": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "convert_element_type_816": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "clamp_max_272": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "clamp_min_409": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "sub_6236": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "round_273": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "div_273": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "convert_element_type_819": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "view_2139": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "mul_13249": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "view_2138": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299"], "mul_13254": ["triton_poi_fused__to_copy_mul_sub_view_10:300"], "sub_6260": ["triton_poi_fused__to_copy_mul_sub_view_10:300"], "convert_element_type_820": ["triton_poi_fused__to_copy_mul_sub_view_10:300"], "view_2141": ["triton_poi_fused__to_copy_mul_sub_view_10:300"], "convert_element_type_821": ["triton_poi_fused__to_copy_mul_sub_view_10:300"], "view_2143": ["triton_poi_fused__to_copy_mul_sub_view_10:300"], "view_2142": ["triton_poi_fused__to_copy_mul_sub_view_10:300"], "amax_137": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "mul_13268": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "mul_13266": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "view_2146": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "add_tensor_46": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "add_20994": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "erf_24": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "mul_13267": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "amin_137": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "add_21077": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "round_276": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "mul_13317": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "mul_13314": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "reciprocal_137": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "view_2149": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "clamp_min_411": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "div_274": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "sub_6273": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "maximum_137": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "full_275": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "minimum_137": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "full_274": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "view_2150": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "convert_element_type_822": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "clamp_max_274": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "clamp_min_412": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "sub_6279": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "round_275": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "div_275": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "mul_13339": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "sub_6299": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "convert_element_type_824": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "convert_element_type_823": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "clamp_max_275": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "clamp_min_413": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "convert_element_type_825": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "view_2154": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "view_2153": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301"], "mul_13344": ["triton_poi_fused__to_copy_mul_sub_view_10:302"], "sub_6303": ["triton_poi_fused__to_copy_mul_sub_view_10:302"], "convert_element_type_826": ["triton_poi_fused__to_copy_mul_sub_view_10:302"], "view_2156": ["triton_poi_fused__to_copy_mul_sub_view_10:302"], "convert_element_type_827": ["triton_poi_fused__to_copy_mul_sub_view_10:302"], "view_2158": ["triton_poi_fused__to_copy_mul_sub_view_10:302"], "view_2157": ["triton_poi_fused__to_copy_mul_sub_view_10:302"], "var_mean_46": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "clone_185": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "add_21140": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "view_2161": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "add_tensor_45": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "add_21146": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "mul_13361": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "mul_13360": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "sub_6309": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "rsqrt_46": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "add_21145": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "amax_138": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "amin_138": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "amax_139": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "amin_139": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "amax_140": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "amin_140": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "sub_6346": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "convert_element_type_830": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "convert_element_type_829": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "clamp_max_277": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "clamp_min_416": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "add_21233": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "round_278": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "mul_13412": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "mul_13409": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "reciprocal_138": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "view_2164": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "clamp_min_414": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "div_276": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "sub_6320": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "maximum_138": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "full_277": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "minimum_138": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "full_276": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "view_2165": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "convert_element_type_828": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "clamp_max_276": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "clamp_min_415": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "sub_6326": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "round_277": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "div_277": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "convert_element_type_831": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "view_2169": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "mul_13434": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "view_2168": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "sub_6391": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "convert_element_type_836": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "convert_element_type_835": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "clamp_max_279": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "clamp_min_419": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "add_21385": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "round_280": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "mul_13508": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "mul_13505": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "reciprocal_139": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "view_2180": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "clamp_min_417": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "div_278": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "sub_6365": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "maximum_139": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "full_279": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "minimum_139": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "full_278": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "view_2181": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "convert_element_type_834": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "clamp_max_278": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "clamp_min_418": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "sub_6371": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "round_279": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "div_279": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "convert_element_type_837": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "view_2185": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "mul_13530": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "view_2184": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "sub_6435": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "convert_element_type_842": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "convert_element_type_841": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "clamp_max_281": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "clamp_min_422": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "add_21533": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "round_282": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "mul_13607": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "mul_13604": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "reciprocal_140": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "view_2196": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "clamp_min_420": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "div_280": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "sub_6409": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "maximum_140": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "full_281": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "minimum_140": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "full_280": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "view_2197": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "convert_element_type_840": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "clamp_max_280": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "clamp_min_421": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "sub_6415": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "round_281": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "div_281": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "convert_element_type_843": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "view_2201": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "mul_13629": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "view_2200": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303"], "mul_13439": ["triton_poi_fused__to_copy_mul_sub_view_4:304"], "sub_6350": ["triton_poi_fused__to_copy_mul_sub_view_4:304"], "convert_element_type_832": ["triton_poi_fused__to_copy_mul_sub_view_4:304"], "view_2171": ["triton_poi_fused__to_copy_mul_sub_view_4:304"], "convert_element_type_833": ["triton_poi_fused__to_copy_mul_sub_view_4:304"], "view_2173": ["triton_poi_fused__to_copy_mul_sub_view_4:304"], "view_2172": ["triton_poi_fused__to_copy_mul_sub_view_4:304"], "mul_13535": ["triton_poi_fused__to_copy_mul_sub_view_4:305"], "sub_6395": ["triton_poi_fused__to_copy_mul_sub_view_4:305"], "convert_element_type_838": ["triton_poi_fused__to_copy_mul_sub_view_4:305"], "view_2187": ["triton_poi_fused__to_copy_mul_sub_view_4:305"], "convert_element_type_839": ["triton_poi_fused__to_copy_mul_sub_view_4:305"], "view_2189": ["triton_poi_fused__to_copy_mul_sub_view_4:305"], "view_2188": ["triton_poi_fused__to_copy_mul_sub_view_4:305"], "mul_13634": ["triton_poi_fused__to_copy_mul_sub_view_4:306"], "sub_6439": ["triton_poi_fused__to_copy_mul_sub_view_4:306"], "convert_element_type_844": ["triton_poi_fused__to_copy_mul_sub_view_4:306"], "view_2203": ["triton_poi_fused__to_copy_mul_sub_view_4:306"], "convert_element_type_845": ["triton_poi_fused__to_copy_mul_sub_view_4:306"], "view_2205": ["triton_poi_fused__to_copy_mul_sub_view_4:306"], "view_2204": ["triton_poi_fused__to_copy_mul_sub_view_4:306"], "_scaled_dot_product_efficient_attention_23": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:307", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:308", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:309", "at::_ops::_scaled_dot_product_efficient_attention::call:591"], "clone_186": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:307", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:308", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:309", "at::_ops::_scaled_dot_product_efficient_attention::call:591"], "permute_232": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:307", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:308", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:309", "at::_ops::_scaled_dot_product_efficient_attention::call:591"], "view_2177": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:307", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:308", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:309", "at::_ops::_scaled_dot_product_efficient_attention::call:591"], "mul_13451": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:307", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:308", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:309", "at::_ops::_scaled_dot_product_efficient_attention::call:591"], "view_2176": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:307", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:308", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:309", "at::_ops::_scaled_dot_product_efficient_attention::call:591"], "add_tensor_44": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:307", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:308", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:309", "at::_ops::_scaled_dot_product_efficient_attention::call:591"], "clone_187": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:307", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:308", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:309", "at::_ops::_scaled_dot_product_efficient_attention::call:591"], "permute_234": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:307", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:308", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:309", "at::_ops::_scaled_dot_product_efficient_attention::call:591"], "view_2193": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:307", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:308", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:309", "at::_ops::_scaled_dot_product_efficient_attention::call:591"], "view_2192": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:307", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:308", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:309", "at::_ops::_scaled_dot_product_efficient_attention::call:591"], "clone_188": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:307", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:308", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:309", "at::_ops::_scaled_dot_product_efficient_attention::call:591"], "permute_236": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:307", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:308", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:309", "at::_ops::_scaled_dot_product_efficient_attention::call:591"], "view_2209": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:307", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:308", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:309", "at::_ops::_scaled_dot_product_efficient_attention::call:591"], "view_2208": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:307", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:308", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:309", "at::_ops::_scaled_dot_product_efficient_attention::call:591"], "add_tensor_43": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:307", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:308", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:309", "at::_ops::_scaled_dot_product_efficient_attention::call:591"], "amax_141": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "view_2210": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "permute_237": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "amin_141": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "sub_6483": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "convert_element_type_848": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "convert_element_type_847": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "clamp_max_283": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "clamp_min_425": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "add_21697": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "round_284": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "mul_13712": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "mul_13709": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "reciprocal_141": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "view_2213": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "clamp_min_423": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "div_282": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "sub_6457": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "maximum_141": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "full_283": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "minimum_141": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "full_282": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "view_2214": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "convert_element_type_846": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "clamp_max_282": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "clamp_min_424": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "sub_6463": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "round_283": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "div_283": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "convert_element_type_849": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "view_2218": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "mul_13734": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "view_2217": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310"], "mul_13739": ["triton_poi_fused__to_copy_mul_sub_view_4:311"], "sub_6487": ["triton_poi_fused__to_copy_mul_sub_view_4:311"], "convert_element_type_850": ["triton_poi_fused__to_copy_mul_sub_view_4:311"], "view_2220": ["triton_poi_fused__to_copy_mul_sub_view_4:311"], "convert_element_type_851": ["triton_poi_fused__to_copy_mul_sub_view_4:311"], "view_2222": ["triton_poi_fused__to_copy_mul_sub_view_4:311"], "view_2221": ["triton_poi_fused__to_copy_mul_sub_view_4:311"], "add_21760": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "view_2225": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "add_tensor_42": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "var_mean_47": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "clone_190": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "add_21766": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "mul_13756": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "mul_13755": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "sub_6493": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "rsqrt_47": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "add_21765": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "amax_142": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "amin_142": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "sub_6530": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "convert_element_type_854": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "convert_element_type_853": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "clamp_max_285": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "clamp_min_428": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "add_21853": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "round_286": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "mul_13807": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "mul_13804": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "reciprocal_142": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "view_2228": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "clamp_min_426": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "div_284": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "sub_6504": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "maximum_142": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "full_285": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "minimum_142": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "full_284": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "view_2229": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "convert_element_type_852": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "clamp_max_284": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "clamp_min_427": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "sub_6510": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "round_285": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "div_285": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "convert_element_type_855": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "view_2233": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "mul_13829": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "view_2232": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312"], "mul_13834": ["triton_poi_fused__to_copy_mul_sub_view_10:313"], "sub_6534": ["triton_poi_fused__to_copy_mul_sub_view_10:313"], "convert_element_type_856": ["triton_poi_fused__to_copy_mul_sub_view_10:313"], "view_2235": ["triton_poi_fused__to_copy_mul_sub_view_10:313"], "convert_element_type_857": ["triton_poi_fused__to_copy_mul_sub_view_10:313"], "view_2237": ["triton_poi_fused__to_copy_mul_sub_view_10:313"], "view_2236": ["triton_poi_fused__to_copy_mul_sub_view_10:313"], "amax_143": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "mul_13848": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "mul_13846": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "view_2240": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "add_tensor_41": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "add_21912": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "erf_25": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "mul_13847": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "amin_143": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "add_21995": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "round_288": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "mul_13897": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "mul_13894": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "reciprocal_143": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "view_2243": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "clamp_min_429": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "div_286": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "sub_6547": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "maximum_143": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "full_287": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "minimum_143": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "full_286": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "view_2244": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "convert_element_type_858": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "clamp_max_286": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "clamp_min_430": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "sub_6553": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "round_287": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "div_287": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "mul_13919": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "sub_6573": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "convert_element_type_860": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "convert_element_type_859": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "clamp_max_287": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "clamp_min_431": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "convert_element_type_861": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "view_2248": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "view_2247": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314"], "mul_13924": ["triton_poi_fused__to_copy_mul_sub_view_10:315"], "sub_6577": ["triton_poi_fused__to_copy_mul_sub_view_10:315"], "convert_element_type_862": ["triton_poi_fused__to_copy_mul_sub_view_10:315"], "view_2250": ["triton_poi_fused__to_copy_mul_sub_view_10:315"], "convert_element_type_863": ["triton_poi_fused__to_copy_mul_sub_view_10:315"], "view_2252": ["triton_poi_fused__to_copy_mul_sub_view_10:315"], "view_2251": ["triton_poi_fused__to_copy_mul_sub_view_10:315"], "var_mean_48": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "clone_193": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "add_22058": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "view_2255": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "add_tensor_40": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "add_22064": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "mul_13941": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "mul_13940": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "sub_6583": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "rsqrt_48": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "add_22063": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "amax_144": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "amin_144": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "amax_145": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "amin_145": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "amax_146": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "amin_146": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "sub_6620": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "convert_element_type_866": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "convert_element_type_865": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "clamp_max_289": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "clamp_min_434": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "add_22151": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "round_290": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "mul_13992": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "mul_13989": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "reciprocal_144": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "view_2258": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "clamp_min_432": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "div_288": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "sub_6594": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "maximum_144": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "full_289": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "minimum_144": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "full_288": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "view_2259": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "convert_element_type_864": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "clamp_max_288": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "clamp_min_433": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "sub_6600": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "round_289": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "div_289": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "convert_element_type_867": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "view_2263": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "mul_14014": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "view_2262": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "sub_6665": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "convert_element_type_872": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "convert_element_type_871": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "clamp_max_291": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "clamp_min_437": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "add_22303": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "round_292": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "mul_14088": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "mul_14085": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "reciprocal_145": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "view_2274": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "clamp_min_435": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "div_290": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "sub_6639": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "maximum_145": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "full_291": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "minimum_145": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "full_290": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "view_2275": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "convert_element_type_870": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "clamp_max_290": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "clamp_min_436": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "sub_6645": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "round_291": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "div_291": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "convert_element_type_873": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "view_2279": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "mul_14110": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "view_2278": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "sub_6709": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "convert_element_type_878": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "convert_element_type_877": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "clamp_max_293": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "clamp_min_440": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "add_22451": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "round_294": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "mul_14187": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "mul_14184": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "reciprocal_146": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "view_2290": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "clamp_min_438": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "div_292": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "sub_6683": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "maximum_146": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "full_293": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "minimum_146": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "full_292": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "view_2291": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "convert_element_type_876": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "clamp_max_292": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "clamp_min_439": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "sub_6689": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "round_293": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "div_293": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "convert_element_type_879": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "view_2295": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "mul_14209": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "view_2294": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316"], "mul_14019": ["triton_poi_fused__to_copy_mul_sub_view_4:317"], "sub_6624": ["triton_poi_fused__to_copy_mul_sub_view_4:317"], "convert_element_type_868": ["triton_poi_fused__to_copy_mul_sub_view_4:317"], "view_2265": ["triton_poi_fused__to_copy_mul_sub_view_4:317"], "convert_element_type_869": ["triton_poi_fused__to_copy_mul_sub_view_4:317"], "view_2267": ["triton_poi_fused__to_copy_mul_sub_view_4:317"], "view_2266": ["triton_poi_fused__to_copy_mul_sub_view_4:317"], "mul_14115": ["triton_poi_fused__to_copy_mul_sub_view_4:318"], "sub_6669": ["triton_poi_fused__to_copy_mul_sub_view_4:318"], "convert_element_type_874": ["triton_poi_fused__to_copy_mul_sub_view_4:318"], "view_2281": ["triton_poi_fused__to_copy_mul_sub_view_4:318"], "convert_element_type_875": ["triton_poi_fused__to_copy_mul_sub_view_4:318"], "view_2283": ["triton_poi_fused__to_copy_mul_sub_view_4:318"], "view_2282": ["triton_poi_fused__to_copy_mul_sub_view_4:318"], "mul_14214": ["triton_poi_fused__to_copy_mul_sub_view_4:319"], "sub_6713": ["triton_poi_fused__to_copy_mul_sub_view_4:319"], "convert_element_type_880": ["triton_poi_fused__to_copy_mul_sub_view_4:319"], "view_2297": ["triton_poi_fused__to_copy_mul_sub_view_4:319"], "convert_element_type_881": ["triton_poi_fused__to_copy_mul_sub_view_4:319"], "view_2299": ["triton_poi_fused__to_copy_mul_sub_view_4:319"], "view_2298": ["triton_poi_fused__to_copy_mul_sub_view_4:319"], "_scaled_dot_product_efficient_attention_24": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:320", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:321", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:322", "at::_ops::_scaled_dot_product_efficient_attention::call:598"], "clone_194": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:320", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:321", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:322", "at::_ops::_scaled_dot_product_efficient_attention::call:598"], "permute_242": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:320", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:321", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:322", "at::_ops::_scaled_dot_product_efficient_attention::call:598"], "view_2271": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:320", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:321", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:322", "at::_ops::_scaled_dot_product_efficient_attention::call:598"], "mul_14031": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:320", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:321", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:322", "at::_ops::_scaled_dot_product_efficient_attention::call:598"], "view_2270": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:320", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:321", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:322", "at::_ops::_scaled_dot_product_efficient_attention::call:598"], "add_tensor_39": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:320", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:321", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:322", "at::_ops::_scaled_dot_product_efficient_attention::call:598"], "clone_195": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:320", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:321", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:322", "at::_ops::_scaled_dot_product_efficient_attention::call:598"], "permute_244": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:320", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:321", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:322", "at::_ops::_scaled_dot_product_efficient_attention::call:598"], "view_2287": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:320", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:321", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:322", "at::_ops::_scaled_dot_product_efficient_attention::call:598"], "view_2286": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:320", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:321", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:322", "at::_ops::_scaled_dot_product_efficient_attention::call:598"], "clone_196": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:320", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:321", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:322", "at::_ops::_scaled_dot_product_efficient_attention::call:598"], "permute_246": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:320", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:321", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:322", "at::_ops::_scaled_dot_product_efficient_attention::call:598"], "view_2303": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:320", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:321", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:322", "at::_ops::_scaled_dot_product_efficient_attention::call:598"], "view_2302": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:320", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:321", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:322", "at::_ops::_scaled_dot_product_efficient_attention::call:598"], "add_tensor_38": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:320", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:321", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:322", "at::_ops::_scaled_dot_product_efficient_attention::call:598"], "amax_147": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "view_2304": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "permute_247": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "amin_147": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "sub_6757": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "convert_element_type_884": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "convert_element_type_883": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "clamp_max_295": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "clamp_min_443": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "add_22615": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "round_296": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "mul_14292": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "mul_14289": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "reciprocal_147": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "view_2307": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "clamp_min_441": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "div_294": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "sub_6731": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "maximum_147": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "full_295": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "minimum_147": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "full_294": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "view_2308": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "convert_element_type_882": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "clamp_max_294": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "clamp_min_442": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "sub_6737": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "round_295": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "div_295": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "convert_element_type_885": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "view_2312": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "mul_14314": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "view_2311": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323"], "mul_14319": ["triton_poi_fused__to_copy_mul_sub_view_4:324"], "sub_6761": ["triton_poi_fused__to_copy_mul_sub_view_4:324"], "convert_element_type_886": ["triton_poi_fused__to_copy_mul_sub_view_4:324"], "view_2314": ["triton_poi_fused__to_copy_mul_sub_view_4:324"], "convert_element_type_887": ["triton_poi_fused__to_copy_mul_sub_view_4:324"], "view_2316": ["triton_poi_fused__to_copy_mul_sub_view_4:324"], "view_2315": ["triton_poi_fused__to_copy_mul_sub_view_4:324"], "add_22678": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "view_2319": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "add_tensor_37": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "var_mean_49": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "clone_198": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "add_22684": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "mul_14336": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "mul_14335": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "sub_6767": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "rsqrt_49": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "add_22683": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "amax_148": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "amin_148": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "sub_6804": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "convert_element_type_890": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "convert_element_type_889": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "clamp_max_297": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "clamp_min_446": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "add_22771": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "round_298": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "mul_14387": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "mul_14384": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "reciprocal_148": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "view_2322": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "clamp_min_444": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "div_296": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "sub_6778": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "maximum_148": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "full_297": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "minimum_148": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "full_296": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "view_2323": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "convert_element_type_888": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "clamp_max_296": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "clamp_min_445": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "sub_6784": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "round_297": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "div_297": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "convert_element_type_891": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "view_2327": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "mul_14409": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "view_2326": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325"], "mul_14414": ["triton_poi_fused__to_copy_mul_sub_view_10:326"], "sub_6808": ["triton_poi_fused__to_copy_mul_sub_view_10:326"], "convert_element_type_892": ["triton_poi_fused__to_copy_mul_sub_view_10:326"], "view_2329": ["triton_poi_fused__to_copy_mul_sub_view_10:326"], "convert_element_type_893": ["triton_poi_fused__to_copy_mul_sub_view_10:326"], "view_2331": ["triton_poi_fused__to_copy_mul_sub_view_10:326"], "view_2330": ["triton_poi_fused__to_copy_mul_sub_view_10:326"], "amax_149": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "mul_14428": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "mul_14426": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "view_2334": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "add_tensor_36": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "add_22830": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "erf_26": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "mul_14427": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "amin_149": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "add_22913": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "round_300": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "mul_14477": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "mul_14474": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "reciprocal_149": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "view_2337": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "clamp_min_447": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "div_298": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "sub_6821": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "maximum_149": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "full_299": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "minimum_149": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "full_298": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "view_2338": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "convert_element_type_894": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "clamp_max_298": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "clamp_min_448": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "sub_6827": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "round_299": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "div_299": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "mul_14499": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "sub_6847": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "convert_element_type_896": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "convert_element_type_895": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "clamp_max_299": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "clamp_min_449": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "convert_element_type_897": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "view_2342": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "view_2341": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327"], "mul_14504": ["triton_poi_fused__to_copy_mul_sub_view_10:328"], "sub_6851": ["triton_poi_fused__to_copy_mul_sub_view_10:328"], "convert_element_type_898": ["triton_poi_fused__to_copy_mul_sub_view_10:328"], "view_2344": ["triton_poi_fused__to_copy_mul_sub_view_10:328"], "convert_element_type_899": ["triton_poi_fused__to_copy_mul_sub_view_10:328"], "view_2346": ["triton_poi_fused__to_copy_mul_sub_view_10:328"], "view_2345": ["triton_poi_fused__to_copy_mul_sub_view_10:328"], "var_mean_50": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "clone_201": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "add_22976": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "view_2349": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "add_tensor_35": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "add_22982": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "mul_14521": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "mul_14520": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "sub_6857": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "rsqrt_50": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "add_22981": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "amax_150": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "amin_150": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "amax_151": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "amin_151": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "amax_152": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "amin_152": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "sub_6894": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "convert_element_type_902": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "convert_element_type_901": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "clamp_max_301": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "clamp_min_452": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "add_23069": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "round_302": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "mul_14572": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "mul_14569": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "reciprocal_150": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "view_2352": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "clamp_min_450": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "div_300": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "sub_6868": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "maximum_150": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "full_301": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "minimum_150": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "full_300": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "view_2353": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "convert_element_type_900": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "clamp_max_300": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "clamp_min_451": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "sub_6874": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "round_301": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "div_301": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "convert_element_type_903": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "view_2357": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "mul_14594": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "view_2356": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "sub_6939": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "convert_element_type_908": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "convert_element_type_907": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "clamp_max_303": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "clamp_min_455": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "add_23221": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "round_304": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "mul_14668": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "mul_14665": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "reciprocal_151": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "view_2368": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "clamp_min_453": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "div_302": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "sub_6913": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "maximum_151": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "full_303": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "minimum_151": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "full_302": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "view_2369": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "convert_element_type_906": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "clamp_max_302": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "clamp_min_454": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "sub_6919": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "round_303": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "div_303": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "convert_element_type_909": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "view_2373": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "mul_14690": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "view_2372": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "sub_6983": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "convert_element_type_914": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "convert_element_type_913": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "clamp_max_305": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "clamp_min_458": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "add_23369": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "round_306": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "mul_14767": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "mul_14764": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "reciprocal_152": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "view_2384": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "clamp_min_456": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "div_304": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "sub_6957": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "maximum_152": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "full_305": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "minimum_152": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "full_304": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "view_2385": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "convert_element_type_912": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "clamp_max_304": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "clamp_min_457": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "sub_6963": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "round_305": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "div_305": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "convert_element_type_915": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "view_2389": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "mul_14789": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "view_2388": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329"], "mul_14599": ["triton_poi_fused__to_copy_mul_sub_view_4:330"], "sub_6898": ["triton_poi_fused__to_copy_mul_sub_view_4:330"], "convert_element_type_904": ["triton_poi_fused__to_copy_mul_sub_view_4:330"], "view_2359": ["triton_poi_fused__to_copy_mul_sub_view_4:330"], "convert_element_type_905": ["triton_poi_fused__to_copy_mul_sub_view_4:330"], "view_2361": ["triton_poi_fused__to_copy_mul_sub_view_4:330"], "view_2360": ["triton_poi_fused__to_copy_mul_sub_view_4:330"], "mul_14695": ["triton_poi_fused__to_copy_mul_sub_view_4:331"], "sub_6943": ["triton_poi_fused__to_copy_mul_sub_view_4:331"], "convert_element_type_910": ["triton_poi_fused__to_copy_mul_sub_view_4:331"], "view_2375": ["triton_poi_fused__to_copy_mul_sub_view_4:331"], "convert_element_type_911": ["triton_poi_fused__to_copy_mul_sub_view_4:331"], "view_2377": ["triton_poi_fused__to_copy_mul_sub_view_4:331"], "view_2376": ["triton_poi_fused__to_copy_mul_sub_view_4:331"], "mul_14794": ["triton_poi_fused__to_copy_mul_sub_view_4:332"], "sub_6987": ["triton_poi_fused__to_copy_mul_sub_view_4:332"], "convert_element_type_916": ["triton_poi_fused__to_copy_mul_sub_view_4:332"], "view_2391": ["triton_poi_fused__to_copy_mul_sub_view_4:332"], "convert_element_type_917": ["triton_poi_fused__to_copy_mul_sub_view_4:332"], "view_2393": ["triton_poi_fused__to_copy_mul_sub_view_4:332"], "view_2392": ["triton_poi_fused__to_copy_mul_sub_view_4:332"], "_scaled_dot_product_efficient_attention_25": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:333", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:334", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:335", "at::_ops::_scaled_dot_product_efficient_attention::call:605"], "clone_202": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:333", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:334", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:335", "at::_ops::_scaled_dot_product_efficient_attention::call:605"], "permute_252": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:333", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:334", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:335", "at::_ops::_scaled_dot_product_efficient_attention::call:605"], "view_2365": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:333", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:334", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:335", "at::_ops::_scaled_dot_product_efficient_attention::call:605"], "mul_14611": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:333", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:334", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:335", "at::_ops::_scaled_dot_product_efficient_attention::call:605"], "view_2364": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:333", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:334", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:335", "at::_ops::_scaled_dot_product_efficient_attention::call:605"], "add_tensor_34": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:333", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:334", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:335", "at::_ops::_scaled_dot_product_efficient_attention::call:605"], "clone_203": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:333", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:334", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:335", "at::_ops::_scaled_dot_product_efficient_attention::call:605"], "permute_254": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:333", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:334", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:335", "at::_ops::_scaled_dot_product_efficient_attention::call:605"], "view_2381": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:333", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:334", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:335", "at::_ops::_scaled_dot_product_efficient_attention::call:605"], "view_2380": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:333", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:334", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:335", "at::_ops::_scaled_dot_product_efficient_attention::call:605"], "clone_204": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:333", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:334", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:335", "at::_ops::_scaled_dot_product_efficient_attention::call:605"], "permute_256": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:333", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:334", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:335", "at::_ops::_scaled_dot_product_efficient_attention::call:605"], "view_2397": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:333", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:334", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:335", "at::_ops::_scaled_dot_product_efficient_attention::call:605"], "view_2396": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:333", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:334", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:335", "at::_ops::_scaled_dot_product_efficient_attention::call:605"], "add_tensor_33": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:333", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:334", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:335", "at::_ops::_scaled_dot_product_efficient_attention::call:605"], "amax_153": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "view_2398": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "permute_257": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "amin_153": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "sub_7031": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "convert_element_type_920": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "convert_element_type_919": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "clamp_max_307": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "clamp_min_461": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "add_23533": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "round_308": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "mul_14872": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "mul_14869": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "reciprocal_153": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "view_2401": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "clamp_min_459": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "div_306": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "sub_7005": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "maximum_153": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "full_307": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "minimum_153": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "full_306": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "view_2402": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "convert_element_type_918": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "clamp_max_306": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "clamp_min_460": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "sub_7011": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "round_307": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "div_307": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "convert_element_type_921": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "view_2406": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "mul_14894": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "view_2405": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336"], "mul_14899": ["triton_poi_fused__to_copy_mul_sub_view_4:337"], "sub_7035": ["triton_poi_fused__to_copy_mul_sub_view_4:337"], "convert_element_type_922": ["triton_poi_fused__to_copy_mul_sub_view_4:337"], "view_2408": ["triton_poi_fused__to_copy_mul_sub_view_4:337"], "convert_element_type_923": ["triton_poi_fused__to_copy_mul_sub_view_4:337"], "view_2410": ["triton_poi_fused__to_copy_mul_sub_view_4:337"], "view_2409": ["triton_poi_fused__to_copy_mul_sub_view_4:337"], "add_23596": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "view_2413": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "add_tensor_32": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "var_mean_51": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "clone_206": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "add_23602": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "mul_14916": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "mul_14915": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "sub_7041": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "rsqrt_51": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "add_23601": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "amax_154": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "amin_154": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "sub_7078": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "convert_element_type_926": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "convert_element_type_925": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "clamp_max_309": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "clamp_min_464": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "add_23689": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "round_310": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "mul_14967": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "mul_14964": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "reciprocal_154": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "view_2416": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "clamp_min_462": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "div_308": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "sub_7052": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "maximum_154": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "full_309": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "minimum_154": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "full_308": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "view_2417": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "convert_element_type_924": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "clamp_max_308": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "clamp_min_463": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "sub_7058": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "round_309": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "div_309": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "convert_element_type_927": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "view_2421": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "mul_14989": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "view_2420": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338"], "mul_14994": ["triton_poi_fused__to_copy_mul_sub_view_10:339"], "sub_7082": ["triton_poi_fused__to_copy_mul_sub_view_10:339"], "convert_element_type_928": ["triton_poi_fused__to_copy_mul_sub_view_10:339"], "view_2423": ["triton_poi_fused__to_copy_mul_sub_view_10:339"], "convert_element_type_929": ["triton_poi_fused__to_copy_mul_sub_view_10:339"], "view_2425": ["triton_poi_fused__to_copy_mul_sub_view_10:339"], "view_2424": ["triton_poi_fused__to_copy_mul_sub_view_10:339"], "amax_155": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "mul_15008": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "mul_15006": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "view_2428": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "add_tensor_31": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "add_23748": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "erf_27": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "mul_15007": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "amin_155": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "add_23831": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "round_312": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "mul_15057": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "mul_15054": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "reciprocal_155": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "view_2431": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "clamp_min_465": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "div_310": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "sub_7095": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "maximum_155": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "full_311": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "minimum_155": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "full_310": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "view_2432": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "convert_element_type_930": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "clamp_max_310": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "clamp_min_466": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "sub_7101": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "round_311": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "div_311": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "mul_15079": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "sub_7121": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "convert_element_type_932": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "convert_element_type_931": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "clamp_max_311": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "clamp_min_467": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "convert_element_type_933": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "view_2436": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "view_2435": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340"], "mul_15084": ["triton_poi_fused__to_copy_mul_sub_view_10:341"], "sub_7125": ["triton_poi_fused__to_copy_mul_sub_view_10:341"], "convert_element_type_934": ["triton_poi_fused__to_copy_mul_sub_view_10:341"], "view_2438": ["triton_poi_fused__to_copy_mul_sub_view_10:341"], "convert_element_type_935": ["triton_poi_fused__to_copy_mul_sub_view_10:341"], "view_2440": ["triton_poi_fused__to_copy_mul_sub_view_10:341"], "view_2439": ["triton_poi_fused__to_copy_mul_sub_view_10:341"], "var_mean_52": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "clone_209": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "add_23894": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "view_2443": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "add_tensor_30": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "add_23900": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "mul_15101": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "mul_15100": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "sub_7131": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "rsqrt_52": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "add_23899": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "amax_156": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "amin_156": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "amax_157": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "amin_157": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "amax_158": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "amin_158": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "sub_7168": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "convert_element_type_938": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "convert_element_type_937": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "clamp_max_313": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "clamp_min_470": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "add_23987": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "round_314": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "mul_15152": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "mul_15149": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "reciprocal_156": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "view_2446": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "clamp_min_468": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "div_312": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "sub_7142": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "maximum_156": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "full_313": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "minimum_156": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "full_312": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "view_2447": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "convert_element_type_936": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "clamp_max_312": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "clamp_min_469": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "sub_7148": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "round_313": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "div_313": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "convert_element_type_939": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "view_2451": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "mul_15174": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "view_2450": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "sub_7213": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "convert_element_type_944": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "convert_element_type_943": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "clamp_max_315": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "clamp_min_473": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "add_24139": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "round_316": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "mul_15248": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "mul_15245": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "reciprocal_157": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "view_2462": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "clamp_min_471": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "div_314": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "sub_7187": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "maximum_157": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "full_315": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "minimum_157": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "full_314": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "view_2463": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "convert_element_type_942": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "clamp_max_314": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "clamp_min_472": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "sub_7193": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "round_315": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "div_315": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "convert_element_type_945": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "view_2467": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "mul_15270": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "view_2466": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "sub_7257": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "convert_element_type_950": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "convert_element_type_949": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "clamp_max_317": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "clamp_min_476": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "add_24287": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "round_318": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "mul_15347": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "mul_15344": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "reciprocal_158": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "view_2478": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "clamp_min_474": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "div_316": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "sub_7231": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "maximum_158": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "full_317": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "minimum_158": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "full_316": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "view_2479": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "convert_element_type_948": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "clamp_max_316": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "clamp_min_475": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "sub_7237": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "round_317": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "div_317": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "convert_element_type_951": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "view_2483": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "mul_15369": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "view_2482": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342"], "mul_15179": ["triton_poi_fused__to_copy_mul_sub_view_4:343"], "sub_7172": ["triton_poi_fused__to_copy_mul_sub_view_4:343"], "convert_element_type_940": ["triton_poi_fused__to_copy_mul_sub_view_4:343"], "view_2453": ["triton_poi_fused__to_copy_mul_sub_view_4:343"], "convert_element_type_941": ["triton_poi_fused__to_copy_mul_sub_view_4:343"], "view_2455": ["triton_poi_fused__to_copy_mul_sub_view_4:343"], "view_2454": ["triton_poi_fused__to_copy_mul_sub_view_4:343"], "mul_15275": ["triton_poi_fused__to_copy_mul_sub_view_4:344"], "sub_7217": ["triton_poi_fused__to_copy_mul_sub_view_4:344"], "convert_element_type_946": ["triton_poi_fused__to_copy_mul_sub_view_4:344"], "view_2469": ["triton_poi_fused__to_copy_mul_sub_view_4:344"], "convert_element_type_947": ["triton_poi_fused__to_copy_mul_sub_view_4:344"], "view_2471": ["triton_poi_fused__to_copy_mul_sub_view_4:344"], "view_2470": ["triton_poi_fused__to_copy_mul_sub_view_4:344"], "mul_15374": ["triton_poi_fused__to_copy_mul_sub_view_4:345"], "sub_7261": ["triton_poi_fused__to_copy_mul_sub_view_4:345"], "convert_element_type_952": ["triton_poi_fused__to_copy_mul_sub_view_4:345"], "view_2485": ["triton_poi_fused__to_copy_mul_sub_view_4:345"], "convert_element_type_953": ["triton_poi_fused__to_copy_mul_sub_view_4:345"], "view_2487": ["triton_poi_fused__to_copy_mul_sub_view_4:345"], "view_2486": ["triton_poi_fused__to_copy_mul_sub_view_4:345"], "_scaled_dot_product_efficient_attention_26": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:346", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:347", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:348", "at::_ops::_scaled_dot_product_efficient_attention::call:612"], "clone_210": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:346", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:347", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:348", "at::_ops::_scaled_dot_product_efficient_attention::call:612"], "permute_262": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:346", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:347", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:348", "at::_ops::_scaled_dot_product_efficient_attention::call:612"], "view_2459": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:346", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:347", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:348", "at::_ops::_scaled_dot_product_efficient_attention::call:612"], "mul_15191": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:346", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:347", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:348", "at::_ops::_scaled_dot_product_efficient_attention::call:612"], "view_2458": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:346", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:347", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:348", "at::_ops::_scaled_dot_product_efficient_attention::call:612"], "add_tensor_29": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:346", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:347", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:348", "at::_ops::_scaled_dot_product_efficient_attention::call:612"], "clone_211": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:346", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:347", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:348", "at::_ops::_scaled_dot_product_efficient_attention::call:612"], "permute_264": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:346", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:347", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:348", "at::_ops::_scaled_dot_product_efficient_attention::call:612"], "view_2475": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:346", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:347", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:348", "at::_ops::_scaled_dot_product_efficient_attention::call:612"], "view_2474": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:346", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:347", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:348", "at::_ops::_scaled_dot_product_efficient_attention::call:612"], "clone_212": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:346", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:347", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:348", "at::_ops::_scaled_dot_product_efficient_attention::call:612"], "permute_266": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:346", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:347", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:348", "at::_ops::_scaled_dot_product_efficient_attention::call:612"], "view_2491": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:346", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:347", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:348", "at::_ops::_scaled_dot_product_efficient_attention::call:612"], "view_2490": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:346", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:347", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:348", "at::_ops::_scaled_dot_product_efficient_attention::call:612"], "add_tensor_28": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:346", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:347", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:348", "at::_ops::_scaled_dot_product_efficient_attention::call:612"], "amax_159": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "view_2492": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "permute_267": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "amin_159": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "sub_7305": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "convert_element_type_956": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "convert_element_type_955": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "clamp_max_319": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "clamp_min_479": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "add_24451": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "round_320": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "mul_15452": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "mul_15449": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "reciprocal_159": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "view_2495": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "clamp_min_477": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "div_318": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "sub_7279": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "maximum_159": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "full_319": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "minimum_159": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "full_318": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "view_2496": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "convert_element_type_954": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "clamp_max_318": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "clamp_min_478": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "sub_7285": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "round_319": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "div_319": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "convert_element_type_957": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "view_2500": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "mul_15474": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "view_2499": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349"], "mul_15479": ["triton_poi_fused__to_copy_mul_sub_view_4:350"], "sub_7309": ["triton_poi_fused__to_copy_mul_sub_view_4:350"], "convert_element_type_958": ["triton_poi_fused__to_copy_mul_sub_view_4:350"], "view_2502": ["triton_poi_fused__to_copy_mul_sub_view_4:350"], "convert_element_type_959": ["triton_poi_fused__to_copy_mul_sub_view_4:350"], "view_2504": ["triton_poi_fused__to_copy_mul_sub_view_4:350"], "view_2503": ["triton_poi_fused__to_copy_mul_sub_view_4:350"], "add_24514": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "view_2507": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "add_tensor_27": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "var_mean_53": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "clone_214": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "add_24520": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "mul_15496": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "mul_15495": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "sub_7315": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "rsqrt_53": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "add_24519": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "amax_160": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "amin_160": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "sub_7352": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "convert_element_type_962": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "convert_element_type_961": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "clamp_max_321": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "clamp_min_482": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "add_24607": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "round_322": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "mul_15547": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "mul_15544": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "reciprocal_160": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "view_2510": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "clamp_min_480": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "div_320": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "sub_7326": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "maximum_160": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "full_321": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "minimum_160": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "full_320": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "view_2511": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "convert_element_type_960": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "clamp_max_320": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "clamp_min_481": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "sub_7332": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "round_321": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "div_321": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "convert_element_type_963": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "view_2515": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "mul_15569": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "view_2514": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351"], "mul_15574": ["triton_poi_fused__to_copy_mul_sub_view_10:352"], "sub_7356": ["triton_poi_fused__to_copy_mul_sub_view_10:352"], "convert_element_type_964": ["triton_poi_fused__to_copy_mul_sub_view_10:352"], "view_2517": ["triton_poi_fused__to_copy_mul_sub_view_10:352"], "convert_element_type_965": ["triton_poi_fused__to_copy_mul_sub_view_10:352"], "view_2519": ["triton_poi_fused__to_copy_mul_sub_view_10:352"], "view_2518": ["triton_poi_fused__to_copy_mul_sub_view_10:352"], "amax_161": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "mul_15588": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "mul_15586": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "view_2522": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "add_tensor_26": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "add_24666": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "erf_28": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "mul_15587": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "amin_161": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "add_24749": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "round_324": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "mul_15637": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "mul_15634": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "reciprocal_161": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "view_2525": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "clamp_min_483": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "div_322": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "sub_7369": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "maximum_161": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "full_323": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "minimum_161": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "full_322": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "view_2526": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "convert_element_type_966": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "clamp_max_322": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "clamp_min_484": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "sub_7375": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "round_323": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "div_323": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "mul_15659": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "sub_7395": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "convert_element_type_968": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "convert_element_type_967": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "clamp_max_323": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "clamp_min_485": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "convert_element_type_969": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "view_2530": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "view_2529": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353"], "mul_15664": ["triton_poi_fused__to_copy_mul_sub_view_10:354"], "sub_7399": ["triton_poi_fused__to_copy_mul_sub_view_10:354"], "convert_element_type_970": ["triton_poi_fused__to_copy_mul_sub_view_10:354"], "view_2532": ["triton_poi_fused__to_copy_mul_sub_view_10:354"], "convert_element_type_971": ["triton_poi_fused__to_copy_mul_sub_view_10:354"], "view_2534": ["triton_poi_fused__to_copy_mul_sub_view_10:354"], "view_2533": ["triton_poi_fused__to_copy_mul_sub_view_10:354"], "var_mean_54": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "clone_217": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "add_24812": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "view_2537": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "add_tensor_25": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "add_24818": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "mul_15681": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "mul_15680": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "sub_7405": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "rsqrt_54": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "add_24817": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "amax_162": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "amin_162": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "amax_163": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "amin_163": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "amax_164": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "amin_164": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "sub_7442": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "convert_element_type_974": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "convert_element_type_973": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "clamp_max_325": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "clamp_min_488": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "add_24905": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "round_326": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "mul_15732": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "mul_15729": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "reciprocal_162": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "view_2540": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "clamp_min_486": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "div_324": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "sub_7416": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "maximum_162": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "full_325": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "minimum_162": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "full_324": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "view_2541": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "convert_element_type_972": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "clamp_max_324": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "clamp_min_487": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "sub_7422": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "round_325": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "div_325": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "convert_element_type_975": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "view_2545": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "mul_15754": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "view_2544": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "sub_7487": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "convert_element_type_980": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "convert_element_type_979": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "clamp_max_327": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "clamp_min_491": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "add_25057": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "round_328": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "mul_15828": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "mul_15825": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "reciprocal_163": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "view_2556": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "clamp_min_489": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "div_326": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "sub_7461": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "maximum_163": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "full_327": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "minimum_163": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "full_326": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "view_2557": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "convert_element_type_978": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "clamp_max_326": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "clamp_min_490": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "sub_7467": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "round_327": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "div_327": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "convert_element_type_981": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "view_2561": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "mul_15850": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "view_2560": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "sub_7531": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "convert_element_type_986": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "convert_element_type_985": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "clamp_max_329": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "clamp_min_494": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "add_25205": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "round_330": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "mul_15927": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "mul_15924": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "reciprocal_164": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "view_2572": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "clamp_min_492": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "div_328": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "sub_7505": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "maximum_164": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "full_329": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "minimum_164": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "full_328": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "view_2573": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "convert_element_type_984": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "clamp_max_328": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "clamp_min_493": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "sub_7511": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "round_329": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "div_329": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "convert_element_type_987": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "view_2577": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "mul_15949": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "view_2576": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355"], "mul_15759": ["triton_poi_fused__to_copy_mul_sub_view_4:356"], "sub_7446": ["triton_poi_fused__to_copy_mul_sub_view_4:356"], "convert_element_type_976": ["triton_poi_fused__to_copy_mul_sub_view_4:356"], "view_2547": ["triton_poi_fused__to_copy_mul_sub_view_4:356"], "convert_element_type_977": ["triton_poi_fused__to_copy_mul_sub_view_4:356"], "view_2549": ["triton_poi_fused__to_copy_mul_sub_view_4:356"], "view_2548": ["triton_poi_fused__to_copy_mul_sub_view_4:356"], "mul_15855": ["triton_poi_fused__to_copy_mul_sub_view_4:357"], "sub_7491": ["triton_poi_fused__to_copy_mul_sub_view_4:357"], "convert_element_type_982": ["triton_poi_fused__to_copy_mul_sub_view_4:357"], "view_2563": ["triton_poi_fused__to_copy_mul_sub_view_4:357"], "convert_element_type_983": ["triton_poi_fused__to_copy_mul_sub_view_4:357"], "view_2565": ["triton_poi_fused__to_copy_mul_sub_view_4:357"], "view_2564": ["triton_poi_fused__to_copy_mul_sub_view_4:357"], "mul_15954": ["triton_poi_fused__to_copy_mul_sub_view_4:358"], "sub_7535": ["triton_poi_fused__to_copy_mul_sub_view_4:358"], "convert_element_type_988": ["triton_poi_fused__to_copy_mul_sub_view_4:358"], "view_2579": ["triton_poi_fused__to_copy_mul_sub_view_4:358"], "convert_element_type_989": ["triton_poi_fused__to_copy_mul_sub_view_4:358"], "view_2581": ["triton_poi_fused__to_copy_mul_sub_view_4:358"], "view_2580": ["triton_poi_fused__to_copy_mul_sub_view_4:358"], "_scaled_dot_product_efficient_attention_27": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:359", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:360", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:361", "at::_ops::_scaled_dot_product_efficient_attention::call:619"], "clone_218": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:359", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:360", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:361", "at::_ops::_scaled_dot_product_efficient_attention::call:619"], "permute_272": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:359", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:360", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:361", "at::_ops::_scaled_dot_product_efficient_attention::call:619"], "view_2553": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:359", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:360", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:361", "at::_ops::_scaled_dot_product_efficient_attention::call:619"], "mul_15771": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:359", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:360", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:361", "at::_ops::_scaled_dot_product_efficient_attention::call:619"], "view_2552": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:359", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:360", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:361", "at::_ops::_scaled_dot_product_efficient_attention::call:619"], "add_tensor_24": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:359", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:360", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:361", "at::_ops::_scaled_dot_product_efficient_attention::call:619"], "clone_219": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:359", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:360", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:361", "at::_ops::_scaled_dot_product_efficient_attention::call:619"], "permute_274": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:359", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:360", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:361", "at::_ops::_scaled_dot_product_efficient_attention::call:619"], "view_2569": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:359", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:360", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:361", "at::_ops::_scaled_dot_product_efficient_attention::call:619"], "view_2568": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:359", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:360", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:361", "at::_ops::_scaled_dot_product_efficient_attention::call:619"], "clone_220": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:359", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:360", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:361", "at::_ops::_scaled_dot_product_efficient_attention::call:619"], "permute_276": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:359", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:360", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:361", "at::_ops::_scaled_dot_product_efficient_attention::call:619"], "view_2585": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:359", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:360", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:361", "at::_ops::_scaled_dot_product_efficient_attention::call:619"], "view_2584": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:359", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:360", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:361", "at::_ops::_scaled_dot_product_efficient_attention::call:619"], "add_tensor_23": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:359", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:360", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:361", "at::_ops::_scaled_dot_product_efficient_attention::call:619"], "amax_165": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "view_2586": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "permute_277": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "amin_165": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "sub_7579": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "convert_element_type_992": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "convert_element_type_991": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "clamp_max_331": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "clamp_min_497": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "add_25369": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "round_332": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "mul_16032": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "mul_16029": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "reciprocal_165": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "view_2589": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "clamp_min_495": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "div_330": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "sub_7553": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "maximum_165": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "full_331": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "minimum_165": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "full_330": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "view_2590": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "convert_element_type_990": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "clamp_max_330": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "clamp_min_496": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "sub_7559": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "round_331": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "div_331": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "convert_element_type_993": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "view_2594": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "mul_16054": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "view_2593": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362"], "mul_16059": ["triton_poi_fused__to_copy_mul_sub_view_4:363"], "sub_7583": ["triton_poi_fused__to_copy_mul_sub_view_4:363"], "convert_element_type_994": ["triton_poi_fused__to_copy_mul_sub_view_4:363"], "view_2596": ["triton_poi_fused__to_copy_mul_sub_view_4:363"], "convert_element_type_995": ["triton_poi_fused__to_copy_mul_sub_view_4:363"], "view_2598": ["triton_poi_fused__to_copy_mul_sub_view_4:363"], "view_2597": ["triton_poi_fused__to_copy_mul_sub_view_4:363"], "add_25432": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "view_2601": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "add_tensor_22": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "var_mean_55": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "clone_222": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "add_25438": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "mul_16076": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "mul_16075": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "sub_7589": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "rsqrt_55": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "add_25437": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "amax_166": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "amin_166": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "sub_7626": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "convert_element_type_998": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "convert_element_type_997": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "clamp_max_333": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "clamp_min_500": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "add_25525": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "round_334": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "mul_16127": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "mul_16124": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "reciprocal_166": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "view_2604": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "clamp_min_498": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "div_332": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "sub_7600": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "maximum_166": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "full_333": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "minimum_166": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "full_332": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "view_2605": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "convert_element_type_996": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "clamp_max_332": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "clamp_min_499": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "sub_7606": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "round_333": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "div_333": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "convert_element_type_999": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "view_2609": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "mul_16149": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "view_2608": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364"], "mul_16154": ["triton_poi_fused__to_copy_mul_sub_view_10:365"], "sub_7630": ["triton_poi_fused__to_copy_mul_sub_view_10:365"], "convert_element_type_1000": ["triton_poi_fused__to_copy_mul_sub_view_10:365"], "view_2611": ["triton_poi_fused__to_copy_mul_sub_view_10:365"], "convert_element_type_1001": ["triton_poi_fused__to_copy_mul_sub_view_10:365"], "view_2613": ["triton_poi_fused__to_copy_mul_sub_view_10:365"], "view_2612": ["triton_poi_fused__to_copy_mul_sub_view_10:365"], "amax_167": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "mul_16168": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "mul_16166": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "view_2616": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "add_tensor_21": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "add_25584": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "erf_29": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "mul_16167": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "amin_167": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "add_25667": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "round_336": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "mul_16217": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "mul_16214": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "reciprocal_167": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "view_2619": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "clamp_min_501": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "div_334": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "sub_7643": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "maximum_167": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "full_335": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "minimum_167": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "full_334": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "view_2620": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "convert_element_type_1002": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "clamp_max_334": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "clamp_min_502": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "sub_7649": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "round_335": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "div_335": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "mul_16239": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "sub_7669": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "convert_element_type_1004": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "convert_element_type_1003": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "clamp_max_335": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "clamp_min_503": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "convert_element_type_1005": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "view_2624": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "view_2623": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366"], "mul_16244": ["triton_poi_fused__to_copy_mul_sub_view_10:367"], "sub_7673": ["triton_poi_fused__to_copy_mul_sub_view_10:367"], "convert_element_type_1006": ["triton_poi_fused__to_copy_mul_sub_view_10:367"], "view_2626": ["triton_poi_fused__to_copy_mul_sub_view_10:367"], "convert_element_type_1007": ["triton_poi_fused__to_copy_mul_sub_view_10:367"], "view_2628": ["triton_poi_fused__to_copy_mul_sub_view_10:367"], "view_2627": ["triton_poi_fused__to_copy_mul_sub_view_10:367"], "var_mean_56": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "clone_225": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "add_25730": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "view_2631": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "add_tensor_20": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "add_25736": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "mul_16261": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "mul_16260": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "sub_7679": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "rsqrt_56": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "add_25735": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "amax_168": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "amin_168": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "amax_169": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "amin_169": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "amax_170": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "amin_170": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "sub_7716": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "convert_element_type_1010": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "convert_element_type_1009": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "clamp_max_337": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "clamp_min_506": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "add_25823": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "round_338": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "mul_16312": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "mul_16309": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "reciprocal_168": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "view_2634": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "clamp_min_504": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "div_336": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "sub_7690": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "maximum_168": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "full_337": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "minimum_168": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "full_336": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "view_2635": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "convert_element_type_1008": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "clamp_max_336": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "clamp_min_505": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "sub_7696": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "round_337": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "div_337": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "convert_element_type_1011": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "view_2639": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "mul_16334": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "view_2638": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "sub_7761": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "convert_element_type_1016": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "convert_element_type_1015": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "clamp_max_339": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "clamp_min_509": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "add_25975": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "round_340": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "mul_16408": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "mul_16405": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "reciprocal_169": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "view_2650": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "clamp_min_507": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "div_338": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "sub_7735": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "maximum_169": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "full_339": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "minimum_169": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "full_338": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "view_2651": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "convert_element_type_1014": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "clamp_max_338": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "clamp_min_508": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "sub_7741": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "round_339": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "div_339": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "convert_element_type_1017": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "view_2655": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "mul_16430": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "view_2654": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "sub_7805": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "convert_element_type_1022": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "convert_element_type_1021": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "clamp_max_341": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "clamp_min_512": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "add_26123": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "round_342": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "mul_16507": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "mul_16504": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "reciprocal_170": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "view_2666": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "clamp_min_510": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "div_340": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "sub_7779": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "maximum_170": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "full_341": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "minimum_170": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "full_340": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "view_2667": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "convert_element_type_1020": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "clamp_max_340": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "clamp_min_511": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "sub_7785": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "round_341": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "div_341": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "convert_element_type_1023": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "view_2671": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "mul_16529": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "view_2670": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368"], "mul_16339": ["triton_poi_fused__to_copy_mul_sub_view_4:369"], "sub_7720": ["triton_poi_fused__to_copy_mul_sub_view_4:369"], "convert_element_type_1012": ["triton_poi_fused__to_copy_mul_sub_view_4:369"], "view_2641": ["triton_poi_fused__to_copy_mul_sub_view_4:369"], "convert_element_type_1013": ["triton_poi_fused__to_copy_mul_sub_view_4:369"], "view_2643": ["triton_poi_fused__to_copy_mul_sub_view_4:369"], "view_2642": ["triton_poi_fused__to_copy_mul_sub_view_4:369"], "mul_16435": ["triton_poi_fused__to_copy_mul_sub_view_4:370"], "sub_7765": ["triton_poi_fused__to_copy_mul_sub_view_4:370"], "convert_element_type_1018": ["triton_poi_fused__to_copy_mul_sub_view_4:370"], "view_2657": ["triton_poi_fused__to_copy_mul_sub_view_4:370"], "convert_element_type_1019": ["triton_poi_fused__to_copy_mul_sub_view_4:370"], "view_2659": ["triton_poi_fused__to_copy_mul_sub_view_4:370"], "view_2658": ["triton_poi_fused__to_copy_mul_sub_view_4:370"], "mul_16534": ["triton_poi_fused__to_copy_mul_sub_view_4:371"], "sub_7809": ["triton_poi_fused__to_copy_mul_sub_view_4:371"], "convert_element_type_1024": ["triton_poi_fused__to_copy_mul_sub_view_4:371"], "view_2673": ["triton_poi_fused__to_copy_mul_sub_view_4:371"], "convert_element_type_1025": ["triton_poi_fused__to_copy_mul_sub_view_4:371"], "view_2675": ["triton_poi_fused__to_copy_mul_sub_view_4:371"], "view_2674": ["triton_poi_fused__to_copy_mul_sub_view_4:371"], "_scaled_dot_product_efficient_attention_28": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:372", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:373", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:374", "at::_ops::_scaled_dot_product_efficient_attention::call:626"], "clone_226": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:372", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:373", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:374", "at::_ops::_scaled_dot_product_efficient_attention::call:626"], "permute_282": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:372", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:373", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:374", "at::_ops::_scaled_dot_product_efficient_attention::call:626"], "view_2647": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:372", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:373", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:374", "at::_ops::_scaled_dot_product_efficient_attention::call:626"], "mul_16351": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:372", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:373", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:374", "at::_ops::_scaled_dot_product_efficient_attention::call:626"], "view_2646": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:372", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:373", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:374", "at::_ops::_scaled_dot_product_efficient_attention::call:626"], "add_tensor_19": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:372", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:373", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:374", "at::_ops::_scaled_dot_product_efficient_attention::call:626"], "clone_227": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:372", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:373", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:374", "at::_ops::_scaled_dot_product_efficient_attention::call:626"], "permute_284": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:372", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:373", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:374", "at::_ops::_scaled_dot_product_efficient_attention::call:626"], "view_2663": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:372", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:373", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:374", "at::_ops::_scaled_dot_product_efficient_attention::call:626"], "view_2662": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:372", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:373", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:374", "at::_ops::_scaled_dot_product_efficient_attention::call:626"], "clone_228": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:372", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:373", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:374", "at::_ops::_scaled_dot_product_efficient_attention::call:626"], "permute_286": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:372", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:373", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:374", "at::_ops::_scaled_dot_product_efficient_attention::call:626"], "view_2679": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:372", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:373", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:374", "at::_ops::_scaled_dot_product_efficient_attention::call:626"], "view_2678": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:372", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:373", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:374", "at::_ops::_scaled_dot_product_efficient_attention::call:626"], "add_tensor_18": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:372", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:373", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:374", "at::_ops::_scaled_dot_product_efficient_attention::call:626"], "amax_171": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "view_2680": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "permute_287": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "amin_171": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "sub_7853": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "convert_element_type_1028": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "convert_element_type_1027": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "clamp_max_343": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "clamp_min_515": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "add_26287": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "round_344": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "mul_16612": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "mul_16609": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "reciprocal_171": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "view_2683": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "clamp_min_513": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "div_342": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "sub_7827": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "maximum_171": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "full_343": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "minimum_171": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "full_342": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "view_2684": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "convert_element_type_1026": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "clamp_max_342": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "clamp_min_514": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "sub_7833": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "round_343": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "div_343": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "convert_element_type_1029": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "view_2688": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "mul_16634": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "view_2687": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375"], "mul_16639": ["triton_poi_fused__to_copy_mul_sub_view_4:376"], "sub_7857": ["triton_poi_fused__to_copy_mul_sub_view_4:376"], "convert_element_type_1030": ["triton_poi_fused__to_copy_mul_sub_view_4:376"], "view_2690": ["triton_poi_fused__to_copy_mul_sub_view_4:376"], "convert_element_type_1031": ["triton_poi_fused__to_copy_mul_sub_view_4:376"], "view_2692": ["triton_poi_fused__to_copy_mul_sub_view_4:376"], "view_2691": ["triton_poi_fused__to_copy_mul_sub_view_4:376"], "add_26350": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "view_2695": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "add_tensor_17": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "var_mean_57": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "clone_230": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "add_26356": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "mul_16656": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "mul_16655": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "sub_7863": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "rsqrt_57": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "add_26355": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "amax_172": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "amin_172": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "sub_7900": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "convert_element_type_1034": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "convert_element_type_1033": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "clamp_max_345": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "clamp_min_518": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "add_26443": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "round_346": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "mul_16707": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "mul_16704": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "reciprocal_172": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "view_2698": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "clamp_min_516": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "div_344": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "sub_7874": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "maximum_172": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "full_345": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "minimum_172": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "full_344": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "view_2699": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "convert_element_type_1032": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "clamp_max_344": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "clamp_min_517": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "sub_7880": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "round_345": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "div_345": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "convert_element_type_1035": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "view_2703": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "mul_16729": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "view_2702": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377"], "mul_16734": ["triton_poi_fused__to_copy_mul_sub_view_10:378"], "sub_7904": ["triton_poi_fused__to_copy_mul_sub_view_10:378"], "convert_element_type_1036": ["triton_poi_fused__to_copy_mul_sub_view_10:378"], "view_2705": ["triton_poi_fused__to_copy_mul_sub_view_10:378"], "convert_element_type_1037": ["triton_poi_fused__to_copy_mul_sub_view_10:378"], "view_2707": ["triton_poi_fused__to_copy_mul_sub_view_10:378"], "view_2706": ["triton_poi_fused__to_copy_mul_sub_view_10:378"], "amax_173": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "mul_16748": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "mul_16746": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "view_2710": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "add_tensor_16": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "add_26502": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "erf_30": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "mul_16747": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "amin_173": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "add_26585": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "round_348": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "mul_16797": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "mul_16794": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "reciprocal_173": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "view_2713": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "clamp_min_519": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "div_346": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "sub_7917": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "maximum_173": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "full_347": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "minimum_173": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "full_346": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "view_2714": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "convert_element_type_1038": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "clamp_max_346": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "clamp_min_520": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "sub_7923": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "round_347": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "div_347": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "mul_16819": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "sub_7943": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "convert_element_type_1040": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "convert_element_type_1039": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "clamp_max_347": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "clamp_min_521": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "convert_element_type_1041": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "view_2718": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "view_2717": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379"], "mul_16824": ["triton_poi_fused__to_copy_mul_sub_view_10:380"], "sub_7947": ["triton_poi_fused__to_copy_mul_sub_view_10:380"], "convert_element_type_1042": ["triton_poi_fused__to_copy_mul_sub_view_10:380"], "view_2720": ["triton_poi_fused__to_copy_mul_sub_view_10:380"], "convert_element_type_1043": ["triton_poi_fused__to_copy_mul_sub_view_10:380"], "view_2722": ["triton_poi_fused__to_copy_mul_sub_view_10:380"], "view_2721": ["triton_poi_fused__to_copy_mul_sub_view_10:380"], "var_mean_58": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "clone_233": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "add_26648": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "view_2725": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "add_tensor_15": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "add_26654": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "mul_16841": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "mul_16840": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "sub_7953": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "rsqrt_58": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "add_26653": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "amax_174": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "amin_174": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "amax_175": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "amin_175": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "amax_176": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "amin_176": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "sub_7990": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "convert_element_type_1046": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "convert_element_type_1045": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "clamp_max_349": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "clamp_min_524": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "add_26741": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "round_350": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "mul_16892": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "mul_16889": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "reciprocal_174": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "view_2728": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "clamp_min_522": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "div_348": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "sub_7964": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "maximum_174": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "full_349": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "minimum_174": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "full_348": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "view_2729": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "convert_element_type_1044": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "clamp_max_348": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "clamp_min_523": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "sub_7970": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "round_349": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "div_349": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "convert_element_type_1047": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "view_2733": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "mul_16914": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "view_2732": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "sub_8035": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "convert_element_type_1052": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "convert_element_type_1051": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "clamp_max_351": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "clamp_min_527": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "add_26893": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "round_352": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "mul_16988": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "mul_16985": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "reciprocal_175": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "view_2744": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "clamp_min_525": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "div_350": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "sub_8009": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "maximum_175": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "full_351": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "minimum_175": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "full_350": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "view_2745": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "convert_element_type_1050": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "clamp_max_350": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "clamp_min_526": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "sub_8015": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "round_351": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "div_351": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "convert_element_type_1053": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "view_2749": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "mul_17010": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "view_2748": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "sub_8079": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "convert_element_type_1058": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "convert_element_type_1057": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "clamp_max_353": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "clamp_min_530": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "add_27041": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "round_354": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "mul_17087": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "mul_17084": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "reciprocal_176": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "view_2760": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "clamp_min_528": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "div_352": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "sub_8053": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "maximum_176": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "full_353": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "minimum_176": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "full_352": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "view_2761": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "convert_element_type_1056": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "clamp_max_352": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "clamp_min_529": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "sub_8059": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "round_353": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "div_353": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "convert_element_type_1059": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "view_2765": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "mul_17109": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "view_2764": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381"], "mul_16919": ["triton_poi_fused__to_copy_mul_sub_view_4:382"], "sub_7994": ["triton_poi_fused__to_copy_mul_sub_view_4:382"], "convert_element_type_1048": ["triton_poi_fused__to_copy_mul_sub_view_4:382"], "view_2735": ["triton_poi_fused__to_copy_mul_sub_view_4:382"], "convert_element_type_1049": ["triton_poi_fused__to_copy_mul_sub_view_4:382"], "view_2737": ["triton_poi_fused__to_copy_mul_sub_view_4:382"], "view_2736": ["triton_poi_fused__to_copy_mul_sub_view_4:382"], "mul_17015": ["triton_poi_fused__to_copy_mul_sub_view_4:383"], "sub_8039": ["triton_poi_fused__to_copy_mul_sub_view_4:383"], "convert_element_type_1054": ["triton_poi_fused__to_copy_mul_sub_view_4:383"], "view_2751": ["triton_poi_fused__to_copy_mul_sub_view_4:383"], "convert_element_type_1055": ["triton_poi_fused__to_copy_mul_sub_view_4:383"], "view_2753": ["triton_poi_fused__to_copy_mul_sub_view_4:383"], "view_2752": ["triton_poi_fused__to_copy_mul_sub_view_4:383"], "mul_17114": ["triton_poi_fused__to_copy_mul_sub_view_4:384"], "sub_8083": ["triton_poi_fused__to_copy_mul_sub_view_4:384"], "convert_element_type_1060": ["triton_poi_fused__to_copy_mul_sub_view_4:384"], "view_2767": ["triton_poi_fused__to_copy_mul_sub_view_4:384"], "convert_element_type_1061": ["triton_poi_fused__to_copy_mul_sub_view_4:384"], "view_2769": ["triton_poi_fused__to_copy_mul_sub_view_4:384"], "view_2768": ["triton_poi_fused__to_copy_mul_sub_view_4:384"], "_scaled_dot_product_efficient_attention_29": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:385", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:386", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:387", "at::_ops::_scaled_dot_product_efficient_attention::call:633"], "clone_234": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:385", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:386", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:387", "at::_ops::_scaled_dot_product_efficient_attention::call:633"], "permute_292": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:385", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:386", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:387", "at::_ops::_scaled_dot_product_efficient_attention::call:633"], "view_2741": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:385", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:386", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:387", "at::_ops::_scaled_dot_product_efficient_attention::call:633"], "mul_16931": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:385", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:386", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:387", "at::_ops::_scaled_dot_product_efficient_attention::call:633"], "view_2740": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:385", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:386", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:387", "at::_ops::_scaled_dot_product_efficient_attention::call:633"], "add_tensor_14": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:385", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:386", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:387", "at::_ops::_scaled_dot_product_efficient_attention::call:633"], "clone_235": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:385", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:386", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:387", "at::_ops::_scaled_dot_product_efficient_attention::call:633"], "permute_294": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:385", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:386", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:387", "at::_ops::_scaled_dot_product_efficient_attention::call:633"], "view_2757": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:385", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:386", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:387", "at::_ops::_scaled_dot_product_efficient_attention::call:633"], "view_2756": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:385", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:386", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:387", "at::_ops::_scaled_dot_product_efficient_attention::call:633"], "clone_236": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:385", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:386", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:387", "at::_ops::_scaled_dot_product_efficient_attention::call:633"], "permute_296": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:385", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:386", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:387", "at::_ops::_scaled_dot_product_efficient_attention::call:633"], "view_2773": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:385", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:386", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:387", "at::_ops::_scaled_dot_product_efficient_attention::call:633"], "view_2772": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:385", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:386", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:387", "at::_ops::_scaled_dot_product_efficient_attention::call:633"], "add_tensor_13": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:385", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:386", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:387", "at::_ops::_scaled_dot_product_efficient_attention::call:633"], "amax_177": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "view_2774": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "permute_297": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "amin_177": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "sub_8127": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "convert_element_type_1064": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "convert_element_type_1063": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "clamp_max_355": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "clamp_min_533": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "add_27205": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "round_356": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "mul_17192": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "mul_17189": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "reciprocal_177": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "view_2777": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "clamp_min_531": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "div_354": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "sub_8101": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "maximum_177": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "full_355": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "minimum_177": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "full_354": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "view_2778": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "convert_element_type_1062": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "clamp_max_354": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "clamp_min_532": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "sub_8107": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "round_355": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "div_355": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "convert_element_type_1065": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "view_2782": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "mul_17214": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "view_2781": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388"], "mul_17219": ["triton_poi_fused__to_copy_mul_sub_view_4:389"], "sub_8131": ["triton_poi_fused__to_copy_mul_sub_view_4:389"], "convert_element_type_1066": ["triton_poi_fused__to_copy_mul_sub_view_4:389"], "view_2784": ["triton_poi_fused__to_copy_mul_sub_view_4:389"], "convert_element_type_1067": ["triton_poi_fused__to_copy_mul_sub_view_4:389"], "view_2786": ["triton_poi_fused__to_copy_mul_sub_view_4:389"], "view_2785": ["triton_poi_fused__to_copy_mul_sub_view_4:389"], "add_27268": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "view_2789": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "add_tensor_12": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "var_mean_59": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "clone_238": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "add_27274": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "mul_17236": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "mul_17235": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "sub_8137": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "rsqrt_59": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "add_27273": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "amax_178": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "amin_178": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "sub_8174": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "convert_element_type_1070": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "convert_element_type_1069": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "clamp_max_357": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "clamp_min_536": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "add_27361": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "round_358": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "mul_17287": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "mul_17284": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "reciprocal_178": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "view_2792": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "clamp_min_534": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "div_356": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "sub_8148": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "maximum_178": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "full_357": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "minimum_178": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "full_356": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "view_2793": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "convert_element_type_1068": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "clamp_max_356": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "clamp_min_535": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "sub_8154": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "round_357": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "div_357": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "convert_element_type_1071": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "view_2797": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "mul_17309": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "view_2796": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390"], "mul_17314": ["triton_poi_fused__to_copy_mul_sub_view_10:391"], "sub_8178": ["triton_poi_fused__to_copy_mul_sub_view_10:391"], "convert_element_type_1072": ["triton_poi_fused__to_copy_mul_sub_view_10:391"], "view_2799": ["triton_poi_fused__to_copy_mul_sub_view_10:391"], "convert_element_type_1073": ["triton_poi_fused__to_copy_mul_sub_view_10:391"], "view_2801": ["triton_poi_fused__to_copy_mul_sub_view_10:391"], "view_2800": ["triton_poi_fused__to_copy_mul_sub_view_10:391"], "amax_179": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "mul_17328": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "mul_17326": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "view_2804": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "add_tensor_11": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "add_27420": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "erf_31": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "mul_17327": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "amin_179": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "add_27503": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "round_360": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "mul_17377": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "mul_17374": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "reciprocal_179": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "view_2807": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "clamp_min_537": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "div_358": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "sub_8191": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "maximum_179": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "full_359": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "minimum_179": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "full_358": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "view_2808": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "convert_element_type_1074": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "clamp_max_358": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "clamp_min_538": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "sub_8197": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "round_359": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "div_359": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "mul_17399": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "sub_8217": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "convert_element_type_1076": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "convert_element_type_1075": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "clamp_max_359": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "clamp_min_539": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "convert_element_type_1077": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "view_2812": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "view_2811": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392"], "mul_17404": ["triton_poi_fused__to_copy_mul_sub_view_10:393"], "sub_8221": ["triton_poi_fused__to_copy_mul_sub_view_10:393"], "convert_element_type_1078": ["triton_poi_fused__to_copy_mul_sub_view_10:393"], "view_2814": ["triton_poi_fused__to_copy_mul_sub_view_10:393"], "convert_element_type_1079": ["triton_poi_fused__to_copy_mul_sub_view_10:393"], "view_2816": ["triton_poi_fused__to_copy_mul_sub_view_10:393"], "view_2815": ["triton_poi_fused__to_copy_mul_sub_view_10:393"], "var_mean_60": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "clone_241": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "add_27566": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "view_2819": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "add_tensor_10": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "add_27572": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "mul_17421": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "mul_17420": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "sub_8227": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "rsqrt_60": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "add_27571": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "amax_180": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "amin_180": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "amax_181": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "amin_181": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "amax_182": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "amin_182": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "sub_8264": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "convert_element_type_1082": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "convert_element_type_1081": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "clamp_max_361": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "clamp_min_542": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "add_27659": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "round_362": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "mul_17472": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "mul_17469": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "reciprocal_180": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "view_2822": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "clamp_min_540": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "div_360": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "sub_8238": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "maximum_180": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "full_361": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "minimum_180": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "full_360": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "view_2823": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "convert_element_type_1080": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "clamp_max_360": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "clamp_min_541": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "sub_8244": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "round_361": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "div_361": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "convert_element_type_1083": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "view_2827": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "mul_17494": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "view_2826": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "sub_8309": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "convert_element_type_1088": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "convert_element_type_1087": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "clamp_max_363": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "clamp_min_545": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "add_27811": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "round_364": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "mul_17568": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "mul_17565": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "reciprocal_181": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "view_2838": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "clamp_min_543": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "div_362": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "sub_8283": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "maximum_181": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "full_363": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "minimum_181": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "full_362": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "view_2839": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "convert_element_type_1086": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "clamp_max_362": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "clamp_min_544": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "sub_8289": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "round_363": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "div_363": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "convert_element_type_1089": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "view_2843": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "mul_17590": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "view_2842": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "sub_8353": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "convert_element_type_1094": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "convert_element_type_1093": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "clamp_max_365": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "clamp_min_548": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "add_27959": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "round_366": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "mul_17667": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "mul_17664": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "reciprocal_182": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "view_2854": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "clamp_min_546": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "div_364": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "sub_8327": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "maximum_182": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "full_365": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "minimum_182": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "full_364": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "view_2855": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "convert_element_type_1092": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "clamp_max_364": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "clamp_min_547": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "sub_8333": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "round_365": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "div_365": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "convert_element_type_1095": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "view_2859": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "mul_17689": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "view_2858": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394"], "mul_17499": ["triton_poi_fused__to_copy_mul_sub_view_4:395"], "sub_8268": ["triton_poi_fused__to_copy_mul_sub_view_4:395"], "convert_element_type_1084": ["triton_poi_fused__to_copy_mul_sub_view_4:395"], "view_2829": ["triton_poi_fused__to_copy_mul_sub_view_4:395"], "convert_element_type_1085": ["triton_poi_fused__to_copy_mul_sub_view_4:395"], "view_2831": ["triton_poi_fused__to_copy_mul_sub_view_4:395"], "view_2830": ["triton_poi_fused__to_copy_mul_sub_view_4:395"], "mul_17595": ["triton_poi_fused__to_copy_mul_sub_view_4:396"], "sub_8313": ["triton_poi_fused__to_copy_mul_sub_view_4:396"], "convert_element_type_1090": ["triton_poi_fused__to_copy_mul_sub_view_4:396"], "view_2845": ["triton_poi_fused__to_copy_mul_sub_view_4:396"], "convert_element_type_1091": ["triton_poi_fused__to_copy_mul_sub_view_4:396"], "view_2847": ["triton_poi_fused__to_copy_mul_sub_view_4:396"], "view_2846": ["triton_poi_fused__to_copy_mul_sub_view_4:396"], "mul_17694": ["triton_poi_fused__to_copy_mul_sub_view_4:397"], "sub_8357": ["triton_poi_fused__to_copy_mul_sub_view_4:397"], "convert_element_type_1096": ["triton_poi_fused__to_copy_mul_sub_view_4:397"], "view_2861": ["triton_poi_fused__to_copy_mul_sub_view_4:397"], "convert_element_type_1097": ["triton_poi_fused__to_copy_mul_sub_view_4:397"], "view_2863": ["triton_poi_fused__to_copy_mul_sub_view_4:397"], "view_2862": ["triton_poi_fused__to_copy_mul_sub_view_4:397"], "_scaled_dot_product_efficient_attention_30": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:398", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:399", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:400", "at::_ops::_scaled_dot_product_efficient_attention::call:640"], "clone_242": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:398", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:399", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:400", "at::_ops::_scaled_dot_product_efficient_attention::call:640"], "permute_302": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:398", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:399", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:400", "at::_ops::_scaled_dot_product_efficient_attention::call:640"], "view_2835": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:398", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:399", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:400", "at::_ops::_scaled_dot_product_efficient_attention::call:640"], "mul_17511": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:398", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:399", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:400", "at::_ops::_scaled_dot_product_efficient_attention::call:640"], "view_2834": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:398", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:399", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:400", "at::_ops::_scaled_dot_product_efficient_attention::call:640"], "add_tensor_9": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:398", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:399", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:400", "at::_ops::_scaled_dot_product_efficient_attention::call:640"], "clone_243": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:398", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:399", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:400", "at::_ops::_scaled_dot_product_efficient_attention::call:640"], "permute_304": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:398", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:399", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:400", "at::_ops::_scaled_dot_product_efficient_attention::call:640"], "view_2851": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:398", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:399", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:400", "at::_ops::_scaled_dot_product_efficient_attention::call:640"], "view_2850": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:398", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:399", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:400", "at::_ops::_scaled_dot_product_efficient_attention::call:640"], "clone_244": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:398", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:399", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:400", "at::_ops::_scaled_dot_product_efficient_attention::call:640"], "permute_306": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:398", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:399", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:400", "at::_ops::_scaled_dot_product_efficient_attention::call:640"], "view_2867": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:398", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:399", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:400", "at::_ops::_scaled_dot_product_efficient_attention::call:640"], "view_2866": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:398", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:399", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:400", "at::_ops::_scaled_dot_product_efficient_attention::call:640"], "add_tensor_8": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:398", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:399", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:400", "at::_ops::_scaled_dot_product_efficient_attention::call:640"], "amax_183": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "view_2868": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "permute_307": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "amin_183": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "sub_8401": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "convert_element_type_1100": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "convert_element_type_1099": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "clamp_max_367": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "clamp_min_551": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "add_28123": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "round_368": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "mul_17772": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "mul_17769": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "reciprocal_183": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "view_2871": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "clamp_min_549": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "div_366": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "sub_8375": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "maximum_183": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "full_367": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "minimum_183": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "full_366": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "view_2872": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "convert_element_type_1098": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "clamp_max_366": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "clamp_min_550": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "sub_8381": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "round_367": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "div_367": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "convert_element_type_1101": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "view_2876": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "mul_17794": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "view_2875": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401"], "mul_17799": ["triton_poi_fused__to_copy_mul_sub_view_4:402"], "sub_8405": ["triton_poi_fused__to_copy_mul_sub_view_4:402"], "convert_element_type_1102": ["triton_poi_fused__to_copy_mul_sub_view_4:402"], "view_2878": ["triton_poi_fused__to_copy_mul_sub_view_4:402"], "convert_element_type_1103": ["triton_poi_fused__to_copy_mul_sub_view_4:402"], "view_2880": ["triton_poi_fused__to_copy_mul_sub_view_4:402"], "view_2879": ["triton_poi_fused__to_copy_mul_sub_view_4:402"], "add_28186": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "view_2883": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "add_tensor_7": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "var_mean_61": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "clone_246": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "add_28192": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "mul_17816": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "mul_17815": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "sub_8411": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "rsqrt_61": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "add_28191": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "amax_184": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "amin_184": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "sub_8448": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "convert_element_type_1106": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "convert_element_type_1105": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "clamp_max_369": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "clamp_min_554": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "add_28279": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "round_370": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "mul_17867": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "mul_17864": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "reciprocal_184": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "view_2886": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "clamp_min_552": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "div_368": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "sub_8422": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "maximum_184": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "full_369": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "minimum_184": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "full_368": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "view_2887": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "convert_element_type_1104": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "clamp_max_368": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "clamp_min_553": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "sub_8428": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "round_369": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "div_369": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "convert_element_type_1107": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "view_2891": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "mul_17889": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "view_2890": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403"], "mul_17894": ["triton_poi_fused__to_copy_mul_sub_view_10:404"], "sub_8452": ["triton_poi_fused__to_copy_mul_sub_view_10:404"], "convert_element_type_1108": ["triton_poi_fused__to_copy_mul_sub_view_10:404"], "view_2893": ["triton_poi_fused__to_copy_mul_sub_view_10:404"], "convert_element_type_1109": ["triton_poi_fused__to_copy_mul_sub_view_10:404"], "view_2895": ["triton_poi_fused__to_copy_mul_sub_view_10:404"], "view_2894": ["triton_poi_fused__to_copy_mul_sub_view_10:404"], "amax_185": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "mul_17908": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "mul_17906": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "view_2898": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "add_tensor_6": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "add_28338": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "erf_32": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "mul_17907": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "amin_185": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "add_28421": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "round_372": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "mul_17957": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "mul_17954": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "reciprocal_185": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "view_2901": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "clamp_min_555": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "div_370": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "sub_8465": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "maximum_185": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "full_371": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "minimum_185": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "full_370": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "view_2902": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "convert_element_type_1110": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "clamp_max_370": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "clamp_min_556": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "sub_8471": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "round_371": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "div_371": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "mul_17979": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "sub_8491": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "convert_element_type_1112": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "convert_element_type_1111": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "clamp_max_371": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "clamp_min_557": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "convert_element_type_1113": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "view_2906": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "view_2905": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405"], "mul_17984": ["triton_poi_fused__to_copy_mul_sub_view_10:406"], "sub_8495": ["triton_poi_fused__to_copy_mul_sub_view_10:406"], "convert_element_type_1114": ["triton_poi_fused__to_copy_mul_sub_view_10:406"], "view_2908": ["triton_poi_fused__to_copy_mul_sub_view_10:406"], "convert_element_type_1115": ["triton_poi_fused__to_copy_mul_sub_view_10:406"], "view_2910": ["triton_poi_fused__to_copy_mul_sub_view_10:406"], "view_2909": ["triton_poi_fused__to_copy_mul_sub_view_10:406"], "var_mean_62": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "clone_249": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "add_28484": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "view_2913": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "add_tensor_5": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "add_28490": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "mul_18001": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "mul_18000": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "sub_8501": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "rsqrt_62": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "add_28489": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "amax_186": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "amin_186": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "amax_187": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "amin_187": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "amax_188": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "amin_188": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "sub_8538": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "convert_element_type_1118": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "convert_element_type_1117": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "clamp_max_373": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "clamp_min_560": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "add_28577": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "round_374": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "mul_18052": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "mul_18049": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "reciprocal_186": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "view_2916": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "clamp_min_558": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "div_372": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "sub_8512": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "maximum_186": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "full_373": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "minimum_186": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "full_372": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "view_2917": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "convert_element_type_1116": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "clamp_max_372": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "clamp_min_559": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "sub_8518": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "round_373": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "div_373": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "convert_element_type_1119": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "view_2921": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "mul_18074": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "view_2920": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "sub_8583": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "convert_element_type_1124": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "convert_element_type_1123": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "clamp_max_375": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "clamp_min_563": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "add_28729": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "round_376": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "mul_18148": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "mul_18145": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "reciprocal_187": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "view_2932": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "clamp_min_561": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "div_374": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "sub_8557": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "maximum_187": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "full_375": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "minimum_187": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "full_374": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "view_2933": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "convert_element_type_1122": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "clamp_max_374": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "clamp_min_562": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "sub_8563": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "round_375": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "div_375": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "convert_element_type_1125": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "view_2937": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "mul_18170": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "view_2936": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "sub_8627": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "convert_element_type_1130": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "convert_element_type_1129": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "clamp_max_377": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "clamp_min_566": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "add_28877": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "round_378": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "mul_18247": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "mul_18244": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "reciprocal_188": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "view_2948": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "clamp_min_564": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "div_376": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "sub_8601": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "maximum_188": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "full_377": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "minimum_188": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "full_376": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "view_2949": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "convert_element_type_1128": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "clamp_max_376": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "clamp_min_565": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "sub_8607": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "round_377": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "div_377": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "convert_element_type_1131": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "view_2953": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "mul_18269": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "view_2952": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407"], "mul_18079": ["triton_poi_fused__to_copy_mul_sub_view_4:408"], "sub_8542": ["triton_poi_fused__to_copy_mul_sub_view_4:408"], "convert_element_type_1120": ["triton_poi_fused__to_copy_mul_sub_view_4:408"], "view_2923": ["triton_poi_fused__to_copy_mul_sub_view_4:408"], "convert_element_type_1121": ["triton_poi_fused__to_copy_mul_sub_view_4:408"], "view_2925": ["triton_poi_fused__to_copy_mul_sub_view_4:408"], "view_2924": ["triton_poi_fused__to_copy_mul_sub_view_4:408"], "mul_18175": ["triton_poi_fused__to_copy_mul_sub_view_4:409"], "sub_8587": ["triton_poi_fused__to_copy_mul_sub_view_4:409"], "convert_element_type_1126": ["triton_poi_fused__to_copy_mul_sub_view_4:409"], "view_2939": ["triton_poi_fused__to_copy_mul_sub_view_4:409"], "convert_element_type_1127": ["triton_poi_fused__to_copy_mul_sub_view_4:409"], "view_2941": ["triton_poi_fused__to_copy_mul_sub_view_4:409"], "view_2940": ["triton_poi_fused__to_copy_mul_sub_view_4:409"], "mul_18274": ["triton_poi_fused__to_copy_mul_sub_view_4:410"], "sub_8631": ["triton_poi_fused__to_copy_mul_sub_view_4:410"], "convert_element_type_1132": ["triton_poi_fused__to_copy_mul_sub_view_4:410"], "view_2955": ["triton_poi_fused__to_copy_mul_sub_view_4:410"], "convert_element_type_1133": ["triton_poi_fused__to_copy_mul_sub_view_4:410"], "view_2957": ["triton_poi_fused__to_copy_mul_sub_view_4:410"], "view_2956": ["triton_poi_fused__to_copy_mul_sub_view_4:410"], "_scaled_dot_product_efficient_attention_31": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:411", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:412", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:413", "at::_ops::_scaled_dot_product_efficient_attention::call:647"], "clone_250": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:411", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:412", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:413", "at::_ops::_scaled_dot_product_efficient_attention::call:647"], "permute_312": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:411", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:412", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:413", "at::_ops::_scaled_dot_product_efficient_attention::call:647"], "view_2929": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:411", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:412", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:413", "at::_ops::_scaled_dot_product_efficient_attention::call:647"], "mul_18091": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:411", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:412", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:413", "at::_ops::_scaled_dot_product_efficient_attention::call:647"], "view_2928": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:411", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:412", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:413", "at::_ops::_scaled_dot_product_efficient_attention::call:647"], "add_tensor_4": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:411", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:412", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:413", "at::_ops::_scaled_dot_product_efficient_attention::call:647"], "clone_251": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:411", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:412", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:413", "at::_ops::_scaled_dot_product_efficient_attention::call:647"], "permute_314": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:411", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:412", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:413", "at::_ops::_scaled_dot_product_efficient_attention::call:647"], "view_2945": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:411", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:412", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:413", "at::_ops::_scaled_dot_product_efficient_attention::call:647"], "view_2944": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:411", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:412", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:413", "at::_ops::_scaled_dot_product_efficient_attention::call:647"], "clone_252": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:411", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:412", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:413", "at::_ops::_scaled_dot_product_efficient_attention::call:647"], "permute_316": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:411", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:412", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:413", "at::_ops::_scaled_dot_product_efficient_attention::call:647"], "view_2961": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:411", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:412", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:413", "at::_ops::_scaled_dot_product_efficient_attention::call:647"], "view_2960": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:411", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:412", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:413", "at::_ops::_scaled_dot_product_efficient_attention::call:647"], "add_tensor_3": ["triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:411", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:412", "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:413", "at::_ops::_scaled_dot_product_efficient_attention::call:647"], "amax_189": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "view_2962": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "permute_317": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "amin_189": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "sub_8675": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "convert_element_type_1136": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "convert_element_type_1135": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "clamp_max_379": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "clamp_min_569": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "add_29041": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "round_380": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "mul_18352": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "mul_18349": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "reciprocal_189": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "view_2965": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "clamp_min_567": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "div_378": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "sub_8649": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "maximum_189": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "full_379": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "minimum_189": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "full_378": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "view_2966": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "convert_element_type_1134": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "clamp_max_378": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "clamp_min_568": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "sub_8655": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "round_379": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "div_379": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "convert_element_type_1137": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "view_2970": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "mul_18374": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "view_2969": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414"], "mul_18379": ["triton_poi_fused__to_copy_mul_sub_view_4:415"], "sub_8679": ["triton_poi_fused__to_copy_mul_sub_view_4:415"], "convert_element_type_1138": ["triton_poi_fused__to_copy_mul_sub_view_4:415"], "view_2972": ["triton_poi_fused__to_copy_mul_sub_view_4:415"], "convert_element_type_1139": ["triton_poi_fused__to_copy_mul_sub_view_4:415"], "view_2974": ["triton_poi_fused__to_copy_mul_sub_view_4:415"], "view_2973": ["triton_poi_fused__to_copy_mul_sub_view_4:415"], "add_29104": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "view_2977": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "add_tensor_2": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "var_mean_63": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "clone_254": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "add_29110": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "mul_18396": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "mul_18395": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "sub_8685": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "rsqrt_63": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "add_29109": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "amax_190": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "amin_190": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "sub_8722": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "convert_element_type_1142": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "convert_element_type_1141": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "clamp_max_381": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "clamp_min_572": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "add_29197": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "round_382": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "mul_18447": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "mul_18444": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "reciprocal_190": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "view_2980": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "clamp_min_570": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "div_380": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "sub_8696": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "maximum_190": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "full_381": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "minimum_190": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "full_380": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "view_2981": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "convert_element_type_1140": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "clamp_max_380": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "clamp_min_571": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "sub_8702": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "round_381": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "div_381": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "convert_element_type_1143": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "view_2985": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "mul_18469": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "view_2984": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416"], "mul_18474": ["triton_poi_fused__to_copy_mul_sub_view_10:417"], "sub_8726": ["triton_poi_fused__to_copy_mul_sub_view_10:417"], "convert_element_type_1144": ["triton_poi_fused__to_copy_mul_sub_view_10:417"], "view_2987": ["triton_poi_fused__to_copy_mul_sub_view_10:417"], "convert_element_type_1145": ["triton_poi_fused__to_copy_mul_sub_view_10:417"], "view_2989": ["triton_poi_fused__to_copy_mul_sub_view_10:417"], "view_2988": ["triton_poi_fused__to_copy_mul_sub_view_10:417"], "amax_191": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "mul_18488": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "mul_18486": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "view_2992": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "add_tensor_1": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "add_29256": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "erf_33": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "mul_18487": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "amin_191": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "add_29339": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "round_384": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "mul_18537": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "mul_18534": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "reciprocal_191": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "view_2995": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "clamp_min_573": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "div_382": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "sub_8739": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "maximum_191": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "full_383": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "minimum_191": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "full_382": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "view_2996": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "convert_element_type_1146": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "clamp_max_382": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "clamp_min_574": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "sub_8745": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "round_383": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "div_383": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "mul_18559": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "sub_8765": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "convert_element_type_1148": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "convert_element_type_1147": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "clamp_max_383": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "clamp_min_575": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "convert_element_type_1149": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "view_3000": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "view_2999": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418"], "mul_18564": ["triton_poi_fused__to_copy_mul_sub_view_10:419"], "sub_8769": ["triton_poi_fused__to_copy_mul_sub_view_10:419"], "convert_element_type_1150": ["triton_poi_fused__to_copy_mul_sub_view_10:419"], "view_3002": ["triton_poi_fused__to_copy_mul_sub_view_10:419"], "convert_element_type_1151": ["triton_poi_fused__to_copy_mul_sub_view_10:419"], "view_3004": ["triton_poi_fused__to_copy_mul_sub_view_10:419"], "view_3003": ["triton_poi_fused__to_copy_mul_sub_view_10:419"], "var_mean_64": ["triton_red_fused_add_addmm_native_layer_norm_view_14:420", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "clone_257": ["triton_red_fused_add_addmm_native_layer_norm_view_14:420", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "add_29402": ["triton_red_fused_add_addmm_native_layer_norm_view_14:420", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "view_3007": ["triton_red_fused_add_addmm_native_layer_norm_view_14:420", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "add_tensor": ["triton_red_fused_add_addmm_native_layer_norm_view_14:420", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "add_29408": ["triton_red_fused_add_addmm_native_layer_norm_view_14:420", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "mul_18581": ["triton_red_fused_add_addmm_native_layer_norm_view_14:420", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "mul_18580": ["triton_red_fused_add_addmm_native_layer_norm_view_14:420", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "sub_8775": ["triton_red_fused_add_addmm_native_layer_norm_view_14:420", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "rsqrt_64": ["triton_red_fused_add_addmm_native_layer_norm_view_14:420", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "add_29407": ["triton_red_fused_add_addmm_native_layer_norm_view_14:420", "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "amax_192": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "view_3008": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "amin_192": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "sub_8813": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "convert_element_type_1154": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "convert_element_type_1153": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "clamp_max_385": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "clamp_min_578": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "add_29476": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "round_386": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "mul_18615": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "mul_18613": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "reciprocal_192": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "view_3011": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "clamp_min_576": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "div_384": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "sub_8787": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "maximum_192": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "full_385": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "minimum_192": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "full_384": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "view_3012": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "convert_element_type_1152": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "clamp_max_384": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "clamp_min_577": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "sub_8793": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "round_385": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "div_385": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "convert_element_type_1155": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "view_3016": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "mul_18634": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "view_3015": ["triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421"], "mul_18639": ["triton_poi_fused__to_copy_mul_sub_view_16:422"], "sub_8817": ["triton_poi_fused__to_copy_mul_sub_view_16:422"], "convert_element_type_1156": ["triton_poi_fused__to_copy_mul_sub_view_16:422"], "view_3018": ["triton_poi_fused__to_copy_mul_sub_view_16:422"], "convert_element_type_1157": ["triton_poi_fused__to_copy_mul_sub_view_16:422"], "view_3020": ["triton_poi_fused__to_copy_mul_sub_view_16:422"], "view_3019": ["triton_poi_fused__to_copy_mul_sub_view_16:422"], "amax_193": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "mul_18644": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "mul_18642": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "add_29516": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "erf_34": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "mul_18643": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "amin_193": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "add_29572": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "round_388": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "mul_18668": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "mul_18666": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "reciprocal_193": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "view_3024": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "clamp_min_579": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "div_386": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "sub_8827": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "maximum_193": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "full_387": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "minimum_193": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "full_386": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "view_3025": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "convert_element_type_1158": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "clamp_max_386": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "clamp_min_580": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "sub_8833": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "round_387": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "div_387": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "mul_18687": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "sub_8853": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "convert_element_type_1160": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "convert_element_type_1159": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "clamp_max_387": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "clamp_min_581": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "convert_element_type_1161": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "view_3029": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "view_3028": ["triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423"], "mul_18692": ["triton_poi_fused__to_copy_mul_sub_view_18:424"], "sub_8857": ["triton_poi_fused__to_copy_mul_sub_view_18:424"], "convert_element_type_1162": ["triton_poi_fused__to_copy_mul_sub_view_18:424"], "view_3031": ["triton_poi_fused__to_copy_mul_sub_view_18:424"], "convert_element_type_1163": ["triton_poi_fused__to_copy_mul_sub_view_18:424"], "view_3033": ["triton_poi_fused__to_copy_mul_sub_view_18:424"], "view_3032": ["triton_poi_fused__to_copy_mul_sub_view_18:424"], "mm_default_159": ["aoti_torch_cuda_mm_out:427"], "mm": ["aoti_torch_cuda_mm_out:428"], "mm_default_158": ["aoti_torch_cuda_mm_out:429"], "mm_default_157": ["aoti_torch_cuda_mm_out:431"], "mm_default_156": ["aoti_torch_cuda_mm_out:432"], "mm_default_155": ["aoti_torch_cuda_mm_out:433"], "mm_default_154": ["aoti_torch_cuda_mm_out:434"], "mm_1": ["aoti_torch_cuda_mm_out:435"], "mm_default_153": ["aoti_torch_cuda_mm_out:436"], "mm_default_152": ["aoti_torch_cuda_mm_out:438"], "mm_default_151": ["aoti_torch_cuda_mm_out:439"], "mm_default_150": ["aoti_torch_cuda_mm_out:440"], "mm_default_149": ["aoti_torch_cuda_mm_out:441"], "mm_2": ["aoti_torch_cuda_mm_out:442"], "mm_default_148": ["aoti_torch_cuda_mm_out:443"], "mm_default_147": ["aoti_torch_cuda_mm_out:445"], "mm_default_146": ["aoti_torch_cuda_mm_out:446"], "mm_default_145": ["aoti_torch_cuda_mm_out:447"], "mm_default_144": ["aoti_torch_cuda_mm_out:448"], "mm_3": ["aoti_torch_cuda_mm_out:449"], "mm_default_143": ["aoti_torch_cuda_mm_out:450"], "mm_default_142": ["aoti_torch_cuda_mm_out:452"], "mm_default_141": ["aoti_torch_cuda_mm_out:453"], "mm_default_140": ["aoti_torch_cuda_mm_out:454"], "mm_default_139": ["aoti_torch_cuda_mm_out:455"], "mm_4": ["aoti_torch_cuda_mm_out:456"], "mm_default_138": ["aoti_torch_cuda_mm_out:457"], "mm_default_137": ["aoti_torch_cuda_mm_out:459"], "mm_default_136": ["aoti_torch_cuda_mm_out:460"], "mm_default_135": ["aoti_torch_cuda_mm_out:461"], "mm_default_134": ["aoti_torch_cuda_mm_out:462"], "mm_5": ["aoti_torch_cuda_mm_out:463"], "mm_default_133": ["aoti_torch_cuda_mm_out:464"], "mm_default_132": ["aoti_torch_cuda_mm_out:466"], "mm_default_131": ["aoti_torch_cuda_mm_out:467"], "mm_default_130": ["aoti_torch_cuda_mm_out:468"], "mm_default_129": ["aoti_torch_cuda_mm_out:469"], "mm_6": ["aoti_torch_cuda_mm_out:470"], "mm_default_128": ["aoti_torch_cuda_mm_out:471"], "mm_default_127": ["aoti_torch_cuda_mm_out:473"], "mm_default_126": ["aoti_torch_cuda_mm_out:474"], "mm_default_125": ["aoti_torch_cuda_mm_out:475"], "mm_default_124": ["aoti_torch_cuda_mm_out:476"], "mm_7": ["aoti_torch_cuda_mm_out:477"], "mm_default_123": ["aoti_torch_cuda_mm_out:478"], "mm_default_122": ["aoti_torch_cuda_mm_out:480"], "mm_default_121": ["aoti_torch_cuda_mm_out:481"], "mm_default_120": ["aoti_torch_cuda_mm_out:482"], "mm_default_119": ["aoti_torch_cuda_mm_out:483"], "mm_8": ["aoti_torch_cuda_mm_out:484"], "mm_default_118": ["aoti_torch_cuda_mm_out:485"], "mm_default_117": ["aoti_torch_cuda_mm_out:487"], "mm_default_116": ["aoti_torch_cuda_mm_out:488"], "mm_default_115": ["aoti_torch_cuda_mm_out:489"], "mm_default_114": ["aoti_torch_cuda_mm_out:490"], "mm_9": ["aoti_torch_cuda_mm_out:491"], "mm_default_113": ["aoti_torch_cuda_mm_out:492"], "mm_default_112": ["aoti_torch_cuda_mm_out:494"], "mm_default_111": ["aoti_torch_cuda_mm_out:495"], "mm_default_110": ["aoti_torch_cuda_mm_out:496"], "mm_default_109": ["aoti_torch_cuda_mm_out:497"], "mm_10": ["aoti_torch_cuda_mm_out:498"], "mm_default_108": ["aoti_torch_cuda_mm_out:499"], "mm_default_107": ["aoti_torch_cuda_mm_out:501"], "mm_default_106": ["aoti_torch_cuda_mm_out:502"], "mm_default_105": ["aoti_torch_cuda_mm_out:503"], "mm_default_104": ["aoti_torch_cuda_mm_out:504"], "mm_11": ["aoti_torch_cuda_mm_out:505"], "mm_default_103": ["aoti_torch_cuda_mm_out:506"], "mm_default_102": ["aoti_torch_cuda_mm_out:508"], "mm_default_101": ["aoti_torch_cuda_mm_out:509"], "mm_default_100": ["aoti_torch_cuda_mm_out:510"], "mm_default_99": ["aoti_torch_cuda_mm_out:511"], "mm_12": ["aoti_torch_cuda_mm_out:512"], "mm_default_98": ["aoti_torch_cuda_mm_out:513"], "mm_default_97": ["aoti_torch_cuda_mm_out:515"], "mm_default_96": ["aoti_torch_cuda_mm_out:516"], "mm_default_95": ["aoti_torch_cuda_mm_out:517"], "mm_default_94": ["aoti_torch_cuda_mm_out:518"], "mm_13": ["aoti_torch_cuda_mm_out:519"], "mm_default_93": ["aoti_torch_cuda_mm_out:520"], "mm_default_92": ["aoti_torch_cuda_mm_out:522"], "mm_default_91": ["aoti_torch_cuda_mm_out:523"], "mm_default_90": ["aoti_torch_cuda_mm_out:524"], "mm_default_89": ["aoti_torch_cuda_mm_out:525"], "mm_14": ["aoti_torch_cuda_mm_out:526"], "mm_default_88": ["aoti_torch_cuda_mm_out:527"], "mm_default_87": ["aoti_torch_cuda_mm_out:529"], "mm_default_86": ["aoti_torch_cuda_mm_out:530"], "mm_default_85": ["aoti_torch_cuda_mm_out:531"], "mm_default_84": ["aoti_torch_cuda_mm_out:532"], "mm_15": ["aoti_torch_cuda_mm_out:533"], "mm_default_83": ["aoti_torch_cuda_mm_out:534"], "mm_default_82": ["aoti_torch_cuda_mm_out:536"], "mm_default_81": ["aoti_torch_cuda_mm_out:537"], "mm_default_80": ["aoti_torch_cuda_mm_out:538"], "mm_default_79": ["aoti_torch_cuda_mm_out:539"], "mm_16": ["aoti_torch_cuda_mm_out:540"], "mm_default_78": ["aoti_torch_cuda_mm_out:541"], "mm_default_77": ["aoti_torch_cuda_mm_out:543"], "mm_default_76": ["aoti_torch_cuda_mm_out:544"], "mm_default_75": ["aoti_torch_cuda_mm_out:545"], "mm_default_74": ["aoti_torch_cuda_mm_out:546"], "mm_17": ["aoti_torch_cuda_mm_out:547"], "mm_default_73": ["aoti_torch_cuda_mm_out:548"], "mm_default_72": ["aoti_torch_cuda_mm_out:550"], "mm_default_71": ["aoti_torch_cuda_mm_out:551"], "mm_default_70": ["aoti_torch_cuda_mm_out:552"], "mm_default_69": ["aoti_torch_cuda_mm_out:553"], "mm_18": ["aoti_torch_cuda_mm_out:554"], "mm_default_68": ["aoti_torch_cuda_mm_out:555"], "mm_default_67": ["aoti_torch_cuda_mm_out:557"], "mm_default_66": ["aoti_torch_cuda_mm_out:558"], "mm_default_65": ["aoti_torch_cuda_mm_out:559"], "mm_default_64": ["aoti_torch_cuda_mm_out:560"], "mm_19": ["aoti_torch_cuda_mm_out:561"], "mm_default_63": ["aoti_torch_cuda_mm_out:562"], "mm_default_62": ["aoti_torch_cuda_mm_out:564"], "mm_default_61": ["aoti_torch_cuda_mm_out:565"], "mm_default_60": ["aoti_torch_cuda_mm_out:566"], "mm_default_59": ["aoti_torch_cuda_mm_out:567"], "mm_20": ["aoti_torch_cuda_mm_out:568"], "mm_default_58": ["aoti_torch_cuda_mm_out:569"], "mm_default_57": ["aoti_torch_cuda_mm_out:571"], "mm_default_56": ["aoti_torch_cuda_mm_out:572"], "mm_default_55": ["aoti_torch_cuda_mm_out:573"], "mm_default_54": ["aoti_torch_cuda_mm_out:574"], "mm_21": ["aoti_torch_cuda_mm_out:575"], "mm_default_53": ["aoti_torch_cuda_mm_out:576"], "mm_default_52": ["aoti_torch_cuda_mm_out:578"], "mm_default_51": ["aoti_torch_cuda_mm_out:579"], "mm_default_50": ["aoti_torch_cuda_mm_out:580"], "mm_default_49": ["aoti_torch_cuda_mm_out:581"], "mm_22": ["aoti_torch_cuda_mm_out:582"], "mm_default_48": ["aoti_torch_cuda_mm_out:583"], "mm_default_47": ["aoti_torch_cuda_mm_out:585"], "mm_default_46": ["aoti_torch_cuda_mm_out:586"], "mm_default_45": ["aoti_torch_cuda_mm_out:587"], "mm_default_44": ["aoti_torch_cuda_mm_out:588"], "mm_23": ["aoti_torch_cuda_mm_out:589"], "mm_default_43": ["aoti_torch_cuda_mm_out:590"], "mm_default_42": ["aoti_torch_cuda_mm_out:592"], "mm_default_41": ["aoti_torch_cuda_mm_out:593"], "mm_default_40": ["aoti_torch_cuda_mm_out:594"], "mm_default_39": ["aoti_torch_cuda_mm_out:595"], "mm_24": ["aoti_torch_cuda_mm_out:596"], "mm_default_38": ["aoti_torch_cuda_mm_out:597"], "mm_default_37": ["aoti_torch_cuda_mm_out:599"], "mm_default_36": ["aoti_torch_cuda_mm_out:600"], "mm_default_35": ["aoti_torch_cuda_mm_out:601"], "mm_default_34": ["aoti_torch_cuda_mm_out:602"], "mm_25": ["aoti_torch_cuda_mm_out:603"], "mm_default_33": ["aoti_torch_cuda_mm_out:604"], "mm_default_32": ["aoti_torch_cuda_mm_out:606"], "mm_default_31": ["aoti_torch_cuda_mm_out:607"], "mm_default_30": ["aoti_torch_cuda_mm_out:608"], "mm_default_29": ["aoti_torch_cuda_mm_out:609"], "mm_26": ["aoti_torch_cuda_mm_out:610"], "mm_default_28": ["aoti_torch_cuda_mm_out:611"], "mm_default_27": ["aoti_torch_cuda_mm_out:613"], "mm_default_26": ["aoti_torch_cuda_mm_out:614"], "mm_default_25": ["aoti_torch_cuda_mm_out:615"], "mm_default_24": ["aoti_torch_cuda_mm_out:616"], "mm_27": ["aoti_torch_cuda_mm_out:617"], "mm_default_23": ["aoti_torch_cuda_mm_out:618"], "mm_default_22": ["aoti_torch_cuda_mm_out:620"], "mm_default_21": ["aoti_torch_cuda_mm_out:621"], "mm_default_20": ["aoti_torch_cuda_mm_out:622"], "mm_default_19": ["aoti_torch_cuda_mm_out:623"], "mm_28": ["aoti_torch_cuda_mm_out:624"], "mm_default_18": ["aoti_torch_cuda_mm_out:625"], "mm_default_17": ["aoti_torch_cuda_mm_out:627"], "mm_default_16": ["aoti_torch_cuda_mm_out:628"], "mm_default_15": ["aoti_torch_cuda_mm_out:629"], "mm_default_14": ["aoti_torch_cuda_mm_out:630"], "mm_29": ["aoti_torch_cuda_mm_out:631"], "mm_default_13": ["aoti_torch_cuda_mm_out:632"], "mm_default_12": ["aoti_torch_cuda_mm_out:634"], "mm_default_11": ["aoti_torch_cuda_mm_out:635"], "mm_default_10": ["aoti_torch_cuda_mm_out:636"], "mm_default_9": ["aoti_torch_cuda_mm_out:637"], "mm_30": ["aoti_torch_cuda_mm_out:638"], "mm_default_8": ["aoti_torch_cuda_mm_out:639"], "mm_default_7": ["aoti_torch_cuda_mm_out:641"], "mm_default_6": ["aoti_torch_cuda_mm_out:642"], "mm_default_5": ["aoti_torch_cuda_mm_out:643"], "mm_default_4": ["aoti_torch_cuda_mm_out:644"], "mm_31": ["aoti_torch_cuda_mm_out:645"], "mm_default_3": ["aoti_torch_cuda_mm_out:646"], "mm_default_2": ["aoti_torch_cuda_mm_out:648"], "mm_default_1": ["aoti_torch_cuda_mm_out:649"], "mm_default": ["aoti_torch_cuda_mm_out:650"], "mm_32": ["aoti_torch_cuda_mm_out:651"], "mm_33": ["aoti_torch_cuda_mm_out:652"]}, "version": 2.0}
V0910 09:44:16.904000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_inductor/compile_fx.py:1528] {"artifact": {"name": "inductor_provenance_tracking_kernel_stack_traces", "encoding": "json"}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2657, "name": "compile_fx", "filename": 19, "loc": "return inference_compiler(unlifted_gm, example_inputs_)"}, {"line": 1267, "name": "__call__", "filename": 33, "loc": "return self.compiler_fn(gm, example_inputs)"}, {"line": 2543, "name": "fw_compiler_base", "filename": 19, "loc": "return compile_fx_forward("}, {"line": 2260, "name": "compile_fx_forward", "filename": 19, "loc": "return inner_compile("}, {"line": 81, "name": "inner", "filename": 34, "loc": "return func(*args, **kwds)"}, {"line": 781, "name": "compile_fx_inner", "filename": 19, "loc": "return wrap_compiler_debug(_compile_fx_inner, compiler_name=\"inductor\")("}, {"line": 144, "name": "debug_wrapper", "filename": 35, "loc": "inner_compiled_fn = compiler_fn(gm, example_inputs)"}, {"line": 167, "name": "newFunction", "filename": 36, "loc": "return old_func(*args, **kwargs)"}, {"line": 962, "name": "_compile_fx_inner", "filename": 19, "loc": "mb_compiled_graph = fx_codegen_and_compile("}, {"line": 1694, "name": "fx_codegen_and_compile", "filename": 19, "loc": "return scheme.codegen_and_compile(gm, example_inputs, inputs_to_check, graph_kwargs)"}, {"line": 1528, "name": "codegen_and_compile", "filename": 19, "loc": "trace_structured("}], "has_payload": "91d7774f0fe8b8680e2a371198dd55af"}
	{"triton_poi_fused_convolution_gelu_0:1": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 350, in forward\n    inputs_embeds = nn.functional.gelu(self.conv2(inputs_embeds))\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 349, in forward\n    inputs_embeds = nn.functional.gelu(self.conv1(input_features))\n"], "triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1:2": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 354, in forward\n    hidden_states = (inputs_embeds + embed_pos).to(inputs_embeds.dtype)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 351, in forward\n    inputs_embeds = inputs_embeds.permute(0, 2, 1)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 350, in forward\n    inputs_embeds = nn.functional.gelu(self.conv2(inputs_embeds))\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 349, in forward\n    inputs_embeds = nn.functional.gelu(self.conv1(input_features))\n"], "triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2:3": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 354, in forward\n    hidden_states = (inputs_embeds + embed_pos).to(inputs_embeds.dtype)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 351, in forward\n    inputs_embeds = inputs_embeds.permute(0, 2, 1)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 350, in forward\n    inputs_embeds = nn.functional.gelu(self.conv2(inputs_embeds))\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 349, in forward\n    inputs_embeds = nn.functional.gelu(self.conv1(input_features))\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3:4": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 354, in forward\n    hidden_states = (inputs_embeds + embed_pos).to(inputs_embeds.dtype)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 351, in forward\n    inputs_embeds = inputs_embeds.permute(0, 2, 1)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 350, in forward\n    inputs_embeds = nn.functional.gelu(self.conv2(inputs_embeds))\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 349, in forward\n    inputs_embeds = nn.functional.gelu(self.conv1(input_features))\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:5": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:6": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:7": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:8": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:9": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:10": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:11": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:12": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9:13": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 354, in forward\n    hidden_states = (inputs_embeds + embed_pos).to(inputs_embeds.dtype)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 351, in forward\n    inputs_embeds = inputs_embeds.permute(0, 2, 1)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 350, in forward\n    inputs_embeds = nn.functional.gelu(self.conv2(inputs_embeds))\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 349, in forward\n    inputs_embeds = nn.functional.gelu(self.conv1(input_features))\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:14": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:15": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:16": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:17": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:18": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:19": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:20": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:21": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:22": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:23": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:24": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:25": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:26": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:27": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:28": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:29": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:30": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:31": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:32": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:33": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:34": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:35": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:36": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:37": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:38": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:39": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:40": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:41": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:42": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:43": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:44": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:45": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:46": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:47": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:48": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:49": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:50": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:51": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:52": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:53": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:54": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:55": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:56": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:57": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:58": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:59": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:60": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:61": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:62": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:63": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:64": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:65": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:66": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:67": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:68": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:69": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:70": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:71": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:72": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:73": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:74": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:75": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:76": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:77": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:78": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:79": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:80": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:81": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:82": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:83": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:84": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:85": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:86": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:87": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:88": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:89": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:90": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:91": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:92": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:93": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:94": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:95": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:96": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:97": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:98": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:99": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:100": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:101": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:102": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:103": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:104": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:105": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:106": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:107": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:108": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:109": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:110": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:111": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:112": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:113": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:114": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:115": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:116": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:117": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:118": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:119": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:120": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:121": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:122": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:123": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:124": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:125": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:126": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:127": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:128": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:129": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:130": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:131": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:132": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:133": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:134": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:135": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:136": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:137": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:138": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:139": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:140": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:141": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:142": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:143": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:144": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:145": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:146": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:147": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:148": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:149": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:150": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:151": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:152": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:153": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:154": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:155": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:156": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:157": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:158": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:159": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:160": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:161": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:162": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:163": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:164": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:165": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:166": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:167": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:168": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:169": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:170": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:171": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:172": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:173": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:174": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:175": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:176": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:177": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:178": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:179": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:180": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:181": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:182": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:183": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:184": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:185": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:186": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:187": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:188": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:189": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:190": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:191": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:192": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:193": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:194": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:195": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:196": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:197": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:198": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:199": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:200": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:201": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:202": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:203": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:204": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:205": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:206": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:207": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:208": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:209": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:210": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:211": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:212": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:213": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:214": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:215": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:216": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:217": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:218": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:219": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:220": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:221": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:222": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:223": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:224": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:225": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:226": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:227": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:228": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:229": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:230": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:231": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:232": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:233": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:234": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:235": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:236": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:237": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:238": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:239": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:240": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:241": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:242": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:243": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:244": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:245": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:246": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:247": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:248": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:249": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:250": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:251": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:252": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:253": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:254": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:255": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:256": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:257": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:258": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:259": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:260": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:261": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:262": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:263": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:264": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:265": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:266": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:267": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:268": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:269": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:270": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:271": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:272": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:273": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:274": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:275": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:276": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:277": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:278": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:279": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:280": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:281": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:282": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:283": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:284": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:285": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:286": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:287": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:288": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:289": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:290": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:291": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:292": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:293": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:294": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:295": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:296": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:297": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:298": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:299": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:300": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:301": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:302": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:303": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:304": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:305": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:306": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:307": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:308": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:309": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:310": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:311": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:312": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:313": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:314": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:315": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:316": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:317": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:318": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:319": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:320": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:321": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:322": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:323": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:324": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:325": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:326": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:327": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:328": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:329": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:330": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:331": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:332": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:333": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:334": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:335": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:336": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:337": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:338": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:339": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:340": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:341": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:342": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:343": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:344": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:345": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:346": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:347": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:348": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:349": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:350": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:351": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:352": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:353": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:354": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:355": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:356": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:357": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:358": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:359": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:360": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:361": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:362": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:363": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:364": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:365": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:366": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:367": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:368": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:369": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:370": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:371": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:372": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:373": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:374": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:375": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:376": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:377": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:378": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:379": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:380": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:381": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:382": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:383": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:384": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:385": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:386": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:387": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:388": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:389": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:390": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:391": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:392": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:393": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:394": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:395": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:396": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:397": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:398": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:399": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:400": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:401": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:402": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:403": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:404": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:405": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:406": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12:407": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 202, in forward\n    hidden_states = self.self_attn_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:408": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:409": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__to_copy_mul_sub_view_4:410": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5:411": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6:412": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7:413": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8:414": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 158, in forward\n    attn_output = attn_output.reshape(bsz, tgt_len, -1).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 99, in sdpa_attention_forward\n    attn_output = attn_output.transpose(1, 2).contiguous()\n"], "triton_poi_fused__to_copy_mul_sub_view_4:415": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13:416": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 210, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 213, in forward\n    hidden_states = self.final_layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:417": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11:418": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "triton_poi_fused__to_copy_mul_sub_view_10:419": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused_add_addmm_native_layer_norm_view_14:420": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 365, in forward\n    hidden_states = self.layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15:421": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 452, in get_audio_embeds\n    audio_embeds = self.multi_modal_projector(audio_hidden_states)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 389, in forward\n    hidden_states = self.linear_1(audio_features)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 451, in get_audio_embeds\n    audio_hidden_states = audio_hidden_states.reshape(-1, self.config.audio_config.intermediate_size)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 365, in forward\n    hidden_states = self.layer_norm(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 218, in forward\n    hidden_states = residual + hidden_states\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "triton_poi_fused__to_copy_mul_sub_view_16:422": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 452, in get_audio_embeds\n    audio_embeds = self.multi_modal_projector(audio_hidden_states)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 389, in forward\n    hidden_states = self.linear_1(audio_features)\n"], "triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17:423": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 452, in get_audio_embeds\n    audio_embeds = self.multi_modal_projector(audio_hidden_states)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 391, in forward\n    hidden_states = self.linear_2(hidden_states)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 452, in get_audio_embeds\n    audio_embeds = self.multi_modal_projector(audio_hidden_states)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 390, in forward\n    hidden_states = self.act(hidden_states)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/activations.py\", line 69, in forward\n    return self.act(input)\n"], "triton_poi_fused__to_copy_mul_sub_view_18:424": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 452, in get_audio_embeds\n    audio_embeds = self.multi_modal_projector(audio_hidden_states)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 391, in forward\n    hidden_states = self.linear_2(hidden_states)\n"], "aoti_torch_cuda_convolution:425": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 349, in forward\n    inputs_embeds = nn.functional.gelu(self.conv1(input_features))\n"], "aoti_torch_cuda_convolution:426": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 350, in forward\n    inputs_embeds = nn.functional.gelu(self.conv2(inputs_embeds))\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 349, in forward\n    inputs_embeds = nn.functional.gelu(self.conv1(input_features))\n"], "aoti_torch_cuda_mm_out:427": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:428": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:429": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:430": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:431": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:432": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:433": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:434": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:435": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:436": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:437": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:438": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:439": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:440": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:441": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:442": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:443": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:444": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:445": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:446": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:447": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:448": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:449": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:450": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:451": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:452": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:453": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:454": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:455": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:456": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:457": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:458": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:459": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:460": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:461": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:462": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:463": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:464": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:465": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:466": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:467": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:468": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:469": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:470": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:471": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:472": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:473": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:474": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:475": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:476": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:477": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:478": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:479": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:480": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:481": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:482": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:483": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:484": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:485": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:486": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:487": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:488": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:489": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:490": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:491": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:492": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:493": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:494": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:495": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:496": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:497": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:498": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:499": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:500": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:501": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:502": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:503": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:504": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:505": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:506": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:507": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:508": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:509": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:510": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:511": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:512": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:513": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:514": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:515": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:516": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:517": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:518": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:519": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:520": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:521": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:522": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:523": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:524": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:525": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:526": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:527": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:528": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:529": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:530": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:531": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:532": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:533": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:534": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:535": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:536": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:537": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:538": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:539": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:540": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:541": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:542": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:543": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:544": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:545": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:546": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:547": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:548": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:549": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:550": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:551": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:552": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:553": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:554": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:555": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:556": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:557": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:558": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:559": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:560": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:561": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:562": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:563": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:564": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:565": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:566": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:567": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:568": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:569": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:570": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:571": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:572": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:573": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:574": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:575": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:576": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:577": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:578": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:579": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:580": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:581": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:582": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:583": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:584": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:585": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:586": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:587": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:588": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:589": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:590": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:591": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:592": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:593": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:594": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:595": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:596": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:597": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:598": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:599": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:600": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:601": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:602": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:603": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:604": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:605": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:606": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:607": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:608": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:609": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:610": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:611": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:612": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:613": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:614": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:615": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:616": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:617": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:618": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:619": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:620": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:621": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:622": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:623": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:624": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:625": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:626": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:627": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:628": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:629": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:630": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:631": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:632": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:633": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:634": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:635": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:636": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:637": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:638": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:639": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:640": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:641": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:642": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:643": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:644": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n"], "aoti_torch_cuda_mm_out:645": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:646": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "at::_ops::_scaled_dot_product_efficient_attention::call:647": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 145, in forward\n    attn_output, attn_weights = attention_interface(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/integrations/sdpa_attention.py\", line 89, in sdpa_attention_forward\n    attn_output = torch.nn.functional.scaled_dot_product_attention(\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 137, in forward\n    query_states = self._shape(self.q_proj(hidden_states) * self.scaling, tgt_len, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 138, in forward\n    key_states = self._shape(self.k_proj(hidden_states), -1, bsz)\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 118, in _shape\n    return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()\n", "  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 139, in forward\n    value_states = self._shape(self.v_proj(hidden_states), -1, bsz)\n"], "aoti_torch_cuda_mm_out:648": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 203, in forward\n    hidden_states, attn_weights = self.self_attn(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 159, in forward\n    attn_output = self.out_proj(attn_output)\n"], "aoti_torch_cuda_mm_out:649": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 214, in forward\n    hidden_states = self.activation_fn(self.fc1(hidden_states))\n"], "aoti_torch_cuda_mm_out:650": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 449, in get_audio_embeds\n    audio_outputs = self.audio_tower(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/utils/generic.py\", line 1083, in wrapper\n    outputs = func(self, *args, **kwargs)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 358, in forward\n    layer_outputs = encoder_layer(\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 216, in forward\n    hidden_states = self.fc2(hidden_states)\n"], "aoti_torch_cuda_mm_out:651": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 452, in get_audio_embeds\n    audio_embeds = self.multi_modal_projector(audio_hidden_states)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 389, in forward\n    hidden_states = self.linear_1(audio_features)\n"], "aoti_torch_cuda_mm_out:652": ["  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/optimum/exporters/executorch/integrations.py\", line 82, in forward\n    audio_embeds = self.model.get_audio_embeds(input_features)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 452, in get_audio_embeds\n    audio_embeds = self.multi_modal_projector(audio_hidden_states)\n  File \"/home/yimingzhou/.conda/envs/executorch/lib/python3.12/site-packages/transformers/models/voxtral/modeling_voxtral.py\", line 391, in forward\n    hidden_states = self.linear_2(hidden_states)\n"]}
V0910 09:44:16.916000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_dynamo/utils.py", 46]}
V0910 09:44:16.917000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1991] {"chromium_event": {}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1940, "name": "compile_fx_aot", "filename": 19, "loc": "compiled_artifacts = compile_fx("}, {"line": 2398, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2459, "name": "compile_fx", "filename": 19, "loc": "return compile_fx("}, {"line": 2657, "name": "compile_fx", "filename": 19, "loc": "return inference_compiler(unlifted_gm, example_inputs_)"}, {"line": 1267, "name": "__call__", "filename": 33, "loc": "return self.compiler_fn(gm, example_inputs)"}, {"line": 2543, "name": "fw_compiler_base", "filename": 19, "loc": "return compile_fx_forward("}, {"line": 2260, "name": "compile_fx_forward", "filename": 19, "loc": "return inner_compile("}, {"line": 81, "name": "inner", "filename": 34, "loc": "return func(*args, **kwds)"}, {"line": 781, "name": "compile_fx_inner", "filename": 19, "loc": "return wrap_compiler_debug(_compile_fx_inner, compiler_name=\"inductor\")("}, {"line": 144, "name": "debug_wrapper", "filename": 35, "loc": "inner_compiled_fn = compiler_fn(gm, example_inputs)"}, {"line": 167, "name": "newFunction", "filename": 36, "loc": "return old_func(*args, **kwargs)"}, {"line": 1032, "name": "_compile_fx_inner", "filename": 19, "loc": "CompileEventLogger.instant("}, {"line": 616, "name": "instant", "filename": 46, "loc": "CompileEventLogger.log_instant_event("}, {"line": 410, "name": "log_instant_event", "filename": 46, "loc": "chromium_log.log_instant_event("}, {"line": 1991, "name": "log_instant_event", "filename": 46, "loc": "torch._logging.trace_structured("}], "has_payload": "ecec03fd92f980eeb2e25321d76d997d"}
	{
	"name": "fx_graph_cache_disabled",
	"ts": 1757522566216830.5,
	"args": {
	"compile_id": "None"
	},
	"ph": "i",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0,
	"s": "p"
	}
V0910 09:44:16.918000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "6418b7648f98507d8ff56326b3c24464"}
	{
	"name": "fx_codegen_and_compile",
	"ts": 1757522656918194.0,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:44:16.922000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "1ed03e590b2fb92dada7a59252e4f107"}
	{
	"name": "inductor_compile",
	"ts": 1757522656922608.0,
	"args": {
	"fn_name": "compile_fx_inner",
	"compile_id": "None",
	"is_backward": false,
	"cache_state": "disabled",
	"cache_event_time": 1757522566216830576,
	"key": null,
	"components": null,
	"cache_bypass_reason": "cache not enabled",
	"remote_cache_enabled": true,
	"local_cache_enabled": true
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:44:16.926000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "0acf564bc4bbad5ae3b1d673693f7916"}
	{
	"name": "compile_fx.<locals>.fw_compiler_base",
	"ts": 1757522656926562.8,
	"args": {
	"compile_id": "None"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
V0910 09:44:16.934000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_logging/structured.py:28] {"str": ["/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/torch/_dynamo/metrics_context.py", 47]}
V0910 09:44:16.935000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1647] {"compilation_metrics": {"compile_id": null, "frame_key": null, "co_name": null, "co_filename": null, "co_firstlineno": null, "cache_size": null, "accumulated_cache_size": null, "guard_count": null, "shape_env_guard_count": null, "graph_op_count": null, "graph_node_count": null, "graph_input_count": null, "start_time": 1757522508.242218, "entire_frame_compile_time_s": null, "backend_compile_time_s": null, "inductor_compile_time_s": null, "code_gen_time_s": null, "fail_type": null, "fail_reason": null, "fail_user_frame_filename": null, "fail_user_frame_lineno": null, "non_compliant_ops": null, "compliant_custom_ops": null, "restart_reasons": null, "dynamo_time_before_restart_s": null, "stack_trace": null, "exception_stack_trace": null, "graph_node_shapes": null, "has_guarded_code": null, "remote_cache_time_saved_s": null, "structured_logging_overhead_s": null, "config_suppress_errors": false, "config_inline_inbuilt_nn_modules": true, "specialize_float": null, "dynamo_config": "{\"_autograd_backward_strict_mode_conditional_banned_ops\": [\"stride\", \"storage_offset\", \"is_contiguous\"], \"_unsafe_skip_fsdp_module_guards\": false, \"accumulated_recompile_limit\": 256, \"allow_empty_graphs\": false, \"allow_ignore_mark_dynamic\": false, \"allow_rnn\": false, \"allow_unspec_int_on_nn_module\": false, \"allowed_functions_module_string_ignorelist\": [\"torch._decomp\", \"torch._prims\", \"torch._refs\", \"torch.distributions\", \"torch.testing\"], \"assume_dunder_attributes_remain_unchanged\": true, \"assume_static_by_default\": true, \"automatic_dynamic_local_pgo\": true, \"automatic_dynamic_remote_pgo\": null, \"automatic_dynamic_shapes\": true, \"automatic_dynamic_shapes_mark_as\": \"dynamic\", \"caching_precompile\": false, \"capture_autograd_function\": true, \"capture_dynamic_output_shape_ops\": false, \"capture_func_transforms\": true, \"capture_scalar_outputs\": false, \"capture_sparse_compute\": false, \"compiled_autograd\": false, \"compiled_autograd_kwargs_override\": {}, \"cprofile\": false, \"cudagraph_backend_keep_input_mutation\": false, \"cudagraph_backend_support_input_mutation\": false, \"dead_code_elimination\": true, \"debug_disable_compile_counter\": false, \"debug_force_graph_break_on_leaf_return\": false, \"debug_force_nested_calls\": false, \"disable\": false, \"do_not_emit_runtime_asserts\": false, \"dont_skip_tracing\": false, \"dynamic_shapes\": true, \"enable_aot_compile\": false, \"enable_compiler_collectives\": false, \"enable_cpp_framelocals_guard_eval\": true, \"enable_cpp_guard_manager\": true, \"enable_cpp_symbolic_shape_guards\": false, \"enable_faithful_generator_behavior\": true, \"enable_trace_contextlib\": true, \"enable_trace_unittest\": false, \"error_on_nested_fx_trace\": true, \"error_on_nested_jit_trace\": true, \"error_on_recompile\": false, \"fail_on_recompile_limit_hit\": false, \"fake_tensor_cache_crosscheck_enabled\": false, \"fake_tensor_cache_enabled\": true, \"fake_tensor_disable_inference_mode\": true, \"force_nn_module_property_static_shapes\": true, \"force_parameter_static_shapes\": true, \"force_unspec_int_unbacked_size_like_on_torchrec_kjt\": false, \"graph_break_on_nn_param_ctor\": true, \"graph_deduplication_lint\": false, \"guard_nn_modules\": true, \"guard_nn_modules_using_dict_tags\": true, \"inline_inbuilt_nn_modules\": true, \"install_free_tensors\": false, \"issue_3_13_0_warning\": true, \"max_saved_pointers_for_recursive_dict_tags_check\": 256, \"minimum_call_count\": 1, \"nested_graph_breaks\": false, \"numpy_default_complex\": \"complex128\", \"numpy_default_float\": \"float64\", \"numpy_default_int\": \"int64\", \"only_allow_pt2_compliant_ops\": false, \"optimize_ddp\": true, \"optimize_ddp_lazy_compile\": false, \"prefer_deferred_runtime_asserts_over_guards\": false, \"prepare_freezing\": false, \"pt2_compile_id_prefix\": null, \"raise_on_ctx_manager_usage\": true, \"raise_on_unsafe_aot_autograd\": false, \"recompile_limit\": 8, \"record_compile_time_instruction_count\": false, \"record_runtime_overhead\": true, \"replay_record_enabled\": false, \"report_guard_failures\": true, \"rewrite_assert_with_torch_assert\": true, \"run_gc_after_compile\": true, \"skip_code_recursive_on_recompile_limit_hit\": true, \"skip_fsdp_guards\": true, \"skip_fsdp_hooks\": true, \"skip_guards_on_constant_func_defaults\": true, \"skip_nnmodule_hook_guards\": true, \"skip_no_tensor_aliasing_guards_on_parameters\": true, \"skip_tensor_guards_with_matching_dict_tags\": true, \"skip_torchrec\": true, \"skipfiles_inline_module_allowlist\": {}, \"specialize_float\": false, \"specialize_int\": false, \"strict_precompile\": false, \"suppress_errors\": false, \"trace_numpy\": true, \"track_nodes_for_deduplication\": false, \"use_graph_deduplication\": false, \"use_lamba_guard_for_object_aliasing\": true, \"use_lazy_graph_module\": true, \"use_numpy_random_stream\": false, \"use_recursive_dict_tags_for_guards\": true, \"verify_correctness\": false, \"wrap_top_frame\": false}", "is_forward": null, "num_triton_bundles": null, "remote_fx_graph_cache_get_time_ms": null, "remote_fx_graph_cache_put_time_ms": null, "start_time_us": 1757522508242218, "duration_us": null, "dynamo_cumulative_compile_time_us": null, "aot_autograd_cumulative_compile_time_us": null, "inductor_cumulative_compile_time_us": null, "inductor_code_gen_cumulative_compile_time_us": null, "triton_compile_time_us": null, "runtime_cudagraphify_time_us": null, "runtime_triton_autotune_time_us": null, "dynamo_compile_time_before_restart_us": null, "distributed_ephemeral_timeout_us": null, "structured_logging_overhead_us": null, "remote_fx_graph_cache_get_time_us": null, "remote_fx_graph_cache_put_time_us": null, "backward_cumulative_compile_time_us": null, "end_time_us": 1757522656931580, "pre_grad_pass_time_us": null, "post_grad_pass_time_us": null, "joint_graph_pass_time_us": null, "log_format_version": 3, "inductor_config": "{\"TYPE_CHECKING\": false, \"_cache_config_ignore_prefix\": [\"trace\", \"cuda.cutlass_dir\", \"worker_start_method\", \"compile_threads\", \"post_grad_custom_post_pass\", \"post_grad_custom_pre_pass\", \"joint_custom_pre_pass\", \"joint_custom_post_pass\", \"_fuse_ddp_communication_passes\", \"_pre_fusion_custom_pass\", \"always_complex_memory_overlap_TESTING_ONLY\", \"fx_graph_cache\", \"fx_graph_remote_cache\", \"autotune_local_cache\", \"autotune_remote_cache\"], \"_collective.auto_select\": false, \"_collective.one_shot_all_reduce_threshold_bytes\": 131072, \"_fuse_ddp_bucket_size\": 25, \"_fuse_ddp_communication\": false, \"_fuse_ddp_communication_passes\": [\"fuse_ddp_with_concat_op\", \"schedule_comm_wait\"], \"_micro_pipeline_tp\": false, \"_post_fusion_custom_pass\": null, \"_pre_fusion_custom_pass\": null, \"_profile_var\": \"\", \"_raise_error_for_testing\": false, \"_save_config_ignore\": [\"trace.upload_tar\", \"joint_custom_pre_pass\", \"joint_custom_post_pass\", \"pre_grad_custom_pass\", \"aot_inductor.repro_level\", \"aot_inductor.dump_aoti_minifier\", \"post_grad_custom_pre_pass\", \"post_grad_custom_post_pass\", \"_fuse_ddp_communication_passes\", \"_pre_fusion_custom_pass\"], \"add_pre_grad_passes\": null, \"aggressive_fusion\": false, \"alignment_asserts\": false, \"allow_buffer_reuse\": true, \"always_complex_memory_overlap_TESTING_ONLY\": false, \"always_keep_tensor_constants\": false, \"annotate_training\": false, \"aot_inductor.allow_stack_allocation\": false, \"aot_inductor.compile_standalone\": false, \"aot_inductor.compile_wrapper_opt_level\": \"O1\", \"aot_inductor.custom_op_libs\": null, \"aot_inductor.custom_ops_to_c_shims\": {}, \"aot_inductor.debug_compile\": false, \"aot_inductor.debug_intermediate_value_printer\": \"0\", \"aot_inductor.dump_aoti_minifier\": false, \"aot_inductor.embed_kernel_binary\": null, \"aot_inductor.emit_multi_arch_kernel\": null, \"aot_inductor.enable_lto\": false, \"aot_inductor.filtered_kernel_names\": null, \"aot_inductor.force_mmap_weights\": false, \"aot_inductor.metadata\": {\"AOTI_DEVICE_KEY\": \"cuda\"}, \"aot_inductor.model_name_for_generated_files\": null, \"aot_inductor.output_path\": \"\", \"aot_inductor.package\": false, \"aot_inductor.package_constants_in_so\": true, \"aot_inductor.package_constants_on_disk\": false, \"aot_inductor.package_cpp_only\": null, \"aot_inductor.precompile_headers\": false, \"aot_inductor.presets\": {}, \"aot_inductor.raise_error_on_ignored_optimization\": true, \"aot_inductor.repro_level\": 2, \"aot_inductor.serialized_in_spec\": \"\", \"aot_inductor.serialized_out_spec\": \"\", \"aot_inductor.use_consts_asm_build\": true, \"aot_inductor.use_minimal_arrayref_interface\": false, \"aot_inductor.use_runtime_constant_folding\": false, \"aot_inductor.weight_use_caching_allocator\": false, \"assert_indirect_indexing\": true, \"assume_aligned_inputs\": false, \"assume_unaligned_fallback_output\": false, \"autoheuristic_collect\": \"\", \"autoheuristic_log_path\": \"DEFAULT\", \"autoheuristic_use\": \"mixed_mm\", \"autotune_fallback_to_aten\": false, \"autotune_in_subproc\": false, \"autotune_local_cache\": true, \"autotune_lookup_table\": {}, \"autotune_multi_device\": false, \"autotune_num_choices_displayed\": 10, \"autotune_remote_cache\": null, \"b2b_gemm_pass\": false, \"batch_fusion\": true, \"benchmark_combo_kernel\": false, \"benchmark_epilogue_fusion\": true, \"benchmark_fusion\": false, \"benchmark_harness\": true, \"benchmark_kernel\": false, \"bfloat16_atomic_adds_enabled\": true, \"bucket_all_gathers_fx\": \"none\", \"bucket_all_gathers_fx_bucket_size_determinator\": null, \"bucket_reduce_scatters_fx\": \"none\", \"bucket_reduce_scatters_fx_bucket_size_determinator\": null, \"bundle_triton_into_fx_graph_cache\": null, \"bundled_autotune_remote_cache\": null, \"bw_outputs_user_visible\": true, \"can_inplace_pad_graph_input\": false, \"check_stack_no_cycles_TESTING_ONLY\": false, \"combo_kernel_allow_mixed_sizes\": 1, \"combo_kernel_foreach_dynamic_shapes\": true, \"combo_kernels\": false, \"combo_kernels_autotune\": 1, \"comment_origin\": false, \"compile_threads\": 32, \"comprehensive_padding\": true, \"compute_all_bounds\": false, \"constant_and_index_propagation\": true, \"conv_1x1_as_mm\": false, \"coordinate_descent_check_all_directions\": false, \"coordinate_descent_search_radius\": 1, \"coordinate_descent_tuning\": false, \"cpp.cxx\": [null, \"g++\"], \"cpp.descriptive_names\": \"original_aten\", \"cpp.dynamic_threads\": false, \"cpp.enable_concat_linear\": false, \"cpp.enable_floating_point_contract_flag\": \"off\", \"cpp.enable_grouped_gemm_template\": false, \"cpp.enable_kernel_profile\": false, \"cpp.enable_loop_tail_vec\": true, \"cpp.enable_tiling_heuristics\": true, \"cpp.enable_unsafe_math_opt_flag\": false, \"cpp.fallback_scatter_reduce_sum\": true, \"cpp.force_inline_kernel\": false, \"cpp.gemm_cache_blocking\": null, \"cpp.gemm_max_k_slices\": 1, \"cpp.gemm_thread_factors\": null, \"cpp.inject_log1p_bug_TESTING_ONLY\": null, \"cpp.inject_relu_bug_TESTING_ONLY\": null, \"cpp.max_horizontal_fusion_size\": 16, \"cpp.min_chunk_size\": 512, \"cpp.no_redundant_loops\": true, \"cpp.simdlen\": null, \"cpp.threads\": -1, \"cpp.use_constexpr_for_int_array\": true, \"cpp.use_decompose_tanh\": false, \"cpp.use_small_dequant_buffer\": false, \"cpp.vec_isa_ok\": null, \"cpp.weight_prepack\": true, \"cpp_cache_precompile_headers\": false, \"cpp_wrapper\": false, \"cpp_wrapper_build_separate\": false, \"cpu_backend\": \"cpp\", \"cuda.arch\": null, \"cuda.binary_remote_cache_force_write\": false, \"cuda.compile_opt_level\": \"-O1\", \"cuda.cuda_cxx\": null, \"cuda.cutlass_backend_min_gemm_size\": 1, \"cuda.cutlass_dir\": \"/data/users/shangdiy/fbsource/buck-out/v2/gen/fbcode/f6cdda1a5c5f7fc9/scripts/yimingzhou/__voxtral_lowering__/voxtral_lowering#link-tree/third_party/cutlass\", \"cuda.cutlass_enabled_ops\": \"all\", \"cuda.cutlass_epilogue_fusion_enabled\": false, \"cuda.cutlass_hash_with_compile_cmd\": false, \"cuda.cutlass_instantiation_level\": \"0\", \"cuda.cutlass_max_profiling_configs\": null, \"cuda.cutlass_max_profiling_swizzle_options\": [1, 2, 4, 8], \"cuda.cutlass_op_allowlist_regex\": null, \"cuda.cutlass_op_denylist_regex\": null, \"cuda.cutlass_prescreening\": true, \"cuda.cutlass_presets\": null, \"cuda.cutlass_tma_only\": false, \"cuda.enable_caching_codegen\": true, \"cuda.enable_cuda_lto\": false, \"cuda.enable_debug_info\": false, \"cuda.enable_ptxas_info\": false, \"cuda.generate_test_runner\": false, \"cuda.upload_to_binary_remote_cache\": false, \"cuda.use_binary_remote_cache\": true, \"cuda.use_fast_math\": false, \"cuda.version\": null, \"cuda_backend\": \"triton\", \"dce\": false, \"debug\": false, \"debug_fusion\": false, \"debug_index_asserts\": false, \"debug_ir_traceback\": false, \"decompose_mem_bound_mm\": false, \"developer_warnings\": true, \"disable_cpp_codegen\": false, \"disable_padding_cpu\": true, \"disable_progress\": true, \"dynamic_scale_rblock\": true, \"efficient_conv_bn_eval_fx_passes\": false, \"emulate_precision_casts\": false, \"enable_auto_functionalized_v2\": true, \"enable_caching_generated_triton_templates\": true, \"enable_linear_binary_folding\": false, \"enabled_metric_tables\": \"\", \"env_str\": null, \"epilogue_fusion\": true, \"epilogue_fusion_first\": false, \"estimate_op_runtime\": \"default\", \"external_matmul\": [], \"fallback_random\": false, \"force_fuse_int_mm_with_mul\": false, \"force_layout_optimization\": false, \"force_pointwise_cat\": false, \"force_same_precision\": false, \"force_shape_pad\": false, \"freezing\": false, \"freezing_discard_parameters\": false, \"fx_graph_cache\": true, \"fx_graph_remote_cache\": null, \"fx_passes_numeric_check\": {\"num_iterations\": 1, \"pre_grad\": false, \"precision\": 0.0001, \"requires_optimizer\": true}, \"fx_wrapper\": false, \"generate_intermediate_hooks\": false, \"global_cache_dir\": null, \"graph_partition\": false, \"group_fusion\": false, \"halide.asserts\": false, \"halide.cpu_target\": \"host\", \"halide.debug\": false, \"halide.gpu_target\": \"host-cuda\", \"halide.scan_kernels\": false, \"halide.scheduler_cpu\": \"Adams2019\", \"halide.scheduler_cuda\": \"Anderson2021\", \"implicit_fallbacks\": true, \"inplace_buffers\": true, \"inplace_padding\": true, \"inter_node_bw\": 25, \"intra_node_bw\": 300, \"is_nightly_or_source\": false, \"is_predispatch\": false, \"joint_custom_post_pass\": null, \"joint_custom_pre_pass\": null, \"joint_graph_constant_folding\": true, \"keep_output_stride\": true, \"kernel_name_max_ops\": 10, \"layout_opt_default\": \"1\", \"layout_optimization\": true, \"log_tlparse\": false, \"loop_ordering_after_fusion\": false, \"max_autotune\": false, \"max_autotune_conv_backends\": \"ATEN,TRITON\", \"max_autotune_flex_search_space\": \"DEFAULT\", \"max_autotune_gemm\": false, \"max_autotune_gemm_backends\": \"ATEN,TRITON,CPP\", \"max_autotune_gemm_search_space\": \"DEFAULT\", \"max_autotune_pointwise\": false, \"max_autotune_prune_choices_based_on_shared_mem\": true, \"max_autotune_report_choices_stats\": true, \"max_autotune_subproc_graceful_timeout_seconds\": 0.0, \"max_autotune_subproc_result_timeout_seconds\": 60.0, \"max_autotune_subproc_terminate_timeout_seconds\": 0.0, \"max_epilogue_benchmarked_choices\": 1, \"max_fusion_buffer_group_pairwise_attempts\": 64, \"max_fusion_size\": 64, \"max_pointwise_cat_inputs\": 8, \"memory_planning\": false, \"memory_pool\": \"intermediates\", \"min_num_split\": 0, \"mixed_mm_choice\": \"heuristic\", \"multi_kernel_hints\": [], \"nan_asserts\": false, \"non_blocking_remote_cache_write\": true, \"online_softmax\": true, \"optimize_scatter_upon_const_tensor\": true, \"pad_channels_last\": false, \"pad_dynamic_shapes\": false, \"pad_outputs\": false, \"padding_alignment_bytes\": 128, \"padding_stride_threshold\": 1024, \"pattern_matcher\": true, \"permute_fusion\": false, \"pick_loop_orders\": true, \"post_grad_custom_post_pass\": null, \"post_grad_custom_pre_pass\": null, \"post_grad_fusion_options\": {}, \"pre_grad_custom_pass\": null, \"pre_grad_fusion_options\": {}, \"precompilation_timeout_seconds\": 3600, \"profile_bandwidth\": false, \"profile_bandwidth_output\": null, \"profile_bandwidth_regex\": \"\", \"profile_bandwidth_with_do_bench_using_profiling\": false, \"profiler_mark_wrapper_call\": false, \"prologue_fusion\": true, \"quiesce_async_compile_pool\": false, \"realize_acc_reads_size_threshold\": null, \"realize_acc_reads_threshold\": 8, \"realize_opcount_threshold\": 30, \"realize_reads_threshold\": 4, \"remote_gemm_autotune_cache\": false, \"remove_pre_grad_passes\": null, \"reorder_for_compute_comm_overlap\": false, \"reorder_for_compute_comm_overlap_passes\": [\"reorder_compute_for_overlap\", \"sink_waits\", \"raise_comms\"], \"reorder_for_locality\": true, \"reorder_for_peak_memory\": true, \"reorder_iterative_debug_limit_to_reorder\": null, \"reorder_iterative_debug_memory_recompute\": false, \"reorder_prefetch_limit\": null, \"rocm.arch\": [], \"rocm.ck_dir\": null, \"rocm.ck_max_profiling_configs\": null, \"rocm.ck_supported_arch\": [\"gfx90a\", \"gfx942\", \"gfx950\"], \"rocm.ck_tile_max_profiling_configs\": null, \"rocm.compile_opt_level\": \"-O2\", \"rocm.flush_denormals\": true, \"rocm.generate_test_runner\": false, \"rocm.is_debug\": false, \"rocm.kBatch_sweep\": null, \"rocm.n_max_profiling_configs\": null, \"rocm.print_kernel_resource_usage\": false, \"rocm.rocm_home\": null, \"rocm.save_temps\": false, \"rocm.split_k_threshold\": 16, \"rocm.use_fast_math\": true, \"rocm.use_preselected_instances\": false, \"run_jit_post_compile_hook\": false, \"save_args\": false, \"scalar_asserts\": true, \"score_fusion_memory_threshold\": 10, \"search_autotune_cache\": false, \"shape_padding\": true, \"sink_waits_iterative_debug_limit_to_sink\": null, \"size_asserts\": true, \"sleep_sec_TESTING_ONLY\": null, \"split_cat_fx_passes\": true, \"split_reductions\": true, \"static_launch_user_defined_triton_kernels\": false, \"static_weight_shapes\": true, \"strict_static_cuda_launcher\": false, \"test_configs.autotune_choice_desc_regex\": null, \"test_configs.autotune_choice_name_regex\": null, \"test_configs.force_extern_kernel_in_multi_template\": false, \"test_configs.graphsafe_rng_func_ignores_fallback_random\": false, \"test_configs.max_mm_configs\": null, \"test_configs.runtime_triton_dtype_assert\": false, \"test_configs.static_cpp_dtype_assert\": false, \"test_configs.track_memory_lifecycle\": null, \"test_configs.use_libtorch\": false, \"torchinductor_worker_logpath\": \"\", \"trace.compile_profile\": false, \"trace.debug_dir\": null, \"trace.debug_log\": false, \"trace.dot_graph_shape\": null, \"trace.draw_orig_fx_graph\": false, \"trace.enabled\": false, \"trace.fx_graph\": true, \"trace.fx_graph_transformed\": true, \"trace.graph_diagram\": false, \"trace.info_log\": false, \"trace.ir_post_fusion\": true, \"trace.ir_pre_fusion\": true, \"trace.log_autotuning_results\": false, \"trace.log_url_for_graph_xform\": null, \"trace.output_code\": true, \"trace.provenance_tracking_level\": 1, \"trace.save_real_tensors\": false, \"trace.upload_tar\": null, \"triton.autotune_at_compile_time\": null, \"triton.autotune_cublasLt\": true, \"triton.autotune_pointwise\": true, \"triton.autotune_with_sample_inputs\": false, \"triton.coalesce_tiling_analysis\": false, \"triton.codegen_upcast_to_fp32\": true, \"triton.cooperative_reductions\": false, \"triton.cudagraph_capture_sizes\": null, \"triton.cudagraph_dynamic_shape_warn_limit\": 50, \"triton.cudagraph_or_error\": false, \"triton.cudagraph_skip_dynamic_graphs\": false, \"triton.cudagraph_support_input_mutation\": false, \"triton.cudagraph_trees\": true, \"triton.cudagraph_trees_history_recording\": false, \"triton.cudagraph_unexpected_rerecord_limit\": 128, \"triton.cudagraphs\": false, \"triton.debug_sync_graph\": false, \"triton.debug_sync_kernel\": false, \"triton.decompose_k_threshold\": 32, \"triton.dense_indexing\": false, \"triton.descriptive_names\": \"original_aten\", \"triton.disallow_failing_autotune_kernels_TESTING_ONLY\": false, \"triton.divisible_by_16\": true, \"triton.enable_pdl\": false, \"triton.enable_persistent_tma_matmul\": false, \"triton.fast_path_cudagraph_asserts\": false, \"triton.force_cooperative_reductions\": false, \"triton.force_cudagraph_sync\": false, \"triton.force_cudagraphs_warmup\": false, \"triton.inject_relu_bug_TESTING_ONLY\": null, \"triton.max_tiles\": null, \"triton.min_split_scan_rblock\": 256, \"triton.multi_kernel\": 0, \"triton.num_decompose_k_splits\": 10, \"triton.persistent_reductions\": true, \"triton.prefer_nd_tiling\": false, \"triton.skip_cudagraph_warmup\": false, \"triton.skip_l1_cache\": false, \"triton.slow_path_cudagraph_asserts\": true, \"triton.spill_threshold\": 16, \"triton.store_cubin\": false, \"triton.tile_reductions\": false, \"triton.tiling_prevents_pointwise_fusion\": true, \"triton.tiling_prevents_reduction_fusion\": true, \"triton.unique_kernel_names\": true, \"triton.unique_user_kernel_names\": false, \"triton.use_block_ptr\": false, \"triton.use_tensor_descriptor\": false, \"triton_kernel_default_layout_constraint\": \"needs_fixed_stride_order\", \"unbacked_symint_fallback\": 8192, \"unroll_reductions_threshold\": 8, \"unsafe_ignore_unsupported_triton_autotune_args\": false, \"unsafe_marked_cacheable_functions\": {}, \"unsafe_skip_cache_dynamic_shape_guards\": false, \"use_experimental_benchmarker\": false, \"use_fast_math\": false, \"use_mixed_mm\": true, \"use_static_cuda_launcher\": true, \"verbose_progress\": false, \"warn_mix_layout\": false, \"worker_log_path\": \"/logs/dedicated_log_torch_compile_worker_rank\", \"worker_start_method\": \"subprocess\", \"worker_suppress_logging\": true}", "remote_cache_version": 14, "inductor_fx_remote_cache_hit_count": null, "inductor_fx_remote_cache_miss_count": null, "inductor_fx_remote_cache_backend_type": "_ManifoldCache", "inductor_fx_remote_cache_hit_keys": null, "inductor_fx_remote_cache_miss_keys": null, "cuda_version": "12.4.0", "triton_version": "3.3.2+fb", "feature_usage": {"fx_cache": false, "parallel_compile_post_warmup": true}, "compile_time_autotune_time_us": null, "is_runtime": false, "gc_time_us": null, "tensorify_float_attempt": null, "tensorify_float_success": null, "tensorify_float_failure": null, "guard_latency_us": null, "recompile_reason": null, "num_graph_breaks": 0, "triton_kernel_compile_times_us": "[[\"triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_7\", 1308640], [\"triton_poi_fused_convolution_gelu_0\", 1254263], [\"triton_red_fused__to_copy_add_addmm_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_9\", 1230700], [\"triton_red_fused__to_copy_add_amax_amin_clamp_convolution_div_gelu_maximum_minimum_mul_native_layer_norm_permute_reciprocal_round_rsub_sub_view_zeros_like_3\", 1227301], [\"triton_poi_fused__to_copy_mul_sub_view_4\", 1196146], [\"triton_poi_fused__to_copy_mul_sub_view_18\", 1186781], [\"triton_poi_fused__to_copy_mul_sub_view_10\", 1185446], [\"triton_poi_fused__to_copy_mul_sub_view_16\", 1184074], [\"triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_5\", 1183541], [\"triton_poi_fused__scaled_dot_product_efficient_attention__unsafe_view_addmm_clone_mul_transpose_view_6\", 1171123], [\"triton_red_fused_add_addmm_native_layer_norm_view_14\", 94186], [\"triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_12\", 75710], [\"triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_13\", 75641], [\"triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_11\", 70518], [\"triton_red_fused_add_convolution_gelu_native_layer_norm_permute_1\", 68299], [\"triton_red_fused__to_copy_add_amax_amin_clamp_div_gelu_maximum_minimum_mul_reciprocal_round_rsub_sub_view_zeros_like_17\", 67450], [\"triton_red_fused__to_copy_add_amax_amin_clamp_div_maximum_minimum_mul_reciprocal_round_rsub_sub_transpose_view_zeros_like_8\", 67031], [\"triton_red_fused__to_copy_add_addmm_amax_amin_clamp_div_maximum_minimum_mul_native_layer_norm_reciprocal_round_rsub_sub_view_zeros_like_15\", 66948], [\"triton_per_fused_add_convolution_gelu_native_layer_norm_permute_2\", 63806]]", "ir_count": null, "cudagraph_skip_reason": null, "python_version": "3.12.11+meta (3.12:55fee9c, Jun 03 2025, 15:41:33) [Clang 17.0.4 (mononoke://mononoke.internal.tfbnw.net/fbsource a35d7733ac346d2f5", "pgo_put_remote_code_state_time_us": null, "pgo_get_remote_code_state_time_us": null, "param_numel": null, "param_bytes": null, "param_count": null, "recompile_user_contexts": null, "inline_inbuilt_nn_modules_candidate": false}, "stack": [{"line": 39, "name": "<module>", "filename": 0, "loc": "__invoke_main()"}, {"line": 36, "name": "__invoke_main", "filename": 0, "loc": "run_as_main(module, main_function)"}, {"line": 105, "name": "run_as_main", "filename": 1, "loc": "oss_run_as_main("}, {"line": 70, "name": "run_as_main", "filename": 2, "loc": "runpy._run_module_as_main(main_module, alter_argv=False)"}, {"line": 198, "name": "_run_module_as_main", "filename": 3, "loc": ""}, {"line": 88, "name": "_run_code", "filename": 3, "loc": ""}, {"line": 151, "name": "<module>", "filename": 4, "loc": "main()"}, {"line": 135, "name": "main", "filename": 4, "loc": "original_out, aoti_out = process_model(model_file, model_name)"}, {"line": 72, "name": "process_model", "filename": 4, "loc": "torch._inductor.aoti_compile_and_package(cuda_ep, package_path=output_path) # , inductor_configs={\"aot_inductor.force_mmap_weights\": True}"}, {"line": 151, "name": "aoti_compile_and_package", "filename": 17, "loc": "return aot_inductor_minifier_wrapper("}, {"line": 1290, "name": "aot_inductor_minifier_wrapper", "filename": 18, "loc": "return func("}, {"line": 194, "name": "_aoti_compile_and_package_inner", "filename": 17, "loc": "aoti_files = aot_compile(gm, args, kwargs, options=inductor_configs)"}, {"line": 301, "name": "aot_compile", "filename": 17, "loc": "return compile_fx_aot("}, {"line": 1938, "name": "compile_fx_aot", "filename": 19, "loc": "get_metrics_context(),"}, {"line": 101, "name": "__exit__", "filename": 47, "loc": "self._on_exit("}, {"line": 1647, "name": "record_compilation_metrics", "filename": 46, "loc": "torch._logging.trace_structured("}]}
V0910 09:44:16.940000 2780682 /data/users/shangdiy/fbsource/fbcode/caffe2/torch/_dynamo/utils.py:1952] {"chromium_event": {}, "has_payload": "8ca67e5b7a68d1ec1d876969d000eb5e"}
	{
	"name": "compile_fx_aot",
	"ts": 1757522656940762.5,
	"args": {
	"compile_id": "None",
	"num_graph_breaks": 0,
	"frame_key": null,
	"co_name": null,
	"co_filename": null,
	"co_firstlineno": null,
	"cache_size": null,
	"accumulated_cache_size": null,
	"guard_count": null,
	"shape_env_guard_count": null,
	"graph_op_count": null,
	"graph_node_count": null,
	"graph_input_count": null,
	"fail_type": null,
	"fail_reason": null,
	"fail_user_frame_filename": null,
	"fail_user_frame_lineno": null,
	"non_compliant_ops": null,
	"compliant_custom_ops": null,
	"restart_reasons": null,
	"dynamo_time_before_restart_s": null,
	"has_guarded_code": null,
	"dynamo_config": "{\"_autograd_backward_strict_mode_conditional_banned_ops\": [\"stride\", \"storage_offset\", \"is_contiguous\"], \"_unsafe_skip_fsdp_module_guards\": false, \"accumulated_recompile_limit\": 256, \"allow_empty_graphs\": false, \"allow_ignore_mark_dynamic\": false, \"allow_rnn\": false, \"allow_unspec_int_on_nn_module\": false, \"allowed_functions_module_string_ignorelist\": [\"torch._decomp\", \"torch._prims\", \"torch._refs\", \"torch.distributions\", \"torch.testing\"], \"assume_dunder_attributes_remain_unchanged\": true, \"assume_static_by_default\": true, \"automatic_dynamic_local_pgo\": true, \"automatic_dynamic_remote_pgo\": null, \"automatic_dynamic_shapes\": true, \"automatic_dynamic_shapes_mark_as\": \"dynamic\", \"caching_precompile\": false, \"capture_autograd_function\": true, \"capture_dynamic_output_shape_ops\": false, \"capture_func_transforms\": true, \"capture_scalar_outputs\": false, \"capture_sparse_compute\": false, \"compiled_autograd\": false, \"compiled_autograd_kwargs_override\": {}, \"cprofile\": false, \"cudagraph_backend_keep_input_mutation\": false, \"cudagraph_backend_support_input_mutation\": false, \"dead_code_elimination\": true, \"debug_disable_compile_counter\": false, \"debug_force_graph_break_on_leaf_return\": false, \"debug_force_nested_calls\": false, \"disable\": false, \"do_not_emit_runtime_asserts\": false, \"dont_skip_tracing\": false, \"dynamic_shapes\": true, \"enable_aot_compile\": false, \"enable_compiler_collectives\": false, \"enable_cpp_framelocals_guard_eval\": true, \"enable_cpp_guard_manager\": true, \"enable_cpp_symbolic_shape_guards\": false, \"enable_faithful_generator_behavior\": true, \"enable_trace_contextlib\": true, \"enable_trace_unittest\": false, \"error_on_nested_fx_trace\": true, \"error_on_nested_jit_trace\": true, \"error_on_recompile\": false, \"fail_on_recompile_limit_hit\": false, \"fake_tensor_cache_crosscheck_enabled\": false, \"fake_tensor_cache_enabled\": true, \"fake_tensor_disable_inference_mode\": true, \"force_nn_module_property_static_shapes\": true, \"force_parameter_static_shapes\": true, \"force_unspec_int_unbacked_size_like_on_torchrec_kjt\": false, \"graph_break_on_nn_param_ctor\": true, \"graph_deduplication_lint\": false, \"guard_nn_modules\": true, \"guard_nn_modules_using_dict_tags\": true, \"inline_inbuilt_nn_modules\": true, \"install_free_tensors\": false, \"issue_3_13_0_warning\": true, \"max_saved_pointers_for_recursive_dict_tags_check\": 256, \"minimum_call_count\": 1, \"nested_graph_breaks\": false, \"numpy_default_complex\": \"complex128\", \"numpy_default_float\": \"float64\", \"numpy_default_int\": \"int64\", \"only_allow_pt2_compliant_ops\": false, \"optimize_ddp\": true, \"optimize_ddp_lazy_compile\": false, \"prefer_deferred_runtime_asserts_over_guards\": false, \"prepare_freezing\": false, \"pt2_compile_id_prefix\": null, \"raise_on_ctx_manager_usage\": true, \"raise_on_unsafe_aot_autograd\": false, \"recompile_limit\": 8, \"record_compile_time_instruction_count\": false, \"record_runtime_overhead\": true, \"replay_record_enabled\": false, \"report_guard_failures\": true, \"rewrite_assert_with_torch_assert\": true, \"run_gc_after_compile\": true, \"skip_code_recursive_on_recompile_limit_hit\": true, \"skip_fsdp_guards\": true, \"skip_fsdp_hooks\": true, \"skip_guards_on_constant_func_defaults\": true, \"skip_nnmodule_hook_guards\": true, \"skip_no_tensor_aliasing_guards_on_parameters\": true, \"skip_tensor_guards_with_matching_dict_tags\": true, \"skip_torchrec\": true, \"skipfiles_inline_module_allowlist\": {}, \"specialize_float\": false, \"specialize_int\": false, \"strict_precompile\": false, \"suppress_errors\": false, \"trace_numpy\": true, \"track_nodes_for_deduplication\": false, \"use_graph_deduplication\": false, \"use_lamba_guard_for_object_aliasing\": true, \"use_lazy_graph_module\": true, \"use_numpy_random_stream\": false, \"use_recursive_dict_tags_for_guards\": true, \"verify_correctness\": false, \"wrap_top_frame\": false}"
	},
	"ph": "E",
	"cat": "dynamo_timed",
	"tid": 0,
	"pid": 0
	}
